Datasets:

kilian-group
/

arxiv-classifier

Dataset card Data Studio Files Files and versions Community

Subset (4)

default · ~99.6k rows (showing the first 120k)

Split (2)

train · ~72.4k rows (showing the first 93.1k)

paper_id stringlengths 9 16	version stringclasses 26 values	yymm stringclasses 311 values	created timestamp[s]	title stringlengths 6 335	secondary_subfield sequencelengths 1 8	abstract stringlengths 25 3.93k	primary_subfield stringclasses 124 values	field stringclasses 20 values	fulltext stringlengths 0 2.84M
1806.09053	2	1806	2019-06-11T10:00:21	Arveson extreme points span free spectrahedra	[ "math.OA" ]	Let $ SM_n(\mathbb{R})^g$ denote $g$-tuples of $n \times n$ real symmetric matrices. Given tuples $X=(X_1, \dots, X_g) \in SM_{n_1}(\mathbb{R})^g$ and $Y=(Y_1, \dots, Y_g) \in SM_{n_2}(\mathbb{R})^g$, a matrix convex combination of $X$ and $Y$ is a sum of the form \[ V_1^* XV_1+V_2^* Y V_2 \quad \quad \quad V_1^* V_1+V_2^* V_2=I_n \] where $V_1:\mathbb{R}^n \to \mathbb{R}^{n_1}$ and $V_2:\mathbb{R}^n \to \mathbb{R}^{n_2}$ are contractions. Matrix convex sets are sets which are closed under matrix convex combinations. A key feature of matrix convex combinations is that the $g$-tuples $X, Y$, and $V_1^* XV_1+V_2^* Y V_2$ do not need to have the same size. As a result, matrix convex sets are a dimension free analog of convex sets. While in the classical setting there is only one notion of an extreme point, there are three main notions of extreme points for matrix convex sets: ordinary, matrix, and absolute extreme points. Absolute extreme points are closely related to the classical Arveson boundary. A central goal in the theory of matrix convex sets is to determine if one of these types of extreme points for a matrix convex set minimally recovers the set through matrix convex combinations. This article shows that every real compact matrix convex set which is defined by a linear matrix inequality is the matrix convex hull of its absolute extreme points, and that the absolute extreme points are the minimal set with this property. Furthermore, we give an algorithm which expresses a tuple as a matrix convex combination of absolute extreme points with optimal bounds. Similar results hold when working over the field of complex numbers rather than the reals.	math.OA	math	ARVESON EXTREME POINTS SPAN FREE SPECTRAHEDRA ERIC EVERT1 AND J. WILLIAM HELTON1 Abstract. Let SMn(R)g denote g-tuples of n × n real symmetric matrices. Given tu- ples X = (X1, . . . , Xg) ∈ SMn1(R)g and Y = (Y1, . . . , Yg) ∈ SMn2(R)g, a matrix convex combination of X and Y is a sum of the form V ∗ 1 XV1 + V ∗ 2 Y V2 V ∗ 1 V1 + V ∗ 2 V2 = In where V1 : Rn → Rn1 and V2 : Rn → Rn2 are contractions. Matrix convex sets are sets which are closed under matrix convex combinations. A key feature of matrix convex combinations 2 Y V2 do not need to have the same size. As a is that the g-tuples X, Y , and V ∗ result, matrix convex sets are a dimension free analog of convex sets. 1 XV1 + V ∗ While in the classical setting there is only one notion of an extreme point, there are three main notions of extreme points for matrix convex sets: ordinary, matrix, and absolute ex- treme points. Absolute extreme points are closely related to the classical Arveson boundary. A central goal in the theory of matrix convex sets is to determine if one of these types of extreme points for a matrix convex set minimally recovers the set through matrix convex combinations. This article shows that every real compact matrix convex set which is defined by a linear matrix inequality is the matrix convex hull of its absolute extreme points, and that the absolute extreme points are the minimal set with this property. Furthermore, we give an algorithm which expresses a tuple as a matrix convex combination of absolute extreme points with optimal bounds. Similar results hold when working over the field of complex numbers rather than the reals. 1. Introduction This paper concerns extreme points of noncommutative (free) convex sets. In the free setting there are three major notions of an extreme point. We shall study the most restricted class of extreme points, the absolute extreme points, a notion introduced by Kleski [KLS14]. This class of extreme points is closely related to Arveson's notion [A69] of an irreducible Date: June 12, 2019. 2010 Mathematics Subject Classification. Primary 46L07. Secondary 46L07, 90C22. Key words and phrases. matrix convex set, extreme point, dilation theory,linear matrix inequality (LMI), spectrahedron, Arveson boundary, real algebraic geometry. 1Research supported by the NSF grant DMS-1500835. 1 2 E. EVERT AND J.W. HELTON boundary representation of an operator system [KLS14, EHKM18]. Hence the subject at hand goes back about 50 years. Noncommutative convex sets can be described as solution sets to types of linear matrix inequalities (LMIs), the workhorse of semidefinite programming. Next we introduce this special type of LMI. Let A = (A1, A2, . . . , Ag) be a g-tuple of bounded self-adjoint operators on a real or complex Hilbert space H and let H be a real or complex Hilbert space with a nested sequences of subspaces { Hℓ}∞ ℓ=1. We define an affine linear function LA on tuples X = (X1, X2, . . . , Xg) of bounded self-adjoint operators acting on Hℓ for some ℓ by LA(X) = IH ⊗ I H + ΛA(X) = IH ⊗ I H + A1 ⊗ X1 + · · · + Ag ⊗ Xg, and we define DA( Hℓ) to be the set of solutions to the LMI DA( H) := {X ∈ S( Hℓ)g LA(X) pos semidef }. (1.1) Here S( Hℓ)g denotes g-tuples of self-adjoint operators on Hℓ. The set ℓ=1DA( Hℓ) ∪∞ which we arrive at is a type of dimension free set which is operator convex and contains tuples of operators acting on each Hℓ. ℓ=1DA( Hℓ) span ∪∞ A central question is whether operator convex combinations of the absolute extreme ℓ=1DA( Hℓ). We will define absolute extreme points (in a limited points of ∪∞ context) in Section 1.1.1. We remark that every closed matrix convex set can be expressed in the form of equation (1.1) [EW97]. Furthermore, matrix convex sets defined by noncom- mutative polynomial inequalities in matrix variables ("noncommutative semialgebraic sets") can be defined in this form where H is finite dimensional [HM12]. Arveson conjectured that the irreducible boundary representations (in our language the absolute extreme points) span when H and H are Hilbert spaces, see [A69] and [A72]. More on this viewpoint to extreme points is found in Section 4.3. Many years later Dritschel and McCullough [DM05] showed if H is separable and H has cardinality of the second uncountable ordinal, then uncountable combinations of absolute extreme points span. In that paper they say their dilation ideas were seriously influenced by a construction used in Agler's approach to model theory, see [A88]. A decade later Davidson and Kennedy [DK15] gave a complete and positive answer to Arveson's original question. As a consequence, [DK15] shows that when H and H are both separable the absolute extreme points span. The finite dimensional version of the problem has been pursued for some time but until now has remained unsettled. In this paper we prove the finite dimensional version of Arveson's conjecture in the real and complex setting, see Theorem 1.3: If H = Rd and X is a g-tuple of self-adjoint n × n matrices over K = R or C with X in ARVESON EXTREME POINTS SPAN FREE SPECTRAHEDRA 3 DA := ∪nDA(Kn), then X is a finite matrix convex combination of absolute extreme points of DA whose sum of sizes is bounded by n(g + 1) when K = R and by 2n(g + 1) when K = C. The proof is constructive and yields an algorithm for construction, see Section 2.4. In the remainder of this section we introduce our basic definitions and notation and give a precise statement of our main results, Theorem 1.1 and Theorem 1.3. Some definitions just given will be repeated to provide a complete list. 1.1. Notation and definitions. Let K denote either R or C. We will say a matrix is self- adjoint over K to mean the matrix is self-adjoint if K = C or symmetric if K = R. For any positive integers g and n, let SMn(K)g denote the set of g-tuples X = (X1, . . . , Xg) of n × n self-adjoint matrices over K and let SM(K)g denote the set SM(K)g = ∪nSMn(K)g. Simi- larly, for positive integers n, ℓ and g let Mn×ℓ(R)g denote the set of g-tuples β = (β1, . . . , βg) of n × ℓ matrices over K. Say a matrix U ∈ Mn(K) is a unitary if U ∗U = In. Similarly, a matrix V ∈ Mn×m(K) is an isometry if V ∗V = Im. Given a matrix M ∈ Kn×n, a subspace N ⊆ Kn is a reducing subspace if both N and N ⊥ are invariant subspaces of M. That is, N is a reducing subspace for M if MN ⊆ N and MN ⊥ ⊆ N ⊥. A tuple X ∈ SMn(K)g is irreducible over K if the matrices X1, . . . , Xg have no common reducing subspaces in Kn; a tuple is reducible over K if it is not irreducible over K. Given a g-tuple X ∈ SMn(K)g and a matrix W ∈ Mn×m(K) we define the conjugation of X by W by W ∗XW = (W ∗X1W, . . . , W ∗XgW ). If W is a unitary (resp. isometric) conjugation. Given tuples X, Y ∈ SMn(K)g say X and Y are unitarily equivalent, denoted by X ∼u Y , if there exists a unitary matrix U ∈ Mn(K) such that isometry) then we say W ∗XW is a unitary (resp. U ∗XU = Y. A subset Γ ⊆ SM(K)g is closed under unitary conjugation if X ∈ Γ and Y ∼u X implies Y ∈ Γ. We define the set Γ at level n, denoted Γ(n), by Γ(n) = Γ ∩ SMn(K)g. That is, Γ(n) is the set of g-tuples of n × n self-adjoint matrices in Γ. 1.1.1. Matrix convex sets and extreme points. Let K ⊆ SM(K)g. A matrix convex com- bination of elements of K is a finite sum of the form V ∗ i Y iVi k Xi=1 V ∗ i Vi = In k Xi=1 E. EVERT AND J.W. HELTON 4 where Y i ∈ K(ni) for i = 1, . . . , k and Vi is an ni × n matrix with entries in K for each i. If additionally Vi 6= 0 for each i, then the matrix convex combination is said to be weakly proper. If K is closed under matrix convex combinations then K is matrix convex. Matrix convex combinations can equivalently be expressed via isometric conjugation. i=1 be a i Vi = In. Define the g-tuple Y and i=1 ⊆ K be a finite collection of elements of K and let {Vi}k i=1 V ∗ the isometry V by As before, let {Y i}k collection of mappings from Kn to Kni such that Pk V ∗ =(cid:16)V ∗ Y = ⊕k Then i=1Y i 1 k(cid:17) . · · · V ∗ V ∗ i Vi = In. k Xi=1 (1.2) V ∗Y V = V ∗ i Y iVi V ∗V = k Xi=1 i=1 V ∗ In words, V ∗Y V is an isometric conjugation which is equal to the matrix convex combination i Y iVi. A matrix convex combination of the form V ∗Y V is called a compression of Pk Y . Given a set K ⊆ SM(K)g, define the matrix convex hull of K, denoted comatK, to be the smallest matrix convex set in SM(K)g that contains K. Equivalently, comatK is the set of all matrix convex combinations of elements of K. Given a matrix convex set K, say X ∈ K(n) is an absolute extreme point of K if whenever X is written as a weakly proper matrix convex combination X = Pk i Y iVi, then for all i either ni = n and X ∼u Y i or ni > n and there exists a tuple Z i ∈ K such that X ⊕ Z i ∼u Y i. We let ∂absK denote the set of absolute extreme points of K and we call ∂absK the absolute boundary of K. We remark that an absolute extreme point X = (X1, . . . , Xg) has the property that X1, . . . , Xg is an irreducible collection of operators. i=1 V ∗ A matrix convex set K is bounded if there is a real number C > 0 such that C − g Xi=1 X 2 i (cid:23) 0 for every tuple X ∈ K. We say K is closed if K(n) is closed for all n ∈ N and we say K is compact if K is closed and bounded. We emphasize that comatK is not assumed to be closed. 1.1.2. Free spectrahedra. Free spectrahedra are a class of matrix convex sets; they are the solution set of a linear matrix inequality. ARVESON EXTREME POINTS SPAN FREE SPECTRAHEDRA 5 Given a g-tuple A of d × d self-adjoint matrices with entries in K, let ΛA denote the homogeneous linear pencil and let LA denote the monic linear pencil ΛA(x) = A1x1 + · · · + Agxg (1.3) Given a positive integer n ∈ N and an X ∈ SMn(K)g, the evaluation of the monic linear pencil LA on X is defined by LA(x) = Id + A1x1 + · · · + Agxg. LA(X) = Idn + ΛA(X) = Idn + A1 ⊗ X1 + · · · + Ag ⊗ Xg where ⊗ denotes the Kronecker product. The free spectrahedron at level n, denoted DA(Kn), will typically be abbreviated A(n) = {X ∈ SMn(K)g LA(X) (cid:23) 0}. DK The corresponding free spectrahedron is the set ∪nDK A(n) ⊆ SM(K)g. In other words, A = {X ∈ SM(K)g LA(X) (cid:23) 0}. DK For emphasis, the elements of the real free spectrahedron DR metric matrices, while the elements of the complex free spectrahedron DC of complex self-adjoint matrices. A are g-tuples of real sym- A are g-tuples We say a free spectrahedron DK A is closed under complex conjugation if X ∈ DK A implies X = (X 1, . . . , X g) ∈ DK A. Note that when K = R the real free spectrahedron DR A is trivially closed under complex conjugation. See [HKM13], [Z17] and [K+] for further discussion of linear pencils and free spectrahedra. 1.2. Absolute extreme points span. The following theorem, our first main result, shows that every compact free spectrahedron which is closed under complex conjugation is the matrix convex hull of its absolute extreme points. Furthermore, it shows that the absolute boundary is the smallest set of irreducible tuples which is closed under unitary conjugation and spans the free spectrahedron. Theorem 1.1. Assume K = R or C and let DK closed under complex conjugation. Then DK points. In notation, A be a compact free spectrahedron which is A is the matrix convex hull of its absolute extreme A = comat∂absDK DK A. 6 E. EVERT AND J.W. HELTON Furthermore, if K ⊆ SM(K)g is any closed matrix convex set and if E ⊆ K is a set of irreducible tuples which is closed under unitary conjugation and whose matrix convex hull is equal to K, then E contains the absolute boundary of K. In other words, K = comatE ⇒ ∂absK ⊆ E. In this sense the absolute extreme points are the minimal spanning set of a free spectrahedron. Proof. The fact that DK immediately from the forthcoming Theorem 1.3. A is the matrix convex hull of its absolute extreme points follows We now prove the second part of the result. K ⊆ SM(K)g be any closed matrix convex set and let E ⊆ DK A be a set of irreducible tuples which is closed under unitary conjugation and satisfies comatE = K. If ∂absK = ∅ we are done. Otherwise, there is a positive integer n and a tuple X ∈ SMn(K)g such that X ∈ ∂absK(n). By assumption X ∈ comatE, so there must exist a finite collection of tuples {Y i} ⊆ E and contractions Vi : Kn → Kni such that X = finite Xi=1 V ∗ i Y iVi. Since X is an absolute extreme point of K and each Y i is irreducible we conclude that i Y iUi = X. By for each i we have ni = n and there is a unitary Ui : Kn → Kn such that U ∗ assumption E is closed under unitary conjugation, so it follows that X ∈ E. 1.3. Dilations to Arveson extreme points. Our second main result is a more quantita- tive version of Theorem 1.1. 1.3.1. Dilations. Let K ⊆ SM(K)g be a matrix convex set and let X ∈ K(n). If there exists a positive integer ℓ ∈ N and g-tuples β ∈ Mn×ℓ(K)g and γ ∈ SMℓ(K)g such that Y = X β β∗ γ! = X1 β1 β∗ 1 γ1! ,· · · , Xg βg γg!! ∈ K, β∗ g then we say Y is an ℓ-dilation of X. The tuple Y is said to be a trivial dilation of X matrix convex combination of Y in the spirit of equation (1.2). if β = 0. Note that, if V ∗ = (cid:16)In 0(cid:17) , then X = V ∗Y V with V ∗V = In. That is, X is a Given tuples A ∈ SMd(K)g and X ∈ SMn(K)g, we define the dilation subspace of A at X, denoted KK DK A,X, to be A,X = {β ∈ Mn×1(K)g ker LA(X) ⊆ ker ΛA(β∗)}. KK In this definition ker LA(X) and ker ΛA(X) are subspaces of Kdn. The dilation subspace is examined in greater detail in Section 2.1. ARVESON EXTREME POINTS SPAN FREE SPECTRAHEDRA 7 1.3.2. Arveson extreme points span. The Arveson boundary of a matrix convex set K is a classical dilation theoretic object which is closely related to the absolute boundary of K. We say a tuple X ∈ K is an Arveson extreme point of K if K does not contain a nontrivial dilation of X. In other words, X ∈ K is an Arveson extreme point of K if and only if, if (1.4) X β β∗ γ! ∈ K(n + ℓ) for some tuples β ∈ Mn×ℓ(K)g and γ ∈ SMℓ(K)g, then β = 0. A coordinate free definition is as follows. The point X ∈ K(n) is an Arveson boundary point of K if for each m each Y in K(m) and isometry V : Kn → Km such that X = V ∗Y V it follows that V X = Y V . The set of Arveson extreme points of K, denoted by ∂ArvK, is called the Arveson boundary of K. If Y is an Arveson extreme point of K and Y is an (ℓ-)dilation of X ∈ K, then we will say Y is an Arveson (ℓ-)dilation of X. The Arveson and absolute extreme points of a matrix convex set are closely related. Indeed the following theorem shows that a tuple is an absolute extreme point if and only if it is an irreducible Arveson extreme point. Theorem 1.2. Let DK Then X ∈ DK X is Arveson extreme point of DK A. Proof. The original statement and proof of this result is given as [EHKM18, Theorem 1.1 (3)] over the field of complex numbers. A proof for the case where K = R is given in Section 5.2. We comment that the original statement handles more general complex dimension free sets; however, this version is well suited to our needs. A be a free spectrahedron which is closed under complex conjugation. A if and only if X is irreducible over K and A is an absolute extreme point of DK Our next theorem shows that every element of a compact free spectrahedron DK is closed under complex conjugation dilates to the Arveson boundary of DK A. Theorem 1.3. Let A be a g-tuple of self-adjoint matrices with entries in K and let DK compact free spectrahedron which is closed under complex conjugation. Let X ∈ DK A which A be a A(n) with dim KK A,X = ℓ. (1) There exists an integer k ≤ 2ℓ + n ≤ 2ng + n and k-dilation Y of X such that Y is an A. Thus, X is a matrix convex combination of absolute Arveson extreme point of DC extreme points of DC A whose sum of sizes is equal to n + k. (2) Suppose X is a tuple of real symmetric matrices, then there exists an integer k ≤ ℓ ≤ A. Thus, X A whose sum of sizes ng and k-dilation Y of X such that Y is an Arveson extreme point of DK is a matrix convex combination of absolute extreme points of DK is equal to n + k. E. EVERT AND J.W. HELTON A is the matrix convex hull of its absolute extreme points. 8 As an immediate consequence, DK Proof. The proof that X ∈ DR 2.3.1. We prove that X ∈ DC We now prove that DK X ∈ DK Arveson extreme point Y ∈ DK compression of Y . A dilates to an Arveson extreme point of DR A dilates to an Arveson extreme point of DC A is the matrix convex hull of its absolute extreme points. Let A. The first part of Theorem 1.3 shows that, in the complex setting, there is an A,X + n such that X is a A is given in Section A in Section 3. A(n + k) for some k ≤ 2 dim KK The g-tuple Y is unitarily equivalent to a direct sum of m irreducible tuples {Y i}m i=1 for some integer m. These too are Arveson, hence absolute, extreme points, see Theorem 1.2. i=1Y i. Equivalently, Since X is a compression of Y , it follows that X is a compression of ⊕m there is an isometry V : Kn → Kn+k such that X = V ∗(⊕m i=1Y i)V . Decomposing V ∗ = (cid:16)V ∗ i=1Y i) gives · · · V ∗ 1 m(cid:17) with respect to the block structure of (⊕m Xi=1 V ∗ i Vi = In Xi=1 V ∗ i Y iVi m m m ni = n + k. Xi=1 (1.5) X = with Y i ∈ DK A(ni) and That is, X is a matrix convex combination of the absolute extreme points Y 1, . . . , Y m. The proof when X is a g-tuple of n × n real symmetric matrices is identical with n + k replaced by n + k where k ≤ dim KK A,X. We comment that there are examples of a free spectrahedron DK A and an Arveson dilation Y of X that has minimal size such that Y is reducible. A and an irreducible tuple X ∈ DK 1.4. Reader's guide. Section 2 introduces the notion of a maximal 1-dilation of an element of a free spectrahedron. The main result of this section is Theorem 2.4 which implies that, in the real setting, Arveson dilations of a tuple X ∈ DR A can be constructed by taking a sequence of maximal 1-dilations of X. This result is then used to prove Theorem 1.3 (2) when K = R. The section ends with Proposition 2.5 which gives a numerical algorithm that can be used to construct Arveson dilations of elements of a real free spectrahedron. Section 3 considers the case where K = C and completes the proof of Theorem 1.3. This is accomplished by showing that, when DC A is closed under complex conjugation, the absolute extreme points of DR A. We then show that every element of a complex free spectrahedron which is closed under complex conjugation is a compression of an element of the associated real free spectrahedron. An appeal to Theorem 2.4 completes the proof. In addition this section gives a classification of free spectrahedra which are closed under complex conjugation. A are absolute extreme points of DC ARVESON EXTREME POINTS SPAN FREE SPECTRAHEDRA 9 Section 4 expands on the historical context of our main results. Section 4.1 describes a count on the number of parameters needed to express a tuple as a matrix convex combination of absolute extreme points which is given by Theorem 1.3. Section 4.2 compares our results to results for general matrix convex sets, and Section 4.3 discusses the original terminology and viewpoint of [A69], [DM05], and [DK15]. An appendix, Section 5.1, contains a discussion of the NC LDL∗ calculation which appears in the proof of Theorem 2.4. In addition, the appendix contains a proof of the real analogue of [EHKM18, Theorem 1.1 (3)]. The authors thank Igor Klep and Scott McCullough for comments on the original version of this manuscript. 2. Real free spectrahedra We first consider the case of Theorem 1.3 where X is an element of DR A . We begin with a collection of lemmas and definitions which will play an important role in the proof of this case. 2.1. The dilation subspace. The subspace KK considering the Schur complement, a tuple β ∈ Mn×1(K)g is an element of KK if there is a real number c > 0 and a tuple γ ∈ Rg such that (2.1) A,X is called the dilation subspace since, by A,X if and only Y = X cβ γ! ∈ DK cβ∗ A. The following lemma explains the relationship between the dilation subspace KK A,X and dilations of the tuple X ∈ DK Lemma 2.1. Let DK A in greater detail. A be a free spectrahedron and let X ∈ DK A(n). (1) If β ∈ Mn×1(K)g and Y = X β β∗ γ! ∈ DK A(n + 1) number cγ > 0 such that is a 1-dilation of X, then β ∈ KK A,X. (2) Let β ∈ Mn×1(K)g. Then β ∈ KK A,X if and only if there is a tuple γ ∈ DK γ ! ∈ DK X cγβ A(n + 1). cγβ∗ A(1) real In particular, one may take γ = 0 ∈ Kg. 10 E. EVERT AND J.W. HELTON (3) X is an Arveson extreme point of DK A if and only if dim KK A,X = 0. Proof. Item (1) follows from considering the Schur complement of LA(Y ) for a dilation Y = X β β∗ γ! ∈ DK A(n + 1) Indeed, multiplying LA(X) by permutation matrices, sometimes called canonical of X. shuffles, see [P02, Chapter 8], shows (2.2) LA(Y ) (cid:23) 0 if and only if LA(X) ΛA(β) ΛA(β∗) LA(γ)! (cid:23) 0. Taking the appropriate Schur complement then implies that (2.3) LA(Y ) (cid:23) 0 if and only if LA(γ) (cid:23) 0 and LA(X) − ΛA(β)LA(γ)†ΛA(β∗) (cid:23) 0 where † denotes the Moore-Penrose pseudoinverse. Item (1) is an immediate consequence of equation (2.3). See [EHKM18, Corollary 2.3] for a related argument. We now prove item (2). Note that LA(0) = I, so similar to before using the Schur complement shows if and only if (2.4) Y0 = X cβ 0! ∈ DK cβ∗ A(n + 1) LA(X) − c2ΛA(β)ΛA(β∗) (cid:23) 0. A,X then ker LA(X) ⊆ ker ΛA(β)ΛA(β∗), so picking c small enough so that kc2ΛA(β)ΛA(β∗)k2 If β ∈ KK is less than the smallest nonzero eigenvalue of LA(X) guarantees that inequality (2.4) holds, hence Y0 ∈ DK A(n + 1). The reverse direction is a consequence of item (1). Item (3) follows from items (1) and (2). Remark 2.2. The choice of γ = 0 ∈ Kg in Lemma 2.1 (2) helps simplify the NC LDL∗ calculations used in the forthcoming proof of Theorem 2.4. 2.2. Maximal 1-dilations. An important aspect of the proof of our main result is con- structing dilations which satisfy a notion of maximality. Given a matrix convex set K and a tuple X ∈ K(n), say the dilation Y = X c β γ! ∈ K(n + 1) c β∗ ARVESON EXTREME POINTS SPAN FREE SPECTRAHEDRA 11 is a maximal 1-dilation of X if Y is a 1-dilation of X and β is nonzero and the real number c and tuple γ ∈ Rg are solutions to the sequence of maximization problems c := Maximizer α∈R,γ∈Rg α s.t. X α β γ ! ∈ K(n + 1) α β∗ and γ := A Local Maximizer s.t. X c β c β∗ γ∈Rg kγk γ! ∈ K(n + 1) where k·k denotes the usual norm on Rg. We note that maximal 1-dilations can be computed numerically, see Proposition 2.5. We emphasize that γ produced by the second optimization need only be any local maximizer, and global maximality is not required. Remark 2.3. If K is a compact matrix convex set and X ∈ K is not an Arveson extreme point of K, then a routine compactness argument shows the existence of nontrivial maximal 1-dilations of X. Other notions of maximal dilations (in the infinite dimensional setting) are discussed in [DM05], [A08, Section 2] and [DK15, Section 1]. 2.3. Maximal dilations reduce the dimension of the dilation subspace. Let A ∈ SMd(R)g, let DR A. The following theorem shows that maximal 1-dilations of X reduce the dimension of the dilation subspace. A be a compact real free spectrahedron, and let X ∈ DR Theorem 2.4. Let A ∈ SMd(R)g be a g-tuple of self-adjoint matrices over R such that DR A is a compact real free spectrahedron and let X ∈ DR A(n). Assume X is not an Arveson extreme point of DR A(n + 1) of X. Furthermore, any such Y satisfies A. Then there exists a nontrivial maximal 1-dilation Y ∈ DR dim KR A, Y < dim KR A,X. Proof. Let Y be a maximal 1-dilation of X. Equivalently, choose the dilation Y (choose β and γ) such that and if Y = X β Yc = X c β β∗ γ! is in DR γ! is in DR c β∗ A(n + 1), A(n + 1) E. EVERT AND J.W. HELTON 12 for a tuple γ ∈ Rg and a real number c ∈ R, then c ≤ 1.1 Furthermore, if c = 1 and Y ∈ DR A(n + 1), then there exists an ε > 0 such that kγ − γk < ε implies kγk ≤ kγk. As mentioned in Remark 2.3, the existence of such a Y follows from the assumptions that X is not an Arveson extreme point of DR A is level-wise compact. A and that DR We will show that First consider the subspace dim KR A, Y < dim KR A,X. EA, Y := {η ∈ Mn×1(R)g there exists a σ ∈ Rg so that ker LA( Y ) ⊆ ker ΛA(cid:16)η∗ σ(cid:17)}. In other words EA, Y is the projection ι of KR defined by A, Y σ! = η for η ∈ Mn×1(R)g and σ ∈ Rg. We will show dim EA, Y < dim KR A,X) where ι η EA, Y := ι(KR If η ∈ EA, Y , then there exists a tuple σ ∈ Rg such that A,X. From Lemma 2.1 (2), it follows that there is a real number c > 0 so that setting σ = cσ gives (cid:16)η∗ σ(cid:17) ∈ KR A, Y . Since DR A is matrix convex it follows that X β cη β∗ σ γ cη∗ σ∗ 0 0 0 1! 1 0 0  so Lemma 2.1 (1) shows η ∈ KR (2.5) A,X. In particular this shows X β cη β∗ σ γ cη∗ σ∗ 0   A.   ∈ DR   1 0 0 0 0 1     EA, Y ⊆ KR A,X. = X cη 0! ∈ DR cη∗ A, Now, assume towards a contradiction that dim EA, Y = dim KR A,X. Using equation (2.5) this implies that EA, Y = KR A,X. 1 If Yc is an element of DR A(n + 1) then so is Y−c. For this reason, it is equivalent to require c ≤ 1. ARVESON EXTREME POINTS SPAN FREE SPECTRAHEDRA 13 In particular we have β ∈ EA, Y . It follows that there is a real number c 6= 0 and a tuple σ ∈ Rg so that (2.6) LA  X β c β β∗ σ γ c β∗ σ 0   (cid:23) 0. Using the NC LDL∗-decomposition (up to canonical shuffles) shows that inequality (2.6) holds if and only if LA(X) (cid:23) 0 and the Schur complements (2.7) Id − c2Q (cid:23) 0 and (2.8) where (2.9) It follows that (2.10) and (2.11) LA(γ) − Q − (ΛA(σ) − cQ)∗(cid:0)Id − c2Q(cid:1)† (ΛA(σ) − cQ) (cid:23) 0 Q := ΛA( β∗)LA(X)†ΛA( β). LA(γ) − Q (cid:23) 0 ker[LA(γ) − Q] ⊆ ker[ΛA(σ) − cQ]. Inequalities (2.10) and (2.11) imply that there exists a real number α > 0 such that 0 < α ≤ α implies It follows from this that LA(γ) − Q ± α (ΛA(σ) − cQ) (cid:23) 0. (2.12) LA(γ ± ασ) − (1 ± cα)Q = LA(γ ± ασ) −(cid:16)ΛA(√1 ± cα β∗)LA(X)†ΛA(√1 ± cα β)(cid:17) (cid:23) 0. Since LA(X) (cid:23) 0, equation (2.12) implies X (2.13) LA √1 ± cα β∗ √1 ± cα β γ ± ασ ! (cid:23) 0. Therefore, from our choice of Y , hence of β, we must have √1 ± cα ≤ 1. It follows that cα = 0. However, we have assumed α > 0 and c 6= 0, so this is a contradiction. We conclude dim EA, Y < dim KR A,X. 14 E. EVERT AND J.W. HELTON Now seeking a contradiction assume dim KR A, Y = dim KR A,X. Since dim EA, Y < dim KR A,X, there must exist tuples η ∈ Mn×1(R)g and σ1, σ2 ∈ Rg such that σ1 6= σ2 and so σ1! , η η σ2! ∈ KR A, Y . It follows that (2.14) σ1 − σ2! ∈ KR 0 A, Y . Set σ = σ1 − σ2 6= 0 ∈ Rg. As before, equation (2.14) with Lemma 2.1 (2) implies that there is a real number c 6= 0 ∈ R so that (2.15) X β β∗ γ cσ 0 0 cσ 0 LA    (cid:23) 0. Considering the NC LDL∗ decomposition shows that equation (2.15) holds if and only if (2.16) LA(X) (cid:23) 0 and LA(γ) − Q − c2ΛA(σ)ΛA(σ) (cid:23) 0, where Q = ΛA( β∗)LA(X)†ΛA( β) as before. It follows from this that (2.17) ker[LA(γ) − Q] ⊆ ker ΛA(σ) and LA(γ) − Q (cid:23) 0. This implies that there is a real number α > 0 so that, for all α ∈ R satisfying 0 < α ≤ α, we have LA(γ) − Q ± ΛA(ασ) = LA(γ ± ασ) − Q (cid:23) 0. Since this is the appropriate Schur complement and since LA(X) (cid:23) 0 it follows that (2.18) β LA X β∗ γ ± ασ! (cid:23) 0 whenever 0 < α ≤ α. Therefore, the local maximality of γ implies kγ + ασk ≤ kγk and kγ − ασk ≤ kγk for sufficiently small α ∈ (0, α], a contradiction to the assumptions that α 6= 0 and σ 6= 0. We conclude that dim KR A,X as asserted by Theorem 2.4. < dim KR A, Y ARVESON EXTREME POINTS SPAN FREE SPECTRAHEDRA 15 2.3.1. Proof of Theorem 1.3 for real free spectrahedra. We are now in position to prove Theorem 1.3 in the case where DR Proof of Theorem 1.3 when K = R. Given a tuple X ∈ DR of a k-dilation Y of X such that Y ∈ ∂ArvDR of Theorem 2.4 and Lemma 2.1 (3). A,X = ℓ, the existence A for some k ≤ ℓ is an immediate consequence A is a compact real free spectrahedron. A with dim KR The fact that DR A is the matrix convex hull of its Arveson extreme points, hence of its absolute extreme points, is proved immediately after the statement of Theorem 1.3. 2.4. Numerical computation. Given a compact real free spectrahedron DR algorithm dilates a tuple X ∈ DR or less. A to an Arveson extreme point Y ∈ DR A, the following A,X steps A in dim KR Proposition 2.5. Let A ∈ SMd(R)g be a g-tuple of self-adjoint matrices over R such that A(n), set Y 0 = X. For integers DR A is a compact real free spectrahedron. Given a tuple X ∈ DR k = 0, 1, 2 . . . and while dim KR A,Y k > 0 define Y k+1 := Y k ck( βk)∗ ck βk γk ! where βk is any nonzero element of KR A,Y k and ck := Maximizer c∈R,γ∈Rg c s.t. LA Y k c( βk)∗ c βk γ ! (cid:23) 0, and γk := A Local Maximizer γ∈Rg s.t. LA Y k A,Y ℓ = 0 for some integer ℓ ≤ dim KR ck( βk)∗ ck βk kγk γ ! (cid:23) 0. Then dim KR X. A,X ≤ ng and Y ℓ is an Arveson ℓ-dilation of Proof. This follows from the proof of Theorem 2.4. The optimization over c in Proposition 2.5 is a semidefinite program, while the opti- mization over γ is a local maximization of a convex quadratic over a spectrahedron. 16 E. EVERT AND J.W. HELTON 3. Complex free spectrahedra This section will prove that every element of a compact complex free spectrahedron which is closed under complex conjugation is the matrix convex hull of its absolute extreme points. We begin with a lemma which shows that the set of real elements in the absolute boundary of a complex free spectrahedron DC A which is closed under complex conjugation is exactly equal to the absolute boundary of DR A. Lemma 3.1. Let A be a g-tuple of d × d real symmetric matrices and let X ∈ DC g-tuple of n× n real symmetric matrices. Then X is an Arveson extreme point of DC only if X is an Arveson extreme point of DR A. Proof. It is straightforward to show that X is an Arveson extreme point of DR Arveson extreme point of DC of DR A if X is an A. To prove the converse, assume X is an Arveson extreme point A and let β ∈ Mn×1(C)g be a tuple such that A be a A if and X β β∗ γ! ∈ DC A. By assumption A is a tuple of real symmetric matrices so DC A is closed under complex conjugation. It follows that X β β∗ γ! = X β γ! ∈ DC β ∗ A. Since DC A is convex we conclude that Re(β) Re(β)∗ X 2 X β This matrix has real entries so it is an element of DR Arveson extreme point of DR γ ! = A so we must have Re(β) = 0. 1 β∗ γ! + X β γ!! ∈ DC β ∗ A. A. However, X was assumed to be an Now, DC A is closed under unitary conjugation so we know 0 (iβ)∗ γ! = 1 X iβ However, this matrix is in DR have assumed that X is an Arveson extreme point of DR conclude that X is an Arveson extreme point of DC 0 −i! X β β∗ γ! 1 0 A, as claimed. 0 i! ∈ DC A. A since Re(β) = 0 from which it follows that Im(iβ) = 0. We A, so iβ = 0, hence β = 0. We ARVESON EXTREME POINTS SPAN FREE SPECTRAHEDRA 17 Our next lemma gives a list of equalities for the dilation subspace which will be used in proving the bound on the dimension of the absolute extreme points appearing in Theorem 1.3. Lemma 3.2. Let DK for the dilation subspace: A be a real or complex free spectrahedron. The following equalities hold (1) Let X ∈ DK A(n1) and Z ∈ DK A,X⊕Z =n(cid:16)β∗ σ∗(cid:17)∗ KK Additionally, A(n2). Then ∈ M(n1+n2)×1(K)g(cid:12)(cid:12) β ∈ KK A,X and σ ∈ KK A,Zo . (2) Let X ∈ DK KK A,X⊕Z = dim KK A,Z. A(n) and let U ∈ Mn(K) be a unitary. Then KK A,X + dim KK A,X = U ∗KK dim KK A,U ∗XU and A,X = dim KK A,U ∗XU . (3) Assume DK A is closed under complex conjugation. Then KK A,X = KK A,X and dim KK A,X = dim KK A,X . Proof. The proof of item (1) is immediate from the fact that ker LA(X⊕Z) ⊆ ker ΛA(cid:16)β∗ σ∗(cid:17) if and only if ker LA(X) ⊆ ker ΛA(β∗) and ker LA(Z) ⊆ ker ΛA(σ∗). To prove item (2) let U ∈ Mn(K) be a unitary and observe that 1! X β X β β∗ γ! ∈ DK To prove item (3): assume DK β∗ γ! ∈ DK X β A ⇐⇒ U ∗XU U ∗β A is closed under complex conjugation. Then A ⇐⇒ X β β∗ γ! ∈ DK γ ! = U ∗ 0 β∗U β∗ γ! U 0 0 1! ∈ DK A. A. 0 We now give a classification of free spectrahedra which are closed under complex conju- gation. Lemma 3.3. Let A be a g-tuple of d × d complex self-adjoint matrices. Then the complex free spectrahedron DC A is closed under complex conjugation if and only if there is a g-tuple B of real symmetric matrices of size less than or equal to 2d × 2d such that DC A = DC B. 18 E. EVERT AND J.W. HELTON Proof. We first prove the forwards direction. Let X be a g-tuple of complex self-adjoint matrices. Since DC A if and only if A is closed under complex conjugation we know that X ∈ DC (3.1) LA(X) (cid:23) 0 and LA(X) (cid:23) 0. Thus X ∈ DC A if and only if LA⊕A(X) (cid:23) 0. Write A = S + iT where S is a tuple of n × n real symmetric matrices and T is a tuple of n × n real skew symmetric matrices. Then A ⊕ A is unitarily equivalent to the g-tuple of real symmetric matrices B defined by (3.2) B := S −T S ! = U ∗ S + iT T 0 0 S − iT! U where U ∈ M2n(C) is the unitary U = √2 2 In iIn iIn In! . We conclude that X ∈ DC A if and only if LB(X) (cid:23) 0. It follows that DC A = DC B. The converse is straightforward. We are now in position to complete the proof of the Theorem 1.3. Proof of Theorem 1.3. Let DC under complex conjugation and let X ∈ DC of generality assume that A is a g-tuple of real symmetric matrices. Set ℓ = dim KC is an element of DR X dilates to an Arveson extreme point Y ∈ DC from Theorem 2.4 with Lemma 3.1. A be a compact complex free spectrahedron which is closed A(n). In light of Lemma 3.3, we may without loss A,X. If X A, that is, if X is a tuple of real symmetric matrices, then the proof that A(n + k) for some integer k ≤ ℓ is immediate To handle the general case where Im(X) 6= 0, write X = S + iT where S is a g-tuple of n× n real symmetric matrices and T is a g-tuple of n× n real skew symmetric matrices. By assumption DC A. As shown in equation (3.2), the tuple (S + iT )⊕ (S − iT ) is unitarily equivalent to the tuple Z ∈ DC A(2n) defined by A is closed under complex conjugation so we know S − iT ∈ DC It follows that X is a compression of Z. Z := S −T S ! . T ARVESON EXTREME POINTS SPAN FREE SPECTRAHEDRA Observe that Z is a tuple of 2n×2n real symmetric matrices so Z ∈ DC Furthermore, an application of Lemma 3.2 shows that dim KC Theorem 2.4 shows that Z dilates to an Arveson extreme point Z ∈ DR integer k ≤ 2ℓ ≤ 2ng and Lemma 3.1 implies that Z is an Arveson extreme point of DC follows that X is a compression of the Arveson extreme point Z. A,Z = 2ℓ, hence dim KR 19 A implies Z ∈ DR A. A,Z ≤ 2ℓ. A(2n + k) for some A. It As in the real case, the proof that DC points is given immediately after the statement of Theorem 1.3. A is the matrix convex hull of its absolute extreme 4. Remarks This section contains remarks which expand on the historical context of our results. Section 4.1 discusses the number of parameters needed to express a tuple as a matrix convex combination of absolute extreme points, while Section 4.2 explores the relationship between the absolute extreme points of free spectrahedra and of general matrix convex sets. Section 4.3 discusses infinite dimensional operator convex sets in Arveson's original context. 4.1. Parameter counts for (matrix) convex combinations of extreme points. The classical Caratheodory Theorem gives an upper bound on how many terms are required to represent an element of a convex set as a convex combination of its extreme points. Theorem 1.3 is the analog of this for a free convex set. In addition to giving a bound on the number of absolute extreme points needed to express an arbitrary tuple X ∈ DK A(n), Theorem 1.3 gives a bound on the number of parameters needed to express the absolute extreme points appearing in the matrix convex combination for X. Given a compact free spectrahedron DK A, the classical Caratheodory Theorem states that A(n) ⊆ SMn(K)g can be written as a convex combination of dim SMn(K)g + 1 A(n), each an element of SMn(K)g. The maximum number of a tuple X ∈ DK classical extreme points of DK parameters in the extreme points required by this classical representation is (dim SMn(K)g + 1)(dim SMn(K)g) = (n(n + 1)g/2 + 1)(n(n + 1)g/2) = O(n4g2). In contrast, Theorem 1.3 shows that X ∈ DK A(n) can be written as a matrix convex combination of a single Arveson extreme point Y ∈ DK A(n + k) for some integer k ≤ 2ng + n. The maximum parameter count on the Arveson extreme point required in this dimension free representation is dim SM2n(g+1)(K)g = 2(n + ng)(n + ng + 1)g = O(n2g3). This suggests that matrix convex combinations are advantageous over classical convex combinations in terms of the number of parameters needed to store the representation of a 20 E. EVERT AND J.W. HELTON tuple as a (matrix) convex combination of extreme points when n is large but that they are disadvantageous if g is large. 4.2. Absolute extreme points of general matrix convex sets. Let K ⊆ SM(K)g be a compact matrix convex set. It is well known that there is a Hilbert space H and a self-adjoint operator A ∈ B(H) such that K = DK A, i.e., where LA(X) is defined as in the introduction [EW97]. K = {X ∈ SM(K)g LA(X) (cid:23) 0}, While Theorem 1.3 shows every compact real free spectrahedron DR A is spanned by its absolute extreme points, [E18, Theorem 1.2] shows the existence of a compact real matrix convex set DR A which has no finite dimensional absolute extreme points. The critical failure of our proof for a general matrix convex set DR A occurs at equation (2.11) in Theorem 2.4. In Theorem 2.4 the tuple A is finite dimensional, while A being discussed here in Section 4.2 is a tuple of operators acting on H which may be infinite dimensional. Thus, the kernel containment ker[LA(γ) − Q] ⊆ ker[ΛA(σ) − cQ] along with LA(γ) − Q (cid:23) 0 does not imply the existence of a real number α > 0 such that Here Q = ΛA( β∗)LA(X)†ΛA( β) similar to before. LA(γ) − Q ± α(ΛA(σ) − cQ) (cid:23) 0. A concrete example of this failure follows. Let H = ℓ2(N), let M = diag(1/n2) ∈ B(H), and let N = diag(1/n) ∈ B(H). Then M (cid:23) 0 and {0} = ker M ⊆ ker N, however M − αN 6(cid:23) 0 for any real number α > 0. 4.3. Alternative contexts. Much of the literature such as [A69], [DM05], and [DK15] referred to in the introduction takes a different viewpoint than the one here. We now briefly describe the correspondence. Operator convex sets are in one to one correspondence with the set of completely positive maps on an operator system [WW99], an area which has received great interest over the last several decades. Under this correspondence, an absolute extreme point of an operator convex set becomes a boundary representation of an operator system [KLS14]. Arveson's original question was phrased in the setting of completely positive maps on an operator system. In this language, Arveson conjectured that every operator system has ARVESON EXTREME POINTS SPAN FREE SPECTRAHEDRA 21 sufficiently many boundary representations to "completely norm it". Additionally, Arveson conjectured that these boundary representations generate the C ∗-envelope. Roughly speak- ing, the C ∗-envelope of an operator system is the "smallest" C ∗-algebra containing that op- erator system [P02]. In this language, Theorem 1.1 shows that every operator system with a finite-dimensional realization (see [FNT17]) is completely normed by its finite dimensional boundary representations. For further material related to operator systems, completely pos- itive maps, boundary representations, and the C ∗-envelope we direct the reader to [Ham79], [D96], [MS98], [F00], [F04], [FHL18], and [PSS18]. References [A88] J. Agler, An abstract approach to model theory, Surveys of some recent results in operator theory, Vol. II, 1-23, Pitman Res. Notes Math. Ser., 192, Longman Sci. Tech., Harlow, 1988. 2 [A69] W. Arveson: Subalgebras of C ∗-algebras, Acta Math. 123 (1969) 141-224. 1, 2, 9, 20 [A72] W. Arveson: Subalgebras of C ∗-algebras, II, Acta Math. 128 (1972) 271-308. 2 [A08] W. Arveson: The noncommutative Choquet boundary, J. Amer. Math. Soc. 21 (2008) 1065-1084. 11 [D96] K.R. Davidson: C ∗-algebras by example, American Mathematical Soc., 1996. 21 [DK15] K.R. Davidson, M. Kennedy: The Choquet boundary of an operator system, Duke Math. J. 164 (2015) 2989-3004. 2, 9, 11, 20 [DM05] M.A. Dritschel, S.A. McCullough: Boundary representations for families of representations of op- erator algebras and spaces, J. Operator Theory 53 (2005) 159-168. 2, 9, 11, 20 [EW97] E.G. Effros, S. Winkler: Matrix convexity: operator analogues of the bipolar and Hahn-Banach theorems, J. Funct. Anal. 144 (1997) 117-152. 2, 20 [E18] E. Evert: Matrix convex sets without absolute extreme points, Linear Algebra Appl. 537 (2018) 287- 301. 20 [EHKM18] E. Evert, J.W. Helton, I. Klep, S. McCullough: Extreme points of matrix convex sets, free spectrahedra and dilation theory, J. of Geom. Anal. 28 (2018) 1373-1498. 2, 7, 9, 10, 24, 25 [F00] D.R. Farenick: Extremal matrix states on operator systems, J. London Math. Soc. 61 (2000) 885-892. 21 [F04] D.R. Farenick: Pure matrix states on operator systems, Linear Algebra Appl. 393 (2004) 149-173. 21 [FNT17] T. Fritz, T. Netzer, A. Thom: Spectrahedral Containment and Operator Systems with Finite- dimensional Realization, SIAM J. Appl. Algebra Geom. 1 (2017) 556-574. 21 [FHL18] A.H. Fuller, M. Hartz, M. Lupini: Boundary representations of operator spaces, and compact rect- angular matrix convex sets, J. Operator Theory 79 (2018) 139-172. 21 [Ham79] M. Hamana: Injective envelopes of operator systems, Publ. Res. Inst. Math. Sci. 15 (1979) 773-785. 21 [HKM13] J.W. Helton, I. Klep, S. McCullough: The matricial relaxation of a linear matrix inequality, Math. Program. 138 (2013) 401-445. 5 [HM12] J.W. Helton, S. McCullough: Every free basic convex semi-algebraic set has an LMI representation, Ann. of Math. (2) 176 (2012) 979-1013. 2 [HJ12] R.A. Horn, C.R. Johnson: Matrix analysis, Cambridge university press, 2012. 22 E. EVERT AND J.W. HELTON [KLS14] C. Kleski: Boundary representations and pure completely positive maps, J. Operator Theory 71 (2014) 45-62. 1, 2, 20 [K+] T. Kriel: Free spectrahedra, determinants of monic linear pencils and decompositions of pencils, preprint https://arxiv.org/abs/1611.03103. 5 [MS98] P.S. Muhly, B. Solel: "An algebraic characterization of boundary representations" In Nonselfadjoint Operator Algebras, Operator Theory, and Related Topics, Oper. Theory Adv. Appl. 104, Birkauser, Basel, 1998, 189-196. 21 [PSS18] B. Passer, O. Shalit, B. Solel: Minimal and maximal matrix convex sets, J. Funct. Anal. 274 (2018) 3197-3253. 21 [P02] V. Paulsen: Completely bounded maps and operator algebras, Cambridge Studies in Advanced Math- ematics 78, Cambridge University Press, 2002. 10, 21 [WW99] C. Webster and S. Winkler: The Krein-Milman Theorem in Operator Convexity, Trans Amer. Math. Soc. 351 (1999) 307-322. 20 [Z17] A. Zalar: Operator Positivstellenatze for noncommutative polynomials positive on matrix convex sets, J. Math. Anal. Appl. 445 (2017) 32-80. 5 ARVESON EXTREME POINTS SPAN FREE SPECTRAHEDRA 23 5. Appendix The appendix contains an NC LDL∗ formula and the proof of Theorem 1.2 over the reals. 5.1. The NC LDL∗ of block 3 × 3 matrices. This subsection contains a brief discussion of the NC LDL∗ decomposition of the evaluation of a linear pencil LA on a block 3 × 3 matrix. Consider a general block 3 × 3 tuple Z :=  η X β β∗ σ γ η∗ σ∗ ψ where X ∈ SMn1(K)g and γ ∈ SMn2(K)g and ψ ∈ SMn3(K)g and β, η, and σ are each g-tuples of matrices of appropriate size. We know that   η X β β∗ σ γ η∗ σ∗ ψ LA   ∼c.s.    LA(X) ΛA(β) ΛA(η) ΛA(β∗) LA(γ) ΛA(σ) ΛA(η∗) ΛA(σ∗) LA(ψ) =: Z   where ∼c.s. denotes equivalence up to permutations (canonical shuffles). It follows that LA(Z) (cid:23) 0 if and only if Z (cid:23) 0. The NC LDL∗ of Z has as its block diagonal factor D the matrix D =  LA(X) 0 S 0 LA(γ) − ΛA(β∗)LA(X)†ΛA(β) − W ∗S†W 0 0 0 0   where S = LA(ψ) − ΛA(η∗)LA(X)†ΛA(η) W = ΛA(σ∗) − ΛA(η∗)LA(X)†ΛA(β). It follows that LA(Z) (cid:23) 0 if and only if LA(X) (cid:23) 0 and S (cid:23) 0 and LA(γ) − ΛA(β∗)LA(X)†ΛA(β) − W ∗S†W (cid:23) 0. Considering the case where K = R and γ ∈ Rg and ψ = 0 ∈ Rg, hence σ = σ∗ ∈ Rg, and substituting η = c β or η = 0 gives equations (2.8) and (2.16), respectively. 24 E. EVERT AND J.W. HELTON 5.2. Proof of Theorem 1.2 over the real numbers. We now give a proof of Theorem 1.2 over the real numbers. To emphasize the real setting in this subsection we will now use the terms symmetric and orthogonal in favor of self-adjoint and unitary. Recall that a tuple X ∈ SMn(R)g is irreducible over R if the matrices X1, . . . , Xg have no common reducing subspaces in Rn; a tuple is reducible over R if it is not irreducible over R. Lemma 5.1. Let X ∈ SMn(R)g be a g-tuple of real symmetric matrices which is irreducible over R and let W ∈ SMn(R) be a real symmetric matrix which commutes with X. Then W is a constant multiple of the identity. Proof. Let W ∈ SMn(R) be a real symmetric matrix such that W X = XW and let E1, . . . ,Ek ⊆ Rn denote the real eigenspaces of W corresponding to the eigenvalues λ1, . . . , λk of W , respectively. Since X is real and W X = XW , each Ej is a reducing subspace for X. If k > 2, then each Ej is a nontrivial real reducing subspace of X which would imply that X is reducible over R. It follows that k = 1 and W = λ1I. We now prove Theorem 1.2 which is our real analogue of [EHKM18, Theorem 1.1 (3)], Theorem 1.2. The proof over R follows exactly the proof over C in [EHKM18] as we now outline. That an irreducible Arveson extreme point is absolute extreme is a simple argument given in [EHKM18, Section 3.4] based on [EHKM18, Lemma 3.14] which (over R) says the following. Lemma 5.2. Fix positive integer n and m and suppose C ∈ Rn×m is a nonzero matrix, the tuple X ∈ SMn(R)g is irreducible over R and E ∈ SMm(R)g. If CXj = EjC for each j, then C T C is a nonzero multiple of the identity. Moreover, the range of C reduces the set {E1, . . . , Eg} so there is an orthogonal matrix U so that for each j we have U T EjU = X ⊕ Zj for some Zj ∈ SMk(R), where k = m − n. Proof. To prove this statement note that XjC T = C T Ej. It follows that XjC T C = C T EjC = C T CXj. Using Lemma 5.1 with the irreduciblity of {X1, . . . , Xg} shows C T C is a nonzero multiple of the identity and therefore C is a real multiple of an isometry. Further, since CX = EC, the range of C is invariant for E. Since each Ej is symmetric, the range of C reduces each Ej and C, as an isometric mapping into its range is a multiple of an orthogonal matrix. Proof of Theorem 1.2 when K = R. Suppose X is both irreducible over R and in the Arveson boundary of DR A. To prove i EiCi, where each Ci is nonzero, X is an absolute extreme point, suppose X = Pν i=1 C T ARVESON EXTREME POINTS SPAN FREE SPECTRAHEDRA 25 i=1 C T Pν A. In this case, let i Ci = I and Ei ∈ DR C =  C1 ... Cν   and E = E1 ⊕ E2 ⊕ · · · ⊕ Eν and observe that C is an isometry and X = C T EC. Hence, as X is in the Arveson boundary, CX = EC. It follows that CiXk = Ei kCi for each i and k. Thus, by Lemma 5.2, it follows that for each i there is an orthogonal matrix Ui such that U T i EiUi = X ⊕ Z i for some Z i ∈ DR A. Therefore X is an absolute extreme point of DR A. The converse proof that an absolute extreme point of DR A is irreducible over R and Arveson extreme is [EHKM18, Lemma 3.11] and [EHKM18, Lemma 3.13] which while stated over C is unchanged over R. 26 E. EVERT AND J.W. HELTON NOT FOR PUBLICATION Contents 1. Introduction 1.1. Notation and definitions 1.1.1. Matrix convex sets and extreme points 1.1.2. Free spectrahedra 1.2. Absolute extreme points span 1.3. Dilations to Arveson extreme points 1.3.1. Dilations 1.3.2. Arveson extreme points span 1.4. Reader's guide 2. Real free spectrahedra 2.1. The dilation subspace 2.2. Maximal 1-dilations 2.3. Maximal dilations reduce the dimension of the dilation subspace 2.3.1. Proof of Theorem 1.3 for real free spectrahedra 2.4. Numerical computation 3. Complex free spectrahedra 4. Remarks 4.1. Parameter counts for (matrix) convex combinations of extreme points 4.2. Absolute extreme points of general matrix convex sets 4.3. Alternative contexts References 5. Appendix 5.1. The NC LDL∗ of block 3 × 3 matrices 5.2. Proof of Theorem 1.2 over the real numbers Index 1 3 3 4 5 6 6 7 8 9 9 10 11 15 15 16 19 19 20 20 21 23 23 24 27 LA(X), 5 Y is an ℓ-dilation of X, 6 Γ at level n, 3 Γ(n), 3 ∂ absK, 4 ∂ ArvK, 7 DC A, 5 DK A(n), 5 DK A, 5 DR A, 5 comatK, 4 A, Y , 12 E KK A,X , 6 absolute boundary, 4 absolute extreme point, 4 Arveson (ℓ-)dilation, 7 Arveson boundary, 7 Arveson extreme point, 7 bounded, 4 closed, 4 closed under complex conjugation, 5 closed under unitary conjugation, 3 compact, 4 complex free spectrahedron, 5 compression, 4 conjugation of X by W , 3 dilation subspace of DK evaluation, 5 A at X, 6 free spectrahedron, 5 free spectrahedron at level n, 5 homogeneous linear pencil, 5 irreducible, 3 isometry, 3 matrix convex, 4 matrix convex combination, 3 Index matrix convex hull, 4 maximal 1-dilation, 11 monic linear pencil, 5 real free spectrahedron, 5 reducible, 3 reducing subspace, 3 self-adjoint, 3 trivial dilation, 6 unitarily equivalent, 3 unitary, 3 unitary (resp. isometric) conjugation, 3 weakly proper, 4 27 28 E. EVERT AND J.W. HELTON Eric Evert, Department of Mathematics, University of California, San Diego E-mail address: [email protected] J. William Helton, Department of Mathematics, University of California, San Diego E-mail address: [email protected]
1402.0135	2	1402	2016-09-29T01:20:56	Property RD and the Classification of Traces on Reduced Group $C^*$-algebras of Hyperbolic Groups	[ "math.OA" ]	In this paper, we show that if $G$ by a non-elementary word hyperbolic group and $a \in G$ an element, if the conjugacy class of $a$ is infinite, then all traces $\tau:C^_{\text{red}}(G) \to \mathbb{C}$ vanish on $a$. We show that all traces $\theta:C^_{\text{red}}(G) \to \mathbb{C}$ are linear combinations of traces $\chi_g:C^*_{\text{red}}(G) \to \mathbb{C}$ given by \[\chi_{g}=\begin{cases} 1 & g \in C(g) \\ 0 & \text{else}\end{cases}\] where $C(g)$ is the conjugacy class of $g$. To do this, we introduce a new method to study traces, using Sobolev norms and property RD.	math.OA	math	Property RD and the Classification of Traces on Reduced Group C∗-algebras of Hyperbolic Groups Sherry Gong∗ Abstract In this paper, we show that if G is a non-elementary word hyperbolic group, a ∈ G red(G) → C is an element, and the conjugacy class of a is infinite, then all traces τ : C∗ vanish on a. Moreover, we completely classify all traces by showing that traces θ : C∗ red(G) → C are linear combinations of traces χg : C∗ red(G) → C given by 1 h ∈ C(g) else 0 χg(h) = , where g is an element with finite conjugacy class, denoted C(g). We demonstrate these two statements by introducing a new method to study traces that uses Sobolev norms and the rapid decay property. 1 Introduction Let G be a countable discrete group, and let (cid:96)2(G) denote the Hilbert space of square- summable functions G → C. Let B((cid:96)2(G)) denominate the algebra of bounded linear operators on (cid:96)2(G). Elements of G act on (cid:96)2(G) as bounded linear operators, so there is a natural map C[G] → B((cid:96)2(G)), where C[G] refers to the group algebra of G. We define the reduced group C∗-algebra of G, written C∗ redG, to be the operator norm closure of C[G] in B((cid:96)2(G)). The reduced group C∗-algebra plays an important role in noncommutative geometry. Connes's book, [4], provides a panoramic account of the research surrounding such algebras. In this paper, we delve into the study of traces on reduced C∗-algebras on groups. A redG → C, not It is easy to see trace on C∗ necessarily positive, that satisfies τ (xy) = τ (yx) for all x, y ∈ C∗ redG is a (not necessarily positive) bounded linear map τ : C∗ redG. ∗Supported by NSF. 1 that if τ is such a map, then for any g, h ∈ G, we have τ (ghg−1) = τ (h). Moreover, any bounded linear map τ : C∗ redG → C satisfying τ (ghg−1) = τ (h) is a trace. Observe that if τ is a trace on a C∗ algebra, then for any complex number λ, λτ is also a trace. In this article, when we say that a C∗ algebra has unique trace, we will mean it has unique trace up to scalar multiples. In [14], Weinberger and Yu studied the degree of non-rigidity of manifolds using idempo- tents in C∗-algebras of groups associated to torsion elements in those groups, and explored whether they are linearly independent in K-theory. One approach to the question they introduced is to use traces to show that idempotents arising from torsion elements of the group, which constitute what Weinberger and Yu called the "finite part of K-theory" in [14], are linearly independent in K-theory. This approach was explored in [7]. The anal- ysis of idempotents also led to results in the study of the moduli space of positive scalar curvature metrics, as in [15] and [16]. In [13], Powers introduced the question of exploring traces on reduced group C∗- algebras. Furthermore, he proved that if G is a non-abelian free group, then there is only one trace on C∗ redG. De la Harpe also investigated this subject in [2], [5], and [6], where he showed that if G is a torsion-free non-elementary hyperbolic group, then C∗ redG has only one trace. In [1], Arzhantseva and Minasyan derived similar conclusions for relatively hyperbolic groups. In this paper, we introduce a new method to examine traces using property RD. Recall that a length function on a group G is a function l : G → Z≥0, satisfying • l(f g) ≤ l(f ) + l(g), • l(g−1) = l(g), • and l(e) = 0, such that for S a finite subset of the non-negative integers, l−1(S) is also finite. Given such a length function on a group, there is a norm (cid:107) · (cid:107)H s on CG, namely (cid:107)(cid:88) cgg(cid:107)2 H s = (cid:88)cg2(1 + l(g))2s. Let H s(G) denote the completion of CG with respect to this norm. Then we say G has rapid decay property, or property RD, if for some s there is a length function on G and a constant C such that (cid:107)(cid:88) cgg(cid:107)red ≤ C(cid:107)(cid:88) cgg(cid:107)H s. 2 Property RD has important applications to developments pertaining to the Novikov conjecture and to the Baum-Connes conjecture, as studied by Connes and Moscovici in [3]. Jolissaint researched this property extensively in [9] and [10], as did Haagerup, who showed earlier in 1979, in the famous paper [8], that free groups have property RD. For a countable discrete group G with length function l and a subset X ⊂ G, we say X has polynomial growth if for Xl = {gg ∈ X, l(g) = l} and nl = Xl, there is a polynomial P satisfying nl < P (l). When such a P does not exist, X is said to grow super-polynomially. We say X grows at least exponentially if there are a, b, c with a, b > 0 such that nl > aebl + c. In [7], we showed that if the conjugacy class of an element g in a group G with property RD grows super-polynomially, then any trace τ : C∗ redG. We will explain this conclusion in more detail and apply it to hyperbolic groups in section 3, after exploring the growths of conjugacy classes in such groups in section 2. In section 4, we apply the results of section 3 to give some examples of hyperbolic groups with unique trace and some with more than one trace. redG → C vanishes on g ∈ C∗ We will use this method to classify all traces on all non-elementary hyperbolic groups, extending de la Harpe's result on torsion-free non-elementary hyperbolic groups. The author would like to thank Prof. Guoliang Yu for his helpful advice and guidance, and Prof. David Kerr for his suggestion to look into questions regarding unique traces on reduced group C∗-algebras. 2 On conjugacy classes In this section, we prove that if a conjugacy class in a hyperbolic group is infinite, then it grows exponentially. In the next section, we will use this development to find all the aforementioned traces. Let G be a finitely generated discrete group and δ be a positive real number. Consider the metric space obtained by taking the Cayley graph of G, and identifying each edge with the unit interval. We say G is δ-hyperbolic if for any x, y, z ∈ G, [x, y] ⊂ Bδ([y, z] ∪ [z, x]), where [x, y] is the geodesic between x and y in the metric space, and for a subset X ⊂ G, Bδ(X) denotes the set of points a such that there is b ∈ X with d(a, b) < δ. We say that a group G is elementary whenever G contains Z as a finite index subgroup. We analyse the problem of conjugacy class growth in non-elementary hyperbolic groups. 3 Theorem 2.1. Let G be a non-elementary hyperbolic group and let a ∈ G be an element. Assume that C(a), the conjugacy class of a in G is infinite. Then C(a) grows at least exponentially. Remark. Note that there is a well known theorem regarding exponential conjugacy growth in hyperbolic groups, but in that theorem, "conjugacy growth" refers to the growth of the number of conjugacy classes that intersect a certain ball, not the number of elements in a particular conjugacy class, rendering the theorem akin to an opposite of this one. Proof. For t a non-negative real number, we designate Bt the ball in the Cayley graph of G around the identity of radius t. is large, and examine its conjugates gxg−1, where (cid:96)(g) < n grows exponentially in n. Here, (cid:96)(gxg−1) is in the range(cid:2) 5n To show this theorem, we want to consider some element x ∈ C(a) of length n, where n 7 . The number of g with (cid:96)(g) < n 7 , 9n (cid:3), whereupon if all different g in (cid:96)(g) < n that the conjugacy class of a has exponentially many elements in the range. 7 gave rise to different gxg−1, we would be done, by virtue of having shown 7 7 Unfortunately, this is not quite true; there could be g and h with (cid:96)(g), (cid:96)(h) < n 7 , where gxg−1 = hxh−1. To surmount this problem, we will bound how much it can happen. More precisely, we will prove the following lemma: Lemma 2.2. Let G be a non-elementary hyperbolic group and a ∈ G be a fixed element. Let x be in C(a) and n = (cid:96)(x). Let y ∈ C(a) satisfy 5n 7 , m = (cid:96)(y), and k < n be an integer. Then the number of h in G such that 7 < (cid:96)(y) < 9n 7 1. (cid:96)(h) = k 2. hxh−1 = y is O(k). That is, it is bounded above by C · k + C1, where C and C1 are constants that depend only on G and a, and independent of n and k. Thus, the number of elements h ∈ G such that 1. (cid:96)(h) < n 7 2. hxh−1 = y is O(n2), meaning it is bounded by C2·n2 +C3, where C2 and C3 are constants that depend only on G and a. 4 Proof of Lemma. In triangle ABC in a graph, let a, b, and c denote the lengths of the sides; that is, c = (cid:96)(AB), b = (cid:96)(AC) and a = (cid:96)(BC). We say that triangle ABC is δ1-thin if for points D ∈ BC, E ∈ AC and F ∈ AB with (cid:96)(AE) = (cid:96)(AF ) = b+c−a , (cid:96)(BD) = (cid:96)(BF ) = a+c−b , we have (cid:96)(DE), (cid:96)(DF ), (cid:96)(EF ) < δ1. Proposition 2.12 in [12] says that for a δ-hyperbolic graph, there exists δ1 such that all triangles are δ1-thin. Let us replace δ with the max of δ and δ1, rendering both the graph δ-hyperbolic and and (cid:96)(CD) = (cid:96)(CE) = a+b−c 2 2 2 all triangles are δ-thin. Moreover, let us assume n > 5δ, which we may, for there are finitely many cases for n ≤ 5δ, and we can thus shove those cases into the constant C1. Figure 1: Choose and fix geodesics γ from 1 to x, ρ from 1 to h, and ν from 1 to y. Let ρ−1 denote the geodesic obtained by traversing ρ backwards. Note that geodesics may not be unique, especially when there are torsion elements, but here we have fixed them ahead of time. Consider the geodesic picture shown in the diagram, where A = e ∈ G, B = h−1, C = xh−1 and D = hxh−1, and AB, BC, CD, and AD are the geodesics ρ−1, γ, ρ, and ν respectively. (In the picture, the Cayley graph is formed by taking generators S and joining g to sg for s ∈ S. The picture is therefore invariant under the right action of G, so, for example, when we say that BC is γ, we mean that BC is the right translate of γ 5 by h−1.) 2 2 2 Now look at triangle ABC. Let i = (cid:96)(AB)+(cid:96)(AC)−(cid:96)(BC) and G ∈ AC such that (cid:96)(AE) = (cid:96)(AG) = i = (cid:96)(AB)+(cid:96)(AC)−(cid:96)(BC) (cid:96)(AB)+(cid:96)(BC)−(cid:96)(AC) . There are E ∈ AB, F ∈ BC , (cid:96)(BE) = (cid:96)(BF ) = k − i = . Applying that triangle ABC is δ-thin, we see that (cid:96)(EF ), (cid:96)(EG), (cid:96)(F G) < δ. Applying hyperbolicity condition triangle ACD, we obtain that for G ∈ AC, there is H ∈ AD ∪ CD such that (cid:96)(GH) < δ. , and (cid:96)(CG) = (cid:96)(CF ) = (cid:96)(AC)+(cid:96)(BC)−(cid:96)(AB) 2 We claim that H ∈ AD. If not, H would be in CD, but then (cid:96)(AC) ≤ (cid:96)(AG)+(cid:96)(GH)+ (cid:96)(HC), (cid:96)(AG) = (cid:96)(AE) < k, and (cid:96)(HC) < k, whereupon (cid:96)(AC) < 2k + δ. However, (cid:96)(AC) > n − k, from which we deduce n − k < 2k + δ, which implies n < 3k + δ < 3n 7 + δ. The latter is impossible because n > 5δ. Thus H ∈ AD. Ergo, we have H ∈ AD with (cid:96)(EH) < (cid:96)(EG) + (cid:96)(GH) < 2δ. Also, (cid:96)(AE) = (cid:96)(AG), and consequently, (cid:96)(AE) − δ < (cid:96)(AE) − (cid:96)(GH) = (cid:96)(AG) − (cid:96)(GH) < (cid:96)(AH) < (cid:96)(AG) + (cid:96)(GH) = (cid:96)(AE) + (cid:96)(GH) < (cid:96)(AE) + δ. Hence, for fixed x and y, and for fixed i with 0 ≤ i ≤ k, which is the length of AE, we will estimate an upper bound for the number of possible configurations as in the diagram. We have fixed F ∈ BC, by way of the condition (cid:96)(BF ) = (cid:96)(BE) = k − (cid:96)(AE) = k − i, and BC has to be the h−1 right translate of the geodesic for x. In consequence, BF is fixed; it is just the product of the first k − i edges of the geodesic of x. BE, as a group element, has distance at most δ from BF , meaning that there are at most Bδ possibilities for BE. For i fixed, as above, there are at most 2δ possibilities for AH, due to the fact that H is along AD, which is the geodesic of y, and we showed above that i − δ = (cid:96)(AE) − δ < (cid:96)(AH) < (cid:96)(AE) + δ = i + δ. Accordingly, AH is the product of the first t edges of the geodesic of y. There are at most 2δ choices of AH, and for each such choice there are at most B2δ choices of AE, for AE and AH are within 2δ of each other as group elements. Thus, in total, for fixed i, there are at most 2δB2δ possibilities for AE. Consequently, for fixed i = (cid:96)(AE) = k − (cid:96)(BE), there are at most 2δB2δ choices of AE and Bδ choices of BE, which implies that there are at most 2δBδB2δ choices of AB = EB · AE. But AB = h−1, meaning there are at most 2δBδB2δ = O(1) choices of h for a given k and i. There are k choices of i ∈ [0, k], and accordingly there are in total at most k(2δBδB2δ) 6 possible h for a given k, allowing us to deduce the first statement in the lemma. Summing up these values for different k < n/7 is adding up n/7 summands, each of which is O(n), resulting in O(n2) possible h with (cid:96)(h) < n/7 and hxh−1 = y. From here, we complete the proof of the theorem. Observe that for any n > (cid:96)(a), there is an element x ∈ C(a) with (cid:96)(x) = n or (cid:96)(x) = n + 1. This existence appears as a consequence of the presence of y ∈ C(a) with (cid:96)(y) > n + 1, and our ability to write y = sN sN−1 ··· s1as−1 ··· s−1 1 N , where each si is one of the generators. Then k ) ≤ 2 k+1) − (cid:96)(sk ··· s1as−1 ··· s−1 ··· s−1 1 1 (cid:96)(sk+1 ··· s1as−1 and (cid:96)(a) < n, from which we deduce that there is k such that (cid:96)(sk ··· s1as−1 {n, n + 1}. 1 ··· s−1 k ) ∈ Let n > (cid:96)(a) + 1, satisfy that there is some x ∈ C(a) with (cid:96)(x) = n. Then for g ∈ Bn/7, we may consider gxg−1 ∈ C(a). Then 5n/7 < (cid:96)(gxg−1) < 9n/7, as a result of which we can map Bn/7 to the elements of the conjugacy class of a of length in the range(cid:2) 5n (cid:3): 7 , 9n 7 φ : Bn/7 → C(a) ∩ (B9n/7 − B5n/7). By the above lemma, the pre-image of any element under φ has size at most C2·n2 +C3, for constants C2, C3 depending only on a and G. However, Bn/7 grows exponentially by Koubi's theorem (Theorem 1.1 in [11]), which states that non-elementary hyperbolic groups have exponential growth. As a result, the number of elements in the image of φ is bounded below by C · eαn/n2 + D for constants C and D. Since for any n, there is x ∈ C(a) with length either n or n + 1, we get that the conjugacy class has exponential growth, as desired. Corollary 2.3. For G a non-elementary hyperbolic group and a ∈ G a non-torsion element, the conjugacy class of a grows exponentially. Proof. By the theorem, it suffices to show that the conjugacy class is infinite. Let C(a) be the conjugacy class of a and Z(a) be the centraliser. The fact that C(a) = G/Z(a) implies that if the conjugacy class is finite, then Z(a) ⊂ G has finite index. By Theorem 3.35 of [12], however, the centraliser of an infinite order element in a hyperbolic group is a finite extension of the subgroup generated by the element. As a result, Z(a) is a finite extension of (cid:104)a(cid:105), and is thus virtually cyclic. Applying that any group is quasi-isometric 7 to any of its finite index subgroups, we would then have derived that G is also virtually cyclic, resulting in the desired contradiction. 3 On traces In this section, we show the main result; we classify all traces on non-elementary word- hyperbolic groups. Theorem 3.1. For G a non-elementary word-hyperbolic group, the traces θ : C∗ are exactly the linear combinations of χg : C∗ red(G) → C given by red(G) → C 1 g ∈ C(g) 0 else , χg = where g is an element of finite conjugacy class and C(g) is its conjugacy class. Proof of Theorem 3.1. As was shown in [12] [Theorem 3.27], in a hyperbolic group, there are only finitely many conjugacy classes of torsion elements. We demonstrated in the proof of Corollary 2.3 that conjugacy classes of non-torsion elements are infinite. As consequence, there are only finitely many elements that have finite conjugacy class. For each such element g, it is easy to see that the map 1 g ∈ C(g) 0 else χg = defines a bounded linear functional C∗ red(G) → C, and is consequently a trace. To verify that every bounded trace is a linear combination of the finitely many afore- mentioned traces, we will use the fact that word hyperbolic groups have property RD: Definition 3.2. Given a length function on a group, there is a norm (cid:107) · (cid:107)H s on CG given by (cid:107)(cid:88) cgg(cid:107)2 H s = (cid:88)cg2(1 + l(g))2s. Let H s(G) denote the completion of CG with respect to the aforementioned norm. Then we say G has rapid decay property, or property RD if for some s, there is a length function on it and a constant C such that (cid:107)(cid:88) cgg(cid:107)red ≤ C(cid:107)(cid:88) cgg(cid:107)H s. 8 It was shown in [9] that all hyperbolic groups have property RD. We will now apply Lemma 2.8 in [7], which states the following: Theorem 3.3. Let G be a group with property RD and g ∈ G be an element whose redG → C vanish on g. conjugacy class grows faster than polynomially. Then all traces C∗ In our case, word-hyperbolic groups have property RD. Ergo, applying Theorem 2.1, redG → C vanish on all elements with infinite conjugacy class, we have that all traces C∗ and they must be linear combinations of the above traces, as desired. In particular, this gives another proof of de la Harpe's result regarding traces on torsion- free non-elementary word-hyperbolic groups, which can be found in [5]: Corollary 3.4. For G a torsion-free, non-elementary, word-hyperbolic group, there is only one trace τ : C∗ redG → C. Theorem 3.1 also implies that every hyperbolic group can be written as an extension of a hyperbolic group with unique trace by a finite group. More precisely: Theorem 3.5. Let G be a hyperbolic group. Then it has a finite normal subgroup N such that G/N is a hyperbolic group with unique trace. Proof. Let N ⊂ G be the subset consisting of elements with finite conjugacy class. Note that all elements in it have finite order, as we showed in the proof of Theorem 3.1. Lemma 3.6. We claim that N is a subgroup. Proof. It is easy to see that N is closed under inversion, and it therefore suffices to show that if g1 and g2 are in N , then g1g2 is in N . For an element g ∈ G, let Z(g) ⊂ G denote the subgroup consisting of elements that commute with g. Observe that g is in N if and only if Z(g) has finite index as a subgroup of G. Thus, it suffices to show that if Z(g1) and Z(g2) have finite index, then Z(g1g2) does as well. Note that Z(g1g2) contains the intersection of Z(g1) and Z(g2), rendering it sufficient to show that if H1 and H2 are finite index subgroups of G, then H1 ∩ H2 is as well. There is a natural injection H1/H1 ∩ H2 → G/H2. Thus, H1/H1 ∩ H2 < ∞. Let a1, a2, . . . an and b1, b2, . . . bm satisfy G = a1H1 ∪ a2H1 ∪ ··· ∪ anH1 and Then G =(cid:83) H1 = b1(H1 ∩ H2) ∪ b2(H1 ∩ H2) ∪ ··· ∪ bm(H1 ∩ H2). i,j aibj(H1 ∩ H2). Therefore, H1 ∩ H2 has finite index, as desired. 9 By the lemma, we have that N is a subgroup of G. As we determined in the proof of Theorem 3.1, N is finite. Also, it is easy to see that N is closed under conjugation. Hence, N is a finite normal subgroup. Note that G/N is hyperbolic, on the grounds that hyperbolicity is preserved under quasi-isometries, and it is easy to see that the map G → G/N is a quasi isometry. Thus, red(G/N ) → C. G/N is hyperbolic. It remains to show that G/N has unique trace τ : C∗ red(G) → C, because N is a finite normal subgroup. By Such a trace τ induces a trace C∗ red(G) → C vanishes on all elements outside of N ; ergo the map Theorem 3.1, the trace C∗ τ : C∗ red(G/N ) → C vanishes on all elements except the identity, as desired. 4 Examples In this section, we apply Theorem 3.1 to give some examples of hyperbolic groups with torsion whose reduced C∗-algebras have unique trace, and some examples in which the reduced C∗-algebras have more than one trace. Example. Let G = G1 ∗ G2 be the free product of two non-trivial hyperbolic groups G1 and G2. Suppose that G1 > 2.1 Then G is hyperbolic and all its conjugacy classes are infinite. The hyperbolicity results from the general fact that the free product of any two hyperbolic groups is hyperbolic. To see that the conjugacy classes have infinite order: Observe that every element x ∈ G can be uniquely expressed as x = g1h1g2h2 ··· gnhn, where gi ∈ G1, hi ∈ G2 and gi (cid:54)= 1 for i > 1 and hi (cid:54)= 1 for i < n. In the case g1 = 1 and hn (cid:54)= 1, consider some element g (cid:54)= 1 ∈ G1 and h (cid:54)= 1 ∈ G2, and consider hghg ··· hghgxg−1h−1g−1h−1 ··· g−1h−1 = (hg)tx(hg)−t, that is conjugation of x by (hg)t. As t varies amongst positive integers, (hg)tx(hg)−t goes through pairwise distinct elements of the conjugacy class of x in G. The case where g1 (cid:54)= 1 and hn = 1 is similar. For the case g1 (cid:54)= 1, and hn (cid:54)= 1, conjugate by (hg)t, where g ∈ G1 has g (cid:54)= 1 and 1 . Then the same argument as above applies. Similarly, for the case g1 = 1 and g (cid:54)= g−1 hn = 1, pick g ∈ G1, with g (cid:54)= 1 and g (cid:54)= g−1 n ∈ G1. Consequently, we have that G is a hyperbolic group with infinite conjugacy classes, implying, by Theorem 3.1, that whenever such a group is non-elementary, its reduced C∗-algebra has unique trace. 1This assumption does not lose any generality, because Z/2 ∗ Z/2 is elementary. 10 This example verifies that many hyperbolic groups with torsion elements still have reduced C∗-algebras with unique traces, extending de la Harpe's results in [2], [5], and [6], which states that torsion-free non-elementary hyperbolic groups have reduced C∗-algebras with unique traces. Now let us give an example of a non-elementary hyperbolic group that does not have unique trace. For this, by Theorem 3.1, it suffices that there are non-identity elements with finite conjugacy class. There are obvious examples of such groups: Example. Let G by a non-elementary hyperbolic group and G1 be a non-trivial finite group. Then G1× G, the direct product of G1 and G, is hyperbolic, for it is quasi-isometric to G. The subset G1 × e of G1 × G, where e is the identity in G, is clearly preserved under conjugation. It is moreover finite, implying that any element in it has finite conjugacy class. Here is an example of a non-elementary hyperbolic group that does not have only one trace, but which is not the direct product of a finite group and a hyperbolic group with unique trace. Example. Let G = (cid:104)x, y, a(cid:105)/(cid:104)a3, aya−1y−1, xax−1a−2(cid:105). This group G is hyperbolic, in that it is quasi-isometric to the free group on two generators, (cid:104)x, y,(cid:105) ⊂ G. In G, a and a2 clearly have finite conjugacy class, as their conjugacy class is {a, a2}. Thus, by Theorem 3.1, G does not have unique trace. However, it is not the direct product of a finite group and a hyperbolic group whose reduced C∗-algebra has unique trace, because the only elements of finite order in G are a and a2; ergo, if G were such a direct product, it would be Z/3Z × G1 for some hyperbolic group G1, but in that case, a would not be conjugate to a2. References [1] G.N. Arzhantseva, A. Minasyan, Relatively hyperbolic groups are C∗-simple, Journal of Functional Analysis, 243(1) (2007), 345-351. [2] M. Bekka, M. Cowling, P. de la Harpe Some groups whose reduced C∗-algebra is simple, Publications math´ematiques de l'I.H.´E.S., tome 80 (1994), p. 117-134. [3] A. Connes, H. Moscovici. Cyclic cohomology, the novikov conjecture and hyperbolic groups, Topology, 29(3):345-388, 1990 11 [4] A. Connes, Noncommutative Geometry, Academic Press, San Diego, CA, 1994, 661 p., ISBN 0-12-185860-X. [5] P. de la Harpe, Groupes hyperboliques, alg`ebres dop´erateurs et un th´eor`eme de Jolis- saint, C. R. Acad. Sci. Paris S´er. I Math. 307 (1988), no. 14, 771-774. [6] P. de la Harpe. On operator algebras, free groups, and other groups, Ast´erisque 232 (S.M.F. 1995) 121-153. (1995), 121-153. [7] S. Gong. Finite part of operator K-theory for groups with rapid decay, to appear in Journal of Noncommutative Geometry. [8] U. Haagerup, An example of a non-nuclear C∗-algebra which has the metric approxi- mation property, Inventiones Mathematicae 50, (1979) 279-293. [9] P. Jolissaint. Rapidly decreasing functions in reduced C∗-algebras of groups, Trans. Amer. Math. Soc. 317 (1990), 167-196. [10] P. Jolissaint. K-theory of reduced C∗-algebras and rapidly decreasing functions on groups, K-theory, (1989), no. 6, 723-735. [11] M. Koubi, Croissance uniforme dans les groupes hyperboliques, Annales de linstitut Fourier, tome 48, no 5 (1998), p. 1441-1453. [12] P. Papasoglu. Notes on hyperbolic and automatic groups, https://www.math. ucdavis.edu/~kapovich/280-2009/hyplectures_papasoglu.pdf. [13] R.T. Powers, Simplicity of the C∗-algebra associated with the free group on two gen- erators, Duke Math. J. 42 (1975), 151-156. [14] S. Weinberger, G. Yu Finite part of operator K-theory for groups finitely embeddable into Hilbert space and the degree of non-rigidity of manifolds, 2013. [15] Z. Xie and G. Yu. A relative higher index theorem, diffeomorphisms and positive scalar curvature, Advances in Mathematics, 250 (2014) 35-73. [16] Z. Xie and G. Yu. Higher rho invariants and the moduli space of positive scalar cur- vature metrics, arXiv:1310.1136, 2012 Department of Mathematics, Massachusetts Institute of Technology, Cambridge, MA 02139, USA E-mail: [email protected] 12
1201.0709	3	1201	2012-11-28T21:27:35	On enveloping C*-algebras of Hecke algebras	[ "math.OA" ]	We give a sufficient condition for a $^$-algebra with a specified basis to have an enveloping $C^$-algebra. Particularizing to the setting of a Hecke algebra $\h(G,\gm)$, we show that under a suitable assumption not only we can assure that an enveloping $C^$-algebra $C^(G, \gm)$ exists, but also that it coincides with $C^(L^1(G,\gm))$, the enveloping $C^$-algebra of the $L^1$-Hecke algebra. Our methods are used to show the existence of $C^(G, \gm)$ and isomorphism with $C^(L^1(G,\gm))$ for several classes of Hecke algebras. Most of the classes which are known to satisfy these properties are covered by this approach, and we also describe some new ones.	math.OA	math	On Enveloping C∗-Algebras of Hecke Algebras Rui Palma Department of Mathematics, University of Oslo, P.O. Box 1053 Blindern, NO-0316 Oslo, Norway E-mail: [email protected] Abstract We give a sufficient condition for a ∗-algebra with a specified basis to have an enveloping C ∗-algebra. Particularizing to the setting of a Hecke algebra H(G, Γ), we show that under a suitable assumption not only we can assure that an enveloping C ∗-algebra C ∗(G, Γ) exists, but also that it coincides with C ∗(L1(G, Γ)), the enveloping C ∗-algebra of the L1-Hecke algebra. Our methods are used to show the existence of C ∗(G, Γ) and isomorphism with C ∗(L1(G, Γ)) for several classes of Hecke algebras. Most of the classes which are known to satisfy these properties are covered by this approach, and we also describe some new ones. Introduction A Hecke pair (G, Γ) consists of a group G and a subgroup Γ ⊆ G for which every double coset ΓgΓ is the union of finitely many left cosets. In this case Γ is also said to be a Hecke subgroup of G. Examples of Hecke subgroups include finite subgroups, finite-index subgroups and normal subgroups. Hecke subgroups are also sometimes called almost normal subgroups (although we will not use this terminology here) and it is in fact many times insightful to think of this definition as a generalization of the notion of normality of a subgroup. Given a Hecke pair (G, Γ) the Hecke algebra H(G, Γ) is a ∗-algebra of func- tions over the set of double cosets Γ\G/Γ, with a suitable convolution product and involution. It generalizes the definition of the group algebra C(G/Γ) of the quotient group when Γ is a normal subgroup. The interest in Hecke algebras in the realm of operator algebras was to a large extent raised through the work of Bost and Connes [4] on phase transi- tions in number theory, and since then several authors have studied C∗-algebras which arise as completions of Hecke algebras. There are several canonical C∗- completions of a Hecke algebra ([21], [9]) and the question of existence of a maximal one (i.e. an enveloping C∗-algebra) has been of particular interest ([4], [1], [5], [8], [21], [7], [12], [3], [9], [6]). One of the reasons for that, firstly explored Date: November 15, 2018 Research supported by the Research Council of Norway, the Nordforsk research net- work "Operator Algebra and Dynamics" and Fundação para a Ciência e Tecnologia grant SFRH/BD/43276/2008. 1 by Hall [8], has to do with how well ∗-representations of a Hecke algebra H(G, Γ) correspond to unitary representations of the group G that are generated by their Γ-fixed vectors. It was shown by Hall [8] that for such a correspondence to hold it is necessary that the Hecke algebra has an enveloping C∗-algebra, which does not always happen. It was later clarified by Kaliszewski, Landstad and Quigg [9] that such a correspondence holds precisely when an enveloping C∗-algebra exists and coincides with other canonical C∗-completions. The problem of deciding if a Hecke algebra has an enveloping C∗-algebra seems to be of a non-trivial nature, with satisfactory answers, arising from various distinct methods, known only for certain classes of Hecke pairs. The main motivation for the present article is to give a unified approach to this problem for a large class of Hecke pairs. We recover most of the known cases in the literature but also several new ones. We achieve this by associating a directed graph to a Hecke algebra H(G, Γ), whose vertices are the double cosets and whose directed edges are determined by how products of the form (ΓgΓ)∗ ∗ ΓgΓ decompose as sums of double cosets. We prove that finiteness of the co-hereditary set generated by a vertex ΓgΓ, i.e. the set of vertices one encounters by moving forward in the graph starting from ΓgΓ, implies that sup π kπ(ΓgΓ)k < ∞ , where the supremum runs over the ∗-representations of the Hecke algebra. Thus, analysing these co-hereditary sets gives valuable information regarding the ex- istence of enveloping C∗-algebras. Moreover, we prove that if all double cosets generate finite co-hereditary sets, then the enveloping C∗-algebra of H(G, Γ) ex- ists and coincides with C∗(L1(G, Γ)), the enveloping C∗-algebra of the L1-Hecke algebra (one of the canonical C∗-completions). We develop certain tools, based on iterated commutators on the group G, that allow us to show that our assumptions hold in a variety of classes of Hecke pairs, and thus enable us to answer affirmatively the question of existence of enveloping C∗-algebras of the corresponding Hecke algebras. Some of the new results we prove state that if a group G satisfies some generalized nilpotency property, then for any Hecke subgroup Γ the Hecke algebra H(G, Γ) has an enveloping C∗-algebra which coincides with C∗(L1(G, Γ)). These results will enable us to show, in a following article (see [16]), that for any G satisfying such properties, Hall's correspondence holds for any Hecke subgroup. We also notice that the classes of Hecke algebras studied in the present work, and therefore most of those studied in the literature, satisfy a stronger property than just having an enveloping C∗-algebra: they are in fact BG∗-algebras. The standard reference for this class of ∗-algebras is Palmer [18], but we also give a short description in Section 1. The reason for considering this stronger property is not only because of how well-behaved these ∗-algebras are, but also because it is natural to consider BG∗-Hecke algebras in the context of crossed products by Hecke pairs (see [15]). The paper is organized as follows: in Section 1 we set up the conventions, notation, and background results regarding ∗-algebras, Hecke algebras and also directed graphs, that will be used throughout the article. In the setting of directed graphs the most important notion will be that of a co-hereditary set. In Section 2 we associate a directed graph to any ∗-algebra with a given basis and prove our first main result, which states that we can put a bound on 2 the norm of all representations of the elements of a finite co-hereditary set. As a consequence, if the ∗-algebra with the given basis is generated by its finite co-hereditary sets, it must have an enveloping C∗-algebra. In Section 3 we prove the second main result of this article: that in the case of a Hecke algebra H(G, Γ) if all double cosets generate finite co-hereditary sets, then an enveloping C∗-algebra exists and it coincides with C∗(L1(G, Γ)). In Section 4 we present some tools for determining if a co-hereditary set generated by a double coset is finite. These methods will be then used in Section 5 to study the existence of an enveloping C∗-algebra, and the isomorphism with C∗(L1(G, Γ)), for several classes of Hecke pairs (G, Γ). In Section 6 we show that the problem of existence of an enveloping C∗- algebra for H(G, Γ) can be reduced to the same problem but for a smaller Hecke subalgebra H(H, Γ), where Γ ⊆ H ⊆ G is an ascendant subgroup. Finally in Section 7 we give some concluding remarks and state some open questions. The present work is part of the author's Ph.D. thesis [15] written at the University of Oslo. The author would like to thank his advisor Nadia Larsen for the very helpful discussions, suggestions and comments during the elaboration of this work. 1 Preliminaries 1.1 Preliminaries on ∗-Algebras Let V be an inner product space over C. Recall that a function T : V → V is said to be adjointable if there exists a function T ∗ : V → V such that hT ξ , ηi = hξ , T ∗ηi , for all ξ, η ∈ V . Recall also that every adjointable operator T is necessarily linear and that T ∗ is unique and adjointable with T ∗∗ = T . We will use the following notation: • L(V ) denotes the ∗-algebra of all adjointable operators in V • B(V ) denotes the ∗-algebra of all bounded adjointable operators in V . Of course, we always have B(V ) ⊆ L(V ), with both ∗-algebras coinciding when V is a Hilbert space (see, for example, [18, Proposition 9.1.11]). Following [18, Def. 9.2.1], we define a pre-∗-representation of a ∗-algebra A on an inner product space V to be a ∗-homomorphism π : A → L(V ) and a ∗-representation of A on a Hilbert space H to be a ∗-homomorphism π : A → B(H ). As in [17, Def. 4.2.1], a pre-∗-representation π : A → L(V ) is said to be normed if π(A) ⊆ B(V ), i.e. if π(a) is a bounded operator for all a ∈ A. We now make a seemingly similar definition, but where the focus is on the elements of the ∗-algebra, instead of its pre-∗-representations: Definition 1.1. Let A be a ∗-algebra. We will say that an element a ∈ A is au- tomatically bounded if π(a) ∈ B(V ) for any pre-∗-representation π : A → L(V ). 3 Easy examples of automatically bounded elements in a ∗-algebra are uni- taries, projections, or more generally, partial isometries. Given a ∗-algebra A let Ab := {a ∈ A : a is automatically bounded} . Definition 1.2 ([18], Def. 10.1.17). A ∗-algebra A is called a BG∗-algebra if every element a ∈ A is automatically bounded, i.e. if Ab = A. Equivalently, A is a BG∗-algebra if all pre-∗-representations of A are normed. The function k · ku : A → R+ 0 ∪ {∞} defined by π kπ(a)k , kaku := sup where the supremum is taken over all ∗-representations of A, will be called the universal norm of A. An element a ∈ A will be said to have a bounded universal norm if kaku < ∞, and the set of all elements a ∈ A which have a bounded universal norm will be denoted by Au, i.e. Au := {a ∈ A : kaku < ∞} . When Au = A the universal norm becomes a true C∗-seminorm, being actually the largest possible C∗-seminorm in A. The Hausdorff completion of A in the universal norm is then a C∗-algebra called the enveloping C∗-algebra of A, which enjoys a number of universal properties (see [18, Theorem 10.1.11] and [18, Theorem 10.1.12]). For this reason, when every element a ∈ A has a bounded universal norm, i.e. Au = A, it is said that A has an enveloping C∗-algebra. In general, a ∗-algebra does not necessarily have an enveloping C∗-algebra. Perhaps the most basic example is that of a polynomial ∗-algebra in a single self-adjoint variable. We now look at the relation between automatically bounded elements and elements with a bounded universal norm. It is known that every BG∗-algebra has an enveloping C∗-algebra ([18, Proposition 10.1.19]), and the same proof yields this slightly more general result, that an automatically bounded element has a bounded universal norm: Proposition 1.3. Let A be a ∗-algebra. We have that Ab ⊆ Au. In particular if A is a BG∗-algebra, then A has an enveloping C∗-algebra. Proof: Suppose a /∈ Au. Then there is a sequence of representations {πi}i∈N, on Hilbert spaces {Hi}i∈N, such that kπi(a)k → ∞. Consider now the inner product space V defined as the algebraic direct sum V :=Mi∈N Hi , and the pre-∗-representation π := Li∈N πi of A on L(V ). It is clear by con- struction that π(a) /∈ B(V ). Hence, a /∈ Ab. 4 1.2 Preliminaries on Hecke Algebras We will mostly follow [10] and [9] in what regards Hecke pairs and Hecke alge- bras and refer to these references for more details. Definition 1.4. Let G be a group and Γ a subgroup. The pair (G, Γ) is called a Hecke pair if every double coset ΓgΓ is the union of finitely many right (and left) cosets. In this case, Γ is also called a Hecke subgroup of G. Given a Hecke pair (G, Γ) we will denote by L and R, respectively, the left and right coset counting functions, i.e. L(g) := ΓgΓ/Γ = [Γ : Γ ∩ gΓg−1] < ∞ R(g) := Γ\ΓgΓ = [Γ : Γ ∩ g−1Γg] < ∞ . We recall that L and R are Γ-biinvariant functions which satisfy L(g) = R(g−1) for all g ∈ G. Moreover, the function ∆ : G → Q+ given by ∆(g) := L(g) R(g) , is a group homomorphism, usually called the modular function of (G, Γ). Definition 1.5. Given a Hecke pair (G, Γ), the Hecke algebra H(G, Γ) is the ∗-algebra of finitely supported C-valued functions on the double coset space Γ\G/Γ with the product and involution defined by (f1 ∗ f2)(ΓgΓ) := XhΓ∈G/Γ f ∗(ΓgΓ) := ∆(g−1)f (Γg−1Γ) . f1(ΓhΓ)f2(Γh−1gΓ) , Remark 1.6. Some authors, including Krieg [10], do not include the factor ∆ in the involution. Here we adopt the convention of Kaliszewski, Landstad and Quigg [9] in doing so, as it gives rise to a more natural L1-norm. We note, nevertheless, that there is no loss (or gain) in doing so, because these two dif- ferent involutions give ∗-isomorphic Hecke algebras. In particular, the question of existence of an enveloping C∗-algebra is not perturbed by this. The Hecke algebra has a natural basis, as a vector space, given by the charac- teristic functions of double cosets. We will henceforward identify a characteristic function of a double coset χΓgΓ with the double coset ΓgΓ itself. It will be useful to know how to write a product ΓgΓ ∗ ΓhΓ of two double cosets in the unique linear combination of double cosets: Lemma 1.7. The expression for the product ΓgΓ∗ ΓhΓ of two double cosets in the unique linear combination of double cosets is given by: ΓgΓ ∗ ΓhΓ = XΓsΓ∈Γ\G/Γ L(g) Cg,h(s) L(s) ΓsΓ , 5 where Cg,h(s) := #{wΓ ⊆ ΓhΓ : ΓgwΓ = ΓsΓ}. Proof: Let us first check that Cg,h(s) is well-defined. It is clear that Cg,h(s) does not depend on the representatives h and s of the chosen double cosets, so it remains to verify that it is also independent on g. Given any other representative βgγ of the double coset ΓgΓ, with β, γ ∈ Γ, it is not difficult to see that the map wΓ 7→ γ−1wΓ , gives a bijective correspondence between the sets {wΓ ⊆ ΓhΓ : ΓgwΓ = ΓsΓ} and {uΓ ⊆ ΓhΓ : ΓβgγuΓ = ΓsΓ}. Hence we have Cβgγ,h(s) = Cg,h(s). Now, to check the product formula we recall (for example from [9]) that ΓgΓ ∗ ΓhΓ = XwΓ∈ΓhΓ/Γ L(g) L(gw) ΓgwΓ , (1) where the sum runs over a set of representatives for left cosets in ΓhΓ. Let us fix a representative g for the double coset ΓgΓ and let S be the set of double cosets S := {ΓgwΓ : wΓ ∈ ΓhΓ/Γ}, i.e. the set of double cosets that appear as summands in (1). The number of times an element ΓsΓ ∈ S appears repeated in the sum (1) is precisely the number Cg,h(s). Hence we can write ΓgΓ ∗ ΓhΓ = XΓsΓ∈S L(g) Cg,h(s) L(s) ΓsΓ . Also, if a double coset ΓrΓ does not belong to S we have Cg,h(r) = 0, thus we get ΓgΓ ∗ ΓhΓ = XΓsΓ∈Γ\G/Γ L(g) Cg,h(s) L(s) ΓsΓ . The reader can find alternative ways of describing the coefficients of this unique linear combination in [10, Lemma 4.4]. In particular, the characteriza- tion (iii) of the cited lemma is very similar to the one we just described. Remark 1.8. A direct computation or Lemma 1.7 imply that the double cosets that appear in the expression for ΓgΓ ∗ ΓhΓ as a unique linear combination of double cosets are all of the form ΓgγhΓ, for some γ ∈ Γ. Conversely, all double cosets of the form ΓgγhΓ, with γ ∈ Γ, appear in this linear combination, be- cause Cg,h(gγh) 6= 0. Another basic property of Hecke algebras which we will need is the following: given a Hecke pair (G, Γ) and a subgroup K such that Γ ⊆ K ⊆ G, then (K, Γ) is a Hecke pair and H(K, Γ) is naturally seen as a ∗-subalgebra of H(G, Γ). This is a particular case of [10, Lemma 4.9]. 6 Definition 1.9. The L1-norm on H(G, Γ), denoted k · kL1, is given by f (ΓgΓ) kfkL1 = XΓgΓ∈Γ\G/Γ f (ΓgΓ) L(g) = XgΓ∈G/Γ The completion L1(G, Γ) of H(G, Γ) under this norm is a Banach ∗-algebra. As observed in [4], [21] and [9] there are several canonical C∗-completions of H(G, Γ). These are: • C∗ r (G, Γ) - Usually called the reduced Hecke C∗-algebra, is the completion of H(G, Γ) under the C∗-norm arising from a regular representation (see, for example, [21] for details). • pC∗(G)p - The corner of the full group C∗-algebra of the Schlichting com- pletion (G, Γ) of the pair (G, Γ), for the projection p := χΓ. We will not describe this construction here since it is well documented in the literature ([21], [7], [9]) and because we will not make use of Schlichting completions in this work. • C∗(L1(G, Γ)) - The enveloping C∗-algebra of L1(G, Γ). • C∗(G, Γ) - The enveloping C∗-algebra (if it exists!) of H(G, Γ). When it exists, it is usually called the full Hecke C∗-algebra. These different C∗-completions of H(G, Γ) are related in the following way, through canonical surjective maps: C∗(G, Γ) 99K C∗(L1(G, Γ)) −→ pC∗(G)p −→ C∗ r (G, Γ) . As was pointed out by Hall [8, Proposition 2.21], a Hecke algebra does not need to have an enveloping C∗-algebra in general , with the Hecke algebra of the pair (SL2(Qp), SL2(Zp)) being one such example, where p is a prime number and Qp, Zp denote respectively the field of p-adic numbers and the ring of p-adic integers. 1.3 Preliminaries on Directed Graphs Recall that a simple directed graph G := (B, E) consists of a set B, whose elements are called vertices, and a subset E ⊆ B2, whose elements are called edges. An edge is thus a pair of vertices (a, b), which we see as directed from a to b. Since we are only interested in directed graphs that are simple, i.e. such that there is at most one edge directed from one vertex to another, we will henceforward drop the word simple and simply write directed graph. Let us now set some notation. Let G := (B, E) be a directed graph. If the ordered pair (a, b) belongs to E we say that b is a successor of a. Definition 1.10. Let G := (B, E) be a directed graph. A set of vertices S ⊆ B is said to be co-hereditary if it contains the successors of all of its elements, i.e. if a ∈ S and b ∈ B is a successor of a, then b ∈ S. 7 It is easy to see that an arbitrary intersection of co-hereditary sets is still a co-hereditary set. Hence, we can talk about the co-hereditary set generated by a subset X ⊆ B of vertices: Definition 1.11. Let G := (B, E) be a directed graph and X ⊆ B a set of vertices. The co-hereditary set generated by X is the smallest co-hereditary set that contains X. Given a directed graph G := (B, E) and a set of vertices X ⊆ B, we will denote by S(X) the set of all the successors of all elements of X, i.e. for some x ∈ X} . Similarly, we define the n-th successor set of X inductively as follows: S(X) := {a ∈ B : a is a successor of x, S 0(X) := X , Sn(X) := S(Sn−1(X)) , for n ≥ 1 . In this way, the 0-th successor set is simply the set X, the 1-st successor set is S(X), the 2-nd successor set is S(S(X)), etc. We will often consider X to be a singleton set X = {b}, and in this case we will use the notation S(b) instead of S({b}). The following result follows easily from the definitions: Lemma 1.12. Let G := (B, E) be a directed graph and X ⊆ B a set of vertices. The co-hereditary set generated by X is the setSn∈N0 Sn(X). Remark 1.13. The sets of vertices we are going to consider in our applications will be sets with specific additional structure (for instance, the set of vertices will typically be a basis of a vector space), and we are interested in proving results of the type: all elements of the co-hereditary set generated by X have a certain property P . To do so, we use a certain form of "induction". Namely, if we prove that all elements of X have the property P , and if we prove that the property P is preserved upon taking successors, then by Lemma 1.12 and the usual induction on N, all elements of the co-hereditary set generated by X will also satisfy P . 2 Graph Associated with a ∗-Algebra Let A be a ∗-algebra. Suppose that we are given a finite set of elements {b1, . . . , bn} ⊆ A satisfying a set of relations of the form b∗ 1b1 = λ11b1 + ··· + λ1nbn (2) b∗ nbn = λn1b1 + ··· + λnnbn , where λij ∈ C for each i, j ∈ {1, . . . , n}. We claim that the elements b1, . . . , bn are automatically bounded, and this fact will pave the way for our study of existence of enveloping C∗-algebras: ...  8 Theorem 2.1. Let A be a ∗-algebra and {b1, . . . , bn} ⊆ A a finite set of ele- ments satisfying relations as in (2). Then the elements b1, . . . , bn are automat- ically bounded. In particular they have a bounded universal norm. In order to prove Theorem 2.1 we will need the following lemma: Lemma 2.2. Let n ∈ N and kij ∈ R+ B := {(x1, . . . , xn) ∈ (R+ 0 )n : x2 0 for every 1 ≤ i, j ≤ n. The set i ≤ ki1x1 + ··· + kinxn ∀1 ≤ i ≤ n} is bounded in Rn. Proof: Let us denote by β the real number β := nXi=1 vuut nXj=1 kij , and let eB be the set defined by eB :=n(x1, . . . , xn) ∈ (R+ We claim that B ⊆ eB. To see this, let (x1, . . . , xn) ∈ B. We have x1 + ··· + xn ≤ 0 )n : x1 + ··· + xn ≤ β √x1 + ··· + xno , nXi=1pki1x1 + ··· + kinxn vuut(cid:16) nXj=1 nXi=1 nXi=1vuut(cid:16) nXj=1 kij(cid:17)x1 + ··· +(cid:16) nXj=1 kij(cid:17)(cid:16)x1 + ··· + xn(cid:17) kij(cid:17)xn ≤ = = β √x1 + ··· + xn , and therefore (x1, . . . , xn) ∈ eB. Hence, it is enough to prove that the set eB is bounded. As it is well known, linear functions in R grow faster than square roots, thus it is clear that the set is bounded in R. Let S : (R+ the pre-image by S of a bounded set in R is also a bounded set in (R+ Y := {x ∈ R+ 0 )n → R be the function S(x1, . . . , xn) :=Pn We have that eB ⊆ S−1(Y ). Since S is only defined for elements in (R+ conclude that eB, and therefore B, is bounded. relations as in (2) and B ⊆ (R+ Proof of Theorem 2.1: Let {b1, . . . , bn} ⊆ A be a finite set in A satisfying 0 : x ≤ β√x} i=1 xi. 0 )n, 0 )n. We 0 )n the set defined by B := {(x1, . . . , xn) ∈ (R+ 0 )n : x2 i ≤ λi1x1 + ··· + λinxn ∀1 ≤ i ≤ n} . 9 Let π : A → L(V ) be a pre-∗-representation and ξ ∈ V a vector such that kξk = 1. We have that i bi)ξ, ξi ≤ kπ(b∗ nXj=1 i bi)ξk = k i bi)ξkkξk λij π(bj)ξk λijkπ(bj)ξk . kπ(bi)ξk2 = hπ(b∗ = kπ(b∗ nXj=1 ≤ Hence it follows that(cid:0)kπ(b1)ξk, . . . ,kπ(bn)ξk(cid:1) ∈ B. Since the definition of the set B is independent of π and ξ, and since by Lemma 2.2 we know that B is bounded in Rn, it follows that sup kξk=1kπ(bi)ξk < ∞ , i.e. π(bi) ∈ B(V ), for all 1 ≤ i ≤ n, and all pre-∗-representations π. Thus, the elements b1, . . . , bn are all automatically bounded and therefore have bounded universal norms by Proposition 1.3 In practice though, Theorem 2.1 can be difficult to apply, as in general one is not given a set of elements {b1, . . . , bn} satisfying the prescribed relations, especially if the structure of the ∗-algebra A is not well understood. For this reason we will describe a more algorithmic approach to Theorem 2.1 where the set {b1, . . . , bn} is not given from the start, but it is instead constructed step-by- step starting from one element b1. This method will be explained through the language of graphs and will be especially useful when applied to Hecke algebras, where knowledge from the Hecke pair can many times be used to show that sets of elements {b1, . . . , bn} satisfying (2) abound. Let A be a ∗-algebra and B a basis of A as a vector space. Given a basis element b0 ∈ B we will denote by Φb0 the unique linear functional Φb0 : A → C such that Φb0 (b) :=(1, 0, if b = b0 if b 6= b0 (3) for every b ∈ B. Definition 2.3. Given a ∗-algebra A with a specified basis B, we define its associated graph as the directed graph G := (B, E), whose set of vertices is the set B and whose set of edges is the set E := {(a, b) ∈ B2 : Φb (a∗a) 6= 0} . Thus, given a vertex a ∈ B, its successors are precisely those basis elements that have non-zero coefficients in the unique expression of a∗a as a linear com- bination of elements of B, i.e. if a∗a = k1b1 + ··· + knbn , 10 where each ki ∈ C is non-zero and the basis elements bi are all different, then the successors of a are precisely b1, . . . , bn. Proposition 2.4. Let A be a ∗-algebra with basis B and G its associated graph. If X ⊆ B is a finite co-hereditary set in G, then all elements of X are automati- cally bounded. In particular, all elements of X have a bounded universal norm. Proof: Let X = {b1, . . . , bn} ⊆ B. Since X contains the successors of all its elements, we must necessarily have b∗ i bi = λi1b1 + ··· + λinbn , 1 ≤ i ≤ n , for some elements λij ∈ C (possibly being zero). It then follows form Theorem 2.1 that all elements b1, . . . , bn are automatically bounded and in particular have a bounded universal norm. Corollary 2.5. Let A be a ∗-algebra, B a basis for A and G its associated graph. If A is generated as a ∗-algebra by the elements of the finite co-hereditary sets of G, then A is a BG∗-algebra. In particular A has an enveloping C∗-algebra. Proof: Let B0 be the set of elements of the finite co-hereditary sets of G. By Proposition 2.4, all elements in B0 are contained in Ab. Since the elements of B0 generate the ∗-algebra A, we conclude that A = Ab, i.e. A is a BG∗-algebra. We can interpret the above corollary in the following (equivalent) way: sup- pose we have a ∗-algebra A with a basis B. Suppose additionally that we have a particular set B0 ⊂ B which generates A. If all the elements of B0 generate finite co-hereditary sets of the associated graph, then A has an enveloping C∗- algebra. Let us now give a couple of immediate examples: Example 2.6. Let A be a finite-dimensional ∗-algebra. If we take any basis B, the associated graph necessarily has finitely many vertices (and edges). Thus, the co-hereditary set generated by any b ∈ B is finite. Example 2.7. Let G be a discrete group, C(G) its group algebra with basis {δg ∈ C(G) : g ∈ G}. Since in the group algebra we have δ∗ g ∗ δg = δe, the only successor of δg in the associated graph is δe. Since δe is the only successor of itself, the co-hereditary set generated by δg has only two elements, δg and δe. Some non-trivial examples, arising from Hecke algebras, will be computed later in section 5. 11 3 A sufficient condition implying the isomorphism C∗(G, Γ) ∼= C∗(L1(G, Γ)) In Corollary 2.5 of the previous section we obtained a sufficient condition for a ∗-algebra to have an enveloping C∗-algebra, namely when it is generated by elements of the finite co-hereditary sets (with respect to a given basis). In this section we will improve this result in the case of a Hecke algebra H(G, Γ): under a suitable assumption we will not only assure an enveloping C∗-algebra of C∗(G, Γ) exists, but we will also be able to identify it with C∗(L1(G, Γ)). Throughout this section and henceforward (G, Γ) will denote a Hecke pair. We will always consider the canonical basis in the Hecke algebra H(G, Γ), con- sisting of double cosets {ΓgΓ : g ∈ G}. This section is devoted to the proof of the following result: Theorem 3.1. Let (G, Γ) be a Hecke pair. If all double cosets generate finite co-hereditary sets, then the enveloping C∗-algebra of H(G, Γ) exists and coin- cides with C∗(L1(G, Γ)). In order to give a proof of Theorem 3.1 we will make use of several lemmas. Lemma 3.2. Let (G, Γ) be a Hecke pair and f1, f2 ∈ H(G, Γ) be two elements such that fi(ΓgΓ) ≥ 0 for all ΓgΓ ∈ Γ\G/Γ. The L1-norm satisfies the equality kf1 ∗ f2kL1 = kf1kL1kf2kL1 . In particular the following equality is also satisfied for any f ∈ H(G, Γ) such that f (ΓgΓ) ≥ 0 for all ΓgΓ ∈ Γ\G/Γ: kf ∗ ∗ fkL1 = kfk2 L1 . Proof: We have that kf1 ∗ f2kL1 = XgΓ∈G/Γ = XgΓ∈G/Γ = XhΓ∈G/Γ = XhΓ∈G/Γ (f1 ∗ f2)(ΓgΓ) XhΓ∈G/Γ f1(ΓhΓ) XgΓ∈G/Γ f1(ΓhΓ) XgΓ∈G/Γ f1(ΓhΓ)f2(Γh−1gΓ) f2(Γh−1gΓ) f2(ΓgΓ) = kf1kL1kf2kL1 . The second claim in this lemma follows directly from the first statement: kf ∗ ∗ fkL1 = kf ∗kL1kfkL1 = kfk2 L1 . 12 Lemma 3.3. Let n ∈ N and A = [aij] be an n× n matrix, whose entries satisfy: aii ∈ R+ and aij ∈ R− 0 for all i 6= j. If there are vectors d = (d1, . . . , dn) and z = (z1, . . . , zn) both in (R+)n satisfying the system Az = d , (4) then A is non-singular. Proof: Let z ∈ (R+)n be a solution to the above system. Suppose that Ker A 6= {0}. Then, the set of solutions to the system (4) contains a line L. Consider now the set S of all the (finitely many) points which are the intersections of L with the canonical hyperplanes of the form xi = 0, and take a point y ∈ S (not necessarily unique) which is closest to z. The point y is the intersection of L with one of the hyperplanes xi = 0, say xi0 = 0 with 1 ≤ i0 ≤ n. Since y = (y1, . . . , yn) is in L, it is also a solution of the system (4) and therefore must satisfy ai0kyk = di0 , nXk=1 k6=i0 implying that there exists at least one number yk which is negative. But on the other hand, the open segment between z and y lies inside (R+)n because z ∈ (R+)n and this segment does not intersect any hyperplane xi = 0 (by choice of the point y). Thus the entries of y = (y1, . . . , yn) are all non-negative, which is a contradiction. Therefore Ker A = {0}. In preparation for the next lemma we set some notation. Given two vectors a = (a1, . . . , an) and b = (b1, . . . , bn) ∈ Rn, we will write a ≤ b whenever ai ≤ bi for every 1 ≤ i ≤ n. We will denote the zero vector by 0 = (0, . . . , 0). Also, given a set of vectors S ⊆ Rn, we will denote by C(S) the cone generated by S, i.e. the set of all linear combinations with coefficients in R+ 0 of the elements of S. Lemma 3.4. Let n ∈ N and A = [aij ] be an n × n matrix whose entries satisfy: aii ∈ R+ and aij ∈ R− 0 for all i 6= j. Assume that there are vectors d = (d1, . . . , dn) > 0 and z = (z1, . . . , zn) > 0 satisfying the system Az = d. Then, if for some y ∈ Rn, we must have y ≥ 0. Ay ≥ 0 , Proof: As we are in the conditions of Lemma 3.3, the matrix A is non- singular. First we claim that {y : Ay ≥ 0} = C(A−1e1, . . . , A−1en), where e1, . . . , en ∈ Rn are the canonical unit vectors. The inclusion ⊇ is obvious, while the inclusion ⊆ follows from the fact that if Ay ≥ 0 then we can write Ay as a positive linear combination of e1, . . . , en. Thus, to prove this lemma it suffices to prove that A−1ek ≥ 0 for every 1 ≤ k ≤ n, and we will show this by induction on n. The case n = 1 is obvious since a11 ∈ R+. Let us now assume 13 that the result holds for n− 1, and prove it for n. Let Bk be the matrix obtained from A by deleting the k-th row and column. Since Az = d, it follows readily that Bk = (5) z1 ... zk−1 zk+1 ... zn   d1 − a1kzk ... dk−1 − ak−1 kzk dk+1 − ak+1 kzk ... dn − an kzk  Since the right hand side of (5) is a vector in (R+)n−1, and moreover the entries of the matrix Bk satisfy the conditions in the statement of the lemma, we can use the induction hypothesis on the matrix Bk. Let v := (v1, . . . , vk−1, vk+1, . . . , vn) ∈ Rn−1 be a solution to the equation   Bk v = d1 ... dk−1 dk+1 ... dn  ,    which exists by Lemma 3.3 (the reason for the chosen indexing of the entries of v will become clear in the remaining part of the proof). The induction hypothesis tells us that v ≥ 0. We also have that d1 − a1kzk −a1kzk  ... ... dn − an kzk d1 ... dk−1 dk+1 ... dn  =  ... −ak−1 kzk −ak+1 kzk ... −an kzk  ≥ 0 . Bk − Bk v = dk−1 − ak−1 kzk dk+1 − ak+1 kzk − z1 ... zk−1 zk+1 ···zn   By the induction hypothesis again, we have zi − vi ≥ 0, for i 6= k. We have that Consider now the vectorev ∈ Rn given byev := (v1, . . . , vk−1, 0, vk+1, . . . , vn). A (z −ev) =  = dk −Pn 0 ... 0 i6=k akivi 0 ... 0 ,  d1 ... dk−1 dk dk+1 ... dn   −  d1 ... dk−1Pn i6=k akivi dk+1 ... dn 14 or in other words, z −ev = A−1  0 ... 0 i6=k akivi 0 ... 0  dk −Pn = (dk − nXi6=k akivi) A−1ek . 0 for k 6= i, vi ≥ 0 as we saw before, and dk > 0. We have already proven that zi − vi ≥ 0, that We now notice that dk −Pn for i 6= k, from which it readily follows that z −ev ≥ 0. We can now conclude i6=k akivi > 0, because all the aki ∈ R− A−1ek = 1 i6=k akivi dk −Pn (z −ev) ≥ 0 . Proof of Theorem 3.1: We already know that if all double cosets generate finite co-hereditary sets, then H(G, Γ) has an enveloping C∗-algebra. Thus, it remains to see that this enveloping C∗-algebra is the enveloping C∗-algebra of L1(G, Γ), and for this we only need to show that kaku ≤ kakL1, (6) for any a ∈ H(G, Γ). Actually we only need to prove (6) when a is a double coset a = ΓsΓ, since the result for a general a ∈ H(G, Γ) follows from the if we write a in the unique linear combination of double following argument: cosets, a =Pn i=1 λiΓsiΓ, then we have kaku = k λiΓsiΓku ≤ nXi=1 λikΓsiΓkL1 = k nXi=1 nXi=1 ≤ = kakL1 . λikΓsiΓku nXi=1 λiΓsiΓkL1 Let therefore ΓsΓ be a double coset and {Γs1Γ, . . . , ΓsnΓ} the finite co-hereditary set it generates. By Lemma 1.7 we have (ΓsiΓ)∗ ∗ ΓsiΓ = nXj=1 λij ΓsjΓ , (7) where the coefficients λij are given by λij := ∆(si) L(s−1 i ,si(sj) )Cs−1 L(sj) i = 15 ,si (sj) L(si)Cs−1 L(sj) i . Let B be the set B := {(x1, . . . , xn) ∈ (R+ 0 )n : x2 i ≤ λi1x1 + ··· + λinxn , ∀ 1 ≤ i ≤ n} . Let us also denote by C the subset of B determined by 0 )n : x2 It follows immediately from the triangle inequality applied to (7) that the univer- C := {(x1, . . . , xn) ∈ (R+ sal norm (in fact, any C∗-norm) satisfies(cid:0)kΓs1Γku, . . . ,kΓsnΓku(cid:1) ∈ B. More- i = λi1x1 + ··· + λinxn , ∀ 1 ≤ i ≤ n} . over, from Lemma 3.2, the L1-norm satisfies kΓsiΓk2 L1 = k(ΓsiΓ)∗ ∗ ΓsiΓkL1 = λij kΓsjΓkL1 . nXj=1 Thus,(cid:0)kΓs1ΓkL1, . . . ,kΓsnΓkL1(cid:1) ∈ C. For ease of reading we will denote by z := (z1, . . . , zn) the point (cid:0)kΓs1ΓkL1, . . . ,kΓsnΓkL1(cid:1). The idea for the remaining For each 1 ≤ i ≤ n let gi : (R+ part of the proof is to argue that z ∈ C is the point with the largest coordinates in the whole set B. 0 )n → R be the function gi(x1, . . . , xn) := x2 i − λij xj . nXj=1 The tangent hyperplane to the graph of gi at the point (z1, . . . , zn) is given by the equation (2zi − λii)(xi − zi) − nXj=1 j6=i λij (xj − zj) = 0 , which, using the fact that (z1, . . . , zn) is a zero of gi, we can reduce to (2zi − λii)xi − nXj=1 j6=i λij xj = z2 i . (8) We claim that 2zi − λii > 0. To see this we notice that 2zi − λii = 2kΓsiΓkL1 − ,si (si) L(si)Cs−1 L(si) i = 2L(si) − Cs−1 i ,si(si) ≥ 2L(si) − L(si) = L(si) > 0 . Let us now take A = [aij ] to be the n × n matrix whose entries are given by 0 for i 6= j and aii ∈ R+. 1, . . . , z2 n). Consider now aij := −λij for i 6= j, and aii := 2zi − λii, thus aij ∈ R− We can easily see from (8) that Az = z2, where z2 = (z2 the set W defined by W :=(cid:8)x ∈ (R+ 0 )n : Ax ≤ z2} . 16 We claim that W contains the set B. To see this, let (y1, . . . , yn) ∈ B. We then have −z2 i + (2zi − λii)yi − nXj=1 j6=i λij yj = −z2 i + 2ziyi − λiiyi − λij yj j6=i nXj=1 nXj=1 λij yj = −(yi − zi)2 + y2 i − λij yj nXj=1 i − ≤ y2 ≤ 0 , which implies that (2zi − λii)yi − nXj=1 j6=i λij yj ≤ z2 i , and thus (y1, . . . , yn) ∈ W . In other words, if y ∈ B, then Ay ≤ z2. We can rewrite this inequality as: Ay ≤ z2 ⇔ 0 ≤ z2 − Ay ⇔ 0 ≤ A(z − y) . Noting that we are under the conditions of Lemma 3.4, because the entries of A satisfy the required conditions and Az = z2, we conclude that 0 ≤ z − y, i.e. y ≤ z. Thus, we conclude that z has bigger coordinates than any other point in B. As we know, we have (kΓs1Γku, . . . ,kΓsnΓku) ∈ B, so by the above we must have kΓsiΓku ≤ zi = kΓsiΓkL1 for any 1 ≤ i ≤ n. Thus, in particular, kΓsΓku ≤ kΓsΓkL1, for the initial double coset ΓsΓ. Since all double cosets generate finite co-hereditary sets we conclude that this inequality holds for any double coset ΓsΓ, and as we explained in the beginning of the proof, this implies that the enveloping C∗-algebra of H(G, Γ) is C∗(L1(G, Γ)). 4 Methods for Hecke Algebras The basis of our study of enveloping C∗-algebras of Hecke algebras will be Corol- lary 2.5 and Theorem 3.1. Our goal is to apply these results to several classes of Hecke pairs, but so far we have not given any hint on how to actually ensure that a given double coset generates a finite co-hereditary set. The objective of this section is to provide some tools, based on iterated commutators, to help us accomplish this task. Given a group G we will denote by [s, t] the commutator of s, t ∈ G, i.e. [s, t] := s−1t−1st . More generally, given elements s1, . . . , sn ∈ G we will denote by [s1, . . . , sn] the iterated commutator defined inductively by [s1, . . . , sn] := [[s1, . . . , sn−1], sn] . 17 Let us now return to Hecke pairs (G, Γ). We will be mostly interested in commutators of the form [g, γ1, . . . , γn], where g ∈ G and γ1, . . . , γn ∈ Γ, and the reason for that is given by the following result: Proposition 4.1. Let (G, Γ) be a Hecke pair and g ∈ G. Let {ΓxnΓ}n∈N0 be a sequence of double cosets satisfying the properties: i) Γx0Γ = ΓgΓ, ii) Γxn+1Γ is a successor of ΓxnΓ, for all n ≥ 0. Then, there exists a sequence {γn}n∈N ⊆ Γ such that ΓxnΓ = Γ[g, γ1, . . . , γn]Γ , for all n ≥ 1. In particular, all elements in Sn(ΓgΓ) have a representative of the form Γ[g, γ1, . . . , γn]Γ, for some γ1, . . . , γn ∈ Γ. Proof: We will choose such a sequence {γn}n∈N inductively on n ∈ N. Suppose n = 1. Since Γx1Γ is a successor of ΓgΓ, it must be of the form Γx1Γ = Γg−1γgΓ for some γ ∈ Γ (see Remark 1.8). Now we notice that g−1γg = g−1γgγ−1γ = [g, γ−1]γ . Hence, we have Γx1Γ = Γg−1γgΓ = Γ[g, γ−1]γΓ = Γ[g, γ−1]Γ . Choosing γ1 := γ−1 yields the desired result. Now let us suppose that there exist elements γ1, . . . , γn ∈ Γ such that ΓxkΓ = Γ[g, γ1, . . . , γk]Γ, for every 1 ≤ k ≤ n. Then, since Γxn+1Γ is a successor of ΓxnΓ = Γ[g, γ1, . . . , γn]Γ, we can write Γxn+1Γ = Γ[g, γ1, . . . , γn]−1γ[g, γ1, . . . , γn]Γ , for some γ ∈ Γ (again by Remark 1.8). We have Γxn+1Γ = Γ[g, γ1, . . . , γn]−1γ[g, γ1, . . . , γn]γ−1Γ = Γ[g, γ1, . . . , γn, γ−1]Γ . Choosing γn+1 := γ−1 yields the desired result for Γxn+1Γ. Hence, since we can extend any finite sequence γ1, . . . , γn satisfying the stated conditions to a sequence γ1, . . . , γn, γn+1 still satisfying the stated condi- tions, it follows that there must be an infinite sequence {γn}n∈N with the desired requirements. We will now establish a sufficient condition to ensure the finiteness of the co-hereditary set generated by an element ΓgΓ based on the iterated commuta- tors we considered above: 18 Theorem 4.2. Let (G, Γ) be a Hecke pair and g ∈ G. Suppose that for any sequence of elements {γk}k∈N ⊆ Γ the total number of double cosets is finite. Then ΓgΓ generates a finite co-hereditary set. #(cid:8)Γ[g, γ1, . . . , γn]Γ : n ∈ N(cid:9) Proof: Suppose the co-hereditary set generated by ΓgΓ is infinite. Then, there must exist a sequence {ΓxnΓ}n∈N0 such that i) Γx0Γ = ΓgΓ, ii) Γxn+1Γ is a successor of ΓxnΓ, for all n ≥ 0. iii) Γxn+1Γ /∈Sn i=0 Si(ΓgΓ), for all n ≥ 0. In particular, we have that ΓxiΓ 6= ΓxjΓ for i 6= j, implying that the set {ΓxnΓ : n ∈ N} is infinite. By Proposition 4.1 there exists a sequence {γn}n∈N ⊆ Γ such that ΓxnΓ = Γ[g, γ1, . . . , γn]Γ for all n ≥ 1. But, by assumption, the number of double cosets in {Γ[g, γ1, . . . , γn]Γ : n ∈ N} is finite. Thus we arrive at a contradiction and therefore the co-hereditary set generated by ΓgΓ must be finite. Corollary 4.3. Let (G, Γ) be a Hecke pair and g ∈ G. If one of the following conditions holds, then ΓgΓ generates a finite co-hereditary set: a) For every sequence {γn}n∈N ⊆ Γ there exists a finite set F ⊆ G and N0 ∈ N such that [g, γ1, . . . , γk] ∈ F for all k ≥ N0. b) For every sequence {γn}n∈N ⊆ Γ there exists N ∈ N such that [g, γ1, . . . , γN ] ∈ Γ. Proof: The result follows directly from Theorem 4.2. For a) we notice that we can write {Γ[g, γ1, . . . , γn]Γ : n ∈ N} as the union of the two finite sets {Γ[g, γ1, . . . , γn]Γ : n < N0} and {Γ[g, γ1, . . . , γn]Γ : n ≥ N0}. For b), one can easily show, by induction, that [g, γ1, . . . , γn] ∈ Γ, for any n ≥ N . Thus, we have (cid:8)Γ[g, γ1, . . . , γn]Γ : n ∈ N(cid:9) =(cid:8)Γ[g, γ1, . . . , γn]Γ : n ≤ N(cid:9) , which is a finite set. There are different classes of Hecke pairs that satisfy conditions a) and b) of the above corollary. As we shall see in more detail in the next section, condi- tion a) is satisfied by groups satisfying certain generalized nilpotency properties, whereas b) is satisfied when Γ is a subnormal subgroup of G, for example. 19 5 Classes of Hecke Pairs We will now use the methods developed in the previous sections to study the existence of enveloping C∗-algebras for several classes of Hecke algebras. Many of the well known results about the existence of a full Hecke C∗-algebra for some classes of Hecke pairs will be recovered in a unified approach and some new classes will also be described. The isomorphism C∗(G, Γ) ∼= C∗(L1(G, Γ)) will also be established in many of the considered classes. It should also be noted that all the classes of Hecke algebras considered here are in fact BG∗-algebras, since our methods can be traced back to Corollary 2.5, but since the focus is mostly on the existence of C∗(G, Γ) we will not mention this in every case. This section is organized as follows: the classes of Hecke pairs from 5.1 to 5.4 have been studied in the operator algebraic literature and results about the corresponding full Hecke C∗-algebras are known. The results about the remaining classes, 5.5 to 5.12, are essentially new, with the results for the classes 5.5, 5.6 and 5.7 generalizing known results in the literature. The classes we consider are presumably all different (in the sense of contain- ment), with the notable exceptions of 5.5 which is a particular case of 5.6, and 5.1 which is a particular case of 5.7. We would like to remark that the results discussed in this section illustrate how our methods apply for natural classes of Hecke pairs and that we have not, by any means, exhausted all the possible classes of Hecke pairs one can study through these methods. 5.1 Γ has Finite Index in G When Γ has finite index in G, the pair (G, Γ) is automatically a Hecke pair, and the Hecke ∗-algebra is finite dimensional (actually, H(G, Γ) is finite dimensional if and only if Γ has finite index in G). As we have seen in Example 2.6, the co-hereditary set generated by a double coset is finite because the graph of H(G, Γ) is itself finite. Hence, Theorem 3.1 tells us that C∗(G, Γ) exists and C∗(G, Γ) ∼= C∗(L1(G, Γ)). Of course this example, investigated by Hall [8, Section 4.2], is well-known and completely understood, because a finite dimensional ∗-algebra is automati- cally complete for any ∗-algebra norm. Hence we necessarily have C∗(G, Γ) ∼= C∗(L1(G, Γ)) ∼= pC∗(G)p ∼= C∗ r (G, Γ) , and all these C∗-algebras are isomorphic to H(G, Γ), without having to invoke our Theorem 3.1. 5.2 (G, Γ) is Directed Recall that (G, Γ) is said to be directed if G = T −1T , where T := {t ∈ G : Γ ⊆ tΓt−1} . Directed Hecke pairs have been widely studied in the literature ([5], [8], [13], [12], [2], [9], for example), in particular because of their association with the theory 20 of semigroup C∗-crossed products. It is known that when (G, Γ) is directed the Hecke algebra has an enveloping C∗-algebra and moreover one has C∗(G, Γ) ∼= C∗(L1(G, Γ)) ∼= pC∗(G)p , (see, for example, [9, Theorem 7.4]). With our methods we can show that C∗(G, Γ) exists, since the Hecke algebra is in fact generated by finite co-hereditary sets. To see this, we first notice that, for t ∈ T , we have ΓtΓ = tΓ. Hence, we also have (ΓsΓ)∗ ∗ ΓtΓ = Γs−1tΓ (9) for every s, t ∈ T , which means that the Hecke ∗-algebra is generated by the set of double cosets {ΓtΓ : t ∈ T}. Taking s = t in equality (9) we see that (ΓtΓ)∗ ∗ ΓtΓ = Γ Thus, the only successor of the double coset ΓtΓ is Γ. Since Γ is the only successor of itself, it follows that the co-hereditary set generated by ΓtΓ has only two elements, ΓtΓ and Γ, and is therefore finite. We conclude that H(G, Γ) is generated by finite co-hereditary sets and therefore C∗(G, Γ) exists by Corollary 2.5. 5.3 Iwahori Hecke Algebras Let (G, Γ) be a Hecke pair such that H(G, Γ) is an Iwahori Hecke algebra (see [8, Definition 5.12] for a precise definition of this concept). Sets of generators and relations have been given for this class of Hecke algebras, but for our purposes we will only need to know that: 1. There is a set S ⊆ G of elements of order two such that H(G, Γ) is gener- ated (as a ∗-algebra) by Γ and the double cosets ΓsΓ, with s ∈ S. 2. for every s ∈ S the following relation holds: (ΓsΓ)2 = L(s)Γ + (L(s) − 1)ΓsΓ . in this work, we refer the reader to Hall's thesis [8, Section 5.3.1]. For the remaining relations in H(G, Γ), of which we will not make any use It was proven by Hall [8, Proposition 2.24], through an estimate on the spectral radius of certain elements, that an Iwahori Hecke algebra has an en- veloping C∗-algebra (actually Hall proved this for the case (SLn(Qp), B), with B ⊆ SLn(Qp) an Iwahori subgroup, but her proof is completely general). We can also conclude this from our methods, by proving that H(G, Γ) is generated by finite co-hereditary sets. By point 1) we only need to see that each double coset ΓsΓ with s ∈ S generates a finite co-hereditary set. So let ΓsΓ ∈ H(G, Γ) with s ∈ S. Since s has order two we see that ΓsΓ is self-adjoint and therefore relation 2) can be rewritten as (ΓsΓ)∗ ∗ ΓsΓ = L(s)Γ + (L(s) − 1)ΓsΓ Hence, the successors of ΓsΓ are only Γ and ΓsΓ itself. Thus, the co-hereditary set generated by ΓsΓ has only two elements, Γ and ΓsΓ, and is therefore finite. We conclude that H(G, Γ) is generated by finite co-hereditary sets, and is there- fore a BG∗-algebra and has an enveloping C∗-algebra. 21 Remark 5.1. By a result of Hall [8, Theorem 6.10] and a result of Kaliszewski, Landstad and Quigg [9, Corollary 6.11] it is known that, for G = SL2(Qp) and Γ an Iwahori subgroup, we necessarily have C∗(G, Γ) ∼= C∗(L1(G, Γ)) ∼= pC∗(G)p . The analogous result for SLn(Qp) with n ≥ 3 is still open, as far as we know. 5.4 Γ is a Protonormal Subgroup of G We recall that Γ is a protonormal subgroup of G (in the sense of Exel [6]), if for every s ∈ G we have Γs−1Γs = s−1ΓsΓ . Subgroups with this property are also called conjugate permutable subgroups in the literature. It was proven by Exel ([6, Proposition 12.1]) that when Γ is a protonor- mal subgroup of G the enveloping C∗-algebra C∗(G, Γ) exists. Moreover, it is completely clear from his proof that C∗(G, Γ) ∼= C∗(L1(G, Γ)), since the bound he uses for the universal norm is actually the L1-norm. Our methods can also recover this result, because in fact any double coset ΓgΓ generates a finite co- hereditary set. We will actually prove that the co-hereditary set generated by ΓgΓ consists only of ΓgΓ and S(ΓgΓ) and is therefore finite. In other words, we will prove that Sn(ΓgΓ) ⊆ S(ΓgΓ) , for every n ∈ N. It suffices to prove that S 2(ΓgΓ) ⊆ S(ΓgΓ). The elements of S 2(ΓgΓ) are of the form Γ[g, γ1, γ2]Γ, where γ1, γ2 ∈ Γ, by Proposition 4.1. We have that [g, γ1, γ2] = [g, γ1]−1γ−1 2 [g, γ1]γ2 = γ−1 1 g−1(γ1gγ−1 2 g−1)γ−1 1 gγ1γ2 . Since Γ is a protonormal subgroup there exist θ, ω ∈ Γ such that γ1gγ−1 gθg−1ω. Thus, we get 2 g−1 = [g, γ1, γ2] = γ−1 = γ−1 1 g−1(gθg−1ω)γ−1 1 θg−1ωγ−1 1 gγ1γ2 , 1 gγ1γ2 and therefore Γ[g, γ1, γ2]Γ = Γγ−1 1 θg−1ωγ−1 1 gΓ . = Γg−1ωγ−1 1 gγ1γ2Γ By Remark 1.8, Γg−1ωγ−1 1 gΓ ∈ S(ΓgΓ). This finishes the proof. 22 5.5 Γ is Subnormal in G Hecke pairs (G, Γ) in which Γ is normal in a normal subgroup of G have been widely studied in the literature, in particular when G is a semi-direct product ([5], [13], [12], [9]), and it is known that in this case H(G, Γ) has an enveloping C∗-algebra and moreover C∗(G, Γ) ∼= C∗(L1(G, Γ)) ∼= pC∗(G)p , (see, for example, [9, Theorem 6.13]). We are now going to prove that when Γ is a subnormal subgroup of G, C∗(G, Γ) exists and C∗(G, Γ) ∼= C∗(L1(G, Γ)). Recall that Γ is subnormal in G if there are subgroups H0, H1, . . . , Hn such that Γ = Hn E Hn−1 E ··· E H0 = G , where the notation Hi+1 E Hi means that Hi+1 is a normal subgroup of Hi. We claim that when Γ is subnormal in G, all double cosets ΓsΓ generate finite co-hereditary sets. To see this we will use Corollary 4.3. Let s ∈ G and {γk}k∈N ⊆ Γ. We will prove by induction that [s, γ1, . . . , γk] ∈ Hk for 1 ≤ k ≤ n. For k = 1 this follows from the following observation: [s, γ1] = s−1γ−1 1 sγ1 ∈ s−1Γsγ1 ⊆ s−1H1sγ1 = H1γ1 = H1 . Now, let us prove that k ⇒ k+1. For simplicity, let us write xk := [s, γ1, . . . , γk], which by induction hypothesis is an element of Hk. Thus, we have [s, γ1, . . . , γk, γk+1] = [xk, γk+1] ∈ x−1 k Γxkγk+1 k Hk+1xkγk+1 = Hk+1γk+1 ⊆ x−1 = Hk+1 . Thus, for any sequence {γk}k∈N we have [s, γ1, . . . , γn] ∈ Γ, which by Corol- lary 4.3 b) implies that ΓsΓ generates a finite co-hereditary set. Since this is true for all double cosets, Theorem 3.1 tells us that C∗(G, Γ) exists and C∗(G, Γ) ∼= C∗(L1(G, Γ)). Remark 5.2. It is known that any subgroup Γ of a nilpotent group G is nec- essarily a subnormal subgroup (see, for example, [11, §62]). Hence already from this we can conclude that the Hecke algebra of any Hecke pair (G, Γ), with G a nilpotent group, has an enveloping C∗-algebra (which coincides with C∗(L1(G, Γ))). In fact, this holds for any group G whose subgroups are all subnormal. Groups with this property form a class that strictly contains the class of nilpotent groups ([20, Theorem 6.11]). We will prove similar results for other classes of groups which strictly generalize the class of nilpotent groups. Example 5.3. Let G be the group of n × n upper triangular matrices with 1's on the diagonal and with entries in Q and let Γ be the subgroup of those matrices with entries in Z. It can be checked, although we will not do so here, that (G, Γ) forms a Hecke pair. The subgroup Γ is subnormal with Γ = Hn E Hn−1 E ··· E H1 = G , 23 where Hk is the subgroup of matrices in G whose first k − 1 upper diagonals have entries in Z. The group G is nilpotent and its 3 × 3 version is the rational Heisenberg group discussed in [9, Example 11.7]. 5.6 Γ is Ascendant in G Recall that Γ is said to be ascendant in G if there is a normal series {Hi}i∈N0 , Γ = H0 E H1 E ··· E Hi E . . . is finite precisely when Γ is subnormal in G. that ends in the group G, in the sense thatSi∈N0 Hi = G. Of course, the series We will now prove that if Γ is ascendant in G, then every double coset generates a finite co-hereditary set, therefore implying that C∗(G, Γ) exists and is isomorphic to C∗(L1(G, Γ)). Let ΓsΓ be any double coset in H(G, Γ), with representative s ∈ G. Since Γ is ascendant, s must belong to one of the subgroups Hn, with n ∈ N0. Of course, Γ is a subnormal subgroup of Hn, and as we saw in the subnormal case, this implies that the co-hereditary set generated by ΓsΓ is necessarily finite. 5.7 Γ has Finitely Many Conjugates in G Suppose Γ has finitely many conjugates in G, or equivalently, the normalizer of Γ has finite index in G. Then, C∗(G, Γ) exists and C∗(G, Γ) ∼= C∗(L1(G, Γ)) because any double coset generates a finite co-hereditary set. To see this, let ΓgΓ be a double coset and let g−1 n Γgn be the conjugates of Γ. With the possible exception of ΓgΓ itself, any element in the co-hereditary set generated by ΓgΓ is a successor of another element. Hence, by Remark 1.8, any such element is of the form 1 Γg1, . . . , g−1 Γx−1γxΓ , where x ∈ G and γ ∈ Γ. We can then write x−1γx = g−1 i θgi, for some i ∈ {1, . . . , n} and θ ∈ Γ, and therefore Γx−1γxΓ = Γg−1 i θgiΓ. Thus, apart possibly from ΓgΓ, all elements in the co-hereditary set generated by ΓgΓ are successors of some ΓgiΓ, with 1 ≤ i ≤ n, by Remark 1.8 again. Thus, this co-hereditary set must be finite. 5.8 G is Finite-by-Nilpotent Recall that a group G is called nilpotent if its lower central series stabilizes at {e} after finitely many steps, i.e. if the normal series defined inductively by G0 := G , Gn+1 := [Gn, G] , is such that Gk = {e}, for some k ∈ N. normal subgroup K such that G/K is nilpotent, i.e. Recall also that a group G is said to be finite-by-nilpotent if G has a finite if G is an extension of a 24 finite group by a nilpotent group. In particular, all nilpotent groups are finite- by-nilpotent (taking K = {e}). Moreover, the class of finite-by-nilpotent groups is strictly larger than the class of nilpotent groups, as every finite group belongs to the former class but not to the latter. Finite-by-nilpotent groups also admit a nice description in terms of their it is known that finite-by-nilpotent groups are precisely lower central series: those whose lower central series stabilizes at a finite group. We are now going to show that for any Hecke pair (G, Γ) where G is finite-by- nilpotent, every double coset ΓsΓ generates a finite co-hereditary set, implying that C∗(G, Γ) exists and coincides with C∗(L1(G, Γ)). Let s ∈ G and {γk}k∈N ⊆ Γ. It is clear that [s, γ1, . . . , γk] ∈ Gk. Since the series {Gk} eventually stabilizes at a finite subgroup, it follows directly from Corollary 4.3 a) that ΓsΓ generates a finite co-hereditary set. This concludes the proof. 5.9 G is Hypercentral Recall that a group G is said to be a hypercentral group (also called a ZA-group) if its upper central series, possibly continued transfinitely, stabilizes at the whole group G. For a rigorous definition of this concept, we refer the reader to [19, sec- tion 12.2] for example. Another characterization of hypercentral groups, which is the one we will use, is given by the following result: Theorem 5.4 (Lemma, page 219, §63, [11]). A group G is hypercentral if and only if it satisfies the following property: for any s ∈ G and any sequence {xn}n∈N ⊂ G there is a k ∈ N such that [s, x1, . . . , xk] = e . We will now prove that if (G, Γ) is a Hecke pair with G a hypercentral group, then every double coset ΓsΓ generates a finite co-hereditary set, so that C∗(G, Γ) exists and C∗(G, Γ) ∼= L1(G, Γ). This is a direct application of Corollary 4.3 a), taking F = {e}, given the characterization of hypercentral groups of Theorem 5.4. Remark 5.5. The class of hypercentral groups also strictly contains the class of nilpotent groups (see Example 5.6), and moreover it is known that every hy- percentral group is locally nilpotent (but not vice-versa). Thus, we have found another class of groups G, satisfying a nilpotent-type property, for which the Hecke algebra H(G, Γ) of any Hecke pair (G, Γ) has an enveloping C∗-algebra (which coincides with C∗(L1(G, Γ))). Example 5.6. Let Z2∞ be the 2-quasicyclic group, i.e. the group of all the 2n-th roots of unity for all n ∈ N. This group is the Pontryagin dual of group of 2-adic integers. The group Z/2Z acts on Z2∞ by mapping an element to its inverse. The generalized dihedral group G := Z2∞ ⋊(cid:0)Z/2Z(cid:1) 25 is a group which is hypercentral, but not nilpotent. 5.10 G is an F C-group and Γ is Finite Recall that a group G is said to be F C if every element s has finitely many conjugates, i.e. the set Cs := {t−1st : t ∈ G} is finite. It can be seen that every subgroup Γ ⊆ G of an F C-group is a Hecke subgroup, because ΓsΓ = [γ∈Γ γsΓ = [γ∈Γ γsγ−1Γ ⊆ [x∈Cs xΓ , and the last union is finite. F C groups are a generalization of both finite and abelian groups, and share many common properties with these classes. They were extensively studied by B. H. Neumann and others, starting with the article [14]. The analogous class of groups in the locally compact setting (groups in which the conjugacy class of any element has compact closure) is usually denoted by F C− and has also been widely studied, since it is a direct generalization of both compact and abelian locally compact groups (see [18, Chapter 12] for an account). When G is a F C-group and Γ ⊆ G is a finite subgroup, we can prove that every double coset ΓsΓ generates a finite co-hereditary set, so that C∗(G, Γ) exists and C∗(G, Γ) ∼= C∗(L1(G, Γ)). To see this, let s ∈ G and {γk}k∈N ⊆ Γ. Also, let Γ = {θ1, . . . , θn} and for each 1 ≤ i ≤ n let us denote by Si ⊆ N the set Of course, the sets Si are mutually disjoint and their union is N. We have that Si := {j ∈ N : γj = θi} . {Γ[s, γ1, . . . , γk]Γ : k ∈ N} = = Now we notice that n[i=1 n[i=1 {Γ[s, γ1, . . . , γk]Γ : k ∈ Si} {Γ[s, γ1, . . . , γk−1, θi]Γ : k ∈ Si} . Γ[s, γ1, . . . , γk−1, θi]Γ = Γ[s, γ1, . . . , γk−1]−1θ−1 i [s, γ1, . . . , γk−1]Γ . Since there are only finitely many conjugates of θ−1 , it follows that the set {Γ[s, γ1, . . . , γk−1, θi]Γ : k ∈ Si} is finite, and therefore {Γ[s, γ1, . . . , γk]Γ : k ∈ N} is finite. Thus, by Theorem 4.2, the co-hereditary set generated by ΓsΓ is finite. i 5.11 G is Locally-Nilpotent and Γ is Finite Recall that a group G is said to be locally-nilpotent if every finitely generated subgroup of G is nilpotent. Let G be a locally-nilpotent group and Γ a finite subgroup. The pair (G, Γ) is automatically a Hecke pair since Γ is finite. We are now going to prove 26 that each double coset ΓsΓ generates a finite co-hereditary set, implying that C∗(G, Γ) exists and coincides with C∗(L1(G, Γ)). To see this, let hs, Γi ⊆ G be the subgroup generated by s and Γ. This subgroup is finitely generated, hence nilpotent. Thus, as we have proven above, ΓsΓ ∈ H(hs, Γi, Γ) ⊆ H(G, Γ) generates a finite co-hereditary set. 5.12 G is Locally-Finite and Γ is Finite Recall that a group G is said to be locally-finite if every finitely generated subgroup of G is finite. Let G be a locally-finite group and Γ a finite subgroup. The pair (G, Γ) is automatically a Hecke pair since Γ is finite. We are now going to prove that each double coset ΓsΓ generates a finite co-hereditary set, implying that C∗(G, Γ) exists and coincides with C∗(L1(G, Γ)). To see this, let hs, Γi ⊆ G be the sub- group generated by s and Γ. This subgroup is finitely generated, hence finite. Thus, as we have proven above, ΓsΓ ∈ H(hs, Γi, Γ) ⊆ H(G, Γ) generates a finite co-hereditary set. An interesting feature of Hecke pairs arising from locally finite groups is that they give rise to AF Hecke algebras. In that regard we have the following result: Proposition 5.7. Let (G, Γ) be a Hecke pair where G is countable and Γ is a fi- nite subgroup. Then H(G, Γ) is an AF ∗-algebra if and only if G is locally finite. Proof: (⇐=) Assume G is locally finite. Since G is assumed countable, let us fix an enumeration of its elements G = {g1, g2, . . .} and for each n ∈ It is clear that N let us define Hn as the subgroup Hn := hΓ, g1, . . . , gni. {Hn}n∈N forms an increasing sequence of finitely generated subgroups, such that contains Γ. Hence, we have a sequence of finite dimensional Hecke algebras S Hn = G. Moreover, since G is locally finite, each Hn is a finite group which {H(Hn, Γ)}n∈N ⊆ H(G, Γ) satisfyingSH(Hn, Γ) = H(G, Γ). Thus, H(G, Γ) is (=⇒) Assume that H(G, Γ) is an AF ∗-algebra. Then any element f ∈ H(G, Γ) lies in a finite dimensional ∗-subalgebra, and is therefore algebraic over C. It then follows from [10, Proposition 2.6] that G is locally finite. an AF ∗-algebra. Example 5.8. Similarly to Example 5.6, let p be a prime number and Zp∞ be the p-quasicyclic group (which is the Pontryagin dual of the group of p-adic integers). The generalized dihedral group is locally finite (but not locally nilpotent unless p = 2). G := Zp∞ ⋊(cid:0)Z/2Z(cid:1) , 27 6 Reduction Techniques Suppose we have Γ ⊆ N for some normal subgroup N of G, i.e. Γ ⊆ N E G . We will see that we can "reduce" the problem of knowing whether H(G, Γ) has an enveloping C∗-algebra to the smaller Hecke ∗-algebra H(N, Γ). This kind of problem has been considered in the literature (see [3, Proposition 2.6] and also [12, Theorem 1.11] with a more general version given in [2, Theorem 5.5]). Sometimes, as presented in the last two references, the structure of H(G, Γ) can be described in terms of H(N, Γ). At this point, however, we will address solely the question of existence of enveloping C∗-algebras, for which our approach is more general. We have the following result: Proposition 6.1. Let (G, Γ) be a Hecke pair and N a normal subgroup of G containing Γ. If H(N, Γ) has an enveloping C∗-algebra (resp. is a BG∗-algebra), then the same is true for H(G, Γ). Moreover, all double cosets of H(N, Γ) generate finite co-hereditary sets if and only if all double cosets of H(G, Γ) do so. Proof: Since H(N, Γ) is a ∗-subalgebra of H(G, Γ), all ∗-representations of H(G, Γ) restrict to ∗-representations of H(N, Γ) (and similarly for pre-∗- representations). Suppose that H(N, Γ) has an enveloping C∗-algebra. Let ΓsΓ ∈ H(G, Γ). By Remark 1.8 we know that (ΓsΓ)∗ ∗ ΓsΓ = nXi=1 λi Γs−1γisΓ for some n ∈ N, λ1, . . . , λn ∈ N and γ1, . . . , γn ∈ Γ. If π is a ∗-representation of H(G, Γ), we have that kπ(ΓsΓ)k = pkπ((ΓsΓ)∗ ∗ ΓsΓ)k ≤ vuut nXi=1 λikπ(Γs−1γ−1 i sΓ)k . We notice that, since N is normal in G and Γ ⊆ N , we have s−1γ−1 i.e. all elements Γs−1γ−1 ∗-representation of H(N, Γ). Hence, we must have i s ∈ N , i sΓ ∈ H(N, Γ). Also, by restriction, π gives rise to a kπ(ΓsΓ)k ≤vuut nXi=1 λikΓs−1γ−1 i sΓku,N , (10) where k · ku,N is the universal norm in H(N, Γ). Since the inequality in (10) holds for every ∗-representation π, we see that ΓsΓ has a bounded universal norm. Thus, H(G, Γ) has an enveloping C∗-algebra. shows that if H(N, Γ) is a BG∗-algebra, then so is H(G, Γ). A similar computation as the above, but with pre-∗-representations instead, 28 Let us now prove the second statement. First we notice that if all dou- ble cosets of H(G, Γ) generate finite co-hereditary sets, then the same is true for H(N, Γ), since H(N, Γ) ⊆ H(G, Γ). Now suppose that all double cosets of H(N, Γ) generate finite co-hereditary sets. Then, given any ΓsΓ ∈ H(G, Γ), we have that all of its successors are of the form Γs−1γsΓ for some γ ∈ Γ, and therefore belong to H(N, Γ), as we saw above. Notice that the co-hereditary set generated by ΓsΓ is the union of the co-hereditary sets generated by its successors (along with the element ΓsΓ itself). Thus, since its successors belong to H(N, Γ), they generate finite co-hereditary sets, and since ΓsΓ has finitely many successors, ΓsΓ also generates a finite co-hereditary set. An easy induction argument allows us to extend the previous result to as- cendant subgroups, and thus, in particular, to subnormal subgroups (recall the definitions given in the sections 5.6 and 5.5): Corollary 6.2. Let (G, Γ) be a Hecke pair and K an ascendant subgroup of is a BG∗- G containing Γ. algebra), then the same is true for H(G, Γ). Moreover, all double cosets of H(K, Γ) generate finite co-hereditary sets if and only if all double cosets of H(G, Γ) do so. If H(K, Γ) has an enveloping C∗-algebra (resp. By combining Proposition 6.1, and the more general Corollary 6.2, together with the results of the previous section, we obtain many more classes of Hecke pairs for which the existence of an enveloping C∗-algebra can be assured. For example, if (G, Γ) is a Hecke pair for which there exists an ascendant nilpotent subgroup K with Γ ⊆ K ⊆ G, then H(G, Γ) has an enveloping C∗-algebra. Of course, one can replace "nilpotent" with any other property discussed in the previous section. As yet another example, we can recover (and slightly improve) a result by Baumgartner, Ramagge and Willis ([3, Proposition 2.6]), concerning Hecke pairs (G, Γ) where Γ has finite index in a normal subgroup N E G. They proved that in this case H(G, Γ) has an enveloping C∗-algebra. Through our Corollary 6.2 we can improve this result in the following way: Corollary 6.3. Let (G, Γ) be a Hecke pair such that Γ has finite index in an ascendant subgroup K ⊆ G. Then H(G, Γ) has an enveloping C∗-algebra which coincides with C∗(L1(G, Γ)). 7 Final Remarks and Questions 1. In this article we studied conditions that imply the existence of a full Hecke C∗-algebra. What about the opposite question: when can we ensure that C∗(G, Γ) does not exist for a given Hecke pair? Even though there are examples of Hecke pairs (G, Γ) for which it is known that C∗(G, Γ) does not exist (see [8, Proposition 2.21] or [21, Example 3.4]), this question seems to be quite delicate in full generality. It would be interesting to have sufficient (group-theoretic) conditions for the non-existence of C∗(G, Γ). 29 2. If C∗(G, Γ) exists, does it necessarily follow that C∗(G, Γ) ∼= C∗(L1(G, Γ))? With the notable exception of the Iwahori Hecke algebras (section 5.3), for which we do not know the answer, this is true for all the other examples we know of. 3. An even stronger question, posed in [9], is still open: if C∗(G, Γ) exists, does it necessarily follow that C∗(G, Γ) ∼= C∗(L1(G, Γ)) ∼= pC∗(G)p ? Many of the new classes studied in this article will be shown in a following article (see [16]) to have this property. 4. Is it true that if C∗(G, Γ) exists, then H(G, Γ) is a BG∗-algebra? All examples of Hecke algebras for which we know that an enveloping C∗- algebra exists are actually BG∗-algebras. It would be interesting to have a counter-example or a proof of this statement. References [1] J. Arledge, M. Laca, I. Raeburn, Semigroup crossed products and Hecke algebras arising from number fields, Doc. Math. 2 (1997), 115-138. [2] U. Baumgartner, J. Foster, J. Hicks, H. Lindsay, B. Maloney, I. Raeburn, J. Ramagge, S. Richardson, Hecke algebras of group extensions, Comm. Algebra 33 (2005), no. 11, 4135-4147. [3] U. Baumgartner, J. Ramagge, G.A. Willis, A compactly generated group whose Hecke algebras admit no bounds on their representations, Glasg. Math. J. 48 (2006), 193-201. [4] J.-B. Bost, A. Connes, Hecke algebras, type III factors and phase transitions with spontaneous symmetry breaking in number theory, Selecta Math. (New Series) 1 (1995), 411-457. [5] B. Brenken, Hecke algebras and semigroup crossed products, Pacific J. Math. 187 (1999), no. 2, 241-262. [6] R. Exel, Hecke algebras for protonormal subgroups, J. Algebra 320 (2008), 1771-1813. [7] H. Glöckner, G. A. Willis, Topologization of Hecke pairs and Hecke C∗- algebras, Topology Proceedings 26 (2001-2002), 565-591. [8] R. W. Hall, Hecke C∗-algebras, Ph.D. thesis, The Pennsylvania State Uni- versity, December 1999. [9] S. Kaliszewski, M. B. Landstad and J. Quigg, Hecke C∗-algebras, Schlicht- ing completions, and Morita equivalence, Proc. Edinb. Math. Soc. (Series 2) 51 (2008), no. 3, 657-695. [10] A. Krieg, Hecke algebras, Mem. Amer. Math. Soc. 87 (1990), No. 435. [11] A. G. Kurosh, The theory of groups, Vol. II, Chelsea Pub. Co., New York, N.Y. (1956), (Translated from the Russian and edited by K. A. Hirsch). 30 [12] M. Laca, N. S. Larsen, Hecke algebras of semidirect products, Proc. Amer. Math. Soc. 131 (2003), 2189-2199. [13] N. S. Larsen, I. Raeburn, Representations of Hecke algebras and dilations of semigroup crossed products, J. London Math. Soc. (2) 66 (2002), 198-212. [14] B. H. Neumann, Groups with finite classes of conjugate elements, Proc. London Math. Soc. (3) 1 (1951), 178-187. [15] R. Palma, C∗-completions of Hecke algebras and crossed products by Hecke pairs, Ph.D. thesis, University of Oslo (2012). [16] R. Palma, Quasi-symmetric group algebras and C∗-completions of Hecke algebras, pre-print (2012), arXiv:1210.3807 . [17] T. W. Palmer, Banach algebras and the general theory of ∗-algebras. Vol. I Algebras and Banach Algebras, Encyclopedia of Mathematics and its Ap- plications, vol. 49, Cambridge University Press, Cambridge (1994). [18] T. W. Palmer, Banach algebras and the general theory of ∗-algebras. Vol. II ∗-Algebras, Encyclopedia of Mathematics and its Applications, vol. 79, Cambridge University Press, Cambridge (2001). [19] D. J. S. Robinson, A course in the theory of groups, Graduate Texts in Mathematics, vol. 80, Springer-Verlag, New York (1993). [20] D. J. S. Robinson, Finiteness conditions and generalized soluble groups. Part 2, Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 63, Springer-Verlag, New York-Berlin (1972). [21] K. Tzanev, Hecke C∗-algebras and amenability, J. Operator Theory 50 (2003), 169-178. 31
1706.08236	1	1706	2017-06-26T05:30:31	The noncommutative L\"owner theorem for matrix monotone functions over operator systems	[ "math.OA", "math.CV", "math.FA" ]	Given a function $f: (a,b) \rightarrow \mathbb{R},$ L\"owner's theorem states $f$ is monotone when extended to self-adjoint matrices via the functional calculus, if and only if $f$ extends to a self-map of the complex upper half plane. In recent years, several generalizations of L\"owner's theorem have been proven in several variables. We use the relaxed Agler, McCarthy and Young theorem on locally matrix monotone functions in several commuting variables to generalize results in the noncommutative case. Specifically, we show that a real free function defined over an operator system must analytically continue to a noncommutative upper half plane as map into another noncommutative upper half plane.	math.OA	math	THE NONCOMMUTATIVE L OWNER THEOREM FOR MATRIX MONOTONE FUNCTIONS OVER OPERATOR SYSTEMS J. E. PASCOE Abstract. Given a function f : (a, b) → R, Lowner's theorem states f is monotone when extended to self-adjoint matrices via the functional calculus, if and only if f extends to a self-map of the complex upper half plane. In recent years, several generaliza- tions of Lowner's theorem have been proven in several variables. We use the relaxed Agler, McCarthy, and Young theorem on locally matrix monotone functions in several commuting variables to gen- eralize results in the noncommutative case. Specifically, we show that a real free function defined over an operator system must an- alytically continue to a noncommutative upper half plane as map into another noncommutative upper half plane. 1. Introduction Let f : (a, b) → R. Lowner answered the question of when such a function is monotone when f is extended to self-adjoint matrices (or even operators in general) via the functional calculus, which has found various applications. Specifically, we say f is matrix monotone on (a, b) ⊂ R if A ≤ B ⇒ f (A) ≤ f (B) whenever A and B are self-adjoint matrices of the same size with spec- trum in (a, b), f is being applied in the sense of the functional calculus on self-adjoint operators, and ≤ is interpreted to mean that the differ- ence is positive semidefinite. Let Π be the upper half plane in C. Lowner's theorem states the following: Theorem 1.1 (Lowner [15]). Let f : (a, b) → R be a bounded Borel function. The function f is matrix monotone if and only if f is real Date: March 19, 2018. 2010 Mathematics Subject Classification. 46L52, 32A70, 30H10. † Partially supported by National Science Foundation Mathematical Science Postdoctoral Research Fellowship DMS 1606260. 1 2 J. E. PASCOE analytic and analytically continues to the upper half plane as a function from Π ∪ (a, b) into Π. For a modern treatment of Lowner's theorem, see e.g. [7, 2, 3]. For applications, see e. g. [5, 14, 24, 23] . Now, one could ask what kind of functions preserve inequalities of, say, two block matrices of size 2 by 2. That is, given a function f (X11, X12, X21, X22) and an inequality (cid:18)A11 A12 A21 A22(cid:19) ≤ (cid:18)B11 B12 B21 B22(cid:19) , when can we say f (A11, A12, A21, A22) ≤ f (B11, B12, B21, B22)? For example, is the function given by the formula X11 − X12X −1 22 X21, the Schur complement, monotone in the above sense on positive block 2 by 2 matrices? It turns out the Schur complement is indeed mono- tone, which has certainly been known for some time[18], and can be shown via elementary arguments– for example, its inverse appears in the formula for block inversion of a matrix as a diagonal entry. How- ever, we are interested in an effective systematic way of classifying such functions as in Lowner's theorem, and that is what we will establish in the noncommutative context. 2. The noncommutative context We now describe the noncommutative context in which we desire to prove a generalization of Lowner's theorem. First, we must give the appropriate generalization of the functional calculus (see [16] for a more thorough introduction). We also note various noncommutative generalizations to the free functional calculus of Lowner's theorem were considered by the current author and Tully-Doyle [22], and by Palfia [19] previously, and to other functional calculi by Hansen [8] and Agler, McCarthy, and Young [1]. Moreover, this work fits into a greater effort to systematize the theory of matrix inequalities [9, 13, 11, 12, 10]. Let R be a real topological vector space. We define the matrix universe over R, denoted M(R), to be M(R) = [n∈N Mn(C) ⊗R R, where Mn(C) denotes the space of n by n matrices. We endow M(R) with the disjoint union topology. Given U ⊂ M(R), we use Un to L OWNER'S THEOREM 3 denote U ∩ Mn(C) ⊗ R. We define the Hermitian matrix universe over R, denoted S(R), to be S(R) = [n∈N Sn(C) ⊗R R, where Sn(C) denotes the space of n by n Hermitian matrices. For a concrete example, if we take R = S2(C), the 2 by 2 Hermitian matrices over C, M(S2(C)) consists of all block 2 by 2 matrices and S(S2(C)) consists of all block 2 by 2 Hermitian matrices. For another example, taking R = R2, the set M(R2) consists of all pairs of same- sized matrices and S(R2) consists of all pairs of same-sized Hermitian matrices. We define a domain D ⊂ M(R) to satisfy the following two axioms: (1) X ⊕ Y ∈ D ⇔ X, Y ∈ D (2) X ∈ Dn ⇒ U ∗XU ∈ D for all n by n unitaries U over C. Let D ⊂ M(R1) be a free domain. We say a function f : D → M(R2) is a free function if (1) f Dn maps into M(R2)n (2) f (X ⊕ Y ) = f (X) ⊕ f (Y ), (3) S −1f (X)S = f (S −1XS) for all n by n invertible matrices S over C such that X, S −1XS ∈ Dn. We note any noncommutative rational expression gives a free function on its domain of definition. For example, the Schur complement, X11 − X12X −1 22 X21, gives a free function on the subset D ⊆ M(S2(C)) where X −1 is defined. For another example, the matrix geometric mean, X 1/2 (X −1/2 defines a free function on the subset D ⊆ S(R2) of all pairs of positive definite matrices. 1 X2X −1/2 22 1 )1/2X 1/2 1 1 If R is a real operator system, that is, R is a real subspace containing 1 in a C ∗ algebra of self-adjoint elements, for each n there is a natural ordering on Sn(C) ⊗ R, since matrices over R are themselves elements of a larger C ∗-algebra. That is, given A, B ∈ Sn(C) ⊗ R, we say A ≤ B if B − A is positive semidefinite as an element of Sn(C) ⊗ R. Accordingly, given R1 and R2 real operator systems and a domain D ⊆ S(R1), we say a free function f : D → S(R2) is matrix mono- tone if A ≤ B ⇒ f (A) ≤ f (B) whenever A and B have the same size. Define Π(R) = {Im X > 0} where A > B if the difference is strictly positive definite, that is, it is self-adjoint and its spectrum is a subset of (0, ∞), and Im X = (X − X ∗)/2i. We show the following theorem. 4 J. E. PASCOE Theorem 2.1 (Noncommutative Lowner theorem over operator sys- tems). Let R1 and R2 be closed real operator systems. Let D ⊆ S(R1) be a free domain. Suppose each Dn is convex and open as a subset of Sn(C) ⊗ R. A function f : D → S(R2) is matrix monotone if and only if f extends to a continuous free function F : Π(R1) ∪ D → Π(R2). We note that such a function must be analytic on each Π(R1)n due to the draconian nature of free functions. See [16]. We also point out that the case where R1 = Rd as a diagonal algebra and R2 = R was explored in [22, 19], and that the current work simplifies the proof of the main result of those works if we are willing to use the commutative Lowner theorem from [1] as a black box. Moreover, if we are given a rational expression, such as the Schur complement, on a nice finite dimensional operator system, such as a matrix algebra, one can apply the algorithms in [10] which make the rational convex Positivstellensatz [21] effective to check that a function is matrix monotone in our sense. Finally, we should comment that the setting of operator systems is equivalent to defining an Archimedian matrix ordering on S(R), where R is an abstract real vector space, by the Choi-Effros Theorem [6]. That is, we might have alternatively defined an ordering on S(R) using any proper closed Archimedian matrix convex cone, but the result is the same. Before we arrive at the proof of our Theorem, we should revisit our Schur complement. Our domain D ⊂ S(S2(C)) is the set of positive definite block 2 by 2 matrices upon which our function, defined by the formula f (cid:18)X11 X12 X21 X22(cid:19) = X11 − X12X −1 22 X21, is a free function f : D → S(R). According to our Theorem, f will be matrix monotone if and only if f extends to a continuous free function from D ∪ Π(S2(C)) to Π(R). It is clear that extension of f to the new domain must still be given by the same formula as before. Either using the algorithms in [10, 21] or by brute force, one can see that Im f = (cid:18) 1 12(cid:19) (X ∗ 22)−1X ∗ 1 12(cid:19)∗(cid:20)Im (cid:18)X11 X12 X21 X22(cid:19)(cid:21)(cid:18) (X ∗ 22)−1X ∗ which is manifestly positive definite whenever Im (cid:18)X11 X12 X21 X22(cid:19) is pos- itive definite– that is f maps Π(S2(C)) to Π(R). That is, our Theorem now implies that the Schur complement is matrix monotone. L OWNER'S THEOREM 5 Another example of a matrix monotone function, is the matrix geo- metric mean and various generalizations, see [17, 4]. In the two pa- rameter case it is not immediately clear to the author how to show the function X 1/2 continues to a map from Π(R2) to Π(R) without going through the generalization of Lowner's theorem. 1 X2X −1/2 (X −1/2 1 )1/2X 1/2 1 1 3. The proof of the main result (⇒) The proof will go by viewing, for each n, f Dn as a matrix monotone function in several commuting variables in the sense of Agler, McCarthy, and Young. Agler, McCarthy, and Young extended Lowner's theorem to several commuting variables for the class of locally matrix monotone func- tions [1]. Subsequently, it was generalized to remove some technical assumptions by the author in [20]. Let E be an open subset of Rd. Let CSAM d n(E) denote the d-tuples of commuting self-adjoint ma- trices of size n with joint spectrum contained in E. We say that a function f : E → R is locally matrix monotone if for any C 1 path γ : (−1, 1) → CSAM d n(E) such that γ ′(0)i > 0, there exists an ǫ > 0 such that for all −ǫ < t1 < t2 < ǫ, g(γ(t1)) ≤ g(γ(t2)). We recall the following theorem. Theorem 3.1 (Agler, McCarthy, and Young [1], Pascoe [20]). Let E be an open subset of Rd. Let g : E → R be a locally matrix monotone function. Then g is analytic, and g extends to a (unique) continuous function on Πd ∪ E which maps into Π which is analytic on Πd. We note that the original formulation of Agler, McCarthy, and Young applied only to C 1 functions g, and via an argument using mollifiers it was shown that the theorem holds for arbitrary functions. We note that is sufficient to show that on each nonempty Dn, our function f analytically continues to Π(R1)n taking values in Π(R2)n. It is an elementary, but perhaps somewhat involved, exercise to show that the induced extension of f will be a free function on Π(R1). Namely, the edge-of-the-wedge theorem will ensure that the extension of f actually analytically continues through each Dn as a function on Π(R1)n ∪ Dn and the rest of the properties will follow by analytic continuation. Now, we note that it is sufficient to show that for every (completely) positive unital linear functional l : M(R2)n → C that fn,l = l ◦ f Dn extends analytically to Dn∪Π(R1)n as a map taking values with positive imaginary part. This is obvious when R2 is finite dimensional, and an exercise in functional analysis otherwise. 6 J. E. PASCOE Fix P ∈ Dn. Let K1, . . . Km > 0 be positive elements of S(R1)n. Let C be the cone generated by K1, . . . , Km and let S be the span of K1, . . . , Km. We will show that fl,n uniquely analytically continues to P + S + iC. Taking larger and larger sets of Ki will give an analytic continuation fn,l to the whole of Π(R1)n. That is, the sets P + S + iC exhaust Π(R1)n. Define the function g(h) = fl(P +Pi hiKi)). Now g(h) is a locally matrix monotone function in the sense of Theorem 3.1 as a function on Rm which induces the unique analytic continuation of fn,l to the desired space taking values in Π. So, we are done. (⇐) The converse direction is easy and follows from a computation of the derivative for directions pointing into the upper half plane. See [22, Lemma 4.8] where the details are essentially the same. References [1] J. Agler, J.E. McCarthy, and N.J. Young. Operator monotone functions and Lowner functions of several variables. Ann. of Math., 176:1783–1826, 2012. [2] R. Bhatia. Matrix Analysis. Princeton University Press, Princeton, 2007. [3] R. Bhatia. Positive Definite Matrices. Springer-Verlag, New York, 2007. [4] Rajendra Bhatia and Rajeeva L. Karandikar. Monotonicity of the matrix geo- metric mean. Mathematische Annalen, 353(4):1453–1467, 2012. [5] Holger Boche and Eduard Axel Jorswieck. Optimization of matrix monotone functions: Saddle-point, worst case noise analysis, and applications. In Proc. of ISIT, 2004. [6] Man-Duen Choi and Edward G. Effros. Injectivity and operator spaces. Journal of Functional Analysis, 24(2):156 – 209, 1977. [7] W.F. Donoghue. Monotone matrix functions and analytic continuation. Springer, Berlin, 1974. [8] F. Hansen. Operator monotone functions of several variables. Math. Inequal. Appl., 6:1–17, 2003. [9] J. W. Helton. Positive noncommuative polynomials are sums of squares. Ann. of Math., 156(2):675–694, 2002. [10] J.W. Helton, I. Klep, S. McCullough, and C. Nelson. Noncommutative poly- nomials nonnegative on a variety intersect a convex set. J. Funct. Anal., 266(12):6684–6752, 2014. [11] J.W. Helton and S. McCullough. Convex noncommutative polynomials have degree two or less. SIAM J. Matrix Anal. & Appl., 25(4):11241139, 2004. [12] J.W. Helton and S. McCullough. A Positivstellensatz for non-commutative polynomials. Trans. AMS, 356:3721–3737, 2004. [13] J.W. Helton and S. McCullough. Every convex free basic semi-algebraic set has an LMI representation. Ann. of Math., 176(2):979–1013, 2012. [14] Eduard Axel Jorswieck and Holger Boche. Majorization and Matrix Monotone Functions in Wireless Communications, volume 3 of Foundations and Trends in Communications and Information Theory. Now publishers, July 2007. pp. 553–701. [15] K. Lowner. Uber monotone Matrixfunktionen. Math. Z., 38:177–216, 1934. L OWNER'S THEOREM 7 [16] D. S. Kaliuzhnyi-Verbovetskyi and V. Vinnikov. Foundations of Noncommu- tative Function Theory. ArXiv e-prints, 2012. [17] Jimmie Lawson and Yongdo Lim. Monotonic properties of the least squares mean. Mathematische Annalen, 351(2):267–279, 2011. [18] Jianzhou Liu and Jian Wang. Some inequalities for schur complements. Linear Algebra and its Applications, 293(1):233 – 241, 1999. [19] Miklos Palfia. Loewner's theorem in several variables. arXiv:1405.5076, 2016. [20] J E. Pascoe. Note on Lowner's theorem on matrix monotone functions in sev- eral commuting variables of Agler, Mccarthy and Young. submitted. [21] J E. Pascoe. Positivstellensatze for noncommutative rational expressions. Proc. Amer. Math. Soc., 2017. to appear. [22] J E. Pascoe and Ryan Tully-Doyle. Free Pick functions: representations, as- ymptotic behavior and matrix monotonicity in several noncommuting vari- ables. J. Funct. Anal., 273:283–328, 2017. [23] E. Wigner. On a class of analytic functions from the quantum theory of colli- sions. Ann. of Math., 53(1):36–67, 1951. [24] E. Wigner and J. v. Neumann. Significance of Lowner's theorem in the quan- tum theory of collisions. Ann. of Math., 59(2):418–433, 1954.
1004.0904	3	1004	2010-10-28T14:11:10	Langlands reciprocity for the even dimensional noncommutative tori	[ "math.OA", "math.NT" ]	We conjecture an explicit formula for the higher dimensional Dirichlet character; the formula is based on the K-theory of the so-called noncommutative tori. It is proved, that our conjecture is true for the two-dimensional and one-dimensional (degenerate) noncommutative tori; in the second case, one gets a noncommutative analog of the Artin reciprocity law.	math.OA	math	Langlands reciprocity for the even dimensional noncommutative tori Igor Nikolaev ∗ Abstract We conjecture an explicit formula for the higher dimensional Dirich- let character; the formula is based on the K-theory of the so-called noncommutative tori. It is proved, that our conjecture is true for the two-dimensional and one-dimensional (degenerate) noncommuta- tive tori; in the second case, one gets a noncommutative analog of the Artin reciprocity law. Key words and phrases: Langlands program, noncommutative tori AMS Subj. Class.: 11M55; 46L85 Introduction The aim of the underlying note is to bring some evidence in favor of the following analog of the Langlands reciprocity [5]: Conjecture 1 (Langlands conjecture for noncommutative tori) Let K be a finite extension of the rational numbers Q with the Galois group Gal (KQ); for an irreducible representation σn+1 : Gal (KQ) → GLn+1(C), there exists a 2n-dimensional noncommutative torus with real multiplication, A2n RM , s), where L(σn+1, s) is the Artin L- function and L(A2n RM . Moreover, A2n RM is the image of an n-dimensional abelian variety Vn(K) under the (general- ized) Teichmuller functor Fn. RM , s) an L-function attached to the A2n RM , such that L(σn+1, s) ≡ L(A2n ∗Partially supported by NSERC. 1 For the notation and terminology we refer the reader to sections 1 and 3; the noncommutative torus A2n RM can be regarded as a substitute of the "auto- morphic cuspidal representation πσn+1 of the group GL(n + 1)" in terms of the Langlands theory. Roughly speaking, conjecture 1 says, that the Galois extensions of the field of rational numbers come from the even dimensional noncommutative tori with real multiplication. Note, that the noncommu- tative tori are intrinsic to the problem, since they classify the irreducible (infinite-dimensional) representations of the Lie group GL(n + 1) [11]; such representations are the heart of the Langlands program [5]. Our conjecture is supported by the following evidence. Theorem 1 Conjecture 1 is true for n = 1 (resp., n = 0) and K abelian extension of an imaginary quadratic field k (resp., the rational field Q). The structure of the note is as follows. A minimal necessary notation is introduced in section 1 and a brief summary of the Teichmuller functor(s) is given in section 3. Theorem 1 is proved in section 2. 1 Preliminaries 1.1 Noncommutative tori i u−1 j A. The k-dimensional noncommutative tori ([4],[12]). A noncommu- tative k-torus is the universal C ∗-algebra generated by k unitary operators u1, . . . , uk; the operators do not commute with each other, but their com- mutators uiuju−1 are fixed scalar multiples exp (2πiθij), θij ∈ R of the identity operator. The k-dimensional noncommutative torus, Ak Θ, is defined by a skew symmetric real matrix Θ = (θij), 1 ≤ i, j ≤ k. Further, we think of the Ak Θ as a noncommutative topological space, whose algebraic K-theory Θ) ∼= Z2k−1. The canonical trace τ on the yields K0(Ak C ∗-algebra Ak Θ) to the real line R; under the homomorphism, the image of K0(Ak Θ) is a Z-module, whose gener- ators τ = (τi) are polynomials in θij. (More precisely, τ = exp(Θ), where the exterior algebra of θij is nilpotent.) Recall, that the C ∗-algebras A and A′ are said to be stably isomorphic (Morita equivalent), if A ⊗ K ∼= A′ ⊗ K for the C ∗-algebra K of compact operators; such an isomorphism indicates, that the C ∗-algebras are homeomorphic as noncommutative topological spaces. By a result of Rieffel and Schwarz [13], the noncommutative tori Ak Θ and Θ) ∼= Z2k−1 and K1(Ak Θ defines a homomorphism from K0(Ak 2 Θ′ are stably isomorphic, if the matrices Θ and Θ′ belong to the same orbit Ak of a subgroup SO(k, k Z) of the group GL2k(Z), which acts on Θ by the formula Θ′ = (AΘ + B) / (CΘ + D), where (A, B, C, D) ∈ GL2k(Z) and the matrices A, B, C, D ∈ GLk(Z) satisfy the conditions: AtD + C tB = I, AtC + C tA = 0 = BtD + DtB. (1) (Here I is the unit matrix and t at the upper right of a matrix means a transpose of the matrix.) The group SO(k, k Z) can be equivalently defined as a subgroup of the group SO(k, k R) consisting of linear transformations of the space R2k, which preserve the quadratic form x1xk+1+x2xk+2+. . .+xkx2k. B. The even dimensional normal tori. Further, we restrict to the case k = 2n (the even dimensional noncommutative tori). It is known, that by the orthogonal linear transformations every (generic) real even dimensional skew symmetric matrix can be brought to the normal form: 0 −θ1 θ1 0 . . . Θ0 =     0 −θn θn 0 (2) Θ0)) = Z + θ1Z + . . . + θnZ + P22n−1 where θi > 0 are linearly independent over Q. We shall consider the non- commutative tori A2n Θ0, given by the matrix (2); we refer to the family as a normal family. Recall, that any noncommutative torus has a canonical trace Θ) ∼= Z2k−1 to R; it follows from τ , which defines a homomorphism from K0(Ak [4], that the image of K0(A2n Θ0) under the homomorphism has a basis, given by the formula τ (K0(A2n i=n+1 pi(θ)Z, where pi(θ) ∈ Z[1, θ1, . . . , θn]. C. The real multiplication ([8]). The noncommutative torus Ak Θ is said to have a real multiplication, if the endomorphism ring End (τ (K0(Ak Θ))) ex- ceeds the trivial ring Z. Since any endomorphism of the Z-module τ (K0(Ak Θ)) is the multiplication by a real number, it is easy to deduce, that all the entries of Θ = (θij) are algebraic integers. (Indeed, the endomorphism is described by an integer matrix, which defines a polynomial equation involving θij.) Thus, the noncommutative tori with real multiplication is a countable sub- set of all k-dimensional tori; any element of the set we shall denote by Ak RM . Notice, that for the even dimensional normal tori with real multiplication, 3 the polynomials pi(θ) produce the algebraic integers in the extension of Q by θi; any such an integer is a linear combination (over Z) of the θi. Thus, the trace formula reduces to τ (K0(A2n RM )) = Z + θ1Z + . . . + θnZ. 1.2 L-function of noncommutative tori We consider even dimensional normal tori with real multiplication. Denote by A a positive integer matrix, whose (normalized) Perron-Frobenius eigenvector coincides with the vector θ = (1, θ1, . . . , θn) and A is not a power of a positive integer matrix; in other words, Aθ = λAθ, where A ∈ GLn+1(Z) and λA is the corresponding eigenvalue. (Explicitly, A can be obtained from vector θ as the matrix of minimal period of the Jacobi-Perron continued fraction of θ [2].) Let p be a prime number; take the matrix Ap and consider its characteristic polynomial char (Ap) = xn+1 + a1xn + . . . + anx + 1. We introduce the following notation: Ln+1 p := a1 −1 ... 0   a2 0 ... 0 . . . an . . . 0 ... . . . . . . −1 p 0 ... 0   . (3) A local zeta function of the A2n Ln+1 z); in other words, p RM is defined as the reciprocal of det (In+1 − ζp(A2n RM , z) := 1 − a1z + a2z2 − . . . − anzn + pzn+1 , 1 z ∈ C. (4) An L-function of A2n number, of primes L(A2n RM is a product of the local zetas over all, but a finite RM , s) = Qp ∤ tr2(A)−(n+1)2 ζp(A2n RM , p−s), s ∈ C. Remark 1 It will be shown, that for n = 0 and n = 1 formula (4) fits conjecture 1; for n ≥ 2 it is an open problem based on an observation, that the crossed product A2n Z is a proper noncommutative analog of the (higher dimensional) Tate module, where matrix Ln+1 corresponds to the Frobenius automorphism of the module [14], p.172. RM ⋊Ln+1 p p 4 2 Proof of theorem 1 2.1 Case n = 1 Each one-dimensional abelian variety is a non-singular elliptic curve; choose this curve to have complex multiplication by (an order in) the imaginary quadratic field k and denote such a curve by ECM . Then, by theory of com- plex multiplication, the (maximal) abelian extension of k coincides with the ∼= E(K) [14]. The minimal field of definition of the curve ECM , i.e. ECM Teichmuller functor F := F1 maps E(K) into a two-dimensional noncommu- tative torus with real multiplication (section 3); we shall denote the torus by A2 RM , s), let A be a 2 × 2 positive integer matrix, whose normalized Perron-Frobenius eigenvenctor is (1, θ1). For a prime p, the characteristic polynomial of the matrix Ap writes as char (Ap) = x2 + tr (Ap)x + 1 and the matrix L2 RM . To calculate the corresponding L-function L(A2 p takes the form: p = (cid:18) tr (Ap) p 0(cid:19) . L2 RM , z) = (1 − tr (Ap)z + pz2)−1. The corresponding local zeta function ζp(A2 We have to prove, that ζp(A2 RM , z) = ζp(ECM , z), where ζp(ECM , z) is the local zeta function for the elliptic curve ECM ; the proof will be arranged into a series of lemmas 1-5. −1 (5) Recall, that ζp(ECM , z) = (1−tr (ψE(K)(P))z +pz2)−1, where ψE(K) is the Grossencharacter on K, P the prime ideal of K over p and tr is the trace of algebraic number [14], Ch.2, §9. Roughly, our proof consists in construction of representation ρ of ψE(K) into the group of invertible elements (units) of End (τ (K0(A2 RM ))), such that tr (ψE(K)(P)) = tr (ρ(ψE(K)(P))) = tr (Ap). This will be achieved with the help of an explicit formula for the Teichmuller functor F ([10], p.524): F : (cid:18) a c b d(cid:19) ∈ End (ECM ) 7−→ (cid:18) a −c −d(cid:19) ∈ End (ARM ). b (6) Lemma 1 Let A = (a, b, c, d) be an integer matrix with ad − bc 6= 0 and b = 1. Then A is similar to the matrix (a + d, 1, c − ad, 0). Proof. Indeed, consider a matrix (1, 0, d, 1) ∈ SL2(Z); it is verified directly, that the matrix realizes the required similarity. (cid:3) 5 Lemma 2 The matrix A = (a + d, 1, c − ad, 0) is similar to its transpose At = (a + d, c − ad, 1, 0). Proof. We shall use the following criterion: the (integer) matrices A and B are similar, if and only if the characteristic matrices xI − A and xI − B have the same Smith normal form. The calculation for the matrix xI − A gives: ad − c (cid:18) x − a − d −1 ∼ (cid:18) 1 x (cid:19) ∼ (cid:18) x2 − (a + d)x + ad − c 0 x2 − (a + d)x + ad − c(cid:19) , x − a − d 0 −1 0 (cid:19) ∼ where ∼ are the elementary operations between the rows (columns) of the matrix. Similarly, a calculation for the matrix xI − At gives: x −1 (cid:18) x − a − d ad − c ∼ (cid:18) 1 (cid:19) ∼ (cid:18) x − a − d x2 − (a + d)x + ad − c 0 x2 − (a + d)x + ad − c(cid:19) . Thus, (xI − A) ∼ (xI − At) and lemma 2 follows. (cid:3) −1 0 0 (cid:19) ∼ Corollary 1 The matrices (a, 1, c, d) and (a + d, c − ad, 1, 0) are similar. Let ECM be elliptic curve with complex multiplication by an order R in the ring of integers of the imaginary quadratic field k. Then ARM = F (ECM ) is a noncommutative torus with real multiplication by the order R in the ring of integers of a real quadratic field k (section 3). Let tr (α) = α + ¯α be the trace function of a (quadratic) algebraic number field. Lemma 3 Each α ∈ R goes under F into an ω ∈ R, such that tr (α) = tr (ω). Proof. Recall that each α ∈ R can be written in a matrix form for a given base {ω1, ω2} of the lattice Λ. Namely, αω1 = aω1 +bω2 and αω2 = cω1 +dω2, where (a, b, c, d) is an integer matrix with ad−bc 6= 0. Note that tr (α) = a+d and bτ 2 + (a − d)τ − c = 0, where τ = ω2/ω1. Since τ is an algebraic integer, we conclude that b = 1. In view of corollary 1, in a base {ω′ 2}, the α has a matrix form (a + d, c − ad, 1, 0). To calculate ω ∈ R corresponding to α, we apply formula (6), which gives us: 1, ω′ F : (cid:18) a + d c − ad 1 0 (cid:19) 7−→ (cid:18) a + d c − ad −1 0 (cid:19) . (7) 6 In a given base {λ1, λ2} of the pseudo-lattice Z + Zθ one can write ωλ1 = (a + d)λ1 + (c − ad)λ2 and ωλ2 = −λ1. It is an easy exercise to verify that ω is a real quadratic integer with tr (ω) = a + d; the latter coincides with the tr (α). (cid:3) Let ω ∈ R be an endomorphism of the pseudo-lattice Z + Zθ of degree deg (ω) := ω ¯ω = n. The endomorphism maps pseudo-lattice to a sub-lattice of index n. Any such has a form Z + (nθ)Z [3], p.131. Let us calculate ω in a base {1, nθ}, when ω is given by the matrix (a + d, c − ad, −1, 0). In this case n = c − ad and ω induces an automorphism ω∗ = (a + d, 1, −1, 0) of the sublattice Z + (nθ)Z according to the matrix equation: (cid:18) a + d n −1 0 (cid:19)(cid:18) 1 θ(cid:19) = (cid:18) a + d 1 −1 0(cid:19)(cid:18) 1 nθ(cid:19) . (8) Thus, one gets a map ρ : R → R∗ given by the formula ω = (a+d, n, −1, 0) 7→ ω∗ = (a + d, 1, −1, 0), where R∗ is the group of units of R. Since tr (ω∗) = a + d = tr (ω) and ω∗ = ρ(ω), one gets the following Corollary 2 For all ω ∈ R, it holds tr (ω) = tr (ρ(ω)). ρ Note, that R∗ = {±εk k ∈ Z}, where ε > 1 is a fundamental unit of the order R ⊆ Ok; here Ok means the ring of integers of a real quadratic field k = Q(θ). Choosing a sign in front of εk, the following index map is defined ι : R F−→ R −→ R∗ −→ Z. Let α ∈ R and deg (α) = −n. To calculate the ι(α), recall some notation from Hasse [6], §16.5.C. Let Z/nZ be a cyclic group of order n. For brevity, let I = Z + Zθ be a pseudo-lattice and In = Z + (nθ)Z its sub-lattice of index n; the fundamental units of I and In are ε and εn, respectively. By Gn one understands a subgroup of Z/nZ of prime residue classes mod n. The gn ⊂ Gn is a subgroup of the non-zero divisors of the Gn. Finally, let gn be the smallest number, such that it divides Gn/gn and εgn ∈ In. (The notation drastically simplifies in the case n = p is a prime number.) Lemma 4 ι(α) = gn. Proof. Notice, that deg (ω) = −deg (α) = n, where ω = F (α). Then the map ρ defines I and In; one can now apply the calculation of [6], pp 296-300. Namely, Theorem XIII ′ on p. 298 yields the required result. (We kept the notation of the original.) (cid:3) 7 Corollary 3 ι(ψE(K)(P)) = p. Proof. It is known, that deg (ψE(K)(P)) = −p, where ψE(K)(P) ∈ R is the Grossencharacter. To calculate the gn in the case n = p, notice that the ∼= Z/pZ and gp is trivial. Thus, Gp/gp = p is divisible only by 1 or p. Gp Since ε1 is not in In, one concludes that gp = p. The corollary follows. (cid:3) Lemma 5 tr (ψE(K)(P)) = tr (Ap). Proof. It is not hard to see, that A is a hyperbolic matrix with the eigenvector (1, θ); the corresponding (Perron-Frobenius) eigenvalue is a fundamental unit ε > 1 of the pseudo-lattice Z + Zθ. In other words, A is a matrix form It is immediate, that Ap is the matrix form of the algebraic number ε. for the εp and tr (Ap) = tr (εp). In view of lemma 3 and corollary 3, tr (α) = tr (F (α)) = tr (ρ(F (α))) for ∀α ∈ R. In particular, if α = ψE(K)(P) then, by corollary 3, one gets ρ(F (ψE(K)(P))) = εp. Taking traces in the last equation, we obtain the conclusion of lemma 5. (cid:3) The fact ζp(A2 RM , z) = ζp(ECM , z) follows from lemma 5, since the trace of the Grossencharacter coincides with such for the matrix Ap. Let ECM be an elliptic curve with complex multiplication by an order in the imaginary quadratic field k and K the minimal field of definition of the ECM . Lemma 6 L(ECM , s) ≡ L(σ2, s), where L(ECM , s) is the Hasse-Weil L- function of ECM and L(σ2, s) the Artin L-function for an irreducible repre- sentation σ2 : Gal (Kk) → GL2(C). e.g. [14], By the Deuring theorem (see K → C∗; here A∗ Proof. p.175), L(ECM , s) = L(ψK, s)L(ψK, s), where L(ψK, s) is the Hecke L-series at- tached to the Grossencharacter ψ : A∗ K denotes the adele ring of the field K and the bar means a complex conjugation. Notice, that since our elliptic curve has complex multiplication, the group Gal (Kk) is abelian; one can apply Theorem 5.1 [7], which says that the Hecke L-series L(σ1 ◦θKk, s) equals the Artin L-function L(σ1, s), where ψK = σ ◦θKk is the Grossencharacter and θKk : A∗ K → Gal (Kk) the canonical homomorphism. Thus, one gets L(ECM , s) ≡ L(σ1, s)L(σ1, s), where σ1 : Gal (Kk) → C means a (complex) conjugate representation of the Galois group. Consider the local factors of the Artin L-functions L(σ1, s) and L(σ1, s); it is immedi- ate, that they are (1 − σ1(F rp)p−s)−1 and (1 − σ1(F rp)p−s)−1, respectively. 8 Let us consider a representation σ2 : Gal (Kk) → GL2(C), such that σ2(F rp) = (cid:18) σ1(F rp) 0 0 σ1(F rp)(cid:19) . (9) It can be verified, that det−1(I2 − σ2(F rp)p−s) = (1 − σ1(F rp)p−s)−1(1 − σ1(F rp)p−s)−1, i.e. L(σ2, s) = L(σ1, s)L(σ1, s). Lemma 6 follows. (cid:3) By lemma 6, we conclude, that L(A2 RM , s) ≡ L(σ2, s) for an irreducible representation σ2 : Gal (Kk) → GL2(C). It remains to notice that L(σ2, s) = L(σ′ 2 : Gal (KQ) → GL2(C) [1], §3. Case n = 1 of theorem 1 follows. (cid:3) 2, s), where σ′ 2.2 Case n = 0 When n = 0, one gets a one-dimensional (degenerate) noncommutative torus; such an object, AQ, can be obtained from the 2-dimensional torus A2 θ by forcing θ = p/q ∈ Q be a rational number (hence our notation). One can always assume θ = 0 and, thus, τ (K0(AQ)) = Z. To calculate matrix L1 p, notice that the group of automorphisms of the Z-module τ (K0(AQ)) = Z is trivial, i.e. is a multiplication by ±1; hence our 1 × 1 (real) matrix A is either 1 or −1. Since A must be positive, we conclude, that A = 1. However, A = 1 is not a prime matrix, if one allows the complex entries; indeed, for any N > 1 matrix A′ = ζN gives us A = (A′)N , where ζN = e N is the N-th p = tr (Ap) = Ap = ζ p root of unity. Therefore, A = ζN and L1 N . A degenerate noncommutative torus, corresponding to the matrix A = ζN , we shall write as AN Q ; in turn, such a torus is the image (under the Teichmuller functor) of a zero-dimensional abelian variety, which we denote by V N 0 . Suppose that Gal (KQ) is abelian and let σ : Gal (KQ) → C× be a homomorphism. Then, by the Artin reciprocity [5], there exists an integer Nσ and a Dirichlet character χσ : (Z/NσZ)× → C×, such that σ(F rp) = χσ(p); choose our zero- dimensional variety be V Nσ Nσ; on the other hand, it is verified directly, that ζ p p = χσ(p). To obtain a local zeta function, we substitute a1 = L1 p into the formula (4) and get . In view of the notation, L1 p = χσ(p). Thus, L1 Nσ = e 2πi Nσ p = ζ p 2πi 0 ζp(ANσ Q , z) = 1 1 − χσ(p)z , (10) where χσ(p) is the Dirichlet character. Therefore, L(ANσ Q , s) ≡ L(s, χσ) is the Dirichlet L-series; such a series, by construction, coincides with the Artin 9 L-series of the representation σ : Gal (KQ) → C×. Case n = 0 of theorem 1 follows. (cid:3) 3 Teichmuller functors Denote by Λ a lattice of rank 2n; recall, that an n-dimensional (principally polarized) abelian variety, Vn, is the complex torus Cn/Λ, which admits an embedding into a projective space [9]. 3.1 Abelian varieties of dimension n = 1 ∼= C/(Z + A. Basic example. Let n = 1 and consider the complex torus V1 τ Z); it always embeds (via the Weierstrass ℘ function) into a projective space as a non-singular elliptic curve. Let H = {τ = x + iy ∈ C y > 0} be the upper half-plane and ∂H = {θ ∈ R y = 0} its (topological) boundary. We identify V1(τ ) with the points of H and A2 θ with the points of ∂H. Let us show, that the boundary is natural; the latter means, that the action of the modular group SL2(Z) extends to the boundary, where it coincides with the stable isomorphisms of tori. Indeed, conditions (1) are equivalent to A = (cid:18) a 0 0 a(cid:19) , B = (cid:18) 0 −b b 0(cid:19) , C = (cid:18) 0 −c 0 (cid:19) , D = (cid:18) d 0 0 d(cid:19) , cθ+d, − aθ+b where ad−bc = 1, a, b, c, d ∈ Z and Θ′ = (AΘ+B)/(CΘ+D) = (0, aθ+b Therefore, θ′ = (aθ + b)(cθ + d)−1 for a matrix (a, b, c, d) ∈ SL2(Z). Thus, the action of SL2(Z) extends to the boundary ∂H, where it induces stable isomorphisms of the noncommutative tori. (11) c cθ+d, 0). B. The Teichmuller functor ([10]). There exists a continuous map F1 : H → ∂H, which sends isomorphic complex tori to the stably isomor- phic noncommutative tori. An exact result is this. Let φ be a closed form on the torus, whose trajectories define a measured foliation; according to the Hubbard-Masur theorem (applied to the complex tori), this foliation corre- sponds to a point τ ∈ H. The map F1 : H → ∂H is defined by the formula τ 7→ θ = Rγ2 φ/Rγ1 φ, where γ1 and γ2 are generators of the first homology of the torus. The following is true: (i) H = ∂H × (0, ∞) is a trivial fiber bundle, whose projection map coincides with F1; (ii) F1 is a functor, which sends isomorphic complex tori to the stably isomorphic noncommutative tori. We 10 shall refer to F1 as the Teichmuller functor. Recall, that the complex torus C/(Z + τ Z) is said to have a complex multiplication, if the endomorphism ring of the lattice Λ = Z + τ Z exceeds the trivial ring Z; the complex mul- tiplication happens if and only if τ is an algebraic number in an imaginary quadratic field. The following is true: F1(V CM is a torus with complex multiplication. RM , where V CM ) = A2 1 1 3.2 Abelian varieties of dimension n ≥ 1 2 n(n+1) A. The Siegel upper half-space ([9]). The space Hn := {τ = (τi) ∈ Im (τi) > 0} of symmetric n × n matrices with complex entries C is called a Siegel upper half-space; the points of Hn are one-to-one with the n-dimensional principally polarized abelian varieties. Let Sp(2n, R) be the symplectic group; it acts on Hn by the linear fractional transformations τ → τ ′ = (aτ + b)/(cτ + d), where (a, b, c, d) ∈ Sp(2n, R) and a, b, c and d are the n × n matrices with real entries. The abelian varieties Vn and V ′ n are isomorphic, if and only if, τ and τ ′ belong to the same orbit of the group Sp(2n, Z); the action is discontinuous on Hn [9], Ch.2, §4. Denote by Σ2n a space of the 2n-dimensional normal noncommutative tori. The following lemma is critical. Lemma 7 Sp(2n, R) ⊆ O(n, nR). Proof. (i) The group O(n, n R) can be defined as a subgroup of GL2(R), which preserves the quadratic form f (x1, . . . , x2n) = x1xn+1 + x2xn+2 + . . . + xnx2n [13]. We shall denote ui = x1 for 1 ≤ i ≤ n and vi = xi for n + 1 ≤ i ≤ 2n. Consider the following skew symmetric bilinear form q(u, v) = u1vn+1 + . . . + unv2n − un+1v1 − . . . − u2nvn, where u, v ∈ R2n. It is known, that each linear substitution g ∈ Sp(2n, R) preserves the form q(u, v). Since q(u, v) = f (x1, . . . , x2n) − un+1v1 − . . . − u2nvn, one concludes that g also preserves the form f (x1, . . . , x2n), i.e. g ∈ O(n, nR). It is easy to see, that the inclusion is proper except the case n = 1, i.e. when Sp(2, R) ∼= O(1, 1R) ∼= SL2(R). Lemma follows. (ii) We wish to give a second proof of this important fact, which is based on the explicit formulas for the block matrices A, B, C and D. The fact that a symplectic linear transformation preserves the skew symmetric bilinear form 11 q(u, v) can be written in a matrix form: (cid:18) a c b d(cid:19)t (cid:18) 0 −I I 0(cid:19)(cid:18) a c b d(cid:19) = (cid:18) 0 −I I 0(cid:19) , (12) where t is the transpose of a matrix. Performing the matrix multiplication, one gets the following matrix identities atd−ctb = I, atc−cta = 0 = btd−dtb. Let us show, that these identities imply the Rieffel-Schwarz identities (1) imposed on the matrices A, B, C and D. Indeed, in view of the formulas (11), the Rieffel-Schwarz identities can be written as: 0 0 0 (cid:18) at (cid:18) at (cid:18) 0 −bt bt at (cid:19)(cid:18) d 0 at (cid:19)(cid:18) 0 −c 0 (cid:19)(cid:18) d 0 0 d(cid:19) +(cid:18) 0 0 (cid:19) +(cid:18) 0 0 d(cid:19) +(cid:18) dt −ct 0 0 c −ct ct b 0 ct −b 0 (cid:19)(cid:18) 0 0 (cid:19)(cid:18) a 0 dt (cid:19)(cid:18) 0 0(cid:19) = (cid:18) I 0 a(cid:19) = (cid:18) 0 0(cid:19) = (cid:18) 0 −b 0 0 0 b 0 0 I (cid:19) 0(cid:19) 0(cid:19) . 0 (13)   A step by step matrix multiplication in (13) shows that the identities atd − ctb = I, atc − cta = 0 = btd − dtb imply the identities (13). (Beware: the operation is not commutative.) Thus, any symplectic transformation satisfies the Rieffel-Schwarz identities, i.e. belongs to the group O(n, nR). Lemma 7 follows. (cid:3) B. The generalized Teichmuller functors. By lemma 7, the action of Sp(2n, Z) on the Hn extends to the Σ2n, where it acts by stable isomorphisms of the noncommutative tori; thus, Σ2n is a natural boundary of the Siegel upper half-space Hn. However, unless n = 1, the Σ2n is not a topological boundary of Hn. Indeed, dimR(Hn) = n(n + 1) and dimR(∂Hn) = n2 + n − 1, while dimR(Σ2n) = n. Thus, Σ2n is an n-dimensional subspace of the topo- logical boundary of Hn; this subspace is everywhere dense in ∂Hn, since the Sp(2n, Z)-orbit of an element of Σ2n is everywhere dense in ∂Hn [13]. A (conjectural) continuous map Fn : Hn → Σ2n, we shall call a generalized Te- ichmuller functor. The Fn has the following properties: (i) it sends each pair of isomorphic abelian varieties to a pair of the stably isomorphic even dimen- sional normal tori; (ii) the range of Fn on the abelian varieties with complex multiplication consists of the noncommutative tori with real multiplication. As explained, such a functor has been constructed only in the case n = 1; the difficulties in higher dimensions are due to the lack (so far) of a proper Teichmuller theory for the abelian varieties of dimension n ≥ 2. Acknowledgment. I am grateful to the referee for thoughtful comments. 12 References [1] E. Artin, Uber eine neue Art von L-Reihen, Abhandlungen aus dem Mathematischen Seminar der Hamburgerischen Universitat, Bd. 3 (1924), 89-108. [2] L. Bernstein, The Jacobi-Perron Algorithm, its Theory and Applica- tions, Lect. Notes in Math. 207, Springer 1971. [3] Z. I. Borevich and I. R. Shafarevich, Number Theory, Acad. Press, 1966. [4] G. A. Elliott, On the K-theory of the C ∗-algebra generated by a pro- jective representation of Zn, Proceedings of Symposia in Pure Math. 38 (1982), Part I, 177-180. [5] S. Gelbart, An elementary introduction to the Langlands program, Bull. Amer. Math. Soc. 10 (1984), 177-219. [6] H. Hasse, Vorlesungen uber Zahlentheorie, Springer, 1950. [7] A. W. Knapp, Introduction to the Langlands program, Proceedings of Symposia in Pure Mathematics, Vol. 61 (1997), 245-302. [8] Yu. I. Manin, Real multiplication and noncommutative geometry, in "Legacy of Niels Hendrik Abel", 685-727, Springer, 2004. [9] D. Mumford, Tata Lectures on Theta I, Birkhauser, 1983. [10] I. Nikolaev, Remark on the rank of elliptic curves, Osaka J. Math. 46 (2009), 515-527. [11] D. Poguntke, Simple quotients of group C ∗-algebras for two step nilpo- tent groups and connected Lie groups, Ann. Scient. ´Ec. Norm. Sup. 16 (1983), 151-172. [12] M. A. Rieffel, Projective modules over higher-dimensional non- commutative tori, Canadian J. Math. 40 (1988), 257-338. [13] M. A. Rieffel and A. Schwarz, Morita equivalence of multidimensional noncommutative tori, Internat. J. Math. 10 (1999), 289-299. 13 [14] J. H. Silverman, Advanced Topics in the Arithmetic of Elliptic Curves, GTM 151, Springer 1994. The Fields Institute for Mathematical Sciences, Toronto, ON, Canada, E-mail: [email protected] Current address: 616-315 Holmwood Ave., Ottawa, ON, Canada, K1S 2R2 14
1503.03046	2	1503	2016-02-05T04:38:39	Regular Representations of Lattice Ordered Semigroups	[ "math.OA" ]	We establish a necessary and sufficient condition for a representation of a lattice ordered semigroup to be regular, in the sense that certain extensions are completely positive definite. This result generalizes a theorem due to Brehmer where the lattice ordered group was taken to be $\mathbb{Z}_+^\Omega$. As an immediate consequence, we prove that contractive Nica-covariant representations on lattice ordered semigroups are regular, and therefore, its minimal isometric dilation is also Nica-covariant. We also introduce an analog of commuting row contractions on lattice ordered group and show that such a representation is regular.	math.OA	math	REGULAR REPRESENTATIONS OF LATTICE ORDERED SEMIGROUPS BOYU LI Abstract. We establish a necessary and sufficient condition for a rep- resentation of a lattice ordered semigroup to be regular, in the sense that certain extensions are completely positive definite. This result gen- eralizes a theorem due to Brehmer where the lattice ordered group was taken to be ZΩ +. As an immediate consequence, we prove that contractive Nica-covariant representations are regular. We also introduce an analog of commuting row contractions on lattice ordered group and show that such a representation is regular. 1. Introduction A contractive map of a group has a unitary dilation if and only if it is completely positive definite, in the sense that certain operator matrices are positive. Consequently, for a semigroup P contained in a group G, a contractive representation of P has a unitary dilation if and only if it can be extended to a completely positive definite map on G. Introduced in [6], such representations on a semigroup are called completely positive definite. In particular, when the group is lattice-ordered, a representation is called regular if a certain natural extension to the group is completely positive definite. Nica [14] introduced the study of isometric representations of quasi-lattice ordered semigroups. This generalized the notion of doubly commuting rep- resentations of semigroups with nice generators. Laca and Raeburn [10] developed the theory, and showed there is a universal C ∗-algebra for iso- metric Nica covariant representations. This field has also been explored in [16]. Davidson, Fuller, and Kakariadis [8, 6] defined and studied contractive Nica-covariant representation on lattice ordered semigroups. The regularity of such representations was seen as a critical property in describing the C ∗-envelope of semicrossed products. They posed a question [6, Question 2.5.11] of whether regularity is automatic for Nica-covariant representations. Fuller [8] established this for certain abelian semigroups. Date: June 29, 2021. 2010 Mathematics Subject Classification. 43A35 ,47A20 ,47D03 . Key words and phrases. Nica covariant, regular dilation, positive definite, lattice ordered group. 1 2 BOYU LI This paper answers this question affirmatively by establishing a necessary and sufficient condition for a representation of a lattice ordered semigroup to be regular. This condition generalizes a result of Brehmer [3], where he gave a necessary and sufficient condition for a representation of ZΩ + to be regular. As an immediate consequence of Brehmer's condition, it is known that doubly commuting representations and commuting column contractions are both regular [12, Proposition I.9.2]. This paper generalizes both results in the lattice ordered group settings. We first show that a Nica-covariant representation, which is an analog of a doubly commuting representation, is regular. We then introduce an analog of commuting column contractions, which is shown to be regular as well. Let G be a group. A unital semigroup P ⊆ G is called a cone. A cone P 2. Preliminaries cone P defines a partial order on G via x ≤ y when x−1y ∈ P . (G, P ) is spanning if P P −1 = G, and is positive when P T P −1 = {e}. A positive is called totally ordered if G = P S P −1, in which case the partial order on G is a total order. If any finite subset of G with a upper bound in P also has a least upper bound in P , the pair (G, P ) is called a quasi-lattice ordered group. We call this partial order compatible with the group if for any x ≤ y and g ∈ G, we always have gx ≤ gy and xg ≤ yg. Equivalently, the corresponding positive cone satisfies a normality condition that gP g−1 ⊆ P for any g ∈ G, and thus x ≤ y whenever yx−1 ∈ P as well. When P is a positive spanning cone of G whose partial order is compatible with the group, if every two elements x, y ∈ G have a least upper bound (denoted by x ∨ y) and a greatest lower bound (denoted by x ∧ y), the pair (G, P ) is called a lattice ordered group. It is immediate that a lattice ordered group is also a quasi-lattice ordered group. Example 2.1. (Examples of Lattice Ordered Groups) (1) (Z, Z≥0) is a lattice ordered group. In fact, this partial order is also a total order. More generally, any totally ordered group (G, P ) is also a lattice ordered group. (2) If (Gi, Pi)i∈I is a family of lattice ordered group, their direct product (Q Gi,Q Pi) is also a lattice ordered group. (3) Let G = CR[0, 1], the set of all continuous functions on [0, 1]. Let P be the set of all non-negative functions in G. Then (G, P ) is a lattice ordered group. (4) Let T be a totally ordered set. A permutation α on T is called order preserving if for any p, q ∈ T , p ≤ q, we also have α(p) ≤ α(q). Let G be the set of all order preserving permutations, which is clearly a group under composition. Let P = {α ∈ G : α(t) ≥ t, for all t ∈ T }. Then (G, P ) is a non-abelian lattice ordered group [1]. (5) Let Fn be the free group of n generators, and F+ generated by the n-generators. Then (Fn, F+ n be the semigroup n ) defines a quasi-lattice REGULAR REPRESENTATIONS OF LATTICE ORDERED SEMIGROUPS 3 ordered group [14, Examples 2.3]. However, the induced partial order is not compatible with the group and the pair is not a lattice ordered group. For any element g ∈ G of a lattice ordered group (G, P ), g can be written uniquely as g = g+g−1 − where g+, g− ∈ P , and g+ ∧g− = e. In fact, g+ = g∨e and g− = g−1 ∨ e. Here are some important properties of a lattice ordered group: Lemma 2.2. Let (G, P ) be a lattice order group, and a, b, c ∈ G. (1) a(b∨ c) = (ab)∨ (ac) and (b∨ c)a = (ba)∨ (ca). A similar distributive law holds for ∧. (2) (a ∧ b)−1 = a−1 ∨ b−1 and similarly (a ∨ b)−1 = a−1 ∧ b−1. (3) a ≥ b if and only if a−1 ≤ b−1. (4) a(a ∧ b)−1b = a ∨ b. In particular, when a ∧ b = e, ab = ba = a ∨ b. (5) If a, b, c ∈ P , then a ∧ (bc) ≤ (a ∧ b)(a ∧ c). statement (4) of Lemma 2.2 g+, g− commute and thus g = g+g−1 One may refer to [4] for a detailed discussion of this subject. Notice by − g+. For a group G, a unital map S : G → B(H) is called completely positive − = g−1 definite if for any g1, g2, · · · , gn ∈ G (cid:2)S(g−1 i gj)(cid:3)1≤i,j≤n ≥ 0. Here, i denotes the row index and j the column index, and we shall follow this convention throughout this paper. A well known result ([13], see also [12, Proposition I.7.1]) stated that a completely positive definite map of G has a unitary dilation. The converse is elementary. Theorem 2.3. If S : G → B(H) is a unital completely positive definite map. Then there exists a unitary representation U : G → B(K) where H is a subspace of K, and that PHU (g)H = S(g). Moreover, this unitary representation can be chosen to be minimal in the sense of K =Wg∈G U (g)H. When (G, P ) is a lattice ordered group, we may simultaneously increase or decrease gi so that it would suffices to take gi ∈ P : Lemma 2.4. Let S : G → B(H) be a map, then the following are equivalent: (1) (cid:2)S(g−1 (2) (cid:2)S(gig−1 (3) (cid:2)S(p−1 (4) (cid:2)S(pip−1 i gj)(cid:3)1≤i,j≤n ≥ 0 for any g1, g2, · · · , gn ∈ G; j )(cid:3)1≤i,j≤n ≥ 0 for any g1, g2, · · · , gn ∈ G; i pj)(cid:3)1≤i,j≤n ≥ 0 for any p1, p2, · · · , pn ∈ P ; j )(cid:3)1≤i,j≤n ≥ 0 for any p1, p2, · · · , pn ∈ P . Proof. Since G is a group, by considering gi and g−1 , it is clear that (1) and (2) are equivalent. Statement (1) clearly implies statement (3), and conversely when statement (3) holds true, for any g1, · · · , gn ∈ G, take g = ∨n i=1 (gi)−. Denote pi = g · gi and notice that from our choice of g, g ≥ (gi)−. Hence, i pi = g · (gi)−1 − (gi)+ ∈ P. 4 BOYU LI But notice that for each i, j, p−1 i pj = g−1 i g−1ggj = g−1 i gj. Therefore, Similarly, statements (2) and (4) are equivalent. (cid:2)S(g−1 i gj)(cid:3)1≤i,j≤n =(cid:2)S(p−1 i pj)(cid:3)1≤i,j≤n ≥ 0. (cid:3) For the convenience of computation, when (G, P ) is a lattice ordered group, S : G → B(H) is called completely positive definite when (cid:2)S(pip−1 j )(cid:3)1≤i,j≤n ≥ 0. For a spanning cone P ⊂ G, a contractive representation T : P → B(H) is called completely positive definite when it can be extended to some com- pletely positive definite map on G. There is a well-known result due to Sz.Nagy that every contraction has a unitary dilation, and therefore, every contractive representation of Z+ is completely positive definite. Ando [2] further showed that every contractive representation of Z2 + is completely positive definite. However, Parrott [15] provided an counterexample where a contractive representations on Z3 + is not completely positive definite. For a completely positive definite representation T on a lattice ordered semigroup, one might wonder what its extension looks like. In a lattice ordered group (G, P ), any element g ∈ G can be uniquely written as g = g+g−1 − where g± ∈ P and g+ ∧ g− = e. Suppose U : G → B(K) is a unitary dilation of T , we can make the following observation. T (g) = PHU (g)(cid:12)(cid:12)H = PHU (g−)∗U (g+)(cid:12)(cid:12)H. This motivates the question of whether the extension T (g) = T (g−)∗T (g+) is completely positive definite. We call a contractive representation T right reg- ular whenever T defined in such way is completely positive definite. There is a dual definition that call T left regular (or ∗-regular ) if T (g) = T (g+)T (g−)∗ is completely positive definite. When (G, P ) is a lattice ordered group, (G, P −1) is also a lattice ordered group. A representation T : P → B(H) give raise to a dual representa- tion T ∗ : P −1 → B(H) where T ∗(p−1) = T (p)∗. Consider g = g+g−1 − = g−1 , we have + (cid:1)−1 − (cid:0)g−1 T (g) = T (g−)∗T (g+) = T ∗(g−1 − )T ∗(g−1 + )∗ = T ∗(g). Hence, T agrees with T ∗ on G. Therefore, we obtain the following Proposi- tion. Proposition 2.5. Let (G, P ) be a lattice ordered group, and T : P → B(H) be a representation and T ∗ defined as above. Then the following are equivalent (1) T is right regular. (2) T ∗ is left regular. REGULAR REPRESENTATIONS OF LATTICE ORDERED SEMIGROUPS 5 (3) For any p1, · · · , pn ∈ P , [ T (pip−1 j )] ≥ 0 (equivalently, [T ∗(pip−1 j )] ≥ 0). Due to this equivalence, we shall focus on the right regularity and call a representation regular when it is right regular. Regular dilations were first studied by Brehmer [3], and they were also studied in [17, 9]. A necessary and sufficient condition for regularity for the abelian group ZΩ was proven by Brehmer [12, Theorem I.9.1]. Theorem 2.6 (Brehmer). Let Ω be a set, and denote ZΩ to be the set of (tω)ω∈Ω where tω ∈ Z and tω = 0 except for finitely many ω. Also, for a finite set V ⊂ Ω, denote eV ∈ ZΩ to be 1 at those ω ∈ V and 0 elsewhere. If {Tω}ω∈Ω is a family of commuting contractions, we may define a contractive representation T : ZΩ + → B(H) by Then T is right regular if and only if for any finite U ⊆ Ω, the operator T (tω)ω∈Ω = Yω∈Ω T tω ω . (−1)V T (eV )∗T (eV ) ≥ 0. XV ⊆U It turns out that not all completely positive definite representations are regular. Example 2.7. It follows from Brehmer's theorem that a representation T on Z2 + is regular if and only if T1 = T (e1), T2 = T (e2) are contractions that satisfy I − T ∗ 1 T1 − T ∗ 2 T2 + (T1T2)∗T1T2 ≥ 0. Take T1 = T2 =(cid:20) 0 1 0 0 (cid:21) and notice, I − T ∗ 1 T1 − T ∗ 2 T2 + (T1T2)∗T1T2 =(cid:20) 1 0 −1 (cid:21) . 0 Brehmer's result implies that T is not regular. However, from Ando's + has a unitary dilation theorem [2], any contractive representation on Z2 and thus is completely definite definite. Isometric Nica-covariant representations on quasi-lattice ordered groups were first introduced by Nica [14]: an isometric representation W : G → B(H) is Nica-covariant if for any x, y with an upper bound, WxW ∗ y = Wx∨yW ∗ x∨y. When the order is a lattice order, it is equivalent to the property that Ws, W ∗ t commute whenever s ∧ t = e. Therefore, the notion of Nica- covariant is extended to abelian lattice ordered groups in [6], and we shall further extend such definition to non-abelian lattice ordered groups and call a representation T : P → B(H) Nica-covariant if T (s)T (t)∗ = T (t)∗T (s) whenever s ∧ t = e. For a Nica-covariant representation T , since T (g+) commutes with T (g−)∗ for any g ∈ G, there is no difference between left x WyW ∗ 6 BOYU LI and right regularity. It observed in [6] that Nica-covariant representations are regular in many cases. Example 2.8. (Examples of Nica covariant representations) (1) On (Z, Z+), a contractive representation T on Z+ only depends on T1 = T (1) since T (n) = T n 1 . This representation is always Nica- covariant since for any s, t ≥ 0, s ∧ t = 0 if and only if one of s, t is 0. A well known result due to Sz.Nagy shows that its extension to Z by T (−n) = T ∗n is completely positive definite and thus T is regular. (2) Similarly, any contractive representation of a totally ordered group (G, P ) is Nica-covariant. A theorem of Mlak [11] shows that such representations are regular. (3) (Zn, Zn +), the finite Cartesian product of (Z, Z+) is a lattice ordered group. A representation T on Zn + depends on n contractions T1 = T (1, 0, · · · , 0), T2 = T (0, 1, 0, · · · , 0),· · · , Tn = T (0, · · · , 0, 1). Notice T is Nica covariant if and only if Ti, Tj ∗-commute whenever i 6= j. Hence Nica covariant representations are equivalent to doubly commuting. It is known [12, Section I.9] that doubly commuting contractive representations are regular. (4) For a lattice ordered group made from a direct product of totally or- dered groups, Fuller [8] showed that their contractive Nica-covariant representations are regular. A question posed in [6, Question 2.5.11] asks whether contractive Nica- covariant representations on abelian lattice ordered groups are regular in general. For example, for G = CR[0, 1] and P equal to the set of non- negative continuous functions, there are no known results on whether con- tractive Nica-covariant representations are regular on such semigroup. Little is known for the non-abelian lattice ordered groups as well. In this paper, we establish that all Nica-covariant representations of lattice ordered semi- groups are regular. Let (G, P ) be a lattice-ordered group, not necessarily abelian. Recall that the regularity conditions require a matrix involving entries in the form of T (pq−1) to be positive, where p, q ∈ P . We start by investigating this quantity of pq−1. Lemma 2.9. Let p, q ∈ P . Then, (pq−1)+ = p(p ∧ q)−1 and, (pq−1)− = q(p ∧ q)−1. Proof. By property (1) and (2) in Lemma 2.2, (pq−1)+ = (pq−1 ∨ e) = p(q−1 ∨ p−1) = p(p ∧ q)−1. REGULAR REPRESENTATIONS OF LATTICE ORDERED SEMIGROUPS Similarly, (pq−1)− = q(p ∧ q)−1. 7 (cid:3) Lemma 2.10. Let p, q, g ∈ P such that g ∧ q = e. Then (pg) ∧ q = p ∧ q. Proof. By the property (5) of Lemma 2.2, we have that (pg) ∧ q ≤ (p ∧ q)(g ∧ q) = p ∧ q. On the other hand, p ∧ q is clearly a lower bound for both p ≤ pg and q, and hence p ∧ q ≤ (pg) ∧ q. This proves the equality. (cid:3) Lemma 2.11. Let p, q ∈ P . If g ∈ P is another element where g ∧ q = 0, then (pgq−1)− = (pq−1)− and, (pgq−1)+ = (pq−1)+g. In particular, if 0 ≤ g ≤ p, then (pg−1q−1)− = (pq−1)− and, (pg−1q−1)+ = (pq−1)+g−1. Proof. By Lemma 2.9, we get (pgq−1)+ = pg(q ∧ pg)−1. Apply Lemma 2.10 to get (q ∧ pg)−1 = (q ∧ p)−1. Now g ∧ (p ∧ q) = e and thus g commutes with p ∧ q by property (4) of Lemma 2.2. Therefore, (pgq−1)+ = pg(q ∧ pg)−1 = p(q ∧ p)−1g = (pq−1)+g. The statement (pgq−1)− = (pq−1)−g can be proven in a similar way. Finally, for the case where 0 ≤ g ≤ p, it follows immediately by consider- (cid:3) ing p′ = pg−1 and thus p = p′g. i=1pig−1 i ≤ ∧n i=1pi. Lemma 2.12. If p1, p2, · · · , pn ∈ P and g1, · · · , gn ∈ P be such that gi ≤ pi for all i = 1, 2, · · · , n. Then ∧n In particular, when i=1pig−1 ∧n i=1pi = e, we have ∧n i = e. Proof. It is clear that e ≤ pig−1 e ≤ ∧n i ≤ pi, and thus i ≤ ∧n i=1pi. Therefore, the equality holds when the later is e. i=1pig−1 3. A Necessary and Sufficient Condition For Regularity When T : P → B(H) is a representation of lattice ordered semigroup, we denote T (g) = T (g−)∗T (g+). Recall that T is regular if T is completely positive definite. The main result is the following necessary and sufficient condition for regularity: (cid:3) 8 BOYU LI Theorem 3.1. Let (G, P ) be a lattice ordered group and T : P → B(H) be a contractive representation. Then T is regular if and only if for any p1, · · · , pn ∈ P and g ∈ P where g ∧ pi = e for all i = 1, 2, · · · , n, we have (⋆) Remark 3.2. If we denote hT (g)∗ T (pip−1 j )T (g)i ≤h T (pip−1 j )i . X =h T (pip−1 j )i and D = diag(T (g), T (g), · · · , T (g)), Condition (⋆) is equivalent of saying D∗XD ≤ X. Notice that we made no assumption on X ≥ 0. Indeed, it follows from the main result that Condition (⋆) is equivalent of saying the representation T is regular, which in turn implies X ≥ 0. Therefore, when checking the Condition (⋆), we may assume X ≥ 0. Remark 3.3. By setting p1 = e and picking any g ∈ P , Condition (⋆) implies that T (g)∗T (g) ≤ I, and thus T must be contractive. The following Lemma is taken from [5, Lemma 14.13]. Lemma 3.4. If A, X, D are operators in B(H) where A ≥ 0. Then a matrix of the form (cid:20) A X∗A1/2 A1/2X D (cid:21) is positive if and only if D ≥ X∗X. Condition (⋆) can thus be interpreted in the following equivalent defini- tion. Lemma 3.5. Let p1, · · · , pn ∈ P and g ∈ P with g ∧ pi = e for all 1 ≤ i ≤ n. Denote q1 = p1g, · · · , qn = png and qn+1 = p1, · · · , q2n = pn. Then Condition (⋆) is equivalent to h T (qiq−1 j )i ≥ 0. Proof. Let X = [ T (pip−1 by Lemma 2.11 that j )] ≥ 0 and D = diag(T (g), T (g), · · · , T (g)). Notice and thus T (pigp−1 j ) = T (pip−1 (pigp−1 (pigp−1 j )+ = (pip−1 j )− = (pip−1 j )T (g). Therefore, j )+g j )−, h T (qiq−1 j )i =(cid:20) X XD D∗X X (cid:21) . Lemma 3.4 implies that this matrix is positive if and only if D∗XD ≤ X, which is Condition ⋆. (cid:3) We shall first show that h T (pip−1 j )i ≥ 0 given pi ∧ pj = e and Condition (⋆). This will serve as a base case in the proof of the main result. Lemma 3.6. Let (G, P ) be a lattice ordered group, and T be a represen- tation on P that satisfies Condition (⋆). If pi ∧ pj = e for all i 6= j, then [ T (pip−1 j )] ≥ 0. REGULAR REPRESENTATIONS OF LATTICE ORDERED SEMIGROUPS 9 Proof. Let q1 = e, q2 = p1 and for each 1 < m ≤ n, recursively define q2m−1+k = pmqk where 1 ≤ k ≤ 2m−1. Since T is contractive, [ T (qiq−1 j )]1≤i,j≤2 =(cid:20) I T (q2q−1 1 ) T (q1q−1 2 ) I (cid:21) ≥ 0. By Lemma 3.5, for each m, [ T (qiq−1 for each 1 ≤ m ≤ n. Therefore, [ T (pip−1 thus must be positive. j )]1≤i,j≤2m ≥ 0. Notice that q2m−1 = pm j )] ≥ 0, and (cid:3) j )] is a corner of [ T (qiq−1 For arbitrary choices of p1, · · · , pn ∈ P , the goal is to reduce it to the case where pi ∧ pj = e. The following lemma does the reduction. Lemma 3.7. Let (G, P ) be a lattice ordered group. Assuming T is a repre- sentation that satisfies Condition (⋆). Assume there exists 2 ≤ k < n where for each J ⊂ {1, 2, · · · , n} with j=1pj and q1 = p1g−1, · · · , qk = pkg−1, J > k, ∧j∈J pj = e. Then let g = ∧k and qk+1 = pk+1, · · · , qn = pn. Then [ T (pip−1 j )] ≥ 0 if [ T (qiq−1 j )] ≥ 0. Proof. Let us denote X = [ T (qjq−1 corner to be Y . Notice first of all, when i, j ∈ {1, 2, · · · , k}, )] ≥ 0 and its lower right (n − k)× (n − k) i qiq−1 j = pig−1gp−1 j = pip−1 j . So the upper left k×k corner of [ T (qiq−1 corner of X are both the same as those in [ T (pip−1 j )]. j )] and the lower right (n−k)×(n−k) Now consider i ∈ {1, 2, · · · , k} and j ∈ {k + 1, · · · , n}. It follows from the assumption that g ∧ pj = (cid:0)∧k apply Lemma 2.11 to get s=1ps(cid:1) ∧ pj = e and g ≤ pi. Therefore, we can (pig−1p−1 (pig−1p−1 j )− = (pip−1 j )+ = (pip−1 j )− j )+g−1. Now g ∈ P , so that T ((qiq−1 T ((pip−1 j )−). Hence, j )+)T (g) = T ((pip−1 j )+) and T ((qiq−1 j )−) = T (qiq−1 j )T (g) = T (pip−1 j ). Similarly, for i ∈ {k + 1, · · · , n}, j ∈ {1, 2, · · · , k}, we have T (pip−1 j ) = T (g)∗ T (qjq−1 ). Now define D = diag(I, · · · , I, T (g), · · · , T (g)) be the block diagonal matrix with k copies of I followed by n−k copies of T (g). Consider DXD∗: it follows immediately from the assumption that D∗XD ≥ 0. We i BOYU LI · · · · · · · · · T (pip−1 j ) · · · · · · · · · T (g)∗ T (qiq−1 j ) · · · · · · · · · · · · ... T (qiq−1 j )T (g) ... [T (g)∗ T (pip−1 j )T (g)]   ≥ 0.   10 have, D∗[ T (qiq−1 j )]D = It follows from our previous computation that each entry in the lower left (n − k) × k corner and upper right k × (n − k) corner are the same as those in [ T (pip−1 j )] on the lower right (n − k) × (n − k) corner. It follows from Condition (⋆) that j )]. Hence, DXD∗ only differs from [ T (pip−1 [T (g)∗ T (pip−1 j )T (g)] ≤ [ T (pip−1 j )]. Hence, the matrix remains positive when the lower right corner [T (g)∗ T (pip−1 in D∗XD is replaced by [ T (pip−1 j )]. The resulting matrix is exactly [ T (pip−1 j )], which must be positive. (cid:3) j )T (g)] Now the main result (Theorem 3.1) can be deduced inductively: Proof. First assume that T : P → B(H) is a representation that satisfies Condition (⋆), which has to be contractive. The goal is to show for any n elements p1, p2, · · · , pn ∈ P , the operator matrix [ T (pip−1 j )] ≥ 0 and thus T is regular. We proceed by induction on n. For n = 1, T (p1p−1 For n = 2, we have, 1 ) = I ≥ 0. [ T (pip−1 j )] =(cid:20) I T (p2p−1 1 ) T (p1p−1 2 ) I (cid:21) . Here, T (p2p−1 tive. Therefore, this 2 × 2 operator matrix is positive. 1 ) = T (p1p−1 2 )∗, and they are contractions since T is contrac- Now assume that there is an N such that for any n < N , we have [ T (pip−1 j )] ≥ 0 for any p1, p2, · · · , pn ∈ P . Consider the case when n = N : For arbitrary choices p1, · · · , pN ∈ P , let g = ∧N pig−1. By doing so, pig−1(cid:0)pjg−1(cid:1)−1 = pip−1 matrix [ T (pip−1 loss of generality, we may assume ∧N i=1pig−1 = (∧N i=1pi = e. j )]. Moreover, ∧n j i=1pi, and replace pi by , and thus they give the same i=1pi)g−1 = e. Hence, without Let m be the smallest integer such that for all J ⊆ {1, 2, · · · , N } and J > m, we have ∧j∈J pj = e. It is clear that m ≤ N − 1. Now do induction on m: For the base case when m = 1, we have pi ∧ pj = e for all i 6= j. Lemma 3.6 tells that Condition (⋆) implies [ T (pip−1 j )] ≥ 0. Now assume [ T (pip−1 j )] ≥ 0 whenever m ≤ M − 1 < N − 1 and consider the case when m = M : For a subset J ⊆ {1, 2, · · · , n} with J = M , REGULAR REPRESENTATIONS OF LATTICE ORDERED SEMIGROUPS 11 let g = ∧j∈J pj and set qj = pjg−1 for all j ∈ J, and qj = pj otherwise. Lemma 3.7 concluded that [ T (pip−1 j )] ≥ 0 and the sub-matrix [ T (pip−1 j )] ≥ 0 whenever [ T (qiq−1 j )]i,j /∈J ≥ 0. j )]i,j /∈J ≥ 0. Therefore, [ T (pip−1 Since {1, 2, · · · , N }\J = N − M < N , the induction hypothesis on n implies that [ T (pip−1 j )] ≥ 0 whenever [ T (qiq−1 j )] ≥ 0, and by dropping from pi to qi, we may, without loss of generality, assume that ∧j∈J pj = e. Repeat this process for all subsets J ⊂ {1, 2, · · · , n} where J = M , and with Lemma 2.12, we eventually reach a state when ∧j∈J pj = e for all J ⊆ {1, 2, · · · , N }, J = M . But in such case, for all J ≥ M , we have ∧j∈J pj = e. Therefore, we are in a situation where m ≤ M −1. The result follows from the induction hypothesis on m. Conversely, suppose that T is regular. Fix g ∈ P and p1, p2, · · · , pk ∈ P where g ∧ pi = e for all i = 1, 2, · · · , k. Denote q1 = p1g, q2 = p2g, · · · , qk = pkg, and qk+1 = p1, qk+2 = p2, · · · , q2k = pk. It follows from regularity that [ T (qiq−1 (cid:3) j )] ≥ 0, which is equivalent to Condition (⋆) by Lemma 3.5. As an immediate consequence of Theorem 3.1, we can show that isometric representations on any lattice ordered group must be regular. Corollary 3.8. Let T : P → B(H) be an isometric representation of a lattice ordered semigroup. Then T is regular. Proof. Take p1, · · · , pn ∈ P and g ∈ P with g ∧ pi = e. It is clear that . Hence, j (cid:17)± g ∧(cid:16)pip−1 j (cid:17)± = e and therefore g commutes with each (cid:16)pip−1 j )T (g) = T (g)∗T ((pip−1 j )−)∗T ((pip−1 T (g)∗ T (pip−1 = T ((pip−1 = T ((pip−1 j )−)∗T (g)∗T (g)T ((pip−1 j )−)∗T ((pip−1 j )+)T (g) j )+) j )+) = T (pip−1 j ). Therefore, [T (g)∗ T (pip−1 j )T (g)] = [ T (pip−1 j )] and Condition (⋆) is satisfied. (cid:3) For a contractive representation T , it would suffice to dilate it to an isometric representation. This provides an analog of [6, Proposition 2.5.4] on non-abelian lattice ordered groups. Corollary 3.9. Let T : P → B(H) be a contractive representation. Then T is completely positive definite if and only if there exists an isometric repre- In particular, T is regular if and only if there exists such isometric dilation sentation V : P → B(K) such that PHV (p)(cid:12)(cid:12)H = T (p) for all p ∈ P . Such V can be taken to be minimal in the sense that K =Wp∈P V (p)H. V and in addition, PHV (p)∗V (q)(cid:12)(cid:12)H = T (p)∗T (q) for all p, q ∈ P with p∧q = S to G has minimal unitary dilation U : G → B(L), let K =Wp∈P U (p)H. It Proof. When T : P → B(H) is completely positive definite and its extension e. 12 BOYU LI is clear that K is invariant for any U (p), p ∈ P . Define a map V : P → B(K) K. V is an isometric dilation of T that satisfies PHV (p)H = T (p), and via V (p) = PKU (p)(cid:12)(cid:12)K, which must be isometric due to the invariance of K = Wp∈P V (p)H. In other words, V is a minimal isometric dilation of T . In particular, when T is regular, for any p, q ∈ P with p ∧ q = e T (p)∗T (q) = PHU (p)∗U (q)(cid:12)(cid:12)H = PHPKU (p)∗U (q)(cid:12)(cid:12)K(cid:12)(cid:12)H = PHV (p)∗V (q)(cid:12)(cid:12)H. Conversely, when V : P → B(K) is a minimal isometric dilation of T , Corollary 3.8 implies that V is regular and thus completely positive definite. There exists a unitary dilation U : G → B(L) where PKU (p)(cid:12)(cid:12)K = V (p). Therefore, PHU (p)(cid:12)(cid:12)H = PHPKU (p)(cid:12)(cid:12)H = PHV (p)(cid:12)(cid:12)H = T (p). Hence, U is also a unitary dilation of T and thus T is completely positive p ∧ q = e, by the regularity of V , definite. Moreover, when PHV (p)∗V (q)(cid:12)(cid:12)H = T (p)∗T (q) for all p, q ∈ P with PHU (p)∗U (q)(cid:12)(cid:12)H = PHPKU (p)∗U (q)(cid:12)(cid:12)K(cid:12)(cid:12)H = T (p)∗T (q). Therefore, T (g) = T (g−)∗T (g+) is completely positive definite and T is regular. (cid:3) 4. Nica-covariant Representations In this section, we answer the question of whether contractive Nica- covariant representations are regular. It suffices to show contractive Nica- covariant representations on lattice ordered groups satisfy Condition (⋆). Theorem 4.1. A contractive Nica-covariant representation on a lattice or- dered group is regular. Proof. Let p1, · · · , pk ∈ P and g ∈ P with g ∧ pi = e for all i = 1, 2, · · · , k. X = [ T (pip−1 j )] and D = diag(T (g), T (g), · · · , T (g)). By Remark 3.2, we may assume X ≥ 0. j ) = T (p− Since for each pi, pj ∈ P , T (pip−1 i,j ≤ pi, pj. Hence, g ∧ p± i,j = e and thus g commutes with p± i,j. Therefore T (g) commutes with T (p+ i,j) because T is a representation and it also commutes with T (p− i,j)∗ by the Nica-covariant condition. As a result, T (g) commutes with each entry in X, and thus D commutes with X. Similarly, D∗ com- mutes with X as well. i,j) where e ≤ p± i,j)∗T (p+ By continuous functional calculus, since X ≥ 0, we know D, D∗ also commutes with X 1/2. Hence, in such case, D∗XD = D∗X 1/2X 1/2D = X 1/2D∗DX 1/2 ≤ X. (cid:3) REGULAR REPRESENTATIONS OF LATTICE ORDERED SEMIGROUPS 13 It was shown in [6, Proposition 2.5.10] that a contractive Nica-covariant representation on abelian lattice ordered groups can be dilated to an iso- metric Nica-covariant representation. Here, we shall extend this result to non-abelian case. Corollary 4.2. Any minimal isometric dilation V : P → B(K) of a con- tractive Nica-covariant representation T : P → B(H) is also Nica-covariant. Proof. Let T : P → B(H) be a contractive Nica-covariant representation. Theorem 4.1 implies that T is regular, and thus by Theorem 2.3, it has a minimal unitary dilation U : G → B(L), which gives rise to a minimal isometric dilation V : P → B(K). Here K = Wp∈P V (p)H and V (p) = PKU (p)K. Notice that K is invariant for U and therefore, PKU (p)∗U (q)K = V (p)∗V (q) for any p, q ∈ P . In particular, if p ∧ q = e, p, q ∈ P , we have from the regularity that T (p)∗T (q) = PHU (p)∗U (q)H = PH(PKU (p)∗U (q)K)H = PHV (p)∗V (q)H. Now let s, t ∈ P be such that s ∧ t = e. First, we shall prove V (s)∗V (t)H = V (t)V (s)∗H: Since {V (p)h : p ∈ P, h ∈ H} is dense in K, it suffices to show for any h, k ∈ H and p ∈ P , hV (s)∗V (t)h, V (p)ki = hV (t)V (s)∗h, V (p)ki . Start from the left, hV (s)∗V (t)h, V (p)ki = hV (p)∗V (s)∗V (t)h, ki = hV (sp)∗V (t)h, ki = (cid:10)V ((sp ∧ t)−1sp)∗V (sp ∧ t)∗V (sp ∧ t)V ((sp ∧ t)−1t)h, k(cid:11) = (cid:10)V ((sp ∧ t)−1sp)∗V ((sp ∧ t)−1t)h, k(cid:11) = (cid:10)T ((sp ∧ t)−1sp)∗T ((sp ∧ t)−1t)h, k(cid:11) . The last equality follows from (cid:0)(sp ∧ t)−1sp(cid:1) ∧(cid:0)(sp ∧ t)−1t(cid:1) = e and thus, T ((sp ∧ t)−1sp)∗T ((sp ∧ t)−1t) = PHV ((sp ∧ t)−1sp)∗V ((sp ∧ t)−1t)H. Since s ∧ t = e, Lemma 2.10 implies that sp ∧ t = p ∧ t. Notice (p ∧ t) ∧ s ≤ t ∧ s = e, and thus by Property (4) of Lemma 2.2, s commutes with p ∧ t. By the Nica-covariance of T , this also implies T (s)∗ commutes with T ((p∧t)−1t). 14 BOYU LI Put all these back to the equation: (cid:10)T ((sp ∧ t)−1sp)∗T ((sp ∧ t)−1t)h, k(cid:11) = (cid:10)T (s(p ∧ t)−1p)∗T ((p ∧ t)−1t)h, k(cid:11) = (cid:10)T ((p ∧ t)−1p)∗T (s)∗T ((p ∧ t)−1t)h, k(cid:11) = (cid:10)T ((p ∧ t)−1p)∗T ((p ∧ t)−1t) (T (s)∗h) , k(cid:11) = (cid:10)V ((p ∧ t)−1p)∗V ((p ∧ t)−1t) (T (s)∗h) , k(cid:11) = (cid:10)V ((p ∧ t)−1p)∗V ((p ∧ t)−1t) (V (s)∗h) , k(cid:11) = hV (p)∗V (t) (V (s)∗h) , ki = hV (t)V (s)∗h, V (p)ki . Here we used the fact that PHV (p)∗V (q)H = T (p)∗T (q) whenever p ∧ q = e. Also, that H is invariant under V (s)∗, so that T (s)∗h ∈ K is the same as V (s)∗h. Now to show V (s)∗V (t) = V (t)V (s)∗ in general, it suffices to show for every p ∈ P , V (s)∗V (t)V (p)H = V (t)V (s)∗V (p)H. Start with the left hand side and repeatedly use similar argument as above, V (s)∗V (t)V (p)H = V (s)∗VtpH = V ((s ∧ tp)−1s)∗V ((s ∧ tp)−1tp)H = V (t(s ∧ p)−1p)V ((s ∧ p)−1s)∗H = V (t(s ∧ p)−1p)V ((s ∧ p)−1s)∗H = V (t)V ((s ∧ p)−1s)∗V ((s ∧ p)−1p)H = V (t)V (s)∗V (p)H. This finishes the proof. (cid:3) 5. Row and Column Contractions A commuting n-tuple (T1, · · · , Tn) where each Ti ∈ B(H) is called a row i ≤ I. Equivalently, the operator [T1, T2, · · · , Tn] ∈ B(Hn, H) is contractive. It can be naturally associated with a contractive representation T : Zn + → B(H) that sends the i-th generator ei to Ti. There i Ti ≤ It is clear that T is a row contraction if and only if T ∗ is a column contraction if Pn is a dual definition called column contractions, when Ti satisfiesPn i=1 T ∗ i=1 TiT ∗ I. contraction. As an immediate corollary to Brehmer's theorem (Theorem 2.6), a column contraction T is always right regular [12, Proposition I.9.2], and therefore a row contraction T is always left regular. This section generalizes the notion of row contraction to arbitrary lattice ordered groups and establishes a similar result. Definition 5.1. Let T : P → B(H) be a contractive representation of a lat- tice ordered group (G, P ). T is called row contractive if for any p1, · · · , pn ∈ P where pi 6= e and pi ∧ pj = e for all i 6= j, T (pi)T (pi)∗ ≤ I. n Xi=1 REGULAR REPRESENTATIONS OF LATTICE ORDERED SEMIGROUPS 15 Dually, T is called column contractive if for such pi, n T (pi)∗T (pi) ≤ I. Xi=1 Remark 5.2. Definition 5.1 indeed generalized the notion of commuting row contractions: when the group is (ZΩ, ZΩ +) where Ω is countable, a represen- tation T : ZΩ + → B(H) is uniquely determined by its value on the generators ω ≤ I. For any p1, · · · , pk ∈ ZΩ 6= 0, each pi can be seen as a function from Ω to Z+ with finite support. Let Si ⊆ Ω be the support of pi, which is non-empty since pi 6= 0. We have Tω = T (eω). T is called commuting row contraction when Pω∈Ω TωT ∗ + where pi ∧ pj = 0 for all i 6= j and pi SiT Sj = ∅ since pi ∧ pj = 0. Therefore, pick any ωi ∈ Si and by T con- tractive, T (ωi)T (ωi)∗ ≥ T (pi)T (pi)∗. Since Si are pairwise-disjoint, ωi are distinct. Therefore, we get that n n T (pi)T (pi)∗ ≤ T (ωi)T (ωi)∗ ≤ I. Xi=1 Xi=1 and thus T satisfies the Definition 5.1. Hence, two definitions coincides on (ZΩ, ZΩ +). Our goal is to prove the following result: Theorem 5.3. A column contractive representation is right regular. There- fore, a row contractive representation is left regular. We shall proceed with a method similar to the proof of Theorem 3.1. Lemma 5.4. Let T be a column contractive representation. Let p1, · · · , pn ∈ P and g1, · · · , gk ∈ P where pi ∧ pi′ = pi ∧ gj = gj ∧ gj ′ = e for all 1 ≤ i 6= i′ ≤ n and 1 ≤ j 6= j′ ≤ k. Moreover, assume that gi 6= e. Denote X = [ T (pip−1 j )] and Di = diag(T (gi), · · · , T (gi)). Then, k D∗ i XDi ≤ X. Xi=1 Proof. The statement is clearly true for all k when n = 1. Now assuming it is true for all k whenever n < N , and consider the case when n = N : i=1 D∗ It is clear that when all of the pi are equal to e, then X −Pk is a n × n matrix whose entries are all equal to I −Pk i XDi i=1 T (gi)∗T (gi) ≥ 0, and thus the statement is true. Otherwise, we may assume without loss of generality that p1 6= e. Let q1 = e and q2 = p2, · · · , qn = pn. Denote X0 = [ T (qiq−1 j )] and E = diag(I, T (p1), · · · , T (p1)) be a n×n block diagonal matrix. Denote Y = [ T (pip−1 j )]2≤i,j≤n and set Ei = diag(T (gi), · · · , T (gi)) be a (n−1)×(n−1) block diagonal matrix. Finally, set Ek+1 = diag(T (p1), · · · , T (p1)) be a (n − 1) × (n − 1) block diagonal matrix. Also notice that E commutes with Di and therefore, ifPk we have i=1 D∗ i X0Di ≤ X0, D∗ i XDi k Xi=1 D∗ = E∗ k Xi=1 ≤ E∗X0E +(cid:20)0 i X0Di! E +(cid:20)0 0 i (Y − E∗ k+1Y Ek+1)Ei(cid:21) i=1 E∗ 0 Pk 0 k+1Y Ek+1(cid:21) = X. 0 Y − E∗ 16 BOYU LI From the proof of Theorem 3.1, X = E∗X0E +(cid:20)0 0 Y − E∗ 0 k+1Y Ek+1(cid:21) . Now Y is a matrix of smaller size and thus by induction hypothesis, Hence, E∗ i Y Ei ≤ Y. k+1 Xi=1 k Y − E∗ k+1Y Ek+1 ≥ ≥ E∗ i Y Ei E∗ i (Y − E∗ k+1Y Ek+1)Ei. k Xi=1 Xi=1 i=1 D∗ Hence, Pk i X0Di ≤ X0. This reduction from X to X0 changes one pi 6= e to e, and therefore by repeating this process, we eventually reach a state where all pi = e. (cid:3) i XDi ≤ X if Pk i=1 D∗ The main result can be deduced immediately from the following Propo- sition: Proposition 5.5. Let T be a column contractive representation on a lattice ordered semigroup P . Let p1, · · · , pn ∈ P and g1, · · · , gk ∈ P where gi ∧ pj = e and gi ∧ gl = e for all i ≤ l. Assuming gi 6= e and denote X = [ T (pip−1 j )] and Di = diag(T (gi), · · · , T (gi)). Then k In particular, Condition (⋆) is satisfied when k = 1. D∗ i XDi ≤ X. Xi=1 Proof. The statement is clear when n = 1. Assuming it's true for n < N , and consider the case when n = N : Let m be the smallest integer such that for all J ⊆ {1, 2, · · · , N } and J > m, ∧j∈J pj = e. It was observed in the proof of Theorem 3.1 that m ≤ N − 1. Proceed by induction on m: REGULAR REPRESENTATIONS OF LATTICE ORDERED SEMIGROUPS 17 In the base case when m = 1, pi ∧ pj = e for all i 6= j, the statement is shown in Lemma 5.4. Assuming the statement is true for m < M −1 < N −1 and consider the case when m = M . For each J ⊆ {1, 2, · · · , N } with j=1pj = g 6= e, denote qi = pi when i /∈ J and qi = qig−1 J = M and ∧M when i ∈ J. Let X0 = [ T (qiq−1 j )] and E be a block diagonal matrix whose i-th diagonal entry is I when i /∈ J and T (g) otherwise. Denote Y = [ T (qiq−1 j )]i,j /∈J and Ei = diag(T (gi), · · · , T (gi)) with N − M copies of T (gi). Finally, let Ek+1 = diag(T (g), · · · , T (g)) with N − M copies of T (g). From the proof of Theorem 3.1, by assuming without loss of generality that J = {1, 2, · · · , M }, we have X = E∗X0E +(cid:20)0 0 Y − E∗ 0 k+1Y Ek+1(cid:21) . Now Y has a smaller size and thus by induction hypothesis on n, E∗ i Y Ei ≤ Y. k+1 Xi=1 and thus Y − E∗ k+1Y Ek+1 ≥ ≥ E∗ i Y Ei E∗ i (Y − E∗ k+1Y Ek+1)Ei. k k Xi=1 Xi=1 i=1 D∗ i X0Di ≤ X0, Therefore, if Pk Xi=1 k D∗ i XDi D∗ = E∗ k Xi=1 ≤ E∗X0E +(cid:20)0 i X0Di! E +(cid:20)0 0 Y − E∗ i=1 E∗ 0 Pk 0 k+1Y Ek+1(cid:21) = X. 0 i (Y − E∗ k+1Y Ek+1)Ei(cid:21) Hence, the statement is true for pi if it is true for qi, where ∧j∈J qj = e. Repeat the process until all such J = M has ∧j∈J pj = e, which reduces to a case where m < M . This finishes the induction. Notice Condition (⋆) is clearly true when g = e, and when g 6= e, it is shown by the case when m = 1. This finishes the proof. (cid:3) 6. Brehmer's Condition Brehmer [3] established a necessary and sufficient condition for a repre- + to be regular (see Theorem 2.6). This section explores sentation on P = ZΩ 18 BOYU LI how Brehmer's result relates to Condition (⋆) without invoking their equiv- alence to regularity. In particular, we show that Brehmer's condition allows us to decompose certain X = [ T (pi − pj)] as a product R∗R, where R is an upper triangular matrix. Let {Tω}ω∈Ω be a family of commuting contractions, which leads to a + by sending each eω to Tω. For each U ⊆ Ω, contractive representation on ZΩ denote (−1)V T (eV )∗T (eV ). ZU = XV ⊆U For example, Z∅ = I Z{1} = I − T ∗ 1 T1 Z{1,2} = Z{1} − T ∗ 2 Z{1}T2 = I − T ∗ 1 T1 − T ∗ 2 T2 + T ∗ 2 T ∗ 1 T1T2 ... Brehmer's theorem stated that T is regular if and only if ZU ≥ 0 for any finite subset U ⊆ Ω. We shall first transform Brehmer's condition into an equivalent form. Lemma 6.1. ZU ≥ 0 for each finite subset U ⊆ Ω if and only if for any finite set J ⊆ Ω and ω ∈ Ω, ω /∈ J, T ∗ ω ZJ Tω ≤ ZJ . Proof. Take any finite subset J ⊆ Ω and ω ∈ Ω, ω /∈ J. ZJ − T ∗ ω ZJ Tω = XV ⊆J (−1)V T (eV )∗T (eV ) + XV ⊆J (−1)V +1T ∗ ω T (eV )∗T (eV )Tω = XV ⊆{ω} S J,ω /∈V (−1)V T (eV )∗T (eV ) + XV ⊆{ω} S J,ω∈V (−1)V T (eV )∗T (eV ) = Z{ω} S J . Therefore, T ∗ proof. ωZJ Tω ≤ ZJ if and only if Z{ω} S J ≥ 0. This finishes the (cid:3) A major tool is the following version of Douglas Lemma [7]: Lemma 6.2 (Douglas). For A, B ∈ B(H), A∗A ≤ B∗B if and only if there exists a contraction C such that A = CB. As an immediate consequence of Lemma 6.2, T ∗ ω ZJ Tω ≤ ZJ is satisfied J Tω = Wω,J Z 1/2 if and only if there is a contraction Wω,J such that Z 1/2 . Therefore, it would suffices to find such contraction Wω,J for each finite subset J ⊆ Ω and ω ∈ Ω, ω /∈ J. By symmetry, it would suffices to do so for each Jn = {1, 2, · · · , n} and ωn = n + 1. Without loss of generality, we shall assume that Ω = N. J REGULAR REPRESENTATIONS OF LATTICE ORDERED SEMIGROUPS 19 Consider P(Jn) = {U ⊆ Jn}, and denote pU = Pi∈U ei ∈ ZΩ Xn = [ T (pU − pV )] where U is the row index and V is the column index. Lemma 6.3. Assuming ZJ ≥ 0 for all J ⊆ Jn. Then for a fixed F ⊆ Jn, we have, +. Denote T ∗ U ZF \U TU = I. XU ⊆F Proof. We first notice that by definition, ZJ = PU ⊆J (−1)U T ∗ fore, U TU . There- (−1)V T ∗ U S V TU S V . XU ⊆F T ∗ U ZF \U TU = XU ⊆F XV ⊆F \U For a fixed set W ⊆ F , consider the coefficient of T ∗ W TW in the double summation. It appears in the expansion of every T ∗ U ZF \U TU , where U ⊆ W , and its coefficient in the expansion of such term is equal to (−1)W \U . Therefore, the coefficient of T ∗ W TW is equal to (−1)W \U = XU ⊆W W Xi=0(cid:18)W i (cid:19)(−1)i. This evaluates to 0 when W > 0 and 1 when W = 0, in which case, W = ∅ and TW = I. (cid:3) Now can now decompose Xn = R∗ nRn explicitly. Proposition 6.4. Assuming ZJ ≥ 0 for all J ⊆ Jn. Define a block ma- trix Rn, whose rows and columns are indexed by P(Jn), by Rn(U, V ) = Z 1/2 Jn\U TU \V whenever V ⊆ U and 0 otherwise. Then Xn = R∗ Proof. Fix U, V ⊆ Jn, the (U, V )-entry in Xn is T (pU − pV ) = T ∗ Now the (U, V )-entry in R∗ V \U TU \V . nRn nRn is equal to Rn(W, U )∗Rn(W, V ). It follows from the definition that Rn(W, U )∗Rn(W, V ) = 0 unless U, V ⊆ W , XW ⊆Jn and thus US V ⊆ W . Hence, Rn(W, U )∗Rn(W, V ) T ∗ W \U ZJn\W TW \V XW ∈P(Jn) = XU S V ⊆W = XU S V ⊆W V \U  XU S V ⊆W = T ∗ T ∗ V \U T ∗ W \(U S V )ZJn\W TW \(U S V )TW \U W \(U S V )ZJn\W TW \(U S V ) T ∗  TW \U . 20 BOYU LI If we denote F = Jn\(US V ) and W ′ = W \(US V ), since US V ⊆ W , we have Jn\W = F \W ′. Hence the summation becomes XU S V ⊆W T ∗ W \(U S V )ZJn\W TW \(U S V ) = XW ′⊆F T ∗ W ′ZF \W ′TW ′, which by Lemma 6.3 is equal to I. Therefore, the (U, V )-entry in R∗ equal to T ∗ V \U TW \U and Xn = R∗ nRn nRn is (cid:3) Remark 6.5. If we order the subsets of Jn by cardinality and put larger sets first, then since Rn(U, V ) 6= 0 only when V ⊆ U , Rn becomes a lower triangular matrix. In particular, the row of ∅ contains exactly one non-zero entry, which is Z 1/2 Jn at (∅, ∅). Example 6.6. Let us consider the case when n = 2, and J2 has 4 subsets {1, 2}, {2},{1},∅. Under this ordering, Proposition 6.4 gives that Xn =  Rn =  I T ∗ 1 T ∗ 2 1 T ∗ T ∗ 2 T1 I T ∗ 2 T1 T ∗ 2 T2 T ∗ 1 T2 I T ∗ 1 T1T2 T2 T1 I . I T1 0 Z 1/2 0 0 1 0 0 T2 0 Z 1/2 2 0 T1T2 Z 1/2 1 T2 Z 1/2 2 T1 Z 1/2 1,2     satisfies R∗ nRn = Xn. We can now prove Brehmer's condition from Condition (⋆) without in- voking their equivalence to regularity. Proposition 6.7. In the case of T : ZΩ Brehmer's condition. + → B(H), Condition (⋆) implies Proof. Without loss of generality, we may assume Ω = N. We shall proceed by induction on the size of J ⊆ N. For J = 1 (i.e. J = {ω}), Condition (⋆) implies T is contractive. Hence, ZJ = I − T ∗ ω Tω ≥ 0. Assuming ZJ ≥ 0 for all J ≤ n, and consider the case when J = n + 1. By symmetry, it would suffices to show this for J = Jn+1 = {1, 2, · · · , n + 1}. By Proposition 6.4, Xn = R∗ nRn where the (∅, ∅)-entry of Rn is equal to Z 1/2 Jn . Let Dn be a block diagonal matrix with 2n copies of Tn+1 along the diagonal. Condition (⋆) implies that nR∗ nRnDn ≤ Xn = R∗ nXnDn = D∗ nRn. D∗ Therefore, by Lemma 6.2, there exists a contraction Wn such that WnRn = RnDn. By comparing the (∅, ∅)-entry on both sides, there exists Cn such REGULAR REPRESENTATIONS OF LATTICE ORDERED SEMIGROUPS 21 that CnZ 1/2 contractive as well. Hence, by Lemma 6.1 and 6.2, Jn = Z 1/2 Jn Tn+1, where Cn is the (∅, ∅)-entry of Wn, which must be ZJn+1 = ZJn − T ∗ n+1ZJnTn+1 ≥ 0. This finishes the proof. (cid:3) 7. Covariant Representations The semicrossed products of a dynamical system by Nica-covariant rep- resentations was discussed in [8, 6], where its regularity is seen as a key to many results. Our result on the regularity of Nica-covariant representations (Theorem 4.1 and Corollary 4.2) allows us to generalize some of the results to arbitrary lattice ordered abelian groups. Definition 7.1. A C ∗-dynamical system is a triple (A, α, P ) where (1) A is a C ∗-algebra; (2) α : P → End(A) maps each p ∈ P to a ∗-endomorphism on A; (3) P is a spanning cone of some group G. Definition 7.2. A pair (π, T ) is called a covariant pair for a C ∗-dynamical system if (1) π : A → B(H) is a ∗-representation; (2) T : P → B(H) is a contractive representation of P ; (3) π(a)T (s) = T (s)π(αs(a)) for all s ∈ P and a ∈ A. In particular, a covariant pair (π, T ) is called Nica-covariant/isometric, if T is Nica-covariant/isometric. The main goal is to prove that Nica-covariant pairs on C ∗-dynamical systems can be lifted to isometric Nica-covariant pairs. This can be seen from [6, Theorem 4.1.2] and Corollary 4.2. However, we shall present a slightly different approach by taking the advantage of the structure of lattice ordered abelian group. Theorem 7.3. Let (A, α, P ) be a C ∗-dynamical system over a positive cone P of a lattice ordered abelian group G. Let π : A → B(H) and T : P → B(H) form a Nica-covariant pair (π, T ) for this C ∗-dynamical system. If V : P → K is a minimal isometric dilation of T , then there is an isometric Nica-covariant pair (ρ, V ) such that for all a ∈ A, Moreover, H is invariant for ρ(a). PHρ(a)(cid:12)(cid:12)H = π(a). Proof. Fix a minimal dilation V of T and consider any h ∈ H, p ∈ P , and a ∈ A: define ρ(a)V (p)h = V (p)π(αp(a))h We shall first show that this is a well defined map. First of all, since V is a minimal isometric dilation, the set {V (p)h} is dense in K. Suppose 22 BOYU LI V (p)h1 = V (s)h2 for some p, s ∈ P and h1, h2 ∈ H. It suffices to show that for any t ∈ P and h ∈ H, we have hV (p)π(αp(a))h1, V (t)hi = hV (s)π(αs(a))h2, V (t)hi . (1) Since A is a C ∗-dynamical system, it follows from the covariant condition π(a)T (s) = T (s)π(αs(a)) that T (s)∗π(a) = π(αs(a))T (s)∗. Hence, hV (p)π(αp(a))h1, V (t)hi = hV (t)∗V (p)π(αp(a))h1, hi = hV (t − t ∧ p)∗V (p − t ∧ p)π(αp(a))h1, hi = hT (t − t ∧ p)∗T (p − t ∧ p)π(αp(a))h1, hi Here we used that fact that V is regular and thus = hπ(αt(a))T (t − t ∧ p)∗T (p − t ∧ p)h1, hi . = (cid:10)π(αp−(p−t∧p)+(t−t∧p)(a))T (t − t ∧ p)∗T (p − t ∧ p)h1, h(cid:11) PHV (t − t ∧ p)∗V (p − t ∧ p)(cid:12)(cid:12)H = T (t − t ∧ p)∗T (p − t ∧ p). Now notice that T (t − t ∧ p)∗T (p − t ∧ p)h1 = PHV (t − t ∧ p)∗V (p − t ∧ p)h1 = PHV (t)∗V (p)h1. Similarly, hV (s)π(αs(a))h2, V (t)hi = hπ(αt(a))T (t − t ∧ s)∗T (s − t ∧ s)h2, hi , where T (t − t ∧ s)∗T (s − t ∧ s)h2 = PHV (t)∗V (s)h2 = PHV (t)∗V (p)h1. Therefore, ρ is well defined on the dense subset {V (p)h}. Since V (p) is isometric and π, α are completely contractive, kV (p)π(αp(a))hk = kπ(αp(a))hk ≤ khk = kV (p)hk, and thus ρ(a) is contractive on {V (p)h}. Hence, ρ(a) can be extended to a contractive map on K. Moreover, for any h ∈ H and a ∈ A, we have ρ(a)h = π(a)h ∈ H, and thus H is invariant for ρ. For any a, b ∈ A, p ∈ P , and h ∈ H, ρ(a)ρ(b)V (p)h = V (p)π(αp(a))π(αp(b))h = V (p)π(αp(ab))h = ρ(ab)V (p)h. Therefore, ρ is a contractive representation of A and thus a ∗-representation. Now for any p, t ∈ P and h ∈ H, ρ(a)V (p)V (t)h = V (p + t)π(αp+t(a))h = V (p)V (t)ρ(αp+t(a))h = V (p)ρ(αp(a))V (t)h. Hence, (ρ, V ) is an isometric Nica-covariant pair. (cid:3) REGULAR REPRESENTATIONS OF LATTICE ORDERED SEMIGROUPS 23 This lifting of contractive Nica-covariant pairs to isometric Nica-covariant pairs has significant implication in its associated semi-crossed product. A family of covariant pairs gives rise to a semi-crossed product algebra in the following way [8, 6]. For a C ∗-dynamical system (A, α, P ), denote P(A, P ) be the algebra of all formal polynomials q of the form n q = epiapi, Xi=1 where pi ∈ P and api ∈ A. The multiplication on such polynomials follows the rule that aes = esα(a) and epeq = epq. For a covariant pair (σ, T ) on this dynamical system, define a representation of P(A, P ) by (σ × T ) n Xi=1 epiapi! = n Xi=1 T (pi)σ(api). Now let F be a family of covariant pairs on this dynamical system. We may define a norm on P(A, S) by kpkF = sup{(σ × T )(p) : (σ, T ) ∈ F}, and the semi-crossed product algebra is defined as k·kF . α P = P(A, S) A ×F In particular, A ×nc and A×nc,iso As an immediate corollary from Theorem 3.1 and 7.3, α P is determined by the Nica-covariant representations, P is determined by the isometric Nica-covariant representation. α Corollary 7.4. For a C ∗-dynamical system (A, α, P ), the semi-crossed prod- uct algebra given by Nica-covariant pairs agrees with that given by isometric Nica-covariant pairs. In other words, A ×nc α P ∼= A ×nc,iso α P. Acknowledgements I would like to thank Professor Ken Davidson for pointing out this area of research and giving me directions. I would also like to thank Adam Fuller and Evgenios Kakariadis for many valuable comments. References [1] M. Anderson and T. Feil. Lattice Ordered Groups: An Introduction. Reidel Texts in the Mathematical Sciences, 1988. [2] T. Ando. On a pair of commuting contractions. Acta Sci. Math. (Szeged), pages 88 -- 90, 1963. [3] S. Brehmer. Uber vertauschbare kontraktionen des hilbertschen raumes. Acta Sci. Math. (Szeged), pages 106 -- 111, 1961. [4] M. Darnel. Theory of Lattice-Ordered Groups. CRC Press, 1994. [5] K. Davidson. Nest Algebras. Longman Scientific and Technical, 1988. [6] K. Davidson, A. Fuller, and E.T.A. Kakariadis. Semicrossed products of operator algebras by semigroups, 2014. http://arxiv.org/abs/1404.1906. 24 BOYU LI [7] R.G. Douglas. On majorization, factorization, and range inclusion of operators on hilbert space. Proc. Amer. Math. Soc., pages 413 -- 415, 1966. [8] A. Fuller. Nonself-adjoint semicrossed products by abelian semigroups. Canad. J. Math., 64:768 -- 782, 2013. [9] I. Halperin. Sz.-Nagy-Brehmer dilations. Acta Sci. Math. (Szeged), pages 279 -- 289, 1962. [10] M. Laca and I. Raeburn. Semigroup cross products and the Toeplitz algebras of nonabelian groups. J. Funct. Anal., 139:415 -- 440, 1996. [11] W. Mlak. Unitary dilations in case of ordered groups. Ann. Polon. Math., 17:321 -- 328, 1966. [12] B. Nagy and C. Foias. Harmonic Analysis of Operators on Hilbert Space. Springer Science and Business Media, 2010. [13] M. Neumark. Positive definite operator functions on a commutative group. Izv. Akad. Nauk SSSR Ser. Mat., 7:237 -- 244, 1943. [14] A. Nica. C*-algebras generate by isometries and Wiener Hopf operators. J. Operator Theory, 27:17 -- 52, 1992. [15] S. Parrott. Unitary dilations for commuting contractions. Paciffc J. Math., pages 481 -- 490, 1970. [16] B. Solel. Regular dilations of representations of product systems. Proc. Roy. Irish Acad. Sect. A, 108A:89 -- 110, 2008. [17] B. Sz.Nagy. Bemerkungen zur vorstehenden arbeit des herrn S. Brehmer. Acta Sci. Math. (Szeged), pages 112 -- 114, 1961. Pure Mathematics Department, University of Waterloo, Waterloo, ON, Canada N2L -- 3G1 E-mail address: [email protected]
1707.07317	3	1707	2018-10-19T02:45:30	Asymptotic orthogonalization of subalgebras in II$_1$ factors	[ "math.OA" ]	Let $M$ be a II$_1$ factor with a von Neumann subalgebra $Q\subset M$ that has infinite index under any projection in $Q'\cap M$ (e.g., $Q$ abelian; or $Q$ an irreducible subfactor with infinite Jones index). We prove that given any separable subalgebra $B$ of the ultrapower II$_1$ factor $M^\omega$, for a non-principal ultrafilter $\omega$ on $\Bbb N$, there exists a unitary element $u\in M^\omega$ such that $uBu^*$ is orthogonal to $Q^\omega$.	math.OA	math	ASYMPTOTIC ORTHOGONALIZATION OF SUBALGEBRAS IN II1 FACTORS SORIN POPA University of California, Los Angeles Abstract. Let M be a II1 factor with a von Neumann subalgebra Q ⊂ M that has infinite index under any projection in Q′ ∩ M (e.g., if Q′ ∩ M is diffuse; or if Q is an irreducible subfactor with infinite Jones index). We prove that given any separable subalgebra B of the ultrapower II1 factor M ω, for a non-principal ultrafilter ω on N, there exists a unitary element u ∈ M ω such that uBu∗ is orthogonal to Qω. 1. Introduction We continue in this paper the study of approximate independence properties for subalgebras in II1 factors from [P7,8]. This time, we investigate the possibility of "orthogonalizing" two subalgebras of a II1 factor via asymptotic unitary conjugacy of one of them, but uniformly with respect to the other. Recall in this respect that two ∗-subalgebras N1, N2 in a II1 factor N are called orthogonal (as in [P1]), or 1-independent (as in [P7]), if τ (x1x2) = τ (x1)τ (x2), ∀x1 ∈ N1, x2 ∈ N2, τ denoting the (unique) trace state on the ambient II1 factor. Thus, given the von Neumann subalgebras B, Q of a II1 factor M , the problem we are interested in is to find unitary elements (un)n ⊂ M such that, when viewing u = (un)n as an element in the ultrapower II1 factor M ω for some non-principal ultrafilter ω on N ([W]), the algebras uBu∗ and Qω are orthogonal. While this cannot of course be done if Q is equal to M and B 6= C, or Q merely "virtually equal" to M with dim(B) large enough, we will prove that once Q has "uniform infinite index" in M and B is separable, then such asymptotic orthogonalisation can be obtained. For instance, if Q is an irreducible subfactor of infinite Jones Supported by NSF Grant DMS-1700344, a Simons Fellowship and Chaire FSMP-FMJH 2016 1 Typeset by AMS-TEX 2 SORIN POPA index and M itself is separable, then there exists a unitary element u ∈ M ω such that uM u∗ ⊥ Qω. The uniform infinite index condition for a subalgebra Q in a II1 factor M that we'll require is that for any non-zero projection p ∈ Q′ ∩ M , the [PP]-index of the inclusion Qp ⊂ pM p be infinite. This condition, which we have also used in [P9], is equivalent to the condition that the centralizer of M in the Jones basic construction algebra hM, Qi for Q ⊂ M contains no non-zero finite projections. This amounts to M 6≺M Q in the sense of "intertwining by bimodules" terminology (2.4 in [P6]). We will in fact investigate this asymptotic orthogonalization problem in the more general case when the unitaries (un)n are subject to constraints, being required to lie in some other given von Neumann subalgebra P ⊂ M . Thus, we will prove that if P ⊂ M is any irreducible subfactor such that P 6≺M Q, then one can indeed find u ∈ U(P ω) such that uBu∗ ⊥ Qω. An example of such a situation is when P, Q are irreducible subfactors of M with [M : P ] < ∞ and [M : Q] = ∞. More generally, we prove the following: 1.1. Theorem. Let Mn be a sequence of finite factors, with dim Mn → ∞. For each n, let Qn ⊂ Mn be a von Neumann subalgebra and Pn ⊂ Mn be an irreducible subfactor. Let ω be a non-principal ultrafilter on N. Denote by M the ultraproduct II1 factor ΠωMn with Q := ΠωQn, P := ΠωPn viewed as von Neumann subalgebras in M. Assume P 6≺M Q. Then, given any separable von Neumann subalgebra B ⊂ M, there exists a unitary element u ∈ P such that uBu∗ is orthogonal to Q. For the above condition P 6≺M Q to be satisfied, it is sufficient that Pn 6≺Mn Qn, ∀n, or that Q′ ∩ P = Πω(Q′n ∩ Mn) be diffuse (see Proposition 2.1 below). The condition is also satisfied if Mn are II1 factors, with Pn = Mn and Qn ⊂ Mn are irreducible subfactors satisfying limn[Mn : Qn] = ∞. It is of course satisfied as well when Pn = Mn and Qn are abelian, ∀n. But in fact, as we will show in Remark 2.4, this particular case of Theorem 1.1 can be immediately derived from results in [P4]. As mentioned before, when applied to the case all Mn are equal to the same II1 factor M and all Qn ⊂ M are equal, with Pn = M , the above theorem gives: 1.2. Corollary. Let M be a II1 factor, Q ⊂ M a von Neumann subalgebra such that M 6≺M Q and B ⊂ M a separable von Neumann subalgebra. Then there exists a unitary element u ∈ M ω such that uBu∗ ⊥ Qω. The above result shows in particular that once a countably generated II1 factor can be embedded in the ultrapower Rω of the hyperfinite II1 factor R, then one can actually embed it so that to be orthogonal to the ultraproduct of an arbitrary ORTHOGONALIZATION OF SUBALGEBRAS 3 sequence of irreducible subfactors Qn ⊂ R with limn[R : Qn] = ∞. This fact may be of interest in relation to Connes Approximate Embedding conjecture. Since orthogonality (or 1-independence) between subalgebras is the first stage of n-independence, it is natural to push the above statement even further, trying to find the unitary u ∈ M ω so that uBu∗ becomes n ≥ 2 independent (or even free independent) to Qω. This interesting problem remains open for now. Questions about "rotating" via unitary conjugacy a subalgebra B in a II1 factor M so that to become (approximately) orthogonal to another subalgebra Q ⊂ M have been first considered in [P1]. The case when the algebra B is 2-dimensional (so "smallest possible") has been studied in [P3], notably in the case Q is a subfactor of finite Jones index. We will comment on this and other related problems in the last section of this paper, where more applications to Theorem 1.1 will be mentioned as well. Acknowledgement. This work has been completed during my stay at the D´eparte- ment de Math´ematique de l'Universit´e de Paris Sud and Institut de Math´ematique de Jussieu during the academic year 2016-2017. I want to thank C. Houdayer, G. Pisier, G. Skandalis and S. Vassout for their kind hospitality and support. I am also grateful to F. Le Maitre for his interest in orthogonalization problems, which renewed my own interest in this topic and motivated me to write this paper. 2. Proof of the main results For notations and terminology used hereafter, we send the reader to ([P7,8]; also [AP] for basics in II1 factors theory). We begin by proving some of the criteria for the condition P 6≺M Q to hold true, that we mentioned in the introduction. We will be under the same general assumptions and notations as in Theorem 1.1, which are recalled for convenience. 2.1. Proposition. Let Mn be a sequence of finite factors, with dim Mn → ∞. For each n, let Qn ⊂ Mn be a von Neumann subalgebra and Pn ⊂ Mn be an irreducible subfactor. Let ω be a non-principal ultrafilter on N and denote by M the ultraproduct II1 factor ΠωMn, with Q := ΠωQn, P := ΠωPn viewed as von Neumann subalgebras in M. Assume one of the following conditions is satisfied: 1◦ Pn 6≺Mn Qn, ∀n. 2◦ Πω(Q′n ∩Mn) is diffuse (E.g.: all Qn abelian; or all Mn are finite dimensional factors with Pn = Mn and Qn subfactors with limn(dim Mn/ dim Qn) = ∞). 3◦ Mn are II1 factors, Pn is equal to Mn and Qn ⊂ Mn are irreducible subfactors with limn[Mn : Qn] = ∞. Then P 6≺M Q. 4 SORIN POPA 2.2. Remark. Another criterion for the condition P 6≺M Q to hold true is that Pn, Qn ⊂ Mn be irreducible II1 subfactors of finite Jones index satisfying limn[Mn : Qn]/[Mn : Pn] = ∞. This result, whose proof requires a lengthier analysis, will be discussed in a forthcoming paper, which will in fact address a variety of intertwining problems. Proof of Proposition 2.1. 1◦ Let x1, ..., xm ∈ (M)1 and ε > 0. Let xi,n ∈ (Mn)1 be so that xi = (xi,n)n, 1 ≤ i ≤ m. Since Pn 6≺Mn Qn, by (Theorem 2.1 in [P6]) there exists a unitary element un ∈ Pn such that kEQn(x∗i,nunxj,n)k2 ≤ 2−n, 1 ≤ i, j ≤ m. Thus, if we let u = (un)n then u is a unitary element in ΠωPn = P satisfying EQ(x∗i uxj) = 0, ∀i, j. By (Theorem 2.1), this shows that P 6≺M Q. 2◦ By (Theorem 2.1 in [P6]), the condition P ≺M Q would imply that there exists an intertwining partial isometry v ∈ M between P and Q. Since P ⊂ M is irreducible and Q′ ∩ M = Πω(Q′n ∩ Mn) is diffuse, this implies v∗v ∈ P and vPv∗ = Q0q′ for some wo-closed subalgebra Q0 ⊂ Q and some projection q′ ∈ Q′ ∩M, with vv∗ = 1Q0q′. But the relative commutant (Q0q′)′ ∩ vv∗Mvv∗ contains vv∗(Q′ ∩ M)vv∗ and is thus diffuse, while by spatiality vPv∗ has trivial relative commutant in vv∗Mvv∗, a contradiction. 3◦ Since limn[Mn : Qn] = ∞, we have [M : Q] = ∞ (see e.g. [PP]). Since we also have Q′ ∩ M = C, this implies M′ ∩ hM, Qi = C. But hM, Qi is type II∞, so the only non-zero projection in M′ ∩ hM, Qi is 1hM,Qi, which is not finite in hM, Qi. (cid:3) To prove Theorem 1.1, we first show that for any F ⊂ M ⊖ C finite and any ε > 0 there exists a unitary element v ∈ P such that the expectation onto Q of any element in vF v∗ is ε-close to 0 in the Hilbert norm given by the trace. Such unitary v will be constructed by patching together "incremental pieces" of it, along the lines of the technique developed in [P5,7,8]. The theorem then follows by a "diagonalisation along ω" procedure of this local result, as in ([P7], [P8]). 2.3. Lemma. Let M be a II1 factor, Q ⊂ M a von Neumann subalgebra and P ⊂ M an irreducible subfactor such that P 6≺M Q. Given any finite set F = F ∗ ⊂ (M ⊖ C1)1 and any ε0 > 0, there exists a unitary element v0 ∈ P such that kEQ(v0xv∗0)k2 2 ≤ ε0, for all x ∈ F . Proof. Let ω be a non-principal ultrafilter on N and denote by M = hM ω, Qωi the semi finite von Neumann algebra associated with Jones basic construction for Qω ⊂ M ω. Thus, M = JQω′J = spM ωeM ω w ⊂ B(L2M ω), where e = eQω ∈ B(L2M ω) denotes the orthogonal projection of L2M ω onto L2Qω and J is the canonical canonical conjugation on L2M ω. Recall that the projection e satisfies the condition eye = EQω (y) for any y ∈ M ω. ORTHOGONALIZATION OF SUBALGEBRAS 5 Recall also that the semi-finite von Neumann algebra M is endowed with a canonical normal faithful semi-finite trace T r, satisfying the condition T r(xey) = τ (xy), for all x, y ∈ M ω. Fix ε > 0 such that ε < ε0. Denote by W the set of partial isometries v ∈ P ω with the property that vv∗ = v∗v and which satisfy the conditions: (1) kEQω (vxv∗)k2 2 ≤ ετ (v∗v), τ (vv∗x) = 0, for all x ∈ F . We endow W with the order ≤ in which v1 ≤ v2 if v1 = v2v∗1 v1. (W , ≤) is then clearly inductively ordered and we let v ∈ W be a maximal element. Assume τ (v∗v) < 1 and denote p = 1 − v∗v. Notice right away that τ (pF p) = 0. Let w be a partial isometry in pP ωp with w∗w = ww∗ and denote u = v + w. Then u is a partial isometry in P ω with u∗u = uu∗. We will show that one can make an appropriate choice w 6= 0 such that u = v + w lies in W . We will construct such a w by first choosing its support q = ww∗ = w∗w, then choosing the "phase w" above q. Note first that by writing eux∗u∗euxu∗ as e(v + w)x∗(v + w)∗e(v + w)x(v + w)∗ and developing into the sum of 16 terms, we get (2) kEQω (uxu∗)k2 2 = T r(eux∗u∗euxu∗) = T r(evx∗v∗evxv∗) + Σ1 + Σ2 + Σ3 + Σ4, where Σi denotes the sum of terms having i appearances of elements from {w, w∗}, 1 ≤ i ≤ 4. Thus, there are four terms in Σ1, six in Σ2, four in Σ3, and one in Σ4. Let us first take care of the terms T r(X) with X containing a pattern of the form ...ewxw∗e..., or ...ewx∗w∗e..., for a given x ∈ F . There are seven such terms: the one in Σ4, all four in Σ3, and two in Σ2. We denote by Σ′ the sum of the absolute values of these terms. Note that for each such X, we have T r(X) = T r(wxw∗ey) for some y ∈ (M ω)1 and thus, by applying the Cauchy-Schwartz inequality and taking into account the definition of T r, we get the estimate (3) T r(X) = T r(....ewxw∗e..) = T r(wxw∗ey) ≤ (T r(ewx∗w∗wxw∗e))1/2(T r(qey∗yeq))1/2 ≤ kqxqk2kqk2, where the last inequality is due to the fact that T r(qey∗yeq) ≤ T r(qeq) = τ (q) and T r(ewx∗w∗wxw∗e) = T r(ewx∗qxw∗e) = τ (wx∗qxw) = τ (qx∗qxq). By (Corollary 2.2.(i) in [P4]), the irreducible subfactor pP ωp of the II1 fac- tor pM ωp contains a diffuse abelian subalgebra that's 2-independent to pF p with 6 SORIN POPA respect to the trace state τ (·)/τ (p) on pM ωp. This implies that there exists a projection q ∈ P(pP ωp) of trace τ (q) = ε2τ (p)2/64 such that τ (qx) = 0 and 2/τ (p) = (τ (q)/τ (p))2τ (x∗x)/τ (p), for all x ∈ pF p and thus for all x ∈ kqxqk2 F as well (because q ≤ p). Thus, for each such x ∈ F one has kqxqk2 2 = (ε4τ (p)2/642)τ (x∗x) ≤ ε2τ (q)/64. It follows that kqxqk2 ≤ ετ (q)1/2/8, ∀x ∈ F . Hence, for this choice of q, the right hand side term in (3) will be majorized by ετ (q)/8. By summing up over the seven terms in Σ′, we get Σ′ ≤ 7ετ (q)/8. We now estimate the sum Σ′′ of T r(X) with X running over the remaining four terms in Σ2, and the sum Σ′′′ of four terms T r(X) with X having only one occurrence of w, w∗ (i.e., the sum of the absolute values of the terms in Σ1). We'll show that one can choose the "phase w" above the (fixed by now) projection q in a way that makes Σ′′ + Σ′′′ be majorized by ετ (q)/16. At this point, it is convenient to enumerate the elements in F = {x1, ..., xn}. For each i = 1, 2, ..., n we have (4) Σ′′ = T r(ewx∗i v∗evxiw∗) + T r(evx∗i w∗ewxiv∗) +T r(ewx∗i v∗ewxiv∗) + T r(evx∗i w∗evxiw∗) = T r(w∗ewY1,i) + T r(w∗ewY2,i) +T r(wY3,iwY4,i) + T r(w∗Y5,iw∗Y6,i) where each one of the terms Yj,i depends on xi ∈ F and belongs to the set S0 := q((M ω)1e(M ω)1)q ⊂ qL2(M, T r)q. Note that, as i = 1, 2, ..., n, the number of possible indices (j, i) in (4) is 6n. Note also that there are 2n terms of the form T r(w∗ewY ), n terms of the form T r(wXwY ) and n terms of the form T r(w∗Xw∗Y ), which by using the fact that T r(w∗Xw∗Y ) = T r(wX∗wY ∗) we can view as n additional terms of the form T r(wXwY ). In all this, the elements X, Y belong to S0 ⊂ qL2(M, T r)q, and are thus bounded in operator norm by 1 and are supported (from left and right) by projections of trace T r majorized by 1. Similarly, as i runs over {1, 2, ..., n}, the four terms in Σ′′′ give rise to 4n terms of the form T r(wX), for some X ∈ S0. Note that by the definition of T r, each one of these terms is equal to τ (wy) for some y ∈ (qM ωq)1. We want to prove that for any δ > 0 there exists w ∈ U(qP ωq) such that each one of the above 8n terms is less than δ. To take care of the terms in Σ′′′, note that by results in ([P4] or [P8]) for any finite subset E ⊂ qM ωq, there exists a finite dimensional subfactor P0 ⊂ qP ωq such that kEP ′ 0∩qM ωq(y) − τ (y)/τ (q)qk2 ≤ δτ (q)/2, ∀y ∈ E. By applying this to ORTHOGONALIZATION OF SUBALGEBRAS 7 the elements in Σ′′′, which are of the form τ (wy) with y running over a certain finite set E ⊂ (qM ωq)1, and using the Cauchy-Schwartz inequality, one obtains that for each unitary element w ∈ N := P ′0 ∩ qP ωq of trace satisfying τ (w) ≤ δτ (q)/2, we have τ (wy) = τ (EP ′ ≤ τ (w(EP ′ 0∩qM ωq(wy)) = τ (wEP ′ 0∩qM ωq(y)) 0∩qM ωq(y) − (τ (y)/τ (q))q) + τ (w)τ (y) ≤ δτ (q)/2 + δτ (q)/2 = δτ (q), forall y ∈ E. Taking δ sufficiently small, one obtains that for any 1 ≤ i ≤ n one has Σ′′′ ≤ ετ (q)/32, for any unitary element w ∈ N satisfying τ (w) ≤ δτ (q)/2. Finally, let us take care of the terms in Σ′′. To do this, recall that we are under the assumption P 6≺M Q, which in turn implies P ω 6≺M ω Qω. Thus, (P ω)′ ∩ M contains no finite non-zero projections of M = hM ω, Qωi and so N′ ∩qMq contains no finite non-zero projections of M either. To estimate the terms in Σ′′, we first show that for any δ0 > 0 and any two m-tuples of elements (Z1, ..., Zm), (Z′1, ..., Z′m) in S0 ∩ M+, there exists a unitary element w ∈ N such that (5) ΣiT r(w∗ZiwZ′i) ≤ δ0. To see this, let H denote the Hilbert space L2(qMq, T r)⊕m and note that we have a unitary representation U(N ) ∋ w 7→ π(w) ∈ U(H), which on an m-tuple X = (Xi)m i=1 ∈ H acts by π(w)(X) = (w∗Xiw)i. Now note that this representation has no (non-zero) fixed point. Indeed, for if X ∈ H satisfies π(w)(X) = X, ∀w ∈ U(N ), then on each component Xi ∈ L2(qMq, T r) of X we would have w∗Xiw = Xi, ∀w. Thus Xiw = wXi and since the unitaries of N span linearly the algebra N , this would imply Xi ∈ N′∩L2(qMq, T r). Hence, X∗i Xi ∈ N′ ∩ L1(qMq, T r) and therefore all spectral projections of X∗i Xi corresponding to intervals [t, ∞) with t > 0 would be projections of finite trace in N′ ∩ qMq, forcing them all to be equal to 0. Thus, Xi = 0 for all i. With this in mind, denote by KZ ⊂ H the weak closure of the convex hull of the set {π(w)(Z) w ∈ U(N )}, where Z = (Z1, ..., Zm) is viewed as an element in H. Since KZ is bounded and weakly closed, it is weakly compact, so it has a unique element Z 0 ∈ KZ of minimal norm k k2,T r. Since KZ is invariant to π(w) and kπ(w)(Z 0)k2,T r = kZ 0k2,T r, it follows that π(w)(Z 0) = Z 0. But we have shown that π has no non-zero fixed points, and so 0 = Z 0 ∈ KZ. Let us deduce from this that if Z = (Zi)i, Z′ = (Z′i)i are the two m-tuples of positive elements in S0, then we can find w ∈ U(N ) such that (5) holds true. Indeed, for if there would exist δ0 > 0 such that ΣiT r(π(w)(Zi)Z′i) ≥ δ0, ∀w ∈ U(N ), then 8 SORIN POPA by taking convex combinations and weak closure, one would get 0 = hZ 0, Z′i ≥ δ0, a contradiction. This finishes the proof of (5). Note that by taking for one of the i the elements Yi, Y ′i to be equal to e, one can get w ∈ U(N ) to also satisfy τ (w)2 ≤ δ0. We will now use this fact to prove that, given any m-tuples (Xi)i, (Yi)i, (X′i)i, (Y ′i )i ∈ Sm 0 (not necessarily having positive operators as entries), there exists w ∈ U(N ) such that T r(w∗XiwX′i) ≤ δ0, T r(wYiwY ′i ) ≤ δ0, for all i. Indeed, because if we denote by ei the left support of X′i and fi the left support of Y ′i , then by the Cauchy-Schwartz inequality we simultaneously have for all i the estimates T r(w∗XiwX′i)2 ≤ T r(w∗X∗i XiwX′iX′i∗)T r(ei) ≤ T r(w∗X∗i XiwX′iX′i∗), and respectively T r(wYiwY ′i )2 ≤ T r(w∗Y ∗i YiwY ′i Y ′i ∗)T r(fi) ≤ T r(w∗Y ∗i YiwY ′i Y ′i ∗). Since all X∗i Xi, X′iX′i∗, Y ∗i Yi, Y ′i Y ′i ∗ are positive elements in S0, we can now apply (5) to deduce that there exist w ∈ U(N ) of arbitrarily small trace such that all the 4n elements appearing in Σ′′ for i = 1, 2, ..., n are arbitrarily small, making Σ′′ ≤ ετ (q)/32, ∀i. Altogether, we then get for u = v + w the estimate kEQω (uxu∗)k2 2 = kEQω (vxv∗)k2 2 + Σ1 + Σ2 + Σ3 + Σ4 ≤ ετ (vv∗) + Σ′ + Σ′′ + Σ′′′ ≤ ετ (vv∗) + 15ετ (ww∗)/16 + 2ετ (ww∗)/32 = ετ (uu∗), for any x ∈ F . Thus, u ∈ W and u ≥ v, u 6= v, contradicting the maximality of v. This shows that v must be a unitary element. Thus, if we represent v ∈ P ω as a sequence of unitary elements (vn)n in P , then we have kEQ(vnxv∗n)k2 2 = kEQ(vxv∗)k2 2 ≤ ε < ε0, lim n→ω for all x ∈ F . Thus, if we let v0 = vn for some large enough n, then v0 satisfies kEQ(v0xv∗0)k2 2 ≤ ε0, for all x ∈ F . (cid:3) Proof of Theorem 1.1. Let {bj}j ⊂ B be a sequence of elements that's dense in (B)1 in the Hilbert norm k k2. By applying the above lemma to the factor M = ΠωMn, with its von Neumann subalgebra Q = ΠωQn and its irreducible ORTHOGONALIZATION OF SUBALGEBRAS 9 subfactor P = ΠωPn, the finite set F = {b1, ..., bm} and ε = 2−m−1, one gets a unitary element um ∈ P such that kEQ(umbju∗m)k2 2 ≤ 2−m−1 for all 1 ≤ j ≤ m. Let us now take representations bj = (bj,n)n ∈ B ⊂ M and um = (um,n)n, with um,n ∈ U(Pn) and bj,n ∈ (Mn)1. Thus, we have kEQn(um,nbj,nu∗m,n)k2 2 ≤ 2−m−1. lim n→ω Denote by Vm the set of all n ∈ N such that kEQn(um,nbj,nu∗m,n)k2 2 < 2−m, for all 1 ≤ j ≤ m. Note that Vm corresponds to an open closed neighborhood of ω in Ω, under the identification ℓ∞N = C(Ω). Let now Wm, m ≥ 0, be defined recursively as follows: W0 = N and Wm+1 = Wn ∩ Vv+1 ∩ {n ∈ N n > min Wn}. Note that, with the same identification as before, Wm is a strictly decreasing sequence of neighborhoods of ω in Ω. Define u = (um)m by letting un = um,n for n ∈ Wm−1 \ Wm. It is then easy to see that u is a unitary element in P and that limω kEQn(unbj,nu∗n)k2 2 = 0 for any j. In other words, EQ(ubju∗) = 0, ∀j. By the density of the set {bj}j in (B)1, it follows that uBu∗ ⊥ Q. (cid:3) Proof of Corollary 1.2. This is just the case where all Mn are equal to the same II1 factor M , all Pn ⊂ M are equal to M and Qn = Q, ∀n, of Theorem 1.1. (cid:3) 2.4. Remark. Let us note here that the case Pn = Mn and Qn ⊂ Mn abelian, ∀n, in Theorem 1.1 can be easily deduced directly from the main Theorem in [P4]. To see this, note first that it is sufficient to prove the statement for a larger Q, and since we can embed each Qn in a MASA (maximal abelian ∗-subalgebra) of Mn, it follows that it is sufficient to settle the case when Q = ΠωQn has diffuse center Z. Then note that by [P4] there exists a Haar unitary v ∈ M that's free independent to B. Since in an ultrapower II1 factor any two Haar unitaries are unitary conjugate, since Z was assumed diffuse, it follows that there exists u ∈ U(M) such that A0 := u{v}′′u∗ ⊂ Z = Z(Q). Thus, uBu∗ is free independent to A0 ⊂ Q. But then, by Kesten's Theorem [K], if {pk}k is any finite partition of 1 with projections in A0 of trace ≤ ε2/2, then one has as in [P7] the estimate kΣkpkxpkk ≤ ε, ∀x ∈ (uBu∗)1 with τ (x) = 0. Thus, since [A0, Q] = 0, we have EQ(x) = ΣkpkEQ(x)pk = EQ(Σkpkxpk) has norm ≤ ε, with ε > 0 arbitrary, implying that x ⊥ Q, i.e., uBu∗ ⊥ Q. 3.1. Initial work on orthogonal subalgebras. The orthogonalization relation between 3. Further comments 10 SORIN POPA subalgebras B, Q of a II1 factor M , as well as questions about conjugating a sub- algebra B by a unitary element u ∈ M such that uBu∗ becomes orthogonal to Q, have been first considered in [P1]. They were used in that paper as a tool for calcu- lating normalizers, and more generally the intertwining space between subalgebras, in the spirit of what has later become the intertwining by bimodules techniques [P6]. For instance, it was shown in [P1] that if a unitary u ∈ M has the property that uA0u∗ ⊥ Q for some diffuse abelian subalgebra A0 ⊂ Q, then u is perpendicular to the normalizer of Q in M . This allowed to prove several indecomposability proper- ties (e.g., absence of Cartan subalgebras) for ultraproduct II1 factors and for free group factors with uncountable number of generators. Related to asymptotic orthogonalization, it has been shown in (Lemma 2.5 in [P1] and Corollary 2.4 in [P9]) that in order for a unitary u to be orthogonal to the intertwining space I(P, Q) between subalgebras P, Q ⊂ M , it is necessary and sufficient that there exists a diffuse subalgebra B ⊂ Qω such that uBu∗ ⊥ P ω. 3.2. The orthogonalization problem. Given von Neumann subalgebras B, Q in a II1 factor M , the problem of finding a unitary u ∈ M such that uBu∗ ⊥ Q will be called the orthogonalization problem for B, Q ⊂ M . For a given von Neumann subalgebra Q ⊂ M and a fixed finite dimensional subalgebra B, with its trace inherited from M , all isomorphic copies of B are unitary conjugate in M . Thus, the orthogonalization problem becomes a question about whether there exist copies of B that are perpendicular to Q. This provides a source of isomorphism invariants for the inclusion Q ⊂ M . 3.3. The two dimensional case. The simplest case of this problem is when B is two dimensional, i.e., B = Cf + C(1 − f ), for some projection f ∈ M of trace τ (f ) = α. Thus, since all projections of same trace are unitary conjugate in M , the question of whether B can be unitarily conjugated to an algebra orthogonal to the von Neumann subalgebra Q ⊂ M becomes: for what α ∈ (0, 1] does there exist f ∈ P(M ) such that EQ(f ) = α1. This problem has been systematically investigated in [P3], where the answers are formulated in terms of the invariant Λ(Q ⊂ M ) := {α ∈ [0, 1] ∃f ∈ P(M ), EQ(f ) = α1}. Similarly, one denotes its approximate version by Λapp(Q ⊂ M ) := {α ∈ ([0, 1] ∀ε > 0, ∃f ∈ P(M ), kEQ(f ) − α1k2 ≤ ε}. It obviously coincides with Λ(Qω ⊂ M ω) and satisfies the property: α ∈ Λapp if and only if the algebra B = Cf + C(1 − f ), with f a projection of trace α, ORTHOGONALIZATION OF SUBALGEBRAS 11 can be asymptotically conjugated to an algebra orthogonal to Q. Note also that Λapp(Qω ⊂ M ω) = Λ(Qω ⊂ M ω). It was already noticed in [P3] that if Q ⊂ M is an irreducible subfactor of infinite index, then (2.4 in [PP]) can be used to show that Λapp(Q ⊂ M ) = Λ(Qω ⊂ M ω) = [0, 1]. This is of course implied by Theorem 1.1, which in fact applies to all cases when M 6≺M Q, for instance when Q = A is a MASA in M . The calculation of the invariant Λ(Q ⊂ M ) is in general quite difficult, but some partial answers could be obtained in [P3] in several particular cases. For instance, if Q = A is a MASA then Λ(A ⊂ M ) contains all rationals in [0, 1] and if in addition NM (A)′′ is of type II1 then Λ(A ⊂ M ) = [0, 1] (this ought to be the case for any MASA). In the finite index case, the results obtained in [P3] depend on weather [M : Q] < 4 (thus [M : Q] = 4 cos2(π/n+2) for some n ≥ 1, by Jones celebrated results in [J]), or [M : Q] ≥ 4. To describe them, denote [M : Q]−1 = λ and define recursively the polynomials Pk(x) by P−1 ≡ 1, P0 ≡ 1, Pk+1(x) = Pk(x) − xPk−1(x), k ≥ 0. Then, (Theorem in [P3]) shows that if [M : Q] = λ−1 = 4 cos2(π/n + 2) then we have Λ(Q ⊂ M ) = {0} ∪ {λPk−1/λPk(λ) 0 ≤ k ≤ n − 1}. While if [M : Q] ≥ 4 and we let 0 < t ≤ 1/2 be so that t(1−t) = λ, then Λ(Q ⊂ M )∩(0, t) = {λPk−1(λ)/Pk(λ) k ≥ 0}. Thus, the situation is quite rigid when the index is under the threshold 4, with the set Λ being finite and completely understood. While above the threshold 4 the set is always infinite, being completely determined in the intervals [0, t) ∪ (1 − t, 1], but with the calculation of Λ(Q ⊂ M )∩[t, 1−t] still open in general. It is interesting to note that in both cases (Theorem in [P3]) provides the following uniqueness result as well: any two projections f1, f2 ∈ M satisfying EQ(f1) = α1 = EQ(f2), with α = λPk−1(λ)/Pk(λ) for some k ≥ 0, are conjugate by a unitary element in Q. Moreover, since [M ω : Qω] = [M : Q], the results in [P3] show that Λ(Q ⊂ M ) = Λapp(Q ⊂ M ) for index < 4 and Λ(Q ⊂ M ) ∩ (0, t) = Λapp(Q ⊂ M ) ∩ (0, t) for index ≥ 4, with any projection that's close to expect on a scalar in these sets being close to a projection that actually expects on a scalar. The results in [P3] also show that for subfactors Q ⊂ M of index ≥ 4 that are locally trivial, i.e., for which Q′ ∩ M = Cf + C(1 − f ), f M f = Qf , (1 − f )M (1 − f ) = Q(1 − f ), with τ (f ) = t ≤ 1/2 where t(1 − t) = λ = [M : Q]−1, the invariant Λ(Q ⊂ M ) contains no points in the interval (t, 1 − t), being equal to the set {0, t} ∪ {λPk−1(λ)/Pk(λ) k ≥ 0} when intersected with [0, 1/2]. This is in particular the case when Q = {en n ≥ 1}′′ ⊂ {en n ≥ 0}′′ = M is a subfactor generated by Jones projections of trace τ (en) = λ < 1/4. The opposite phenomenon may hold true for subfactors Q ⊂ M with graph A∞ and [M : Q] = λ−1 > 4. Namely, it is quite possible that in all such cases 12 SORIN POPA one has [t, 1 − t] ⊂ Λ(Q ⊂ M ). As a supporting evidence, consider the standard representation Qst ⊂E Mst of Q ⊂ M , as described in [P10]. This is an inclusion of discrete type I von Neumann algebras (i.e., direct sums of type I∞ factors) with inclusion graph A∞ and a conditional expectation E having the property that EM = EM Q . Also, E is the unique expectation preserving the trace T r on Mst whose weights are proportional to the square roots of indices of irreducible subfactors in the Jones tower. Since Q ⊂ M is embedded with commuting squares into Qst ⊂E Mst, one has Λ(Q ⊂ M ) ⊂ Λ(Qst ⊂E Mst), and it is an easy exercise to see that the latter contains the entire interval [t, 1 − t] (however, it is not clear how one could "push down" into M a projection p ∈ Mst that satisfies E(p) = s1 with s ∈ [t, 1 − t]). 3.4. The finite dimensional case. The orthogonalization problem is certainly in- teresting for any finite dimensional abelian algebra B = Σn Cfi, with f1, ..., fn a partition of 1 with projections in M of trace τ (fi) = αi, beyond the case n = 2. To state this problem properly, for each n ≥ 2 we consider the set Fn of all n-tuples α = (α1, ..., αn) with 0 ≤ αi ≤ 1 and Σiαi = 1. If Q is a von Neumann subalgebra of M , we denote i=1 Λn−1(Q ⊂ M ) := {(αi)i ∈ Fn ∃f1, ..., fn ∈ P(M ), Σifi = 1, EQ(fi) = αi, ∀i}. j=1 Thus, Λ1(Q ⊂ M ) = {(β, 1−β) β ∈ Λ(Q ⊂ M )}. Also, any (αi)i ∈ Λn−1(Q ⊂ M ) produces elements in Λk−1(Q ⊂ M ) with k ≤ n − 2 by taking k-tuples (βj)k corresponding to partitions of {1, ..., n} into k subsets Sj and letting βj = Σi∈Sj αi. Thus, the restrictions on Λ = Λ1 propagate into a set of restrictions for Λn, n ≥ 2. In particular, any entry of an element in Λn(Q ⊂ M ) for n ≥ 1 is at least [M : Q]−1. The question here is to calculate (or at least estimate) the invariants Λn(Q ⊂ M ) for all n ≥ 1. The case when Q is an irreducible subfactor of finite index is particularly interesting. One source of (n + 1)-tuples α ∈ Λn(Q ⊂ M ) is to take irreducible subfactors P ⊂ Q and look for partitions of 1 with n + 1 projections f1, ..., fn, fn+1 in P ′ ∩ M . Q (fi) ∈ P ′ ∩ Q = C1 must be scalars. Such P ⊂ Q can be taken to be an irreducible subfactor obtained by reducing inclusions from a Jones tunnel Q−m ⊂ Q−m+1 ⊂ ... ⊂ Q−1 ⊂ Q ⊂ M associated with Q ⊂ M by a minimal projection in Q′ −m ∩ Q. Thus, an (n + 1)-tuple in Λn will appear whenever one has an (n + 1)-point in the principal graphs of Q ⊂ M , M ⊂ hM, Qi. For instance, any triple point in ΓQ⊂M , will produce an element in Λ2(Q ⊂ M ), whose entries are proportional to square roots of indices of the corresponding irreducible subfactors in the Jones tunnel/tower. When combined with the restrictions on entries coming from the obstructions on Λ(Q ⊂ M ) = Λ1(Q ⊂ M ), this can provide restrictions on the existence of triple Indeed, because then EM ORTHOGONALIZATION OF SUBALGEBRAS 13 points (and thus of graphs of subfactors). More generally, one can apply this same reasoning to the universal graph of Γu Q⊂M , as defined in [P10]. Another interesting question for subfactors of finite index Q ⊂ M is whether Λ(Q ⊂ M ) (more generally Λn(Q ⊂ M ), n ≥ 1) depends solely on the standard invariant GQ⊂M of the subfactor Q ⊂ M . The results in [P3] show that if [M : Q] < 4, then in fact Λ(Q ⊂ M ) = Λ1(Q ⊂ M ) only depends on [M : Q]. This may be the case for all Λn(Q ⊂ M ), when [M : Q] ≤ 4, but it is quite unclear for index > 4, where however the irreducibly condition Q′ ∩ M = C should be imposed. A test case is when ΓQ⊂M = A∞ (i.e., when Q ⊂ M has TLJ standard invariant) with the index running over the interval (4, ∞) (cf. [P5]). 3.5. Unitaries in orthonormal basis. One case of particular interest is to decide whether for some given 2 ≤ k ≤ n one can have α = (αi)i ∈ Λn(Q ⊂ M ) with α1 = α2 = ... = αk = λ = [M : Q]−1. Thus, in such a case one has sλ ∈ Λ(Q ⊂ M ) for any s = 0, 1, 2, ..., k. By (Proposition 1.7 in [PP]), this is equivalent to whether Q−1 ⊂ Q has an orthonormal basis {mi}i with the first k terms 1 = m1, ..., mk being unitary elements. If [M : Q] 6∈ N and n denotes its integer part then, as pointed out in (1.4.2◦ of [PP]), the formula Σimim∗i = [M : Q]1 implies k ≤ n − 1. The [P3] restrictions on Λ(Q ⊂ M ) can be used to obtain further restrictions on the maximal number of unitaries that can appear in an orthonormal basis of M over Q. Indeed, if the index [M : Q] is less than 4 but 6= 3, then the equations λPm−1(λ)/Pm(λ) = 2λ do not have solutions for λ−1 = 4 cos2(π/n+2) and n 6= 2, 4. If in turn [M : Q] > 4 and we let [M : Q] = n + ε, with 1 > ε > 0, then having n − 1 mutually orthogonal projections f1, ..., fn−1 ∈ M with EQ(fi) = λ = [M : Q]−1 would imply that f = 1 − Σifi satisfies τ (f ) = 1 − (n − 1)λ ∈ Λ(Q ⊂ M ) and τ (f ) ≥ λ/(1 − λ). This in turn implies ε ≥ λ/(1 − λ). Thus, ε < λ/(1 − λ) forces 2 ≈ 4.3, then k ≤ n − 2. In particular, this shows that if 4 < [M : Q] < 4 + k ≤ 2, i.e., there exists at most one unitary u ∈ M with EQ(u) = 0. √13−3 In turn, it is an open problem whether an irreducible subfactor Q ⊂ M with integer index n ≥ 5 always has an orthonormal basis with n unitaries. Note that by a result in [P2], if a subfactor Q of a II1 factor M contains a Cartan subalgebra of M , then Q ⊂ M does have an orthonormal basis of unitaries (even if its index is infinite). However, if a subfactor Q ⊂ M has A∞-graph (e.g., when Q ⊂ M is constructed by the universal amalgamated free product method in [P5]), then finding even one single unitary u ∈ M with EQ(u) = 0 is an open question (see also the conjecture at the end of 3.3 above). 3.6. A related dilation problem. The question about whether a scalar α ∈ (0, 1) is the expected value on Q of a projection in M is viewed in [P3] as a dilation 14 SORIN POPA problem. More generally, one can ask this same question for an arbitrary element b ∈ Q with 0 ≤ b ≤ 1: can b be dilated to a projection p ∈ M , i.e., does there exist a projection p ∈ M such that EQ(p) = b ? Alternatively, one can attempt the calculation of the entire set EQ(P(M )), or of the set Λ(Q ⊂ M ) of all functions g : [0, 1] → [0, 1] that can be spectral distributions of elements in EQ(P(M )), i.e., with the property that there exists p ∈ P(M ) with g(α) = τ (eα(EQ(p)), ∀α ∈ [0, 1]. These sets should be calculable for subfactors of index < 4. On the other hand, note that if Q ⊂ M satisfies the infinite index condition M 6≺M Q, then Corollary 1.2 easily yields a calculation of the approximate versions of these sets, showing that EQω (P(M ω)) = {b ∈ Qω 0 ≤ b ≤ 1} and thus Λ(Qω ⊂ M ω) is equal to the set of non-increasing functions from [0, 1] to [0, 1]. To see this, note first that EQω (P(M ω)) is k k2-closed (by the usual ω-diagonalisation procedure). Then note that any b ∈ Qω with 0 ≤ b ≤ 1 can be approximated uniformly by elements of finite spectrum b′ = Σiαiqi with qi ∈ P(Qω), 0 ≤ αi ≤ 1. Finally, as noticed in 3.3 above, any αiqi can be dilated to a projection pi ∈ qiM ωqi, thus p = Σipi satisfies EQω (p) = b′. References [AP] C. Anantharaman, S. Popa: "An introduction to II1 factors", www.math.ucla.edu/∼popa/Books/ [C] A. Connes: Classification of injective factors, Ann. of Math., 104 (1976), 73-115. [J] V.F.R. Jones: Index for subfactors, Invent. Math. 72 (1983), 1-25. [K] H. Kesten: Symmetric random walks on groups, Trans. Amer. Math. Soc. 92 (1959), 336-354. [PP] M. Pimsner, S. Popa, Entropy and index for subfactors, Annales Scient. Ecole Norm. Sup., 19 (1986), 57-106. [P1] S. Popa, Orthogonal pairs of *-subalgebras in finite von Neumann algebras, J. Operator Theory, 9 (1983), 253-268. [P2] S. Popa, Notes on Cartan subalgebras in type II1 factors. Mathematica Scandi- navica, 57 (1985), 171-188. [P3] S. Popa: Relative dimensions, towers of projections and commuting squares of subfactors, Pacific J. Math., 137 (1989), 181-207, [P4] S. Popa: Free independent sequences in type II1 factors and related problems, Asterisque, 232 (1995), 187-202. [P5] S. Popa, An axiomatization of the lattice of higher relative commutants of a subfactor, Invent. Math., 120 (1995), 427-445. ORTHOGONALIZATION OF SUBALGEBRAS 15 [P6] S. Popa: Strong Rigidity of II1 Factors Arising from Malleable Actions of w- Rigid Groups I, Invent. Math., 165 (2006), 369-408. [P7] S. Popa: A II1 factor approach to the Kadison-Singer problem, Comm. Math. Physics. 332 (2014), 379-414 (math.OA/1303.1424). [P8] S. Popa: Independence properties in subalgebras of ultraproduct II1 factors, Jour- nal of Functional Analysis 266 (2014), 5818-5846 (math.OA/1308.3982) [P9] S. Popa: Constructing MASAs with prescribed properties, to appear in Kyoto J. of Math, math.OA/1610.08945 [P10] S. Popa: Some properties of the symmetric enveloping algebras with applications to amenability and property T, Documenta Mathematica, 4 (1999), 665-744. [V] D. Voiculescu: Symmetries of some reduced free product C∗-algebras, In: "Op- erator algebras and their connections with topology and ergodic theory", Lect. Notes in Math. Vol. 1132, 566-588 (1985). [W] F. B. Wright: A reduction for algebras of finite type, Ann. of Math. 60 (1954), 560 - 570. Math.Dept., UCLA, Los Angeles, CA 90095-1555, [email protected]
1708.05796	1	1708	2017-08-19T03:12:55	A New Index Theorem for Monomial Ideals by Resolutions	[ "math.OA", "math.FA", "math.KT" ]	We prove an index theorem for the quotient module of a monomial ideal. We obtain this result by resolving the monomial ideal by a sequence of Bergman space like essentially normal Hilbert modules.	math.OA	math	A NEW INDEX THEOREM FOR MONOMIAL IDEALS BY RESOLUTIONS RONALD G. DOUGLAS, MOHAMMAD JABBARI, XIANG TANG, AND GUOLIANG YU Abstract. We prove an index theorem for the quotient module of a monomial ideal. We obtain this result by resolving the monomial ideal by a sequence of Bergman space like essentially normal Hilbert modules. 1. Introduction Let Bm be the unit ball in the complex space Cm, and L2 a(Bm) be the Bergman space of square integrable holomorphic functions on Bm. Denote A to be the algebra C[z1,··· , zm] of polynomials of m generators. The algebra A plays two roles in our study. One is that a(Bm) A is a dense subspace of the Hilbert space L2 by Toeplitz operators. a(Bm), the other is that A acts on L2 In this article we are interested in an ideal I of A generated by monomials. Let I be a(Bm)/I. The first the closure of I in L2 author proved in [4, Theorem 2.1] that the the Toeplitz operators Tzi on I, i = 1,··· , m, and the quotient QI are essentially normal1, i.e. the following commutators are compact, a(Bm), and QI be the quotient Hilbert space L2 [TziI,(cid:0)TzjI (cid:1)∗ ] ∈ K(I), and [TziQI ,(cid:0)TzjQI (cid:1)∗ ] ∈ K(QI), i, j = 1,··· , m. Let T(QI) be the unital C∗-algebra generated by the Toeplitz operators TziQI , i = 1,··· , m. The above essentially normal property of the Toeplitz operators gives the following extension sequence 0 −→ K −→ T(QI) −→ C(σe I) −→ 0, I is the essential spectrum space of the algebra T(QI) and K is the algebra of where σe compact operators. The index problem we want to answer in this article is to provide a good description of the above K-homology class. The main difficulty in answering the above question is that the ideal I in general fails to be radical. This makes the geometric ideas introduced in [6] and [9] impossible to apply directly. The seed of the main idea in this article is the following observation discussed 1(cid:105) (cid:47) A = C[z1, z2]. The quotient in [6, Section 5.2]. For m = 2, consider the ideal I = (cid:104)z2 1Arveson [2, Corollary 2.2] proved the similar result on the Drury-Arveson space. 1 2 RONALD G. DOUGLAS, MOHAMMAD JABBARI, XIANG TANG, AND GUOLIANG YU QI can be written as the sum of two space a,1(D) ⊕ L2 L2 a,2(D), a,2(−)) is the where D is the unit disk inside the complex plane C, and L2 weighted Bergman space with respect to the weight defined by the defining function 1−z2 (and (1 − z2)2). Define the restriction map RI : L2 ∂f ∂z1 a,1(−) (and L2 a,1(D) ⊕ L2 a(B2) → L2 z1=0). RI(f ) := (fz1=0, a,2(D) by Then it is not hard to introduce a Hilbert A = C[z1, z2]-module structure on L2 a,2(D) so that the following exact sequence of Hilbert modules holds, L2 a,2(D) −→ 0. 0 −→ I −→ L2 a(B2) −→ L2 a,1(D) ⊕ L2 a,1(D) ⊕ a,1(D) ⊕ L2 a,2(D). a,1(D) and It is well known that the Toeplitz operators on the (weighted) Bergman space L2 a,2(D) are essentially normal. In [6, Section 5.2], we observe that one can conclude the L2 essential normal property of the Hilbert module I and QI from the above exact sequence. Furthermore, it is not hard to prove that the quotient Hilbert module QI is isomorphic to a,2(D). And we can identify the extension class [T(QI)] associated the module L2 to QI with the one associated to L2 a,1(D)⊕ L2 In this article, we extend the above example to an arbitrary ideal I of L2 a(Bm) generated by monomials. By considering an ideal generated by two monomials, we realize that it is more natural to work with long exact sequences of Hilbert modules, instead of short ones. The following is our main theorem. Theorem 1.1. Let I be an ideal of C[z1,··· , zm] generated by monomials, and I be its a(Bm). There are Bergman space like Hilbert A-modules closure in the Bergman space L2 A0 = L2 a(Bm),A1,··· ,Ak together with bounded A-module morphisms Ψi : Ai → Ai+1, i = 0,··· , k − 1 such that the following exact sequence of Hilbert modules holds 0 −→ I (cid:44)→ L2 a(Bm) Ψ0−→ A1 Ψ1−→ ··· Ψk−1−→ Ak −→ 0. As a corollary of Theorem 1.1, we obtain a new proof of essentially normal property of the ideal I and its quotient QI. Moreover, Theorem 1.1 allows us to identify the extension class associated to T(QI) geometrically. We compute it in the following theorem. Theorem 1.2. (Theorem 3.10) Let T(Ai) be the unital C∗-algebra generated by Toeplitz operators on Ai, and σe 1 ∪ ··· ∪ σe k), the following equation holds, i be the associated essential spectrum space. In K1(σe [T(QI)] = [T(A1)] − [T(A2)] + ··· + (−1)k−1[T(Ak)], INDEX THEOREM AND RESOLUTION 3 As is explained in Section 2.3 and Remark 3.2, every algebra T(Ai), i = 1,··· , k, can be identified as the algebra of Toeplitz operators on square integrable holomorphic sections of a hermitian vector bundle on a disjoint union of subsets of Bm. This geometric interpretation allows us to use the ideas developed in [6] and [9] to study the "geometry" of the algebra T(QI). We would like to remark that although our Theorem 1.1 is stated for the closure of a(Bm), the same proof also works for the a monomial ideal inside the Bergman space L2 closure of any monomial ideal inside more general spaces, e.g. weighted Bergman spaces a,s(Bm) and the Drury-Arveson space. L2 Our result in Theorem 1.1 can be viewed as a "resolution" ([11]) of the ideal I by essentially normal Hilbert modules A0,··· ,Ak. Such an idea of "resolution" goes back the first author's work in [4], and the study in this article should not be limited to monomial ideals. In Section 4.3, we explain that a similar construction also works on a more general ideal in C[z1, z2, z3]. We hope to report in the near future about a systematic study on extending ideas from Theorem 1.1 to more general ideals I. This article is organized as follows. In Section 2, we will introduce the main building block in our construction. In Section 3, we will construct the Hilbert modules Ai and morphisms Ψi, i = 0,··· , k in Theorem 1.1, and prove the main Theorem. Our proof is inspired by the corresponding algebraic study on monomial ideals [3], [10]. We will end our paper by exhibiting our constructions on concrete examples in Section 4. In particu- lar, a non-monomial ideal example is discussed in Section 4.3. Acknowledgements: All authors are partially supported by U.S. NSF. Yu is partially supported by NSFC11420101001. 2. Generalized Bergman Space and the Associated Toeplitz Operators In this section, we introduce and study a building block in the construction of the resolution in Theorem 1.1. 2.1. Notations. We start with fixing some notations. For a positive integer q, we use the symbol Sq(m) to denote the set of q-shuffles of the set [m] = {1,··· , m}, i.e. Sq(m) := {j := (j1,··· , jq)1 ≤ j1 < j2 < ··· < jq ≤ m}. Let N be the set of all nonnegative integers. For any i = (i1,··· , iq) ∈ Nq, we use i to denote the sum i1 + ··· + iq. Fix j = (j1,··· , jq) ∈ Sq(m), and b = (b1,··· , bq) ∈ Nq. We associate a subset Bb j ⊆ Nm defined by j := {(n1,··· , nm) ∈ Nmni ∈ N, i /∈ j, 0 ≤ ni ≤ bk, i = jk ∈ j}. Bb 4 RONALD G. DOUGLAS, MOHAMMAD JABBARI, XIANG TANG, AND GUOLIANG YU We call Bb j the box associated to j and b. In the following, we introduce a Hilbert space Hb On Bm, consider the weighted Bergman space L2 holomorphic functions on Bm with respect to the norm (cid:90) f2 L2 a,s = f (z)2(1 − z2)s (m + s)! m!s! dV (z), Bm j as a closed subspace of L2 a,s(Bm) defined by square integrable a(Bm). where dV is the normalized Lebesgue measure on Bm. This Hilbert space has the following standard orthonormal basis (cid:12)(cid:12)(cid:12)(cid:12) n = (n1,··· , nm) ∈ Nm (cid:41) (cid:40) (1) zn := 1 ··· znm zn1 m(cid:112)ωs(n) a(Bm) whose expansion (cid:80) where ωs(n) := n1!···nm!(m+s)! Definition 2.1. The Hilbert space Hb tions f ∈ L2 satisfies (n1+···+nm+s+m)!. j is a closed subspace of L2 a(Bm) consisting of func- n fnzn with respect to the orthonormal basis zn We have the following orthonormal basis for the Hilbert space Hb j , fn = 0, for n /∈ Bb j . zn := , zn1 1 ...znm m(cid:112)ω0(n) (cid:88) 0≤nj1≤b1,··· ,0≤njq≤bq where 0 ≤ nj1 ≤ b1,··· , 0 ≤ njq ≤ bq, and the index ni for i /∈ j belongs to N. In terms of this basis, an element X ∈ Hb j can be written as (2) X = Xn1···nmzn. In the following, we define a representation of the polynomial algebra A = C[z1,··· , zm] zp on the Hilbert space j . For p = 1,··· , m, define the operator T j,b on the Hilbert space Hb Hb j by (cid:40) T j,b zp (zn) := zpzn, 0, Lemma 2.2. The above operators {T j,b A on the space Hb j . zi if (n1,··· , np + 1,··· , nm) ∈ Bb j , otherwise. }m i=1 define a bounded representation of the algebra Proof. Let Tzp be the standard Toeplitz representation of zp on the Bergman space L2 and P b operator T j,b a(Bm) onto the closed subspace Hb j . It follows that P b j , and the corresponding basis vector zn. j be the orthogonal projection from L2 j TzpP b For p (cid:54)= p(cid:48), consider n = (n1,··· , nm) ∈ Bb zp can be identified with P b is bounded. j TzpP b j a(Bm), j . Then the INDEX THEOREM AND RESOLUTION 5 ,··· , nm) and If (n1,··· , np + 1,··· , np(cid:48) (n1,··· , np,··· , np(cid:48) + 1,··· , nm) ∈ Bb + 1,··· , nm) belong to Bb (cid:0)T j,b zp (zn)(cid:1) = T j,b j , both (n1,··· , np + 1,··· , np(cid:48) j . Hence, we have (cid:0)T j,b zp(cid:48) (zn)(cid:1) = zpzp(cid:48)zn. zp T j,b zp(cid:48) If (n1,··· , np + 1,··· , np(cid:48) + 1,··· , nm) /∈ Bb either p = js and np = bs for some s or p(cid:48) = js and np(cid:48) we assume that p = js and np = bs for some s. Then (n1,··· , np + 1,··· , np(cid:48) does not belong to the box Bb j . We conclude that j but (n1,··· , np,··· , np(cid:48) ,··· , nm) ∈ Bb j , = bs. Without loss of generality, ,··· , nm) (cid:0)T j,b zp (zn)(cid:1) = T j,b zp (cid:0)T j,b zp(cid:48) (zn)(cid:1) = 0. zp (zn) = 0, and T j,b T j,b zp(cid:48) Summarizing the above discussion, we conclude that T j,b zp(cid:48) T j,b zp = T j,b zp T j,b zp(cid:48) . (cid:3) 2.2. Essentially normal property. In this subsection, we prove the following property of the Hilbert A-module Hb j . Proposition 2.3. The following commutators are compact. ∗ [T j,b zs zt ] ∈ K(Hb j ), ∀s, t = 1,··· , m. , T j,b Therefore, Hb j is an essentially normal Hilbert A-module. Consider the monomial ideal I = (cid:104)zb1+1 (cid:105) of the polynomial algebra A. Then the Hilbert module Hb a(Bm)/I. And Proposition 2.3 follows from [2] and [5]. Here, we outline its proof here for completeness. j can be identified with the quotient module QI := L2 ,··· , zbq+1 j1 jq j be the orthogonal projection from L2 j , it suffices to prove the commutator [P b a(Bm) onto the closed subspace Hb j . As j , Tzs] is compact. Proof. Let P b zs can be written as P b T j,b For n ∈ Bb j , P b P b j Tzs(zn) = j Tzs(zn) is computed as follows, 0, j TzsP b ω0(n1···nm) (cid:40) (cid:113) ω0(n1···(ns+1)···nm) (cid:40) (cid:113) ω0(n1···(ns+1)···nm) −(cid:113) ω0(n1···(bk+1)···nm) ω0(n1···nm) 0,  0, Hence, [P b j , Tzs](zn) is computed as follows, ω0(n1···bk···nm) zn1···(ns+1)···nm, [P b j , Tzs](zn) := zn1···(ns+1)···nm, if (n1 ··· (ns + 1)··· nm) ∈ Bb otherwise. j Similarly, TzsP b j (zn) is computed as follows, TzsP b j (zn) = zn1···(ns+1)···nm, if (n1 ··· ns ··· nm) ∈ Bb otherwise. j if (n1 ··· ns ··· nm) ∈ Bb j , and s = jk, and ns = bk for some k, otherwise. 6 RONALD G. DOUGLAS, MOHAMMAD JABBARI, XIANG TANG, AND GUOLIANG YU We observe that the weight ω0(n1···(bk+1)···nm) ω0(n1···bk···nm) ∞. From this, we can conclude that the commutator [P b converges to zero as (n1,··· , bk,··· , nm) → (cid:3) j , Tzs] is compact. j , Tzs] and therefore also [T j,b zs Remark 2.4. It is not hard to check in the above proof of Proposition 2.3 that the commu- zt ] belong to the Schatten-p ideal for p > m − q. tator [P b 2.3. Geometry of the Hilbert space Hb j . In this subsection, we discuss briefly the geometry of the Hilbert space Hb Let Bj be the subset of Bm cut by the hyperplanes Hji := {zji = 0}, i = 1,··· , q, i.e. j introduced in Section 2.1. , T j,b ∗ Bj := {(z1,··· , zm) ∈ Bm zj1 = ··· = zjq = 0}. Observe that Bj is the unit ball inside the subspace Hj := {(z1,··· , zm)zj1 = ··· = zjq = 0}, which is isomorphic to the standard complex space Cm−q. For i = (i1,··· , iq) ∈ Nq, we consider the following weighted Bergman space L2 a,q+i(Bj). And given any b ∈ Nq, we consider the following Hilbert space a,q+i(Bj). L2 j := (cid:77) (cid:101)Hb j → (cid:101)Hb We define a map Rb j : Hb j by Rb j (f ) := where Ri j(f ) ∈ L2 a,q+i(Bj) is defined as i∈Nq,i1≤b1,··· ,iq≤bq (cid:88) Ri j(f ), i1≤b1,··· ,iq≤bq Ri j(f ) := ∂i1+···+iq f ··· ∂ziq ∂zi1 j1 jq (cid:12)(cid:12)(cid:12)Bj . A straightforward computation on the orthonormal basis gives the following property, and we leave the detail proof to the reader. Proposition 2.5. The map Rb j is an isomorphism of Hilbert spaces. Remark 2.6. Proposition 2.5 suggests that our general construction is a proper general- ization of the Example of ideal (cid:104)z2 1(cid:105) in the Introduction. We consider the trivial vector bundle Eb structure on Eb j C(b1+1)···(bq+1). The Hermitian metric on Eb j := C(b1+1)···(bq+1) × Bj over Bj. The hermitian is defined as follows. We choose the standard basis of {ei}i1≤b1,··· ,iq≤bq of j at z ∈ Bj is It is not hard to see that the Hilbert space (cid:101)Hb (cid:104)ei, ei(cid:48)(cid:105)z = δi,i(cid:48)(1 − z2)q+i. of L2-holomorphic sections of the bundle Eb j can be identified with the Bergman space j . We consider the Toeplitz algebra T(Eb j ) INDEX THEOREM AND RESOLUTION 7 generated by matrix valued Toeplitz operators on the Bergman space of L2-holomorphic j , one can easily identify the Toeplitz algebra Tb sections. Under the isomorphism Rb j , i = 1,··· , m on Hb generated by T j,b j , with the Toeplitz algebra T(Eb zi j ) on (cid:101)Hb j . 3. Resolutions of Monomial Ideals In this section, we present the proof of the main theorem (Theorem 1.1) of this article. In the first three subsections, we generalize the discussion in Section 2 to construct an exact sequence of Hilbert A-modules associated to k boxes in Nm. And in Section 3.4, we apply our construction to prove Theorem 1.1. ,··· ,Bbk 3.1. Construction of the Hilbert C[z1,··· , zm]-modules. Let Bb1 ber of boxes in Nm. We start with the following easy property of boxes in Nm. Lemma 3.1. Intersections of boxes in Nm are again boxes. Proof. Let us consider two boxes Bb1 . Let j12 be the union of j1 and j2. Suppose that there are q12 elements in j12. Then j12 can be viewed as an element of Sq12(m). We 12 ) in Nq12 write j12 as j12 := (j1 by 12 ). Define b12 to be an element b12 := (b1 12,··· , jq12 12,··· , bq12 and Bb2 be k num- jk j2 j1 j1 1, bs(cid:48) 2 ), 12 = js jk jk 12 = js 12 = js(cid:48) jk and Bb2 It is easy to check that the intersection of Bb1 lemma can be proved by induction from the above proof for two boxes. 1 = js(cid:48) 1 /∈ j1 ∩ j2 2 /∈ j1 ∩ j2. is Bb12 bs 1, bs(cid:48) 2 , bk 12 := j12 j1 j2 2 . The general case of the (cid:3)  min(bs For any subset I ⊂ {1,··· , k}, we use BbI jI BbI jI := (cid:92) i∈I Bbi ji . to denote the following intersection, For each box BbI , we consider the corresponding Hilbert A = C[z1,··· , zm] module HbI as is introduced in Section 2. It is not hard to see that subsets of size q in {1,··· , k} are in 1-1 correspondence with elements in Sq(k). jI jI Given Bb1 ,··· ,Bbk , for 1 ≤ q ≤ k, we define a Hilbert module Aq as follows. j1 jk (cid:77) I∈Sq(k) Aq := HbI jI . For convenience, we use A0 to denote the Bergman space L2 a(Bm). We remark that every Hilbert space Aq is equipped with a Hilbert A-module structure from the corresponding A-module structure on each component HbI . It follows from Proposition 2.3 that each Aq is an essentially normal Hilbert A-module. jI 8 RONALD G. DOUGLAS, MOHAMMAD JABBARI, XIANG TANG, AND GUOLIANG YU 3.2. Morphisms. In this subsection, we define the boundary morphism Ψq : Aq → Aq+1 for q = 0,··· , k−1. In the following, we heavily use the expression introduced in Equation (2). To explain our construction, we start with a few examples with a small number k of boxes Bb1 ,··· ,Bbk in Nm. j1 When k = 1, there is only one box Bb jk j . We have two Hilbert modules A0 = L2 j . We define Ψ0 : A0 → A1 as follows. For X ∈ A0 = L2 a(Bm) a(Bm), the map Ψ0 and A1 = Hb maps X to Y := Ψ0(X) ∈ Hb j of the following form, Yn1···nm := Xn1···nm, n ∈ Bb 0, and Bb2 j j2 When k = 2, there are two boxes Bb1 a(Bm), Ψ0(X) is written as Y1 + Y2, where Y1 ∈ Hb1 j1 j1 otherwise. . We denote their intersection by Bb12 . are of and Y2 ∈ Hb2 j12 j2 For X ∈ A0 = L2 the following form, (cid:40) (cid:40) (cid:40) , (cid:40) (Y1)n := For (X1, X2) ∈ Hb1 j1 Xn, n ∈ Bb1 0, ⊕ Hb2 Xn, n ∈ Bb2 otherwise, 0, = A1, define Ψ1(X1, X2) ∈ A2 by (Y2)n := j1 j2 , , otherwise. j2 Ψ1(X1, X2)n := (X1)n − (X2)n, n ∈ Bb12 0, otherwise. j12 , For a general k, in order to define the morphism Ψq : Aq → Aq+1, we introduce the q+1 : Sq+1(k) → Sq(k) for i = 1,··· , q + 1. An element in Sq+1(k) is a q+1(Iq+1) is the subset of {1,··· , k} of following maps f i subset Iq+1 of {1,··· , k} of size q + 1. The map f i size q by dropping the i-th smallest element in Iq+1. Define Ψq : Aq → Aq+1 by for X =(cid:80) Ψq(X) := (cid:88) (cid:40) (cid:80)q+1 Jk∈Sq(k) X Jk with X Jk ∈ HbJk jJk (Y Iq+1)n := Iq+1∈Sq+1(k) Y Iq+1, Y Iq+1 ∈ HbIq+1 jIq+1 , . The function Y Iq+1 ∈ HbIq+1 jIq+1 q+1(Iq+1))n, n ∈ BbIq+1 jIq+1 otherwise. , i=1 (−1)i−1(X f i 0, is defined by Remark 3.2. Similar to the explanation in Section 2.3, the Hilbert module Ai, i = 1,··· , k, can be identified with the Bergman space of L2-holomorphic sections of a her- mitian vector bundle on a disjoint union of subsets of the unit ball Bm. Under this iden- tification, the module morphisms Ψi, i = 0,··· , k − 1, can be realized as restriction maps of jets of holomorphic sections to the subsets. Though we are not using this geometric picture heavily in this paper, we would like to mention it to the reader as we believe that such a geometric picture will play a crucial role in the future study about more general ideals. INDEX THEOREM AND RESOLUTION 9 3.3. Properties of the box resolution. In this subsection, we study properties of the box resolutions. Proposition 3.3. ∀q ≥ 0, the morphism Ψq : Aq → Aq+1 is bounded. Proof. We write X ∈ Aq as a sum (cid:88) Then by the above definition, Ψq(X) =(cid:80) Iq∈Sq(k) X = X Iq , X Iq ∈ HbIq jIq . Y I(cid:48) q+1, where Y I(cid:48) q+1 ∈ HbI(cid:48) jI(cid:48) q+1 q+1 is equal to (Y I(cid:48) q+1)n :=  (cid:88) I(cid:48) q+1 The norm of Ψq(X) is computed as q+12 = Ψq(X)2 = Y I(cid:48) q+1(I(cid:48) q+1))n, n ∈ BbI(cid:48) jI(cid:48) q+1 , q+1 otherwise. Y I(cid:48) n q+1 2 q+1(cid:88) i=1 (−1)i−1(X f i q+1(I(cid:48) q+1))n2 q+1 I(cid:48) 0, (cid:80)q+1 i=1 (−1)i−1(X f i (cid:88) (cid:88) (cid:88) (cid:88) I(cid:48) n∈Bb I(cid:48) I(cid:48) q+1 j q+1 q+1 = I(cid:48) q+1 ≤(cid:88) I(cid:48) q+1 I(cid:48) n∈Bb I(cid:48) j q+1 q+1 (cid:88) I(cid:48) n∈Bb I(cid:48) j q+1 q+1 by the Cauchy-Schwartz inequality q+1))n2 (q + 1)(X f i q+1(I(cid:48) as BbI(cid:48) jI(cid:48) ≤(cid:88) q+1 q+1 ⊆ BbIq jIq (cid:88) I(cid:48) q+1 n∈BbIq jIq (q + 1)(X Iq )n2 as every Iq is contained in at most (k − q) number of I(cid:48) ≤ (k − q)(q + 1) (X Iq )n2 q+1 (cid:88) (cid:88) I∈Sq(k) = (k − q)(q + 1)X2. n∈BbIq jIq (cid:3) Proposition 3.4. The map Ψq : Aq → Aq+1 is an A = C[z1,··· , zm]-module morphism. 10 RONALD G. DOUGLAS, MOHAMMAD JABBARI, XIANG TANG, AND GUOLIANG YU Proof. For every I ∈ Sq(k), for X I ∈ HbI where Y I∪{s} ∈ HbI∪{s} and the function Y I∪{s} has the following form, jI∪{s} , and s is the α-th smallest number in I ∪{s}, and sign(I, s)=α−1, (cid:88) 1≤s≤k,s /∈I jI , Ψq(X I) is a sum (−1)sign(I,s)Y I∪{s}, (cid:40) Y I∪{s} n = n , n ∈ BbI∪{s} X I jI∪{s} , 0, otherwise. jI (cid:113) ω0(n1···(np+1)···nm) (cid:113) ω0(n1···(np+1)···nm) ω0(n1···nm) X I ω0(n1···nm) X I If p ∈ {1,··· , m}, the zp action on HbI is as follows, T jI ,bI zp (X I)n1···(np+1)···nm = From this, we observe that the operator T jI ,bI the zp action on HbI∪{s} jI∪{s} is as follows, zp jI∪{s},bI∪{s} zp T (Y I∪{s})n1···(np+1)···nm 0,  (cid:113) ω0(n1···(np+1)···nm) (cid:113) ω0(n1···(np+1)···nm) ω0(n1···nm) ω0(n1···nm)  = n1···np···nm, p /∈ jI, p (cid:54)= s, Y I∪{s} n1···np···nm, p = jt ∈ jI∪{s}, np + 1 ≤ bt, Y I∪{s} 0, otherwise. n1···np···nm, p /∈ jI, n1···np···nm, p = js ∈ jI, and np + 1 ≤ bs, otherwise. preserves the component HbI jI . Similarly, Using the above definition of Ψq(X I), we can directly check that on each component HbI∪{s} jI∪{s} , (cid:16) Ψq (cid:0)T jI ,bI zp (X I)(cid:1)(cid:17)I∪{s} (cid:16) (cid:0)X I(cid:1)I∪{s}(cid:17) = T jI∪{s},bI∪{s} zp Ψq , (cid:3) which shows that Ψq is compatible with the A-module structure. Proposition 3.5. Im(Ψq−1) ⊆ ker(Ψq). Proof. For every I ∈ Sq−1(k) and any X I ∈ HbI form jI (−1)sign(I,s)Y I∪{s}, , the image of X I under Ψq−1 is of the where Y I∪{s} ∈ HbI∪{s} and the function Y I∪{s} has the following form, jI∪{s} , and s is the α-th smallest number in I ∪{s}, and sign(I, s)=α−1, (cid:88) 1≤s≤k,s /∈I (cid:40) Y I∪{s} n = n , n ∈ BbI∪{s} X I jI∪{s} , 0, otherwise. Similarly, the image of Y I∪{s} under the map Ψq is of the form INDEX THEOREM AND RESOLUTION 11 (−1)sign(I∪{s},t)Z I∪{s,t}, where Z I∪{s,t} ∈ HbI∪{s,t} {s}, t)=β − 1, and the function Z I∪{s,t} has the following form, jI∪{s,t} , and t is the β-th smallest number in I ∪ {s, t}, and sign(I ∪ (cid:88) 1≤t≤k,t /∈I∪{s} (cid:40) Z I∪{s,t} n = Y I∪{s} n 0, , n ∈ BbI∪{s,t} jI∪{s,t} , otherwise. Combining the above computation, we have the following expression for Ψq(Ψq−1(X I)), Ψq(Ψq−1(X I)) = = = = 1≤s≤k,s /∈I 1≤t≤k,t /∈I∪{s} (−1)sign(I,s)Ψq(Y I∪{s}) (−1)sign(I∪{s},t)Z I∪{s,t} (−1)sign(I,s) (cid:88) (−1)sign(I,s)+sign(I∪{s},t) + (−1)sign(I,t)+sign(I∪{t},s)(cid:17) (cid:16) (−1)sign(I,s)+sign(I∪{s},t)Z I∪{s,t} Z I∪{s,t}. 1≤s≤k,s /∈I (cid:88) (cid:88) (cid:88) (cid:88) 1≤s(cid:54)=t≤k,s,t /∈I 1≤s<t≤k,s,t /∈I When s < t, it is not hard to check the following equations sign(I, s) = sign(I ∪ {t}, s), sign(I ∪ {s}, t) = sign(I, t) + 1. And we conclude that Ψq(Ψq−1(X I)) = 0, and complete the proof of this proposition. (cid:3) Proposition 3.6. Im(Ψq−1) ⊇ ker(Ψq), q = 1,··· , k. Proof. We prove the proposition by induction on the number k. For k = 1, we consider the map Ψ0 : A0 → A1. With the orthonormal basis, it is not hard to observe that A1 can be identified with a closed subspace of A0 = L2 a(Bm), and the map Ψ0 is the corresponding orthogonal projection map. Therefore, Ψ0 is a surjective map. Suppose that the following is true Im(Ψq−1) ⊇ ker(Ψq), q = 1,··· , k, for all 1 ≤ k < p. We look at the case that k = p, and separate the proof into two different cases. 12 RONALD G. DOUGLAS, MOHAMMAD JABBARI, XIANG TANG, AND GUOLIANG YU Case I: q = 1. Consider X = (X 1,··· , X p) ∈ A1 such that X ∈ ker Ψ1. Define a function ξ ∈ A0 = L2 a(Bm) as follows. (cid:40) ξn := X s n, 0, there is s such that n ∈ Bbs otherwise. js , We observe that if there are s, t such that n belongs to both Bbs component of Ψ1(X) is js and Bbt jt , then the Hbst jst n − X t X s n = 0, as X ∈ ker(Ψ1). Hence, the value ξn is independent of the choices of s. And therefore ξ is well defined. Furthermore, ξ2 is no more than the sum (3) Hence, ξ ∈ A0 and Ψ0(ξ) = X ∈ ker(Ψ1). Case II: 2 ≤ q ≤ p − 1. We consider the following two collections of p − 1 boxes, X 12 + ··· + X p2. (1) the first p − 1 boxes {Bb1 j1 ,··· ,Bbp−1 jp−1 }. We follow the construction in Section 3.1-3.2 and consider the associated A- modules A1 s+1, s = 1,··· , p − 2. Set A1 s together with the A-module morphisms Ψ1 s p := {0}, and Ψ1 s → A1 p−1 = 0. : A1 (2) the intersections of the first p − 1 boxes with the last one Bbp , jp {Bb1p j1p ,··· ,Bbp−1p jp−1p }. We follow the construction in Section 3.1-3.2 and consider the associated A- modules A2 s+1, s = 1,··· , p − 2. Set A2 s together with the A-module morphisms Ψ2 s s → A2 : A2 p−1 = 0. p := {0}, and Ψ2 By the induction assumption, we know that Im(Ψ1 q−1) ⊇ ker(Ψ1 q), Im(Ψ2 q−1) ⊇ ker(Ψ2 q), q = 1,··· , p − 1. We define a map Φt : A1 t → A2 t by Φt(X I) = Y I∪{p}, I ∈ St(p − 1) where by Y I∪{p} we refer to the component corresponding to the intersection of the boxes Bbi1p ji1p ,··· ,Bbitp jitp with (cid:40) Y I∪{p} n := (−1)tX I 0, n , n ∈ BbI∪{p} jI∪{p} otherwise. Similar to Proposition 3.4, Φt is an A-module morphism. We leave the detail to the reader to check. With the above construction, we can easily check the following identities. INDEX THEOREM AND RESOLUTION 13 (1) Aq = A1 (cid:32) (cid:33) q−1, for q = 2,··· , p, where A1 q ⊕ A2 Ψ1 0 q Φq Ψ2 q−1 p = {0}. , for q = 2,··· , p − 1, where Ψ1 (2) Ψq = We use the above identifications to prove Im(Ψq−1) ⊇ ker(Ψq). The proof consists of p−1 = 0. the following three cases. i) q = 2, ii) 3 ≤ q ≤ p − 2, iii) q = p − 1. i) q = 2. Suppose (X1, X2) ∈ A1 2 ⊕ A2 1 = A2 is in the kernel of the morphism Ψ2. By the above identification of Ψq, we have Ψ1 2(X1) = 0, Φ2(X1) + Ψ2 1(X2) = 0. (0, 0) = Ψ2 By the induction assumption, ker(Ψ1 Ψ1 1(Y1) = X1. By Proposition 3.5 for the morphism Ψ•, we have (cid:0)Ψ1(Y1, 0)(cid:1) = Ψ2 (cid:16) 2) ⊆ Im(Ψ1 1). So there exists Y1 ∈ A1 (cid:16) (cid:0)Ψ1 1(Y1), Φ1(Y1)(cid:1) = (cid:0)Ψ1 1(Y1)(cid:1), Φ2 (cid:0)Ψ1 1(Y1)(cid:1) = 0 (cid:0)Φ1(Y1)(cid:1)(cid:17) (cid:0)Φ1(Y1)(cid:1) = 0. Consider X(cid:48) (cid:0)Ψ1 1(Y1)(cid:1) + Ψ2 0, Φ2(X1) + Ψ2 1 1(Y1) = X1, Ψ1 2 as Ψ1 Ψ1 2 = Hence, Φ2(X1) + Ψ2 1 . 1 2 = X2 − Φ1(Y1). We compute 1(Φ1(Y1)) = Ψ2 1(X2) + Φ2(X1) = 0, 1 such that (cid:0)Φ1(Y1)(cid:1)(cid:17) 1(X(cid:48) Ψ2 2) = Ψ2 1(X2 − Φ1(Y1)) = Ψ2 2(X1), Φ2(X1) + Ψ2 1(X2) − Ψ2 1(X2)) = 0. as Ψ2(X1, X2) = (Ψ1 1(X(cid:48) Using the property that Ψ2 (X(cid:48) 0, (Y2)n := 2 (cid:40) 2) = 0, we construct an element Y2 ∈ Hbp , for some i = 1, ..., p − 1, ip)n, n ∈ Bbip jp jip by setting otherwise. 1(X(cid:48) As Ψ2 2) = 0, the above definition of Y2 is independent of the choices of i. It is not hard to check the norm of Y2 is bounded following the similar estimate as Equation (3). Therefore, Y2 is in Hbp a(Bm), and Ψ2 0(Y2) = X(cid:48) 2. ⊂ L2 jp In summary, we have constructed an element (Y1, Y2) ∈ A1 = A1 . And it is not 1 ⊕ Hbp jp hard to check that Ψ1(Y1, Y2) = (Ψ1 1(Y1), Φ1(Y1) + Ψ2 0(Y2)) = (X1, Φ1(Y1) + X(cid:48) 2) = (X1, X2), which shows that (X1, X2) ∈ Im(Ψ1). ii) 3 ≤ q ≤ p − 2. Suppose (X1, X2) ∈ A1 q ⊕A2 identification of Ψq, we have q−1 = Aq is in the kernel of the morphism Ψq. By the above Ψ1 q(X1) = 0, Φq(X1) + Ψ2 q−1(X2) = 0. 14 RONALD G. DOUGLAS, MOHAMMAD JABBARI, XIANG TANG, AND GUOLIANG YU q−1) ⊇ ker(Ψ1 q), there is Y1 ∈ A1 As Im(Ψ1 0, Φq(X1) + Ψ2 q−1(Φq−1(Y1)) = 0. Hence, we have q−1 such that X1 = Ψ1 q−1(Y1). As Ψq(Ψq−1(Y1, 0)) = As Im(Ψ2 q−2) ⊇ ker(Ψ2 q−1), there exists Y2 ∈ A2 q−1(X2 − Φq−1(Y1)) = 0. Ψ2 q−2 such that q−2(Y2) = X2 − Φq−1(Y1). Ψ2 Therefore, we have found (Y1, Y2) ∈ Aq such that Ψq−1(Y1, Y2) =(cid:0)Ψ1 q−1(Y1), Φq−1(Y1) + Ψq−2(Y2)(cid:1) = (X1, X2). iii) q = p − 1. We notice that Ap is the same as A2 it follows that the map Ψp−1 : Ap−1 = A1 p−1. As the map Ψ2 p−1 ⊕ A2 p−3 : A2 p−2 → Ap = A2 p−2 → A2 p−1 is surjective. p−1 is surjective, (cid:3) 3.4. Proof of Theorem 1.1. In this subsection, we prove the main theorem of the paper. We assume that I is an ideal of the polynomial algebra C[z1,··· , zm] generated by monomials zαi := zα1 m , for αi ∈ Nm, i = 1,··· , l. 1 ··· zαm i i 1 ··· znm 1 ··· znm 1 ··· znm m . Equivalently, zn1 Following [10, Theorem 1.1.2], monomials inside the ideal I form a linear basis of the ideal I over C. We consider the lattice Nm and the subset C(I) consisting of exponents of monomials that do not belong to I. According to [10, Proposition 1.1.5], a monomial v belongs to I if and only if there is a monomial w such that v = wzαi for some i = 1,··· , l. m does not belong to I if and only for any i = 1,··· , l, Therefore, the monomial zn1 zαi is not a factor of zn1 m does not belong to I if and only if for every i, there is si such that nsi < αsi i . Consider the finite collection S(α1,··· , αl) of l-tuple of natural numbers s = (s1,··· , sl) such that 1 ≤ si ≤ m. For each s, let js ⊆ {1,··· , m} be the subset consisting of those numbers appearing in the array (s1,··· , sl). For every k ∈ js, let bk be the minimum of i such that si = k for i = 1,··· , l. From the above conditions on monomials not in all αsi I, we conclude that C(I) is the union of all boxes Bs. We refer the reader to [3, Section 9.2, Theorem 3] for a related discussion. We conclude from the above discussion that a polynomial f belongs to I if and only if f has no nonzero component in any of the boxes Bbs be the collection of nonempty boxes in {Bbs for any s ∈ S(α1,··· , αl). Let Bb1 : s ∈ S(α1,··· , αl)} associated to the ideal I. Associated to the above collection of boxes Bb1 , we apply the constructions in Section 3.1-3.2 to construct a sequence of Hilbert A-modules A0,··· ,Ak together with module morphisms Ψq : Aq → Aq+1, i.e. ,··· ,Bbk ,··· ,Bbk jk jk j1 j1 js js A0 = L2 a(Bm) Ψ0−→ A1 Ψ1−→ ··· Ψk−1−→ Ak −→ 0. INDEX THEOREM AND RESOLUTION 15 Proposition 2.3 shows that each Aq (q = 0, 1, ..., k) is a Hilbert A = C[z1,··· , zm]-module. Proposition 3.3-3.4 show that Ψq (q = 0,··· , k − 1) is a bounded module morphism. Proposition 3.5-3.6 show that the above sequence is exact at Aq for q = 1,··· , k. We are left to prove that the kernel of the morphism Ψ0 is the completion of I in a(Bm), i.e. L2 I = ker(Ψ0). By the above discussion, if f ∈ I, then f has no nonzero component in any of the boxes for any s ∈ S(α1,··· , αl). This shows that f ∈ ker(Ψ0). Therefore, I and its closure Bbs I are contained inside ker(Ψ0). js Suppose that f is in the kernel ker(Ψ0). Write f in terms of the orthonormal basis, As Ψ0(f ) = 0, by the definition of Ψ0, for any s = 1,··· , k, and any n ∈ Bbs , fn = 0. For any positive integer M , let fM be the truncation of the above expansion of f by requiring n1,··· , nm < M , i.e. js f = fnzn. (cid:88) n∈Nm (cid:88) fM := fnzn. n∈Nm,n1<M,··· ,nm<M j1 ,··· ,Bbk . By the construction of the boxes Bb1 It is not hard to see that fM is a polynomial, and has no component in the boxes Bb1 , fM belongs to the ideal I. As M → ∞, fM converges to f in L2 a(Bm). Hence, we have shown that f belongs to the closure I, and ker(Ψ0) is a subset of I. So we conclude that I = ker(Ψ0). ,··· ,Bbk jk jk j1 In summary, we have completed the proof of Theorem 1.1 for general monomial ideals. 3.5. K-homology class. As a corollary of Theorem 1.1, by applying [5, Theorem 1], we a(Bm)/I are both can conclude that the closure I of the ideal I and the quotient QI := L2 essentially normal Hilbert modules. Let T(I) (and T(QI)) be the unital C∗-algebra generated by Teoplitz operators on the module I (and the quotient module QI). We would like to discuss properties of the K-homology class associated to the following Toeplitz extension, I) −→ 0, 0 −→ K −→ T(QI) −→ C(σe I is the essential spectrum space of the algebra T(QI) and K is the algebra of where σe compact operators. By Theorem 1.1, for i = 1,··· , k, we introduce the following closed subspace of Ai, As Ψk−1 is surjective, A− A− k = Ak. i := Im(Ψi−1) = ker(Ψi). 16 RONALD G. DOUGLAS, MOHAMMAD JABBARI, XIANG TANG, AND GUOLIANG YU As Ψi : Ai → Ai+1 is a morphism of A = C[z1,··· , zm]-modules, the kernel A− i = ker(Ψi) is naturally an A-module. Furthermore, we have the following exact sequence of Hilbert A-modules, 0 −→ A− i −→ Ai −→ A− i+1 −→ 0, i = 1,··· , k − 1, where the first map is the inclusion, and the second map is Ψi. Lemma 3.7. ∀i = 1,··· , k, the A-module A− quotient module Qi := Ai/A− Proof. When i = k − 1, as Ψk−1 is surjective, we have the following short exact sequence, is essentially normal, and therefore the is also essentially normal. i i 0 −→ A− k−1 −→ Ak−1 −→ Ak −→ 0. As is explained in Section 3.1, both Ak−1 and Ak are essentially normal A-modules. It follows from [5, Theorem 1] that A− k−1 is an essentially normal A-module. Repeating the above arguments to the exact sequence k−2 −→ Ak−1 −→ A− 0 −→ A− k−1 −→ 0, we conclude that A− ments show that every A− i k−2 is an essentially normal A-module. Similarly, the iterated argu- (cid:3) is an essentially normal A-module. Modeled by the above short exact sequence of essentially normal Hilbert A-modules A− i and Ai, we prove the following property. Proposition 3.8. Let M1, M2, M3 be essentially normal Hilbert A-modules, and W1 : M1 → M2, W2 : M2 → M3 be morphisms of Hilbert A-modules satisfying the following short exact sequence W1−→ M2 0 −→ M1 W2−→ M3 −→ 0. be the closure of the unit ball in Cm, and αi : C(Bm Let Bm i = 1, 2, 3 be the induced representation of C(Bm partial co-isometry operators U : M2 → M1 and V : M2 → M3 such that U U∗ = I = V V ∗, U V ∗ = 0 = V U∗, U∗U + V ∗V = I, ) → C(Mi) := B(Mi)/K(Mi), ) on the Calkin algebra C(Mi). There are and commute with A-module structures up to compact operators, i.e. U α2U∗ = α1, V α2V ∗ = α3. Proof. As W2 is surjective, W2W ∗ 2 is positive definite. Let W2 = A3V be the polar decomposition of the morphism W2 such that A3 is a positive definite operator on M3, s.t. A3 = (W2W ∗ 2 ) 1 2 , and V is a partial coisometry, i.e. V V ∗ = I. As W2 is an A-module morphism, for any f ∈ A = C[z1,··· , zm], we have A3V T 2 f = W2T 2 f = T 3 f W2 = T 3 f A3V, INDEX THEOREM AND RESOLUTION 17 f and T 3 where T 2 f are the Toeplitz operators on M2 and M3 associated to f . As M2 and M3 are essentially normal, T 2 f are both normal in the respective Calkin algebras. It follows from the Fuglede-Putnam theorem that the following equation holds modulo compact operators f and T 3 A3V (T 2 f )∗ = W2(T 2 f )∗ = (T 3 f )∗W2 = (T 3 f )∗A3V. Taking the adjoint of both sides of the above equation, we reach f V ∗A3 = V ∗A3T 3 T 2 f . Multiplying A3V to the left of each term in the above equation, with the property V V ∗ = I, we have A3V T 2 f V ∗A3 = A3V V ∗A3T 3 f = A2 3T 3 f . As we conclude from the above equation that modulo compact operators A3V T 2 f = T 3 f A3V, A3V T 2 f V ∗A3 = T 3 f A3V V ∗A3 = T 3 f A2 3 = A2 3T 3 f . As A3 is positive definite, it is safe to conclude that modulo compact operators f A3 = A3T 3 T 3 f . The above commutativity plus the equation A3V T 2 modulo compact operators, f = T 3 f A3V gives the following identity, The property of V V ∗ = I confirms that modulo compact operators V T 2 f = T 3 f V. which is exactly V T 2 f V ∗ = T 3 f , V α2V ∗ = α3. As W1 : M1 → M2 is injective, W ∗ 1 W1 is positive definite. Let W1 = W A1 be the polar 2 and W : M1 → M2 is a partial decomposition of the operator W1, where A1 = (W ∗ isometry, i.e. W ∗W = I. A similar argument as above for W2 show that modulo compact operators, for any f ∈ A = C[z1,··· , zm], A1T 1 1 W1) 1 f = T 1 f A1, and If we set U = W ∗, then U T 2 f U∗ = T 1 f W = T 1 f . W ∗T 2 f , and U U∗ = I, which shows U α2U∗ = α1. 18 RONALD G. DOUGLAS, MOHAMMAD JABBARI, XIANG TANG, AND GUOLIANG YU As W2W1 = A3V U∗A1 = 0, V U∗ = 0 follows from the invertibility of A1 and A3. Therefore, U∗U and V ∗V are commuting two orthogonal projections on M2. To prove that their sum is the identity operator, it is sufficient to prove that the kernel of their If ξ ∈ M2 satisfies U∗U ξ + V ∗V ξ = 0, then U∗U ξ = V ∗V ξ = 0, and sum is trivial. U ξ = V ξ = 0. Then W2ξ = A3V ξ = 0, and W ∗ 1 ξ = A1U ξ = 0. As W2ξ = 0, ξ belongs to the kernel of W2, and the exactness of the morphisms shows that there is η ∈ W1 such that W1η = ξ. As W ∗ 1 W1η = 0, and ξ = W1η = 0. Hence the kernel of (cid:3) U∗U + V ∗V is trivial, and U∗U + V ∗V = I. 1 ξ = 0, W ∗ Let T(Mi) be the unital C∗-algebra generated by Toeplitz operators on Mi, and σe i be the associated essential spectrum space. The following property is a quick corollary of Proposition 3.8. Corollary 3.9. Under the same assumption as Proposition 3.8, the following equation holds in K1(σe 2), [α2] = [α1] + [α3], where [α1] and [α3] are identified as classes in K1(σe U and V introduced in Proposition 3.8. 2) by the partial coisometry operators Proof. This follows the property for K-homology associated to the three representations 2) → C(M2) with U U∗ = I = V V ∗, U V ∗ = 0 = V U∗, α2, α1 = U α2U∗, α3 = V α2V ∗ : C(σe (cid:3) and U∗U + V ∗V = I. We are now ready to apply Proposition 3.8 and Corollary 3.9 to the exact sequence i )) be the unital C∗-algebra generated i (and σe i−) be the associated essential i−)) constructed in Theorem 1.1. Let T(Ai) (and T(A− by Toeplitz operators on Ai (and A− i ), and σe spectrum space, and αi (and α− into the Calkin algebra C(Ai) (and C(A− i ) be the associated representation of C(σe i ) (and C(σe i )). Theorem 3.10. (Theorem 1.2) The K-homology class associated to [T(QI)] is the same as [α− 1 ∪ ··· ∪ σe k), 1 ], and in K1(σe [T(QI)] = [α1] − [α2] + ··· + (−1)k−1[αk]. Proof. We apply Corollary 3.9 to the following short exact sequence of essentially normal Hilbert A-modules 0 −→ A− i −→ Ai −→ A− i+1 −→ 0, i = 1,··· , k − 1. We have the following equation in K1(σi), [αi] = [α− i ] + [α− i+1]. INDEX THEOREM AND RESOLUTION 19 When i = k − 1, A− k = Ak, and in K1(σe k−1), [αk−1] = [α− k−1] + [αk]. Similarly, for i = k − 2, in K1(σe k−2), [αk−2] = [α− k−2] + [α− k−1]. Combining the above two equations on K-homology groups, we conclude that in K1(σe σe k−2), k−1∪ [αk−1] + [α− k−2] = [αk] + [αk−2], by pushing forward the respective equations in K1(σe k−1, σe K1(σe k−2) via the natural inclusion maps σe k−2. Repeating the above arguments inductively, we conclude that in K1(σe k−1) and K1(σe k−1 ∪ σe k−2 (cid:44)→ σe k−1 ∪ σe 1 ∪ ··· ∪ σe k), k−2) into the ones in 1 ] = [α1] − [α2] + ··· + (−1)k−1[αk]. [α− By the short exact sequence of essentially normal Hilbert A-modules, 0 −→ I −→ L2 a(Bm) −→ A− 1 −→ 0, we conclude that there is a natural A-module isomorphism between the quotient Hilbert modules QI = L2 1 . We conclude from [6, Proposition 4.4] that they must define the same K-homology class. Therefore, we conclude that in K1(σe a(Bm)/I and A− 1 ∪ ··· ∪ σe k), [T(QI)] = [α1] − [α2] + ··· + (−1)k−1[αk]. (cid:3) In this section, we explain our construction of the boxes in examples. 4. Examples 4.1. Ideal I = (cid:104)z2 comprises the region {(n1, n2)n1, n2 ≥ 2}. (2, 2). There are two boxes associated to the ideal, Bb1 {(n1, n2)n2 ≤ 1}. The intersection Bb12 2(cid:105) ⊂ C[z1, z2]. The exponents of monomials in the ideal I = (cid:104)z2 2(cid:105) 1z2 1z2 In this example, there is only one α = := {(n1, n2)n1 ≤ 1}, Bb2 := is of the form {(n1, n2)n1, n2 ≤ 1}. and Bb2 of Bb1 j1 j2 We have following subsets Bj1 := {(0, z2) z2 < 1}, Bj2 := {(z1, 0) z1 < 1}. The Hilbert module A1 is the direct sum of two submodules A1 1 and A2 1, where A1 j12 j1 j2 the closed subspace of L2 a(B2) spanned by {zn L2 1 , zn The Hilbert A-module on A2 is the subspace of L2 is easy to see that A2 = A1 a(B2) spanned by {zn 1 z2}n∈N. 1 ∩ A2 1. 2 , z1zn 2}n∈N, and A2 1 is 1 is the closed subspace of a(B2) spanned by {1, z1, z2, z1z2}. It 20 RONALD G. DOUGLAS, MOHAMMAD JABBARI, XIANG TANG, AND GUOLIANG YU Figure 1. Staircase diagram corresponding to I = (cid:104)zp 1zq 2(cid:105) ⊂ C[z1, z2]. We consider the ideal I = (cid:104)zp 1zq 1zs 2(cid:105). 1zs 4.2. Ideal I = (cid:104)zp 2(cid:105) ⊂ C[z1, z2], 1zq 1zs p, q, r, s ≥ 0. We explain our construction of boxes and the associated Hilbert modules in Theorem 1.1 in this example. 2, zr 2, zr 2, zr The exponents of monomials in I belong to the white region of the Figure (1) marked by I. The subset C(I) ⊂ N2 consisting of exponents of monomials not in the ideal I is the blue region in Figure (1). In this example, α1 = (r, s), α2 = (p, q), S(α1, α2) consists of 4 arrays (1, 1), (1, 2), (2, 1), (2, 2). The boxes associated to these arrays are described below. (1) For s = (1, 1), the box Bb11 (2) For s = (1, 2), the box Bb12 (3) For s = (2, 1), the box Bb21 (4) For s = (2, 2), the box Bb22 j12 j11 j21 is {(n1, n2)n1 < r}; = {(n1, n2)n1 < p, n2 < s}; = {(n1, n2)n1 < r, n2 < q}; = {(n1, n2)n2 < q}. j22 In the Figure (1), the box Bb11 region C, and the box Bb22 Bb12 still works even if we include it. j22 j12 , we do not need to include the box Bb21 j11 is marked as region A, and the box Bb12 is marked as region B. As the box Bb21 is marked as is contained inside in our construction. However, our theorem j12 j21 j21 1 ,A12 1 ,A22 1 . The Hilbert space 2}n∈N. And the Hilbert 2 ,··· , zr−1 zn 2 , z1zn a(B2) spanned by 1 1 The Hilbert space A1 is a direct sum of three spaces A11 is the subspace of L2 A11 a(B2) spanned by {zn space A12 is the finite dimension subspace of L2 ,··· ,··· ··· ,··· a(B2) spanned by {zn And the Hilbert space A22 is the subspace of L2 , z1 , z1z2 ··· , z1zs−1 , zp−1 , zp−1 , zp−1 zs−1 zs−1 , z2, 1 z2 . 2 1 1 2 1 2 1 z2,··· , zn 1 zq−1 2 }n∈N. 1 , zn INDEX THEOREM AND RESOLUTION 21 The Hilbert space A2 is a direct sum of three spaces, A11 The Hilbert space A3 is the subspace of L2 1 , A11 1 ∩A22 1 , and A12 1 ∩A22 1 . 1 ∩A12 a(B2) spanned by , zr−1 , zr−1 , z2, 1 1 , zr−1 1 zq−1 2 . ,··· ,··· ··· ,··· 1 z2 zq−1 2 1, z3 − z2 4.3. Ideal (cid:104)z2 T : C3 → C3 by (z1, z2, z3) (cid:55)→ (u1, u2, u3) = (z1, z2, z3 − z2 transformation T , the ideal I = (cid:104)z2 1, z3 − z2 ball B3 to the domain Ω defined as follows, 2(cid:105) ⊂ C[z1, z2, z3]. We consider the biholomorphic transformation 2). Under the biholomorphic 1, z3(cid:105). T maps the unit 2(cid:105) is changed to I(cid:48) = (cid:104)z2 Ω = {(u1, u2, u3)u12 + u22 + u3 + u2 22 < 1} ⊂ C3. (cid:48) (cid:48) of the ideal I(cid:48) in L2 a(B3) is mapped to the closure I a(B3) is mapped isometrically to L2 Let dVΩ be the normalized Lebesgue measure on Ω, and L2 a(Ω) be the Bergman space on Ω with respect to the measure dVΩ. Under the biholomorphic transformation T , the Bergman space L2 a(Ω). And the closure I of the ideal I in L2 a(Ω). We notice that the a(B3) (and I) transformation T maps the polynomial algebra to itself. As A-modules, L2 and L2 1, z3(cid:105), we apply the similar construction as in Section 3. There is only one box Bb j associated to the ideal I(cid:48), where j = (1, 3) and b = (1, 0), and B = {(n1, n2, 0)n1 ≤ 1}. We consider the subset Ωj = {(z1, z2, z3) ∈ Ωz1 = z3 = 0}. Ωj can be identified with an open subset of C of the form {z2 z22 +z24 < 1}. Let L2 a,s(Ωj) be the Hilbert space of square integrable holomorphic functions on Ωj with respect to the norm a(Ω) (and I ) are isomorphic. a(Ω), for the ideal I(cid:48) = (cid:104)z2 On L2 (1 − z22 − z24)sdVΩj, where dVΩj is the normalized Lebesgue measure on Ωj. Define the Hilbert space A to be the following direct sum of two weighted Bergman spaces, i.e. A = L2 a,2(Ωj) ⊕ L2 a,3(Ωj). We define the A = C[z1, z2, z3]-module structure on AI(cid:48) by the following formulas. For (X, Y ) ∈ L2 a,2(Ωj) ⊕ L2 Tz1(X, Y ) = (0, X), Tz2(X, Y ) = (z2X, z2Y ), Tz3(X, Y ) = (0, 0). a,3(Ωj), We notice that monomials {zi a,3(Ωj). A straightforward computation shows that AI(cid:48) is an essentially normal Hilbert C[z1, z2, z3]- module (See also [8]). 2}i∈N form an orthogonal basis of both L2 a,2(Ωj) and L2 22 RONALD G. DOUGLAS, MOHAMMAD JABBARI, XIANG TANG, AND GUOLIANG YU We define a map ΨI(cid:48) : L2 a,2(Ωj) ⊕ L2 a(Ω) → AI(cid:48) = L2 Ψ(f ) := (fz1=z3=0, a,3(Ωj) by z1=z3=0). ∂f ∂z1 Similar to the argument in the proof of Theorem 1.1 in Section 3.4, we can prove that ΨI(cid:48) is a bounded surjective A-module morphism and its kernel is exactly I . Hence we have the following short exact sequence of essentially normal A-modules, (cid:48) (cid:48) −→ L2 a(Ω) 0 −→ I ΨI(cid:48)−→ AI(cid:48) −→ 0. The biholomorphic transformation T maps W := {(0, z2, z2 2) ∈ B3} isomorphically to Ωj = {(0, z2, 0) ∈ Ω}. We can replace Ωj by W in the corresponding definitions of AI the map ΨI. Using the transformation T , we have a short exact sequence of essentially normal A-modules, 0 −→ I −→ L2 a(B3) ΨI−→ AI −→ 0. As a corollary, we can conclude that the ideal I and associated quotient module QI are both essentially normal. And we have the following geometric description for the corresponding Teoplitz extension, [T(QI)] = [T(cid:0)L2 a,3(Ωj)(cid:1)]. a,2(Ωj) ⊕ L2 References [1] Arveson, W., The Dirac operator of a commuting d-tuple. J. Funct. Anal. 189 (2002), no. 1, 53–79. [2] Arveson, W., p-Summable commutators in dimension d, J. Operator Theory 54 (2005), 101-117. [3] Cox, D., Little, J., O'Shea, D., Ideals, varieties, and algorithms. An introduction to computational algebraic geometry and commutative algebra. Third edition. Undergraduate Texts in Mathematics. Springer, New York, (2007). [4] Douglas, R., A new kind of index theorem. Analysis, geometry and topology of elliptic operators, 369–382, World Sci. Publ., Hackensack, NJ, (2006). [5] Douglas, R., Essentially reductive Hilbert modules. J. Operator Theory. 55 (2006), no. 1, 117–133. [6] Douglas, R., Tang, X., and Yu, G., An analytic Grothendieck Riemann Roch theorem, Adv. Math 294 (2016), 307-331. [7] Douglas, R., Wang, K., A harmonic analysis approach to essential normality of principal submodules. J. Funct. Anal. 261 (2011), no. 11, 3155–3180. [8] Douglas, R, Guo, K., Wang, Y., On the p-essential normality of principal submodules of the Bergman module on strongly pseudoconvex domains, arXiv:1708.04949. [9] Englis, M., Eschmeier, J., Geometric Arveson-Douglas conjecture. Adv. Math. 274 (2015), 606–630. [10] Herzog, J., Hibi, T., Monomial ideals. Graduate Texts in Mathematics, 260. Springer-Verlag London, Ltd., London, (2011). [11] Hironaka, H., Resolution of singularities of an algebraic variety over a field of characteristic zero. I, II. Ann. of Math. (2) 79 (1964), 109–203; (2) 79, (1964) 205–326. INDEX THEOREM AND RESOLUTION 23 Department of Mathematics, Texas A&M University, College Station, TX, 77843 E-mail address: [email protected] Department of Mathematics, Washington University, St. Louis, Missouri, USA, 63130 E-mail address: [email protected] Department of Mathematics, Washington University, St. Louis, Missouri, USA, 63130 E-mail address: [email protected] Department of Mathematics, Texas A&M University, College Station, TX, 77843, Shanghai Center for Mathematical Sciences, Fudan University, Shanghai, China, 200433 E-mail address: [email protected]
1205.1590	2	1205	2013-07-29T16:50:19	Z-stability of crossed products by strongly outer actions II	[ "math.OA" ]	We consider a crossed product of a unital simple separable nuclear stably finite Z-stable C-algebra A by a strongly outer cocycle action of a discrete countable amenable group \Gamma. Under the assumption that A has finitely many extremal tracial states and \Gamma is elementary amenable, we show that the twisted crossed product C-algebra is Z-stable. As an application, we also prove that all strongly outer cocycle actions of the Klein bottle group on Z are cocycle conjugate to each other. This is the first classification result for actions of non-abelian infinite groups on stably finite C*-algebras.	math.OA	math	Z-stability of crossed products by strongly outer actions II Hiroki Matui Yasuhiko Sato Graduate School of Science Graduate School of Science Chiba University Kyoto University Inage-ku, Chiba 263-8522, Japan Sakyo-ku, Kyoto 606-8502, Japan Abstract We consider a crossed product of a unital simple separable nuclear stably finite Z- stable C ∗-algebra A by a strongly outer cocycle action of a discrete countable amenable group Γ. Under the assumption that A has finitely many extremal tracial states and Γ is elementary amenable, we show that the twisted crossed product C ∗-algebra is Z-stable. As an application, we also prove that all strongly outer cocycle actions of the Klein bottle group on Z are cocycle conjugate to each other. This is the first classification result for actions of non-abelian infinite groups on stably finite C ∗- algebras. 1 Introduction This is a continuation of our previous paper [27] studying Z-stability of crossed product C∗-algebras and classification of group actions on Z. Here Z denotes the Jiang-Su algebra introduced by X. Jiang and H. Su in [11], and a unital C∗-algebra A is said to be Z-stable if A is isomorphic to A ⊗ Z. In Elliott's program to classify nuclear C∗-algebras via K- theoretic invariants (see [34] for an introduction to this subject), the Jiang-Su algebra Z In fact, all unital simple separable nuclear has recently become to play a central role. C∗-algebras classified so far by their Elliott invariants are Z-stable. Also, Z-stability is known to be closely related to other regularity properties, such as finite nuclear dimension, strict comparison of positive elements and the property (SI) (see [35, 46, 28]). We refer the reader to [36, 45] for several characterizations of Z. In the first half of the paper, we study Z-stability of crossed products arising from strongly outer cocycle actions of discrete countable groups. Let A be a unital, simple, separable, nuclear, stably finite, Z-stable C∗-algebra with finitely many extremal tracial states. Let (α, u) : Γ y A be a cocycle action of a discrete group Γ (see Definition 2.1). We say that (α, u) is strongly outer when its extension to the weak closure in any tracial representation is outer (see Definition 2.5 for the precise definition). In this setting, we prove that if Γ is elementary amenable and (α, u) is strongly outer, then the twisted crossed product A ⋊(α,u) Γ is Z-stable (Corollary 4.11). In our previous paper [27], we had to make extra hypotheses on the C∗-algebra A and on the group Γ in order to deduce the weak Rohlin property (see Definition 2.5) from strong outerness. The hypothesis on A was necessary to replace central sequences in the sense of von Neumann algebras with central sequences in the sense of C∗-algebras. In this paper, we have removed it by using amenability of nuclear C∗-algebras proved by U. Haagerup [7] (see Proposition 3.5). As 1 for the group Γ, we dealt with only ZN and finite groups in the previous paper. Now we extend our arguments to all elementary amenable groups by showing that they admit outer actions on the hyperfinite II1-factor with a 'nice' Rohlin property (see Definition 2.4 and Proposition 3.3). In order to prove Z-stability of crossed products, we need one more ingredient, namely the property (SI). In [28], we have already shown that Z-stability of A implies the property (SI) (Theorem 2.8). Furthermore, by using the weak Rohlin property of (α, u), we will prove that A satisfies an α-equivariant version of the property (SI) (Proposition 4.5). Together with an exact sequence of central sequence algebras (Theorem 4.3), this allows us to conclude that A ⋊(α,u) Γ is Z-stable. In Section 5, we will also show that if a unital separable nuclear C∗-algebra A is not of type I, then the central sequence algebra has a subquotient which is isomorphic to a II1-factor (Theorem 5.1). This may be thought of as an analogue of Glimm's theorem. In the latter half of the paper, we study cocycle actions of the Klein bottle group on UHF algebras and the Jiang-Su algebra. The Klein bottle group is the group generated by two elements a, b satisfying bab−1 = a−1, and is isomorphic to the fundamental group of the Klein bottle. Classification of group actions is one of the most fundamental subjects in the theory of operator algebras. We briefly review classification results of automor- phisms or group actions on simple stably finite C∗-algebras known so far. For AF and AT algebras, A. Kishimoto [15, 16, 17] showed the Rohlin property for a certain class of automorphisms and obtained a cocycle conjugacy result. The first-named author [24] extended this result to unital simple AH algebras with real rank zero and slow dimension growth. The second-named author [38] proved that strongly outer Z-actions on Z are unique up to strong cocycle conjugacy. T. Katsura and the first-named author [14] gave a complete classification of strongly outer Z2-actions on UHF algebras by using the Rohlin property, and then this result was extended to a certain class of strongly outer Z2-actions on unital simple AF algebras [24]. The uniqueness of strongly outer ZN -actions on UHF algebras of infinite type was obtained in [26]. In our previous paper [27], we proved that strongly outer Z2-actions on Z are unique up to strong cocycle conjugacy. In the present paper, we first show that strongly outer cocycle actions of the Klein bottle group on UHF algebras are unique up to cocycle conjugacy (Theorem 6.16). The proof is along the same lines as the proof of [24, Theorem 6.6], but we need to look at the OrderExt invariant carefully and to check that it always vanishes, unlike the case of Z2-actions. Then we establish a cohomology vanishing type result for cocycle actions of the Klein bottle group on the Jiang-Su algebra Z (Proposition 7.7). The Z-stability theorem proved in the first half of the paper plays an essential role. Finally, by using this cohomology vanishing, we obtain the uniqueness of strongly outer cocycle actions of the Klein bottle group on Z (Theorem 7.12). The uniqueness up to strong cocycle conjugacy of strongly outer actions is also given (Theorem 7.15). These are the first classification results of (cocycle) actions of non-abelian infinite groups on C∗-algebras. 2 2 Preliminaries 2.1 Notations Let A be a C∗-algebra. For a, b ∈ A, we mean by [a, b] the commutator ab − ba. The set of tracial states on A is denoted by T (A). For a ∈ A, we define kak2 = sup τ∈T (A) τ (a∗a)1/2 If A is simple and T (A) is non-empty, then k·k2 is a norm. For τ ∈ T (A), we let (πτ , Hτ ) denote the GNS representation of A associated with τ . For τ ∈ T (A), the dimension function dτ associated with τ is given by dτ (a) = lim n→∞ τ (a1/n), for a positive element a ∈ A. When A is unital, we mean by U (A) the set of all unitaries in A. For u ∈ U (A), the inner automorphism induced by u is written Ad u. An automorphism α ∈ Aut(A) is called outer, when it is not inner. For α ∈ Aut(A), the fixed point subalgebra {a ∈ A α(a) = a} is written Aα. We let T (A)α = {τ ∈ T (A) τ ◦ α = τ}. A single automorphism is often identified with a Z-action generated by it. Let A be a separable C∗-algebra and let ω ∈ βN \ N be a free ultrafilter. Set n→∞kank = 0}, A∞ = ℓ∞(N, A)/c0(A), n→ωkank = 0}, Aω = ℓ∞(N, A)/cω(A). c0(A) = {(an) ∈ ℓ∞(N, A) cω(A) = {(an) ∈ ℓ∞(N, A) lim lim We identify A with the C∗-subalgebra of A∞ (resp. Aω) consisting of equivalence classes of constant sequences. We let A∞ = A∞ ∩ A′, Aω = Aω ∩ A′ and call them the central sequence algebras of A. A sequence (xn)n ∈ ℓ∞(N, A) is called a central sequence if k[a, xn]k → 0 as n → ∞ for all a ∈ A. A central sequence is a representative of an element in A∞. An ω-central sequence is defined in a similar way. For τ ∈ T (A), we define τω ∈ T (Aω) by τω(x) = lim n→ω τ (xn), where (xn)n is a representative sequence of x ∈ Aω. When α is an automorphism of A, we can consider its natural extension on A∞, Aω, A∞ and Aω. We denote it by the same symbol α. We let Z denote the Jiang-Su algebra introduced by X. Jiang and H. Su [11]. They proved that any unital endomorphisms of Z are approximately inner and that Z is isomor- phic to the infinite tensor product of its replicas. In particular, Z is strongly self-absorbing, cf. [43]. When a C∗-algebra A satisfies A ∼= A ⊗ Z, we say that A absorbs Z tensorially, or A is Z-stable. 3 2.2 Group actions and cocycle conjugacy We set up some terminologies for group actions. For a discrete group Γ, the neutral element is denoted by 1 ∈ Γ. Definition 2.1. Let A be a unital C∗-algebra and let Γ be a discrete group. (1) A pair (α, u) of a map α : Γ → Aut(A) and a map u : Γ × Γ → U (A) is called a cocycle action of Γ on A if and αg ◦ αh = Ad u(g, h) ◦ αgh u(g, h)u(gh, k) = αg(u(h, k))u(g, hk) hold for any g, h, k ∈ Γ. We always assume α1 = id, u(g, 1) = u(1, g) = 1 for all g ∈ Γ. We denote the cocycle action by (α, u) : Γ y A. Notice that α gives rise to a (genuine) action of Γ on A∞ and Aω. (2) A cocycle action (α, u) is said to be outer if αg is outer for every g ∈ Γ except for the neutral element. (3) Two cocycle actions (α, u) : Γ y A and (β, v) : Γ y B are said to be cocycle conjugate if there exist a family of unitaries (wg)g∈Γ in B and an isomorphism θ : A → B such that θ ◦ αg ◦ θ−1 = Ad wg ◦ βg and θ(u(g, h)) = wgβg(wh)v(g, h)w∗gh for every g, h ∈ Γ. Furthermore, when there exists a sequence (xn)n of unitaries in B such that xnβg(x∗n) → wg as n → ∞ for every g ∈ Γ, we say that (α, u) and (β, v) are strongly cocycle conjugate. (4) Let α : Γ y A be an action. The fixed point subalgebra {a ∈ A αg(a) = a ∀g ∈ Γ} is written Aα. (5) Let α : Γ y A be an action. A family of unitaries (xg)g∈Γ in A is called an α-cocycle if one has xgαg(xh) = xgh for all g, h ∈ Γ. Definition 2.2 ([33, Definition 2.4]). Let (α, u) : Γ y A be a cocycle action of a discrete group Γ on a unital C∗-algebra A. The twisted crossed product A ⋊(α,u) Γ is the universal C∗-algebra generated by A and a family of unitaries (λα g )g∈Γ satisfying λα g λα for all g, h ∈ Γ and a ∈ A. h = u(g, h)λα gh and λα g aλα∗g = αg(a) If two cocycle actions (α, u) : Γ y A and (β, v) : Γ y B are cocycle conjugate, then A ⋊(α,u) Γ and B ⋊(β,v) Γ are canonically isomorphic. Conversely, if there exists an isomorphism θ : A ⋊(α,u) Γ → B ⋊(β,v) Γ satisfying g )(λβ θ(A) = B and θ(λα g )∗ ∈ B ∀g ∈ Γ, then it is easy to check that (α, u) and (β, v) are cocycle conjugate. When Γ is amenable, A ⋊(α,u) Γ is naturally isomorphic to the reduced twisted crossed product ([33, Theorem 3.11]). 4 2.3 The weak Rohlin property We recall the definitions of amenable groups and of elementary amenable groups (in the sense of M. M. Day [5]). The cardinality of a set F is written F. Definition 2.3. Let Γ be a countable discrete group. (1) For a finite subset F ⊂ Γ and ε > 0, we say that a finite subset K ⊂ Γ is (F, ε)- invariant if K ∩Tg∈F g−1K ≥ (1−ε)K. exists an (F, ε)-invariant finite subset K ⊂ Γ. (2) The group Γ is said to be amenable if for any finite subset F ⊂ Γ and ε > 0 there (3) The class of elementary amenable groups is defined as the smallest family of groups containing all finite groups and all abelian groups, and closed under the processes of taking subgroups, quotients, group extensions and increasing unions. For countable discrete amenable groups, we introduce the property (Q) as follows. Definition 2.4. Let Γ be a countable discrete amenable group and let α : Γ y R be an outer action on the AFD II1-factor R. We say that Γ has the property (Q) if the following holds: For any finite subset F ⊂ Γ and ε > 0, there exists an (F, ε)-invariant finite subset K ⊂ Γ and a sequence of projections (pn)n in R such that 1 −Xg∈K (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) αg(pn)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2 as n → ∞. → 0 and k[x, pn]k2 → 0 ∀x ∈ R The definition above does not depend on the choice of the outer action α : Γ y R, because all outer actions of Γ on R are mutually cocycle conjugate ([4, 13, 30, 23]). In Proposition 3.3, it will be shown that any elementary amenable group has the property (Q). Next we introduce the notions of strong outerness, the weak Rohlin property and the tracial Rohlin property for group actions on C∗-algebras. Definition 2.5 ([27, Definition 2.7]). Let A be a unital simple C∗-algebra with T (A) non-empty and let Γ be a discrete countable amenable group. (1) We say that an automorphism α ∈ Aut(A) is not weakly inner for τ ∈ T (A)α if the weak extension of α to an automorphism of πτ (A)′′ is outer. (2) A cocycle action (α, u) of Γ on A is said to be strongly outer if αg is not weakly inner for any τ ∈ T (A)αg and g ∈ Γ \ {e}. (3) We say that a cocycle action (α, u) : Γ y A has the weak Rohlin property if for any finite subset F ⊂ Γ and ε > 0 there exist an (F, ε)-invariant finite subset K ⊂ Γ and a central sequence (fn)n in A such that 0 ≤ fn ≤ 1, n→∞kαg(fn)αh(fn)k = 0 lim 5 for all g, h ∈ K with g 6= h and lim n→∞ max τ∈T (A)τ (fn) − K−1 = 0. When G = Z, in addition to the conditions above, we further impose the restriction that K is of the form {0, 1, . . . , k} for some k ∈ N. (4) Let A be a unital simple C∗-algebra with tracial rank zero. We say that a cocycle action (α, u) : Γ y A has the tracial Rohlin property if the central sequence (fn)n in the definition above can be chosen as a central sequence consisting of projections. We remark that the definition of the tracial Rohlin property which we have introduced here is slightly stronger than that given in [31, 32] 2.4 The property (SI) We give the definition of the property (SI) which plays an important role in Section 4. The original definition given in [38] is slightly different from this version, but they are actually equivalent (see the discussion following [27, Definition 4.1]). Definition 2.6 ([27, Definition 4.1]). We say that a separable C∗-algebra A has the property (SI) when for any central sequences (xn)n and (yn)n in A satisfying 0 ≤ xn ≤ 1, 0 ≤ yn ≤ 1, lim n→∞ max τ∈T (A) τ (xn) = 0 and inf m∈N lim inf n→∞ min τ∈T (A) τ (ym n ) > 0, there exists a central sequence (sn)n in A such that lim n→∞ks∗nsn − xnk = 0 and lim n→∞kynsn − snk = 0. Remark 2.7. The property (SI) introduced in the definition above can be reformulated in terms of ω-central sequences. Namely, A has the property (SI) if and only if for any ω-central sequences (xn)n and (yn)n of positive contractions in A satisfying lim n→ω there exists an ω-central sequence (sn)n in A such that τ (xn) = 0 and max τ∈T (A) lim n→ω inf m∈N min τ∈T (A) τ (ym n ) > 0, lim n→ωks∗nsn − xnk = 0 and lim n→ωkynsn − snk = 0. Indeed, these two formulations are both equivalent to the following condition: for any finite subset F ⊂ A, ε > 0 and c > 0, there exist a finite subset G ⊂ A, δ > 0 and m ∈ N such that the following holds. If positive contractions x, y ∈ A satisfy k[x, a]k < δ, τ (x) < δ max τ∈T (A) k[y, a]k < δ ∀a ∈ G, and τ (ym) > c, min τ∈T (A) then one can find s ∈ A such that k[s, a]k < ε ∀a ∈ F, ks∗s − xk < ε and kys − sk < ε. 6 In [27, Lemma 4.3], we proved that certain Z-stable C∗-algebras have the property (SI). Recently we have generalized this result and obtained the following ([28]). Theorem 2.8 ([28, Theorem 1.1, Theorem 4.2]). Let A be a unital, simple, separable, nuclear, stably finite, infinite dimensional C∗-algebra. (1) If A is Z-stable, then A has the property (SI). (2) Assume further that A has finitely many extremal tracial states. If A has the property (SI), then A is Z-stable. In Section 4, we will give a slightly different proof of (2) of Theorem 2.8 (Theorem 4.9). 2.5 Approximate representability of cocycle actions In this subsection, we collect notations and terminologies which will be used in Section 6 and Section 7. For a Lipschitz continuous function f , we denote its Lipschitz constant by Lip(f ). Let A be a unital C∗-algebra. The collection of all continuous bounded affine maps from T (A) to R is denoted by Aff(T (A)). The dimension map DA : K0(A) → Aff(T (A)) is defined by DA([p])(τ ) = τ (p). The connected component of the identity in U (A) is denoted by U (A)0. For τ ∈ T (A), the de la Harpe-Skandalis determinant ([8]) associated with τ is written ∆τ : U (A)0 → R/τ (K0(A)). We let log be the standard branch defined on the complement of the negative real axis. If ku−1k < 2, then ∆τ (u) = (2π√−1)−1τ (log u)+τ (K0(A)). We frequently use the following fact: when u, v ∈ U (A) satisfy ku−1k+kv−1k < 2, one has τ (log(uv)) = τ (log u)+τ (log v) for any τ ∈ T (A) (see [8, Lemma 1]). For a homomorphism ϕ between C∗-algebras, K0(ϕ) and K1(ϕ) are the induced homo- morphisms on K-groups. Let A and B be unital C∗-algebras. Two unital homomorphisms ϕ, ψ : A → B are said to be asymptotically unitarily equivalent if there exists a continuous family of unitaries (ut)t∈[0,∞) in B such that ϕ(a) = lim t→∞ Ad ut(ψ(a)) for all a ∈ A. When there exists a sequence of unitaries (un)n∈N in B such that ϕ(a) = lim n→∞ Ad un(ψ(a)) for all a ∈ A, ϕ and ψ are said to be approximately unitarily equivalent. If ϕ and ψ are approximately unitarily equivalent, then Ki(ϕ) = Ki(ψ) for i = 0, 1. An automorphism α ∈ Aut(A) is said to be asymptotically (resp. approximately) inner if α is asymptotically (resp. approximately) unitarily equivalent to the identity map. As usual, we denote by Inn(A) the set of all approximately inner automorphisms of A. Evidently Inn(A) is a normal subgroup of Aut(A). 7 Let ϕ ∈ Aut(A) be an automorphism of a unital C∗-algebra A. The implementing uni- tary in the crossed product C∗-algebra A ⋊ϕ Z is written λϕ. The set of all automorphisms ψ ∈ Aut(A ⋊ϕ Z) satisfying ψ(A) = A and ψ(λϕ)λϕ∗ ∈ A is denoted by AutT(A ⋊ϕ Z) (similar definitions are found in [9, Section 2] and [26, Section 2]). An automorphism ψ ∈ AutT(A ⋊ϕ Z) is said to be T-asymptotically inner if there exists a continuous family of unitaries (ut)t∈[0,∞) in A such that ψ(x) = lim t→∞ Ad ut(x) for all x ∈ A ⋊ϕ Z. In an analogous way, one can define T-approximate innerness. group actions. Next, we recall the notion of approximate (resp. asymptotical) representability of Definition 2.9 ([9, Definition 2.2]). Let Γ be a countable discrete group and let A be a unital C∗-algebra. A cocycle action (α, u) : Γ y A is said to be approximately repre- sentable if there exists a family of unitaries (vg)g∈Γ in A∞ such that vgvh = u(g, h)vgh, αg(vh) = u(g, h)u(ghg−1, g)∗vghg−1 and vgxv∗g = αg(x) hold for all g, h ∈ Γ and x ∈ A. Asymptotical representability is defined in an analogous way. It is routine to check that approximate (resp. asymptotical) representability is pre- served under cocycle conjugacy. The following proposition lists non-trivial examples of strongly outer, asymptotically representable actions on stably finite C∗-algebras. Proposition 2.10. Let α : Γ y A be a strongly outer action. Assume that one of the following conditions holds. (1) Γ = Z, αg is asymptotically inner for every non-trivial g ∈ Z, and A is a unital simple AH algebra with real rank zero, slow dimension growth and finitely many extremal tracial states. (2) Γ = ZN and A is a UHF algebra of infinite type. (3) Γ = Z and A is the Jiang-Su algebra. (4) Γ = Z2 and A is the Jiang-Su algebra. Then α : Γ y A is asymptotically representable. Proof. (1) follows from [24, Theorem 4.8, 4.9], because one can easily construct a strongly outer, asymptotically representable action of Z on A. (2) follows from [26, Theorem 3.4, 4.5]. (3) follows from [37, Theorem 1.2] and [38, Theorem 1.3], because one can easily construct a strongly outer, asymptotically representable action of Z on Z. In a similar fashion, (4) follows from [27, Theorem 7.9]. 8 3 The weak Rohlin property In this section we prove Theorem 3.6 and Theorem 3.7. See Definition 2.4 for the definition of the property (Q). Lemma 3.1. (1) The group of integers Z has the property (Q). (2) Any finite group has the property (Q). (3) If Γ =Sn Γn is an increasing union of countable discrete amenable groups Γn with the property (Q), then Γ also has the property (Q). (4) If 1 → N → Γ → H → 1 is a short exact sequence of countable discrete amenable groups and N, H have the property (Q), then Γ also has the property (Q). (5) Any countable abelian group has the property (Q). Proof. (1) follows from [4] and (2) follows from [13]. (3) is obvious. (5) is an immediate consequence of the other assertions, because any countable abelian group is an increasing union of finitely generated abelian groups. It remains for us to show (4). Let N and H be countable discrete amenable groups with the property (Q) and let 1 → N → Γ → H → 1 be a short exact sequence. Let π : Γ → H be the quotient map and let σ : H → Γ be its right inverse. Let α : Γ y R be an outer action of Γ on the AFD II1-factor R and let β : H y R be an outer action of H on R. Suppose that we are given a finite subset F ⊂ Γ and ε > 0. Without loss of generality, we may assume that there exist finite subsets F1 ⊂ N and F2 ⊂ H such that F = F1 ∪ σ(F2). Since H has the property (Q), we can find an (F2, ε/2)-invariant finite subset K2 ⊂ H and a sequence of projections (qn)n in R satisfying → 0 and k[x, qn]k2 → 0 ∀x ∈ R. 1 − Xh∈K2 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) βh(qn)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2 Define a finite subset F1 ⊂ N by F1 = {σ(h)−1gσ(h) h ∈ K2, g ∈ F1} ∪ {σ(h2h1)−1σ(h2)σ(h1) h1 ∈ K2, h2 ∈ F2}. Since N has the property (Q), we can find an ( F1, ε/2)-invariant finite subset K1 ⊂ N and a sequence of projections (pn)n in R satisfying → 0 and k[x, pn]k2 → 0 ∀x ∈ R. Put K = {σ(h)g ∈ Γ g ∈ K1, h ∈ K2}. For any t ∈ F1, h ∈ K2 and g ∈ K1 ∩ (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 1 − Xg∈K1 αg(pn)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2 Ts∈ F1 Also, for any t ∈ F2, h ∈ K2 ∩Ts∈F2 s−1K1, one has tσ(h)g = σ(h)σ(h)−1tσ(h)g ∈ K. s−1K2 and g ∈ K1 ∩Ts∈ F1 σ(t)σ(h)g = σ(th)σ(th)−1σ(t)σ(h)g ∈ K. s−1K1, one has 9 Hence (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) K ∩\t∈F t−1K(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ≥(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) K2 ∩ \s∈F2 s−1K2(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ·(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) K1 ∩ \s∈ F1 s−1K1(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ≥ (1−ε/2)K2 · (1−ε/2)K1 > (1−ε)K, thus K is (F, ε)-invariant. Define an outer action γ : Γ y R ¯⊗R by γs = αs ⊗ βπ(s) for s ∈ Γ. It is easy to see lim n→∞k[x, pn ⊗ qn]k2 = 0 ∀x ∈ R ¯⊗R and lim n→∞(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 1 −Xs∈K γs(pn ⊗ qn)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2 which completes the proof. = lim = lim n→∞(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 1 − Xh∈K2,g∈K1 n→∞(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 1 − Xh∈K2 γσ(h)(αg(pn) ⊗ qn)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2 1 ⊗ βh(qn)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2 = 0, Remark 3.2. Let N be a countable discrete amenable group with the property (Q) and let 1 → N → Γ → Z → 1 be a short exact sequence. Thus, Γ is isomorphic to a semidirect product N ⋊ Z. Letting a ∈ Z denote a generator, we write Γ = {gai g ∈ N, i ∈ Z}. From the lemma above, Γ has the property (Q). Moreover, its proof tells us that the 'invariant' set in Γ can be chosen of the form {gai g ∈ K, 0 ≤ i ≤ k−1}, where K ⊂ N is an 'invariant' set. More precisely, we get the following. Let α : Γ y R be an outer action of Γ on the hyperfinite II1-factor R. Then, for any finite subset F ⊂ N , ε > 0 and k ∈ N, there exists an (F, ε)-invariant finite subset K ⊂ N and a sequence (pn)n of projections in R such that 1 −Xg∈K k−1Xi=0 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) αgai(pn)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2 → 0 and k[x, pn]k2 → 0 ∀x ∈ R as n → ∞. Proposition 3.3. Every countable discrete elementary amenable group has the property (Q). Proof. This follows from the lemma above and [2, Proposition 2.2 (b)]. In order to state and prove Proposition 3.5 below, we need to recall U. Haagerup's result [7, Theorem 3.1], which says that any nuclear C∗-algebra has a virtual diagonal in the sense of [12]. In particular, any nuclear C∗-algebra is amenable. Here we quote the following theorem from [6], which is an interpretation of [7, Theorem 3.1]. See also [19, Lemma 4.2]. 10 Theorem 3.4 ([6, Theorem 2.1]). Let A be a unital nuclear C∗-algebra. For any finite subset F ⊂ A and ε > 0, there exist w1, w2, . . . , wm ∈ A such that wiw∗i = 1 and mXi=1 mXi=1 awixw∗i − mXi=1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) wixw∗i a(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) < εkxk ∀a ∈ F, x ∈ A. The following is a variant of [39, Lemma 2.1]. We include the proof for completeness. Proposition 3.5. Let A be a unital nuclear C∗-algebra. For any finite subset F ⊂ A and ε > 0, there exist a finite subset G ⊂ A and δ > 0 such that the following hold. If x ∈ A is a positive element such that kxk ≤ 1 and k[x, a]k2 < δ ∀a ∈ G, then there exists a positive element y ∈ A such that kyk ≤ 1, kx − yk2 < ε and k[y, a]k < ε ∀a ∈ F. Proof. By applying the theorem above to F ⊂ A and ε > 0, we get G = {w1, w2, . . . , wm} ⊂ A. Let δ = ε/m. Suppose that x ∈ A is a positive contraction satisfying k[x, wi]k2 < δ for any wi ∈ G. Put y =Pi wixw∗i . Clearly wiw∗i = 1 and kx − yk2 < mδ = ε. Furthermore, for any a ∈ F , 0 ≤ y ≤Xi k[y, a]k =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mXi=1 awixw∗i − mXi=1 wixw∗i a(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) < εkxk ≤ ε. We are now ready to show the equivalence of strong outerness and the weak Rohlin property (see Definition 2.5 for the definitions) for cocycle actions of groups with the property (Q). Theorem 3.6. Let A be a unital, simple, separable, nuclear, stably finite, infinite di- mensional C∗-algebra with finitely many extremal tracial states and let Γ be a countable discrete amenable group with the property (Q). For a cocycle action (α, u) of Γ on A, the following conditions are equivalent. (1) (α, u) is strongly outer. (2) (α, u) has the weak Rohlin property. Proof. The implication (2)⇒(1) was proved in [27, Remark 2.8]. We show the other implication. Suppose that (α, u) : Γ y A is strongly outer. Let E be the set of extremal tracial states on A. For τ ∈ E, it is well-known that πτ (A)′′ is the AFD II1-factor. Set π = Lτ∈E πτ and M = π(A)′′ = Lτ∈E πτ (A)′′. We identify A with π(A) and omit π. The cocycle action (α, u) on A naturally extends to the cocycle action (¯α, u) on M . Note 11 that, for a bounded sequence (xn)n in M , xn converges to zero in the strong operator topology if and only if kxnk2 converges to zero. Suppose that we are given a finite subset F ⊂ Γ and ε > 0. Let R be the AFD II1-factor and let γ : Γ y R be an outer action. Since Γ has the property (Q), there exist an (F, ε)-invariant finite subset K ⊂ Γ and a sequence of projections (pn)n in R such that (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 1 −Xg∈K γg(pn)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2 → 0 and k[x, pn]k2 → 0 ∀x ∈ R as n → ∞. such that First, we would like to show that there exists a sequence of projections (qn)n in M k[x, qn]k2 → 0, k¯αg(qn)¯αh(qn)k2 → 0 and τ (qn) → 1/K for all x ∈ M , g, h ∈ K with g 6= h and τ ∈ T (M ) as n → ∞. The cocycle action ¯α induces an action of Γ on the set of minimal central projections in M . For each Γ-orbit we choose and fix a minimal central projection. Let E0 denote the set of such projections. There exist finite subsets He ⊂ Γ for each e ∈ E0 such that Put Γe = {g ∈ Γ ¯αg(e) = e}. Then Γe is a subgroup of Γ and ¯αh(e) = 1. Xe∈E0 Xh∈He Γ = [h∈He hΓe holds for every e ∈ E0. Let (¯αe, ue) be the restriction of (¯α, u) to Γe and M e, i.e. ¯αe g = ¯αgM e, ue(g, h) = u(g, h)e for g, h ∈ Γe. Since (α, u) is strongly outer, the cocycle action (¯αe, ue) of Γe on M e is outer. Likewise, we let γe denote the restriction of γ : Γ y R to Γe. It follows from [30, Chapter 1] (see also [23, Theorem 2.3]) that (¯αe, ue) is cocycle conjugate to (¯αe ⊗ γe, ue ⊗ 1). Hence, there exist an isomorphism πe : M e ¯⊗R → M e and a family of unitaries (ve g)g∈Γe in M e such that Ad ve g ◦ ¯αg = πe ◦ (¯αe We define an isomorphism π : M ¯⊗R → M by g ⊗ γe g) ◦ π−1 e ∀g ∈ Γe. π((¯αh ⊗ γh)(x)) = ¯αh(πe(x)) ∀x ∈ M e ¯⊗R, h ∈ He. Let qn = π(1 ⊗ pn). We would like to show k¯αg(qn) − π(1 ⊗ γg(pn))k2 → 0 for every g ∈ Γ as n → ∞. Once this is done, it is evident that (qn)n is the desired sequence of projections. We have qn = π(1 ⊗ pn) = Xe∈E0 Xh∈He = Xe∈E0 Xh∈He = Xe∈E0 Xh∈He π(¯αh(e) ⊗ pn) π((¯αh ⊗ γh)(e ⊗ γ−1 ¯αh(πe(e ⊗ γ−1 h (pn))). h (pn))) 12 Take g ∈ Γ. For each h ∈ He there exists k ∈ He such that gh ∈ kΓe, and the map h 7→ k is a bijection. Then one can verify lim n→∞k¯αg(¯αh(πe(e ⊗ γ−1 h (pn)))) − π(¯αk(e) ⊗ γg(pn))k2 = lim h (pn)))) − π(¯αk(e) ⊗ γg(pn))k2 n→∞k¯αk(¯αk−1gh(πe(e ⊗ γ−1 n→∞k¯αk(πe((¯αk−1gh ⊗ γk−1gh)(e ⊗ γ−1 n→∞k¯αk(πe(e ⊗ γk−1g(pn)) − π(¯αk(e) ⊗ γg(pn))k2 = 0. = lim = lim h (pn)))) − π(¯αk(e) ⊗ γg(pn))k2 Therefore we get lim n→∞k¯αg(qn) − π(1 ⊗ γg(pn))k2 = lim = lim n→∞(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) n→∞(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xe∈E0 Xh∈He Xe∈E0 Xk∈He ¯αg(¯αh(πe(e ⊗ γ−1 h (pn)))) − π(1 ⊗ γg(pn))(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2 π(¯αk(e) ⊗ γg(pn)) − π(1 ⊗ γg(pn))(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)2 = 0. By Kaplansky's density theorem, we may replace (qn)n with a sequence (xn)n of pos- itive contractions in A. Thanks to Proposition 3.5, we can further replace (xn)n with a sequence (yn)n of positive contractions in A satisfying k[a, yn]k → 0, kαg(yn)αh(yn)k2 → 0 and τ (yn) → 1/K for all a ∈ A, g, h ∈ K with g 6= h and τ ∈ T (A) as n → ∞. Finally, by the same way as [27, Proposition 3.3], we obtain a sequence (zn)n of positive contractions in A satisfying k[a, zn]k → 0, kαg(zn)αh(zn)k → 0 and τ (zn) → 1/K for all a ∈ A, g, h ∈ K with g 6= h and τ ∈ T (A) as n → ∞. Consequently (α, u) has the weak Rohlin property. For a C∗-algebra A with tracial rank zero, we can prove that strong outerness is equivalent to the tracial Rohlin property in the sense of Definition 2.5 (4). Theorem 3.7. Let A be a unital simple separable C∗-algebra with tracial rank zero and with finitely many extremal tracial states. Let Γ be a countable discrete amenable group with the property (Q). For a cocycle action (α, u) of Γ on A, the following conditions are equivalent. (1) (α, u) is strongly outer. (2) (α, u) has the tracial Rohlin property in the sense of Definition 2.5 (4). Proof. The implication (2)⇒(1) can be shown in a similar fashion to [27, Remark 2.8]. We prove (1)⇒(2). Suppose that (α, u) : Γ y A is strongly outer. Define M and (¯α, u) : Γ y M in the same way as Theorem 3.6. It is well-known that M is isomorphic to a finite direct 13 sum of the hyperfinite II1-factor (see [31, Lemma 2.16], [27, Remark 2.6 (8)]). Exactly the same argument as Theorem 3.6 shows that there exists a sequence of projections (qn)n in M such that k[x, qn]k2 → 0, k¯αg(qn)¯αh(qn)k2 → 0 and τ (qn) → 1/K for all x ∈ M , g, h ∈ K with g 6= h and τ ∈ T (M ) as n → ∞. By [31, Lemma 2.15], we may replace qn with projections en in A, that is, there exists a sequence of projections (en)n in A such that k[a, en]k2 → 0, kαg(en)αh(en)k2 → 0 and τ (en) → 1/K for all a ∈ A, g, h ∈ K with g 6= h and τ ∈ T (A) as n → ∞. Then, from [24, Proposition 4.1], we can find a sequence (fn)n of projections in A such that k[a, fn]k → 0, kαg(fn)αh(fn)k → 0 and τ (fn) → 1/K for all a ∈ A, g, h ∈ K with g 6= h and τ ∈ T (A) as n → ∞. Thus (α, u) has the tracial Rohlin property. 4 Z-stability of crossed products In this section, we prove Theorem 4.9. We also prove that the fixed point subalgebra of the central sequence algebra has strict comparison (Theorem 4.8). 4.1 Fixed point subalgebras of central sequence algebras Throughout this subsection, we let A be a unital, simple, separable, nuclear, stably finite, infinite dimensional C∗-algebra with finitely many extremal tracial states. We also fix a countable discrete amenable group Γ and a cocycle action (α, u) : Γ y A with the weak Rohlin property. Note that we allow Γ to be trivial. Let E be the set of extremal tracial states on A. Set π =Lτ∈E πτ and M = π(A)′′ = Lτ∈E πτ (A)′′. It is well-known that πτ (A)′′ is the AFD II1-factor R for τ ∈ E. We identify A with π(A) and omit π. It is clear that T (A) is identified with T (M ). We let R = (ℓ∞(N, R)/{(xn)n kxnk2 → 0 as n → ω}) ∩ R′, be the central sequence algebra of R in the sense of von Neumann algebra, where R is identified with the subalgebra consisting of equivalence classes of constant sequences. It is also well-known that R is a II1-factor (see [41, Theorem XIV.4.18] for example). Similarly, we define M by M = (ℓ∞(N, M )/{(xn)n kxnk2 → 0 as n → ω}) ∩ M′. Clearly M is isomorphic to Lτ∈E R. In particular, T (M) is canonically identified with T (M ), and hence with T (A). We let T (A)/Γ ∼= T (M )/Γ ∼= T (M)/Γ denote the quotient space by the Γ-action τ 7→ τ ◦ αg. The cocycle action (α, u) : Γ y A gives rise to the cocycle action (¯α, u) : Γ y M . Notice that α : Γ y Aω and ¯α : Γ y M are genuine actions. 14 The inclusion map A ֒→ M induces the homomorphism ρ : Aω → M. Put J = Ker ρ. For an ω-central sequence (xn)n in A, its image in Aω belongs to J if and only if kxnk2 → 0 as n → ω. In other words, for x ∈ Aω, x belongs to J if and only if τω(x∗x) = 0 for any τ ∈ T (A) (see Section 2.1 for the definition of τω). The action α globally preserves J, and so the fixed point subalgebra J α is well-defined. Throughout this subsection, we keep the above notations. The following lemma is folklore among experts. For the reader's convenience, we include a proof here. Lemma 4.1. Let (γ, w) : Γ y R be an outer cocycle action of Γ on the AFD II1-factor R. Then Rγ is a II1-factor. Proof. By [30, Chapter 1] (see also [23, Theorem 2.3]), (γ, w) is cocycle conjugate to (γ ⊗ id, w ⊗ 1) : Γ y R ¯⊗R. Hence there exists a unital homomorphism ϕ : R → Rγ (or one may use [30, Lemma 8.3], which says that Rγ is of type II1). Let τ be the unique trace on R. Suppose that p is a projection belonging to the center of Rγ. Take a projection q ∈ R such that τ (ϕ(q)) = τ (p). There exists a unitary x ∈ R such that xpx∗ = ϕ(q), because R is a II1-factor. Then (xγg(x∗))g is a γ-cocycle in R∩{ϕ(q)}′. By the cohomology vanishing theorem [30, Proposition 7.2], we can find a unitary y ∈ R ∩ {ϕ(q)}′ such that xγg(x∗) = yγg(y∗) for all g ∈ Γ. Therefore y∗xpx∗y = ϕ(q) and y∗x ∈ Rγ. Since p is in the center of Rγ, we obtain p = ϕ(q). This means p = 0 or p = 1, and whence Rγ is a II1-factor. Lemma 4.2. (1) The fixed point algebra M ¯α is a finite direct sum of II1-factors. (2) The map τ 7→ τM ¯α induces an isomorphism from T (M)/Γ to T (M ¯α). Proof. (1) The cocycle action ¯α induces an action of Γ on the set of minimal central projections of M . For each Γ-orbit we choose and fix a minimal central projection. Let {e1, e2, . . . , en} denote the set of such projections. For each i = 1, 2, . . . , n, there exist finite subsets Hi ⊂ Γ such that nXi=1 Xh∈Hi ¯αh(ei) = 1. Put Γi = {g ∈ Γ ¯αg(ei) = ei}. Then Γi is a subgroup of Γ and holds. Let (¯α(i), u(i)) : Γi y M ei be the restriction of (¯α, u) to Γi and M ei. It is not so hard to see hΓi Γ = [h∈Hi ¯αh(x) x ∈ (Mei) ¯α(i) , M ¯α = nXi=1 Xh∈Hi 15 which is isomorphic to nMi=1 (Mei) ¯α(i) . Since (¯α(i), u(i)) : Γi y M ei is an outer cocycle action of a countable discrete amenable group on the AFD II1-factor, by the lemma above, the fixed point algebra (Mei) ¯α(i) is a II1-factor, which completes the proof. (2) immediately follows from the proof of (1). Theorem 4.3. (1) The homomorphism ρ : Aω → M is surjective, that is, 0 −−−−→ J −−−−→ Aω ρ −−−−→ M −−−−→ 0 is exact. (2) The restriction of ρ to (Aω)α induces the following short exact sequence: 0 −−−−→ J α −−−−→ (Aω)α ρ −−−−→ M ¯α −−−−→ 0. Proof. (1) Take a positive contraction x ∈ M. Let (xn)n ∈ ℓ∞(N, M ) be a representative sequence of x consisting of positive contractions. By Kaplansky's density theorem, we may assume that xn is in A. We choose an increasing sequence (Fm)m of finite subsets of A whose union is dense in A. Applying Proposition 3.5 to Fm and 1/m, we get a finite subset Gm ⊂ A and δm > 0. Define Ωm = {n ∈ N n ≥ m, k[xn, a]k2 < 1/δm ∀a ∈ Gm}. Then Ωm belongs to ω because x is in M. For each n ∈ Ωm\Ωm+1, by means of Proposition 3.5, we obtain a positive contraction yn ∈ A such that kxn − ynk2 < 1/m and k[yn, a]k < 1/m ∀a ∈ Fm. Put yn = 0 for n ∈ N \ Ω1. It is easy to see that (yn)n is an ω-central sequence. Let y ∈ Aω be the image of (yn)n. Since kxn − ynk2 → 0 as n → ω, we have ρ(y) = x which completes the proof. It (2) Clearly ρ((Aω)α) is contained in M ¯α and the kernel of ρ(Aω)α equals J α. remains to show the surjectivity of ρ(Aω)α. Take a positive element x ∈ M ¯α. We would like to prove that ρ((Aω)α) contains x. Since ρ is surjective, there exists a ∈ Aω such that ρ(a) = x. Choose a finite subset F ⊂ Γ, ε > 0 and m ∈ N arbitrarily. Choose δ > 0 so that δ(1 + F + ··· + Fm−1) < ε. Since (α, u) has the weak Rohlin property, there exists an (F, δ)-invariant finite subset K ⊂ Γ and a positive contraction f ∈ Aω such that τω(f ) = K−1 and αg(f )αh(f ) = 0 16 for all τ ∈ T (A) and g, h ∈ K with g 6= h. Evidently (ρ(αg(f )))g∈K forms a partition of unity consisting of projections in M. By the standard reindexation trick, we may assume that f commutes with a. Define a function ℓ : K → N by ℓ(g) = min{n ∈ N ∃g1, g2, . . . , gn ∈ F, gg1 . . . gn /∈ K}. One has ℓ−1(1) ≤ δK, because K is (F, δ)-invariant. Then ℓ−1(n) ≤ δFn−1K is obtained inductively, so that ℓ−1({1, 2, . . . , m}) ≤ δ(1 + F + ··· + Fm−1)K < εK. Let h : K → [0, 1] be a function defined by h(g) = m−1 min{ℓ(g)−1, m}. The estimate above implies h−1(1) > (1−ε)K. We define positive contractions b, c ∈ Aω by b =Xg∈K αg(af ) and c =Xg∈K h(g)αg(af ) It is easy to see that kc − αg(c)k is less than 1/m for all g ∈ F . Also, we have ρ(b) =Xg∈K ¯αg(x)¯αg(ρ(f )) = x, and so a − b ∈ J. Furthermore, c ≤ b and τω(b − c) ≤ Xh(g)6=1 τω(αg(af )) ≤ Xh(g)6=1 τω(αg(f )) < ε for all τ ∈ T (A). Hence τω((a− c)∗(a− c)) = τω((b− c)∗(b− c)) < ε holds for all τ ∈ T (A). Since F , ε and m were arbitrary, the standard trick on central sequences implies that there exists a positive contraction c ∈ (Aω)α such that a − c ∈ J, which completes the proof. Lemma 4.4. For any x ∈ J α, there exists a positive contraction e ∈ J α such that ex = xe = x. Proof. Take x ∈ J α. Let (xn)n be a representative sequence of x. We choose an increasing sequence (Fm)m of finite subsets of A whose union is dense in A. Let Γ = {gm m ∈ N}. For m ∈ N, define a continuous function hm on [0,∞) by hm(t) =(mt 0 ≤ t ≤ 1/m 1 1/m ≤ t. Put yn = x∗nxn + xnx∗n. It is easy to see that (yn)n is an ω-central sequence of positive elements satisfying τ (yn) → 0 and αg(yn)− yn → 0 as n → ω for any τ ∈ T (A) and g ∈ Γ. Let Ωm be the set of all natural numbers n ≥ m such that τ (hm(yn)) < 1 m , kαgi(hm(yn)) − hm(yn)k < 1 m and k[hm(yn), a]k < 1 m 17 for all τ ∈ T (A), i = 1, 2, . . . , m and a ∈ Fm. Then Ωm belongs to ω. Set eΩm = Ω1 ∩ Ω2 ∩ ··· ∩ Ωm. Define en ∈ A by en =(hm(yn) n ∈eΩm \eΩm+1 n ∈ N \eΩ1. 0 Then (en)n is an ω-central sequence satisfying τ (en) → 0 and αg(en) − en → 0 as n → ω for all τ ∈ T (A) and g ∈ Γ. Thus, the image e ∈ Aω of (en)n is in J α. Moreover, for every n ∈eΩm \eΩm+1, we have kxn − enxnk2 = k(1 − hm(yn))xnx∗n(1 − hm(yn))k ≤ k(1 − hm(yn))yn(1 − hm(yn))k < m−1. Hence x = ex. Likewise, we obtain x = xe, which completes the proof. In the following proposition, we consider an equivariant version of the property (SI). Notice that we need not assume that A has finitely many extremal tracial states for this proposition. See Definition 2.6 for the definition of the property (SI). Proposition 4.5. Suppose that A has the property (SI). For any ω-central sequences (xn)n and (yn)n of positive contractions in A satisfying lim n→ω max τ∈T (A) τ (xn) = 0, inf m∈N lim n→ω min τ∈T (A) τ (ym n ) > 0, and lim n→ωkαg(xn) − xnk = lim n→ωkαg(yn) − ynk = 0 ∀g ∈ Γ, there exists an ω-central sequence (sn)n in A such that and lim n→ωks∗nsn − xnk = 0, lim n→ωkynsn − snk = 0, lim n→ωkαg(sn) − snk = 0 ∀g ∈ Γ. Proof. The assertion is actually contained in the proof of [27, Theorem 4.7], but we include a proof here for completeness. We first note that, by Remark 2.7, the assertion is equivalent to the property (SI) itself, when Γ is trivial. Let us consider the general case. Let (xn)n and (yn)n be as in the statement. Put ν = inf m∈N lim n→ω min τ∈T (A) τ (ym n ) > 0. Take a finite subset F ⊂ Γ and ε > 0 arbitrarily. Since (α, u) : Γ y A has the weak Rohlin property, we can find an (F, ε)-invariant finite subset K ⊂ Γ and a central sequence (fl)l of positive contractions satisfying lim l→∞ max τ∈T (A)τ (fl) − K−1 = 0 and lim l→∞ αg(fl)αh(fl) = 0 for all g, h ∈ K with g 6= h. 18 We claim that there exists an ω-central sequence (yn)n of positive contractions in A such that yn ≤ yn, inf m∈N lim n→ω min τ∈T (A) τ (ym n ) = νK−1 and lim n→ω αg(yn)αh(yn) = 0 for all g, h ∈ K with g 6= h. To this end, it suffices to show that for any finite subset G ⊂ A, δ > 0 and m ∈ N, there exists a sequence (yn)n of positive contractions in A such that yn ≤ yn, min τ∈T (A) for all g, h ∈ K with g 6= h and lim n→ω τ (ym n ) > νK−1 − 2δ, lim n→ωkαg(yn)αh(yn)k < δ lim n→ωk[yn, a]k < δ ∀a ∈ G. Choose l ∈ N so that max τ∈T (A)τ (fl) − K−1 < δ, kαg(f 1/m l )αh(f 1/m l )k < δ for all g, h ∈ K with g 6= h and Set yn = y1/2 n f 1/m l k[f 1/m l , a]k < δ ∀a ∈ G. y1/2 n . Clearly yn ≤ yn. By means of [27, Lemma 4.6], we get lim n→ω min τ∈T (A) τ (ym n ) = lim n→ω > lim n→ω min τ∈T (A) min τ∈T (A) τ (ym τ (ym n fl) n )K−1 − 2δ ≥ ν − 2δ. We also have lim n→ωkαg(yn)αh(yn)k ≤ kαg(f 1/m l )αh(f 1/m l )k < δ for all g, h ∈ K with g 6= h and lim n→ωk[yn, a]k ≤ k[f 1/m l , a]k < δ for all a ∈ G. Thus, the proof of the claim is completed. sequence (rn)n in A such that Because A has the property (SI) (see also Remark 2.7), there exists an ω-central r∗nrn − xn → 0 and ynrn − rn → 0 as n → ω. Define sn ∈ A by Then it is easy to see sn = 1 pKXg∈K αg(rn). s∗nsn − xn → 0 and ynsn − sn → 0 19 as n → ω. Besides, for any h ∈ F , 1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xg∈K\h−1K + K \ hK1/2 αg(rn)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 1 + Xg∈K\hK (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) αg(rn)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) lim K1/2 K1/2 n→ωksn − αh(sn)k ≤ lim n→ω pK ≤ K \ h−1K1/2 pK ≤ 2√ε. Since F ⊂ Γ and ε > 0 were arbitrary, the proof is completed. Lemma 4.6. Suppose that A has the property (SI). For any τ ∈ T ((Aω)α) and x ∈ J α, one has τ (x) = 0. Proof. Let x ∈ J α be a positive contraction. It suffices to show τ (x) = 0 for all τ ∈ T ((Aω)α). By Lemma 4.4, we can find a positive contraction e1 ∈ J α such that xe1 = x. By Proposition 4.5, there exists s1 ∈ J α such that s∗1s1 = x and (1− e1)s1 = s1. Applying Lemma 4.4 to the positive contraction x + s1s∗1, we obtain a positive contraction e2 ∈ J α such that (x + s1s∗1)e2 = x + s1s∗1. By Proposition 4.5, there exists s2 ∈ J α such that s∗2s2 = x and (1 − e2)s2 = s2. Then x + s1s∗1 + s2s∗2 is again a positive contraction in J α. Repeating this argument, one can find (si)i ⊂ J α satisfying s∗i si = x and x + s1s∗1 + s2s∗2 + ··· + sns∗n ≤ 1 for every n ∈ N. Hence τ (x) = 0 for all τ ∈ T ((Aω)α). Lemma 4.7. Suppose that A has the property (SI). (1) The map τ 7→ τω(Aω)α induces an isomorphism from T (A)/Γ to T ((Aω)α). (2) The map τ 7→ τω is an isomorphism from T (A) to T (Aω). (3) If A has a unique tracial state, then so does (Aω)α. Proof. (1) immediately follows from Lemma 4.2, Theorem 4.3 and Lemma 4.6. (2) and (3) are direct consequences of (1). The following proposition says that (Aω)α has strict comparison in a certain sense. We remark that the condition dτω (a) < dτω (b) for all τ ∈ T (A) is equivalent to the condition dτ (a) < dτ (b) for all τ ∈ T ((Aω)α), thanks to the lemma above. Proposition 4.8. Suppose that A has the property (SI). Let a, b ∈ (Aω)α be positive elements satisfying dτω (a) < dτω (b) for all τ ∈ T (A). Then there exists r ∈ (Aω)α such that rbr∗ = a. Proof. For a subset X ⊂ R, we let 1X be the characteristic function of X. Notice that we keep the identification between T (A), T (M ) and T (M). By τ ◦ ρ = τω, we have dτω (a) = lim n→∞ τω(a1/n) = lim n→∞ τ (ρ(a)1/n) = τ (1(0,∞)(ρ(a))), where the support projection 1(0,∞)(ρ(a)) is well-defined because M ¯α is a von Neumann algebra. Similarly we get dτω (b) = τ (1(0,∞)(ρ(b))). 20 Choose k ∈ N so that dτω (a) < k − 1 k dτω (b) ∀τ ∈ T (A). By Lemma 4.2 (1), there exists a unital homomorphism ϕ : Mk → M ¯α. By [22, Theorem 10.2.1], C0((0, 1]) ⊗ Mk is projective. It follows from Theorem 4.3 that we can find a homomorphism ϕ : C0((0, 1]) ⊗ Mk → (Aω)α such that ρ( ϕ(ι ⊗ x)) = ϕ(x) holds for every x ∈ Mk, where ι ∈ C0((0, 1]) is the identity map. By the usual fast reindexation trick, we may and do assume that the range of ϕ commutes with b. Let p ∈ Mk be a rank one projection. Since ρ(b) commutes with elements of ϕ(Mk), we have τ (1(0,∞)(ρ(b)ϕ(p))) = τ (1(0,∞)(ρ(b))ϕ(p)) = 1 k τ (1(0,∞)(ρ(b))) and τ (1(0,∞)(ρ(b)ϕ(1−p))) = τ (1(0,∞)(ρ(b))ϕ(1−p)) = k−1 k τ (1(0,∞)(ρ(b))) for all τ ∈ T (M). As T (M) has finitely many extremal points, there exists ε > 0 such that (4.1) min τ (1(ε,∞)(ρ(b)ϕ(p))) > 0 τ∈T (M) and τ (1(0,∞)(ρ(a))) < τ (1(ε,∞)(ρ(b)ϕ(1−p))) ∀τ ∈ T (M). (4.2) Define continuous functions g and h on [0,∞) by g(t) = min{ε−1, t−1} and h(t) = tg(t). By Lemma 4.2, M ¯α is a finite direct sum of II1-factors and every trace on M ¯α is the restriction of a trace on M. It follows from (4.2) that we can find a unitary v ∈ M ¯α such that By Theorem 4.3, there exists a unitary w ∈ (Aω)α such that ρ(w) = v. Set 1(0,∞)(ρ(a)) ≤ v1(ε,∞)(ρ(b)ϕ(1−p))v∗. r = a1/2wg(b)1/2 ϕ(ι ⊗ (1−p))1/2. Then and rbr∗ = a1/2wg(b)1/2 ϕ(ι ⊗ (1−p))1/2b ϕ(ι ⊗ (1−p))1/2g(b)1/2w∗a1/2 = a1/2wh(b) ϕ(ι ⊗ (1−p))w∗a1/2 ≤ a ρ(rbr∗) = ρ(a1/2wg(b)1/2 ϕ(ι ⊗ (1−p))1/2b ϕ(ι ⊗ (1−p))1/2g(b)1/2w∗a1/2) = ρ(a1/2)(v1(ε,∞)(ρ(b)ϕ(1−p))v∗)vρ(h(b))ϕ(1−p)v∗ρ(a1/2) = ρ(a1/2)v1(ε,∞)(ρ(b)ϕ(1−p))h(ρ(b)ϕ(1−p))v∗ρ(a1/2) = ρ(a1/2)v1(ε,∞)(ρ(b)ϕ(1−p))v∗ρ(a1/2) = ρ(a). 21 Thus, a − rbr∗ is a positive element of J α. On the other hand, one has τω((h(b) ϕ(ι ⊗ p))m) = τ ((ρ(h(b))ϕ(p))m) = τ (h(ρ(b)ϕ(p))m) ≥ τ (1(ε,∞)(ρ(b)ϕ(p))). This, together with (4.1), implies inf m∈N min τ∈T (M) τω((h(b) ϕ(ι ⊗ p))m) > 0. Therefore, by appealing to Proposition 4.5, we can find s ∈ (Aω)α such that s∗s = a − rbr∗ and h(b) ϕ(ι ⊗ p)s = s. Put Clearly rbt∗ = 0, and so we get t = s∗g(b)1/2 ϕ(ι ⊗ p)1/2. (r + t)b(r + t)∗ = rbr∗ + tbt∗ = rbr∗ + s∗g(b)1/2 ϕ(ι ⊗ p)1/2b ϕ(ι ⊗ p)1/2g(b)1/2s = rbr∗ + s∗h(b) ϕ(ι ⊗ p)s = a. 4.2 Z-stability of crossed products In this subsection, we state direct consequences of the results obtained in Section 3 and Section 4.1. Throughout this subsection, we let A be a unital, simple, separable, nuclear, stably finite, infinite dimensional C∗-algebra with finitely many extremal tracial states. See Definition 2.1 and Definition 2.2 for the definitions of (strong) cocycle conjugacy and twisted crossed products. We note that the following theorem contains Theorem 2.8 (2) as a special case (namely Γ = {1}). Theorem 4.9. Suppose that A has the property (SI). Let (α, u) : Γ y A be a cocycle action of a countable discrete amenable group Γ with the weak Rohlin property. Then there exists a unital homomorphism from Z to (Aω)α, and hence (α, u) : Γ y A is strongly cocycle conjugate to (α⊗id, u⊗1) : Γ y A⊗Z. In particular, the twisted crossed product A⋊(α,u) Γ is Z-stable. Proof. We first prove the existence of a unital homomorphism from Z to (Aω)α. As shown in [11], Z is an inductive limit of the prime dimension drop algebras I(k, k+1)'s. Hence, in the light of [44, Proposition 2.2], it suffices to construct a unital homomorphism from I(k, k+1) to (Aω)α. Let (ei,j)i,j be a system of matrix units for Mk. As in the proof of Proposition 4.8, we can find a homomorphism ϕ : C0((0, 1]) ⊗ Mk → (Aω)α such that 22 τω( ϕ(ιm ⊗ e1,1)) = 1/k for every τ ∈ T (A) and m ∈ N, where ι ∈ C0((0, 1]) is the identity map. Clearly ϕ(ι ⊗ e1,1) ≥ 0 and ϕ(ι ⊗ e1,i) ϕ(ι ⊗ e1,j)∗ =( ϕ(ι ⊗ e1,1)2 0 i = j i 6= j. By Lemma 4.5, there exists s ∈ (Aω)α such that and s∗s = 1 − ϕ(ι2 ⊗ 1) ϕ(ι ⊗ e1,1)s = s. It follows from [38, Proposition 2.1] that there exists a unital homomorphism from I(k, k+1) to (Aω)α. Once we get a unital embedding Z → (Aω)α, strong cocycle conjugacy between (α, u) and (α ⊗ id, u ⊗ 1) follows by essentially the same method as [34, Theorem 7.2.2]. Then we can conclude that the twisted crossed product A ⋊(α,u) Γ is Z-stable. Corollary 4.10. Let Γ be a countable discrete amenable group with the property (Q). Suppose that A has the property (SI). Let (α, u) : Γ y A be a strongly outer cocycle action. Then there exists a unital homomorphism from Z to (Aω)α, and hence (α, u) is strongly cocycle conjugate to (α⊗ id, u⊗ 1) : Γ y A⊗Z. In particular, the twisted crossed product A ⋊(α,u) Γ is Z-stable. Proof. By Theorem 3.6, the cocycle action (α, u) has the weak Rohlin property. Then the conclusion follows from the theorem above. Corollary 4.11. Let Γ be an elementary amenable group. Suppose that A has the property (SI). Let (α, u) : Γ y A be a strongly outer cocycle action. Then there exists a unital homomorphism from Z to (Aω)α, and hence (α, u) is strongly cocycle conjugate to (α ⊗ id, u ⊗ 1) : Γ y A ⊗ Z. In particular, the twisted crossed product A ⋊(α,u) Γ is Z-stable. Proof. By Proposition 3.3, Γ has the property (Q). Then the conclusion follows from the corollary above. Remark 4.12. In the corollaries above, the assumption of the property (SI) can be replaced with Z-stability of A, thanks to Theorem 2.8 (1). 5 A remark on central sequence algebras In this section, we show that for any unital separable nuclear C∗-algebra A which is not of type I, the central sequence algebra Aω has a subquotient which is isomorphic to a II1-factor (Theorem 5.1). This may be regarded as a variant of Glimm's theorem, which says that if a separable C∗-algebra is not of type I, then it has a subquotient isomorphic to a UHF algebra. Let M be a von Neumann algebra with separable predual M∗. We recall the definition of the central sequence algebra of M in the sense of von Neumann algebra from [3, Section II] (see also [41, Chapter XIV §4]). 23 For a normal positive functional ϕ ∈ M∗, we let ϕ =(cid:18) ϕ(x∗x) + ϕ(xx∗) kxk# (cid:19)1/2 . 2 Since x 7→ ϕ(x∗x)1/2 is a seminorm, x 7→ kxk# convex topology on M defined by the family of these seminorms k·k# topology (see [40, Definition II.2.3] for example). A bounded sequence (xn)n in M is called strongly ω-central if ϕ is also a seminorm on M . The locally ϕ is called the ∗-strong lim n→ω sup{ϕ(xny − yxn) y ∈ M, kyk ≤ 1} = 0 for every ϕ ∈ M∗. Let C denote the set of all strongly ω-central sequences. Clearly C is a unital C∗-subalgebra of ℓ∞(N, M ). Let I denote the set of all bounded sequences in M ω-converging to zero in the ∗-strong topology. It is easy to see that I is a norm-closed two-sided ideal of C. We call the quotient M = C/I the central sequence algebra of M . It is well-known that if M is a hyperfinite type II factor then M is a II1-factor (see [41, Theorem XIV.4.18] and [3, Lemma 2.11.(b)]). Theorem 5.1. Let A be a unital separable nuclear C∗-algebra which is not of type I. Then there exists a subquotient of Aω = Aω ∩ A′ which is isomorphic to a II1-factor. Proof. Since A is not of type I, by Glimm's theorem, there exist a type II factor M with separable predual and a unital homomorphism π : A → M such that π(A) is ∗-strongly dense in M . Moreover, M is hyperfinite, because A is assumed to be nuclear. Let C, I and M be as above. We let B ⊂ ℓ∞(N, A) be the set of all ω-central sequences (an)n such that the sequence (π(an))n in M is strongly ω-central. It is routine to check that B is a unital C∗-subalgebra of ℓ∞(N, A) and that B contains cω(A). Let π : ℓ∞(N, A) → ℓ∞(N, M ) be the homomorphism which is naturally induced by π. Then π(B) is contained in C and π(cω(A)) is contained in I. Hence πB induces a unital homomorphism from B/cω(A) ⊂ Aω to M. As mentioned above, M is known to be a II1-factor. Therefore, it remains for us to show that the homomorphism B/cω(A) → M is surjective. The proof is essentially the same as [39, Lemma 2.1], but we include it for completeness. Take a strongly ω-central sequence (xn)n ∈ C arbitrarily. We assume kxnk ≤ 1. Let ψ ∈ M∗ be a normal positive faithful functional. Then k·k# ψ is a norm and the ∗-strong topology agrees with the topology induced by the norm k·k# ψ on any bounded subset of M (see [40, Proposition III.5.3] for example). Hence it suffices to show that there exists an ω-central sequence (an)n in A such that kxn − π(an)k# ψ → 0 as n → ω. By Kaplansky's density theorem (see [40, Theorem II.4.8] for example), we can find (bn)n ∈ ℓ∞(N, A) such that kbnk ≤ 1 and kxn − π(bn)k# ψ → 0 as n → ∞. We choose an increasing sequence (Fm)m of finite subsets of A whose union is dense in A. Since A is nuclear, we can apply Theorem 3.4 to Fm and 1/m, and get a finite subset Gm ⊂ A. Define Ωm = {n ∈ N n ≥ m, k[π(bn), π(w)]π(w∗)k# ψ < m−1Gm−1 ∀w ∈ Gm}. As (xn)n is strongly ω-central, (π(bn))n is also strongly ω-central. Therefore, by [41, Lemma XIV.3.4 (i)], [π(bn), π(w)] tends to zero in the ∗-strong topology as n → ω for 24 every w ∈ A. It follows that Ωm belongs to ω. For each n ∈ Ωm \ Ωm+1, we define an ∈ A by By the choice of Gm, we have k[an, c]k < 1/m for all c ∈ Fm, that is, (an)n is an ω-central sequence. Besides, kxn − π(an)k# wbnw∗. an = Xw∈Gm ψ =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) xn − Xw∈Gm ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) xn − Xw∈Gm = kxn − π(bn)k# # ψ π(wbnw∗)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) π(bn)π(ww∗)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ψ + 1/m, # ψ + 1/m ψ → 0 as n → ω. and so kxn − π(an)k# Remark 5.2. In the theorem above, suppose that M is the hyperfinite II1-factor. Then, by [41, Lemma XIV.3.4 (ii)], we can see that for any ω-central sequence (an)n in A, the sequence (π(an))n in M is strongly ω-central. Thus, B above equals Aω = Aω ∩ A′, and whence there exists a unital homomorphism from Aω to the II1-factor M. 6 Actions of the Klein bottle group I In this section, we would like to prove that all strongly outer cocycle actions of the Klein bottle group on a UHF algebra are mutually cocycle conjugate (Theorem 6.16). We note that some of the ideas and techniques used in this section are borrowed from [10]. 6.1 The OrderExt invariant In this subsection, we would like to collect some basic facts on the OrderExt group for later use. The OrderExt invariant was introduced by A. Kishimoto and A. Kumjian [18] in order to determine when an approximately inner automorphism of a unital simple AT algebra with real rank zero is asymptotically inner. In the later subsections, we use this invariant for classification of certain group actions on AF algebras. We begin with the definition of the OrderExt group ([18, Section 2]). Let G0, G1, F be abelian groups and let D : G0 → F be a homomorphism. When ξ : 0 −−−−→ G0 ι−−−−→ Eξ q −−−−→ G1 −−−−→ 0 is exact, R is in Hom(Eξ, F ) and R ◦ ι = D, the pair (ξ, R) is called an order-extension. Two order-extensions (ξ, R) and (ξ′, R′) are equivalent if there exists an isomorphism θ : Eξ → Eξ ′ such that R = R′ ◦ θ and 0 −−−−→ G0 −−−−→ Eξ −−−−→ G1 −−−−→ 0 ξ : ξ′ : 0 −−−−→ G0 −−−−→ Eξ ′ −−−−→ G1 −−−−→ 0 (cid:13)(cid:13)(cid:13) yθ 25 (cid:13)(cid:13)(cid:13) is commutative. Then OrderExt(G1, G0, D) consists of equivalence classes of all order- extensions. As shown in [18], OrderExt(G1, G0, D) is equipped with an abelian group structure. The map sending (ξ, R) to ξ induces a homomorphism from OrderExt(G1, G0, D) onto Ext(G1, G0). When (ξ, R) is an order-extension, there exists r ∈ Hom(G1, F/ Im D) such that the following diagram is commutative: ι−−−−→ Eξ q −−−−→ G1 −−−−→ 0 yR yr 0 −−−−→ G0 yD 0 −−−−→ Im D −−−−→ F −−−−→ F/ Im D −−−−→ 0, Clearly r depends only on the OrderExt class of (ξ, R) and the map (ξ, R) 7→ r gives rise to a homomorphism. We denote it by ρ : OrderExt(G1, G0, D) → Hom(G1, F/ Im D). Lemma 6.1. Let G0, G1, F be abelian groups and let D : G0 → F be a homomorphism. If D is injective, then the homomorphism ρ : OrderExt(G1, G0, D) → Hom(G1, F/ Im D) is an isomorphism. Proof. Suppose that r ∈ Hom(G1, F/ Im D) is given. Define ξ ∈ Ext(G1, G0) by Eξ = {(g, f ) ∈ G1 × F r(g) = f + Im D}, ι(h) = (0, D(h)), q(g, f ) = g. Note that ι is injective, because D is injective. We define R : Eξ → F by R(g, f ) = f . Then one has ρ([(ξ, R)]) = r. To prove the injectivity of ρ, we assume that an order- extension (ξ, R) satisfies ρ([(ξ, R)]) = 0. Then R(Eξ) is contained in Im D. Since Ker D = 0, by Lemma 2.4 and Proposition 2.5 of [18], we get [(ξ, R)] = 0. Therefore ρ is an isomorphism. We would like to recall the main result of [18]. Let B be a unital simple AT algebra with real rank zero. As described in [18], there exist natural homomorphisms η0 : Inn(B) → OrderExt(K1(B), K0(B), DB) and The following is the main result of [18]. η1 : Inn(B) → Ext(K0(B), K1(B)). Theorem 6.2 ([18, Theorem 4.4]). Suppose that B is a unital simple AT algebra with real rank zero. Then the homomorphism η0 ⊕ η1 : Inn(B) → OrderExt(K1(B), K0(B), DB) ⊕ Ext(K0(B), K1(B)) is surjective and its kernel equals the set of all asymptotically inner automorphisms of B. In the rest of this subsection, we let A be a unital simple AF algebra and let ϕ ∈ Inn(A) be an approximately inner automorphism such that ϕn is not weakly inner for every n ∈ N. By the Pimsner-Voiculescu exact sequence, (K0(A⋊ϕ Z), K1(A⋊ϕ Z)) are naturally identified with (K0(A), K0(A)) (see [24, Section 6]). The space of tracial states T (A ⋊ϕ Z) is also identified with T (A). These identifications induce the natural isomorphism OrderExt(K1(A ⋊ϕ Z), K0(A ⋊ϕ Z), DA⋊ϕZ) ∼= OrderExt(K0(A), K0(A), DA). From now on we will freely identify these two OrderExt groups. 26 Lemma 6.3. For any approximately inner automorphism ψ ∈ AutT(A ⋊ϕ Z), one has η1(ψ) = 0. Proof. This follows from the proof of [24, Lemma 6.3]. From Theorem 6.2 and the lemma above, we obtain the following. Proposition 6.4. Suppose that A ⋊ϕ Z is a unital simple AT algebra with real rank zero. Let ψ ∈ AutT(A ⋊ϕ Z) be an approximately inner automorphism. If η0(ψ) = 0, then ψ is asymptotically inner. For t ∈ R we define ϕt ∈ AutT(A ⋊ϕ Z) by ϕt(x) = x ∀x ∈ A and ϕt(λϕ) = exp(2π√−1t)λϕ. Lemma 6.5. Let ρ : OrderExt(K0(A), K0(A), DA) → Hom(K0(A), Aff(T (A))/ Im DA) be the homomorphism described above. For every t ∈ R and x ∈ K0(A), one has ρ(η0( ϕt))(x) = tDA(x) + Im DA. Proof. Let t ∈ R and let p ∈ A be a projection. It suffices to show the equation for x = [p]. Under the identification between K0(A) and K1(A⋊ϕ Z), [p] corresponds to [λϕp+v(1−p)], where v is a unitary in A satisfying vpv∗ = ϕ(p). Define f : [0, 1] → U (A ⋊ϕ Z) by f (s) = exp(2π√−1st)λϕp + v(1 − p). Then f is a unitary in the mapping torus of ϕt and 1 2π√−1Z 1 0 τ ( f (s)f (s)∗) ds = 1 2π√−1Z 1 0 τ (2π√−1tp) ds = tτ (p) for τ ∈ T (A). Hence ρ(η0( ϕt))([p]) = tDA([p]) + Im DA. We conclude this subsection by the following lemma, which will be used in Section 6.3. Lemma 6.6. Suppose that σ ∈ Aut(A ⋊ϕ Z) satisfies σA ∈ Inn(A) and σ(λϕ)λϕ ∈ A. Let ψ ∈ AutT(A ⋊ϕ Z) be an approximately inner automorphism. Then σ ◦ ψ ◦ σ−1 is in AutT(A ⋊ϕ Z) and η0(σ ◦ ψ ◦ σ−1) = −η0(ψ). Proof. Let B = A ⋊ϕ Z. It is straightforward to check σ ◦ ψ ◦ σ−1 ∈ AutT(B). Let M (B, ψ) = {f ∈ C([0, 1]) ⊗ B ψ(f (0)) = f (1)} and M (B, σ ◦ ψ ◦ σ−1) = {f ∈ C([0, 1]) ⊗ B (σ ◦ ψ ◦ σ−1)(f (0)) = f (1)} be mapping tori. Then we have the following commutative diagram: 0 −−−−→ C0(0, 1) ⊗ B −−−−→ M (B, ψ) −−−−→ B −−−−→ 0 yid⊗σ y 27 0 −−−−→ C0(0, 1) ⊗ B −−−−→ M (B, σ ◦ ψ ◦ σ−1) −−−−→ B −−−−→ 0, yσ ξ : ξ′ : yid yθ y− id where the horizontal sequences are exact. As mentioned above, K∗(B) = K∗(A ⋊ϕ Z) is naturally isomorphic to K0(A). Since σA is approximately inner, we get K0(σ) = id on K0(B). Also, from σ(λϕ)λϕ ∈ A, we obtain K1(σ) = − id on K1(B). Thus, we get the commutative diagram: 0 −−−−→ K0(B) −−−−→ K1(M (B, ψ)) −−−−→ K1(B) −−−−→ 0 0 −−−−→ K0(B) −−−−→ K1(M (B, σ ◦ ψ ◦ σ−1)) −−−−→ K1(B) −−−−→ 0, where the horizontal sequences ξ and ξ′ are exact. Hence ξ′ = −ξ in Ext(K1(B), K0(B)). Moreover, letting R and R′ be the corresponding rotation maps for ψ and σ ◦ ψ ◦ σ−1 respectively, we can easily check R = R′◦θ. Then the conclusion follows from the definition of η0(·). 6.2 Cocycle actions of Z2 on UHF algebras In this subsection, we would like to summarize the results obtained in [24, Section 6] and generalize them to cocycle actions of Z2. Throughout this subsection, we write Z2 = ha, b bab−1 = ai. For a cocycle action (α, u) : Z2 y A, we put so that u = u(b, a)u(a, b)∗, αb ◦ αa = Ad u ◦ αa ◦ αb holds. When A is a UHF algebra or the Jiang-Su algebra, (α, u) is cocycle conjugate to a genuine action if and only if ∆τ (u) = 0 ([27, Theorem 8.6, 8.8]). First, let us recall the invariant of cocycle actions of Z2 taking its values in an OrderExt group ([24, Definition 6.4]). Let (α, u) : Z2 y A be a strongly outer and locally KK-trivial (i.e. K0(αg) = id) cocycle action of Z2 on a unital simple AF algebra A. The automor- phism αb ∈ Aut(A) extends to αb ∈ AutT(A ⋊αa Z) by letting αb(x) = αb(x) for all x ∈ A and αb(λαa ) = uλαa. As mentioned in the previous subsection, by the Pimsner-Voiculescu exact sequence, (K0(A ⋊αa Z), K1(A ⋊αa Z)) are naturally identified with (K0(A), K0(A)). The space of tracial states T (A ⋊αa Z) is also identified with T (A). Under these identifi- cations, we regard η0(αb) as an element in OrderExt(K0(A), K0(A), DA), and define our invariant c(α, u) ∈ OrderExt(K0(A), K0(A), DA) by c(α, u) = η0(αb) ∈ OrderExt(K0(A), K0(A), DA). Remark 6.7. Let ρ : OrderExt(K0(A), K0(A), DA) → Hom(K0(A), Aff(T (A))/ Im D) be the homomorphism described in the previous subsection. Then one has ρ(c(α, u))([1]) = ∆τ (u). Indeed, under the identification between K0(A) and K1(A ⋊αa Z), [1] corresponds to [λαa ]. Let z : [0, 1] → U (A) be a piecewise smooth path such that z(0) = 1 and z(1) = u. Then f (t) = z(t)λαa is a unitary in the mapping torus of αb. Therefore ρ(c(α, u))([1]) = ∆τ (u). Lemma 6.8. Let (α, u), (β, v) : Z2 y A be strongly outer and locally KK-trivial cocycle actions of Z2 on a unital simple AF algebra A. If (α, u) and (β, v) are cocycle conjugate via an approximately inner automorphism θ ∈ Inn(A), then c(α, u) = c(β, v). 28 Proof. There exists a family of unitaries (wg)g∈Z2 in A such that θ ◦ αg ◦ θ−1 = Ad wg ◦ βg and θ(u(g, h)) = wgβg(wh)v(g, h)w∗gh hold for every g, h ∈ Z2. Then θ extends to an isomorphism from A ⋊αa Z to A ⋊βa Z by letting θ(λαa) = waλβa. As in [24, Section 6], we let M (A ⋊αa Z, αb) and M (A ⋊βa Z, βb) denote the mapping tori. Take a continuous path of unitaries z : [0, 1] → U (A) such that z(0) = 1 and z(1) = wb. Define an isomorphism θ : M (A ⋊αa Z, αb) → M (A ⋊βa Z, βb) by θ(f )(t) = z(t)∗θ(f (t))z(t). Since for any x ∈ A and w∗b θ(αb(x))wb = (Ad w∗b ◦ θ ◦ αb)(x) = βb(θ(x)) w∗b θ(αb(λαa))wb = w∗b θ(uλαa)wb = w∗b θ(u)waλβawb = w∗b wbβb(wa)vβa(w∗b )w∗awaλβawb = βb(wa)vλβa = βb(waλβa) = βb(θ(λαa )), θ is well-defined. Hence we obtain the following commutative diagram: 0 −−−−→ C0(0, 1) ⊗ (A ⋊αa Z) −−−−→ M (A ⋊αa Z, αb) −−−−→ A ⋊αa Z −−−−→ 0 0 −−−−→ C0(0, 1) ⊗ (A ⋊βa Z) −−−−→ M (A ⋊βa Z, βb) −−−−→ A ⋊βa Z −−−−→ 0. From this it is not so hard to see η0(αb) = η0( βb), because K0(θ) = id. Therefore we get c(α, u) = c(β, v). In the same way as [24, Theorem 6.6], we can prove the following. Theorem 6.9. Let (α, u), (β, v) : Z2 y A be strongly outer, locally KK-trivial cocycle actions of Z2 on a unital simple AF algebra A with finitely many extremal tracial states. The following are equivalent. (1) c(α, u) = c(β, v) in OrderExt(K0(A), K0(A), DA). (2) (α, u) and (β, v) are cocycle conjugate via an approximately inner automorphism θ ∈ Aut(A). 29 yθ yθ yθ Proof. We give only sketchy arguments as the proof of [24, Theorem 6.6] applies almost verbatim. (2)⇒(1) was shown in the lemma above without assuming that A has finitely many extremal traces. Let us consider the other implication (1)⇒(2). It is well-known (see [24, Section 4] for example) that A ⋊αa Z and A ⋊βa Z are unital simple AT algebras with real rank zero. By [24, Theorem 4.8, 4.9], we may assume that there exists w ∈ U (A) such that βa = Ad w ◦ αa. Define an isomorphism θ : A ⋊βa Z → A ⋊αa Z by θ(x) = x for x ∈ A and θ(λβa) = wλαa . We can observe η0(αb) = c(α, u) = c(β, v) = η0( βb) = η0(θ ◦ βb ◦ θ−1). It follows from Proposition 6.4 that αb and θ ◦ βb ◦ θ−1 are asymptotically unitarily equiv- alent. Then [24, Theorem 6.1] applies and yields an approximately inner automorphism µ ∈ AutT(A ⋊αa Z) and t ∈ U (A) such that µA is in Inn(A) and µ ◦ θ ◦ βb ◦ θ−1 ◦ µ−1 = Ad t ◦ αb. (Actually, in [24, Theorem 6.1], A is assumed to have a unique trace. We can weaken this hypothesis by using Theorem 3.7 instead of [24, Lemma 5.1]. See also Theorem 6.14.) Let s = µ(λαa)(λαa )∗ ∈ U (A). One can deduce and Moreover, we can verify (µA) ◦ βb ◦ (µA)−1 = Ad t ◦ αb (µA) ◦ βa ◦ (µA)−1 = (Ad µ(w)s) ◦ αa. tαb(µ(w)s)uαa(t∗)s∗µ(w∗) = µ(v), which implies that (α, u) and (β, v) are cocycle conjugate. Let us consider the case that A is a UHF algebra. As in [14, Section 5], for every prime number p, we define ζA(p) = sup{k ∈ N ∪ {0} [1] is divisible by pk in K0(A)} ∈ {0, 1, 2, . . . ,∞}. In [14], it was shown that there exists a bijective correspondence between the set of (strong) cocycle conjugacy classes of strongly outer Z2-actions on A and the abelian group Y1≤ζA(p)<∞ Z/pζA(p)Z. As mentioned in [24, Remark 6.7], it is easy to see that this group is isomorphic to {r ∈ Hom(K0(A), R/ Im DA) r([1]) = 0}. It is also known that this group is isomorphic to the fundamental group π1(Aut(A)) ([42]). The following theorem is a strengthened version of the main result of [14] (see also [27, Theorem 8.6]). 30 Theorem 6.10. Assume that A is a UHF algebra. There exist bijective correspondences between the following three sets. (1) Cocycle conjugacy classes of strongly outer cocycle actions of Z2 on A. (2) OrderExt(K0(A), K0(A), DA). (3) Hom(K0(A), R/K0(A)). The correspondences are given by (α, u) 7→ c(α, u) and (α, u) 7→ ρ(c(α, u)). Furthermore, a strongly outer cocycle action (α, u) : Z2 y A is cocycle conjugate to a genuine action if and only if ρ(c(α, u))([1]) = 0. Proof. Since DA : K0(A) → Aff(T (A)) ∼= R is injective, by Lemma 6.1, the homomorphism ρ : OrderExt(K0(A), K0(A), DA) → Hom(K0(A), R/K0(A)) is an isomorphism. By virtue of the theorem above, the correspondence from (1) to (2) is injective. Take r ∈ Hom(K0(A), R/ Im DA). Let t ∈ R be such that r([1]) = t + Im DA. Define r′ ∈ Hom(K0(A), R/ Im DA) by r′(x) = r(x) − tDA(x) + Im DA so that r′([1]) = 0. By [14, Lemma 5.3] and the observation above, there exists a strongly outer action γ : Z2 y A such that ρ(c(γ, 1)) = r′. By Lemma 6.5, we have ρ(η0((bγa)t ◦ γb))(x) = tDA(x) + r′(x) = r(x). Take a scalar valued 2-cocycle u : Z2 × Z2 → T ⊂ U (A) such that u = exp(2π√−1t) (note that the 2-cohomology group H 2(Z2, T) is isomorphic to T). Then ρ(c(γ, u)) = r. Thus the correspondence from (1) to (2) is surjective. Let (α, u) : Z2 y A be a strongly outer cocycle action. Evidently, if (α, u) is cocycle conjugate to a genuine action, then ρ(c(α, u))([1]) = 0 thanks to the theorem above and Remark 6.7. If ρ(c(α, u))([1]) = 0, then by [14, Lemma 5.3] and the observation above, there exists a strongly outer action γ : Z2 y A such that c(γ, 1) = c(α, u). By using the theorem above again, we can conclude that (α, u) is cocycle conjugate to the genuine action γ. We will use the following lemma in the next subsection. Lemma 6.11. Let A be a UHF algebra and let ϕ be an automorphism of A such that ϕn is not weakly inner for every n ∈ N. For any c ∈ OrderExt(K0(A), K0(A), DA), there exists ψ ∈ AutT(A ⋊ϕ Z) such that η0(ψ) = c. Proof. From the theorem above, there exists a strongly outer cocycle action (α, u) : Z2 y A such that c(α, u) = c. By [15, Theorem 1.3], we may assume that there exists w ∈ U (A) such that αa = Ad w ◦ ϕ. Define an isomorphism θ : A ⋊αa Z → A ⋊ϕ Z by θ(x) = x for x ∈ A and θ(λαa) = wλϕ. Then θ ◦ αb ◦ θ−1 is in AutT(A ⋊ϕ Z) and η0(θ ◦ αb ◦ θ−1) = η0(αb) = c(α, u) = c. 31 6.3 Actions of the Klein bottle group on UHF algebras Throughout this subsection, we let Γ = ha, b bab−1 = a−1i. The group Γ is isomorphic to the fundamental group of the Klein bottle and is called the Klein bottle group. In this subsection, we will show that all strongly outer cocycle actions of Γ on a UHF algebra are cocycle conjugate to each other. First, we would like to show the existence of strongly outer, approximately repre- sentable Γ-actions on the Jiang-Su algebra Z. See Definition 2.9 for the definition of approximate representability. Lemma 6.12. There exists a homomorphism π : Γ → U (C([0, 1]) ⊗ M2) such that π(g) is not a scalar multiple of 1 for any g ∈ Γ \ {1} and π(g)(0) = 1 for all g ∈ Γ. Proof. Put Let v : [0, 1] → M2 be a continuous path of unitaries such that 0 u(t) =(cid:20)exp π√−1t v(0) =(cid:20)1 0 0 1(cid:21) , 0 exp(−π√−1t)(cid:21) . v(1) =(cid:20)0 1 1 0(cid:21) . One can define a homomorphism π : Γ → U (C([0, 1]) ⊗ M2) by π(a)(t) =(1 u(t − 1/2) t ≤ 1/2 1/2 ≤ t and π(b)(t) =(v(2t) v(1) t ≤ 1/2 1/2 ≤ t. Clearly π(g) is not a scalar multiple of 1 for any g ∈ Γ\{1} and π(a)(0) = π(b)(0) = 1. Proposition 6.13. There exists a strongly outer, approximately representable action α : Γ y Z. In particular, for any unital Z-stable C∗-algebra A, there exists a strongly outer, approximately representable action α : Γ y A. Proof. Since the C∗-algebra {f ∈ C([0, 1]) ⊗ M2 f (0) ∈ C} is embeddable to Z, by the lemma above, there exists a homomorphism π : Γ → U (Z) such that Ad π(g) 6= id for all g ∈ Γ \ {1}. Define an action α of Γ on the infinite tensor product of Z by αg =Ok∈N Ad π(g). It is clear that α is strongly outer and approximately representable. For any unital C∗- algebra A, the action id⊗α : Γ y A⊗Z is strongly outer and approximately representable. The following is a Rohlin type theorem for Γ-actions on unital simple AF algebras. The proof is almost the same as [29, Theorem 3] and [24, Theorem 5.5]. 32 Theorem 6.14. Let A be a unital simple AF algebra with finitely many extremal tracial states and let (α, u) : Γ y A be a strongly outer cocycle action. Assume further that αr a and αs b are in Inn(A) for some r, s ∈ N. Then for any m ∈ N, there exist projections e, f ∈ (A∞)αa such that (f ) = f b (e) = e, αm+1 αm b and αi b(e) + m−1Xi=0 mXj=0 αj b(f ) = 1. Proof. For any ε > 0, by Proposition 3.3 and Theorem 3.7, there exists an ({a, b}, ε)- invariant finite subset K ⊂ Γ and a central sequence (en)n of projections in A such that lim n→∞kαg(fn)αh(fn)k = 0 for all g, h ∈ K with g 6= h and lim n→∞ max τ∈T (A)τ (fn) − K−1 = 0. In addition, by the proof of Lemma 3.1 (4), we may assume that K is of the form {bjai ∈ Γ 0 ≤ i ≤ m1 − 1, 0 ≤ j ≤ m2 − 1} for some m1, m2 ∈ N. Since αr construct a central sequence of projections (f′n)n in A such that a is approximately inner, when m1 is large enough, we can lim n→∞kf′n(fn + αa(fn) + ··· + αm1−1 αa(f′n) ≈ f′n (fn)) − f′nk = 0, and τ (f′n) ≈ m1τ (fn) ∀τ ∈ T (A), a by using the arguments in [15, Lemma 3.1] (see also [29, Lemma 6]). Consequently we obtain the following: for any m ∈ N, there exists a central sequence of projections (en)n in A such that lim n→∞ max τ∈T (A)τ (en) − m−1 = 0, lim n→∞ken − αa(en)k = 0 and lim n→∞kenαj b(en)k = 0 for all j = 1, 2, . . . , m−1. By using this instead of [24, Lemma 5.2], one can complete the proof in exactly the same way as that of [24, Theorem 5.5]. Next we would like to establish a Γ-action version of [24, Theorem 6.1] (or [9, Theorem 4.11]). We note that the idea of the proof is taken from [10]. Theorem 6.15. Let A be a unital simple AF algebra with finitely many extremal tracial states and let ϕ ∈ Inn(A). For i = 1, 2, assume that ψi ∈ Aut(A ⋊ϕ Z) satisfies the following conditions. (1) ψi(A) = A and ψi(λϕ)λϕ ∈ A. 33 i A is approximately inner for some r ∈ N. (2) ϕn ◦ (ψiA)m ∈ Aut(A) is not weakly inner for each (n, m) ∈ Z2 \ {(0, 0)}. (3) ψr Assume further that ψ1 ◦ ψ−1 is asymptotically inner. Then there exists an approximately inner automorphism µ ∈ AutT(A ⋊ϕ Z) and a unitary v ∈ U (A) such that µA is also approximately inner and 2 µ ◦ ψ1 ◦ µ−1 = Ad v ◦ ψ2. 2 Proof. We would like to apply the Evans-Kishimoto intertwining argument to ψ1 and ψ2. It is easy to see that ψ1 ◦ ψ−1 is in AutT(A ⋊ϕ Z). By Proposition 2.10 (1), (the Z-action generated by) ϕ is asymptotically representable. Then [9, Theorem 4.8] implies that ψ1◦ψ−1 is T-asymptotically inner. For each i = 1, 2, the cocycle action of Γ generated by ϕ and ψiA is strongly outer. It follows from the theorem above that we can find Rohlin projections for ψi in the fixed point algebra (A∞)ϕ. Hence, with the help of [24, Lemma 4.10] (or [26, Lemma 3.3]), the (equivariant version of) Evans-Kishimoto intertwining argument shows the statement. 2 We are now ready to prove the uniqueness of strongly outer cocycle actions of the Klein bottle group on UHF algebras. Theorem 6.16. Let (α, u) and (β, v) be strongly outer cocycle actions of Γ on a UHF algebra A. Then (α, u) and (β, v) are cocycle conjugate. Proof. Put u = u(b, a)u(a−1, b)∗u(a−1, a) and v = v(b, a)v(a−1, b)∗v(a−1, a). Because u(λαa )∗αb(x)λαa u∗ = Ad u(α−1 a (αb(x)) = αb(αa(x)) for x ∈ A, the automorphism αb extends to αb ∈ Aut(A ⋊αa Z) by letting αb(λαa ) = u(λαa )∗. Likewise, we can extend βb to βb ∈ Aut(A ⋊βa Z). By [15, Theorem 1.3], we may assume that there exists w ∈ U (A) such that βa = Ad w ◦ αa. Define an isomorphism θ : A ⋊βa Z → A ⋊αa Z by θ(x) = x for x ∈ A and θ(λβa) = wλαa . Then θ ◦ βb ◦ θ−1 ◦ α−1 is in AutT(A ⋊αa Z). Since b OrderExt(K0(A), K0(A), DA) ∼= Hom(K0(A), R/ Im DA) is divisible, there exists c ∈ OrderExt(K0(A), K0(A), DA) such that −2c = η0(θ ◦ βb ◦ θ−1 ◦ α−1 b ). By Lemma 6.11, there exists ψ ∈ AutT(A ⋊αa Z) such that η0(ψ) = c. Then η0(ψ ◦ θ ◦ βb ◦ θ−1 ◦ ψ−1 ◦ α−1 b ) = η0(ψ ◦ θ ◦ βb ◦ θ−1 ◦ ψ−1 ◦ θ ◦ β−1 = η0(ψ) + η0(θ ◦ βb ◦ θ−1 ◦ ψ−1 ◦ θ ◦ β−1 = c + c + η0(θ ◦ βb ◦ θ−1 ◦ α−1 b ) = 0, b ◦ θ−1 ◦ θ ◦ βb ◦ θ−1 ◦ α−1 b ) b ◦ θ−1) + η0(θ ◦ βb ◦ θ−1 ◦ α−1 b ) where we have used Lemma 6.6. Therefore, by Lemma 6.4, ψ ◦ θ ◦ βb ◦ θ−1 ◦ ψ−1 ◦ α−1 b ∈ AutT(A ⋊αa Z) 34 is asymptotically inner. It follows from the theorem above that there exist an approxi- mately inner automorphism µ ∈ AutT(A ⋊αa Z) and a unitary t ∈ U (A) such that µ ◦ ψ ◦ θ ◦ βb ◦ θ−1 ◦ ψ−1 ◦ µ−1 = Ad t ◦ αb. Let ν = µ ◦ ψ and let s = ν(λαa )(λαa )∗ ∈ U (A). One can deduce and Moreover, from (6.1), (νA) ◦ βb ◦ (νA)−1 = Ad t ◦ αb (νA) ◦ βa ◦ (νA)−1 = (Ad ν(w)s) ◦ αa. (6.1) (6.2) (6.3) (ν ◦ θ ◦ βb ◦ θ−1 ◦ ν−1)(λαa ) = (ν ◦ θ ◦ βb ◦ θ−1)(ν−1(s∗)λαa) = (ν ◦ θ ◦ βb)(ν−1(s∗)w∗λβa) = (ν ◦ θ)(βb(ν−1(ν(w)s)∗)v(λβa)∗) = ν(βb(ν−1(ν(w)s)∗)v(λαa )∗w∗) = Ad t(αb(ν(w)s)∗)ν(v)(λαa )∗(ν(w)s)∗ = Ad t(αb(ν(w)s)∗)ν(v)α−1 a (ν(w)s)∗(λαa )∗ is equal to Hence we can verify Ad t(αb(λαa )) = tuα−1 a (t∗)(λαa )∗. ν(v) = tαb(ν(w)s)uα−1 a (t∗)α−1 a (ν(w)s). This, together with (6.2), (6.3), implies that (α, u) and (β, v) are cocycle conjugate. Corollary 6.17. Any strongly outer cocycle action of Γ on a UHF algebra A is approxi- mately representable. Proof. This immediately follows from the theorem above and Proposition 6.13. Remark 6.18. Let (α, u) : Γ y A be a strongly outer cocycle action on a UHF algebra A. By [27, Corollary 5.9], A ⋊(α,u) Γ ∼= (A ⋊αa Z) ⋊ αb Z is a unital simple AH algebra with real rank zero and slow dimension growth. By means of the Pimsner-Voiculescu exact sequence, we get K0(A ⋊(α,u) Γ) ∼= K0(A), K1(A ⋊(α,u) Γ) ∼=(K0(A) ζA(2) = ∞ K0(A) ⊕ Z/2Z otherwise. Note that ζA(2) = ∞ if and only if A absorbs the UHF algebra of type 2∞ tensorially. 7 Actions of the Klein bottle group II Throughout this section, we let Γ denote the Klein bottle group ha, b bab−1 = a−1i. In this section we will prove that all strongly outer cocycle actions of Γ on the Jiang-Su algebra are cocycle conjugate to each other (Theorem 7.12). 35 7.1 Homotopy of unitaries In this subsection we establish Lemma 7.1 and Lemma 7.3, which will be used in the next subsection to obtain certain cohomology vanishing type results. The following lemma is a variant of [24, Lemma 4.10] or [27, Lemma 6.3], and is essentially contained in the proof of [38, Theorem 5.3]. But we include a proof here for completeness. Lemma 7.1. Let B be a UHF algebra with a unique trace τ and let ϕ be an automorphism of B such that ϕn is not weakly inner for all n ∈ N. For any finite subset F ⊂ B and ε > 0, there exist a finite subset G ⊂ B and δ > 0 such that the following holds. If x ∈ U (B) satisfies k[x, c]k < δ ∀c ∈ G, kϕ(x) − xk < δ, τ (log(xϕ(x∗))) = 0, then we can find a continuous map z : [0, 1] → U (B) such that z(0) = 1, z(1) = x, and Lip(z) < 4. kϕ(z(t)) − z(t)k < ε ∀t ∈ [0, 1] k[z(t), c]k < ε ∀c ∈ F, In addition, if F = ∅, then G = ∅ is possible. Proof. The case F = G = ∅ is contained in [27, Lemma 8.1], and so we treat the general case here. Let (dn)n be a sequence of natural numbers such that dn < dn+1 and B ∼= Md1 ⊗ Md2 ⊗ Md3 ⊗ . . . . We regard Bn = Md1 ⊗ ··· ⊗ Mdn as a subalgebra of B. Let yn ∈ Bn ∩ B′n−1 be a n k = 1, 2, . . . , dn}, where ζn = exp(2π√−1/dn). Define an unitary whose spectrum is {ζ k automorphism σ of B by σ = limn→∞ Ad(y1y2 . . . yn). Then σm is not weakly inner for all m ∈ N. By [15, Theorem 1.3, 1.4], the Z-actions generated by ϕ and σ are strongly cocycle conjugate. Hence it suffices to show the assertion for σ. Suppose that we are given F ⊂ B and ε > 0. Without loss of generality, we may assume that there exists n ∈ N such that F is contained in the unit ball of Bn. Applying [14, Lemma 4.2] to ε/2, we obtain a positive real number δ1 > 0. We may assume δ1 is less than min{2, ε}. Choose a finite subset G ⊂ B and δ2 > 0 so that if u ∈ U (B) satisfies k[u, c]k < δ2 for all c ∈ G, then there exists u0 ∈ U (B ∩ B′n) such that ku − u0k < δ1/4. Let δ = min{δ1/2, δ2}. Suppose that x ∈ U (B) satisfies k[x, c]k < δ ∀c ∈ G, kσ(x) − xk < δ, τ (log(xσ(x∗))) = 0. By the choice of δ, we can find x0 ∈ B ∩ B′n such that kx − x0k < δ1/4. We may assume that there exists m > n such that x0 ∈ Bm ∩ B′n. Put y = yn+1yn+2 . . . ym ∈ Bm ∩ B′n. Then k[y, x0]k = kσ(x0) − x0k < kσ(x) − xk + δ1/2 < δ + δ1/2 ≤ δ1. Also, from kx − x0k < δ1/4 < 1/2 and kx − σ(x)k < δ < 1, one has τ (log(x0yx∗0y∗)) = τ (log(x0σ(x∗0))) = τ (log(x0x∗xσ(x∗)σ(xx∗0))) = τ (log(xσ(x∗))) = 0. 36 It follows from [14, Lemma 4.2] that one can find a path of unitaries w : [0, 1] → Bm ∩ B′n such that w(0) = 1, w(1) = x0, Lip(w) ≤ π + ε and k[y, w(t)]k < ε/2 for every t ∈ [0, 1]. Note that yw(t)y∗ is equal to σ(w(t)). Take a path of unitaries w′ : [0, 1] → B such that w′(0) = 1, w′(1) = xx∗0 and Lip(w′) < δ1/4. Then the path z(t) = w′(t)w(t) meets the requirement. In order to prove Lemma 7.3, we need the following lemma. The proof uses [21]. Lemma 7.2. Let C be a unital simple AT algebra with a unique trace τ . Assume that Ki(C) is isomorphic to a non-zero subgroup of Q for each i = 0, 1. Let w ∈ U (C) be such that [w] 6= 0. Suppose that (xn)n is a central sequence of unitaries in C such that [xn] = 0 and τ (log(xnwx∗nw∗)) = 0 for all sufficiently large n. Then there exists a central sequence (zn)n of unitaries in C([0, 1], C) such that zn(0) = 1, zn(1) = xn and Lip(zn) < 7 for all sufficiently large n. Proof. For almost commuting unitaries u, v in a unital C∗-algebra A, we denote their Bott class by Bott(u, v) ∈ K0(A) ([1, Section 1]). When A is a unital simple separable C∗-algebra with tracial rank zero, by [20, Theorem 3.6], we have τ (Bott(u, v)) = 1 2π√−1 τ (log(vuv∗u∗)) ∀τ ∈ T (A). Let x ∈ C∞ be the image of (xn)n. Since the unitary x commutes with C, there exists a homomorphism π : C0(T\{1})⊗ C → C∞ such that π(f ⊗ c) = f (x)c for every f ∈ C0(T\ {1}) and c ∈ C. We regard Ki(π) as homomorphisms from K1−i(C) to Ki(C∞). Then K1(π)([1]) = [x] = 0, because [xn] = 0 in K1(C) for all sufficiently large n. Also, one has K0(π)([w]) = Bott(w, x) = 0, because τ (Bott(w, xn)) = (2π√−1)−1τ (log(xnwx∗nw∗)) = 0 for all sufficiently large n and K0(C) has no infinitesimal. As Ki(C) is isomorphic to a non-zero subgroup of Q for each i = 0, 1, we obtain K1(π) = 0 and K0(π) = 0. Hence we have the following. • Let p ∈ C be a projection. Then for all sufficiently large n, the K1-class of xnp+1−p is zero. • Let u ∈ C be a unitary. Then for all sufficiently large n, Bott(u, xn) equals zero. Once this is done, one can construct (zn)n as follows. Let F ⊂ C be a finite subset and let ε > 0. By applying [21, Corollary 17.9] (see also [25, Theorem 5.4]), we get a finite subset G ⊂ C, δ > 0, a finite subset P of projections in C and a finite subset U of unitaries in C. Notice that we do not need to work with the entire K-group K(C) of C, because Ki(C) are torsion free. For all sufficiently large n, k[c, xn]k < δ, [xnp + 1 − p] = 0 and Bott(u, xn) = 0 hold for all c ∈ G, p ∈ P and u ∈ U . It follows from [21, Corollary 17.9] that there exists a continuous map zn : [0, 1] → U (C) such that zn(0) = 1, zn(1) = xn, Lip(zn) < 2π + ε and k[c, zn(t)]k < ε for all c ∈ F and t ∈ [0, 1]. Since F and ε were arbitrary, the proof is completed. 37 Lemma 7.3. Let B be a UHF algebra with a unique trace τ and let (α, u) : Γ y B be a strongly outer cocycle action. Suppose that B absorbs the UHF algebra of type 2∞ tensorially. For any finite subset F ⊂ B and ε > 0, there exist a finite subset G ⊂ B and δ > 0 such that the following holds. If x ∈ U (B) satisfies k[x, c]k < δ ∀c ∈ G, kαa(x) − xk < δ, kαb(x) − xk < δ, τ (log(xαb(x∗))) = 0, then we can find a continuous map z : [0, 1] → U (B) such that z(0) = 1, z(1) = x, Lip(z) < 7, kαa(z(t)) − z(t)k < ε and k[z(t), c]k < ε, for all c ∈ F and t ∈ [0, 1]. Proof. The proof is by contradiction. Assume that there exist a finite subset F ⊂ B and ε > 0 such that the assertion does not hold. Then we would have a central sequence (xn)n of unitaries in B satisfying kαb(z(t)) − z(t)k < ε lim n→∞kαg(xn) − xnk = 0 ∀g ∈ Γ and τ (log(xnαb(x∗n))) = 0, and such that there does not exist a unitary z ∈ C([0, 1], B) as described in the statement. Let C = B ⋊(α,u) Γ be the twisted crossed product C∗-algebra. Since B absorbs the UHF algebra of type 2∞, by Remark 6.18, C is a unital simple AT algebra with real rank zero and K0(C) ∼= K1(C) ∼= K0(B). Evidently [λα b ] is not zero in K1(C). We regard B as a subalgebra of C and think of τ as the unique tracial state on C. Then the sequence (xn)n consisting of unitaries in B is a central sequence in C such that [xn] = 0 in K1(C) and τ (log(xnλα b x∗n(λα b )∗))) = τ (log(xnαb(x∗n))) = 0. It follows from the lemma above that there exists a central sequence (zn)n of unitaries in C([0, 1], C) such that zn(0) = 1, zn(1) = xn and Lip(zn) < 7 for all sufficiently large n. By Corollary 6.17, (α, u) : Γ y B is approximately representable. Therefore, in the same way as [26, Lemma 3.3], we may further assume that zn(t) is in B for all n ∈ N and t ∈ [0, 1]. When n is sufficiently large, we have k[zn(t), c]k < ε, kαa(zn(t)) − zn(t)k < ε and kαb(zn(t)) − zn(t)k < ε for all c ∈ F and t ∈ [0, 1], which is a contradiction. 7.2 Cohomology vanishing In this subsection we prove Proposition 7.7, which is a cohomology vanishing type result for cocycle actions of Γ on Z. We also prove that all strongly outer actions of Γ on a certain UHF algebra are strongly cocycle conjugate (Theorem 7.9). Lemma 7.4. Let (α, u) : Γ y B be a strongly outer cocycle action on a UHF algebra B with a unique trace τ . For any finite subset F ⊂ B and ε > 0, there exist a finite subset G ⊂ B and δ > 0 such that the following holds. If x ∈ U (B) satisfies k[x, c]k < δ ∀c ∈ G, kαa(x) − xk < δ, τ (log(xαa(x∗))) = 0, 38 then there exists y ∈ U (B) such that k[y, c]k < ε ∀c ∈ F, kαa(y) − yk < ε and kx − yαb(y∗)k < ε. In addition, if F = ∅ and u(g, h) = 1 for all g, h ∈ Γ, then G = ∅ is possible. Proof. The proof is by contradiction. Assume that there exist a finite subset F ⊂ B and ε > 0 such that the assertion does not hold. Then we would have a central sequence (xn)n of unitaries in B satisfying αa(xn) − xn → 0 as n → ∞ and τ (log(xnαa(x∗n))) = 0, and such that there does not exist a unitary y ∈ B as described in the statement. Let x ∈ U ((B∞)αa) be the image of (xn)n. Choose m ∈ N so that 4/m < ε. Define xi,n ∈ U (B) for i = 0, 1, 2, . . . by x0,n = 1 and xi+1,n = xnαb(xi,n). It is easy to see that (xi,n)n is also a central sequence and αa(xi,n) − xi,n → 0 as n → ∞. Put u = u(b, a)u(a−1, b)∗u(a−1, a) (we need u(a−1, a) because αa−1 is not necessarily equal to α−1 a ). For each i, when n is sufficiently large, τ (log(xi+1,nαa(x∗i+1,n))) = τ (log(xnαb(xi,n)αa(αb(x∗i,n)x∗n))) = τ (log(αb(xi,n)αa(αb(x∗i,n)))) + τ (log(αa(x∗n)xn)) = τ (log(α−1 = τ (log(u∗αb(αa(xi,n))uαb(x∗i,n))) = τ (log(u∗αb(αa(xi,n))αb(x∗i,n)u)) + τ (log(u∗αb(xi,n)uαb(x∗i,n))) = τ (log(αa(xi,n)x∗i,n)), a (αb(xi,n))αb(x∗i,n))) where the last equality follows since (αb(xi,n))n is central and u is homotopic to 1. There- fore, for each i, we have τ (log(xi,nαa(x∗i,n))) = 0 for all sufficiently large n. Let xi ∈ U ((B∞)αa) be the image of (xi,n)n. By Lemma 7.1, there exist paths of unitaries z0, z1 ∈ C([0, 1], (B∞)αa ) such that z0(1) = z1(1) = 1, z0(0) = xm, z1(0) = xm+1, Lip(z0) ≤ 4 and Lip(z1) ≤ 4. By Theorem 6.14, there exist projections e, f ∈ (B∞)αa such that and b (e) = e, αm+1 αm b (f ) = f αi b(e) + m−1Xi=0 mXj=0 αj b(f ) = 1. We may further assume that the projections e, f commute with {αk Z, i ∈ N, t ∈ [0, 1]}. Define a unitary y ∈ (B∞)αa by b (xi), z0(t), z1(t) k ∈ y = m−1Xi=0 xiαi b(z0(i/m))αi b(e) + mXj=0 xjαj b(z1(j/(m+1)))αj b(f ). It is easy to check kx − yαb(y∗)k ≤ 4/m < ε. This is a contradiction. We consider the case F = ∅ and u(g, h) = 1 for all g, h ∈ Γ. In this case we would have a unitary x in (B∞)αa, and u defined above is equal to 1. By replacing (B∞)αa with (B∞)αa, the same proof works. 39 Lemma 7.5. Let B be a UHF algebra of infinite type with a unique trace τ and let (α, u) : Γ y B be a strongly outer cocycle action. For any finite subset F ⊂ B, ε > 0 and r ∈ τ (K0(B)) with r < 1/2, there exists z ∈ U (B) such that kz − αb(z)k < e2π√−1r − 1 + ε k[z, c]k < ε ∀c ∈ F, and kz − αa(z)k < ε, τ (log(zαb(z∗))) = 2π√−1r. Proof. Suppose that we are given r ∈ τ (K0(B)) with r < 1/2. Let σ = lim Ad(y1y2 . . . yn) be the automorphism of B = Md1 ⊗ Md2 ⊗ Md3 ⊗ . . . constructed in the proof of Lemma 7.1. Since B is of infinite type, we may further impose the condition that dn divides dn+1. We regard Mdn as a subalgebra of B. When n is sufficiently large, there exists a unitary zn ∈ Mdn such that znynz∗ny∗n = e2π√−1r. Hence we have τ (log(znσ(z∗n))) = 2π√−1r and kzn − σ(zn)k = e2π√−1r − 1. Moreover (zn)n is a central sequence in B. Take a strongly outer action γ : Γ y Z. Define β : Γ y Z ⊗ B by βa = γa ⊗ id and βb = γb ⊗ σ. Set z′n = 1⊗ zn ∈ Z ⊗ B. It is clear that (z′n)n is a central sequence in Z ⊗ B satisfying z′n = βa(z′n), kz′n − βb(z′n)k = e2π√−1r − 1 and τ (log(z′nβb(z′n)∗)) = 2π√−1r. Let (α, u) : Γ y Z ⊗ B be a strongly outer cocycle action. By Theorem 6.16, (α, u) is cocycle conjugate to β. Thus, there exist θ ∈ Aut(Z ⊗ B) and a family of unitaries (wg)g∈Γ in Z ⊗ B such that θ ◦ αg ◦ θ−1 = Ad wg ◦ βg ∀g ∈ Γ. Put z′′n = θ−1(z′n). Then (z′′n)n is again a central sequence in Z ⊗ B. It is straightforward to see lim n→∞kz′′n − αa(z′′n)k = 0 and In addition, for sufficiently large n, lim n→∞kz′′n − αb(z′′n)k = e2π√−1r − 1. τ (log(z′′nαb(z′′n)∗)) = τ (log(z′nwbβb(z′n)∗w∗b )) = τ (log(z′nβb(z′n)∗)) = 2π√−1r, because (z′n)n is central and wb is homotopic to 1. The proof is completed. Lemma 7.6. Let (α, u) : Γ y B be a strongly outer cocycle action on a UHF algebra B of infinite type with a unique trace τ . For any finite subset F ⊂ B and ε > 0, there exist a finite subset G ⊂ B and δ > 0 such that the following holds. If x ∈ U (B) satisfies τ (log(xαa(x∗))) = 0, ∆τ (x) = 0 k[x, c]k < δ ∀c ∈ G, kαa(x) − xk < δ, then there exists y ∈ U (B) such that k[y, c]k < ε ∀c ∈ F, and kαa(y) − yk < ε, kx − yαb(y∗)k < ε τ (log(y∗xαb(y))) = 0. 40 Proof. Suppose that we are given F ⊂ B and ε > 0. By applying Lemma 7.4 to F and ε/2, we obtain a finite subset G ⊂ B and δ > 0. We would like to show that G and δ meet the requirement. For x as in the statement of the lemma, there exists y ∈ U (B) such that k[y, c]k < ε/2 ∀c ∈ F, kαa(y) − yk < ε/2 and kx − yαb(y∗)k < ε/2. Set Then r = 1 2π√−1 τ (log(y∗xαb(y))). e2π√−1r − 1 ≤ kx − yαb(y∗)k < ε/2. From ∆τ (x) = 0, one also obtains r ∈ τ (K0(B)). By using the lemma above, we can find a unitary z ∈ B such that k[z, c]k < ε/2 ∀c ∈ F, kz − αb(z)k < e2π√−1r − 1 + ε/2 and Then kz − αa(z)k < ε/2, τ (log(zαb(z∗))) = 2π√−1r. τ (log(z∗y∗xαb(yz))) = τ (log(y∗xαb(y))) + τ (log(αb(z)z∗)) = 0. Thus yz does the job. The following is a cohomology vanishing type result for cocycle actions of the Klein bottle group Γ on the Jiang-Su algebra Z. We will use this proposition in the next subsection to prove the uniqueness of strongly outer cocycle actions of Γ on Z. In the proof, Corollary 4.11 plays a central role. We remark that the basic idea of the proof is the same as that of [27, Lemma 7.5, 7.6] or [38, Theorem 5.3]. Proposition 7.7. Let (α, u) : Γ y Z be a strongly outer cocycle action and let τ denote the unique trace on Z. For any finite subset F ⊂ Z and ε > 0, there exist a finite subset G ⊂ Z and δ > 0 such that the following holds. If x ∈ U (Z) satisfies k[x, c]k < δ ∀c ∈ G, kαa(x) − xk < δ, ∆τ (x) = 0 then there exists y ∈ U (Z) such that k[y, c]k < ε ∀c ∈ F, kαa(y) − yk < ε and kx − yαb(y∗)k < ε. Proof. According to [36, Proposition 3.3], the Jiang-Su algebra Z contains a unital subal- gebra isomorphic to Z = {f ∈ C([0, 1], B0 ⊗ B1) f (0) ∈ B0 ⊗ C, f (1) ∈ C ⊗ B1}, where B0 and B1 are the UHF algebras of type 2∞ and 3∞, respectively. We identify B0 and B1 with B0 ⊗ C and C⊗ B1. Set B = B0 ⊗ B1 and denote the unique trace on Z ⊗ B by τ . Clearly (α ⊗ id, u ⊗ 1) : Γ y Z ⊗ Bj and (α ⊗ id, u ⊗ 1) : Γ y Z ⊗ B are strongly outer. 41 Suppose that we are given a finite subset F ⊂ Z and ε > 0. By applying Lemma 7.3 to (α⊗id, u⊗1) : Γ y Z⊗B, {c⊗1 c ∈ F} and ε/2, we obtain a finite subset G1 ⊂ Z⊗B and δ1 > 0. By taking a smaller δ1, we may assume that there exist finite subsets G′1 ⊂ Z ⊗ C, G1,0 ⊂ C ⊗ B0 and G1,1 ⊂ C ⊗ B1 such that G1 = G′1 ∪ G1,0 ∪ G1,1, and that δ1 is smaller than ε. For each j = 0, 1, by applying Lemma 7.6 to (α ⊗ id, u ⊗ 1) : Γ y Z ⊗ Bj, G′1 ∪ G1,j ∪{c⊗ 1 c ∈ F} and δ1/2, we obtain a finite subset G2,j ⊂ Z ⊗ Bj and δ2,j > 0. We may assume that G2,j is contained in {c ⊗ 1 c ∈ G3,j} ∪ {1 ⊗ d d ∈ Bj} for some finite subsets G3,j ⊂ Z. Let G = G3,0 ∪ G3,1 and δ = min{δ2,0, δ2,1, 2}. Suppose that we are given a unitary x ∈ Z satisfying k[x, c]k < δ ∀c ∈ G, kαa(x) − xk < δ, ∆τ (x) = 0. For each j = 0, 1, we have k[x ⊗ 1, c]k < δ2,j ∀c ∈ G2,j, k(αa ⊗ id)(x ⊗ 1) − x ⊗ 1k < δ2,j and Since τ (K0(Z)) equals Z, we also have ∆τ (x ⊗ 1) = 0. τ (log((x ⊗ 1)((αa ⊗ id)(x ⊗ 1))∗)) = 0. Then Lemma 7.6 implies that there exists yj ∈ U (Z ⊗ Bj) such that k[yj, c]k < δ1/2 ∀c ∈ G′1 ∪ G1,j ∪ {c ⊗ 1 c ∈ F}, k(αa ⊗ id)(yj) − yjk < δ1/2, kx ⊗ 1 − yj(αb ⊗ id)(y∗j )k < δ1/2 and τ (log(y∗j (x ⊗ 1)(αb ⊗ id)(yj))) = 0. Put y = y∗0y1 ∈ Z ⊗ B. Then one can verify k[y, c]k < δ1 ∀c ∈ G1, k(αa ⊗ id)(y) − yk < δ1, k(αb ⊗ id)(y) − yk < δ1 and τ (log(y(αb ⊗ id)(y∗))) = τ (log(y∗0y1(αb ⊗ id)(y∗1y0))) = τ (log(y1(αb ⊗ id)(y∗1y0)y∗0)) = τ (log((x ⊗ 1)∗y1(αb ⊗ id)(y∗1y0)y∗0(x ⊗ 1))) = τ (log((x ⊗ 1)∗y1(αb ⊗ id)(y∗1))) + τ (log((αb ⊗ id)(y0)y∗0(x ⊗ 1))) = 0. It follows from Lemma 7.3 that there exists a continuous map z : [0, 1] → U (Z ⊗ B) such that z(0) = 1, z(1) = y, k[z(t), c ⊗ 1]k < ε/2, k(αa ⊗ id)(z(t)) − z(t)k < ε/2 and k(αb ⊗ id)(z(t)) − z(t)k < ε/2 42 for all c ∈ F and t ∈ [0, 1]. Let w(t) = y0z(t). Then one may think of w as a unitary in Z ⊗ Z ⊂ Z ⊗ Z because w(j) = yj belongs to Z ⊗ Bj for each j = 0, 1. Furthermore, k[w, c ⊗ 1]k < ε ∀c ∈ F, k(αa ⊗ id)(w) − wk < ε and kx ⊗ 1 − w(αb ⊗ id)(w∗)k < ε. By Corollary 4.11, there exists a unital embedding of Z into (Z∞)α. Hence we can find a unital homomorphism π : Z ⊗ Z → Z∞ such that π(c ⊗ 1) = c ∀c ∈ Z, αg ◦ π = π ◦ (αg ⊗ id) ∀g ∈ Γ. Consequently, we obtain k[π(w), c]k < ε ∀c ∈ F, kαa(π(w)) − π(w)k < ε and thereby completing the proof. kx − π(w)αb(π(w)∗)k < ε, As a byproduct of Lemma 7.4, we get the following. Proposition 7.8. Let α : Γ y B be a strongly outer action on a UHF algebra B with a unique trace τ . For an α-cocycle (xg)g∈Γ in B, the following conditions are equivalent. (1) ∆τ (xa) = 0. (2) There exists a sequence (wn)n of unitaries in B such that wnαg(w∗n) → xg as n → ∞ for every g ∈ Γ. Proof. We show (2)⇒(1). Suppose that there exists a unitary w ∈ B such that kxg − wαg(w∗)k < 1 for each g = a, b, a−1. Let x′g = w∗xgαg(w). Then (x′g)g is also an α- cocycle, and so we have This, together with x′a−1 = α−1 a (x′a)∗, implies x′aαa(x′b) = x′bαb(x′a−1). τ (log x′a) = τ (log x′a−1) = −τ (log x′a). Thus τ (log x′a) = 0. Hence ∆τ (xa) = ∆τ (x′a) = 0. Let us consider the other implication. Take ε > 0 arbitrarily. By applying Lemma 7.4 to α : Γ y B with the empty set and ε/3 in place of F and ε, we obtain δ > 0. Since the Z-action generated by αa has the Rohlin property, there exists a unitary u ∈ B such that kxa − uαa(u∗)k < min{δ/4, ε/3}. Put x′g = u∗xgαg(u) for all g ∈ Γ. From x′aαa(x′b) = x′bαb(x′a−1 ), one obtains kαa(x′b) − x′bk < δ/2. 43 Besides, τ (log(x′bαa(x′b)∗)) = τ (log(x′aαa(x′b)αb(x′a−1)∗αa(x′b)∗)) = τ (log x′a) + τ (log(αb(x′a−1)∗)) = 2τ (log x′a). By ∆τ (x′a) = ∆τ (xa) = 0, we have r = 1 2π√−1 τ (log x′a)) ∈ τ (K0(B)). It follows from [27, Lemma 8.3] that there exists a unitary v ∈ B such that τ (log(vαa(v∗))) = 2π√−1r. kv − αa(v)k < min{δ/4, ε/3}, Put x′′g = v∗x′gαg(v) for all g ∈ Γ. Then one has kαa(x′′b ) − x′′bk = kαa(v∗x′bαb(v)) − v∗x′bαb(v)k < δ/4 + δ/2 + δ/4 = δ and τ (log(x′′b αa(x′′b )∗)) = τ (log(v∗x′bαb(v)αa(αb(v∗))αa(x′b)∗αa(v))) = τ (log(x′bαb(v)αa(αb(v∗))αa(x′b)∗)) + τ (log(αa(v)v∗)) = τ (log(vαa−1 (v∗))) + τ (log(αa(x′b)∗x′b)) + τ (log(αa(v)v∗)) = τ (log(x′bαa(x′b)∗)) − 4π√−1r = 0. Therefore, by Lemma 7.4, we can find a unitary w ∈ B such that kαa(w) − wk < ε/3 and kx′′b − wαb(w∗)k < ε/3. Consequently we get kxa − uvwαa(w∗v∗u∗)k = ε/3 + ε/3 + kxa − uαa(u∗)k < ε and kxb − uvwαb(w∗v∗u∗)k = kx′′b − wαb(w∗)k < ε/3. Since ε was arbitrary, (2) has been shown. In the setting of the proposition above, the map g 7→ ∆τ (xg) is a homomorphism from Γ to R/τ (K0(B)). Therefore we always have 2∆τ (xa) = 0 because of bab−1 = a−1. Moreover, if K0(B) is 2-divisible (or equivalently, if B absorbs the UHF algebra of type 2∞ tensorially), then we always have ∆τ (xa) = 0. If B absorbs the UHF algebra of type 2∞ Theorem 7.9. Let B be a UHF algebra. tensorially or B does not contain a unital copy of M2, then all strongly outer actions of Γ on B are strongly cocycle conjugate. 44 Proof. By Theorem 6.16, all strongly outer actions of Γ on B are cocycle conjugate. If B absorbs the UHF algebra of type 2∞ tensorially, then the conclusion follows from Proposition 7.8 and the observation above. Suppose that B does not contain a unital copy of M2. Then {c ∈ R/τ (K0(B)) 2c = 0} is isomorphic to Z/2Z and 1/2 + τ (K0(B)) is the generator. Let α : Γ y B be a strongly outer action and let (xg)g be an α-cocycle in B such that ∆τ (xa) 6= 0. By the observation above, ∆τ (xa) equals 1/2 + τ (K0(B)). For g = anbm ∈ Γ, set ζg = (−1)n ∈ C ⊂ B. It is easy to see that (ζgxg)g is an α-cocycle satisfying Ad(ζgxg) = Ad xg and ∆τ (ζaxa) = 0. Hence the conclusion follows. 7.3 Actions of the Klein bottle group on the Jiang-Su algebra In this subsection we prove that all strongly outer cocycle actions of Γ on Z are mutually cocycle conjugate (Theorem 7.12). We also show that all strongly outer actions of Γ on Z are strongly cocycle conjugate to each other (Theorem 7.15). The following is an analogue of Theorem 6.15 for the Jiang-Su algebra. Proposition 7.10. Let ϕ ∈ Aut(Z). For i = 1, 2, assume that ψi ∈ Aut(Z ⋊ϕ Z) satisfies the following conditions. (1) ψi(Z) = Z and ψi(λϕ)λϕ ∈ Z. (2) ϕn ◦ (ψiZ)m ∈ Aut(Z) is not weakly inner for each (n, m) ∈ Z2 \ {(0, 0)}. Assume further that ψ1 ◦ ψ−1 inner automorphism µ ∈ AutT(Z ⋊ϕ Z) and a unitary v ∈ U (Z) such that is approximately inner. Then there exists an approximately 2 µ ◦ ψ1 ◦ µ−1 = Ad v ◦ ψ2. 2 2 It is easy to see that ψ1 ◦ ψ−1 Proof. We denote the unique trace on Z by τ . is in AutT(Z ⋊ϕ Z). By Proposition 2.10 (3), (the Z-action generated by) ϕ is asymptotically representable. Then [9, Theorem 4.8] implies that ψ1 ◦ ψ−1 is T-approximately inner, that is, for any finite subset F ⊂ Z ⋊ϕ Z and ε > 0, there exists a unitary x ∈ Z such that kψ1(c) − xψ2(c)x∗k < ε for all c ∈ F , and we may also assume ∆τ (x) = 0 by replacing x with its suitable scalar multiple. Moreover, for each i = 1, 2, the cocycle action of Γ on Z generated by ϕ and ψiZ is strongly outer. Hence, by virtue of Proposition 7.7, we have the following for each i = 1, 2: if a sequence (xn)n of unitaries in Z is a central sequence in Z ⋊ϕ Z and ∆τ (xn) = 0, then there exists a sequence (yn)n of unitaries in Z such that (yn)n is a central sequence in Z ⋊ϕ Z and satisfies xn − ynψi(y∗n) → 0 as n → ∞. Then the Evans-Kishimoto intertwining argument shows the statement (see [17, Theorem 5.1] for example). Lemma 7.11. Let ϕ be an automorphism of Z such that ϕn is not weakly inner for every n ∈ N. Then an automorphism ψ ∈ AutT(Z ⋊ϕ Z) is approximately inner if and only if ∆τ (ψ(λϕ)(λϕ)∗) = 0, where τ denotes the tracial state on Z. Proof. Let A = Z ⋊ϕ Z. By Corollary 4.11, A is Z-stable. It is well-known that for any UHF algebra B, A⊗ B ∼= (Z ⊗ B) ⋊ϕ⊗id Z is a unital simple AT algebra with real rank zero (see [15, Theorem 1.3]). Therefore one can apply [25, Theorem 7.1] to automorphisms of A. 45 Let ψ ∈ AutT(A). By the Pimsner-Voiculescu exact sequence, K0(A) ∼= K1(A) ∼= Z. Clearly K0(ψ)([1]) = [1] and K1(ψ)([λϕ]) = [λϕ]. It follows that Ki(ψ) is the identity for each i = 0, 1, and hence KL(ψ) = KL(id). Let Θψ,id : K1(A) → Aff(T (A))/ Im DA ∼= R/Z be the homomorphism described in [25, Section 3]. Then, by definition, Θψ,id([λϕ]) is equal to ∆τ (ψ(λϕ)(λϕ)∗), and so the conclusion follows from [25, Theorem 7.1]. Theorem 7.12. Let (α, u) and (β, v) be strongly outer cocycle actions of Γ on the Jiang-Su algebra Z. Then (α, u) and (β, v) are cocycle conjugate. Proof. Put u = u(b, a)u(a−1, b)∗u(a−1, a) and v = v(b, a)v(a−1, b)∗v(a−1, a). The auto- morphism αb extends to αb ∈ Aut(Z ⋊αa Z) by letting αb(λαa) = u(λαa)∗. Likewise, we can extend βb to βb ∈ Aut(Z ⋊βa Z). By [38, Theorem 1.3], we may assume that there exists w ∈ U (Z) such that βa = Ad w ◦ αa. Define an isomorphism θ : Z ⋊βa Z → Z ⋊αa Z by θ(x) = x for x ∈ Z and θ(λβa) = wλαa. Then θ ◦ βb ◦ θ−1 ◦ α−1 is in AutT(Z ⋊αa Z). Let ζ ∈ C be such that ζ = 1 and b ∆τ (ζ−21) = ∆τ ((θ ◦ βb ◦ θ−1 ◦ α−1 b )(λαa )(λαa )∗). Define ψ ∈ AutT(Z ⋊αa Z) by ψ(x) = x for x ∈ Z and ψ(λαa ) = ζλαa. Then b )(λαa ), b )(λαa ) = ζ 2(θ ◦ βb ◦ θ−1 ◦ α−1 (ψ ◦ θ ◦ βb ◦ θ−1 ◦ ψ−1 ◦ α−1 and so ∆τ ((ψ ◦ θ ◦ βb ◦ θ−1 ◦ ψ−1 ◦ α−1 = ∆τ (ζ 21) + ∆τ ((θ ◦ βb ◦ θ−1 ◦ α−1 = 0. b )(λαa )(λαa )∗) b )(λαa )(λαa )∗) It follows from the lemma above that ψ ◦ θ ◦ βb ◦ θ−1 ◦ ψ−1 ◦ α−1 is approximately inner. Therefore, Proposition 7.10 applies to the automorphisms ψ◦ θ◦ βb ◦ θ−1◦ ψ−1 and αb, and yields an approximately inner automorphism µ ∈ AutT(Z ⋊αa Z) and a unitary t ∈ U (Z) such that b µ ◦ ψ ◦ θ ◦ βb ◦ θ−1 ◦ ψ−1 ◦ µ−1 = Ad t ◦ αb. In the same way as the proof of Theorem 6.15, we can conclude that (α, u) and (β, v) are cocycle conjugate. Corollary 7.13. Any strongly outer cocycle action of Γ on the Jiang-Su algebra Z is approximately representable. Proof. This immediately follows from the theorem above and Proposition 6.13. Finally, we would like to prove that all strongly outer actions of Γ on Z are mutually strongly cocycle conjugate (Theorem 7.15). Proposition 7.14. Let α : Γ y Z be a strongly outer action. For two α-cocycles (xg)g and (yg)g in Z, the following conditions are equivalent. (1) ∆τ (xg) = ∆τ (yg) for all g ∈ Γ, where τ is the unique tracial state on Z. 46 (2) There exists a sequence (wn)n of unitaries in Z such that wnxgαg(w∗n) → yg as n → ∞ for every g ∈ Γ. Proof. (2)⇒(1) is obvious. We prove the other implication. Let α⊗ : Γ y Z⊗ be the two-sided infinite tensor product of α : Γ y Z, that is, Z⊗ =OZ Z, α⊗g =OZ αg ∀g ∈ Γ. First, we claim that the assertion holds for α⊗. Let (xg)g be an α⊗-cocycle in Z⊗. Let ζg ∈ C ⊂ Z⊗ be the unitary satisfying ∆τ (xg) = ∆τ (ζg). Then (ζg)g is clearly an α⊗- cocycle, and so it suffices to show that there exists a sequence (wn)n of unitaries in Z such that wnxgαg(w∗n) → ζg as n → ∞. Note that g 7→ ζg is a homomorphism. In particular, ζa = 1 or ζa = −1. Take ε > 0 arbitrarily. Apply Proposition 7.7 to α⊗ : Γ y Z⊗ with the empty set and ε/2 in place of F and ε > 0 to obtain a finite subset G ⊂ Z⊗ and δ > 0. By [38, Theorem 5.3], there exists a unitary y ∈ Z⊗ such that kζ−1 a xa − y∗α⊗a (y)k < min{δ/2, ε/2}, and hence From xaα⊗a (xa−1 ) = 1, it is easy to see kyxaα⊗a (y∗) − ζak < min{δ/2, ε/2}. kyxa−1α⊗a−1(y∗) − ζ−1 a k < min{δ/2, ε/2}. Therefore, defining an α⊗-cocycle (x′g)g by x′g = yxgα⊗g (y∗) for g ∈ Γ, one gets kα⊗a (x′b) − x′bk = kζaα⊗a (x′b) − ζ−1 a x′bk < kx′aα⊗a (x′b) − x′bα⊗b (x′a−1)k + δ = δ. Let σ ∈ Aut(Z⊗) be the shift automorphism. Then we have σ ◦ α⊗g = α⊗g ◦ σ for any g ∈ Γ and [σn(s), t] → 0 as n → ∞ for any s, t ∈ Z⊗. Choose n ∈ N so that k[σn(x′b), c]k < δ ∀c ∈ G. Evidently kα⊗a (σn(x′b)) − σn(x′b)k = kσn(α⊗a (x′b)) − σn(x′b)k < δ. It follows from Lemma 7.7 that there exists z ∈ U (Z⊗) such that kα⊗a (z) − zk < ε/2 Put w = σ−n(z)y. Then we obtain and kζ−1 b σn(x′b) − z∗α⊗b (z)k < ε/2. kwxaα⊗a (w∗) − ζak = kσ−n(z)x′aα⊗a (σ−n(z∗)) − ζak = kσ−n(z)x′aσ−n(α⊗a (z∗)) − ζak < kσ−n(z)σ−n(α⊗a (z∗)) − 1k + ε/2 < ε 47 and kwxbα⊗b (w∗) − ζbk = kσ−n(z)x′bα⊗b (σ−n(z∗)) − ζbk = kzσn(x′b)α⊗b (z∗) − ζbk < ε/2. Since ε > 0 was arbitrary, the proof of the claim is completed. Now we consider α-cocycles (xg)g and (yg)g in Z satisfying ∆τ (xg) = ∆τ (yg) for all g ∈ Γ. By Theorem 7.12, α is cocycle conjugate to α⊗. Thus, there exist an isomorphism θ : Z → Z⊗ and an α⊗-cocycle (ug)g in Z⊗ such that θ ◦ αg ◦ θ−1 = Ad ug ◦ α⊗g ∀g ∈ Γ. Then (θ(xg)ug)g and (θ(yg)ug)g are α⊗-cocycles satisfying ∆τ (θ(xg)ug) = ∆τ (θ(yg)ug). By the claim above, we can find a sequence (wn)n of unitaries in Z⊗ such that wnθ(xg)ugα⊗g (w∗n) → θ(yg)ug ∀g ∈ Γ, and whence θ−1(wn)xgαg(θ−1(w∗n)) = θ−1(wn)xgθ−1(ugα⊗g (w∗n)u∗g) = θ−1(wnθ(xg)ugα⊗g (w∗n)u∗g) → yg for all g ∈ Γ as desired. Theorem 7.15. Any two strongly outer actions α : Γ y Z and β : Γ y Z are strongly cocycle conjugate. Proof. By Theorem 7.12, α and β are cocycle conjugate, that is, there exist an isomorphism θ : Z → Z and a β-cocycle (xg)g in Z such that θ ◦ αg ◦ θ−1 = Ad xg ◦ βg ∀g ∈ Γ. let ζg ∈ C ⊂ Z be the unitary satisfying As in the proof of the proposition above, ∆τ (xg) = ∆τ (ζg). Then (ζ−1 g xg, we may assume ∆τ (xg) = 0. It follows from the proposition above that there exists (wn)n in U (Z) such that wnβg(w∗n) → xg as n → ∞ for all g ∈ Γ, which implies that α is strongly cocycle conjugate to β. g xg)g is also a β-cocycle. Hence, by replacing xg with ζ−1 Acknowledgement. Ozawa and Reiji Tomatsu for several helpful conversations and valuable comments. The authors would like to thank Toshihiko Masuda, Narutaka References [1] O. Bratteli, G. A. Elliott, D. E. Evans and A. Kishimoto, Homotopy of a pair of approximately commuting unitaries in a simple C∗-algebra, J. Funct. Anal. 160 (1998), 466 -- 523. 48 [2] C. Chou, Elementary amenable groups, Illinois J. Math. 24 (1980), 396 -- 407. [3] A. Connes, Almost periodic states and factors of type III, J. Funct. Anal. 16 (1974), 415 -- 445. [4] A. Connes, Outer conjugacy classes of automorphisms of factors, Ann. Sci. ´Ecole Norm. Sup. (4) 8 (1975), 383 -- 419. [5] M. M. Day, Amenable semigroups, Illinois J. Math. 1 (1957), 509 -- 544. [6] H. Futamura, N. Kataoka and A. Kishimoto, Type III representations and automor- phisms of some separable nuclear C∗-algebras, J. Funct. Anal. 197 (2003), 560 -- 575. [7] U. Haagerup, All nuclear C∗-algebras are amenable, Invent. Math. 74 (1983), 305 -- 319. [8] P. de la Harpe and G. Skandalis, D´eterminant associ´e `a une trace sur une alg´ebre de Banach, Ann. Inst. Fourier (Grenoble) 34 (1984), 241 -- 260. [9] M. Izumi and H. Matui, Z2-actions on Kirchberg algebras, Adv. Math. 224 (2010), 355 -- 400. arXiv:0902.0194 [10] M. Izumi and H. Matui, in preparation. [11] X. Jiang and H. Su, On a simple unital projectionless C∗-algebra, Amer. J. Math. 121 (1999), 359 -- 413. [12] B. E. Johnson, Approximate diagonals and cohomology of certain annihilator Banach algebras, Amer. J. Math. 94 (1972), 685 -- 698. [13] V. F. R. Jones, Actions of finite groups on the hyperfinite type II1 factor, Mem. Amer. Math. Soc. 28 (1980), no. 237. [14] T. Katsura and H. Matui, Classification of uniformly outer actions of Z2 on UHF algebras, Adv. Math. 218 (2008), 940 -- 968. arXiv:0708.4073 [15] A. Kishimoto, The Rohlin property for automorphisms of UHF algebras, J. Reine Angew. Math. 465 (1995), 183 -- 196. [16] A. Kishimoto, Automorphisms of AT algebras with the Rohlin property, J. Operator Theory 40 (1998), 277 -- 294. [17] A. Kishimoto, Unbounded derivations in AT algebras, J. Funct. Anal. 160 (1998), 270 -- 311. [18] A. Kishimoto and A. Kumjian, The Ext class of an approximately inner automor- phism, II, J. Operator Theory, 46 (2001), 99-122. [19] A. Kishimoto, N. Ozawa and S. Sakai, Homogeneity of the pure state space of a separable C∗-algebra, Canad. Math. Bull. 46 (2003), 365 -- 372. [20] H. Lin, Asymptotically unitary equivalence and asymptotically inner automorphisms, Amer. J. Math. 131 (2009), 1589-1677. arXiv:math/0703610 49 [21] H. Lin, Approximate homotopy of homomorphisms from C(X) into a simple C∗- algebra, Mem. Amer. Math. Soc. 205 (2010), no. 963. arXiv:math/0612125 [22] T. A. Loring, Lifting solutions to perturbing problems in C∗-algebras, Fields Institute Monographs, 8. American Mathematical Society, Providence, RI, 1997. [23] T. Masuda, Unified approach to classification of actions of discrete amenable groups on injective factors, preprint. arXiv:1006.1176 [24] H. Matui, Z-actions on AH algebras and Z2-actions on AF algebras, Comm. Math. Phys. 297 (2010), 529 -- 551. arXiv:0907.2474 [25] H. Matui, Classification of homomorphisms into simple Z-stable C∗-algebras, J. Funct. Anal. 260 (2011), 797 -- 831. arXiv:1002.4790 [26] H. Matui, ZN -actions on UHF algebras of infinite type, J. Reine Angew. Math. 657 (2011), 225 -- 244. arXiv:1004.3103 [27] H. Matui and Y. Sato, Z-stability of crossed products by strongly outer actions, to appear in Comm. Math. Phys. arXiv:0912.4804 [28] H. Matui and Y. Sato, Strict comparison and Z-absorption of nuclear C∗-algebras, to appear in Acta Math. arXiv:1111.1637 [29] H. Nakamura, The Rohlin property for Z2-actions on UHF algebras, J. Math. Soc. Japan 51 (1999), 583 -- 612. [30] A. Ocneanu, Actions of discrete amenable groups on von Neumann algebras, Lecture Notes in Mathematics 1138, Springer-Verlag, Berlin, 1985. [31] H. Osaka and N. C. Phillips, Furstenberg transformations on irrational rotation alge- bras, Ergodic Theory Dynam. Systems 26 (2006), 1623 -- 1651. arXiv:math/0409169 [32] N. C. Phillips, The tracial Rokhlin property for actions of finite groups on C∗-algebras, Amer. J. Math. 133 (2011), 581 -- 636. arXiv:math/0609782 [33] J. A. Packer and I. Raeburn, Twisted crossed products of C∗-algebras, Math. Proc. Cambridge Philos. Soc. 106 (1989), 293 -- 311. [34] M. Rørdam, Classification of nuclear, simple C∗-algebras, Classification of nuclear C∗-algebras. Entropy in operator algebras, 1 -- 145. Encyclopaedia Math. Sci. 126, Springer, Berlin, 2002. [35] M. Rørdam, The stable and the real rank of Z-absorbing C∗-algebras, Internat. J. Math. 15 (2004), 1065 -- 1084. arXiv:math/0408020 [36] M. Rørdam and W. Winter, The Jiang-Su algebra revisited, J. Reine. Angew. Math. 642 (2010), 129 -- 155. arXiv:0801.2259 [37] Y. Sato, A generalization of the Jiang-Su construction, preprint. arXiv:0903.5286 50 [38] Y. Sato, The Rohlin property for automorphisms of the Jiang-Su algebra, J. Funct. Anal. 259 (2010), 453 -- 476. arXiv:0908.0135 [39] Y. Sato, Discrete amenable group actions on von Neumann algebras and invariant nuclear C∗-subalgebras, preprint. arXiv:1104.4339 [40] M. Takesaki, Theory of operator algebras I, Encyclopaedia of Mathematical Sciences, 124. Operator Algebras and Non-commutative Geometry, 5. Springer-Verlag, Berlin, 2002. [41] M. Takesaki, Theory of operator algebras III, Encyclopaedia of Mathematical Sciences, 127. Operator Algebras and Non-commutative Geometry, 8. Springer-Verlag, Berlin, 2003. [42] K. Thomsen, The homotopy type of the group of automorphisms of a UHF-algebra, J. Funct. Anal. 72 (1987), 182 -- 207. [43] A. S. Toms and W. Winter, Strongly self-absorbing C∗-algebras, Trans. Amer. Math. Soc. 359 (2007), 3999 -- 4029. arXiv:math/0502211 [44] A. S. Toms and W. Winter, Z-stable ASH algebras, Canad. J. Math. 60 (2008), 703 -- 720. arXiv:math/0508218 [45] W. Winter, Strongly self-absorbing C∗-algebras are Z-stable, J. Noncommut. Geom. 5 (2011), 253 -- 264. [46] W. Winter, Nuclear dimension and Z-stability of pure C∗-algebras, Invent. Math. 187 (2012), 259 -- 342. arXiv:1006.2731 51
1510.03472	1	1510	2015-10-12T21:57:07	$L_1$-space for a positive operator affiliated with von Neumann algebra	[ "math.OA", "math.FA" ]	In this paper we suggest an approach for constructing an L1-type space for a positive selfadjoint operator affiliated with von Neumann algebra. For such operator we intro- duce a seminorm, and prove that it is a norm if and only if the operator is injective. For this norm we construct an L1 -type space as the complition of the space of hermitian ultraweakly continuous linear functionals on von Neumann algebra, and represent L1- type space as a space of continuous linear functionals on the space of special sesquilinear forms. Also, we prove that L1 -type space is isometrically isomorphic to the predual of von Neumann algebra in a natural way. We give a small list of alternate definitions of the seminorm, and a special definition for the case of semifinite von Neumann algebra, in particular. We study order properties of L1-type space, and demonstrate the con- nection between semifinite normal weights and positive elements of this space. At last, we construct a similar L-space for the positive element of C-algebra, and study the connection between this L-space and the L1 -type space in case when this C-algebra is a von Neumann algebra.	math.OA	math	L1-space for a positive operator affiliated with von Neumann algebra ANDREJ NOVIKOV ∗ October 15, 2018 Abstract In this paper we suggest an approach for constructing an L1-type space for a positive selfadjoint operator affiliated with von Neumann algebra. For such operator we introduce a seminorm, and prove that it is a norm if and only if the operator is injective. For this norm we construct an L1-type space as the complition of the space of hermitian ultraweakly continuous linear functionals on von Neumann algebra, and represent L1-type space as a space of continuous linear functionals on the space of special sesquilinear forms. Also, we prove that L1-type space is isometrically isomorphic to the predual of von Neumann algebra in a natural way. We give a small list of alternate definitions of the seminorm, and a special definition for the case of semifinite von Neumann algebra, in particular. We study order properties of L1-type space, and demonstrate the connection between semifinite normal weights and positive elements of this space. At last, we construct a similar L-space for the positive element of C-algebra, and study the connection between this L-space and the L1-type space in case when this C-algebra is a von Neumann algebra. Keywords:operator algebra; von Neumann algebra; C-algebra; noncommutative integration; L1-space; pos- itive operator; semifinite normal weight; unbounded operator. AMS Subject Classification: 46L05, 46L10, 46L51, 47B65, 47C15, 47L50 1 Introduction In 1953 I.E. Segal proposed the foundations of noncommutative integration theory with respect to a faithful normal semi-finite trace τ , based on the notion of measurability of an unbounded operator affiliated with a von Neumann algebra, and used it to define noncommutative L1, L2 and L∞ spaces. [1] Thereafter, this definition was succesfully extended to the full range of Lp, and the theory of this Lp-spaces was studied by various authors. Extension of the theory to the case of a von Neumann algebra equipped with an arbitrary weight became possible only after development of the Tomita -- Takesaki modular theory. The main results in this field are reviewed in [2]. One of the approaches for noncommutative integration in von Neumann algebras with respect to the normal weight was proposed by A.N. Sherstnev in 1970s (see [3] and [4]). Later it was developed by A.M. Bikchentaev, O.E. Tikhonov, N.V. Trunov, A.A. Zolotarev and others. The main results in this area are reviewed in the [5] and [6]. In this paper we propose an approach for the construction of L1 space, which is associated with a positive selfadjoint operator affiliated with a von Neumann algebra. In some sence the statement of a question in this paper is dual to Sherstnev's approach, and many of the theorems of this paper have their duals, as it is noted in the text. ∗Lobachevskii Institute of Mathematics and Mechanics, Kazan (Volga region) Federal University, [email protected] 1 2 Definitions and notation Throughout this paper we adhere to the following notation. By M we denote a von Neumann algebra that acts on a Hilbert space H with the scalar product h·, ·i. We denote its selfadjoint part by Msa, and the set of all projections in M by Mpr. Let p ∈ Mpr and x ∈ M, then by xp we denote the restriction of pxp to pH (xp := pxppH ), also we denote the reduction of M to pH by Mp. By C(M) we denote the center of M. By M∗ and Mh ∗ we denote the predual of M and its Hermitian part, respectively. If an operator x is affiliated with M then we write xηM. We denote the domain of an operator x by D(x). We denote the closure of an operator x by x, adjoint operator is denoted as x∗. We denote the identity operator, the zero operator and the zero vector by 1, 0 and 0 , respectively. We use standard notation for multiplication of a functional ϕ ∈ M∗ by an operator x ∈ M, namely, xϕ, ϕx and xϕx denote the linear functionals y 7→ ϕ(xy), y 7→ ϕ(yx) and y 7→ ϕ(xyx), respectively. For a normed space X we use X ∗ to denote its continuous dual space and X al to denote its algebraic dual space. For an ordered normed space X we denote its positive cone by X +. We also consider partial order for positive selfadjoint operators affiliated with M. For positive selfadjoint 2 ). If for an 2 f k2 for all f ∈ D(y 2 f k2 ≤ ky 2 ) ⊂ D(x 1 1 1 1 1 x, yηM we write x ≤ y if and only if D(y increasing net (xj )j∈J of operators affiliated with M there exists x = sup j∈J 2 ) and kx (xj ), then we write xj ր x. For a positive selfadjoint operator xηM we use xλ to denote λx(λ + x)−1 with λ ∈ R+ \ {0}. From the Spectral theorem it follows that the mapping λ 7→ xλ ∈ M+ is monotone operator-valued function and ∗ we define ϕ(x) as 2 ), therefore xλ ր x. For an unbounded x and ϕ ∈ M+ 2 f for all f ∈ D(x λ f = x x lim 1 1 1 2 λ→+∞ ϕ(x) := lim λ→+∞ ϕ(xλ). 3 Construction and Representation of L1-spaces From now on a stands for a positive selfadjoint operator affiliated with M. In this section we give the definition of L1(a) and its dual L∞(a). Also we construct natural isomorphism of L∞(a) onto the space of special kind sesquilinear forms Sa(M). We consider D+ bounded, then D+ a ≡ {ϕ ∈ M+ a = M+ ∗ , Dh ∗ ϕ(a) < +∞}, Dh a ≡ D+ a − D+ a and Da ≡ linCD+ a . Note that if operator a is a = Mh ∗ and Da = M∗. We define a seminorm k · ka on Dh a as kϕka := inf{ϕ1(a) + ϕ2(a) ϕ = ϕ1 − ϕ2; ϕ1, ϕ2 ∈ D+ a }. Also, from [7, Theorem 2] states, that if operator a is bounded, then kϕka = ka 2 k. If k · ka is a norm, then we call it the a-norm. Note that the 1-norm coincides with the restriction of the standard norm in M∗ onto Mh ∗ . 2 ϕa 1 1 Lemma 1. For an injective bounded operator a the sets a {a 2 = {a 2 x ∈ Msa} are σ-weakly dense subsets of M and Msa, respectively. 2 Ma 2 xa 1 1 1 1 1 2 xa 1 2 x ∈ M} and a 1 2 Msaa 1 2 = 1 2 ( 1 2 ) = 1 in the strong operator topology. Let y ∈ M and xn := ( 1 Proof. According to the Spectral Theorem the sequence qn ≡ a rp(a yn = qnyqn = a σ-weakly converges to y. Hence, a note that if y is selfadjoint, then yn is selfadjoint. 2 )−1 is increasing and converges to 2 )−1y( 1 2 )−1, then the sequence 2 converges to y in the weak operator topology. Since kynk ≤ kyk, it follows that yn 2 is dense in M in the σ-weak operator topology. To finish the proof n + a n + a n + a 2 Ma 2 xna 1 1 1 1 1 1 1 1 A slight change in the latter proof actually shows that the sets {a 1 2 xa 1 2 x ∈ M, a 1 2 xa 1 2 ∈ M} and {a 1 2 xa 1 2 x ∈ Msa, a 1 2 xa 1 2 ∈ M} are σ-weakly dense subsets of M and Msa, respectively. Theorem 1. k · ka is a norm on Dh a if and only if operator a is injective. 2 Proof. If operator a is not injective, then there exists a non-zero f , such that af = 0 , hence kh·f, f ika = haf, f i = 0 and k · ka is not a norm. Conversely, if k · ka is a norm and operator a is bounded, then kϕka = ka 2 k by [7, Theorem 2], and from 2 k = 0 only if ϕ = 0. If operator a is unbounded, then for each λ ∈ R+ the Lemma 1, it follows that ka positive bounded operator aλ ∈ M is injective. Evidently, ϕ1(aλ) + ϕ2(aλ) ≤ ϕ1(a) + ϕ2(a) for each λ ∈ R+ and all ϕ1, ϕ2 ∈ D+ a. Since operator aλ is injective, it follows that k · kaλ is a norm, hence k · ka is a norm. a , therefore the inequality kϕkaλ ≤ kϕka holds for each ϕ ∈ Dh 2 ϕa 2 ϕa 1 1 1 1 For an injective operator a by Lh 1 (a) we denote the completion of the normed space (Dh Proposition 1] the dual of Lh 1 (a) is (Lsa ∞(a), kxka), where Lsa ∞(a) ≡ {x ∈ (Dh a, k · ka). By [7, a)al∃λ ∈ R, −λa ≤ x ≤ λa} and kxka ≡ inf{λ ∈ R − λa ≤ x ≤ λa}. We identify the elements of Dh Further for an injective operator a we always assume that Lsa For x ∈ M we define the sesquilinear form 1 \ a 2 xa 1 2 on D(a a with the corresponding elements in Lh 1 (a). 1 1 2 ) × D(a ∞(a) is equiped with the a-norm. 2 ) by the equality \ 2 xa a \ a 2 ya 1 1 1 1 1 1 1 \ a 2 xa 1 2 (f, g) := 1 1 2 f, a 2 gi. The set of all such sesquilinear forms is denoted by Sa(M) ≡ { hxa \ order on Sa(Msa), such that 2 xa a Sa(Msa) we denote the seminormed space of sesquilinear forms { 2 (f, f ) ≤ \ 2 xa a 2 if and only if \ a 2 xa \ a 2 ya 2 ≤ 1 1 1 1 1 1 2 x ∈ M}. We consider partial 2 (f, f ) for all f ∈ D(a 2 ). By 2 x ∈ Msa} equiped with the seminorm 1 1 \ a 2 xa \ 2 ) := inf{λ ∈ R+ − λ a 2 1a 1 1 1 2 ≤ pa( 1 \ a 2 xa 1 2 ≤ λ 1 \ a 2 1a 1 2 }. Theorem 2. For ϕ ∈ Da the equality a 1 2 ϕa 1 2 (x) = lim λ→+∞ 1 1 ϕ(a 2 λ xa 2 λ ) with x ∈ M defines the normal functional 1 2 ϕa a 1 2 ∈ M∗. Proof. Let ϕ ∈ D+ a and +∞Pi=1 1 2 ), +∞Pi=1 ka 1 2 fik2 < hxa 1 2 fi, a 1 2 fii. For h·fi, fii be a representation of ϕ. It is evident that {fi}+∞ i=1 ⊂ D(a +∞ and lim λ→+∞ 1 a 2 λ f = a 1 2 f for all f ∈ D(a 1 2 ), which implies lim λ→+∞ nPi=1 hxa 1 1 2 λ fi, a 2 λ fii = nPi=1 the fixed x ∈ M the inequality hxa 1 1 2 λ fi, a 2 λ fii ≤ kxkka 1 2 fik2 holds, hence for the fixed x ∈ M the series hxa 1 1 2 λ fi, a 2 λ fii converge uniformly with repect to λ on (0, +∞), therefore a 1 2 ϕa 1 2 (x) ≡ lim λ→+∞ 1 1 ϕ(a 2 λ xa 2 λ ) = +∞Pi=1 hxa 1 1 2 λ fi, a 2 λ fii = +∞Pi=1 hxa 1 2 fi, a 1 2 fii. The latter implies that a 1 2 ϕa 1 2 (x) is well-defined for any x ∈ M +∞Pi=1 lim λ→+∞ and a 1 2 ϕa 1 2 ∈ M+ ∗ . For ϕ ∈ Da there exist ϕ1, ϕ2, ϕ3, ϕ4 ∈ D+ a 1 2 ϕia 1 2 ∈ M+ ∗ , i = 1, 4, hence a 1 2 ϕ1a 1 2 (x) − a a , such that ϕ = ϕ1 − ϕ2 + iϕ3 − iϕ4. We have proved that λ ) − 2 ϕ2a 2 (x) = lim 2 (x) − ia 2 (x) + ia 2 ϕ4a 2 ϕ3a ϕ1(a λ xa 1 2 2 1 1 1 1 1 1 1 λ→+∞ 1 1 1 1 1 1 ϕ2(a 2 λ xa 2 λ ) + iϕ3(a 2 λ xa 2 λ ) − iϕ4(a 2 λ xa any x ∈ M and a 1 2 ϕa 1 2 ∈ M∗. 2 λ ) = lim λ→+∞ 1 1 ϕ(a 2 λ xa 2 λ ) ≡ a 1 2 ϕa 1 2 (x), therefore a 1 2 ϕa 1 2 (x) is well-defined for Corollary 1. For ϕ ∈ D+ a , such that ϕ = hold for all x ∈ M. +∞Pi=1 h·fi, fii, the equalities a 1 2 ϕa 1 2 (x) = +∞Pi=1 hxa 1 2 fi, a 1 2 fii = +∞Pi=1 1 \ 2 xa a 1 2 (fi, fi) Definition 1. For ϕ ∈ Da and a sesquilinear form a 2 (x). 2 ϕa 1 1 1 \ a 2 xa \ 2 ∈ Sa(M) we consider ϕ( 2 xa a 1 1 1 2 ) ≡ 1 \ 2 xa a 1 2 (ϕ) := Lemma 2. The latter definition indentifies a sesquilinear form functional on (Dh a, k · ka) 3 1 \ 2 xa a 1 2 ∈ Sa(M) with the continuous linear Proof. The linearity of 1 \ a 2 xa 1 2 is evident. For ϕ ∈ Dh a consider arbitary ϕ1, ϕ2 ∈ D+ then 1 \ a 2 xa 1 2 (ϕ) = a 1 2 ϕa 1 2 (x) ≤ a 1 2 ϕ1a 1 2 (x)+a 1 2 ϕ2a 1 2 (x) ≤ kxk(ϕ1(a)+ ϕ2(a)), which implies a , such that ϕ = ϕ1 − ϕ2, 2 (ϕ) ≤ \ a 2 xa 1 1 kxkkϕka and therefore 1 \ 2 xa a 1 2 is a continuous linear functional. Theorem 3. For an injective operator a the idenitification of the sesquilinear forms in Sa(Msa) with the continuous linear functionals on (Dh a, k · ka) lead to the identification of (Sa(Msa), pa) with Lsa ∞(a). Proof. Lemma 2 states that arbitary \ a 2 xa 2 is real-valued on Dh a, hence 1 1 1 \ a 2 xa 1 2 ∈ Lsa ∞(a). 1 \ 2 xa a 1 2 ∈ Sa(Msa) is a continuous linear functional on (Dh a, k · ka). Also, Suppose x ∈ Lh ∞(a) and kxka = 1. For any f ∈ D(a 1 x(ψf ) ≤ ψf (a) = ka 1 2 f k2. Define a form bx(f ) on D(a 2 ) × D(a 1 1 1 2 ) there exists a positive functional ψf := h·f, f i and 2 ) we define 2 ) as bx(f ) := x(ψf ). Now, for f, g in D(a 1 ∞(a) there exists a sesquilinear form bx on D(a 4 (bx(f + g) − bx(f − g) + a sesquilinear form bx(f, g) on D(a ibx(f + ig) − ibx(f − ig)). Note that x(ϕ) is real-valued for every ϕ in Dh the equality bx(f, g) = bx(g, f ) and the linearity by the first argument. From the above arguments we conclude 2 ) by the polarization identity bx(f, g) := 1 a. A straightforward calculation gives that for any x in Lh 2 ) × D(a 1 1 1 2 f, a− 1 2 f ) ≤ ka We consider a sesquilinear form by defined with the equality by(f, g) = bx(a− 1 From by(f, f ) = bx(a− 1 such that by(f, g) = hyf, gi. Also, bx(f, g) = by(a by definition. Since hyf, gi = by(f, g) = bx(a− 1 2 f, a 2 f, a− 1 that y is selfadjoint. Therefore, we identify x ∈ Lsa 2 f, a 2 g, a− 1 \ ∞(a) with 2 ya a \ The equality kxka = inf{λ−λa ≤ x ≤ λa} = inf{λ−λ a 2 1a 2 f k2 = kf k2, it follows that by is bounded, hence there exists y ∈ M, 2 f ) = by(g, f ) = hyg, f i = hf, ygi it follows 2 g) = bx(a− 1 2 ) and bx = 2 gi for any f and g in D(a \ 2 } = pa( a 2 ya 2 ∈ Sa(Msa). 2 ) completes 2 ) × Im(a 2 g) = hya \ a 2 1a \ a 2 ya \ a 2 ya 2 a− 1 2 ≤ λ 2 ≤ 1 2 ). 1 1 1 1 2 1 1 1 1 1 1 1 1 1 1 1 1 1 2 ), such that bx(f, g) = bx(g, f ). 2 g) on Im(a 2 f, a− 1 1 the proof. Remark 1. If an operator a is injective then inf{λ − λa ≤ a 1 2 xa 1 2 ≤ λa} = kxk and the latter implies that the mapping x 7→ 1 \ 2 xa a 1 2 is an isometrical isomorphism of Msa onto (Sa(Msa), pa). Corollary 2. For an injective operator a the mapping u : x ∈ Msa 7→ isomporphism of Msa onto Lsa onto (M∗)h. ∞(a). Moreover, the adjoint mapping ut is an isometrical isomorphism of (Lsa ∞(a) is an isometrical ∞(a))∗ 1 \ 2 xa a 1 2 ∈ Lsa Remark 2. For an injective operator a we already identify the elements of Dh of the complition Lh 1 (a). Also, since (Lh convinient to identify the elements of Lh Corollary 2 and ϕ ∈ Dh for all x ∈ Msa. a the equality ut(ϕ) = a 2 ϕa 1 ∞(a))∗ is isometrically isomorphic to the second dual of Lh 1 (a) with the corresponding elements of (Lh a with the corresponding elements 1 (a), it is ∞(a))∗. Then for ut from 2 (x) 2 ) = a 2 ϕa 1 1 1 1 \ 2 holds, since ut(ϕ)(x) = ϕ(u(x)) = ϕ( 2 xa a 1 Definition 2. For an injective operator a and a linear functional ϕ ∈ (Lsa a 2 , where ut is the isomorphism from Corollary 2. 2 ϕa 1 1 ∞(a))∗ we denote ut(ϕ) ∈ (M∗)h as Theorem 4. For an injective operator a the mapping v : ϕ ∈ Lh isomorphism of Lh 1 (a) onto Mh ∗ . 1 (a) 7→ a 1 2 ϕa 1 2 ∈ Mh ∗ is an isometrical Proof. Note that v is the restriction of ut from Corollary 2 to Lh ψn = ( 1 ψ pointwise (i.e. 2 )−1ψ( 1 2 )−1 lies in Dh n + a 2 ψna a lim 2 (x) = ψ(x) for all x ∈ M), hence v(Dh a. Since a n + a n + a 2 ( 1 1 1 1 1 1 n→+∞ 1 2 )−1 ր 1, it follows that v(ψn) = a 1 (a). Let ψ be an element of Mh ∗ . The sequence 2 converges to 2 ψna ∗ . By [8, Theorem 1 (a) is isometrically isomorphic to a) is weakly dense in Mh 1 1 a) is also dense in Mh ∗ in the 1-norm topology, and therefore Lh 3.12], v(Dh Mh ∗ . 4 Summarizing the facts, that if operator a is injective, then (Sa(Msa), pa) is isometrically isomorphic to Msa and that M is the natural complexification of Msa, it seems reasonable to identify the complexification of \ Sa(Msa) with Sa(M) extending pa onto Sa(M) by the equality pa( a 2 xa further we identify L∞(a) with Sa(M), also we denote complexification of Lh L1(a) by the equality kϕka = ka 2 ) = kxk. For an injective operator a 1 (a) as L1(a) extending k · ka onto 2 ϕa 2 k. 1 1 1 1 The folowing corollary evidently follows from Theorem 3 and Corollary 2. Corollary 3. For an injective operator a the mapping U : x ∈ M 7→ isomorphism of M onto L∞(a), the adjoint mapping U t : ϕ ∈ L∗ ∞(a) 7→ a isomorphism of L∗ L1(a) onto M∗. 2 ∈ L∞(a) is an isometrical 2 ∈ M∗, also is an isometrical ∞(a) onto M∗. Moreover, the restriction V := U tL1(a) is an isometrical isomorphism of 1 1 1 \ 2 xa a 2 ϕa 1 The latter Corollary is an analog of the result of [9]. 4 Alternate Definition of the Norm 1 1 2 ϕa We have defined kϕka as inf{ϕ1(a) + ϕ2(a)ϕ = ϕ1 − ϕ2, ϕ1, ϕ2 ∈ D+ a }. Also, if operator a is bounded, then the equality kϕka = ka 2 k holds by [7, Theorem 2]. Thus, if an operator a is bounded, then we are able 2 k ∈ R maintaining all to change the definintion of k · ka to that k · ka is the mapping ϕ ∈ Da 7→ ka the properties described in the previous section. Also, it is evident, that if operator a is bounded and M is 2 k = ϕ(a) holds for all ϕ ∈ Da. Thus, in case of commutative M we commutative, then the equality ka are able to define k · ka even simplier, as the mapping ϕ ∈ Da 7→ ϕ(a) ∈ R, maintaining all the properties of k · ka. 2 ϕa 2 ϕa 1 1 1 1 In this section we demonstrate that in general case the equality kϕka = ka 2 ϕa 2 k also holds for all ϕ ∈ Dh a, 2 ϕa 2 k ∈ R. Also, we show that the a → ϕ(a) ∈ R is a seminorm if and only if operator a is affiliated with the center C(M) of M. 2 k = ϕ(a) for all ϕ ∈ Da. At last, we consider case of a therefore k · ka may be alternatively defined as the mapping ϕ ∈ Da → ka mapping ϕ ∈ Dh Moreover, we show that aηC(M) if and only if ka semifinite M and M = B(H) as an example. 2 ϕa 1 1 1 1 1 1 Let a = λde(λ) be the spectral decomposition of a. The equality ma(ϕ) = +∞R0 λdϕ(e(λ)) (ϕ ∈ M+ ∗ ) +∞R0 +∞R0 defines the additive positively homogeneous mapping ma : M+ M+ (pmap)(ϕ) = ma(pϕp) (ϕ ∈ M+ ∗ . The mapping ma is the element of extended positive part cM+ of M. [10] For p ∈ Mpr the equality ∗ ) defines the element pmap in cM+, where (pϕp)(x) = ϕ(pxp)(x ∈ M)). [10] ∗ 7→ [0, +∞], which is lower semicontinuous on Lemma 3. Let a = λde(λ) be the spectral decomposition of an operator a and for any natural n the restriction ae(n) of ae(n) to e(n)H belongs to C(Me(n)), then aηC(M). Proof. For a natural m ≥ n the restiction ae(n) of ae(n) to e(m)H belongs to C(Me(m)) as the function of the operator am. Therefore, for any x ∈ M the equality ae(n)e(m)xe(m) = e(m)xe(m)ae(n) holds. Passing to the limit m → +∞ we conclude that ae(n) belongs to C(M), hence aηC(M). Theorem 5. Let a = are equivalent: +∞R0 λde(λ) be the spectral decomposition of an operator a, then the following statements (i) aηC(M); (ii) ∀p ∈ Mpr the inequality pmap ≤ ma holds; (iii) ∀ϕ ∈ Mh (iv) the mapping ϕ 7→ ma(ϕ+) (ϕ ∈ Mh i.e., ϕ, ψ ∈ Mh ∗ , ϕ ≤ ψ imply ma(ϕ+) ≤ ma(ψ+); ∗) is monotone, ∗ and any decomposition ϕ = ϕ1 − ϕ2 (ϕ1, ϕ2 ∈ M+ ∗ ) the inequality ma(ϕ+) ≤ ma(ϕ1) holds; 5 ∗ ) is subadditive, (v) the mapping ϕ 7→ ma(ϕ+) (ϕ ∈ Mh i.e., ma((ϕ + ψ)+) ≤ ma(ϕ+) + ma(ψ+) for all ϕ, ψ ∈ Mh ∗ ; (vi) the mapping ϕ 7→ ma(ϕ+) (ϕ ∈ Mh ma((λϕ + (1 − λ)ψ)+) ≤ λma(ϕ+) + (1 − λ)ma(ψ+) for all ϕ, ψ ∈ Mh (vii) the mapping ϕ 7→ ma(ϕ) (ϕ ∈ Mh ∗ ) is subadditive, i.e., ma(ϕ + ψ) ≤ ma(ϕ) + ma(ψ) for all ϕ, ψ ∈ Mh ∗ (viii) the mapping ϕ 7→ ma(ϕ) (ϕ ∈ Mh ∗) is convex, i.e., ma(λϕ + (1 − λ)ψ) ≤ λma(ϕ) + (1 − λ)ma(ψ) for all ϕ, ψ ∈ Mh ∗) is convex, i.e., ∗ , λ ∈ [0, 1]; ∗ , λ ∈ [0, 1]. Proof. First prove the implication (i) ⇒ (ii). Since a = λde(λ) is affiliated with M, it follows that for any natural n the bounded operator ae(n) belongs to C(M), hence pae(n)p ≤ ae(n) for all p ∈ Mpr by [11, Theorem 1]. For an arbitary ϕ ∈ M+ ∗ and p ∈ Mpr we have: +∞R0 ma(ϕ) = lim n→∞ ϕ(ae(n)) ≥ lim n→∞ ϕ(pae(n)p) = = lim n→∞ (pϕp)(ae(n)) = ma(pϕp) = (pmap)(ϕ) (i.e. pmap ≤ ma). The proof of the implications (ii) ⇒ (iii) ⇒ (iv) ⇒ (v) ⇒ (vii), (v) ⇔ (vi), (vii) ⇔ (viii) literally repeats the proof of respective implications in [11], therefore it is sufficient to prove the impication (vii) ⇒ (i). According to Lemma 3, it is sufficient to prove, that for any natural n the operator ae(n) belongs to C(Me(n)). Let (vii) hold and eϕ, eψ ∈ (Me(n))h e(n)ψe(n), ψe(n) = eψ. Then ∗ . Construct ϕ, ψ ∈ Mh ∗ , such that ϕ = e(n)ϕe(n), ϕe(n) = eϕ, ψ = eϕ + eψ(ae(n)) = ϕ + ψ(ae(n)) = = ma(ϕ + ψ) ≤ ma(ϕ) + ma(ψ) = eϕ(ae(n)) + eψ(ae(n)). By [11, Theorem 1], ae(n) ∈ C(Me(n)). Remark 3. According to [11, Corollary 1] if a ∈ C(M), then ka the latter equality to the case of unbounded operator a in Corollary 5. 2 ϕa 1 1 2 k = ϕ(a) for all ϕ ∈ M∗. We generalize In Theorem 2 we have defined a a as a normal functional on M by the equality a λ ) with aλ ր a. Further we define L∞(a) for an arbitary a and define the meaning of a 2 for ϕ ∈ Dh 2 ϕa 2 λ xa ϕ(a 1 2 1 2 ϕa 2 ϕa 2 ≡ 2 for 1 1 lim 1 1 1 1 λ→+∞ ϕ ∈ L∗ ∞(a). Definition 3. For a non-injective operator a we define Lsa 0}. With [bx] we denote the equivalence class of the sesquilinear form bx. The norm of this quotient space is denoted by k[bx]ka := pa(bx). Let p be the projection onto ker a, then for q = 1 − p the equalities aq = qa = a hold. ∞(a) as the quotient space Sa(Msa)/{bx ∈ Sa(Msa)pa(bx) = Theorem 6. The mapping [ 1 \ a 2 xa 1 2 ] 7→ 1 \ a q xqa 2 1 2 q is an isometrical isomorphism of Lsa ∞(a) onto Saq (Msa q ) Proof. For all f, g ∈ D(a 1 2 ) the chain of equalities 1 \ a 2 xa 1 2 (f, g) ≡ hxa 1 2 f, a 1 2 gi = hxqa 1 2 qf, qa 1 2 qgi = 1 \ a q xqa 2 1 2 q (qf, qg) holds. Also, note that 1 1 \ 2 xa a \ a 2 ya 2 f k2 ≤ h(x − y)a \ 2 xa a 2 − 1 1 1 1 1 1 If pa( −λka 2 (f, g) = 1 \ a 2 xa 1 2 (qf, qg) for all f, g ∈ D(a \ a 2 1a \ a 2 xa 2 ≤ 1 1 1 1 1 2 ). 2 ) = 0, then −λ 2 f, a 2 f i ≤ λka 1 1 1 \ 2 ≤ λ 2 for all λ ≥ 0, which implies a 2 1a 2 f k2 for all λ ≥ 0 and all f ∈ D(a 2 ). The latter implies that xq = yq, \ a 2 ya 2 − 1 1 1 1 since Im(a 1 2 ) is dense in qH. Therefore, the mapping [ 1 \ 2 xa a 1 2 ] 7→ 6 1 \ q xqa a 2 1 2 q is well-defined. 2 ] = [ \ 1 1 \ 2 (λx)a 1 a \ 2 xa a a 2 ] 7→ q (λx)qa \ 2 ]ka = pa( a 2 xa 2 1 1 1 1 The mapping is linear, since [ 1 \ 2 xa a 1 2 ] + [ 1 \ a 2 ya 1 1 \ a 2 (x + y)a 1 2 ] 7→ a 1 \ 1 1 \ 2 q (x + y)qa 2 q = a 2 q (xq + yq)a 1 2 q = 1 \ a q xqa 2 1 2 q + 1 \ q yqa a 2 1 2 q and λ[ 1 \ 2 xa a 1 2 ] = [ 1 1 \ 1 2 q = a 2 q (λxq)a 2 q = λ 1 \ q xqa a 2 1 2 q . Also, the chain of equalities k[ 2 ) = inf{λ∀f ∈ D(a \ 2 f k2} = inf{λ∀f ∈ qH − λkf k2 ≤ hxqf, f i ≤ λkf k2} = kxqk = paq ( a q xqa 2 1 1 1 λka 1 2 ) − λka 1 2 f k2 ≤ hxa 1 2 f, a 1 2 f i ≤ 2 q ) holds, therefore the mapping 1 \ 2 xa a 1 2 ] 7→ [ 1 \ q xqa 2 a 1 2 q is an isometrical isomorphism. We denote the complexification of Lsa ∞(a) by L∞(a). Since k[ 1 \ a 2 xa 1 2 ]ka = kxqk for all [ 1 \ a 2 xa 1 2 ] ∈ Lsa ∞(a), it is natural to extend the norm k · ka onto L∞(a) with the equality k[ 1 \ a 2 xa 1 2 ]ka = kxqk. Definition 4. For ϕ ∈ L∗ ∞(a) let bϕ be the corresponding element in S∗ 1 1 1 linear functional on M with the equality a 2 ϕa 2 (x) := ϕ([ 2 q )). Note that a 1 2 ϕa 1 2 (x) = aq (Mq). We define a 1 2 ϕa 1 2 as a bounded 1 \ 2 xa a 2 ])(≡ bϕ( 1 \ q xqa a 2 1 2 q (xq). 1 1 a 2 q bϕa Theorem 7. For any ϕ in L∗ ∞(a) the equality kϕka = ka 1 2 ϕa 1 2 k holds. Proof. From Theorems 3, 6 and Remark 1 L∞(a) is isometrically isomorphic to Mq. Therefore, the dual L∗ is isometrically isomorphic to M∗ isomorphism from Corollary 3. q and kϕka = kbϕkaq = kU t(bϕ)k = ka ∞(a) 2 k, where U t is the q bϕa q k = ka 2 ϕa 1 2 1 2 1 1 Theorem 8. For any ϕ in Dh a the equality kϕka = ka 1 2 ϕa 1 2 k holds. Proof. If an operator a is injective, then since Dh equality kϕka = ka 2 k holds for any ϕ in Dh 2 ϕa 1 1 a by Theorem 7. a is isometrically embedded into L∗ ∞(a), it follows that the If operator a is not injective, then kϕka = inf{ϕ1(a) + ϕ2(a)ϕ = ϕ1 − ϕ2, ϕ1, ϕ2 ∈ D+ ϕ2(qaq)ϕ = ϕ1 − ϕ2, ϕ1, ϕ2 ∈ D+ ka 2 ϕa 2 k. 1 1 a } = inf{cϕ1(aq) + cϕ2(aq)bϕ = cϕ1 − cϕ2,cϕ1,cϕ2 ∈ D+ a } = inf{ϕ1(qaq) + q k = 1 2 1 2 a } = kbϕkaq = ka q bϕa Remark 4. According to Theorem 3, for ϕ ∈ Da the normal functional a 2 defined in Theorem 2 coincide as the functionals on M. functional a 2 ϕa 1 1 1 2 ϕa 1 2 from Definition 4 and normal Corollary 4. For each ϕ ∈ Dh a the equality kϕka = lim kϕkaλ holds. λ→+∞ Proof. Evidently, for the fixed ϕ ∈ Dh each ϕ ∈ Dh a there exists lim kϕkaλ ≤ kϕka. λ→+∞ a the mapping λ ∈ R+ 7→ kϕkaλ is monotone and kϕkaλ ≤ kϕka, thus for According to Theorem 8 the chain of equalities kϕka = ka 1 2 ϕa 1 2 k = a 1 2 ϕa 1 holds for each ϕ ∈ Dh a, where ua a a λ ϕa 2 ϕa λ (u) = lim 2 (u) = 2 ϕa lim 2 1 1 2 1 1 1 1 λ→+∞ kϕkaλ, it follows kϕka ≤ lim λ→+∞ 2 a λ (u) ≤ lim λ→+∞ kϕkaλ ≤ kϕka for each ϕ ∈ Dh a. λ ϕa λ→+∞ 2 2 is the polar decomposition of a 1 1 1 1 1 2 ϕa 1 1 1 1 1 2 (1) = a 2 ϕa 2 (u) = a 2 (u) 2 . According to Definition 2, λ k = a. Since ka λ ϕa 2 ϕa 2 1 1 2 ka 2 λ ϕa 2 λ k for each ϕ ∈ Dh Corollary 5. If aηC(M), then ϕ(a) = ka 1 2 ϕa 1 2 k for each ϕ ∈ Dh a, Proof. Note that ϕ(a) ≡ lim λ→+∞ ϕ(aλ) and aλ ∈ C(M). According to [11, Corollary 1], ϕ(aλ) = ka 1 1 2 λ ϕa 2 λ k for each ϕ ∈ M∗. At last, by Corollary 4 lim λ→+∞ ka 2 λ ϕa 2 λ k = lim λ→+∞ 1 1 kϕkaλ = kϕka for each ϕ ∈ Dh a. 7 Summarizing all the facts of this section, it is possible to define k · ka on Dh are equivalent. First and basic, as the mapping ϕ ∈ Dh which actually coincides with the mappings ϕ ∈ Dh a 7→ inf{ lim and ϕ ∈ Dh a 7→ lim inf{ϕ1(aλ) + ϕ2(aλ)ϕ = ϕ1 − ϕ2, ϕ1, ϕ2 ∈ M+ a in the different ways, which a 7→ inf{ϕ1(a) + ϕ2(a)ϕ = ϕ1 − ϕ2, ϕ1, ϕ2 ∈ D+ a }, (ϕ1(aλ) + ϕ2(aλ))ϕ = ϕ1 − ϕ2, ϕ1, ϕ2 ∈ M+ ∗ } ∗ }. Another one as the mappings ϕ ∈ λ→+∞ Dh a 7→ sup kxk=1 lim λ→+∞ λ→+∞ ϕ(a 1 1 2 λ xa 2 λ ), which actually coincides with the mapping ϕ ∈ Dh a 7→ lim 1 1 ϕ(a 2 λ xa 2 λ ). sup kxk=1 λ→+∞ At last if a is affiliated with the center C(M) of an algebra M, then we are able to define k · ka as the mapping ϕ ∈ Dh ϕ(aλ). The latter definition is dual to Segal's definition of k · k1 as the mapping a 7→ lim λ→+∞ x ∈ mτ 7→ τ (x). [12] linRm+ There is one more possible equivalent definition of k·ka for the case of a semifinite M with a faithfull semifinite τ = {x ∈ M+τ (x) < +∞}. We denote τ . By [12, Theorem V.2.18], xeτ is a normal functional for any x ∈ mτ and kxeτ k = kxkτ = τ (x). normal trace τ . Let eτ be the extension of τ onto mτ = linCm+ Theorem 9. For any ϕ = keτ in Da (k ∈ mτ ), if a 2 ∈ M, then the equality kϕka = τ (a τ , where m+ τ as msa 2 ) holds. 2 ka 2 ka 1 1 1 1 1 1 1 a ). For all f ∈ D(a 2 ) the inequality kk 1 2 a 1 2 f k2 ≤ Proof. First prove a 2 ka 2 ∈ mτ . Assume k positive (keτ ∈ D+ 2 ) is dense in H, it follows that k 1 1 2 a 1 2 ∈ M and kk 1 2 a 2 k ≤ qka 1 1 2 ka 1 2 k. ka 1 2 ka 1 2 kkf k2 holds. Since D(a Since lim λ→+∞ k 1 2 a 1 2 λ f = k 1 2 a 1 2 f for all f ∈ D(a 1 2 ) and D(a 1 2 ) is dense in H, it follows that k 1 2 a 1 2 λ f con- verges to k 1 2 a 1 2 f and a 1 2 λ k 1 2 f converges to (k 1 2 a 1 2 )∗f for all f ∈ H. At the same time, lim 1 ha 2 λ k 1 2 f, a 1 2 λ k 1 2 f i = 1 λ→+∞ 2 , hence τ (a 2 ∈ m+ τ . 2 ka 1 1 1 hk τ (k 2 ak 2 a 1 1 2 f, f i, therefore k 2 )∗) = τ (k 2 (k 2 a 2 ak 2 ak 2 = k 2 a 2 ) = lim 1 1 1 1 1 1 1 1 1 2 (k 1 2 a τ (k 1 2 )∗. Moreover, a 2 aλk 2 a 2 ) = kτ (a) < +∞. Therefore, a 2 = (k 2 )∗k 2 ka 2 a 1 1 1 1 1 1 1 λ→+∞ 1 2 λ k 2 k ≤ qka 1 1 1 2 λ k = ka λ xa 1 2 2 λ k λ ) = lim 1 2 For all f ∈ H the chain of inequalities ka 1 2 f k2 = hk 1 2 aλk 1 2 f, f i ≤ hk 1 2 ak 1 2 f, f i = k(k 1 2 a 1 2 )∗f k2 ≤ ka 1 2 ka 1 2 kkf k2 holds. Hence, kk 1 2 a 1 2 ka 1 2 k and a 1 1 1 1 2 λ ka 2 λ σ-weakly converges to a 2 ka 2 ) = a 2 ka 1 1 1 1 1 2 ka 1 2 . Let 1 1 x ∈ mτ , then a dense in M, therefore a keτ (a 2 (x) = lim 2 keτ a 2eτ (x). mτ is σ-weakly By [12, Theorem V.2.18] the equality kkeτ k = τ (k) holds for any k ∈ mτ . Using Theorem 8 kϕka = 2 keτ a 2 keτ a 2eτ k = τ (a λ ) = xeτ (a λ→+∞ 2 = a xeτ (a 2 k = ka 2eτ . λ ka λ→+∞ 2 ka 2 ka 2 ka 2 ). ka 1 2 2 1 1 1 1 1 1 1 1 1 1 2 ka 1 2 ) = τ ((k 1 2 a 1 2 )∗k 1 2 a 1 2 ) = Let Tr be the canonical trace in the space B(H) of bounded linear operators in Hilbert space H and C1(H) denote the space of trace class operators in H. Corollary 6. For any ϕ = kfTr in Da (k ∈ C1(H)) the equality kϕka = Tra Proof. Assume ϕ ∈ D+ a (k ∈ C+ 1 (H)). Since Tr(k 1 1 2 ak 2 ) = lim 1 2 ka 1 2 holds. +∞, it follows that k 1 2 ak 1 2 ∈ C+ 1 (H). Hence, a 1 2 k 1 2 is a Hilbert-Schmidt operator and a If ϕ ∈ Da (k ∈ C1(H)), then there exist k1, k2, k3, k4 ∈ C+ and a 1 2 k1a 1 2 − a 1 2 k2a 1 2 + ia 1 2 k3a 1 2 − ia 1 2 k4a 1 2 = a 1 2 ka 1 2 ∈ C1(H). 1 1 λ→+∞ Tr(k 2 aλk 2 ) = lim kfTr(aλ) = kfTr(a) < 1 (H), such that kfTr = k1fTr−k2fTr+ik3fTr−ik4fTr, 2 ∈ C+ 1 (H). λ→+∞ 2 ka 1 1 a 7→ τ (a for the case of a semifinite M with a faithfull normal trace τ . This result is similar to the result of [13]. The latter Theorem and Corollary make it posible to define k · ka as the mapping keτ ∈ Dh 1 2 ka 1 2 ) 5 Embedding of Normal Weights into L+ 1 (a) and Generation of L1(a) Since our approach is influenced by the theory of noncommutative integration with respect to a weight, one of the natural questions is how our L1-spaces are related to the weights on M. In this section we show that all 8 semifinite weights, for which ϕ(a) < +∞, can be embedded into L+ be represented as the difference of two elements corresponding to embeddings of semifinite normal weights. 1 (a). Moreover, every element of Lh 1 (a) can Definition 5. We write [ (equivalently xq ≥ 0). 1 \ a 2 xa 1 2 ] ∈ L+ ∞(a) and call [ 1 \ a 2 xa 1 2 ] ∈ L∞(a) positive if and only if 1 \ a 2 xa 1 2 ≥ 1 \ a 2 0a 2 1 Definition 6. We write ϕ ∈ (L∗ \ a 2 xa 2 ] ∈ L+ [ ∞(a). 1 1 ∞(a))+ and call ϕ ∈ L∗ \ ∞(a) positive if and only if ϕ([ 2 xa a 1 1 2 ]) ≥ 0 for all For an injective operator a we identify elements of L1(a) with the corresponding elements of L∞(a). By L+ 1 (a) we denote the intersection of (L∗ ∞(a))+ with L1(a). Lemma 4. ϕ ∈ (L∗ ∞(a))+ if and only if a 1 2 ϕa 1 2 ∈ (M∗)+. Proof. Note that [ for all f ∈ D(a 1 1 1 \ 2 xa a \ 2 f i ≥ 0 a 2 xa 2 ) is dense in qH. Also, if xq ≥ 0, then 2 ); which implies hxqf, f i ≥ 0 for all f ∈ qH, since Im(a ∞(a) if and only if xq ≥ 0. Indeed, if 2 , then hxa 2 ] ∈ L+ \ a 2 0a 2 f, xa 2 ≥ 1 1 1 1 1 1 1 1 1 1 1 hxa 1 2 2 q qf i ≥ 0. 2 (x) ≡ ϕ([a 2 f i = hxqa 2 f, a q qf, a Assume ϕ ∈ (L∗ ∞(a))+. Note that if x ∈ M+ then xq ∈ M+ 2 xa 2 ∈ (M∗)+. If [ \ 2 x′a a q , hence [ 2 ]) we deduce, that for all x ∈ M+ the inequality a ∞(a), then xq ∈ M+ 2 (x′) ≥ 0 for all [ 2 ]. Hence, ϕ([ Assume a \ a 2 xa \ 2 xa a \ 2 xa a 2 ] ∈ L+ 2 ]) = a 2 ] = [ 2 ϕa 2 ϕa 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 a 2 ϕa that [ 1 \ a 2 xa 1 2 ] ∈ L+ ∞(a). 1 2 ] ∈ L+ ∞(a). Using the equality 1 \ a 2 xa 2 ϕa 1 1 2 (x) ≥ 0 holds. q and there exists x′ = xq ⊕ 01−q ∈ M+, such Corollary 7. Let ϕ ∈ L∗ ∞(a). ϕ is positive if and only if the equality kϕka = ϕ(a) holds. Proof. If ϕ ∈ L+ 2 (1) = ϕ(a). a 2 ϕa 1 1 ∞(a), then a 1 2 ϕa 1 2 ∈ (M∗)+ and according to Theorem 7 kϕka = ka 1 2 ϕa 1 2 k. Hence, kϕka = Conversely, if kϕka = ϕ(a) then ka 1 2 ϕa 1 2 ka = a 1 2 ϕa 1 2 (1), hence a 1 2 ϕa Lemma 4. 1 2 ∈ (M∗)+ and ϕ ∈ (L∗ ∞(a))+ by Corollary 8. For an injective operator a the isometrical isomorphisms U, U t and V from Corollary 3 preserve the order. Moreover, U (M+) = L+ ∞(a))+) = M∗+, V (L+ ∞(a), U t((L∗ 1 (a)) = M+ ∗ . Theorem 10. For an injective operator a the set D+ a is the dense subset of L+ 1 (a). 1 (a) ⊂ Lh 1 (a) and there exists the sequence {ϕn} ⊂ Dh 1 1 2n . Since Proof. Let ϕ ∈ L+ 1 (a). Evidently, L+ \ ϕka ≤ 1 2 1a a be such that ϕn = ϕ1 ϕ1 2 ∈ L∞(a)(∼= L∗ n − ϕ2 n(a) − kϕka ≤ 1 2 ) = ϕ1 n(a) + ϕ2 \ ϕ1 a n( 2 1a \ 2 ) − ϕ2 a n( 2 1a n(a) − ϕ2 1 1 1 1 \ 1(a)), it follows ϕn( a 2 1a n(a) ≤ kϕnka + 1 n(a) + ϕ2 \ 2n−1 . Therefore, ϕn( 2 1a a n and ϕ1 2 ) + 2ϕ2 1 1 1 1 n(a) → ϕ(a). Hence, kϕ2 1 \ a 2 1a 2 ) → ϕ( 2n . Since kϕn − ϕka ≤ 1 1 2 ) = ϕ(a). Let ϕ1 a, such that kϕn − n ∈ D+ 2n , it follows that n(a) → kϕka = ϕ(a). Evidently, n, ϕ2 a n(a) = ϕ1 nka = ϕ2 n(a) + ϕ2 n(a) → 0, which implies kϕ1 n − ϕka → 0. Let Φ be a weight on M+. It is natural to assume that the embedding ϕ of Φ into L1(a) must be positive, therefore kϕka = Φ(a) < +∞ by Corollary 7. Theorem 11. For an injective operator a any normal weight Φ, such that sup Φ(aλ) ≡ Φ(a) < +∞, λ∈(0,+∞) defines element of L+ 1 (a). 9 Proof. The condition Φ(a) < +∞ along with Lemma 1 implies that Φ is semifinite. According to [10] Φ(x) = ∗ for all x ∈ M+, which implies that Φ(x) is the ωi(x)σ ⊂ I, cardσ ∈ N}, x ∈ M+, ωi ∈ M+ Pi∈J tight upper bound of all finite sums indexed with the elements of σ ⊂ J. ωi(x) = sup{Pi∈σ If operator a is bounded, then it is evident, that Φ(a) = Pi∈J ωi(aλ) ≤ Pi∈J Pi∈σ Pi∈J Φ(aλ) holds, therefore Pi∈σ ωi(a) ≤ Φ(a). Hence, Pi∈J ωi(a) and Φ(a) ≤ Pi∈J ωi(a) = sup σ⊂J, ωi(a). On the other hand, for the finite σ ⊂ J the inequality Pi∈σ ωi(aλ) ≤ ωi(a). If a is unbounded, then Φ(aλ) = ωi(a) ≤ Φ(a), so Φ(a) = Pi∈J ωi(a). Since Φ(a) < +∞, it follows that ωi ∈ D+ a . Also, since ker a = {0 } and a ≥ 0, it follows ωi(a) = 0 if and only if ωi = 0. Without loss of generality, we consider Φ(a) = ωi(a). Consider ϕn := ωi ∈ D+ a . The sequence ϕn is converging in the topology of a-norm. If kϕ − ϕnka → 0 then ϕ ∈ L+ ∞(a) and kϕka = lim n kϕnka = Φ(a). Moreover, for all 1 \ a 2 xa 1 2 ∈ L∞(a) the chain of equalities +∞Pi=1 1 \ a 2 xa 1 ωi( 2 ) = lim n 1 \ a 2 xa 1 ϕn( 2 ) = ϕ( 1 \ a 2 xa 1 2 ) holds. If 1 \ a 2 xa 1 2 ∈ L+ ∞(a) is bounded, then there exists bounded operator xa ∈ M+, such that for all f, g ∈ D(a 2 ) and ϕ( 1 1 \ a 2 xa 1 2 ) = Φ(xa). 1 \ a 2 xa 1 2 (f, g) = hxaf, g) nPi=1 cardσ∈N +∞Pi=1 Corollary 9. Let M = B(H) and a ∈ C+ 1 (H) ⊂ B(H) be injective. Then (i) for any x ∈ B+(H) there exists sequence (xn) in C+ 1 (H), (ii) for any x ∈ Bsa(H) there exists sequence (xn) in C sa 1 (H), (iii) for any x ∈ B(H) there exists sequence (xn) in C1(H), such that kxfTr − xnfTrka → 0; such that kxfTr − xnfTrka → 0; such that kxfTr − xnfTrka → 0. Proof. Since an operator a is bounded, it follows that D+ ∗ . Note that if x ∈ B+(H), then xfTr is normal semifinite weight, such that xfTr(a) < +∞. Applying Theorem 10 and 11 we deduce the statement (i). The statements (ii) and (iii) trivialy follow from (i). a = M+ Theorem 12. Any element ϕ of Lh embeddings of normal semifinite weights into L1(a). 1 (a) can be represented as the difference of two elements L+ 1 (a), which are Proof. Let (ωi) be a sequence in Dh are able to assume that kω1ka + a, such that kωn − ϕka → 0. Passing to the subsequence if it is necessary, we kωi+1 − ωika < ∞. Denoting ϕ1 = ω1 and ϕn = ωn − ωn−1 if n = 2, +∞, we obtain ∞Pi=1 kϕika < ∞, hence k ϕi − ϕka → 0. Now choose ϕ1 n, ϕ2 n such that ϕn = ϕ1 n − ϕ2 n, ϕ1 n, ϕ2 n ∈ D+ a and ϕ1 n(a) + ϕ2 n(a) ≤ kϕnka + 1 2n . The sequences ( ϕ2 i ) are fundamental in the topology of k · ka and nPi=1 ϕ1 i ), ( nPi=1 ∞Pi=1 nPi=1 all elements of these sequences are positive, therefore there exist ϕ1, ϕ2 ∈ L+ (k = 1, 2). For 1 \ a 2 xa \ 2 ∈ L∞(a) the chain of equalities ϕ( a 2 xa 1 1 1 2 ) = lim n→∞ nPi=1 1 \ ϕ2 2 xa a i ( 1 2 )) = lim n→∞ 1 \ ϕ1 a i ( 2 xa 1 2 ) − lim n→∞ nPi=1 1 \ ϕ2 a i ( 2 xa 1 \ 2 ) = ϕ1( a 2 xa \ 2 ) − ϕ2( a 2 xa 1 1 1 1 2 ) holds. 1 (a), such that kϕk − nPi=1 2 ) = lim ( n→∞ \ a 2 xa 1 1 ωn( nPi=1 ϕk i ka → 0 \ ϕ1 a i ( 2 xa 1 1 2 ) − Let Φk(x) := ϕ1 n(x) and Φ2(x) := ϕ2 n(x) for x ∈ M+, k = 1, 2. By [10], Φ1, Φ2 are normal weights. nPi=1 +∞Pn=1 Moreover, Φk(a) = ϕ1 n(a) < +∞ (k = 1, 2). By Lemma 1 Φ1, Φ2 are semifinite and by Theorem 11 elements +∞Pn=1 +∞Pn=1 ϕ1, ϕ2 ∈ L+ ∞(a) are embeddings of Φ1, Φ2, respectively. 10 The latter result is analogue of [14, Theorem 1]. Definition 7. [6] Let Φ be a normal semifinite weight on M. We call it regular if for any ϕ ∈ M+ there exists ω ∈ M+ ∗ (ω 6= 0), such that ω ≤ ϕ and ω ≤ Φ. ∗ (ϕ 6= 0) By [6, Theorem 4] a normal semifinite weight Φ on M is regular if and only if each sesquilinear form in 1 (Φ) is closable in the sense of [15]. By [6, Theorem 6] a normal semifinite weight on B(H) is regular if and L+ only if Φ = kfTr, where k is a positive selfadjoint operator in H, such that it has the bounded inverse operator. Theorem 13. Let dimH = ∞. For an injective operator a in C+ that ψ cannot be represented as an embedding of a normal semifinite weight. 1 (H), there exists an element ψ ∈ L+ 1 (a), such Proof. Let Φ = afTr be a positive normal functional on B(H). From Corollaries 9, 6 and [13], it follows that the mapping x ∈ B(H) 7→ xfTr determines an isometrical isomorphism of L1(Φ) described in [6] onto Lh If each element of L+ 1 (a) can be represented as an embedding of a positive normal semifinite weight, then 1 (a) there exists the correspoding selfadjoint operator kψ ≥ 0, ψ gi) is closable ψ f, k by [16, Theorem 5.12] for each element ψ of L+ such that ψ = kψfTr. The corresponding sesquilinear form ckψ ∈ L+ by [15, Theorem 1.27]. 1 (Φ) (ckψ(f, g) := hk 1 (a) 1 2 1 2 By [6, Theorem 6] the weight Φ is not regular. Hence, by [6, Theorem 4] there exists a positive nonclosable sesquilinear form in L+ 1 (ϕ), so we get a contradiction. 6 Case of C-algebras It is notable that the same approach can be applied to the case of C-algebras. Let A be a C-algebra. By A+, A∗+ we denote its positive cone and the positive cone of its continuous dual. By Asa we denote the set of all selfadjoint operators in A, by A∗h we denote the set of all continuous hermitian functionals on A. For a ∈ A+ we define k · ka on A∗ as the mapping f ∈ A∗ 7→ ka 2 f a 2 k. 1 1 Theorem 14. For all f ∈ A∗h the equality kf ka = inf{f1(a) + f2(a)f = f1 − f2, f1, f2 ∈ A∗+} holds. Proof. Let π be the embedding of A into the universal enveloping von Neumann algebra N . For f ∈ A∗h let ϕ be the corresponding element of N h 2 k. Hence, according to [7, Theorem 2] the equality kf ka = kϕkπ(a) holds. Therefore, ∗ , then kf ka = ka 2 k = kπ(a) 2 ϕπ(a) 2 f a 1 1 1 1 kf ka = inf{ϕ1(π(a)) + ϕ2(π(a))ϕ = ϕ1 − ϕ2, ϕ1, ϕ2 ∈ N + ∗ } = = inf{f1(a) + f2(a)f = f1 − f2, f1, f2 ∈ A∗+} . Theorem 15. k · ka is a norm on A∗ if and only if ϕ(a) > 0 for all ϕ ∈ A∗+ \ {0}. Proof. If ϕ is positive, then ka norm. 1 2 ϕa 1 2 k = a 1 2 ϕa 1 2 (1) = ϕ(a). Hence, if ϕ(a) = 0 and ϕ 6= 0 then k · ka is not a Using the embedding π of A into the universal enveloping von Neumann algebra N we get a positive selfadjoint injective operator π(a) ∈ N . Using Theorem 1 we get that k · kπ(a) is a norm on N∗. Since A∗ ∼= N∗ and from [7, Theorem 2], it follows that k · ka is a norm. 11 For an operator a, such that ϕ(a) > 0 for all ϕ ∈ A∗+ \ {0}, we construct M (a) as the complition of 2 ∈ A∗ is an isometrical isomorphism (A∗, k · ka). Also, according to Theorem 3 the mapping ϕ ∈ M (a) 7→ a of M (a) onto A∗. 2 ϕa 1 1 Remark 5. If an operator a has the bounded inverse operator, then k · ka is a norm on A∗, since ka−1k k · k ≤ k · ka ≤ kakk · k. Also, the latter inequalities imply, that L1(a) coincides with A∗ as the topological vector spaces. 1 The condition ϕ(a) > 0 for all ϕ ∈ A∗+ \{0} has varios interpretations for various C-algebras. If we consider a = (an) ∈ A = c0, then this condition is equivalent to ∀n ∈ N an > 0. If we consider a = (an) ∈ A = c, then this condition implies ∀n ∈ N an > 0 and lim an > 0, hence for operator a = (an) there exists the bounded inverse operator a−1 = ( 1 an ) ∈ c. Remark 6. Assume A is a von Neumann algebra. If an operator a ∈ A satisfies the conditions of Theorem 15, then it also satisfies the conditions of Theorem 1. Therefore, we are able to construct L1(a) and M (a). Evidently, we are able to naturally embed L1(a) into M (a) as a linear subspace. If dim(H) = +∞, then A∗ and A∗ do not coincide. According to Corollary 2 L1(a) and M (a) are isometrically isomorphic to A∗ and A∗, respectively. Therefore, if dim H = +∞, then L1(a) and M (a) do not coincide. Example 1. To give the example of such operator a, that satisfy the condition of Theorem 1, but does not satisfy the condition of Theorem 15, consider A = ℓ∞. Since ℓ∞ is an abelian von Neumann algebra, which acts on the Hilbert space H = ℓ2, we are able to construct L1(a) for an injective operator a. The injectiveness of an operator a is equivalent to the condition, that an > 0 for each n ∈ N. For example, (an) = ( 1 ∞ is injective. According to Remark 5 if a has a bounded inverse operator, then a satisfies the conditions of Theorem 15. Let us prove that if a satisfies the conditions of Theorem 15, then a has the bounded inverse operator. If we assume the contrary, then eather there exists an = 0, or there exists a subsequence ank , such that lim ank = 0. Evidently, for each an there exists a functional ϕn ∈ ℓ∗+ ∞ there exists a Banach limit ϕa ∈ ℓ∗+ an > 0. Therefore, a has the bounded inverse operator. However, ( 1 ∞ , such that ϕn(a) := an, therefore ∀n ∈ N an > 0. Also for each a ∈ ℓ+ n ) does not have a bounded inverse operator, therefore it does not satisfy the conditions of ∞ , such that ϕa(a) = lim inf n→∞ an. Hence, lim inf n→∞ n ) ∈ ℓ+ Theorem 15. Example 2. To give a noncommutative example, assume A = B(H) and dimH = ∞, then there exists an injective positive trace-class operator a ∈ C+ 1 (H). For any injective operator a we are able to construct L1(a), but for any trace-class operator there exists a Dixmier trace ϕ ∈ B∗+(H), for which ϕ(a) = 0. Therefore, such operator a does not satisfy the conditions of Theorem 15. Acknowledgment Research supported in part by Russian Foundation for Basic Research grant 14-01-31358. I would like to thank Dr. Oleg Tikhonov from Kazan Federal University for his expert advice (especially on the Theorem 5) and encouragement. He kindly read my paper and offered invaluable detailed advices on grammar, organization, and the theme of the paper. References [1] I. E. Segal, A non-commutative extension of abstract integration, Ann. Math. 57(3) (1953) 401 -- 457. [2] R.P. Kostecki, W-algebras and noncommutative integration, preprint http: // arxiv. org/ abs/ 1307. 4818 (2014). 12 [3] A. N. Sherstnev, States on von Neumann algebras, Funct. Anal. Appl. 8(3) (1974) 272 -- 273; Translated from Russian: К общей теории состояний на алгебрах фон Неймана, Функц. анализ и его прил., 8(3) (1974) 89 -- 90. [4] A. N. Sherstnev, A non-commutative analogue to the space L1, Rus. Math. Surveys 33(1) (1978) 217 -- 218; Translated from Russian: О некоммутативном аналоге пространства L1, УМН, 33(1) (1978) 231 -- 232. [5] A. N. Sherstnev, On the general theory of the measure and integral in von Neumann algebras, Sov. Math. (Iz. VUZ) 26(8) (1982) 21 -- 40; Translated from Russian: К общей теории меры и интеграла в алгебрах Неймана, Изв. вузов. Матем., 26(8) (1982) 20 -- 35. [6] N. V. Trunov, A. N. Sherstnev, Introduction to the theory of noncommutative integration, Math. Sci., 37(6) (1987) 1504 -- 1523; Translated from Russian: Введение в теорию некоммутативного интегрирования Итоги науки и техн. Сер. Соврем. пробл. мат. Нов. достиж., 27 (1985), 167 -- 190. [7] G. Sh. Skvortsova, O. E. Tikhonov, Convex sets in noncommutative L1-spaces that are closed in the topol- ogy of local convergence in measure, Russian Math. (Iz. VUZ) 42(8) (1982) 21 -- 40; Translated from Rus- sian: Выпуклые множества в некоммутативных L1-пространствах, замкнутые в топологии локальной сходимости по мере, Изв. вузов. Матем., 42(8) (1998) 48 -- 55. [8] W. Rudin, Functional Analysis (McGraw-Hill, Inc., 1991). [9] N. V. Trunov, A.N. Sherstnev On the general theory of integration with respect to a weight in algebras of operators. I, Sov. Math. (Iz. VUZ), 22(7) (1978) 79 -- 88; Translated from Russian: К общей теории интегрирования в алгебрах операторов относительно веса, I, Изв. вузов. Матем., 22(7) (1978), 65 -- 72. [10] U. Haagerupp, Normal weights on W*-algebras J. Funct. Anal. 19(3) (1975) 302 -- 317. [11] A. A.Novikov, O. E. Tikhonov, Characterization of central elements of operator algebras by inequalities, Lobachevskii J. Math., 36(2) (2015) 208 -- 210. [12] M. Takesaki, Theory of operator algebras I (Springer-Verlag, 2002). [13] G.D. Lugovaya, A.N. Sherstnev, Realization of the space L1 with respect to an unbounded measure on projectors, Sov. Math (Iz. VUZ), 28(12) (1984) 42 -- 50; Translated from Russian: О реализации пространства L1 относительно неограниченной меры на проекторах, Изв. вузов. Матем., 28(12) (1984) 35 -- 42. [14] O. E. Tikhonov, Integrable bilinear forms and an integral over an operator-valued measure, Sov. Math. (Iz. VUZ), 26(3) (1982) 94 -- 100; Translated from Russian: Интегрируемые билинейные формы и интеграл по операторнозначной мере, Изв. вузов. Матем., 26(3) (1982) 76 -- 80. [15] T. Kato, Perturbation Theory for Linear Operators (Springer-Verlag, 1980). [16] G. K. Pedersen, M. Takesaki, The Radon-Nikodym theorem for von Neumann algebras, Acta. Math. 130(1) (1973) 53 -- 87. 13
1711.08802	1	1711	2017-11-23T18:56:30	Poncar\'e half-space of a C*-algebra	[ "math.OA" ]	Let $A$ be a C$*^$-algebra. Given a representation $A\subset B(L)$ in a Hilbert space $L$, the set $G^+\subset A$ of positive invertible elements can be thought as the set of inner products in $L$, related to $A$, which are equivalent to the original inner product. The set $G^+$ has a rich geometry, it is a homogeneous space of the invertible group $G$ of $A$, with an invariant Finsler metric. In the present paper we study the tangent bundle $TG^+$ of $G^+$, as a homogenous Finsler space of a natural group of invertible matrices in $M_2(A)$, identifying $TG^+$ with the {\it Poincar\'e halfspace} $H$ of $A$, $$ H=\{h\in A: Im(h)\ge 0, Im(h) \hbox{ invertible}\}. $$ We show that $\h\simeq TG^+$ has properties similar to those of a space of non-positive constant curvature.	math.OA	math	Poincar´e half-space of a C∗-algebra E. Andruchow, G. Corach, L. Recht October 8, 2018 Abstract Let A be a C∗-algebra. Given a representation A ⊂ B(L) in a Hilbert space L, the set G+ ⊂ A of positive invertible elements can be thought as the set of inner products in L, related to A, which are equivalent to the original inner product. The set G+ has a rich geometry, it is a homogeneous space of the invertible group G of A, with an invariant Finsler metric. In the present paper we study the tangent bundle T G+ of G+, as a homogenous Finsler space of a natural group of invertible matrices in M2(A), identifying T G+ with the Poincar´e halfspace H of A, H = {h ∈ A : Im(h) ≥ 0, Im(h) invertible}. We show that H ≃ T G+ has properties similar to those of a space of non-positive constant curvature. 2010 MSC: 46L05, 58B20, 22E65, 46L08 Keywords: Positive invertible operator, inner product. 1 Introduction Let A be a unital C∗-algebra, G the group of invertible elements in A, G+ the subset of G of positive elements. Observe that G+, as an open subset of As := {x ∈ A : x∗ = x}, is an open submanifold of As and its tangent space at any point is identified with As. If Gs = G ∩ As, then it can be proven that G+ is the component of the identity in Gs. Also, there is a left action of G on Gs given by g · a = (g−1)∗ag−1, and G+ is the orbit of 1 under this action. With this dual nature, G+ carries a natural structure of homogeneous space of G, with a linear connection, and a Finsler metric. These facts, and many others concerning the differential geometry of G+, have been studied in [6], [8], [9]. The goal of the present paper is the study of the tangent bundle T G+, in particular its presentation as a homogeneous space of the unitary group of a natural quadratic form in A2 = A × A. There are several motivations for this study. We consider G+ as the configuration space of a quantum mechanical system whose elements represent a family of equivalent metrics over a Hilbert space. The tangent bundle T G+ is the phase space of the configuration space G+ of the quantum system. Note also that T G+ identifies in a natural way with the bundle of observables associated with the different metrics. We shall explain this with more detail below. 1 Note the correspondence (a, X) ←→ X + ia, where a ∈ G+ and X ∈ (T G+)a; here X is a selfadjoint element of A, because G+ is open in the space of selfadjoint elements of A. This correspondence establishes a clear identification between T G+ and H, the Poincar´e half-space of A, H = {h ∈ A : Im(h) ∈ G+}. Following ideas from C.L. Siegel [19], [20] and [21], we claim that there is a natural form θH in A × A = A2, determined by H, whose unitary group U (θH) acts transitively in H. Namely, if we put projective coordinates (cid:18) x1 1 ∈ H are characterized by the 1 2i{x2x−1 x2 (cid:19) ∈ A2, elements h = x2x−1 1 − (x∗1)−1x∗2} = (x∗1)−1{x∗1x2 − x∗2x1}x−1 1 ∈ G+, condition Im(h) = 1 2i or equivalently 1 2i{x∗1x2 − x∗2x1} ∈ G+. Thus, if we put 1 i {x∗1y2 − x∗2y1}, x2 (cid:19) ,(cid:18) y1 x2 (cid:19) ,(cid:18) x1 y2 (cid:19)) = x2 (cid:19)) ∈ G+. θH((cid:18) x1 the condition Im(h) ∈ G+ is θH((cid:18) x1 The unitary group U (θH) of θH, i.e., the group of invertible matrices in M2(A) which pre- serve θH, acts transitively on H, and makes H a homogeneous space. Therefore, T G+ can be considered as the phase space of the mentioned quantum system, with the group U (θH ) of symmetries. The homogenous space H identifies in turn, also in a natural fashion, with the space QρH of selfadjoint projections in A2 which decompose the quadratic form θH (where ρH is the reflection in A2 induced by θH). This identification is equivariant with the respective actions of the symmetry group U (θH). The space Qρ, for arbitrary symmetries ρ, was studied thoroughly in [7]; we can deduce the geometric and metric properties of T G+ from the properties established for QρH . In a forthcoming second part of this paper, we shall study additional geometric structures related to a pre-quantization of T G+, in the framework of a Hilbert-C∗-module structure. We also establish a natural bijection between T G+ and the space D of strict contractions of D = {a ∈ A : kak < 1}. KH = {(cid:18) x1 Besides the form θH and the group UU (θH) of invertible elements in M2(A) which leave θH invariant, an important role will be played by the unit sphere KH, x2 (cid:19) ,(cid:18) x1 x2 (cid:19) ∈ A2 : θH((cid:18) x1 The sphere KH will serve the role of a coordinate space for H, and these data will be related by the commutative diagram of U (θH)-homogeneous spaces "❊❊❊❊❊❊❊❊ x2 (cid:19)) = 1}. KH ~⑥⑥⑥⑥⑥⑥⑥⑥ ¯ϕ ϕ A, H / QρH ΦH 2 ~ " / where ϕ and ϕ are submersions and ΦH is a diffeomorphism. There is an analogous commutative triangle for the open unit disk model, by means of the form θD, θD((cid:18) x1 x2 (cid:19) ,(cid:18) y1 y2 (cid:19)) = x∗1y1 − x∗2y2. Both forms (and therefore both sets of data) are covariantly related by the unitary matrix in M2(A) 1 U = 1 √2(cid:18) 1 i −i (cid:19) . For instance, the bijection between H and D is a Moebius transformation. Let us summarize the contents of the paper. In Section 2 we define the space H, the form θH, the unitary group U (θH) and the sphere KH of this form. In Section 3 we introduce an alternative model for H, namely the open unit disk D, along with the form θD and the unitary group U (θD) and sphere KD of this form. The spaces, forms, groups and unit spheres of both models are intertwined by the unitary matrix U ∈ M2(A). The reason to have two models for the same space, is that some computations are easier or more natural with one or the other. For instance, in Section 4 we study the local (Banach-Lie group) structure of U (θH). An important role is played by the subgroup B ⊂ U (θH), which we call the Borel subgroup of U (θH), and which has a natural meaning in this setting. In Section 5, we introduce the action of U (θD) on the unit sphere KD, which implies that also U (θH) acts on KH (this is another example, where a fact is easier to establish in model than the other). These actions enable us to introduce the actions of U (θH) and U (θD) on H and D, respectively. It is shown that these actions are transitive, and the isotropy subgroups are computed. In Section 7, we recall from [7] the space Qρ of signed decompositions of a form induced by a symmetry ρ, and prove that there are natural diffeomorphisms between Qρ, H and D, which are equivariant with respect to the corresponding group actions. This is a key fact, which allows us to import from Qρ to H and D the main geometric features of that space: a linear connection, a Finsler metric and its properties. Among these, that H behaves as a non-positively curved metric length space [12]. In Section 9 we compute the specific form of the linear connection induced in H. In Section 10 we compute special cases of geodesics in H. Finally, in Section 11, as an Appendix, we outline an intrinsic, coordinate free, version for the Poncar´e half-space, in terms of a Hilbertizable space endowed with a coherent family of inner products. 2 Poincar´e halfspace We define the following forms in A2: Definition 2.1. The A-valued inner product: a2 (cid:19) ,(cid:18) b1 <(cid:18) a1 b2 (cid:19) >= a∗1b1 + a∗2b2 and the A-valued simplectic form ω((cid:18) a1 a2 (cid:19) ,(cid:18) b1 b2 (cid:19)) = a∗2b1 − a∗1b2. 3 Let us denote Then, it is apparent that 1 J =(cid:18) 0 −1 0 (cid:19) ∈ M2(A). ω((cid:18) a1 a2 (cid:19) ,(cid:18) b1 b2 (cid:19)) =< J(cid:18) a1 a2 (cid:19) ,(cid:18) b1 b2 (cid:19) > . We shall denote by a, b, c, etc. the elements of M2(A). Let us denote by Gl2(A) the group We shall use the selfadjoint reflection ρH (i.e., ρ2 of invertible elements and by U2(A) the group of unitary elements in M2(A). H = 1, ρ∗H = ρH, ), ρH = −iJ =(cid:18) 0 −i 0 (cid:19) . i and the form θH θH((cid:18) x1 x2 (cid:19) ,(cid:18) y1 y2 (cid:19)) =< ρH(cid:18) x1 x2 (cid:19) ,(cid:18) y1 y2 (cid:19) >= 1 i (x∗1y2 − x∗2y1), i.e., θH = iω. The following group, which is the group of invertible matrices in M2(A) which preserve the form θH (equivalently, the form ω), will play an important role in this study: U (θH ) = {a ∈ Gl2(A) : θH(a(cid:18) b1 for all (cid:18) b1 b2 (cid:19) , a(cid:18) c1 b2 (cid:19) ,(cid:18) c1 c2 (cid:19)) = θH((cid:18) b1 c2 (cid:19) ∈ A2}. b2 (cid:19) ,(cid:18) c1 c2 (cid:19)), (1) Clearly, a ∈ U (θH) if and only if ρH a∗ρH = a−1. Definition 2.2. Let KH ⊂ A2 be the set KH = {(cid:18) a1 a2 (cid:19) ∈ A2 : θH((cid:18) a1 a2 (cid:19) ,(cid:18) a1 a2 (cid:19)) = 1, with a1 ∈ G}, i.e., (cid:18) a1 a2 (cid:19) ∈ KH if 1 i (a∗1a2 − a∗2a1) = 2Im(a∗1a2) = 1. Notice that a2 ∈ G, automatically. space for H. We shall use this hyperboloid KH to understand the geometry of H; it shall be a coordinate There is a natural fibration of KH over H: ϕH : KH → H, ϕH ((cid:18) x1 x2 (cid:19)) = x2x−1 1 . 4 3 The open unit disk model for T G+ We shall study an alternative model for H (i.e., for T G+). The main reason to give this alternative version is that many computations will be clearer (and easier) in this model. Let D be the open unit ball of A: D = {z ∈ A : z∗z < 1} = {z ∈ A : kzk < 1}. Consider the selfadjoint reflection ρD: 0 ρD =(cid:18) 1 0 −1 (cid:19) . The induced A-valued indefinite inner product on A2 y2 (cid:19)) =< ρD(cid:18) x1 x2 (cid:19) ,(cid:18) y1 θD((cid:18) x1 x2 (cid:19) ,(cid:18) y1 y2 (cid:19) >= x∗1y1 − x∗2y2. The group U (θD) of elements in Gl2(A) which preserve the form θD: y2 (cid:19)) = θD((cid:18) x1 U (θD) = {a ∈ Gl2(A) : θD(a(cid:18) x1 Equivalently, a ∈ U (θD) if ρDa∗ρD = a−1. x2 (cid:19) , a(cid:18) y1 x2 (cid:19) ,(cid:18) y1 y2 (cid:19))}. Before proceeding any further, note that ρD and ρH are conjugate via the unitary element U = i.e., 1 1 √2(cid:18) 1 i −i (cid:19) , (2) (3) Therefore, the groups U (θH) and U (θD) are conjugate, and the properties and features of The sphere KD U (θH) are translated to U (θD). U ρDU∗ = ρH. KD := {(cid:18) x1 = {(cid:18) x1 x2 (cid:19) ,(cid:18) x1 x2 (cid:19) ∈ A2 : θD((cid:18) x1 x2 (cid:19) ∈ A2 : x∗1x1 − x∗2x2 = 1, x1 ∈ G}. x2 (cid:19)) = 1, x1 ∈ G} There is an analogous fibration of KD over D: ϕD : KD → D, ϕD((cid:18) x1 x2 (cid:19)) = x2x−1 1 . In fact, if (cid:18) x1 x2 (cid:19) ∈ KD, then x∗1x1 = 1 + x∗2x2, 5 Let us check the invertibility conditions. If(cid:18) x1 x2 (cid:19) is invertible. Equivalently, that 1 + x2x−1 x2 (cid:19) ∈ KD, then x1 ∈ G. We must verify that 1 ∈ G. 1 = (x1 + x2)x−1 the first coordinate of U(cid:18) x1 Since (cid:18) x1 x2 (cid:19) ∈ KD, it follows that x2x−1 z2 (cid:19) = U(cid:18) x1 Conversely, suppose that(cid:18) z1 invertible. 1 ∈ D, and thus kx2x−1 1 k < 1. Then 1 + x2x−1 x2 (cid:19) ∈ KH. Then z1, z2 are invertible, and we must is 1 and it follows that thus i.e., x2x−1 1 ∈ D 1 = (x−1 1 )∗(1 + x∗2x2)x−1 1 = (x1x∗1)−1 + (x2x−1 1 )∗x2x−1 1 ; (x2x−1 1 )∗x2x−1 1 < 1, x2 (cid:19) ∈ KD if and only if U(cid:18) x1 Lemma 3.1. (cid:18) x1 Proof. The fact that U ρDU∗ = ρH implies that the equation θD((cid:18) x1 equivalent to θH(U(cid:18) x1 x2 (cid:19) , U(cid:18) x1 x2 (cid:19) ∈ KH . x2 (cid:19)) = 1. x2 (cid:19) ,(cid:18) x1 x2 (cid:19)) = 1 is check that z1 − iz2 is also invertible. The fact that (cid:18) z1 Therefore a straightforward computation shows that −i /∈ σ(z2z−1 iz2)z−1 1 ∈ G, i.e., z1 − iz2 ∈ G z2 (cid:19) ∈ KH implies that Im(z2z−1 1 ). Then z2z−1 1 ) > 0. 1 + i = i(z1 − 4 The groups U (θH) and U (θD) We shall describe in the next subsections the basic properties shared by U (θH) and U (θD). Some computations are easier or more natural in one of the two presentations of these isomorphic groups. 4.1 The unitary group U(θH ) of the form θH In order to study U (θH), we consider the following subgroups Definition 4.1. Let B be the group of elements b ∈ U (θH) which are of the form and T consisting of c, with τ∗ = τ . 0 b21 b =(cid:18) b11 c =(cid:18) 1 τ b22 (cid:19) 0 1 (cid:19) , 6 It is apparent that B, which we will call the Borel subgroup of U (θH), is indeed a group. Also it is clear that T is a group. Note that J c∗J−1 =(cid:18) 0 −1 0 (cid:19)(cid:18) 1 0 1 (cid:19)(cid:18) 0 −1 0 (cid:19) =(cid:18) 1 −τ 1 (cid:19) = c−1, 1 τ 1 0 i.e., c ∈ U (θH). We shall see that these groups are complemented Banach-Lie subgroups of Gl2(A). An elementary computation shows that U (θH) is closed under the involution of M2(A): if a ∈ U (θH) then a∗ ∈ U (θH ). It follows that the factors of the polar decomposition of a = ua remains inside U (θH): u,a ∈ U (θH). It suffices to show that a ∈ U (θH). Clearly a2 = a∗a ∈ U (θH). Note that both JaJ−1 and a−1 are positive elements with the same square: (JaJ−1)2 = Ja2J−1 = J a∗aJ−1 = (a∗a)−1 = a−2. Then JaJ−1 = a−1. The same is true for the other polar decomposition a = a∗ w. Note also the elementary fact that a unitary element u belongs to U (θH) if and only if it We shall denote by UU (θH) the (subgroup of) unitary elements of U (θH), and by U (θH)+ commutes with J. Similarly, a positive element b ∈ U (θH) if and only if JbJ−1 = b−1. the set of positive elements in U (θH). Proposition 4.2. The group U (θH) is a C∞ Banach-Lie group, and a complemented subman- ifold of M2(A). Proof. Let us exhibit a local chart for 1 ∈ U (θH). Denote by M2(A)s and M2(A)as the spaces of selfadjoint and anti-selfadjoint elements of M2(A). Consider the space X = Xas ⊕ Xs = { β ∈ M2(A)as : βJ = J β} ⊕ {γ ∈ M2(A)s : γJ = −J γ}. (4) Elements X ∈ X are of the form X = β + γ, β12 γ12 −γ11 (cid:19) with β∗11 = −β11, and all other entries selfadjoint. Consider the map −β12 −β11 (cid:19) +(cid:18) γ11 X =(cid:18) β11 γ12 E : X → U (θH), E( β + γ) = e βeγ. Note that e β is a unitary element which commutes with J, i.e., e β ∈ UU (θH). The element eγ is a positive invertible element of M2(A); the fact that the exponent γ anticommutes with J means that eγJ = Je−γ, i.e., eγ ∈ U (θH)+, and thus E is well defined. If one restricts E to V = { β ∈ Xas : k βk < π} ⊕ Xs, then EV is a homeomorphism onto W = {a ∈ U (θH) : ku − 1k < 2, where u = aa−1}. Clearly, V and W are open sets of X and U (θH), respectively. If a ∈ W, the fact that ku−1k < 2 implies that u = eδ for a unique δ ∈ M2(A)as with kδk < π (and thus δ is a series in powers 7 of u). Since u is unitary and belongs to U (θH ), it commutes with J. Then, its logarithm δ commutes with J, i.e., δ ∈ X1. On the other hand, since a is positive and invertible, it has a unique selfadjoint logarithm log(a) = ǫ. The fact that JaJ−1 = a−1 implies that J ǫJ−1 = J log(a)J−1 = log(JaJ−1) = log(a−1) = − log(a) = −ǫ, i.e., ǫ ∈ X2, and thus a = E(δ, ǫ). The inverse of E is E−1 : W → V, E−1(a) = log(aa−1) + log(a) ∈ X1 ⊕ X2. It is apparent that both E and E−1 are C∞ maps. The Banach space X on which the neighbour- hood W of 1 in U (θH) is modelled, is complemented in M2(A). Indeed, X can be also presented as X = {X ∈ M2(A) : X =(cid:18) x11 x21 −x∗11 (cid:19) , with x12, x21 ∈ As}. x12 A supplement for X is, for instance: {(cid:18) a b c a∗ (cid:19) : b∗ = −b, c∗ = −c}. Charts around other elements of U (θH) are obtained by translation, using the left action of U (θH) on itself. Remark 4.3. The differential at the origin of the map E is the identity. It follows that the Banach-Lie algebra µ(θH) of U (θH) coincides with X : X ∈ µ(θH) if X =(cid:18) x11 x21 −x∗11 (cid:19) , x12 where x12 and x21 are selfadjoint. Proposition 4.4. B and T are Banach-Lie subgroups of U (θH), and complemented submanifolds of M2(A). They generate an open and closed subgroup of U (θH), which contains the connected component of the identity. Proof. First note that the diagonal entries of elements in B must be invertible elements in A. Elementary matrix computations show that b ∈ B if and only if it is of the form 0 b =(cid:18) b x (b∗)−1 (cid:19) , with b∗xb−1 = x∗, or, equivalently, b∗x selfadjoint. Thus, B can be parametrized by B = {(b, x) : b invertible in A and b∗x ∈ As}. This set B is globally diffeomorphic to the set G × As = {(a, y) : a invertible in A, y = y∗}. 8 which is a complemented submanifold of A2. The diffeomorphism and its inverse are given by (b, x) 7→ (b, b∗x) and (a, y) 7→ (a, (a∗)−1y). Thus, B is globally diffeomorphic to a submanifold of A2. Moreover, this diffeomorphism extends to an open subset of M2(A): c M2(A) = {(cid:18) a y + z d (cid:19) : (a, y) ∈ G × As, z, c, d ∈ A, z∗ = −z} → M2(A), (cid:18) a y + z d (cid:19) 7→(cid:18) which maps G × As onto B. The Banach-Lie algebra aB of B can be computed using this parametrization. If b(t) ∈ GA and x(t) ∈ A are smooth curves such that b∗(t)x(t) is selfadjoint, b(0) = 1, b(0) = y11, x(0) = 0 and x(0) = y21, then (a∗)−1 + d (cid:19) (a∗)−1y + z a c c is a smooth curve in B with b(0) = 1 and b(t) =(cid:18) b(t) 0 (b∗(t))−1 (cid:19) x(t) b(0) = Y , Y =(cid:18) y11 y21 −y∗11 (cid:19) , 0 dt{b∗(t)x(t)} = b∗(t)x(t) + b∗(t) x(t) ∈ As; in particular, at t = 0, this implies that y21 is where d selfadjoint. Thus, the Banach-Lie algebra aB of B is 0 aB = {Y =(cid:18) y11 y21 −y∗11 (cid:19) : y∗21 = y21}. The subgroup T is parametrized by As, and can be proved to be a submanifold of M2(A) in a similar (simpler) fashion. Also, it is apparent that its Banach-Lie algebra aT of T is aT = {Z =(cid:18) 0 z12 0 (cid:19) : z∗12 = z12}. 0 We claim that the subgroups B and T generate a closed and open subgroup of U (θH). To this effect, note that the Banach-Lie algebras of these groups are in direct sum, and its sum is the Banach-Lie algebra µ(θH) of UU (θH ): µ(θH) = aB ⊕ aT . It follows that there is a neighbourhood W of 1 in U (θH) where any element is the product of elements in B and T . Let a0 ∈ U (θH) which is a product of elements in B and T . Then U0 = {a : a−1 0 a ∈ U} is an open neighbourhood of a0 in U (θH). It follows that the set of these products is an open subgroup. The relation a ∼ b if and only if a−1b ∈ BT is an equivalence relation. It follows that this subgroup is a union of connected components of U (θH), containing the connected component of the identity. 9 Remark 4.5. If A is a von Neumann algebra, then U (θH ) is connected. Since U (θH)+ is clearly connected (in fact, contractible), one needs to show that U2(A) ∩ {J}′ is connected. Since J is anti-selfadjoint, it follows that {J}′ ⊂ M2(A) is a von Neumann algebra, and therefore U2(A) ∩ {J}′ is the unitary group of a von Neumann algebra, thus connected. Theorem 4.6. The unitary part U2(A) ∩ {J}′ of U (θH) is isomorphic to UA × UA. The group Π0(U (θH )) of connected components of U (θH), is isomorphic to Π0(UA) × Π0(UA). Proof. Consider the map {J}′ → A, (cid:18) a −b a (cid:19) 7→ a + ib. b This map is an injective C∗-homomorphism; thus, its restriction to U2(A) ∩ {J}′ Γ : UU (θH ) → UA is a group homomorphism. It is a retraction: the map u 7→(cid:18) u 0 0 u (cid:19) is a cross section for Γ and a group homomorphism. By straightforward computations, the kernel of Γ consist of matrices (cid:18) a −i(a − 1) i(a − 1) a (cid:19) with a∗a = aa∗ = 1 2 (a + a∗). Then a is a normal element, which is a zero of the continuous function f (z) = z2 − Re(z). Therefore, the spectrum of a is contained in the zero set of f , namely {z ∈ C : z − 1 2}. The map z 7→ 2z − 1 sends this circle to the unit circle, and since it is a polynomial map, it sends elements a as above onto normal elements with spectrum in the unit circle, i.e., unitary elements of A. Conversely, if u ∈ UA, elementary computations show that a = 1 2 (a + a∗). Moreover, it is easy to verify that the map 2 (u + 1) satisfies a∗a = aa∗ = 1 2 = 1 ker Γ → UA, (cid:18) a −i(a − 1) i(a − 1) a (cid:19) 7→ 2a − 1 is a group homomorphism, thus a bicontinuous isomorphism. Therefore, since Γ splits, one has that (by means of an explicit isomorphism) UU (θH) ≃→ UA × A. The polar decomposition induces the isomorphism between Π0(U (θH)) and Π0(UU (θH)), because the positive part U +(θH) is contractible. Corollary 4.7. B and T generate U (θH) if and only if UA is connected. 10 5 The unitary group U (θD) of the form θD As remarked above, U (θD) and U (θH) are conjugate via the unitary operator U given in (2). Therefore the same properties proved for U (θH) also hold for U (θD). Thus, UU (θD) is a Banach- Lie subgroup of Gl2(A) and it is closed under the polar decomposition: if a ∈ U (θD) and a = ua is its polar decomposition, then u,a ∈ U (θD). The fact that the unitary part u belongs to U (θD) means that U (θD) commutes with ρD. It is elementary that this implies that the matrix of u is of the form u =(cid:18) u1 0 0 u2 (cid:19) , with u1, u2 in UA. Denote by λ = log a the unique selfadjoint logarithm of the (positive invertible) element a. The fact that ρa = a−1ρD, means that ρD λ = −λρD. On the other hand, positive elements r ∈ U (θD) satisfy that ρD rρD r = 1. In particular, if r12 r22 (cid:19) , r∗12 r =(cid:18) r11 11 − r12r∗12 = 1. x2 (cid:19) ∈ KD. then r11 ≥ 1, because r11 ≥ 0 and r2 The main issue in introducing this description of U (θD) is the following result: Theorem 5.1. U (θD) acts on KD by left multiplication: if a ∈ U (θD) and (cid:18) x1 a(cid:18) x1 Proof. Recall that (cid:18) x1 x1 ∈ G. Since a preserves θD, it is clear that θD(a(cid:18) x1 that the first coordinate of a(cid:18) x1 separately for the unitary part u and the absolute value a. The first assertion is clear: x2 (cid:19) ,(cid:18) x1 x2 (cid:19)) = 1 and that x2 (cid:19)) = 1. We must show x2 (cid:19) is invertible in A. Clearly, it suffices to prove this fact x2 (cid:19) = (x1, x2) ∈ KD means that θD((cid:18) x1 x2 (cid:19) , a(cid:18) x1 x2 (cid:19) ∈ KD, then u(cid:18) x1 x2 (cid:19) =(cid:18) u1 0 0 u2 (cid:19)(cid:18) x1 x2 (cid:19) =(cid:18) u1x1 u2x2 (cid:19) , and u1x1 is invertible. For the second assertion, we claim that if r is a positive element in U (θD), r =(cid:18) r11 r∗12 r12 r22 (cid:19) x2 (cid:19) ∈ which satisfies that kr12k < 1, then the first coordinate of r(cid:18) x1 KD. Indeed, the first coordinate of this product is r11x1 + r12x2. Since x1 is invertible, this sum x2 (cid:19) is invertible for any(cid:18) x1 11 is invertible if and only if r11 +r12x2x−1 1 that kr12x2x−1 invertible. 1 k < 1. Recall from above that r11 ≥ 1. These facts imply that r11 + r12x2x−1 1 is invertible. Since kx2x−1 1 k < 1 and kr12k < 1, it follows is 1 Note that for any n ≥ 1, a1/n = e n λ ∈ U (θD), because 1 n λ anti-commutes with ρD. Note also that a1/n → 1 as n → ∞. Thus, there exists n ≥ 0 such that r = a1/n satisfies that kr12k < 1. It follows from the above observation, that for any (cid:18) x1 x2 (cid:19) ∈ KD. Then, x2 (cid:19) = a1/n(. . . (a1/n(cid:18) x1 x2 (cid:19) ∈ KD, a1/n(cid:18) x1 x2 (cid:19)) ∈ KD. a(cid:18) x1 inductively, Therefore: Corollary 5.2. U (θH) acts on KH by left multiplication. 6 The actions of U (θD) and U (θH) on D and H In this section we prove that U (θH) acts in T G+ (in fact, we prove this fact for the disk model D of T G+). First let us note that there is the natural inmersion G+ ֒→ T G+, a 7→ 0 + ia. G+ is a homogeneous space of G, with the left action g · a = (g∗)−1ag−1. This action is the main feature in studying the geometry of G+. We shall see that it can be regarded as a restriction of the action of UU (θH) on T G+ (see [6], [8], [9]). Note that G is a subgroup of UU (θH ), via the injective group homomorphism G ֒→ UU (θH) , g 7→(cid:18) g 0 (g∗)−1 (cid:19) . 0 In order to introduce the actions of U (θD) and U (θH ) on D and H, respectively, we need the maps: and Definition 6.1. ϕD : KD → D , ϕ(x1, x2) = x2x−1 1 , ϕH : KH → H , ϕ(x1, x2) = x2x−1 1 . • Given z ∈ D, consider (1, z) and compute θD((1, z), (1.z)) = 1 − z∗z. Since kzk < 1, this element is positive and invertible. Then (1 − z∗z)−1/2, z(1 − z∗z)−1/2) ∈ KD. If a =(cid:18) a11 a12 a21 a22 (cid:19) ∈ U (θD), put a · z := ϕD(a(cid:18) (1 − z∗z)−1/2 z(1 − z∗z)−1/2) (cid:19)) = (a21 + a22z)(a11 + a12z)−1. • Analogously, given h ∈ H, θH((1, h), (1, h)) = 2 Im(h). Then ( 1√2 Im(h)−1/2, 1√2 a · h := ϕH (a 1√2 1√2 hIm(h)−1/2) ∈ KH. For a =(cid:18) a11 a12 a21 a22 (cid:19) ∈ U (θH ), put hIm(h)−1/2 !) = (a21 + a22h)(a11 + a12h)−1. Im(h)−1/2 12 Remark 6.2. Both actions are well defined, and they are, indeed, left actions. Let us prove that these left actions are transitive. More specifically: Proposition 6.3. The action of the subgroup B on H is transitive. Proof. Let h ∈ H and b ∈ B, b =(cid:18) b x (b∗)−1 (cid:19) , 0 with b∗x selfadjoint. Then: b · h = xb−1 + (b∗)−1hb−1 = (b∗)−1(b∗x + h)b−1. The map x 7→ (b∗)−1xb−1 is a linear isomorphism in A which preserves positivity (and, thus, selfadjointness). Thus, Im(b · h) = (b∗)−1Im(h)b−1. Fix h ∈ H and let h′ be another element in H. Put z = Im(h) and z′ = Im(h′); both are in G+. The action of G on G+ is transitive, thus there exists g ∈ G such that z′ = (g∗)−1zg−1. Put y = h′g − (g∗)−1z. Note that g∗y = g∗h′g − z is selfadjoint: Im(g∗y) = g∗Im(h′)g − z = g∗z′g − z = 0. A direct computation shows that if then g =(cid:18) g 0 y (g∗)−1 (cid:19) , g · h = h′. Remark 6.4. As remarked, the unitary matrix U maps KD onto KH, and intertwines the groups UU (θD) and UU (θH): a ∈ UU (θH ) if and only if U∗aU ∈ UU (θD). We shall see later (Remark 7.4), that the Moebius transformation Γ : H → D induced by these transformations, maps i ∈ H to 0 ∈ D. Let us denote by IH the isotropy group (of the action of UU (θH)) of i ∈ H: IH i = {c ∈ UU (θH) : c · i = i}. i Accordingly, the isotropy group (of the action of UU (θD)) of 0 ∈ D is On the other hand, note that d · 0 = 0 if and only if d21 = 0, since d ∈ UU (θD), this implies The above facts imply that U∗IH ID 0 = { d ∈ UU (θD) : d · 0 = 0}. i U = ID 0 . that also d12 = 0 and d11, d22 ∈ UA, i.e., ID 0 = {(cid:18) u 0 0 u (cid:19) : u ∈ UA}, and, therefore, IH i = U ID 0 U∗ = ID 0 . 13 Theorem 6.5. 1. The subgroup B acts freely and transitively on KH. In particular, U (θH) acts transitively on KH . 2. KH is an C∞ submanifold of A2. 3. The restriction of ϕ to KH , is an C∞ epimorphism, with a global C∞ cross section. ϕKH : KH → H 4. The group U (θH) acts covariantly with respect to ϕ (on the left): (cid:18) b1 b2 (cid:19) ∈ KH, ϕ(a(cid:18) b1 b2 (cid:19)) = a · ϕ((cid:18) a1 a2 (cid:19)). Proof. The first assertion: consider the element (1, i) ∈ KH , and pick b ∈ B, if a ∈ U (θH) and 0 with b∗x selfadjoint. Then b =(cid:18) b x (b∗)−1 (cid:19) b · (1, i) = (b, x + i(b∗)−1). x2 (cid:19) = (x1, x2) ∈ KH . Then b = x1 and x = x2 − i(x∗1)−1 These pairs parametrize KH . Pick (cid:18) x1 determine a matrix b in B: x1 is invertible and a straightforward computation shows that the x2 (cid:19), and b is fact that (cid:18) x1 x2 (cid:19) ∈ KH implies that b∗x is selfadjoint. Clearly b · (1, i) = (cid:18) x1 determined by this condition. The second assertion: the action of B provides a homeomorphism σ : B → KH, σ(b) = b · (1, i), with inverse σ−1 : KH → B, σ−1((cid:18) x1 x2 (cid:19)) =(cid:18) x1 x2 − i(x∗1)−1 0 (x∗1)−1 (cid:19) . a2 (cid:19) ∈ A2 : Im(a∗1a2), a1 ∈ G}, which is open in A2, and the The map σ−1 extends to H := {(cid:18) a1 map σ extends to M2(A). Therefore σ provides a global C∞ adapted chart for KH (modelled in the manifold B). The third assertion: consider the map ψ : H → KH , ψ(h) = 1√2 1√2 Im(h)−1/2 h Im(h)−1/2 ! = 14 1 √2(cid:18) 1 h (cid:19) Im(h)−1/2. By this description, it is apparent that ψ takes values in H, and moreover, after elementary computations (involving right multiplication by elements of of GA), iω(ψ(h)) = Im((Im(h)−1/2)∗h Im(h)−1/2) = Im(h)−1/2Im(h)Im(h)−1/2 = 1. Finally, we get ϕ(ψ(h)) = ϕ( hIm(h)−1/2) = h. Im(h)−1/2, 1 1 √2 √2 b2 (cid:19)) = ϕ(a(cid:18) 1 a · ϕ((cid:18) b1 The fourth assertion: 1 (cid:19)), using the invariance of ϕ under the right action of GA, this equals b2 (cid:19)). 1 )) = ϕ(a(cid:18) b1 ϕ(a(cid:18) b1 b2 (cid:19) b−1 b2b−1 Corollary 6.6. KD is a C∞-submanifold of A2 Proof. In Theorem 6.5 it is shown that KH is a C∞-submanifold of H, which is an open subset of A2. Thus, KD = U∗KH is also a submanifold of A2. Let us finish this section, by proving the claim made at the beginning of it, that the action of G in G+ is the restriction of the action of UU (θH) on T G+ (using the model H). Remark 6.7. The injective group homomorphism G ֒→ UU (θH), described at the beginning of this section, allows one to regard g ∈ G as an element in UU (θH), namely (cid:18) g 0 (g∗)−1 (cid:19). An element a ∈ G+ lies in H as 0 + ia. Then 0 0 (cid:18) g 0 (g∗)−1 (cid:19) · (0 + ia) = (g∗)−1iag−1 = ig · a, which is the guise under which g · a ∈ G+ appears in T G+. 7 The space D as decompositions of an indefinite form. The fact that ρD is selfadjoint in M2(A) and satisfies ρ2 D = 1 implies that the form θD induces a non degenerate A-valued indefinite quadratic form in A2. We shall consider the following set, which was studied in [7] (Sections 3,4,6): QρD = {ǫ ∈ M2(A) : ǫ2 = 1 and ρDǫ ∈ G+}. In particular, ρDǫ is selfadjoint, which implies that θD(ǫ(cid:18) a1 a2 (cid:19) ,(cid:18) b1 b2 (cid:19)) = θD((cid:18) a1 a2 (cid:19) , ǫ(cid:18) b1 b2 (cid:19)) for all (cid:18) a1 a2 (cid:19) ,(cid:18) b1 b2 (cid:19) ∈ A2, 15 x2 (cid:19) =(cid:18) x1 i.e., ǫ is symmetric for the form θD. The fact that ǫ2 = 1 implies that A2 is decomposed in two eigenspaces: + = {(cid:18) x1 x2 (cid:19) ∈ A2 : ǫ(cid:18) x1 y2 (cid:19)}. A2 The fact that ρDǫ ∈ G+ means that the quadratic form induced by θD is positive definite in A2 and negative definite in A2 −, and that the eigenspaces are θD-orthogonal. Conversely, any such decomposition of A2 induces a non selfadjoint reflection in QρD . Therefore, it is appropriate to think of QρD as the set of positive-negative decompositions of the quadratic form (given by) θD. The set QρD is a submanifold of M2(A). It has yet another important characterization in Section 3 of [7]: in the polar decomposition of ǫ, the unitary part is precisely ρD: y2 (cid:19) ∈ A2 : ǫ(cid:18) y1 y2 (cid:19) = −(cid:18) y1 x2 (cid:19)} , A2 − = {(cid:18) y1 + ǫ = ǫ∗ρD. Also, any nonselfadjoint reflection with this latter property belongs to QρD . Therefore, QρD is parametrized by a subset of Gl2(A)+, the set positive invertible elements in M2(A). In particular, this endows QρD with an C∞ submanifold structure, with a rich metric geometry of non-positive type. The group U (θD) acts transitively in QρD , x, y, etc. notation, when the elements (cid:18) x1 Definition 7.1. Any element (cid:18) x1 projection in M2(A): ρD =(cid:18) x1 g · ǫ = gǫg−1. We shall see below that that QρD is naturally diffeomorphic to D. In order to lighten the x2 (cid:19) ,(cid:18) y1 y2 (cid:19), etc., appear as subindices, let us denote them by x2 (cid:19) ∈ KD defines a (modular) rank one (non selfadjoint) x2 (cid:19)(cid:0) x∗1 x∗2 (cid:1)(cid:18) 1 a2 (cid:19) ,(cid:18) x1 x2 (cid:19)) =(cid:18) x1 x2 (cid:19)) = 1, it it clear that px is a projection, which x2 (cid:19) ,(cid:18) x1 x2 (cid:19) θD((cid:18) a1 With this description, since θD((cid:18) x1 0 −1 (cid:19) =(cid:18) x1x∗1 −x1x∗2 x2x∗1 −x2x∗2 (cid:19) . x2 (cid:19) < ρD(cid:18) a1 x2 (cid:19)∗ x2 (cid:19)(cid:18) x1 a2 (cid:19)) =(cid:18) x1 a2 (cid:19) ,(cid:18) x1 x2 (cid:19) > . px =(cid:18) x1 0 Or, equivalently, px((cid:18) a1 is θD-symmetric. Consider the following maps: ΦD : D → QρD , ΦD(h) = 2px − 1, where (cid:18) x1 x2 (cid:19) ∈ KD satisfies that ϕ((cid:18) x1 ϕ : KD → QρD , x2 (cid:19)) = h, and ϕ((cid:18) x1 x2 (cid:19)) = 2px − 1. 16 (5) (6) Lemma 7.2. The map ΦD is well defined, C∞ and equivariant with respect to the actions of U (θD). Proof. Suppose(cid:18) x1 y2 (cid:19) ∈ KD such that x2x−1 1 , i.e., (cid:18) y1 y2 (cid:19) =(cid:18) x1 x2 (cid:19) · g, for x2 (cid:19) ,(cid:18) y1 1 = y2y−1 g = x−1 1 y1 ∈ G. Note that 1 = y∗1y1 − y∗2y2 = (x1g)∗x1g − (x2g)∗x2g = g∗(x∗1x1 − x∗2x2)g = g∗g, i.e., g ∈ UA. Then py =(cid:18) y1 y2 (cid:19)∗ y2 (cid:19)(cid:18) y1 ρD =(cid:18) x1 x2 (cid:19)∗ x2 (cid:19) gg∗(cid:18) x1 ρD =(cid:18) x1 x2 (cid:19)∗ x2 (cid:19)(cid:18) x1 ρD = px. Let us prove that the reflection 2px− 1 belongs to QρD . It is symmetric for the form θD. Clearly, ρD(2px − 1) is invertible, and it is non negative if and only if Explicitly, Put γ = x1x∗2. Since (cid:18) x1 (2px − 1))ρD = ρD(ρD(2px − 1))ρD ≥ 0. (2px − 1)ρD =(cid:18) x1x∗1 − 1 x2x∗2 + 1 (cid:19) . x2 (cid:19) ∈ KD, one has that x∗1x1 = 1 + x∗2x2. Then x1x∗2 x2x∗1 γ∗γ = x2x∗1x1x∗2 = x2(1 + x∗2x2)x∗2 = x2x∗2 + (x2x∗2)2, thus γ∗γ + 1 = (x2x∗2 + 1)2, i.e., x∗2x2 + 1 = (γ∗γ + 1)1/2. Similarly, we get x1x∗1− 1 = (γγ∗ + 1)1/2 (using now that x1x1 > 1, because x∗1x1 > 1 and x1 is invertible) . Then (2px − 1)ρD =(cid:18) (γγ∗ + 1)1/2 γ∗ γ (γ∗γ + 1)1/2 (cid:19) . In order to prove that this matrix is invertible, denote Clearly m is selfadjoint and γ m =(cid:18) 0 γ∗ 0 (cid:19) . (2px − 1)ρD = (1 + m2)1/2 + m ≥ 0, because the real function f (t) = (1 + t2)1/2 + t ≥ 0 for all t ∈ R. The fact that ΦD is C∞ follows from a standard argument in fibrations: clearly, the formula that defines ΦD in terms of coordinates in KD is C∞, therefore, using C∞ local cross sections for the fibration ϕ : KD → D, one obtains that ΦD is C∞. Finally, if g ∈ U (θD) (i.e., g∗ρD = ρD g−1), x2 (cid:19) < ρD · , g(cid:18) x1 pg·x = g(cid:18) x1 x2 (cid:19) >= g(cid:18) x1 x2 (cid:19) < g∗ρD · ,(cid:18) x1 x2 (cid:19) > 17 x2 (cid:19) < ρD g−1 · ,(cid:18) x1 which means that ΦD is U (θD)-equivariant: ΦD(g(cid:18) x1 = g(cid:18) x1 x2 (cid:19) >= gpxg−1, x2 (cid:19)) = gΦD((cid:18) x1 x2 (cid:19))g−1. In the next theorem we summarize several results about the maps considered in the following diagram of homogeneous spaces of the group U (θD): "❊❊❊❊❊❊❊❊ KD ~⑥⑥⑥⑥⑥⑥⑥⑥ ϕ ϕ ΦD D / QρD (7) Theorem 7.3. The diagram (7) commutes. The maps ϕ : KD → D and ϕ : KD → QρD are C∞ submersions. The map ΦD : D → QρD is a C∞-diffeomorphism. All maps are equivariant under the action of U (θD). Proof. Clearly, ΦD ◦ ϕ((cid:18) x1 x2 (cid:19)) = px = ϕ((cid:18) x1 x2 (cid:19)), thus the diagram commutes. Let us prove that every reflection ǫ in M2(A) which is θD symmetric and such that ρDǫ (equivalently, ǫρD is positive) must be of the form ǫ = 2px − 1 for x ∈ KD. That is, given ǫ of the form with ǫ∗ii = ǫii such that ǫ12 ǫ22 (cid:19) 0 ≤ ǫρD =(cid:18) ǫ11 −ǫ12 ǫ =(cid:18) ǫ11 −ǫ∗12 −ǫ∗12 −ǫ22 (cid:19) , there exists (cid:18) x1 x2 (cid:19) ∈ KD such that (cid:18) ǫ11 −ǫ∗12 ǫ12 ǫ22 (cid:19) =(cid:18) x1x∗1 − 1 x2x∗1 −x1x∗2 −x2x∗2 − 1 (cid:19) . We shall look for (cid:18) x1 with x′2(x′1)−1 = x2x−1 1 decomposition of x∗1, and take x′1 = x∗1. It is not hard to see that such x′1 is unique. x2 (cid:19) ∈ KD there exists (cid:18) x′1 x′2 (cid:19) and x′1 > 0. Indeed, since x1 is invertible, put x1 = x∗1u the polar x2 (cid:19) with x1 > 0. Note that for any (cid:18) x1 The fact that ǫ2 = 1 means that ǫ2 11 − ǫ12ǫ∗12 = 1 ǫ11ǫ12 + ǫ12ǫ22 = 0 −ǫ∗12ǫ12 + ǫ2 22 = 1 .   Thus, x1 = (1 + ǫ11)1/2 and x2 = −ǫ∗12(1 + ǫ11)−1/2. We must check that x2x∗1 = −ǫ∗12, which is apparent, and that −x2x∗2 − 1 = ǫ22. Indeed, by the above relation on eij, −x2x∗2 − 1 = ǫ∗12(1 + ǫ11)−1ǫ12 − 1 = ǫ∗12(1 + (ǫ12ǫ∗12 + 1)1/2)−1ǫ12 − 1. 18 ~ " / Since ǫ∗12(ǫ12ǫ∗12)n = (ǫ∗12ǫ12)nǫ∗12, then for any continuous function f : [0, +∞) → C, ǫ∗12f (ǫ12ǫ∗12) = f (ǫ∗12ǫ12)ǫ∗12. Thus, again using the relations on ǫij, −x2x∗2 − 1 = (1 + (ǫ∗12ǫ12 + 1)1/2)−1ǫ∗12ǫ12 − 1 = (1 + ǫ22)−1(ǫ2 22 − 1) − 1 = ǫ22. Next, we must check that (cid:18) x1 x2 (cid:19) ∈ KD. Clearly x1 = (1 + ǫ11)1/2 is invertible (and positive). x1x∗1 − x2x∗2 = 1 + ǫ11 − (1 + ǫ11)−1ǫ12ǫ∗ −12(1 + ǫ11)−1/2. Since −ǫ12ǫ∗12 = (1 + ǫ11)2, this equals The formula(cid:18) x1 x2 (cid:19) =(cid:18) 1 + ǫ11 − (−1 + ǫ11) = 2. (1 + ǫ11)1/2 −ǫ∗12(1 + ǫ11)−1/2 (cid:19), regarded as a map QρD → KD, provides a global x2 (cid:19) ∈ KD with x1 > 0 C∞ cross section for ϕ, proving that it is retraction, thus a submersion. (as computed above, Let us exhibit the inverse of ΦD: since ǫ = 2px − 1 for a unique (cid:18) x1 1 = −ǫ∗12(1 + ǫ11)−1. D : QρD → D , Φ−1 Φ−1 D (ǫ) = x2x−1 Clearly, it is a C∞ map. The map ΦD was computed using coordinates in KD. It can be also computed in terms of z ∈ D. Analogously as ϕ, ϕ has also a global cross section given by the unique element in each fiber with positive first coordinate. Namely, δ : D → KD , δ(z) =(cid:18) 1 z (cid:19) (1 − z∗z)−1/2. Then z (cid:19) (1 − z∗z)−1(cid:0) 1 z∗ (cid:1) ρD − 1 −2(1 − z∗z)−1z∗ −2z(1 − z∗z)−1z∗ − 1 (cid:19) . Remark 7.4. There is an analogous diagram as (7) for the space H: ΦD(z) = 2pδ(z) − 1 = 2(cid:18) 1 =(cid:18) 2(1 − z∗z)−1 − 1 2z(1 − z∗z)−1 ϕ KH ~⑥⑥⑥⑥⑥⑥⑥⑥ ΦH ¯ϕ "❊❊❊❊❊❊❊❊ / QρH (8) H Recall the unitary operator U which intertwines ρD and ρH: U ρDU∗ = ρH. We saw (Lemma 3.1) that U maps KD onto KH . Also, it is clear that ǫ ∈ QρH if and only if U∗ǫU ∈ QρD . Also it is clear, by construction, that ΦH(h) = U ΦD(γ(h))U∗. (9) 19 ~ " / Let us state another consequence obtained from the equivalence of diagrams (7) and (8 Theorem 7.5. The map Γ : H → D, Γ(h) = (1 + ih)(1 − ih)−1 is a (well defined) diffeomorphism with inverse Γ−1 : D → H Γ−1(z) = i(1 − z)(1 + z)−1. Proof. One passes from H to D with the cross section h 7→ 1 √2(cid:18) 1 h (cid:19) Im(h)−1/2 ∈ KH composed with left multiplication by U∗, followed by ϕ, namely, h 7→ 1 √2(cid:18) 1 h (cid:19) Im(h)−1/2 7→ 1 2(cid:18) 1 −i i (cid:19)(cid:18) 1 h (cid:19) Im(h)−1/2 7→ (1 + ih)(1 − ih)−1. 1 Which means that Γ : H → D is well defined and smooth. Its inverse is computed analogously: z 7→(cid:18) 1 z (cid:19) (1 − z∗z)−1/2 7→ U(cid:18) 1 z (cid:19) (1 − z∗z)−1/2 ϕ → (i − iz)(1 + z)−1. A straightforward computation shows that these maps are each other inverses. since px((cid:18) y1 Let us finish this section by recalling the action of the group U (θH). If (cid:18) x1 y2 (cid:19)), one has that for any a ∈ U (θH ), x2 (cid:19) , a−1·) = apxa−1. x2 (cid:19) ,(cid:18) y1 x2 (cid:19) ,·) = a.(cid:18) x1 x2 (cid:19) θH((cid:18) x1 y2 (cid:19)) =(cid:18) x1 pa·x = a.(cid:18) x1 x2 (cid:19) θH((cid:18) x1 x2 (cid:19) θH(a.(cid:18) x1 x2 (cid:19) ∈ H, then, Then 2pa·x − 1 = a(2px − 1)a−1. Therefore: Proposition 7.6. If h ∈ H and g ∈ UU (θH), ΦH(g · h) = gΦH(h)g−1, i.e., ΦH is equivariant for the action of U (θH). Therefore Γ(g · h) = (U∗gU ) · Γ(h) (10) 20 8 The hyperbolic geometry of Qρ In previous papers [6], [8], [9] the geometry of the set of positive invertible elements of a C∗- algebra was studied. As is the case with the classical example of positive definite complex matrices (see for instance [16]), this space, with the appropriate Finsler metric, behaves as a non-positively curved manifold. In [17] it is proven that the set Qρ embedds in the space of positive invertible elements, in a way that the main geometric features remain invariant. Let us briefly describe below these constructions and results. We shall use these results (and therefore also state them) for the case of the C∗-algebra M2(A); the space of positive and invertible matrices shall be denoted by Gl2(A)+. Remark 8.1. (see [6]) Gl2(A)+ is an open subset of M2(A)s, so it has a natural differentiable structure. We consider in Gl2(A)+ the following left action of Gl2(A): g · a = (g∗)−1ag−1, g ∈ Gl2(A), a ∈ Gl2(A)+. This action is transitive. The isotropy subgrup of an element a ∈ Gl2(A)+ is the group of a-unitary operators. Thus, the Banach-Lie isotropy algebra of a is the space of a-anti-Hermitian elements of M2(A): x∗a + ax = 0. A natural complement for this space is the space of a- Hermitian elements: y∗a − ay = 0. This decomposition is equivariant under the action of Gl2(A), and induces a linear connection in Gl2(A)+. The covariant derivative of this connection is given by DY dt = dY dt − 1 2{ γγ−1Y + Y γ−1 γ}, where Y (t) is a tangent field along the curve γ(t) in Gl2(A)+; due to the trivial local structure of Gl2(A)+ ⊂ M2(A)s, this simply means that Y (t) is a curve of selfadjoint elements in M2(A). A geodesic is a curve γ such that D γ dt = 0. The geodesic γ with γ(0) = a and γ(0) = X is given by γ(t) = e t 2 Xa−1 ae t 2 Xa−1 . The exponential map expa : M2(A)s → Gl2(A)+, expa(X) = e 1 2 Xa−1 ae 1 2 Xa−1 , is everywhere a diffeomorphism. Any pair a, b ∈ Gl2(A)+ is joined by a unique geodesic, which is γa,b(t) = a1/2(a−1/2ba−1/2)ta1/2. The space Gl2(A)+ carries a Finsler metric. By this we mean a continuous distribution Gl2(A)+ ∋ a 7→ k ka of norms defined in the corresponding tangent spaces T (Gl2(A)+)a = M2(A)s. Notice that we do not require that this distribution be smooth, as in the finite di- mensional setting (see for instance [1], [2], [6], [7], for other examples of Finsler metrics in the context of operator theory). If a ∈ Gl2(A)+ and X ∈ M2(A)s, put kXka = ka− 1 2 Xa− 1 2k. (11) 21 Theorem 8.2. ([6], Section 6) With the metric defined in (11), the geodesics of the connection for any a, b ∈ Gl2(A)+, the unique geodesic γa,b of the connection has are globally minimal: minimal length among all smooth curves in Gl2(A)+ joining a and b. The (geodesic) distance between a and b can be computed: dg(a, b) = k log(a− 1 2 ba− 1 2 )k, where log denotes the unique selfadjoint logarithm of a positive invertible element. Moreover, the metric has the following property, which in Riemannian geometry is equivalent to non positive curvature, and is used as a definition of non positive curvature for metric length spaces (i.e., metric spaces with given short curves, see [4], [5], [12]). Theorem 8.3. ([9]) If γ(t), δ(t) are two geodesics in G+, then f (t) = dg(γ(t), δ(t)) is a convex function. In particular, if the geodesics start at the same point, i.e., γ(0) = δ(0), then dg(γ(t), δ(t)) ≤ t dg(γ(1), δ(1)). There is a natural embedding of Qρ in Gl+ 2 (A): Theorem 8.4. ([17]) The embedding of Qρ in Gl2(A)+ is given by [17] Qρ ֒→ Gl+ This embedding has the following properties 2 (A), ǫ → ρǫ. 1. Let ǫ1, ǫ2 ∈ Qρ. Then the unique geodesic of Gl2(A)+ joining ρǫ1 and ρǫ2 lies in (the image of) Qρ (under the above embedding). 2. Qρ is an homogeneous space under the action of U (θρ), Gl+ under the action of Gl2(A). The embedding is equivariant for these actions. 2 (A) is an homogeneous space Thus, if we endow Qρ with the geometry induced by this embedding, it becomes a non positively curved metric length space. We use these facts to translate to H and D the metric structure of Qρ, by means of the equivariant diffeomorphisms ΦH and ΦD, respectively. Definition 8.5. For any h1, h2 ∈ H and z1, z2 ∈ D, dH(h1, h2) = dg(ρHΦH(h1), ρH ΦH(h2)) and Therefore one has: dD(z1, z2) = dg(ρDΦD(z1), ρDΦD(z2)). 22 Corollary 8.6. Both (H, dH ) and (D, dD) are non positively curved metric length spaces. In other words, if δ1, δ2 are two geodesics in H (resp. D), the function f (t) = dH (δ1(t), δ2(t)) (resp. f (t) = dD(δ1(t), δ2(t))) are convex. If, additionally, δ1(0) = δ2(0) then, for t ∈ [0, 1] dH (δ1(t), δ2(t)) ≤ t dH (δ1(1), δ2(1)) ( resp. dD(δ1(t), δ2(t)) ≤ t dD(δ1(1), δ2(1)) ). The diffeomorphism (12) is an isometry. Γ : H → D Proof. Recall formula (3.1) in Remark (7.4): ΦH(h) = U ΦD(γ(h))U∗. A straightforward com- putation shows that this implies that if h1, h2 ∈ H, then dD(Γ(h), Γ(h2)) = dH(h1, h2). In [15], treating other kind of problems, a similar homeomorphism was established, between D and the set of elements in A with positive and invertible real part (i.e. −iH) Recall that the group U (θH) acts transitively on H. This group is, in turn, isomorphic to U (θD), which acts transitively on D. Also recall (Proposition (7.6)), that these actions are equivariant for ΦH and ΦD, respectively. Then one has: Corollary 8.7. The actions of U (θH) and U (θD) on H and D, respectively, are isometric. Proof. The diffeomorphism ΦH, followed by the embedding of QρH , is a composition of maps which are equivariant for the action of U (θH) ⊂ Gl2(A). On the other hand, the action of this latter group on Gl+ 2 (A) is isometric ([7], p.66). Example 8.8. Let us compute the dD distance between 0 and z in D: dD(0, z) = k log(ΦD(z)ρD)k = k log(cid:18) 2(1 − z∗z)−1 2z(1 − z∗z)−1 2z(1 − z∗z)−1z∗ + 1 (cid:19)k. 2(1 − z∗z)−1z∗ Straightforward computations show that this matrix above can be factorized (cid:18) 2(1 − z∗z)−1 2z(1 − z∗z)−1 2z(1 − z∗z)−1z∗ + 1 (cid:19) = ∆1∆2, 2(1 − z∗z)−1z∗ where ∆1 =(cid:18) (1 − z∗z)−1 0 0 (1 − zz∗)−1 (cid:19) and ∆2 =(cid:18) 1 + z∗z 2z 2z∗ 1 + zz∗ (cid:19) . (A key fact in this computation is that z(1− z∗z)−1 = (1− zz∗)−1z). Denote by Ω =(cid:18) 0 z∗ 0 (cid:19). Note that Ω∗ = Ω and kΩk < 1. Then z ∆1 = (1 − Ω2)−1 and ∆2 = (1 + Ω)2. 23 Therefore log(∆1∆2) = log(1 + Ω) − log(1 − Ω). Using the power series log(1 + t) − log(1 − t) = 2 t2k+1 2k + 1 ∞ Xk=0 we get that log(1 + Ω) − log(1 − Ω) = 2(cid:18) The norm of this matrix equals 0 1 2k+1 (z∗z)k z∗P∞k=0 1 2k+1 (zz∗)k 0 (cid:19) zP∞k=0 k log(1 + Ω) − log(1 − Ω)k = k(log(1 + Ω) − log(1 − Ω))2k1/2. Note that (log(1 + Ω) − log(1 − Ω))2 equals 4(cid:18) (z∗P∞k=0 0 1 1 2k+1 (z∗z)k) 2k+1 (zz∗)k)(zP∞k=0 = 4(cid:18) z∗z(P∞k=0 =(cid:18) log(1 + z) − log(1 − z) 1 2k+1 (z∗z)k)2 0 0 0 1 0 2k+1 (z∗z)k)(z∗P∞k=0 (zP∞k=0 2k+1 (zz∗)k)2 (cid:19) zz∗(P∞k=0 log(1 + z∗) − log(1 − z∗) (cid:19)2 0 1 2k+1 (zz∗)k) (cid:19) 1 . The square root of the norm of this matrix is max{k log(1 + z) − log(1 − z)k,k log(1 + z∗) − log(1 − z∗)k}. The function f (t) = log(1 + t) − log(1 − t) is strictly increasing in [0, 1), with f (0) = 0. Thus (using that kzk = kzk), klog(1 + z) − log(1 − z)k = max{f (t) : t ∈ σ(z)} = f (kzk). Analogously, klog(1 + z∗) − log(1 − z∗)k = f (kzk). Then 1 + kzk 1 − kzk dD(0, z) = log( ). In the scalar case A = C, this norm equals which is the Poincar´e distance in the open unit disk D. dD(0, z) = log(cid:18) 1 + z 1 − z(cid:19) , 24 9 The covariant derivative in H In this section we compute explicitly the covariant derivative induced by the reductive structure. Recall the decomposition (4) of the Banach-Lie algebra of U (θH )), X = Xas ⊕ Xs = { β ∈ M2(A)as : βJ = J β} ⊕ {γ ∈ M2(A)s : γJ = −J γ}. Elements X ∈ X are of the form X = X0 + Xh, x12 X =(cid:18) x11 −x12 −x11 (cid:19) +(cid:18) α β −α (cid:19) β with x∗11 = −x11, and all other entries selfadjoint. The left hand subspace Xas is the Banach-Lie algebra of the isotropy group of the action at the element i ∈ H. The right hand subspace Xs is the horizontal space at this point. The computation of the covariant derivative will be done in several steps. Step 1. First we compute the differential of the map πi : U (θH) → H, πi(g) = g · i, at the identity 1 ∈ U (θH). Recall that πi(g) = (g22i + g21(g12i + g11)−1. Then, differentiating at 1, we get d(πi)1(γ) = γ21 + γ12 + i(γ22 − γ11). If γ is horizontal, d(πi)1(γ) = 2γ12 − 2iγ11. Then: Theorem 9.1. The 1-form of the reductive connection at i ∈ H is χ Υ (cid:19) , if ζ = χ + iΥ. 2(cid:18) −Υ χ κi(ζ) = 1 Step 2. For any given h = x + iy, one can find an element b ∈ B ⊂ U (θH) (the Borel subgroup of U (θH)) such that b · i = h. For instance, 0 b =(cid:18) y−1/2 xy−1/2 y1/2 (cid:19) with inverse b−1 =(cid:18) y1/2 Straightforward computations show that b · i = h (and b−1 · h = i). of H. If d dt h = ζ, 0 −y−1/2x y−1/2 (cid:19) . Step 3 Let us compute now the differential of the action of g ∈ U (θH) on tangent vectors d dt (g · h) = (g22 − wg12)ζ(g12h = g11)−1, where w = g · h = (g22h + g21)(g12z + g11)−1. Using the transformation in Step 2, we can carry a tangent vector ζ at h to the tangent vector y−1/2ζy−1/2 at i. If ζ + χ + iΥ, κi(y−1/2ζy−1/2) + 1 2 y−1/2(cid:18) −Υ χ χ Υ (cid:19) y−1/2. Step 4 To obtain κh(ζ), we use the inner automorphism Adb: (cid:18) y−1/2 xy−1/2 y−1/2 (cid:19){ 0 1 2 y−1/2(cid:18) −Υ χ χ Υ (cid:19) y−1/2}(cid:18) y1/2 −y−1/2x y−1/2 (cid:19) 0 25 Then κ(ζ)′ = κi( Dζ dt t=0) = κ(ζ)′ + [κi(Z(i)), κi(h′)]. χ′ Υ′ (cid:19) + 2(cid:18) y′Υ − χx′ −y′χ − χy′ x′χ − Υy′ (cid:19) −x′χ − Υx′ 2(cid:18) −Υ′ χ′ 1 1 = 1 2(cid:18) −y−1χy−1x χ − xy−1χy−1x xy−1χy−1 (cid:19) + y−1χy−1 1 2(cid:18) −y−1Υ −xy−1Υ − Υy−1x Υy−1 (cid:19) . 0 Step 5 Next, we compute the covariant derivative of the reductive connection in H, at the point i ∈ H. Consider the curve h(t) = x(t) + iy(t) in H, with z(0) = i, and the vector field ζ = χ + iΥ defined on a neighbourhood of i. Denote by Dζ dt t=0 the covariant derivative at i ∈ H. According to [14], one has that κi( Dζ dt t=0) = d dt κh(t)Z(h(t))t=0 + [κi(Z(i)), κi( d dt h(t)t=0]. To lighten the notation, we shall denote by κ(ζ)′, h′ the usual derivatives at t = 0. So the formula above reads Let us write the sum of right hand matrix above with the bracket [κi(Z(i)), κi(h′)]. After strenuous but elementary computations one gets 1 4(cid:18) Υy′ − χx′ −Υx′ − χy′ χx′ − Υy′ (cid:19) + −χy′ − Υx′ 1 4(cid:18) y′Υ − x′χ −y′χ − x′Υ x′χ − y′Υ (cid:19) . −x′Υ − y′χ Now we must apply κ−1 i , or else realize that the above term is the value of κi at −Re(x′Υ + y′χ) + iRe(x′χ − y′Υ). Therefore, dζ dt t=0 = χ′ + iΥ′ + {−Re(x′Υ + y′χ) + iRe(x′χ − y′Υ)}. Dζ Step 6 Let h0 = x0 + iy0 ∈ H, and h(t) = x(t) + iy(t) be a smooth curve in H with h(0) = h0. Let ζ = χ + iΥ be a smooth vector field defined on a neighbourhood of H0. Let us compute dt t=0. To do this, we shall use the invariance of the connection under the action of the group U (θH), and the fact that we know this formula in the case h0 = i. As seen in Step 2, the (explicit) element b defined there, performs b−1 · h0 = i. Thus we can carry the data to the point i, perform the covariant derivative, and translate it back to h0 with the action (note for instance, that y1/2 , and so forth). We get: 0 = Re(y1/2 0 uy1/2 0 Re(u)y1/2 0 Dζ dt = ζ′ − Re(x′y−1 0 Υ + y′y−1 0 χ) + iRe(x′y−1 0 χ − y′y−1 0 Υ). (13) We devote the next section to describe examples of geodesics in the halfspace model H: 10 Examples of geodesics in H In terms of the decomposition (4) of the Banach-Lie algebra of U (θH), X = Xas ⊕ Xs = { β ∈ M2(A)as : βJ = J β} ⊕ {γ ∈ M2(A)s : γJ = −J γ}. 26 geodesics of H starting at i for t = 0 have the form δ(t) = etXh · i, where Xh is a horizontal element, i.e., an antihermitian element in M2(A) of the form Xh =(cid:18) α β −α (cid:19) , β with α∗ = −α, β∗ = −β. We shall compute these geodesics in two particular cases: Example 10.1. Suppose that the entries α and β in Xh commute. Put γ = (α2 + β2)1/2. The element γ may not be invertible; however, if we make the assumption that A is weakly closed (i.e., a von Neumann algebra), then there exist unique selfadjoint elements x and y in A such that xγ = γx = α and yγ = γy = β. Moreover, there exists a selfadjoint element χ ∈ A such that x = cos(χ) and y = sin(χ). Accordingly, Xh/γ = (cid:18) cos(χ) sin(χ) − cos(χ) (cid:19). Put 12 =(cid:18) 1 0 0 1 (cid:19). Straightforward computations show that for n ≥ 0 sin(χ) (tXh)2n = (tγ)2n12 and (tXh)2n+1 = (tγ)2n+1Xh/γ. Then etXh = cosh(tγ)12+sinh(tγ)Xh/γ =(cid:18) cosh(tγ) + cos(χ) sinh(tγ) Therefore δ(t) = etXh · i equals sin(χ) sinh(tγ) sin(χ) sinh(tγ) cosh(tγ) − cos(χ) sinh(tγ) (cid:19) . (sin(χ) sinh(tγ) + i(cosh(tγ) − cos(χ)) (cosh(tγ) + cos(χ) sinh(tγ) + i sin(χ) sinh(tγ))−1 . After straightforward calculations, involving well known identities concerning cosh and sinh, we arrive at δ(t) = (sin(χ) sinh(2tγ) + i) (cosh(2tγ) + cos(χ) sinh(2tγ))−1 . (14) Since all elements involved commute and ξ, γ are selfadjoint, it follows that Re(δ(t)) = sin(χ) sinh(2tγ) (cosh(2tγ) + cos(χ) sinh(2tγ))−1 and Im(δ(t)) = (cosh(2tγ) + cos(χ) sinh(2tγ))−1 . Therefore, if one regards δ as a geodesic in T G+, i.e. δ = (Re(δ), Im(δ)), it is given by δ(t) =(cid:16)sin(χ) sinh(2tγ) (cosh(2tγ) + cos(χ) sinh(2tγ))−1 , (cosh(2tγ) + cos(χ) sinh(2tγ))−1(cid:17) . Additionally, if there exists µ ∈ A, µ = µ∗, such that µβ = −α (for instance, if β is invertible), then (Re(δ(t) − µ)2 + (Im(δ(t))2 = µ2 + 1, that is, geodesics δ of the Poincar´e halfspace (with commuting α, β) satisfy the equation of an A-valued circle, centered in the real axis (at µ∗ = µ) with radius (1 + µ2)1/2. 27 Another special case which can be explicitly computed occurs when α and β anti-commute: αβ = −βα Example 10.2. Suppose now that the entries α, β in Xh anti-commute. Then (cid:18) α and (cid:18) 0 β β 0 (cid:19) commute. Thus 0 0 −α (cid:19) t  etXh = e α 0 0 −α  e t  0 β β 0 Therefore   =(cid:18) etα 0 0 e−tα (cid:19)(cid:18) cosh(tβ) sinh(tβ) sinh(tβ) cosh(tβ) (cid:19) . δ(t) = etXh · i = e−tα(sinh(tβ) + i cosh(tβ))(cosh(tβ) + i sinh(tβ))−1e−tα. By straightforward computations, (sinh(tβ) + i cosh(tβ))(cosh(tβ) + i sinh(tβ))−1 = (sinh(2tβ) + i) cosh−1(2tβ). Since cosh is even, cosh(2tβ) commutes with etα, analogously, since sinh is odd, e−tα sinh(2tβ) = sinh(2tβ)etα. Then δ(t) = sinh(2tβ) cosh−1(2tβ) + i cosh−1(2tβ)e−2tα. Clearly, this is the real/imaginary part decomposition of δ. Note that the real part of δ does not depend on α. Thus, as a curve in T G+, this geodesic is given by δ(t) =(cid:0)sinh(2tβ) cosh−1(2tβ), cosh−1(2tβ)e−2tα(cid:1) . Suppose that α, β have a polar decompositions in A, α = µα = αµ, β = νβ = βν. Then after elementary calculations, one has that the imaginary and real parts of δ satisfy the equation (Re(δ − iµ)2 + (νIm(δ))2 = 0. 11 Appendix: The Poincar´e half-space of a Hilbertizable space. A complex locally convex topological vector space V is Hilbertizable if there exists an inner product β which makes V a Hilbert space. In this section we fix such a space V. We shall denote by S(V) the space of all sesquilinear forms σ on V, which are continuous in both variables. We consider in S(V) the topology whose basis of neighbourhoods of the origin are the sets W (V, ǫ) = {σ ∈ S(V) : σ(ξ, η) < ǫ, ξ, η ∈ V }, where V is a neighbourhoog of 0 in V and ǫ > 0. Let us denote by I(V) the set of all β ∈ S(V) which are positive definite and which reproduce the topology of V. There is a natural involution in S(V), which we shall call the conjugation in S(V), which is givan by σc(ξ, η) = σ(η, ξ). 28 Then S(V) decomposes as S(V) = S0(V) ⊕ S1(V), 0 = σ0) and anti-Hermitian (σc the Hermitian (σc open subset of S0(V). algebra, with the topology given by 1 = −σ1) forms, respectively. The set I(V) is an Denote by L(V) the algebra of continuous linear operators acting in V. L(V) is a topological W (V, V ′) = {a ∈ L(V) : a(ξ) ∈ V ′ for ξ ∈ V } Given β ∈ I(V), V becomes a Hilbert space, we shall denote it by Vβ. Likewise, L(V) as a system of neighbourhoods of 0 ∈ L(V), for V, V ′ neighbourhoods of 0 ∈ V. becomes a C∗-algebra, which will be denoted by Lβ(V). 11.1 Charts in S(V) Given β ∈ I(V), put Φβ : Lβ → S(V) , Φβ(a)(ξ, η) = β(aξ, η). By Riesz' Theorem, it is clear that Φβ is a bijection. We shall call Φβ the chart for S(V) centered at β. Denote by G(V) the group of bijective elements of L(V). The group G(V) acts on S(V) as changes of variables, that is, if g ∈ G(V) and σ ∈ S(V), Lgσ(ξ, η) = σ(g−1ξ, g−1η). It is apparent that this action restricts to an action of G(V) on I(V), and that it is transitive on I(V). Let us describe a change of charts by means on an element g ∈ G(V). Let β and β = Lgβ. Then the following diagram commutes Lβ(V) ↓ Adg L β(V) Lg−→ Lβ(V) ↓ Φβ Φ β−→ S(V) . (15) Here • Lg : Lβ(V) → Lβ(V) is the action given by Lga = gag−1, where g = (g−1)∗, with ∗ the It is also an β (V) of positive invertible operators, preserving its metric and involution of Lβ(V). This map Lg is a ∗-preserving linear isomorphism. isomorphism of the set L+ its linear connection (see Section 6). • Adg : Lβ(V) → L β(V) is a C∗-algebra isomorphism. The diagram (15) can be read as follows: ΦLgβ = Φβ LgAdg−1. As was noted in Section 6, the group G(V) acts on L+ β (V) by means of the action Lg. G(V) acts also on I(V) as noted above. The commutativity of the diagram (15) implies that Φβ intertwines both actions: Φβ : L+ β (V) → I(V) , LgΦβ = Φβ Lg. 29 11.2 The Poincar´e half-space We shall denote the Poincar´e half-space of V by Π(V), which is the set Π(V) = {σ ∈ S(V) : Im(σ) ∈ I(V)}, 2 (σ + σc) and Im(σ) = i where Re(σ) = 1 2 (σc − σ). We define charts in Π(V). Let β ∈ I(V). Clearly Φβ(a∗) = Φβ(a)c. Therefore Φβ maps the Poincar´e halfspace space H(Lβ(V)) of the C∗-algebra Lβ(V) as defined in Section 1, onto Π(V). The purpose of this Appendix is to introduce the geometry of the bundle of observables associated with the space of metrics I(V) of the Hilbertizable space V. Specifically, the bundle O → I(V), where the fiber Oβ over β ∈ I(V) is the vector space Oβ = {a ∈ Lβ(V) : a∗ = a}. We claim that the natural way to present the bundle of observables is as the tangent bundle TI(V). tangent vector X ∈ (TI(V)β is, canonically, an element of S0(V). Therefore the map But the space TI(V) can be identified with the Poincar´e half-space Π(V) as follows. A (β, X) ∈ (TI(V)β ←→ X + iβ ∈ Π(V) identifies TI(V) with Π(V) in a natural way. 11.3 The group of movements in Π(V). Let Φβ and Φ β, with β = Lgβ for g ∈ G(V), be a pair of charts. We shall denote by φ the change of charts given by Φβ→ S(V) Φ β ր Lβ(V) ↑ ψ L β(V) , Φ−1 β Φ β = LgAdg−1 = ψ. The C∗-isomorphism Adg−1 : L β(V) → Lβ(V) enables one to relate any construction done in both algebras. For instance, the forms θH and their unitary groups U (θH). Consider the group U (θH)β, which acts on the half-space Π(Lβ(V)) ⊂ Lβ(V). If h ∈ U (θH)β and z ∈ Π(Lβ(V)), denote by Λhz the action of h on z. We have the following diagram: Π(Lβ(V)) ↑ ψ Π(L β(V)) ψΛhψ−1 −→ Π(Lβ(V)) Λh−→ Π(L β(V)) ↑ ψ . Here Λh denotes the action of h ∈ U (θH) β on Π(L β(V)). We want to exhibit how the transition maps ψ translate this action on Π(Lβ(V)). Note that ψΛhψ−1 = LgAdg−1ΛhAdgL g−1 = LgΛhL g−1. (16) 30 Denote by Gβ(V) the group of invertible operators in Lβ(V). There is a representation u : Gβ(V) → U (θH)β , u(g) =(cid:18) g 0 0 g (cid:19) . (Recall that g = (g−1)∗). Using this representation we may write (16) above as ψΛhψ−1 = u(g)Λhu(g)−1. Moreover, if we denote by Adg : U (θH)β → U (θH) β, Adg = u(g)hu(g)−1, it is straightforward to verify that ψΛhψ−1 = AdgλAdg−1 h. (17) Therefore we have obtained the following construction: • To each β ∈ I(V) we associate the group U (θH )β. • If β = Lgβ, to the pair (β, β) we associate the group isomorphism ωβ β : U (θH) β → U (θH )β, ω β,β(h) = AdgλAdg−1 h. if β = Lgβ = Remarkably, this isomorphism ωβ β does not depend on the choice of g: Lg′β, then g and g′ give rise to the same isomorphism. This is another straightforward computation left to the reader. We shall refer to this system of groups and group isomorphism as the group of movements U (θ) of the Poincar´e half-space Π(V). This group of movements acts on Π(V) and is independent on the choice of coordinates. Thus, Π(V) is a homogeneous space of the group of movements U (θ). 11.4 The relative case So far, the building blocks of this construction are a Hilbertizable space V, the space of forms S(V) and its subset of inner products I(V). Now we want to restrict these constructions to a subalgebra A ⊂ L(V). Let us define: Definition 11.1. A C∗-pair is a pair (V,A), where V is a Hilbertizable space and A ⊂ L(V) is a subalgebra, such that there exists β ∈ I(V) such that A is a sub-C∗-algebra of Lβ(V). In such case, we say that β is adapted to A. Let us denote by I(A) = {β ∈ I(V) : β is adapted to A}. Denote by GA the invertible group of A. Clearly GA acts on I(V), and Lgβ ∈ I(A) if β ∈ I(A) and g ∈ GA. Denote by IO(A) ⊂ I(A) an orbit of this action. A C∗-triple is a triple (V,A,IO(A)) consisting of a C∗-pair and a orbit. With these data, we can repeat the former constructions, substituting L(V) by A, and I(V) by IO(A). inary part. The (restricted) group of movements U (θ) acts on Π(A) accordingly. The Poincar´e half-space Π(A) of A is formed by the elements of SO(A) with positive imag- 31 References [1] Atkin, C. J., The Finsler geometry of groups of isometries of Hilbert space, J. Austral. Math. Soc. Ser. A 42 (1987), 19 -222. [2] Atkin, C. J.,The Finsler geometry of certain covering groups of operator groups, Hokkaido Math. J. 18 (1989), 45 -77. [3] Beltita, D., Smooth homogeneous structures in operator theory. Chapman & Hall/CRC Monographs and Surveys in Pure and Applied Mathematics, 137. Chapman & Hall/CRC, Boca Raton, FL, 2006. [4] Bridson, M. R.; Haefliger, A., Metric spaces of non-positive curvature. Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], 319. Springer-Verlag, Berlin, 1999. [5] Burago, D.; Burago, Y.; Ivanov, S., A course in metric geometry. Graduate Studies in Mathematics, 33. American Mathematical Society, Providence, RI, 2001. [6] Corach, G.; Porta, H.; Recht, L., The geometry of the space of selfadjoint invertible elements in a C∗-algebra. Integral Equations Operator Theory 16 (1993), 333 -- 359. [7] Corach, G.; Porta, H.; Recht, L., The geometry of spaces of projections in C∗-algebras, Adv. Math. 101 (1993), 59 -- 77. [8] Corach, G.; Porta, H.; Recht, L., Geodesics and operator means in the space of positive operators, Internat. J. Math. 4 (1993), 193 -- 202. [9] Corach, G.; Porta, H.; Recht, L., Convexity of the geodesic distance on spaces of positive operators. Illinois J. Math. 38 (1994), no. 1, 87 -- 94. [10] de la Harpe, P., Classical Banach-Lie algebras and Banach-Lie groups of operators in Hilbert space. Lecture Notes in Mathematics, Vol. 285. Springer-Verlag, Berlin-New York, 1972. [11] de la Harpe, P., Classical groups and classical Lie algebras of operators.Operator algebras and applications, Part I (Kingston, Ont., 1980), pp. 477 -- 513, Proc. Sympos. Pure Math., 38, Amer. Math. Soc., Providence, R.I., 1982. [12] Gromov, M., Metric structures for Riemannian and non-Riemannian spaces. Based on the 1981 French original. With appendices by M. Katz, P. Pansu and S. Semmes. Translated from the French by Sean Michael Bates. Reprint of the 2001 English edition. Modern Birkhuser Classics. Birkhuser Boston, Inc., Boston, MA, 2007. [13] Larotonda, A. R., Notas sobre variedades diferenciables. (Spanish) [[Notes on differentiable manifolds]] Notas de Geometr´ıa y Topolog´ıa [Notes on Geometry and Topology], 1. Univer- sidad Nacional del Sur, Instituto de Matem´atica, Bah´ıa Blanca, 1980. [14] Mata-Lorenzo, L. E.; Recht, L., Infinite-dimensional homogeneous reductive spaces. Acta Cient. Venezolana 43 (1992), 76 -- 90. [15] Meyer, R., Adjoining a unit to an operator algebra, J. Operator Theory 46 (2001), 281 -- 288. 32 [16] Mostow, G. D., Some new decomposition theorems for semi-simple groups, Mem. Amer. Math. Soc. No. 14 (1955), 31 -- 54. [17] Porta, H.; Recht, L., Geometric embeddings of operator spaces, Illinois J. Math. 40 (1996), 151 -- 161. [18] Raeburn, I., The relationship between a commutative Banach algebra and its maximal ideal space, J. Functional Analysis 25 (1977), 366-390. [19] Siegel, C.L., Topics in complex function theory. Vol. I: Elliptic functions and uniformization theory. Translated from the original German by A. Shenitzer and D. Solitar. Interscience Tracts in Pure and Applied Mathematics, No. 25 Wiley-Interscience A Division of John Wiley & Sons, New York-London-Sydney 1969 ix+186 pp. [20] Siegel, C.L., Topics in complex function theory. Vol. II. Automorphic functions and abelian integrals. Translated from the German by A. Shenitzer and M. Tretkoff. With a preface by Wilhelm Magnus. Reprint of the 1971 edition. Wiley Classics Library. A Wiley-Interscience Publication. John Wiley & Sons, Inc., New York, 1988. xii+193 pp. [21] Siegel, C.L., Topics in complex function theory. Vol. III. Abelian functions and modu- lar functions of several variables. Translated from the German by E. Gottschling and M. Tretkoff. With a preface by Wilhelm Magnus. Reprint of the 1973 original. Wiley Clas- sics Library. A Wiley-Interscience Publication. John Wiley & Sons, Inc., New York, 1989. x+244 pp. [22] Upmeier, H., Symmetric Banach manifolds and Jordan C∗-algebras. North-Holland Math- ematics Studies, 104. Notas de Matem´atica [Mathematical Notes], 96. North-Holland Pub- lishing Co., Amsterdam, 1985. Esteban Andruchow Instituto de Ciencias, Universidad Nacional de Gral. Sarmiento, J.M. Gutierrez 1150, (1613) Los Polvorines, Argentina and Instituto Argentino de Matem´atica, 'Alberto P. Calder´on', CONICET, Saavedra 15 3er. piso, (1083) Buenos Aires, Argentina. e-mail: [email protected] Gustavo Corach Instituto Argentino de Matem´atica, 'Alberto P. Calder´on', CONICET, Saavedra 15 3er. piso, (1083) Buenos Aires, Argentina, and Depto. de Matem´atica, Facultad de Ingenier´ıa, Universidad de Buenos Aires, Argentina. e-mail: [email protected] L´azaro Recht Departamento de Matem´atica P y A, Universidad Sim´on Bol´ıvar Apartado 89000, Caracas 1080A, Venezuela e-mail: [email protected] 33
1906.05548	3	1906	2019-07-06T13:41:04	The cb-norm approximation of generalized skew derivations by elementary operators	[ "math.OA" ]	Let $A$ be a ring and $\sigma: A \to A$ a ring endomorphism. A generalized skew (or $\sigma$-)derivation of $A$ is an additive map $d: A \to A$ for which there exists a map $\delta:A \to A$ such that $d(xy)=\delta(x)y+\sigma(x)d(y)$ for all $x,y \in A$. If $A$ is a prime $C^$-algebra and $\sigma$ is surjective, we determine the structure of generalized $\sigma$-derivations of $A$ that belong to the cb-norm closure of elementary operators $\mathcal{E}\ell(A)$ on $A$; all such maps are of the form $d(x)=bx+axc$ for suitable elements $a,b,c$ of the multiplier algebra $M(A)$. As a consequence, if an epimorphism $\sigma: A \to A$ lies in the cb-norm closure of $\mathcal{E}\ell(A)$, then $\sigma$ must be an inner automorphism. We also show that these results cannot be extended even to relatively well-behaved non-prime $C^$-algebras like $C(X,\mathbb{M}_2 )$.	math.OA	math	THE CB-NORM APPROXIMATION OF GENERALIZED SKEW DERIVATIONS BY ELEMENTARY OPERATORS ILJA GOGI ´C Dedicated to the memory of my mentor and a friend, Professor R. M. Timoney Abstract. Let A be a ring and σ : A → A a ring endomorphism. A general- ized skew (or σ-)derivation of A is an additive map d : A → A for which there exists a map δ : A → A such that d(xy) = δ(x)y + σ(x)d(y) for all x, y ∈ A. If A is a prime C ∗-algebra and σ is surjective, we determine the structure of gen- eralized σ-derivations of A that belong to the cb-norm closure of elementary operators Eℓ(A) on A; all such maps are of the form d(x) = bx + axc for suit- able elements a, b, c of the multiplier algebra M (A). As a consequence, if an epimorphism σ : A → A lies in the cb-norm closure of Eℓ(A), then σ must be an inner automorphism. We also show that these results cannot be extended even to relatively well-behaved non-prime C ∗-algebras like C(X, M2). 1. Introduction A well-known consequence of Skolem-Noether theorem (see e.g. [8, Theorem 4.46]) is that a finite-dimensional central simple algebra A over a field F admits only inner derivations and inner automorphisms. This fact can be also proved by observing that all F-linear maps φ : A → A are elementary operators, i.e. they can be writ- ten as finite sums of two-sided multiplications x 7→ axb, with a, b ∈ A (see [8, Lemma 1.25, Theorem 1.30]). Hence, one can ask the following general question: Problem 1.1. Under which conditions on a semiprime ring (or an algebra) R are all derivations and/or automorphisms of R that are also elementary operators necessarily inner? In order to investigate Problem 1.1 it is sometimes convenient to consider maps d : R → R that comprise both (generalized) derivations and automorphisms. One particularly interesting class of such maps d is the following (see e.g. [19, 20]): Definition 1.2. Let R be a ring and let σ : R → R be a ring endomorphism. An additive map d : R → R is called a generalized σ-derivation (or a generalized skew derivation) if there exists a map δ : R → R such that (1.1) d(xy) = δ(x)y + σ(x)d(y) for all x, y ∈ R. Date: 3 July 2019. 2010 Mathematics Subject Classification. Primary 16W20, 47B47. Secondary 46L07, 16N60. Key words and phrases. Elementary operator, generalized skew derivation, epimorphism, prime C ∗-algebra. This work was supported by the Croatian Science Foundation under the project IP-2016-06- 1046. 1 2 ILJA GOGI ´C In case when R is semiprime and σ is surjective, in [10] we considered the problem of determining the structure of generalized σ-derivations d : R → R that are also elementary operators. In order to give a description of such maps, we used standard techniques of the theory of rings of quotients. As a consequence of the proof of [10, Theorem 1.2], which is the main result of that paper, we showed that if R is semiprime and centrally closed, then a derivation δ : R → R (resp. a ring epimorphism σ : R → R) is an elementary operator if and only if δ (resp. σ) is an inner derivation (resp. inner automorphism), thus giving an affirmative answer to Problem 1.1 for this class of rings. It is also interesting to consider Problem 1.1 in the setting of C ∗-algebras. More- over, when working with C ∗-algebras, it is well-known that their derivations, auto- morphisms and elementary operators are completely bounded. This motivates us to consider the following analytic variation of Problem 1.1: Problem 1.3. Under which conditions on a C ∗-algebra A are all derivations and/or automorphisms of A that admit a cb-norm approximation by elementary operators necessarily inner? In our previous work we considered Problem 1.3 only for derivations. More precisely, we showed that all such derivations of A are inner in a case when A is prime [12, Theorem 4.3] or central [12, Theorem 5.6]. This result was further extended in [14, Theorem 1.5] for unital C ∗-algebras whose every Glimm ideal is prime. The latter result in particular applies to derivations of local multiplier algebras (see e.g. [2]), since their Glimm ideals are prime [2, Corollary 3.5.10]. Hopefully, this result might be useful in order to give an answer to Pedersen's problem from 1978, which asks whether all derivations of local multiplier algebras are inner [24]. Motivated by these results, in this paper we consider the problem of determining the structure of generalized σ-derivations of C ∗-algebras A, with σ surjective, that can be approximated by elementary operators in the cb-norm. The main result of this paper is Theorem 3.1, where we fully describe the struc- ture of such maps when A is a prime C ∗-algebra. In particular, if A is prime, we show that if an epimorphism σ : A → A admits a cb-norm approximation by elementary operators, then σ must be an inner automorphism of A (Corollary 3.8). The proof of Theorem 3.1 is given through several steps in Section 3. In Section 4 we consider the possible generalization of Theorem 3.1 and its conse- quences for C ∗-algebras that are not necessarily prime. However, this generalization will not be possible even for some well-behaved C ∗-algebras, like homogeneous C ∗- algebras, even thought they have only inner derivations (see [31, Theorem 1] for unital case and [13, Proposition 3.2] for general case). In fact, we show that for each n ≥ 2 there is a compact manifold Xn such that the C ∗-algebra C(Xn, Mn) (where Mn is the algebra of n × n complex matrices) admits outer automorphisms that are simultaneously elementary operators (Proposition 4.2). We also give an example of a unital separable C ∗-algebra A that admits both outer derivations and outer automorphisms that are also elementary operators (Proposition 4.6). 2. Preliminaries Let R be a ring. As usual, by Z(R) we denote its centre. By an ideal of R we always mean a two-sided ideal. An ideal I of R is said to be essential if I has a 3 non-zero intersection with every other non-zero ideal of R. If R is unital, by R× we denote the set of all invertible elements of R. Recall that a ring R is said to be semiprime if for x ∈ R, xRx = {0} implies x = 0. If in addition for x, y ∈ R, xRy = {0} implies x = 0 or y = 0, R is said to be prime. If R is a semiprime ring by M (R) we denote its multiplier ring (see [2, Sec- tion 1.1]). Note that M (R) is also a semiprime ring and that R is prime if and only if M (R) is prime [2, Lemma 1.1.7]. For each a ∈ M (R)× we denote by Ad(a) the automorphism x 7→ axa−1. We call such automorphisms inner (when R is unital this coincides with the standard notion of the inner automorphism). Remark 2.1. If R is a semiprime ring and d : R → R a generalized σ-derivation (Definition 1.2), then a map δ is obviously uniquely determined by d. Moreover, δ is a σ-derivation, that is δ is an additive map that satisfies (2.1) for all x, y ∈ R. Indeed, using (1.1) and additivity of d and σ, for all x, y, z ∈ R we have δ(xy) = δ(x)y + σ(x)δ(y) d((x + y)z) = δ(x + y)z + σ(x + y)d(z) = δ(x + y)z + σ(x)d(z) + σ(y)d(z) and d(xz + yz) = d(xz) + d(yz) = δ(x)z + σ(x)d(z) + δ(y)z + σ(y)d(z). Subtracting these two equations we get the additivity of δ. Similarly, subtracting the next two equations d(xyz) = δ(xy)z + σ(xy)d(z) = δ(xy)z + σ(x)σ(y)d(z), d(xyz) = δ(x)yz + σ(x)d(yz) = δ(x)yz + σ(x)δ(y)z + σ(x)σ(y)d(z) shows (2.1). Further, it is now easy to verify that the map ρ := d − δ is a left R-module σ-homomorphism, that is ρ : R → R is an additive map that satisfies (2.2) for all x, y ∈ R. Therefore, every generalized σ-derivation can be uniquely decom- posed as ρ(xy) = σ(x)ρ(y) d = δ + ρ, where δ is a σ-derivation and ρ is a left R-module σ-homomorphism. In particular, generalized σ-derivations simultaneously generalize σ-derivations of R (we get them for d = δ) and left R-module σ-homomorphisms of R (we get them for δ = 0). Example 2.2. The simplest examples of σ-derivations δ : R → R are inner σ- derivations, i.e. those of the form where a is some element of M (R). Further, any map of the form δ(x) = ax − σ(x)a, ρ(x) = σ(x)a, where a ∈ M (R), is a left R-module σ-homomorphism. If R is unital, then all left R-module σ-homomorphism of R are of this form (M (R) = R in this case). 4 ILJA GOGI ´C Throughout this paper A will be a C ∗-algebra. Then M (A) has a structure of a C ∗-algebra and is called the multiplier algebra of A. It is well-known (and easily checked) that A, as a ring, is semiprime. As usual, by CB(A) we denote the set of all completely bounded maps φ : A → A (see e.g. [23]). For S ⊆ CB(A) we denote by Scb the cb-norm closure of S. The most prominent class of completely bounded maps on A are elementary operators, i.e. those that can be expressed as finite sums of two-sided multiplications Ma,b : x 7→ axb, where a and b are elements of M (A). We denote the set of all elementary operators on A by Eℓ(A). It is well-known that elementary operators on C ∗-algebras are completely bounded. In fact, we have the following estimate for their cb-norm: (2.3) , where k · kh is the Haagerup tensor norm on the algebraic tensor product M (A) ⊗ M (A), i.e. (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi ai ⊗ bi(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)h Mai,bi(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)cb ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi i(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) i bi(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi : t =Xi aia∗ b∗ 1 1 2 ktkh = inf  By inequality (2.3) the mapping . ai ⊗ bi  (M (A) ⊗ M (A),k · kh) → (Eℓ(A),k · kcb) given by Xi is a well-defined contraction. Its continuous extension to the completed Haagerup tensor product M (A)⊗h M (A) is known as a canonical contraction from M (A)⊗h M (A) to CB(A) and is denoted by ΘA. We have the following result (see [2, Proposition 5.4.11]): ai ⊗ bi 7→Xi Mai,bi . Theorem 2.3 (Mathieu). ΘA is isometric if and only if A is a prime C ∗-algebra. If A is not a prime note that ΘA is not even injective. Indeed, in this case there are non-zero elements a, b ∈ A such that aAb = {0}. Then a ⊗ b defines a non-zero tensor in M (A) ⊗ M (A) but Ma,b = 0. For a unital but not necessarily prime C ∗-algebra one can construct a central Haagerup tensor product A ⊗Z,h A and consider the induced contraction ΘZ A : A⊗Z,h A → CB(A) (the questions when ΘZ A is isometric or injective were treated in [30, 5, 4]). 3. Results We begin this section by stating the main result of this paper: Theorem 3.1. Let A be a prime C ∗-algebra and let d : A → A be a generalized σ-derivation, with σ surjective. The following conditions are equivalent: (i) d ∈ Eℓ(A)cb. (ii) Either d is a left multiplication implemented by some element of M (A) or σ is an inner automorphism of A. In the latter case, if d = δ + ρ is a decomposition as in Remark 2.1, then δ is an inner σ-derivation and ρ is a right multiplication of σ by some element of M (A). 5 Remark 3.2. (i) Note that any left multiplication d = Ml,1, where l ∈ M (A), is a generalized σ-derivation with respect to any ring endomorphism σ : A → A. Indeed, for any such σ, let δ(x) = lx − σ(x)l and ρ(x) = σ(x)l. Then obviously d = δ + ρ, so in this case we cannot say anything about the epimorphism σ and the corresponding maps δ and ρ. (ii) If d is not a left multiplication, then following the second case of part (ii) of Theorem 3.1 we have σ = Ad(a), δ(x) = bx − σ(x)b and ρ(x) = σ(x)b′, for some a ∈ M (A)× and b, b′ ∈ M (A), b 6= b′. In particular, d is of the form d(x) = bx + axc, where c := a−1(b′ − b). Remark 3.3. In the sequel of this section we assume that A is an infinite-dimensional prime C ∗-algebra, since otherwise [22, Theorem 6.3.8] and the primeness of A would imply that A is isomorphic to the matrix algebra Mn for some non-negative inte- [8, ger n. Then every linear map φ : A → A is an elementary operator (see e.g. Lemma 1.25]), so Theorem 3.1 is just a simple consequence of [10, Theorem 1.2] (the maximal right ring of quotients Qmr(A) in this case coincides with A). For the proof of Theorem 3.1 we will need some auxiliary results. We start with the following: Lemma 3.4. Let A be a prime C ∗-algebra. Suppose that σ : A → A is a ring epimorphism for which there exists a non-zero element a ∈ M (A) such that (3.1) ax = σ(x)a ∀x ∈ A. Then a is invertible in M (A), so that σ = Ad(a) is an inner automorphism of A. Before proving Lemma 3.4 recall from [2] (see also [6]) that an essentially defined double centralizer on a semiprime ring R is a triple (L,R, I), where I is an essential ideal of R, L : I → R is a left R-module homomorphism, R : I → R is a right R-module homomorphism such that L(x)y = xR(y) for all x ∈ I. One can form the symmetric ring of quotients Qs(R) which is characterized (up to isomorphism) by the following properties: (i) R is a subring of Qs(R); (ii) for any q ∈ Qs(R) there is an essential ideal I of R such that qI + Iq ⊆ R; (iii) if 0 6= q ∈ Qs(R) and I is an essential ideal of R, then qI 6= 0 and Iq 6= 0; (iv) for any essentially defined double centralizer (L,R, I) on R there exists q ∈ Qs(R) such that L(x) = qx and R(x) = xq for all x ∈ I. In a case when R = A is a C ∗-algebra, Qs(A) has a natural structure as a unital complex ∗-algebra, whose involution is positive definite. An element q ∈ Qs(A) is called bounded if there is λ ∈ R+ such that q∗q ≤ λ1, in a sense that there is a finite number of elements q1, . . . , qn ∈ Qs(A) such that The set Qb(A) of all bounded elements of Qs(A) has a pre-C ∗-algebra structure with respect to the norm q∗q + q∗ i qi = λ1. n Xi=1 kqk2 = inf{λ ∈ R+ : q∗q ≤ λ1}, 6 ILJA GOGI ´C which clearly extends the norm of A. One can easily check that an element q ∈ Qs(A) is bounded if and only if it can be represented by a bounded (continuous) essentially defined double centralizer (see [2, p. 57]). We call Qb(A) the bounded symmetric algebra of quotients of A and its completion Mloc(A) the local multiplier algebra of A. Note that Mloc(A) has a structure of a C ∗-algebra as a completion of a pre-C ∗-algebra. Proof of Lemma 3.4. First note that a non-zero element a ∈ M (A) that satisfies (3.1) cannot be a zero-divisor. Indeed, if there exists x ∈ M (A) such that ax = 0 then for each y ∈ A we have ayx = σ(y)ax = 0, so that aAx = {0}. Since A is prime and since A is an essential ideal of M (A), a 6= 0 implies x = 0. Similarly, if xa = 0 then for all y ∈ A we have xσ(y)a = xay = 0. Since σ is surjective, this is equivalent to xAa = {0}, so the primeness of A again implies x = 0. We now show that a is invertible in M (A). Since M (A) is a unital C ∗-subalgebra of Mloc(A), we have M (A)× = M (A) ∩ Mloc(A)×, so it suffices to show that a is invertible in Mloc(A). In order to do this, first note that aA is a non-zero ideal of A, hence essential, since A is prime. Indeed, since σ is surjective, we have A = σ(A), hence AaA = {σ(x)ay : x, y ∈ A} = {axy : x, y ∈ A} ⊆ aA. In particular, for α ∈ C and x ∈ A we have (σ(αx)−ασ(x))aA = {0}, which implies σ(αx) = ασ(x). Therefore σ is a linear map, hence an algebra epimorphism. We define maps L,R : aA → A by L(ax) = σ(x) and R(ax) = x. That the maps L and R are well-defined follows from the fact that a is not a (left) zero-divisor. Clearly, R is a right A-module homomorphism. Next, for x, y ∈ A we have L(σ(x)ay) = L(axy) = σ(xy) = σ(x)σ(y) = σ(x)L(ay) and L(ax)ay = σ(x)ay = axy = axR(ay). This shows that (L,R, aI) is an essentially defined double centralizer on A. Since A is prime, by [2, Corollary 2.2.15] all essentially defined double centralizers are In particular, there exists an element b ∈ Qb(A) ⊆ automatically continuous. Mloc(A) such that σ(x) = L(ax) = axb = σ(x)ab and x = R(ax) = bax for all x ∈ A. Since σ(A) = A, this is equivalent to A(1− ab) = {0} and (1− ba)A = {0}. Hence, a is invertible in Mloc(A) and a−1 = b. This completes the proof. (cid:3) Recall from [29, Definition 3.2] that a sequence (an) in an infinite-dimensional nan is norm convergent is said to be n=1 αnan = 0 C ∗-algebra B such that the series P∞ strongly independent if for every sequence (αn) ∈ ℓ2, equality P∞ implies αn = 0 for all n ∈ N. [1, Lemma 2.3]. n=1 a∗ The next fact can be deduced from [7, Proposition 1.5.6], [29, Lemma 4.1] and Remark 3.5. Let B be an infinite-dimensional C ∗-algebra. 7 (i) Every tensor t ∈ B ⊗h B has a representation as a convergent series t = P∞ n=1 an ⊗ bn, where (an) and (bn) are sequences in B such that the series P∞ n=1 ana∗ nbn are norm convergent. Moreover, the sequence independent, then t = 0 if and only if an = 0 for all n ∈ N. n and P∞ n=1 an ⊗ bn is a representation of t as above, with (bn) strongly (bn) can be chosen to be strongly independent. (ii) If t = P∞ n=1 b∗ Corollary 3.6. Let A be a prime C ∗-algebra and suppose that (an), (bn) are se- nbn are norm conver- n=1 ana∗ n=1 b∗ quences in M (A) such that the series P∞ gent, with (bn) strongly independent. If ∞ n and P∞ (3.2) anxbn = 0 Xn=1 for all x ∈ A, then an = 0 for all n ∈ N. Proof. If t :=P∞ Proposition 3.7. Let A be a prime C ∗-algebra and let σ : A → A be a ring epimorphism. If ρ : A → A is a non-zero left A-module σ-homomorphism, then the following conditions are equivalent: n=1 an⊗bn ∈ M (A)⊗hM (A), then (3.2) is equivalent to ΘA(t) = 0. Since A is prime, by Theorem 2.3 ΘA is isometric (hence injective), so t = 0. The claim now follows from part (ii) of Remark 3.5. (cid:3) (i) ρ ∈ Eℓ(A)cb. (ii) There are elements a, p ∈ M (A), with a invertible and p 6= 0, such that σ = Ad(a) and ρ(x) = σ(x)p = axa−1p for all x ∈ A. Proof. Since A is prime, by Theorem 2.3 the canonical contraction ΘA : M (A) ⊗h M (A) → CB(A) is isometric. In particular, the image of ΘA is closed in the cb-norm so Eℓ(A)cb coincides with the image of ΘA. Hence, there is a tensor t ∈ M (A) ⊗h M (A) such that ρ = ΘA(t). By Remark 3.5, we can write t = P∞ n=1 an ⊗ bn, where (an) and (bn) are sequences in M (A) such that the series P∞ n=1 ana∗ nbn are norm convergent, with (bn) strongly independent. n and P∞ Then (2.2) implies n=1 b∗ ∞ (anx − σ(x)an)ybn = 0 Xn=1 anx = σ(x)an for all x, y ∈ A. By Corollary 3.6 we have (3.3) for all n ∈ N. Since ρ is non-zero, there is n0 ∈ N such that an0 6= 0. By Lemma 3.4 a := an0 is invertible in M (A). Hence σ = Ad(a) is an inner automorphism of A. Finally, if p :=P∞ n=1 anbn ∈ M (A), using (3.3) we get anxbn = σ(x) ∞ Xn=1 Xn=1 anbn! = σ(x)p = axa−1p. ρ(x) = ∞ (cid:3) As a direct consequence of Proposition 3.7 we get: 8 ILJA GOGI ´C Corollary 3.8. If A is a prime C ∗-algebra then every ring epimorphism σ : A → A that lies in Eℓ(A)cb must be an inner automorphism of A. The next fact can be deduced from the proof of [12, Theorem 4.3]. For com- pleteness, we include a proof. ∞ f (x) ⊗ 1 = Lemma 3.9. Let B be a unital infinite-dimensional C ∗-algebra and let f, g, h : B → B be any functions with f 6= 0. Suppose that for all x ∈ B we have the following equality of tensors in B ⊗h B Xn=1 (3.4) (ang(x) + h(x)an) ⊗ bn, where (an) and (bn) are sequences in B such that the seriesP∞ n=1 b∗ are norm convergent, with (bn) strongly independent. Then there is a non-zero el- ement b ∈ B such that (3.5) for all x ∈ B. Proof. Choose x0 ∈ B such that f (x0) 6= 0 and let ϕ ∈ B ∗ be an arbitrary bounded linear functional such that ϕ(f (x0)) 6= 0. If for x = x0 we act on the equality (3.4) with the right slice map Rϕ : B⊗h B → B, Rϕ : a⊗ b 7→ ϕ(a)b (see e.g. [29, Section 4]), we obtain n andP∞ f (x) = bg(x) + h(x)b n=1 ana∗ nbn (3.6) ϕ(f (x0)) · 1 = For n ∈ N let ϕ(ang(x0) + h(x0)an)bn. ∞ Xn=1 αn := ϕ(ang(x0) + h(x0)an) ϕ(f (x0)) . Note that (αn) ∈ ℓ2, since all bounded linear functionals on C ∗-algebras are com- pletely bounded (see e.g. n=1(ang(x0) + h(x0)an)(ang(x0) + h(x0)an)∗ is norm convergent. Then (3.6) can be rewritten as [23, Proposition 3.8]) and the series P∞ n=1 αnbn = 1, so by (3.4) we have P∞ ∞ Xn=1 (αnf (x) − ang(x) − h(x)an) ⊗ bn = 0 for all x ∈ B. Consequently, since (bn) is strongly independent, Remark 3.5 (ii) implies that αnf (x) = ang(x) + h(x)an for all n ∈ N and x ∈ B. Since P∞ n=1 αnbn = 1, there is some n0 ∈ N such that αn0 6= 0. If b := (1/αn0)an0 , then the above equation is obviously equivalent to (3.5). Also, b 6= 0 since f 6= 0. Proof of Theorem 3.1. (ii) =⇒ (i). This is trivial (see also Remark 3.2). (i) =⇒ (ii). Assume that d ∈ Eℓ(A)cb and that d 6= Ml,1 for all l ∈ M (A). In particular d 6= 0. Using the same arguments from the beginning of the proof of Proposition 3.7, we see that there is a tensor t ∈ M (A) ⊗h M (A) such that d = ΘA(t). By Remark 3.5, we can write t = P∞ n=1 an ⊗ bn, where (an) and (bn) (cid:3) are sequences of M (A) such that the series P∞ convergent, with (bn) strongly independent. Using (1.1) for all x, y ∈ A we get n and P∞ n=1 ana∗ n=1 b∗ nbn are norm ∞ 9 δ(x)y = (anx − σ(x)an)ybn, Xn=1 or equivalently (3.7) ΘA(δ(x) ⊗ 1)) = ΘA ∞ Xn=1 (anx − σ(x)an) ⊗ bn! . Again, since ΘA is isometric (hence injective), (3.7) is equivalent to the equality δ(x) ⊗ 1 = (anx − σ(x)an) ⊗ bn ∞ Xn=1 of tensors in M (A) ⊗h M (A) for all x ∈ A. If δ = 0, then d must be a non-zero left A-module σ-homomorphism of A (see Remark 2.1) so the claim follows directly from Proposition 3.7. If δ 6= 0, Lemma 3.9 implies that there is a non-zero element b ∈ M (A) such that δ(x) = bx − σ(x)b for all x ∈ A. If we decompose d = δ + ρ as in Remark 2.1, the map ρ′ : A → A defined by ρ′(x) := ρ(x) − σ(x)b = d(x) − bx is obviously a left A-module σ-homomorphism of A that lies in Eℓ(A)cb (since d does). Since, by assumption, d is not a left multiplication, ρ′ is non-zero. Hence, by Proposition 3.7 there are elements a, p ∈ M (A) with a invertible and p 6= 0 such that σ = Ad(a) and ρ′(x) = σ(x)p for all x ∈ A. In particular, if we put b′ := b + p, we get ρ(x) = σ(x)b′, which completes the proof. (cid:3) 4. Counterexamples and further remarks In [12, 14] we considered derivations of unital C ∗-algebras A that lie in Eℓ(A)cb. We showed that all such derivations are inner in a case when A is prime [12, Theorem 4.3] or central [12, Theorem 5.6], or more generally, when A is a unital C ∗-algebra whose every Glimm ideal is prime [14, Theorem 1.5]. In light of this, it is natural to ask if one can extend Corollary 3.8 in its original form (and consequently Theorem 3.1) for similar classes of C ∗-algebras. However, this will not be possible, even for relatively well-behaved C ∗-algebras like homo- geneous C ∗-algebras. In fact, we will now show that for all n ≥ 2, a C ∗-algebra An = C(P U (n), Mn), where P U (n) = U (n)/S1 is the projective unitary group, admits outer automorphisms which are simultaneously elementary operators on An (Proposition 4.2). In order to show this, first suppose that A is a general separable n-homogeneous C ∗-algebra (i.e. all irreducible representations of A have the same finite dimension n). Then by [17, Theorem 4.2] the primitive spectrum X := Prim(A) is a (locally compact) Hausdorff space and by a well-known theorem of Fell [11, Theorem 3.2] and Tomiyama-Takesaki [32, Theorem 5] there is a locally trivial bundle E over X with fibre Mn and structure group Aut∗(Mn) ∼= P U (n) such that A is isomor- phic to the C ∗-algebra Γ0(E) of continuous sections of E that vanish at infinity. Moreover, any two such algebras Ai = Γ0(Ei) with primitive spectra Xi (i = 1, 2) 10 ILJA GOGI ´C η Z (A) Let us denote by Aut∗ are isomorphic if and only if there is a homeomorphism f : X1 → X2 such that E1 ∼= f ∗(E2) (the pullback bundle) as bundles over X1 (see [32, Theorem 6]). Thus, we may identify A with Γ0(E). Further, by Dauns-Hofmann theorem [27, Theorem A.34] we can identify Z(A) with C0(X) and Z(M (A)) with Cb(X). Z (A) the set of all Z(M (A))-linear ∗-automorphisms of A (i.e. σ(x∗) = σ(x)∗ and σ(zx) = zσ(x) for all z ∈ Z(M (A)) and x ∈ A) and by InnAut∗(A) the set of automorphisms σ of A that are of the form σ = Ad(u), for some unitary element u ∈ M (A). Obviously InnAut∗(A) ⊆ Aut∗ Z(A). It is a very interesting (and non-trivial) problem to describe when do we have Aut∗ Z (A) = InnAut∗(A) (see e.g. [28, 18, 25, 26] for some results regarding this question). In particular, we have the following consequence of [25, Theorem 2.1] (see also [25, 2.19]): Theorem 4.1 (Phillips-Raeburn). If A = Γ0(E) is a separable n-homogeneous C ∗-algebra with primitive spectrum X, we have the exact sequence −→ H 2(X; Z) 0 −→ InnAut∗(A) −→ Aut∗ of abelian groups, where H 2(X; Z) is the second integral Cech cohomology group of X. Further, the image of η is contained in the torsion subgroup of H 2(X; Z). Proposition 4.2. For n ≥ 2 let An = C(P U (n), Mn). Then every derivation of An is inner, but An admits an outer automorphism that is also an elementary operator. Remark 4.3. A C ∗-algebra A is said to be quasicentral if any element a ∈ A can be decomposed as a = zb for some b ∈ A and z ∈ Z(A) (see [9, 3, 12] for other characterizations of such algebras). If every closed ideal of A is quasicentral, note that any Z(A)-linear map φ : A → A preserves all closed ideals of A (i.e. φ(I) ⊆ I for any such ideal I). Indeed, since any a ∈ I can be decomposed as a = zb, with z ∈ Z(I) and b ∈ I and since Z(I) ⊆ Z(A), we have φ(a) = φ(zb) = zφ(b) ∈ I. This observation in particular applies to n-homogeneous C ∗-algebras A ∼= Γ0(E). Indeed, using the fact that the bundle E is locally trivial one can easily check that n-homogeneous C ∗-algebras are quasicentral. Also, every closed ideal of an n- homogeneous C ∗-algebra is also an n-homogeneous C ∗-algebra, hence quasicentral. Further, if A is unital (and n-homogeneous), every bounded Z(A)-linear map φ : A → A is an elementary operator on A. This follows directly from the above observation and Magajna's theorem [21, Theorem 1.1]. Therefore, for every unital n-homogeneous C ∗-algebra A we have Aut∗ Z (A) ⊆ Eℓ(A). Proof of Proposition 4.2. That all derivations of An are inner follows from [31, The- orem 1]. On the other hand, by [16, Section IV] An admits an automorphism Z(An) \ InnAut∗(An) (note that H 2(P U (n); Z) = Zn, so all elements of σ ∈ Aut∗ H 2(P U (n); Z) are torsion elements). By Remark 4.3, σ ∈ Eℓ(An). Suppose that σ = Ad(a) for some a ∈ A× n . Then, since σ is ∗-preserving, for all x ∈ An we have Hence a∗a ∈ Z(An), so a = √a∗a ∈ Z(An). Therefore, if u := a−1a, then u is a unitary element of An and σ = Ad(u) ∈ InnAut∗(An); a contradiction. On the other hand, if An = C(P U (n), Mn) as before, every σ ∈ Aut∗ Z (An) is implemented by some unitary element of Qb(A). This follows from the following fact: ax∗a−1 = σ(x∗) = σ(x)∗ = (a∗)−1x∗a∗. (cid:3) 11 Proposition 4.4. Let A = Γ0(E) be a separable n-homogeneous C ∗-algebra whose primitive spectrum X is locally contractable. Then for every σ ∈ Aut∗ Z (A) there is a unitary element u ∈ Qb(A) such that σ = Ad(u). Proof. We first show that there is a dense open subset U of X such that H 2(U ; Z) = 0. This can be shown by using the similar arguments as in the proof of [15, Indeed, let F be a collection of all families V consisting of mutu- Lemma 3.1]. ally disjoint open contractable subsets of X. We use the standard set-theoretic inclusion for partial ordering. If C is a chain in F, then obviously S C is an upper bound of C in F. Therefore, applying Zorn's lemma, we obtain a maximal family M in F. Let U be the union of all members of M. Since U is a disjoint union of contractable spaces, we have H 2(U ; Z) = 0. Since X is locally contractable (and regular), the maximality of M implies that U is a dense (evidently open) subset of X. Now let I := Γ0(EU ), where EU is a restriction bundle of E to U . Since U is dense in X, I is an essential closed ideal of A. If σ ∈ Aut∗ Z(A), by Remark 4.3 Z (I). Since Prim(I) = U and H 2(U ; Z) = 0, by we have σ(I) ⊆ I, so σI ∈ Aut∗ Theorem 4.1 there is a unitary element u ∈ M (I) such that σ(x) = uxu∗ for all x ∈ I. Since I is an essential ideal of A, we also have σ(x) = uxu∗ for all x ∈ A. If we define Lu,Ru : I → A by Lu(x) = xu and Ru(x) = ux, then obviously (Lu,Ru, I) is a bounded essentially defined double centralizer of A, so u ∈ Qb(A) and σ = Ad(u). (cid:3) Problem 4.5. Can we omit the assumption of local contractibility of the space X in Proposition 4.4, that is if A is a general separable n-homogeneous C ∗-algebra, are all automorphisms σ ∈ Aut∗ Z(A) of the form σ = Ad(u) for some unitary u ∈ Qb(A)? In [12, Example 6.1] we gave an example of a unital separable C ∗-algebra A which admits outer derivations that are also elementary operators. We now show that the same C ∗-algebra admits outer automorphisms that are also elementary operators: Proposition 4.6. Let B := C([1,∞], M2) be a C ∗-algebra that consists of all continuous functions from the extended interval [1,∞] to the C ∗-algebra M2. If A is a C ∗-subalgebra of B that consists of all a ∈ B such that a(n) =(cid:20) λn(a) 0 0 λn+1(a) (cid:21) (n ∈ N) for some convergent sequence (λn(a)) of complex numbers, then A admits an outer automorphism which is also an elementary operator on A. Proof. Let f : [1,∞] → C be a continuous function such the series P∞ n=1 f (n) does not converge and such that the range of f is a subset of the imaginary axis. We define an element b ∈ B by b :=(cid:20) f 0 0 0 (cid:21) . It was observed in [12, Section 6] that δ := ad(b) defines an outer derivation of A which lies in the operator norm closure of the space of all inner derivations of A. Since b∗ = −b, δ is a ∗-derivation (i.e. δ(x∗) = δ(x)∗ for all x ∈ A). Hence 12 ILJA GOGI ´C σ := exp(δ) defines a ∗-automorphism of A (see e.g. [2, Section 4.3]). Note that σ = Ad(u), where u := exp b =(cid:20) exp f 0 0 1 (cid:21) is a unitary element of B. Since the exponential map is continuous, and since δ in fact lies in the operator norm closure of the space of all inner ∗-derivations of A, we conclude that σ lies in the operator norm closure of the set of all inner ∗-automorphisms of A. Claim 1. σ is an outer automorphism of A. On the contrary, suppose that there exists an invertible element a ∈ A such that σ = Ad(a). Then u∗ax = xu∗a for all x ∈ A. Since J := {a ∈ A : a(n) = 0 for all n ∈ N} defines an closed essential ideal of both A and B, we conclude that u∗a is an invertible central element of B. Hence, there exists an invertible continuous function ϕ ∈ C([1,∞]) such that a = (ϕ ⊕ ϕ)u =(cid:20) exp f · ϕ 0 ϕ (cid:21) . 0 ϕ(n + 1) = exp n Xk=1 f (k)! ϕ(1) Since a ∈ A, we conclude that ϕ(n + 1) = (exp f (n))ϕ(n), and consequently (4.1) n=1 f (n) converges, a contradiction. for all n ∈ N. Finally, since limx→∞ ϕ(x) exists and since ϕ(1) 6= 0, (4.1) implies that the series P∞ Claim 2. σ is an elementary operator on A. As noted, σ lies in the operator norm closure of the set of all inner ∗-auto- morphisms of A. In particular, σ lies in the operator norm closure of the set of all elementary operators on A. But the latter set is closed in the operator norm by [12, Lemma 6.6]. Hence, σ is an elementary operator. (cid:3) We end this paper with the following question: Problem 4.7. Is Corollary 3.8 true for all von Neumann algebras? In particular, if an automorphism σ of a von Neumann algebra A is also an elementary operator, is σ necessarily an inner automorphism? References [1] S. D. Allen, A. M. Sinclair and R. R. Smith, The ideal structure of the Haagerup tensor product of C ∗-algebras, J. reine angew. Math. 442 (1993), 111 -- 148. [2] P. Ara and M. Mathieu, Local Multipliers of C ∗-algebras, Springer, London, 2003. [3] R. J. Archbold, Density theorems for the centre of a C ∗-algebra, J. London Math. Soc. (2), 10 (1975), 189 -- 197. [4] R. J. Archbold, D. W. B. Somerset and R. M. Timoney, On the central Haagerup tensor product and completely bounded mappings of a C ∗-algebra, J. Funct. Anal. 226 (2005), 406 -- 428. [5] R. J. Archbold, D. W. B. Somerset and R. M. Timoney, Completely bounded mappings and simplicial complex structure in the primitive ideal space of a C ∗-algebra, Trans. Amer. Math. Soc. 361 (2009), 1397 -- 1427. [6] K. I. Beidar, W. S. Martindale III, A. V. Mikhalev, Rings with generalized identities, Marcel Dekker, Inc., 1996. 13 [7] D. P. Blecher and C. Le Merdy, Operator algebras and Their modules, Clarendon Press, Oxford, 2004. [8] M. Bresar, Introduction to Noncommutative Algebra, Springer International Publishing, 2014. [9] C. Delaroche, Sur les centres des C ∗-alg`ebres II, Bull. Sc. Math., 92 (1968), 111 -- 128. [10] D. Eremita, I. Gogi´c, D. Ilisevi´c, Generalized skew derivations implemented by elementary operators, Algebr. Represent. Theory 17 (2014), 983 -- 996. [11] J. M. G. Fell, The structure of algebras of operator fields, Acta Math. 106 (1961) 233 -- 280. [12] I. Gogi´c, Derivations which are inner as completely bounded maps, Oper. Matrices 4 (2010), 193 -- 211. [13] I. Gogi´c, Derivations of subhomogeneous C ∗-algebras are implemented by local multipliers, Proc. Amer. Math. Soc., 141 (2013), 3925 -- 3928. [14] I. Gogi´c, On derivations and elementary operators on C ∗-algebras, Proc. Edinb. Math. Soc., 56 (2013), 515 -- 534. [15] I. Gogi´c, The local multiplier algebra of a C ∗-algebra with finite dimensional irreducible representations, J. Math. Anal. Appl., 408 (2013), 789 -- 794. [16] R. V. Kadison and J. R. Ringrose, Derivations and Automorphisms of Operator Algebras, Commun. math. Phys. 4 (1967), 32 -- 63. [17] I. Kaplansky, The structure of certain operator algebras, Trans. Amer. Math. Soc. 70 (1951) 219 -- 255. [18] E. C. Lance, Automorphisms of certain operator algebras, Amer. J. Math. 91 (1969) 160 -- 174. [19] T.-K. Lee, Generalized skew derivations characterized by acting on zero products. Pacific J. Math. 216 (2004), 293 -- 301. [20] T.-K. Lee and K.-S. Liu, Generalized skew derivations with algebraic values of bounded degree, Houston J. Math. 39 (2013) 733 -- 740. [21] B. Magajna, Uniform approximation by elementary operators, Proc. Edinb. Math. Soc. (2) 52 (2009), no. 3, 731 -- 749. [22] G. J. Murphy, C ∗-Algebras and Operator Theory, Academic Press, San Diego, 1990. [23] V. Paulsen, Completely Bounded Maps and Operator Algebras, Cambridge University Press, 2003. [24] G. K. Pedersen, Approximating derivations on ideals of C ∗-algebras, Invent. Math. 45 (1978), 299 -- 305. [25] J. Phillips and I. Raeburn, Automorphisms of C ∗-algebras and second Cech cohomology, Indiana Univ. Math. J. 29 (1980) 799 -- 822. [26] J. Phillips, I. Raeburn and J. L. Taylor, Automorphisms of certain C ∗-algebras and torsion in second Cech cohomology, Bull. London Math. Soc. 14 (1982) 33 -- 38. [27] I. Raeburn and D. P. Williams, Morita Equivalence and Continuous-Trace C ∗-Algebras, Mathematical Surveys and Monographs 60, Amer. Math. Soc., Providence, RI, 1998. [28] M.-s. B. Smith, On automorphism groups of C ∗-algebras, Trans. Amer. Math. Soc. 152 (1970) 623 -- 648. [29] R. R. Smith, Completely Bounded Maps and the Haagerup Tensor Product, J. Functional Analysis 102 (1991), 156 -- 175. [30] D. W. Somerset, The central Haagerup tensor product of a C ∗-algebra, J. Oper. Theory 39 (1998), 113 -- 121. [31] J. P. Sproston, Derivations and automorphisms of homogeneous C ∗-algebras, Proc. Lond. Math. Soc. (3) 32 (1976), 521 -- 536. [32] J. Tomiyama and M. Takesaki, Applications of fibre bundles to the certain class of C ∗- algebras, Tohoku Math. J. (2) 13 (1961) 498 -- 522. Department of Mathematics, University of Zagreb, Bijenicka 30, 10000 Zagreb, Croa- tia E-mail address: [email protected]
0904.2104	5	0904	2012-12-07T21:07:20	Translation invariant pure state and its split property	[ "math.OA", "math-ph", "math-ph" ]	We prove Haag duality property of any translation invariant pure state on $\clb = \otimes_{\IZ}M_d(C), \;d \ge 2$, where $M_d(C)$ is the set of $d \times d$ dimensional matrices over field of complex numbers. We also prove a necessary and sufficient condition for a translation invariant factor state to be pure on $\clb$. This result makes it possible to study such a pure state with additional symmetry. We prove that exponentially decaying two point spacial correlation function of a real lattice symmetric reflection positive translation invariant pure state is a split state. Further there exists no translation invariant pure state on $\clb$ that is real, lattice symmetric, refection positive and $su(2)$ invariant when $d$ is an even integer. This in particular says that Heisenberg iso-spin anti-ferromagnets model for 1/2-odd integer spin degrees of freedom admits spontaneous symmetry breaking at it's ground states	math.OA	math	TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY ANILESH MOHARI Abstract We prove Haag duality property of any translation invariant pure state on B = ⊗ZZ Md(C), d ≥ 2, where Md(C) is the set of d × d dimensional matrices over field of complex numbers. We also prove a necessary and sufficient condition for a translation invariant factor state to be pure on B. This result makes it possible to study such a pure state with additional symmetry. We prove that exponentially decaying two point spacial correlation function of a real lattice symmetric reflection positive translation invariant pure state is a split state. Further there exists no translation invari- ant pure state on B that is real, lattice symmetric, refection positive and su(2) invariant when d is an even integer. This in particular says that Heisenberg iso-spin anti-ferromagnets model for 1/2-odd integer spin degrees of freedom admits spontaneous symmetry breaking at it's ground states 2 1 0 2 c e D 7 ] . A O h t a m [ 5 v 4 0 1 2 . 4 0 9 0 : v i X r a 1. Introduction 0 = Q0πω(BR)′′Q0 and M1 0 ⊆ (M1 Let ω be a translation invariant factor state on B = ⊗ZZMd(C) and (H, πω, Ω) be the GNS space associated with (B, ω). Let BR = ⊗n≥1Md(C) ( respectively BL = ⊗n<1Md(C) be the right (left) sided C∗-sub-algebra of B and e0 ( respectively e0 ) be the support projection of ω in πω(BR)′′ (respectively in πω(BL)′′. Thus we have two commuting projections e0, e0 defined by e0 = [πω(BR)′Ω] and e0 = [πω(BL)′Ω] as πω(BR)′′ commutes with πω(BL)′′. We set Q0 = e0e0 and take K0 to be the Hilbert subspace of H determined by the projection Q0. Also set von-Neumann 0 = Q0πω(BL)′′Q0. So by our construction algebras M1 we have M1 0)′. Further the vector state φ0(x) =< Ω, xΩ > on B(K0) is faithful and normal on M1 0. Further ω being a factor state, we will also have factor property of M1 0 ( Theorem 2.4 ). What is less obvious when [M1 0Ω] = IK0 , identity of K0. This we can expect cyclic property i.e. question is far from obvious to prove that the following statements are equivalent: (a) [M1 (b) (M1)′ (c) πω(BR)′ = πω(BL)′′; (d) [πω(BR)′Ω] = [πω(BL)′′Ω]; (e) ω is pure. 0 and M1 0 and M1 0Ω] = [ M1 0Ω] = IK0 , [ M1 0Ω] = IK0 0 = M1 0; 1991 Mathematics Subject Classification. 46L. Key words and phrases. Uniformly hyperfinite factors. Cuntz algebra, Popescu dilation, Kol- mogorov's property, Arveson's spectrum, Haag duality . ... This paper has grown over the years starting with initial work in the middle of 2005. The author gratefully acknowledge discussion with Ola Bratteli and Palle E. T. Jorgensen for inspiring participation in sharing the intricacy of the present problem. Finally the author is indebted to Taku Matsui for valuable comments on an earlier draft of the present problem where the author made an attempt to prove Haag duality. 1 2 ANILESH MOHARI Before we elaborate further on equivalence of above statements we briefly recall results on translation invariant pure state on B = ⊗ZZMd(C) that finds it's relevance while proving Haag duality (c). There is a one to one affine map between translation invariant states on B and translation invariant states on BR = ⊗ZZ+ Md(C) by ω → ωR = ωBR. The inverse map is the inductive limit state of (BR, ψR) →λn (BR, ψR) where (λn : n ≥ 0) is the canonical semi-group of right shifts on BR. Pure states on a UHF algebra are studied in the general framework of [Pow1]. Such a situation has been investigated also in details at various degrees of generality in [BJP] and [BJKW] with primary motivation to develop a C∗ algebraic method to study iterative function systems and its associated wavelet theory. One interesting result in [BJP] says that any translation invariant pure state on BR is also a product state and the canonical endomorphism associated with two such states are unitary equivalent. However such a statement is not true for two translation invariant pure states on B as their restriction to BR need not be isomorphic. Theorem 3.4 in [Mo3] says that ωR is either a type-I or a type-III factor state on BR. Both type of factors are known to exists in the literature arises in statistical mechanics [BR2,Si,Ma1]. Thus the classification problem of translation dynamics on B with invariant pure states on B up to unitary isomorphism is a delicate one. However in [Mo3] we have got partial success when such states admits Kolmogorov's property introduces in [AM]. Since a θ invariant states ω on B is completely determined by its restriction ωR to BR, in principle it is possible to describe various property of ω including pureness by studying certain properties of their restriction ωR. Since pureness of ωR is not necessary for pureness of ω, a valid fundamental question that arises here: What are the parameters that determines both ω and ωR uniquely and how these parameters determines properties of ω and ωR? Theorem 7.1 in [BJKW] has aimed towards a sufficient condition on the associ- ated minimal data Popescu elements for purity of the translation invariant state. However the proof is faulty as certain argument used in the proof is not time rever- sal symmetric. Thus Lemma 7.6 in [BJKW] needs additional assumption related to the support projection of the dual Cuntz's state. Besides this additional struc- ture proof of Lemma 7.8 in [BJKW] is also not complete unless we find a proof for M = M′ ( we retained same notations here in the text) for such a factor state ω. Such a problem could have been solved if there were any method which shows di- rectly that Takesaki's conditional expectation exists from M′ onto M. Thus main body of the proof for Theorem 7.1 in [BJKW] is incomplete. Translation invariant states on B as an inductive limit states are investigated in a series of papers [Mo1],[Mo2],[Mo3]. Though we will be working here within the framework of amalgamated representation of Od ⊗ Od introduced in [BJKW], criteria on asymptotic behavior of translation invariant pure states is used to prove first that (d) indeed implies (e). The statement (d) implies (e) has originated from section 7 of [BJKW]. Converse problem is directly related to the main result of section 3 i.e. Haag duality πω(BR)′ = πω(BL)′′ for a pure translation invariant state (Theorem 3.6) ω on B. We explore the set of representation of B quasi-equivalent to πω and equip it with a partial ordering to prove Haag duality. Mackey's system of imprimitivity plays a crucial role even though pure state not necessarily give rises a Mackey's system of imprimitivity generated by support projection E0 with respect shift. Though we have worked here with amalgamated representation of Od ⊗ Od, it seems that just for Haag duality one can avoid doing so. It seems that the underlining group ZZ can easily be replaced by ZZ k for some k ≥ 2 and wedge duality for a pointed TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 3 cone can be proved following the same ideas. We defer this line of analysis leaving it for a possible future direction of work as it's relation with problems in quantum spin chain in higher dimensional lattice needs some additional structure. Haag duality property plays an important role in studying the factor state ωR or ωL when ω admits some additional symmetry apart from translation symmetry [Mo5] to determine split property [Ma1,Ma2] of a pure translation invariant state on B and it's relation with decaying property of special correlation functions. We briefly set the standard notations and known relations in the following text. The quantum spin chain that we consider here is described by a UHF C∗-algebra denoted by B = ⊗ZZ Md(C). Here B is the C∗ -completion of the infinite tensor product of the algebra Md(C) of d by d complex matrices, each component of the tensor product element is indexed by an integer j. Let Q be a matrix in Md(C). By Q(j) we denote the element ... ⊗ 1 ⊗ 1...1 ⊗ Q ⊗ 1 ⊗ ...1⊗, , ., where Q appears in the j-th component. Given a subset Λ of Z, BΛ is defined as the C∗-sub-algebra of B generated by all Q(j) with Q ∈ Md(C), j ∈ Λ. We also set Bloc = [Λ:Λ<∞ BΛ where Λ is the cardinality of Λ. Let ω be a state on B. The restriction of ω to BΛ is denoted by ωΛ. We also set ωR = ω[1,∞) and ωL = ω(−∞,0]. The translation θk is an automorphism of B defined by θk(Q(j)) = Q(j+k). Thus θ1, θ−1 are unital ∗-endomorphism on BR and BL respectively. We say ω is translation invariant if ω ◦ θk = ω on B ( ω ◦ θ1 = ω on B ). In such a case (BR, θ1, ωR) and (BL, θ−1, ωL) are two unital ∗-endomorphisms with invariant states. Main result obtained in this paper given below. Theorem 1.1. Let ω be a translation invariant factor state on B. Then the following holds: (a) ω is pure if and only if [πω(BL)′′Ω] = [πω(BR)′Ω] where πω is the GN S repre- sentation of B associated with ω; (b) ω is pure if and only if it admits Haag duality property i.e. πω(BR)′ = πω(BL)′′ (Theorem 3.6); A general mathematical question that is central here now: when and how can we guarantee that ωR(ωL) are type-I factors or type-III factor by studying additional symmetry of the state? To that end we first recall [Ma2] a standard definition of a state to be split in the following. Definition 1.2. Let ω be a translation invariant state on B. We say that ω is split if the following condition is valid: Given any ǫ > 0 there exists a k ≥ 1 so that (1) supQ<1ω(Q) − ωL ⊗ ωR(Q) ≤ ǫ where the above supremum is taken over all local elements Q ∈ B(−∞,−k]∪[k,∞) with the norm less than 1. Here we recall few well known facts from [Pow1,BR,Ma1,Ma2]. The uniform cluster condition is valid if and only if the state ω is quasi-equivalent to the product state ψL ⊗ ψR of a state ψL of BL and another state ψR of BR. Thus a Gibbs state of a finite range interaction is split. If ω is a pure translation invariant state, then ωR(ωL) is type-I if and only if ω is also a split state. The canonical trace is a non- pure split state and unique ground state of XY model [AMa,Ma2] is a non-split 4 ANILESH MOHARI pure state. One central aim is to find a criteria for a pure translation invariant state to be split. To that end we present a precise definition for exponential decay. Definition 1.3. Let ω be a translation invariant state on one dimensional spin chain B. We say the two point spacial correlation functions for ω decay exponentially if there exists a δ > 0 so that (2) eδkω(Q1θk(Q2)) − ω(Q1)ω(Q2) → 0 as k → ∞ for any local elements Q1, Q2 ∈ B. For any translation invariant state ω on B we set translation invariant state ω by reflecting around the point 1 2 on B by ω(Q(−l) −l ⊗ Q(−l+1) −l+1 ⊗ ... ⊗ Q(−1) −1 ⊗ Q(0) 0 ⊗ Q(1) 1 ... ⊗ Q(n) n ) (3) = ω(Q(−n+1) n ... ⊗ Q(0) 1 ⊗ Q(1) 0 ⊗ Q(2) −1 ⊗ ...Q(l) −l+1 ⊗ Q(l+1) −l ) for all n, l ≥ 1 and Q−l, ..Q−1, Q0, Q1, .., Qn ∈ Mn(IC) where Q(k) is the matrix Q at lattice point k. We define ω on B by extending linearly to any Q ∈ Bloc. Thus ω → ω is an affine one to one onto map on the set of translation invariant states on B. Thus the state ω is translation invariant, ergodic, factor state if and only if ω is translation invariant, ergodic, factor state respectively. We say ω is lattice reflection symmetric if ω = ω. 1 1 m If Q = Q(l) 0 ⊗Q(l+1) ⊗....⊗Q(l+m) where Q0, Q1, ..., Qm are arbitrary elements in Md and Qt 1, .. stands for transpose with respect to an orthonormal basis (ei) for ICd (not complex conjugate) of Q0, Q1, .. respectively. We define Qt by extending linearly for any Q ∈ Bloc. For a state ω on UHFd C∗ algebra ⊗ZZMd we define a state ¯ω on ⊗ZZ Md by the following prescription we set Qt = Qt(l) 0 ⊗Qt(l+1) 0, Qt ⊗..⊗Qt(l+m) m (4) ¯ω(Q) = ω(Qt) Thus the state ¯ω is translation invariant, ergodic, factor state if and only if ω is translation invariant, ergodic, factor state respectively. We say ω is real if ¯ω = ω. A translation invariant state ω is said to be in detailed balance if ω is lattice reflection symmetric and real (for further details see section 3 ). The canonical trace on B is both real and lattice symmetric. This notion of detailed balance state is introduced as an reminiscence of Onsager's reciprocal relations explored in recent articles [AM,Mo1,Mo2,Mo3] on non-commutative probability theory. Here we also recall well known notion [DLS] that a state ω on B is called reflection positive if (5) ω(J (x)x) ≥ 0 for all x ∈ BR where J is the reflection map with a twist g0 ∈ Ud(C) from BR onto BL and x stands for complex conjugation with respect to a basis (ei) for Cd applies globally on each matrices of the lattice simultaneously. Let G be a compact group and g → v(g) be a d−dimensional unitary representa- tion of G. By γg we denote the product action of G on the infinite tensor product B induced by v(g), (6) γg(Q) = (.. ⊗ v(g) ⊗ v(g) ⊗ v(g)...)Q(... ⊗ v(g)∗ ⊗ v(g)∗ ⊗ v(g)∗...) for any Q ∈ B. We say ω is G-invariant if ω(γg(Q)) = ω(Q) for all Q ∈ Bloc. TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 5 Our main results on symmetry in section 4 uses Haag duality property crucially to study lattice reflection symmetry property of a translation invariant pure state ω. Here we make a short list of end results obtained in this paper. Theorem 1.4. Let ω be a pure lattice symmetric translation invariant real ( with respect to a basis (ei) of ICd ) state on B. Then the following holds: (a) If ω is also reflection positive with a twist g0 and two point spacial correlation function for ω decays exponentially then ω is a split state i.e. πω(BR)′′ is a type-I factor (Theorem 5.4). (b) If ω is also reflection positive and G-invariant then g → u(g) is a real repre- sentation with respect to the orthonormal basis (ei) for ICd provided the invariant subspace of the representation g → u(g) ⊗ u(g) is one dimensional. Pk∈ZZ θk(h0) with h0 = h∗ Theorem 1.5. Let H be a translation invariant Hamiltonian of the form H = 0 ∈ Bloc and H be real, lattice reflection symmetric and reflection positive with a twist g0 ∈ Ud(IC). Let H be also SU (2) invariant where g → u(g) is an irreducible representation of SU (2). Then ground state for H is not unique when d = 2s + 1 and s is a half-odd integer. We end this paper with a brief application of our results to antiffero-magnet Heisenberg models which includes prime examples such as HXY and HXXX models. The paper is organized as follows. In section 2 we study Popescu's dilationas- sociated with a translation invariant state on Cuntz algebra Od and review 'com- mutant lifting theorem' investigated in [BJKW]. The proof presented here remove the murky part of the proof of Theorem 5.1 in [BJKW]. In section 3 we explore both the notion of Kolmogorov's shift and it's intimate relation with Mackey's im- primitivity system to explore a duality argument introduced in [BJKW]. We find a useful necessary and sufficient condition (Theorem 1.1 (a) ) in terms of support projection of Cuntz's state for a translation invariant factor state ω on B to be pure. The criteria on support projection is crucial to prove our main mathematical result Theorem 3.6. Section 4 studies discrete symmetry and section 5 gives the proof of the statement (a) of Theorem 1.4. Section 6 studies continuous symmetry and gives proof of the statement (b) of Theorem 1.4 and also proof of Theorem 1.5. Remark 1.6. The paper tittled "On Haag Duality for Pure States of Quantum Spin Chain" by authors: M. Keyl, Taku Matsui, D. Schlingemann, R. F. Werner, Rev. Math. Phys. 20:707-724,2008 has an incomplete proof for Haag duality property as Lemma 4.3 in that paper has a faulty argument. 2. States on Od and the commutant lifting theorem In this section we essentially recall results from [BJKW] and organize it with ad- ditional remarks and arguments as it demands to understand the present problem investigated in section 3. First we recall that the Cuntz algebra Od(d ∈ {2, 3, .., }) is the universal C∗-algebra generated by the elements {s1, s2, ..., sd} subject to the relations: s∗ i sj = δi j1 sis∗ i = 1. X1≤i≤d 6 ANILESH MOHARI There is a canonical action of the group U (d) of unitary d × d matrices on Od given by βg(si) = X1≤j≤d gj i sj for g = ((gi j) ∈ U (d). In particular the gauge action is defined by βz(si) = zsi, z ∈ IT = S1 = {z ∈ C : z = 1}. If UHFd is the fixed point sub-algebra under the gauge action, then UHFd is the closure of the linear span of all wick ordered monomials of the form si1 ...sik s∗ jk ...s∗ j1 which is also isomorphic to the UHFd algebra Md∞ = ⊗∞ 1 Md so that the isomorphism carries the wick ordered monomial above into the matrix element and the restriction of βg to U HFd is then carried into action ei1 j1(1) ⊗ ei2 j2 (2) ⊗ .... ⊗ eik jk (k) ⊗ 1 ⊗ 1.... Ad(g) ⊗ Ad(g) ⊗ Ad(g) ⊗ .... We also define the canonical endomorphism λ on Od by λ(x) = X1≤i≤d sixs∗ i and the isomorphism carries λ restricted to UHFd into the one-sided shift on ⊗∞ 1 Md. Note that λβg = βgλ on UHFd. y1 ⊗ y2 ⊗ ... → 1 ⊗ y1 ⊗ y2.... Let d ∈ {2, 3, .., , ..} and ZZd be a set of d elements. I be the set of finite sequences I = (i1, i2, ..., im) where ik ∈ ZZd and m ≥ 1. We also include empty set ∅ ∈ I and set s∅ = 1 = s∗ i1 ∈ Od. In the following we recall a commutant lifting theorem ( Theorem 5.1 in [BJKW] ), crucial for our purpose. ∅, sI = si1 ......sim ∈ Od and s∗ I = s∗ im ...s∗ space K so that P1≤k≤d vkv∗ Theorem 2.1. Let v1, v2, ..., vd be a family of bounded operators on a Hilbert k = I. Then there exists a unique up to isomorphism Hilbert space H, a projection P on K and a family of isometries {Sk :, 1 ≤ k ≤ d, P } satisfying Cuntz's relation so that (7) P S∗ kP = S∗ kP = v∗ k for all 1 ≤ k ≤ d and K is cyclic for the representation i.e. the vectors {SIK : I < ∞} are total in H. Moreover the following holds: (a) Λn(P ) ↑ I as n ↑ ∞; (b) For any D ∈ Bτ (K), Λn(D) → X ′ weakly as n → ∞ for some X ′ in the commutant {Sk, S∗ k : 1 ≤ k ≤ d}′ so that P X ′P = D. Moreover the self adjoint elements in the commutant {Sk, S∗ k : 1 ≤ k ≤ d}′ is isometrically order isomorphic with the self adjoint elements in Bτ (K) via the surjective map X ′ → P X ′P , where (c) {vk, v∗ k ≤ d}′′. Bτ (K) = {x ∈ B(K) : P1≤k≤d vkxv∗ k = x}. k, 1 ≤ k ≤ d}′ ⊆ Bτ (K) and equality holds if and only if P ∈ {Sk, Sk, 1 ≤ TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 7 exists an operator u : K → K′ so that Pk wkuv∗ If (wi) be another such an Popescu elements on a Hilbert space K′ such that there k = u then there exists an operator U : Hv → Hw so that π′(x)U = U π(x) where (Hw, π′, S′ i) are Popescu dilation of (wi) and π′ is the associated minimal representation of Od. In particular U is isometry, unitary if u is so respectively. If u is unitary and K = K′ then we can as well take Hv = Hw. PROOF: Following Popescu [Po] we define a completely positive map R : Od → B(K) by (8) R(sI s∗ J ) = vI v∗ J for all I, J < ∞. The representation S1, .., Sd of Od on H thus may be taken to be the Stinespring dilation of R [BR, vol-2] and uniqueness up to unitary equivalence follows from uniqueness of the Stinespring representation. That K is cyclic for the representation follows from the minimality property of the Stinespring dilation. For (a) let Q be the limiting projection. Then we have Λ(Q) = Q, hence Q ∈ {Sk, S∗ k}′ and Q ≥ P . In particular QSI f = SI f for all f ∈ K and I < ∞. Hence Q = I by the cyclicity of K. For (b) essentially we deffer from the argument used in Theorem 5.1 in [BJKW]. We fix any D ∈ Bτ (K) and note that P Λk(D)P = τk(D) = D for any k ≥ 1. Thus for any integers n > m we have Λm(P )Λn(D)Λm(P ) = Λm(P Λn−m(D)P ) = Λm(D) Hence for any fix m ≥ 1 limit < f, Λn(D)g > as n → ∞ exists for all f, g ∈ Λm(P ). Since the family of operators Λn(D) is uniformly bounded and Λm(P ) ↑ I as m → ∞, a standard density argument guarantees that the weak operator limit of Λn(D) exists as n → ∞. Let X ′ be the limit. So Λ(X ′) = X ′, by Cuntz's relation, X ′ ∈ {Sk, S∗ k : 1 ≤ k ≤ k}′. Since P Λn(D)P = D for all n ≥ 1, we also conclude that P X ′P = D by taking limit n → ∞. Conversely it is obvious that P {Sk, S∗ k : k ≥ 1}′P with Bτ (K). k : k ≥ 1}′P ⊆ Bτ (K). Hence we can identify P {Sk, S∗ Further it is obvious that X ′ is self-adjoint if and only if D = P X ′P is self- adjoint. Now fix any self-adjoint element D ∈ Bτ (K). Since identity operator on K is an element in Bτ (K) for any α ≥ 0 for which −αP ≤ D ≤ αP , we have αΛn(P ) ≤ Λn(D) ≤ αΛn(P ) for all n ≥ 1. By taking limit n → ∞ we conclude that −αI ≤ X ′ ≤ αI, where P X ′P = D. Since operator norm of a self-adjoint element A in a Hilbert space is given by A = infα≥0{α : −αI ≤ A ≤ αI} we conclude that X ′ ≤ D. That D = P X ′P ≤ X ′ is obvious, P being a projection. Thus the map is isometrically order isomorphic taking self-adjoint elements of the commutant to self-adjoint elements of Bτ (K). kX ′P = P S∗ kP X ′P = v∗ We are left to prove (c). Inclusion is trivial. For the last part note that for any k, 1 ≤ k ≤ d}′ so invariant element D in B(K) there exists an element X ′ in {Sk, S∗ kP = P X ′S∗ that P X ′P = D. In such a case we verify that Dv∗ k = P X ′P S∗ kP = k = v∗ kD∗. P S∗ k : 1 ≤ k ≤ d}′. Since P πω(Od)′P = B(K)τ , we conclude Hence D ∈ {vk, v∗ that B(K)τ ⊆ M′. Thus equality holds whenever P ∈ {Sk, S∗ k, 1 ≤ k ≤ d}′′. For converse note that by commutant lifting property self-adjoint elements of the commutant {Sk, S∗ k, 1 ≤ k ≤ d}′ is order isometric with the algebra M′ via the k, 1 ≤ k ≤ d}′′ by Proposition 4.2 in [BJKW]. map X ′ → P X ′P . Hence P ∈ {Sk, S∗ kD. We also have D∗ ∈ Bτ (K) and thus D∗v∗ 8 ANILESH MOHARI For the proof of intertwining relation and their property we refer to main body of the proof of Theorem 5.1 in [BJKW] A family (vk, 1 ≤ k ≤ d) of contractive operators on a Hilbert space K is called Popescu's elements and dilation (H, P, K, Sk, 1 ≤ k ≤ d) in Theorem 2.1 is called Popescu's dilation to Cuntz elements. In the following proposition we deal with a family of minimal Popescu elements for a state on Od. Proposition 2.2. There exists a canonical one-one correspondence between the following objects: (a) States ψ on Od (b) Function C : I × I → C with the following properties: (i) C(∅, ∅) = 1; (ii) for any function λ : I → C with finite support we have λ(I)C(I, J)λ(J) ≥ 0 XI,J∈I C(Ii, Ji) = C(I, J) for all I, J ∈ I. (iii) Pi∈ZZd (c) Unitary equivalence class of objects (K, Ω, v1, .., vd) where (i) K is a Hilbert space and Ω is an unit vector in K; (ii) v1, .., vd ∈ B(K) so that Pi∈ZZd (iii) the linear span of the vectors of the form v∗ i = 1; viv∗ I Ω, where I ∈ I, is dense in K. Where the correspondence is given by a unique completely positive map R : Od → J ) = vI v∗ J ; B(K) so that (i) R(sI s∗ (ii) ψ(x) =< Ω, R(x)Ω >; (iii) ψ(sI s∗ (iv) For any fix g ∈ Ud and the completely positive map Rg : Od → B(K) defined by Rg = R ◦ βg give rises to a Popescu system given by (K, Ω, βg(vi), .., βg(vd)) where J ) = C(I, J) =< v∗ I Ω, v∗ J Ω >; βg(vi) = P1≤j≤d gi jvj. PROOF: For a proof we simply refer to Proposition 2.1 in [BJKW]. The following is a simple consequence of Theorem 2.1 valid for a λ-invariant state ψ on Od. This proposition will have very little application in main body of this paper but this gives a clear picture explaining the delicacy of the present problems. Proposition 2.3. Let ψ be a state on Od and (H, π, Ω) be the GNS space as- sociated with (Od, ψ). We set Si = π(si) and normal state ψΩ on π(Od)′′ defined by Let P be the projection on the closed subspace K generated by the vectors {S∗ I < ∞} and I Ω : ψΩ(X) =< Ω, XΩ > (9) vk = P SkP for 1 ≤ k ≤ d. Then following holds: (a) {v∗ I Ω : I < ∞} is total in K. (b) P1≤k≤d vkv∗ kP = P S∗ (c) S∗ k = I; kP for all 1 ≤ k ≤ d; TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 9 (d) For any I = (i1, i2, ..., ik), J = (j1, j2, ..., jl) with I, J < ∞ we have (10) ψ(sI s∗ J ) =< Ω, vI v∗ J Ω > and the vectors {SIf : f ∈ K, I < ∞} are total in the GNS Hilbert space associated with (Od, ψ). Further such a family (K, vk, 1 ≤ k ≤ d, ω) satisfying (a) to (d) are determined uniquely up to isomorphism. Conversely given a Popescu system (K, vk, 1 ≤ k ≤ d, Ω) satisfying (a) and (b) there exists a unique state ψ on Od so that (c) and (d) are satisfied. Furthermore the following statements are valid: (e) If the normal state φ0(x) =< Ω, xΩ > on the von-Neumann algebra M = {vi, v∗ i , x ∈ M then ψ is λ invariant and φ0 is faithful on M. i }′′ is invariant for the Markov map τ (x) = P1≤k≤d vixv∗ (f ) If P ∈ π(O)′′ then following are equivalent: (i) ψ is an ergodic state for (Od, λ); (ii) (M, τ, φ0) is ergodic. In such a case M is a factor. PROOF: We fix a state ψ and consider the GNS space (H, π, Ω) associated with (Od, ψ) and set Si = π(si). It is obvious that S∗ kP ⊆ P for all 1 ≤ k ≤ d, thus P is the minimal subspace containing Ω and invariant by all {S∗ k; 1 ≤ k ≤ d} i.e. (11) Thus v∗ operator in K. This completes the proof of (a) (b) and (c). kP and so Pk vkv∗ k = Pk P SkS∗ kP k = P S∗ kP = S∗ P S∗ kP = S∗ kP = P which is identity For (d) we note that ψ(sI s∗ =< Ω, P SI S∗ J ) =< Ω, SIS∗ J Ω > J P Ω >=< Ω, vI v∗ J Ω > . Since H is spanned by the vectors {SIS∗ J Ω : I, J < ∞} and K is spanned by the vectors {S∗ J Ω : I < ∞}, K is cyclic for SI i.e. the vectors {SIK : I < ∞} spans H. Uniqueness up to isomorphism follows as usual by total property of vectors v∗ I Ω in K. J Ω = v∗ Conversely for a Popescu's elements (K, vi, Ω) satisfying (a) and (b), we consider the family (H, Sk, 1 ≤ k ≤ d, P ) of Cuntz's elements defined as in Theorem 2.1. We claim that Ω is a cyclic vector for the representation π(si) → Si. Note that by our construction vectors {SI f, f ∈ K : I < ∞} are total in H and v∗ J Ω for all J < ∞. Thus by our hypothesis that vectors {v∗ J Ω : I < ∞} are total in K, we verify that vectors {SIS∗ J Ω : I, J < ∞} are total in H. Hence Ω is a cyclic for the representation si → Si of Od. J Ω = S∗ We left to prove (e) and (f). It simple to note by (d) that ψλ = ψ i.e. < Ω, SiSI S∗ Xi =< Ω, vI v∗ < Ω, vivI v∗ i Ω >= Xi J S∗ J Ω >=< Ω, SI S∗ J Ω > J v∗ i Ω > for all I, J < ∞ where in the second equality we have used our hypothesis that the vector state φ0 on M is τ -invariant. In such case we aim now to show that φ0 is faithful on M. To that end let p′ be the support projection in M for τ invariant state φ0. Thus φ0(1 − p′) = 0 i.e. p′Ω = Ω and by invariance we also 10 ANILESH MOHARI have φ0(p′τ (1 − p′)p′) = φ0(1 − p′) = 0. Since p′τ (1 − p′)p′ ≥ 0 and an element in M, by minimality of support projection, we conclude that p′τ (1 − p′)p′ = 0. Hence p′Ω = Ω and p′v∗ I Ω for all I < ∞. As K is the closed linear span of the vectors {v∗ I Ω : I < ∞}, we conclude that p′ = p. In other words φ0 is faithful on M. This completes the proof for (e). kp′ for all 1 ≤ k ≤ d. Thus p′v∗ kp′ = v∗ I Ω = v∗ We are left to show (f) where we assume that P ∈ π(Od)′′. Ω being a cyclic vector for π(Od)′′, the weak∗ limit of the increasing projection Λk(P ) is I. Thus by Theorem 3.6 in [Mo1] we have (π(Od)′′, Λ, ψΩ) is ergodic if and only if the reduced dynamics (M, τ, φ0) is ergodic. Last part of the statement is an easy consequence of a theorem in [Fr] (see also [BJKW],[Mo1]). Before we move to next result we comment here that in general for a λ invariant k : 1 ≤ k ≤ d}′′ need not be invariant state on Od the normal state φ0 on M = {vk, v∗ for τ . To that end we consider ( [BR] vol-II page 110 ) the unique KMS state ψ = ψβ for the automorphism αt(si) = eitsi on Od. ψ is λ invariant and ψUHFd is the unique faithful trace. ψ being a KMS state for an automorphism, the normal state induced by the cyclic vector on πψ(Od)′′ is also separating for π(Od)′′. As ψβz = ψ for all z ∈ S1 we have < Ω, π(sI )Ω >=< Ω, βz(sI )Ω >= zI < Ω, π(sI )Ω > for all z ∈ S1 and so < Ω, π(sI )Ω >= 0 for all I ≥ 1. In particular < Ω, v∗ I Ω >= 0 where (vi) are defined as Proposition 2.3 and thus < viΩ, v∗ i vI Ω >= 0 for all 1 ≤ i ≤ d. Hence viΩ = 0. By Proposition 2.3 (e), Ω is separating for M i = 1. Thus we conclude by Proposition 2.3 (e) that φ0 is not τ invariant on M. This example also indicates that the support projection of a λ invariant state ψ in π(Od)′′ need not be equal to the minimal sub-harmonic projection P i.e. the closed span of vectors {S∗ J : I, J < ∞} need not be even an algebra. I Ω >=< Ω, v∗ and so we get vi = 0 for all 1 ≤ i ≤ d and this contradicts that Pi viv∗ I Ω : I < ∞} containing Ω and {vI v∗ Now we aim to deal with another class of Popescu elements associated with an λ-invariant state on Od. In fact this class of Popescu elements will play a significant role for the rest of the text and we will repeatedly use this proposition! Proposition 2.4. Let (H, π, Ω) be the GNS representation of a λ invariant state ψ on Od and P be the support projection of the normal state ψΩ(X) =< Ω, XΩ > in the von-Neumann algebra π(Od)′′. Then the following holds: (a) P is a sub-harmonic projection for the endomorphism Λ(X) = Pk SkXS∗ k on π(Od)′′ i.e. Λ(P ) ≥ P satisfying the following: (i) Λn(P ) ↑ I as n ↑ ∞; (ii) P S∗ kP, 1 ≤ k ≤ d; kP = S∗ (iii) P1≤k≤d vkv∗ k = I where Sk = π(sk) and vk = P SkP for 1 ≤ k ≤ d; (b) For any I = (i1, i2, ..., ik), J = (j1, j2, ..., jl) with I, J < ∞ we have ψ(sI s∗ Ω, vI v∗ J Ω > and the vectors {SI f : f ∈ K, I < ∞} are total in H; J ) =< range of P , is generated by {vk, v∗ (c) The von-Neumann algebra M = P π(Od)′′P , acting on the Hilbert space K k : 1 ≤ k ≤ d}′′ and the normal state i.e. φ0(x) =< Ω, xΩ > is faithful on the von-Neumann algebra M. (d) The self-adjoint part of the commutant of π(Od)′ is norm and order isomor- phic to the space of self-adjoint fixed points of the completely positive map τ . The TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 11 isomorphism takes X ′ ∈ π(Od)′ onto P X ′P ∈ Bτ (K), where Bτ (K) = {x ∈ B(K) : k = x}. Furthermore M′ = Bτ (K). Pk vkxv∗ k ≤ d} of bounded operators on a Hilbert space K so that Pk vkv∗ the commutant M′ = {x ∈ B(K) : Pk vkxv∗ Conversely let M be a von-Neumann algebra generated by a family {vk : 1 ≤ k = 1 and k = x}. Then the Popescu dilation (H, P, Sk, 1 ≤ k ≤ d) described in Theorem 2.1 satisfies the following: (i) P ∈ {Sk, S∗ (ii) For any faithful normal invariant state φ0 on M there exists a state ψ on Od defined by k, 1 ≤ k ≤ d}′′; ψ(sI s∗ J ) = φ0(vI v∗ J ), I, J < ∞ so that the GNS space associated with (M, φ0) is the support projection for ψ in π(Od)′′ satisfying (a)-(d). Further for a given λ-invariant state ψ, the family (K, M, vk 1 ≤ k ≤ d, φ0) satisfying (a)-(d) is determined uniquely up to unitary conjugation. (e) φ0 is a faithful normal τ -invariant state on M. Furthermore the following statements are equivalent: (i) (Od, λ, ψ) is ergodic; (ii) (M, τ, φ0) is ergodic; (iii) M is a factor. PROOF: Λ(P ) is also a projection in πψ(Od)′′ so that ψΩ(Λ(P )) = 1 by invariance property. Thus we have Λ(P ) ≥ P i.e. P Λ(I − P )P = 0. Hence we have (12) P S∗ kP = S∗ kP Moreover by λ invariance property we also note that the faithful normal state φ0(x) =< Ω, xΩ > on the von-Neumann algebra M = P πψ(Od)′′P is invariant for the reduce Markov map [Mo1] on M given by (13) τ (x) = P Λ(P xP )P We claim that limn↑∞Λn(P ) = I. That {Λn(P ) : n ≥ 1} is a sequence of increasing projections follows from sub-harmonic property of P and endomorphism property of Λ. Let the limiting projection be Y . Then Λ(Y ) = Y and so Y ∈ {Sk, S∗ J Ω, Y is a scaler, being a non-zero projection, it is the identity operator in Hπψ . k}′. Since by our construction GNS Hilbert space Hπ ω is generated by SI S∗ Now it is routine to verify (a) (b) and (c). For the first part of (d) we appeal to Theorem 2.2. For the last part note that for any invariant element D in B(K) there exists an element X ′ in π(Od)′ so that P X ′P = D. Since P ∈ π(Od)′′ we note that (1 − P )X ′P = 0. Now since X ′ ∈ {Sk, S∗ kP = P XS∗ k = v∗ k : 1 ≤ k ≤ d}′ = M′. Since P πω(Od)′P = B(K)τ , we kD∗. Thus D ∈ {vk, v∗ conclude that B(K)τ ⊆ M′. The reverse inclusion is trivial. This completes the proof for (d). kD. Since D∗ ∈ Bτ (K) we also have D∗v∗ k}′, we verify that Dv∗ k = P XP S∗ kXP = P S∗ kP XP = v∗ kP = P S∗ For the converse part of (i), since by our assumption and commutant lifting k, 1 ≤ k ≤ d}′ is order property self-adjoint elements of the commutant {Sk, S∗ isometric with the algebra M′ via the map X ′ → P X ′P , P ∈ {Sk, S∗ k, 1 ≤ k ≤ d}′′ by Proposition 4.2 in [BJKW]. For (ii) without loss of generality assume that φ0(x) =< Ω, xΩ > for all x ∈ M and Ω is a cyclic and separating vector for M. 12 ANILESH MOHARI J ) = φ0(vI v∗ ( otherwise we set state ψ(sI s∗ J ) and consider it's GNS representation ) We are left to show that Ω is a cyclic vector for the representation π(si) → Si. To that end let Y ∈ π(Od)′ be the projection on the subspace generated by the vectors {SI S∗ J Ω : I, J < ∞}. Note that P being an element in π(Od)′′, Y also commutes with all the element P π(Od)′′P = P MP . Hence Y xΩ = xΩ for all x ∈ M. Thus Y ≥ P . Since Λn(P ) ↑ I as n ↑ ∞ by our construction, we conclude that Y = Λn(Y ) ≥ Λn(P ) ↑ I as n ↑ ∞. Hence Y = I. In other words Ω is cyclic for the representation si → Si. This completes the proof for (ii). Uniqueness up to unitary isomorphism follows as GNS representation is deter- mined uniquely unto unitary conjugation and so its support projection. The first part of (e) we note that P SIS∗ J P = vI v∗ M = P π(Od)′′P is the von-Neumann algebra generated by {vk, v∗ J for all I, J < ∞ and thus k : 1 ≤ k ≤ d} k invariant follows as ψ is λ-invariant. We are left to prove equivalence of statements (i)-(iii). and thus τ (x) = P Λ(P xP )P for all x ∈ M. That φ0 is τ (x) = Pk vkxv∗ By Theorem 3.6 in [Mo1] Markov semi-group (M, τ, φ0) is ergodic if and only if (π(Od)′′, Λ, ψΩ) is ergodic ( here we need to recall by (a) that Λn(P ) ↑ I as n ↑ ∞ ). By a standard result [Fr, also BJKW] (M, τ, φ0) is ergodic if and only if there is no non trivial projection e invariant for τ i.e. I τ = {e ∈ M : e∗ = e, e2 = e, τ (e) = e} = {0, 1}. If τ (e) = e for some projection e ∈ M then (1 − e)τ (e)(1 − e) = 0 and so ev∗ k(1 − e) = 0. Same is true if we replace e by 1 − e as τ (1) = 1 and ke = 0. Thus e commutes with vk, v∗ τ (1 − e) = 1 − τ (e) = 1 − e and thus (1 − e)v∗ k Inequality in the reverse direction is trivial and thus I τ is trivial if and only if M is a factor. Thus equivalence of (ii) and (iii) follows by a standard result [Fr] in non-commutative ergodic theory. This completes the proof. for all 1 ≤ k ≤ d. Hence I τ ⊆ MT M′. The following two propositions are essentially easy adaptations of results ap- peared in [BJKW, Section 6 and Section 7], crucial in our present framework. Proposition 2.5. Let ψ be a λ invariant factor state on Od and (H, π, Ω) be it's GNS representation. Then the following holds: (a) The closed subgroup H = {z ∈ S1 : ψβz = ψ} is equal to {z ∈ S1 : βzextends to an automorphism of π(Od)′′} (b) Let OH Then π(OH (c) If H is a finite cyclic group of k many elements and π(UHFd)′′ is a factor, then d be the fixed point sub-algebra in Od under the gauge group {βz : z ∈ H}. d )′′ = π(UHFd)′′. π(Od)′′T π(UHFd)′ ≡ Cm where 1 ≤ m ≤ k. It is simple that H is a closed subgroup. For any fix z ∈ H we define PROOF: unitary operator Uz extending the map π(x)Ω → π(βz(x))Ω and check that the z extends βz to an automorphism of π(Od)′′. For the converse map X → UzXU ∗ we will use the hypothesis that ψ is a λ-invariant factor state and βzλ = λβz to guarantee that ψβz(X) = 1 n P1≤k≤n ψβzλk(X) → ψ(X) as n → ∞ for any X ∈ π(Od)′′, where we have used the same symbol βz for the extension. Hence z ∈ H. n P1≤k≤n ψλkβz(X) = 1 For any z1, z2 ∈ S1 we extend both ψβz1 and ψβz2 to its inductive limit state on O∗ d using the canonical endomorphism Od →λ Od. Inductive limit state being an affine map, their inductive limit states are also factors. The inductive limit of TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 13 the canonical endomorphism became an automorphism. (B, θ) is asymptotically abelian i.e. xθn(y) − θn(y)x → 0 as n → ∞ for all x, y ∈ B (see also page 240 in [BR, vol2]). Thus in particular (B, ZZ, ω) is ZZ−central for any translation invariant ergodic state ω (see page 380 in [BR vol-2]). Thus we may appeal to a general result in C∗-non-commutative ergodic theory to conclude that their inductive limit, being translation invariant factor states, are either same or orthogonal (Theorem 4.3.19 in [BR vol2]). In the following instead of working with Od we should be working with the inductive limit C∗ algebra and their inductive limit states. For simplicity of notation we still use U HFd, Od for its inductive limit of Od →λ Od and U HFd →λ U HFd respectively and so for its inductive limit states. Now we aim to prove (b). H being a closed subgroup of S1, it is either entire S1 or a finite subgroup {exp( 2iπl k )l = 0, 1, ..., k − 1} where the integer k ≥ 1. If H = S1 we have nothing to prove for (b). When H is a finite closed subgroup, we identify [0, 1) with S1 by the usual map and note that if βt is restricted to t ∈ [0, 1 k ), then by scaling we check that βt defines a representation of S1 in automorphisms of OH d . Now we consider the direct integral representation π′ defined by π′ = Z ⊕ [0, 1 k ) k ) ), where HOH d dtπOH d βt d ⊗ L2([0, 1 d on HOH is the cyclic space of π(OH of OH d ) generated by Ω. That it is indeed direct integral follows as states ψβt1 and ψβt2 are either same or orthogonal for a factor state ψ (see the above paragraph). Interesting point here to note that the new representation π′ is (βt) co-variant i.e. π′βt = βtπ′, hence by simplicity of the C∗ algebra Od we conclude that π′(UHFd)′′ = π′(OH d )′′βt By exploring the hypothesis that ψ is a factor state, we also have as in Lemma 6.11 in [BJKW] I ⊗ L∞([0, 1 k ) ) ⊂ π′(OH d )′′. Hence we also have π′(OH d )′′ = π(OH d )′′ ⊗ L∞([0, 1 k ) ). Since βt is acting as translation on I ⊗ L∞([0, 1 we have k ) ) which being an ergodic action, Since π′(UHFd)′′ = π(UHFd)′′ ⊗ 1, we conclude that π(UHFd)′′ = π(OH d )′′. π′(UHFd)′′ = π(OH d )′′ ⊗ 1 A proof for the statement (c) follows from Lemma 7.12 in [BJKW]. The orig- inal idea of the proof can be traced back to Arveson's work on spectrum of an automorphism of a commutative compact group [Ar1]. Let ω′ be an λ-invariant state on the UHFd sub-algebra of Od. Following [BJKW, section 7], we consider the set Kω′ = {ψ : ψ is a state on Od such that ψλ = ψ and ψUHFd = ω′} By taking invariant mean on an extension of ω′ to Od, we verify that Kω′ is non empty and Kω′ is clearly convex and compact in the weak topology. In case ω′ is an ergodic state ( extremal state ) Kω′ is a face in the λ invariant states. Before we proceed to the next section here we recall Lemma 7.4 of [BJKW] in the following proposition. 14 ANILESH MOHARI Proposition 2.6. Let ω′ be ergodic. Then ψ ∈ Kω′ is an extremal point in Kω′ if and only if ψ is a factor state and moreover any other extremal point in Kω′ have the form ψβz for some z ∈ S1. PROOF: Though Proposition 7.4 in [BJKW] appeared in a different set up, same proof goes through for the present case. We omit the details and refer to the original work for a proof. 3. Dual Popescu system and pure translation invariant states: In this section we review the amalgamated Hilbert space developed in [BJKW] and prove a powerful criteria for a translation invariant factor state to be pure. {vk, 1 ≤ k ≤ d} be a family of bounded operators on K so that M = {vk, v∗ To that end let M be a von-Neumann algebra acting on a Hilbert space K and k, 1 ≤ k = 1. Furthermore let Ω be a cyclic and separating vector for M so that the normal state φ0(x) =< Ω, xΩ > on M is invariant for the Markov k for x ∈ M. Let ω be the translation k ≤ d}′′ and Pk vkv∗ map τ on M defined by τ (x) = Pk vkxv∗ invariant state on UHFd = ⊗ZZ Md defined by j2(l + 1) ⊗ .... ⊗ ein jn (l + n − 1)) = φ0(vI v∗ J ) j1(l) ⊗ ei2 ω(ei1 where ei j(l) is the elementary matrix at lattice sight l ∈ ZZ. We set vk = J σ i k)J ∈ M′ ( see [BJKW] for details ) where J and σ = (σt, t ∈ IR) are Tomita's conjugation operator and modular automorphisms asso- ciated with φ0. (v∗ 2 By KMS or modular relation [BR vol-1] we verify that vk v∗ k = 1 Xk and φ0(vI v∗ J ) = φ0(v I v∗ J ) (14) where I = (in, .., i2, i1) if I = (i1, i2, ..., in). Moreover v∗ (v I )∗J Ω = Ω. We also set M to be the von-Neumann algebra generated by 2 v I Ω = v∗ J ∆ I {vk : 1 ≤ k ≤ d}. Thus M ⊆ M′. A major problem that we will have to address when equality holds. I Ω = J σ i 1 2 Let (H, P, Sk, 1 ≤ k ≤ d) and ( H, P, Sk, 1 ≤ k ≤ d) be the Popescu dilation described as in Theorem 2.1 associated with (K, vk, 1 ≤ k ≤ d) and K, vk, 1 ≤ k ≤ d) respectively. Following [BJKW] we consider the amalgamated tensor product H ⊗K H of H with H over the joint subspace K. It is the completion of the quotient of the set C ¯I ⊗CI ⊗ K, where ¯I, I both consist of all finite sequences with elements in {1, 2, .., d}, by the equivalence relation defined by a semi-inner product defined on the set by requiring < ¯I ⊗ I ⊗ f, ¯I ¯J ⊗ IJ ⊗ g >=< f, v ¯J vJ g >, < ¯I ¯J ⊗ I ⊗ f, ¯I ⊗ IJ ⊗ g >=< v ¯J f, vJ g > and all inner product that are not of these form are zero. We also define two com- muting representations (Si) and ( Si) of Od on H⊗K H by the following prescription: SI λ( ¯J ⊗ J ⊗ f ) = λ( ¯J ⊗ IJ ⊗ f ), TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 15 S ¯I λ( ¯J ⊗ J ⊗ f ) = λ( ¯J ¯I ⊗ J ⊗ f ), where λ is the quotient map from the index set to the Hilbert space. Note that the subspace generated by λ(∅ ⊗ I ⊗ K) can be identified with H and earlier SI can be identified with the restriction of SI defined here. Same is valid for S ¯I . The subspace K is identified here with λ(∅ ⊗ ∅ ⊗ K). Thus K is a cyclic subspace for the representation sj ⊗ si → SjSi of Od ⊗ Od in the amalgamated Hilbert space. Let P be the projection on K. Then we have S∗ i P = P S∗ S∗ i P = P S∗ i P = v∗ i i P = v∗ i for all 1 ≤ i ≤ d. We start with a simple proposition. Proposition 3.1. The following holds: (a) For any 1 ≤ i, j ≤ d and I, J < ∞ and ¯I, ¯J < ∞ S∗ S∗ j SI S∗ ¯J j Ω >=< Ω, Si S ¯I S∗ ¯J SiSI S∗ < Ω, S ¯I J S∗ J Ω >; (b) The vector state ψΩ on UHFd ⊗ UHFd ≡ ⊗0 −∞Md ⊗∞ 1 Md ≡ ⊗ZZ Md is equal to ω; (c) π( Od ⊗ Od)′′ = B( H ⊗K H) if and only if {x ∈ B(K) : τ (x) = x, τ (x) = x} = {zI : z ∈ C}. i Ω = S∗ i Ω = v∗ i Ω = S∗ PROOF: By our construction S∗ i Ω = v∗ i Ω. Now (a) and (b) follows by repeated application of S∗ i Ω and commuting property of the two representation π(Od ⊗ I) and π(I ⊗ Od). The last statement (c) follows from a more general fact proved below that the commutant of π(Od ⊗ Od)′′ is order isomorphic with the set {x ∈ B(K) : τ (x) = x, τ (x) = x} = {zI : z ∈ C} via the map X → P XP where X is the weak∗ limit of {Λm Λn(x) as (m, n) → (∞, ∞). For details let Y be the strong limit of increasing sequence of projections (Λ Λ)n(P ) as S∗ n → ∞. Then Y commutes with Si Sj, S∗ j for all 1 ≤ i, j ≤ d. As Λ(P )) ≥ P , we i i Y = 0. As Y commutes with Si Sj we get also have Λ(Y ) ≥ Y . Hence (1 − Y )S∗ i Si SjY = 0 i.e. (1 − Y ) SjY = 0 for all 1 ≤ j ≤ d. By symmetry of the (1 − Y )S∗ argument we also get (1 − Y )SiY = 0 for all 1 ≤ i ≤ d. Hence Y commutes with π(Od)′′ and by symmetry of the argument Y commutes as well with π( Od)′′. As Y f = f for all f ∈ K and K is cyclic for the representation π( Od ⊗ Od) we conclude that Y = I in H ⊗K H. Let x ∈ B(K) so that τ (x) = x and τ (x) = x then as in the proof of Theorem 2.1 we also check that (Λ Λ)k(P )Λm Λn(x)(Λ Λ)k(P ) is independent of m, n as long as m, n ≥ k. Hence weak∗ limit Λm Λn(x) → X exists as m, n → ∞. Furthermore limiting element X ∈ π(Od ⊗ Od)′ and P XP = x. That the map X → P XP is an order-isomorphic on the set of self adjoint elements follows as in Theorem 2.1. This completes the proof. Proposition 3.1 in brief says that ( H ⊗K H, Si Sj 1 ≤ i, j ≤ d, P ) is the Popescu dilation associated with Popescu elements (K, vi vj , 1 ≤ i, j ≤ d}. Now we will be more specific in our starting Popescu's elements in order to explore the representa- tion π of Od ⊗ Od in the amalgamated Hilbert space H ⊗K H. 16 ANILESH MOHARI Let ω be a translation invariant factor state on B and ψ be an extremal point in Kω′. We consider the Popescu's elements (K, M, vk, 1 ≤ k ≤ d, Ω) described as in Proposition 2.4 associated with support projection of the state ψ in πψ(Od)′′ and also consider associated dual Popescu's elements (K, M, vk, 1 ≤ k ≤ d) where M is the von-Neumann algebra generated by {vk : 1 ≤ k ≤ d}. Thus in general M ⊆ M′ and an interesting question: when do we have M′ = M? Going back to our starting example of unique KMS state for the automorphisms βt(si) = tsi, t ∈ S1, we check d Sk and thus equality holds i.e. M = M′. But the corner that v∗ vector space Mc = P π( Od)′′P generated by the elements {vI v∗ J : I, J < ∞} fails to be an algebra. Thus two questions sounds reasonable here. kJ = 1 k = S∗ k, J v∗ (a) Does equality M′ = M holds in general for an extremal element ψ ∈ Kω′ and a factor state ω? (b) When can we expect Mc to be a ∗-algebra and so equal to M? The dual condition on support projection and equality M = M′ are rather deep and will lead us to a far reaching consequence on the state ω. In the paper [BJKW] these two conditions are implicitly assumed to give a criteria for a translation invariant factor state to be pure. Apart from this refined interest, we will address the converse problem that terns out to be crucial for our main results. In the following we prove a crucial step towards that goal fixing the basic structure which will be repeatedly used in the computation using Cuntz relations. Proposition 3.2. Let ω be a translation invariant factor state on B and ψ be an extremal point in Kω′. We consider the amalgamated representation π of Od ⊗ Od in H ⊗K H where the Popescu's elements (K, M, vk, 1 ≤ k ≤ d) are taken as in Proposition 2.4. Then the following statements hold: (a) π( Od ⊗ Od)′′ = B( H ⊗K H). Furthermore π(Od)′′ and π( Od)′′ are factors and the following sets are equal: (i) H = {z ∈ S1 : ψβz = ψ}; (ii) Hπ = {z : βz extends to an automorphisms of π(Od)′′}; (iii) Hπ = {z : βz extends to an automorphisms of π( Od)′′}. Moreover π( UHFd ⊗ I)′′ and π(I ⊗ UHFd)′′ are factors. (b) z → Uz is the unitary representation of H in the Hilbert space H ⊗K H defined by Uz(π(sj ⊗ si)Ω = π(zsj ⊗ zsi)Ω (c) The commutant of π( UHFd ⊗ UHFd)′′ is invariant by the canonical endomor- i and Λ(X) = Pi phisms Λ(X) = Pi SiXS∗ i . Same is true for each i that the i XSi keeps the commutant of π( UHFd ⊗ UHFd)′′ invariant. SiX S∗ surjective map X → S∗ Same holdss for the map X → S∗ i X Si. (d) The centre of π( UHFd ⊗ UHFd)′′ is invariant by the canonical endomorphisms Λ(X) = Pi SiXS∗ i . Moreover for each i the surjective map i XSi keeps the centre of π( UHFd ⊗ UHFd)′′ invariant. Same holdss for the i X Si. i and Λ(X) = Pi X → S∗ map X → S∗ SiX S∗ PROOF: P being the support projection by Proposition 2.4 we have {x ∈ B(K) : k = x} = M′. That (M′, τ , φ0) is ergodic follows from a general result [Mo1] ( see also [BJKW] for a different proof ) as (M, τ, φ0) is ergodic for a factor Pk vkxv∗ TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 17 state ψ being extremal in Kω′ (Proposition 2.6). Hence {x ∈ B(K) : τ (x) = τ (x) = x} = C. Hence by Proposition 3.1 we conclude that π( Od ⊗ Od)′′ = B( H ⊗K H). That both π(Od)′′ and π( Od)′′ are factors follows trivially as π( Od⊗Od)′′ = B( H⊗K H) and π(Od)′′ ⊆ π( Od)′. By our discussion above we first recall that Ω is a cyclic vector for the representa- tion of π( Od⊗Od). Let G = {z = (z1, z2) ∈ S1×S1 : βz extends to an automorphism on π( Od ⊗ Od)′′} be the closed subgroup where β(z1,z2)(sj ⊗ si) = z1sj ⊗ z2si. By repeated application of the fact that π(Od)′′ commutes with π( Od)′′ and S∗ i Ω = i Ω as in Proposition 3.1 (a) we verify that ψβ(z,z) = ψ on Od ⊗ Od if z ∈ H. For S∗ z ∈ H we set unitary operator Uzπ(x ⊗ y)Ω = π(βz(x) ⊗ βz(y))Ω for all x ∈ Od and y ∈ Od. Thus we have Uzπ(si)U ∗ z = zsi. By taking it's restriction to π(Od)′′ and π( Od)′′ respectively we check that H ⊆ Hπ and H ⊆ Hπ. z = zπ(si) and also Uzπ(si)U ∗ the canonical endomorphism Λ defined by Λ(X) = Pk SkXS∗ For the converse let z ∈ Hπ and we use the same symbol βz for the extension to an automorphism of π(Od)′′. By taking the inverse map we check easily that ¯z ∈ Hπ and in fact Hπ is a subgroup of S1. Since λ commutes with βz on Od, k also commutes with extension of βz on π(Od)′′. Note that the map π(x)H → π(βz(x))H for x ∈ Od is a well defined linear ∗-homomorphism. Since same is true for ¯z and βzβ¯z = I, the map is an isomorphism. Hence βz extends uniquely to an automorphism of π(Od)′′ H commuting with the restriction of the canonical endomorphism on π(Od)′′ H. Since π(Od)′′ H is a factor, we conclude as in Proposition 2.5 (a) that z ∈ H. Thus Hπ ⊆ H. As π( Od)′′ is also a factor, we also have Hπ ⊆ H. Hence we have H = Hπ = Hπ and {(z, z) : z ∈ H} ⊆ G ⊆ H × H. For the second part of (a) we will adopt the argument used for Proposition 2.5. To that end we first note that Ω being a cyclic vector for the representation Od ⊗Od in the Hilbert space H ⊗K H, by Lemma 7.11 in [BJKW] (note that the proof only needs the cyclic property ) the representation of UHFd on H⊗KH is quasi-equivalent to it's sub-representation on the cyclic space generated by Ω. On the other hand by our hypothesis that ω is a factor state, Power's theorem [Po1] ensures that the state ω′ (i.e. the restriction of ω to BR which is identified here with UHFd ) is also a factor state on UHFd. Hence quasi-equivalence ensures that π(I ⊗ UHFd)′′ is a factor. We also note that the argument used in Lemma 7.11 in [BJKW] is UHFd. Thus π( UHFd ⊗ I)′′ is also symmetric i.e. same argument is also valid for a factor. This completes the proof of (a). We have proved (b) while giving proof of (a). For X ∈ B( H ⊗K H), as Λ(X) commutes with π(λ( UHFd ⊗ UHFd))′′ and {SiS∗ j : 1 ≤ i, j ≤ d} we verify by Cuntz's relation that Λ(X) is also an element in the commutant of π(λ( UHFd ⊗ UHFd))′′ once X is so. It is also obvious that Λ(X) is an element in π( UHFd ⊗ UHFd)′′ if X is so. Thus Λ(X) is an element in the commutant/centre of π(λ( UHFd ⊗ UHFd)′′ once X is so. For the last statement i XSi on π( UHFd ⊗UHFd)′′ which is clearly onto by Cuntz consider the map X → S∗ relation. Hence we need to show that S∗ i XSi is an element in the commutant when- ever X is so. To that end note that S∗ i XSiS∗ i Y XSi = S∗ i Y SiS∗ i . Thus onto property of the map ensures i XSi since X commutes with SiS∗ i XY Si = S∗ i Y Si = S∗ i SiS∗ 18 ANILESH MOHARI i XSi is an element in the commutant/centre of π( UHFd ⊗ UHFd)′′ once X that S∗ is so. This completes the proof of (c) and (d). One interesting problem here how to describe the von-Neumann algebra I con- sists of invariant elements of the gauge action {βz : z ∈ H} in B( H ⊗K H). A general result due to E. Stormer [So] says that the algebra of invariant elements are von-Neumann algebra of type-I with centre completely atomic. Here the situation is much simple because we know explicitly that I = {Uz : z ∈ H}′ and we may write spectral decomposition as Uz = Pk∈ H zkFk for z ∈ H, H is the dual group of H either H = {z : zn = 1} or ZZ. Thus the centre of I is equal to {Fk : k ∈ H}. As a first step we describe the center Z of π( UHFd ⊗ UHFd)′′ by exploring Cuntz relation that it is also non-atomic even for a factor state ω. In fact we will show that the centre Z is a sub-algebra of the centre of I. In the following proposition we give an explicit description. Proposition 3.3. Let ω, ψ be as in Proposition 3.2 with Popescu system (K, M, vk, Ω) be taken as in Proposition 2.4 i.e. on support projection. Then the centre of π( UHFd ⊗ UHFd)′′ is completely atomic and the element E0 = [π( UHFd ⊗ UHFd)′ ∨ π( UHFd ⊗ UHFd)′′Ω] is a minimal projection in the centre of π( UHFd ⊗ UHFd)′′ and centre is invariant for both Λ and Λ. Furthermore the following holds: (a) The centre of π( UHFd ⊗ UHFd)′′ has the following two disjoint possibilities: (i) There exists a positive integer m ≥ 1 such that the centre is generated by the family of minimal orthogonal projections {Λk(E0) : 0 ≤ k ≤ m − 1} where m ≥ 1 is the least positive integer so that Λm(E0) = E0. In such a case {z : zm = 1} ⊆ H; (ii) The family of minimal nonzero orthogonal projections {Ek : k ∈ ZZ} where Ek = Λk(E0) for k ≥ 0 and Ek = S∗ I E0SI for k < 0 where I = −k and independent of multi-index I generates the centre and H = S1; (b) Λ(E) = Λ(E) for any E in the centre of π( UHFd ⊗ UHFd)′′ (c) If Λ(E0) = E0 then E0 = 1. PROOF: Let E′ ∈ π( UHFd ⊗UHFd)′ be the projection on the subspace generated SI ′ S∗ by the vectors {SIS∗ J ′ Ω, I = J, I ′ = J ′ < ∞} and πΩ be the restriction J UHFd ⊗ UHFd to the cyclic subspace HΩ generated by of the representation π of UHFd ⊗ UHFd we check that πω is unitary equivalent with Ω. Identifying B with πΩ. Thus πΩ is a factor representation. For any projection E in the centre of π( UHFd ⊗ UHFd)′′, via the unitary equiva- lence we note that EE′ = E′EE′ is an element in the centre of πΩ( UHFd ⊗UHFd)′′. ω being a factor state we conclude that EE′ is a scaler multiple of E′ and so we have (15) EE′ = ω(E)E′ Thus we also have EY E′ = ω(E)Y E′ for all Y ∈ π( UHFd ⊗ UHFd)′ and so (16) EE0 = ω(E)E0 TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 19 Since EE′ is a projection and E′ 6= 0, we have ω(E) = ω(E)2. Thus ω(E) = 1 or 0. So for such an element E the following is true: (i) If E ≤ E0 then either E = 0 or E = E0 i.e. E0 is a minimal projection in the centre of π( UHFd ⊗ UHFd)′′ (ii) ω(E) = 1 if and only if E ≥ E0 (iii) ω(E) = 0 if and only if EE0 = 0. As Λ(E0) is a projection in the centre of π( UHFd ⊗ UHFd)′′ by our last proposi- tion i.e. Proposition 3.2 (c), we have either ω(Λ(E0)) = 1 or 0. Since Λ(E0) 6= 0 by injective property of the endomorphism, we have either Λ(E0) ≥ E0 or Λ(E0)E0 = 0. In case Λ(E0) ≥ E0 we have S∗ i Λ(E0)Si = E0 for all 1 ≤ i ≤ d. If i E0Si being a non-zero projection in the centre of π(UHFd ⊗ UHFd)′′ (Propo- so S∗ sition 3.2 (c) ), by (i) we have E0 = Λ(E0). Thus we have either Λ(E0) = E0 or Λ(E0)E0 = 0. i E0Si ≤ S∗ i E0Si and S∗ j S∗ If Λ(E0)E0 = 0, we have Λ(E0) ≤ I − E0 and by Cuntz's relation we check that E0 ≤ I − S∗ j E0Sj for all 1 ≤ i, j ≤ d. So we also have E0S∗ j E0SjE0 = E0. Thus we have either E0S∗ i E0SiSj is an element in the centre by Proposition 3.2 (c). So either we have Λ2(E0)E0 = 0 or Λ2(E0) ≤ E0. Λ being an injective map we either have Λ2(E0)E0 = 0 or Λ2(E0) = E0. i E0SiSjE0 ≤ E0 − E0S∗ i E0SiSjE0 = 0 or E0S∗ i E0SiSjE0 = E0 as S∗ j S∗ j S∗ i E0SiSj ≤ I − S∗ j S∗ j S∗ More generally we check that if Λ(E0)E0 = 0, Λ2(E0)E0 = 0, ..Λk(E0)E0 = 0 for some k ≥ 1 then either Λk+1(E0)E0 = 0 or Λk+1(E0) = E0. To verify that first we check that in such a case E0 ≤ I − S∗ I E0SI for all I = n and then following the same steps as before to check that S∗ i S∗ I E0SI Si ≤ I − S∗E0Si for all i. Thus we have E0S∗ I E0SI SiE0 ≤ E0 and arguing as before we complete the proof of the claim that either Λk+1(E0)E0 = 0 or Λk+1(E0) = E0. i S∗ We summarize now by saying that E0, Λ(E0), .., Λm−1(E0) are mutually orthog- onal projections with m ≥ 1 possibly be infinite if not then Λm(E0) = E0. Let πk, k ≥ 0 be the representation π of UHFd ⊗ UHFd restricted to the sub- UHFd ⊗ UHFd is isomorphic to the rep- space Λk(E0). The representation π0 of UHFd ⊗ UHFd restricted to E′ and thus quasi-equivalent. For a resentation π of general discussion on quasi-equivalence we refer to section 2.4.4 in [BR vol-1]. ω being a factor state, π0 is a factor representation. We claim now that each πk is a factor representation. We fix any k ≥ 1 and let X be an element in the cen- tre of πk(UHFd ⊗ UHFd). Then for any I = k, S∗ I XSI is an element in the centre of π0(UHFd ⊗ UHFd) by Proposition 3.2 (d). Further S∗ I XSI = S∗ I XSIS∗ J SJ = S∗ J XSJ for all J = I = k. π0 being a factor represen- tation, we have S∗ I XSI = cE0 for some scaler c independent of the multi-index we choose I = k. Hence cΛk(E0) = PJ=k SJ S∗ J X = X as X is an element in the centre of π( UHFd ⊗ UHFd). Thus for each k ≥ 1, πk is a factor representation as π0 is so. J = PJ=k SJ S∗ I EkSI = E0 and so S∗ I XSIS∗ I SI S∗ We also note that Λ(E0)Λ(E0) 6= 0. Otherwise we have < SiΩ, SjΩ >= 0 for all i Ω >= 0 for all i, j as π(Od)′′ commutes with π( Od)′′. However i, j and so < Ω, SjS∗ i = 1 which leads a contradiction. Hence Λ(E0)Λ(E0) 6= 0. Si S∗ S∗ i Ω = S∗ i Ω and Pi As π restricted to Λ(E0) is a factor state and both Λ(E0) and Λ(E0) are elements in the centre of π( UHFd ⊗ UHF)′′ by Proposition 3.2 (d), we conclude that Λ(E0) = Λ(E0). Using commuting property of the endomorphisms Λ and Λ, we verify by a 20 ANILESH MOHARI simple induction method that Λk(E0) = Λk(E0) for all k ≥ 1. Thus the sequence of orthogonal projections E0, Λ(E0), ... are also periodic with same period or aperiodic according as the sequence of orthogonal projections E0, Λ(E0), .... If Λm(E0) = E0 for some m ≥ 1 then we check that P0≤k≤m−1 Λk(E0) is a Λ and as well Λ invariant projection and thus equal to 1 by cyclic property of Ω for π(Od ⊗ Od)′′. In such a case we set Vz = P0≤k≤m−1 zkEk for z ∈ S1 for which zm = 1 and check that Λ(Vz) = P0≤k≤m−1 zkΛ(Ek) = P0≤k≤m−1 zkEk+1 = ¯zVz where Em = E0 and so by Cuntz relation we have V ∗ z SiVz = ¯zSi for all 1 ≤ i ≤ d. z = ¯z Si for Following the same steps we also have Λ(Vz) = ¯zVz and so V ∗ z 1 ≤ i ≤ d. Thus Vz = Uz for all z ∈ H0 = {z : zm = 1} ⊆ H. SiV ∗ Now we consider the case where E0, Λ(E0), ..Λk(E0), .. is a sequence of aperiodic orthogonal projections. We extend family of projections {Ek : k ∈ Z} to all integers by and Ek = Λk(E0) for all k ≥ 1 Ek = S∗ I E0SI for all k ≤ 1, where I = −k We claim that the definition of {Ek; k ≤ −1} does depends only on length of the multi-index I that we choose. We may choose any other J so that J = I and check the following identity: S∗ I E0SI = S∗ J E0SJ = S∗ J SJ = S∗ I E0SI S∗ I SI S∗ J E0SJ J as I = J. Further Λk(E0) = Λk(E0) ensures that SI S∗ where E0, being an element in the centre of π( UHFd ⊗ UHFd)′′, commutes with SI S∗ J commutes with E0 for all I = J = k and k ≥ 1. Hence we also have SJ = S∗ for all J = I = k and k ≥ 1. Now we claim that I E0SI S∗ E−k = S∗ J E0 SJ J Λ(Ek) = Λ(Ek) = Ek+1 for all k ∈ Z. For k ≥ 0 we have nothing to prove. For k ≤ −1 we check that the following steps I E0SI ) = Xk S∗ I ′ E0SI ′ SkS∗ Λ(S∗ = Xk SkS∗ i S∗ I ′ E0SI ′ SiS∗ k i SiS∗ k = S∗ I ′ E0SI ′ where we wrote I = (I ′, i) and used elements SkS∗ i commutes with {Ek : k ∈ Z}, elements in the centre of π(UHFd ⊗ UHFd)′′. For a proof that Λ(Ek) = Ek+1 we may follow the same steps as Ek = S∗ I X SI where I = −k and k ≤ −1. We also claim that {Ek : k ∈ Z} is an orthogonal family of non-zero projections. To that end we choose any two elements say Ek, Em, k 6= m and use endomor- phism Λn for n large enough so that both n + k ≥ 0, n + m ≥ 0 to conclude that Λn(EkEm) = Ek+nEk+m = 0 as k + n 6= k + m. Λ being an injective map we get the required orthogonal property. Thus Pk∈Z Ek being an invariant projection for both Λ and Λ we get by cyclicity of Ω that Pk∈Z Ek = I. Let πk, k ≤ −1 be the UHFd ⊗ UHFd restricted to the subspace Ek. Going along the representation π of same line as above, we verify that for each k ≤ −1, πk is a factor representation UHFd ⊗ UHFd. We also set Vz = P−∞<k<∞ zkEk for all z ∈ S1 and check of that Λ(Vz) = ¯zVz and also Λ(Vz) = ¯zVz. Hence S1 = H and as H is a closed TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 21 subset of S1. This completes the proof of (a). Proof for (b) and (c) are now simple consequence of the proof of (a). It is clear that I contains I0 :=def π( UHFd ⊗ UHFd)′′ ∨ {Uz : z ∈ H}′′. By the last proposition the centre of I, which is equal to {Uz : z ∈ H}′′, contains the centre of π( UHFd ⊗ UHFd)′′ and thus by taking commutant we also have I ⊆ π( UHFd ⊗ UHFd)′′ ∨ π( UHFd ⊗ UHFd)′. In the last proposition we have described explicitly the factor decomposition of the representation π of π(UHFd ⊗ UHFd)′′. One central issue when such an factor decomposition is also an extremal decomposition. A clear answer at this stage seems to be bit hard. However the following proposition makes an attempt for our purpose. To that end we set few more notations and elementary properties. k is pure once π′ For each k ∈ H, let π′ k be the representation π of k( UHFd ⊗ UHFd)′′ then S∗ UHFd ⊗ UHFd restricted 0 is pure. Fix any k ∈ H and to Fk. We claim that each π′ let X be an element in the commutant of π′ I FkSI = I Λk(F0)SI = F0 as S∗ S∗ J SI commutes with F0 for I = J = k and further for any I = J, S∗ I XSI S∗ J SJ = S∗ J XSJ = S∗ J XSJ as X commutes with SI S∗ J with I = J. Thus by Proposition 3.2 (c) S∗ I XSI is an element in commutant of π′ I XSI = cF0 for some scaler c independent of I = k as π′ 0 is pure. We use commuting property of X with π( UHFd ⊗ UHFd)′′ to conclude that X = cΛk(E0) for some scaler c. If k ≤ −1 we employ the same method but with endomorphism Λ−k so that Λ−k(X) is an element in the commutant of π′ I = cI and by injective property of the endomorphism we get X is a scaler. Thus we conclude that each π′ 0( UHFd ⊗UHFd)′′. Thus PI:I=−k SI XS∗ 0( UHFd ⊗ UHFd)′′ for any I = k and thus S∗ I SI S∗ k is a pure once π′ 0 is pure. UHFd ⊗ UHFd Next we claim that for each fix k ∈ H0, representation πk of defined in Proposition 3.3 is quasi-equivalent to representation π′ k ( here we recall H0 ⊆ H as H0 ⊆ H ). That π′ 0 is quasi-equivalent to π0 follows as they are isomorphic. More generally for any k ∈ H, we abuse the notation and extend πk for the restriction of π to the minimal central projections Ek on the subspace span by {π( UHFd ⊗ UHFd)′f : ∀ f ∈ H ⊗K H, Fkf = f }. Each Ek is a minimal central element containing Fk and however two such elements i.e. Ek and Ej are either equal or mutually orthogonal. Thus {Ek : k ∈ H} = {Ek : k ∈ H0} and quasi-equivalence follows as πk is isomorphic with π′ k for all k ∈ H0. At this stage we also set for the time being 0 = [π( UHFd ⊗ UHFd)′′Ω] F ′ It is obvious that F ′ pure. 0 ≤ F0. We prove in following text that equality holds if ω is First we consider the case when H = {z : zn = 1}. Projections Λ(F ′ 0) and 0) are elements in π( UHFd ⊗ UHFd)′ by Proposition 3.3. The representation 0) are as well pure. i XSi with any 1 ≤ i ≤ d will do the job for the 0) for some scaler. By pulling 0)SiF ′ Λ(F ′ π( UHFd ⊗ UHFd)′′ restricted to both the projections Λ(F ′ A pull back by the map X → S∗ projection Λ(F ′ 0). Thus Λ(F ′ back with the action X → S∗ 0)Λ(F ′ i XSi we get F ′ 0) = cΛ(F ′ Λ(F ′ 0), Λ(F ′ 0 = cF ′ 0 and so 0)Λ(F ′ 0S∗ i c =< Ω, S∗ i Λ(F ′ 0)SiΩ > 22 ANILESH MOHARI < Ω, SkS∗ i F ′ 0SiS∗ kΩ > = Xk 0) ≥ Λ(F ′ 0). kΩ = S∗ as S∗ Ω, SkS∗ and Λ we conclude that Λ(F ′ line for Λk(F ′ Λn(F ′ 0) = Λk(F ′ kΩ >= 1. This shows that Λ(F ′ kΩ and further F0 commutes with π(UHFd and thus c = Pk < Interchanging the role of Λ 0). Proof essentially follows along the same 0) for all k ≥ 1. By Proposition 2.5 we also note that 0) = Λ(F ′ 0 = Λn(F0) as H = {z : zn = 1}. 0) = F ′ 0) is a Λ and as well Λ invariant projection. Since Thus F ′ = P0≤k≤n−1 Λ(F ′ F ′Ω = Ω we conclude by the cyclic property of Ω for π(Od ⊗ Od)′′ that F ′ = 1. Since Λk(F ′ 0) ≤ Fk and Pk Fk = 1 we conclude that Λk(F0) = Fk. In such a case we may check that Fk = [π( UHFd ⊗ UHFd)′′S∗ I Ω : I = n − k] for 1 ≤ k ≤ n − 1. Similarly in case H = S1 and ω is pure we also have F0 = F ′ 0 and for k ≥ 1 Fk = [π( UHFd ⊗ UHFd)′′SI Ω : I = k] F−k = [π( UHFd ⊗ UHFd)′′S∗ I Ω : I = k] Thus we have got an explicit description of the complete atomic centre of I when ω is a pure state. Proposition 3.4. Let ω, ψ and Popescu system (K, M, vk, Ω) be as in Proposi- tion 3.3. Then (a) {βz : z ∈ H} invariant elements in π( UHFd ⊗ Od)′′ ( as well as in π( Od ⊗ UHFd)′′ ) are equal to π(UHFd ⊗ UHFd)′′. (b) I = I0 if and only if ω is pure. Further the following statements are equivalent: (c) I = π( UHFd ⊗ UHFd)′′; (d) π( UHFd ⊗ Od)′′ = B( H ⊗K H); (e) π( Od ⊗ UHFd)′′ = B( H ⊗K H); In such a case ( if any of (c),(d) and (e) is true ) the following statements are also true: (f ) π( UHFd ⊗ UHFd)′′ is a type-I von-Neumann algebra with centre equal to {Uz : z ∈ H}′′ where Uz is defined in Proposition 3.2. (g) ω is a pure state on B. Conversely if ω is a pure state then π( UHFd ⊗ UHFd)′′ is a type-I von-Neumann algebra with centre equal to {Uz : z ∈ H0}′′ where H0 is a subgroup of H. PROOF: Along the same line of the proof of Proposition 2.5 (b) we get {βz : z ∈ H} invariant elements in π(Od ⊗ UHFd)′′ is π( UHFd ⊗ UHFd)′′ where factor property of π(Od)′′ is crucial as in proof of Proposition 2.5 (b). Same holds for π(UHFd ⊗ Od)′′ as π( Od)′′ is a factor. Here we comment that factor property of π( Od)′′ can be ensured whenever ψ is an extremal element in Kω′ (See Proposition 3.2 (a) ). For (b) we will first prove I0 = I if ω is pure. As by definition I0 ⊆ I, it is 0 i.e. X commutes with {Uz : z ∈ H}′′ enough if we show I′ and π(UHFd ⊗ UHFd)′′. For each k ∈ H, FkXFk is an element in the commutant of Fkπ(UHFd ⊗ UHFd)′′Fk. ω being pure each representation π restricted to Fk is 0 ⊆ I′. Let X ∈ I′ TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 23 irreducible and thus FkXFk = ckFk for some scalers ck. Hence X = Pk ckFk ∈ I′ = {Uz : z ∈ H}′′. For the converse we need to show that the restriction of π(UHFd ⊗ UHFd)′′ to F ′ 0 is pure. Let X be an element on the subspace F ′ 0 and in the commutant of 0π(UHFd ⊗ UHFd)′′F ′ 0, ( which in our earlier notation E′ in Proposition 3.3 ). F ′ Then X commutes with each Fk for k ∈ H and π(UHFd ⊗ UHFd)′′Fk as F ′ 0 ≤ F0 0 for k 6= 0. So X commutes with {Uz : z ∈ H}′′ and and Fk are orthogonal to F ′ π(UHFd ⊗ UHFd)′′ i.e. X ∈ I′ 0. By our assumption I0 = I, we have now X ∈ I′ which is equal to {Fk : k ∈ H}′′ and so X = cF0 for some scaler c0. This shows that F ′ 0 = F0 and ω is pure. (c) implies (d): {Uz : z ∈ H} is a commuting family of unitaries such that z and thus by (c) {Uz : z ∈ H}′′ ⊆ π( UHFd ⊗ UHFd)′′. Let X βz(X) = UzXU ∗ be an element in the commutant of π( UHFd ⊗ Od)′′. Then X commutes also with {Uz : z ∈ H}′′ and thus X ∈ π( UHFd ⊗ UHFd)′′ by (c). Hence X is an element in the centre of π( UHFd ⊗ UHFd)′′ and so X = Pk ckEk where Ek are the minimal projections in the centre of π( UHFd ⊗ UHFd)′′ given in Proposition 3.3. However X also commutes with π(Od)′′ by our assumption (c) and Λ(Ek) = Ek+1 for k ∈ H. So ck = ck+1 and X is a scaler multiple of unit operator. Hence (d) follows from (c). Along the same line we prove (c) implies (e). For a proof for (d) implies (c) and (e) implies (c), we simply apply (a). Now we will prove (f) and (g). That π( UHFd ⊗UHFd)′′ is a type-I von-Neumann algebra ( with completely atomic centre ) follows by a theorem of [So] once we use (c). In the proof of Proposition 3.3 we have proved that the centre of π( UHFd ⊗ UHFd)′′ is {Uz : z ∈ H0}′′ where H0 ⊆ H. For equality in the present situation we simply use (c), as βw(Uz) = Uz for all w, z ∈ H, to conclude that Uz is in the centre of π( UHFd ⊗ UHFd)′′. If (c) holds then I0 = I and thus (g) follows by (b). Here we will give another proof using the same idea to prove (f). Let X be an element in the commutant of π0( UHFd ⊗ UHFd)′′, where π0 is the factor representation on the minimal central projection E0 defined in Proposition 3.3. Then X commutes with {Uz : z ∈ H}′′ and so by (c) X in an element in π( UHFd ⊗ UHFd)′′. So X is in the centre of π0( UHFd ⊗ UHFd)′′. π0 being a factor representation X is a scaler multiple of E0. Thus π0 is an irreducible representation and so ω is pure. By Proposition 3.1 we recall that π′ of (B, ω). Thus π′ each k ∈ H0, πk being quasi-equivalent to π′ of π(UHFd ⊗ UHFd)′′. This completes the proof. 0 is unitarily equivalent to GNS representation 0 is irreducible if and only if ω is pure. So for a pure state ω, for k, πk is a type-I factor representation The following theorem is the central step that will be used repeatedly. Proposition 3.5. Let ω be an extremal translation invariant state on B and ψ be an extremal element ψ in Kω. We consider the Popescu elements (K, vk : 1 ≤ k ≤ d, M, Ω) as in Proposition 2.4 for the dual Popescu elements and associated amalgamated representation π of Od ⊗ Od as described in Proposition 3.1. Then the following holds: (a) π( Od ⊗ Od)′′ = B( H ⊗K H); 24 ANILESH MOHARI (b) π( Od)′′ = π(Od)′ if and only if π( Od)′′E = π(Od)′E. (c) Q = E E is the support projection of the state ψ in π(Od)′′ E and also in π( Od)′′E where E and E are the support projections of the state ψ in π(Od)′′ and π( Od)′′ respectively; (d) If EF = E F then E = F , E = F , P = Q. (e) If P = Q then the following statements are true: (i) M′ = M where M = {P SiP : 1 ≤ i ≤ d}′′; (ii) π(Od)′ = π( Od)′′. (f ) If P = [ MΩ] then M′ = M. (g) ω is pure on B if and only if there exists a sequence of elements xn ∈ M such that for each m ≥ 0 xn+mτn(x) → φ0(x)1 as n → ∞ in strong operator topology, equivalently φ0(τn(x)x∗ n+mxn+mτn(y)) → φ0(x)φ0(y) as n → ∞ for all x, y ∈ M k, x ∈ M; Same where M = {vi = P SiP : 1 ≤ i ≤ d}′′ and τ (x) = P1≤k≤d vkxv∗ holds true if we replace M0 for M where M0 = {x ∈ M : βz(x) = x; z ∈ H}. (a) is a restatement of Proposition 3.2 (a). E ( E ) being the support PROOF: projection of the state ψ in π(Od)′′ ( π( Od)′′ ) and ψ = ψΛ we have Λ(E) ≥ E and further we have E = [π(Od)′Ω] ≥ [π( Od)′′Ω] and hence Λn(E) ↑ I as n → ∞ because Ω is cyclic for π(Od ⊗ Od)′′ in H ⊗K H. We set von-Neumann algebras N1 = π(Od)′E and N2 = π( Od)′′E. By our construction in general π( Od)′′ ⊆ π(Od)′ and so N2 ⊆ N1. Since Λn(E) ↑ I as n → ∞ in strong operator topology, two operators in π(Od)′ are same if their actions are same on E. So (b) is true. For (c) we note that Q = E E ∈ N2 ⊆ N1 and claim that Q is the support projection of the state ψ in N2. To that end let xE ≥ 0 for some x ∈ π( Od)′′ so that ψ(QxQ) = 0. As Λk(xE) ≥ 0 for all k ≥ 1 and Λk(E) → I we conclude that x ≥ 0. As EΩ = Ω and thus ψ( Ex E) = ψ(QxQ) = 0, we conclude Ex E = 0, E being the support projection for π( Od)′′. Hence QxQ = 0. As ψ(Q) = 1, we complete the proof of the claim that Q is the support of ψ in N2. Similarly Q is also the support projection of the state ψ in π(Od)′′ E. This completes the proof of (c). Thus if EF = E F , we get Λn(E)F = EΛn( F ) and E Λ(F ) = Λ( E) F and thus taking limit we get F = E and E = F . It is obvious now that P = EF = E E = Q. This completes the proof of (d). As E ∈ π(Od)′′ and E ∈ π( Od)′′ we check that von-Neumann algebras M1 = Qπ(Od)′′Q and M1 = Qπ( Od)Q acting on Q satisfies M1 ⊆ M1′ . Now we explore that π( Od ⊗ Od)′′ = B(H ⊗K H) and note that in such a case Qπ( Od ⊗ Od)′′Q is the set of all bounded operators on the Hilbert subspace Q. As E ∈ π(Od)′′ and E ∈ π( Od)′′ we check that together M1 = Qπ(Od)′′Q and M1 = Qπ( Od)Q generate all bounded operators on Q. Thus both M1 and M1 are factors. The canonical states ψ on M1 and M1 are faithful and normal. We set lk = QSkQ and lk = Q SkQ, 1 ≤ k ≤ d and recall that vk = P SkP and vk = P SkP, 1 ≤ k ≤ d. We note that P lkP = vk and P lkP = vk where we recall by our construction P TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 25 is the support projection of the state ψ in π(Od)′′ [π(Od)Ω]. Q being the support projection of π(Od) E, by Theorem 2.4 applied to Cuntz elements {Si E : 1 ≤ i ≤ d}, Eπ(Od)′ E is order isomorphic to M1′ via the map X → QXQ. As the projection F = [π(Od)′′Ω] ∈ π(Od)′, we check that the element QF EQ ∈ M1′ . However QF EQ = E EF EE = QP Q = P and thus P ∈ M1′ . We also check that M1Ω = M1P Ω = P M1Ω = MΩ and thus P = [M1Ω]. We set M for the von-Neumann algebra generated by {vk : 1 ≤ k ≤ d}. In such a case M1 = M and M1 = M. By order isomorphic property we get (i) is equivalent to Eπ(Od)′ E = Eπ( Od)′′ E and taking commutant again we get π(Od)′′ E = π( Od)′ E. Now we invoke the first part of the argument changing the role or using the endomorphism Λ we conclude that π(Od)′′ = π( Od)′. This completes the proof of (e) provided we find a proof for (i) which is not so evident. Now we explore the representation π of Od ⊗ Od which is pure to prove (i). To that end we note since P = Q by our assumption, Ω is a common cyclic and separating vector for M and M′. Thus we can get an endomorphism α : M′ → M defined by α(y) = J J y J J where J is the Tomita's conjugate operator associated with cyclic and separating vector Ω for M. We note that the general theory does not guarantee [AcC] that the endomorphism be Takesaki's canonical conditional expectation associated with φ0. If so then the modular automorphism group (σt) of M′ also preserves M. Thus σz(x) ∈ M for −1 ≤ Im(z) ≤ 1 if x is an analytic element in M. Thus we would have got J vk(δ)J = σ i (vk(δ)) ∈ M where x(δ) is average of σt(x) with respect to Gaussian measure with variance δ > 0. That vk(δ) is an analytic element follows from the general Tomita-Takesaki theory [BR1]. Since vk(δ) → vk in strong operator topology as δ → 0 and M = {vk : 1 ≤ k ≤ d}′′ together with J MJ = M′ we arrived at M = M′. In the following we avoid this tempted route and aim to explore the general representation theory of C∗-algebras [BR1,chapter 2]. 2 We claim that M′ = M. Suppose not. Then α( M) is a proper von-Neumann subalgebras of α(M′) ⊆ M being an into map and hence α( M) is a proper von- Neumann subalgebra of M. Now consider the Popescu elements (K, α(vi), Ω) and its dilation as in Theorem 2.1. Then by the commutant lifting theorem applied to pairs (vi), α(vi) we find an unitaryoperator U on H so that U π( Od)′′U ∗ is strictly contained in π( Od)′′ ( Without loss of generality we can take the dilated Hilbert space for (K, α(vi), Ω) to be same as H as there exists an isomorphism preserving K, see the remark that follows after Theorem 2.1 ). We extend U to an unitary operator on H ⊗K H and denote πu(x) = U π(x)U ∗ for x ∈ Od ⊗Od which is unitary equivalent to the pure representation π and πu( Od)′′ is strictly contained in π( Od)′′. Now πu is also an amalgamated representation over the subspace K with Pu = Qu. Thus we can repeat now same with πu and so on. Note that the process won't terminate in finite time. Our aim is to bring a contradiction from this using formal set theory. To that end we reset for π as π0 as temporary notation as π will be used as notation for a generic representation. Let P be the collection of representation (π, Hπ, Ω) quasi-equivalent to π0 : Od ⊗ Od → B( H ⊗K H) with a shift invariant vector state ω(x) =< Ω, π(x)Ω > i.e. ω(π(θ(x)) = ω(π(x)). So there exists cardinal numbers nπ, n0(π) so that nπHπ is unitary equivalent to n0(π)π0. Thus given an element (π, Hπ, Ω) we can associate two cardinal numbers nπ and n0(π) and without 26 ANILESH MOHARI loss of generality we assume that Hπ ⊆ n0(π)H0 and nπHπ = n0(π)H0. π0 being a pure representation, any element π ∈ P is a type-I factor representation of Od ⊗Od. The interesting point here that ⊕π∈P π is also an element in P with associated cardinal numbers Pπ nπ and Pπ n0(π). We say (π1, H1, Ω1) << (π2, H2, Ω2) if there exists an isometry U : nπ1 H1 → nπ2H2 so that (C1) For each 1 ≤ α ≤ nπ1 we have U Ω1 (C2) nπ2π2(x)E′ (C3) U ⊕1≤α≤nπ1 2 = U nπ1π1(x)U ∗ where E ′ 1 ( Od)′′U ∗ ⊂ ⊕1≤α≤nπ2 πα 2 ∈ nπ2 π2( Od)′; 1 ( Od)′′E′ πα 2. α = Ω2 α′ for some 1 ≤ α′ ≤ nπ2; That the partial order is non-reflexive follows as (π, H, Ω) << (π, H, Ω) contra- 2. By our starting assumption that M′ 6= M we check that dicts (C3) as I = E′ π0 << πu. Thus going via the isomorphism we also check that for a given element π ∈ P there exists an element π′ ∈ P so that π << π′. Thus P0 is a non empty set and has at least one infinite chain. Partial order property follows easily. If π1 << π2 and π2 << π3 then π1 << π3. If U12 and U23 are isometric operators that satisfies (C1)-(C3) respectively, then U13 = U23U12 will do the job for π1 and π3. However by Hausdorff maximality theorem there exists a non-empty maximal totally ordered subset P0 of P. We claim that πmax = ⊕π∈P0π on Hπmax = ⊕π∈P0Hπ is an upper bound in P0. That πmax ∈ P is obvious. Further given an element (H1, π1, Ω1) ∈ P0 there exists an element (H2, π2, Ω2) ∈ P0 so that π1 << π2 by our starting remark as π0 << πu. By extending isometry U12 to an isometry from H1 → nπmaxHπmax trivially we get the required isometry that satisfies (C1),(C2) and (C3) where cardinal numbers nπmax = Pπ∈P0 nπ ∈ ℵ0. Thus by maximal property of P0 we have πmax ∈ P0. This brings a contradiction as by our construction (πmax, Hπmax, Ω) << (πmax, Hπmax, Ω) as πmax ∈ P0 but partial order is strict. This contradicts our starting hypothesis that M is a proper subset of M′. This completes the proof for (i) of (e) M′ = M when P = Q. In the proof of M′ = M in (e), we have used equality P = Q just to ensure that Ω is also a cyclic for M and P = Q is used to prove π(Od)′ = π( Od)′′. So (f) follows by the proof of (e). A proof for (g) is given in [Mo3] with M0. Here we will also give an alternative proof relating the criteria obtained in Proposition 3.4. To that end we claim that \n≥1 Λn(π(UHFd)′) = π( UHFd)′\ π(UHFd)′. That Λn(π(UHFd)′) ⊆ { SI S∗ J : I = J < ∞}′ follows by Cuntz relation and Λn(π(UHFd)′) ⊆ π( UHFd)′T π(UHFd)′. For the reverse inclusion let thus Tn≥1 X ∈ Eπ( UHFd)′T π(UHFd)′E. For n ≥ 1, we choose I = n and set Yn = I X SI. We check that it is independent of the index that we have chosen as S∗ J X SJ where in second equality we have Yn = S∗ used X ∈ π( UHFd)′ and also Λn(Yn) = PJ=n J = X. This proves the I X SI S∗ equality in the claim. Going along the same line we also get J X SJ = S∗ SJ = S∗ I SJ S∗ I X SI S∗ J SI S∗ \n≥1 Λn(π(Od)′) = π( UHFd)′\ π(Od)′ = π( UHFd ⊗ Od)′. By Proposition 3.4 ω is pure if and only if the set above is trivial. Thus once more by Proposition 1.1 in [Ar2] and Theorem 2.4 in [Mo2], purity is equivalent to TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 27 asymptotic relation Ψτ n − φ0 → 0 as n → ∞ for any normal state on M′ ( Here we recall by Proposition 2.4 P π(Od)′P = M′ as P is also the support projection in π(Od)′′F ), where commutant is taken in B(K). By duality argument [Mo3] we conclude that ω is pure if and only if there exists a sequence of elements xn ∈ M so that for each m ≥ 0, xm+nτn(x) → φ0(x)1 as n → ∞ for all x ∈ M ⊆ B(K). This completes the proof of (d) with M. For the proof with M0 we need to show if part as only if part follows M0 being a subset of M and τ takes elements of M0 to itself. For if part we refer to Theorem 3.2 in [Mo3]. We set (M′)0 = {x ∈ M′ : βz(x) = x, z ∈ H}. Similarly we also set M0 and ( M′)0 as (βz : z ∈ H) invariant elements of M and ( M′) respectively. We note that as a set ( M0)′ could be different from ( M′)0. We note also that P M1P ⊆ M and unless P is an element in M1, equality is not guaranteed for a factor state ω. The major problem is to show that P is indeed an element in M1 when ω is a pure state. We warn here an attentive reader that in general for a factor state ω, the set F π( Od)′′F , which is a subset of F π(Od)′F , need not be an algebra. However by commutant lifting theorem applied to dilation vi → SiF , π(Od)′F is order iso- morphic to M′ as P = FE is the support projection. Thus the von-Neumann sub-algebra generated by the elements F π( Od)′′F is order isomorphic to M. How- ever M0 may properly include M00 = {P π( UHFd)P }′′ ( as an example take ψ to be the unique KMS state on Od and ω be the unique trace on B for which we get M00 = C and P π( Od)′′P is the linear span of {v∗ J , I, vJ : J < ∞}. Existence of a φ0 preserving norm one projectionRz∈H βzdz ensures that modular operator of φ0 preserves M0 [Ta] and so does on (M′)0. However there is no reason to take it granted for M0 to be invariant by the modular group of ((M′)0, φ0). By Takesaki's theorem such a property is true if and only if there exists a φ0-invariant norm one projection from (M′)0 onto M0. In the following we avoid this tempted route. At this stage it is not clear how we can ensure existence of a norm one projection from M′ to M directly and so the equality M′ = M when ω is a pure state. Further interesting point here that the equality M = M′ holds when ω is the unique trace on B as v∗ d Sk for all 1 ≤ k ≤ d where P 6= Q and π(Od)′ ⊃ π( Od)′′. In the last proposition we have also proved if [ MΩ] = P then M′ = M. Thus a natural question that arises here: how the equality P = Q is related to purity of ω? We are now in a position to state the main mathematical result of this section. k and J v∗ k = S∗ kJ = 1 Theorem 3.6. Let ω be as in Theorem 3.5. Then the following holds: (a) P is also the support projection of ψ in π( Od))′′ H if and only if ω is pure. (b) If ω is pure then the following holds: (i) M′ = M where M = {P SiP : 1 ≤ i ≤ d}′′; (ii) π(Od)′ = π( Od)′′. (iii) πω(BR)′ = πω(BL)′′; 28 ANILESH MOHARI PROOF: First we will prove that ω is pure if P is also the support projection of the state ψ in π( Od)′′ F , where F = [π( Od)′′Ω]. The support projection of ψ in π( Od)′′ F is E F and thus we also have P = E F by our hypothesis. Since Λn(P ) = Λn(E)F ↑ F and now Λn(P ) = EΛn( F ) ↑ E as n ↑ ∞, we also have F = E. Similarly we also have for each n, E Λn(F ) = Λn( E) F and thus taking limit we also get E = F . So we have P = EF = E E = Q. M = P π( Od)′′P is cyclic in K i.e. [ MΩ] = [P π( Od)′′P Ω] = P F = P E = P as F = E. Λn(π( UHFd)) = π( UHFd)′′T π( UHFd)′ ( for a proof which is a However Tn→∞ simple application of Cuntz relation, we refer to section 5 of [Mo2]). Further ψ being a factor state in Kω′, by Proposition 3.2 π( UHFd)′′ is a factor. In particular we Λn(π( UHFd)) F = C F . Thus by Proposition 1.1 in [Ar2] we conclude have Tn→∞ that Ψ ◦ τn − φ0 → 0 as n → ∞ for all normal state Ψ on M0 where M0 = P π( UHFd)′′P as F = E and support projection of ψ in π( UHFd)′′ is E and P = E EE. Note that M0 ⊆ M′ 0 where M0 = P π(UHFd)′′P . Further by Proposition 2.5 and M0 = {x ∈ M : βz(x) = x; z ∈ H}. M0 = {x ∈ M : βz(x) = x; z ∈ H} Once we set P0 = [M0Ω] then we also have P0 = [ M0Ω] as [ MΩ] = P = [MΩ] by expending uz = Pk∈ H zkPk where z → uz = P UzP is an unitary representation of For x ∈ M, y ∈ M′ we have φ0(τ (x)y) = Xk kyΩ >= Xk kΩ, xyv∗ kΩ, xv∗ group H. kΩ > < v∗ < v∗ (as v∗ kΩ = v∗ kΩ ) < Ω, xvkyv∗ kΩ >= φ0(xτ (y)) = Xk The dual group of ( M, τ , φ0) is given on the commutant by ( M′, τ, φ0) where k for x ∈ M′. where commutant is taken in B(K). Now moving to τ (x) = Pk vkxv∗ {βz : z ∈ H} invariant elements in the duality relation above, we verify that adjoint Markov map of ( M0, τ , φ0) is given by ( M′ 0, the commutant of M0 is taken in B(K0) and K0 is the Hilbert subspace P0 with Ω as cyclic and separating vector for M0 in K0. Thus by Theorem 2.4 in [Mo3], there exists a sequence of elements yn ∈ M′ 0 such that ynτn(y) → φ0(y)1 as n → ∞ for all 0 ⊆ B(K0). Thus ω is pure by Proposition 3.5 (e) once we recall M′ = M y ∈ M′ as P = Q ( and so M′ 0 = M0 ) by Proposition 3.5 (d). 0, τ, φ0) where M′ Now we aim to prove F = E and F = E if ω is pure. We set unitary operator V = Pk Sk S∗ k. That V is an unitaryoperator follows by Cuntz's relations and commuting property of (Si) and ( Si). Further a simple computation shows that UHFd ⊗ UHFd and θ V π(x)V ∗ = π(θ(x)) for all x ∈ B = BL ⊗ BR, identified with is the right shift. We also have V EV ∗ = Xk,k′ Sk S∗ kES∗ k′ Sk′ = Λ(E) ≥ E So V (I − E)V ∗ ≤ I − E, i.e. (I − E)V ∗E = 0. Also for any X ∈ π(Od)′ we have kΩ. Thus (I − F)V ∗ F = 0 i.e. V F V ∗ = kΩ = S∗ kΩ as S∗ SkX S∗ V ∗ F XΩ = F Pk TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 29 Λ( F ) ≥ F . Similarly we also have V ∗ EV ≥ E and V ∗F V ≥ F . We set two family of increasing projections for all natural numbers n ∈ Z as follows En = V nE(V n)∗, Fn = V n F (V n)∗ Since βz(V ) = V for all z ∈ H, V ∈ π( UHFd ⊗ UHFd)′′ by Proposition 3.4 as ω is pure. ω being also a factor state, we have < f, V ng >→< f, Ω >< Ω, g > as n → +or − ∞ for any f, g ∈ π(Bloc))Ω by Power's criteria [Po1]. Since such vectors are dense in the Hilbert space topology and the family {V n : n ≥ 1} is uniformly bounded, we get V n → Ω >< Ω in weak operator topology as n → +or − ∞. For the time being we assume that H is trivial. Otherwise the argument that UHFd ⊗ UHFd i.e. π restricted follows here we can use for the representation π0 of to [π( UHFd ⊗ UHFd)Ω]. We have following distinct cases: Case 1. E 6= I( E 6= I). Let En → E−∞ as n → −∞ and thus V E−∞V = E−∞. We claim that either E−∞ = Ω >< Ω or E−∞ is a proper infinite dimensional if E−∞ is a finite projection then E−∞ = Ω >< Ω. Suppose not projection i.e. then the finite subspace is shift invariant. In particular there exists an unit vector f orthogonal to Ω such that V f = zf for some z ∈ S1 and this contradicts weak mixing property i.e. V n → Ω >< Ω in weak operator topology proved above as point spectrum of V has only 1 with spectral multiplicity 1. 0 = V nU0EU ∗ 0 in E−∞ and note that U0EnU ∗ If E−∞ is infinite dimensional we can get an unitaryoperator U0 from F0 = H ⊗K H onto E−∞ and via the unitary map we can get a sequence of increasing projections 0 (V n)∗. Note that if E−∞ is U0EnU ∗ infinite dimension the process will not stop in finite step. Thus we have F0 ⊖ Ω = ⊕1≤k≤nE F (k) where the index set is either singleton or infinity and each F (k) will give a system of imprimitivity with respect to V , where F (1) = F0 − E−∞. Further UHFd being a simple C∗-algebra, each such imprimitivity sysyem is of Mackey index ℵ0 [Mo3,section 4]: We fix a nonzero f ∈ E − θ−1(E) 6= 0 otherwise E = I as UHFd = BL and we θn(E) ↑ I as n → ∞. πf : x → θ−1(x)f gives a representation of check that [πf (θ−1(BL)′′f ] ≤ E − θ−1(E) as f ⊥ [θ−1(π(BR)′)Ω] ≥ [θ−1(π(BL))Ω]. Thus simplicity ensures that E − θ−1(E) is a projection of dimension ℵ0. Further nE is either 1 or ℵ0 since F0 is separable. Since F is also a proper projection, same argument is valid for F with F−∞ = F (k), where each H(k) give limn→−∞θn( F ) i.e. we can write F0 ⊖ Ω = ⊕1≤k≤n F rises to a system of imprimitivity with respect to V where each system of imprim- itivity is of Mackey index ℵ0 where F (1) = F0 − F−∞ and n F is either 1 or ℵ0. In the following we use temporary notation H for Hilbert subspace F0. For a cardinal number n, we amplify a representation π : B → B(H) of the C∗ algebra B to n fold direct sum nπ = ⊕1≤k≤nπk acting on nH = ⊕1≤k≤nHk defining by nπ(x)(⊕ζk) = ⊕(π(x)ζk) where πk = π is the representation of B = UHFd ⊗ UHFd on Hk = H where H = [π( UHFd⊗UHFd)Ω]. We also extend ¯F = ⊕ Fα, ¯E = ⊕Eα and ¯V = ⊕1≤k≤nVk respectively. We also set notation Ωk = ⊕1≤k≤nδk j Ω. 30 ANILESH MOHARI Thus by Mackey's theorem, there exists a cardinal number n ∈ ℵ0 and an unitary operator U : nH → nH so that ¯V = U ¯V U ∗ and ¯E = U ¯F U ∗. We set a represen- tation πU : B → B(nH) by πU (x) = U nπ(x)U ∗ and rewrite the above identity as ⊕1≤k≤n[πk(UHFd)′Ωk] = ⊕1≤k≤n[πU k ( UHFd)′′Ωk] k (x) = U πk(x)U ∗. Note that by our construction we can ensure U Ωk = Ωk where πU for all 1 ≤ k ≤ n as the operator intertwining between two imprimitivity systems are acting on the orthogonal subspace of the projection generated by vectors {Ωk : 1 ≤ k ≤ n}. We claim E = F . Suppose not i.e. F < E. In such a case we have ⊕1≤k≤n[πk(UHFd)′Ωk] < ⊕1≤k≤n[πu k (UHFd)′Ωk] Alternatively ⊕1≤k≤n[πk( UHFd)′′Ωk] < ⊕1≤k≤n[πu k ( UHFd)′′Ωk] Thus in principle we can repeat our construction now with πU and so we get a strict partial ordered set of quasi-equivalent representation of B. In the following we now aim to employ formal set theory to bring a contradiction on our starting assumption that F < E. To that end we need to deal with more then one representation of B. For the rest of the proof we reset notation π0 for π used for the pure representation of B in H0 = [π0(B)Ω0] where Ω0 is the cyclic vector, the reset notation for Ω. Let P be the collection of representation (π, Hπ, Ω) quasi-equivalent to π0 : B → B(H0) with a shift invariant vector state ω(x) =< Ω, π(x)Ω > i.e. ω(π(θ(x)) = ω(π(x)). So there exists minimal cardinal numbers nπ, n0(π) so that nπHπ is unitary equivalent to n0(π)π0. Thus for such an element (π, Hπ, Ωπ) we can associate two cardinal numbers nπ and n0(π) and without loss of generality we assume that Hπ ⊆ n0(π)H0 and nπHπ = n0(π)H0. π0 being a pure representation, any element π ∈ P is a type-I factor representation of B. The interesting point here that ⊕π∈P π is also an element in P with associated cardinal numbers Pπ nπ and Pπ n0(π). We say (π1, H1, Ω1) << (π2, H2, Ω2) if there exists an isometry U : nπ1H1 → nπ2H2 so that (C1) For each 1 ≤ α ≤ nπ1 we have U Ω1 (C2) nπ2π2(x)E′ (C3) ⊕1≤α≤nπ1 2 = U nπ1π1(x)U ∗ where E ′ 1 (UHFd)′Ω1 [πα α] < ⊕1≤α≤nπ2 α = Ω2 α′ for some 1 ≤ α′ ≤ nπ2; 2 ∈ nπ2 π2(B)′; [πα 2 (UHFd)′Ω2 α]E′ 2. In the inequality we explicitly used that both Hilbert spaces are subspaces of nH0 for some possibly larger cardinal number n. That the partial order is non- reflexive follows as (π, H, Ω) << (π, H, Ω) contradicts (C3) as I = E′ 2. Partial order property follows easily. If π1 << π2 and π2 << π3 then π1 << π3. If U12 and U23 are isometric operators that satisfies (C1)-(C3) respectively, then U13 = U23U12 will do the job for π1 and π3. Thus πU ∈ P and by our starting assumption that F 6= E we also check that π0 << πU . Thus going via the isomorphism we also check that for a given element π ∈ P there exists an element π′ ∈ P so that π << π′. Thus P0 is a non empty set and has at least one infinite chain containing π0. However by Hausdorff maximality theorem there exists a non-empty maximal totally ordered subset P0 of P containing π0. We claim that πmax = ⊕π∈P0π on Hπmax = ⊕π∈P0Hπ is an upper bound in P0. That πmax ∈ P is obvious. Further TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 31 given an element (H1, π1, Ω1) ∈ P0 there exists an element (H2, π2, Ω2) ∈ P0 so that π1 << π2 by our starting remark as π0 << πU . By extending isometry U12 to an isometry from H1 → nπmaxHπmax trivially we get the required isometry that satisfies (C1),(C2) and (C3) where cardinal numbers nπmax = Pπ∈P0 nπ ∈ ℵ0. Thus by maximal property of P0 we have πmax ∈ P0. This brings a contradiction as by our construction (πmax, Hπmax, Ω) << (πmax, Hπmax, Ω) as πmax ∈ P0 but partial order is strict. This contradicts our starting hypothesis that F < E. This completes the proof that F = E when E 6= I. By symmetry of the argument we also get F = E when E < 1. Case 2: E = I( E = I). We need to show F = I(F = I) respectively. Suppose not and assume that both F is a proper non-zero projection. We set projection G on the closed linear span of elements in the subspaces [θ−n(F )π( UHFd)′′Ω] for all n ≥ 0. We recall that θ(X) = V XV ∗ where V = Pk Sk S∗ k and θ−1(X) = Λ(X) for X ∈ π( UHFd)′′. Thus we have V ∗θ−n(F )π( UHFd)′′Ω = θ−n−1(F )V ∗π( UHFd)′′V Ω = θ−n−1(F )Λ(π( UHFd)′′Ω. Thus (1 − G)V ∗G = 0 i.e. θ(G) ≥ G. It is also clear that F ≤ G as the defining sequence of subspaces of G goes to precisely F as n → ∞ ( recall that θ−n(F ) = Λn(F ) ↑ I strongly as n ↑ ∞ ). Once more we have θn(G) ≥ θn( F ) = Λn( F ) ↑ I as n ↑ ∞. If G is a proper projection we can follow the steps as in the case 1 to find an unitary operator U : nH → nH with U Ωk = Ωk and U ¯V U ∗ = ¯V so that U ¯GU ∗ = ¯F . We consider the subset PG of elements in P for which Eπ = 1 and {θ−n(Fπ) : n ≥ 0} commutes with Fπ and modify the strict partial ordering by modifying (C3) as [πα 1 ( UHFd)′′Ω1 α]E′ 2 (C3') ⊕1≤α≤nπ1 So we also get πU ∈ PG and π0 << πU and going along the same line we conclude that G = F . Thus we conclude that G is either equal to 1 or G = F . α] < ⊕1≤α≤nπ2 [πα 2 ( UHFd)′′Ω2 Sub-case 1 of case 2: If G = I then F G = F and so [F π( UHFd)′′Ω] = F as θ−n(F ) ≥ F . Thus F ≥ F . So F ≥ Λn(F ) for all n ≥ 1 and taking limit we get F ≥ I i.e. F = I. This contradicts our starting assumption that F is a proper projection. Sub-case 2 of case 2: Now we consider the case G = F < I. In such a case we have (1 − F )θ−n(F ) F = 0 and so θ−n(F ) commutes with F for all n ≥ 0. Now we set projection F ′ defined by F ′ = F − F F + Ω >< Ω and check by commuting property of F with F that F ′θ−1(F ′)F ′ = F (I − F )θ−1(F )(I − θ−1( F ))F (I − F ) + Ω >< Ω = F (I − F ) + Ω >< Ω as θ−1(F ) ≥ F and θ−1( F ) ≤ F . If F ′ − Ω >< Ω = F (I − F ) 6= 0 and so nor equal to I. If θ−n(F ′) ↑ I + Ω >< Ω − limn→∞θ−n( F ) 6= I, we get the orthogonal projection i.e. limn→∞θ−n( F ) − Ω >< Ω is θ invariant and as in case-1 it would be an infinite dimensional separable Hilbert subspace. Thus we can follow the steps 32 ANILESH MOHARI of case-1 with elements F ′, F replacing the role of F , E to get an unitary operator U : nH → nH so that U ¯V = ¯V U and U ¯FU ∗ = ¯F ′ for a cardinal number n. Now we consider a further subset PG′ of PG consist of quasi-equivalent rep- resentations π to π0 of B where π admits the additional property: Eπ = I and {θ−n( Fπ) : n ≥ 0} commutes with Fπ satisfying Fπ Fπ 6= Fπ with the strict partial ordering π1 << π2 given by modifying condition (C3') as (C3") ⊕1≤α≤nπ1 [πα 1 ( UHFd)′′Ω1 α] > ⊕1≤α≤nπ2 [πα 2 ( UHFd)′′Ω2 α]E′ 2 Since πU also satisfies the conditions that of π0 ∈ PG′ by covariance relation of U with respect to shifts once more we get πU ∈ PG′ and π0 << πU . Thus we can repeat the process and so PG′ has at least one infinite chain of totally ordered containing π0. Once more by Hausdorff maximality principle we bring a contradiction to our starting assumption that F ′ < F . Thus we have shown either FF = F or FF = Ω >< Ω. However Λ( FF ) = Λ(F FF ) = F Λ( F)F ≥ F FF = FF and similarly Λ( FF ) ≥ FF i.e. subharmonic projections for both the endomorphisms. If F F = Ω >< Ω we get by sub-harmonic property that ω(sI s∗ J where (li) is a scalers i.e. ω is a pure product state on AR [BJP]. So EF = Ω >< Ω and F = Ω >< Ω ≤ F as E = 1 and F Ω = Ω. J ) = lIl∗ On the other hand F commutes with F and F F = F also says that F ≤ F . Thus in any case we have F ≥ F and so Λn(F ) ≤ F for all n ≥ 1. Taking limit n → ∞, we get I ≤ F i.e. F = I. This brings a contradiction to our starting hypothesis that F is a proper projection i.e. F < E = I. This completes the proof of F = E for the case when H is trivial closed subgroup of S1. Now we remove the assumption that H is trivial. By our construction Q0 = [π0(UHF)′Ω][π0( UHFd)′Ω] as E = [π(UHF)′Ω], E = [π( UHFd)′Ω] and Q = E E. As F = [π(Od)′′Ω] and P = EF we get P0 = [π0(UHF)′Ω][π0(UHFd)′′Ω]. For the situation case-1 (i.e. E 6= I ) one can simply replace π by π0 to get a proof from above that we have P0 = Q0. For case-2 we need to modify the argument with F and F replaced by F F0 and F F0. Since F and F are (βz : z ∈ H) invariant, by purity we have F and F are elements in π( UHFd ⊗ UHFd)′′ and thus commutes with F0. As V commutes with F0, the argument will go through for the situation to conclude that either F F0 ≥ F F0 or θ−n(F )F0 commutes with F F0. In case F F0 ≥ F F0 we get F Fk ≥ Λk(F )Fk ≥ F Fk for all k ∈ H but k ≥ 0. I F0 SI ≥ F F−k where I = k. However In case H = ZZ, we also deduce S∗ I F ≥ S∗ F SI and thus we also have F F−k ≥ F F−k for all k ≥ 1. Thus summing up I we get F ≥ F and as before we conclude that F = I. This bring a contradiction. F SI S∗ Thus we are left to consider the situation case -2 where E = I and {θ−n(F )F0 : n ≥} commutes with F F0 ( H 6= {1} ). In such a case we will have as before F F F0 = Ω >< Ω and so ω is a pure product state and P = EF = Ω >< Ω. Thus F ≤ F as E = I and F Ω = Ω. So Λn(F ) ≤ F for all n ≥ 1. Taking limit we conclude F = I which contradicts our starting assumption that F < E = I. TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 33 Now we write the equality P0 = Q0 as EF F0 = E EF0 and apply Λ on both side to conclude that Λ(E)F F1 = Λ(E) EF1 and multiplying by E from left we get EF F1 = E EF1 as Λ(E)E = E and thus we get P F1 = QF1. By repeated application of Λ, we get P Fm = QFm. Thus we get P = Pk P Fk = Pk QFk = Q. This F Fk = Pk EFk = E and also completes the proof for P = Q. Similarly F = Pk F = E. Now we are left to prove those three statements given in (b). ω being pure we have P = Q and thus by Proposition 3.5 we have M = M′ and π(Od)′ = π( Od)′′. We are left to show πω(BR)′ = πω(BL)′′. For that we recall F0 and check few obvious relation F0π(Od)′F0 = F0π( Od)′′F0 and πω(BR)′ ⊆ F0π(Od)′F0. Since F0π( UHFd)′′F0 is equal to (βz : z ∈ H) invariant elements in F0π( Od)′′F0 and elements in πω(BR)′ are (βz : z ∈ H) invariant we conclude that πω(BR)′ ⊆ πω(BL)′′. Inclusion in other direction is obvious and thus Haag duality property (iii) holds. 4. Symmetry of a translation invariant pure state on B In this section we investigate ω on B with some additional natural discrete symme- try. Let ψ be a λ-invariant state on Od and ψ be the state on Od defined by ψ(sI s∗ J ) = ψ(s J s∗ I ) for all I, J < ∞ and (H ψ, π ψ, Ω ψ) be the GNS space associated with (Od, ψ). That ψ is well defined follows once we check by (14) that I ) = φ0(vI v∗ J ) I ) = φ0(v J v∗ ψ(s J s∗ and appeal to Proposition 2.3 by observing that cyclicity condition i.e. the closed linear span P0 of the set of vectors {v∗ I Ω : I < ∞} is K, can be ensured if not true already by taking a new set of Popescu elements {P0vkP0 : 1 ≤ k ≤ d}. Otherwise one may also recall Proposition 2.3 that the map sI s∗ J being unital and completely positive [Po] ( in particular positive ), ψ is a well defined state on Od. J → vI v∗ Similarly for any translation invariant state ω on B we set translation invariant state ω by reflecting around the point 1 2 on B by ω(Q(−l) −l ⊗ Q(−l+1) −l+1 ⊗ ... ⊗ Q(−1) −1 ⊗ Q(0) 0 ⊗ Q(1) 1 ... ⊗ Q(n) n ) (17) = ω(Q(−n+1) n ... ⊗ Q(0) 1 ⊗ Q(1) 0 ⊗ Q(2) −1 ⊗ ...Q(l) −l+1 ⊗ Q(l+1) −l ) for all n, l ≥ 1 and Q−l, ..Q−1, Q0, Q1, .., Qn ∈ Mn(IC) where Q(k) is the matrix Q at lattice point k. We define ω on B by extending linearly to any Q ∈ Bloc. Note first that the map ψ → ψ is a one to one and onto affine map in the convex set of λ invariant state on Od. In particular the map ψ → ψ takes an element from Kω to Kω and the map is once more one to one and onto. Hence for any extremal point ψ ∈ Kω, ψ is also an extremal point in Kω. Using Power's criteria we also verify here that ω is an extremal state if and only if ω is an extremal state. However such a conclusion for a pure state ω is not so obvious. We have the following useful proposition. Proposition 4.1. Let ω be an extremal translation invariant state on B and ψ → ψ be the map defined for λ invariant states on Od. Then the following holds: (a) ψ ∈ Kω is a factor state if and only if ψ ∈ Kω is a factor state. 34 ANILESH MOHARI (b) ω is pure if and only if ω is pure. (c) A Popescu systems (K, M, vk, Ω) of ψ satisfies Proposition 2.4 with (πψ(sk), 1 ≤ k ≤ d, P, Ω) i.e. the projection P on the subspace K is the support projection of the state ψ in π(Od)′′ and vi = P πψ(si)P for all 1 ≤ i ≤ d, then the dual Popescu systems (K, M′, vk, Ω) satisfies Proposition 2.4 with (π ψ(sk), 1 ≤ k ≤ d, P, Ω) the projection P on the subspace K is the support projection of the state ψ i.e. in π ψ(Od)′′ and vi = P π ψ(si)P for all 1 ≤ i ≤ d, if and only if {x ∈ B(K) : Pk vkxv∗ k = x} = M. PROOF: Since ω is an extremal translation invariant state, by Power's criteria ω is also an extremal state. As an extremal point of Kω is map to an extremal point in Kω by one to one property of the map ψ → ψ, we conclude by Proposition 2.6 that ψ is a factor state if and only if ψ is a factor state. For (b) note that xy = xy and x∗ = x∗ by our definition. Thus ω(x∗y) = ω( x∗y) = ω((x)∗ y). Thus one can easily construct an unitary operator between the two GNS spaces associated with (B, ω) and (B, ω) intertwining two representation modulo a reflection i.e. U πω(x)U ∗ = πω(x) and U Ωω = Ωω. Thus (b) is now obvious. (c) follows by the converse part of the Proposition 2.4 once applied to the dual Popescu systems (K, M′, vk, Ω). Thus the state ω is translation invariant, ergodic, factor state, pure if and only if ω is translation invariant, ergodic, factor state, pure respectively. We say ω is lattice symmetric if ω = ω. For a λ invariant state ψ on Od we define as before a λ invariant state ψ by (18) ψ(sI s∗ J ) = ψ(s I s∗ J ) for all I, J < ∞. It is obvious that ψ ∈ Kω′ if and only if ψ ∈ Kω′ and the map ψ → ψ is an affine map. In particular an extremal point in Kω′ is also mapped It is also clear that ψ ∈ Kω′ if and only if ω is to an extremal point of Kω′. lattice symmetric. Hence a lattice symmetric state ω determines an affine map ψ → ψ on the compact convex set Kω′. Furthermore, if ω is also extremal on B, then the affine map takes extremal elements to extremal elements of Kω′. The set of extremal elements in Kω′ can be identified with S1/H ≡ S1 or {1} and the restriction of the affine map on the set of extremal element is continuous in weak topology ( by Proposition 2.6 the map z → ψβz is one to one and onto the set of extremal elements of Kω′ for a fixed extremal element ψ ∈ Kω′ ). Thus there exists z0 ∈ S1 so that ψ = ψβz0 and as ββz = ββz for all z ∈ S1, we get the affine map taking ψβz → ψβz0βz and thus determines a continuous one to ψ = ψ its inverse is itself. Thus either the affine one and onto map on S1/H and as map has a fixed point or z2 it is a rotation map by an angle 2π ( Here we have identified S1/H with S1 in case H 6= S1 ). Thus there exists an extremal element ψ ∈ Kω′ so that either ψ = ψβζ where ζ is either 1 or −1 where we recall that we have identified S1/H = S1 when H 6= S1. Note that if we wish to remove the identification, then for H = {z : zn = 1} for some n ≥ 1, ζ is either 1 or exp πi n . Note that in case H = S1 then ψ = ψ for ψ ∈ Kω′ as Kω is a singleton set by Proposition 2.6. 0 = 1 i.e. Proposition 4.2. Let ω be a translation invariant lattice symmetric state on B. Then the following holds: (a) If ω is also an extremal translation invariant state on B then H = {z ∈ S1 : ψβz = ψ} is independent of ψ ∈ Kω′. TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 35 πi (b) If H = {z : zn = 1} for some n ≥ 0 then ψ = ψβζ for all ψ ∈ Kω′ where ζ is n . Let (H, Sk, 1 ≤ k ≤ d, Ω) be the GNS space associated with fixed either 1 or exp (Od, ψ), P be the support projection of the state ψ in π(Od)′′ and K = P H with Popescu systems (K, M, vk, 1 ≤ k ≤ d, Ω) as in Proposition 2.4 where vk = P SkP k : 1 ≤ k ≤ d}′′ is invariant for and associated normal state φ0 on M = {vk, v∗ k. Let ( H, Sk 1 ≤ k ≤ d, Ω) be the Popescu minimal dilation in τ (x) = Pk vkxv∗ Theorem 2.1 of the dual Popescu systems (K, M, vk, 1 ≤ k ≤ d, Ω) defined in Proposition 3.2. Then there exists an unitaryoperator Uζ : H ⊗K H so that (19) for all 1 ≤ k ≤ d. U ∗ ζ = U ¯ζ, UζΩ = Ω, UζSkU ∗ ζ = β ¯ζ( Sk) Furthermore if ω is also pure then there exists an unitaryoperator uζ : K → K so that (20) uζΩ = Ω, uζvku∗ ζ = β ¯ζ(vk) ζ = ∆− 1 2 u∗ 1 ζ = J , uζ∆ ζ = ζMuζ = M. Moreover M′ = M. Further if ζ = 1 then uζ is self-adjoint for all 1 ≤ k ≤ d and uζJ u∗ M′, u∗ and otherwise if ζ 6= 1 then u2n ζ (c) If H = S1 then Kω′ is having only one element ψ, so ψ = ψ and (19) and (20) are valid with ζ = 1. ζ = u ¯ζ and uζMu∗ is self adjoint. 2 , u∗ PROOF: (a) follows by Proposition 2.6. Now we aim to prove (b). For existence of an extremal state ψ ∈ Kω′ so that ψ = ψβζ we refer to the paragraph preceding the (ψβz) = ψβz for all z ∈ S1, a simple application statement of this proposition. As of Proposition 2.6 says that ψ = ψβζ for all extremal points in Kω′ if it holds for one extremal element. Hence existence part in (b) is true by Krein-Millmann theorem. Ω is a cyclic vector for π(Od ⊗ Od) and thus we define Uζ : H ⊗K H → H ⊗K H by Uζ : SI S∗ J SI ′ S∗ J ′ Ω → β ¯ζ(SI ′ S∗ J ′ SI S∗ J ) Ω That Uζ is an unitary operator follows from (14) and the dual relation (18) along with our condition that ψ = ψβζ . By our construction we also have UζSk = β ¯ζ( Sk)Uζ for all 1 ≤ k ≤ d. In particular Uζπ(Od)′′U ∗ ζ = π( Od)′′. ω being pure we have P = Q by Theorem 3.6 as F = E and so U P U ∗ = U QU ∗ = U E EU ∗ = EE = Q = P which ensures an unitary operator uζ = P UζP on K and a routine calculation shows that (21) uζv∗ ku∗ ζ = β ¯ζ(v∗ k) for all 1 ≤ k ≤ d. As U ∗ U 2n (b). is inverse of its own. Thus u2n ζ ζ ζ = U ¯ζ we have u∗ ζ = u ¯ζ. If ζ 6= 1, then ζ2n = 1 and thus is self-adjoint. That M′ = M by Theorem 3.6 In the following we consider the case ζ = 1 for simplicity of notation and other- wise for the case ζ 6= 1 very little modification is needed in the symbols or simply reset temporary notation vk for ¯ζ vk i.e. include the phase factor. We denote u1 = u in the following for simplicity. It is simple to verify now the J Ω where SxΩ = x∗Ω, x ∈ M following steps uSvI v∗ I Ω = F vI v∗ J Ω = uvJ v∗ I Ω = vJ v∗ 36 ANILESH MOHARI 1 2 u, i.e uJ u∗u∆ and F x′Ω = x′∗Ω, x′ ∈ M′ are the Tomita's conjugate operator. Hence uJ ∆ 2 = J ∆− 1 2 u∗ = J ∆− 1 2 and by uniqueness of polar decomposition we 2 . That uMu∗ = M is obvious. For 2 u∗ = ∆− 1 conclude that uJ u∗ = J and u∆ uMu∗ = M we note that by our construction U SkU ∗ = Sk and so U π( Od)U ∗ = π(Od) and hence projecting to its support projection we get the required relation. 1 1 m 1 we set Qt = Qt(l) Now we introduce another useful symmetry on ω. If Q = Q(l) ⊗ .... ⊗ Q(l+m) where Q0, Q1, ..., Qm are arbitrary elements in Md and Qt 1, .. stands for transpose with respect to an orthonormal basis (ei) for ICd (not complex conjugate) of Q0, Q1, .. respectively. We define Qt by extending linearly for any Q ∈ Bloc. For a state ω on UHFd C∗ algebra ⊗ZZMd we define a state ¯ω on ⊗ZZ Md by the following prescription 0 ⊗ Qt(l+1) 0, Qt ⊗ .. ⊗ Qt(l+m) 0 ⊗ Q(l+1) 1 m (22) ¯ω(Q) = ω(Qt) Thus the state ¯ω is translation invariant, ergodic, factor state if and only if ω is translation invariant, ergodic, factor state respectively. We say ω is real if ¯ω = ω. In this section we study a translation invariant real state. For a λ invariant state ψ on Od we define a λ invariant state ¯ψ by (23) ¯ψ(sI s∗ J ) = ψ(sJ s∗ I ) for all I, J < ∞ and extend linearly. That it defines a state follows as for an J we have ¯ψ(x∗x) = ψ(y∗y) ≥ 0 where y = P c(I, J)sJ s∗ element x = P c(I, J)sI s∗ I . It is obvious that ψ ∈ Kω′ if and only if ¯ψ ∈ K¯ω′ and the map ψ → ¯ψ is an affine map. In particular an extremal point in Kω′ is also mapped to an extremal point in K¯ω′. It is also clear that ¯ψ ∈ Kω′ if and only if ω is real. Hence a real state ω determines an affine map ψ → ¯ψ on the compact convex set Kω′. Furthermore, if ω is also extremal on B, then the affine map, being continuous on the set of extremal elements in Kω′, which can be identified with S1/H ≡ S1 or {1} ( by Proposition 2.6 ) by fixing an extremal element ψ0 ∈ Kω′. In such a case there exists a unique z0 ∈ S1 so that ¯ψ0 = ψ0βz0 Now ¯ψ0βz = ¯ψ0β¯z for all z ∈ S1, the affine map takes ψ0βz → ψ0βz0 ¯z. If z0 = 1 we get that the map fixes two point namely ψ0 and ψ0β−1. Even otherwise we can choose z ∈ S1 so that z2 = z0 and for such a choice we get an extremal element namely ψ0βz gets fixed by the map. What is also crucial here that we can as well choose z ∈ S1 so that z2 = −z0, if so then ψ0βz gets mapped into ψ0βz0β¯z = ψ0β−z = ψ0βzβ−1. Thus in any case we also have an extremal element ψ ∈ Kω′ so that ¯ψ = ψβζ where ζ ∈ {1, −1}. Thus going back to the original set up, we sum up the above by saying that if n } then there exists an extremal element iπ H = {z : zn = 1} ⊆ S1 and ζ ∈ {1, exp ψ ∈ Kω′ so that ¯ψ = ψβζ. Proposition 4.3. Let ω be a translation invariant real factor state on ⊗ZZMd. Then the following holds: (a) if H = {z : zn = 1} ⊆ S1 and ζ ∈ {1, exp iπ n } then there exists an extremal element ψ ∈ Kω′ so that ¯ψ = ψβζ . Let (H, πψ(sk) = Sk, 1 ≤ k ≤ d, Ω) be the GNS representation of (Od, ψ), P be the support projection of the state ψ in π(Od)′′ and (K, M, vk, 1 ≤ k ≤ d, Ω) be the associated Popescu systems as in Proposition 2.4. Let ¯vk = J vkJ for all 1 ≤ k ≤ d and ( ¯H, ¯Sk, P, Ω) be the Popescu minimal dilation TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 37 as described in Theorem 2.1 associated with the systems (K, M′, ¯vk, 1 ≤ k ≤ d, Ω). Then there exists an unitary operator Wζ : H → ¯H so that ζ = β ¯ζ( ¯Sk) (24) for all 1 ≤ k ≤ d. Furthermore P is the support projection of the state ¯ψ in ¯π(Od)′′ and there exists an unitaryoperator wζ on K so that Wζ Ω = Ω, WζSkW ∗ (25) wζΩ = Ω, wζ vkw∗ ζ = β ¯ζ(¯vk) = J βζ(vk)J for all 1 ≤ k ≤ d and wζJ w∗ only if ζ = 1; (b) If H = S1, Kω′ is a set with unique element ψ so that ¯ψ = ψ and relations (24) and (25) are valid with ζ = 1. 2 . wζ is self adjoint if and ζ = J and wζ ∆ ζ = ∆− 1 2 w∗ 1 PROOF: For existence part in (a) we refer the paragraph above preceded the statement of the proposition. We fix a state ψ ∈ Kω′ so that ¯ψ = ψβζ and define W : H → ¯H by Wζ : SI S∗ J Ω → β ¯ζ ( ¯S∗ I ¯S∗ J )Ω That Wζ is an unitaryoperator follows from (3.10) and thus Wζ Sk = β ¯ζ( ¯Sk)Wζ for all 1 ≤ k ≤ d. For simplicity of notation we take the case ζ = 1 as very little modification is needed to include the case when ζ 6= 1 or reset Cuntz elements by absorbing the phase factor in the following computation and use notation W for Wζ. P being the support projection we have by Proposition 2.4 that M′ = {x ∈ B(H) : Pk vkxv∗ kJ = x}. Hence by the converse part of Proposition 2.4 we conclude that P is also the support projection of the state ¯ψ in ¯π(Od)′′. Hence Wζ P W ∗ ζ = P . Thus we define an unitary operator wζ : K → K by wζ = P WζP and verify that k = x} and thus M = {x ∈ B(K) : Pk J vkJ xJ v∗ = P Wζβζ (S∗ k = P ¯S∗ ¯v∗ kP k)W ∗ ζ P = P WζP βζ(S∗ = P WζP βζ(v∗ k)P W ∗ ζ P = wζ βζ(v∗ k)P W ∗ ζ P k)w∗ ζ . We recall that Tomita's conjugate linear operators S, F [BR] are the closure of the linear operators defined by S : xΩ → x∗Ω for x ∈ M and F : yΩ → y∗Ω for y ∈ M′. We check the following relations for ζ = 1 with simplified notation w1 = w, wSvI v∗ J Ω = wvJ v∗ = F ¯vI ¯v∗ J Ω = F wvI v∗ I Ω = ¯vJ ¯v∗ J Ω I Ω for I, J < ∞. Since such vectors are total, we have wS = F w on the domain of S. Thus wSw∗ = F on the domain of F . We write S = J ∆ 2 as the unique polar 2 w∗ = J ∆− 1 decomposition. Then F = S∗ = ∆ 2 . 2 w∗ = ∆− 1 By the uniqueness of polar decomposition we get wJ w∗ = J and w∆ 2 . Same algebra is valid in case ζ 6= 1 if we reset the notations vk on the right hand side absorbing the phase factor. In the following we repeat it for completeness of the proof as the phase factor could be delicate. 2 . Hence wJ w∗w∆ 2 J = J ∆− 1 1 1 1 1 wζ SvI v∗ J Ω = wζ vJ v∗ = ζI−J¯vJ ¯v∗ = F ζ−I+J¯vI ¯v∗ ζ Ω I Ω = wζ vJ v∗ I Ω = ζI−JF ¯vI ¯v∗ I w∗ J Ω J Ω for I, J < ∞ = F wζ vI v∗ J Ω 38 ANILESH MOHARI for all I, J < ∞. Now we are going to show that wζ is self-adjoint if and only if ζ = 1. Note ζ and thus applying β ¯ζ on both side of the following ζ ) = wζ β¯z(x)w∗ that βz(wζ xw∗ identity (26) for all 1 ≤ k ≤ d, we also get wζ βζ(vk)w∗ wζJ βζ(vk)J w∗ ζ = J βζ(vk)J ζ = J β2 ζ = β ¯ζ 2(vk) as J commutes with wζ . wζvkw∗ ¯ζ (vk)J and thus w2 ζ vk(w∗ ζ )2 = ζ2 = 1 if and only if ζ = 1 ( as ζ = 1 or exp ζ ∈ M′ and further as wζ commutes with J , w2 n where n ≥ 2 ). In such a case we get w2 ζ ∈ M. ω being an extremal element in Kω′ we have M ∨ M = B(K) by Proposition 3.5 and as M ⊆ M′, we get that M is a factor. Thus for a factor M, w2 ζ is a scaler. Since wζΩ = Ω we get w2 ζ = wζ . This completes the proof. ζ = 1 i.e. w∗ iπ A state ω on ⊗ZZ Md is said be in detailed balance if ω is both lattice symmetric and real. In the following proposition as before we identified once more S1/H ≡ S1 in case H 6= S1 and set ζ be the least value in S1 H ≡ S1 so that ζ2 ∈ H. Theorem 4.4. Let ω be a translation invariant factor state on B = ⊗ZZMd. Then the following are equivalent: (a) ω is real and lattice symmetric; (b) There exists an extremal element ψ ∈ Kω′ so that ψ = ψβζ and ¯ψ = ψβζ, where ζ is either 1 or exp n . iπ Furthermore if ω is a pure state then the following holds: (c) There exists a Popescu elements (K, vk, 1 ≤ k ≤ d, Ω) for ω with relation vk = Jv vkJv for all 1 ≤ k ≤ d, where Jv = vJ and v is a self-adjoint unitary operator on K commuting with modular operators ∆ 2 and conjugate operator J associated with cyclic and separating vector Ω for M. Further βz(v) = v for all z ∈ H and H ⊆ {1, −1}; (d) The map Jv : H ⊗K H → H ⊗K H defined by J ′ )Ω → π(sI ′ s∗ J ′ sI s∗ J sI ′ s∗ π(sI s∗ J )Ω, 1 I, J, I ′, J ′ < ∞ extends the map Jv : K → K to an anti-unitary map so that Jvπ(si)Jv = ¯π(si) for all 1 ≤ i ≤ d where ¯π is the conjugate linear extension of π from the generating set (si), i.e. ¯π(sI s∗ J ) for I, J < ∞ and then extend it anti-linearly for its linear combinations. J ) = π(sI s∗ PROOF: Since ω is lattice symmetric, by Proposition 4.2 ψ = ψβζ for all ψ ∈ Kω′ where ζ is fixed number either 1 or exp iπ n for some n ≥ 1. Now we use real property of ω and choose by Proposition 4.3 an extremal element ψ ∈ Kω′ so that ¯ψ = ψβζ. This proves that (a) implies (b). That (b) implies (a) is obvious. Now we aim to prove the last statements which is the main point of the propo- sition. For simplicity of notation we consider the case ζ = 1 and leave it to reader to check that a little modification needed to include the case ζ 6= 1 and all the algebra stays valid if vk is replaced by βζ(vk). We consider the Popescu system (K, M, vk, 1 ≤ k ≤ d, Ω) as in Proposition 2.4 associated with ψ. Thus by Propo- sition 4.3 and Proposition 4.4 there exists unitary operators uζ, wζ on K so that uζvku∗ ζ = β ¯ζ(vk) TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 39 where uζJ u∗ (27) uζwζ vkw∗ ζ = J , wζ J w∗ ζ u∗ ζ = uζJ βζ (vk)J u∗ ζ = J βζ(uζvku∗ ζ)J = J βζ(β ¯ζ(vk)J = J vkJ wζ vkw∗ ζ = β ¯ζ(¯vk) = J βζ(vk)J ζ = wζ ∆ ζ = J and uζ∆ 2 u∗ 1 1 2 w∗ ζ = ∆− 1 2 . Thus We also compute that wζuζvku∗ (28) ζw∗ ζ = wζ β ¯ζ(vk)w∗ ζ = J βζβ ¯ζ(vk)J = J vkJ By Proposition 3.2, for a factor state ω we also have M ∨ M = B(K). As M ⊆ M′, in particular we note that M is a factor. So u∗ ζ uζwζ ∈ M′ commuting also with J and thus a scaler as M is a factor. As uζΩ = wζΩ = Ω, we conclude that uζ commutes with wζ . ζw∗ Now we set vζ = uζwζ which is an unitary operator commuting with both J 2 uζwζ . 2 . That vζ commuting with ∆ 2 follows as uζwζ∆ 2 = uζ∆− 1 2 wζ = ∆ 1 1 1 1 and ∆ Next claim that we make now that vζ is a self-adjoint element. To that end note ζ ⊆ M′. We check the kJ Ω = J vkJ Ω and thus that the relations (28) says that vζMv∗ following identity: vζ v∗ ζ Ω = vζ v∗ separating property we deduce that ζ ⊆ M and so vζM′v∗ kΩ = vζv∗ ζ Ω = J v∗ kv∗ kv∗ vζ v∗ kv∗ for all 1 ≤ k ≤ d. So we conclude that v2 ζ is an element in the centre of M. The centre of M being trivial as ω is a factor state ( here we have more namely pure ) and vζ Ω = Ω, we conclude that v2 ζ is the unit operator. Hence vζ is a self-adjoint element. ζ = J vkJ ζ ∈ M′ and as vζ commutes with J , v2 For simplicity of notation we set v for vζ. βz(v) = v for all z ∈ H is equivalent J J J Ω. βz being an automor- 2 commutes with J J Ω : I − J = J Ω : I − J = k] = P Fk we get vFkP = J FkP = FkP as J to the property that v keeps the subspaces P Fk invariant. Since vvI v∗ and v is self-adjoint and vΩ = Ω we get vvI v∗ phism on M preserving the state φ0, modular elements J , ∆ and in particular J commutes with FkP for all k ∈ H. Since [vI v∗ k] = P Fk and [vI v∗ commutes with Fk. J v∗ = J vv∗ J Ω = J vv∗ 1 Fix any z ∈ H. By taking action of βz on both side of the relation vvkv∗ = J vkJ , we have vvkv∗ = ¯z2J vkJ = ¯z2vvkv∗. Thus z2vk = vk for all 1 ≤ k ≤ d. Since k = 1, we have z2 = 1. Pk vkv∗ The last statement (d) follows by a routine calculation as shown below for a special vectors. ( as JvviJv = vi) (Jv being anti-unitary ) < Ω, π(sI s∗ =< Ω, vI v∗ =< Ω, Jv vI v∗ J sI ′ s∗ J ′ Ω > J vI ′ vJ ′ Ω > J vI ′v∗ J ′ JvΩ > =< vI v∗ J vI ′ v∗ J ′ Ω, Ω > =< π(sI ′ s∗ J ′ sI s∗ J )Ω, Ω > For anti-unitary relation involving more general vectors, we use Cuntz relations and the above special cases. The statement is obvious as Jv is anti-linear. This completes the proof. 40 ANILESH MOHARI We set an anti-linear ∗-automorphism Jv : Od ⊗ Od → Od ⊗ Od defined by J for I, J, I ′, J ′ < ∞ by extending anti-linearly. J ′) = sI ′ s∗ J ⊗ sI ′ s∗ J ′ ⊗ sI s∗ Jv(sI s∗ We say a state ψ on Od ⊗ Od is reflection positive if ψ(Jv(x)x) ≥ 0 for all x ∈ Od and equality holds if and only if x = 0. Similarly for a state ω on B we define reflection positivity. Note that this notion extended to Od ⊗Od is an abstract version of the concept "reflection positivity" of a state on B introduced in [FILS] for any involution (linear or conjugate linear ) taking element from future algebra to past algebra. However this notion is different from Lieb's spin flip reflection symmetric [FILS]. This hidden symmetry v will play an important role to determine properties of ω (Section 5). Now we aim to make v little more general primarily motivated with reflection symmetry with a twist introduced in [FILS]. To that end we fix any g0 ∈ Ud(C) so that g2 0 = 1 and βg0 is the natural action on Od and Od. We say ω is lattice reflection symmetric with twist g0 if ω(βg0 (r(x)) = ω(x) for all x ∈ B where r is the refection automorphism around − 1 2 . So when g0 = 1 we get back to our notion of lattice reflection symmetric. We fix now such a lattice reflection g0-twisted factor state ω. Since βg0βz = βzβg0 for all z ∈ S1, by going along the same line as in Proposition 4.2, any extremal element in ψ in Kω will admit ψg0 = ψ ◦ ζ where ζ = 1 or ζ = exp πi n where H = {z ∈ S1 : zn = 1} and ψg0 = ψβg0 . Thus we can follow the same steps that of Proposition 4.4 to have a modified statements in the proof of Proposition 4.4 with vk replaced by βg0 (vk) for such a pure real state i.e. there exists unitary operators uζ, wζ on K so that uζβg0 (vk)u∗ ζ = β ¯ζ(vk) wζ vkw∗ ζ = β ¯ζ(¯vk) = J βζ(vk)J where uζJ u∗ ζ = J , wζ J w∗ ζ = J and uζ∆ 1 2 u∗ ζ = wζ ∆ 1 2 w∗ ζ = ∆− 1 2 . Thus (29) uζwζvkw∗ ζ u∗ ζ = uζJ βζ(vk)J u∗ ζ = J βζ(uζvku∗ ζ)J = J βζ(βg0 (β ¯ζ (vk)))J = J βg0 (vk)J We also compute that (30) wζuζvku∗ ζw∗ ζ = wζ β ¯ζ(βg0 (vk))w∗ Thus taking vg0 = wζuζ, as g2 ζ = β ¯ζ(βg0 (wζ vkw∗ ζ )) = J βζβ ¯g0 (β ¯ζ(vk))J = J β ¯g0 (vk)J 0 = g0 we also have (31) vg0 βg0 (vk)v∗ g0 = J vkJ for all 1 ≤ k ≤ d where vg0 is an unitary operator commuting with ∆ 2 and J . Unlike the twist free case, self-adjoint property of vg0 is not guaranteed in general. In fact we get from the following computation 1 g0 vk(v∗ v2 g0 )2 = vg0 J β ¯g0(vk)J v∗ g0 = J β ¯g0 (J β ¯g0 (vk)J )J = β ¯g0g0 (vk) Thus vg0 is self adjoint if and only if g0 = ¯g0 as g2 0 = 1. ¯βg(r(x))x) ≥ 0 for all x ∈ BR where Such a ω is called reflection positive if ω( r is the reflection around the lattice point 1 2 so that r(BR) = BL and ¯x stands for complex conjugation i.e. ¯x = J0xJ0 where J0 : (z1, ..zd) = (¯z1, ..¯zd) with respect to a basis. Such an involution are included within the abstract framework of positive reflection symmetric with twist introduced in [FILS]. TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 41 Theorem 4.5. Let ω be a translation invariant, reflection symmetric with twist g0, pure state on B and ψ be an extremal point Kω′ and π as described as in Theorem 4.4. Then the following statements are true: (a) ψ is reflection positive with twist g0 on π( Od ⊗Od) if and only if vg0 in Theorem 3.10 is equal to 1 i.e. we have J vkJ = βg0 (vk) for all 1 ≤ k ≤ d; (b) ψ is reflection positive with twist on B if and only if J vI v∗ J ) for all I = J < ∞; In such a case vg0 commutes with P0 and vg0 P0 = P0 and τ (y) = J τ (J yJ )J for y ∈ M′ 0 ⊆ B(K0) where we recall P0 = [M0Ω] and K0 is the Hilbert subspace for P0K. (c) ∆ = I if and only if vg0 βg0 (vk)v∗ k, 1 ≤ k ≤ d. In such a case H is trivial and M is finite type-I and spacial correlation functions of ω decays exponentially. Further if ω is reflection positive with twist g0, then vg0 = 1. J J = βg0 (vI v∗ g0 = v∗ PROOF: We will prove for g0 = 1 as a proof for g0 6= 1 needs no difference except involving a twist action ¯g0 on BL to accomodate conjugate linearity on Popescu elements of the map Jv. We recall from Theorem 3.8 that P π(Od)′′P = M and P π( Od)′′P = M ⊆ M′ ( we do not need equality here ) and P = E E. Thus for any x ∈ Od we may write ω(Jv(x)x) =< Ω, ¯π(Jv(x))π(x)Ω > =< Ω, P π(Jv(x))P π(x)P Ω >=< Ω, JvP π(x)P JvP π(x)P Ω > where we have used equality ¯π(Jv(x)) = Jvπ(x)Jv from Theorem 3.10. If v = 1 i.e. Jv = J we have ω(J xJ x) ≥ 0 by the self-dual property of Tomita's positive cone {J aJ aΩ : a ∈ M} [BR1] and being a pointed cone equality holds if and only x = 0. Thus ω is a reflection positive map on π( Od ⊗ Od). Conversely if ω is reflection positive on π( Od⊗Od)′′ we have < Ω, yJvyJvΩ >≥ 0 2 we may rewrite 2 is a non-negative operator. Since 2 , we conclude that v is a where y ∈ M = P π(Od)′′P . Since v commutes with J and ∆ < Ω, yvJ yΩ >=< y∗Ω, v∆ ∆− 1 non-negative operator. v being unitary we conclude that v = 1. 2 is also a non-negative operator commuting with v∆ 2 y∗Ω >≥ 0 i.e. v∆ 1 1 1 1 Proof for (b) follows the same route that of (a) replacing the role of M and M by M0 and M0 respectively. We will deal with the non-trivial part of (c). Assume vg0 βg0 (vk)v∗ 1 ≤ k ≤ d. So ∆ is affiliated to M′. As J ∆ 1 Hence ∆ = I as M is a factor and ∆Ω = Ω. 2 J = ∆− 1 k for all 2 , ∆ is also affiliated to M. g0 = v∗ In general ω being a pure state M is either a type-I or type-III factor [Mo3, Theorem 3.4]. Thus we conclude that M is a finite type-I factor if ∆ = 1 ( i.e. φ0 is a tracial state on M ). This completes the proof of the first part of (c). The last part of (c) is rather elementary. We note that purity of ω ensures that the point spectrum of the self-adjoint contractive operator T , defined by T xΩ = τ (x)Ω on the KMS Hilbert space, in the unit circle is trivial i.e. {z ∈ S1 : T f = zf for some non zero f ∈ K} is the trivial set {1} ( as an consequence of strong mixing property ). Thus T being a contractive matrix on a finite dimensional Hilbert space, the spectral radius of T − Ω >< Ω is α for some α < 1. Now we use Proposition 3.1 for any Xl ∈ BL and Xr ∈ BR to verify the following eδkω(Xlθk(Xr)) − ω(Xl)ω(Xr) = eδkφ0(JvxlJvτk(xr)) − φ0(xl)φ0(xr) → 0 42 ANILESH MOHARI as k → ∞ for any δ > 0 so that eδα < 1 where J xlJ = P XlP and xr = P XrP for some xl, xr ∈ M. As α < 1 such a δ > 0 exists. This completes the proof for (c) as last statement follows from (b). 5. Translation invariant twisted reflection positive pure state and it's split property: Let ω be a translation invariant real lattice symmetric with a twist g0 pure state on B as in Theorem 4.5. We fix an extremal element ψ ∈ Kω′ so that ¯ψ = ψ = ψβζ and consider the Popescu elements (K, M, vi, Ω) as in Theorem 4.5. P being the support projection of a factor state ψ we have M = P π(Od)′′P = {vk, v∗ k : 1 ≤ k ≤ d}′′ ( Proposition 2.4 ). So the dual Popescu elements (K, M′, vk, 1 ≤ k ≤ d, Ω) satisfy the relation vvkv = J βg0 (vk)J ( recall that the factor ζ won't show up as two symmetry will kill each other as given in Theorem 4.5 ) for all 1 ≤ k ≤ d. We quickly recall as M0 is the {βz : z ∈ H} invariant elements of M(= P π(Od)′′P ), the norm one projection x → Rz∈H βz(x)dz from M onto M0 preserves restriction of the completely positive map τ (x) = Pk vkxv∗ map on M0. Hence the completely positive map τ (x) = Pk vkxv∗ the faithful normal state φ0. So by Takesaki's theorem modular group associated with φ0 preserves B0. Further since βz(τ (x)) = τ (βz(x)) for all x ∈ M, the k to M0 is a well defined k on M0 is also KMS symmetric modulo an unitaryconjugation by v i.e. << x, τ (y) >>=<< τv(x), y >> where x, y ∈ M0 and << x, y >>= φ0(x∗σ i (y)) and (σt) is the modular auto- morphism group on M0 associated with φ0 and [M0Ω] = P0 where P0 = P F0 and τv(x) = v∗τ (vxv∗)v for all x ∈ M0. Thus τv = τ if and only if ω is reflection positive on B with twist g0 (Theorem 4.5). 2 However the inclusion M0 ⊆ M need not be an equality in general unless H is trivial. The unique ground state of XY model in absence of magnetic field give rise to a non-split translation invariant real lattice symmetric pure state ω and further H = {1, −1} (see next section). We now fix a translation invariant real lattice symmetric pure state ω which is also reflection positive with a twist g0 on B and explore KMS-symmetric property of (M0, τ, φ0) and the extended Tomita's conjugation operator J on H⊗K H defined in Theorem 4.5 to study the relation between split property and exponential decaying property of spacial correlation functions of ω. For any fix n ≥ 1 let Q ∈ π(B[−k+1,k]). We write Q = XI=J=I ′=J ′=n q(I ′, J ′I, J)βg0 ( SI ′ S∗ J ′ )SI S∗ J and q be the matrix q = ((q(I ′, J ′I, J))) of order d2n × d2n. Proposition 5.1. The matrix norm of q is equal to operator norm of Q in π(B[−n+1,n]). PROOF: C∗ completions of π( UHFd ⊗ UHFd) is isomorphic to B. Thus we note that the operator norm of Q is equal to the matrix norm of q where q = ((q(I ′, IJ ′, J))) is a d2n × d2n matrix with q(I ′, IJ ′, J) = q(I ′, J ′I, J). However TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 43 the map L(q) = q is linear and identity preserving. Moreover L2(q) = q. Thus L = 1. Hence q = q. This completes the proof. Proposition 5.2. Let ω be a translation invariant real lattice symmetric pure state on UHFd ⊗ZZMd . Then there exists an extremal point ψ ∈ Kω′ so that ψβζ = ψ = ¯ψ where ζ ∈ {1, exp iπ 2 } and the associated Popescu systems (H, Sk, 1 ≤ k ≤ d, Ω) and (H, Sk, 1 ≤ k ≤ d, Ω) described in Proposition 3.2 and Theorem 4.5 satisfies the following: (a) For any n ≥ 1 and Q ∈ π(B[−n+1,n]) we write Q = XI ′=J ′=I=J=n q(I ′, J ′I, J)βg0 ( S∗ I ′ S∗ J ′)S∗ I SJ and set a notation for simplicity as θk(Q) = XI=J=I ′=J ′=n q(I ′, J ′I, J)βg0 ( SI ′ S∗ J ′)Λ2k(SI S∗ J ). Then θk(Q) ∈ B(−∞,−k] S[k+1,∞). (b) Q = Jg0 QJg0 if and only if q(I ′, J ′I, J) = q(I, JI ′, J ′); (c) If the matrix q = ((q(I ′, J ′I, J))) is non-negative then there exists a matrix b = ((b(I ′, J ′I, J))) so that q = b∗b and then q = P QP = XK=K ′=n JvxK,K ′ JvxK,K ′ J ∈ M0 (d) In such a case i.e. if Q = J QJ the following holds: where xK,K ′ = PI,J: I=J=n b(K, K ′I, J)vI v∗ (i) ω(Q) = PK=K ′=n φ0(JvxK,K ′ JvxK,K ′) (ii) ω(θ2k(Q)) = PK=K ′=n φ0(JvxK,K ′ Jvτ2k(xK,K ′ )). PROOF: Since the elements βg0 ( SI ′ S∗ I SJ : I = J = I ′ = J ′ = n form an linear independent basis for π(B[−n+1,n]), (a) follows. (b) is also a sim- ple consequence of linear independence of the basis elements and the relation J βg0( SI ′ S∗ J ) as described in Theorem 4.5. J ′ )SI S∗ J J = SI ′S∗ J ′ βg0 ( SI S∗ J ′)S∗ For (c) we write Q = XK=K ′=n J βg0 (QK,K ′)J QK,K ′ where QK.K ′ = PI,J: I=J=n b(K, K ′I, J)SI S∗ J . ω being pure we have ( The- orem 3.6) P = E E where E and E are support projection of ψ in π(Od)′′ and π( Od)′′ respectively. So for any X ∈ π(Od)′′ and Y ∈ π( Od)′′ we have P XY P = EEXY EE = EEY E EX EE = P XP Y P . Thus (c) follows as ω(Q) = φ0(q) by Proposition 3.1 (b) and Theorem 4.5 as ω is reflection symmetry with twist g0. For (d) we use (a) and (c). This completes the proof. Proposition 5.3. Let ω, a translation invariant pure state on B, be in detailed balance and reflection positive with a twist g0. Then the following are equivalent: (a) ω is decaying exponentially. (b) The spectrum of T − Ω >< Ω is a subset of [−α, α] for some 0 ≤ α < 1 where T is the self-adjoint contractive operator defined by T xΩ = τ (x)Ω, x ∈ M0 44 ANILESH MOHARI on the KMS-Hilbert space << x, y >>= φ0(x∗σ i 2 (y) >>. PROOF: Since T kxΩ = τk(x)Ω for x ∈ M0 and for any L ∈ BL and R ∈ BR we have ω(Lθk(R)) = φ0(J yJ τk(x)) =<< y, T kx >> where x = P π(R)P and y = J P π(L)P J are elements in M0. Since P π(BR)′′P = M0 and P π(BL)′′P = M0 = M′ 0 as M = M′ by Theorem 4.4, we conclude that (a) holds if and only if ekδ < f, T kg > − < f, Ω >< Ω, g > → 0 as k → ∞ for any vectors f, g in a dense subset D of the KMS Hilbert space. That (b) implies (a) is now obvious since ekδαk = (eδα)k → 0 whenever we choose a δ > 0 so that eδα < 1 where α < 1. For the converse suppose that (a) holds and T 2 − Ω >< Ω is not bounded away from 1. Since T 2 − Ω >< Ω is a positive self-adjoint contractive operator, for each n ≥ 1, we find an unit vector fn in the Hilbert space so that E[1−1/n,1]fn = fn and fn ∈ D, where s → E[s,1] is the spectral family of the positive self-adjoint operator T 2 − Ω >< Ω and in order to ensure fn ∈ D we also note that E[s,1]D = {Es,1]f : f ∈ D} is dense in E[s,1] for any 0 ≤ s ≤ 1. Thus by exponential decay there exists a δ > 0 so that e2kδ(1 − 1 n )k ≤ e2kδZ[0,1] sk < fn, dEsfn >= e2kδ < fn, [T 2k − Ω >< Ω]fn >→ 0 as k → ∞ for each n ≥ 1. Hence e2δ(1 − 1 e2δ ≤ 1. This contradicts that δ > 0. This completes the proof. n ) < 1. Since n is any integer, we have For the simplicity of notation we take in the following g0 = 1. Now we are set to state our main result in this section. For any Q ∈ π(B) we set J (Q) = J QJ . Recall that J 2 = I. Any element Q = 1 2 (Q − J (Q)) is a sum of an even element in {Q : J (Q) = Q} and an odd element in {Q : J (Q) = −Q}. Moreover iQ is an even element if Q is an odd element. Also note that Qeven ≤ Q and Qodd ≤ Q. Hence it is enough if we verify (1) for all even elements for split property. We fix any n ≥ 1 and an even element Q ∈ B[−n+1,n]. We I SJ . The matrix q = (q(I ′, J ′I, J) is symmetric and thus q = q+ − q− where q+ and q− are the unique non-negative matrix contributing it's positive and negative parts of q. Hence q+ ≤ q and q− ≤ q. We set a notation for simplicity that write as in Proposition 5.2 Q = PI ′=J ′=I=J=n q(I ′, J ′I, J) S∗ 2 (Q + J (Q)) + 1 I ′ SJ ′ S∗ θk(Q) = XI=J=I ′=J ′=n q(J ′, I ′I, J)Λk( SI ′ S∗ J ′ )Λk(SI S∗ J ) which is an element in B(−∞,−k] S[k,∞) and by Proposition 5.2 (d) ω(θk(Q)) = XK=K ′=n φ0(J xK,K ′ J τ2k(xK,K ′ )) provided q = (q(I ′, J ′I, J) is positive, where P QP = PK=K ′=n J xK,K ′ J xK,K ′ J and q = b∗b. Thus in such a case we have by and xK,K ′ = PI,J b(K, K ′I, J)vI v∗ Proposition 5.2 (d) that ω(θk(Q)) − ωL ⊗ ωR(θk(Q)) = XK=K ′=n φ0(J xK,K ′ J (τ2k − φ0)(xK,K ′ )) = XK=K ′=n << xK,K ′ , (T − Ω >< Ω)2kxK,K ′ >> TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 45 ≤ α2k XK=K ′=n << xK,K ′ , xK,K ′ >> provided T − Ω >< Ω ≤ α and so In the last identity we have used Proposition 5.1. ≤ α2kω(Q) ≤ α2kq = α2kq Hence for an arbitrary Q for which J (Q) = Q we have ω(θk(Q)) − ωL ⊗ ωR(θk(Q)) ≤ α2k(q+ + q−) ≤ 2α2kq = 2α2kQ where in the last identity we have used once more Proposition 5.1. Thus we have arrived at our main result by a well know criteria [BR1] on split property. Theorem 5.4. Let ω be a translation invariant pure state. Let ω be also real ( with respect to a basis for Cd ) and lattice symmetric with twist g0. If ω is reflection positive with twist a g0 ( g0 ∈ Ud(C), g2 0 = 1) and the spatial correlation function of ω decays exponentially then ω is split i.e. πω(BR)′′ is a type-I factor. 6. Spontaneous symmetry breaking in quantum spin chain We aim to investigate the methodology developed in section 3 to study properties of the ground states of a translation invariant Hamiltonian for one lattice dimensional quantum spin chain B = ⊗ZZMd. Let G be a compact group and g → v(g) be a d−dimensional unitary representa- tion of G. By γg we denote the product action of G on the infinite tensor product B induced by v(g), γg(Q) = (.. ⊗ v(g) ⊗ v(g) ⊗ v(g)...)Q(... ⊗ v(g)∗ ⊗ v(g)∗ ⊗ v(g)∗...) for any Q ∈ B. We recall now that the canonical action of the group U (d) of d × d matrices on Od is given by βv(g)(sj) = X1≤i≤d siv(g)i j and thus βv(g)(s∗ j ) = X1≤i≤d ¯v(g)i js∗ i Note that v(g)ei >< ejv(g)∗ = v(g)ei >< v(g)ej = Pk,l v(g)l k j el >< ek, where e1, .., ed are the standard basis for ICd. Identifying ei >< ej with sis∗ j we verify that on BR the gauge action βv(g) of the Cuntz algebra Od and γg coincide i.e. γg(Q) = βv(g)(Q) for all Q ∈ BR. ¯v(g) i Proposition 6.1. Let ω be a translation invariant factor state on B. Suppose that ω is G−invariant, ω(γg(Q)) = ω(Q) for all g ∈ G and any Q ∈ B. Let ψ be an extremal point in Kω′ and (K, M, vk, 1 ≤ k ≤ d, φ0) be the Popescu system associated with (H, Si = π(si), Ω) described as in Proposition 2.4. Then we have the following: (a) There exists an unitaryrepresentation g → U (g) in B(H) and a representation g → ζ(g) so that (32) U (g)SiU (g)∗ = ζ(g)βv(g)(Si), 1 ≤ i ≤ d 46 ANILESH MOHARI for all g ∈ G and (b) There exists an unitaryrepresentation g → u(g) in B(K) so that u(g)Mu(g)∗ = M for all g ∈ G and φ0(u(g)xu(g)∗) = φ0(x) for all x ∈ M. Furthermore the operator V ∗ = (v∗ d)tr : K → ICd ⊗ K is an isometry which intertwines the representation of G, 1 , .., v∗ (33) (ζ(g)v(g) ⊗ u(g))V ∗ = V ∗u(g) for all g ∈ G, where g → ζ(g) is the representation of G in U (1). (c) J u(g)J = u(g) and ∆itu(g)∆−it = u(g) for all g ∈ G. PROOF: ω being a factor state by Proposition 3.3, H is a closed subgroup of S1. Thus H is either S1 or a finite cyclic subgroup. We also recall that λβg = βgλ for all g ∈ G and ω being G-invariant we have ψβg ∈ Kω′ for all ψ ∈ Kω′ and g ∈ G. In case H = S1, by Proposition 2.6 Kω′ is having a unique element and thus by our starting remark we have ψβg = ψ for the unique extremal element ψ ∈ Kω′. In such a case we define unitary operator U (g)π(x)Ω = π(βg(x))Ω and verify (a) with ζ(g) = 1 for all g ∈ G. Now we are left to deal with the more delicate case. Let H = {z : zn = 1}′′ In such a case by Proposition 2.5 and Proposition 3.2 (a) we d )′′ = π( UHFd)′′. Thus for any 0 ≤ k ≤ KΩ : I ′ = for some n ≥ 1. have π(OH n − 1 orthogonal projection Fk is spanned by the vectors { SI ′ S∗ J ′, I = J, and K = k}. We set unitary operator U (g)′ on Fk : k ≥ 0 by d )′′ = π(UHFd)′′ and π( OH J ′SI S∗ J S∗ U (g)′π(sI ′ s∗ J ′ sI s∗ J s∗ K)Ω = π(βv(g)(sI ′ s∗ J ′ sI s∗ J s∗ K))Ω where I ′ = J ′, I = J and K = k. It is a routine work to check that U (g)′ is indeed an inner product preserving map on the total vectors in Ek. The family {Fk : 0 ≤ k ≤ n − 1} being an orthogonal family projection with Pk Fk = I, U (g)′ extends uniquely to an unitary operator on H ⊗K H. It is obvious by our construction that g → U (g)′ is a representation of G in H ⊗K H. For each g ∈ G the Popescu element (H, βv(g)(Sk), 1 ≤ k ≤ d, Ω) determines an extremal point ψg ∈ Kω′ and thus by Proposition 2.6 there exists a complex number ζ(g) with modulus 1 so that ψg = ψβζ(g). Note that for another such a choice ζ′(g), we have ¯ζ(g)ζ(g)′ ∈ H. As H is a finite cyclic subgroup of S1, we have a unique choice once we take ζ(g) to be an element in the group S1/H which we identify with S1. That g → ζ(g) is a representation of G in S1 = {z ∈ IC : z = 1} follows as the choice in S1/H of ζ(g) is unique. Hence there exists an unitary operator U (g) and a representation g → ζ(g) in S1 so that U (g)Ω = Ω, U (g)SiU (g)∗ = ζ(g)βv(g)(Si) for all 1 ≤ i ≤ d. Thus U (g) = Pk ζ(g)kU (g)′Ek for all g ∈ G as their actions on any typical vector SI S∗ J SKΩ, I = J, K = k < ∞ are same. Both g → U ′(g) and g → ζ(g) being representations of G, we conclude that g → U (g) is an unitary representation of G. The above covariance relation ensures that U (g)π(Od)′′U (g)∗ = π(Od)′′ for all g ∈ G and thus also U (g)π(Od)′U (g)∗ = π(Od)′ for all g ∈ G. Now it is also routine work to check that U (g)F U (g)∗ = F , where we call F = {π(Od)′′Ω] and U (g)EU (g)∗ = E where E = [π(Od)′Ω] is the support projection of the state ψ in π(Od)′′. Hence the support projection P = EF of the state ψ in the cyclic TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 47 subspace F is also G invariant i.e. U (g)P U (g)∗ = P for all g ∈ G. Thus we define g → u(g) = P U (g)P an unitary representation of g in K. Hence we have (34) u(g)vju(g)∗ = ζ(g)βv(g)(vj) = ζ(g)viv(g)i j for all 1 ≤ k ≤ d. By taking adjoint we get u(g)v∗ 1 ≤ j ≤ d. j u(g)∗ = ¯ζ(g) ¯v(g)i jv∗ i for all We are now left to prove (c). To that end we first verify that S0u(g) = u(g)S0 as J Ω are same, where S0xΩ = x∗Ω for x ∈ M. their actions on any typical vector vI v∗ Hence by uniqueness of the polar decomposition we conclude that (c) holds. Proposition 6.2. Let ω be a G-invariant translation invariant factor state on B as in Proposition 4.1 and ω be also pure real (with respect to an orthonormal basis for Cd ) and lattice symmetric with a twist g0. We fix an extremal element ψ ∈ Kω so that ψg0 = ¯ψ = ψβζ where ζ = 1 or exp iπ n as in Theorem 4.5. Let the associated family {vk : 1 ≤ k ≤ d} of operators defined in Theorem 3.10 be linearly independent (i.e. Pk ckvk = 0 if and only if ck = 0 for all 1 ≤ k ≤ d ). Then g0 intertwines the representation g → ζ(g)v(g) with it's complex conjugate matrix representation with respect to the orthonormal basis (ei) in ICd if and only if vg0 commutes with u(g) for all g ∈ G; PROOF: For simplicity of notation in the proof we use v for vg0 . For (a) we fix an extremal element ψ ∈ Kω′ as described in Theorem 4.5 and consider the Popescu elements {vj, 1 ≤ j ≤ d} satisfying for all 1 ≤ j ≤ d. We set temporary notation vβg0 (vj)v∗ = J vjJ vg = u(g)vu(g)∗ i j = vi and v(g) for complex conjugation matrix of v(g) i.e. v(g) the following simple steps. j(g). We compute u(g)vβg0 (vj)v∗u(g)∗ = J u(g)vju(g)∗J = ζ(g)βv(g)(J vj J ) and so i.e. vgζ(g)βv(g)(βg0 (vj ))(vg)∗ = u(g)vu(g)∗u(g)βg0 (vj )u(g)∗u(g)v∗u(g)∗ = ζ(g)βv(g)(J vjJ ) = ζ(g)βv(g)(vβg0 (vj )v∗) v∗vgζ(g)βv(g)(βg0 (vj))(vg)∗v = ζ(g)βv(g)(βg0 (vj)) Explicitly if v commutes with u(g), by the covariance relation we have [ζ(g)vi j (g) − ζ(g)vi j (g)]vj = 0 X1≤j≤d for all 1 ≤ i ≤ d. By linear independence we conclude that there exists an or- thonormal basis for ICd so that the matrix representation (ζ(g)vi j (g)) of ζ(g)v(g) with respect to the basis having real entries. Conversely we have v∗vg commuting with all vi i.e. v∗vg ∈ M′. But v, u(g) and so v∗vg commutes with J and thus v∗vg ∈ M and as v∗vgΩ = Ω we have v∗vg = 1 by factor property of M. For 48 ANILESH MOHARI g0 6= 1 we get the relation g0v(g)g∗ 0 = v(g) where conjugation with respect to the basis (ei) as a necessary and sufficient condition for commuting property of u(g) and v. We are left to deal with another class of example. Let ω be a translation invariant pure state on B = ⊗ZZ M2. If ω is G = U (1) ⊆ SU (2) invariant then by a Theorem [Ma2] ω is either a product state or a non-split state. The following theorem says more when ω is also real and lattice symmetric. The following theorem is originated from reviewer's remark on an earlier version. Theorem 6.3. Let ω be a translation invariant pure state on B = ⊗ZZ M2 which also be reflection symmetric with a twist g0 and real with respect to a basis (ei). If ω is also U (1) ⊆ SU (2) invariant and reflection positive on B with twist then following holds: (a) ω is a non-split state if and only if H = {1, −1}; (b) ω is a split state ( hence product state ) if and only if H = {1}. PROOF: Assume for the time being that {v0, v1} are linearly independent. In such a case by Proposition 6.1 there exists a representation z → uz in K of U (1) so that uzv0u∗ z = ζ(z)zv1 where z → ζ(z) = zk for some k ∈ ZZ. ω being lattice symmetric and real with respect to a basis, by Theorem 5.4 we also have H ⊆ {1, −1}. z = ζ(z)zv0 and uzv1u∗ j (z))−1((vi j(z))) = ((z−kvi Once we also assume ω to reflection positive with twist, z → ζ(z)v(z) has a real representation of U (1) ⊆ SU (2) in a basis and v commutes with uz for all z ∈ H j(z))) for all z ∈ S1. Hence z2kI = by Proposition 5.2. Thus ((zkvi ((vi j (z))) where the right hand side is a multiple of two elements in SU (2). By taking determinants of both the side we conclude that z4k = 1 for all z ∈ S1. Hence k = 0. Now by U (1) ⊆ SU (2) invariance we also check that ψ(sI s∗ J ) = 0 if I + J is not a multiple of 2. Further as ψ(βz(sI s∗ J )) = zI−Jψ(sI s∗ J ) and I − J = I + J − 2J is a multiple of 2 whenever I + J is an even number, we check that −1 ∈ H. Thus we conclude that {−1, 1} ⊆ H whenever {v0, v1} are linearly independent. In such a case ω is a non-split state. 0 = 1. Thus viv∗ Now we consider the case where {v0, v1} are linearly dependent. We assume without loss of generality that v0 6= 0 and v1 = αv0. v0v∗ 1 = 1 ensures that (1 + α2)v0v∗ j ) for all 0 ≤ i, j ≤ 1. Hence ω(ei1 >< ej1 ⊗ ... ⊗ ein >< ejn) = φ0(vI v∗ jn ). This clearly shows that ω is a product state. ω being pure, we conclude that there is an extremal point ψ so that the associated Popescu elements are given by v0 = 1 and v1 = 0. In such a case ω is a split state and H = {1}. J ) = φ0(vi1 v∗ j = φ0(viv∗ 0 + v1v∗ j1 )...φ0(vin v∗ Thus we can sum up from the above argument that such a state ω is split ( non-split ) if and only if H = {1} (H = {1, −1}). Theorem 6.4. Let G be a simply connected compact Lie group and g → v(g) be an irreducible representation on ICd so that the invariance vectors of ICd ⊗ ICd with respect to the representation g → v(g) ⊗ v(g) is one dimensional. Let ω be a translation and {βg : g ∈ G} invariant factor state on B = ⊗ZZ Md. Then the following holds: (a) The family (vk) is linearly independent and {v∗ orthogonal vectors in K; (b) The following statements are equivalent: kΩ : 1 ≤ k ≤ d} is a set of TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 49 (i) Pk v∗ kvk = 1; (ii) ∆ = I; (iii) The action g : x → u(g)xu(g)∗ on M is ergodic. Further in such a case M is a finite type-In for some n ≥ 1 and the unique normalized trace φ0 on M is strongly mixing for τ : M → M, where (M, τ, φ0) is defined as in Proposition 2.4; Further there exists a unique representation of G, g → u′(g) ∈ M so that u′(g)vku′(g)∗ = X1≤j≤d vk j (g)vj for all 1 ≤ k ≤ d. The unique representation g → u′(g) is irreducible i.e. (M, αg, φ0) is G-ergodic. PROOF: As a first step we explore the hypothesis that the invariant vectors in ICd⊗ICd of the representation g → v(g)⊗v(g) of G is one dimensional. By appealing to the assumption with invariant vectors ((φ0(v∗ i SjΩ >)) we have i vj))) and ((< Ω, S∗ φ0(v∗ i vj ) = δi j λ d for all 1 ≤ i, j ≤ d and scaler λ > 0. That the scaler λ is indeed non-zero follows by separating property of Ω ( otherwise we will have vk = 0 for all k ). In particular we get the vectors {viΩ : 1 ≤ i ≤ d} are linearly independent and so by separating property of Ω for M (a) follows. (b) is immediate from (a) and Proposition 6.2. Vectors ((< Ω, S∗ i SjΩ >)) and ((φ0(viv∗ j ))) are also invariant for the product representation g → v(g) ⊗ v(g) and thus we also have φ0(viv∗ j ) = 1 d δi j for all 1 ≤ i, j ≤ d, where we have used Popescu's relation Pk vkv∗ we also have φ0(v∗ commutes with g → u(g) as the induced automorphism preserves the state φ0. j for some λ > 0. Same is true for φ0(vi∆sv∗ k = 1. Similarly j ) as ∆ i vj) = λ d δi Now we will prove the equivalence of those three statements. (ii) implies (i) is k vk = 1. obvious. Now we will prove (i) implies (ii): (i) ensures that λ = 1 and Pk v∗ Thus a simple computation shows that 1 ∆ kΩ − ∆− 1 2 v∗ 2 v∗ kΩ2 kΩ > −2 < Ω, vkv∗ kΩ, ∆−1v∗ kΩ > + < v∗ kΩ, J ∆− 1 =< vkΩ, vkΩ > + < J ∆− 1 2 v∗ 2 v∗ kΩ, ∆v∗ =< v∗ =< vkΩ, vkΩ > + < vkΩ, vkΩ > −2 < Ω, vkv∗ kΩ > kΩ > −2 < Ω, vkv∗ kΩ >= 0 kΩ > Hence by separating property of Ω for M we conclude that ∆v∗ k for all 1 ≤ k ≤ d. So ∆ is affiliated to M′. As J ∆J = ∆−1, ∆ is also affiliated to M. Hence ∆ = I as M is a factor. This completes the proof for (i) implies (ii). k∆−1 = v∗ (iii) implies (i) follows as ergodic action ensures that M is a type-I finite factor kvk is a G-invariant element, we conclude that the element is a scaler and hence by trace property of φ0 it is equal to 1. [Wa] and φ0 is the normalized trace. Since Pk v∗ That irreducibility g → u′(g) is equivalent to G-ergodicity of (M, αg, φ0) follows from a more general result [BR1], however here one can verify easily as M is a 50 ANILESH MOHARI type-I factor. For the non-trivial part of last statement we will use once more the property (b) i.e. self-adjointness of the Popescu elements {vk, 1 ≤ k ≤ d}. To that end let E be a G- irreducible projection in M and set E′ = J EJ . We set von-Neumann algebra ME = E′ME′. ME is a type-I finite factor as M is so and u(g)MEu(g)∗ = ME for all g as E′ is G-invariant. Further ME is also G-irreducible as E is an G-irreducible projection in M. The vector state φE(X) =< Ω, XΩ > on ME being G-invariant and irreducible, we conclude that φE is a scaler multiple of the unique normalized trace on ME ( see [HLS] for details ). That the non-normalized state ψE on Od defined by ψE(sI s∗ J ) =< Ω, E′vI v∗ J E′Ω > I, J < ∞ is λ-invariant follows by the tracial property of φE on ME and by (i) we make the following computation as follows: ψE(λ(sI s∗ < Ω, E′vkvI v∗ J v∗ kE′Ω > I, J < ∞ J )) = X1≤k≤d = X1≤k≤d < Ω, E′v∗ kE′vkE′vI v∗ J E′Ω > = ψE(sI s∗ J ) where Pk E′v∗ kE′vkE′ = E′ by (i). Further λ-invariance of ψE ensures that < Ω, E′xE′Ω >=< Ω, E′τ (x)E′Ω > for all x ∈ M. Tomita's modular operator being trivial we also have E′Ω = EΩ and so by duality relation we have < Ω, J EJ xΩ >=< Ω, τ (J EJ )xΩ > for all x ∈ M. Thus by cyclic and separating property of Ω for M, we get τ (J EJ ) = J EJ . Now by ergodic property of (M, τ, φ0) ( See Proposition 2.4 (e) ) and so the property for the dual Markov map, we conclude that J EJ is either 0 or 1. This completes the proof for (i) implies (iii). Thus in such a case i.e. if any of the statement (i)-(iii) is true, φ0 is a tracial state on M. That it is the unique trace follows as M is either a type-I finite factor or a type-II1 factor. However ω being pure, M can not be a type-II1 factor [Mo3]. M being a type-I factor and G being a simply connected group, by a general result [Ki] any continuous action of the G is implemented by an inner conjugation, i.e. there exists an unitary operator g → u′(g) ∈ M so that u(g)′xu(g)′∗ = u(g)xu(g)∗ for all g ∈ G. Thus we have u′(g)vku′(g)∗ = Pj vk j (g)vj for all 1 ≤ k ≤ d. Now for uniqueness let g → u′′(g) ∈ M be another such representation. Then u′′(g)u′(g)∗ ∈ MT M′ and M being a factor, λ(g) = u′′(g)u′(g)∗ is a scaler and as u′(g) = λ(g)u′′(g) and each one being a representation we also get g → λ(g) is a representation and so λ(g) = 1 as G is simply connected. Hence uniqueness follows. We are left to discuss few motivating examples for this abstract framework to study symmetry. For basic facts about thermodynamic limit, KMS states, ground states on quantum spin chain we refer to [Ru,BR2,EK]. We consider the following standard ( irreducible ) representation of Lie algebra su(2) in C2 TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 51 i 1 , , −i 1 , σx = (cid:18) 0 , σy = (cid:18) 0 σz = (cid:18) 1 g0 = (cid:18) 0 0 (cid:19) , 0 (cid:19) , , −1 (cid:19) . 0 (cid:19) , −i , , 0 0 i , Note also that g2 0 = −σx, g0σyσ∗ 0 = 1 and ig0 ∈ SU (2) and g0σxg∗ 0 = σy and g0σzg∗ 0 = −σz. By taking exponential of iσx, iσy, iσz, we conclude that g0 inter- twins g → vi j(g) with it's complex conjugate representation. The representation being irreducible, such an intertwining g0 is unique modulo a phase factor. Since we also need g2 0 = 1, we only have the choice given above or −g0. Same is true if we work with any irreducible representation of su(2) in d = 2s + 1 dimension and one such g0 ∈ Ud(C) exists which inter-twins the representation with it's complex conjugate. Now we aim to deal with two instructive examples. To that end we will now study the relation (33) for 1 2 -integer quantum spin chain. Theorem 6.5. Let ω be reflection positive with twist g0 as in Theorem 4.5 (b) and d be 2 with g0 = σy given above. Then (35) v1 = ǫJ v2J where ǫ is either 1 or −1. Further if the family {vk : 1 ≤ k ≤ 2} are linearly independent and ω is G-invariant where g → vi j(g) is a representation taking values j (g)) ∈ S1 ⊆ SU (2). There exists no SU (2)-invariant in SU2(C) then the range (vi state ω on B satisfying conditions for Theorem 4.5 (b) for d = 2 with g → vi j (g) irreducible. PROOF: By Theorem 4.5 (b) we have vg0 Xv∗ g0 = X for all x ∈ M0 ∨ M0 (36) and (37) Since vg0Mv∗ vg0 v1v∗ g0 = iJ v2J vg0 v2v∗ g0 = −iJ v1J g0 = M by separating property of Ω for M, we also claim that (38) which we verify as vg0 v1v∗ g0 = iJ v2J vg0 v∗ 1 v∗ g0 Ω = vg0 v∗ 1Ω = vg0 v∗ 1Ω = vg0 v∗ 1v∗ g0 Ω = −iJ v∗ 2J Ω = −iJ v∗ 2Ω = −iJ v∗ 2J Ω So by separating property we have vg0 v∗ other relation 1v∗ g0 = −iJ v∗ 2J . Similarly we have the Further we recall vg0 commutes with J and so vg0 v2v∗ g0 = −iJ v1J (39) We also have v2 g0 vk(v2 vg0 J v2J v∗ g0 = iv1 g0 )∗ = β−1(vk) for 1 ≤ k ≤ 2. 52 ANILESH MOHARI A simple computation now shows that vg0 (v1 + J v2J )v∗ g0 = i(v1 + J v2J ) and So we have vg0 (v∗ 1 − J v∗ 2J )v∗ g0 = −iJ v∗ 2J + iv∗ 1 vg0 (v1 + J v2J )X(v∗ = −(v1 + J v2J )X(v∗ 1 − J v∗ 2J )v∗ g0 2J ) 1 − J v∗ 1 − J v∗ where X ∈ M0 ∨ M0. The element (v1 + J v2J )X(v∗ invariant, is an element in M0 ∨ M0 and thus (v1 + J v2J )X(v∗ all X ∈ M0 ∨ M0. 2J ) being {βz : z ∈ H} 2J ) = 0 for 1 − J v∗ 2 J )(v∗ 1 − J v∗ 1 − J v∗ 2J )∗Ω is uz-invariant where βz(x) = uzxu∗ Assume now that v1 − J v2J 6= 0 and then action of (v1 + J v2J )∗(v1 + J v2J ) 2 J )∗Ω] is equal to 0 vector. By purity of ω we also have on [X(v∗ [M0 ∨ M0f ] = P0 for any f 6= 0 and P0f = f . Note that f = (v∗ 1 − J v∗ z defined as in Proposition 3.2 and so P0f = f . Thus by our assumption and separating property of Ω, we have f 6= 0. Hence we conclude that (v1 + J v2J )∗(v1 + J v2J )P0 = 0. Since P0Ω = Ω, by separating property for M we conclude that v1 + J v2J = 0. Interchanging the role of elements involve we conclude the first part of the result. 1 − J v∗ 2 J )(v∗ 1(g) = v2 1(g). ((vi 2(g) = v2 2(g) = −v2 2(g) and v1 By covariance relation (33) and linear independence of {vk : 1 ≤ k ≤ 2} we have now v1 j (g))) being an element in SU (2) we also have v1 1(g) = 0 for all g ∈ G. Thus (vi j (g)) ∈ S1 as a subgroup of SU (2). The last statement is now obvious once we use Proposition 6.3 to ensure linear independence hypothesis on the family {vk : 1 ≤ k ≤ d} since G = SU (2) satisfies the hypothesis by Clebsch Gordan theory and so brings a contradiction if we claim to exist such a state which is also SU (2) invariant with g → v(g) irreducible. This completes the proof. 1(g). Hence we conclude that v1 2(g) = v2 XY model: We consider the exactly solvable XY model. The Hamiltonian HXY of the XY model is determined by the following prescription: HXY = J(Xj∈ZZ {σ(j) x σ(j+1) x + σ(j) y σ(j+1) y } − 2λ Xj∈ZZ σ(j) z ), x , σ(j) y and σ(j) z where λ is a real parameter stand for external magnetic field and J is a non-zero real number, σ(j) are Pauli spin matrices at site j. It is well known [AMa] that ground state exists and unique. It is simple to verify that H = H since we can rewrite HXY as sum over element of the form σ(j−1) σ(j) y . Since the transpose of σx is itself, transpose of σy is −σy and transpose of σz is itself, we also verify that H t XY = HXY . Hence HXY is in detailed balance. x + σ(j−1) σ(j) x y For J < 0, it is also well known that for λ ≥ 1 the unique ground state is a product state thus split state. On the other hand for λ < 1 the unique ground state is not a split state [Ma2 Theorem 4.3]. For J > 0 HXY is reflection symmetric with a twist g0 which rotates an angle π with respect Y -axis. Further by a general theorem [FILS] ω is also reflection positive with a twist g0 when J > 0 and λ = 0. Thus by Theorem 6.3 the unique ground state is a non split state and H = {1, −1}. In such a case a simple application of Theorem 5.4 says that the correlation functions of the ground state does not decay exponentially. Theorem 6.5 gives functional relation between v1 and v2 and ∆ is non-trivial. TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 53 XXX MODEL: Here we consider the prime example where very little exact results were known. The Hamiltonian HXXX of the spin s anti-ferromagnetic chain i.e. the Heisenberg's XXX model is determined by the following formula: {S(j) x S(j+1) x + S(j) y S(j+1) y + S(j) z S(j+1) z } HXXX = J Xj∈ZZ and S(j) x , S(j) y where S(j) are representation in d = 2s + 1 dimensional of Pauli spin matrices σx, σy and σz respectively at site j. Existence of ground state for XXX model follows from more general theory [BR vol-2]. Since HXXX can be rewritten as sum of elements of the form z {S(j−1) x x + S(j−1) S(j) y y + S(j−1) S(j) z S(j) z } , it is simple to check that H XXX = HXXX . We also claim that H t XXX = HXXX . To that end we consider the space Vd of homogeneous polynomials in two complex variable with degree m, m ≥ 0 i.e. Vd is the space of functions of the form f (z1, z2) = a0zd 1 + a1zd−1 1 z2 + ... + adzd 2 with z1, z2 ∈ IC and a′ is are arbitrary complex constants. Thus Vd is a d-dimensional complex vector space. The d−dimensional irreducible representation πd of the Lie- algebra su(2) is given by πd(X)f = − ∂f ∂z1 (X11z1 + X12z2) + ∂f ∂z2 (X21z1 + X22z2) where X in any element in Lie-algebra su(2). It is simple to verify that the transpose of Sx = πd(σx) is itself, transpose of Sy = πd(σy) is −Sy and transpose of Sz = πd(σz) is itself. Thus H t XXX = HXXX for any d. Further for J > 0, the unique positive temperature state (KMS state ) for HXXX is also reflection positive [FILS] with twist g0 where g0 is as in XY model. Thus any limit point of KMS states as temperature goes to zero is also reflection positive. Thus for J > 0, if ground state is unique then it is also pure and reflection positive with twist g0. This brings a contradiction to our hypothesis on ω ( by the last statement in Theorem 6.5 ) that is real lattice symmetric positive with twist g0. We can expect same result to hold for any s with half-odd integer spin as g0 acts non trivially and g0g0 = −I. Thus in particular for J > 0 ( anti-ferromagnet ) and s is half-integer, there is a spontaneous symmetry breaking in the ground state. So we conclude that there are solutions other then the well known Bethe's ansatz [Be] solution. As an indirect consequence we conclude that Bethe's solution is not pure as it has all other property as infinite volume limit and low temperature limit preserves those property. It is not hard to prove now that the set of ground states does not even form a simplex. A valid question that we can ask at this stage whether the symmetry that we have got so far for HXXX is enough to make action transitive on the set of it's extremal points? Now we briefly discuss the situation when d = 3 i.e. s = 1. In such a case Pauli spin matrices are given by σx = 2− 1 σy = 2− 1 2   2   0 1 0 0 i 0 , , , 1, 0 0, 1 1, 0 , −i, 0 0, −i , i, 0 ,   ,   , 54 ANILESH MOHARI σz =   1 0 0 , , , 0, 0 0, 0 0, −1 g0 =   , 0 0 , −1 , 0, −1 1, 0 0, 0   .   . A direct calculation shows that the intertwiner g0 is a matrix with real entries given below Thus we have g0 = g0 and unitary operator vg0 given in relation (31) is self adjoint. So far we have not used reflection positivity. In such case we will investigate now vg0 . We write relation (31) now below: vg0 v1v∗ vg0 v2v∗ vg0 v3v∗ g0 = −J v3J g0 = J v2J g0 = −J v1J Going along the line of the proof for d = 2, using reflection positivity and purity of the state ω we show also now that either v1 = J v3J or v1 = −J v3J . Thus v1Ω and v2Ω are eigen vectors of vg0 with both eigenvalues are equal to 1 or −1 simultaneously. By the same reason v2 = J v2J or v2 = −J v2J i.e. v2Ω is an eigenvector with eigenvalue either 1 or −1. Now we use SU (2)-invariance property of ω to conclude from consistency of covariance relation that eigen value for v2Ω is equal to that of v1Ω. Thus vg0 is either 1 or vg0 = β−1. Thus we find ǫv1 = −J v3J ǫv2 = J v2J ǫv3 = −J v1J where ǫ is either 1 or −1. In case H is trivial, then ǫ = 1 and at this point it is not clear whether converse is true. Our result says very little about integer spin HXXX model as G− ergodic prop- erty in Theorem 6.4 is not evident as of now for group G = SU (2). REFERENCES • [Ac] Accardi, L. : A non-commutative Markov property, (in Russian), Func- tional. anal. i Prilozen 9, 1-8 (1975). • [AcC] Accardi, Luigi; Cecchini, Carlo: Conditional expectations in von Neumann algebras and a theorem of Takesaki. J. Funct. Anal. 45 (1982), no. 2, 245273. • [AM] Accardi, L., Mohari, A.: Time reflected Markov processes. Infin. Dimens. Anal. Quantum Probab. Relat. Top., vol-2 ,no-3, 397-425 (1999). • [AKLT] Affleck, L.,Kennedy, T., Lieb, E.H., Tasaki, H.: Valence Bond States in Isotropic Quantum Antiferromagnets, Commun. Math. Phys. 115, 477-528 (1988). • [AL] Affleck, L., Lieb, E.H.: A Proof of Part of Haldane's Conjecture on Spin Chains, Lett. Math. Phys, 12, 57-69 (1986). • [AMa] Araki, H., Matsui, T.: Ground states of the XY model, Commun. Math. Phys. 101, 213-245 (1985). • [Ar1] Arveson, W.: On groups of automorphisms of operator algebras, J. Func. Anal. 15, 217-243 (1974). • [Ar2] Arveson, W.: Pure E0-semigroups and absorbing states. Comm. Math. Phys 187 , no.1, 19-43, (1997) TRANSLATION INVARIANT PURE STATE ON ⊗ZMd(C) AND IT'S SPLIT PROPERTY 55 • [Be] Bethe, H.: Zur Theorie der Metalle. I. Eigenwerte und Eigenfunktionen der linearen Atomkette. I. Eigenvalues and eigenfunctions of the linear atom chain), Zeitschrift fr Physik A, Vol. 71, pp. 205226 (1931). (On the theory of metals. • [BR] Bratteli, Ola,: Robinson, D.W. : Operator algebras and quantum statistical mechanics, I,II, Springer 1981. • [BJP] Bratteli, Ola,: Jorgensen, Palle E.T. and Price, G.L.: Endomorphism of B(H), Quantisation, nonlinear partial differential equations, Operator algebras, ( Cambridge, MA, 1994), 93-138, Proc. Sympos. Pure Math 59, Amer. Math. Soc. Providence, RT 1996. • [BJKW] Bratteli, Ola,: Jorgensen, Palle E.T., Kishimoto, Akitaka and Werner Reinhard F.: Pure states on Od, J.Operator Theory 43 (2000), no-1, 97-143. • [BJ] Bratteli, Ola: Jorgensen, Palle E.T. Endomorphism of B(H), II, Finitely correlated states on ON , J. Functional Analysis 145, 323-373 (1997). • [Cu] Cuntz, J.: Simple C∗-algebras generated by isometries. Comm. Math. Phys. 57, no. 2, 173 -- 185 (1977). • [Ev] Evans, D.E.: Irreducible quantum dynamical semigroups, Commun. Math. Phys. 54, 293-297 (1977). • [EvK] Evans, David E.; Kawahigashi, Yasuyuki: Quantum symmetries on operator algebras. Oxford Mathematical Monographs. Oxford Science Publications. The Clarendon Press, Oxford University Press, New York, 1998. xvi+829 pp. • [Ex] Exel, Ruy : A new look at the crossed-product of a C∗-algebra by an endomorphism. (English summary) Ergodic Theory Dynam. Systems 23 (2003), no. 6, 17331750. • [FNW1] Fannes, M., Nachtergaele, B., Werner, R.: Finitely correlated states on quantum spin chains, Commun. Math. Phys. 144, 443-490(1992). • [FNW2] Fannes, M., Nachtergaele, B., Werner, R.: Finitely correlated pure states, J. Funct. Anal. 120, 511- 534 (1994). • [Fr] Frigerio, A.: Stationary states of quantum dynamical semigroups, Comm. Math. Phys. 63 (1978) 269-276. • [FILS] Frohlich, J., Israel, R., Lieb, E.H., Simon, B.: Phase Transitions and Reflection Positivity-I general theory and long range lattice models, Comm. Math. Phys. 62 (1978), 1-34 • [Ha] Hall, B.C. : Lie groups, Lie algebras and representations: an elemen- tary introduction, Springer 2003. • [Li] Liggett, T.M. : Interacting Particle Systems, Springer 1985. • [Mo1] Mohari, A.: Markov shift in non-commutative probability, J. Funct. Anal. vol- 199 , no-1, 190-210 (2003) Elsevier Sciences. • [Mo2] Mohari, A.: Pure inductive limit state and Kolmogorov's property, J. Funct. Anal. vol 253, no-2, 584-604 (2007) Elsevier Sciences. • [Mo3] Mohari, A.: Pure inductive limit state and Kolmogorov's property-II, http://arxiv.org/abs/1101.5961 • [Mo4] Mohari, A: Jones index of a completely positive map, Acta Appli- candae Mathematicae. Vol 108, Number 3, 665-677 (2009). • [Mo5] Mohari, A.: Translation invariant pure states in quantum spin chain and it's split property -II under preparation. • [Ma1] Matsui, T.: A characterization of pure finitely correlated states. Infin. Dimens. Anal. Quantum Probab. Relat. Top. 1, no. 4, 647 -- 661 (1998). 56 ANILESH MOHARI • [Ma2] Matsui, T.: The split property and the symmetry breaking of the quantum spin chain, Comm. Maths. Phys vol-218, 293-416 (2001) • [Po] Popescu, G.: Isometric dilations for infinite sequences of non-commutating operators, Trans. Amer. Math. Soc. 316 no-2, 523-536 (1989) • [Pow1] Powers, Robert T.: Representations of uniformly hyper-finite alge- bras and their associated von Neumann. rings, Annals of Math. 86 (1967), 138-171. • [Pow2] Powers, Robert T.: An index theory for semigroups of ∗-endomorphisms of B(H) and type II1 factors. Canad. J. Math. 40 (1988), no. 1, 86 -- 114. • [Ru] Ruelle, D. : Statistical Mechanics, Benjamin, New York-Amsterdam (1969) . • [Sa] Sakai, S. : Operator algebras in dynamical systems. The theory of unbounded derivations in C∗-algebras. Encyclopedia of Mathematics and its Applications, 41. Cambridge University Press, Cambridge, 1991. • [Ta] Takesaki, M. : Theory of Operator algebras II, Springer, 2001. • [Wa] Wassermann, Antony : Ergodic actions of compact groups on operator algebras. I. General theory. Ann. of Math. (2) 130 (1989), no. 2, 273319 The Institute of Mathematical Sciences, CIT Campus, Taramani, Chennai-600113 E-mail : [email protected]
1605.03568	1	1605	2016-05-11T19:04:01	Noncommutative Ergodic Theorems for Connected Amenable Groups	[ "math.OA", "math.DS" ]	This paper is devoted to the study of noncommutative ergodic theorems for connected amenable locally compact groups. For a dynamical system $(\mathcal{M},\tau,G,\sigma)$, where $(\mathcal{M},\tau)$ is a von Neumann algebra with a normal faithful finite trace and $(G,\sigma)$ is a connected amenable locally compact group with a well defined representation on $\mathcal{M}$, we try to find the largest noncommutative function spaces constructed from $\mathcal{M}$ on which the individual ergodic theorems hold. By using the Emerson-Greenleaf's structure theorem, we transfer the key question to proving the ergodic theorems for $\mathbb{R}^d$ group actions. Splitting the $\mathbb{R}^d$ actions problem in two cases according to different multi-parameter convergence types---cube convergence and unrestricted convergence, we can give maximal ergodic inequalities on $L_1(\mathcal{M})$ and on noncommutative Orlicz space $L_1\log^{2(d-1)}L(\mathcal{M})$, each of which is deduced from the result already known in discrete case. Finally we give the individual ergodic theorems for $G$ acting on $L_1(\mathcal{M})$ and on $L_1\log^{2(d-1)}L(\mathcal{M})$, where the ergodic averages are taken along certain sequences of measurable subsets of $G$.	math.OA	math	NONCOMMUTATIVE ERGODIC THEOREMS FOR CONNECTED AMENABLE GROUPS MU SUN Abstract. This paper is devoted to the study of noncommutative ergodic theorems for con- nected amenable locally compact groups. For a dynamical system (M, τ, G, σ), where (M, τ ) is a von Neumann algebra with a normal faithful finite trace and (G, σ) is a connected amenable locally compact group with a well defined representation on M, we try to find the largest non- commutative function spaces constructed from M on which the individual ergodic theorems hold. By using the Emerson-Greenleaf's structure theorem, we transfer the key question to proving the ergodic theorems for Rd group actions. Splitting the Rd actions problem in two cases accord- ing to different multi-parameter convergence types -- cube convergence and unrestricted conver- gence, we can give maximal ergodic inequalities on L1(M) and on noncommutative Orlicz space L1 log2(d−1) L(M), each of which is deduced from the result already known in discrete case. Fi- nally we give the individual ergodic theorems for G acting on L1(M) and on L1 log2(d−1) L(M), where the ergodic averages are taken along certain sequences of measurable subsets of G. 1. Introduction The study of ergodic theorems is an old branch of dynamical system theory which was started in 1931 by von Neumann and Birkhoff, having its origins in statistical mechanics. While new applications to mathematical physics continued to come in, the theory soon earned its own rights as an important chapter in functional analysis and probability. In the classical situation the sta- tionarity is described by a measure preserving transformation T , and one considers averages taken along a sequence f, f ◦ T, f ◦ T 2, . . . for integrable f. This corresponds to the probabilistic con- cept for stationarity. The modes of convergence under consideration mostly are norm convergence for"mean" ergodic theorems, and convergegence almost everywhere for "individual" (or pointwise) ergodic theorems. More generally, we study semigroups (or groups) {Tg, g ∈ G} of operators and limits of averages of Tgf over subsets In ⊂ G. In particular, group-theoretic considerations seem to be inherent in this analysis subject, and usually more interesting according to the abundance of structures for different group types. Given a locally compact group G with left Haar measure · , a sequence of measurable subsets {Fn}∞ n=1 in G is called a Følner sequence (similarly for Følner net), if 0 < Fn < ∞ for each n and ∀g ∈ G, lim n→∞ gFn △ Fn Fn = 0. A locally compact group is called amenable if it has a Følner sequence. There are many other equiv- alent ways to define amenability, however in the context of ergodic theory, the above definition is the most convenient. One of the ultimate aims in this direction is to establish the ergodic theorems for amenable group which is treated as a substitute for Abelian group. Using the group structure theorem, Emerson and Greenleaf gave a result for connected amenable locally compact group in [EG74]. It is until 2001 that Lindenstrauss's paper in Inventiones [Lin01] gave the pointwise ergodic theorem for the general amenable group and the averages is taken along certain tempered Følner sequence. The research on the existence and choice of the convergence sequence in the group is also quite an interesting and delicate subject. While the observable quantities in classical systems can be described by real functions on proba- bility space, they can be given for quantum systems by the operators in noncommutative probability space. In this case we have the so called noncommutative ergodic theory which has been developed since the very beginning of the theory of "rings of operators". But only mean ergodic theorems 2010 Mathematics Subject Classification: Primary 46L53, 46L55; Secondary 47A35, 37A99. Key words and phrases: Noncommutative Lp-spaces, noncommutative dynamical systems, connected amenable group, locally compact group, maximal ergodic inequalities for Rd group, individual ergodic theorem. 1 2 MU SUN have been obtained at the early stage. It is until 1976 that a substitute of "a.e. convergence" was first treated in Lance's work [Lan76], which is called almost uniform convergence. This conception unsealed the study of individual ergodic theorems for noncommutative case. In the same period, Conze and Dang-Ngoc was inspired by Lance's work and generalized the pointwise ergodic result in [EG74] to the actions on von Neumann algebra, which is now the p = ∞ case of noncommutative Lp spaces. In this paper, we denote M as a von Neumann algebra equipped with a normal faithful finite trace τ . In this case, we will often assume that τ is normalized, i.e., τ (1) = 1, thus we call (M, τ ) a noncommutative probability space. Let Lp(M)(1 ≤ p ≤ ∞) be the associated noncommutative Lp-space and Lp logr L(M)(r > 0) the associated noncommutative Orlicz space. The set of all projections in M is denoted as P (M). A multi-parameter net {xα1,...,αd } is said to converge bilaterally almost uniformly (resp. almost uniformly) to x, if for any ε > 0, there exists e ∈ P (M), such that τ (e⊥) ≤ ε and {e(xα1,...,αd − x)e} (resp. {(xα1,...,αd − x)e}) converges to 0 in M. Usually we denote it as b.a.u. (resp. a.u.) convergence. Let T : M → M be a linear map, it is called a kernel if it is a positive contraction and satisfy τ ◦ T ≤ τ for all x ∈ L1(M) ∩M+. We know the domain of kernels can naturally extend to Lp(M) for any 1 ≤ p < ∞ (c.f. Lemma 1.1 [JX07]), and we denote U as the set of all kernels. Now given a locally compact semigroup G, we define a homomorphism σ : G → U where correspondingly g 7→ Tg as G-actions on Lp(M), and the mapping g → Tg(x) is strongly continuous, for any x ∈ Lp(M), 1 ≤ p < ∞. So putting together (M, τ, G, σ) we call it a noncommutative dynamical system, and we we denote by Fσ the projection from Lp(M) onto the invariant subspace of G-actions. Ergodic averages along increasing measurable net {Vα} is defined as MVα(x) = 1 Vα ZVα Tg(x) dg. Our aim is to further generalize the noncommutative individual ergodic theorem for the system (M, τ, G, σ) if G is a connected amenable locally compact group. The main step of our method is also inherited from [EG74] that we decompose the connected amenable Lie group into the product of one compact group and Rd group. However, it is usually difficult to use iteration to obtain the result for p ∼ 1 case for multi-parameter R actions. To describe the situation, we here introduce two types of convergence for multi-parameter se- quence {ak1,...,kd }(k1,...,kd)∈Nd (similar for net) in any Banach space B. The first type is called cube convergence, if the sequence {an,...,n}n∈N converges in B when n tends to ∞. The second type is called unrestricted convergence, if {ak1,...,kd }(k1,...,kd)∈Nd converges in B when k1, . . . , kd tend to ∞ arbitrarily. Thus when we consider the ergodic averages for Rd actions, we have separate maxi- mal ergodic inequalities according to different types of convergence, for which we refer to section 3. We emphasize here that for the more general case of unrestricted convergence, it is kind of a breakthrough since there is no weak type (1, 1) inequality for multi-parameter noncommutative dynamical system. It is only recently in my joint work with Hong [HS16] the maximal inequality for the case of Zd group acting on noncommutative Orlicz space L1 log2(d−1) L(M) is obtained, in the light of which we can give a result in continuous case. It is in section 2 we give some more detailed introduction of noncommutative probability space, and in section 4 we present the main result of this paper. Since it is obvious that cube convergence is just a special case of unrestricted convergence, the sequences of measurable subsets of G taken along for keeping cube convergence of the ergodic averages should be included in unrestricted convergence case. This idea is specified through the narrative and proof Theorem 4.2. 2. Preliminaries We use standard notions for the theory of noncommutative Lp spaces. Our main references are [PX03] and [Xu07]. Let M be a von Neumann algebra equipped with a normal faithful finite trace τ . Let L0(M) be the spaces of measurable operators associated to (M, τ ). For a measurable operator x, its generalized singular number is defined as µt(x) = inf{λ > 0 : τ(cid:0)1(λ,∞)(x)(cid:1) ≤ t}, t > 0. NONCOMMUTATIVE ERGODIC THEOREMS FOR CONNECTED AMENABLE GROUPS 3 The trace τ can be extended to the positive cone L+ 0 (M) of L0(M), still denoted by τ , Given 0 < p < ∞, let and for x ∈ Lp(M), τ (x) = Z ∞ 0 µt(x)dt, x ∈ L+ 0 (M). Lp(M) = {x ∈ L0(M) : τ (xp) < ∞} kxkp = (cid:0)τ (xp)(cid:1) 1 p = (cid:0)Z ∞ 0 (µt(x))pdt(cid:1) 1 p . Then (Lp(M), k · kp) is a Banach space (or quasi-Banach space when p < 1). This is the noncom- mutative Lp space associated with (M, τ ), denoted by Lp(M, τ ) or simply by Lp(M). As usual, we set L∞(M, τ ) = M equipped with the operator norm. The noncommutative Orlicz spaces are defined in a similar way as commutative ones. Given an Orlicz function Φ, the Orlicz space LΦ(M) is defined as the set of all measurable operators x such that Φ( x λ ) ∈ L1(M) for some λ > 0. Equipped with the norm kxkΦ = inf (cid:26)λ > 0 : τ (cid:20)Φ(cid:18) x λ (cid:19)(cid:21) ≤ 1(cid:27) , LΦ(M) is a Banach space. When Φ(t) = tp with 1 ≤ p < ∞, the space LΦ(M) coincides with Lp(M). If Φ(t) = tp(1 + log+ t)r for 1 ≤ p < ∞ and r > 0, we get the space Lp logr L(M). From the definition, it is easy to check that whenever (M, τ ) is a probability space, we have Lq(M) ⊂ Lp logr L(M) ⊂ Ls(M) for q > p ≥ s ≥ 1. The spaces Lp(M; ℓ∞) and Lp(M; ℓc ∞) play an important role in the formulation of noncom- mutative maximal inequalities. A sequence {xn}n≥0 ⊂ Lp(M) belongs to Lp(M; ℓ∞) if and only if it can be factored as xn = aynb with a, b ∈ L2p(M) and a bounded sequence {yn} ⊂ L∞(M). We then define Following [JX07], this norm is symbolically denoted by ksupn positive sequence {xn}n belongs to Lp(M; ℓ∞) if and only if there exists a ∈ L+ xn ≤ a for all n ≥ 0. Moreover k{xn}nkLp(ℓ∞) = inf xn=aynb(cid:8)kak2p sup n≥0 kynk∞ kbk2p(cid:9). +xnkp. It is shown in [JX07] that a p (M) such that ksup n +xnkp = inf (cid:8)kakp : a ∈ L+ p (M) s.t. xn ≤ a, ∀ n ≥ 0(cid:9). Here and in the rest of the paper, L+ ∞) is defined to be the set of sequences {xn}n≥0 with {xn2}n≥0 belonging to Lp/2(M; ℓ∞) equipped with (quasi) norm p (M) denotes the positive cone of Lp(M). The space Lp(M; ℓc k{xn}nkLp(ℓc ∞) = k{xn2}nk 1 2 L p 2 (ℓ∞). We refer to [Jun02], [JX07] and [Mus03] for more information on these spaces and related facts. The vector-valued Orlicz spaces Lp logr L(M; ℓ∞) (1 ≤ p < ∞, r > 0) are firstly introduced by Bekjan et al in [BCO13]. The key observation is that Lp(ℓ∞)-norm has an equivalent formulation: k{xn}nkLp(ℓ∞) = inf (cid:8) 1 2 (kak2 2p + kbk2 2p) sup n≥0 kynk∞(cid:9), where the infimum is taken over the same parameter as before. Given an Orlicz function Φ, let {xn} be a sequence of operators in LΦ(M), we define τ(cid:0)Φ(sup n +xn)(cid:1) = inf (cid:8) 1 2(cid:0)τ (Φ(a2)) + τ (Φ(b2))(cid:1) sup n≥0 kynk∞(cid:9), where the infimum is taken over all the decompositions xn = aynb for a, b ∈ L0(M) and yn ∈ L∞(M) with a2, b2 ∈ LΦ(M) and supn kynk∞ ≤ 1. Then LΦ(M; ℓ∞) is defined to be the set of sequences {xn}n ⊂ LΦ(M) such that there exists one λ > 0 satisfying τ(cid:0)Φ(sup n + xn λ )(cid:1) < ∞ 4 MU SUN equipped with the norm k{xn}nkLΦ(ℓ∞) = inf (cid:26)λ > 0 : τ(cid:0)Φ(sup n + xn λ )(cid:1) < 1(cid:27) . Then (LΦ(M; ℓ∞), k · kLΦ(ℓ∞)) is a Banach space. It was proved in [BCO13] that a similar charac- terization holds for sequences of positive operators: τ(cid:0)Φ(sup n +xn)(cid:1) ≈ inf (cid:8)τ (Φ(a)) : a ∈ L+ Φ(M) s.t. xn ≤ a, ∀ n ≥ 0(cid:9) which implies a similar characterization for the norm k{xn}nkLΦ(ℓ∞) ≈ inf (cid:8)kakΦ : a ∈ L+ Φ(M) s.t. xn ≤ a, ∀ n ≥ 0(cid:9). From the definition, it is not difficult to check that whenever (M, τ ) is a probability space, we have Lq(M; ℓ∞) ⊂ Lp logr L(M; ℓ∞) ⊂ Ls(M; ℓ∞) for q > p ≥ s ≥ 1. We refer the reader to [BCO13] for more information on the vector-valued Orlicz spaces. 3. Maximal ergodic inequalities for multi-parameter R+ semigroups In this section, we are going to give two versions of maximal ergodic inequalities for d-parameter (d ≥ 2) R+ semigroups according to different types of convergence. We give the cube convergence version first which is directly a generalization of weak type (1, 1) inequality to multi-parameter and continuous case. Theorem 3.1. Let (M, τ, R+, σi) i = 1, . . . , d be d noncommutative dynamical systems. Set σi(s) = T (i) s i = 1, . . . , d and M[t] = 1 td Z t 0 T (d) sd dsd · · · Z t 0 T (1) s1 ds1. Then for every x ∈ L1(M) and any λ > 0, there exists e ∈ P (M) such that τ (e⊥) ≤ C kxk1 λ and sup t>0 ke M[t](x) ek∞ ≤ λ , where C is a positive constant independent of x and λ. Proof. Recall that the semigroup (T (i) i = 1, . . . , d is strongly continuous on L1(M), i.e., s )s≥0 for any x ∈ L1(M) the function s 7→ T (i) s (x) is continuous from [0, ∞) to L1(M), and so is the function t 7→ M[t](x). Thus to prove the inequality it suffices to consider M[t](x) for t in a dense subset of (0, ∞), for instance, the subset {n2−m : m, n ∈ N}. Using once more the strong continuity of (T (i) s )s≥0, we can replace the integral defining M[t](x) by a Riemann sum. Thus we have approximately M[n2−m](x) = (n2−m)d n−1 1 Z (kd+1)2−m kd2−m Xkd=0 T (d) kd2−m(x) · · · T (d) sd (x) dsd · · · n−1 Xk1=0 Z (k1+1)2−m k12−m T (1) s1 (x) ds1 ≈ 1 nd n−1 Xkd=0 n−1 Xk1=0 T (1) k12−m (x). As we have transformed our semigroup ergodic averages into the discrete case which is already proved in Theorem 1.2 [Ska05], thus we conclude the wanted weak type (1, 1) inequality. (cid:3) Next we give the unrestricted version. It is on the other hand an extension to continuous case of the multi-parameter maximal ergodic inequalities on noncommutative Orlicz spaces recently proved in [HS16]. Though we make loose of the trace preserving condition, the proof in [HS16] still hold in our kernel case. The rest is to use the same transference as in Theorem 3.1, so we omit the proof here. NONCOMMUTATIVE ERGODIC THEOREMS FOR CONNECTED AMENABLE GROUPS 5 Theorem 3.2. Let (M, τ, R+, σi) i = 1, . . . , d be d noncommutative dynamical systems. Set σi(s) = T (i) s i = 1, . . . , d and Mt1,... ,td = 1 t1 · · · td Z td 0 T (d) sd dsd · · · Z t1 0 T (1) s1 ds1. Then for every x ∈ L1 log2(d−1) L(M) and any λ > 0, there exists e ∈ P (M) such that τ (e⊥) ≤ C kxkL1 log2(d−1) L λ and sup t1>0,... ,td>0 ke Mt1,... ,td (x) ek∞ ≤ λ , where C is a positive constant independent of x and λ; moreover, for x ∈ L2 log2(d−1) L(M) we have the following estimates τ (e⊥) ≤ (cid:18)C kxkL2 log2(d−1) L λ 2 (cid:19) and sup t1>0,... ,td>0 kMt1,... ,td(x) ek∞ ≤ λ. 4. Individual ergodic theorems In this section, we are going to treat different types of almost uniform convergence of ergodic averages for connected amenable group actions. As a preparation, we first give the following lemma which is a direct conclusion from last section. Lemma 4.1. Let (M, τ, R+, σi) i = 1, . . . , d be d noncommutative dynamical systems. i) For every x ∈ L1(M), M[t](x) converges b.a.u. to Fσd · · · Fσ1 (x) as t tends to ∞. ii) For every x ∈ L1 log2(d−1) L(M), Mt1,... ,td (x) converges b.a.u. to Fσd · · · Fσ1 (x) as t1, . . . , td tend to ∞ arbitrarily. Moreover, if x ∈ L2 log2(d−1) L(M), then we have the a.u. convergence. Proof. We first prove ii). It is already known that for any y ∈ L∞(M), Mt1,... ,td (x) converges [CN78], [JX07]), and L∞(M) is dense in L1 log2(d−1) L(M). Thus b.a.u. to Fσd · · · Fσ1 (x) (c.f. taking any x ∈ L1 log2(d−1) L(M), any ε > 0 and any λ > 0, we can always find a y ∈ L∞(M), where kx − ykL1 log2(d−1) L(M) ≤ 1 2C λε, kFσd · · · Fσ1 (x − y)k∞ ≤ λ, and C is the positive constant from the application of the first maximal inequality in Theorem 3.2 to the element x − y: there exists a projection e1 ∈ P (M) such that τ (e⊥ 1 ) ≤ C kx − ykL1 log2(d−1) L(M) λ ≤ ε 2 and sup t1>0,... ,td>0 ke1 Mt1,... ,td (x − y) e1k∞ ≤ λ . On the other hand, there exists a projection e2, such that τ (e⊥ 2 ) ≤ ε/2 and ke2 (Mt1,... ,td (y) − Fσd · · · Fσ1 (y)) e2k∞ converges to 0, which means that there exists r1, . . . , rd ∈ R+, when ti ≥ ri, i = 1, . . . , d, we have Now take e = e1 ∧ e2, then we have τ (e⊥) ≤ ε and ke2Mt1,... ,td (y)e2 − e2 Fσd · · · Fσ1 (y) e2k∞ < λ. ke Mt1,... ,td (x) e − e Fσd · · · Fσ1 (x) ek∞ ≤ke Mt1,... ,td (x − y) ek∞ + ke Mt1,... ,td (y) e − e Fσd · · · Fσ1 (y) ek∞ + ke Fσd · · · Fσ1 (y − x) ek∞ <3λ. So we have proved Mt1,... ,td (x) converges b.a.u. to Fσd · · · Fσ1 (x) for every x ∈ L1 log2(d−1) L(M), and using the same process with the second inequality in Theorem 3.2, we have all the conclusions of ii). As L1 log2(d−1) L(M) is dense in L1(M), and the cube convergence is just a special case of unre- stricted convergence, then apply the maximal inequality in Theorem 3.1 and repeat the reasoning as above, it follows the result of i). Thus we complete the proof. (cid:3) We are now in position to give our main result by using a structure theorem of connected amenable locally compact group of Emerson and Greenleaf [EG74]. Theorem 4.2. Given (M, τ, G, σ) a noncommutative dynamical system with G a connected amenable locally compact group, there exists V and W that are classes of increasing sequences of measurable subsets of G, W is strictly included in V such that: 6 MU SUN i) for each (Vn)n≥1 ∈ V, [n≥1 L2 log2(d−1) L(M)), we have Vn generates G and any x ∈ L1 log2(d−1) L(M) (resp. x ∈ 1 Vn ZVn Tg(x)dg converges to Fσ(x) b.a.u. (resp. a.u.) when n → ∞; ii) for each (Wn)n≥1 ∈ W and any x ∈ L1(M), we have 1 Wn ZWn Tg(x)dg converges to Fσ(x) b.a.u. when n → ∞. Proof. In any connected amenable locally compact group G, there is a largest compact normal subgroup K(G). The quotient G′ = G/K(G) is an amenable Lie group with K(G′) trivial. a) We first treat the case of amenable Lie group G′ with K(G′) trivial. Following Lemma 4.2 from [EG74] by Emerson and Greenleaf: there are closed one-parameter subgroups L1, . . . , Ld(Li ≃ R) and a compact connected subgroup K such that, 1) The map j : K × Ld × · · · × L1 → G′ is a homeomorphism; 2) The subsets Hi = j(Li × · · · × L1) are closed subgroups in G′, 1 ≤ i ≤ d; 3) Hi is normal in Hi+1 for i = 1, 2, . . . , d (we take Hd+1 = G′); 4) The map j identifies the product Haar measures mK × md × · · · × m1 on K × Ld × · · · × L1 with a left Haar measure dg′ on G′. Let σi be the restriction of σ to Li (i = 1, . . . , d), and σ the restriction to K. Therefore each subgroup forms a dynamical system, and for simplicity we denote by Fi and F the corresponding projections from B(M) onto the fixed point subspaces with respect to every subgroup actions. Here B(M) denotes some noncommutative function space constructed from M. Without saying in particular, we usually identify group G′ with its parameterized form K × Ld × · · · × L1. Let V ′(td, . . . , t1) = K × [−td, td] × · · · × [−t1, t1], td, . . . , t1 > 0, in the following, our subsets sequence have this kind of form. Choosing B(M) and the sequence ti(ni) −−−−→ ni→∞ we can achieve the following convergence ∞, i = 1, 2, . . . , d properly, for any x ∈ B(M) (4.1) 1 V ′(td(nd), . . . , t1(n1)) Z V ′(td(nd),...,t1(n1)) Tg′(x)dg′ −→ F Fd · · · F1(x) in different senses, and correspondingly we get the sequence classes V and W from the construction. By the former decomposition properties we can easily deduce that Tg′ = σ(g′) = σ(k, sd, . . . , s1) = σ(k) T (d) sd · · · T (1) s1 , and we note Mnd,... ,n1(x) = 1 2d t1 · · · td td(nd) t1(n1) Z dsd · · · Z −td(nd) −t1(n1) ds1T (d) sd · · · T (1) s1 (x) Mnd,... ,n1(x) = Z K dk σ(k) Mnd,... ,n1(x) Typically we consider (cid:13)(cid:13)(cid:13)(cid:16)Z = (cid:13)(cid:13)(cid:13)(cid:16)Z ≤ Z K K σ(k) Mnd,... ,n1(x) dk − F Fd · · · F1(x)(cid:17)e(cid:13)(cid:13)(cid:13)∞ dk σ(k)(cid:2)Mnd,... ,n1(x) − Fd · · · F1(x)(cid:3)(cid:17)e(cid:13)(cid:13)(cid:13)∞ dk(cid:13)(cid:13)(cid:13)(cid:16)Mnd,... ,n1(x) − Fd · · · F1(x)(cid:17)e(cid:13)(cid:13)(cid:13)∞ K , NONCOMMUTATIVE ERGODIC THEOREMS FOR CONNECTED AMENABLE GROUPS 7 and e denotes any projection in P (M). So we know that the convergence of Mnd,... ,n1(x) is dominated by Mnd,... ,n1 (x). Here is when we apply Lemma 4.1 to obtain the b.a.u. (resp. a.u.) convergence of (4.1) for ∞, i = 1, 2, . . . , d every x ∈ L1 log2(d−1) L(M) (resp. x ∈ L2 log2(d−1) L(M)) as ti(ni) −−−−→ ni→∞ arbitrarily; in particular, if we let the sequence t1(n1) = · · · = td(nd) → ∞, then we have the b.a.u. convergence of (4.1) for every x ∈ L1(M). It remains to prove F Fd · · · F1(x) = Fσ(x). However, take any h1 ∈ H1 and h ∈ H2. From the normality of H1 in H2, we know there exists h′ 1. Then we F1(x) = ThF1(x). As F2F1(x) is a limit(in strong operator topology) have Th1ThF1(x) = ThTh′ of combinations of ThF1(x), so F2F1(x) is invariant of H1-actions. Similarly, we can prove that F Fd · · · F1(x) is invariant under each L1, . . . , Ld, K subgroup actions, thus invariant of G′, so we finish this part of proof. 1 ∈ H1, such that h1h = hh′ 1 b) Now we return to the general case of G. Let u be the canonical surjection of G onto G′ = G/K(G), and define We also define V (td, . . . , t1) = u−1(cid:0)V ′(td, . . . , t1)(cid:1). V = {(Vn = V (td(n), . . . , t1(n)))n≥1 : ti(n) → +∞ as n → +∞, i = 1, . . . , d} and W = {(Wn)n≥1 ∈ V : Wn = W (td(n), . . . , t1(n)) satisfies td(n) = · · · = t1(n) for each n}. In the following, we can also write K in short for K(G) without confusion to part a) (since K is also compact subgroup) and let dk be the normalized Haar measure on K, dg be the left Haar measure on G such that We have then Z G f (g) dg = Z G′ dg′Z K f (g′k) dk, ∀f ∈ C0(G). V (td, . . . , t1) = V ′(td, . . . , t1). Like before, for K and G′, there are corresponding subgroup action systems as (M, τ, K, σ) and (M, τ, G′, σ′), also we denote F and F ′ as the corresponding projections onto each fixed point subspace. Since K is compact, we similarly consider the subgroup action, note that F (x) = Z σk(x) dk and we have K 1 V (td, . . . , t1) Z Tg(x) dg = 1 V ′(td, . . . , t1) Z V (td, ... ,t1) V ′(td, ... ,t1) dg′σ′ g′ F (x). Then by the result in part a), we get the corresponding convergence results of the above equation with respect to sequences in V and W, and this completes the proof. (cid:3) References [BCO13] T. Bekjan, Z. Chen, and A. Osekowski: Noncommutative maximal inequalities associated with convex functions, arXiv: 1108.2795. [Bir31] G.D. Birkhoff: Proof of the ergodic theorem, Proc. Nat. Acad. Sci. 17 (1931), 656-660. [CLS05] V. Chilin, S. Litvinov, A. Skalski: A few remarks in non-commutative ergodic theory, J. Operator Theory 53 (2005), 331-350. [CN78] J.P. Conze and N. Dang-Ngoc: Ergodic theorems for noncommutative dynamical systems, Inventiones mathematicae 46 (1978), 1-15. [Cuc71] I. Cuculescu: Martingales on von Neumann algebras, J. Multivariate Anal. 1 (1971), 17-27. [DS88] N. Dunford and J.T. Schwartz: Linear operators: General theory, Wiley Classics Library. New York: Wiley (1988). [EG74] Emerson, W.R., Greenleaf, F.P.: Group structure and the pointwise ergodic theorem for connected amebale groups, Advances in Math. 14 (1974), 153-172. [GL00] M. Goldstein and S. Litvinov: Banach principle in the space of τ -measurable operators, Studia Math. 143 (2000), 33-41. 8 MU SUN [Hen83] E. Hensz: On some ergodic theorems for von Neumann algebras, Probability Theory on Vector Spaces, III, lublin, Springer's LNM , vol. 119-123 (August 1983). [HS16] G. Hong, M. Sun, Noncommutative multi-parameter Wiener-Wintner type ergodic theorem, arXiv:1602.00927, [math.OA], 2016: 1-25. [Hu09] Y. Hu: Noncommutative extrapolation theorems and applications, Illinois J. of Math. 53 (2009), 463-482. [Jun02] M. Junge: Doob's inequality for non-commutative martingales, J. Reine Angew. Math. 549 (2002), 149-190. [JX07] M. Junge and Q. Xu: Noncommucative maximal ergodic theorems, J. Amer. Math. Soc. 20 (2007), 385-439. [Kum78] B. Kummerer: A non-commutative individual ergodic theorem, Invent. Math. 46(1978), 139-145. [Lan76] E.C. Lance: Ergodic theorems for convex sets and operator algebras, Invent. Math., 37(3)(1976), 201-214. [Lin01] E.Lindenstrauss: Pointwise ergodic theorems for amenable groups, Invent. Math., 146(2)(2001), 259-295. [MMT08] F. Mukhamedov, M. Mukhamedov, S. Temir: On multiparameter weighted ergodic theorem for noncom- mutative Lp-spaces, J. Math. Anal. Appl. 343 (2008), 226-232. [Mus03] M. Musat: Interpolation Between Non-commutative BM O and Non-commutative Lp-spaces, J. Funct. Anal., 202 (2003), 195-225. [Pis98] G. Pisier: Non-commutative vector valued Lp-spaces and completey p-summing maps, Ast´erisque 247 (1998), 129p. [PX03] G. Pisier and Q. Xu: Non-commutative Lp-spaces, in Handbook of the Geometry of Banach Spaces vol. 2, North-Holland, Amsterdam (2003), 1459-1517. [Rud62] W. Rudin: Fourier analysis on groups, interscience (1962). [Ska05] A. G.Skalski: On a classical scheme in noncommutative multiparameter ergodic theory, Quantum Probability and Infinite Dimensional Analysis, QP-PQ: Quantum Probab. White Noise Anal. , vol. 18, World Sci. Publ., Hackensack, NJ (2005), 473-491. [Xu07] Q. Xu, Noncommutative Lp spaces and Martingale inequalities, Book Manuscript, (2007). [Yea77] F.J. Yeadon: Ergodic theorems for semifinite von Neumann algebras. I. J. London Math. Soc. 16(2) (1977), 326-332.
1905.12451	1	1905	2019-05-28T14:31:54	On the inductive limit of direct sums of simple TAI algebras	[ "math.OA" ]	An ATAI (or ATAF, respectively) algebra, introduced in [Jiang1] (or in [Fa] respectively) is an inductive limit $\lim\limits_{n\rightarrow\infty}(A_{n}=\bigoplus\limits_{i=1}A_{n}^{i},\phi_{nm})$, where each $A_{n}^{i}$ is a simple separable nuclear TAI (or TAF) C*-algebra with UCT property. In [Jiang1], the second author classified all ATAI algebras by an invariant consisting orderd total K-theory and tracial state spaces of cut down algebras under an extra restriction that all element in $K_{1}(A)$ are torsion. In this paper, we remove this restriction, and obtained the classification for all ATAI algebras with the Hausdorffized algebraic $K_{1}$-group as an addition to the invariant used in [Jiang1]. The theorem is proved by reducing the class to the classification theorem of $\mathcal{AHD}$ algebras with ideal property which is done in [GJL1]. Our theorem generalizes the main theorem of [Fa] and [Jiang1] (see corollary 4.3).	math.OA	math	ON THE INDUCTIVE LIMIT OF DIRECT SUMS OF SIMPLE TAI ALGEBRAS BO CUI, CHUNLAN JIANG, AND LIANGQING LI Abstract An ATAI (or ATAF, respectively) algebra, introduced in [Jiang1] (or in [Fa] respectively) is an inductive n is a simple separable nuclear TAI (or TAF) C-algebra with n, φnm), where each Ai Ai limit lim n→∞ (An = Li=1 UCT property. In [Jiang1], the second author classified all ATAI algebras by an invariant consisting orderd total K-theory and tracial state spaces of cut down algebras under an extra restriction that all element in K1(A) are torsion. In this paper, we remove this restriction, and obtained the classification for all ATAI algebras with the Hausdorffized algebraic K1-group as an addition to the invariant used in [Jiang1]. The theorem is proved by reducing the class to the classification theorem of AHD algebras with ideal property which is done in [GJL1]. Our theorem generalizes the main theorem of [Fa] and [Jiang1] (see corollary 4.3). 1. Introduction In [Lin2], Lin gave an abstract description of simple AH algebras (with no dimension growth) classified in [EGL1]. He described the decomposition property of simple AH algebras in [G3] as TAI property and proved that all simple separable nuclear TAI algebras with UCT are classifiable and therefore in the class of [EGL1]. For simple AH algebras of real rank zero, the corresponding decomposition property is called TAF by [Lin1], which is partially inspired by Popa's paper [Popa]. As proved by Lin ([Lin1-2]), a simple separable nuclear C-algebra A with UCT property is a TAI (or TAF, respectively) algebra if and only if A is a simple AH algebra (or simple AH algebra of real rank zero) with no dimension growth, which is classified in [EGL1]. As in [EG2], let TII,k be the 2-dimensional connected simplicial complex with H1(TII,k) = 0 and H2(TII,k) = Z/kZ, and let Ik be the subalgebra of Mk(C[0, 1]) = C([0, 1], Mk(C)) consisting of all functions f with the properties f (0) ∈ C1k and f (1) ∈ C1k (this algebra is called an Elliott dimension drop interval alge- bra). Denote HD the class of algebras consisting of direct sums of building blocks of the forms Ml(Ik) and P Mn(C(X))P , with X being one of the spaces {pt}, [0, 1], S1, and TII,k, and with P ∈ Mn(C(X)) being a projection. (In [DG], this class is denoted by SH(2), and in [Jiang1], this class is denoted by B). A C-algebra is called an AHD algebra, if it is an inductive limit of algebras in HD. In [GJL2], the authors classified all AHD algebras with the ideal property. In [Jiang1], Jiang classified ATAI algebras A under the extra restriction that K1(A) = torK1(A). In this classification, Jiang used the scaled ordered total K-group (from [DG]) and the tracial state spaces T (pAp) 1 of cut-down algebras pAp, with certain compatibility conditions (from [Stev] and [Ji-Jiang]) as the invariant -- we will call it Inv0(A). In this paper, we will use the invariant Inv0(A) together with the Hausdorffized algebraic K1-group to deal with the torsion-free part of K1(A), with certain compatibility conditions -- we will call this Inv(A). We will prove that this invariant reduces to Jiang's invariant in the case that K1(A) is a torsion group, that is, we removed the extra restriction that K1(A) = torK1(A) of Jiang's classification in [Jiang1] and prove our theorem by reduced to the classification of AHD algebra with ideal property, which is done in [GJL1]. 2. Preliminaries And Definitions 2.1. Let A be a C-algebra. Two projections in A are said to be equivalent if they are Murray -- von Neumann equivalent. We write p (cid:22) q if p is equivalent to a projection in qAq. We denote by [p] the equivalent class of projections equivalent to p. Let a ∈ A+, we write p (cid:22) a if p (cid:22) q for some projection q ∈ aAa. Definition 2.2. Let G ⊂ A be a finite set and δ > 0. We shall say that φ ∈ M ap(A, B) is G−δ multiplicative if kφ(ab) − φ(a)φ(b)k < δ for all a, b ∈ G. We also use M apG−δ(A, B) to denote all G − δ multiplicative maps. If two maps ϕ, φ ∈ M ap(A, B) satisfy the condition kφ(a) − ψ(a)k < δ for all a ∈ G, then we will write φ ≈δ ϕ on G. If G, H ⊂ A are two subsets of a C-algebra A and for any g ∈ G, there is a b ∈ H with kg − bk < δ, then we denote G ⊂δ H. Definition 2.3. [Jiang1] We denote by I the class of all unital C-algebras with the formLn each Bi ∼= Mki or Bi ∼= Mki(C[0, 1]) for some k(i). Let A ∈ I, we have the following: (1) Every C-algebra in I is of stable rank one. i=1 Bi, where (2) Two projections p and q in a C-algebra A ∈ I are equivalent if and only if τ (p) = τ (q) for all τ ∈ T (A), where T (A) denotes the space of all tracial states. (3) For any ε > 0 and any finite subset F ⊂ A, there exist a number δ > 0 and a finite subset G ⊂ A satisfying the following: If L : A → B is a G − δ multiplicative contractive completely positive linear map, where B is a C-algebra, then there exists a homomorphism h : A → B such that kh(a) − L(a)k ≤ ε, ∀a ∈ F Definition 2.4. ([Lin1]) A unital simple C-algebra in A is said to be tracially AI (TAI) if for any finite subset F ⊂ A containing a nonzero element b, ε > 0, integer n > 0 and any full element a ∈ A+, there exist nonzero projection p ∈ A and a C-algebra I ⊂ A with I ∈ I and 1I = p such that: (1) kxp − pxk < ε for all x ∈ F ; (2) pxp ∈ε I for all x ∈ F ; (3) n[1 − p] (cid:22) [p] and 1 − p (cid:22) a. 2 Definition 2.5. [Jiang1] A C-algebra A (not necessary unital) is said to be ATAI algebra (approximately TAI algebra) if it is the inductive limit of a sequence of direct sums of simple unital TAI algebras with UCT. 2.6. Let A and B be two C-algebras. We use M ap(A, B) to denote the space of all completely positive ∗- contractions from A to B. If both A and B are unital, then M ap(A, B)1 will denote the subset of M ap(A, B) consisting of all such unital maps. 2.7. In the notation for an inductive limit system lim(An, φn,m), we understand that φn,m = φm−1,m ◦ φm−2,m−1 ◦ · · · ◦ φn,n+1 where all φn,m : An → Am are homomorphisms. We shall assume that, for any summand Ai i=1 Ai n, necessarily, φn,n+1(1Ai ) 6= 0, n since, otherwise, we could simply delete Ai n from An, without changing the limit algebra. n in the direct sum An =Ltn m, we use φi,j n,m to denote the partial map of φn,m from the i-th block Ai An to the j-th block Aj is, φ−,j m of Am. Also, we use φ−,j φi,j n,m = πjφn,m, where πj : Am → Aj n,m to denote the partial map of φn,m from An to Aj m is the canonical projection. We also use φi,− n of m. That n,m to denote n, Am =Lj Aj If An =Li Ai n,m =Li : Ai φn,mAi n n → Am. 2.8. An AH algebra is a C-algebra which is lim(An =Lkn i=1 pn,iM[n,i](C(Xn,i))pn,i, φn,m), where each Xn,i is a compact metrizable space, and pn,i ∈ M[n,i](C(Xn,i)) is a projection. Recall in [G1], Gong proved that a simple AH algebra with uniformly bounded dimension of local spectra, i.e. supn,i dim(Xn,i) < ∞ can be rewritten as AH inductive limit with spaces being [0, 1], S1, S2, TII,k, TIII,k, where each TII,k (or TIII,k respectively) is two -- dimensional (or three -- dimensional respectively) connected simplicial complexes with H 2(TII,k) = Z/kZ and H 1(TII,k) = 0 (or with H 3(TIII,k) = Z/kZ and H 1(TIII,k) = 0, H 2(TIII,k) = 0). For any positive k the dimension drop interval algebra Ik is defined as 2.9. [GJL1] Let X be a compact space and ψ : C(X) → P Mk1(C(Y ))P (rank(P ) = k) be a unital homomorphism. For any point y ∈ Y , there are k mutually orthogonal rank-1 projections p1, p2, · · · , pk with Ik =(cid:8)f ∈ Mk(C[0, 1])(cid:12)(cid:12)f (0) = λ1k, f (1) = µ1k, λ, µ ∈ C(cid:9). pi = P (y) and(cid:8)x1(y), x2(y), · · · , xk(y)(cid:9) ⊂ X (may be repeat) such that ψ(f )(y) = kPi=1 C(X). We denote the set(cid:8)x1(y), x2(y), · · · , xk(y)(cid:9) (counting multiplicities), by Spψy. We shall call Spψy For any f ∈ Ik, let function f : [0, 1] → CF Mk(C) (disjoint union) be defined by the spectrum of ψ at the point y. f (xi(y))pi, ∀f ∈ kPi=1 f (t) = λ, if t = 0 and f (0) = λ1k µ, if t = 1 and f (1) = µ1k f (t), if 0 < t < 1 3 That is, f (t) is the value of irreducible representation of f corresponding to the point t. Similarly, for f ∈ Ml(Ik), we can define f : [0, 1] −→ Ml(C)F Mlk(C), by f (t) = a, if t = 0 and f (0) = a ⊗ 1k b, if t = 1 and f (1) = b ⊗ 1k f (t), if 0 < t < 1 Suppose that φ : Ik → P Mn(C(Y ))P is a unital homomorphism. Let r = rank(P ). For each y ∈ Y , there are t1, t2, · · · , tm ∈ [0, 1] and a unitary u ∈ Mn(C) such that P (y) = u 1rank(P ) 0 0 0 ! u∗ f (t1) f (t2) . . .   f (tm) 0n−r and (∗) φ(f )(y) = u for all f ∈ Ik. u∗ ∈ P (y)Mn(C)P (y) ⊂ Mn(C) We define the set Spφy to be the points t1, t2, · · · , tm with possible fraction multiplicity. If ti = 0 or 1, For example if we assume t1 = t2 = t3 = 0 < t4 ≤ t5 ≤ · · · ≤ tm−2 < 1 = tm−1 = tm, then Spφy =(cid:8)0∼ 1 k , 0∼ 1 k , 0∼ 1 k , t4, t5, · · · , tm−2, 1∼ 1 k , 1∼ 1 k(cid:9), which can also be written as k , t4, t5, · · · , tm−2, 1∼ 2 k(cid:9) Spφy =(cid:8)0∼ 3 Here we emphasize that, for t ∈ (0, 1), we do not allow the multiplicity of t to be non-integral. Also for 0 or 1, the multiplicity must be multiple of 1 k (other fraction numbers are not allowed). Let ψ : C[0, 1] → P Mn(C(Y ))P be defined by the following composition ψ : C[0, 1] ֒→ Ik φ −→ P Mn(C(Y ))P, where the first map is the canonical inclusion. Then we have Spψy = {Spφy}∼k -- that is, for each element t ∈ (0, 1), its multiplicity in Spψy is exactly k times of the multiplicity in φy. Recall that for A = Ml(Ik), every point t ∈ (0, 1) corresponds to an irreducible representation πt, defined by πt(f ) = f (t). The representations π0 and π1 defined by π0 = f (0) and π1 = f (1) are no longer irreducible. We use 0 and 1 to denote the corresponding points for the irreducible representa- tions. That is, π0(f ) = f (0) and π1(f ) = f (1). 4 Or we can also write f (0) , f (0) and f (1) , f (1). Then the equation (∗) could be written as  f (t1) f (t2) . . . f (tm) 0n−r u∗  φ(f )(y) = u equivalence. where some of ti may be 0 or 1. In this notation, f (0) is equal to diag(cid:0) f (0), f (0), · · · , f (0) } k(cid:9) =(cid:8)0, 0, 0, t4, t5, · · · , tm−2, 1, 1(cid:9) Spφy =(cid:8)0∼ 1 Under this notation, we can also write 0∼ k as 0. Then the example of Spφy can be written as k , 0∼ 1 k , 0∼ 1 k , t4, t5, · · · , tm−2, 1∼ 1 k , 1∼ 1 k {z 1 2.10. We use HD to denote all C-algebras C =L C i, where each C i is of the forms Ml(Ik) or P MnC(X)P with X being one of the spaces {pt}, [0, 1], S1, TII,k. Each block C i will be called a basic HD block or a basic building block. (cid:1) up to unitary By AHD algebra, we mean the inductive limit of A1 φ1,2−→ A2 φ2,3−→ · · · −→ An · · · −→ · · · , where An ∈ HD for each n. For an AHD inductive limit A = lim(An, φn,m), we write An =Ltn n = M[n,i](Ikn,i ). For convenience, even for a block Ai i=1 Ai n, where Ai n = Pn,iM[n,i](C(Xn,i))Pn,i n = M[n,i](Ikn,i ), we still use Xn,i for Sp(Ai n) = n is regarded as a homogeneous algebra or a sub-homogeneous algebra over Xn,i. of Ai [0, 1] -- that is, Ai Definition 2.11. ([EG2], [DG]) Let X be a compact connected space and let P ∈ MN (C(X)) be a projection of rank n. The weak variation of a finite set F ⊂ P MN (C(X))P is defined by ω(F ) = sup π,σ inf u∈U(n) max a∈F kuπ(a)u∗ − σ(a)k, where π, σ run through the set of irreducible representations of P MN C(X)P into Mn(C). For F ⊂ Mr(Ik), we define ω(F ) = ω(ı(F )), where ı : Mr(Ik) −→ Mrk(C[0, 1]) is the canonical embedding and ı(F ) is regarded as a finite subset of Mrk(C[0, 1]). Let A be a basic HD block, a finite set F ⊂ A is said to be weakly approximately constant to within ε if ω(F ) < ε. 2.12. Let K(A) = K∗(A) ⊕L∞ K(A)s (see 4.1 of [DG]). It is well known that k=2 K∗(A, Z/k) be as in [DG]. Let Λ be the Bockstein operation between As in [DG], let K∗(A, Z ⊕ Z/k) ∼= K0(A ⊗ C(TII,k × S1)). K∗(A, Z ⊕ Z/k)+ ∼= K0(A ⊗ C(TII,k × S1))+. and let K(A)+ to be the semigroup of K(A) generated by {K∗(A, Z ⊕ Z/kZ)+, k = 2, 3, · · · }. 5 2.13. Let HomΛ(K(A), K(B)) be the homomorphism between K(A) and K(B) compatible with Bockstein operation Λ. Associativity of Kasparov KK-product gives a map Γ : KK(A, B) → HomΛ(K(A), K(B)) Recall that every element α ∈ KK(A, B) defines a map α∗ ∈ HomΛ(K(A), K(B)). That is, it gives a sequence of homomorphisms αi : Ki(A) −→ Ki(B) f or i = 0, 1, and αi k : Ki(A, Z/k) −→ Ki(B, Z/k) f or i = 0, 1, which are compatible with the Bockstein operation Λ. 2.14. Let K(A)+ be defined in 4.6 of [DG]. Notice that, from [DG, 4.7], K(A)+ ⊂ K(A)+, where for any unital C-algebra B, let KK(A, B)+ = {α ∈ KK(A, B), α(K(A)+) ⊆ K(B)+}, KK(A, B)D,+ = {α ∈ KK(A, B)+, α[1A] ≤ [1A]}. 2.15. Let A = P M•C(TII,k)P , where we use • to denote any possible positive integers. From 5.14 of [G3], we know that an element α ∈ KK(A, B) is completely determined by α0 : K0(C(TII,n)) −→ K0(B) and k : K1(C(TII,k), Z/k) −→ K1(B, Z/k), k = 0, 1, 2, · · · . α1 For any positive integer k ≥ 2, denote K(k)(A) = K0(A) ⊕ K1(A) ⊕ K0(A, Z/k) ⊕ K1(A, Z/k) and K(k)(A)+ = K(A)+ ∩ K(k)(A) Then for A = P M•C(TII,k)P , an element α ∈ KK(A, B) is in KK(A, B)+ if and only if α(K(k)(A)+) ⊂ K(k)(B)+, k = 0, 1. Note that K(k)(A)+ is finitely generated. 2.16. Note that, for any C-algebra A, it follows from Kunneth Theorem that K0(A ⊗ C(Wk × S1)) = K0(A) ⊕ K1(A) ⊕ K0(A, Z/k) ⊕ K1(A, Z/k) where Wk = TII,k. One can choose finite set P ⊂ M•(A ⊗ C(Wk × S1)) of projections such that {[P ] ∈ K0(A) ⊕ K1(A) ⊕ K0(A, Z/k) ⊕ K1(A, Z/k)P ∈ P(A)} ⊂ K(A) generate K(k)(A) = K0(A) ⊕ K1(A) ⊕ K0(A, Z/k) ⊕ K1(A, Z/k), where Wk = TII,k. If we choose finite set G(P) ⊂ A large enough and δ(P) > 0 small enough, then every G(P) − δ(P) multiplicative contraction φ : A → B determines a map φ∗ : PK(A) → K(B) which is compatible with the Bockstein operation Λ (see [GL]). If A = P M•C(TII,k)P , then it also defines a KK element [φ] ∈ KK(A, B). Let A = P M•C(TII,k)P , recall that for G ⊇ G(P), δ ≤ δ(P), a G − δ multiplicative map φ : A → B is called a quasi PK homomorphism, if there is a homomorphism ψ : A → B satisfying φ∗ = ψ∗ : PK(A) → K(B). If φ is a quasi PK homomorphism, then [φ] ∈ KK(A, B)+. Suppose that A = P M•C(TII,k)P and suppose that B = Bi is a direct sum of basic HD bulding blocks. Let α ∈ KK(A, B), then α ∈ KK(A, B)+ if and only if for each i ∈ {1, 2, · · · , n}, αi ∈ KK(A, Bi) is either nLi=1 6 zero or αi(1C(TII,k)) > 0 in K0(Bi). Recall that in [DG], a KK element α ∈ KK(A, B = Bi) is called m-large if rank(αi(1A)) ≥ m · rank(1A), for all i ∈ {1, 2, · · · , n}. By 5.5 and 5.6 of [DG] if α ∈ KK(A, B)+ is 6-large and α([1A]) ≤ [1B], then α can be realized by a homomorphism from A to B. nLi=1 2.17. Recall that the scale of A, denoted byP A, is a subset of K0(A) consisting of [p], where p ∈ A is a projection. As all the C-algebras A in this paper have cancellation of projections, if A has unit 1A, then For two C-algebras A, B, by a "homomorphism" α from (K(A), K(A)+,P A) to (K(B), K(B)+,P B), it means a system of maps X A =(cid:8)x ∈ K0(A); 0 ≤ x ≤ [1A](cid:9). αi k : Ki(A, Z/k) → Ki(B, Z/k); i = 0, 1, k = 0, 2, 3, 4, · · · which are compatible with Bockstein operation and α =Lk,i αi 0(P A) ⊆P B. α0 k satisfies α(K(A)+) ⊆ K(B)+ and finally 2.18. For a unital C-algebra A, let T A denote the space of tracial states of A, i.e. τ ∈ T A, if and only if τ is a positive linear map from A to the complex plane C, with τ (xy) = τ (yx) and τ (1A) = 1. Af f T A is the Banach space of all the continuous affine maps from T A to C. (In most references, Af f T A is defined to be the set of all the affine maps from T A to R. Our Af f T A is a complexification of the standard Af f T A.) The element 1 ∈ Af f T A, defined by 1(τ ) = 1, for all τ ∈ T A, is called the unit of Af f T A. Af f T A, together with the positive cone Af f T A+ and the unit element 1 forms a scaled ordered complex Banach space. (Notice that for any element x ∈ Af f T A, there are x1, x2, x3, x4 ∈ Af f T A+ such that x = x1 − x2 + ix3 − ix4.) There is a natural homomorphism ρA : K0(A) → Af f T A defined by ρA([p])(τ ) =Pn and [p] ∈ K0(A) represented by projection p = (pi,j) ∈ Mn(A). i=1 τ (pii) for τ ∈ T A Any unital homomorphism φ : A → B induces a continuous affine map T φ : T B → T A, It turns out that T φ induces a unital positive linear map If φ : A → B is not unital, we still have the positive linear map Af f T φ : Af f T A → Af f T B. Af f T φ : Af f T A → Af f T B. but it will not preserve the unit 1, only has property Af f T φ(1Af f T A) ≤ 1Af f T B. 2.19. If α : (K(A), K(A)+,P A) → (K(B), K(B)+,P B) is a homomorphism as in 2.17, then for each projection p ∈ A, there is a projection q ∈ B such that α[p] = [q]. Notice that for all the C-algebras A considered in this paper, the following is true: if p1, p2 are projections and [p1] = [p2] in K0(A), then there is u ∈ A with up1u∗ = p2. Therefore both T (pAp) and T (qBq) depend only on the class [p] ∈ K0(A) and [q] ∈ K0(B). For the classes [p] ∈P A(⊂ K0(A)), T (pAp) is taken as part of the invariant Inv0(A). For two classes [p] ∈P A, [q] ∈P B, with α([p]) = [q], we will consider the system of continuous affine maps ξp,q : T (qBq) → T (pAp). Such system of maps is said to be compatible if for any two projections p1 ≤ p2 with α([p1]) = [q1], α([p2]) = [q2] and q1 ≤ q2, the following diagram is 7 Af f T (p1Ap1) ξp1,q1 −−−−→ Af f T (q1Bq1) Af f T (p2Ap2) ξp2,q2 −−−−→ Af f T (q2Bq2) ıTy ıTy commutative: where the horizontal maps are induced by ξp1,q1 and ξp2,q2 respectively, and the vertical maps are induced by the inclusions p1Ap1 → p2Ap2, q1Bq1 → q2Bq2. 2.20. We denote (cid:0)K(A), K(A)+,P A, {T (pAp)}[p]∈P A(cid:1) by Inv0(A). By a "map" between the invariants (cid:0)K(A), K(A)+,P A, {T (pAp)}[p]∈P A(cid:1) and(cid:0)K(B), K(B)+,P B, {T (qBq)}[q]∈P B(cid:1), we mean a map α :(cid:16)K(A), K(A)+,X A(cid:17) →(cid:16)K(B), K(B)+,X B(cid:17) as in 2.17 and maps ξp,q : T (qBq) → T (pAp) which are compatible as 2.19. We denote this map by (α, ξ) :(cid:16)K(A), K(A)+,X A, {T (pAp)}[p]∈P A(cid:17) →(cid:16)K(B), K(B)+,X B, {T (qBq)}[q]∈P B(cid:17) or simply (α, ξ) : Inv0(A) → Inv0(B) 2.21. Let A be a unital C-algebra. Let U (A) denote the group of unitaries of A and let U0(A) denote the connected component of 1A in U (A). Let DU (A) and DU0(A) denote the commutator subgroups of U (A) and U0(A), respectively. (Recall that the commutator subgroup of a group G is the subgroup generated by all elements of the form aba−1b−1, where a, b ∈ G.) One introduces the following metric DA on U (A)/DU (A) (see [NT,§3]). For u, v ∈ U (A)/DU (A) DA(u, v) = inf{kuv∗ − ck : c ∈ DU (A)} where, on the right hand side of the equation, we use u, v to denote any elements in U (A), which represent the elements u, v ∈ U (A)/DU (A). Denote the extended commutator group DU +(A), which is generated by DU (A) ⊂ U (A) and the set (cid:8)e2πitp = e2πitp + (1 − p) ∈ U (A)(cid:12)(cid:12)t ∈ R, p ∈ A is a projection(cid:9). Let ^DU (A) denote the closure of DU +(A). That is, ^DU (A) = DU +(A) 2.22. Let A be a unital C-algebra. Let Af f T A and ρA : K0(A) → Af f T A be as defined as in 2.18. For simplicity, we will use ρK0(A) to denote the set ρA(K0(A)). The metric dA on Af f T A/ρK0(A) is defined as follows (see [NT, §3]). ′ Let d denote the quotient metric on Af f T A/ρK0(A), i.e, for f, g ∈ Af f T A/ρK0(A), Define dA by Obviously, dA(f, g) ≤ 2πd ′ ′ d (f, g) = inf{kf − g − hk, h ∈ ρK0(A)}. dA(f, g) = (f, g). 2, ′ e2πid (f,g) − 1, ′ if d ′ if d (f, g) ≥ 1 2 , (f, g) < 1 2 . 8 Let ^ρK0(A) denote the closed real vector space spanned by ρK0(A). That is, For any u, v ∈ U (A)/ ^DU (A), define ^ρK0(A) :=nX λiφi(cid:12)(cid:12)(cid:12)λi ∈ R, φi ∈ ρK0(A)o. DA(u, v) = inf{kuv∗ − ck : c ∈ ^DU (A)}. denote the quotient metric on Af f T A/ ^ρK0(A), that is, (f, g) = inf{kf − g − hk, h ∈ ^ρK0(A)}, ′ f, g ∈ Af f T A/ ^ρK0(A). ′ Let ed Define edA by ed edA(f, g) = 2, e2πied ′ (f,g) − 1, (f, g) ≥ 1 2 , (f, g) < 1 2 . ′ ′ if ed if ed 2.23. Let ^SU (A) :=nx ∈ U (A)(cid:12)(cid:12)(cid:12)xn ∈ ^DU (A) f or some n ∈ Z+ \ {0}o. and for simpler, denote gρK0(A) , ^ρK0(A), gDU (A) = ^DU (A),gSU (A) = ^SU (A). A unital homomorphism φ : A → B induces a contractive group homomorphism φ♮ : U (A)/gSU (A) −→ U (B)/gSU (B). If φ is not unital, then the map φ♮ : U (A)/gSU (A) −→ U (φ(1A)Bφ(1A))/gSU (φ(1A)Bφ(1A)) is induced by the corresponding unital homomorphism. In this case, φ also induces the map ı∗ ◦ φ♮ : U (A)/gSU (A) −→ U (B)/gSU (B), which is denoted by φ∗ to avoid confusion. 2.24. In [GJL1] and [GJL2], denote (cid:16)K(A), K(A)+,X A,(cid:8)Af f T (pAp)(cid:9)[p]∈P A,(cid:8)U (pAp)/gSU (pAp)(cid:9)[p]∈P A(cid:17) by Inv(A). By a map from Inv(A) to Inv(B), one means α :(cid:16)K(A), K(A)+,X A(cid:17) −→(cid:16)K(B), K(B)+,X B(cid:17) as in 2.17, and for each pair ([p], [p]) ∈P A ×P B with α([p]) = [p], there are an associate unital positive linear map ξp,p : Af f T (pAp) −→ Af f T (pBp) and an associate contractive group homomorphism χp,p : U (pAp)/gSU (pAp) −→ U (pBp)/gSU (pBp) 9 satisfying the following compatibility conditions: (a) If p < q, then the diagrams Af f T (pAp) −−−−→ Af f T (pBp) ıTy ξp,p ξq,q ıTy Af f T (qAq) −−−−→ Af f T (qBq) and commutes, where the vertical maps are induced by inclusions. (b) The following diagram commutes ı∗y U (pAp)/gSU (pAp) U (qAq)/gSU (qAq) αy K0(pAp) K0(pBp) χp,p −−−−→ U (pAp)/gSU (pBp) −−−−→ U (qAq)/gSU (qBq) ı∗y χq,q −−−−→ Af f T (pAp) ρ ρ ξp,py −−−−→ Af f T (pBp) and therefore ξp,p induces a map (still denoted by ξp,p): (The commutativity of (III) follows from the commutativity of (I), by 1.20 of [Ji-Jiang]. So this is not an ξp,p : Af f T (pAp)/gρK0(pAp) −→ Af f T (pBp)/gρK0(pBp) extra requirement.) (c) The following diagrams (I) (II) (III) (IV ) (V ) and Af f T (pAp)/gρK0(pAp) −−−−→ U (pAp)/gSU (pAp) ξp,py Af f T (pBp)/gρK0(pBp) −−−−→ U (pBp)/gSU (pBp) U (pAp)/gSU (pAp) −−−−→ K1(pAp)/torK1(pAp) χp,py U (pBp)/gSU (pBp) −−−−→ K1(pBp)/torK1(pBp) χp,py α1y commute, where α1 is induced by α. We will denote the map from Inv(A) to Inv(B) by (α, ξ, χ) :(cid:16)K(A), K(A)+,X A,(cid:8)Af f T (pAp)(cid:9)[p]∈P A,(cid:8)U (pAp)/gSU (pAp)(cid:9)[p]∈P A(cid:17) −→ (cid:16)K(B), K(B)+,X B,(cid:8)Af f T (pBp)(cid:9)[p]∈P B,(cid:8)U (pBp)/gSU (pBp)(cid:9)[p]∈P B(cid:17) Note that Inv0 is part of Inv(A). 10 3. Local approximation lemma Proposition 3.1. Let A = P M•C(TII,k)P or Ml(Ik) and P be as in 2.16. For any finite set F ⊂ A, ε > 0, there exist a finite set G ⊂ A (G ⊃ G(P) large enough), a positive number δ > 0(δ < δ(P) small enough) such that the following statement is true: If B ∈ HD, φ, ψ ∈ M ap(A, B) are G − δ multiplicative and φ∗ = ψ∗ : PK(A) → K(B), then there is a homomorphism ν ∈ Hom(A, ML(B)) defined by point evaluations and there is a unitary u ∈ ML+1(B) such that k(φ ⊕ ν)(a) − u(ψ ⊕ ν)(a)u∗k < ε for all a ∈ F Proof. For the special case that A = P M•C(TII,k)P and B is homogeneous, this is [G3, Theorem 5.18]. The proof of the general case is completely same (see 2.16 above). Note that, the calculation of K0(Π∞ K0(Π∞ some of Bn being of the form Ml(Ik). n=1 Bn, Z/k) in 5.12 of [G3] works well for the case that n=1Bn/L∞ n=1Bn/L∞ n=1 Bn), n=1 Bn, Z/k) and K1(Π∞ n=1Bn/L∞ Lemma 3.2. Let A = P M•(C(X))P or Ml(Ik). Let F ⊂ A be approximately constant to within ε (i.e., ω(F ) < ε). Then for any two homomorphisms φ, ψ : A → B defined by point evaluations with K0φ = K0ψ, there exists a unitary u ∈ B such that kφ(f ) − uψ(f )u∗k < 2ε ∀f ∈ F . (cid:3) Proof. The case A = P M•(C(X))P is Lemma 3.5 of [GJLP2]. For the case A = Ml(Ik), one can also find a homomorphism φ : Ml(C) → B such that ′ kφ(f ) − φ ′ (f (0))k < ε ∀f ∈ F, where f (0) as in 2.9. Then one follows the same argument as the proof of Lemma 3.5 of [GJLP2] to get the result. (cid:3) Lemma 3.3. Let A = P M•(C(X))P or Ml(Ik), ε > 0, finite set F ⊂ A with ω(A) < ε. There exist a finite set G ⊂ A (G ⊃ G(P)), a number δ > 0 (δ < δ(P))and a positive integer L such that the following statement is true. If B ∈ HD, and φ, ψ ∈ M ap(A, B)1 are G − δ multiplicative and φ∗ = ψ∗ : PK(A) → K(B) and ν : A → M∞(B) is a homomorphism defined by point evaluations with ν([1A]) ≥ L · [1B] ∈ K0(B), then there is a unitary u ∈ (1B ⊕ ν(1A))M∞(B)(1B ⊕ ν(1A)) such that k(φ ⊕ ν)(f ) − u(ψ ⊕ ν)(f )u∗k < 5ε ∀f ∈ F. Proof. Let L1 be as in Proposition 3.1 and L = 2L1. Since ν([1A]) ≥ L·[1B] = 2L1 ·[1B] ∈ K0(B), there is a projection Q < ν(1A) such that Q is equivalent to 1ML1 (B). By Proposition 3.1, there exist a homomorphism ν1 : A → QM∞(B)Q defined by point evaluation and a unitary w ∈ (1B ⊕ ν1(1A))M∞(B)(1B ⊕ ν1(1A)) such that k(φ ⊕ ν1)(f ) − w(ψ ⊕ ν1)(f )w∗k < ε ∀f ∈ F. Since ν1 (and ν respectively) is homotopic to a homomorphism factoring through Mrank(P )(C) (for A = P M•(C(X))P ) or factoring through Ml(C) (for A = Ml(Ik)), there is a unital homomorphism ν2 : A → (ν(1A) − ν1(1A))M∞(B)(ν(1A) − ν1(1A)) such that K0(ν1 ⊕ ν2) = K0(ν). 11 By Lemma 3.2, ν is approximately unitarily equivalent to ν1 ⊕ ν2 to within 2ε on F . Hence φ ⊕ ν is approximately unitarily equivalent to ψ ⊕ ν on F to within 2ε + ε + 2ε = 5ε. (cid:3) Lemma 3.4. Let A = P M•C(TII,k)P , and P ⊂ M•(A⊗C(Wk ×S1)) as in 2.16. And let B = lim (Bn, ψnm) n→∞ be a unital simple inductive limit of direct sums of HD building blocks. Let φ : A → B be a unital homomorphism. It follows that for any G ⊃ G(P) and δ < δ(P), there is a Bn and a unital G−δ multiplicative contraction φ1 : A → Bn which is a quasi PK-homomorphism such that kψn,∞ ◦ φ1(g) − φ(g)k < δ 2 multiplicative (see Lemma 4.40 in [G3]). Now if φ Proof. There is a finite set G1 and δ1 > 0 such that if a complete positive linear map ψ : A → C (C is a C-algebra) and a homomorphism φ : A → C satisfy that kφ(g) − ψ(g)k < δ1 for all g ∈ G1, then ψ is G − δ (g) − φ(g)k < δ for all : A → Bn g ∈ G1, then ψl,∞ ◦ φ is G − δ multiplicative. By replacing G by G ∪ G1 and δ by min(δ, δ1), we only need to construct a unital quasi PK-homomorsim φ1 : A → Bn such that 2 multiplicative. Hence for some n > l (large enough), φ1 = ψi,n ◦ φ : A → Bl satisfies kψl,∞ ◦ φ ′ ′ is G − δ ′ ′ (∗) kψn,∞ ◦ φ1(g) − φ(g)k < δ, ∀g ∈ G. Furthermore, if φ1 is a G − δ multiplicative map satisfying the above condition, then KK(φ1) × KK(ψn,∞) = KK(φ) ∈ KK(A, B)+. Note that K(k)(A)+ is finitely generated, there is an m > n such that KK(φ1) × KK(ψn,m1) ∈ KK(A, Bm1)+. Since B is simple, for certain m > m1, KK(ψn,m ◦ φ1) is 6-large and therefore can be realized by a homomorphism. That is, replacing φ1 by ψn,m ◦ φ1, we get a quasi PK -homomorphism. Thus the proof of the lemma is reduced to the construction of φ1 to satisfy (∗). Since B = lim n→∞ (Bn, ψnm), there is a Bn and finite set F ⊂ Bn such that G ⊂ δ ψn,∞(F ). Since Bn(⊂ B) is a nuclear C-algebra, there are two complete positive contractions λ1 : Bn → MN (C) and λ2 : MN (C) → Bn such that 3 kλ2 ◦ λ1(f ) − f k < δ 3 ∀f ∈ F. Since Bn is a subalgebra of B, by Arversons Extension Theorem, one can extend the map λ1 : Bn → MN (C) to β1 : B → MN (C) such that β1 ◦ ψn,∞ = λ1. One can verify that φ1 = λ2 ◦ β1 ◦ φ : A → Bn satisfies the condition (∗) as below. For g ∈ G, there is an f ∈ F such that kφ(g) − ψn,∞(f )k < δ 3 , and therefore kψn,∞ ◦ φ1(g) − φ(g)k = k(ψn,∞ ◦ λ2 ◦ β1 ◦ φ1)(g) − φ(g)k ≤ kψn,∞(λ2 ◦ β1 ◦ φ1(g)) − ψn,∞(λ2 ◦ β1 ◦ ψn,∞(f ))k + k(ψn,∞ ◦ λ2 ◦ λ1(f ) − ψn,∞(f )k + k(ψn,∞(f ) − φ(g)k ≤ δ 3 + δ 3 + δ 3 = δ. 12 This ends the proof of the lemma. (cid:3) Lemma 3.5. For any finite set F ⊂ P M•C(TII,k)P , B with ω(F ) < ε, there are a finite set G ⊃ G(P), a positive number δ < δ(P) and a positive integer L (this L will be denoted by L(F, ε) later), such that if C is a HD basic building block, p, q ∈ C are two projections with p + q = 1C , and φ0 : B → pCp and ϕ1 : B → qCq are two maps satisfying the following conditions: (1) φ0 is quasi PK homomorphism and G − δ multiplicative, and φ1 is defined by point evaluations (or equivalently, factoring through a finite dimensional C-algebra), and (2) rank(q) ≥ L rank(p), then there is a homomorphism φ : B → C such that kφ0 ⊕ φ1(f ) − φ(f )k < 5ε for all f ∈ F . Proof. Since φ0 is a quasi PK-homomorphism and is G − δ multiplicative, there is a homomorphism φ 0∗ = φ0∗ : PK(B) → K(C). By Lemma 3.3, there is unitary u ∈ C such that 0 : B → pCp such that φ ′ ′ k(φ0 ⊕ φ1)(f ) − u(φ 0 ⊕ φ1)(f )u∗k < 5ε ′ The homomorphism φ = Adu∗ ◦ (φ ′ 0 ⊕ φ1) is as desired. ∀f ∈ F. (cid:3) Lemma 3.6. Let ε1 > ε2 > · · · > εn > · · · be a sequence withP εi < +∞. Let A be a simple AH algebra with no dimension growth (as in [EGL] and [Li4]). Then A can be written as an AHD inductive limit A1 = t1Mi=1 t1Mi=1 with the unital subalgebras Bn(=L Bi n) ⊂ An(=L Ai with Fn =L F i n ⊂ Gn =L Gi n is not of type TII, then Bi 1 −→ A2 = n and F i (1) If Ai n = F i n = Ai Ai n = π0(F i n) ⊂ Gi with F i other than type TII, and π0 : Bi n) ⊕ π1(F i n and ω(π0(F i n → P Ai nP , B0,i Ai 2 −→ · · · n such that the following statements hold: n) (that is, Bi n ⊂ Ai n) and Fn ⊂ Bn and Gn ⊂ An n. If Ai n)) < ε, where Di n ⊂ Ai n is of type TII, then Bi n n is a direct sum of the HD building blocks n = P Ai nP ⊕Di n and π1 : Bi n → Di n are canonical projections; (2) Gi n generates Ai n, φn,n+1(An) ⊂ Bn+1, φn,n+1(Gn) ⊂ Fn+1, and ball of A; φn,∞(Gn) = ∞Sn=1 ∞Sn=1 φn,∞(Fn) =unit (3) Suppose that both Ai n and Aj n+1 are of type TII and φ , π0 ◦ φi,j n ) = p0 ⊕ p1 ∈ B0,j φ(1B0,i such that φ1 is defined by point evaluations (or equivalently, φ1(B0,i p1B0,j n+1 and φ = φ0 ⊕ φ1 with φ0 ∈ Hom(B0,i n+1p1) and n , p0B0,j : B0,i n → B0,j n,n+1B0,i n+1. Then n+1p1)1 n ) is a finite dimensioned sub-algebra of n+1p0)1, φ1 ∈ Hom(B0,i n , p1B0,j n where L(π0(F i n), εn) is as in Lemma 3.5 (note that ω(π0(F i n)) < εn). [p1] ≥ L(π0(F i n), εn) · [p0], Proof. From [Li4], we know that A can be written as an inductive limit of the direct sums of HD building blocks A = lim n→∞ (fAn, ψnm). In our construction, we will choose An = gAkn and homomorphisms φn,n+1 : An = 13 gAkn → An+1 = ^Akn+1 satisfying KK(φn,n+1) = KK(ψkn,kn+1) and Af f T φn,n+1 is close to Af f T ψkn,kn+1 within any pregiven small number on a pregiven finite set, in such a way that φn,n+1 also satisfies the desired condition as in the lemma with certain choices of subalgebras Bn ⊂ An and and finite subsets Fn ⊂ Bn and Gn ⊂ An. Suppose that we already have An = gAkn with the unital sub-algebra Bn = ⊕iBi n) ⊂ An = ⊕iAi n, and two subsets Fn ⊂ Bn, Gn ⊂ An. as required in the lemma. We will construct, roughly, the next algebra An+1 = ^Akn+1, the sub-algebra Bn+1 ⊂ An+1, two subsets Fn+1 ⊂ Bn+1, Gn+1 ⊂ An+1, and the homomorphism φn,n+1 : An → An+1 to satisfy all the requirements in the lemma. n = ⊕i(B0,i n ⊕ Di l > kn large enough, and for each block Ai into three parts φ1 Applying the decomposition theorem-[G3, Theorem 4.37], as in the proof of the main theorem in [Li4], for kn,l can be decomposed 0 ⊕ φ1 ⊕ φ2, roughly described as below: There are mutually orthogonal projections l Qk)1, l Q0)1 possessing the ) = Q0 + Q1 + Q2, there are two homomorphisms φk ∈ Hom(Ai (k = 1, 2) and a sufficient multiplicative quasi PK homomorphism φk ∈ M ap(Ai following properties: Q0,Q1,Q2 ∈fAj of type TII, the homomorphism ψi,j n = gAi n, QkfAj l with ψi,j kn,l(1Ai kn n n, Q0fAj (i) φ1 0 ⊕ φ1 ⊕ φ2 is close to ψi,j kn,l to within εn on Gi n; (ii) φ1 0(1B0,i n ) is a projection and [Q1] ≥ L(π0(F i n), ε) · [Q0] and φ1(1B0,i n ) ≥ L(π0(F i n), ε) · [φ1 0(1B0,i n )]; homomorphism φ0 ∈ Hom(Ai (iii) φ1 is defined by point evaluation at a dense enough finite subset of Sp(gAi 2−→fAj (iv) φ2 factors through an interval algebra Di Then we can choose kn+1 = l and An+1 = ^Akn+1. Note that φ1 l Q0)1, we have ω(φ0 ⊕ φ1(Gi n, Q0fAj 1 as φ2 = ξi n)) < εn+1; ξi 1−→ Di 1 1 : Ai n 2 ◦ ξi l . ξi kn can choose φ0 : gAi kn ) = φ0(1B0,i φ1 0(1B0,i n n ^ Ai ) kn+1 0(1 gAi → φ1 ). Modify ψkn,kn+1 by replacing ψi,j φ1 0(1 gAi kn kn ) such that KK(φ1 0 is a quasi PK homomorphism, one 0 ⊕ φ1 ⊕ φ2) = KK(φ0 ⊕ φ1 ⊕ φ2) with kn,kn+1 by φ0 ⊕ φ1 ⊕ φ2 to define φn,n+1 : An → An+1. ) such that for any Fix j with Aj n+1 = Aj kn+1 being of type TII. Let I0 = { i Ai is not of type TII }. Let Pi = (φ0 ⊕ φ1)(1Ai n ) ∈ Aj is of type TII}, I1 = { i Ai Pi ∈ Aj n+1 and B0,j , P Aj n+1 n+1P , n = kn n = gAj n+1, let P = Li∈I0 kn gAj and let Dj n+1 = Li∈I0 1) ⊕ Li∈I1 2(Di ξi n+1 = B0,j ^ φi,j kn,kn+1 ) ⊂ ^ Aj kn+1 kn (gAj n+1 ⊕ Dj than type TII . Let Bj not of type TII . Choose Gn+1 = ⊕Gj F j n+1 = Gj Note that KK(ψkn,kn+1) = KK(φn,n+1). Since the rank of φ1 n+1 is of type TII, let F j n+1 = φ−,j n+1. If Aj n+1 ⊂ Aj n+1 if Aj n,n+1(Gn). which is a direct sum of HD building blocks other n+1(⊃ φn,n+1(Gn)) sufficiently large. If Aj n+1 is of type TII; and Bj n+1 = Aj n+1 is n+1 is not of type TII, let n+1 if Aj 0(1 gAi kn ) = φ0(1 gAi kn ) is much smaller than the ranks of φ1(1 gAi,j kn ) and φ2(1 gAi,j kn ), Af f T φn,n+1 is very close to Af f T ψkn,kn+1. Hence A has the same Elliott invariant as A = lim(gAkn, ψkn,km ). Hence A ′ ∼= A. ′ = lim(An,n+1, φn,m) (cid:3) 14 Corollary 3.7. Let A be a simple AH algebra with no dimension growth. And let P1, P2, · · · , Pk ∈ A be a set of mutually orthogonal projections. Then one can write A as inductive limit A = lim(An, φn,m) with mutually orthogonal projections P 0 k ∈ A1 such that for each i, φ1,∞(P 0 i ) = Pi and 2 , · · · , P 0 1 , P 0 PiAPi = lim i )Anφ1,n(P 0 i ), φn,mφ1,n(P 0 i )Anφ1,n(P 0 n→∞(cid:16)φ1,n(P 0 i )(cid:17) satisfies the properties of Lemma 3.6. Proof. In the proof of Lemma 3.6, one can assume that there are P i ) = Pi for all i = 1, 2, · · · , k. Then the construction can be carried out to get our conclusion. Note that, applying 2, · · · , P 1, P ′ ′ ′ ′ k ∈ fA1 with ψ1,∞(P Lemma 1.6.8 of [G3], one can strengthen [G3, Theorem 4.37] such that the following is true: For a set of pre-given orthogonal projections p1, p2, · · · , pk ∈ An, one can further require that ψ0 ∈ M ap(An, Q0AmQ0) satisfies that ψ0(pi) are projections and (ψ0 ⊕ ψ1 ⊕ ψ2)(pi) = φn,m(pi) for i = 1, 2, · · · , k. (cid:3) The following lemma is the main technique lemma of this section. Lemma 3.8. Let A,A ′ be simple AHD inductive limit algebras with A1 φ1,2−→ A2 A and A 1 ′ ψ1,2−→ A ′ 2 ψ2,3−→ · · · −→ A ′ n · · · −→ A ′ being described in Lemma 3.6. Let Λ : A → A homomorphism. Let F ⊂ Am be a finite set and ε > 0. Then there is an A Λ1 : Am → A l such that ′ φ2,3−→ · · · −→ An · · · −→ be a l and a homomorphism ′ ′ kΛ ◦ φm,∞(f ) − ψl,∞ ◦ Λ1(f )k < ε ∀f ∈ F. Fn and 5εn < ε Proof. One can choose n > m large enough such that φm,n(F ) ⊂ ε Bn, where Bn is as in Lemma 3.6. We will construct a homomorphism φ : Bn → A 4 8 . Note that φm,n(Am) ⊂ l such that ′ (∗) kΛ ◦ φn,∞(f ) − ψl,∞(φ(f ))k < ε 2 ∀f ∈ Fn ⊂ Bn ⊂ An. Then the homomorphism φ ◦ φm,n is as desired. n is of type TII(cid:9) and I1 =(cid:8) i Bi Let I0 =(cid:8) i Bi are mutually orthogonal projections. Hence(cid:8)Λ◦φn,∞(1B0,i tions(cid:8)Pi(cid:9)i∈I0S(cid:8)Qi(cid:9)i∈I0S(cid:8)Ri(cid:9)i∈I1 ))u, ψn1,∞(Qi) = u∗(Λ ◦ φn,∞(1Di are mutually orthogonal projections in A. One can choose n1 (large enough) and mutually orthogonal projec- 16 and ψn1,∞(Pi) = ))u for ))u for i ∈ I0 and ψn,∞(Ri) = u∗(Λ ◦ φn,∞(1Bi n is not of type TII(cid:9). Then(cid:8)1B0,i )(cid:9)i∈I0S(cid:8)Λ◦φn,∞(1Di n (cid:9)i∈I0S(cid:8)1Di )(cid:9)i∈I0S(cid:8)Λ◦φn,∞(1Bi n(cid:9)i∈I0S(cid:8)1Bi )(cid:9)i∈I1 n1 and a unitary u ∈ A such that ku − 1k < ε ⊂ A n n n n n ′ ′ n(cid:9)i∈I1 u∗(Λ ◦ φn,∞(1B0,i i ∈ I1. n Note that kAdu − idk < ε 8 . Replacing Λ : A → A ′ ′ by Λ = Adu ◦ Λ : A Λ−→ A ′ Adu−→ A ′ to make (∗) true, it suffices to construct φ : Bn → A (∗∗) ′ kΛ ′ l for certain l > n1 such that 3ε 8 ◦ φn,∞(f ) − ψl,∞ ◦ φ(f )k < ∀f ∈ Fn ⊂ Bn ⊂ An. Such construction can be carried out for each of the blocks B0,i n and Di n for i ∈ I0, Bi n for i ∈ I1. Namely, we need to construct φB0,i n : B0,i n −→ ψn1,l(Pi)A 15 ′ lψn1,l(Pi) ∀i ∈ I0; φDi n φBi n : Di n −→ ψn1,l(Qi)A lψn1,l(Qi) : Bi n −→ ψn1,l(Ri)A lψn1,l(Ri) ′ ′ separately, to satisfy the condition ′ kΛ ◦ φn,∞(f ) − ψl,∞ ◦ φ(f )k < ∀i ∈ I0; ∀i ∈ I1. 3ε 8 . ∪(cid:8)F i n(cid:9)i∈I1 n)(cid:9)i∈I0 for all f ∈(cid:8)π0(F i n)(cid:9)i∈I0 ∪(cid:8)π1(F i n(cid:9)i∈I1 and(cid:8)Bi n(cid:9)i∈I0 For the blocks(cid:8)Di n+1 is of type TII(cid:9) and J1 =(cid:8) j Bi Let J0 =(cid:8) j Bj n+1 for j ∈ J0, eQi,j = π1(φi,j B0,j n,n+1(1B0,i (Here we only consider the case i ∈ I0). )) ∈ Dj n the domain algebras are stably generated. So we only need to construct φ : B0,i l large enough. , the existence of such homomorphisms follows from the fact that lψn1,l(Pi) for n −→ ψn1,l(Pi)A ′ n+1 is not of type TII(cid:9). Let eP i,j = π0(φi,j n,n+1(1B0,i n+1 for j ∈ J0, eRi,j = π1(φi,j )) ∈ Dj n )) ∈ n,n+1(1B0,i n+1 for j ∈ J1. n As in Lemma 3.6, we have the decomposition π0 ◦ φi,j n,n+1(1B0,i n n+1p0)1 and φ1 ∈ Hom(B0,i n , p1B0,j ) = p0 + p1 ∈ B0,j n+1 and π0 ◦ φi,j n+1p1)1. Denote p0, p1 by pi,j = n,n+1B0,i 0 and pi,j 1 , n φ0 ⊕ φ1 with φ0 ∈ Hom(B0,i n , p0B0,j 1 . It follows that 0 ⊕ pi,j then eP i,j = pi,j nΛ ′(cid:0)φn+1,∞(pi,j 0 )(cid:1)oj∈J0[nΛ ′(cid:0)φn+1,∞(pi,j 1 )(cid:1)oj∈J0[nΛ is a set of mutually orthogonal projections with sum to be Λ (large enough), there are mutually orthogonal projections ′(cid:0)φn+1,∞(eQi,j)(cid:1)oj∈J0[nΛ ′(cid:0)φn,∞(1B0,i )(cid:1) = ψn1,∞(Pi) ∈ A ′(cid:0)φn+1,∞(eRi,j)(cid:1)oj∈J1 . For n2 > n1 n ′ {P i,j 0 }j∈J0 ∪ {P i,j 1 )}j∈J0 ∪ {Qi,j}j∈J0 ∪ {Ri,j}j∈J1 ⊂ ψn1,n2(Pi)A ′ n2 ψn1,n2(Pi) and a unitary v ∈ ψn,∞(Pi)A ′ for all j ∈ J0; and for all j ∈ J1. ψn,∞(Pi) such that kv − 1k < ε 16 , and ψn2,∞(P i,j ψn2,∞(P i,j 0 ) = v∗(cid:16)Λ 1 ) = v∗(cid:16)Λ ψn2,∞(Qi,j) = v∗(cid:16)Λ ψn2,∞(Ri,j) = v∗(cid:16)Λ 0 )(cid:1)(cid:17)v, ′(cid:0)φn+1,∞(pi,j 1 )(cid:1)(cid:17)v, ′(cid:0)φn+1,∞(pi,j ′(cid:0)φn+1,∞(eQi,j)(cid:1)(cid:17)v ′(cid:0)φn+1,∞(eRi,j)(cid:1)(cid:17)v Let eΛ = Adv ◦ Λ construction of homomorphisms ′ . Then the construction of φB0,i n satisfying (∗∗) for all f ∈ π0(F i 0) is reduced to the for all j ∈ J0; and n −→ ψn2,l(P i,j 0 ⊕ P i,j 1 )A ′ lψn2,l(P i,j ξj 0 : B0,j ξj 1 : eQi,jDj ξj : eRi,jBj n+1eQi,j −→ ψn2,l(Qi,j)A n+1eRi,j −→ ψn2,l(Ri,j)A 16 1 ), 0 ⊕ P i,j lψn2,l(Qi,j) ′ ′ lψn2,l(Ri,j) for all j ∈ J1, such that ε 4 n) ⊂ B0,i n and j ∈ J0. ε 4 (∗ ∗ ∗) 0(f )k < 1(f )k < ∀f ∈ π0(F i keΛ ◦ φn+1,∞(f ) − ψl,∞ ◦ ξj keΛ ◦ φn,∞(f ) − ψl,∞ ◦ ξj keΛ ◦ φn+1,∞(f ) − ψl,∞ ◦ ξj(f )k < 1(j ∈ J0) and ξj(j ∈ J1) are homomorphisms from eQi,jDj (Warning: The domain of ξj hand, ξj which are subalgebras of Bn+1). ∀f ∈ π1(cid:16)φi,j n,n+1B0,i n,n+1B0,i ∀f ∈ φi,j 0 is B0,i ε 4 n n (π0(F i (π0(F i n+1eQi,j n+1eRi,j n))(cid:17) ⊂ eQi,jDj n)) ⊂ eRi,jBj n+1eQi,j(j ∈ J0) and eRi,jBj and j ∈ J0. and j ∈ J1. n which is a sub-algebra of Bn but not a sub-algebra of Bn+1. On the other 1(j ∈ J0) and ξj (j ∈ J1) follows from the fact that the corresponding n+1eRi,j(j ∈ J1) n+1eRi,j(j ∈ J1) are stably generatedof course, we need to The existence of the homomorphisms ξj domain algebras eQi,jDj choose l > n2 large enough. n+1eQi,j(j ∈ J0) and eRi,jBj So we only need to construct ξj n) and εn (note that ω(π0(F i for π0(F i that φ0 ∈ Hom(B0,i algebra. There is a homomorphism λ1 : φ1(B0,i n+1pi,j n , pi,j 0 Bj 0 to satisfy (∗ ∗ ∗) above. Let G ⊃ G(P) and δ < δ(P) be as in Lemma 3.5 n), εn) is also from Lemma 3.5. Recall n ) is a finite dimensional 0)) < εn). Recall that L(π0(F i 1 ), and that φ1(B0,i 0 ), φ1 ∈ Hom(B0,i n , pi,j 1 Bj n ) → ψn2,l(P i,j ′ lψn2,l(P i,j 1 ) (for l large enough) such that n+1pi,j 1 )A ε 16 ∀f ∈ φ1(π0(F i n)). n+1pi,j 0 B0,j pi,j large enough, one can obtain a φ0(G) − δ multiplicative quasi-PK homomorphism λ0 : pi,j ψn2,l(P i,j 0 )A ′ 0 ), for l(> n2) 0 → n+1pi,j 0 B0,j ′ 0 )A n+1pi,j mψn2,m(P i,j 0 → ψn2,∞(P i,j 0 ), ψm,l(cid:1), the finite set φ0(π0(F i mψn2,∞(P i,j n)) ⊂ Applying Lemma 3.4 to the inductive limit, lim(cid:0)ψn2,m(P i,j kψl,∞ ◦ λ1(f ) −eΛ(φn+1,∞(f ))k < 0 and the homomorphism eΛ ◦ φn+1,∞ : pi,j kψl,∞ ◦ λ0(f ) − (eΛ ◦ φn+1,∞)(f )k < 0 ) such that mψn2,l(P i,j 0 B0,j 0 )A = λ0 ◦ φ0 ⊕ λ1 ◦ φ1 : B0,i n −→ ψn2,l(P i,j 0 ⊕ P i,j ′ Let ξ 1 )A ′ ε 16 ∀f ∈ φ0(π0(F i n)). ′ lψn2,l(P i,j 0 ⊕ P i,j 1 ). Then ′ (f )k < ε 8 ∀f ∈ π0(F i n). (∗ ∗ ∗∗) keΛ ◦ φn,∞(f ) − ψl,∞ ◦ ξ 1 ] ≥ L(π0(F i On the other hand, since [P i,j finite dimensional image, we know that ξ n), εn) · [P i,j 0 ] in K-theory, and λ1 ◦ φ1 is a homomorphism with = λ0 ◦ φ0 ⊕ λ1 ◦ φ1 satisfies the condition of Lemma 3.6 -- note that λ0 is φ0(G) − δ multiplicative implies that λ0 ◦ φ0 is G − δ multiplicative. By Lemma 3.6, there is a homomorphism ξj n −→ ψn2,l(P i,j (f )k ≤ 5εn < ε 8 . Combining with (∗ ∗ ∗∗) we know that ξj 0 satisfies (∗ ∗ ∗) as desired, and therefore the lemma is proved. 1 ) such that kξj lψn2,l(P i,j 0 ⊕ P i,j 0 ⊕ P i,j 0 : B0,j 0(f ) − ξ 1 )A ′ ′ ′ (cid:3) Definition 3.9. A C-algebra A is said to have the ideal property if each closed two-sided ideal in A is generated by its projections. Remark 3.10. All simple, unital C-algebras have the ideal property. The direct sum of TAI algebras have the ideal property. The inductive limit of C-algebras with the ideal property have the ideal property. 17 The following results is due to Gong, Jiang, Li. Proposition 3.11. ([GJL1]) Suppose that A = lim(An, φn,m) and B = lim(Bn, ψn,m) are two (not neces- sarily unital) AHD inductive limit algebras with the ideal property. Suppose that there is an isomorphism which is compatible with Bockstein operations. Suppose that for each projection p ∈ A and p ∈ B with α :(cid:16)K(A), K(A)+,X A(cid:17) −→(cid:16)K(B), K(B)+,X B(cid:17) α([p]) = [p], there exist a unital positive linear isomorphism ξp,p : Af f T (pAp) −→ Af f T (pBp) and an isometric group isomorphism satisfying the following compatibility conditions: γp,p : U (pAp)/gSU (pAp) −→ U (pBp)/gSU (pBp) (1) For each pair of projections p < q <∈ A and p < q <∈ B with α([p]) = [p], α([q]) = [q], the diagrams Af f T (pAp) −−−−→ Af f T (pBp) ξp,p y ξq,q −−−−→ Af f T (qBq) γ p,p −−−−→ U (pAp)/gSU (pBp) −−−−→ U (qAq)/gSU (qBq) y γ q,q commute, where the vertical maps are induced by the inclusion homomorphisms. (2) The maps α0 and ξp,p are compatible, that is, the diagram and y Af f T (qAq) y U (pAp)/gSU (pAp) U (qAq)/gSU (qAq) α0y K0(pBp) K0(pAp) −−−−→ Af f T (pAp) ρ ρ ξp,py −−−−→ Af f T (pBp) commutes (this is not an extra requirement, since it follows from the commutativity of the first diagram in (1) above by [Ji-Jiang]), and then we have the map (still denoted by ξp,p): ξp,p : Af f T (pAp)/gρK0(pAp) −→ Af f T (pBp)/gρK0(pBp) (3) The map ξp,p and γp,p are compatible, that is, the diagram Af f T (pAp)/gρK0(pAp) Af f T (pBp)/gρK0(pBp) ξp,py eλA−−−−→ U (pAp)/gSU (pAp) eλB−−−−→ U (pBp)/gSU (pBp) γ p,py 18 commutes. (4) The map α1 : K1(pAp)/torK1(pAp) −→ K1(pBp)/torK1(pBp) (note that α keeps the positive cone of K(A)+ and therefore takes K1(pAp) ⊂ K1(A) to K1(pBp) ⊂ K1(B) is compatible with γp,p, that is, the diagram U (pAp)/gSU (pAp) γ p,py U (pBp)/gSU (pBp) eπpAp−−−−→ K1(pAp)/torK1(pAp) eπpBp−−−−→ K1(pBp)/torK1(pBp) α1y commutes. Then there is an isomorphism Γ : A → B such that (a)K(Γ) = α, and (b)If Γp : pAp → Γ(p)BΓ(p) is the restriction of Γ, then Af f T (Γp) = ξp,p and Γ♮ p = γp,p, where [p] = [Γ(p)]. Proposition 3.10. ([GJL2, Proposition 2.38]) Let A, B ∈ HD or AHD be unital C-algebras. Suppose It follows that Inv0(A) ∼= Inv0(B) implies that that K1(A) = tor(K1(A)) and K1(B) = tor(K1(B)). Inv(A) ∼= Inv(B). 4. Main Theorems Theorem 4.1. If A is an ATAI C-algebra, then A is an AHD algebra with ideal property. Proof. By remark 3.10, A has the ideal property. So we will only prove A is an AHD algebra. Suppose that A is the inductive limit (An = n, φn,m), where Ai n are simple AHD algebras. Since all Ai We will construct a sequence of sub-C-algebras Bi n are simple, without lose of generality, we can assume that all the homorphisms φn,m are injective. n which are direct sums of HD building blocks and n ⊂ Ai homomorphisms ψn,n+1 : Bn =L Bi n+1 such that the diagram Ai tnLi=1 n → Bn+1 =L Bi B1 ψ1,2 / B2 ı1 ı2 A1 φ1,2 / A2 / · · · / · · · is approximately commutative in the sense of Elliott and φn,∞(ın(Bn)) = A. Let {an,k}∞ k=1 be a dense subset of the unit ball of An and εn = 1 2n . Let G1 ⊂ A1 be defined by G1 = {πi(a11)} ⊂ Ai 1 = A1 where π1 : A1 → Ai 1 are canonical projections. t1Li=1 t1Li=1 Fix i ∈ {1, 2, · · · , t1}, one can write Ai 1 = lim n→∞ ξn,m. 19 (Cn, ξn,m) as in Lemma 3.6 with injective homomorphisms ∞Sn=1 _ / _ / / / For πi(G1) ⊂ Ai 1, we choose n large enough such that πi(G1) ⊂ε1 ξn,∞(Cn). And let Bi finite set with πi(G1) ⊂ε1 ı(F i 1), where ı is the inclusion homomorphism ξn,∞. Let B = 1 = Cn and F i 1 a Ai 1, 1 ֒→ Bi t1Li=1 t1Li=1 and F1 =L F i 1. We will construct the following diagram 1 1 ı1 T F1 =L F i B1 =L Bi A1 =L Ai S G1 =L Gi 1 1 ψ1,2 φ1,2 2 2 ı2 F2 =L F i T / B2 =L Bi / A2 =L Ai S G2 =L Gi 2 2 n n Fn =L F i T / · · · Bn =L Bi / · · · An =L Ai S Gn =L Gi ın n n such that (1) Gk ⊂εk ık(Fk), Gk ⊃ φk−1,k(Gk−1 ∪ {ak−1,k}) ∪ {ak,j}k j=1; (2) kık ◦ ψk−1,k(f ) − φk−1,k ◦ ık−1(f )k < εk−1, ∀f ∈ Fk−1. The construction will be carried out by induction. Suppose that we have the diagram until Fn ⊂ Bn ֒→ An ⊃ Gn. We will construct the next piece of the diagram. n Let P i,j = φi,j n+1. Then {P i,j}tn ) ∈ Aj n+1 in place of A and {P i,j}tn n+1. n,n+1(1Ai Apply Corollary 3.7 for Aj n+1 = lim(Ck, λk,l) with P i,jAj l ∈ Cl) being AHD inductive limit algebra as described in Lemma 3.6. (Warning: Do not confuse these Ck with the Cn in the expression Ai i=1 in places of {P1, P2, · · · , Pk}, we can write Aj i=1 is a set of mutually orthogonal projections in Aj n+1P i,j = lim k ) (where λk,l(Qi,j k ) = Qi,j k , λk,lQi,j k CkQi,j k CkQi,j (Cn, ξn,m).) (Qi,j k 1 = lim n→∞ For each pair i, j, we apply Lemma 3.8 to the homomorphism φi,j n+1P i,j in place n in place of F ∈ Am, there is an l (large enough) and a homomorphism n,n+1 : Ai n → P i,jAj n ⊂ Bi ′ of Λ : A → A ψi,j n,n+1 : Bi n → Qi,j , and F i l ClQi,j l such that kφi,j n,n+1 ◦ ın(f ) − ın+1 ◦ ψi,j n,n+1(f )k < ε ∀f ∈ F i n, where ın+1Qi,j l ClQi,j Finally choose Gj l = λl,∞Qi,j l ClQi,j l . n+1 ⊃ πj(cid:2)(Gn ∪ {an,n+1}) ∪ {an+1,j}n+1 j=1 ∪ ın+1(Sk ′j n+1 ⊂ Cl such that Gj ψi,j n,n+1(F i n))(cid:3) and Gn+1 = L Gj ′j n+1. Define Bj n+1 to n+1. By increasing l, we can assume that there is an F be Cl which is a subalgebra of Aj n+1. Let F j n+1 ⊂ Bj n+1 be defined by F j n+1 ⊂εn+1 F ′j n+1 = F n+1 ∪(cid:0)Li φi,j n,n+1(F i n)(cid:1) and let Fn+1 = L F j lim(Bn, ψn,m) = lim(An, φn,m). This ends the proof. n+1. This ends our inductive construction. By the Elliott intertwining argument, (cid:3) 20 _ / _ / _ / / / / / / Theorem 4.2. Let A = lim(An, φn,m) and B = lim(Bn, ψn,m) be inductive limit algebras with Ai being unital separable nuclear simple TAI algebras with UCT. Suppose that there is an isomorphism n, Bi n which is compatible with Bockstein operations. Suppose that for each projection p ∈ A and p ∈ B with α :(cid:16)K(A), K(A)+,X A(cid:17) −→(cid:16)K(B), K(B)+,X B(cid:17) α([p]) = [p], there exist a unital positive linear isomorphism ξp,p : Af f T (pAp) −→ Af f T (pBp) and an isometric group isomorphism satisfying the following compatibility conditions: γp,p : U (pAp)/gSU (pAp) −→ U (pBp)/gSU (pBp) (1) For each pair of projections p < q <∈ A and p < q <∈ B with α([p]) = [p], α([q]) = [q], the diagrams Af f T (pAp) −−−−→ Af f T (pBp) ξp,p y Af f T (qAq) y U (pAp)/gSU (pAp) U (qAq)/gSU (qAq) α0y K0(pBp) K0(pAp) y ξq,q −−−−→ Af f T (qBq) γ p,p −−−−→ U (pAp)/gSU (pBp) −−−−→ U (qAq)/gSU (qBq) y γ q,q −−−−→ Af f T (pAp) ρ ρ ξp,py −−−−→ Af f T (pBp) and commute, where the vertical maps are induced by the inclusion homomorphisms. (2) The maps α0 and ξp,p are compatible, that is, the diagram commutes (this is not an extra requirement, since it follows from the commutativity of the first diagram in (1) above by [Ji-Jiang]), and then we have the map (still denoted by ξp,p): ξp,p : Af f T (pAp)/gρK0(pAp) −→ Af f T (pBp)/gρK0(pBp) (3) The map ξp,p and γp,p are compatible, that is, the diagram Af f T (pAp)/gρK0(pAp) Af f T (pBp)/gρK0(pBp) ξp,py eλA−−−−→ U (pAp)/gSU (pAp) eλB−−−−→ U (pBp)/gSU (pBp) γ p,py commutes. (4) The map α1 : K1(pAp)/torK1(pAp) −→ K1(pBp)/torK1(pBp) (note that α keeps the positive cone of K(A)+ and therefore takes K1(pAp) ⊂ K1(A) to K1(pBp) ⊂ K1(B) is compatible with γp,p, that is, the 21 diagram commutes. U (pAp)/gSU (pAp) γ p,py U (pBp)/gSU (pBp) eπpAp−−−−→ K1(pAp)/torK1(pAp) eπpBp−−−−→ K1(pBp)/torK1(pBp) α1y Then there is an isomorphism Γ : A → B such that (a)K(Γ) = α, and (b)If Γp : pAp → Γ(p)BΓ(p) is the restriction of Γ, then Af f T (Γp) = ξp,p and Γ♮ p = γp,p, where [p] = [Γ(p)]. Proof. It follows from the fact of Theorem 4.1 that all ATAI algebras are AHD algebras with ideal property and Proposition 3.11 ([GJL1]) that all AHD algebras with ideal property are classified by Inv(A). (cid:3) and torsion K1-group. Assume that there is an isomorphism α ∈ HomΛ(K(A), K(B)) such that n, ψn,m) be inductive limits with Ai Corollary 4.3. (A)([Jiang1], Theorem 3.11) Let A = limn→∞(An =L Ai L Bi α(X A) =X B. α(K(A)+) = K(B)+, n, φn,m) and B = limn→∞(Bn = n being unital separable nuclear simple TAI algebras with UCT n, Bi and for each pair of projections p ∈ A, q ∈ B with α([p]) = [q], there is a continuous affine homomorphism ξp,q : T (qBq) → T (pAp) which is compatible in the sense of 2.19, then there is an isomorphism Λ : A → B which induces α and ξ above. (B) ([Fa]) Let A and B be two ATAF algebras. Suppose that is an isomorphism of ordered groups which preserves the action of the Bockstein operations. Then there is a ∗ -- isomorphism ϕ : A → B with α :(cid:16)K(A), K(A)+,X A(cid:17) →(cid:16)K(B), K(B)+,X B(cid:17) ϕ∗ = α. Proof. (A) It follows from the fact that K1(A) = torK1(A), K1(B) = torK1(B) and Proposition 3.10 ([GJL2, Proposition 2.38]). (B) From Theorem 4.1, if A is an ATAF C-algebra, then A is an AHD algebra of real rank zero which is classified in [DG] (Note that AHD algebra in [DG] is denoted by ASH algebra). (cid:3) 22 References [1] [D] M. Dadarlat, Reduction to dimension three of local spectra of Real rank zero C-algebras, J. Reine Angew. Math. 460(1995) 189-212 [2] [DG] M. Dadarlat and G. Gong, A classification result for approximately homogeneous C-algebras of real rank zero, Geometric and Functional Analysis, 7(1997) 646-711 [3] [Ell1] G. A. Elliott, On the classification of C-algebras of real rank zero, J. Reine Angew. Math. 443(1993) 263-290 [4] [Ell2] G. A. Elliott, A classification of certain simple C-algebras, Quantum and Non-Commutative Analysis, Kluwer, Dordrecht, (1993), pp, 373-388 [5] [Ell3] G. A. Elliott, A classification of certain simple C-algebras, II, J. Ramaunjan Math. Soc. 12 (1997), 97-134 [6] [EE] G. A. Elliott, D. E. Evans: The structure of the irrational rotation C-algebra. Ann. of Math. (2) 138 (1993), no. 3, 477-501. [7] [EG1] G. A. Elliott and G. Gong, On the inductive limits of matrix algebras over two-tori, American. J. Math 118(1996) 263-290 [8] [EG2] G. A. Elliott and G. Gong, On the classification of C-algebras of real rank zero, II. Ann. of Math 144(1996) 497-610 [9] [EGL1] G. A. Elliott, G. Gong and L. Li, On the classification of simple inductive limit C-algebras, II; The isomorphism Theorem, Invent. Math. 168(2)(2007) 249-320 [10] [EGL2] G. A. Elliott, G. Gong and L. Li, Injectivity of the connecting maps in AH inductive limit systems, Canand. Math. Bull. 26(2004) 4-10 [11] [EGJS] G. A. Elliott, G. Gong, X. Jiang, H. Su: Aclassification of simple limits of dimension drop C-algebras. Fields Inst. Commun. 13, 125-143 (1997) [12] [EGLP] G. A. Elliott, G. Gong, H. Lin, C.Pasnicu: Abelian C-subslgebras of C-algebras of real rank zero and inductive limit C-algebras. Duke Math. J. 83, 511-554 (1996) [13] [EGS] G. A. Elliott, G. Gong, H. Su: On the classification of C-algebras of real rank zero, IV: Reduction to local spectrum of dimension two. Fields Inst. Commun. 20, 73-95 (1998) [14] [Fa] X. C. Fang, The classification of certain non-simple C-algebras of tracial rank zero, J. Funct. Anal. 256(2009) 3861- 3891 [15] [G1-2] G. Gong, On inductive limit of matrix algebras over higher dimension spaces, Part I, II, Math Scand. 80(1997) 45-60, 61-100 [16] [G3] G. Gong, On the classification of simple inductive limit C-algebras, I: Reduction Theorems. Doc. Math. 7(2002) 255-461 [17] [GJL1] G. Gong, C. Jiang, L. Li, A classification of inductive limit C-algebras with ideal property, Preprint, [18] [GJL2] G. Gong, C. Jiang, L. Li, Hausdorffized algebraic K1 group and C-algebras with the ideal property, Preprint, [19] [GJLP1] G. Gong, C. Jiang, L. Li, C. Pasnicu, AT structure of AH algebras with ideal property and torsion free K-theory, J. Func. Anal. 58(2010) 2119-2143 [20] [GJLP2] G. Gong, C. Jiang, L. Li, C. Pasnicu, A Reduction theorem for AH algebras with the ideal property, Int Math Res Notices. 24(2018), 7606-7641 [21] [GL] G. Gong and H. Lin, Almost multiplicative morphisms and K-theory, International J. Math. 11 (2000) 983-1000. [22] [GLN] G. Gong, H. Lin and Z. Niu, Classification of finite simple amenable Z-stable C-algebras, Preprint arXiv:1501.00135 [23] [GLX] G. Gong, H. Lin and Y. Xue, Determinant rank of C-algebras, Pacific J. Math 274 (2015) No. 2 405-436 [24] [Ji-Jiang] K. Ji and C. Jiang, A complete classification of AI algebra with ideal property, Canadian. J. Math, 63(2), (2011), 381-412 [25] [Jiang1] C. Jiang, A classification of non simple C-algebras of tracial rank one: Inductive limit of finite direct sums of simple TAI C-algebras, J. Topol. Anal. 3 No.3(2011), 385-404 [26] [Jiang2] C. Jiang, Reduction to dimension two of local spectrum for AH algebras with ideal property, Canand. Math. Bull. 60(2017) No.4, 791-806 [27] [Li1] L. Li, On the classification of simple C-algebras: Inductive limit of matrix algebras over trees, Mem Amer. Math, Soc. 127(605) 1997 23 [28] [Li2] L. Li, Simple inductive limit C-algebras: Spectra and approximation by interval algebras, J. Reine Angew Math 507(1999) 57-79 [29] [Li3] L. Li, Classification of simple C-algebras: Inductive limit of matrix algebras over 1-dimensional spaces, J. Func. Anal. 192(2002) 1-51 [30] [Li4] L. Li, Reduction to dimension two of local spectrum for simple AH algebras, J. of Ramanujian Math. Soc. 21 No.4(2006) 365-390 [31] [Lin1] H. Lin, Tracially AF C-algebras, Trans. Amer. Math. Soc. 353 (2001) No. 2, 693-722 [32] [Lin2] H. Lin, Simple nuclear C-algrbras of tracial topological rank one. J. Funct. Anal. 251(2007), No. 2, 601-679 [33] [NT] K. E. Nielsen and K. Thomsen, Limits of circle algebras. Expo. Math. 14, 17-56 (1996) [34] [Popa] S. Popa, On locally finite-dimensional approximation of C-algebras, Pacifi J. Math. (1997) 141-158 [35] [Rob] L. Robert, Classification of inductive limits of 1-dimensional NCCW-complexes, Adv. in Math. 231(2012), 2802-2836 [36] [R] M. Rordam, Classification of certain infinite simple C-algebras. J. Funct. Anal. 131(1995), 415-458 [37] [Stev] K. Stevens, The classification of certain non-simple approximate interval algebras. Fields Inst. Commun. 20(1998) 105-148 [38] [Thm1] K. Thomsen, Insuctive limit of interval algebras, American J. of Math 116, 605C620 (1994) [39] [Thm2] K. Thomsen, Limits of certain subhomogeneous C-algebras, Mem. Soc. Math. Fr. (N.S.) 71(1999) 24
1308.4161	6	1308	2018-12-16T21:32:00	Natural transformations associated with a locally compact group and universality of the global Terrell law	[ "math.OA", "math-ph", "math.DS", "math-ph" ]	Via the construction of a functor from $\mathsf{C}_{u}(H)$ to an auxiliary category we associate, with any triplet $(G,F,\rho)$, two natural transformations, $\mathfrak{m}_{\star}$ morphism of $\mathsf{Fct}(\mathsf{C}_{u}(H)^{op},\mathsf{Fct}(H,\mathsf{set}))$ and $\mathfrak{v}_{\natural}$ morphism of $\mathsf{Fct}(\mathsf{C}_{u}^{0}(H)^{op},\mathsf{Fct}(H,\mathsf{set}))$. $G$ and $F$ are locally compact groups, $\rho:F\to Aut(G)$ is a continuous morphism, $H$ is the external topological semidirect product of $G$ and $F$ relative to $\rho$, $\mathsf{C}_{u}^{0}(H)$ is a subcategory of $\mathsf{C}_{u}(H)$ a subcategory of the category of $C^{\ast}-$dynamical systems with symmetry group $H$ and equivariant morphisms. For $\mathfrak{A}$ in $\mathsf{C}_{u}^{0}(H)$ to assemble $\mathfrak{m}_{\star}^{\mathfrak{A}}$ we exploit the Connes characters generated by JLO cocycles $\Phi$ on the unitization of certain $C^{\ast}-$crossed products relative to $\mathfrak{A}$, while to construct $\mathfrak{v}_{\natural}^{\mathfrak{A}}$ we exert the states of the $C^{\ast}-$algebra underlying $\mathfrak{A}$ associated in a convenient manner with the $0-$dimensional components of the $\Phi$'s. Being $\mathsf{C}_{u}(H)$ the category of fissioning systems, we use $\mathfrak{m}_{\star}^{\mathfrak{A}}$ and $\mathfrak{v}_{\natural}^{\mathfrak{A}}$ to define the nucleon phases and the fragment states resulting next the fission processes of the fissioning system $\mathfrak{A}$ occur. We apply the naturality of $\mathfrak{m}_{\star}$ and $\mathfrak{v}_{\natural}$ to establish the universality of the global nucleon masses and the global Terrell law, stated as invariance of the light and heavy nucleon core masses and invariance of the prompt-neutron yield under controvariant action of $\mathsf{C}_{u}^{0}(H)$ and under action of $H$ over the field of fission processes.	math.OA	math	Natural transformations associated with a locally compact group and universality of the global Terrell law Benedetto Silvestri 8 1 0 2 c e D 6 1 ] . A O h t a m [ 6 v 1 6 1 4 . 8 0 3 1 : v i X r a 2010 Mathematics Subject Classification. Primary 46L55, 46M15, 81R15, 81R60; Secondary 82B10, 81V35 Key words and phrases. C∗−crossed products, Chern-Connes character, natural transformations, nuclear binary fission. Abstract. Via the construction of a functor from Cu(H) to an auxiliary category we associate, with any triplet (G, F, ρ), two natural transformations, m⋆ between functors from the opposite of Cu(H) to Fct(H, set) and v♮ between functors from the opposite of C0 u(H) to Fct(H, set). G and F are locally compact groups, ρ : F → AutGr(G) is a group morphism such that the map (g, f ) 7→ ρ f (g) is continuous, H is the external topological semidirect product of G and F relative to ρ, a groupoid when seen as a category, C0 u(H) is a subcategory of Cu(H) a subcategory of the category of C∗−dynamical systems with symmetry group H and equivariant morphisms, and Fct(H, set) is the category of functors from H to set and natural transformations. For any object A of C0 u(H) to assemble mA ⋆ we exploit the Chern-Connes characters generated by JLO cocycles Φ on the unitization of certain C∗−crossed products relative to A, while to construct vA ♮ we exert the states of the C∗−algebra underlying A associated in a convenient manner with the 0−dimensional components of the Φ's. We interpret Cu(H) as the category of fissioning systems and their transformations, we use mA ♮ to define the nucleon phases and the fragment states resulting next the fission processes of the fissioning system A occur and we formulate in a C∗−algebraic framework a new nucleon phase hypothesis originally advanced by Mouze and Ythier. We apply the naturality of m⋆ and v♮ to establish the universality of the global nucleon masses and the global Terrell law, stated as invariance of the light and heavy nucleon core masses and invariance of the prompt-neutron yield under contravariant action of C0 u(H) and under action of H over the field of fission processes. Then under the nucleon phase hypothesis we exhibit the stability of the nucleon core masses at values 82 and 126 and the invariance of the Terrell law, as a particular case of the global ones. ⋆ and vA Contents Introduction 0.1. 0.2. Terminology and preliminaries Part 1. Preliminaries in Equivariance 1.1. Equivariance 1.2. Extensions on the multiplier algebra Part 2. Stability Introduction 2.1. 2.2. The category of nucleon systems 2.3. The canonical nucleon-fragment doublet T• Part 3. Universality Introduction 3.1. 3.2. Natural transformations related to T• 3.3. Universality of the global Terrell law Acknowledgments Bibliography 4 17 33 34 52 59 60 60 91 113 114 114 129 149 151 3 4 CONTENTS 0.1. Introduction Mouze and Ythier advanced the nucleon phase hypothesis in order to explain, among other occurrences, the presence of two fixed values 82 and 126 in the following Terrell law (0.1.1) ν = 0.08 (AL − 82) + 0.1 (AH − 126), describing the mean value of the prompt-neutron yield of two asymmetric fragments resulting next the neutron fission of U233, U235, Pu239 and the spontaneous fission of C f 252, in function of the mass numbers AL and AH of the light and heavy fragments respectively, [26, 27, 33]. Basically they assumed that under the particular circumstances occurring in the fission process, two nucleon cores come into existence, one of mass number 82 and the other of mass number 126. Thus by mean of the closure of the shells at these values, their presence in the above equality can be justified. Since the evenience of closed shells in the presence of the nucleon phase in which the cores happen to appear, it has been claimed by Ricci that "the organization of matter in closed shells should be a universal law" see [33, IIa]. u(H). Here C0 We contend instead that the nucleon phase itself is universal and we grant the tenet that universal in physics has to be understood natural in category theory. The present is the first of a series of three parts where, by introducing the concept of the category G(G, F, ρ) of nucleon systems and the structure of nucleon-fragment doublet, we state and resolve the equivariant and invariant forms of the universality claim, later described, for the positions K = Cu(H) and H = C0 u(H) is a strict subcategory of the category Cu(H) of fissioning systems and their transformations, a subcategory of the category of C∗−dynamical systems with symmetry group H and equivariant morphisms. In short the first asserts that the global fragment state is originated via the global nucleon phase and both are equivariant under contravariant action of C0 u(H) and under action of H. The second establishes that the global nucleon and fragment masses, and the global Terrell law are invariant under contravariant action of C0 u(H) and under action of H over the field of fission processes. As a consequence we obtain the stability of the masses of the nucleon cores and the invariance of the prompt-neutron yield under essential contravariant action of C0 u(H) and under action of H. Then provided a new nucleon phase hypothesis the stability of the nucleon masses at the values 82 and 126 and the invariance of the Terrell law under action of the above transformations, follow. The meaning of the term 'essential' will be clarified in course of the introduction. We anticipate that the introduction of the global nucleon and fragment masses, and the global Terrell law - generalizing and extending the nucleon and fragment masses, and the Terrell law respectively - as extensions serve to switch 'essential' into 'true' contravariance. By balancing the language in order to allocate the mathematical and physical de- scriptions, we organize this introduction as follows. Firstly we describe features of the category G(G, F, ρ) and their physical interpretation, then we exert them to expose the forms of the universality claim. Next we outline the concept of nucleon-fragment doublet T on an arbitrary category and exhibit the T−resolution of the forms of the 0.1. INTRODUCTION 5 universality claim. Then the resolution of the claim is established as T•−resolution, where T• is the canonical nucleon-fragment doublet on Cu(H) constructued in the main theorem of the work. A nucleon-fragment doublet provides to nucleon phases and fragment states the following list of properties, next called initial list, whose meaning will be later clarified: (1) thermal nature of the nucleon phases, (2) noncommutative geometric and thermal origination of fragment states, (3) phase transition via symmetry breaking, (4) contravariance under action of the category Ξ(DDD, L) and covariance under action of the symmetry group H. Let us start with some notation. For any set A let P(A) be the power set of A. If A is a C∗−algebra let EA be the set of states of A, Aob be the set of selfadjoint elements of A and A+ ≔ {a∗a a ∈ A}. If A is a category and x, y are objects of A, then let MorA(x, y) denote the set of morphisms of A whose domain or source is x and whose codomain or target is y, let MorA denote the set of morphisms of A. Let AutA(x) be the group of invertible morphisms of A whose source and target is x. Let op mean opposite category, and for any morphism T of a category let d(T) and c(T) denote the domain and codomain of T respectively. Let set be the category of sets and maps in a fixed universe, moreover all the following categories are implicitly assumed to have the object set a subset of the fixed universe. Let gr : Morset → Morset be such that d(gr( f )) = d( f ) and so defined f 7→ (x 7→ (x, f (x))); by abuse of language let Qx∈X C beQx∈X Cx where Cx = C for all x ∈ X, with X and C sets. Let Gr be the category of groups and group morphisms with map composition, and Ab the subcategory of abelian groups. If G and F are locally compact groups, and ρ : F → AutGr(G) is a group homomorphism such that the map (g, f ) 7→ ρ f (g) on G × F at values in G, is continuous, then let H = G ⋊ρ F be the external topological semidirect product of G and F relative to ρ. Here AutGr(G) is the group of group automorphisms of the group underlying G. Let j1 and j2 be the canonical injections of G and F into H respectively. For any locally compact group K let C(K) be the category of unital C∗−dynamical systems with symmetry group K and surjective equivariant morphisms. For all categories X, Y let Fct(X, Y) be the category of functors from X to Y and natural transformations as morphisms. For any functor F let Fo and Fm denote the object map and morphism map respectively. In particular we consider the group H as the category whose unique object is its unit 1, so Fct(H, set) can be considered as the category of set representations of H and intertwinings as morphisms. Let a field contravariant (respectively covariant) under action of X be a functor from Xop (respectively X) to set, moreover we add via L if L is its morphism If Z is a field contravariant (respectively covariant) under action of X then let map. a subfield J of Z be any field contravariant (respectively covariant) under action of X such that Jo(a) ⊆ Zo(a) and Jm(t) = Zm(t) ↾ Jo(c(t)) (respectively Jm(t) = Zm(t) ↾ Jo(d(t))), for any object a and morphism t of X. Notice that for any field say N covariant under action of the group H via R, R is a representation of H on the set N(1). If M is a field contravariant under action of X, we call x a section of M contravariant under action of 6 CONTENTS X via f and g, if x is a natural transformation between functors from Xop and set such that f is the morphism map of the target functor of x, g is the morphism map of the source functor of x, and M is the target functor of x. Similarly for covariant sections by replacing Xop by X. We call x a section contravariant (covariant) under action of X via f and g, a section x of M contravariant (covariant) under action of X via f and g, for some field M contravariant (covariant) under action of X. We call x a section invariant under contravariant (covariant) action of X via g, if x is a section of M contravariant (covariant) under action of X via f and g, the object map of M is a constant with constant value equal some set Z and f is the constant map with constant value equal the identity map on Z. The term equivariance will be used to mean covariance or contravariance. Let K and D be categories, F ∈ Fct(K, D) and C be an object of D. Let (·)F,C † be the map on MorK such that for any object X, Y of K and T ∈ MorK(X, Y) we have TF,C † : MorD(Fo(Y), C) → MorD(Fo(X), C), g 7→ g ◦ Fm(T). † We call (·)F,C the conjugate of Fm, and call dual of the restriction of an action on the automorphism group of an object of a category the composition of conjugation with inversion. Notice that we do not refer to the object C since we use this notation only in the following cases from which it will be clear by the context which object C is concerned. (1) D = Ab and C the field R of real numbers; (2) K = CA∗ the category of C∗−algebras and ∗−homomorphisms, D = BS the category of complex Banach spaces and linear bounded maps, F the forgetful functor from CA∗ to BS and C the field C of complex numbers. Moreover in this case by abuse of language we shall speak of the conjugate of a ∗−homomorphism, by meaning its image via the conjugate of Fm. The first step is the introduction of the category G(G, F, ρ) of nucleon systems whose data determine what follows. 0.1. INTRODUCTION 7 (A, Pr 1 (TTT, Pr 2 ) ∈ Fct(G(G, F, ρ), Ab), ) ∈ Fct(G(G, F, ρ)op, set), (ψM, bM ◦ inv) ∈ MorGr(H, AutG(G,F,ρ)(M)), ∀M ∈ G(G, F, ρ), (0.1.2) IM : TTTM → set, ∀M ∈ G(G, F, ρ), M IQ M, βc ∈ YM∈G(G,F,ρ) YQ∈TTTM a : YM∈G(G,F,ρ) YQ∈TTTMYβ∈IQ e : YM∈G(G,F,ρ) YQ∈TTTMYβ∈IQ ϕϕϕ ∈ YM∈G(G,F,ρ) YQ∈TTTMYβ∈IQ V ∈ YM∈G(G,F,ρ) YQ∈TTTM Yβ∈PQ M M M C(H), C(R), EA(M)Q β , Yl∈H MorCA∗(A(M)Q β , A(M)bM(l)Q β ); where (aM)T G(G, F, ρ), T ∈ TTTM and α ∈ IT M. α = h A(M)T α , H, (ηM)T α i, (eM)T α = h A(M)T α , R, (εM)T α i, for any object M of Let N be an object of G(G, F, ρ), T ∈ TTTN and α ∈ IT N. Thus the following properties hold (1) (ϕϕϕN)T α is invariant under action of G, i.e. (ϕϕϕN)T α ◦ (ηN)T α (j1(g)) = (ϕϕϕN)T α , ∀g ∈ G, (2) (ϕϕϕN)T α is an (εN)T α −KMS state. Note that if α ∈ R+ then (ϕϕϕN)T the inverse temperature α, with respect to the dynamics (εN)T α is a α−KMS state, i.e. a state of thermal equilibrium at α (−α−1(·)). Next set α ◦ (ηN)T α (j2(h)) = (ϕϕϕN)T α }, ≔ {h ∈ F (ϕϕϕN)T F(N)T α ST α (N) ≔ G ⋊ρ F(N)T α , BT α (N) ≔ A(N)T α (ηN)T α ⋊ P ∈ YM∈G(G,F,ρ) YQ∈TTTM ST α (N), P(IQ M), PT N A∗ N ≔ {β ∈ IT N F(N)T ≔ MorAb(AN, R). β ⊇ F(N)T c )T}, (βN 8 CONTENTS A fundamental set valued map is RepN : A∗ N → set. Let c ∈ A∗ N then 1 RepN(c) ∋ r 7→ Tr ∈ TTTN, RepN(c) ∋ r 7→ αr ∈ PTr N , RepN(c) ∋ r 7→ Ar ≔ A(N)Tr αr , RepN(c) ∋ r 7→ ϕϕϕr ≔ (ϕϕϕN)Tr αr , RepN(c) ∋ r 7→ Br ≔ BTr αr (N), RepN(c) ∋ r 7→ εr ≔ (εN)Tr αr , RepN(c) ∋ r 7→ Vr ≔ V(N)Tr αr , RepN(c) ∋ r 7→ ur ∈ MorAb(AN, K0(B+ r )); RepN(c) ∋ r 7→ Φr, RepN(c) ∋ r 7→ ρr ∈ EAr; (0.1.3) and the maps (0.1.4) such that Φr is an entire normalized even cocycle on the unitization B+ r of Br, c factorizes through the real part, via the standard duality, of the character generated by Φr and the map ur; while ρr is a ϕϕϕr−normal state associated in an appropriate way to the 0−dimensional component of Φr, where the standard duality is between the entire cyclic cohomology and the K0−theory K0(B+ r . Let us define r ) of B+ NN(c) ≔ {ρr r ∈ RepN(c)}. Next we interpret the previous data in a physical context as follows. (1) N is a physical system called nucleon system whose set of observables is AN and whose set of states called nucleon phases is A∗ N; (2) L(a) is the nucleon system generated by the fissioning system a, for any functor L from a category H to G(G, F, ρ) and a ∈ H; (3) for any T ∈ TTTN and α ∈ PT N there exists a physical system O(N)T α called fragment system such that (a) A(N)T O(N)T α is its observable algebra and EA(N)T α evolves in time through the dynamics (εN)T α (−α−1(·)), α is its state space. If α > 0 then (b) T is an operation performable over the states of O(N)T α ; N and r ∈ RepN(c) the following properties hold (4) for all c ∈ A∗ (a) if the operation Tr is performed on O(N)Tr αr when occurring in the state ϕϕϕr, then O(N)Tr αr will occur in the state ρr, (b) if O(N)Tr αr occurs in the state ρr, then N occurred in the phase c; (5) the information about the operations in TTTN are sufficient to organize AN as a group but not as an algebra. 1Actually Br depends also by a Haar measure on STr αr (N) but here we do not need such a generality. 0.1. INTRODUCTION 9 This interpretation is consistent with the fact that ω−normal states are usually considered perturbations of the state ω. Notice that ρr is a state of the fragment system O(N)Tr αr , moreover if α > 0 then ϕϕϕα is a state of thermal equilibrium at inverse temperature α with respect to the dynamical system underlying O(N)T α . We refer to (4b) by saying that the fragment state ρr is originated via the nucleon phase c, in particular NN(c) is the set of all the fragment states originated via the nucleon phase c. Referring only to c the sequence (4a,4b) is summarized by saying that c is the phase of the nucleon system N occurring by performing the operation Tr on the state of thermal equilibrium ϕϕϕr. Now the meaning of (1) and (2) in the initial list appear clear. Note that in general NN(c) contains more than one element and its elements may belong to state spaces of different C∗−algebras. If L is a functor from a category C to G(G, F, ρ), then for any map M whose domain is a subset of the object set of G(G, F, ρ), we let Mx denote ML(x) for any object x of C such that L(x) ∈ Dom(M). Now we can expose the two forms of the universality claim, then we exploit the rest of the introduction to sketch their solutions. We shall describe the equivariant form of the claim firstly in mathematical (M) and then in physical (P) terms. Equivariant form of the universality claim (M). There exist a category K, a subcategory H of K, a functor L from K to G(G, F, ρ), m and W such that m ∈ MorFct(Kop,set), gr ◦ W ∈ MorFct(Hop,set). Let UUU and DDD be the object maps of the source functors of m and gr ◦ W respectively and let a ∈ K and b ∈ H. Thus (1 7→ ma) ∈ MorFct(H,set), (1 7→ gr(Wb)) ∈ MorFct(H,set); UUUa ⊆ TTTa and DDDb ⊆ UUUb. Moreover for all Q ∈ UUUa, β ∈ PQ a ma(Q, β) ∈ A∗ a, and all T ∈ DDDb, α ∈ PT b (0.1.5) ∃ r ∈ Repb(mb(T, α)) Wb(T, α) = ρr, Tr = T, αr = α. We say that L, m and W satisfy the equivariant form of the universality claim (M) w.r.t. K and H. Let m and W be called the global nucleon phase and global fragment state w.r.t. L respectively. Equivariant form of the universality claim (P). There exist a category K of fissioning systems and their transformations, a subcategory H of K, a functor L from K to G(G, F, ρ), a section m contravariant under action of K, and a map W such that gr ◦ W is a section contravariant under action of H. Let UUU and DDD be the object maps of the source functors of m and gr ◦ W respectively and let a in K and b in H. Thus 1 7→ ma and 1 7→ gr(Wb) are covariant under action of H, UUUa ⊆ TTTa and DDDb ⊆ UUUb. Moreover for all Q ∈ UUUa, β ∈ PQ a , and for all T ∈ DDDb, α ∈ PT b 10 CONTENTS (1) ma(Q, β) is a phase of the nucleon system L(a); (2) Wb(T, α) is a fragment state originated via the nucleon phase mb(T, α) through the relation (0.1.5); (3) Wb(T, α) is a state of the fragment system O(b)T α and α (−α−1(·)). mb(T, α) is the phase if α > 0 then the system evolves in time through (εb)T of the nucleon system L(b), occurring by performing the operation T on the state of thermal equilibrium (ϕϕϕb)T α at inverse temperature α with respect to the dynamical system underlying O(b)T α . α whose operator algebra is A(b)T Invariant form of the universality claim. There exist a category K, a subcategory H of K, L, m and W satisfying the equivariant form of the universality claim (M) w.r.t. K and H; moreover there exist Nas, P, R, {µ j, λ j} j∈{m,w} and θ with the following properties. Nas is a set valued map defined on the object set of H, P, R ∈ Fct(Hop, set), P(b) equals the power set of Nas(b) the set of fission processes whose underlying fissioning system is b, for all b ∈ H. R is the unique object of Fct(Hop, set) whose object map is the constant map with constant value equal to the power set of R, and whose morphism map is the constant map with constant value equal to the identity map on the power set of R. For all j ∈ {m, w} µ j, λ j, θ ∈ MorFct(Hop,set)(P, R). (1 7→ (µ j)b), (1 7→ (λ j)b), (1 7→ θb) ∈ MorFct(H,set), Let b ∈ H. Thus and there exist maps (1) Nas(b) ∋ y 7→ Ty ∈ DDDb, (2) Nas(b) ∋ y 7→ αy ∈ PTy b , (3) Nas(b) ∋ y 7→ fy ∈Q j∈{m,w} Ab, (4) Nas(b) ∋ y 7→ Ny ∈Q j∈{m,w}(A(b)Ty αy )+, such that for any Y ∈ P(b) and j ∈ {m, w} we have (µ j)b(Y) = {mb(Ty, αy)(fy (λ j)b(Y) = {Wb(Ty, αy)(Ny j ) y ∈ Y}, j ) y ∈ Y}, θb(Y) =n0.08(Wb(Ty, αy)(Ny 0.1(Wb(Ty, αy)(Ny m) − mb(Ty, αy)(fy m))+ w) − mb(Ty, αy)(fy w)) y ∈ Yo. Here we let DDD be the object map of the source functor of gr ◦ W. Let µm and µw be called global light and global heavy nucleon masses w.r.t. L respectively, let λm and λw be called global light and global heavy fragment masses w.r.t. L respectively, and θ be called global Terrell law w.r.t. L. Next we introduce the crucial structure of the entire work and explain how it re- solves the diverse forms of the universality claim, but let us start with some convention. G(G, F, ρ) acts covariantly on the field A of nucleon observables via Pr1 and contravari- antly on the field A∗ of nucleon phases via the conjugate of Pr1; while G(G, F, ρ) acts 0.1. INTRODUCTION 11 N T ∈ TTTN, α ∈ PT contravariantly on the field TTT of operations via Pr2. Therefore for any object N of G(G, F, ρ), the group H acts on AN via ψN and on A∗ N via duality, while H acts on TTTN via bN. Let l ∈ H, f ∈ AN, c ∈ A∗ α and ω continuous func- α . We let fl = ψN(l)(f), cl = c ◦ ψN(l−1), Tl = bN(l)(T), al = V(N)T tional on A(N)T α (l)(a), and ωl = ω ◦ V(N)Tl α (l−1). Notice that the last two are an element and a continuous functional on A(N)Tl α respectively. For any functor L from a category K to G(G, F, ρ), let Li = Pri ◦Lm for any i ∈ {1, 2}. Hence (A ◦ Lo, L1) is a functor from K to Ab, while (TTT ◦ Lo, L2) is a contravariant field under action of K. For any T ∈ MorK, h ∈ Ad(T), h ∈ A∗ c(T) and Q ∈ TTTc(T) we let hT = L1(T)(h), hT = h ◦ L1(T), and QT = L2(T)(Q). N, a ∈ A(N)T A nucleon-fragment doublet T on a category C is a tuple h S, J, Z, S, L, m, W, R, DDD, UUU i satisfying the following properties 2. (1) UUU and DDD are maps defined on subsets Dom(UUU) and Dom(DDD) of the object set of G(G, F, ρ) respectively, such that Dom(DDD) ⊆ Dom(UUU), UUUN ⊆ TTTN and DDDM ⊆ UUUM for all N ∈ Dom(UUU) and M ∈ Dom(DDD); (2) (UUU, Pr2) and (DDD, Pr2) are fields contravariant under action of Dom(UUU) and Dom(DDD)0 respectively; (3) the third item in (0.1.2) holds true by replacing in all the occurrances G(G, F, ρ) by Dom(UUU) and by Dom(DDD)0; (4) L is a functor from the category C to G(G, F, ρ); (5) S is a field of nucleon phases contravariant under action of Θ(UUU, L) via the conjugate of L1; (6) each fiber of S is covariant under action of H via the dual of ψ ◦ Lo; (7) m is a section of maps valued in S−valued maps, contravariant under action of Θ(UUU, L) via Sm and L2, whose values induce covariant sections under action of H via the dual of ψ ◦ Lo and via b ◦ Lo; (8) Z is a field contravariant under action of Ξ(DDD, L); (9) S is a map defined on the morphism set of Ξ(DDD, L) satisfying properties consistent α , for with Z and such that S(T, T, α) is a ∗−homomorphism from A(b)TT any a, b ∈ Ξ(DDD, L), T ∈ MorΞ(DDD,L)(b, a), T ∈ DDDa and α ∈ PT a ; α to A(a)T (10) J, a subfield of Z, is a field of disjoint union over operations of set of maps - with values fragment states originated via the nucleon phases determined by m - contravariant under action of Ξ(DDD, L) via Jm where J1 = L2 and J2 is a map induced by the conjugate of the evaluation of S; (11) each fiber of J is covariant under action of H via a map induced by the dual of V ◦ Lo; (12) W is a map such that gr ◦ W is a section of J, contravariant under action of Ξ(DDD, L) via Jm and L2, whose values induce sections covariant under action of H via the dual of V ◦ Lo and via b ◦ Lo. 2 the complete and detailed definition is given in Def. 2.2.66, here for simplicity we assume that R is the constant map with constant value equal to H. 12 CONTENTS Here Θ(UUU, L) is the subcategory of C inverse images via L of Dom(UUU) the full subcategy of G(G, F, ρ) whose object set is the domain of UUU; while Ξ(DDD, L) is the subcategory of C inverse images via L of Dom(DDD)0 a strict subcategy of G(G, F, ρ) whose object set is the domain of DDD. In particular Ξ(DDD, L) is a strict subcategory of Θ(UUU, L). J1 is the component of Jm acting on operations, and J2 is the component of Jm acting on maps of fragment states. The symmetries above described in (7) and (12) imply what we said in (4) in the initial list, we shall return later to these properties. For any object a of Ξ(DDD, L) and any T ∈ DDDa, ma(T) and Wa(T) are maps defined on PT a such that for all α ∈ PT a we have that (1) ma(T, α) is a nucleon phase of the nucleon system L(a) generated by the fissioning system a; (2) (0.1.6) ∃ r ∈ Repa(ma(T, α)) Wa(T, α) = ρr, Tr = T, αr = α; in particular (a) Wa(T, α) ∈ Na(ma(T, α)) namely Wa(T, α) is a fragment state originated via ma(T, α), (b) Wa(T, α) is a state of the fragment system O(a)T α and if α > 0 then the system evolves in time through (εa)T α whose operator algebra is α (−α−1(·)). A(a)T ma(T, α) is the phase of the nucleon system L(a), occurring by performing the operation T on the state of thermal equilibrium (ϕϕϕa)T α at inverse temperature α with respect to the dynamical system underlying O(a)T α . Since the definition of the map P we have that m(T, α) and W(T, α) occur by performing the operation T on (ϕϕϕa)T α only for those α ∈ IT a such that the symmetry group of the state of thermal equilibrium (ϕϕϕa)T α is larger than the symmetry group of the state of thermal equilibrium (ϕϕϕa)T , justifying (3) in the initial list. (βc)T a Let us call m and W the T−nucleon phase and T−fragment state respectively. For a , T ∈ MorΞ(DDD,L)(b, a) continuous functional ω on A(a)T α any a, b ∈ Ξ(DDD, L), T ∈ DDDa, α ∈ PT and element B of A(b)TT α we let ωT = ω ◦ S(T, T, α) and BT = S(T, T, α)B. The characteristics core of m and W consists of (0.1.6) relating the two natural trans- formations, and the properties of equivariance described in (7) and (12), namely for any d, e ∈ Θ(UUU, L), Q ∈ UUUd, β ∈ PQ d , Q ∈ MorΘ(UUU,L)(e, d), and a, b ∈ Ξ(DDD, L), T ∈ DDDa, α ∈ PT a , T ∈ MorΞ(DDD,L)(b, a), l ∈ H, we have (0.1.7) me(QQ, β) = md(Q, β)Q, md(Ql, β) = md(Q, β)l, Wb(TT, α) = Wa(T, α)T, Wa(Tl, α) = Wa(T, α)l. 0.1. INTRODUCTION 13 From the above equalities we deduce (4) in the initial list. More in general it is clear by the very definition that nucleon-fragment doublets generate resolutions of the claim, more specifically (0.1.6) and (0.1.7) determine the T−resolution of the equivariant form of the universality claim (M). It is worthwhile emphasizing that the concept of nucleon- fragment doublet is flexible enough to model diverse situations with symmetries H and Ξ(DDD, L), in which an origination mechanism of the type above described occurs. Thus the terms nucleon phase, fragment state, fissioning system and later prompt-neutron yield, have to be understood in a broad sense. A nucleon-fragment doublet on C can be related with what we call an extended C−equivariant stability, and the first main result of the entire work is the construction of the canonical extended Cu(H)−equivariant stability E•. The main difficulty is to construct m and W satisfying (0.1.6) and (0.1.7). Let T• denote the nucleon-fragment doublet on Cu(H) related to E•, in such a case Θ(UUU, L) = Cu(H) and Ξ(DDD, L) = C0 u(H). Let us call global nucleon phase and global fragment state the T•−nucleon phase and T•−fragment state respectively, then (0.1.6) and (0.1.7) applied to T• resolve the equivariant form of the universality claim (M). In the second main result of this work - under a suitable hypothesis ensuring the functoriality of the source and target functors of the second of the below morphisms - we encode (0.1.7) in the following result (0.1.8) m⋆ morphism of Fct(Cu(H)op, Fct(H, set)), u(H)op, Fct(H, set)), v♮ morphism of Fct(C0 related each other, modulo quotient, by the relation in (0.1.6), establishing in this way the compact equivariant form of the universality claim. Finally let us examine how the concept of nucleon-fragment doublet enables also to resolve the invariant form of the universality claim. To this end let us define a Terrell-type law corresponding to T, exhibiting invariances as a result of the above equivariances. For any a ∈ Ξ(DDD, L) let the set of fission processes occurring to the fissioning system a be the set NT as(a) of the 4−tuples n = h a, T, α, {f j, N j} j∈{m,w} i, satisfying properties related to the binary fission phenomenon and such that T, α are as above, fm and fw are observables of the nucleon system L(a), while Nm and Nw are positive observables, namely elements of (A(a)T as of fission processes be the union of the family {NT as(a)}a∈Ξ(DDD,L). Since the semantics outlined we have that α )+. Let the set NT (1) a is the fissioning system for which the fission process n occurs; (2) L(a) is the nucleon system generated by the fissioning system a; (3) O(a)T α is the fragment system whose observable algebra is A(a)T α and if α > 0 evolving in time through (εa)T α (−α−1(·)); (4) (ϕϕϕa)T α is a state of O(a)T α and if α > 0 then it is a state of thermal equilibrium at inverse temperature α with respect to the underlying dynamical system of O(a)T α ; 14 CONTENTS (5) ma(T, α) is the phase of L(a), occurring by performing T on (ϕϕϕa)T α ; (6) Wa(T, α) is the state of O(a)T (7) for j ∈ {m, w} α originated via ma(T, α); (a) ma(T, α)(f j) is the mean value in ma(T, α) of f j, (b) Wa(T, α)(N j) is the mean value in Wa(T, α) of N j. In order to decode the binary fission let us give the following interpretation of the data of n. Let #m =light and #w =heavy, then for all j ∈ {m, w} (1) T is the operation realizing the fission process n whenever performed on (ϕϕϕa)T α ; (2) N j is the observable of O(a)T α relative to the mass of the # j fragment; (3) f j is the observable of L(a) relative to the mass of the # j nucleon core. Next to NT as the action of H naturally extends, since for all l ∈ H we can set kT(l)(n) = h a, Tl, α, {fl j, Nl j} j∈{m,w} i, later simply denoted by nl, as well "essentially" extends the contravariant action of Ξ(DDD, L), meaning that for any object b of Ξ(DDD, L) and any morphism T of Ξ(DDD, L) from b to a we have where x = {f′ modify this map in order to have a "true" contravariant action. j} j∈{m,w} such that (f′ j, N′ j)T = N j. Later we shall see how to n(T,x) = h b, TT, α, {f′ j, N′ j)T = f j and (N′ j} j∈{m,w} i, For all j ∈ {m, w} define κT j and ζT j the functions on NT as mapping any n into j (n) ≔ Wa(T, α)(N j), κT j (n) ≔ ma(T, α)(f j); ζT called restricted T − # j fragment mass and restricted T − # j nucleon mass, collectively called restricted T−fragment masses and T−nucleon masses respectively. Next the restricted T−Terrell law is the function νT on NT as mapping any n into the mean value νT(n) of the prompt-neutron yield in Wa(T, α), said also the prompt-neutron yield of the fission process n, where νT ≔ 0.08(κT m − ζT m) + 0.1(κT w − ζT w). As a result of (0.1.7) we obtain under essential contravariant action of Ξ(DDD, L) and action of H the invariance of the restricted T−fragment and nucleon masses, i.e. for all j ∈ {m, w} (0.1.9) j (n(T,x)) = κT κT j (nl) = κT κT j (n(T,x)) = ζT ζT j (nl) = ζT ζT j (n), j (n), j (n), j (n); 0.1. INTRODUCTION 15 therefore the invariance of νT follows (0.1.10) νT(n(T,x)) = νT(n), νT(nl) = νT(n). (0.1.9) and (0.1.10) constitute the invariance of the restricted T−fragment and nucleon masses, and the restricted T−Terrell law respectively. The properties of invariance of the restricted global fragment and nucleon masses, and restricted global Terrell law follow for T = T•. Now in order to have a true contravariant action and then to resolve the invariant form of the universality claim let us procede as follows. Let PT = (PT m) be the couple of maps defined on the object set and morphism set of Ξ(DDD, L) respectively, such that PT o (a) preserving the union j} j∈{m,w} such that (f′ and mapping any {n} into the set of the n(T,x) where x = {f′ j)T = f j and (N′ m(T) is the map on PT j, N′ o (a) is the power set of NT as(a), while PT o , PT j)T = N j. We call T − # j nucleon mass and T − # j fragment mass the maps µT the object set of Ξ(DDD, L) such that (µT PT we call T-Terrell law the map θT defined on the object set of Ξ(DDD, L) such that θT P(R)−valued extension to PT j defined on j )a are the P(R)−valued extensions to as(a), with j ∈ {m, w}. Moreover a is the j )a and (λT j respectively to NT o (a) of the restrictions of ζT o (a) of the restriction of νT to NT j and κT j and λT Now PT results to be a functor from Ξ(DDD, L)op to set, since (0.1.7). Let RT denote the unique functor from Ξ(DDD, L)op to set such that its object map is the constant map with constant value equal the power set P(R) of R and its morphism map is the constant map with constant value equal IdP(R). Let RH denote the unique functor from H to set such that RH m is the constant map with constant value equal to IdP(R), Finally let τT o (a) for all l ∈ H, and set QT a ). Thus as a result of (0.1.7) the universality of the T−nucleon masses and T−Terrell law follows for all j ∈ {m, w} in terms of a the map on H such that τT a = (1 7→ PT a (l) is the extension of kT(l) to PT o = (1 7→ P(R)) and RH o (a), τT as(a). (1) Invariance of the T−nucleon and fragment masses and T−Terrell law under contravari- ant action of Ξ(DDD, L) µT j , λT j , θT ∈ MorFct(Ξ(DDD,L)op,set)(PT, RT). (0.1.11) (2) Invariance of the T−nucleon and fragment masses and T−Terrell law under action of H. For all a ∈ Ξ(DDD, L) (0.1.12) (1 7→ (µ j)T a), (1 7→ (λ j)T a), (1 7→ θT a) ∈ MorFct(H,set)(QT a, RH). In other words similarly for λT j , and (µT j )d(T) ◦ PT (µT j )a ◦ τT m(T) = (µT a(l) = (µT j )c(T), j )a, ∀l ∈ H θT d(T) ◦ PT a ◦ τT θT m(T) = θT a(l) = θT c(T), a, ∀l ∈ H. 16 CONTENTS Here the contravariance of PT under action of Ξ(DDD, L) means that for all T, S ∈ MorΞ(DDD,L) such that d(T) = c(S) PT m(T)PT PT o (c(T)) ⊆ PT m(T ◦ S) = PT o (d(T)), m(S) ◦ PT m(T). Thus (0.1.11) and (0.1.12) jointly (0.1.6) and (0.1.7) represent the T−resolution of the invariant form of the universality claim. j , λT• If we call global # j nucleon mass, global # j fragment mass and global Terrell law the maps µT• j and θT• respectively, then the universality of the global nucleon and fragment masses and the universality of the global Terrell law follow since (0.1.11) and (0.1.12) for T = T•, resolving jointly (0.1.6) and (0.1.7) the invariant form of the universality claim. This is the third main result of this work. Incidentally the invariant form of the universality claim results as a consequence of its equivariant form. Finally let the # j nucleon and fragment masses, and Terrell law denote the restrictions j , κT• of ζT• j and νT• respectively to the set of all the fission processes n satisfying the revised nucleon phase hypothesis requiring ζT• w (n) = 126 and that H would contain as a subgroup the direct product of the universal covering group of the Poincar´e group with the gauge group of the standard model. Thus under the revised nucleon phase hypothesis as a result of (0.1.9) and (0.1.10) we can state what follows. The light and heavy nucleon masses are invariant with constant values 82 and 126 and the prompt- neutron yield (0.1.1) is invariant under essential contravariant action over the field of fission processes of suitable perturbations of fissioning systems, and under action over the field of fission processes of relativistic transformations of reference frames. m (n) = 82, ζT• In conclusion it is worthwhile remarking that the universality of the global nucleon masses and the universality of the global Terrell law - in the special form of the stability of the values 82 and 126 occurring in the Terrell law and the invariance of the Terrell law itself under the above transformations mainly the relativistic ones - is an experimentally testable property that can be provided in order to furnish an indirect empirical evidence of the existence and universality of the global nucleon phase m. Let us outline the main content of the three parts. In part 2 we introduce the category G(G, F, ρ) of nucleon systems, set its physical in- terpretation, define the concepts of extended C−equivariant stability, nucleon-fragment doublet T on a category C, define the T−nucleon phase, the T−fragment state and the T−resolution of the equivariant form of the universality claim. Then in the first main result of this work we construct, and in this way establish the existence of, the canonical extended Cu(H)−equivariant stability and the canonical nucleon-fragment doublet T• on Cu(H). As a result we resolve the equivariant form of the universality claim, by applying its T−resolution to T•. In part 3 we establish in a more coincise and elegant form the equivariant form of the universality claim. Namely in the second main result of this work we encode the equivariances (0.1.7) applied for T = T• in a unique fashion into the existence of natural transformations m⋆ and v♮ satisfying (0.1.8). Then we define the T−nucleon and 0.2. TERMINOLOGY AND PRELIMINARIES 17 T−fragment light and heavy masses, and the T−Terrell law, establish the T−resolution of the invariant form of the universality claim, and apply it to T• in order to establish in the third main result of this work the invariant form of the universality claim, in particular the universality of the global Terrell law. 0.2. Terminology and preliminaries 0.2.1. Sets and topologies. In all three parts we consider the Zermelo-Fraenkel the- ory together the axiom of universes stating that for any set there exists a universe containing it as an element [23, p. 10]. We consider fixed a universe U and a universe U0 such that U ∈ U0. See [2, Expose I Appendice] for the definition and properties of universes, see also [25, p. 22] and [7, §1.1], in particular if V is a universe, then y ∈ V implies y ⊂ V, hence U ⊂ U0. Let set be the category of sets belonging to the universe U, functions as morphisms with map composition. Whenever we refer to a set unless the contrary is stated, we mean an element of U, moreover for any structure S (e.g. the structure of topological space, topological algebra, etc.) whenever we refer to "the set of the S's", we always mean the subset of those elements of U satisfying the axioms of S. Let Gr be the category of groups and group morphisms with map composition and let Ab be the full subcategory of Gr of abelian groups. If A, B and C are categories, then we let Obj(A) denote the set of objects of A, we let a ∈ A denote a ∈ Obj(A). For any x, y ∈ A let MorA(x, y) be the set of morphisms of A from x to y, let 1x be the identity morphism of x, while InvA(x, y) = { f ∈ MorA(x, y) (∃g ∈ MorA(y, x))( f ◦ g = 1y, g ◦ f = 1x))} denotes the, possibly empty if x , y, set of invertible morphisms from x to y; set AutA(y) = InvA(y, y), for any y ∈ A. Given two sets A, B let P(A) and Pω(A) denote the set of subsets and finite subsets of A respectively. For any map f : A → B let f −1 : P(B) → P(A) such that f −1(Y) ≔ {a ∈ A f (a) ∈ Y} for any Y ⊆ B. Given any set x, often and only if it will not cause confusion, we use the convention to denote {x} by x, so for example if f : A → B and b ∈ B, then f −1(b) stands for f −1({b}). Let ev(·) denote the evaluation map, i.e. if F : A → B is any map and a ∈ A, then eva(F) ≔ F(a). of set and Ba, Ca objects of set for all a ∈ A, then we let x(a, b) denote x(a)(b) for any If x ∈ Qa∈A Morset(Ba, Ca), with A object b ∈ Ba. If Y(a, b) = Y ∈ U0 for all a ∈ A and b ∈ Ba, then we let Qa∈AQb∈Ba Y denote Qa∈AQb∈Ba Y(a, b). If f : X → A and g : X → B then by abuse of the standard language Set N0 ≔ N − {0} and R0 ≔ R − {0}, while eR ≔ R ∪ {∞} provided by the topology we denote by f × g the map on X with values in A × B such that ( f × g)(x) ≔ ( f (x), g(x)) for all x ∈ X. If A is any set then IdA is the identity map on A we of one-point compactification. often use the convention to remove the index A if it is clear the set involved. If S is a topological space, then Cl(S), Op(S) and Comp(S) denote the sets of closed, open and compact subsets of S respectively, while B(S) denotes the σ−field of Borel subsets of S. If T is a locally compact space and E is a Hausdorff locally convex space, let Cc(T, E) denote the linear space of continuous E−valued maps f on T with compact support 18 CONTENTS supp( f ), where supp( f ) ≔ f −1(E − {0}), set Cc(T) ≔ Cc(T, C) provided by the inductive limit topology of uniform convergence over compact subsets of T. Let X and T be a locally compact group and locally compact space respectively, then H(X) is the set of Haar measures on X and M(T) is the set of Radon measures on T. Let µ ∈ M(T) and S be a locally compact subspace of T, set µS : Cc(S) → C such that µS( f ) ≔ µ(ef ), where ef is the 0−extension on T of f ∈ Cc(S), thus µS ∈ M(S). Let T, S be two locally compact spaces, µ ∈ M(T) and ε : T → S be µ−proper, thus ε(µ) denotes the image of µ under ε as defined in [12, Ch. 5, §6, n◦1, Def. 1]. By construction ε(µ) ∈ M(S) such that for all f ∈ Cc(S) (0.2.1) Z f dε(µ) =Z f ◦ ε dµ. Let X be a locally compact group and s ∈ X, set Ls, Rs : X → X, such that Ls(x) ≔ s · x and Rs(x) ≔ x · s, while L∗ s(h) ≔ h ◦ Rs−1 respectively. Whenever we will deal with different groups, it will be clear by the context to which group the maps R and L are referring to. By definition H(X) is the set of left-invariant µ ∈ M(X), i.e. µ ∈ M(X) such that µ ◦ L∗ s : CX → CX such that L∗ s(h) ≔ h ◦ Ls−1 and R∗ s ↾ Cc(X) = µ, for all s ∈ X. s, R∗ If X and Y are two topological linear spaces over K ∈ {R, C}, L(X, Y) denotes the linear space of continuous linear maps from X to Y, set L(X) ≔ L(X, X) and X∗ ≔ L(X, K). Ls(X, Y) is the topological linear space whose underlying linear space is L(X, Y) provided by the topology of pointwise convergence, while Lw(X, Y) is the locally convex linear space whose underlying linear space is L(X, Y) provided by the topology generated by the following set of seminorms {q(φ,x) (φ, x) ∈ Y∗ × X}, where q(φ,x)(A) (cid:17) φ(Ax). All the normed spaces in this work are assumed to be over the complex field. In case X is a normed space we assume L(X) to be provided by the topology generated by the usual sup −norm. If X is any structure including as a substructure the one of normed space say X0, for example the normed space underlying any normed algebra, we let L(X) denote the normed space L(X0) whenever it does not cause conflict of notations. Therefore we never shall use this convention in case X is an Hilbert C∗−module, where L(X) always denotes the set of all adjointable operators on X, as we shall see in the second part. If A and B are two linear operators in X, we set [A, B] ≔ AB − BA, where the composition and sum are to be understood in the context of possibly unbounded operators, i.e. defined on the intersection of the corresponding domains. If X, Y are Hilbert spaces and U ∈ L(X, Y) is unitary then ad(U) ∈ L(L(X), L(Y)) denotes the isometry defined by ad(U)(a) ≔ UaU−1, for all a ∈ L(X). Let HS denote the set of Hilbert spaces. Finally unless the contrary is not stated any convention established in one part has to be understood valid for the remaining of that part and for the following parts. 0.2.2. C∗−algebras, C∗−dynamical systems and their crossed products. For any normed algebra D let RD (·) : D → L(D) denote the right and left multi- plication map on D respectively, i.e. Ra(b) = ba and La(b) = ab for any a, b ∈ D, often we re- move the index D. Let A be a C∗−algebra, let Aob ≔ {a ∈ A a = a∗} and A+ ≔ {a∗a a ∈ A}. EA denotes the set of states of A, if A is a von Neumann algebra NA denotes the set (·) : D → L(D) and LD 0.2. TERMINOLOGY AND PRELIMINARIES 19 of its normal states. If B is a C∗−algebra Hom∗(A, B) is the set of ∗−homomorphisms defined on A and at values in B, Isom∗(A, B) is the subset of bijective elements of Hom∗(A, B) and Aut∗(A) = Isom∗(A, A). Let CA∗ denote the category of C∗−algebras and ∗−homomorphisms with map composition as law of morphism composition, eas- ily we deduce that Isom∗(A, B) = InvCA∗(A, B), so Aut∗(A) = AutCA∗(A). If in addition A and B are unital, then Hom∗ 1(A, B) is the set of unit preserving ∗−homomorphisms defined on A and at values in B. Let A+ denote the ∗−algebra whose underlying lin- ear space is A × C, the involution and product are defined by (a, λ)∗ = (a∗, λ), and (a, λ) · (b, µ) ≔ (a · b + λb + µa, λµ), for all (a, λ), (b, µ) ∈ A × C. ([28, Ch. 2, §7, n◦2, I and §10, n◦1, III]). (0, 1) is the unity of A+, while the map φ : a 7→ (a, 0) is an injective ∗−isomorphism of A into A+, we often shall identify A with its image in A+ under φ. Under this identification A is a two side ideal of A+, so we have that Lx, Rx ↾ A are linear endomorphisms of the vector space underlying A for any x ∈ A+. A+ is a C∗−algebra if provided by the following norm extending the one on A, kxk ≔ kLx ↾ Ak, well-set A the smallest unital since L(a,λ) ↾ A = La + λ· ∈ L(A) ([28, Ch. 3, §16, n◦1, III]). Set subalgebra of A+ containing A, so A = φ(A), if A has the identity, A = A+ otherwise. Let B a ∗−algebra and α ∈ MorCA∗(A, B), set α+ : A+ ∋ (a, λ) 7→ (α(a), λ) ∈ B+, thus (0.2.2) α+ ∈ Hom∗ 1(A+, B+), while in case B has the identity, we set (0.2.3) ( α : A+ → B, (a, λ) 7→ α(a) + 1Bλ, α ∈ Hom∗ thus (∗−representation in case B equals L(H) for some Hilbert space H). called the ∗−homomorphism of A+ induced by α 1(A+, B), Let Aut∗ s(A) denote the topological group of ∗−automorphisms of A provided by the topology of pointwise convergence. Rep(A) denotes the set of ∗−representations of A, while Repc(A) denotes the set of cyclic ∗−representations of A. If HHH = h H, π, Ω i and KKK = h K, ξ, Ψ i are in Repc(A) then we call them unitarily equivalent if their underlying representations of A are unitarily equivalent, say through the unitary operator U : H → K, and UΩ = Ψ. If π is a nondegenerate representation of A on H, then we denote by Nπ and call π−normal states its elements, the set of the φ ◦ π where φ ∈ NL(H). Since π is nondegenerate Nπ ⊂ EA. If ψ is a state of A then ψ−normal means π−normal, where h H, π, Ω i is a cyclic representation associated with ψ. If T ∈ MorCA∗(A, B), set T† : B∗ → A∗ such that T†(ω) = ω ◦ T, if λ ∈ InvCA∗(A, B), then set λ∗ = (λ−1)†. For any map f : D ⊆ A → A define the map ad(λ)(f) : λ(D) → A such that for all b ∈ λ(D) (0.2.4) ad(λ)(f)(b) ≔ (λ ◦ f ◦ λ−1)(b). Let U be any set and γ : U → AutCA∗(A), define EU A(γ) ≔ {ψ ∈ EA (∀u ∈ U)(ψ ◦ γ(u) = ψ)}. Let A = h A, H, σ i and B = h B, H, θ i be C∗−dynamical systems, here called sim- ply dynamical systems or dynamical systems with group symmetry H. T is an 20 CONTENTS (A, B)−equivariant morphism, or equivalently, (σ, θ)−equivariant morphism if T ∈ MorCA∗(A, B) and T ◦ σ(h) = θ(h) ◦ T for all h ∈ H. h H, π, W i is a (nondegenerate) covariant representation of A if h H, π i is a (nondegenerate) ∗−representation of A, W is a strongly continuous unitary representation of H on H such that for all h ∈ H π ◦ σ(h) = ad(W(h)) ◦ π. We denote by Cov(A) the set of nondegenerate covariant representations of A. h HHH, W i is a cyclic covariant representation of A if HHH = h H, π, Ω i is a cyclic representation of A, h H, π, W i is a covariant representation of A and W(H){Ω} = {Ω}. Let ϕ ∈ EH A(σ) and HHH = h H, π, Ω i be a cyclic representation of A associated with ϕ. Set Wσ HHH : H → L(H) such that for all h ∈ H and a ∈ A (0.2.5) Wσ HHH(h)π(a)Ω = π(σ(h)a)Ω. Then h HHH, Wσ i is a cyclic covariant representation of A called the cyclic covariant repre- HHH sentation of A induced by HHH. We convein to remove the index σ whenever it does not cause confusion. Set U(A) ≔ {U ∈ A U∗ = U−1} provided by the group structure inherited by the product on A, and U(H) ≔ U(L(H)) for any Hilbert space H. We say that σ is inner if there exists a group morphism v : H → U(A) such that σ = ad ◦ v, in such a case we say that σ is inner implemented by v, or that v implements unitarily σ. A is said inner implemented by v if σ is so. If A is a von Neumann algebra, it can be always considered in standard form since [5, III.2.2.26] and [36, Def. 9.1.18], called its canonical standard form, then by [36, Thm 9.1.15] we deduce that σ is inner, moreover said Ust(A) the subgroup of U(A) whose elements u satisfy uJu∗ = J and uL2(A)+ = L2(A)+, see [36, Def. 9.1.18] for the notations, there exists a unique group action V : H → Ust(A) inner implementing σ. Let µ ∈ H(H) and let Cµ c (H, A) denote the ∗−algebra whose underlying linear space is Cc(H, A), while the product and involution are respectively ∗µ and ∗ such that for all f, g ∈ Cc(H, A) and s ∈ H ([38, eqs.2.16 − 2.17]) (0.2.6) ( f ∗µ g)(s) ≔Z f (r)σ(r)(cid:16)g(r−1s)(cid:17) dµ(r), f ∗(s) ≔ ∆H(s−1)σ(s)( f (s−1)∗), σ H the where the integration is w.r.t. C∗−crossed product of A by H associated to µ, see [38, Lemma 2.27]. It is defined as the C∗−algebra completion of the normed ∗−algebra CCCµ c (H, A), k · kµ i, where k · kµ is the µ−universal norm such that for any f ∈ Cc(H, A) ([38, eq. 2.22 and Lm. 2.31]) to the norm topology on A. Denote by A ⋊µ c (H, A) ≔ h Cµ k f kµ = supnk(π ⋊µ u)( f )kL(H) h H, π, u i ∈ Cov(A)o, where ([38, Prp. 2.23]) (0.2.7) (π ⋊µ u)( f ) ≔Z π( f (s))u(s) dµ(s). 0.2. TERMINOLOGY AND PRELIMINARIES 21 Here the integral is valued in the locally convex space Ls(H), i.e. it is an element of L(H) and the integration is w.r.t. the strong operator topology on L(H). Its existence is ensured by Rmk. 1.1.37 and by the fact that the product is continuous as a map on Ls(H) × Ls(H)1 with values in Ls(H), where Ls(H)1 is the topological subspace of Ls(H) of its elements with norm less or equal to 1. We call the following L1 µ−norm Cc(H, A) ∋ f 7→Z k f (s)k dµ(s). It is worthwhile a remark about notations. It is well-known that the Haar measure is unique up to a constant factor [13, Th. 1 §1 n◦2], nevertheless in this work, at difference with the standard usage, we prefer to mention expressly which Haar measure we use in (0.2.6,0.2.7) and a fortiori in A ⋊µ σ H. Since [38, Prp. 2.39.] if h H, π, W i is a covariant representation of A, then π ⋊µ W σ H which is nondegenerate extends uniquely by continuity to a ∗−representation of A ⋊µ if it is so h H, π, W i. Let G and F be two topological groups, ρ : F → AutGr(G) a group homomorphism such that the map (g, f ) 7→ ρ f (g) on G × F at values in G, is continuous, where AutGr(G) is the group of automorphisms of the group underlying G. Thus we let G ⋊ρ F denote the external topological semi-direct product of G and F relative to ρ, see [9, III.19]. By definition for all (g1, h1), (g2, h2) ∈ G ⋊ρ F (g1, h1) ·ρ (g2, h2) ≔ (g1 · ρh1(g2), h1 · h2). Moreover j1 : G → G⋊ρF and j2 : F → G⋊ρF will be the (continuous) canonical injections. Since [9, Prp. 14, I.66] G ⋊ρ F is locally compact if and only if G and F are locally compact, so in this case we can consider a dynamical system A = h A, G ⋊ρ F, σ i. In addition let F0 be a topological subgroup of F, ξ : R → G be a continuous group homomorphism and l ∈ G ⋊ρ F, set τσ ≔ σ ◦ j1 γσ ≔ σ ◦ j2 SG F0 σF0 τ(l,ξ) σ ≔ τσ ◦ ad(l) ◦ j1 ◦ ξ. ≔ G ⋊ρ F0 ≔ σ ↾ SG F0 Here and thereafter we use the convention to denote ρ ↾ F0 simply by ρ anytime it is clear by the context w.r.t. which subset F0 of F the restriction has been performed. τ(l,ξ) is well-set, since j1(G) is a normal subgroup of G ⋊ρ F, moreover we have for all h ∈ G ⋊ρ F σ (0.2.8) τ(l·ρh,ξ) σ = ad(σ(l)) ◦ τ(h,ξ) σ . Whenever it is clear by the context which dynamical system is involved we convein to remove the index σ. If F0 is locally compact, for instance closed in F by [9, Prp. 13, I.66], then SG , σF0 i is a F0 is locally compact since [9, Prp. 14, I.66]. Therefore h A, SG F0 22 CONTENTS σ SG F0 dynamical system, and the crossed product A ⋊µ ). For any domain D in C, i.e. open simply connected subset of C, denote by H(D) the set of analytic maps on D with values in C, while by Hw(D, A) and Hs(D, A) the set of weak and strong analytic maps on D with values in A. Let f : D → A, then by definition f ∈ Hw(D, A) if φ ◦ f ∈ H(D) for all φ ∈ A∗, while f ∈ Hs(D, A) if f is C−derivable, thus differentiable. Thus Hs(D, A) ⊆ Hw(D, A). Let h A, R, η i be a dynamical system. Set δη to be the infinitesimal generator of the strongly continuous semigroup η ↾ R+ acting on A, i.e. a ∈ Dom(δη) iff the following limit exists in the norm topology of A is well-set for any µ ∈ H(SG F0 (0.2.9) δη(a) ≔ lim t→0, t,0 η(t)(a) − a t . Aη is the ∗−subalgebra of A of the entire analytic elements of η, see [14, Def. 2.5.20.], and η : C → AAη denote the unique entire analytic extension of η, i.e. for any a ∈ Aη the map C ∋ z 7→ η(z)(a) ∈ A belongs to Hw(C, A), so to Hs(C, A) since [14, Prp. 2.5.21.], and it is the necessarily unique analytic extension of R ∋ t 7→ η(t)(a) ∈ A. The uniqueness follows by [14, Def. 2.5.20.] and by the uniqueness of entire analytic extension of numerical maps, see [34, Cor. of Thm. 10.18]. We use the definition of β − KMS given in [15, Def. 5.3.1], namely for any β ∈ R the set Kη β of the β − KMS states with respect to the dynamical system h A, R, η i is the set of the states ω ∈ EA such that there exists an η−invariant, norm dense ∗−subalgebra D of A such that ω(aη(iβ)b) = ω(ba) for all a, b ∈ D. If we replace D by A we obtain an equivalent statement, see [31, Prp. 8.12.3]. Let the set of η − KMS states denote Kη ∞ is defined in [15, Def. 5.3.18]. −1, finally Kη 0.2.3. Borelian functional calculus of possibly unbounded scalar type spectral operators in Banach spaces. Let S be a set, denote by B(S) the Banach space of bounded complex maps on S with the norm k f k = sups f (s). Let B be a field of subsets of S, a complex map defined on S is B−measurable if f −1(A) ∈ B, for any Borel set A of C. Denote by TM(B) the closure in B(S) of the linear subspace I(B) generated by the set {χδ δ ∈ B}, where χδ is the characteristic map of the set δ. TM(B) as a normed subspace of B(S) is a Banach space, moreover the space of all bounded B−measurable maps on S is contained in TM(B). Let X be a Banach space, a map F : B → X is defined to be additive if F(∅) = 0 and F(∪n i=1 ⊂ B of disjoint sets, while F is said to be bounded if supδ∈B kF(δ)k < ∞. Let F : B → X be a bounded additive map, thus we can define IF : I(B) → X such that i=1 F(σi), for any n ∈ N and any family {σi}n i=1σi) =Pn nXi=1 IF( λiχσi) ≔ nXi=1 λiF(σi), (0.2.10) for any n ∈ N, any family {σi}n IF is a well defined linear bounded operator, so admits the linear extension by continuity to the space TM(B), we shall denote this extension again by IF ([18, 10.1]). If Y is a Banach space and Ψ ∈ L(X, Y), then i=1 ⊂ B of disjoint sets and {λi}n i=1 ⊂ C. (0.2.11) Ψ ◦ IF = IΨ◦F. 0.2. TERMINOLOGY AND PRELIMINARIES 23 Assume now in addition that B is a σ−field, let G be a complex Banach space, Pr(G) the subset of P ∈ L(G) such that PP = P and 1G and 0G be the identity and the zero operator on G respectively, we convein to remove the index G whenever it is clear by the context which space is involved. E is defined to be a countably additive spectral measure in G on B if for all {αn}n∈N ⊂ B disjoint sets and δ1, δ2 ∈ B we have (1) E(B) ⊆ Pr(G), (2) E(δ1 ∩ δ2) = E(δ1)E(δ2), (3) E(δ1 ∪ δ2) = E(δ1) + E(δ2) − E(δ1)E(δ2), (4) E(S) = 1, (5) E(∅) = 0, (6) E(Sn∈N αn) =P∞ n=1 E(αn) w.r.t. the weak operator topology on L(G). the strong operator topology on G It results that the convergence in (6) holds w.r.t. and E is bounded ([19, Cor 15.2.4]), hence IE : B → L(G) is well defined. In case S is a topological space and B is generated as σ−field by a basis of the topology on S containing S, then B(S) ⊆ B so we can set ([3, Ch. 2, pg. 126]) supp(E) ≔ \{δ∈BE(δ)=1, δ∈Cl(S)} δ, in particular supp(E) is closed. By using the same argument to prove [35, eq. (1.5)] we obtain (0.2.12) E(supp(E)) = 1. We are now able to define the functional calculus (shortly f.c.) associated to a countably additive spectral measure. Let • S be a set and B a σ−field of subsets of S; • E be a countably additive spectral measure in G on B; • f a B−measurable map; • fσ ≔ f χσ, for all σ ∈ B; • δn ≔ [−n, n] and fn ≔ f f −1(δn), for all n ∈ N. Thus fn ∈ TM(B) being B−measurable and bounded, hence we are able to set ([19, Def. 18.2.10] and [35, Def. 1.3.]) (0.2.13) (Dom( f (E)) ≔ {x ∈ G limn∈N IE( fn)x exists}, f (E)x ≔ limn∈N IE( fn)x, ∀x ∈ Dom( f (E)). The map f 7→ f (E) is called the functional calculus of the spectral measure E and f (E) is a densely defined closed linear operator in G ([19, Thm. 17.2.11]). It is easy to show that f (E)E(σ) = ( f χσ)(E), for all σ ∈ B. In case S is a topological space and B is generated as σ−field by a basis of the topology on S containing S, then since (0.2.12) we obtain for any f, g : S → C B−measurable maps (0.2.14) f ↾ supp(E) = g ↾ supp(E) ⇒ f (E) = g(E). 24 CONTENTS A linear possibly unbounded operator R in G is called a scalar type spectral operator in G if there exists a countably additive spectral measure F in G on B(C), such that R = ı(F), where ı is the identity map on C. We call F a resolution of the identity (shortly r.o.i) of R. There exists a unique r.o.i of R ([19, Cor. 18.2.14]) denoted by ER, moreover ([19, Lemmas 18.2.13 and 18.2.25]) (0.2.15) sp(R) = supp(ER). Denote by Bor(C) the set of the Borelian maps, i.e. B(C)− measurable maps, thus for any g ∈ Bor(C) we convein to denote the operator g(ER) by g(R) and call the map Bor(C) ∋ g 7→ g(R) the Borelian functional calculus of R ([19, Def. 18.2.15]). If E is a countable additive spectral measure in G on a σ−field B of subsets of a set S, then for any B−measurable map f the operator f (E) is a scalar type spectral operator in G whose resolution of the identity is ([19, Thm. 18.2.17]) (0.2.16) E f (E) = E ◦ f −1. 0.2.4. Joint RI constructed from {Ex}x∈X. The material, and with some adaptation the notations, of the present section 0.2.4 arises from [3, Ch. 2, §1.3]. Let S be a set, B a σ−field of subsets of S, H be a Hilbert space, then E is defined to be a RI in H on B if it is a countably additive spectral measure in H on B such that E(σ) is selfadjoint, for all δ ∈ B. If S is a topological space, a RI in H on B(S) is called a Borel RI in H on S. Let X be a set, Pω(X) the set of its finite parts, R (cid:17) {Rx}x∈X, where Rx is a complete separable metric space for all x ∈ X. {Ex}x∈X is defined to be a family of commuting Borel RI's in H on R, if Ex is a Borel RI in H on Rx, for all x ∈ X, and [Ey(σy), Ez(σz)] = 0, for all y, z ∈ X and σq ∈ B(Rq), q ∈ {y, z}. We convein to avoid mentioning R, whenever Rx = C, for all x ∈ X. For any Y ⊆ X let RY denoteQx∈Y Rx and if Q = {x1, . . . , xp} ⊆ X, Rx1,...,xp denotes RQ. Set (C(RX) ≔S{C(Q, δ) Q ∈ Pω(X), δ ∈ B(RQ)}, C(Q, δ) ≔ {λ ∈ RX λ ↾ Q ∈ δ}, ∀Q ∈ Pω(X), δ ∈ B(RQ). C(RX) is a field of subsets of RX, called the field of cylindrical sets, Cσ(RX) denotes the σ−field generated by C(RX). Let A ∈ P(RX), we recall that B(RA) is the σ−field generated by the set Ψ(Qx∈A B(Rx)), where Ψ :Qx∈A B(Rx) → P(RA) is such that Ψ(δδδ) ≔Qx∈A δδδx, for all δδδ ∈ Qx∈A B(Rx). Let E (cid:17) {Ex}x∈X be a family of commuting Borel RI's in H on R, then there exists a unique RI in H on Cσ(RX), denoted by x∈X Ex and called the joint RI constructed from E, such that for all Q ∈ Pω(X) and δδδ ∈Qy∈Q B(Ry) we have ([3, Ch. 2, §1.3, Thm. 1.3]) Ex)(C(Q,Yy∈Q δδδy)) =Yy∈Q Ey(δδδy), (0.2.17) ( x∈X where Qy∈Q Ex(δδδy) ≔ Ex1(δδδx1) ◦ · · · ◦ Exp(δδδxp), if Q = {x1, . . . , xp}, well-defined since the commutativity property. Let f denotes (x∈X Ex) ◦ f −1 defined on B, and EE denotes EId : RX → S be a (Cσ(RX), B) measurable map, then E f E in case S = RX and B = Cσ(RX). E 0.2. TERMINOLOGY AND PRELIMINARIES 25 0.2.5. Categories. We recall that U and U0 are fixed universes such that U ∈ U0, set is the category of sets belonging to the universe U, functions as morphisms with map composition. Whenever we refer to a set unless the contrary is stated, we mean an element of U, moreover for any structure S whenever we refer to "the set of the S's", we always mean the subset of those elements of U satisfying the axioms of S. We recall that we let Gr be the category of groups and group morphisms with map composition, and Ab be the full subcategory of Gr of abelian groups. For any set X and any group G by abuse of language we let Morset(X, G) denote also the object of Gr with pointwise composition, inversion and identity. If the contrary is not stated, any diagram involving maps between sets of U has to be understood in set. Let A, B and C be categories. Let Obj(A) denote the set of objects of A, we let a ∈ A denote a ∈ Obj(A). For any x, y ∈ A let MorA(x, y) be the set of morphisms of A from x to y, let 1x be the identity morphism of x, while InvA(x, y) = { f ∈ MorA(x, y) (∃g ∈ MorA(y, x))( f ◦ g = 1y, g ◦ f = 1x))} denotes the set of invertible morphisms from x to y. Set MorA =S{MorA(x, y) x, y ∈ A} and InvA = S{InvA(x, y) x, y ∈ A}, while AutA(y) = InvA(y, y), for y ∈ A, and AutA = S{AutA(x) x ∈ A}. We shall consider AutA(y) with its natural group structure. For any T ∈ MorA(x, y) we set d(T) = x called domain or source of T and c(T) = y called codomain or target of T, the composition on MorA is always denoted by ◦. Let Aop denote the opposite category of A, [25, p. 33]. Let Fct(A, B) denote the category of functors from A to B and natural transformations provided by pointwise composition, see [23, 1.3], [25, p.40], [7, p. 10]. Let us identify any F ∈ Fct(A, B) with the couple (Fo, Fm), where Fo : Obj(A) → Obj(B) called the object map of F while Fm : MorA → MorB, called the morphism map of F is such that Fx,y m : MorA(x, y) → MorB(Fo(x), Fo(y)) where Fx,y m = Fm ↾ MorA(x, y) for all x, y ∈ A. For any object x and morphism t of A we always let F(x) and F(t) denote Fo(x) and Fm(t) respectively, while we shall never use this convention for the maps Fo and Fm. Let ◦ : Fct(B, C) × Fct(A, B) → Fct(A, C) denote the standard composition of functors [23, Def. 1.2.10], [25, p. 14], for any σ ∈ Fct(A, B), the identity morphism 1σ of σ in the category Fct(A, B) is such that 1σ(M) = 1σo(M), for all M ∈ A. Let β ∗ α ∈ MorFct(A,C)(H ◦ F, K ◦ G) be the Godement product between the natural transformations β and α, where H, K ∈ Fct(B, C) and F, G ∈ Fct(A, B) while β ∈ MorFct(B,C)(H, K) and α ∈ MorFct(A,B)(F, G), see [7, Prp. 1.3.4] (or [25, p. 42] where it is used the symbol ◦ instead of ∗). The product ∗ is associative, in addition we have the following rule for all γ ∈ MorFct(A,B)(H, L), α ∈ MorFct(A,B)(F, H), and δ ∈ MorFct(B,C)(K, M), β ∈ MorFct(B,C)(G, K), see [7, Prp. 1.3.5] (0.2.18) We have (0.2.19) (δ ∗ γ) ◦ (β ∗ α) = (δ ◦ β) ∗ (γ ◦ α). β ∗ 1F = β ◦ Fo, moreover 1G ∗ 1F = 1G◦F, and 1F ◦ 1F = 1F. For any two categories D, F, any a, b, c ∈ Fct(D, F), any T ∈QO∈D MorF(a(O), b(O)) and S ∈QO∈D MorF(b(O), c(O)) we set S ◦ T ∈ QO∈D MorF(a(O), c(O)) such that (S◦T)(M) ≔ S(M)◦T(M) for all M ∈ D. If S is a nonempty subset of Obj(A) by abuse of language we indicate with S also the full subcategory of 26 CONTENTS A whose object set is S. Let A be a subcategory of B let IA→B ∈ Fct(A, B) be the identity functor defined in the obvious way, set IdA = IA→A. Let M, P be categories, F be a functor from M to P and N be a subcategory of P, thus we let F−1(N) be the possibly empty subcategory of M such that (0.2.20) Obj(F−1(N)) ≔ F−1 MorF−1(N) ≔ F−1 o (Obj(N)) m (MorN). Clearly F ◦ IF−1(N)→M is a functor from F−1(N) to P, which by abuse of language we also consider a functor from F−1(N) to N. We have the following equivalence of categories see [23, (1.3.5)] (0.2.21) Fct(K, D)op ≃ Fct(Kop, Dop); where the morphism map of the equivalence is the identity map while the object map is F 7→ opD ◦ F ◦ opKop, with opX is the contravariant functor from any category X to Xop whose object and morphism maps are the identity maps. By abuse of language in what follows we shall not mention in the above map the functors opD and opKop, so as a matter of fact considering the equivalence (0.2.21) as an identity. Finally let BS be the category of complex Banach spaces, linear bounded maps and map composition, and let us recall that CA∗ is the category of C∗−algebras, ∗−homomorphisms and map composition. 0.2.6. Multiplier algebra and the category of dynamical systems. Let Sd(H) be the set of the possibly unbounded selfadjoint operators in a Hilbert space H, while let U(H, K) be the set of the unitary operators from H to the Hilbert space K. Let A be a C∗−algebra, X be a Hilbert A−module, [32, Def. 2.8], LA(X) or simply L(X) the C∗−algebra of adjointable maps on X, [32, Def. 2.17 and Prp. 2.21], and let K(X) be the algebra of compact operators on X, [32, Def. 2.24]. Let AA denote the Hilbert A−module associated with A, [32, Exm. 2.10], let K(A) denote K(AA), while let M(A) ≔ L(AA) and iA denote the multiplier algebra of A and the canonical embedding of A into M(A) respectively, where iA(a)(b) = ab for all a, b ∈ A, [32, Def. 2.48 and Exm. 2.43]. Let B be a C∗−algebra and α : B → L(X) be a ∗−homomorphism, α is said nondegenerate if span{α(b)x b ∈ B, x ∈ X} is dense in X, [32, Def. 2.49]. iB is nondegenerate and injective since any C∗−algebra admits an approximate identity. As a consequence of [32, Prp. 2.50] we have that if α is nondegenerate then there exists a unique ∗−homomorphism α− : M(B) → L(X) such that (0.2.22) α− ◦ iB = α. α− is nondegenerate since it is so α, thus α−(1) = 1. Since the double conjugate Hilbert space of any Hilbert space H equals H, we deduce by [32, Exm. 2.27] that H is a K(H)−Hilbert module with the same norm, where H is the conjugate Hilbert space of H, and K(K) is the C∗−algebra of the compact operators on K, for any Hilbert space K. Therefore since (0.2.22) it follows that if R is a nondegenerate representation of B, then 0.2. TERMINOLOGY AND PRELIMINARIES 27 R− is the unique extension of R to a representation of M(B) such that (0.2.23) Let HHH = h H, R, Ω i be a cyclic representation of B, thus for all c ∈ M(B) and b ∈ B we deduce by [32, proof of Prp. 2.50] R− ◦ iB = R. R−(c) R(b)Ω = R((iB)−1(c iB(b))) Ω. (0.2.24) HHH− (cid:17) h H, R−, Ω i is a cyclic representation of M(B) since (0.2.23) and since HHH is cyclic. By the uniqueness of the extension to M(B) it follows that if h K, S, Ω i is a cyclic represen- tation of M(B), then h K, S ◦ iB, Ω i is a cyclic one of B such that (S ◦ iB)− = S. (0.2.25) Let A = h A, H, σ i be a C∗−dynamical system, µ ∈ H(H), and R is a nondegenerate ∗−representation of A ⋊µ σ H then there exists a nondegenerate covariant representation h H, π, W i of A such that R = π ⋊µ W. In particular said B = A ⋊µ (0.2.26) where jB A(a)( f )(l) = a f (l), for jB all a ∈ A, A is nondegenerate since Cc(H, A) provided by the sup −norm equals A ⊗Cc(H) and since iA is nondegenerate. We call h H, π, W i the covariant representation of A associated with R. A is the canonical embedding of A into M(B) such that jB f ∈ Cc(H, A) and l ∈ H. R− ◦ jB A = π, σ H 7→ T ◦ f defined on CCCµ The set of dynamical systems with symmetry group H, equivariant morphisms and map composition is a category denoted by C0(H), see [20, pg.26]. For any (σ, θ)−equivariant morphism T the map f c (H, A) extends uniquely by continuity to a ∗−homomorphism cµ(T) from A ⋊µ θ H, moreover cµ(S ◦ T) = cµ(S) ◦ cµ(T) for any dynamical system C and (B, C)−equivariant morphism S. Hence the functions h A, H, σ i 7→ A ⋊µ σ H and T 7→ cµ(T) determine a functor from C0(H) to the category of C∗−algebras and ∗−homomorphisms, see [20, pg.26] or [38, Cor. 2.48]. If H0 is a locally compact subgroup of H then any (σ, θ)−equivariant morphism is (σ ↾ H0, θ ↾ H0)−equivariant, hence the map h A, H, σ i 7→ h A, H0, σ ↾ H0 i and the identity map on MorC0(H) determine a functor from C0(H) to C0(H0). In particular for any ν ∈ H(H0) (0.2.27) cν(T) ∈ MorCA∗(A ⋊ν σ H to B ⋊µ σ H0, B ⋊ν θ H0). 0.2.7. K0−theory for C∗−algebras. In this section A denotes a C∗−algebra. The ∗−algebra M∞(A). Let n ∈ N0, set an ≔ a · ... · a and P(A) ≔ {p ∈ A p = p∗ = p2}. Denote by Mn(A) the set of the square matrices of order n at elements in A, which is a ∗−algebra providing it by the operations such that for all i, j ∈ {1, ..., n} and t ∈ C n−times z } { (1) (M + N)ij ≔ Mij + Nij, (2) (t · M)ij ≔ t · Mij, (3) (M · N)ij ≔Pn (4) (M∗)ij = M∗ ji. s=1 Mis · Ns j. 28 CONTENTS Let H the Hilbert space in which A acts faithfully, e.g. through the universal repre- sentation, see [22, Rmk. 4.5.8], so we can identify A as C∗−algebra with its faithful image acting on H provided with the standard operator norm. Next let Hk = H for any k = 1, . . . , n, set (o : Mn(A) → L(Ln o(M)(v)k ≔Pn Hs), s=1 j=1 Mkjv j, ∀M ∈ Mn(A), v ∈Ln s=1 Hs, k = 1, . . . , n, it is possible to show ([22, pg. 147]) that o is a ∗−isomorphism. Hence the following map define a norm k · kMn(A) : Mn(A) ∋ M 7→ ko(M)k, on Mn(A) for which it is a C∗−algebra since isometric to L(Ln Let A ⊙ Mn(C) be the algebraic tensor product of the ∗−algebras A and Mn(C), which is a ∗−algebra providing the product to be defined by (a ⊗ b) · (c ⊗ d) ≔ (a · c) ⊗ (b · d). For any i, j ∈ {1, . . . , n}, set eij ∈ Mn(C) such that (eij)rs ≔ δirδ js, for all r, s ∈ {1, . . . , n}. It is possible to show ([37, Prp. T.5.20]) that the following map Hs). s=1 A ⊙ Mn(C) → Mn(A), i, j=1 aij ⊗ eij 7→ A, Ars = ars, ∀r, s ∈ {1, ..., n}, Pn  is a well-defined ∗−isomorphism, called the standard isomorphism between A ⊙ Mn(C) and Mn(A). Moreover since [22, Ex. 11.1.5], the fact that o is an isometry of Mn(A) onto its image through o and that the universal representation of A is faithful, we can state that the previous map is an isometry between Mn(A), and the spatial tensor product (see [22, pg. 847]) A ⊗ Mn(C) of the C∗ algebras A and Mn(C). Set ΦA nm : Mn(A) → Mm(A) such that for any m ∈ N0, m ≥ n and A ∈ Mn(A) (ΦA ΦA nm(A)ij ≔ Aij, i, j ∈ {1, ..., n}, nm(A)ij ≔ 0, i, j ∈ {n + 1, ..., m}, clearly we have Φmk ◦ Φnm = Φnk, for all k, m, n ∈ N0 such that k ≥ m ≥ n, moreover Φnm is a ∗−isometry into its image. We call M∞(A) ≔ lim−−→(Ms(A), ΦA rs), the normed inductive limit of the system {(Mn, Φnm)}n,m∈N0,m≥n and it is the normed Mn(A) and λ : U → N0 such that ∗−algebra constructed as follows. Let U (cid:17) Sn∈N0 a ∈ Mλa(A), for all a ∈ U. Define ≃ on U such that for any a, b ∈ U λap(a) = ΦA a ≃ b ⇔ (∃p ∈ N0)(p ≥ λa, p ≥ λb, ΦA λbp(b)), denote by hai the equivalence set of a mod ≃. Next the following khaik∞ ≔ kakMλ(a)(A) 0.2. TERMINOLOGY AND PRELIMINARIES 29 is a well-dedifed norm on U/ ≃, indeed let a, b ∈ U such that a ≃ b, then there exists a p ∈ N0 such that ΦA λbp are isometries. Set by abuse of language λbp(b)), so kakMλ(a)(A) = kbkMλ(b)(A) since ΦA λap(a) = ΦA λap and ΦA (M∞(A) ≔Sn∈N0 Mn(A)/ ≃, f A n : Mn(A) → M∞(A), a 7→ hai, n ∈ N0. We obtain (0.2.28) (M∞(A) =Sn∈N0 fn(Mn(A)), fm ◦ Φnm = fn, m ≥ n. ≔ fn(0A n ) for some n ∈ N0, where 0A Set 0A n is the matrix of order n at elements in A all ∞ equal to 0, well-defined by the second equality in (0.2.28). Let a, b ∈ M∞(A), then since the second equality in (0.2.28) there exist p ∈ N0 and a, b ∈ Mp(A) such that x = fp(a) and y = fp(b), the following x + y ≔ fp(a + b), x · y ≔ fp(a · b), t · x ≔ fp(t · a), t ∈ C x∗ ≔ fp(a∗), are well-defined operations making h M∞(A), +, ·, 0A ∞, ·, ∗, k · k∞ i a (non unital) normed ∗−algebra, denoted again by M∞(A). k · k∞ will be called the standard norm in M∞(A). For any n ∈ N0, since A is identifiable with a two side ideal of A+, we deduce that Mn(A) and M∞(A) is identifiable with a two side ideal of Mn(A+) and M∞(A+) respectively, thus we can consider a ≡ b mod M∞(A), for any a, b ∈ M∞(A+). Moreover for any m ∈ N0, m ≥ n (0.2.29) ( f A n = f A+ nm = ΦA+ ΦA n ↾ Mn(A), nm ↾ Mn(A). The group K(A). Let n ∈ N0, define diagA n : Mn(A) × Mn(A) → M2n(A) such that for any a, b ∈ Mn(A) diagn(a, b) ≔ a 0n b.! 0n Define in P(A) the following equivalences ([4, Def. 4.2.1, Prp. 4.6.3]). Let p, q ∈ P(A) thus algebraic equivalence: p ∼ q iff there exists x, y ∈ A such that xy = p and yx = q, similarity: p ∼s q iff there exists an invertible element z in A such that zpz−1 = q, homotopy: p ∼h q iff there exists a continuous path of projections in A from p to q, i.e. a norm-continuous map f : [0, 1] → A such that f (t) ∈ P(A), for all t ∈ [0, 1] and f (0) = p and f (1) = q. 30 CONTENTS Denote by [e] the equivalence class of e mod ∼. For any e, f ∈ P(M∞(A)) it follows ([4, Ch. II, Sez. 4]) (0.2.30) Set (0.2.31) e ∼ f ⇔ e ∼s f ⇔ e ∼h f. V(A) ≔ h P(M∞(A))/ ∼, + i, [e] + [ f ] ≔ [e′ + f ′], ∀e, f ∈ P(M∞(A)), e′ ∈ [e], f ′ ∈ [ f ], e′ ⊥ f ′,  ([4, Def. 5.1.2 and comments following it]), where a ⊥ b iff a · b = 0 for any a, b ∈ P(A). The operation + is indipendent by the choice of e′ and f ′ which there exist as showed below. Below we convein to denote 0A+ n by diagn, for any n, p ∈ N0, p ≥ n. Let e, f ∈ P(M∞(A)) then there exist m ∈ N0, a, b ∈ Mm(A) such that e = fm(a) and f = fm(b). Note that Φm(2m)(c) = diagm(c, 0m), c ∈ {a, b}, thus e = f2m(diagm(a, 0m)) and f = f2m(diagm(b, 0m)), since (0.2.29-0.2.28). Define g ∈ M2m(A+) as np by Φnp and diagA+ n by fn, ΦA+ ∞ by 0∞ f A+ 1m 0m! , g ≔ 0m 1m f ′ ≔ f2m(cid:16)diagm(0m, b)(cid:17). z ≔ f2m(g),  Thus z = z∗ = z−1, since g = g∗ = g−1, and z f z−1 = z f z = f2m(cid:16)g diagm(b, 0m) g(cid:17) = f ′, moreover f ′ ∈ P(M2m(A+)) thus f ′ ∈ [ f ] since (0.2.30). Finally f ′ · e = f2m(cid:16)diagm(0m, b) · diagm(a, 0m)(cid:17) = 0∞ and the third sentence of (0.2.31) follows. Define the relation ≈ on V(A) × V(A) such that for all p, q, r, s ∈ P(M∞(A)) ([p], [q]) ≈ ([r], [s]) ⇔ (∃z ∈ P(M∞(A)))([p] + [s] + [z] = [q] + [r] + [z]), ≈ is an equivalence relation, let us denote by [([p], [q])] the equivalence class of ([p], [q]) mod ≈. So we can define for any ([p], [q]), ([a], [b]) ∈ V(A) × V(A) and for some ([r], [r]) ∈ V(A) × V(A) K00(A) ≔ (V(A) × V(A))/ ≈, [([p], [q])] + [([a], [b])] ≔ [([p] + [a], [q] + [b])], 0 ≔ [([r], [r])], K00(A) ≔ h K00(A), +, 0 i.  It is possible to show that K00(A) is a well-defined commutative group where if we denote by −[([p], [q])] the inverse of [([p], [q])], we have −[([p], [q])] = [([q], [p])] ([37, Appendix (0.2.32) that r ≡ s mod M∞(A) Here for any a ∈Sn∈N0  while for any p, q ∈ P(M∞(A)) (0.2.33) α# ∈ MorSg(V(A), V(B)), α∗ ∈ MorAb(K00(A), K00(B)), α⋆ ∈ MorAb(K0(A), K0(B)),  α#([hai]) = [hα ◦ ai], α⋆(cid:16)[([r], [s])](cid:17) =h(cid:16)(α+)#([r]), (α+)#([s])(cid:17)i, α∗(cid:16)[([p], [q])](cid:17) =h(cid:16)α#([p]), α#([q])(cid:17)i, 0.2. TERMINOLOGY AND PRELIMINARIES 31 G]). Finally set K0(A) ≔n[([p], [q])] ∈ K00(A+) p ≡ q mod M∞(A)o, K0(A) ≔ h K0(A), + ↾ (K0(A) × K0(A)), 0 i, then K0(A) is a subgroup of K00(A+) ([4, Ch. III, Sec. 5.5]). V is the object part of a functor from the category of ∗−algebras and ∗−homomorphisms to the category Sg of commutative semigroups and semigroup morphisms, while K00 and K0 are the object part of functors from the category of ∗−algebras and ∗−homomorphisms to the category Ab. More exactly for any couple of ∗−algebras A and B and ∗−homomorphism α : A → B, we have that and (β ◦ α)• = β• ◦ α•, for any ∗−algebra C, ∗−homomorphism β : B → C and • ∈ {#, ∗, ⋆}. Mn(A) such that hai ∈ P(M∞(A)), and r, s ∈ P(M∞(A+)) such ([37, Prp. 6.1.3 and Prp. 6.2.4]). If A is unital then K00(A) is isomorphic as a group to K0(A), so we shall use the convention to identify K0(A) with K00(A) whenever A is unital, called here as in [4] the standard picture of K0(A) ([4, Prp. 5.5.5]). Let H be a locally compact group, µ ∈ H(H), A = h A, H, σ i and B = h B, H, θ i be two objects of C0(H) and T be a (A, B)−equivariant morphism. Set kµ(T) ≔ (cµ(T)+)∗, η H)+) → K0((B ⋊µ thus kµ(T) : K0((A ⋊µ θ H)+) is a group morphism, in addition the functions h A, H, η i 7→ K0((A ⋊µ η H)+) and T 7→ kµ(T) define a functor from C0(H) to Ab. In particular for any locally compact subgroup H0 of H and ν ∈ H(H0) since the discussion prior (0.2.27) we have (0.2.34) kν(T) : K0((A ⋊ν η H0)+) → K0((B ⋊ν θ H0)+). 0.2.8. The Chern-Connes character. Let A and B be unital C∗−algebras. h ·, · iA : K0(A) × Hev ε (A) → C denote the pairing as defined in [16, IV.7.δ, Thm. 21]. If T : A → B is a unit preserving ∗−homomorphism, φ = {φ2n}n∈N is an entire even normalized cocycle on B and e an idempotent in M∞(A), then since [16, IV.7.δ, Thm. 21, Lemma 20], we obtain (0.2.35) h (T)∗([e]), [φ] iB = h [e], T†([φ]) iA. 32 CONTENTS ε (B) → Hev Here T† : Hev ε (A) denotes the map [{φ2n}n∈N] 7→ [{φ2n ◦ T[2n]}n∈N], [φ] is the class in Hev ε (B) corresponding to φ, and T[N] : AN → BN such that Pri ◦T[N] = T ◦ Pri for all i ∈ {1, . . . N}. Let ch(A, π, D, Γ ) ∈ Hev ε (A), also denoted by chh R, D, Γ i where R = h A, π i, denote the Chern-Connes character, [16, IV.8.δ, Def. 17], of the even θ−summable K−cycle (A, π, D, Γ ), [16, IV.2.γ, Def. 11 and IV.8, Def. 1]. It is well-known that the class corresponding to the JLO cocycle [21], associated with any even θ−summable K−cycle (A, π, D, Γ ) via [16, IV.8.ǫ, Thm. 21], belongs to ch(A, π, D, Γ ), [16, IV.8.ǫ, Thm. 22]. Hence one can use the JLO cocycle for computing the pairing h x, ch(A, π, D, Γ ) iA. An important result concerning the Connes character is that h x, ch(A, π, D, Γ ) iA ∈ Z for all x ∈ K0(A), [16, IV.8.δ, Thm. 19 and Prp. 18]. In this work whenever we refer to an even θ−summable K−cycle (A, π, D, Γ ), we assume that D , 0 which is automatically true in case dim H = ∞, with H the Hilbert space where π acts. Part 1 Preliminaries in Equivariance In the present part we fix more specific terminology and collect properties of equiv- ariance we employ in the paper. In our knowledge all the results in this part are new, however they are relatively not difficult to obtain by applying general well-known facts. Because of this we use again the term Preliminaries to title this part. In section 1.1.1 we establish that the Borelian functional calculus in a Banach space is equivariant under an isometric action. Then we apply this result to ensure the equivari- ance in case the spectral measure involved is the joint resolution of identity constructed by a family of commuting Borel resolutions of the identity in a Hilbert space. Then we provide a sufficient condition to ensure selfadjointness. In section 1.1.2 we dis- play the equivariance of the KMS−states under the conjugate of surjective equivariant morphisms. Section 1.1.3 is devoted to state the equivariance under action of H of rep- resentations of C∗−crossed products for different symmetry groups. In section 1.2 we prove for any state of a C∗−algebra the existence of a canonical extention to the multiplier algebra, then for any unital dynamical system, with A its underlying algebra, the result is used to associate a state of A with any state of the crossed product A ⋊σ H. 1.1. Equivariance 1.1.1. Equivariance of the functional calculus in Banach spaces. We prove in Thm. 1.1.11 that the Borelian functional calculus in a Banach space is equivariant under an isometric action. We apply this result to ensure in Thm. 1.1.14 the equivariance when the spectral measure E involved is the joint RI constructed by a family of commuting Borel RI's in a Hilbert space. In Thm. 1.1.8 we provide a sufficient condition concerning the Borelian map f to ensure the selfadjointness of the operator f (E). In this section we assume fixed a topological space S, a Banach space G and a Hilbert space H. Lemma 1.1.1. Let E be a countably spectral measure in G on a σ−field R of subsets of a set Z, and σ, δ ∈ R such that δ ⊇ σ and E(σ) = 1. Then E(δ) = 1. Proof. Z = σ∪∁σ implies 1 = E(σ)+E(∁σ), so E(∁σ) = 0. Moreover δ = δ∩(σ∪∁σ) = (cid:3) σ ∪ (δ ∩ ∁σ) hence E(δ) = E(σ) + E(δ)E(∁σ) = 1. Proposition 1.1.2. Let E be a countably additive spectral measure in G on a σ−field R of subsets of S such that R is generated as σ−field by a basis of the topology on S containing S. Thus supp(E) = \{σ∈R E(σ)=1} σ. Proof. The inclusion ⊇ follows since (0.2.12) and since supp(E) is closed. Let σ ∈ R such that E(σ) = 1, then E(σ) = 1 since Lemma 1.1.1 and σ ∈ R, thus {δ ∈ R E(δ) = (cid:3) 1, δ ∈ Cl(S)} ⊇ {σ σ ∈ R, E(σ) = 1} and the inclusion ⊆ follows. Proposition 1.1.3. Let Z be a set, R be a σ−field of subsets of Z, E a countably additive spectral measure on R and f : Z → S be a (R, B(S))−measurable map. Thus (1) supp(E ◦ f −1) ⊆T{σ∈RE(σ)=1} f (σ); 1.1. EQUIVARIANCE 35 (2) if Z is a topological space and R is generated as σ−field by a basis of the topology on Z containing Z, then supp(E ◦ f −1) ⊆ f (supp(E)); (3) if in addition to the conditions in st. (2) f is continuous, then supp(E ◦ f −1) = f (supp(E)). Proof. By Prp. 1.1.2 we have (1.1.1) supp(E ◦ f −1) = \{δ∈B(S)E( f −1(δ))=1} δ. Let σ ∈ R thus f (σ) ∈ B(S) and f −1( f (σ)) ⊇ σ, so E(σ) = 1 implies E( f −1( f (σ)) = 1 since Lemma 1.1.1, and st.(1) follows by (1.1.1). St.(2) follows by st.(1) and (0.2.12). Let f be continuous, set B f (cid:17) {δ ∈ B(S) E( f −1(δ)) = 1}, R f (cid:17) f −1(B f ), and Aσ (cid:17) f (σ) and Bδ (cid:17) δ, for all σ ∈ R f and δ ∈ B f . Thus A f −1(δ) ⊆ Bδ, since f ( f −1(δ)) ⊆ δ, for all δ ∈ B f , therefore T{δ∈B f } A f −1(δ) ⊆T{δ∈B f } Bδ, i.e. (1.1.2) \{σ∈R f } f (σ) ⊆ \{δ∈B f } δ. Moreover (1.1.3) σ(cid:17) f (σ) ⊆ f (supp(E)) = f(cid:16) \{σ∈RE(σ)=1,σ∈Cl(Z)} \{σ∈RE(σ)=1,σ∈Cl(Z)} ⊆ \{σ∈RE(σ)=1} ⊆ \{σ∈R f } ⊆ \{σ∈R f } f (σ). f (σ) f (σ) Here the second inclusion follows by Lemma 1.1.1, the forth by the continuity of the map f . Finally (1.1.3), (1.1.2) and (1.1.1) imply f (supp(E)) ⊆ supp(E ◦ f −1) and st.(3) follows (cid:3) by st.(2). Corollary 1.1.4. Let R (cid:17) {Rx}x∈X be a family of complete separable metric spaces, : RX → S be E (cid:17) {Ex}x∈X a family of commuting Borel RI's in H on R, and f (Cσ(RX), B(S))−measurable. Then for any Q ∈ Pω(X) supp(E f E) ⊆ f (C(Q,Yx∈Q supp(Ex))). Proof. By Prp. 1.1.3(1) and (0.2.17). (cid:3) 36 Remark 1.1.5. Let B be a σ−field of subsets of a set Z and E a RI in H on B. By following the argument in the proof of [18, Thm. 12.2.6.(d)] with the help of [19, Thm. 18.2.11.(i)], we obtain h(E)∗ = h∗(E), for all B−measurable map h, where h∗ is the complex conjugate map of h. Lemma 1.1.6. Let B be a σ−field of subsets of a set Z, E a countably additive spectral measure in G on B and σ ∈ B such that E(σ) = 1. Let f and g be two B−measurable maps, thus f ↾ σ = g ↾ σ implies f (E) = g(E). Proof. Let δ ∈ B, then ( f χδ)(E) = f (E)E(δ) since [19, Thm. 18.2.11( f )] and the fact that (cid:3) χδ(E) = E(δ). Moreover f ↾ δ = g ↾ δ implies f χδ = gχδ and the statement follows. The following result yields sufficient conditions to ensure the selfadjointness of the operator f (EE). Remark 1.1.7. Let R (cid:17) {Rx}x∈X be a family of complete separable metric spaces, : RX → C be E is the reso- E (cid:17) {Ex}x∈X a family of commuting Borel RI's in H on R, and f Cσ(RX)−measurable, then since (0.2.16) we obtain that E f lution of the identity of f (EE). E = E f (EE) i.e. E f Theorem 1.1.8 (Selfadjointness). Let R (cid:17) {Rx}x∈X be a family of complete separable met- : RX → C be ric spaces, E (cid:17) {Ex}x∈X a family of commuting Borel RI's in H on R, and f Cσ(RX)−measurable. Then (1) for all Q ∈ Pω(X) we have sp( f (EE)) ⊆ f (C(Q,Yx∈Q supp(Ex))), (2) if there exists an A ∈ Pω(X) such that f (C(A,Qx∈A supp(Ex))) ⊆ R then f (EE) is selfadjoint. Proof. Since Rmk. 1.1.7 and (0.2.15) (1.1.4) sp( f (EE)) = supp(E f E), then st.(1) follows since Cor. 1.1.4. Since Rmk. 1.1.7 we have (1.1.5) f (EE) = ı(E f E), moreover since (1.1.4) and (0.2.12) (1.1.6) Let h be the 0−extension to C of ı ↾ sp( f (EE)), then since (1.1.6), Lemma 1.1.6 and (1.1.5) we obtain E(sp( f (EE))) = 1. E f (1.1.7) If there exists an A ∈ Pω(X) such that f (C(A,Qx∈A supp(Ex))) ⊆ R then h = h∗ since st.(1), thus st.(2) follows since Rmk. 1.1.5 and (1.1.7). (cid:3) f (EE) = h(E f E). 1.1. EQUIVARIANCE 37 Proposition 1.1.9. Let Ur : R+ → L(H) be a strongly continuous semigroup of unitary operators on H, for r ∈ {1, 2}, such that [U1(t), U2(s)] = 0, for all s, t ∈ R+. Denote by Ar the infinitesimal generator of Ur, then [E−iA1 (σ), E−iA2 (δ)] = 0, for all σ, δ ∈ B(C). Proof. In this proof set Er (cid:17) E−iAr, for r ∈ {1, 2}. Since the Stone Thm., see for example [18, Thm. 12.6.1.] and its proof, we have Ur(t) = ft(Br), where Br (cid:17) −iAr and ft : C ∋ λ 7→ exp(itλ), for all t ∈ R+. Hence, since (0.2.16), Ur(1) is a scalar type spectral operator whose r.o.i. is EUr(1) = Er ◦ f −1 t ln λ, for any t ∈ R+ − {0}, then gt ◦ ft = Id so 1 . Let gt : C − {0} ∋ λ 7→ −i (1.1.8) Moreover since [19, Lemma 18.2.13 and Cor. 18.2.4] we deduce for all δ ∈ B(C) that [U1(1), EU2(1)(δ)] = 0, so by applying again [19, Lemma 18.2.13 and Cor. 18.2.4] we obtain (cid:3) [EU1(1)(σ), EU2(1)(δ)] = 0, for all σ, δ ∈ B(C) and the statement follows by (1.1.8). Er = EUr(1) ◦ g−1 1 . Proposition 1.1.10. Let E be a countably additive spectral measure in G on a σ−field of subsets of a set Z, F a Banach space and α ∈ L(Lw(G), Lw(F)) morphism of unital algebras. Then (1) α ◦ E is a countably additive spectral measure in F on B; (2) if Z is a topological space and B is generated as a σ−field by a basis for the topology of Z containing Z, then supp(α ◦ E) ⊆ supp(E), in addition supp(α ◦ E) = supp(E) if α−1({1}) = {1}. Proof. Trivial. (cid:3) Now we can state the first main result of this part namely the equivariance of the general functional calculus, result that shall be applied to ensure in Thm. 1.1.14 the covariance of the f.c. for of a commuting set of resolutions of the identity. Theorem 1.1.11 (Equivariance of the functional calculus). Let E be a countably additive spectral measure in G on a σ−field B of subsets of a set Z, F a Banach space and U : G → F be a linear isometry. Then (1) ad(U) ◦ E is a countably additive spectral measure in F on B; (2) if Z is a topological space and B is generated as a σ−field by a basis for the topology of Z containing Z, then supp(E) = supp(ad(U) ◦ E); (3) for any B−measurable map f we have Proof. St. (1) & (2) follow since Prp. 1.1.10(1) & (2) and since ad(U) ∈ L(Lw(G), Lw(F)) and it is a morphism of unital algebras. Let f be B−measurable and set EU (cid:17) ad(U) ◦ E. Thus (Dom( f (EU)) = {y ∈ G ∃ limn∈N IEU( fn)y}, f (EU)y = limn∈N IEU( fn)y, y ∈ Dom( f (EU)). (1.1.9) U f (E)U−1 = f (ad(U) ◦ E), sp(U f (E)U−1) = sp( f (E)), E f (ad(U)◦E) = ad(U) ◦ E ◦ f −1.  38 Moreover IEU = ad(U) ◦ I, since (0.2.11), thus since ad(U)−1 = ad(U−1) (1.1.10) in particular Dom( f (EU)U) = Dom( f (E)), and for any y ∈ Dom( f (E)) Dom( f (EU)) = UDom( f (E)), f (EU)Uy = lim n∈N U IE( fn)y = U lim n∈N IE( fn)y = U f (E)y, so f (EU)U = U f (E) and the first equality in (1.1.9) follows. Let T be a scalar type spectral operator in G, since Prp. 1.1.10(2) we have supp(EU T ) = supp(ET). Moreover by the first equality in (1.1.9) i.e. UTU−1 is a scalar type spectral operator in F such that UTU−1 = ı(EU T ), (1.1.11) hence EU T = EUTU−1, supp(EUTU−1) = supp(ET), therefore by (0.2.15) we obtain (1.1.12) sp(UTU−1) = sp(T). Hence the second equality in (1.1.9) follows with the position T = f (E), well-set since f (E) is a scalar type spectral operator by [19, Lemma 18.2.17]. With this position by (1.1.11) and the first equlity in (1.1.9) follows EU = E f (ad(U)◦E), then the third equality in (cid:3) (1.1.9) follows by (0.2.16). f (E) Corollary 1.1.12. Let F be a Banach space U : G → F a linear isometry, T a scalar type spectral operator in G and f a Borelian map. Thus (1) UTU−1 is a scalar type spectral operator in F such that EUTU−1 = ad(U) ◦ ET; (2) sp(UTU−1) = sp(T); (3) f (UTU−1) = U f (T)U−1; (4) E f (UTU−1) = ad(U) ◦ ET ◦ f −1. Proof. St.(1) & (2) follows since (1.1.11) & (1.1.12), st.(3) & (4) follow by st.(1) and since (cid:3) the first and third equality in (1.1.9) applied for the position E = ET. Definition 1.1.13. Let X be a set and T (cid:17) {Tx}x∈X such that Tx is a scalar type spectral operator in G for all x ∈ X, set ET ≔ {ETx}x∈X. Now we can state the main result of this section Theorem 1.1.14 (Equivariance of the f.c. associated with ET). Let X be a set, K a Hilbert space, U : H → K a unitary operator, and f a Cσ(CX)−measurable map. Moreover let T (cid:17) {Tx}x∈X satisfy the following two properties: Tx is a scalar type spectral operator in H for all x ∈ X, and ET is a family of commuting Borel RI's in H. Set T(U) (cid:17) {UTxU−1}x∈X, then 1.1. EQUIVARIANCE 39 (1) EET(U) = ad(U) ◦ EET; (2) f (EET(U)) = U f (EET)U−1; (3) E f (EET(U) Proof. St.(1) follows since Cor. 1.1.12(1) applied to any Tx, x ∈ X and by the unique- ness in [3, Thm. 1.3 pg. 122]. St.(2) follows since st.(1) and the first equality in (1.1.9), (cid:3) while st.(3) follows since the third equality in (1.1.9) and st.(1). ) = ad(U) ◦ EET ◦ f −1. 1.1.2. Equivariance of KMS-states. In Thm. 1.1.19 and Cor. 1.1.21 we prove the equivariance of the KMS−states under the conjugate of the action of surjective equivari- ant morphisms. In what follows with approximate identity of a Banach ∗−algebra we mean a bounded order preserving approximate identity of positive elements bounded by 1. Lemma 1.1.15. Let A be a Banach ∗−algebra, B be C∗−algebra and T be a ∗−homomorphism from A to B such that T(A) is norm dense. If {eα}α∈D is an approximate identity of A then {T(eα)}α∈D is an approximate identity of B. Proof. Let {eα}α∈D be an approximate identity of A. Then kT(eα)k ≤ 1 and α ≤ β ⇒ T(eα) ≤ T(eβ) for all α, β ∈ D since the positivity and the continuity of T with kTk ≤ 1, see [14, Prp. 2.3.1]. Next kT(eα)T(a) − T(a)k = kT(eαa − a)k ≤ keαa − ak, for all a ∈ A and α ∈ D, so for all b ∈ T(A) (1.1.13) lim α∈D kT(eα)b − bk = 0. Let ε > 0 and B ∈ B thus there exists b ∈ T(A) such that kb − Bk ≤ ε 2, and for all α ∈ D kT(eα)B − Bk ≤ kT(eα)(B − b)k + kT(eα)b − bk + kb − Bk ≤ 2kb − Bk + kT(eα)b − bk. Hence lim supα∈D kT(eα)B − Bk ≤ ε and lim infα∈D kT(eα)B − Bk ≤ ε for all ε > 0 since [9, IV.24(14), IV.27(33), IV.23(13)] and (1.1.13), then limα∈D kT(eα)B − Bk = 0 since [9, IV.23, (cid:3) Cor.1] and the statement follows. Remark 1.1.16. Let us assume the hypotheses of Lemma 1.1.15. If A is a C∗−algebra T(A) is norm closed [30, Thm. 9.5.12.(d)], hence in this case the norm density of T(A) is equivalent to the surjectivity of T. While if A and B are unital algebras then T is unit preserving since its continuity. Lemma 1.1.17. Let A, B be C∗−algebras and T be a surjective ∗−homomorphism from A to B, then T†(EB) ⊆ EA. Proof. Since T†(ω) is positive for all ω ∈ EB the statement follows by Lemma 1.1.15 (cid:3) and [14, Prp. 2.3.11]. Lemma 1.1.18. Let h A, H, η i and h B, H, θ i be dynamical systems and T be a (η, θ)−equivariant morphism. Then 40 (1) T(Aη) ⊆ Bθ and the for all z ∈ C the following diagram is commutative (1.1.14) Bθ T Aη θ(z) / B T / A η(z) (2) T ◦ δη ⊆ δθ ◦ T. Proof. Let a ∈ Aη and ηa denote the map C ∋ z 7→ η(z)(a) ∈ A, thus θ(t)(Ta) = Tη(t)(a) = (T ◦ ηa)(t) for all t ∈ R. Next T ◦ ηa is analytic being composition of two analytic maps, then T(a) ∈ Bθ and by the uniqueness of the analytic extension, the commutativity of the diagram follows. Let b ∈ Dom(δη), so since the continuity of T T(δηb) = lim t→0,t,0 Tη(t)b − Tb t = lim t→0,t,0 θ(t)Tb − Tb t = δθ(Tb). (cid:3) The next result states the equivariance of the KMS−states under the conjugate of the action of surjective equivariant maps. Theorem 1.1.19. Let h A, H, η i and h B, H, θ i be dynamical systems and T an (η, θ)−equivariant morphism such that T is surjective or unit preserving in case A and B are unital. Then T†(Kθ β) ⊆ Kη β for all β ∈eR. Proof. If β ∈ R, ω ∈ Kθ β and x, y ∈ Aη then ω(T(x)θ(iβ)T(y)) = ω(T(y)T(x)) since Lemma 1.1.18(1), thus the statement follows by (1.1.14) and Lemma 1.1.17. If β = ∞ the statement follows by the equivariance of T, by [15, Prp. 5.3.19(3)] and Lemma 1.1.17. (cid:3) An alternative proof of Thm. 1.1.19 for β ∈ R follows by Lemma 1.1.17, by the equivariance of T and by [15, Prp. 5.3.7.(2)]. Lemma 1.1.20. Let h A, R, η i be a dynamical system and λ ∈ AutCA∗(A). Then (1) λ(Aη) = Aad(λ)◦η; (2) ad(λ) ◦ η = ad(λ) ◦ η; (3) ad(λ)(δη) = δad(λ)◦η. Proof. In this proof let ηλ denote ad(λ) ◦ η, which is clearly a strongly continuous one-parameter group acting on A by ∗−automorphisms. λ is C−differentiable since it is C−linear and norm continuous, therefore by the chain rule of differentiable maps and by [14, Prp. 2.5.21.] we deduce that (z 7→ λ ◦ η(z)(a)) ∈ Hs(C, A), for all a ∈ Aη. Hence for all c ∈ λ(Aη) (1.1.15) (z 7→ ad(λ) ◦ η(z)(c)) ∈ Hst(C, A), ad(λ) ◦ η ↾ R = ηλ, / O O / O O 1.1. EQUIVARIANCE 41 and the inclusion ⊂ of st.(1) follows. If we apply this result to the system h A, R, ad(λ)◦η i and to the ∗−automorphism λ−1, we deduce the remaining inclusion and st.(1) follows. St.(2) follows by (1.1.15), st.(1) and the uniqueness of the entire analytic extension of ηλ. Let b ∈ λ(Dom(δη)) thus by (0.2.4), (0.2.9) and the norm continuity of λ we deduce that thus (1.1.16) ad(λ)(δη)(b) = lim t↾0, t,0 ηλ(t)(b) − b t , ad(λ)(δη) ⊂ δηλ. By applying (1.1.16) we obtain ad(λ−1)(δηλ) ⊂ δη, which implies δηλ ⊂ ad(λ)(δη), and the statement follows. (cid:3) If h A, R, η i is a dynamical system and λ ∈ AutCA∗(A) then λ−1 is (ad(λ) ◦ η, η)−equivariant, hence by Thm. 1.1.19 we obtain the following result, however we prefer to show it as a consequence of Lemma 1.1.20. Cor. 1.1.21 will be used via Cor. 1.1.28 to prove Thm. 2.3.17 a step toward the construction of the object part of the functor from Cu(H) and G(G, F, ρ). β . Corollary 1.1.21. Let h A, R, η i be a dynamical system, λ ∈ AutCA∗(A) an β ∈eR, then λ∗(cid:16)Kη β(cid:17) = Kad(λ)◦η Proof. In this proof we denote ad(λ) ◦ η by ηλ. Let β ∈ R, ω ∈ Kη β and Dη be a norm dense, η−invariant, ∗−subalgebra of Aη such that ω(a η(iβ)(b)) = ω(b a), for all a, b ∈ Dη. Hence for all c, d ∈ λ(Dη) thus (1.1.17) ω(cid:16)λ−1(c) η(iβ)(λ−1(d))(cid:17) = λ∗(ω)(d c), λ∗(ω)(cid:16)c (ad(λ) ◦ η)(iβ)(d)(cid:17) = λ∗(ω)(d c). Moreover λ(Dη) is a norm dense, since λ is surjective and norm continuous, ηλ−invariant, ∗−subalgebra of λ(Aη), hence the inclusion ⊂ of the statement follows since (1.1.17) and Lemma 1.1.20(1) & (2). The remaining inclusion follows by the previous one applied to the system h A, R, ad(λ) ◦ η i and to the ∗−automorphism λ−1. Let ψ ∈ Kη ∞ and b ∈ λ(Dom(δη)). By definition iψ(aδη(a)) ≥ 0, for all a ∈ Dom(δη), then iλ∗(ψ)(cid:16)b ad(λ)(δη)(b)(cid:17) = iψ(cid:16)λ−1(b)δη(λ−1b)(cid:17) ≥ 0, hence the inclusion ⊂ of the statement follows since Lemma 1.1.20(3). By applying this inclusion to to the system h A, R, ad(λ) ◦ η i and to λ−1 we obtain (λ−1)∗(cid:16)Kad(λ)◦η which is the remaining inclusion and the statement follows. (cid:17) ⊂ Kη β (cid:3) β 42 1.1.3. Equivariance of representations of C∗−crossed products. The main result of this section is Thm. 1.1.54 where we show the equivariance under action of H of representations of C∗−crossed products for different symmetry groups. In this section we assume fixed two locally compact topological groups G and F, a group homomorphism ρ : F → AutGr(G) such that the map (g, f ) 7→ ρ f (g) on G × F at values in G, is continuous, moreover let H denote G ⋊ρ F. Definition 1.1.22. Let A = h A, H, σ i be a dynamical system and ω ∈ A∗ set Fω(A) ≔ {h ∈ F ω ◦ γσ(h) = ω}, and Fω(A) ≔ {F0 closed subgroup of F ω ∈ EF0 A (γσ ↾ F0)}. We convein to remove (A) from Fω(A) and Fω(A) whenever it is clear which dynamical system is involved. Lemma 1.1.23. Let h A, H, σ i be a dynamical system and ω ∈ A∗, thus Fω = max Fω, in particular Fω =SF0∈Fω F0. Proof. By construction Fω ⊇ F0 for any F0 ∈ Fω, thus it is sufficient to show that Fω ∈ Fω. Let h ∈ F such that there exists a net {hα}α∈D in F for which h = limα∈D hα and ω ◦ σ(j2(hα)) = ω for all α ∈ D. Since the σ(A, A∗)−continuity of σ and the continuity of j2, we have for all A ∈ A ω(γ(h)A) = ω(σ(j2(h))A) = lim α∈D ω(σ(j2(hα))A) = ω(A), so Fω is closed. (cid:3) Remark 1.1.24. Since Lemma 1.1.23 the group SG Fω dynamical system h A, H, σ i it is so also h A, SG Fω consider the C∗−crossed product A ⋊µ σ SG Fω for any µ ∈ H(SG Fω ). is locally compact, hence for any , σ i. In particular it makes sense to Lemma 1.1.25. Let h A, H, η i and h B, H, θ i be dynamical systems, ω ∈ B∗ and T a (η, θ)−equivariant morphism such that T(A) is σ(B, B∗)−dense. Then Fω = FT†(ω) and SG Fω = SG . FT† (ω) Proof. FT†(ω) = {h ∈ F ω ◦ T ◦ η(j2(h)) = ω ◦ T} = {h ∈ F ω ◦ θ(j2(h)) ◦ T = ω ◦ T} ⊇ {h ∈ F ω ◦ θ(j2(h)) = ω} = Fω. The inclusion ⊆ follows since the density hypothesis and ω◦θ(j2(h)), ω ∈ B∗. The second (cid:3) equality follows by the first one. For the remaining of the present section we assume fixed a C∗−dynamical system A = h A, H, σ i. 1.1. EQUIVARIANCE 43 Lemma 1.1.26. If ω ∈ EG A(τ) and F0 ∈ Fω, then ω ∈ E SG F0 A (σF0). Proof. The statement follows since σ is a group action and since (g, f ) = (g, 1)·ρ (1, f ), (cid:3) for all (g, f ) ∈ G × F. Since Lemma 1.1.23 & 1.1.26 we can state the following Corollary 1.1.27. ω ∈ EG A(τ) ⇒ ω ∈ E Fω SG A (σFω). Corollary 1.1.28. Let l, h ∈ H, β ∈eR and ξ : R → G be a continuous group morphism. Then σ∗(l)(cid:16)Kτ(h,ξ) (cid:17) = Kτ(l·ρ h,ξ) . β β Proof. By Cor. 1.1.21 and (0.2.8). (cid:3) Lemma 1.1.29. Let ψ ∈ EG (1) Fσ∗(l)(ψ) = ad(Pr2(l))(Fψ); (2) SG = ad(l)(SG Fψ Fσ∗(l)(ψ) ). A(τ) and l ∈ H, then Proof. Since (g, h) = (1, h) ·ρ (ρ(h−1)g, 1) for all g ∈ G and h ∈ F, we have l = j2(Pr 2 (l)) ·ρ j1(cid:18)ρ(Pr 2 (l−1)) Pr 1 (l)(cid:19) , next ψ ∈ EG A(τ) by construction, thus (1.1.18) σ∗(l)(ψ) = γ∗(Pr 2 (l))(ψ). Hence ad(Pr2(l))(Fψ) ∈ Fσ∗(l)(ψ), so by Lemma 1.1.23 (1.1.19) ad(Pr 2 (l))(Fψ) ⊂ Fσ∗(l)(ψ). Now σ∗(l)(ψ) ∈ EG replace ψ by σ∗(l)(ψ) and l by l−1 and obtain A(τ), since j1(G) is a normal subgroup of H, hence (1.1.19) holds if we ad(Pr 2 (l−1))(Fσ∗(l)(ψ)) ⊂ Fψ, hence Fσ∗(l)(ψ) ⊂ ad(Pr2(l))(Fψ) and st.(1) follows by (1.1.19). st.(2) follows by (g, h) = j1(g) ·ρ j2(h) for all g ∈ G and h ∈ F, by st.(1) and since j1(G) is a normal subgroup of (cid:3) H. Proposition 1.1.30. Let V ∈ U(A), ψ ∈ EA and h H, π, Ω i be a cyclic representation of A associated with ψ. Then h H, π, π(V)Ω i is a cyclic representation of A associated with ad(V)∗(ψ). Proof. ψ ◦ ad(V−1) = ωπ(V)Ω ◦ π. Since AV ⊂ A and AV−1 ⊂ A we have A = AV−1V ⊂ (cid:3) AV ⊂ A, so AV = A. Thus π(V)Ω is cyclic for the set π(A). Definition 1.1.31. Let HHH ≔ h H, π, Ω i be a cyclic representation of A, v a group morphism of H into U(A), and l ∈ H. Set HHH(v,l) ≔ h H, π, π(v(l))Ω i. 44 Remark 1.1.32. Let ψ ∈ EA and H be a cyclic representation of A associated with ψ. Since Prp. 1.1.30, if A is inner implemented by v, then HHH(v,l) is a cyclic representation of A associated with σ∗(l)(ψ). Definition 1.1.33. Let A be a nonempty set, ωωω : A → EG A(τ) and HHH : A → Repc(A) be such that HHHα = h Hα, πα, Ωα i is a cyclic representation of A associated with ωωωα, for all α ∈ A. Thus we can set for all α ∈ A Uσ HHH,α σFωωωα ≔ W HHH . Whenever it is clear the dynamical system involved, we convein to remove the index σ, moreover we remove the index α anytime A is a singleton. Remark 1.1.34. Since Lemma 1.1.23 Fωωωα is closed in F for any α ∈ A, so h A, SG , σFωωωα i is a dynamical system, hence Def. 1.1.33 is well-set by Cor. 1.1.27. By construction HHH,α : SG Uσ → L(Hα) such that for all l ∈ SG and a ∈ A we have Fωωωα Fωωωα Fωωωα Uσ HHH,α(l) πα(a)Ωα = πα(σ(l)a)Ωα. Corollary 1.1.35. Let ϕ ∈ EG A(τ). Moreover let HHH = h H, π, Ω i be a cyclic representation of A associated with ϕ. If A is inner implemented by v, then HHH(v,l) is a cyclic representation of A associated with σ∗(l)(ϕ) and A(τ) and l ∈ H, thus σ∗(l)(ϕ) ∈ EG (1.1.20) UHHH(v,l) = (ad ◦ πϕ ◦ v)(l) ◦ UHHH ◦ ad(l−1) ↾ SG Fσ∗(l)(ϕ) . Proof. The first sentence of the statement follows since l−1 ·ρ j1(g) = ad(l−1)(j1(g))·ρ l−1, for all g ∈ G and since j1(G) is a normal subgroup of H. If A is inner, then the second sentence of the statement follows by Rmk. 1.1.32. UHHH(v,l) is well-set, since the first two sentences of the statement and Def. 1.1.33. Moreover since Lemma 1.1.29(2) (1.1.21) ad(l)(SG Fϕ ) = SG Fσ∗(l)(ϕ) . If HHH = h H, π, Ω i thus for all h ∈ SG Fϕ and a ∈ A we have UHHH(v,l)(cid:16)ad(l)h(cid:17) π(a) π(v(l)) Ω = π(cid:16)σ(ad(l)h)a(cid:17) π(v(l)) Ω = π(cid:16)v(ad(l)h) a v(ad(l)h−1)v(l)(cid:17) Ω = π(cid:16)v(l) v(h) v(l−1) a v(l) v(h−1)(cid:17) Ω = π(v(l)) π(cid:16)σ(h)(cid:16)ad(v(l−1))a(cid:17)(cid:17) Ω = π(v(l)) UHHH(h) π(cid:16)ad(v(l−1))a(cid:17) Ω = π(v(l)) UHHH(h) π(v(l−1)) π(a) π(v(l)) Ω. Moreover π(v(l−1)) = π(v(l))−1, hence (1.1.20) follows by (1.1.21) and the cyclicity of the (cid:3) representation HHH(v,l). 1.1. EQUIVARIANCE 45 Remark 1.1.36. Under the hypotheses of Cor. 1.1.35 we have for all s ∈ SG Fσ∗(l)(ϕ) UHHH(v,l)(s) = π(v(l)) UHHH(cid:16)ad(l−1)s(cid:17) π(v(l−1)). Remark 1.1.37. Let T be a locally compact space µ ∈ M(T), H a Hilbert space. Thus for any map f : T → Ls(H) scalarly essentially µ−integrable we haveR f dµ ∈ L(H), since [12, Ch. 6, §1, n◦4, Cor. 2]. Lemma 1.1.38. Let G1, G2 be two locally compact groups, η : G1 → G2 be an isomor- phism of topological groups and µ ∈ H(G1). Then η(µ) is well-set and η(µ) ∈ H(G2). Proof. η is µ−misurable since it is continuous. Let E be a Hausdorff locally convex space and g ∈ Cc(G2, E), thus η−1(supp(g)) = supp(g ◦ η) ∈ Comp(G1) (1.1.22) Indeed η−1(supp(g)) is compact η−1 being continuous, moreover supp(g◦η) ⊇ η−1(supp(g)). Let x ∈ supp(g ◦ η) thus there exists a net {xα}α∈D in G1 such that limα∈D xα = x and g(η(xα)) , 0, for all α ∈ D. But limα∈D η(xα) = η(x) so η(x) ∈ supp(g), i.e. x ∈ η−1(supp(g)) and (1.1.22) follows. Let f ∈ Cc(G2), then by (1.1.22) f ◦η ∈ Cc(G1) and η(µ) is well-defined. Let s ∈ G2 thus Ls ◦ η = η ◦ Lη−1(s) so L∗ s( f ) ◦ η = L∗ η−1(s)( f ◦ η), then Z L∗ s( f ) ◦ η dµ s( f ) dη(µ) =Z L∗ =Z L∗ =Z f ◦ η dµ =Z f dη(µ), η−1(s)( f ◦ η) dµ where the third equality follows by the left invariance of µ, while the first and forth ones (cid:3) follow by (0.2.1). Remark 1.1.39. Let E be a Hausdorff locally convex space, T, S be two locally compact spaces, µ ∈ M(T), and ε : T → S be µ−proper. Since [12, Ch. 6, §1, n◦1, pg. 4 and Ch. 5, §6, n◦2, Thm. 1] we have for any scalarly essentially ε(µ)−integrable map f : S → E that f ◦ ε is scalarly essentially µ−integrable and Moreover if E = Ls(H) for some Hilbert space H, then by Rmk. 1.1.37 Z f ◦ ε dµ =Z f dε(µ). Z f ◦ ε dµ ∈ L(H). Lemma 1.1.40. Let X be a locally compact group, µ ∈ H(X) and Y be a locally compact subgroup of X. Then µY ∈ H(Y) and ∆Y = ∆X ↾ Y. so Thus µY ∈ H(Y). Moreover s(ef ), µY(L∗ s( f )) = µ(]L∗ s( f )) s(ef ) and ]R∗ s(ef )) = µ(L∗ = µ(ef ) = µY( f ). s−1( f )(cid:17) = ∆X(s)µ(cid:16) ]R∗ s−1( f )(cid:17) = ∆X(s)µ(cid:16)R∗ s−1(ef )(cid:17) = µ(ef ) = µY( f ), ∆X(s)µY(cid:16)R∗ 46 Proof. Let e ∈ ∁Y and s ∈ Y then s−1·e, e·s−1 ∈ ∁Y, indeed if by absurdum s−1·e, e·s−1 ∈ Y, then e = s · (s−1 · e) = (e · s−1) · s ∈ Y. Let f ∈ Cc(Y) then s( f ) = R∗ ]L∗ s( f ) = L∗ then since µY ∈ H(Y) and the independence of the modular function by the Haar measure (cid:3) ([38, Lemma 1.61]), we deduce that ∆Y = ∆X ↾ Y. Lemma 1.1.29(2) and Lemma 1.1.38 allow to set the following Definition 1.1.41 (Haar systems). Let A be a nonempty set and ωωω : A → EG A(τ). We define the set of Haar systems associated with ωωω and A, denoted by H(ωωω, A), the subset of the µµµ ∈ Y(α,l)∈A×H H(SG Fσ∗ (l)(ωωωα) ), such that for all (α, l) ∈ A × H (1.1.23) µµµ(α,l) = ad(l)(µµµ(α,1)). Definition 1.1.42. Let ν ∈ H(H) and ωωω : A → EG A(τ), define for all (α, l) ∈ A × H ννν(α,l) ≔ ad(l)(νSG Fωωωα ). ννν will be called the Haar system generated by ν and ωωω. H(ωωω, A) is nonempty, indeed Proposition 1.1.43. Let ν ∈ H(H) and ωωω : A → EG A(τ), then the Haar system generated by ν and ωωω belongs to H(ωωω, A). Proof. Since Lemma 1.1.40, Lemma 1.1.29(2) and Lemma 1.1.38. (cid:3) Definition 1.1.44. Let A be a nonempty set, ωωω : A → EG A(τ) and µµµ ∈ H(ωωω, A). For any l ∈ H and α ∈ A set Bωωω,α,l µµµ (A) ≔ A ⋊ µµµ(α,l) σ SG Fσ∗(l)(ωωωα) Bωωω,α,l,+ µµµ (A) ≔ (Bωωω,α,l µµµ (A))+. 1.1. EQUIVARIANCE 47 We convein to remove l when it equals the identity and to remove α when A is a singleton. Moreover whenever it is clear which dynamical system is involved, we convein to remove A and to denote the map σ∗(l) ◦ ωωω by ωωωl for any l ∈ H. Let HHH : A → Repc(A) be such that HHHα = h Hα, πα, Ωα i is a cyclic representation of A associated with ωωωα, for all α ∈ A. Set for all α ∈ A (1) Rµµµ (2) RRRµµµ (3) RRR HHH,α(A) ≔ πα ⋊µµµ(α,1) Uσ HHHα µµµ (A), Rµµµ HHH,α(A) ≔ (Bωωω,α HHH,α(A)) (A), Rµµµ HHH,α(A) ≔ (Bωωω,α,+ HHH,α(A)). µµµ µµµ If A is inner implemented by v, we can define HHH(v,l) : I → Repc(A), such that β 7→ HHH(v,l) and for all α ∈ A α ≔ (HHHα)(v,l) ≔ Uσ HHH,v,α,l HHH(v,l) α (1) Uσ (2) Rµµµ (3) RRRµµµ (4) RRR HHH,v,α,l(A) ≔ πα ⋊µµµ(α,l) Uσ HHH,v,α,l(A) ≔ (Bωωω,α,l HHH,v,α,l(A) ≔ (Bωωω,α,l,+ (A), Rµµµ (A), Rµµµ HHH,v,α,l µµµ µµµ µµµ HHH,v,α,l(A)) HHH,v,α,l(A)). We convein to remove A and the index σ whenever it is clear the dynamical system involved, to remove the index l whenever it equals the identity, and the index v whenever it is uniquely determined by σ, and to remove α if it is a singleton. Remark 1.1.45. Let α ∈ A and l ∈ H, and ωωω and µµµ as in Def. 1.1.44, then Bωωω,α,l is is the C∗−algebra obtained by adding the unit µµµ well-set since Lemma 1.1.23, and Bωωω,α,l,+ to Bωωω,α,l . Moreover µµµ µµµ • Rµµµ • Rµµµ HHH,v,α,l extends uniquely to a ∗−representation of Bωωω,α,l HHH,v,α,l is the ∗−representation of Bωωω,α,l,+ (0.2.3). µµµ µµµ in Hα induced by Rµµµ in Hα; HHH,v,α,l according to Lemma 1.1.29(2) and (1.1.22) permit to set the following Definition 1.1.46. Define σσσ ∈ Y(ψ,l)∈EG A(τ)×H Morset(cid:16)Cc(SG Fψ , A), Cc(SG Fσ∗(l)(ψ) , A)(cid:17), such that for all (ψ, l) ∈ EG A(τ) × H. σσσ(ψ,l)( f ) ≔ σ(l) ◦ f ◦ ad(l−1) ↾ SG Fσ∗(l)(ψ) , Proposition 1.1.47. supp(σσσ(ψ,l)( f )) = ad(l)(supp( f )), for any f ∈ Cc(SG Fψ , A) and (ψ, l) ∈ EG A(τ) × H, in particular σσσ(ψ,l) is well-defined. Proof. Let (ψ, l) ∈ EG respectively for all φ ∈ EA. If f ∈ Cc(Hψ, A) then σσσ(ψ,l)( f ) is continuous since composition of continuous A(τ) × H, and φl and Hφ denote σ∗(l)(φ) and SG Fφ 48 maps, moreover σ(l) is linear thus supp(σσσ(ψ,l)( f )) ⊆ supp( f ◦ ad(l−1) ↾ Hψl). Clearly ad(l) f −1(0) = ( f ◦ ad(l−1) ↾ Hψl)−1(0), then since ad(l) is injective it follows ad(l)∁ f −1(0) = ∁( f ◦ ad(l−1) ↾ Hψl)−1(0), moreover ad(l)∁ f −1(0) = ad(l)supp( f ) since ad(l) is a homeomorphism, thus supp( f ◦ ad(l−1) ↾ Hψl) = ad(l)supp( f ). Hence supp(σσσ(ψ,l)( f )) ⊆ ad(l)supp( f ), which applied to the position ψl, l−1 and σσσ(ψ,l)( f ), possible according to the first sentence in Cor. 1.1.35, yields supp(cid:16)σσσ(ψl,l−1)(σσσ(ψ,l)( f ))(cid:17) ⊆ ad(l−1)supp(σσσ(ψ,l)( f )), i.e ad(l)supp( f ) ⊆ supp(σσσ(ψ,l)( f )), and the statement follows. (cid:3) Lemma 1.1.48. Let ωωω : A → EG A(τ) and µµµ be a Haar system associated with ωωω and A. Then for any (α, l) ∈ A × H and h ∈ H we have (1) σσσ(ωωωα,l) µµµ(α,l) C c (SG Fσ∗(l)(ωωωα) is a ∗−isomorphism of ∗−algebras µµµ(α,1) from C c (SG Fωωωα , A) onto , A) continuous w.r.t. the inductive limit topologies; (2) µµµh is a Haar system associated with ωωωh and A, and σσσ(ωωωl µµµ(α,1) (3) σσσ(ωωωα,l) is an isometry of CCC c µµµ(α,l) , A) onto CCC c (SG α,l−1) = (σσσ(ωωωα,l))−1; , A). (SG Fωωωα Fσ∗(l)(ωωωα) Here µµµh (β,u) ≔ µµµ(β,u·h), for all (β, u) ∈ A × H and recall that ωωωh ≔ σ∗(h) ◦ ωωω. Proof. Let (α, l) ∈ A × H and f, f1, f2 ∈ Cc(SG Fωωωα , A), then for all s ∈ SG Fσ∗(l)(ωωωα) σσσ(ωωωα,l)( f1 ∗µµµ(α,1) f2)(s) = σ(l)(cid:16)Z f1(r)σ(r)(cid:16) f2(r−1ad(l−1)(s))(cid:17) dµµµ(α,1)(r)(cid:17) =Z σ(l)( f1(r))σ(lr)(cid:16) f2(r−1ad(l−1)(s))(cid:17) dµµµ(α,1)(r) =Z σ(l)( f1(ad(l−1)r))σ(rl)(cid:16) f2(ad(l−1)(r−1s))(cid:17) dµµµ(α,l)(r) =Z σσσ(ωωωα,l)( f1)(r)σ(r)σσσ(ωωωα,l)( f2)(r−1s)dµµµ(α,l)(r) =(cid:16)σσσ(ωωωα,l)( f1) ∗µµµ(α,l) σσσ(ωωωα,l)( f2)(cid:17)(s). Here the second equality follows by the continuity of σ(l) in k · kA−topology and by the fact the integration is w.r.t. the same topology. The third equality follows by the fact that µµµ(α,1) = ad(l−1)(µµµ(α,l)) and by Rmk. 1.1.39. 1.1. EQUIVARIANCE 49 σσσ(ωωωα,l)( f )∗(s) = ∆(s−1)σ(s)(σσσ(ωωωα,l)( f )(s−1)∗) = ∆(s−1)σ(sl)( f (ad(l−1)s−1)∗) = σ(l)∆(ad(l−1)s−1)σ(ad(l−1)s)( f (ad(l−1)s−1)∗) = σσσ(ωωωα,l)( f ∗)(s). Here ∆ = ∆H. The first and last equality follow by ∆Y = ∆ ↾ Y for Y ∈ {SG }, since Lemma 1.1.40, while the third one follows since ∆ is a group morphism. Moreover since Lemma 1.1.29(2) it is easy to see that (σσσ(ωωωα,l))−1 = σσσ(σ∗(l)(ωωωα),l−1), and the first part of st.(1) follows. In the following we let Hα = SG . For any compact subset K of Hα let C(Hα; K, A) be the space of the f ∈ Cc(Hα, A) such that supp( f ) ⊆ K and ıK be the identity map embedding C(Hα; K, A) into Cc(Hα, A). If we prove that for any compact K of H the map σσσ(ωωωα,l) ◦ ıK is continuous w.r.t. the topology of uniform convergence on C(H; K, A) and the inductive limit topology on Cc(Hα,l, A), then the second part of st.(1) will follows since [11, II.27 Prp. 5(ii)]. Next since Prp. 1.1.47 and Hα,l = SG Fσ∗(l)(ωωωα) , SG Fωωωα Fσ∗(l)(ωωωα) Fωωωα supp(σσσ(ωωωα,l)(ıK( f ))) ⊆ ad(l)(K), ∀ f ∈ C(Hα; K, A) (1.1.24) hence since [38, Rmk 1.86] it remains only to show that σσσ(ωωωα,l)(ıK( fβ)) converges to 0 uniformly on Hα,l for any net { fβ}β in C(Hα; K, A) converging uniformly on Hα to 0. But this follows since (1.1.24) and since σ(l) is an isometry. st.(2) is easy to show. st.(3) (cid:3) follows since st.(1) and [38, Cor. 2.47]. Lemma 1.1.49. Let A be inner implemented by v, ψ ∈ EG A(τ) and h H, π, U i be a , σFψ i. Then h H, π, Ul i is a covariant representation covariant representation of h A, SG Fψ of h A, SG Fσ∗(l)(ψ) , σFσ∗(l)(ψ) i, where Ul (cid:17) ad(π(v(l))) ◦ U ◦ ad(l−1) ↾ SG . Fσ∗(l)(ψ) Proof. Since Lemma 1.1.29(2) Ul is well-defined, moreover it is a strongly continuous unitary group action since it is so U, since ad(V) is strongly continuous for any unitary operator V on H, and since ad(·) on H is an isomorphism of topological groups. Let l ∈ H, h ∈ SG and a ∈ A then Fσ∗(l)(ψ) Ul(h)π(a)Ul(h−1) = π(v(l))U(ad(l−1)h)π(v(l−1))π(a)π(v(l))U(ad(l−1)h−1)π(v(l−1)) = π(v(l))U(ad(l−1)h)π(σ(l−1)a)U(ad(l−1)h−1)π(v(l−1)) = π(v(l))π(σ(ad(l−1)(h)l−1)a)π(v(l−1)) = π(σ(lad(l−1)(h)l−1)a) = π(σ(h)a). (cid:3) Remark 1.1.50. Let us prove directly Lemma 1.1.48(3) in case A is inner implemented , σFσ∗(l)(ωωωα) i by v. Let h H, π, U i be a nondegenerate covariant representation of h A, SG Fσ∗(l)(ωωωα) then for any f ∈ Cc(SG Fωωωα (cid:17) ad(π(v(l−1))) ◦ U ◦ ad(l) ↾ SG Fωωωα , Ul−1 , A) 50 set (1.1.25) (cid:13)(cid:13)(cid:13)(π ⋊µµµα,l U)(σσσ(ωωωα,l)( f ))(cid:13)(cid:13)(cid:13) = (cid:13)(cid:13)(cid:13)Z π(σσσ(ωωωα,l)( f )(s))U(s) dµµµ(α,l)(s)(cid:13)(cid:13)(cid:13) = (cid:13)(cid:13)(cid:13)Z π(cid:16)(σ(l) ◦ f ◦ ad(l−1))(s)(cid:17)U(s) dµµµ(α,l)(s)(cid:13)(cid:13)(cid:13) = (cid:13)(cid:13)(cid:13)Z π(cid:16)(σ(l) ◦ f )(s)(cid:17)U(ad(l)s) dµµµ(α,1)(s)(cid:13)(cid:13)(cid:13) = (cid:13)(cid:13)(cid:13)Z π(v(l))π( f (s))π(v(l−1))U(ad(l)s)π(v(l))π(v(l−1)) dµµµ(α,1)(s)(cid:13)(cid:13)(cid:13) = (cid:13)(cid:13)(cid:13)Z ad(π(v(l)))(cid:16)π( f (s))Ul−1 (s)(cid:17) dµµµ(α,1)(s)(cid:13)(cid:13)(cid:13) = (cid:13)(cid:13)(cid:13)ad(π(v(l)))(cid:16)Z π( f (s))Ul−1 (s) dµµµ(α,1)(s)(cid:17)(cid:13)(cid:13)(cid:13) = (cid:13)(cid:13)(cid:13)(π ⋊µµµα,1 Ul−1 )( f )(cid:13)(cid:13)(cid:13) ≤ k f kµµµα,1. Here the third equality follows by Rmk. 1.1.39, the sixth one since ad(V) ∈ L(Ls(H)), for any V ∈ U(H) and the integration is w.r.t. the strong operator topology, the seventh one follows by the fact that ad(V) is an isometry since ad(V) ∈ AutCA∗(L(H)) and the well-known fact that any ∗−homomorphism between C∗−algebras is automatically con- tinuous with norm less or equal to 1. Finally the inequality follows since an application of Lemma 1.1.49 stating that h H, π, Ul−1 i is a nondegenerate covariant representation of h A, SG , σFωωωα i. Therefore Fωωωα (1.1.26) kσσσ(ωωωα,l)( f )kµµµ(α,l) ≤ k f kµµµ(α,1). Let h, s ∈ H and g ∈ Cc(SG F ωωωh α µµµh to obtain , A), since Lemma 1.1.48(2) we can apply (1.1.26) to ωωωh and kσσσ(ωωωh α,s)(g)kµµµh (α,s) ≤ kgkµµµh (α,1). Next µµµl (α,l−1) = µµµ(α,1) and µµµl (α,1) = µµµ(α,l) thus for any g ∈ Cc(SG Fσ∗(l)(ωωωα) , A) kσσσ(ωωωl α,l−1)(g)kµµµ(α,1) ≤ kgkµµµ(α,l), moreover σσσ(ωωωl α,l−1) = (σσσ(ωωωα,l))−1 since Lemma 1.1.48(2), therefore (1.1.27) k f kµµµ(α,1) ≤ kσσσ(ωωωα,l)( f )kµµµ(α,l). Lemma 1.1.48(3) follows by (1.1.26) & (1.1.27). 1.1. EQUIVARIANCE 51 Fωωωα , σFωωωα i, and set V ≔ ad(π(v(l))) ◦ Vl−1 ◦ ad(l−1) ↾ SG Remark 1.1.51. Let h H, π, Vl−1 i be a nondegenerate covariant representation of h A, SG σ∗(l)(Fωωωα ), then since Lemma 1.1.49 h H, π, V i is a nondegenerate covariant representation of h A, SG , σFσ∗(l)(ωωωα) i. Hence we can reload the chain of equalities (1.1.25) in the opposite sense by replacing U by V and Ul−1 by Vl−1, for obtaining (1.1.27). Fσ∗(l)(ωωωα) Corollary 1.1.52. Let ωωω : A → EG A(τ) and µµµ be a Haar system associated with ωωω and A. Then for any (α, l) ∈ A × H there exists a unique extension by continuity of σσσ(α,l) to µµµ , denoted by the same symbol, such that σσσ(ωωωα,l) ∈ InvCA∗(Bωωω,α Bωωω,α µµµ , Bωωω,α,l ). µµµ µµµ µµµ to Bωωω,α,l Proof. Since Lemma 1.1.48(3) there exists a unique extension by continuity of σσσ(ωωωα,l) from Bωωω,α , while since the first sentence of Lemma 1.1.48(2) and Lemma 1.1.48(3) α,l−1) from Bωωω,α,l applied to ωωωl and to µµµl there exists a unique extension by continuity of σσσ(ωωωl to Bωωω,α µµµ . Since the second sentence of Lemma 1.1.48(2) such two extensions are one the inverse of the other. Finally the extension σσσ(ωωωα,l) is a ∗−homomorphism since Lemma 1.1.48(1) and the norm continuity of the product and involution on Bωωω,α (cid:3) µµµ . µµµ Corollary 1.1.53. Let l, h ∈ H and α ∈ A, then σσσ(σ∗(l)(ωωωα),h) ◦ σσσ(ωωωα,l) = σσσ(ωωωα,h·l). Proof. The equality holds if restricted to Cc(SG Fωωωα , A) hence the statement follows by (cid:3) Cor. 1.1.52. We are now in the position to state the second main result of this part namely Theorem 1.1.54 (Equivariance of representations). Let A be inner implemented by v, ωωω : A → EG A(τ), µµµ be a Haar system associated with ωωω and A, while HHH : A → Repc(A) such that HHHα ≔ h Hα, πα, Ωα i is a cyclic representation of A associated with ωωωα, for all α ∈ A. Then for all α ∈ A and l ∈ H the following diagram is commutative Rµµµ HHH,v,α,l Bωωω,α,l µµµ / L(Hα) σσσ(ωωωα,l) ad(πα(v(l))) Bωωω,α µµµ / L(Hα) Rµµµ HHH,α Proof. By construction Cc(SG µµµ , moreover all the maps involved in the statement are norm continuous and linear, since Cor. 1.1.52 and the well-known fact that any ∗−homomorphism between C∗−algebras is automatically con- tinuous. Thus it is sufficient to show the statement for the maps σσσ(ωωωα,l) and Rµµµ HHH,α restricted , A) is k · kµµµ(α,1)−dense in Bωωω,α Fωωωα / O O / O O 52 on Cc(SG Fωωωα , A). Let f ∈ Cc(SG Fωωωα , A), then (cid:16)(πα ⋊µµµ(α,l) UHHH(v,l) Z πα(σσσ(ωωωα,l)( f )(s))UHHH(v,l) α α ) ◦ σσσ(ωωωα,l)(cid:17)( f ) = (s) d µµµ(α,l)(s) = Z πα(σσσ(ωωωα,l)( f )(ad(l)s))UHHH(v,l) α (ad(l)s) d µµµ(α,1)(s) = Z πα(cid:16)(σ(l) ◦ f )(s)(cid:17)πα(v(l))UHHHα(s)πα(v(l−1))d µµµ(α,1)(s) = Z πα(v(l))πα( f (s))UHHHα(s)πα(v(l−1))d µµµ(α,1)(s) = ad(πα(v(l)))(cid:16)Z πα( f (s))UHHHα(s)d µµµ(α,1)(s)(cid:17) = (cid:16)ad(πα(v(l))) ◦ (πα ⋊µµµ(α,1) UHHHα)(cid:17)( f ). Here the second equality follows since Rmk. 1.1.39, the third by Cor. 1.1.35, the fifth by ad(V) ∈ L(Ls(Hα)), for any V ∈ U(Hα) and since the integration is w.r.t. the strong (cid:3) operator topology. 1.2. Extensions on the multiplier algebra We prove in Lemma 1.2.2 that there is a unique canonical manner to extend a state of a C∗−algebra to a state of its multiplier algebra. Then we apply this result in Cor. 1.2.5 to associate a state of A with a state of the crossed product A ⋊µ σ H, where h A, H, σ i is a dynamical system such that A is unital and µ ∈ H(H). Cor. 1.2.5 is used via Lemma 2.2.26 in constructing an equivariant stability in Thm. 2.2.81. Lemma 1.2.6 and Lemma 1.2.12 are used to obtain Thm. 2.3.25 one of the auxiliary results needed to prove in Cor. 2.3.26 the existence of the object part of a functor from Cu(H) to G(G, F, ρ). We prove in a functional analytic setting the convergence formula (1.2.7), and the extension results Lemma 1.2.12, Lemma 1.2.6, Cor. 1.2.10 and from Lemma 1.2.13 to Cor. 1.2.17. Finally we prove in a different way Lemma 1.2.8 and the convergence in (1.2.11). Convention 1.2.1. In the present section we fix a C∗−algebra B and for any nonzero positive trace class operator ρ on a Hilbert space H, we use the convention of denoting with ωρ the state Tr(ρ a) Tr(ρ) , for all a ∈ L(H). Moreover by abuse of language for any unit on L(H) defined by ωρ(a) ≔ vector Ω ∈ H let ωΩ denote ωPΩ, where PΩ is the projector associated with the closed subspace generated by Ω. Let ψ ∈ EB then HHH = h H, R, ρ i is called a representation of B relative to ψ, if h H, R i is a nondegenerate representation of B and ρ is a positive trace class operator on H such that ψ = ωρ ◦ R, define HHH− ≔ h H, R−, ρ i. Lemma 1.2.2. Let ψ ∈ EB, then (1) (∃ !ψ− ∈ EM(B))(ψ− ◦ iB = ψ), 1.2. EXTENSIONS ON THE MULTIPLIER ALGEBRA 53 (2) if HHH is a representation of B relative to ψ then HHH− is a representation of M(B) relative to ψ−, Remark 1.2.3. Since Lemma 1.2.2(2) if HHH is a cyclic representation of B associated with ψ then HHH− is a cyclic representation of M(B) associated with ψ−. Proof of Lemma 1.2.2. In this proof let i denote iB. Let HHH = h H, R, ρ i be a represen- tation relative to ψ, set φ = ωρ ◦ R−, thus φ is a state of M(B) since R− is a representation of M(B) such that R−(1) = 1. Moreover by construction HHH− is a representation associated with φ, while φ ◦ i = ψ, hence the existence part of st.(1) and st.(2) follow. Let us prove the uniqueness part of st.(1). Let ψ− ∈ EM(B) such that ψ− ◦ i = ψ, and h K, S, Ω i be a cyclic associated with ψ−, set K0 = S(i(B))Ω, S↾ : M(B) ∋ c 7→ S(c) ↾ K0,  the norm topology. So S↾ is a representation of M(B) on where the closure is w.r.t. K0, since S(c) is norm continuous for any c ∈ M(B) and i(B) is an ideal of M(B). Next Ω ∈ K0 since a standard argument, [14, p. 56], applied to the state ψ = ωΩ ◦ S ◦ i and the representation S ◦ i. Hence KKK↾ (cid:17) h K0, S↾, Ω i is a cyclic representation of M(B) such that ωΩ ◦ S↾ = ψ−, i.e. (1.2.1) Next let KKK↾ is a cyclic associated with ψ−. KKK0 (cid:17) h K0, S↾ ◦ i, Ω i, then (S↾ ◦ i)(B)Ω = S↾(i(B))Ω = K0, hence KKK0 is a cyclic representation of B since S↾ is a representation of M(B). Moreover ωΩ ◦ S↾ ◦ i = ψ− ◦ i = ψ the first equality coming since (1.2.1), thus (1.2.2) KKK0 is a cyclic associated with ψ. Next let ψj be a state of M(B) such that ψj ◦ i = ψ and KKKj = h K j, S j, Ω j i be a cyclic associated with ψj, for any j ∈ {1, 2}. Thus since (1.2.2) and the uniqueness modulo unitary equivalence of the GNS construction for states on a C∗−algebra there exists a unique unitary operator U : K1 0 → K2 0 such that (1.2.3) UΩ1 = Ω2 ↾ ◦ i = ad(U) ◦ (S1 S2 ↾ ◦ i).  54 Moreover (S j (0.2.24) applied to the cyclic KKK j ↾ ◦ i)− = S j ↾(c)(S2 S2 where d = i−1(ci(b)). Next KKK2 ↾ since (0.2.25) applied to the cyclic KKK j ↾, therefore since (1.2.3) and ↾ ◦ i)(b)Ω2 ↾ ◦ i)−(c)(S2 ↾ ◦ i)(d)Ω2 0 for any j ∈ {1, 2}, we obtain for all c ∈ M(B) and b ∈ B ↾ ◦ i)(b)Ω2 = (S2 = (S2 = U(S1 = US1 = US1 ↾ ◦ i)(b)Ω1 ↾ ◦ i)(b)Ω2, ↾ ◦ i)(d)Ω1 ↾(c)(S1 ↾(c)U−1(S2 ↾ = ad(U) ◦ S1 ↾, therefore 0 is cyclic so S2 ωΩ2 ◦ S2 ↾ = ωΩ1 ◦ S1 ↾, thus ψ− 1 = ψ− 2 since (1.2.1). (cid:3) Definition 1.2.4. Let ψ ∈ EB, then we call the canonical extension of ψ to M(B) the unique state ψ− such that ψ− ◦ iB = ψ. Corollary 1.2.5. Let A = h A, H, σ i be a dynamical system such that A is unital, µ ∈ H(H) and ψ a state of B = A ⋊µ σ H. Thus ψ− ◦ jB A ∈ EA and (1) if h H, R, ρ i is a representation relative to ψ then h H, π, ρ i is a representation relative to ψ− ◦ jB A, (2) if h H, R, Ω i is a cyclic representation associated with ψ then h Hπ Ω, π↾, Ω i is a cyclic representation associated with ψ− ◦ jB A. Here h H, π, U i is the covariant representation of A associated with R, Hπ Ω π↾ : A ∋ a 7→ π(a) ↾ Hπ Ω. Proof. Let B = A ⋊µ A is nondegenerate, thus (ψ− ◦ jB σ H. ψ− is a state of M(B) since Lemma 1.2.2(1), while jB A(1) = 1 since jB A is positive hence it is a state of A. Let h H, R, ρ i be a representation relative to ψ which there exists since ψ is a state, thus π is nondegenerate since R it is so, moreover A = ωρ ◦ R− ◦ jB A)(1) = 1 moreover ψ− ◦ jB ≔ π(A)Ω and ψ− ◦ jB A (1.2.4) = ωρ ◦ π, where the first equality follows since Lemma 1.2.2(2) and the second one since (0.2.26), so st.(1) follows. Next if h H, R, Ω i is a cyclic representation associated to ψ, then ψ− ◦ jB Ω since the argument in [14, p.56] applied (cid:3) to the state ψ− ◦ jB A = ωΩ ◦ π since (1.2.4) therefore Ω ∈ Hπ A and the representation π, so st.(2) follows. The next Lemma 1.2.6 together Lemma 1.2.12 are important in showing Thm. 2.3.25 one of the auxiliary results used in the proof of Cor. 2.3.26 were we construct the object part of a functor from Cu(H) to G(G, F, ρ). 1.2. EXTENSIONS ON THE MULTIPLIER ALGEBRA 55 Lemma 1.2.6. Let h A, H, σ i be a dynamical system, µ ∈ H(H), B = A ⋊µ σ H. Then for any a ∈ A the following is a commutative diagram (1.2.5) M(B) eva(iM(B)◦jB A) / M(B) iB B iB / B eva(jB A) in particular for all f ∈ Cc(H, A) the following diagram is commutative (1.2.6) B iB( f ) B jB A(a) / B ?⑧⑧⑧⑧⑧⑧⑧⑧⑧⑧⑧⑧⑧⑧⑧⑧ iB(jB A(a)( f )) Proof. (1.2.5) trivially implies (1.2.6), moreover all the maps involved in (1.2.5) are norm continuous and Cc(H, A) is norm dense in B hence (1.2.6) implies (1.2.5), thus let us proof (1.2.6). Let f ∈ Cc(H, A) and let here i and j denote iB and jB A respectively, thus for all g ∈ Cc(H, A) and s ∈ H (j(a) ◦ i( f ))(g)(s) = j(a)( f ∗µ g)(s) = aZ f (r)σ(r)(g(r−1s))d µ(r) =Z a f (r)σ(r)(g(r−1s))d µ(r) = (j(a)( f ) ∗µ g)(s) = i(j(a)( f ))(g)(s). (cid:3) Lemma 1.2.7. Let h A, H, σ i be a dynamical system, µ ∈ H(H) and {Eβ}β∈C be an the approximate identity of A. Let B denote A ⋊µ topology on M(M(B)) of simple convergence in iB(B), i.e. for all m ∈ iB(B) σ H then limβ iM(B)(j(Eβ)) = 1 w.r.t. (1.2.7) kjB A(Eβ)m − mkM(B) = 0. lim β∈C Proof. Let A = h A, H, σ i, j = jB A, i = iB and k · kB be the universal norm on B. We have supp(iA(Eβ) ◦ f ) ⊆ supp( f ) for any f ∈ Cc(H, A) and β ∈ C since Eβ is linear and / O O / O O / O O ? 56 supp( f ) (cid:17) ∁ f −1(0), hence sup k(Eβ f − f )(l)k = sup{k(Eβ f − f )(l)k l ∈ supp(iA(Eβ) ◦ f ) ∪ supp( f )} (1.2.8) l∈supp( f ) = sup l∈H k(Eβ f − f )(l)k. Since the definition of the approximate identity we deduce that limβ iA(Eβ) = 1 w.r.t. the topology on M(A) of simple convergence in A, moreover kiA(Eβ)kM(A) ≤ 1 since iA is an isometry into its range, next the unit ball of M(A) is clearly a bounded subset of M(A) w.r.t. the topology of simple convergence in A, hence it is equicontinuous according to [11, III.25 Thm. 1]. Therefore we can apply [11, III.17 Prp. 5(2, 3)] and deduce that limβ iA(Eβ) = 1 w.r.t. the topology on M(A) of uniform convergence in compact subsets of A, i.e. limβ supa∈K kEβa − akA = 0, for any compact subset K of A. Next f (Q) is a compact subset of A for any f ∈ Cc(H, A) and any compact subset Q of H, therefore lim β sup l∈Q (cid:13)(cid:13)(cid:13)(cid:13)(cid:16)j(Eβ)( f ) − f(cid:17)(l)(cid:13)(cid:13)(cid:13)(cid:13)A In particular limβ supl∈H(cid:13)(cid:13)(cid:13)(cid:13)(cid:16)j(Eβ)( f ) − f(cid:17)(l)(cid:13)(cid:13)(cid:13)(cid:13)A limβ j(Eβ)( f ) = f w.r.t. the L1 see [38, Lemma 2.27], therefore = lim β kEβ f (l) − f (l)kA = 0. sup l∈Q µ−norm topology. Next k · kB is majorized by the L1 = 0 since (1.2.8) and supp( f ) is compact, so µ−norm, (1.2.9) lim β kj(Eβ)( f ) − f kB = 0 ∀ f ∈ Cc(H, A). Next i is an isometry onto its range thus by (1.2.9) & (1.2.5) we obtain (1.2.10) iM(B)(j(Eβ)) = 1, lim β the topology on M(M(B)) of simple convergence in i(Cc(H, A)). Next by con- w.r.t. struction, see [38, Lemma 2.27], Cc(H, A) is dense in B w.r.t. the k · kB−topology hence i(Cc(H, A)) is dense in i(B), moreover kiM(B)(j(Eβ))kM(M(B)) ≤ 1 for any β ∈ C since iM(B) ◦ j is an isometry. Next the unit ball of M(M(B)) is clearly a bounded subset w.r.t. the topology on M(M(B)) of simple convergence in M(B) hence it is equicontinuous according to [11, III.25 Thm. 1]. Therefore since (1.2.10) we can apply [11, III.17 Prp. 5(1, 2)] and the (cid:3) statement follows. Lemma 1.2.8. Let A and B be C∗−algebras, X be a Hilbert B−module and π : A → L(X) a ∗−homomorphism. Then π maps approximate identities of A into approximate identities of π(A), if in addition π is nondegenerate and {Eβ}β∈C is an approximate identity of A, then limβ π(Eα) = 1 w.r.t. the topology on L(X) of simple convergence. Proof. The first sentence of the statement follows since π is norm continuous with norm less or equal to 1 and since it is order preserving. If π is nondegenerate then X = span{π(a)x a ∈ A, x ∈ X} thus lim π(Eβ) = 1 w.r.t. the topology on L(X) of simple convergence in a total subset of X, since the first sentence of the statement. Next for any β ∈ C the π(Eβ) lies in the unit ball of L(X) which is a bounded set w.r.t. the topology of 1.2. EXTENSIONS ON THE MULTIPLIER ALGEBRA 57 simple convergence in X hence equicontinuous by [11, III.25 Thm. 1]. Therefore we can apply [11, III.17 Prp. 5(1, 2)] and deduce that lim π(Eβ) = 1 w.r.t. the topology on L(X) (cid:3) of simple convergence in X. Remark 1.2.9. We can deduce for any a ∈ B (1.2.11) kjB A(Eβ)a − akB = 0, lim β∈C as an application of Lemma 1.2.8 to the Hilbert B−module X = B and to the nondegen- erate homomorphism π = jB A. Thus since Lemma 1.2.6 and since iB is an isometry into its range we obtain (1.2.7). Viceversa we can use Lemma 1.2.6 and (1.2.7) to obtain (1.2.11). Corollary 1.2.10. Let A be a C∗−algebra, (H, R) a nondegenerate representation of A and ρ a nonzero positive trace class operator on H. Then ωρ ◦ R ∈ EL(H). Proof. Let {Eβ}β∈C be an approximate identity of A then kR(Eβ)k ≤ 1 for all β ∈ C and limβ R(Eβ) = 1 weakly, since Lemma 1.2.8 and the strong operator topology is stronger than the weak operator one. Therefore limβ R(Eβ) = 1 σ−weakly since [14, Prp. 2.4.2], so (1.2.12) ωρ(R(Eβ)) = ωρ(1) = 1, lim β since ωρ is σ−weakly continuous, see for example [14, Thm. 2.4.21]. Next ωρ ◦ R is (cid:3) positive hence the statement follows since (1.2.12) and [14, Prp. 2.3.11]. Remark 1.2.11. Under the hypotheses of Cor. 1.2.10 and since Tr(ρ) ωρ = Tr ◦ Lρ we have that kTr ◦ Lρ ◦ Rk = Tr(ρ). Lemma 1.2.12. Let A be a C∗−algebra then the map M(A) ∋ u 7→ iM(A)(u) ↾ K(A) is a ∗−isomorphism of M(A) onto M(K(A)). Proof. (M(A), Id ↾ K(A)) is a maximal unitization of K(A) since [32, Cor. 2.54], hence by [32, proof. of Thm. 2.47] there exists a unique ∗−homomorphism Φ : M(A) → M(K(A)) such that Φ ◦ Id ↾ K(A) = iK(A), moreover Φ is a ∗−isomorphism. Next iM(A)(u) ↾ K(A) = iK(A)(u) for any u ∈ K(A) therefore the statement follows since the above uniqueness. (cid:3) Let us end this section by proving useful results concerning the extension of suitable morphisms to multiplier algebras. Lemma 1.2.13. Let A, B and C be C∗−algebras, X and Y be a Hilbert B−module and Hilbert C−module respectively, β : L(X) → L(Y) and α : A → L(X) be ∗−homomorphisms such that β is nondegenerate. If α is surjective, or α(A) is strictly dense in L(X) and β(K(X)) ⊇ K(Y), then β ◦ α is nondegenerate and (β ◦ α)− = β ◦ α−. Proof. Let δ = β ◦ α and y ∈ Y. Case α surjective. Of course β(L(X)) = δ(A), so (1.2.13) β(L(X))Y ⊆ span{δ(a)z a ∈ A, z ∈ Y}, and the first sentence of the statement follows since β is nondegenerate. Remaining case. β is strictly continuous since it is norm continuous and β(K(X)) ⊇ K(Y). Moreover β is continuous w.r.t. the strict topology on L(X) and the strong operator topology on 58 L(Y) since the strict topology on L(Y) is stronger than the ∗−strong topology ([32, Prp. C.7]) so stronger than the strong operator topology. Then since the hypothesis it follows that β(L(X)) ⊆ δ(A) closure w.r.t. the strong operator topology on L(Y), so (1.2.13) follows. Then β ◦ α is nondegenerate and (β ◦ α)− ◦ iA = β ◦ α since (0.2.22). Moreover α is nondegenerate in both cases, indeed 1 ∈ L(X), while the strict topology on L(X) is stronger than the strong operator topology. So α− ◦ iA = α, then the equality follows (cid:3) since the uniqueness of which in (0.2.22). More in general we can state Proposition 1.2.14. Let A, B and C be C∗−algebras, X and Y be a Hilbert B−module and Hilbert C−module respectively, β : L(X) → L(Y) and α : A → L(X) be ∗−homomorphisms. If 1Y belongs to the strong operator closure of the set (β ◦ α)(A) and α is nondegenerate, then β ◦ α is nondegenerate and (β ◦ α)− = β ◦ α−. Proof. α− ◦ iA = α since (0.2.22), while since the hypothesis there exists a net {ai} in A such that y = limi(β ◦ α)(ai)y for all y ∈ Y, thus β ◦ α is nondegenerate. Therefore (β ◦ α)− ◦ iA = β ◦ α and the equality in the statement follows since the uniqueness of (cid:3) which in (0.2.22). Remark 1.2.15. Under the notations of Prp. 1.2.14 we obtain the same statement if α is surjective and β is nondegenerate, indeed in such a case β ◦ α is nondegenerate. Clearly we obtain the same statement if we only require α and β◦α to be nondegenerate. Corollary 1.2.16. Let A, B and C be C∗−algebras, Y be a Hilbert C−module β : B → L(Y) and α : A → B be ∗−homomorphisms. If 1Y belongs to the strong operator closure of the set (β◦α)(A) and α is surjective, then β◦α is nondegenerate and (β◦α)− = β−◦(iB◦α)−. Proof. Since iB is nondegenerate and α is surjective then iB ◦ α is nondegenerate, moreover β− ◦ iB ◦ α = β ◦ α. Hence we can apply Prp. 1.2.14 to the maps β− and iB ◦ α, (cid:3) and the statement follows. Corollary 1.2.17. Let A, B and C be C∗−algebras, Y be a Hilbert C−module β : B → L(Y) and α : A → B be ∗−homomorphisms. If β ◦ α is nondegenerate and α is surjective then and (β ◦ α)− = β− ◦ (iB ◦ α)−. Proof. Since iB is nondegenerate and α is surjective then iB ◦ α is nondegenerate, (cid:3) moreover β− ◦ iB ◦ α = β ◦ α. Thus the statement follows since Rmk. 1.2.15. Part 2 Stability 2.1. Introduction In this part we construct the category G(G, F, ρ) of nucleon systems, introduce its physical interpretation, define nucleon-fragment doublets on a category C and extended C−equivariant stabilities, construct the canonical extended Cu(H)−equivariant stability and the canonical nucleon-fragment doublet on Cu(H). The part is organized as follows. In section 2.2.1 we introduce the category G(G, F, ρ), describe in section 2.2.2 the concept of a state originated via a phase and introduce the physical interpretation of these data in section 2.2.3. In section 2.2.4 we define the main structure of the work namely the nucleon-fragment doublet on a category C, and introduce the auxiliary structures of equivariant stability and extended C−equivariant stability. Extended C−equivariant stabilities are important because of with each of them easily we can associate a nucleon- fragment doublet on C. Our tenet is to translate the physical concept of "universality" into the concept of "natural transformation" in category theory. Therefore our goal of resolving the universality claim justifies the introduction of nucleon-fragment doublets. Indeed in this section we describe the symmetry properties of a doublet and define the expanded equivariances of the T−nucleon phase and T−fragment state, representing the T−resolution of the equivariant form of the universality claim. Then we describe the properties of invariance and the corresponding physical interpretation. Section 2.2.5 is dedicated to establish the existence of an equivariant stability. In section 2.3.1 we construct the object part of a functor from Cu(H) to G(G, F, ρ), where the former is a subcategory of C∗−dynamical systems and equivariant morphisms, and in section 2.3.2 we complete the construction of the functor. It is in section 2.3.2 that we state in our main theorem the existence of the canonical extended Cu(H)−equivariant stability and the canonical nucleon-fragment doublet T• on Cu(H). As a result we resolve the equivariant form of the universality claim stated as T•−resolution. In part 3 we shall provide the compact equivariant form and the invariant form of the universality claim, and as a result we will establish the universality of the global Terrell law. 2.2. The category of nucleon systems Section 2.2.1 is dedicated to the definition of the category of nucleon systems, whose physical interpretation is given in section 2.2.3 by mean of the cardinal concept of origination of fragment states via a nucleon phase developed in section 2.2.2. In section 2.2.4 we introduce the concept of nucleon-fragment doublet on a category, the main structure of this work, and state its symmetries. In the same section we define equivariant stabilities and extended C−equivariant stabilities auxiliary objects to construct nucleon- fragment doublets. In section 2.2.5 we construct an equivariant stability. In the present section 2.2 we assume fixed two locally compact topological groups G and F, a group homomorphism ρ : F → AutGr(G) such that the map (g, f ) 7→ ρ f (g) on G × F at values in G, is continuous, moreover let H denote G ⋊ρ F. 2.2.1. The category G(G, F, ρ). 2.2. THE CATEGORY OF NUCLEON SYSTEMS 61 Definition 2.2.1 (The category of dynamical systems). For any locally compact group V we define the category C(V) whose object set is the set of the dynamical systems h A, V, η i such that A is unital, while for any A, B ∈ Obj(C(V)) we define MorC(V)(A, B) the set of the surjective (A, B)−equivariant morphisms with law of composition the map composition and if A = B the identity map as the identity morphism. Let Cu(V) denote the full subcategory of C(V) whose object set is the set of the dynamical systems A = h A, V, η i such that A is a von Neumann algebra in its canonical standard form. Let vA or vη denote the unique group morphism of H into U(A) unitarily implementing and associated with η and to the canonical standard form of A according to [36, Thm 9.1.15]. Definition 2.2.2 (The category of nucleon systems). Define Obj(G(G, F, ρ)) the set of the G = h TTT, I, βc, P, a, e, ϕϕϕ, A, ψ, b, m, E i such that (1) TTT is a set; (2) I : TTT → set; P(IQ); (3) βc ∈QQ∈TTT IQ; (4) P ∈QQ∈TTT (5) a ∈QQ∈TTTQα∈IQ C(H); (6) e ∈QQ∈TTTQα∈IQ C(R); (7) ϕϕϕ ∈QQ∈TTTQα∈IQ EAQ (11) m ∈ MorAb(A,QQ∈TTT Morset(PQ, R)). (8) A ∈ Ab (9) ψ ∈ MorGr(H, AutAb(A)); (10) b ∈ MorGr(H, Autset(TTT)); α = h AT ; α Here aT denote b(l)(T) for all l ∈ H. Then we require for all T ∈ TTT and α ∈ T α i, in addition let FϕϕϕT denote FϕϕϕT α , H, ηT (aT α α ) for all T ∈ TTT and α ∈ IT, and let Tl α that ITl = IT and (2.2.1) Moreover ϕϕϕT α ∈ EG AT α (τηT α ), βTl c = βT c , ad(Pr 2 (l))(FϕϕϕT α ) = F ϕϕϕTl α . Phase transition via dynamical symmetry breaking. for all T ∈ TTT (2.2.2) PT = {α ∈ IT FϕϕϕT α ⊇ FϕϕϕT βT c }. Equivariance. For any l ∈ H and f ∈ A 3 (2.2.3) evf(m ◦ ψ(l)) = evf(m) ◦ b(l−1). Thermal nature. For any T ∈ TTT and α ∈ IT we have eT α − KMS, α , ∀l ∈ H; α ∈ εT α (l)) ◦ εT εTl α = ad(ηT ϕϕϕT (2.2.4) α = h AT α , R, εT α i and 3 (2.2.3) is well-set since Rmk. 2.2.4 62 Integrality. m(f)(T) is a Z−valued map, for all (f, T) ∈ A × TTT. Equivariant stability. E = h µ, u, HHH, D, Γ , v, w, z i such that β (G)+)), β ) β (G)), (4) HHHT α ∈ PT; α = h HT (1) µ ∈QQ∈TTTQβ∈PQ H(SQ (2) u ∈QQ∈TTTQβ∈PQ ∈ MorAb(A, K0(BQ (3) HHH ∈QQ∈TTTQβ∈PQ Repc(AQ (5) D, Γ ∈QQ∈TTTQβ∈PQ Sd(HQ (6) v ∈QQ∈TTTQβ∈PQQl∈H U(HQ (7) w ∈QQ∈TTTQβ∈PQQl∈H MorCA∗(BQ (8) z ∈QQ∈TTTQβ∈PQQl∈H MorCA∗(AQ β , HQl β ); (9) for all T ∈ TTT, α ∈ PT, α , Γ T T (a) h RRR α (G), DT (b) for all l, h ∈ H and g ∈ {v, w, z} β ); α , πT α , ΩT α i is a cyclic representation of AT α associated with ϕϕϕT α , for all T ∈ TTT, β (G)), β (G), BQl β , AQl β ), α i is an even θ−summable K−cycle; (2.2.5) (2.2.6) (2.2.7) ηTl α (h) ◦ zT α (h), α (l) ◦ ηT α (h) ◦ gT α (l) α (l) = zT α (h · l) = gTl gT gT α (1) = Id. (c) we have DTl α = vT Γ Tl α = vT α (l) DT α (l) Γ T α vT α vT α (l)−1, α (l)−1, while the following (2.2.7,2.2.8,2.2.9) are commutative diagrams BTl α (G) RTl α (G) / L(HTl α ) wT α (l) ad(vT α (l)) BT α (G) / L(HT α ) RT α (G) / O O / O O 2.2. THE CATEGORY OF NUCLEON SYSTEMS 63 A ψ(l) A uTl α / K0(BTl α (G)+) (wT α (l)+)∗ / K0(BT α (G)+) uT α M(BTl α (G)) jTl α ATl α ηTl α (l) M(BT α (G)) (iTl α ◦wT α (l))− >⑥⑥⑥⑥⑥⑥⑥⑥⑥⑥⑥⑥⑥⑥⑥⑥⑥ `❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆❆ jT α (d) for all f ∈ A we have AT α / ATl α zT α (l) m(f)(T, α) = h uT α (f), chh RRR T α (G), DT α , Γ T α i i(G,T,α). (2.2.8) (2.2.9) (2.2.10) Here for any T ∈ TTT and α ∈ PT we set (2.2.11) , F ϕϕϕϕϕϕϕϕϕT α α (G) ≔ SG ST α (G) ≔ BϕϕϕT BT α µT α ≔ iBT α (G), ≔ jBT α (G) α (G) ≔ RµT RT α HHHT α α (G) ≔ RRRµT α HHHT α iT α jT α RRRT AT α , (aT α ), (aT α ), α ), h ·, · i(G,T,α) ≔ h ·, · i(BT α (G))+. (aT We call any object G of G(H) a nucleon system with symmetry group H, or simply nucleon system, moreover we call TTT and m the set of operations and the mean value map associated with G respectively. / O O / O O o o > ` / O O 64 In the following definition and remark let G = h TTT, I, βc, P, a, e, ϕϕϕ, A, ψ, b, m, E i be an object of G(G, F, ρ), l ∈ H, T ∈ TTT and α ∈ PT, where E = h µ, u, HHH, D, Γ , v, w, z i. Definition 2.2.3. Define V(G)T Remark 2.2.4. PT = PTl since (2.2.1). Let g ∈ {v, w, z} then by (2.2.5) we deduce that α ) such that V(G)T α (l) ∈ MorCA∗(AT α (l) ≔ zT α (l) ◦ ηT α (l). α , ATl gT α (l) is bijective and (2.2.12) (gT α (l))−1 = gTl α (l−1), so ad(zT α (l)) ◦ ηT α = ηTl α , then (2.2.13) moreover V(G)T α (1) = Id thus V(G)T α (h · l) = V(G)Tl α (h) ◦ V(G)T α (l), (2.2.14) (V(G)T α (l))−1 = V(G)Tl α (l−1). α (l) is surjective being bijective and iTl Next wT degenerate since Lemma 1.2.13. Therefore (iTl (0.2.22) satisfies (iTl obtain α (l))− ◦ iT α = iTl α ◦ wT α ◦ wT α is nondegenerate, thus iTl α ◦ wT α (l) is non- α (l))− in (2.2.9) is well set and since α (l). Finally by an application of (2.2.9) we α ◦ wT (2.2.15) α ◦ wTl (iT α (l−1))− ◦ jTl α = jT α ◦ ηT α (l−1) ◦ zTl α (l−1). c, Pi, ai, ei, ϕϕϕi, Ai, ψi, bi, mi, Ei i ∈ Definition 2.2.5. For any i ∈ {1, 2, 3} let Gi = h TTTi, Ii, βi Obj(G(G, F, ρ)). Define MorG(G,F,ρ)(G1, G2) to be the set of the (g, d) ∈ MorAb(A1, A2) × Morset(TTT2, TTT1) such that Pd(T) 1 = PT 2 for all T ∈ TTT2, and for all f ∈ A1 we have evf(m2 ◦ g) = evf(m1) ◦ d. Moreover for any (g, d) ∈ MorG(G,F,ρ)(G1, G2) and (h, s) ∈ MorG(G,F,ρ)(G2, G3) define (2.2.16) (h, s) ◦ (g, d) ≔ (h ◦ g, d ◦ s). Thus we have the following Proposition 2.2.6. There exists a unique category G(G, F, ρ) whose set of objects is Obj(G(G, F, ρ)) and for all objects G1 and G2 the set of the morphisms from G1 to G2 is MorG(G,F,ρ)(G1, G2) provided by the law of composition as defined in Def. 2.2.5. Proof. For any i ∈ {1, 2, 3} let Gi ∈ G(G, F, ρ), moreover (g, d) ∈ MorG(G,F,ρ)(G1, G2) and (h, s) ∈ MorG(G,F,ρ)(G2, G3). The identity morphism in MorG(G,F,ρ)(G1, G1) is the couple composed by the identity maps on A1 and on TTT1. Let f ∈ A1, then evf(m3 ◦ h ◦ g) = evg(f)(m3 ◦ h) = evg(f)(m2) ◦ s = evf(m2 ◦ g) ◦ s = evf(m1) ◦ d ◦ s. Hence (h, s) ◦ (g, d) ∈ MorG(G,F,ρ)(G1, G3), it is easy to see that the composition is associative (cid:3) hence the statement follows. 2.2. THE CATEGORY OF NUCLEON SYSTEMS 65 Convention 2.2.7. Let G be an object of G(G, F, ρ), if the contrary is not stated, then we shall use the following notation G = h TTTG, IG, βG EG = h µG, uG, HHHG, DG, Γ G, vG, wG, zG i; c , PG, aG, eG, ϕϕϕG, AG, ψG, bG, mG, EG i, and for any Q ∈ TTTG and β ∈ PQ G . (aG)Q (eG)Q (HHHG)Q β = h A(G)Q β = h A(G)Q β = h (HG)Q β , H, (ηG)Q β , R, (εG)Q β , (πG)Q β i, β i, β , (ΩG)Q β i. Remark 2.2.8. H ∋ l 7→ (ψN(l), bN(l−1)) ∈ AutG(G,F,ρ)(N) is a morphism of groups for any object N of G(G, F, ρ) since (2.2.16), since ψN and bN are morphisms of groups and by (2.2.3). 2.2.2. States originated via a phase. The set NN(c) of states originated via a given phase c ∈ A∗ N is a primary concept needed in order to provide a reasonable physical interpretation of the structural data of any object N of G(G, F, ρ). Let us briefly describe and physically interpret this set, whose precise definition is given in Def. 2.2.24 through Def.2.2.15. Let RepN(c) be the set of the tuples r = h T, α, ν, q, KKK, Φ i, called represen- tations of c, where T ∈ TTTN, α ∈ PT α (N), KKK = h K, θ, Ω i is a GNS−represetation associated with the state (ϕϕϕN)T α , q is a group morphism from AN to the K0−theory of B+ α ) and Φ is an entire normalized even cocycle on B+ r such that c factorizes through the real part, via the standard duality, of the character generated by Φ and the map q, i.e. c = vr with N, ν is a Haar measure on ST r , where Br = B(ϕϕϕN)T ((aN)T ν α vr ≔ ℜh q(·), [Φ] i. α , Φ, A(N)T α , θ ⋊ W, T, α, (εN)T α (h)(a))Ω, for all h ∈ ST Here [Φ] is the entire cyclic cohomology class generated by Φ, h ·, · i is the standard duality between the entire cyclic cohomology and the K0−theory of B+ r , and ℜ is the real part. We let πr, Φr, Ar, Rr, Tr, αr, εr and ϕϕϕr denote (πN)T α and (ϕϕϕN)T α respectively, where W(h)θ(a)Ω = θ((ηN)T α (N) and a ∈ A(N)T α . With any representation r of c we associate a state r of Ar in the following way. 0)♮ associated with the 0−dimensional component of Φr, to do Firstly we get the state (Φr this we use Def. 2.2.11 where we construct the state associated with any functional φ on 0)♮ ↾ Br a unital C∗−algebra. Secondly we get the canonical extension ((Φr to the multiplier algebra M(Br), according to the construction provided in Lemma 1.2.2. Finally we compose the extension so obtained with the canonical injection jr of Ar into M(Br), by obtaining a state r of Ar, which is required to be πr−normal by definition. We are forced to use the multiplier algebra because in general Ar could not be injectively mapped into Br. Now the πr−normality is required in order to interpret r as a state obtained by performing an operation on ϕϕϕr. Note that if αr ∈ R+ 0 , then ϕϕϕr is an αr − KMS state w.r.t. the dynamics εr(−α−1 r (·)), therefore r is a state obtained by perturbing a state 0)♮ ↾ Br)− of (Φr 66 of thermal equilibrium at the inverse temperature αr of the physical system evolving in time via the dynamics εr(−α−1 r (·)). By definition NN(c) is the set of the states s by ranging over all representations s of c, so summing up what said we have (2.2.17) 0)♮ ↾ Br)− ◦ jr, r ≔ ((Φr r ∈ Nπr, NN(c) ≔ns s ∈ RepN(c)o . It is worthwhile noting that the presence of cohomology classes in the definition of RepN(c) it is at the basis of the possible degeneration of NN(c). Indeed for any representa- tion r = h T, α, ν, q, KKK, Φ i of c it is so also r = h T, α, ν, q, KKK, Φ i whenever r is πr−normal and the cocycles Φ and Φ on Br belongs to the same cohomology class, i.e. [Φ] = [ Φ], hence NN(c) will be degenerate as soon as r , r. With the characterization of NN(c) in mind, we will propose in the assumption at page 73 the existence of a physical system N and for any T ∈ TTTN and α ∈ PT of physical systems OT αr in the state r implies the previous occurrance of the system N in the phase c. It is exactly in this sense that has to be understood Def. 2.2.31(18) in which we declare that the state r is originated via the phase c, for any r ∈ RepN(c). α , such that for any r ∈ RepN(c) the occurrance of OTr A natural question arises when c is an integer phase of Chern-Connes type, i.e. c(AN) ⊆ Z and there exists a representation r of c such that Φr is the JLO cocycle r , Rr, D, Γ i: is it the state ωe−D2 ◦ πr originated associated with a θ−summable K−cycle h B+ via c in the sense above specified, namely ωe−D2 ◦ πr = r? Lemma 2.2.26 gives the answer in the positive essentially under the hypothesis that Γ is represented via Rr by an element with norm less or equal to 1. This is an important result employed, in the construction in Thm. 2.2.81 of an equivariant stability, in order to prove the equivariance (2.2.40). Definition 2.2.9. Let M ∈ G(G, F, ρ) set c ∈ A∗ M is said to be integer if c(AM) ⊆ Z, moreover define A∗ M ≔ MorAb(AM, R), m M ∈ YQ∈TTTM Morset(PQ M, A∗ M), M (Q, α)(f) ≔ mM(f)(Q, α), m ∀Q ∈ TTTM, α ∈ PQ M, f ∈ AM, and ψM ∗ : H → Autset(A∗ M), l 7→ (ω 7→ ω ◦ ψM(l−1)). 2.2. THE CATEGORY OF NUCLEON SYSTEMS 67 Definition 2.2.10. Define gr ∈ Yf ∈Morset Morset(d( f ), d( f ) × c( f )), f 7→ (x 7→ (x, f (x))) Definition 2.2.11. Let B be a unital C∗−algebra and φ ∈ B∗ − {0}. We call the state associated with φ the state defined as follows φ♮ ≔ φ1 + φ2 kφ1 + φ2k . Here (φ1, φ2) is the unique couple such that (1) φj is a positive functional on B, j ∈ {1, 2}, (2) 1 (3) 1 2(φ + φ∗) = φ1 − φ2, 2k(φ + φ∗)k = kφ1k + kφ2k, where φ∗ ∈ B∗ such that φ∗(a) ≔ φ(a∗), for all a ∈ B. Remark 2.2.12. The existence and uniqueness of the couple (φ1, φ2) in Def. 2.2.11 2(φ + φ∗), follows since [22, Thm. 4.3.6] applied to the hermitian and bounded functional 1 where a functional ψ is hermitian if ψ∗ = ψ. Remark 2.2.13. Under the notations of Def. 2.2.11 we have that kφ1 + φ2k = kφ1k + kφ2k = 1 2k(φ + φ∗)k, the first equality coming since [14, Cor. 2.3.12]. We recall that for any unital C∗−algebra B, h ·, · iB denotes the standard duality between the K0−theory of B and the even entire cyclic cohomology of B. Moreover for any entire normalized even cocycle Φ on B, [Φ] denotes the element in Hev ε (B) corresponding to Φ. Note that Φ0 ∈ B∗, where Φ0 is the 0−dimension component of Φ, hence if Φ0 , 0 then Φ♮ 0 is the state of B associated with Φ0 according to Def. 2.2.11. In what follows we use Def. 1.1.44 moreover we recall that Nπ is the set of π−normal states, for any representation π of a C∗−algebra. Up to the end of section 2.2.2 we let G be a fixed but arbitrary object of G(G, F, ρ). Convention 2.2.14. For any T ∈ TTTG, α ∈ PT G and ν ∈ H(ST α (G)) let α,ν(G) ≔ B(ϕϕϕG)T BT α,ν(G) ν α jT α,ν ≔ j BT A(G)T α ((aG)T α ) , h ·, · i(G,T,α,ν) ≔ h ·, · iBT α,ν(G)+. The above convention generalizes (2.2.11), indeed BT and h ·, · i(G,T,α) = h ·, · i(G,T,α,(µG)T α ). α (G) = BT α,(µG)T α (G), jT α = jT α,(µG)T α Definition 2.2.15 (G−representations of a phase). Let c ∈ A∗ G define RepG(c) the set of the r = h T, α, ν, u, KKK, Φ i such that 68 α (G)), (1) T ∈ TTTG and α ∈ PT G , (2) ν ∈ H(ST (3) u ∈ MorAb(AG, K0(BT (4) KKK = h K, π, Ω i is a cyclic representation of A(G)T (5) Φ is an entire normalized even cocycle on BT (6) (Φ♮ 0 (7) c = vr. α,ν(G))− ◦ jT α,ν(G)+)), α,ν ∈ Nπ, ↾ BT α associated with (ϕϕϕG)T α , α,ν(G)+ such that Φ0 ↾ BT α,ν(G) , 0, Here Φ♮ defined in Def. 1.2.4, while vr ∈ A∗ 0 is the state associated with Φ0 according to Def. 2.2.11, (·)− is the canonical extension G such that for all f ∈ AG vr(f) ≔ ℜh u(f), [Φ] i(G,T,α,ν), (2.2.18) where ℜλ is the real part of λ ∈ C. We call Φ the representative of c relative to r and call any element of RepG(c) a G−representation of c. Remark 2.2.16. (Φ♮ 0 ↾ BT α,ν(G))− ◦ jT α,ν is a state of A(G)T α since Cor. 1.2.5, thus in Def. 2.2.15(6) we require that this state belongs to Nπ. Definition 2.2.17. Let t = h T, α, ν, u, KKK, L, ∆ i such that (1) Def. 2.2.15(1-4) hold, (2) h RRR α ), L, ∆ i is an even θ−summable K−cycle, G such that for all f ∈ AG KKK((aG)T then we set wt ∈ A∗ ν (2.2.19) wt(f) = h u(f), chh RRR ν KKK((aG)T α ), L, ∆ i i(G,T,α,ν). Definition 2.2.18 (C0−representations of an integer phase). Let c ∈ A∗ G be an integer phase, define CG 0 (c) the set of the t = h T, α, ν, u, KKK, L, ∆ i such that ν KKK((aG)T (1) Def. 2.2.15(1-4) hold, (2) h RRR (3) there exists an element b ∈ BT (4) c = wt. α ), L, ∆ i is an even θ−summable K−cycle α,ν(G)+ such that kbk ≤ 1 and RRR KKK((aG)T α )(b) = ∆, ν We call C0−representation of c any element of CG 0 (c). Definition 2.2.19 (C−representations of an integer phase). Let c ∈ A∗ G be an integer phase, define CG(c) the set of the t = h T, α, ν, u, KKK, L, ∆ i such that ν KKK((aG)T (1) Def. 2.2.15(1-4) hold, (2) h RRR (3) Def. 2.2.15(5-6) hold where Φ is the JLO cocycle associated with h RRR (4) c = wt. α ), L, ∆ i is an even θ−summable K−cycle ν KKK((aG)T α ), L, ∆ i, We call C−representation of c any element of CG(c). Definition 2.2.20. Define VVV•(G) the subset of the T ∈ TTTG such that for any α ∈ PT G there exists an element b ∈ BT α (G)+ such that kbk ≤ 1 and RT α (G)(b) = (Γ G)T α . 2.2. THE CATEGORY OF NUCLEON SYSTEMS 69 Definition 2.2.21. For h T, α, (µG)T α , (uG)T α , (HHHG)T α , (DG)T any T ∈ TTTG α , (Γ G)T α i. and α ∈ PT G , define tG(T, α) ≔ Remark 2.2.22. If VVV•(G) , ∅ then t(T, α) ∈ CG 0 (m G (T, α)) for all T ∈ VVV•(G) and α ∈ PT G since (2.2.10). Definition 2.2.23. Let c ∈ A∗ G and r ∈ RepG(c) let us use the following convention r = h Tr, αr, µr, ur, HHHr, Φr i, and set HHHr = h Hr, πr, Ωr i; Ar ≔ A(G)Tr αr , ηr ≔ (ηG)Tr αr , zr ≔ (zG)Tr αr , εr ≔ (εG)Tr αr , ϕϕϕr ≔ (ϕϕϕG)Tr αr , Br ≔ BTr αr,µr(G), jr ≔ jBr , Ar Ψr ≔ (Φr)♮ 0 ↾ Br, h ·, · ir ≔ h ·, · i(G,Tr,αr,µr). If c ∈ A∗ t = h Tt, αt, µt, ut, HHHt, Dt, Γt i, moreover HHHt = h Ht, πt, Ωt i. G is an integer phase and t ∈ CG(c) ∪ CG 0 (c), let us use the following convention Definition 2.2.24 (States originated via a phase). Let c ∈ A∗ G define if in addition c is an integer phase we set NG(c) ≔nΨ− DG(c) ≔nω r ◦ jr r ∈ RepG(c)o , 0 (c)o . ◦ πt t ∈ CG −D2 t e Remark 2.2.25. According to Def. 2.2.15(6), see also Rmk. 2.2.16, for any c ∈ A∗ G and r ◦ jr ∈ Nπr, which is at the ground for the physical interpretation r ∈ RepG(c) we have Ψ− constructed in Def. 2.2.31. If c is an integer phase, the next result shows that CG 0 (c) ⊆ CG(c), one can associate a G−representation of c with any element t in CG(c), and in the CG 0 (c) case the state in N(c) ◦ πt. This result is at associated with this representation assumes the simple form ω −D2 e t the basis of the proof of the equivariance (2.2.40) in Thm. 2.2.81. It is in the proof of the next result that we use Cor. 1.2.5(1). 70 Lemma 2.2.26 (States originated via the same integer phase). Let c ∈ A∗ G be integer, then the following map h T, α, µ, u, HHH, D, Γ i eG (2.2.20) where Φ is the JLO cocycle asssociated to h RRR CG 0 (c) and mapping CG(c) into RepG(c). Moreover for all t ∈ CG (2.2.21) HHH(a(G)T 7→ h T, α, µ, u, HHH, Φ i, 0 (c) Ψ− ◦ πt, µ eG(t) ◦ jeG(t) = ω e −D2 t α ), D, Γ i, is well-defined on CG(c) ∪ in particular (2.2.22) CG 0 (c) ⊆ CG(c) and DG(c) ⊆ NG(c). Remark 2.2.27. In Assump. 2.1 and Def. 2.2.31 G will be interpreted as a physical system, A∗ G as the set of the states of G, called phases, and NG(c) as the set of the states, of suitable physical systems, originated via the phase c for any c ∈ A∗ G. Thus since Lemma 2.2.26 in case c is integer, any C0−representation of c induces a state originated In particular since Rmk. 2.2.22 we have that via c whose form is given in (2.2.21). α )2 ◦ (πG)T (T, α) for all T ∈ VVV•(G) and ω e−((DG)T α ∈ PT G . This fact is at the basis of the proof in Thm. 2.2.81 that the map V•, defined in Def. 2.2.77, satisfies the H equivariance (2.2.40). α is a state originated via the integral phase m G Proposition 2.2.28. Let H be a Hilbert space, Γ a Z2−grading on H, and D a possibly unbounded selfadjoint operator in H. If ΓDΓ = −D and D is positive, then D = 0. Proof. If D is positive then D = D 1 2 D 1 2 , where D 1 2 is a positive, in particular selfadjoint, operator in H, thus ΓDΓ = ΓD 1 2 D 1 2 Γ = (D 1 2 Γ)∗(D 1 2 Γ), 1 where the second equality follows since a general rule, see for example [1, Prp. 1.2.4.(4)], and since Dom(D 2 ) is dense and Γ is unitary. Therefore ΓDΓ is positive, and its spectrum is a subset of R+, while the spectrum of −D is a subset of R−. Hence ΓDΓ = −D implies that the spectrum of −D equals {0} and the (cid:3) statement follows since (0.2.15) and the spectral theorem. 2 ) is dense indeed Dom(D 2 Γ) = ΓDom(D 1 1 −D2 t Proof of Lemma 2.2.26. The first sentence of the statement is trivial, (2.2.22) follows ◦ πt ∈ Nπt and (2.2.21), so let us prove (2.2.21). Let t = h T, α, µ, u, HHH, D, Γ i, 1 the closed unit ball of B+, while α ) which is nondegenerate since HHH it is so. We deduce that since ω e HHH = h H, π, Ω i, ρ = e−D2, Φ0 = Φe(t) S = Rµ α ) and S0 = Rµ α,µ(G), B+ , B = BT 0 HHH((aG)T HHH((aG)T (2.2.23) Φ0 = Tr ◦ LΓρ ◦ S = Tr ◦ Lρ ◦ RΓ ◦ S, the first equality follows since [16, IV.8.δ], the second since the commutativity property of the trace. Next Γ commutes with D2 since ΓDΓ = −D by construction, and since ΓD2Γ = ΓDΓ ΓDΓ, thus Γ ∈ ED2(B(R))′ since [19, Cor. 18.2.4] or [22, 5.6.17], where ED2 2.2. THE CATEGORY OF NUCLEON SYSTEMS 71 is the resolution of the identity of the selfadjoint operator D2 and B(R) is the set of the Borelian subsets of R. Moreover since the construction of the functional calculus and [17, Thm. 4.10.8( f )], or by [22, 5.6.26] or [1, Thm 1.5.5(vi)], the functional calculus of any possibly unbounded normal operator A takes values in the algebra of the operators affiliated to the von Neumann algebra EA(B(R))′′, in particular the operator function belongs to this algebra if it is bouded, therefore (2.2.24) [Γ, ρ] = 0. It is worthwhile noting that since the map composition law of the functional calculus (D2)1/2 = D which equals D if and only if D is positive, case excluded by Prp. 2.2.28 and since the trivial case D = 0 has been excluded from the beginning. Therefore the fact that Γ commutes with e−D2 does not conflict with ΓDΓ = −D. Next Γ and ρ are selfadjoint then Φ0 is hermitian since (2.2.24) since Tr is hermitian and the commutativity preperty of the trace, thus (2.2.25) Φ0 = 1 2 (Φ0 + Φ∗ 0). Next let b ∈ B+ 1.2.11 and (2.2.23) we have 1 such that S(b) = Γ which exists by hypothesis, then since Γ2 = 1, Rmk. (2.2.26) kΦ0k ≥ sup{(Tr ◦ Lρ ◦ RΓ ◦ S)(ab) a ∈ B+ 1 } = sup{(Tr ◦ Lρ ◦ S)(a) a ∈ B+ 1 } = Tr(ρ). Next let P± be the projector associated with the Hilbert subspace H± = {v ∈ H Γv = ±v}, thus P+ + P− = 1 and ΓP j = (−1) jP j, j ∈ {−1, 1}, hence for all a ∈ B+ Tr(ΓρS(a)) = Xj=−1,1 = Xj=−1,1 Tr(cid:16)ΓP jρS(a)(cid:17) (−1) jTr(P jρS(a)). Therefore since (2.2.23) (2.2.27) in particular by Rmk. 1.2.11 Φ0 = Xj=−1,1 (−1) jTr ◦ LP jρ ◦ S, kΦ0k ≤ Xj=−1,1 = Xj=−1,1 kTr ◦ LP jρ ◦ Sk Tr(P jρ) = Tr(ρ), which together (2.2.26) implies 72 (2.2.28) kΦ0k = Xj=−1,1 kTr ◦ LP jρ ◦ Sk = Tr(ρ). So Φ♮ 0 = ωρ ◦ S since (2.2.25,2.2.27,2.2.28) and Rmk. 2.2.13, hence (2.2.29) Φ♮ 0 ↾ B = ωρ ◦ S0, and (2.2.21) follows since Cor. 1.2.5(1). (cid:3) 2.2.3. Physical Interpretation. In Def. 2.2.31 we define the physical interpretation of the data of the category G(G, F, ρ), expecially we physically decode the states originated via a phase described in Def. 2.2.24. What here introduced will be emploied in the next section to describe the physical invariances of a nucleon-fragment doublet. Remark 2.2.29. If φ is a β−KMS state for some one-parameter dynamics τ with β > 0, i.e. a state of thermal equilibrium at the inverse temperature β of the system whose dynamics is τ, then φ−normal states usually are interpreted as states obtained by performing on φ small perturbations or operations. Thus, according to Def. 2.2.15(6), (2.2.4) and [15, p. 77], for any c ∈ A∗ and r ∈ RepG(c) such that αr ∈ R+ 0 , it follows that Ψ− r ◦ jr is a state generated by an operation performed on the state of thermal equilibrium ϕϕϕr at the inverse temperature αr of the system whose dynamics is εr(−α−1 r (·)). Remark 2.2.30 (Noncommutative geometric nature of the degeneration of NG(c)). Let c ∈ A∗ and χ ∈ N(c), so there exists r ∈ RepG(c) such that (2.2.30) χ = Ψ− r ◦ jr. Next Ψ− r is the canonical extension of Ψr to M(Br), where Ψr is the restriction to Br of the state associated to the 0−dimensional component of the entire normalized cyclic even cocycle Φr on B+ r such that c = vr i.e. (2.2.31) Ψ− r ◦ iBr = Ψr, Ψr = (Φr c(f) = ℜh ur(f), [Φr] ir ∀f ∈ A. 0)♮ ↾ Br, (2.2.30) and (2.2.31) justify Assumtion 2.1(4b) where we propose to consider any element in N(c) as a state whose occurrence signales the occurrence of the phase c. If we get an entire normalized even cocyle Φ on Br such that ( χ (cid:17) (( Φ0)♮ ↾ Br)− ◦ jr ∈ Nπr, [Φr] = [ Φ], then r (cid:17) h T, α, µ, u, HHH, Φ i ∈ N(c). Therefore χ , χ implies that N(c) is degenerate with degeneration of noncommutative geometric nature. 2.2. THE CATEGORY OF NUCLEON SYSTEMS 73 Next we think about the elements of NG(c) as those states of suitable systems O's, signaling the occurrance of the phase c of G, namely such that whenever one of the systems O's occurs in one of these states, then the physical system G previously occurred in the phase c. According to Rmk. 2.2.30 the degeneration of the set of the states originated via the same phase can be of noncommutative geometric nature. We can consider G as the group of spatiotemporal translations and F the group containing the gauge group F0 and the remaining symmetries of the system, in such a case ρ restricted to F0 needs to be trivial. Assumption 2.1. For any N ∈ Obj(G(G, F, ρ)) there exists a unique physical system still denoted by N such that (1) AN is the set of observables of N; (2) A∗ (3) for any T ∈ TTTN and α ∈ PT N is the set of states, said phases of N, N there exists a physical system O(N)T α such that (a) A(N)T α and EA(N)T α are the algebra of observables and the set of states of O(N)T α respectively. If α ∈ R+ 0 then O(N)T (b) T acts as a special type of operation on EAT α intermediation of phases of N; α evolves in time through (εN)T α (−α−1(·)), producing states of O(N)T α via the (4) for all c ∈ A∗ N and r ∈ RepN(c) the following properties hold (a) if the operation Tr is performed on the system O(N)Tr ϕϕϕr, then O(N)Tr αr will occur in the state Ψ− r ◦ jr, αr when occurring in the state (b) if O(N)Tr αr occurs in the state Ψ− r ◦ jr then N occurred in the phase c; (5) the information about the set of operations TTTN are sufficient to provide the set of observ- ables AN with the structure of a group but not an algebra; (6) for any l ∈ H, f ∈ AN, T ∈ TTTN and a ∈ A(N)T α (a) ψN(l)(f) is the observable obtained by transforming f through l, (b) bN(l)(T) is the operation obtained by transforming T through l, (c) V(N)T α (l)(a) is the observable of the system O(N)bN(l)(T) α a through l, , obtained by transforming (7) for all M ∈ Obj(G(G, F, ρ)), (g, d) ∈ MorG(G,F,ρ)(N, M), f ∈ AN and Q ∈ TTTM (a) g(f) is the observable of the system M obtained by transforming f through g, (b) d(Q) is the operation obtained by tranforming Q through d, Now we fix those properties of any physical interpretation satisfying Assumption 2.1 and Rmk. 2.2.29. In what follows let ≡ denote equivalence of propositions. Definition 2.2.31. We call interpretation any couple of maps (s, u) such that if G, M ∈ G , T ∈ G such that α such a = a∗, C is a category, a ∈ C and F is a Obj(G(G, F, ρ)), (g, f) ∈ MorG(G,F,ρ)(G, M), P ∈ TTTM, β ∈ PP β ), b ∈ A(G)f(P) MorCA∗(A(G)f(P) β RepG(c) , ∅, 4 r ∈ RepG(c), χ ∈ EA(G)Q functor from C to G(G, F, ρ), then such that b = b∗, l ∈ H, f ∈ AG, c ∈ A∗ M, Q ∈ TTTG, α ∈ PQ , A(M)P , a ∈ AQ β α 4in particular c = m G (Q, α) 74 α ), α ) u(a), α (l)) ≡ l, (1) s(G) ≡ the nucleon system u(G), (2) u(F(a)) ≡ generated by the fissioning system u(a), (3) s(O(G)Q (4) s(O(G)Q is (εG)Q α ) ≡ the fragment system whose observable algebra is A(G)Q α ; α ) ≡ the fragment system whose observable algebra is A(G)Q α and whose dynamics α (−α−1(·)), if α ∈ R+ 0 , (5) s(χ) ≡ the state of s(O(G)Q α ) u(χ), (6) s(χ) ≡ the state u(χ) of s(O(G)Q α ), (7) s(a) ≡ the observable u(a) of s(O(G)Q (8) s(a) ≡ the observable of s(O(G)Q (9) u(V(G)Q (10) s(Q) ≡ the operation u(Q), (11) u(bG(l)) ≡ l; (12) uM(Tb) ≡ obtained by transforming through the action of u(T) s(b), (13) u(f(P)) ≡ obtained by transforming through the action of u(f) s(P), (14) χ(a) equals the mean value in s(χ) of s(a), (15) u((ϕϕϕG)Q α ) ≡ of thermal equilibrium (ϕϕϕG)Q (16) s(c) ≡ the phase of s(G) occurring by performing s(Tr) on s(ϕϕϕr); (17) s(c) ≡ the phase of s(G) occurring by performing on s(ϕϕϕr) s(Tr), (18) u(Ψ− r ◦ jr) ≡ originated via s(c), (19) s(f) ≡ the observable u(f) of s(G), (20) u(ψG(l)) ≡ l, (21) uM(g(f)) ≡ obtained by transforming through the action of u(g) s(f), (22) c(f) equals the mean value in s(c) of s(f). α at the inverse temperature α, if α ∈ R+ 0 , In order to avoid redundancies, often we convein to remove the expression "of s(OT α)" and "of s(G)". According to Assumption 2.1(4b), Def. 2.2.31(18) has to be understood as "whose occurrence follows the occurrence of s(c)", moreover Def. 2.2.31(16) follows since Rmk. 2.2.29 and Assumption 2.1(4b). The introduction of the map u permits to specify the semantics in actual models, we shall use it in addressing in part 3 the nuclear binary fission phenomenon. This is why we use here in an abstract meaning the concepts of fragment and nucleon systems. In the remaining of the work let (s, u) be a fixed interpretation. Remark 2.2.32. Since Rmk. 2.2.8 we obtain (1) u(V(G)Q (2) u(bG(l)Q) ≡ obtained by transforming through the action of l s(Q), (3) u(ψG(l)f) ≡ obtained by transforming through the action of l s(f). α (l)(a)) ≡ obtained by transforming through the action of l s(a), Let G be an object of G(G, F, ρ) and c ∈ A∗ G, then the next two propositions easily follow by Def. 2.2.31. Proposition 2.2.33 (Thermal nature of the nucleon phases and their stability under variation of the operations.). s(c) ≡ the phase of the nucleon system u(G) occurring by performing the operation u(Tp) on the state of thermal equilibrium ϕϕϕp at the inverse 2.2. THE CATEGORY OF NUCLEON SYSTEMS 75 temperature αp of the fragment system whose observable algebra is Ap and whose dynamics is εp(−α−1 p (·)), for all p ∈ RepG(c) such that αp > 0. Proposition 2.2.34 (Noncommutative geometric and thermal origination of fragment states via a nucleon phase.). Ψ− r ◦ jr is the ϕϕϕr−normal state originated via s(c), of the fragment system whose observable algebra is Ar and whose dynamics is εr(−α−1 r (·)), for all r ∈ RepG(c) such that αr > 0. 2.2.4. Nucleon-fragment doublets on C. We define the auxiliary concepts of equi- variant stability and extended C−equivariant stability for a category C, in Def. 2.2.52 and Def. 2.2.62. Nucleon-fragment doublets are essential in order to solve the universality claim described in introduction 0.1, as we shall establish in Cor. 2.3.56. In Def. 2.2.66 we define the structure of nucleon-fragment doublet T on an arbitrary category, in Def. 2.2.71 we define the T−nucleon phase and T−fragment state and their expanded equiv- ariances which represent the T−resolution of the equivariant form of the universality claim. We describe the properties of equivariance of T in Prp. 2.2.70, the consequent properties of invariance and their physical interpretation in Prp. 2.2.72 and Prp. 2.2.76 with the help of Prp. 2.2.73. In Cor. 2.2.68 we relate a nucleon-fragment doublet on C to any extended C−equivariant stability. The first step is to define equivariant phases in Def. 2.2.43 let us start with preparatory definitions. Definition 2.2.35. Let UUU be defined pre (G, F, ρ)−map if it is a map on Dom(UUU) ⊆ Obj(G(G, F, ρ)) such that UUUN ⊆ TTTN, for all N ∈ Dom(UUU). Define UUU to be a (G, F, ρ)−map if it is a pre (G, F, ρ)−map such that (∀M ∈ Dom(UUU))(∀(h, f) ∈ MorG(G,F,ρ)(N, M))(f(UUUM) ⊆ UUUN). (2.2.32) Let h H, UUU i be defined a pre (G, F, ρ)−couple of maps if UUU is a pre (G, F, ρ)−map and H is a map on Dom(UUU) such that HN is a subgroup of H, for all N ∈ Dom(UUU). H is said to be full if it is the constant map with constant value equal to H. Let h H, UUU i be defined a (G, F, ρ)−couple of maps if it is a pre (G, F, ρ)−couple of maps such that for all N ∈ Dom(UUU) and l ∈ HN (2.2.33) bN(l)(UUUN) ⊆ UUUN. If UUU is a (G, F, ρ)−map, then h H, UUU i is a (G, F, ρ)−couple of maps with H full, since Rmk. 2.2.8. Definition 2.2.36. Let h H, DDD, UUU i be defined a (G, F, ρ)−triplet of maps if UUU is a (G, F, ρ)−map and h H, UUU i is a pre (G, F, ρ)−couple of maps, DDD is a map such that Dom(DDD) ⊆ Dom(UUU), DDDN ⊆ UUUN, for all N ∈ Dom(DDD), and h H ↾ Dom(DDD), DDD i is a (G, F, ρ)−couple of maps. If h H, UUU i is a pre (G, F, ρ)−couple of maps such that UUU is a (G, F, ρ)−map, then h H, UUU, TTT i is a (G, F, ρ)−triplet of maps. Definition 2.2.37. Let DDD be a pre (G, F, ρ)−map, define Dom(DDD)0 to be the unique subcate- gory of G(G, F, ρ) such that Obj(Dom(DDD)0) = Dom(DDD), and MorDom(DDD)0(N, M) = {(h, f) ∈ MorG(G,F,ρ)(N, M) f(DDDM) ⊆ DDDN}, ∀N, M ∈ Dom(DDD). 76 Remark 2.2.38. Let DDD be a pre (G, F, ρ)−map, then (DDD, Pr 2 ) ∈ Fct((Dom(DDD)0)op, set), where Pr2 is the map (g, f) 7→ f defined on MorG(G,F,ρ). If h H, DDD i is a (G, F, ρ)−couple of maps, N ∈ Dom(DDD) and l ∈ HN then since Rmk. 2.2.8 (ψN(l), bN(l−1)) ∈ AutDom(DDD)0(N). If UUU is a pre (G, F, ρ)−map then Dom(UUU)0 is a subcategory of Dom(UUU) the full subcategory of G(G, F, ρ) whose object set equals Dom(UUU); however if UUU is a (G, F, ρ)−map then Dom(UUU)0 = Dom(UUU) as categories. Definition 2.2.39. Let R be an object of G(G, F, ρ) and Q ∈ TTTR, set X(R, Q) ≔ Qβ∈PQ R(cid:16)A(R)Q β(cid:17)∗ . Definition 2.2.40. Let ∆o, Z and V be maps on Obj(G(G, F, ρ)) while ∆m be the map on MorOb j(G(G,F,ρ)) such that for all M, N ∈ Obj(G(G, F, ρ)) and (h, f) ∈ MorG(G,F,ρ)(N, M), we have ∆o(M) ≔ [Q∈TTTM Morset(PQ M, A∗ M), while and ∆m(h, f) : ∆o(M) → ∆o(N), Morset(PI M, A∗ and V(M) : H → Autset(Z(M)) such that Z(M) ≔ aQ∈TTTM V(M)(l) : (T, f ) 7→(cid:16)bM(l)(T), PbM(l)(T) M M) ∋ f 7→(cid:16)Pf(I) N ∋ α 7→ f (α) ◦ h(cid:17) , ∀I ∈ TTTM, X(M, Q), ∋ α 7→ f (α) ◦ V(M)bM(l)(T) α (l−1)(cid:17) . Definition 2.2.41. Let h H, DDD i be a pre (G, F, ρ)−couple of maps. Define ZH and OH to be maps on Dom(DDD) such that for all N ∈ Dom(DDD) ZN H ON H ≔ (1 7→ Z(N), V(N) ↾ HN), ≔(cid:16)1 7→ ∆o(N), HN ∋ l 7→ ∆m(ψN(l−1), bN(l))(cid:17) . If in addition h H, DDD i is a (G, F, ρ)−couple of maps let us define PDDD,H to be the map on Dom(DDD) such that for all O ∈ Dom(DDD) PO ≔ (1O 7→ DDDO, HO ∋ l 7→ bO(l) ↾ DDDO). DDD,H m) where m is the map on MorDom(DDD)0 such that for all N, M ∈ Dom(DDD) and (h, f) ∈ MorDom(DDD)0(N, M) If H is full we let Z, O and PDDD denote ZH, OH and PDDD,H respectively. Let QDDD ≔ (DDD, QDDD QDDD we have QDDD m((h, f)) ≔ f ↾ DDDM. Finally define ∆DDD ≔ (∆o ↾ Dom(DDD), ∆m ↾ MorDom(DDD)). 2.2. THE CATEGORY OF NUCLEON SYSTEMS 77 Remark 2.2.42. Let h H, DDD, UUU i be a (G, F, ρ)−triplet of maps. Thus ∆UUU ∈ Fct(Dom(UUU)op, set) and QDDD ∈ Fct((Dom(DDD)0)op, set), QUUU ∈ Fct(Dom(UUU)op, set), while PN ∈ Fct(H, set), in particular PN H ∈ Fct(HO, set), for UUU any N ∈ Dom(UUU) and O ∈ Dom(DDD). UUU,H ∈ Fct(HN, set), and PO DDD,H, ZO H, OO Next we define an equivariant phase as a section of maps valued in nucleon phase valued maps contravariant under action of Dom(UUU), and as a result the value of this section at any N induces a covariant section under action of HN. Definition 2.2.43 (Equivariant phases). Let h UUU, m i be defined equivariant phase if (1) UUU is a (G, F, ρ)−map; (2) m ∈QN∈Dom(UUU)QQ∈UUUN Morset(PQ (3) m ∈ MorFct(Dom(UUU)op,set)(QUUU, ∆UUU). N, A∗ N); m is called integer if mN(T, α) is integer for all N ∈ Dom(UUU), T ∈ UUUN and α ∈ PT N. Remark 2.2.44. Let h UUU, m i be an equivariant phase then Def. 2.2.43(3) is equivalent to state that for all M, N ∈ Dom(UUU) and (h, f) ∈ MorG(G,F,ρ)(N, M) the following diagram is commutative (2.2.34) UUUN mN / ∆o(N) f↾UUUM ∆m(h,f) UUUM / ∆o(M). mM Therefore for all N ∈ Dom(UUU) and l ∈ H the following diagram is commutative since Rmk. 2.2.8 (2.2.35) UUUN mN / ∆o(N) bN(l)↾UUUN ∆m(ψN(l−1),bN(l)) UUUN / ∆o(N), mN i.e. (1 7→ mN) ∈ MorFct(H,set)(PN UUU, ON), ∀N ∈ Dom(UUU). Definition 2.2.45 (Field determined by an equivariant phase). Let h UUU, m i be an equi- variant phase, we call field determined by m the map S defined on Dom(UUU) such that for all N ∈ Dom(UUU) SN ≔ {mN(Q, α)Q ∈ UUUN, α ∈ PQ N}. / O O / O O / O O / O O 78 Proposition 2.2.46 (Functoriality of S). Let h UUU, m i be an equivariant phase, and S N we be the field determined by m, then for all N ∈ Dom(UUU), l ∈ H, Q ∈ UUUN and α ∈ PQ have ψN ∗ (l)(mN(Q, α)) = mN(bN(l)Q, α). In particular we obtain (2.2.36) ψN ∗ (l)SN = SN, ∀l ∈ H. Moreover for all M ∈ Dom(UUU), (h, f) ∈ MorG(G,F,ρ)(N, M), T ∈ UUUM and β ∈ PT h†(mM(T, β)) = mN(f(T), β). In particular we obtain h†SM ⊆ SN hence M we have (2.2.37) (S, (h, f) 7→ h†) ∈ Fct(Dom(UUU)op, set). Proof. Since Rmk. 2.2.44. (cid:3) Next we define sections of maps valued in maps whose values are fragment states originated via the phases determined by an equivariant phase. The value of this section at any N induces a covariant section under action of HN. Notice that we do not require the contravariance with respect to the action of some subcategory of G(G, F, ρ). A variant of this request is postponed when introducing extended C−equivariant stabilities. Definition 2.2.47 (Equivariant states associated with an equivariant phase). W is a state associated with h H, UUU, m i and equivariant on DDD if h UUU, m i is an equivariant phase, h H, DDD, UUU i is a (G, F, ρ)−triplet of maps and there exist o such that W ∈ YN∈Dom(DDD) YQ∈DDDN Yβ∈PQ o ∈ YN∈Dom(DDD) YQ∈DDDN Yβ∈PQ N N NN(mN(Q, β)), RepN(mN(Q, β)), ◦ joN(Q,β), oN(Q,β) WN(Q, β) = Ψ− ToN(Q,β) = Q, αoN(Q,β) = β, ∀N ∈ Dom(DDD), Q ∈ DDDN, β ∈ PQ N; (1 7→ gr(WN)) ∈ MorFct(HN,set)(PN DDD,H, ZN H), ∀N ∈ Dom(DDD). (2.2.38) moreover (2.2.39) We call W a state associated with h UUU, m i equivariant on DDD if it is a state associated with h H, UUU, m i equivariant on DDD where H is full. Remark 2.2.48. Notice that W ∈ YN∈Dom(DDD) YQ∈DDDN Yβ∈PQ N NA(N)Q β , in particular gr(WN) is a section of the bundle h Z(N), DDDN, Pr1 i. 2.2. THE CATEGORY OF NUCLEON SYSTEMS 79 Remark 2.2.49. (2.2.39) is equivalent to say that for all l ∈ HN the following diagram is commutative (2.2.40) DDDN gr(WN) / Z(N) bN(l)↾DDDN V(N)(l) DDDN / Z(N). gr(WN) Definition 2.2.50 (Field determined by an equivariant state). Let W be a state associated to h H, UUU, m i and equivariant on DDD, we call field determined by W the map J on Dom(DDD) such that for all N ∈ Dom(DDD) JN ≔ gr(WN)(DDDN). Proposition 2.2.51 (Covariance of the field determined by an equivariant state). Let W be a state associated with h H, UUU, m i and equivariant on DDD, and let J be the field determined by W, then for all N ∈ Dom(DDD) and l ∈ HN, V(N)(l)JN = JN, i.e. JN is the space of the representation of HN via the action V(N). In particular if we assume that there exists a von Neumann algebra X(N) and a map X(N) : H → α , for all Q ∈ DDDN and α ∈ PQ AutCA∗(X(N)) such that X(N) = A(N)Q N, : HN → Autset(X(N)∗) such that l 7→ (ω 7→ ω ◦ XN(l−1)) and the map W on and define XN ∗ Dom(DDD) such that α , and X(N) = V(N)Q WN ≔ {WN(Q, α)Q ∈ DDDN, α ∈ PQ N}, then (2.2.41) I.e. WN is the space of a representation of HN via the dual action of XN. ∗ (h)WN = WN, ∀h ∈ HN. XN Proof. Since (2.2.40). (cid:3) Definition 2.2.52 (Equivariant stabilities). h H, UUU, m, W i is an equivariant stability on DDD if W is a state associated with h H, UUU, m i and equivariant on DDD. Moreover h UUU, m, W i is a full equivariant stability on DDD if h H, UUU, m, W i is an equivariant stability on DDD and H is full, while it is integer if m it is so. In order to introduce extended C−equivariant stabilities we need some preparatory definitions. As remarked in section 0.2, for any nonempty subset S of the object set of a category A we let S denote also the full subcategory of A whose object set is S. The general concept of enrichment is introduced for example in [24, Def. 1.3.2]. For what concerns the following definition we need a particular case, and of this case only to know that if A is enriched over B, then MorA(x, y) is an object of B for all x, y objects of A. / O O / O O 80 Definition 2.2.53. Let K and D be categories, F ∈ Fct(K, D), C be an object of D and D be enriched over itself in a way that there exists the map (·)F,C † ∈ YT∈MorK MorD(MorD(Fo(c(T)), C), MorD(Fo(d(T)), C)), such that for any object X, Y of K and T ∈ MorK(X, Y) we have (2.2.42) (TF,C † TF,C † : MorD(Fo(Y), C) → MorD(Fo(X), C), (g) ≔ g ◦ Fm(T), ∀g ∈ MorD(Fo(Y), C). Remark 2.2.54. Since F is a functor we deduce that (cid:16)X 7→ MorD(Fo(X), C), T 7→ TF,C † (cid:17) ∈ Fct(Kop, D). We shall use this notation mainly in the following cases, (1) K = CA∗, D = BS, F = ICA∗→BS and C = C; (2) K = D = Ab, F = IdAb and C = R. Thus consistently with the operation (·)† used in part 1, we let T† denote TF,C † only in these two cases and whenever it is clear by the context which category K is involved. Definition 2.2.55. Let K be a category, E a subcategory of G(G, F, ρ) and L ∈ Fct(K, E). For i ∈ {1, 2} let Li = Pri ◦Lm where Pri is the function defined on MorG(G,F,ρ) projecting over the i−th component. Thus L1 and L2 are maps uniquely determined by (2.2.43) (A ◦ Lo, L1) ∈ Fct(K, Ab), (TTT ◦ Lo, L2) ∈ Fct(Kop, set), Lm(t) = (L1(t), L2(t)), ∀t ∈ MorK. Ab is enriched over itself, since MorAb(x, y) ∈ Ab via the pointwise composition, inversion and identity for any x, y ∈ Ab, moreover (L1)† ∈ Yt∈MorK MorAb(A∗ L(c(t)), A∗ L(d(t))), Remark 2.2.56. Let K be a category, E a subcategory of G(G, F, ρ) and L ∈ Fct(K, E), t 7→ (L1(t))†. then the two inclusions in (2.2.43) mean the following L1 ∈ Yt∈MorK L2 ∈ Yt∈MorK MorAb(AL(d(t)), AL(c(t))), Morset(TTTL(c(t)), TTTL(d(t))), and for all x, y ∈ MorK such that d(x) = c(y) L1(x ◦ y) = L1(x) ◦ L1(y), L2(x ◦ y) = L2(y) ◦ L2(x). 2.2. THE CATEGORY OF NUCLEON SYSTEMS 81 Moreover (A, Pr1) ∈ Fct(G(G, F, ρ), Ab), so (A, Pr1) ◦ IE→G(G,F,ρ) ◦ L ∈ Fct(K, Ab), let us denote it by U thus U = (A ◦ Lo, L1), (L1)† = (·)U,R † , (A∗ ◦ Lo, (L1)†) ∈ Fct(Kop, Ab). Definition 2.2.57. Let UUU be a map such that Dom(UUU) ⊆ Obj(G(G, F, ρ)), C a category and F a functor from C to G(G, F, ρ). Define Θ(UUU, F) ≔ F−1(Dom(UUU)) and FUUU ≔ F ◦ IΘ(UUU,F)→C. Set Fo ≔ (FUUU)m. Dom(UUU) is a subcategory of G(G, F, ρ), so Θ(UUU, F) is a UUU well-defined subcategory of C, and FUUU is a functor from Θ(UUU, F) to Dom(UUU) since the convention established in section 0.2. Thus according to Def. 2.2.55 we can define the map F† ≔ ((FUUU)1)†. UUU ≔ (FUUU)o, and Fm UUU Note that by (0.2.21) we have also that FUUU ∈ Fct(Θ(UUU, F)op, Dom(UUU)op). Definition 2.2.58. Let DDD be a pre (G, F, ρ) map, C be a category and F a functor from C to G(G, F, ρ). Define Ξ(DDD, F) ≔ F−1(Dom(DDD)0), FDDD ≔ F ◦ IΞ(DDD,F)→C, and set FDDD o ≔ (FDDD)o, and FDDD m ≔ (FDDD)m. According to the convention in section 0.2 we consider FDDD as a functor from Ξ(DDD, F) to Dom(DDD)0, hence by (0.2.21) we have also that FDDD ∈ Fct(Ξ(DDD, F)op, (Dom(DDD)0)op). Note that if UUU is a pre (G, F, ρ)−map, then Ξ(UUU, F) is a subcategory of Θ(UUU, F), however if UUU is a (G, F, ρ)−map then Ξ(UUU, F) = Θ(UUU, F). In the next definition we use Z introduced in Def. 2.2.40, Definition 2.2.59. Let K be a category, E a subcategory of G(G, F, ρ), E ∈ Fct(K, E) and Z ∈ Fct(Kop, set) such that Zo = Z ◦ Eo. Define Z1 and Z2 maps on MorK such that Morset(TTTE(c(t)), TTTE(d(t))), Z1 ∈ Yt∈MorK Z2 ∈ Yt∈MorK YQ∈TTTE(c(t)) Morset(cid:16)X(E(c(t)), Q), X(E(d(t)), Z1(t)Q)(cid:17); and for all t ∈ MorK and (Q, f ) ∈ Z(E(c(t))) Z(t)(Q, f ) = (Z1(t)Q, Z2(t, Q) f ). The concept below introduced is fundamental to integrate in the definition of extended C−equivariant stabilities the missing contravariance of W under action of Dom(DDD)0 in a way that this action is induced by the conjugate of an action over observ- ables. 82 Definition 2.2.60. Let H be a category, E a subcategory of G(G, F, ρ) and E ∈ Fct(H, E). Let h Z, S i be defined a conjugate action via E if (1) Z ∈ Fct(Hop, set); (2) Zo = Z ◦ Eo; (3) Z1 = E2; (4) S ∈ Yt∈MorH YQ∈TTTE(c(t)) Yβ∈PQ E(c(t)) MorCA∗(cid:16)A(E(d(t)))E2(t)Q β , A(E(c(t)))Q β(cid:17) ; (5) let t ∈ MorH, Q ∈ TTTE(c(t)) and α ∈ PQ E(c(t)), thus for any f ∈ X(E(c(t)), Q) and s ∈ MorH such that d(t) = c(s) the following diagram is commutative A(E(c(t)))Q α f (α) / C S(t◦s,Q,α) 8♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣ S(s,E2(t)Q,α) S(t,Q,α) ;✈✈✈✈✈✈✈✈✈✈✈✈✈✈✈✈✈✈✈✈ (Z2(t,Q) f )(α) A(E(d(s)))E2(t◦s)Q α A(E(d(t)))E2(t)Q . α In particular (Z2(t, Q) f )(α) = S(t, Q, α)†( f (α)), moreover the left triangle in the 2.2.60(5) is well-set indeed d(s) = d(t ◦ s), c(t) = c(t ◦ s) and diagram in Def. E2(t ◦ s) = E2(s) ◦ E2(t) since (2.2.43). Definition 2.2.61 (C−equivariant is a C−equivariant stability on DDD if h H, UUU, m, W i is an equivariant stability on DDD, C is a cate- gory and F ∈ Fct(C, G(G, F, ρ)). h h UUU, m, W i, F i is a full C−equivariant stability on DDD if K is a C−equivariant stability on DDD and H is full. stabilities). K = h h H, UUU, m, W i, F i Now we are able to give the following Definition 2.2.62 (Extended C−equivariant stabilities). h h H, UUU, m, W i, F i is an ex- tended C−equivariant stability on DDD via Z and S if (1) h h H, UUU, m, W i, F i is a C−equivariant stability on DDD; (2) h Z, S i is a conjugate action via FDDD; (3) gr ◦ W ◦ FDDD o ∈ MorFct(Ξ(DDD,F)op,set)(QDDD ◦ FDDD, Z). Notice that Def. 2.2.62(3) integrates the missing symmetry of W w.r.t. the contravari- ant action of the category Dom(DDD)0, by displaying instead the contravariance of the section gr ◦ W ◦ FDDD o with respect to the action of the inverse image Ξ(DDD, F) of Dom(DDD)0 via the functor F. We shall see its meaning in the more general context of nucleon-fragment doublets in Prp. 2.2.70 and Prp. 2.2.72 and its physical interpretation in Prp. 2.2.76. / / / 8 O O ; 2.2. THE CATEGORY OF NUCLEON SYSTEMS 83 Remark 2.2.63. Let h h H, UUU, m, W i, F i be an extended C−equivariant stability on DDD via Z and S, then Def. 2.2.62(3) is equivalent to state that for all d, e ∈ Ξ(DDD, F) and t ∈ MorΞ(DDD,F)(e, d) the following is a commutative diagram DDDF(e) gr(WF(e)) / Z(F(e)) F2(t)↾DDDF(d) Z(t) DDDF(d) / Z(F(d)). gr(WF(d)) Next we state the symmetry properties corresponding to the data of an extended C−equivariant stability. = m ∗ 1FUUU and Proposition 2.2.64 (Properties of equivariance related to a C−equivariant stability). Let C be a category and E = h h H, UUU, m, W i, F i be a C−equivariant stability on DDD. Then m ◦ Fo UUU (1) m ◦ Fo UUU (2) (1 7→ mF(a)) ∈ MorFct(HF(a),set)(PF(a) (3) (1 7→ gr(WF(b))) ∈ MorFct(HF(b),set)(PF(b) If S denotes the field determined by m, then ∈ MorFct(Θ(UUU,F)op,set)(QUUU ◦ FUUU, ∆UUU ◦ FUUU), H ), for all a ∈ Θ(UUU, F), H ), for all b ∈ Ξ(DDD, F). UUU , OF(a) DDD,H , ZF(b) (2.2.44) (S ◦ Fo ψF(a) ∗ UUU, F† UUU) ∈ Fct(Θ(UUU, F)op, set), (h)SF(a) = SF(a), ∀a ∈ Θ(UUU, F), ∀h ∈ HF(a). If E is an extended C−equivariant stability on DDD via Z and S and J denotes the field determined by W, then (2.2.45) (J ◦ FDDD o , Zm) ∈ Fct(Ξ(DDD, F)op, set), V(F(b))(l)JF(b) = JF(b), ∀b ∈ Ξ(DDD, F), ∀l ∈ HF(b). Proof. The first sentence follows since (0.2.19), then follows st. (2,3) are trivial. (2.2.44) follows since Prp. 2.2.46, while (2.2.45) follows since Rmk. 2.2.63 and (cid:3) Prp. 2.2.51. (1); sts. Remark 2.2.65. The inclusion in (2.2.45) means for all a, b ∈ Ξ(DDD, F) and t ∈ MorΞ(DDD,F)(b, a) that Z(t)JF(a) ⊆ JF(b). Part of the requests in the definition of the category G(G, F, ρ) has been introduced to develop the semantics, i.e. the physical interpretation in section 2.2.3, while the remaining part to ensure the existence of an equivariant stability as we shall see in section 2.2.5. This because, in order to fulfill our aim to resolve the universality claim, we are interested to the symmetry properties related to the category C stated in Prp. / O O / O O 84 2.2.64 and Def. 2.2.62(3), but isolated from all what does not affect the physical meaning according to the semantics. Prp. 2.2.64 itself and Def. 2.2.62 exhibit the clues to extract the physically relevant information of an extended C−equivariant stability, in a context where the use of the functor F reduces at minimum, i.e. where it suffices and operates only to apply the semantics. Henceforth we are in the position to introduce the following definition where each item is titled with the description of the corresponding content. Definition 2.2.66 (Nucleon-fragment doublets). h S, J, Z, S, L, m, W, R, DDD, UUU i is a nucleon-fragment doublet on C if (1) C is a category; (2) h R, DDD, UUU i is a (G, F, ρ)−triplet of maps; (3) L ∈ Fct(C, G(G, F, ρ)); (4) S is a field of nucleon phases contravariant under action of Θ(UUU, L) via the conjugate of L1. S ∈ Fct(Θ(UUU, L)op, set), Sa ⊆ A∗ L(a), ∀a ∈ Θ(UUU, L), S(t) = (L1(t))† ↾ Sc(t), ∀t ∈ MorΘ(UUU,L); (5) Each fiber of S is covariant under action of H via the dual of ψ ◦ Lo. ψL(a) ∗ (h)Sa = Sa, ∀a ∈ Θ(UUU, L), ∀h ∈ RL(a); (6) m is a section of maps valued in S−valued maps, contravariant under action of Θ(UUU, L) via Sm and L2, whose values induce covariant sections under action of H via the dual of ψ ◦ Lo and via b ◦ Lo. m ∈ Ya∈Θ(UUU,L) YQ∈UUUL(a) Morset(PQ L(a), Sa), such that (2.2.46) (2.2.47) and m ∈ MorFct(Θ(UUU,L)op,set)(QUUU ◦ LUUU, ∆UUU ◦ LUUU), (1 7→ ma) ∈ MorFct(RL(a),set)(PL(a) UUU , OL(a) R ), ∀a ∈ Θ(UUU, L). (7) J is a field of disjoint union over operations of set of maps - with values fragment states originated via the nucleon phases determined by m - contravariant under action of Ξ(DDD, L) via Jm where J1 = L2 and J2 is induced by the conjugate of the evaluation of S. (2.2.48) J ∈ Fct(Ξ(DDD, L)op, set), and h Z, S i is a conjugate action via LDDD such that (a) Jb ⊂ Z(b), for all b ∈ Ξ(DDD, L), 2.2. THE CATEGORY OF NUCLEON SYSTEMS 85 (b) J(t) = Z(t) ↾ Jc(t) for all t ∈ MorΞ(DDD,L); and Jb ⊂ aQ∈DDDL(b) Yβ∈PQ L(b) (2.2.49) NL(b)(mb(Q, β)), ∀b ∈ Ξ(DDD, L). (8) Each fiber of J is covariant under action of H via a map induced by the dual of V ◦ Lo. V(L(b))(l)Jb = Jb, ∀b ∈ Ξ(DDD, L), ∀l ∈ RL(b); (9) W is a map such that gr ◦ W is a section of J, contravariant under action of Ξ(DDD, L) via Jm and L2, whose values induce sections covariant under action of H via the dual of V ◦ Lo and via b ◦ Lo. There exists o such that W ∈ Yb∈Ξ(DDD,L) YQ∈DDDL(b) Yβ∈PQ o ∈ Yb∈Ξ(DDD,L) YQ∈DDDL(b) Yβ∈PQ L(b) L(b) NL(b)(mb(Q, β)), RepL(b)(mb(Q, β)), (2.2.50) (2.2.51) (2.2.52) and ◦ job(Q,β), ob(Q,β) Wb(Q, β) = Ψ− Tob(Q,β) = Q, αob(Q,β) = β, ∀b ∈ Ξ(DDD, L), Q ∈ DDDL(b), β ∈ PQ L(b); gr ◦ W ∈ MorFct(Ξ(DDD,L)op,set)(QDDD ◦ LDDD, J) (1 7→ gr(Wb)) ∈ MorFct(RL(b),set)(PL(b) DDD,R , ZL(b) R ), ∀b ∈ Ξ(DDD, L). Remark 2.2.67. Notice that the first request in (2.2.50) implies the existence of an o satisfying the second request and for which the first equality in (2.2.50) holds. However even (2.2.51) does not imply the second and third equalities in (2.2.50). Since Def. 2.2.66 abstracts properties of an extended C−equivariant stability, it is natural to expect that Corollary 2.2.68. Let C be a category and E = h h H, UUU, m, W i, F i be an extended C−equivariant stability on DDD via Z and S, and let S and J denote the fields determined by m and W respectively. Then h (S ◦ Fo 0 , H, DDD, UUU i is a nucleon-fragment doublet on C. o , Zm), Z, S, F, m ∗ 1FU, W ◦ FDDD UUU, F† UUU), (J ◦ FDDD Proof. Since Prp. 2.2.64 and the definitions of S and J. (cid:3) 86 Definition 2.2.69. We call related to E the nucleon-fragment doublet constructed in Cor. 2.2.68. Up to the end of section 2.2.4 we let T = h S, J, Z, S, F, m, W, R, DDD, UUU i be a fixed but arbitrary nucleon-fragment doublet on C. Clearly the following results apply for the nucleon-fragment doublet related to any extended C−equivariant stability on DDD via Z and S. In Prp. 2.2.70 we make explicit the symmetries of S, m, J and W, in this way clarifying the meaning of the titles in the items of Def. 2.2.66. We emphatize that all the actions over functionals involved are conjugate of actions over observables. As a result Prp. 2.2.72 describes the properties of invariance of a nucleon-fragment doublet, physically interpeted as invariance of mean values in Prp. 2.2.76 with the help of Prp. 2.2.73 and Prp. 2.2.75. Proposition 2.2.70 (Properties of equivariance of a nucleon-fragment doublet). Let t ∈ MorΘ(UUU,F) and h ∈ MorΞ(DDD,F). Thus (1) S(t ◦ l) = S(l) ◦ S(t) for all l ∈ MorΘ(UUU,F) such that d(t) = c(l) and Sc(t) ⊆ A∗ F(c(t)), S(t)Sc(t) ⊆ Sd(t), S(t)u = u ◦ F1(t), ∀u ∈ Sc(t). (2) for all Q ∈ UUUF(c(t)) and α ∈ PQ F(c(t)) mc(t)(Q, α) ∈ Sc(t), mc(t)(Q, α) ◦ F1(t) = md(t)(F2(t)Q, α), mc(t)(Q, α) ◦ ψF(c(t))(l−1) = mc(t)(bF(c(t))(l)Q, α), ∀l ∈ RF(c(t)). (3) J(h ◦ i) = J(i) ◦ J(h), for all i ∈ MorΞ(DDD,F) such that d(h) = c(i), J(h)Jc(h) ⊆ Jd(h), and for all (I, f ) ∈ Jc(h) I ∈ DDDF(c(h)), f ∈ Yβ∈PI F(c(h)) NF(c(h))(mc(h)(I, β)) ∩ NA(F(c(h)))I β , J(h)(I, f ) = (F2(h)I, fh), fh : PF2(h)I F(d(h)) ∋ γ 7→ f (γ) ◦ S(h, I, γ). (2.2.53) (4) For all I ∈ DDDF(c(h)), β ∈ PI F(c(h)) and h ∈ RF(c(h)) (I, Wc(h)(I)) ∈ Jc(h), Wc(h)(I, β) ◦ S(h, I, β) = Wd(h)(F2(h)I, β), Wc(h)(I, β) ◦ V(F(c(h)))bF(c(h))(h)I β (h−1) = Wc(h)(bF(c(h))(h)I, β). 2.2. THE CATEGORY OF NUCLEON SYSTEMS 87 Proof. St.(1) follows since Def. 2.2.66(4), st.(2) by Def. 2.2.66(6), the first item in (2.2.53) follows since Def. 2.2.66(7a) and (2.2.49), the remaining of st.(3) by Def. 2.2.66(7). (cid:3) St.(4) follows since (2.2.51) and (2.2.52). As often remarked we introduce the structure of nucleon-fragment doublet on a category in order to resolve the universality claim. The next definition specifies the characteristics of the doublet, equivalently stated in Prp. 2.2.70, which primarly serve to this end, as we will show in Cor. 2.3.56. Definition 2.2.71 (T−nucleon phase and T−fragment state). We call m and W the T−nucleon phase and T−fragment state. Moreover let (2.2.46) be called the equivariance of the T−nucleon phase under contravariant action of Θ(UUU, L), and let (2.2.47) be called the equivariance of the T−nucleon phase under action of H. Let (2.2.51) be called the equivariance of the T−fragment state under contravariant action of Ξ(DDD, L), and let (2.2.52) be called the equivariance of the T−fragment state under action of H. Complexively we call all the previous properties the equivariances of the T−nucleon phase and T−fragment state, expanded if we add also the property (2.2.50). Finally we call T−resolution of the equivariant form of the universality claim, the set of the expanded equivariances of the T−nucleon phase and T−fragment state. Since all the above introduced state-evaluated maps are equivariant under actions implemented by conjugate of actions over observables, easily Prp. 2.2.70 gives rise to invariance of mean values as stated in the following Proposition 2.2.72 (Properties of invariance of nucleon-fragment doublets). Let (1) a, b ∈ Θ(UUU, F), T ∈ UUUF(b), β ∈ PT (2) t ∈ MorΘ(UUU,F)(a, b), (3) I ∈ UUUF(a), α ∈ PI F(a), f ∈ AF(a), l ∈ RF(a); F(b), then Moreover let mb(T, β)(F1(t)f) = ma(F2(t)T, β)(f), ma(I, α)(f) = ma(bF(a)(l)I, α)(ψF(a)(l)f), (1) d, e ∈ Ξ(DDD, F), R ∈ DDDF(e), δ ∈ PR (2) p ∈ MorΞ(DDD,F)(d, e), (3) O ∈ DDDF(d), γ ∈ PO F(d), a ∈ A(F(d))O γ , h ∈ RF(d); F(e), b ∈ A(F(d))F2(p)R δ , then We(R, δ)(S(p, R, δ)b) = Wd(F2(p)R, δ)(b), Wd(O, γ)(a) = Wd(bF(d)(h)O, γ)(V(F(d))O γ (h)a). Proof. Since Prp. 2.2.70 and (2.2.14). (cid:3) Proposition 2.2.73. Under the hypothesis of Prp. 2.2.72 we have (1) s(F(a)) ≡ the nucleon system generated by the fissioning system u(a); (2) s(O(F(a))I α) ≡ the fragment system whose observable algebra is A(F(a))I α and whose dynamics is (εF(a))I α(−α−1(·)), if α > 0; 88 (3) s((ϕϕϕF(a))I α) ≡ the state of thermal equilibrium (ϕϕϕF(a))I α at the inverse temperature (4) s(ma(I, α)) ≡ the phase of s(F(a)) occurring by performing the operation u(I) on α of s(O(F(a))I α), if α > 0; s((ϕϕϕF(a))I α); (5) s(We(R, δ)) ≡ the state of s(OF(e)R (6) s(O(F(a))bF(a)(l)I α δ ) originated via s(me(R, δ)); A(F(a))bF(a)(l)I (7) s((ϕϕϕF(a))bF(a)(l)I α α ) ≡ the fragment and whose dynamics is ad((ηF(a))I ) ≡ the state of thermal equilibrium (ϕϕϕF(a))bF(a)(l)I α(l)) ◦ (εF(a))I α α(−α−1(·)), if α > 0; at the inverse system whose observable algebra is temperature α of s(O(F(a))bF(a)(l)I ); α (8) s(ma(bF(a)(l)I, α)) ≡ the phase of s(F(a)) occurring by performing on ) the operation obtained by transforming through the action of s((ϕϕϕF(a))bF(a)(l)I l the operation u(I); α (9) s(Wd(bF(d)(h)O, γ)) s(md(bF(d)(h)O, γ)); ≡ the state of s(O(F(d))bF(d)(h)O γ ) originated via (10) s(mb(T, β)) ≡ the phase of s(F(b)) occurring by performing the operation u(T) on s((ϕϕϕF(b))T β ); (11) s(ma(F2(t)(T), β)) ≡ the phase of s(F(a)) occurring by performing on ) the operation obtained by transforming through the action of s((ϕϕϕF(a))F2(t)(T) u(F2(t)) the operation u(T); β (12) s(Wd(F2(p)R, δ)) ≡ the state of s(O(F(d))F2(p)R δ ) originated via s(md(F2(p)R, δ)). Proof. Sts. (4,5) follow by (2.2.50), sts. (8-12) follow by sts. statements are trivial. (4,5), the remaining (cid:3) i.e. Convention 2.2.74. Let a ∈ Ξ(DDD, F), I ∈ DDDF(a), δ ∈ PI F(a), set F(a)I δ (cid:17) F(ϕϕϕF(a))I δ ((aF(a))I δ ), F(a)I δ =nh ∈ F (cid:16)(ϕϕϕF(a))I δ ◦ (ηF(a))I δ ◦ j2(cid:17) (h) = (ϕϕϕF(a))I δo . Proposition 2.2.75 (Thermal nature-fragment state NCG origination-Phase transition of the nucleon phase ma). Under the hypothesis of Prp. 2.2.72 we obtain (1) Thermal nature of the nucleon phase ma(I, α) and stability under variation of the operations Tp. s(ma(I, α)) ≡ the phase of the nucleon system generated by the fissioning system u(a), occurring by performing the operation u(Tp) on the state of thermal equilibrium ϕϕϕp at the inverse temperature αp of the fragment system whose observable algebra is Ap and whose dynamics is εp(−α−1 p (·)), for all p ∈ RepF(a)(c) such that αp > 0. (2) Noncommutative geometric and thermal origination of fragment states via the nucleon phase ma(I, α). Ψ− r ◦ jr, 2.2. THE CATEGORY OF NUCLEON SYSTEMS 89 is the ϕϕϕr−normal state originated via s(ma(I, α)), of the fragment system whose observable algebra is Ar and whose dynamics is εr(−α−1 r (·)), for all r ∈ RepF(a)(ma(I, α)) such that αr > 0. (3) Phase transition of ma via symmetry breaking. s(ma(I, ξ)) exists only for those ξ ∈ II F(a) such that F(a)I ξ ⊇ F(a)I (βF(a) c . )I Proof. Since the definition of PN for any N object of G(G, F, ρ), Prp. 2.2.33 and Prp. (cid:3) 2.2.34. Jointly Prp. 2.2.73 the next result physically interpret Prp. 2.2.72 Proposition 2.2.76 (Mean values invariance related to a nucleon-fragment doublet). Under the hypothesis of Prp. 2.2.72 we obtain (1) Ξ(DDD, F)−invariance of the mean value in m. The following values are equal • the mean value in s(mb(T, β)) of the observable obtained by transforming through the action of u(F1(t)) the observable u(f), • the mean value in s(ma(F2(t)(T), β)) of the observable u(f). (2) H−invariance of the mean value in ma. The following values are equal • the mean value in s(ma(I, α)) of the observable u(f), • the mean value in s(ma(bF(a)(l)(I), α)) of the observable obtained by trans- forming through the action of l the observable u(f), (3) Ξ(DDD, F)−invariance of the mean value in W. The following values are equal • the mean value in s(We(R, δ)) of the observable obtained by transforming through the action of u(S(p, R, δ)) the observable u(b), • the mean value in s(Wd(F2(p)R, δ)) of the observable u(b); (4) H−invariance of the mean value in Wd. The following values are equal • the mean value in s(Wd(O, γ)) of the observable u(a), • the mean value in s(Wd(bF(d)(h)(O), γ)) of the observable obtained by trans- forming through the action of h the observable u(a). Proof. Since Prp. 2.2.72. (cid:3) 2.2.5. Construction of an equivariant stability. In this section we construct an equi- variant stability in Thm. 2.2.81. Definition 2.2.77. Define TTT• : Obj(G(G, F, ρ)) ∋ M 7→ TTTM, M m• : Obj(G(G, F, ρ)) ∋ M 7→ m , Dom(VVV•) ≔ {M ∈ Obj(G(G, F, ρ)) VVV•(M) , ∅}, 90 moreover set the map V• defined on Dom(VVV•) such that for all M ∈ Dom(VVV•), T ∈ VVV•(M) and α ∈ PT M V•(M) ∈ YQ∈VVV•(M) Yβ∈PQ M M NM(m (Q, β)), V•(M)(T, α) ≔ Ψ− (eM◦tM)(T,α) ◦ j(eM◦tM)(T,α). Remark 2.2.78. (eM ◦ tM)(T, α) ∈ RepM(m M (T, α)) for all M ∈ Dom(VVV•), T ∈ VVV•(M) and α ∈ PT M since Rmk. 2.2.22 and Lemma 2.2.26, so V• is well-defined. Lemma 2.2.79. Let B, C and D be three ∗−algebras, where D is unital with unit 1. U ∈ U(D), R ∈ MorCA∗(C, D) and η ∈ MorCA∗(B, C). Then (R ◦ η) = R ◦ η+, and (ad(U) ◦ R) = ad(U) ◦ R. Proof. Let b ∈ B, c ∈ C and λ ∈ C, then (R ◦ η)(b, λ) = (R ◦ η)(b) + λ1 = R(η(b), λ) = ( R ◦ η+)(b, λ), and (ad(U) ◦ R)(c, λ) = (ad(U) ◦ R)(c) + λ1 = ad(U)(R(c) + λ1) = (ad(U) ◦ R)(c, λ). (cid:3) Lemma 2.2.80. Let N ∈ G(G, F, ρ), T ∈ TTTN and α ∈ PT α (N))+ for which kbk ≤ 1 and RT α (N))+ such that kblk ≤ 1 and RTl b ∈ (BT bl ∈ (BTl b(l)(VVV•(N)) ⊆ VVV•(N). α (N)(b) = (Γ N)T α (N)(bl) = Γ Tl N such that there exists an element α . If l ∈ H set bl = ((wN)T α )+(l)(b) then α . In particular if VVV•(N) , ∅ then Proof. The inequality follows since ((wN)T α )+(l) is an isometry being (wN)T the equality follows since (2.2.7,2.2.6) and Lemma 2.2.79. α (l) so, while (cid:3) The following important result ensures the existence of an equivariant stability. Here the requests (2.2.5, 2.2.6, 2.2.7, 2.2.9), in the definition of the category G(G, F, ρ), find their justification since are used in its proof directly and via Lemma 2.2.80. Theorem 2.2.81 (Existence of a full integer equivariant stability on VVV•). h TTT•, m•, V• i is a full integer equivariant stability on VVV•. Moreover for all N ∈ Dom(VVV•), T ∈ VVV•(N) and α ∈ PT N (2.2.54) V•(N)(T, α) = ωexp(−((DN)T α )2) ◦ (πN)T α . Proof. In this proof let G be an arbitrary but fixed object of G(G, F, ρ) and let us use (2.2.11) and convention 2.2.7 by removing the index G. (2.2.54) follows since (2.2.21). h TTT•, m• i is a full equivariant phase according to (2.2.3) Def. 2.2.5, while V• satisfies 2.3. THE CANONICAL NUCLEON-FRAGMENT DOUBLET T• 91 (2.2.38) since construction, VVV• satisfies (2.2.33) since Lemma 2.2.80. Let us show that V• satisfies (2.2.40). Let T ∈ VVV•(G), α ∈ PT, l ∈ H, h ∈ {1, l} and set temporarily ρh = exp(−(DTh α )2) α (G) Ψh = Ψ(e◦t)(Th,α) Rh = RTh jh = jTh α π = πT α Hh = HTh α . α (l−1) ηl−1 = ηT zl−1 = zTl α (l−1) wl−1 = wTl α (l−1) vl = vT α (l) i = iT α We remove the index h whenever it equals 1. Since (2.2.29) and Lemma 1.2.2(2) we deduce (2.2.55) l = ωρl ◦ R− Ψ− l . Next Rl = ad(vl) ◦ R ◦ wl−1 by (2.2.12) and (2.2.7), hence (2.2.56) R− l = (ad(vl) ◦ R)− ◦ (i ◦ wl−1)− = ad(vl) ◦ R− ◦ (i ◦ wl−1)−, where the first equality follows since wl−1 is surjective, Rl is nondegenerate and Cor. 1.2.17, while the second one follows since R and ad(vl) ◦ R are nondegenerate and Rmk. 1.2.15. Next R− l ◦ jl = ad(vl) ◦ R− ◦ j ◦ ηl−1 ◦ zl−1 since (2.2.15) and (2.2.56), thus R− l ◦ jl = ad(vl) ◦ π ◦ ηl−1 ◦ zl−1 since (0.2.26), which together (2.2.55) yields (2.2.57) Ψ− l ◦ jl = ωρl ◦ ad(vl) ◦ π ◦ ηl−1 ◦ zl−1. Next ρl = ad(vl)(ρ) since (2.2.6) and (1.1.9), thus for all a ∈ L(H) TrHl(ρlad(vl)(a)) = TrHl(ad(vl)(ρa)) = TrH(ρa), in particular TrHl(ρl) = TrH(ρ) so ωρl ◦ ad(vl) = ωρ, therefore since (2.2.57) and (2.2.12) Ψ− α (l))−1, l ◦ jl = ωρ ◦ π ◦ ηl−1 ◦ zl−1 α (l) ◦ ηT = ωρ ◦ π ◦ (zT and (2.2.40) follows by (2.2.14) and (2.2.21). (cid:3) 2.3. The canonical nucleon-fragment doublet T• In section 2.3.1 we construct the object part of a functor GH △ from Cu(H) to G(G, F, ρ). In section 2.3.2 we complete the construction of GH △ , and employ this result to establish in our main Thm. 2.3.52 the existence of a extended full integer Cu(H)−equivariant stability on VVV•, and consequently of a nucleon-fragment doublet on Cu(H). Finally as a consequence of the main theorem we exhibit the resolution of the equivariant form of the universality claim. In the present section 2.3 we assume fixed two locally compact 92 topological groups G and F, a group homomorphism ρ : F → AutGr(G) such that the map (g, f ) 7→ ρ f (g) on G × F at values in G, is continuous. Let H denote G ⋊ρ F. 2.3.1. The object part GH. The main result in this section is Cor. 2.3.26 were we construct the object part of a functor from Cu(H) to G(G, F, ρ). Auxiliary important results in this directions are Thm. 1.1.54 and (2.3.5), Thm. 2.3.17, Thm. 2.3.21 and In the present section 2.3.1 we fix a dynamical system A = h A, H, σ i, Thm. 2.3.25. while starting from Def. 2.3.13 and except Thm. 2.3.25, A is inner and we fix a group morphism v : H → U(A) such that A is implemented by v. We let KX denote Morset(X, K) for any set X and K ∈ {R, C}. By taking into account (0.2.13), Def. 1.1.13 and notations after (0.2.17) we can give the following Definition 2.3.1. Let ω ∈ EG A(τ). We say that h HHH, µ, ζ, f i is a h A, ω i−selfadjoint system if (1) HHH ≔ h H, π, Ω i is a cyclic representation of A associated with ω; (2) µ ∈ H(SG ); Fω (3) ζ : RX → SG Fω (4) f is a Cσ(CX)−measurable map such that there exists an A ∈ Pω(X) satisfying is a continuous group morphism, where X is a nonempty set; f (C(A,Yx∈A supp(ETx))) ⊆ R. Here T (cid:17) {Tx}x∈X is such that iTx is the infinitesimal generator of the strongly continuous one- parameter semigroup of unitarities UHHH ◦ ζ ◦ ix ↾ R+ on H, where ix : R → RX is such that Pry ◦ix = δx,yIdR, for all x, y ∈ X. Set (2.3.1) Dζ, f HHH (A) ≔ f (EET). We convein to remove (A), whenever it is clear by the context which dynamical system is involved. Remark 2.3.2. Since RX is an abelian group, we deduce by Prp. 1.1.9 that ET is a family of commuting Borel RI's in H. Then we can apply Thm. 1.1.8(2) to ET and state that Dζ, f HHH (A) is a well-defined selfadjoint operator in H. Definition 2.3.3 (Pre thermal phases). We say that T = h h, ξ, βc, I, ωωω i is a pre thermal phase associated with A, if h ∈ H, ξ : R → G is a continuous group morphism, βc ∈ eR, I is a neighbourhood of βc, and A(τ) ∩ Kτ(h,ξ) EG β , (2.3.2) such that for all α ∈ I (2.3.3) Definition 2.3.4. Let ωωω ∈Yβ∈I (α ≤ βc ⇒ Fωωωα ⊇ Fωωωβc , βc < α ⇒ Fωωωβc ∩ ∁Fωωωα , ∅. 2.3. THE CANONICAL NUCLEON-FRAGMENT DOUBLET T• 93 • T = h h, ξ, βc, I, ωωω i be a pre thermal phase associated with A; • µµµ be a Haar system associated with ωωω and A; • HHH : I → Repc(A) be such that HHHβ = h Hβ, πβ, Ωβ i is a cyclic representation of A associated with ωωωβ, for all β ∈ I; is a continuous group morphism, where X is a nonempty set; • ζ : RX → SG Fωωωβc • Γ ∈Qα∈I∩]−∞,βc] L(Hα); • l ∈ H, β ∈ I and α ∈ I∩] − ∞, βc]. Define • ζl ≔ ad(l) ◦ ζ • µµµl : I × H ∋ (γ, h) 7→ µµµl (γ,h) ≔ µµµ(γ,h·ρl). If there exist A ∈ Pω(X) and a Cσ(CX)−measurable map f satisfying (2.3.4) f (C(A,Yx∈A supp(ETα x ))) ⊆ R, where iTα we can set y is the infinitesimal generator of the semigroup UHHHα ◦ ζ ◦ iy ↾ R+ on Hα, for all y ∈ X, • Dζ, f • RRRµµµ,ζ, f HHH,α(A) ≔ Dζ, f HHH,Γ,α(A) ≔ ( RRR HHHα (A), µµµ HHH,α(A), Dζ, f HHH,α(A), Γα). If A is inner implemented by v we can define Γ l HHH,v : I∩] − ∞, βc] ∋ δ 7→ Γ l HHH,v,δ ≔ ad(πδ(v(l)))(Γδ), and if in addition (2.3.4) holds we can set (1) Dζ, f (2) RRRµµµ,ζ, f HHH,v,α,l(A) ≔ Dζl, f HHH(v,l) µµµ HHH,v,Γ,α,l(A) ≔ ( RRR α (A), HHH,v,α,l(A), Dζ, f HHH,v,α,l(A), Γ l HHH,v,α). We convein to remove A whenever it is clear the dynamical system involved, in addition to remove both the indices v and l whenever l equals the unit. The next result shows that Def. 2.3.4 is well-set and the equivariance of the operator Dζ, f HHH,v,α,l under action of H, a step toward the construction in Cor. 2.3.26 of the object part of a functor from Cu(H) to G(G, F, ρ). Here we use the covariance of the functional calculus relative to a commuting set of resolutions of the identity stated in Thm. 1.1.14. Proposition 2.3.5. Def. 2.3.4(1) & (2) are well-defined and we have (2.3.5) Dζ, f HHH,v,α,l = πα(v(l)) Dζ, f HHH,α πα(v(l−1)). Proof. Since (2.3.3), Lemma 1.1.29(2) and the fact that any bijective map on a set Y induces an isomorphism of the order by inclusion on the power set of Y, we have SG F ⊆ SG F , so ωωωl βc ωωωl α ζl : RX → SG F ωωωl α . 94 Therefore UHHH,v,α,l ◦ ζl is well-set and Cor. 1.1.35 yields (2.3.6) Hence letting Tα,l UHHH,v,α,l ◦ ζl ◦ ix ↾ R+ we obtain for all x ∈ X x be such that iTα,l x UHHH,v,α,l ◦ ζl = ad(πα(v(l))) ◦ UHHH,α ◦ ζ. is the infinitesimal generator of the semigroup (2.3.7) Tα,l x = πα(v(l)) Tα x πα(v(l−1)), thus by Cor. 1.1.12(2), (0.2.15) and (2.3.4), (2.3.8) f (C(A,Yx∈A supp(ETα,l x ))) ⊆ R. Therefore according to Def. 2.3.1, the objects in Def. 2.3.4(1) & (2) are well-defined. (cid:3) Finally (2.3.5) follows since (2.3.7) and Thm. 1.1.14(2). Corollary 2.3.6. Let (1) T = h h, ξ, βc, I, ωωω i be a pre thermal phase associated with A; (2) µµµ be a Haar system associated with ωωω and A; (3) HHH : I → Repc(A) be such that HHHβ = h Hβ, πβ, Ωβ i is a cyclic representation of A associated with ωωωβ, for all β ∈ I; is a continuous group morphism, where X is a nonempty set; (4) ζ : RX → SG Fωωωβc (5) Γ ∈Qα∈I∩]−∞,βc] L(Hα); (6) l, u ∈ H, β ∈ I and α ∈] − ∞, βc] ∩ I. Then (1) Fσ∗(l)(ωωωβ) = ad(Pr2(l))(Fωωωβ); (2) SG = ad(l)(SG ); Fσ∗(l)(ωωωβ) Fωωωβ (3) ζl : RX → SG F ωωωl βc is a continuous group morphism; (4) µµµl is a Haar system associated with ωωωl; (5) h l ·ρ h, ξ, βc, I, ωωωl i is a pre thermal phase asociated to A; (6) if A is inner implemented by v then RRRµµµl = RRRµµµ HHH(v,l),v,α,u HHH,v,α,u·ρl moreover if there exists an A ∈ Pω(X) and a Cσ(CX)−measurable map f satisfying (2.3.4), then (2.3.9) (2.3.10) in particular Dζl, f HHH(v,l),v,α,u = Dζ, f HHH,v,α,(u·ρl) RRRµµµl,ζl, f HHH(v,l),Γ l HHH,v,α = RRRµµµ,ζ, f HHH,v,Γ,α,l. Proof. St. (1) & (2) follow by ωωωβ ∈ EG A(τ), and by Lemma 1.1.29. St.(3) follows by st.(2), while st.(4) since Lemma 1.1.48(2). St.(5) follows since Cor. 1.1.28, st.(1) and since any bijective map on a set induces an isomorphism of the order by inclusion on its power 2.3. THE CANONICAL NUCLEON-FRAGMENT DOUBLET T• 95 set. RRRµµµl (2.3.8), thus (2.3.9) and the first equality in st.(6) follow since is well-set since st.(4), Dζl, f HHH(v,l),v,α,u HHH(v,l),v,α,u is well-set since st. (3,4,5), Rmk. 1.1.32 and (2.3.11) (ζl)u = ζu·ρl (µµµl)u = µµµu·ρl (HHH(v,l))(v,u) = HHH(v,u·ρl). (cid:3) Definition 2.3.7. If T = h h, ξ, βc, I, ωωω i is a pre thermal phase associated with A and l ∈ H, then we define Tl ≔ h l ·ρ h, ξ, βc, I, ωωωl i which is a pre thermal phase associated with A according to Cor. 2.3.6. Definition 2.3.8. Let TTTA be the set of the h T, µµµ, HHH, ζ, f, Γ i such that (1) T = h h, ξ, βc, I, ωωω i is a pre thermal phase associated with A; (2) µµµ is a Haar system associated with ωωω and A; (3) HHH : I → Repc(A) is such that HHHβ = h Hβ, πβ, Ωβ i is a cyclic representation of A associated with ωωωβ, for all β ∈ I; is a continuous group morphism, where X is a nonempty set; (4) ζ : RX → SG Fωωωβc (5) Γ ∈Qα∈I∩]−∞,βc] L(Hα); for all α ∈ I∩] − ∞, βc], (6) f is a Cσ(CX)−measurable map such that there exists an A ∈ Pω(X) satisfying (2.3.4) (7) RRR(T, α) ≔ RRRµµµ,ζ, f HHH,Γ,α is an even θ−summable K−cycle for all α ∈ I∩] − ∞, βc]. Remark 2.3.9. We deduce since [16, IV.2.γ Def. 11 and IV.8 Def. 1] that Def. 2.3.8(7) is equivalent to the following two requests, (1) (2.3.15) for h = Id; (2) for all α ∈ I∩] − ∞, βc], Γα is a unitary, selfadjoint operator on Hα (a Z2−grading on Hα), such that for all a ∈ Bωωω,α,+ (a) [Γα, Rµµµ (b) ΓαDζ, f HHH,α HHH,α(a)] = 0, Γα = −Dζ, f HHH,α. µµµ PT A Definition 2.3.10. Let T = h T, µµµ, HHH, ζ, f, Γ i ∈ TTTA, where T = h h, ξ, βc, I, ωωω i, define ≔ I∩] − ∞, βc] and for all α ∈ PT A α (A) ≔ K0(Bωωω,α,+ KT ). µµµ Moreover set and A A K KQ α (A), KA ≔ [Q∈TTTA [α∈PQ ≔ [Q∈TTTAYα∈PQ α (A) ≔(cid:16)(σσσ(ωωωα,l))+(cid:17)∗ α (A). KQ A cA(l) ↾ KT . Finally set cA(l) as the map defined on KA such that for any T ∈ TTTA and α ∈ PT A 96 Convention 2.3.11. If ωωω : A → EG A(τ) with A a nonempty set and µµµ is a Haar system associated with ωωω and A, then for all l ∈ H and α ∈ A we convein to denote the pairing h ·, · iBωωω,α,l,+ by h ·, · iµµµ,ωωω,α,l. Moreover we remove the index l if it equals the identity. µµµ Definition 2.3.12. Let AA be the group whose underlying set is AA ≔ YQ∈TTTAYβ∈PQ A KQ β (A), and whose composition, inversion and identity 0 are the pointwise composition, inversion and identity, namely f · g ∈ AA such that (f · g)(T)(α) ≔ f(T)(α) · g(T)(α), f−1(T)(α) ≔ f(T)(α)−1 and 0(T)(α) is the identity of KT α (A), for all f, g ∈ AA, T ∈ TTTA and α ∈ PT A. Moreover set mA : AA → YQ∈TTTA Morset(PQ A , R), such that for any f ∈ AA and T ∈ TTTA mA(f)(T) : PT A → R, α 7→ h f(T)(α), ch(RRR(T, α)) iµµµ,ωωω,α, where T = h T, µµµ, HHH, ζ, f, Γ i and T = h h, ξ, βc, I, ωωω i. mA is called mean value map associated with A. Definition 2.3.13. Let l ∈ H, define bA,v(l) : TTTA ∋ h T, µµµ, HHH, ζ, f, Γ i 7→ h Tl, µµµl, HHH(v,l), ζl, f, Γ l HHH,v i. Convention 2.3.14. Often in the proof of the statements and only when it is not cause of α (A), for any T ∈ TTTA and confusion we convein to remove A from AA, mA, cA, KA, K α ∈ PT A, moreover we remove A and v from bA,v. and KT A Proposition 2.3.15. Let l ∈ H, thus cA(l) is well-defined and cA(l)(KT α (A)) ⊆ K0(Bωωω,α,l,+ µµµ ), for any T ∈ TTTA and α ∈ PT A. A, β ∈ PQ Proof. Let Q, T ∈ TTTA and α ∈ PT µµµ(α,1) (SG is the completion of CCC c (SG ∅. Now Bωωω,α µµµ(α,1) minimal Cauchy filters of CCC c implies 5 Cc(SG , A) ∩ Cc(SG Fωωω Fωωωα µµµ Fωωωα A such that KT µµµ , , A), thus its underlying set is the set of all , ∅ thus B (cid:17) Bωωω,α α ∩ KQ β µµµ ∩ Bωωω,β , A), see for example [9, II.21], therefore B , ∅ , A) , ∅. Hence a fortiori SG Fωωωα constant C such that k · kµµµ(α,1) = Ck · kµµµ(β,1) then Bωωω,α,+ 1.1.52. Thus KT (0.2.2), (0.2.32), Cor. 1.1.52 and the standard picture of the functor K0. β and(cid:16)(σσσ(ωωωα,l))+(cid:17)∗ =(cid:16)(σσσ(ωωωβ,l))+(cid:17)∗ Fωωωα and σσσωωωα,l = σσσωωωβ,l since Cor. therefore the statement follows since (cid:3) = Bωωω,β,+ α = KQ µµµ µµµ β = SG Fωωω β so there exists a 5This irrespectively by the fact that there could be δ ∈ PT λδ : Bωωω,δ λε(Bωωω,ε µµµ → D and λε : Bωωω,ε µµµ ) , ∅. µµµ → D such that λδ(Cc(SG Fωωω δ A, ε ∈ PQ , A)) ∩ λε(Cc(SG Fωωω A , a C∗−algebra D and ∗−embeddings µµµ ) ∩ , A)) = ∅ although λδ(Bωωω,δ ε 2.3. THE CANONICAL NUCLEON-FRAGMENT DOUBLET T• 97 Remark 2.3.16 (Integrality). Since the general result [16, IV.8.δ, Prp. 18, Thm. 19, and IV.8.ǫ Thm. 22] we deduce that mA(f)(T) is a Z−valued map for any f ∈ AA and T ∈ TTTA. The following Thm. 2.3.17, Thm. 2.3.21 and Thm. 2.3.25 are important steps towards the proof of Cor. 2.3.26 were we construct the object part of a functor from Cu(H) to G(G, F, ρ). Theorem 2.3.17. bA,v ∈ MorGr(H, Autset(TTTA)) and cA ∈ MorGr(H, Autset(KA)). Moreover for any T ∈ TTTA, α ∈ PT A and l ∈ H cA(l)(KT α (A)) = KbA,v(l)(T) α (A). Proof. Let T = h T, µµµ, HHH, ζ, f, Γ i ∈ TTTA, l ∈ H and α ∈ PT A. In this proof for any h ∈ {Id, l} we use the following notations HHH,v,α,h, H (cid:17) Hα, 1 (cid:17) 1H, and Tr (cid:17) TrH, Dh (cid:17) Dζ, f R(λ, Dh) (cid:17) (λ1 − Dh)−1, ∀λ ∈ ρ(Dh), U (cid:17) πα(v(l)), Bh (cid:17) Bωωω,α,h Rh (cid:17) Rµµµ σσσl (cid:17) σσσ(ωωωα,l), Tl (cid:17) b(l)(T), xl (cid:17) c(l)(x), ∀x ∈ KT α , RRRh (cid:17) RRR(Th, α). , HHH,v,α,h, µµµ  Here 1 (cid:17) 1Hα and ρ(T) is the resolvent set of any selfadjoint operator T in H. We convein to remove the index h whenever it equals the unit. (2.3.10) yields (2.3.12) RRRl = RRRµµµ,ζ, f HHH,v,Γ,α,l. Since Cor. 2.3.6 and (2.3.8) to prove Tl ∈ TTTA it is sufficient to show that RRRl is an even θ−summable K−cycle. Since (2.3.5) and Cor. 1.1.12(2) (2.3.13) ρ(Dl) = ρ(D), moreover since Lemma 2.2.79 and Thm. 1.1.54 we obtain (2.3.14) Rl ◦ σσσ+ l = ad(U) ◦ R. Let us consider the following set of statements for h ∈ {Id, l} (2.3.15) R(λ, Dh) is a compact operator on H, ∀λ ∈ ρ(Dh), Dom([Dh, Rh(a)]) = Dom(Dh), ∀a ∈ B+ h [Dh, Rh(a)] ∈ L(Dom(Dh), H), ∀a ∈ B+ h Tr(exp(−D2 h)) < ∞  98 then it holds by hypothesis for h = Id, we claim to show it for h = l. Let λ ∈ ρ(D), thus since (2.3.5), we have λ1 − Dl = U(λ1 − D)U−1, and Dom(Dl) = UDom(D), moreover Dom(R(λ, D)) = H, hence (λ1 − Dl)ad(U)(R(λ, D)) = 1 and ad(U)(R(λ, D))(λ1 − Dl) = IdDom(Dl). Therefore (2.3.16) R(λ, Dl) = ad(U)(R(λ, D)). Since (2.3.16), (2.3.15) for h = Id and since the set of compact operators on H is a two-sided ideal of L(H), we obtain (2.3.17) Let a ∈ B+, thus Dom([Dl, Rl(σσσ+ moreover R(λ, Dl) is a compact operator. l (a))]) = UDom(D) since (2.3.15), (2.3.14) and (2.3.5), (2.3.18) Dl Rl(σσσ+ l (a)) − Rl(σσσ+ l (a))Dl = U[D, R(a)]U−1, hence for all v ∈ Dom(D) k[Dl, Rl(σσσ+ l (a))]Uvk = k[D, R(a)]vk ≤ k[D, R(a)]kL(Dom(D),H)kUvk. l (B+) = B+ l since Cor. 1.1.52, therefore we can state for all b ∈ B+ l Dom([Dl, Rl(b)]) = Dom(Dl), [Dl, Rl(b)] ∈ L(Dom(Dl), H), k[Dl, Rl(b)]kL(Dom(Dl),H) = k[D, R((σσσ+ l )−1(b))]kL(Dom(D),H). Next σσσ+ (2.3.19)  For any h ∈ {Id, l}, sp(Dh) ⊆ R, since Dh is a selfadjoint operator by Rmk. 2.3.2, therefore exp(−D2 h) ∈ L(H), since the spectral theorem, see for example [19, Thm. 18.2.11(c)], and the fact that sp(Dl) ∋ λ 7→ exp(−λ2), is bounded. Therefore Tr(exp(−D2 h)) is a well-set element of R. Since Cor. 1.1.12(3) and (2.3.5) (2.3.20) exp(−D2 l ) = ad(U)(exp(−D2)), hence Tr(exp(−D2 l )) = Tr(exp(−D2)), then by (2.3.15) for h = Id we obtain (2.3.21) Tr(exp(−D2 l )) < ∞. (2.3.15) for h = l follows by (2.3.13), (2.3.17), (2.3.19) and (2.3.21). Next Γ l = α ad(U)(Γα) so is a Z2−grading on H, moreover since (2.3.14), (2.3.5), the bijectivity of σσσ+ l and Rmk. 2.3.9 we obtain for all b ∈ Bl (cid:17) Γ l HHH,v,α (2.3.22) ([Γ l α, Rl(b)] = 0, Γ l αDlΓ l α = −Dl. Thus (2.3.15) for h = l and (2.3.22) yields that RRRl is an even θ−summable K−cycle, thus Tl ∈ TTTA. So b(l) maps TTTA into TTTA and b is a H−action on TTTA since (2.3.11). Let x ∈ KT α thus xl ∈ KTl α since Tl ∈ TTTA and Prp. 2.3.15. Thus c(l) maps KA into itself since Cor. 1.1.52 and (cid:3) c is a H− action on KA since Cor. 1.1.53 Thm. 2.3.17 permits the following 2.3. THE CANONICAL NUCLEON-FRAGMENT DOUBLET T• 99 Definition 2.3.18. Define cA : H → Morset(K A A , K ) such that cA(l)(g) ≔ cA(l) ◦ g for all A l ∈ H and g ∈ K . Proposition 2.3.19. cA ∈ MorGr(H, Autset(K A )). Proof. Since Thm. 2.3.17. (cid:3) A Since AA ⊂ Morset(TTTA, K ), Thm. 2.3.17 and Prp. 2.3.19 permit the following Definition 2.3.20. For any l ∈ H define the map ψA,v(l) on AA such that for all f ∈ AA ψA,v(l)(f) ≔ cA(l) ◦ f ◦ bA,v(l−1). Theorem 2.3.21. We have (1) ψA,v ∈ MorGr(H, AutAb(AA)), (2) for all l ∈ H and f ∈ AA we have evf(mA ◦ ψA,v(l)) ◦ bA,v(l) = evf(mA). Proof. In this proof we convein to denote ψA,v by ψ and cA by c. ψ(l) maps AA into itself, in addition ψ is a H−action since b and c are H−actions by Thm. 2.3.17 and Prp. 2.3.19, finally ψ(l) is a group morphism since the second inclusion in (0.2.32) and since the standard picture used for the K0−groups, hence st.(1) follows. Next let us adopt the notations in proof of Thm. 2.3.17, let l ∈ H, f ∈ AA, T ∈ TTTA and α ∈ PT A then (cid:16)evf(m ◦ ψ(l)) ◦ b(l)(cid:17)(T)(α) = m(ψ(l)(f))(Tl)(α) = h ψ(l)(f)(Tl)(α), ch(RRR(Tl, α)) iµµµl,ωωωl,α = h c(l)(cid:16)f(T)(α)(cid:17), ch(RRRl) iµµµl,ωωωl,α = h(cid:16)f(T)(α)(cid:17)l , ch(RRRl) iµµµl,ωωωl,α. Hence st.(2) follows if we show that for all x ∈ KT α (2.3.23) h xl, ch(RRRl) iµµµl,ωωωl,α = h x, ch(RRR) iµµµ,ωωω,α. 100 Let us prove (2.3.23). By (2.3.14), (2.3.5) and (2.3.20) we obtain for all s0, . . . , s2n ∈ R and a0, . . . , a2n ∈ B+ (2.3.24) Tr(cid:16)Γ l Rl(σσσ+ α [Dl, Rl(σσσ+ l ) . . . l (a1))] exp(−s1D2 l )[Dl, Rl(σσσ+ l (a0)) exp(−s0D2 l )[Dl, Rl(σσσ+ l (a2n−1))] exp(−s2n−1D2 l (a2n))] exp(−s2nD2 l )(cid:17) = (Tr ◦ ad(U))(cid:16)Γα R(a0) exp(−s0D2)[D, R(a1)] exp(−s1D2) . . . [D, R(a2n−1)] exp(−s2n−1D2)[D, R(a2n)] exp(−s2nD2)(cid:17) = Tr(cid:16)Γα R(a0) exp(−s0D2)[D, R(a1)] exp(−s1D2) . . . (2.3.23) and then st.(2) follows since (2.3.24), (2.3.12) and [16, IV.8.ǫ Thm 22, Thm. 21, (cid:3) and IV.7.δ Thm 21]. [D, R(a2n−1)] exp(−s2n−1D2)[D, R(a2n)] exp(−s2nD2)(cid:17). Remark 2.3.22 (Odd case). Let TTT1 A be defined as TTTA by replacing K0 by K1 and setting Γα = 1α for all α ∈ I∩] − ∞, βc]. Thus it is easy to show that Thm. 2.3.21 still holds with TTT1 A in place of TTTA and K1 in place of K0. According to the definition of vA in Def. 2.2.1 and the construction of σσσ(ωωωα,l) in Cor. 1.1.52 we set the following Definition 2.3.23. Let A = h A, H, σ i ∈ Obj(Cu(H)) define (1) IA : TTTA → set (2) (βA c ) ∈QQ∈TTTA IQ A ; A A A C(H); C(R); EA; (6) ψA ≔ ψA,vA; (7) bA ≔ bA,vA; (3) aA ∈QQ∈TTTAQβ∈IQ (4) eA ∈QQ∈TTTAQβ∈IQ (5) ϕϕϕA ∈QQ∈TTTAQβ∈IQ (8) HA ∈QQ∈TTTAQβ∈IQ (9) πA ∈QQ∈TTTAQβ∈IQ (10) ΩA ∈QQ∈TTTAQβ∈IQ (11) EA = h µA, uA, HHHA, DA, Γ A, vA, wA, zA i an 8−tuple of elements ofQQ∈TTTAQβ∈PQ HS; MorCA∗(A, L((HA)Q (HA)Q β ; β )); A A A set; A where if T ∈ TTTA, T = h T, µµµ, HHH, ζ, f, Γ i, T = h h, ξ, βc, I, ωωω i, HHHγ = h Hγ, πγ, Ωγ i, ∀γ ∈ I; then 2.3. THE CANONICAL NUCLEON-FRAGMENT DOUBLET T• 101 (−α(·)) i, α i = h Hα, πα, Ωα i, σ α , (ΩA)T (1) IT = I; A (2) (βA c )T = βc; (3) for all α ∈ I (a) (aA)T α = A, α = h A, R, τ(h,ξ) (b) (eA)T (c) (ϕϕϕA)T α = ωωωα, α , (πA)T (d) h (HA)T (e) (µA)T α = µµµ(α,1), α = evα ◦ evT ↾ AA, (f) (uA)T (g) (HHHA)T α = HHHα, α = Dζ, f (h) (DA)T (i) (Γ A)T α = Γα, (j) for all l ∈ H (i) (vA)T (ii) (wA)T (iii) (zA)T HHH,α(A), α (l) = πα(vA(l)), α (l) = σσσ(ωωωα,l), α (l) = IdA; in addition for all α ∈ I we define (A), BT iT α jT α α (A) ≔ SG ST Fωωωωωωωωωα α (A) ≔ Bωωω,α µµµ (A), ≔ iBT α (A), ≔ jBT α (A) α (A) ≔ Rµµµ α (A) ≔ RRRµµµ HHH,α(A), HHH,α(A), h ·, · i(A,T,α) ≔ h ·, · iBT RT RRRT A , α (A)+. Often and only if it will not cause confusion, we convein to remove (A) from ST α (A) α (A). According to Def. 2.3.8, Def. 2.3.10, Def. 2.3.12 and Def. 2.3.23 we can set and BT the following Definition 2.3.24. Define GH to be the map on Obj(Cu(H)) such that if A ∈ Obj(Cu(H)) then (2.3.25) GH(A) ≔ h TTTA, IA, βA c , PA, aA, eA, ϕϕϕA, AA, ψA, bA, mA, EA i. Theorem 2.3.25. Let T ∈ TTTA, α ∈ PT A and l ∈ H, thus (ibA(l)T α ◦ σσσ(ωωωα,l))− ◦ jT A = jbA(l)T A ◦ σ(l). 102 Proof. In this proof let Tl denote bA(l)T. Let f ∈ Cc(ST α , A) and a ∈ A then (2.3.26) (iTl α ◦ σσσ(ωωωα,l))−(jT A(a)) ◦ (iTl (iTl α ◦ σσσ(ωωωα,l))−((jT (iTl α ◦ σσσ(ωωωα,l))−(iT α ◦ σσσ(ωωωα,l))( f ) = A(a)) ◦ iT α ( f )) = α (jT (iTl α ◦ σσσ(ωωωα,l))(jT A(a)( f ))) = A(a)( f )) = iTl α (cid:16)σ(l) ◦ jT α (a)( f ) ◦ ad(l−1) ↾ STl α(cid:17) , where the first and third equalities follow since (0.2.22), the second one by (1.2.6), the fourth by construction. Next for all h ∈ STl α (cid:16)σ(l) ◦ jT α (a)( f ) ◦ ad(l−1) ↾ STl σ(l)(a f (ad(l−1)(h))) = α(cid:17) (h) = σ(l)(a)σσσ(ωα,l)( f )(h) =(cid:16)jTl A (σ(l)(a)) ◦ σσσ(ωωωα,l)(cid:17) ( f )(h), hence by (2.3.26) we have (2.3.27) (iTl α ◦ σσσ(ωωωα,l))−(jT α ◦ jTl (cid:16)iTl A(a)) ◦ (iTl α ◦ σσσ(ωωωα,l))( f ) = A (σ(l)(a)) ◦ σσσ(ωωωα,l)(cid:17) ( f ) = jTl A (σ(l)(a)) ◦ (iTl α ◦ σσσ(ωωωα,l))( f ), where the last equality follows since (1.2.6). Next Cc(ST σσσ(ωωωα,l) is an isometry since Cor. 1.1.52, thus by (2.3.27) we deduce α , A) is dense in BT α moreover iM(BTl α )((iTl α ◦ σσσ(ωωωα,l))−(jT A(a))) ↾ K(BTl α ) = iM(BTl α )(jTl A (σ(l)(a))) ↾ K(BTl α ), therefore the statement follows since Lemma 1.2.12. (cid:3) Now we are able to state the following Corollary 2.3.26. GH maps Obj(Cu(H)) into Obj(G(G, F, ρ)). Proof. Def. 2.2.2(9,10) follow since Thm. 2.3.21(1) and Thm. 2.3.17. The additivity of mA follows by the additivity of the Chern-Connes character, (2.2.1) follows since Lemma 1.1.29(1), (2.2.2) by (2.3.3), (2.2.3) since Thm. 2.3.21(2), (2.2.4) by (2.3.2) and (0.2.8), the integrality since Rmk. 2.3.16, (uA)T α is by construction a group morphism while (2.2.10) follows since the construction of mA. Next (2.2.5), (2.2.6) and (2.2.7) since Cor. 1.1.53, (2.3.5) and Thm. 1.1.54 respectively, while (2.2.8) follows by the construction of ψA. (cid:3) Finally (2.2.9) follows since Thm. 2.3.25. Notice that V(GH(A))T dynamics underlying A. α = σ for all A ∈ Obj(Cu(H)), T ∈ TTTA and α ∈ PT A, where σ is the 2.3. THE CANONICAL NUCLEON-FRAGMENT DOUBLET T• 103 2.3.2. The functor GH △ . Thm. 2.3.42 completes the result in Cor. 2.3.26 by stating that GH is the object part of a functor from Cu(H) to G(G, F, ρ). Then we utilize this result in our main theorem Thm. 2.3.52 to exhibit the existence of an extended full inte- ger Cu(H)−equivariant stability, and consequently the existence of a nucleon-fragment doublet on Cu(H). As a consequence we obtain in Cor. 2.3.56 the resolution of the equivariant form of the universality claim. Definition 2.3.27. Let T : A → B be a ∗−homomorphism between C∗−algebras and ≔ h Hβ, πβ◦T, Ωβ i, HHH : A ∋ α 7→ h Hα, πα, Ωα i ∈ Repc(B), define the map HHHT on A such that HHHT β for all β ∈ A. Lemma 2.3.28. Let T : A → B be a surjective ∗−homomorphism between C∗−algebras, ω ∈ EB and HHH be a cyclic representation of B associated with ω. Then HHHT is a cyclic representation of A associated with T†(ω). Proof. T†(ω) ∈ EB since Lemma 1.1.17 so the statement is well-set, moreover HHHT is (cid:3) cyclic since T(A) = B. Unless differently specified in the present subsection let A = h A, H, η i, B = h B, H, θ i and C = h C, H, δ i be objects of C(H), T ∈ MorC(H)(A, B) and S ∈ MorC(H)(B, C). Corollary 2.3.29. Let φ ∈ EG A(τη) and Uη with φ, then T†(φ) ∈ EG B(τθ) and HHH be a cyclic representation of B associated HHHT = Uθ HHH. Remark 2.3.30. Uη HHHT makes sense since Def. 1.1.33, the first sentence of the statement of Cor. 2.3.29 and Lemma 2.3.28. The equality in Cor. 2.3.29 is well-set since Lemma 1.1.25. Proof of Cor. 2.3.29. φ ◦ T ◦ η(j1(g)) = φ ◦ θ(j1(g)) ◦ T = φ ◦ T for all g ∈ G, so the first sentence of the statement follows since Lemma 1.1.17. Let HHH = h H, π, Ω i and l ∈ SG then for all a ∈ A FT† (φ) Uη HHHT (l)(π ◦ T)(a)Ω = (π ◦ T)(η(l)a)Ω = (π ◦ θ(l) ◦ T)(a)Ω = Uθ HHH(l)(π ◦ T)(a)Ω, where the last equality follows since Lemma 1.1.25. Thus the equality in the statement (cid:3) follows since HHHT is cyclic by Lemma 2.3.28. Proposition 2.3.31. Let A be a nonempty set and ωωω : A → EG B(τθ). Then H(ωωω, B) = H(T† ◦ ωωω, A). Proof. Since Cor. 2.3.29 the statement is well-set. Let l ∈ H and α ∈ A then η∗(l)(T†(ωωωα)) = ωωωα ◦ θ(l−1) ◦ T = T†(θ∗(l)(ωωωα)), hence by Lemma 1.1.25 (2.3.28) and the statement follows. SG Fη∗(l)(T†(ωωωα)) (A) = SG Fθ∗(l)(ωωωα) (B), (cid:3) 104 all α ∈ A and l ∈ H let ad(ωωωα,θ) ⋆ Remark 2.3.32. Let A be a nonempty set, ωωω : A → EG B(τθ) and µµµ ∈ H(ωωω, B). Thus for (l) on Cc(SG Fωωωα , A) at values in Cc(SG Fθ∗(l)(ωωωα) A) such that f 7→ f ◦ ad(l−1) ↾ SG Fθ∗(l)(ωωωα) , this map is continuous w.r.t. the inductive limit topology following the line in the proof of Lemma 1.1.48, hence there exists according to [38, Cor. 2.47] a unique extension on Bωωω,α µµµ (B) at values in Bωωω,α,l (l). Next it is easy to see that (B) which will be denoted again by the symbol ad(ωωωα,θ) ⋆ µµµ θθθ(ωωωα,l) = cµµµ(α,l)(θ(l)) ◦ ad(ωωωα,θ) ⋆ (l) = ad(ωωωα,θ) ⋆ (l) ◦ cµµµ(α,1)(θ(l)). Remark 2.3.33. Let A be a nonempty set, ωωω : A → EG B(τθ) and µµµ ∈ H(ωωω, B). Thus for all α ∈ A and l ∈ H since (0.2.27), Lemma 1.1.17 and Lemma 1.1.25 we have cµµµ(α,l)(T) ∈ MorCA∗(BT†◦ωωω,α,l µµµ (A), Bωωω,α,l µµµ (B)), where BT†◦ωωω,α,l µµµ (A) is well-set since Prp. 2.3.31. In addition since (0.2.34) we have kµµµ(α,l)(T) : K0(BT†◦ωωω,α,l µµµ (A)) → K0(Bωωω,α,l µµµ (B)). Moreover T† ◦ ωωω : A → EG Finally since (2.3.28) we deduce that A(τη) and µµµ ∈ H(T† ◦ ωωω, A) since Cor. 2.3.29 and Prp. 2.3.31. (2.3.29) ad(ωωωα,θ) ⋆ (l) ◦ cµµµ(α,1)(T) = cµµµ(α,l)(T) ◦ ad(T†(ωωωα),η) ⋆ (l). Definition 2.3.34. Define dH the map on MorC(H) such that dH(T) : TTTB ∋ h T, µµµ, HHH, ζ, f, Γ i 7→ h TT, µµµ, HHHT, ζ, f, Γ i, where if T = h h, ξ, βc, I, ωωω i then TT = h h, ξ, βc, I, T† ◦ ωωω i. In the remaining of this subsection we let d denote dH. In the following result we shall use the equivariance of the KMS−states under the action of surjective equivariant maps stated in Thm. 1.1.19. Lemma 2.3.35. d(T) : TTTB → TTTA and d(S ◦ T) = d(T) ◦ d(S), as a result (TTT• ◦ GH, d ↾ MorCu(H)) is a functor from Cu(H)op to set. Proof. Let T = h T, µµµ, HHH, ζ, f, Γ i ∈ TTTB with T = h h, ξ, βc, I, ωωω i, we claim to show that d(T)(T) satisfies the requests in Def. 2.3.8. TT is a pre thermal phase associated with A since Cor. 2.3.29 and Thm. 1.1.19. Therefore we obtain Def. 2.3.8 (2,3,4,6) since Prp. 2.3.31, Lemma 2.3.28, Lemma 1.1.25 and Cor. 2.3.29 respectively. Next let α ∈ PT B and 2.3. THE CANONICAL NUCLEON-FRAGMENT DOUBLET T• 105 HHHα = h π, Hα, Ωα i, then RRRµµµ HHHT,α(A) = (BT†◦ωωω,α µµµ (A), Rµµµ HHHT,α(A)), where (2.3.30) Rµµµ HHHT,α(A) = (πα ◦ T) ⋊µµµ(α,1) Uη HHHT α = (πα ◦ T) ⋊µµµ(α,1) Uθ HHHα = (πα ⋊µµµ(α,1) Uθ HHHα = Rµµµ HHH,α(B) ◦ cµµµ(α,1)(T), ) ◦ cµµµ(α,1)(T) where the second equality follows by Cor. 2.3.29. Next let X be the nonempty set such that RX is the domain of ζ, thus Dζ, f (A) = f (EEL), where L = {Lx}x∈X such that iLx is the infinitesimal generator of the strongly continuous one-parameter semigroup Uη HHHT α ◦ ζ ◦ ix ↾ R+ on HHHα, for all x ∈ X. Thus since Cor. 2.3.29 we deduce that HHHT,α(A) = Dζ, f HHHT α (2.3.31) HHHT,α(A) = Dζ, f Dζ, f HHH,α(B). Def. (2.3.8)(7) follows since (2.3.30), (2.3.31) and Rmk. 2.3.9 and our claim is proved so (cid:3) d(T)(T) ∈ TTTA. The remaining part of the statement is easy to show. Definition 2.3.36. Define and KA(T) ≔ [T∈TTTB [α∈PT (T) ≔ [T∈TTTB Yα∈PT K A B B Kd(T)(T) α (A), Kd(T)(T) α (A). KA(T) is a well-defined subset of KA since Lemma 2.3.35. Definition 2.3.37. Define hH(T) : KA(T) → KB such that for any T = h T, µµµ, HHH, ζ, f, Γ i ∈ TTTB and α ∈ PT B hH(T) ↾ Kd(T)(T) α (A) ≔ kµµµ(α,1)(T). In the remaining of this subsection we let h denote hH. Lemma 2.3.38. We have (1) h(T) is well-defined and h(T)(Kd(T)(T) (2) KA(S ◦ T) ⊆ KA(T); (3) Let ı be the identity map from KA(S ◦ T) to KA(T), then the following diagram is α (B), for any T ∈ TTTB and α ∈ PT B; (A)) ⊆ KT α commutative (2.3.32) KB(S) h(T)◦ı h(S) / KC :✉✉✉✉✉✉✉✉✉✉ h(S◦T) KA(S ◦ T) / O O : 106 Proof. h(T) is well-defined since the same argument used in Lemma 2.3.15, while the inclusion in st.(1) follows by Rmk 2.3.33. St.(2) follows by Lemma 2.3.35. h(T) ◦ ı is a well-set map since st.(2), with values in KB(S) since Lemma 2.3.35 and st.(1). The diagram is commutative since kµµµ(α,1) is the morphism map of a functor from C0(H) to (cid:3) Ab. Since Lemma 2.3.38(1) we can give the following Definition 2.3.39. Define h H A (T) : K B such that h H (T)(g) ≔ h(T) ◦ g for all A g ∈ K (T). Moreover set ı : K A (S ◦ T) → K A (T) such that ı(g) = ı ◦ g for all g ∈ K (S ◦ T). (T) → K A Proposition 2.3.40. The following diagram is commutative (2.3.33) B K (S) h(T)◦ı h(S) C / K ;✇✇✇✇✇✇✇✇✇ h(S◦T) A K (S ◦ T) Proof. Since (2.3.32). (cid:3) We recall that according to our notation ( f × g) : X → A × B such that ( f × g)(x) = ( f (x), g(x)) for any x ∈ X, where X, A, B are sets while f : X → A and g : X → B. For any f ∈ AA the map f ◦ d(T) is well-set since Lemma 2.3.35, while f ◦ d(T) : TTTB → K since Def. 2.3.39 and Lemma 2.3.38(1), we can give the following (T), thus A Definition 2.3.41. Define the map gH on MorC(H) such that and for any f ∈ AA Finally define gH(T) : AA → AB, H gH(T)(f) ≔ h (T) ◦ f ◦ dH(T). GH △ ≔ (GH, (gH × dH) ↾ MorCu(H)). Now we are in the position of stating that GH △ is a functor namely Theorem 2.3.42. GH △ ∈ Fct(Cu(H), G(G, F, ρ)). Proof. The part of the statement concerning GH follows since Cor. 2.3.26. Let A, B and C be objects of Cu(H), T ∈ MorCu(H)(A, B) and S ∈ MorCu(H)(B, C). Let g denote gH, thus g(T) is a group morphism since the second line in (0.2.32) and since the standard picture used for the K0−groups. Next we claim to show that (2.3.34) g(S ◦ T) = g(S) ◦ g(T), evf(mB ◦ g(T)) = evf(mA) ◦ d(T), ∀f ∈ AA. / O O ; 2.3. THE CANONICAL NUCLEON-FRAGMENT DOUBLET T• 107 The first equality follows since (2.3.33) and Lemma 2.3.35, let us prove the second equality of (2.3.34). Let f ∈ AA, T ∈ TTTB, and α ∈ PT B, moreover let T = h T, µµµ, HHH, ζ, f, Γ i with T = h h, ξ, βc, I, ωωω i, and let TT denote d(T)(T), then evf(mB ◦ g(T))(T)(α) = mB(g(T)(f))(T)(α) = h g(T)(f)(T)(α), ch(RRR(T, α)) iµµµ,ωωω,α = h (c+ µµµ(α,1)(T))∗(cid:16)f(TT)(α)(cid:17), ch(RRR(T, α)) iµµµ,ωωω,α µµµ(α,1)(T))†(cid:16)ch(RRR(T, α))(cid:17) iµµµ,T†◦ωωω,α, = h f(TT)(α), (c+ (2.3.35) (2.3.36) where the last equality follows since Rmk. 2.3.33 and (0.2.35). Next (cid:16)evf(mA) ◦ d(T)(cid:17)(T)(α) = mA(f)(TT)(α) = h f(TT)(α), ch(RRR(TT, α)) iµµµ,T†◦ωωω,α. Moreover by construction HHHT,α(A), Dζ, f thus we obtain since (2.3.30) & (2.3.31) and Lemma 2.2.79 RRR(TT, α) =(cid:16)BT†◦ωωω,α,+ (A), Rµµµ HHHT,α(A), Γα(cid:17), µµµ RRR(TT, α) =(cid:16)BT†◦ωωω,α,+ µµµ (A), Rµµµ HHH,α(B) ◦ c+ µµµ(α,1)(T), Dζ, f HHH,α(B), Γα(cid:17). Hence we deduce since [16, Ch IV.8.ǫ, Thm. 22 and Thm. 21] and [16, Ch IV.7.δ, Thm. 21 and Lemma 20] that (2.3.37) ch(RRR(TT, α)) = (c+ µµµ(α,1)(T))† (ch(RRR(T, α))). (2.3.35) & (2.3.36) & (2.3.37) imply the claimed second equality in (2.3.34). Next since (2.3.34) and Lemma 2.3.35 (g × d)(T) ∈ MorG(G,F,ρ)(GH(A), GH(B)), while since the first equality in (2.3.34) and Lemma 2.3.35 (g × d)(S ◦ T) = (g × d)(S) ◦ (g × d)(T), where ◦ in the right side of the equality is the law of composition in the set of morphisms (cid:3) of G(G, F, ρ), and the statement follows. Remark 2.3.43. According to Cor. 2.3.26, Def. 2.2.2, Cnv. 2.2.7, Def. 2.3.23 and Def. 2.3.24, we have for any D ∈ Cu(H) the following connection between notations (TTTD, ID, βD c , PD, aD, eD) = (TTTGH(D), IGH(D), βGH(D) c , PGH(D), aGH(D), eGH(D)) (ϕϕϕD, AD, ψD, bD, mD, ED) = (ϕϕϕGH(D), AGH(D), ψGH(D), bGH(D), mGH(D), EGH(D)), (µD, uD, HHHD, DD) = (µGH(D), uGH(D), HHHGH(D), DGH(D)) (Γ D, vD, wD, zD) = (Γ GH(D), vGH(D), wGH(D), zGH(D)), (HD, πD, ΩD) = (HGH(D), πGH(D), ΩGH(D)). 108 Moreover for all T ∈ TTTD and α ∈ PT D ST α (D) = ST α (D) = BT BT α (D) = RT α (D) = RRRT RT RRRT α (GH(D)), α (GH(D)), α (GH(D)), α (GH(D)), h ·, · i(D,T,α) = h ·, · i(GH(D),T,α). Finally iT to A = D respectively. α and jT α in Def. 2.2.2 applied to G = GH(D) equal iT α and jT α in Def. 2.3.23 applied According to Def. 2.2.15, Def. 2.2.20, Def. 2.2.21, Def. 2.2.24, Lemma 2.2.26 and Def. 2.2.77 we set Convention 2.3.44. Let D ∈ Cu(H) and B ∈ C0 u(H) set α , ∀T ∈ TTTD, α ∈ PT D, ≔ A(GH(D))T A(D)T α ≔ PGH(D) PD , TTT• OD ≔ OGH(D), TTT• RepD ≔ RepGH(D), VVV(D) ≔ VVV•(GH(D)), tD ≔ tGH(D), ND ≔ NGH(D), eD ≔ eGH(D), ≔ PGH(B) PB , VVV• ZB ≔ ZGH(B), VB ≔ V•(GH(B)). VVV• Clearly A(D)T α equals the C∗−algebra underlying D. Definition 2.3.45. Proposition 2.3.46. moreover C0 u(H) ≔ Ξ(VVV•, GH △ ), GH △ ≔ (GH △ )VVV•. u(H)) GH = GH ↾ Obj(C0 GH △ =(cid:16) GH, (gH × dH) ↾ MorC0 u(H)(cid:17) , Obj(C0 u(H)) = {A ∈ Cu(H) VVV(A) , ∅}, 2.3. THE CANONICAL NUCLEON-FRAGMENT DOUBLET T• 109 and for any A, B ∈ C0 u(H) we have MorC0 u(H)(B, A) = {T ∈ MorCu(H)(B, A) (∀T ∈ VVV(A))(dH(T)(T) ∈ VVV(B))}. Definition 2.3.47. Set EH • ≔ h h TTT•, m•, V• i, GH △ i. Whenever it is clear by the context which group H is involved we use the convention to denote EH • by E•. Now we can state Theorem 2.3.48. (2) let A ∈ Cu(H), T ∈ TTTA and α ∈ PT A, thus (a) for all f ∈ AA (1) E• is a full integer Cu(H)−equivariant stability on VVV•; mA(T, α)(f) = h f(T)(α), ch(RRR(T, α)) i(A,T,α); (b) if A ∈ C0 u(H) and T ∈ VVV(A) then VA(T, α) = ωexp(−((DA)T α )2) ◦ (πA)T α , (c) let B ∈ Cu(H) and T ∈ MorCu(H)(B, A). If A ∈ C0 u(H), T ∈ VVV(A) and (i) if B ∈ C0 u(H) and dH(T)(T) ∈ VVV(B), then VB(cid:16)dH(T)(T), α(cid:17) = VA(T, α) ◦ T, moreover for all l ∈ H and a ∈ B VB(cid:16)bB(l)(dH(T)(T)), α(cid:17) (θ(l)(a)) = VA(T, α)(T(a)); (ii) if for all β ∈ PT β (A)(b) = (Γ A)T A there exist b ∈ BT β , kbk ≤ 1 and b = c+ RT β (A)+ and b ∈ BdH(T)(T) β (B)+ such that (T)(b), then dH(T)(T) ∈ VVV(B). (µA)T β Proof. St.(1) follows since Thm. 2.2.81 and Thm. 2.3.42, st.(2a) follows since the construction of GH, while st.(2b) since (2.2.54). The first equality in st.(2(c)i) follows since st.(2b) applied to B and A, and by (2.3.31) switching A with B. The second equality follows since the first one, st.(1) and Prp. 2.2.64(3). St.(2(c)ii) follows since (cid:3) (2.3.30) switching A with B and by Lemma 2.2.79. Remark 2.3.49. Under the conditions in Thm. 2.3.48(2(c)ii) we deduce that B ∈ u(H), moreover if these conditions are true for all T ∈ VVV(A) then we deduce that C0 T ∈ MorC0 u(H)(B, A). According to the definition of Z in Def. 2.2.40 we can set 110 Definition 2.3.50. Define ∇ ≔ (∇o, ∇m) and XH such that ∇o ≔ Z ◦ GH, ∇m ∈ YT∈MorC0 u(H) Morset(∇o(c(T)), ∇o(d(T))) ∇m(T)(T, f ) ≔ (dH(T)T, α 7→ f (α) ◦ T), ∀T ∈ MorC0 XH ∈ YT∈MorC0 u(H), u(H) YQ∈TTTc(T) Yβ∈PQ c(T) MorCA∗(A(d(T))dH(T)Q β , A(c(T))Q β ), XH(T, Q, β) ≔ T, ∀T ∈ MorC0 u(H), Q ∈ TTTc(T), β ∈ PQ c(T). Proposition 2.3.51. We have that (1) ∇ ∈ Fct(C0 (2) h ∇, XH i is a conjugate action via GH △ . u(H)op, set), Proof. St.(1) follows since Lemma 2.3.35, st.(2) follows since st.(1) and the construc- (cid:3) tion of GH △ . Now we are in the position of stating our Main Theorem 2.3.52 (canonical nucleon-fragment doublet on Cu(H)). E• is an ex- tended full integer Cu(H)−equivariant stability on VVV• via ∇ and XH. In particular there exists a nucleon-fragment doublet on Cu(H) namely the nucleon-fragment doublet T• related to E•. Proof. Since Thm. 2.3.48(1), Prp. 2.3.51(2), Rmk. 2.2.63 and Thm. 2.3.48(2(c)i). (cid:3) Definition 2.3.53. Let E• be called the canonical extended Cu(H)−equivariant stability, and let T• be called the canonical nucleon-fragment doublet on Cu(H). Definition 2.3.54. Let I• denote the field determined by V•, set J• ≔ (I• ◦ GH, ∇m). As a consequence of the main theorem we obtain the properties of equivariance and invariance of T• Corollary 2.3.55. Prp. 2.2.70, Prp. 2.2.72 and Prp. 2.2.76 hold true for T = T•. In particular the next result resolves the equivariant form of the universality claim described in introduction 0.1. Corollary 2.3.56 (Equivariant form of the universality claim). We have (1) m ◦ GH ∈ MorFct(Cu(H)op,set)(QTTT• ◦ GH (2) (1 7→ mA) ∈ MorFct(H,set)(PA TTT• (3) gr ◦ V• ◦ GH ∈ MorFct(C0 (4) (1 7→ gr(VB)) ∈ MorFct(H,set)(PB VVV• , OA), for all A ∈ Cu(H); u(H)op,set)(QVVV• ◦ GH , ZB), for all B ∈ C0 △ , J•); △ , ∆TTT• ◦ GH △ ); u(H); 2.3. THE CANONICAL NUCLEON-FRAGMENT DOUBLET T• 111 (5) V• ◦ GH ∈ YD∈C0 and by setting oD = eD ◦ tD for all D ∈ C0 u(H), we obtain ND(mD(Q, β)) ∩ NA(D)Q β , RepD(mD(Q, β)) u(H) YQ∈VVV(D)Yβ∈PQ o ∈ YD∈C0 u(H) YQ∈VVV(D)Yβ∈PQ DDD DDD oD(Q,β) ◦ joD(Q,β), VD(Q, β) = Ψ− ToD(Q,β) = Q, αoD(Q,β) = β, ∀D ∈ C0 u(H), Q ∈ VVV(D), β ∈ PQ DDD. Proof. Since Thm. 2.3.52, Cor. 2.2.68 and Def. 2.2.66. (cid:3) In part 3 we shall establish the compact equivariant and invariant forms of the universality claim. Part 3 Universality 3.1. Introduction In this part we complete our task started in part 2 of resolving in diverse formula- tions the universality claim described in introduction 0.1, here we provide the compact equivariant and invariant forms. The part is organized as follows. We start section 3.2 by displaying the solution of the equivariant form of the universality claim, i.e. the expanded equivariances of the T•−nucleon phase and T•−fragment state established at the end of part 2. Then in the second main result of the entire work, we resolve the compact equivariant form of the claim by encoding the above properties into two natural transformations. Namely m⋆ between functors from Cu(H)op to Fct(H, set), and v♮ between functors from C0 u(H)op to Fct(H, set), provided that the hypothesis E ensuring the functoriality of the source and target of v♮ holds true. In section 3.3.1 we associate nucleon and fragment masses maps ζT j , and a Terrell-like law νT with any nucleon-fragment doublet T on a category j and κT C, and show how the invariance of these maps follows by the equivariances of the T−nucleon phase and T−fragment state. In section 3.3.2 we establish the T−resolution of the invariant form of the universality claim, representing an equivariant formulation of the result of the previous section. Indeed we introduce the T−nucleon and fragment j and νT j , κT masses µT respectively. Then we establish their universality by showing that µT j and θT are invariant under contravariant action of Ξ(DDD, F) and under action of H. In section 3.3.3 we apply the results of section 3.3.2 when T is the canonical nucleon-fragment doublet on C0 u(H). As a result we obtain, in the third main result of the entire work, the univarsality of the global nucleon and fragment masses and the universality of the global Terrell law, thus resolving the invariant form of the universality claim. j , and the T−Terrell law θT as set-valued extensions of ζT j , λT j and λT 3.2. Natural transformations related to T• In Cor. 2.3.56 we stated the expanded equivariances of the T•−nucleon phase and T•−fragment state, resolving the equivariant form of the universality claim described in introduction 0.1. In Thm. 3.2.28 the main result of this section and the second main result of the entire work, we shall encode these properties in a unique fashion. Shortly we show, modulo a suitable equivalence relation, that the maps m and V• realize natural transformations between functors from the category Cu(H)op to the category Fct(H, set) u(H)op to Fct(H, set) respectively. In the present section we assume fixed two and from C0 locally compact topological groups G and F, a group homomorphism ρ : F → AutGr(G) such that the map (g, f ) 7→ ρ f (g) on G × F at values in G, is continuous, let H denote G ⋊ρ F. For the sake of completeness let us restate Cor. 2.3.56 (1) m ◦ GH ∈ MorFct(Cu(H)op,set)(QTTT• ◦ GH (2) (1 7→ mA) ∈ MorFct(H,set)(PA TTT• (3) gr ◦ V• ◦ GH ∈ MorFct(C0 (4) (1 7→ gr(VB)) ∈ MorFct(H,set)(PB VVV• u(H)op,set)(QVVV• ◦ GH △ , J•); △ , ∆TTT• ◦ GH △ ); , OA), for all A ∈ Cu(H); , ZB), for all B ∈ C0 u(H); 3.2. NATURAL TRANSFORMATIONS RELATED TO T• 115 (5) V• ◦ GH ∈ YD∈C0 u(H) YQ∈VVV(D)Yβ∈PQ DDD ND(mD(Q, β)) ∩ NA(D)Q β , and by setting oD = eD ◦ tD for all D ∈ C0 u(H), we obtain o ∈ YD∈C0 u(H) YQ∈VVV(D)Yβ∈PQ DDD oD(Q,β) ◦ joD(Q,β), RepD(mD(Q, β)) VD(Q, β) = Ψ− ToD(Q,β) = Q, αoD(Q,β) = β, ∀D ∈ C0 u(H), Q ∈ VVV(D), β ∈ PQ DDD. Thm. 3.2.28 organizes in a more concise and elegant form the above properties, but at cost of rearranging the functors modulo a suitable equivalence relation on a subset of TTTA. and A 7→ OA are maps from Obj(Cu(H)) to Obj(Fct(H, set)) it More exactly, since A 7→ PA TTT• is natural to ask if we can arrange them to form the object part of two functors, say M and N, from the category Cu(H)op to the category Fct(H, set), and then to verify if the following claim holds true: A 7→ (1 7→ mA) realizes a natural transformation between M and N. Similarly since B 7→ PB u(H)) to Obj(Fct(H, set))) it VVV• is natural to ask if we can arrange them to form the object part of two functors, say M′ and N′ from the category C0 u(H)op to the category Fct(H, set), and then to verify if the following claim holds true: B 7→ (1 7→ gr(VB)) realizes a natural transformation between the functors M′ and N′. Now the request of the existence of the aforementioned functor M is equivalent to require that for all A, B ∈ Cu(H), T ∈ MorCu(H)(B, A), l ∈ H and T ∈ TTTA and B 7→ ZB are maps from Obj(C0 (bB(l) ◦ dH(T))(T) = (dH(T) ◦ bA(l))(T), while we shall prove in Lemma 3.2.24(1) that for all T ∈ TTT⋄ A (bB(l) ◦ dH(T))(T) B≎ (dH(T) ◦ bA(l))(T), B≎ is a suitable equivalence relation on a subset TTT⋄ B of TTTB. Hence it is clear that where in order to prove our claim we need to pass in a convenient sense to the quotient all A≎'s. The construction culminate in Def. 3.2.25, the involved functors w.r.t. the relations while in Cor. 3.2.26 we prove that the constructed structures realize functors from the categories Cu(H)op and C0 u(H)op to the category Fct(H, set). Finally we succeed in proving our claim by stating in Thm. 3.2.28 that A 7→ (1 7→ mA ♮ )) are natural transformations between the constructed functors where m⋆ and V♮ are m and V after passing in a convenient sense to the quotient w.r.t. the respective equivalence relations. ⋆) and B 7→ (1 7→ gr(VB 116 Convention 3.2.1. In the remaining of this section let A = h A, H, η i such that A ∈ C(H), while by starting from Def. 3.2.13, let B = h B, H, θ i and assume that A, B ∈ Cu(H). If T, Q ∈ TTTA we convein to use the following notation whenever it does not cause confusion T = h T, µµµ, HHH, ζ, f, Γ i and Q = h T′, µµµ′, KKK, ζ′, f ′, ∆ i, where T = h h, ξ, βc, I, ωωω i and T′ = h h′, ξ′, β′ are the domains of the maps ζ and ζ′ respectively, while HHHα = h Hα, πα, Ωα i and KKKβ = h Kβ, υβ, Ψβ i for any α ∈ PT c, I′, ωωω′ i, X and X′ are the sets such that RX and RX′ A and β ∈ PQ A . Definition 3.2.2. Let A and A be the maps on Obj(C(H)) such that A (A), A(A) ∈ QI∈TTTAQα∈PI A P(L(Hα)) and for all T ∈ TTTA and α ∈ PT Definition 3.2.3. Let A (A)T α ≔ πα(A) ∪ UHHHα(H), A ≃ be the relation such that A we have A(A)T α ≔ (A (A)T α )′′. A ≃≔n(T, Q) ∈ TTTA × TTTA (T, µµµ, ζ, f ) = (T′, µµµ′, ζ′, f ′) ∧ (∀α ∈ PT (∃ Vα : Hα → Kα unitary) (υα = ad(Vα) ◦ πα, Ψα = VαΩα, ∆α = ad(Vα)(Γα))o. A) Set [T]A ≃ ≔ {Q T A ≃ Q} and TTT> A ≔ {[T]A ≃ T ∈ TTTA}. Remark 3.2.4. Let T ∈ TTTA and α ∈ PT A then since the bicommutant theorem A(A)T α α . Thus T ∈ VVV(A) implies Γα ∈ α , since the integration in (0.2.7) is w.r.t. the strong operator topology. Moreover is the von Neumann algebra generated by the set A (A)T A(A)T the position f = f ′ is well-set since agrees with (3.2.4). ≔ {T ∈ TTTA (∀α ∈ PT A)(Γα ∈ πα(A)′′)}, Definition 3.2.5. Define TTT⋄ A A≎ ≔ ( moreover [T]A≎ TTT⋆ A TTT A ≔ {Q T ≔ {[T]A≎ A ≃) ∩ (TTT⋄ A), A × TTT⋄ A≎ Q}, T ∈ TTT⋄ A, T ∈ TTT⋄ A}, ≔ {T ∈ TTTA (∀α ∈ PT A)(Γα ∈ A(A)T α )}, A h ≔ ( A ≃) ∩ (TTT A × TTT A), and if A ∈ C0 u(H) set [T]A h ≔ {Q T A h Q}, T ∈ TTT A, TTT A ≔ {[T]A h T ∈ TTT A}, A ≈ ≔ ( A ≃) ∩ (VVV(A) × VVV(A)), ≈ [T]A VVV♮ A ≔ {Q T A ≈ Q}, T ∈ VVV(A) ≔ {[T]A ≈ T ∈ VVV(A)}. 3.2. NATURAL TRANSFORMATIONS RELATED TO T• 117 Lemma 3.2.6. Let T ad(Vα)(A(A)T α ). A ≃ Q and α ∈ PT A then A (A)Q α = ad(Vα)(A (A)T α ) and A(A)Q α = Proof. Let T A ≃ Q and α ∈ PT A. Thus for any h ∈ SG Fωωωα and a ∈ A UKKKα(h)υα(a)Ψα = υα(η(h)a)Ψα = Vαπα(η(h)a)Ωα = VαUHHHα(h)πα(a)Ωα = VαUHHHα(h)V∗ = (ad(Vα) ◦ UHHHα)(h)υα(a)Ψα, αυα(a)Ψα so we obtain since the cyclicity of KKKα (3.2.1) UKKKα = ad(Vα) ◦ UHHHα, and the first equality of the statement follows. The second equality follows since the first one, since the bicommutant theorem and since the continuity of ad(W) w.r.t. the (cid:3) weak operator topology on L(Hα) for any unitary operator W on Hα. A ≃, Proposition 3.2.7. A A≎, h and h, for any T ∈ TTT⋄ A and I ∈ TTT = [I]A u(H) then VVV(A) ⊆ TTT A ≈ are equivalence relations, moreover [T]A = [T]A≎ and ≃ A ⊆ TTT> A. Finally if for any Y ∈ VVV(A), in particular VVV♮ A ⊆ TTT A. A, in particular TTT⋆ A and [Y]A A and TTT A ⊆ TTT> [I]A ≃ A ∈ C0 ≈ = [Y]A ≃ Proof. The first sentence is easy to show, while the first equality follows since the ⊆ [T]A for (cid:3) bicommutant theorem. Next for any T ∈ TTT follows since Lemma 3.2.6. VVV(A) ⊆ TTT any Y ∈ VVV(A) follows since (3.2.2). A follows since Rmk. 3.2.4, while [Y]A ≃ A clearly [T]A ≃ , while [T]A ≃ ⊇ [T]A ⊆ [Y]A ≈ h h Convention 3.2.8. For any T ∈ TTTA we let [T] denote [T]A ≃ and often when it does not cause confusion we let T ≃ Q denote T A ≃ Q. Remark 3.2.9. By using a line similar to the one in the proof of Thm. 2.3.17, by (3.2.4) and taking into account that two unitary equivalent cyclic representations of a C∗−algebra are associated with the same state, we deduce that [T] holds more than one element for any T ∈ TTTA. Lemma 3.2.10. If A, B ∈ Cu(H), T ∈ MorCu(H)(B, A), T ∈ TTTA and α ∈ PT A, then A(B)dH(T)(T) α = A(A)T α . Proof. Since the bicommutant w α (H) = πα(T(B)) ∪ UHHHT the weak operator topology of any subset S of L(Hα). Thus the statement follows since T is (cid:3) surjective and Cor. 2.3.29. theorem we deduce w , where S w α = πα(A) ∪ UHHHα(H) that A(B)dH(T)(T) is the closure w.r.t. and A(A)T α Proposition 3.2.11. If A is unitarily implemented by v and l ∈ H, then bA,v(l) is A ≃)−compatible, moreover bA,v(l)(TTT⋄ A is A hence bA,v(l) ↾ TTT⋄ A and bA,v(l)(TTT A) ⊆ TTT⋄ A) ⊆ TTT A ≃, ( 118 A is ( A≎)−compatible and bA,v(l) ↾ TTT A A A≎, h)−compatible. h, ( A and T ∈ MorCu(H)(B, A), then dH(T) is ( ≃, B hence dH(T) ↾ TTT⋄ and dH(T)(TTT )−compatible. If A ∈ C0 B ∈ C0 u(H) and T ∈ MorC0 A) ⊆ TTT A is ( B ≃)−compatible, moreover dH(T)(TTT⋄ B≎)−compatible and dH(T) ↾ TTT A≎, If in addition A, B ∈ Cu(H) A) ⊆ TTT⋄ B B A h, h A is ( A A u(H) then bA,v(l) ↾ VVV(A) is ( ≈)−compatible, while whenever ≈, A u(H)(B, A) then dH(T) ↾ VVVA is ( ≈, B ≈)−compatible. Proof. Let T B ≃ dH(T)(Q). Let α ∈ PT A ≃ Q and l ∈ H we claim to show that bA,v(l)(T) A ≃ bA,v(l)(Q) and if in ad- dition A, B ∈ Cu(H) that dH(T)(T) A then Vαπα(v(l))Ωα = υα(v(l))Ψα and (ad(Vα) ◦ ad(πα(v(l))))(Γα) = (ad(υα(v(l))) ◦ ad(V))(Γα) = ad(υα(v(l)))(∆α), while clearly A, the case T ∈ TTT⋄ ad(Vα) ◦ πα ◦ T = υα ◦ T, and our claim is proved. Next assume T ∈ TTT A follows similarly. Clearly ad(πα(v(l))) ◦ πα = πα ◦ ad(v(l)), hence ad(πα(v(l)))(A (A)T α ) = A (A)bA,v(l)(T) since the bicommu- tant theorem and since ad(πα(v(l))) is weakly continuous. Therefore bA,v(l)(T) ∈ TTT A. Next dH(T)(T) ∈ TTT B since Lemma 3.2.10. The last sentence of the statement follows since the u(H). (cid:3) first one and Lemma 2.2.80, and since the second one and the definition of MorC0 The following result is fundamental in order to define in Def. 3.2.20 the maps m# and since Cor. 1.1.35 thus ad(πα(v(l)))(A(A)T α ) = A(A)bA,v(l)(T) α α v♮, where # ∈ {>, ⋆, }. Lemma 3.2.12. We have (1) mA is ( (2) if A ∈ C0 A ≃, =)−compatible, i.e. T u(H) then VA is ( A ≃ Q and α ∈ PT Proof. Let T A ≃ Q ⇒ mA(T) = mA(Q), A ≈, =)−compatible. proof of Thm. 1.1.54 we deduce that the next is a commutative diagram A. Since (3.2.1) and following the arguments used in the ad(Vα) L(Hα) / L(Kα) :✈✈✈✈✈✈✈✈✈ Rµµµ KKK,α Rµµµ HHH,α Bωωω,α µµµ , hence since Lemma 2.2.79 we obtain Rµµµ KKK,α x is such that iSα (3.2.2) Next for any x ∈ X if Sα x is the infinitesimal generator of the semigroup UKKKα ◦ ζ ◦ ix ↾ R+ and UHHHα ◦ ζ ◦ ix ↾ R+ respectively then since (3.2.1) we obtain (3.2.3) = ad(Vα) ◦ Rµµµ x and Tα HHH,α. Sα x = VαTα x V∗ α, thus by Cor 1.1.12(2), (0.2.15) and (2.3.4), (3.2.4) f (C(A,Yx∈A supp(ESα x ))) ⊆ R, / O O : 3.2. NATURAL TRANSFORMATIONS RELATED TO T• 119 where A ∈ Pω(X) satisfying Def. 2.3.8(6) applied to T. Since (3.2.3) and Thm. 1.1.14(2) we deduce that (3.2.5) By construction we have (3.2.6) Dζ, f KKK,α = VαDζ, f HHH,αV∗ α. RRR(T, α) =(cid:16)(Bωωω,α,+ RRR(Q, α) =(cid:16)(Bωωω,α,+ µµµ µµµ , Rµµµ , Rµµµ HHH,α), Dζ, f KKK,α), Dζ, f HHH,α, Γα(cid:17) , KKK,α, ∆α(cid:17) . By (3.2.5) and Cor. 1.1.12(3) we have exp(−(Dζ, f (3.2.7) KKK,α)2) = ad(Vα)(exp(−(Dζ, f HHH,α)2)). HHH,α, Dζ, f KKK,α HHH,α respectively, then since (3.2.2) & (3.2.5) & (3.2.7) we obtain for all s0, . . . , s2n ∈ R Next by abuse of language let V, ∆, Γ , RKKK, RHHH, DKKK and DHHH denote Vα, ∆α, Γα, Rµµµ and Dζ, f and a0, . . . , a2n ∈ Bωωω,α,+ KKK,α, Rµµµ µµµ Tr(cid:16)∆ RKKK(a0) exp(−s0D2 KKK)[DKKK, RKKK(a1)] exp(−s1D2 KKK) . . . [DKKK, RKKK(a2n−1)] exp(−s2n−1D2 (Tr ◦ ad(V))(cid:16)Γ RHHH(a0) exp(−s0D2 [DHHH, RHHH(a2n−1)] exp(−s2n−1D2 Tr(cid:16)Γ RHHH(a0) exp(−s0D2 KKK)[DKKK, RKKK(a2n)] exp(−s2nD2 HHH)[DHHH, RHHH(a1)] exp(−s1D2 HHH)[DHHH, RHHH(a2n)] exp(−s2nD2 HHH) . . . HHH)[DHHH, RHHH(a1)] exp(−s1D2 KKK)(cid:17) = HHH)(cid:17) = HHH) . . . HHH)[DHHH, RHHH(a2n)] exp(−s2nD2 Therefore since (3.2.6) and [16, IV.8.ǫ Thm 22, Thm. 21, and IV.7.δ Thm 21] we have [DHHH, RHHH(a2n−1)] exp(−s2n−1D2 HHH)(cid:17). h ·, ch(RRR(Q, α)) iµµµ,ωωω,α = h ·, ch(RRR(T, α)) iµµµ,ωωω,α, and st.(1) follows. St.(2) follows since (3.2.7), the hypothesis υα = ad(Vα) ◦ πα and Thm. (cid:3) 2.3.48(2b) & (2a). Prp. 3.2.11 permits to set the following Definition 3.2.13. Let # ∈ {>, ⋆, }, define bA # to be the map on H and dH # to be the map on MorCu(H) such that for all l ∈ H and T ∈ MorCu(H)(B, A) >(l) : TTT> bA ⋆(l) : TTT⋆ bA bA (l) : TTT A → TTT> A → TTT⋆ A → TTT if in addition A, B ∈ C0 A → VVV♮ ♮ (l) : VVV♮ bA > (T) : TTT> dH ⋆(T) : TTT⋆ dH dH (T) : TTT 7→ [bA(l)(T)]A , ≃ 7→ [bA(l)(T)]A≎, 7→ [bA(l)(T)]A h, A → TTT> A, [T]A ≃ A → TTT⋆ A, [T]A≎ A → TTT A, [T]A u(H) we can set for all l ∈ H and T ∈ MorC0 A → VVV♮ 7→ [bA(l)(T)]A ≈, ♮ (T) : VVV♮ dH A, [T]A ≈ h B, [T]A ≃ B, [T]A≎ B, [T]A h 7→ [dH(T)(T)]B , ≃ 7→ [dH(T)(T)]B≎, 7→ [dH(T)(T)]B h, u(H)(B, A) B, [T]A ≈ 7→ [dH(T)(T)]B ≈. 120 Remark 3.2.14. Since Prp. 3.2.7 we have that bA (l) = bA >(l) ↾ TTT (T) = dH > (T) ↾ TTT ⋆(l) = bA >(l) ↾ TTT⋆ A and dH A, finally if A ∈ C0 ⋆(T) = dH u(H) then bA > (T) ↾ ♮ (l) = TTT⋆ A, while bA (l) ↾ VVV♮ bA A and dH ♮ (T) = dH A and dH (T) ↾ VVV♮ A. Definition 3.2.15. For any q ∈ TTT> A set Pq (∀α ∈ PT A)(KT α (A) = KQ α (A)) we can set for all q ∈ TTT> A to be the set PQ A and α ∈ Pq A , where Q ∈ q. Since T A A ≃ Q ⇒ Kq α(A) ≔ KQ α (A), Q ∈ q. Set s(>) = ∅, s(⋆) = ⋄ and s() = . Let # ∈ {>, ⋆, }, define A# : Obj(Cu(H)) → Obj(Ab) such that A#,A ≔ Yq∈T# A Yα∈Pq A Kq α(A), provided by the pointwise composition, inversion and identity. By abuse of language we shall use the same symbol A#,A to denote the set underlying the group A#,A. Let r# be the map on Obj(Cu(H)) such that rA # : A#,A → AA and for all f ∈ A#,A, Q ∈ TTTs(#) A and α ∈ PQ A Define rA # (f)(Q)(α) ≔ f([Q]A ≃ )(α). ψ# ∈ YD∈Cu(H) Morset(H, AutAb(A#,D)), such that for all l ∈ H and f ∈ A#,A ψA # (l)(f) ≔ cA(l) ◦ f ◦ bA # (l−1), Finally gH # ∈ YL∈MorCu (H) MorAb(A#, d(L), A#, c(L)), such that for all W ∈ MorCu(H)(A, B) and f ∈ A#,A gH # (W)(f) ≔ h H (W) ◦ f ◦ dH # (W). Note that r# is well-defined since Prp. 3.2.7, while ψ# and gH # are well-defined since the second inclusion in (0.2.32) and the standard picture used for the K0−groups. Definition 3.2.16. Define Z♮ : Obj(C0 u(H)) → Obj(set) and Zm ♮ ∈ QT∈MorC0 u(H) Morset(Z♮(c(T)), Z♮(d(T))), such that Z♮ : D 7→ ap∈VVV♮ D Morset(Pp D, A∗ D), 3.2. NATURAL TRANSFORMATIONS RELATED TO T• 121 where AD denotes the C∗−algebra underlying D, and if A, B ∈ C0 then u(H) and T ∈ MorC0 u(H)(B, A) Zm ♮ (T) : Z♮(A) → Z♮(B), (p, g) 7→(cid:18)dH V♮ ∈ YD∈C0 u(H) dH ♮ (T)(p) ♮ (T)(p), P B ∋ α 7→ g(α) ◦ T(cid:19) . MorGr(H, Autset(Z♮(D))), Moreover define such that if A ∈ C0 u(H) and l ∈ H then V♮(A)(l) : (p, g) 7→(cid:18)bA bA ♮ (l)(p) ♮ (l)(p), P A ∋ α 7→ g(α) ◦ η(l−1)(cid:19) . For the remaining of this section let 1 denote the unit element of H. Definition 3.2.17. Let # ∈ {>, ⋆, }. Define P# and Q# o be the maps on Obj(Cu(H)), and Q# m be the map on MorCu(H) such that ≔ (1 7→ TTT# A, bA # ), Q# PA # o(A) ≔ TTT# A, ≔ dH Q# # . m Moreover define P♮, Z♮, and Q♮ such that if A ∈ C0 u(H) then o be the maps on Obj(C0 u(H)), and Q♮ m be the map on MorC0 u(H) ≔ (1 7→ VVV♮ A, bA ♮ ), ≔(cid:16)1 7→ Z♮(A), V♮(A)(cid:17) , PA ♮ ZA ♮ o(A) ≔ VVV♮ A, Q♮ m ≔ dH ♮ . Q♮ m) ∈ Fct(Cu(H)op, set), u(H)op, set), ♮ ) ∈ Fct(C0 > ∈ Fct(H, set), Lemma 3.2.18. We have that (1) (Q> o , Q> (2) (Z♮, Zm (3) PA (4) ZA ♮ (5) ψA > ∈ MorGr(H, AutAb(A>,A)), (6) (A>, gH (7) rA (8) rA (9) gH(T) ◦ rB ∈ Fct(H, set), if A ∈ C0 > ) ∈ Fct(Cu(H), Ab), > = rA > is an injective group morphism, > ◦ ψA >, for all l ∈ H, >(l) = ψA(l) ◦ rA u(H), > ◦ gH > (T), for all T ∈ MorCu(H)(B, A). 122 >(l) maps A>,A into itself since the construction of bA Proof. st.(1) & (2) follow since Lemma 2.3.35, st.(3) & (4) by Thm. 2.3.17. For any l ∈ H the ψA > and by Thm. 2.3.17. > and cA are H−actions by st.(3) and Prp. 2.3.19 respectively, so ψA st.(5) follows. St.(6) follows since (2.3.33) and st.(1). St.(7) is immediate, while st.(8) & (9) (cid:3) follow by direct computation. > is a H−action since bA Since Rmk. 3.2.14 and then following the same argument used in the proof of Lemma 3.2.18 we obtain the following m) ∈ Fct(Cu(H)op, set), o, Q# # ∈ Fct(H, set), # ∈ MorGr(H, AutAb(A#,A)), Lemma 3.2.19. For all # ∈ {⋆, } we obtain (1) (Q# (2) PA (3) ψA (4) (A#, gH (5) rA (6) rA (7) gH(T) ◦ rB # (8) (Q♮ (9) PA ♮ # is an injective group morphism, # ◦ ψA # , for all l ∈ H, m) ∈ Fct(C0 o, Q♮ ∈ Fct(H, set), if A ∈ C0 # ) ∈ Fct(Cu(H), Ab), # (l) = ψA(l) ◦ rA = rA # ◦ gH u(H). # (T), for all T ∈ MorCu(H)(B, A), u(H)op, set), According to Lemma 3.2.12, Lemmas 3.2.18(7) and 3.2.19(5), we can set the following Definition 3.2.20. Let # ∈ {>, ⋆, }, define m# and m# to be the maps on Obj(Cu(H)) such that mA # ∈ Yp∈TTT# A Morset(Pp A, A∗ #,A), mA # (p)(β) ≔ mA(I)(β) ◦ rA # , p ∈ TTT# A, I ∈ p, β ∈ Pp A; mA # ≔ (1 7→ mA # ). Moreover define V♮ and v♮ to be the maps on Obj(C0 section s of the fibered space h VVVA, [·]A ≈, VVV♮ A i u(H)) such that if A ∈ C0 u(H) then for any VA ♮ ∈ Yp∈VVV♮ A Yβ∈Pp A NA(mA(sp)(β)), VA ♮ (p)(β) ≔ VA(sp)(β), p ∈ VVV♮ A, β ∈ Pp A; vA ♮ ≔ (1 7→ gr(VA ♮ )). According to Lemma 3.2.18(5 & 6) and Lemma 3.2.19(3 & 4) we can set the following o and O# be the maps on Obj(Cu(H)), and ∆# Definition 3.2.21. Let # ∈ {>, ⋆, }, define ∆# m be the map on MorCu(H) such that ∆# o(A) ≔ [p∈TTT# A Morset(Pp A, A∗ #,A), 3.2. NATURAL TRANSFORMATIONS RELATED TO T• 123 while for any T ∈ MorCu(H)(B, A) ∆# m(T) : ∆# o(A) → ∆# o(B), Morset(Pp A, A∗ dH # (T)(p) B ∋ β 7→ g(β) ◦ gH # (T)(cid:19) , ∀p ∈ TTT# A. #,A) ∋ g 7→(cid:18)P ψ#,⊲ ∈ YD∈Cu(H) #,A) ∋ g 7→(cid:16)Pp #,⊲(l) : ∆# A o(A) → ∆# ψA A, A∗ MorGr(H, Autset(∆# o(D))) o(A), ∋ β 7→ g(β) ◦ ψA # (l−1)(cid:17) , ∀p ∈ TTT# A, l ∈ H; Moreover define Morset(Pp finally OA # ≔ (1 7→ ∆# o(A), ψA #,⊲). m) ∈ Fct(Cu(H)op, set). follow since Lemma 3.2.19(3 & 4). # ∈ Fct(H, set), o, ∆# Lemma 3.2.22. For all # ∈ {>, ⋆, } we obtain that ψ#,⊲ is well-defined and (1) OA (2) (∆# Proof. The case # = > follows since Lemma 3.2.18(5 & 6), while the cases # ∈ {⋆, } (cid:3) A. We say that the hypothesis E(T, l, T, α) holds true if (ad ◦ πα ◦ T ◦ vB)(l) ↾ U(Hα) = (ad ◦ πα ◦ vA)(l) ↾ U(Hα), while the hypothesis E(T, l) holds true if the hypothesis E(T, l, I, β) holds true for all I ∈ TTTA and β ∈ PI A, finally we say that the hypothesis E holds true if the hypothesis E(T, l) holds true for all l ∈ H and T ∈ MorC0 Definition 3.2.23. Let T ∈ MorCu(H)(B, A), l ∈ H, T ∈ TTTA and α ∈ PT u(H). ⋆(T) = dH Lemma 3.2.24. For all l ∈ H and T ∈ MorCu(H)(B, A) we have ⋆(l) ◦ dH (1) bB H (T) ◦ cB(l) = cA(l) ◦ h (2) h (3) gH ⋆(T) ◦ ψB ⋆,⊲(l) ◦ ∆⋆ (4) ψB (5) if in addition the hypothesis E(T, l) holds true then we obtain ⋆(l), (T), ⋆(T), m(T) ◦ ψA ⋆(l) = ψA m(T) = ∆⋆ ⋆(T) ◦ bA H ⋆(l) ◦ gH ⋆,⊲(l), (a) bB (b) gH (c) ψB (l) ◦ dH (T) ◦ ψB ,⊲(l) ◦ ∆ (T) = dH (l) = ψA m(T) = ∆ (T) ◦ bA (l), (l) ◦ gH (T), m(T) ◦ ψA ,⊲(l). Proof. Let l ∈ H, T ∈ MorCu(H)(B, A) and T ∈ TTT⋄ A then (3.2.8) (dH(T) ◦ bA(l))(T) = h (Tl)T, µµµl, (HHH(vA,l))T, ζl, f, Γ l HHH,vA i (Tl)T = h l · h, ξ, βc, I, T† ◦ η∗(l) ◦ ωωω i (HHH(vA,l))T : PT Γ l HHH,vA : PT A ∋ α 7→ h Hα, πα ◦ T, πα(vA(l))Ωα i, A ∋ α 7→ ad(cid:16)πα(vA(l))(cid:17) (Γα), 124 while (3.2.9) (bB(l) ◦ dH(T))(T) = h (TT)l, µµµl, (HHHT)(vB,l), ζl, f, Γ l HHHT,vB i (TT)l = h l · h, ξ, βc, I, θ∗(l) ◦ T† ◦ ωωω i (HHHT)(vB,l) : PT Γ l HHHT,vB : PT A ∋ α 7→ h Hα, πα ◦ T, (πα ◦ T)(vB(l))Ωα i, A ∋ α 7→ ad(cid:16)(πα ◦ T)(vB(l))(cid:17) (Γα), moreover T is equivariant by hypothesis so θ∗(l) ◦ T† = T† ◦ η∗(l), thus (3.2.10) (TT)l = (Tl)T. A, thus (HHH(vA,l))T α is a cyclic representation associated with (T† ◦ η∗(l))(ωωωα) and is a cyclic representation associated with the state (θ∗(l)◦T†)(ωωωα) = (T†◦η∗(l))(ωωωα), Let α ∈ PT (HHHT)(vB,l) so there exists a unitary operator Vα on Hα such that α (3.2.11) πα ◦ T = ad(Vα) ◦ (πα ◦ T), Vαπα(vA(l))Ωα = (πα ◦ T)(vB(l))Ωα. Next for all b ∈ B we have ad(cid:16)T(vB(l))(cid:17) (T(b)) = (T ◦ θ(l))(b) = (η(l) ◦ T)(b) = (ad(vA(l))(T(b)), but T is surjective so (3.2.12) in particular ad(cid:16)T(vB(l))(cid:17) = ad(vA(l)), ad(cid:16)(πα ◦ T)(vB(l))(cid:17) ↾ πα(A) = ad(cid:16)πα(vA(l))(cid:17) ↾ πα(A), (3.2.13) so since ad(W) is weakly continuous for any unitary operator W on Hα and π(A)′′ = w πα(A) since the bicommutant theorem , we obtain ad(cid:16)(πα ◦ T)(vB(l))(cid:17) ↾ πα(A)′′ = ad(cid:16)πα(vA(l))(cid:17) ↾ πα(A)′′. Moreover Γα ∈ πα(A)′′ by hypothesis so since (3.2.8) & (3.2.9) (3.2.14) Next Vα ∈ πα(A)′ since T is surjective and the first equality of (3.2.11), therefore HHH,vA,α. Γ l HHHT,vB,α = Γ l (3.2.15) Γ l HHH,vA,α = ad(Vα)(Γ l HHH,vA,α). Finally since (3.2.8) & (3.2.9) & (3.2.10) & (3.2.11) & (3.2.14) & (3.2.15) we obtain (dH(T) ◦ bA(l))(T) B ≃ (bB(l) ◦ dH(T))(T), then st.(1) follows since Prp. 3.2.7 & 3.2.11. St.(2) follows since Rmk. 2.3.32, (2.3.29) and Prp. 2.3.31, and the fact that (K0(·), (·)∗) ◦ ((·)+, (·)+) is a functor from CA∗ to Ab. In 3.2. NATURAL TRANSFORMATIONS RELATED TO T• 125 conclusion st.(3) follows since st.(1) & (2), and st.(4) since st.(3). If the hypothesis E(T, l) holds true then since (3.2.13) and the bicommutant theorem we obtain (3.2.16) ad(cid:16)(πα ◦ T)(vB(l))(cid:17) ↾ A(A)T α = ad(cid:16)πα(vA(l))(cid:17) ↾ A(A)T α , then st.(5a) follows under the same argument used to prove st.(1). St.(5b) & (5c) follow (cid:3) since st.(5a) & (2). Definition 3.2.25. Define PPPH ≔(cid:16)P⋆, MorCu(H)op ∋ T 7→ (1 7→ Q⋆ OOOH ≔(cid:16)O⋆, MorCu(H)op ∋ T 7→ (1 7→ ∆⋆ ≔(cid:16)P♮, MorC0 ≔(cid:16)Z♮, MorC0 u(H)op ∋ T 7→ (1 7→ Q♮ u(H)op ∋ T 7→ (1 7→ Zm m(T))(cid:17) , m(T))(cid:17) , m(T))(cid:17) , ♮ (T))(cid:17) . PPPH ♮ ZZZH ♮ Corollary 3.2.26. We have (1) PPPH ∈ Fct(Cu(H)op, Fct(H, set)); (2) OOOH ∈ Fct(Cu(H)op, Fct(H, set)); (3) If the hypothesis E holds true then u(H)op, Fct(H, set)), u(H)op, Fct(H, set)). ♮ ∈ Fct(C0 ∈ Fct(C0 (a) PPPH (b) ZZZH ♮ Proof. Let T ∈ MorCu(H)(B, A). Since Lemma 3.2.19(1) & (2) to prove st.(1) it is suffi- cient that ⋆, PB ⋆), m(T) is a natural transformation from the functor PA m(T)) ∈ MorFct(H,set)(PA i.e. 1 7→ Q⋆ Lemma 3.2.24(1). Since Lemma 3.2.22(1) & (2) to prove st.(2) it is sufficient that ⋆ to PB (1 7→ Q⋆ ⋆ , which is true since (1 7→ ∆⋆ m(T)) ∈ MorFct(H,set)(OA ⋆, OB ⋆), which is true since Lemma 3.2.24(4). Assume A, B ∈ C0 Lemma 3.2.19(8) & (9) to prove st.(3a) it is sufficient that u(H) and T ∈ MorC0 u(H)(B, A). Since (1 7→ Q♮ m(T)) ∈ MorFct(H,set)(PA ♮ , PB ♮ ), which is true since Lemma 3.2.24(5a) and Rmk. 3.2.14. Since Lemma 3.2.18(2) & (4) to prove st.(3b) it is sufficient that (1 7→ Zm ♮ (T)) ∈ MorFct(H,set)(ZA ♮ , ZB ♮ ), which is true since Lemma 3.2.24(5a) and Rmk. 3.2.14, and since T is equivariant i.e. (cid:3) T ◦ θ(l) = η(l) ◦ T, for all l ∈ H. 126 Remark 3.2.27. Let ΓD denote the space of sections of the fibered space h VVVD, [·]D ≈, VVV♮ D i, for any D ∈ C0 u(H) ΓD since Def. 3.2.20 we have that u(H), then for any s ∈QD∈C0 u(H)Yp∈VVV♮ V♮ ∈ YD∈C0 D Yβ∈Pp D ND(mD(sD p )(β)), vD ♮ = (1 7→ gr(VD ♮ )), ∀D ∈ C0 u(H). Now we are able to state in the second main result of this work that m⋆ and, under the hypothesis E, also v♮ are natural transformations. Since Rmk. 3.2.27 this result represents the compact equivariant form of the universality claim described in introduction 0.1, whose invariant form will be established in Thm. 3.3.26. Theorem 3.2.28 (Compact equivariant form of the universality claim). We have (1) m⋆ ∈ MorFct(Cu(H)op,Fct(H,set))(PPPH, OOOH), (2) if hypothesis E holds true then v♮ ∈ MorFct(C0 u(H)op,Fct(H,set))(PPPH ♮ , ZZZH ♮ ) Proof. St.(1) is well-set since Cor. 3.2.26, moreover it ammounts to be equivalent to the claimed statements (3.2.17) & (3.2.18), where (3.2.17) (1 7→ mA ⋆) ∈ MorFct(H,set)(PA ⋆, OA ⋆), and for all T ∈ MorCu(H)(B, A) the following is a commutative diagram in the category Fct(H, set) (3.2.18) PA ⋆ 17→mA ⋆ OA ⋆ 17→Q⋆ m(T) 17→∆⋆ m(T) PB ⋆ / OB ⋆. 17→mB ⋆ / / / 3.2. NATURAL TRANSFORMATIONS RELATED TO T• 127 Next (3.2.17) equivales to the commutativity in set of the following diagram for all l ∈ H (3.2.19) TTT⋆ A mA ⋆ ∆⋆ o (A) bA ⋆(l) ψA ⋆,⊲(l) TTT⋆ A / ∆⋆ o (A). mA ⋆ Let p ∈ TTT⋆ A, I ∈ p and β ∈ Pp A. Then ⋆,⊲(l) ◦ mA (ψA ⋆)(p)(β) = mA ⋆(l−1) ⋆ ◦ ψA ⋆(p)(β) ◦ ψA = mA(I)(β) ◦ rA ⋆(l−1) = mA(I)(β) ◦ ψA(l−1) ◦ rA ⋆ = mA(bA(l)I)(β) ◦ rA ⋆ = (mA ⋆(l))(p)(β), ⋆ ◦ bA where the third equality follows since Lemma 3.2.19(6) and the fourth one since Thm. 2.3.21(2). Hence (3.2.19) and so also (3.2.17) follow. Next (3.2.18) equivales to the commutativity in set of the following diagram (3.2.20) Next TTT⋆ A dH ⋆ (T) TTT⋆ B mA ⋆ mB ⋆ ∆⋆ o (A) ∆⋆ m(T) / ∆⋆ o (B). (mB ⋆ ◦ dH ⋆(T))(p)(β) = mB ⋆ ([dH(T)I]B≎)(β) = mB(dH(T)I)(β) ◦ rB ⋆ = mA(I)(β) ◦ gH(T) ◦ rB ⋆ = mA(I)(β) ◦ rA = mA ⋆(p)(β) ◦ gH m(T) ◦ mA ⋆ ◦ gH ⋆(T) ⋆)(p)(β), = (∆⋆ ⋆(T) where the third equality follows since the second equality in (2.3.34) (in switching A with B), while the fourth one since Lemma 3.2.19(7). Therefore (3.2.20) and (3.2.18) and / / / / / / 128 then st.(1) follow. Up to the end of the present proof we assume that A, B ∈ C0 u(H). St.(2) is well-set since Cor. 3.2.26, moreover it ammounts to be equivalent to the claimed statements (3.2.21) & (3.2.22), where (3.2.21) (1 7→ gr(VA ♮ )) ∈ MorFct(H,set)(PA ♮ , ZA ♮ ), and for all T ∈ MorC0 Fct(H, set) u(H)(B, A) the following is a commutative diagram in the category (3.2.22) PA ♮ 17→gr(VA ♮ ) ZA ♮ 17→Q♮ m(T) 17→Zm ♮ (T) PB ♮ / ZB ♮ . 17→gr(VB ♮ ) Next (3.2.21) equivales to the commutativity in set of the following diagram for all l ∈ H (3.2.23) VVV♮ A gr(VA ♮ ) Z♮(A) bA ♮ (l) V♮(A)(l) VVV♮ A / Z♮(A), gr(VA ♮ ) ♮ (l))(p), for all p ∈ VVV♮ ♮ ♮ (p) = (VA i.e. η∗(l) ◦ VA A, which follows since Cor. 2.3.56(4), hence (3.2.21) is proved. Next (3.2.22) equivales to the commutativity in set of the following diagram ◦ bA (3.2.24) VVV♮ A gr(VA ♮ ) Z♮(A) dH ♮ (T) Zm ♮ (T) VVV♮ B / Z♮(B), gr(VB ♮ ) which follows since Thm. 2.3.48(2(c)i), therefore (3.2.22) as well st.(2) follow. (cid:3) Remark 3.2.29. Notice that Thm. 3.2.28(1) is equivalent to Cor. 2.3.56(1,2), while under the hypothesis E Thm. 3.2.28(2) is equivalent to Cor. 2.3.56(3,4). Indeed Thm. 3.2.28(1) is equivalent to (3.2.19) and (3.2.20), for all A, B ∈ Cu(H), l ∈ H and T ∈ / / / / / / / / / 3.3. UNIVERSALITY OF THE GLOBAL TERRELL LAW 129 MorCu(H)(B, A), while, in case the hypothesis E holds true, Thm. 3.2.28(2) is equivalent to (3.2.23) and (3.2.24) for all A, B ∈ C0 u(H), l ∈ H and T ∈ MorC0 u(H)(B, A). Remark 3.2.30. Let X = h D, H, J, P i be the canonical standard form of A, φ a faithful semi-finite normal weight on A, h πφ, Hφ, ηφ i the semi-cyclic representation of A associ- ated with φ, [36, Def. 7.1.5], and Y = h πφ(A), Hφ, Jφ, Pφ i the standard form associated to φ realized on H, thus X = Y [36, p. 153]; finally let ∆ denote the modular operator associated with φ. Since the construction of vA and since [36, Lemma 6.1.5(vi), Thm. 7.2.6 and Thm. 9.1.15] we deduce that the following holds • vA(l)P = P; • vA(l) J vA(l−1)∆ 1 2 ηφ(x) = ηφ(x∗), for all x ∈ nφ ∩ n∗ φ. Next let A, B ∈ Cu(H) and T ∈ MorCu(H)(B, A) and let A and B denote the underlying von Neumann algebras of A and B respectively, thus the hypothesis E(T, l) holds true if (1) T(vB(l))P = P; (2) T(vB(l)) J T(vB(l−1))∆ 1 2 ηφ(x) = ηφ(x∗), for all x ∈ nφ ∩ n∗ φ. Note that the requests ad(vA(l))A = A and ad(T(vB(l)))A = A are automatically satisfied since vA(l) ∈ U(A) and vB(l) ∈ U(B) by construction. Now (1) should be not difficult to show, while with the help of (3.2.12) it should be possible to prove (2), but up to now we did not succeed. 3.3. Universality of the global Terrell law Any fissioning system U233 + nth, U235 + nth, Pu239 + nth and C f 252, below referred as as−type, exhibits an asymmetric binary fission consisting in obtaining two asymmetric final fragments. Here the asymmetry of the fragments is with respect to their mass numbers in unified atomic mass units. Let AH and AL be the mass numbers of the heavy In addition any fissioning system up to C f 252 for and light fragments respectively. example Fm258 and Hs266, below referred as s−type, exhibits a symmetric binary fission consisting in obtaining two symmetric final fragments. The nucleon phase hypothesis advanced by Mouze and Ythier states the following, see [26, 27, 33] for details. The reaction-time of any binary fission process is 1.77 10−25s thus occurring at temperatures of the order of 1013K, according to the energy-time uncertainty relation. Hence the distinction between the proton and neutron phase disappears and a new nucleon phase occurs. More exactly whenever a fission process involves an as−type fissioning system ς, two nucleon cores come into existence, one of mass number 82 and the other of mass number 126. The two final asymmetric fragments consist by these two cores surrounded by their valence nucleons, and the closure of the shells at 82 and 126 explains the following Terrell law [33, eq. (1)] (3.3.1) ν = 0.08 (AL − 82) + 0.1 (AH − 126), where ν is the mean value of the prompt-neutron yield in the state describing the fragments obtained next the fission process occurs to ς, in what follows simply called 130 prompt-neutron yield of the fission process occurring to ς. Similarly the reaction involv- ing a s−type fissioning system generates two nucleon cores each one of mass number 126, and the symmetric final fragments consist of such a cores surrounded by their valence nucleons, [33, II.a]. From the information extracted by the experiment we deduce the following stability, the values 82 and 126 remain unchanged under variation of the generating fissioning system. Moreover the hypothesis of existence of nucleon cores implies the thermal nature and phase transition of the nucleon phase: the fission process activates and the cores are generated at temperatures higher than 1013K. The universality claim described in introduction 0.1 has beeen resolved in its equi- variant form in Cor. 2.3.56. Here we shall resolve the invariant form of the claim which establishes the invariance of the global light and heavy nucleon and fragment masses, and the global Terrell law under contravariant action over the field of fission processes of the subcategory C0 u(H) of fissioning systems and their transformations, and under covariant action over the field of fission processes of the symmetry group H. As a result the (restricted global) light and heavy nucleon masses as well the prompt-neutron yield are invariant under essential contravariant action of C0 u(H) and under covariant action of H. Finally under the revised nucleon phase hypothesis the stability of the nucleon masses at the values 82 and 126 and the invariance of (3.3.1) under action of the above It ammounts to apply the below T−resolution of the invari- transformations follow. ant form of the universality claim, to the case where T equals T•. Let #m =light and #w =heavy. j , λT o , PT For any nucleon-fragment doublet T on C, there exist sections µT j , j ∈ {m, w} and θT all invariant under contravariant action of Ξ(DDD, F) and such that each of their values induces an invariant section under covariant action of H. There exists a field PT = (PT m) contravariant under action of the subcategory Ξ(DDD, F) of fissioning systems and their transformations, where PT o (a) is the set of subsets of fission processes occurring to the fissioning system a, for any object a of Ξ(DDD, F). The maps (µT o (a) the set of the T − # j nucleon masses, T − # j fragment masses, and the prompt-neutron yields of the fission precesses belonging to Y respectively. As a result of Prp. 2.2.72 the universality of the T − # j nucleon j , λT and fragment masses, and the universality of the T−Terrell law establishes what follows. µT j and θT are natural transformations whose source functor is PT and the morphism map of their common target functor is the constant map with constant value equal to the identity map on the power set of R, moreover 1 7→ (µT a are morphisms in the category of functors from H to set. In other words a associate with each Y in PT j )a and θT j )a, 1 7→ (λT j )a and 1 7→ θT j )a, (λT (3.3.2) θT d(T) ◦ PT a ◦ τT θT m(T) = θT a(l) = θT c(T), ∀T ∈ MorΞ(DDD,F), a, ∀a ∈ Ξ(DDD, F), l ∈ H, 3.3. UNIVERSALITY OF THE GLOBAL TERRELL LAW 131 similarly for µT T, S ∈ MorΞ(DDD,F) such that d(T) = c(S) j and λT j . The contravariance of PT under action of Ξ(DDD, F) means that for all PT m(T)PT PT o (c(T)) ⊆ PT m(T ◦ S) = PT o (d(T)), m(S) ◦ PT m(T). Next we briefly sketch for the canonical case what above outlined. Let NT• as be the set of the n = h A, T, α, {f j, N j} j∈{m,w} i, A ∩ R+ such that A = h A, H, η i is an object of the category C0 0 , N j ∈ A+ and f j ∈ AA, satisfying consistent physical requests. NT• as models the set of all the possible asymmetric fission processes, later called simply fission processes, in the following sense. For any n ∈ NT• u(H), T ∈ VVV(A), α ∈ PT as as above we have that (1) A is the fissioning system to which the asymmetric fission process n occurs, (2) GH(A) is the nucleon system generated by the fissioning system A, (3) O(A)T α is the fragment system whose observable algebra is A and whose dynam- ics is (εA)T α (−α−1(·)), α is the state of thermal equilibrium at inverse temperature α of O(A)T α , (4) ϕϕϕT (5) T is the operation realizing the fission process n whenever performed on ϕϕϕT α , (6) mA(T, α) is the phase of GH(A), occurring by performing T on ϕϕϕT α , (7) VA(T, α) is the state of O(A)T (8) N j is the observable of O(A)T (9) f j is the observable of GH(A) relative to the mass of the # j nucleon core, (10) κT• (11) ζT• j (n) ≔ VA(T, α)(N j) is the mean value in VA(T, α) of the mass of the # j fragment, j (n) ≔ mA(T, α)(f j) is the mean value in mA(T, α) of the mass of the # j nucleon core. α originated via mA(T, α), α relative to the mass of the # j fragment, It is worthy remarking that for each fissioning system the nucleon phase as well the fragment state is one, while the light and heavy cases are ascribable to the observables fm and fw for the nucleon cores and Nm and Nw for the fragments. Now the restricted global Terrell law generalizing (3.3.1) can be defined in the following way. For any n ∈ NT• as the mean value of the prompt-neutron yield in VA(T, α), said also prompt-neutron yield of the fission process n, equals νT•(n) where (3.3.3) νT• ≔ 0.08(κT• m − ζT• m ) + 0.1(κT• w − ζT• w ). Since by construction fission processes depend by fissioning systems and operations, it is natural to expect that the category C0 as . Indeed for any l ∈ H let Tl = bA(l)(T), fl = η(l)(N j). Moreover let B be a j fissioning system in C0 j} j∈{m,w}, where f′ j ∈ B+ with B the C∗−algebra underlying B, satisfying f j = gH(T)(f′ j) u(H)(B, A) such that there exists x = {f′ u(H)op and the group H act on the set NT• = ψA(l)(f j) and Nl j u(H), T ∈ MorC0 j ∈ AB and N′ j, N′ 132 and N j = T(N′ j). Let TT = dH(T)T, define (3.3.4) n(T,x) ≔ h B, TT, α, {f′ nl ≔ h A, Tl, α, {fl j, N′ j, Nl j} j∈{m,w} i, j} j∈{m,w} i. Thus C0 processes since u(H) acts "essentially" contravariantly and H acts covariantly on the set of fission n(T,x) ∈ NT• as , nl ∈ NT• as . As we shall see later, in order to have a "true" contravariant action of C0 implement a power-set extension of the above transformations. Notice that u(H) we need to (1) Tl is the operation obtained by transforming T through the action of l, (2) fl j is the observable of GH(A), obtained by transforming through the action of l the observable f j, (3) Nl j is the observable of O(A)T α obtained by transforming through the action of l the observable N j, (4) TT is the operation obtained by transforming T through the action of T, (5) f j is the observable of GH(A), obtained by transforming through the action of T the observable f′ j of GH(B), (6) N j is the observable of O(A)T j of O(B)TT α . the observable N′ α obtained by transforming through the action of T Now by applying to T• the properties of invariance of doublets stated in Prp. 2.2.72 we establish for any j ∈ {m, w} the invariance of the restricted global nucleon mass ζT• j , the invariance of the restricted global fragment mass κT• j and the invariance of the restricted global Terrell law νT• under essential contravariant action of the category C0 u(H) and under action of the group H. More exactly (3.3.5) hence (3.3.6) j (n(T,x)) = κT• κT• κT• j (nl) = κT• j (n(T,x)) = ζT• ζT• ζT• j (nl) = ζT• j (n), j (n); j (n), j (n), νT•(n(T,x)) = νT•(n), νT•(nl) = νT•(n). The equalities in (3.3.5) and (3.3.6) are equivalent to the universality of the T−nucleon and fragment masses, and the T−Terrell law respectively stated in (3.3.2) for the case T = T•, if PT• as (A) the set of the fission processes whose underlying fissioning system is A, PT• m (T) extends and modifies the first map defined o (A) is the power set of NT• 3.3. UNIVERSALITY OF THE GLOBAL TERRELL LAW 133 in (3.3.4) by transforming the dependence by x into a set-valed map, (µT• j )A and θT• A are the P(R)−valued extensions to PT• j and νT• to NT• as (A) respectively; finally τT• A (l) is the extension of the restriction of the second map in (3.3.4) to NT• as (A). We remark that provided the hypothesis E, the universality of the global nucleon phase and global fragment state (namely the T•−nucleon phase and T•−fragment state) and hence the universality of the global Terrell law follow by the naturality of the transformations m⋆ and v♮ stated in Thm. 3.2.28. o (A) of the restrictions of ζT• j )A, (λT• j , κT• Now we call # j nucleon and fragment masses, and Terrell law the restriction of ζT• j , κT• j and νT• respectively to the set of all the fission processes n satisfying the revised nucleon phase hypothesis requiring ζT• w (n) = 126 and that H would contain as a subgroup the direct product of the universal covering group of the Poincar´e group with the gauge group of the standard model. Thus under the revised nucleon phase hypothesis as a result of (3.3.5) and (3.3.6) we can state what follows. The light and heavy nucleon masses are invariant with constant values 82 and 126 and the prompt-neutron yield (3.3.1) is invariant under essential contravariant action of C0 u(H) and under action of relativistic transformations of reference frames over the field of fission processes. m (n) = 82, ζT• In section 3.3 we assume fixed two locally compact topological groups G and F, a group homomorphism ρ : F → AutGr(G) such that the map (g, f ) 7→ ρ f (g) on G × F at values in G, is continuous, let H denote G ⋊ρ F. 3.3.1. Invariance of the restricted T−Terrell law. In Def. 3.3.3 we define the re- stricted T−nucleon and fragment masses and the restricted T−Terrell law for any nucleon-fragment doublet T on a category C, and provide their invariance in Prp. 3.3.10, physically interpreted in Prp. 3.3.13. In the present section 3.3.1 let C be a category, and T be h S, J, Z, S, F, m, W, R, DDD, UUU i a fixed but arbitrary nucleon-fragment doublet on C where for simplicity we assume R to be the constant map with constant value equal to H. Convention 3.3.1. For any x ∈ Ξ(DDD, F), T ∈ DDDF(x), α ∈ PT F(x) set A(x, T, α) ≔ A(F(x))T α , and for any T ∈ MorΞ(DDD,F) and Q ∈ DDDF(c(T)), set QT ≔ F2(T)Q. Definition 3.3.2 (Fission processes relative to T). Let NT as be the set of the asymmetric fission processes relative to T defined as the set of 4−tuples h a, T, α, {f j, N j} j∈{m,w} i, such that for all j ∈ {m, w} (1) a ∈ Ξ(DDD, F), (2) T ∈ DDDF(a) and α ∈ PT (3) f j ∈ AF(a), (4) N j ∈ A(a, T, α)+, F(a) ∩ R+ 0 , 134 (5) Wa(T, α)(N j) ≥ ma(T, α)(f j), ma(T, α)(fw) ≥ ma(T, α)(fm) ≥ 0, (6) for any T ∈ MorΞ(DDD,F) such that c(T) = a F1(T)−1(f j) , ∅, S(T, T, α)−1(N j) ∩ A(d(T), TT, α)+ , ∅. Let NT be the set of the fission processes relative to T defined as the set of the 3−tuples h a, {Ti, αi}i∈{s,as}, {f j, N j} j∈{m,w} i, such that h a, Tas, αas, {f j, N j} j∈{m,w} i ∈ NT A(a, Tas, αas) and for all j ∈ {m, w} as, Ts ∈ DDDF(a), αs ∈ PTs F(a) satisfying A(a, Ts, αs) = Wa(Ts, αs)(N j) ≥ ma(Ts, αs)(f j), ma(Ts, αs)(fm) = ma(Ts, αs)(fw) = ma(Tas, αas)(fw). Note that by construction with any fission process we can associate the asymmetric fission process extracted from it. : NT : NT as → R and κT j Definition 3.3.3 (Restricted T−nucleon and fragment masses and T−Terrell law). Let #m =light and #w =heavy. For any j ∈ {m, w} let ζT as → R be the j restricted T − # j nucleon mass and the restricted T − # j fragment mass respectively, defined as the maps such that for all n = h a, T, α, {f j, N j} j∈{m,w} i ∈ NT j (n) ≔ ma(T, α)(f j), ζT j (n) ≔ Wa(T, α)(N j). κT as → R be the restricted T−Terrell law such that m) + 0.1(κT νT ≔ 0.08(κT Let νT : NT as we have w − ζT w). m − ζT Remark 3.3.4. For any n = h a, T, α, {f j, N j} j∈{m,w} i ∈ NT as we have νT(n) = 0.08 (Wa(T, α)(Nm) − ma(T, α)(fm)) + 0.1 (Wa(T, α)(Nw) − ma(T, α)(fw)) . Definition 3.3.5. Define kT as the map on H such that kT(l) is a map on NT as such that for all n = h a, T, α, {f j, N j} j∈{m,w} i ∈ NT as we have kT(l)(n) ≔ h a, bF(a)(l)(T), α, { ψF(a)(l)(f j), V(F(a))T α (l)(N j) } j∈{m,w} i. Convention 3.3.6. If n = h a, T, α, {f j, N j} j∈{m,w} i ∈ NT whenever it is not cause of confusion, we let Tl, fl V(F(a))T α (l)(N j) and kT(l)(n) respectively. as, then for any j ∈ {m, w} and l ∈ H j and nl denote bF(a)(l)(T), ψF(a)(l)(f j), j, Nl 3.3. UNIVERSALITY OF THE GLOBAL TERRELL LAW 135 Up to the end of section 3.3.1 let n = h a, T, α, {f j, N j} j∈{m,w} i ∈ NT as. Remark 3.3.7. For all l ∈ H we have (νT ◦ kT(l))(n) = 0.08(cid:16)Wa(Tl, α)(Nl 0.1(cid:16)Wa(Tl, α)(Nl m) − ma(Tl, α)(fl w) − ma(Tl, α)(fl m)(cid:17) + w)(cid:17) . Definition 3.3.8. We say that (n, b, T, x) satisfies the hypothesis S w.r.t. T, if b ∈ Ξ(DDD, F), T ∈ MorΞ(DDD,F)(b, a) and x = {f′ j, N′ j} j∈{m,w} such that for any j ∈ {m, w} (1) f′ (2) N′ j ∈ F1(T)−1(f j), j ∈ S(T, T, α)−1(N j) ∩ A(b, TT, α)+. Let (n, b, T, x) satisfy the hypothesis S w.r.t. T, define (3.3.7) n(T,x) ≔ h b, TT, α, x i. Proposition 3.3.9. We have that (1) νT is a nonnegative map, (2) kT ∈ MorGr(H, Autset(NT as)); (3) for all j ∈ {m, w} we have ma(Tl, α)(fl Wa(Tl, α)(Nl j) = ma(T, α)(f j) j) = Wa(T, α)(N j); (4) let (n, b, T, x) satisfy the hypothesis S w.r.t. T, where x = {f′ j, N′ j} j∈{m,w}, then for all j ∈ {m, w} mb(cid:16)TT, α(cid:17) (f′ Wb(cid:16)TT, α(cid:17) (N′ j) = ma(T, α)(f j), j) = Wa(T, α)(N j), moreover n(T,x) ∈ NT as, and νT(n(T,x)) = 0.08(cid:16)Wb(cid:16)TT, α(cid:17) (N′ 0.1(cid:16)Wb(cid:16)TT, α(cid:17) (N′ m) − mb(cid:16)TT, α(cid:17) (f′ w) − mb(cid:16)TT, α(cid:17) (f′ w)(cid:17) . m)(cid:17) + Proof. St.(1) follows by Def. 3.3.2(5), the equalities in st.(3) and the first two equalities as follows since (cid:3) in st.(4) follow by Prp. 2.2.72; st.(2) follows by st.(3) and (2.2.13). n(T,x) ∈ NT the definition of the morphism set of Ξ(DDD, F) and since the equalities in st.(4). Proposition 3.3.10 (Invariance of the restricted T−nucleon masses, restricted T−fragment masses and restricted T−Terrell law). Let (n, b, T, x) satisfy the hypothe- sis S w.r.t. T, l ∈ H and j ∈ {m, w} then 136 (1) Invariance of the restricted T−nucleon masses and restricted T−fragment masses under essential contravariant action of Ξ(DDD, F) j (n(T,x)) = ζT ζT j (n(T,x)) = κT κT j (n), j (n). (2) Invariance of the restricted T−nucleon masses and restricted T−fragment masses under action of H j ◦ kT(l) = ζT ζT j , κT j ◦ kT(l) = κT j . (3) Invariance of the restricted T−Terrell law under essential contravariant action of Ξ(DDD, F) and under action of H νT(n(T,x)) = νT(n), νT ◦ kT(l) = νT. Proof. Sts.(1,2) follow since Prp. 3.3.9(4,3) respectively, st.(3) since sts.(1,2). (cid:3) Definition 3.3.11. We say that (s, u) is a semantics for T if (s, u) is an interpretation such that for any h a, {T#, α#}#∈{s,as}, {f j, N j} j∈{m,w} i ∈ NT, # ∈ {s, as}, a, b ∈ Ξ(DDD, F), T ∈ DDDF(a), α ∈ PT F(a) and T ∈ MorΞ(DDD,F)(b, a) we have (1) u(T#) ≡ realizing the #ymmetric fission process of the fissioning system u(a), (2) u(Nm) ≡ relative to the light fragment mass, (3) u(Nw) ≡ relative to the heavy fragment mass, (4) u(fm) ≡ relative to the light nucleon core mass, (5) u(fw) ≡ relative to the heavy nucleon core mass, (6) u(S(T, T, α)) ≡ T, (7) u(F1(T)) ≡ T, (8) u(F2(T)) ≡ T. In the present section, let (s, u) be a semantics for T. Prp. 2.2.73 and Def. 3.3.11 justify the following Definition 3.3.12. νT(n) equals the mean value of the prompt neutron-yield in s(Wa(T, α)). Proposition 3.3.13. Let (n, b, T, x) satisfy the hypothesis S w.r.t. T and l ∈ H thus for any j ∈ {m, w} (1) The mean value of the observable relative to the # j nucleon core mass of the nu- cleon system generated by the fissioning system u(b), in the phase occurring by performing on s((ϕϕϕF(b))F2(T)T ) the operation obtained by transforming through the action of T the operation realizing the asymmetric fission process of the fissioning system u(a), equals the mean value of the observable, obtained by transforming through the action of T the observable relative to the # j nucleon α 3.3. UNIVERSALITY OF THE GLOBAL TERRELL LAW 137 core mass of the nucleon system generated by the fissioning system u(b), in the phase of the nucleon system generated by the fissioning system u(a), occur- ring by performing on s((ϕϕϕF(a))T α ) the operation realizing the asymmetric fission process of the fissioning system u(a). (2) The mean value of the observable obtained by transforming through the action of l the observable relative to the # j nucleon core mass, in the phase of the nucleon system generated by the fissioning system u(a), occurring by performing on s((ϕϕϕF(a))Tl α ) the operation obtained by transforming through the action of l the operation realizing the asymmetric fission process of the fissioning system u(a), equals the mean value of the observable relative to the # j nucleon core mass in the phase of the nucleon system generated by the fissioning system u(a), occurring by performing on s((ϕϕϕF(a))T α ) the operation realizing the asymmetric fission process of the fissioning system u(a). (3) The mean value νT(n(T,x)) of the prompt neutron-yield in the state originated via the phase of the nucleon system generated by the fissioning system u(b), occur- ring by performing on s((ϕϕϕF(b))F2(T)T ) the operation obtained by transforming through the action of T the operation realizing the asymmetric fission process of the fissioning system u(a), equals the mean value νT(n) of the prompt neutron- yield in the state originated via the phase of the nucleon system generated by the fissioning system u(a), occurring by performing on s((ϕϕϕF(a))T α ) the operation realizing the asymmetric fission process of the fissioning system u(a). α (4) The mean value νT(nl) of the prompt neutron-yield in the state originated via the phase of the nucleon system generated by the fissioning system u(a), occurring by performing on s((ϕϕϕF(a))Tl α ) the operation obtained by transforming through the action of l the operation realizing the asymmetric fission process of the fissioning system u(a), equals the mean value νT(n) of the prompt neutron-yield in the state originated via the phase of the nucleon system generated by the fissioning system u(a), occurring by performing on s((ϕϕϕF(a))T α ) the operation realizing the asymmetric fission process of the fissioning system u(a). α ) ≡ the state of thermal equilibrium (ϕϕϕF(a))T Here s((ϕϕϕF(a))T α at the inverse temperature α of the fragment system whose observable algebra is A(a, T, α) and whose dynamics is (εF(a))T α (−α−1(·)). Proof. Since Prp. 3.3.10, Prp. 2.2.73 and rearrangements to avoid redundancies, (cid:3) alternatively sts. (1,2) follow since Prp. 2.2.76. The next definition will be used in formulating the nucleon phase hypothesis in section 3.3.4. Definition 3.3.14. l is said to be nucleon equivalent with n if l ∈ NT as and ζT j (l) = ζT j (n) for all j ∈ {m, w}. 138 Remark 3.3.15. For any l = h d, Q, β, {gj, M j} j∈{m,w} i nucleon equivalent with n we have νT(l) = 0.08(cid:16)Wd(Q, β)(Mm) − ma(T, α)(fm)(cid:17) + 0.1(cid:16)Wd(Q, β)(Mw) − ma(T, α)(fw)(cid:17) . 3.3.2. Universality of the T−Terrell law. In order to maintain the present section reasonably independent by the previous one, when feasible we shall not use the results in section 3.3.1. In Def. 3.3.21 and Def. 3.3.22 we define the T−nucleon and fragment masses, and the T−Terrell law, and prove in Thm. 3.3.23 their invariance under con- travariant action of Ξ(DDD, F) and under action of H, by providing the equivariant form of Prp. 3.3.10. This is an essential result representing the T−resolution of the invariant form of the universality claim, since when applied to T• furnishes the invariant form of the claim. Thm. 3.3.23 establishes the existence of invariant sections µT j and θT under contravariant action of Ξ(DDD, F) via the extension of the map defined in (3.3.7) to the power set of NT as. Moreover each value of these sections induces an invariant section under covariant action of H. More exactly we show in Cor. 3.3.18 that there exists a field PT contravariant under action of Ξ(DDD, F) such that its fiber in each a is the power set of the subset NT as(a) of the asymmetric fission processes relative to T whose underlying fis- sioning system equals a; while its morphism map extends and modifies the map defined in (3.3.7) by transforming the dependence by x into a set-valued map. Thus µT j and θT result to be morphisms of the category Fct(Ξ(DDD, F)op, set) whose source functor is PT and such that (µT j , and νT to NT as(a) respectively, and inducing morphisms of Fct(H, set). In the present section 3.3.2 let C be a category, and T = h S, J, Z, S, F, m, W, R, DDD, UUU i be a nucleon-fragment dou- blet on C where for simplicity we assume that R is the constant map with constant value equal to H. Finally up to the end of this part since we will deal only with asymmetric processes, whenever we mention fission processes without any specification we always refer to the asymmetric ones. a are the power set extensions of the restrictions of ζT j , λT j )a, (λT j )a and θT j , κT j , λT Definition 3.3.16. Define NT as(·) : Obj(Ξ(DDD, F)) → P(NT as) such that for all a ∈ Ξ(DDD, F) NT as(a) ≔ {n ∈ NT as Pr 1 (n) = a}, where Pr1 is the map on NT as mapping any 4−tupla into its first element. Moreover define PT o : Obj(Ξ(DDD, F)) ∋ a 7→ P(NT as(a)), and ρT ∈ YT∈MorΞ(DDD,F) Morset(NT as(c(T)), PT o (d(T))), 3.3. UNIVERSALITY OF THE GLOBAL TERRELL LAW 139 such that for any T ∈ MorΞ(DDD,F) and h c(T), T, α, {f j, N j} j∈{m,w} i ∈ NT as(c(T)) ρT(T)(h c(T), T, α, {f j, N j} j∈{m,w} i) ≔ nh d(T), TT, α, {x j, y j} j∈{m,w} i x ∈ Yj∈{m,w} F1(T)−1(f j), y ∈ Yj∈{m,w} S(T, T, α)−1(N j) ∩ A(d(T), TT, α)+o. Finally define PT m ∈ YT∈MorΞ(DDD,F) Morset(PT o (c(T)), PT o (d(T))), such that for all T ∈ MorΞ(DDD,F) and Y ∈ PT o (c(T)) PT m(T)Y ≔[nρT(T)y y ∈ Yo , set PT ≔ (PT o , PT m). In order to prove that PT is a functor we need the following Lemma 3.3.17. (2) ρT(T ◦ S) = PT (1) ρT and PT are well-defined and nonempty set-valued maps; m(S) ◦ ρT(T), for all S, T ∈ MorΞ(DDD,F) such that d(T) = c(S). Proof. St. (1) follows since Prp. 2.2.72, and Def. 3.3.2(6). Let S, T ∈ MorΞ(DDD,F) such as(c(T)) then since Def. 2.2.60(5), that d(T) = c(S) and let h c(T), T, α, {f j, N j} j∈{m,w} i ∈ NT (2.2.43) and ( f ◦ g)−1 = g−1 ◦ f −1 (3.3.8) ρT(T ◦ S)(h c(T), T, α, {f j, N j} j∈{m,w} i) = nh d(S), (TT)S, α, {z j, q j} j∈{m,w} i z ∈ Yj∈{m,w}(cid:16)F1(S)−1 ◦ F1(T)−1(cid:17) (f j), q ∈ Yj∈{m,w}(cid:16)S(S, TT, α)−1 ◦ S(T, T, α)−1(cid:17) (N j) ∩ A(d(S), (TT)S, α)+o. 140 Next (3.3.9) (PT m(S) ◦ ρT(T))(h c(T), T, α, {f j, N j} j∈{m,w} i) = [nρT(S)(h d(T), TT, α, {x j, y j} j∈{m,w} i) x ∈ Yj∈{m,w} y ∈ Yj∈{m,w} S(T, T, α)−1(N j) ∩ A(d(T), TT, α)+o = F1(T)−1(f j), [(cid:26)nh d(S), (TT)S, α, {z j, q j} j∈{m,w} i z ∈ Yj∈{m,w} F1(S)−1(x j), q ∈ Yj∈{m,w} S(S, TT, α)−1(y j) ∩ A(d(S), (TT)S, α)+o (cid:12)(cid:12)(cid:12) x ∈ Yj∈{m,w} F1(T)−1(f j), y ∈ Yj∈{m,w} S(T, T, α)−1(N j) ∩ A(d(T), TT, α)+(cid:27). Next for any j ∈ {m, w} [xj∈F1(T)−1(fj) F1(S)−1(x j) = (F1(S)−1 ◦ F1(T)−1)(f j), (3.3.10) while (3.3.11) [ S(S, TT, α)−1(y j) ∩ A(d(S), (TT)S, α)+ = yj∈S(T,T,α)−1(N j)∩A(d(T),TT,α)+ S(S, TT, α)−1(cid:16)S(T, T, α)−1(N j) ∩ A(d(T), TT, α)+(cid:17) ∩ A(d(S), (TT)S, α)+ = (S(S, TT, α)−1 ◦ S(T, T, α)−1)(N j) ∩ S(S, TT, α)−1(A(d(T), TT, α)+) ∩ A(d(S), (TT)S, α)+ = (S(S, TT, α)−1 ◦ S(T, T, α)−1)(N j) ∩ A(d(S), (TT)S, α)+, where we considered that L−1(B+) ⊇ A+ for any L ∈ MorCA∗(A, B), that f −1(∪x∈Xax) = ∪x∈X f −1(ax), f −1(∩x∈Xax) = ∩x∈X f −1(ax), and ∪x∈X(ax ∩ b) = (∪x∈Xax) ∩ b for any map f : A → B, subset {ax x ∈ X} of P(B) and b ∈ P(B). St. (2) follows since (3.3.8), (3.3.9) [8, (35) (cid:3) Fascicule de resultats; Pr. 9, II.36], and (3.3.10), (3.3.11). Corollary 3.3.18. PT ∈ Fct(Ξ(DDD, F)op, set) whose object map takes nonempty set values. 3.3. UNIVERSALITY OF THE GLOBAL TERRELL LAW 141 Proof. Let S, T ∈ MorΞ(DDD,F) such that d(T) = c(S) and Y ∈ PT o (c(T)), then since Lemma 3.3.17(2) we have (PT m(S) ◦ PT m(T))Y = PT m(S)(cid:16)[{ρT(T)(y) y ∈ Y}(cid:17) =[{(PT =[{ρT(T ◦ S)(y) y ∈ Y} = PT m(S) ◦ ρT(T))(y) y ∈ Y} m(T ◦ S)Y. hence the statement follows since the second assertion of Lemma 3.3.17(1). (cid:3) Prp. 3.3.9 permits to provide the following Definition 3.3.19. Define τT ∈ Ya∈Ξ(DDD,F) MorGr(H, Autset(PT(a))), τT a(l)Y ≔ {kT(l)(y) y ∈ Y}, ∀a ∈ Ξ(DDD, F), l ∈ H, Y ∈ PT(a). Set for all a ∈ Ξ(DDD, F) QT a ≔ (1 7→ PT(a), τT a). Definition 3.3.20. Let RT be the unique element of Fct(Ξ(DDD, F)op, set) such that RT constant map with constant value equal to P(R), and RT value equal to IdP(R). Let RH be the unique element in Fct(H, set) such that RH and RH m is the constant map with constant value equal to IdP(R). o is the m is the constant map with constant o = 1 7→ P(R) Definition 3.3.21 (T−nucleon masses and T−fragment masses). For any j ∈ {m, w} let MT j : P(NT LT j : P(NT as) → P(R), Y 7→ {ζT as) → P(R), Y 7→ {κT j (y) y ∈ Y}; j (y) y ∈ Y}. Define the T − # j nucleon mass to be the map and define the T − # j fragment mass to be the map Morset(PT(a), RT(a)), (µT j )a ≔ MT j ↾ PT(a), ∀a ∈ Ξ(DDD, F); µT j ∈ Ya∈Ξ(DDD,F) λT j ∈ Ya∈Ξ(DDD,F) Morset(PT(a), RT(a)), (λT j )a ≔ LT j ↾ PT(a), ∀a ∈ Ξ(DDD, F). 142 Definition 3.3.22 (T−Terrell law). Let ΘT : P(NT as) → P(R), Y 7→ {νT(y) y ∈ Y}. Define the T−Terrell law to be the map θT ∈ Ya∈Ξ(DDD,F) Morset(PT(a), RT(a)), θT a ≔ ΘT ↾ PT(a), ∀a ∈ Ξ(DDD, F). Now we are in the position of stating the T−resolution of the invariant form of the universality claim. Theorem 3.3.23 (Universality of the T−nucleon masses, T−fragment masses and T−Terrell law). For any j ∈ {m, w} we have (1) Invariance of the T−nucleon masses and T−fragment masses under contravariant action of Ξ(DDD, F) µT j , λT j ∈ MorFct(Ξ(DDD,F)op,set)(PT, RT). (2) Invariance of the T−nucleon masses and T−fragment masses under action of H. For all a ∈ Ξ(DDD, F) (1 7→ (µT j )a), (1 7→ (λT j )a) ∈ MorFct(H,set)(QT a, RH). (3) Invariance of the T−Terrell law under contravariant action of Ξ(DDD, F) θT ∈ MorFct(Ξ(DDD,F)op,set)(PT, RT), and under action of H. For all a ∈ Ξ(DDD, F) (1 7→ θT a) ∈ MorFct(H,set)(QT a, RH). Remark 3.3.24. Since Cor. 3.3.18 the statements of Thm. 3.3.23 are well-set. Moreover they are equivalent to what follows, for all T ∈ MorΞ(DDD,F), a ∈ Ξ(DDD, F), l ∈ H and j ∈ {m, w} we have (µT j )d(T) ◦ PT (µT j )a ◦ τT m(T) = (µT a(l) = (µT j )c(T), j )a, (λT j )d(T) ◦ PT (λT j )a ◦ τT m(T) = (λT a(l) = (λT j )c(T), j )a, and θT d(T) ◦ PT a ◦ τT θT m(T) = θT c(T), a(l) = θT a. 3.3. UNIVERSALITY OF THE GLOBAL TERRELL LAW 143 Proof of Thm. 3.3.23. Since Prp. 2.2.72 and Lemma 3.3.17(1) we obtain for all T ∈ MorΞ(DDD,F), l ∈ H and j ∈ {m, w} (µT j )d(T) ◦ ρT(T) = {·} ◦ ζT j ↾ NT as(c(T)), j ◦ kT(l) = ζT ζT j , (λT j )d(T) ◦ ρT(T) = {·} ◦ κT j ↾ NT as(c(T)), j ◦ kT(l) = κT κT j , d(T) ◦ ρT(T) = {·} ◦ νT ↾ NT θT as(c(T)), νT ◦ kT(l) = νT. Then for any Y ∈ PT o (c(T)) we have (θT d(T) ◦ PT m(T))(Y) =[{(θT d(T) ◦ ρT(T))(y) y ∈ Y} =[{{νT(y)} y ∈ Y} = {νT(y) y ∈ Y} = θT c(T)(Y); moreover for all a ∈ Ξ(DDD, F) and Z ∈ PT(a) (θT a ◦ τT a(l))(Z) = {(νT ◦ kT(l))(z) z ∈ Z} a(Z), = {νT(z) z ∈ Z} = θT similarly for µ and λ. Hence our claim follows since Rmk. 3.3.24. (cid:3) 3.3.3. Universality of the global Terrell law. The resolution of the equivariant form of the universality claim has been established in Cor. 2.3.56 whose compact form has been stated in Thm. 3.2.28. Here we apply the results of section 3.3.2 to the canonical nucleon-fragment doublet T• in order to establish in Thm. 3.3.26 the invariant form of the universality claim, the main result of this section and the final one of the entire project. It states the invariance of the global nucleon and fragment masses and the invariance of the global Terrell law under contravariant action of C0 u(H) and under covariant action of H, over the field of fission processes. As a result we obtain the invariance of the restricted global Terrell law stated in Cor. 3.3.27, whose physical interpretation as invariance of the prompt-neutron yield is established in Cor. 3.3.28; result that can be obtained in a more direct way as consequence of the corresponding ones in section 3.3.1. Let (s, u) be a semantics for T•. Definition 3.3.25 (Global nucleon masses, global fragment masses and global Terrell law). Let the global nucleon phase and the global fragment state be the T•−nucleon phase and T•−fragment state respectively. Moreover let the (restricted) global # j nucleon mass be the (restricted) T• − # j nucleon mass, let the (restricted) global # j fragment mass be the (restricted) T• − # j fragment mass for any j ∈ {m, w}. Let the (restricted) global Terrell law be the (restricted) T•−Terrell law. 144 for all l ∈ H we have Up to the end of section 3.3.3 let n = h A, T, α, {f j, N j} j∈{m,w} i ∈ NT• as . Since section 3.3.1 (νT• ◦ kT•(l))(n) = 0.08(cid:16)VA(Tl, α)(Nl 0.1(cid:16)VA(Tl, α)(Nl m)(cid:17) + w)(cid:17) . m) − mA(Tl, α)(fl w) − mA(Tl, α)(fl u(H)(B, A) and x = {f′ Moreover if (n, B, T, x) satisfies the hypothesis S w.r.t. T•, then B ∈ C0 MorC0 N j = T(N′ satisfy the hypothesis S w.r.t. T•, where x = {f′ u(H), T ∈ j), and j), for any j ∈ {m, w}, where B is the C∗−algebra underlying B. Let (n, B, T, x) j ∈ B+, f j = gH(T)(f′ j} j∈{m,w} such that f′ j ∈ AB, N′ j, N′ j, N′ j} j∈{m,w} then where TT = dH(T)(T), moreover n(T,x) ∈ NT• as and n(T,x) = h B, TT, α, x i, νT•(n(T,x)) = 0.08(cid:16)VB(cid:16)TT, α(cid:17) (N′ 0.1(cid:16)VB(cid:16)TT, α(cid:17) (N′ m) − mB(cid:16)TT, α(cid:17) (f′ w) − mB(cid:16)TT, α(cid:17) (f′ w)(cid:17) . m)(cid:17) + The next is the third main result of this work, jointly Cor. 2.3.56 resolves the invariant form of the universality claim, in particular it presents the stability of the global nucleon and fragment masses and the invariance of the global Terrell law. As an application in the next section we will obtain the invariance of the Terrell law (3.3.1) and the stability of the nucleon masses at the values 82 and 126. Theorem 3.3.26 (Universality of the global nucleon and fragment masses and the global Terrell law). For any j ∈ {m, w} we have (1) Invariance of the global nucleon masses and global fragment masses under contravariant action of C0 u(H) j , λT• µT• j ∈ MorFct(C0 u(H)op,set)(PT•, RT•). (2) Invariance of the global nucleon masses and global fragment masses under action of H. For all D ∈ C0 u(H) (1 7→ (µT• j )D), (1 7→ (λT• j )D) ∈ MorFct(H,set)(QT• D , RH) (3) Invariance of the global Terrell law under contravariant action of C0 u(H) θT• ∈ MorFct(C0 and under action of H. For all D ∈ C0 u(H)op,set)(PT•, RT•), u(H) (1 7→ θT• D ) ∈ MorFct(H,set)(QT• D , RH). Proof. Since Thm. 2.3.52 and Thm. 3.3.23 applied for T = T•. (cid:3) Alternative proof of Thm. 3.3.26. If hypothesis E holds true then it follows since (cid:3) Thm. 3.2.28 and Rmk. 3.2.29. 3.3. UNIVERSALITY OF THE GLOBAL TERRELL LAW 145 As a result we can state the following invariance whose physical interpretation is explained in Cor. 3.3.28. Corollary 3.3.27 (Invariance of the restricted global nucleon masses, global fragment masses and global Terrell law). Let (n, B, T, x) satisfy the hypothesis S w.r.t. T•, l ∈ H and j ∈ {m, w} then (1) Invariance of the restricted global nucleon masses and restricted global fragment masses under essential contravariant action of C0 u(H) ζT• j (n(T,x)) = ζT• κT• j (n(T,x)) = κT• j (n), j (n). (2) Invariance of the restricted global nucleon masses and restricted global fragment masses under action of H j ◦ kT•(l) = ζT• ζT• j , κT• j ◦ kT•(l) = κT• j . (3) Invariance of the restricted global Terrell law under essential contravariant ac- tion of C0 u(H) and action of H νT•(n(T,x)) = νT•(n), νT• ◦ kT•(l) = νT•. Proof. Since Thm. 3.3.26, alternatively by Thm. 2.3.52 and Prp. 3.3.10. (cid:3) Next we clarify the physical meaning of Cor. 3.3.27 concerning the gobal Terrell law, the part concerning the restricted nucleon and fragment masses follows similarly by applying Prp. 3.3.13. Corollary 3.3.28. Let (n, B, T, x) satisfy the hypothesis S w.r.t. T• and l ∈ H, then (1) The mean value of the prompt neutron-yield in the state originated via the phase of the nucleon system generated by the fissioning system u(B), occurring by performing on s((ϕϕϕB)TT α ) the operation obtained by transforming through the action of T the operation realizing the asymmetric fission process of the fissioning system u(A), equals the mean value of the prompt neutron-yield in the state originated via the phase of the nucleon system generated by the fissioning system u(A), occurring by performing on s((ϕϕϕA)T α ) the operation realizing the asymmetric fission process of the fissioning system u(A). (2) The mean value of the prompt neutron-yield in the state originated via the phase of the nucleon system generated by the fissioning system u(A), occurring by performing on s((ϕϕϕA)Tl α ) the operation obtained by transforming through the action of l the operation realizing the asymmetric fission process of the fissioning system u(A), equals the mean value of the prompt neutron-yield in the state originated via the phase of the nucleon system generated by the fissioning 146 system u(A), occurring by performing on s((ϕϕϕA)T asymmetric fission process of the fissioning system u(A). α ) the operation realizing the α ) ≡ the state of thermal equilibrium (ϕϕϕA)T Here s((ϕϕϕA)T fragment system whose observable algebra is A and whose dynamics is (εA)T where A is the underlying C∗−algebra of A. α at the inverse temperature α of the α (−α−1(·)), Proof. Since Cor. 3.3.27 and Prp. 2.2.73; alternatively since Prp. 3.3.13 applied for (cid:3) T = T•. 3.3.4. Nucleon phase hypothesis revised. In this section we provide the revised version in our setting of the nucleon phase hypothesis advanced by G. Mouze and C. It ammounts to specify firstly the group H as the semidirect Ythier see [26, 27, 33]. product of R4 with a group containing the direct product of the universal covering group of the proper Lorentz group on R4 with the gauge group of the standard model. Secondly to hypothize the existence of a suitable element n ∈ NT• as displaying the values 82 and 126. We apply Cor. 3.3.27 to the nucleon phase hypothesis to obtain in Cor. 3.3.31 the invariance of the Terrell law and the stability of the nucleon masses at the above values. To any asymmetric fission process n = h A, T, α, {f j, N j} j∈{m,w} i relative to T• two asymmetric fragments correspond with mass numbers VA(T, α)(Nm) and VA(T, α)(Nw). However the Terrell law (3.3.1) describes the behaviour of a family of couples of fragments each of them arising by a fissioning system belonging to the set B ≔ {U233 + nth, U235 + nth, Pu239 + nth, C f 252}, moreover the values 82 and 126 appears in it. The hypothesis below formulated fixes the group H, assumes the existence of a fission process n relative to T• such that mA(T, α)(fm) = 82 and mA(T, α)(fw) = 126, and for any ς ∈ B supposes the existence of a fission process relative to T• nucleon equivalent with n. SU(2, C) is the universal covering group of the proper Lorentz group L↑ + on R4, see for example [6, p.121], while F0 = U(1)×SU(2)×SU(3) is the gauge group of the standard model. Moreover let the standard action of SU(2, C) on R4 denote the action defined in [6, eqs. (3.39) − (3.33a)], and let g denote the Lorentz metric tensor on R4. Finally let ı1 and ı2 be the canonical injections of SL(2, C) and F0 in SL(2, C) × F0 respectively, and Prµ(λ) = λµ for any λ ∈ C4 and µ ∈ {1, . . . , 4}. Definition 3.3.29. We call h R, η, K i a Poincar´e triplet if (1) R is a locally compact group; (2) SL(2, C) × F0 is a topological subgroup of R; (3) η : R → AutGr(R4) is a group homomorphism; (4) the map (g, f ) 7→ η f (g) on R4 × R at values in R4, is continuous; (5) η ◦ ı1 and η ◦ ı2 are the standard action of SL(2, C) and the trivial action of F0 on R4 respectively; (6) K = R4 ⋊η R. Definition 3.3.30. B ≔ {U233 + nth, U235 + nth, Pu239 + nth, C f 252}. 3.3. UNIVERSALITY OF THE GLOBAL TERRELL LAW 147 Hypothesis 1 (Revised nucleon phase hypothesis). Let h F, ρ, K i be a Poincar´e triplet, we say that n is associated with h F, ρ, K i if • (1) n ∈ NTK as ; c )T ≤ 10−13; (2) (βA TK • m (n) = 82 and ζ (3) ζ (4) for any ς ∈ B there exists a fission process relative to TK TK • w (n) = 126; • nucleon equivalent with n. Here A and T are the dynamical system and the operation underlying n respectively. We say that the revised nucleon phase hypothesis holds true if there exists a Poincar´e triplet h F, ρ, K i such that the set of the fission processes associated with it is not empty. In such a case we say that h F, ρ, K i satisfies the revised nucleon phase hypothesis. If h F, ρ, K i satisfies the revised nucleon phase hypothesis we call # j nucleon mass, TK • j # j fragment mass and Terrell law (associated with the triplet) the restriction of ζ and νTK have • to the set of the fission processes associated with h F, ρ, K i respectively. Thus we TK • j , κ Corollary 3.3.31 (Invariance of the nucleon masses and Terrell law). Let h F, ρ, H i satisfy the revised nucleon phase hypothesis, n be associated with h F, ρ, H i and let T• denote TH • . Thus for all (B, T, x) such that (n, B, T, x) satisfies the hypothesis S w.r.t. T•, j ∈ {m, w} and all l ∈ H we obtain j (kT•(l)n) = ζT• ζT• = Zj, j (n) = ζT• j (n(T,x)) (3.3.12) where Zm = 82 and Zw = 126, and (3.3.13) νT•(kT•(l)n) = νT•(n) = νT•(n(T,x)) Proof. Since Cor. 3.3.27. = 0.08(cid:16)VA(T, α)(Nm) − 82(cid:17) + 0.1(cid:16)VA(T, α)(Nw) − 126(cid:17) . (cid:3) Clearly similar result follows for any fission process nucleon equivalent with n. Remark 3.3.32. (3.3.12) in particular establishes the stability of the light and heavy nucleon masses at the values 82 and 126 under action of the universal covering group of the Poincar´e group over the field of fission processes, implemented as an action over the operations realizing these processes. As a result these values remain unchanged under relativistic transformations of the reference frame. Moreover the values 82 and 126 result to be stable under essential contravariant action over the field of fission processes of the fissioning system transformations belonging to C0 u(H). (3.3.13) establishes the invariance of the prompt-neutron yield under the above described actions. Remark 3.3.33. Although the global nucleon phase is integer, we emphasize that our goal is not to provide an explanation of why the light and heavy nucleon masses hold exactly the values 82 and 126 respectively. Rather we focus our attention, on 148 how and in this way also why, the universality of the global nucleon phase and global fragment state - namely their reciprocal relashionship and their equivariance under contravariant action of C0 u(H) and under covariant action of H precisely established in the resolution of the equivariant form of the universality claim in Cor. 2.3.56 - renders the values 82 and 126 in the Terrell law, and the Terrell law itself invariant under essential contravariant action of suitable perturbations of fissioning systems and under action of relativistic transformations of reference frames, over the field of fission processes. What is fundamental in our context is the existence of what we called the nucleon system generated by the fissioning system, whose nucleon phases are, as they should be, relativistically covariant, contravariant under action of suitable perturbations, and originating states of fragment systems, regardless of the exact values suitable observables assume when measured in them. Remark 3.3.34 (Experimental testability of the universality of the global nucleon phase). The universality of the global nucleon masses and global Terrell law - in the special form of the stability of the values 82 and 126 occurring in the Terrell law and the invariance of the Terrell law itself mainly under the relativistic transformations described in Cor. 3.3.31 - is experimentally testable and can provide an empirical evidence of the existence and universality of the global nucleon phase m. Remark 3.3.35. Let r ∈ {s, as} then provisionally we can set Tr = h T, µµµ, HHH, ζr, f, Γ i ∈ TTTA with T = h h, ξ, βc, I, ωωω i, satisfying (3.3.14) (Dom(ζr) = R4, f =P4 µ,ν=1 Prµ gµ,ν Prν . A, since f maps R4 into f satisfies (2.3.4) for the position A = X = {1, . . . , 4} for any α ∈ PT R and the support of the resolution of the identity of any selfadjoint operator in a Hilbert space is a subset of R. Hence the request (3.3.14) is well-set. Next ζr : R4 → SG since Def. 2.3.8. Set Tα µ,r is the infinitesimal generator of the one- parameter unitary group UHHHα ◦ ζr ◦iµ on HHHα and iµ : R → R4 such that Prν ◦iµ = δν,µIdR, for all µ, ν ∈ X. Since Prµ(EETα µ,r, (3.3.14) and an application of the functional calculus, see for example [22, Thm. 5.6.19] we deduce that ν,r ν ∈ X}, where iTα r = {Tα ) = Tα r Fωωωβc (3.3.15) Dζr, f HHH,α (A) = 4Xµ,ν=1 Tα µ,r gµ,ν Tα ν,r, where S is the closure of any closable operator S in a Hilbert space. Thus in this case the only difference between the phases mA(Ts, α) and mA(Tas, α) results by the fact that we select, via ζs and ζas, two different sets of infinitesimal generators of commutative subgroups of SG HHH,α (A) in general cannot be considered the mass operator even when G = R4 and ζr(x) = (x, 1F) for any x ∈ R4, indeed it needs not to be positive. . Finally it is worthwhile noting that the selfadjoint operator Dζr, f Fωωωβc ACKNOWLEDGMENTS 149 G Remark 3.3.36. Since (2.2.10) it is clear that m (T, α) is represented by the Chern- Connes character of an entire cocycle depending by the element (T, α), while it is prefer- able to have a unique character associated with the map m . We shall analyze in a future work the possibility of constructing a functor L on G(G, F, ρ), encoding for any object G the data {(T, α) T ∈ TTTG, α ∈ PT to a bivariant Chern-Connes character as defined in [29], whose notation we use in what follows. We require L assigning to any object G of G(G, F, ρ) at least the following data: G } into three C∗−algebras and relating the map m G G • C∗−algebras D and R and a smooth subalgebra R∞ of R, • a group morphism v : AG → K0(D), • i ∈ {0, 1}, p, n ∈ N such that (2 − i)n ≥ p − 1, • z ∈QI∈TTTGQβ∈PI • φ ∈ Ext2n+i p(D; R∞, R∞ +), Ei Λ (R∞ ♮, C♮), G such that for all T ∈ TTTG, α ∈ PT G and f ∈ AG we obtain G m (T, α)(f) = φ⋆(v(f) ⊗D z(T, α)), well-set since the left side is integer by construction, the right side is integer since the index theorem [29, Thm. 6.4.]. Acknowledgments I wish to thank Mauro Cappelli who signaled me the paper [33] of Renato Angelo Ricci on the nucleon phase hypothesis of binary fission advanced by G. Mouze and C. Ythier. Bibliography 1. Antoine, J. P.; Inoue, A. and Trapani, C. Partial ∗−Algebras and Their Operator Realizations, Kluwer Academic Publishers 2002. 2. Artin, M.; Grothendieck, A. and Verdier, J. L. Seminaire Geometrie Algebrique du Bois-Marie 1963-1964. Theorie des topos et cohomologie etale des schemas. Tome I Theorie des topos 3. Berezanskii, Y. M. Selfadjoint operators in spaces of functions of infinitely many variables, American Mathematical Society, Translations of mathematical monograph, Vol. 63, 1986 4. Blackadar, B. K−Theory for Operator Algebras, Cambridge Press, Second edition, 1998 5. Blackadar, B. Operator Algebras Springer 2006 6. Bogolubov, N. N.; Logunov, A. A.; Oksak, A. I. and Todorov, I. T. General Principles of Quantum Field Theory Kluwer Academic Publishers 1990 7. Borceux, F. Handbook of Categorical Algebra 1 Cambridge University Press 1994 8. Bourbaki, N. Theorie des Ensambles Diffusion C.C.L.S. 1970 9. Bourbaki, N. Topologie Generale 1 Chapitres 1 a 4, Diffusion C.C.L.S. 1971 10. Bourbaki, N. Topologie Generale Chapitres 5 a 10, Diffusion C.C.L.S. 1974 11. Bourbaki, N. Topological Vector Spaces, Springer, 1987 12. Bourbaki, N. Integration 1, Springer, 2003 13. Bourbaki, N. Integration 2, Springer, 2003 14. Bratteli, O.; Robinson, D. W. Operator Algebras and Quantum Statistical Mechanics 1 Second edition. Springer Verlag 1987. 15. Bratteli, O.; Robinson, D. W. Operator Algebras and Quantum Statistical Mechanics 2 Second edition. Springer Verlag 1996. 16. Connes, A. Noncommutative Geometry, Academic Press, 1994. 17. Dunford, N.; Schwartz, J. Linear Operators Part I Wiley-Interscience, 1957 18. Dunford, N.; Schwartz, J. Linear Operators Part II Wiley-Interscience, 1963 19. Dunford, N.; Schwartz, J. Linear Operators Part III Wiley-Interscience, 1971 20. Guentner, E.; Higson, N. and Trout, J. Equivariant E−Theory for C∗−Algebras Memoirs of the American Mathematical Society, n◦703, AMS 2000 21. Jaffe, A.; Lesniewski, A. and Osterwalder, K. Quantum K-theory: the Chern character Comm. Math. Phys. 118 (1988), 1-14; 22. Kadison, R.V.; Ringrose, J. R. Fundamentals of the Theory of Operator Algebras Vol 1 and 2 AMS 1997 23. Kashiwara, M., Schapira, P. Categories and Sheaves Grundlehren der mathematischen Wissenschaften 332 Springer 2006 24. Leinster, T. Higher Operads, Higher Categories ArXiv:math/0305049v1 25. Mac Lane, S. Categories for the Working Mathematician Springer second ed. 1998 26. Mouze, G.; Hachem, S. and Ythier, C. Intern. Journal of Modern Physics E, 17, 2240 (2008). 27. Mouze, G.; Hachem, S. and Ythier, C. The nucleon phase hypothesis of binary fission http://arxiv.org/abs/1006.4068v2 28. Naimark, M. A. Normed Algebras Wolters-Noordhoff Publishing Groningen, Third american edition, 1972 29. Nistor, V. A bivariant Chern-Connes character Ann. Math. 138 (1993) 151 152 BIBLIOGRAPHY 30. Palmer, T. W. Banach algebras and the general theory of ∗−algebras. Vol II 31. Pedersen, G. K. C∗−Algebras and their Automorphism Groups Academic Press, 1979 32. Raeburn, I.; Williams, D. P. Morita Equivalence and Continuous-Trace C∗−Algebras, Mathematical Surveys and Monographs, Vol. 60, Americal Mathematical Society, 1998 33. Ricci, R. A. A daring interpretation of binary fission, Europhysics News Vol. 40, No. 5, 2009, pp. 13-16, http://dx.doi.org/10.1051/epn/2009701 34. Rudin, W. Real and Complex Analysis Third Edition McGraw-Hill 1987 35. Silvestri, of unbounded functions equalities for in Banach 60 pp. http://dx.doi.org/10.4064/dm464-0-1 , operators spectral B. Integral spaces, Dissertationes Math. 464 (2009), http://arxiv.org/abs/0804.3069 36. Takesaki, M. Theory of Operator Algebras II, Springer-Verlag, 2003. 37. Wegg-Olsen, N. E. K−theory and C∗−algebras 38. Williams, D. P. Crossed Products of C∗−Algebras, Mathematical Surveys and Monographs, Vol. 134, Americal Mathematical Society, 2007.
1105.0867	3	1105	2012-02-16T20:08:36	Isometries between quantum convolution algebras	[ "math.OA" ]	Given locally compact quantum groups $\G_1$ and $\G_2$, we show that if the convolution algebras $L^1(\G_1)$ and $L^1(\G_2)$ are isometrically isomorphic as algebras, then $\G_1$ is isomorphic either to $\G_2$ or the commutant $\G_2'$. Furthermore, given an isometric algebra isomorphism $\theta:L^1(\G_2) \rightarrow L^1(\G_1)$, the adjoint is a -isomorphism between $L^\infty(\G_1)$ and either $L^\infty(\G_2)$ or its commutant, composed with a twist given by a member of the intrinsic group of $L^\infty(\G_2)$. This extends known results for Kac algebras (although our proofs are somewhat different) which in turn generalised classical results of Wendel and Walter. We show that the same result holds for isometric algebra homomorphisms between quantum measure algebras (either reduced or universal). We make some remarks about the intrinsic groups of the enveloping von Neumann algebras of C$^$-algebraic quantum groups.	math.OA	math	Isometries between quantum convolution algebras Matthew Daws, Hung Le Pham November 15, 2018 Abstract Given locally compact quantum groups G1 and G2, we show that if the convolution al- gebras L1(G1) and L1(G2) are isometrically isomorphic as algebras, then G1 is isomorphic either to G2 or the commutant G′ 2. Furthermore, given an isometric algebra isomorphism θ : L1(G2) → L1(G1), the adjoint is a ∗-isomorphism between L∞(G1) and either L∞(G2) or its commutant, composed with a twist given by a member of the intrinsic group of L∞(G2). This extends known results for Kac algebras (although our proofs are somewhat different) which in turn generalised classical results of Wendel and Walter. We show that the same re- sult holds for isometric algebra homomorphisms between quantum measure algebras (either reduced or universal). We make some remarks about the intrinsic groups of the enveloping von Neumann algebras of C∗-algebraic quantum groups. MSC classification: 16T20, 20G42, 22D99, 46L89, 81R50 (Primary); 46L07, 46L10, 46L51 (Secondary). Keywords: Locally compact quantum group, isometric isomorphism, intrinsic group. 1 Introduction Locally compact quantum groups generalise Kac algebras, and form an abstract generalisation of Pontryagin duality. For a locally compact quantum group G, we shall write L∞(G) for the von Neumann algebraic quantum group, and C0(G) for the (reduced) C∗-algebraic quantum group. As one can move between these algebras, we tend to view them as representing the same object G. Let L1(G) be the "quantum convolution algebra", which is the predual of L∞(G), made into a Banach algebra by using the coproduct. We can alternatively identify L1(G) as a certain closed ideal in C0(G)∗. Notice that even in the classical case, where G is even an abelian locally compact group, the algebra L1(G) does not determine G, as if G is finite, then L1(G) is isomorphic to C( G), the continuous functions on the dual group G, and so L1(G) is isomorphic to L1(H) if and only if G and H are of the same cardinality. However, Wendel's theorem [25] shows that if we take the norm into account, then L1(G) com- pletely determines G. To be precise, if θ : L1(G2) → L1(G1) is an isometric algebra isomorphism, then there is a character χ on G1, a positive constant c > 0, and a continuous group homomor- phism α : G1 → G2 such that θ(f )(s) = cχ(s)f (α(s)) almost everywhere for s ∈ G1. The constant c simply reflects the fact that the Haar measure is only unique up to a constant. This was gener- alised to Fourier algebras by Walter, [24]: here notice that A(G) and A(Gop) are also isometrically isomorphic, where Gop is the opposite group to G, and indeed Walter's theorem shows (amongst other things) that A(G1) and A(G2) are isometrically isomorphic if and only if G1 is isomorphic to either G2 or Gop 2 . 1 The Kac algebra case was shown by De Canni`ere, Enock and Schwartz in [5] (see also [6]). The proof in the Kac algebra case uses that the antipode is bounded, which is no longer true in the locally compact quantum group case. We instead use a characterisation of the unitary antipode through the Haar weight (see [14, Proposition 5.20] and Section 3.1 below). The intuitive idea is to show that an isometric algebra isomorphism must intertwine the unitary antipode, although our actual argument is slightly indirect. Our principle result is that when θ : L1(G2) → L1(G1) is an isometric algebra isomorphism, then there is u, a member of the intrinsic group of L∞(G2), such that x 7→ θ∗(x)u is either a ∗-isomorphism, or an anti-∗-isomorphism, from L∞(G1) to L∞(G2). We briefly study the intrinsic group, and prove that it coincides with the collection of characters of L1(G2), as we expect from Wendel's Theorem. An anti-∗-isomorphism to L∞(G2) can be converted to a ∗-isomorphism to the commutant L∞(G2)′ by composing with x 7→ Jx∗J; the possibility of an anti-∗-isomorphism occurring can of course already be seen in Walter's Theorem. In particular, if L1(G1) and L1(G2) are isometrically isomorphic, then G1 is isomorphic to either G2 or G′ 2. We can easily remove the possibility of G′ 2 occurring by restricting to completely isometric (or even just completely contractive) isomorphisms between L1(G1) and L1(G2), see Section 3.3. Having established the result for L1 algebras, we can prove similar results for quantum measure algebras -- for example, for isometric algebra isomorphisms between the dual spaces C0(G2)∗ and C0(G1)∗. Indeed, we work with some generality, and look at C∗-bialgebras (A, ∆) which admit a surjection π : A → C0(G) which intertwines the coproduct, and such that π∗ identifies L1(G) as an ideal in A∗. This includes the reduced and universal C∗-algebraic quantum groups associated with G. As in the Kac algebra case, we use order properties of A∗∗ to determine L1(G) inside A∗. Our characterisation of such isometric isomorphisms involves the intrinsic group of A∗∗, but we show that this is always canonically isomorphic to the intrinsic group of L∞(G). We finish to showing how, in some sense, the picture becomes clearer by embedding everything into L∞( G), and here the interaction between multipliers and the antipode becomes important (compare with [3]). 1.1 Acknowledgements The second named author would like to acknowledge the financial support of the Marsden Fund (the Royal Society of New Zealand). 2 Locally compact quantum groups We give a quick overview of the theory of locally compact quantum groups. For readable intro- ductions, see [12] or [21]. Our main reference is [14], which is a self-contained account of the C∗-algebraic approach to locally compact quantum groups. We shall however mainly work with von Neumann algebras, for which see [13]. However, this paper is not self-contained, and should be read in conjunction with [14]. Indeed, in a number of places, we shall reference [14], where really we need the obvious von Neumann algebraic version of the required result. See also [16] and [23] for the C∗-algebraic and von Neumann algebraic approaches, respectively. A Hopf-von Neumann algebra is a pair (M, ∆) where M is a von Neumann algebra and ∆ : M → M⊗M is a unital norm ∗-homomorphism which is coassociative: (ι ⊗ ∆)∆ = (∆ ⊗ ι)∆. Then ∆ induces a Banach algebra product on the predual M∗. We shall write the product in M∗ by juxtaposition, so hx, ωω′i = h∆(x), ω ⊗ ω′i (x ∈ M, ω, ω′ ∈ M∗). 2 Similarly, the module actions of M on M∗ will be denoted by juxtaposition. Recall the notion of a normal semi-finite faithful weight ϕ on M (see [20, Chapter VII] for example). We let + ϕ = {x ∈ M + : ϕ(x) < ∞}. nϕ = {x ∈ M : ϕ(x∗x) < ∞}, mϕ = lin{x∗y : x, y ∈ nϕ}, m Then mϕ is a hereditary ∗-subalgebra of M, nϕ is a left ideal, and m+ ϕ is indeed M + ∩ mϕ. We can perform the GNS construction for ϕ, which leads to a Hilbert space H, a dense range map Λ : nϕ → H and a unital normal ∗-representation π : M → B(H) with π(x)Λ(y) = Λ(xy). In future, we shall tend to suppress π. Then Λ(nϕ ∩ n∗ ϕ) is full left Hilbert algebra, and this contains a maximal Tomita algebra (see [20, Section 2, Chapter VI]); denote by Tϕ ⊆ nϕ ∩ n∗ ϕ the inverse image under Λ of this maximal Tomita algebra. Tomita-Takesaki theory gives us the modular conjugation J and the modular automorphism group (σt). Then Tϕ is a ∗-algebra, dense in M for the σ-weak topology, all of whose elements are analytic for (σt). A von Neumann algebraic quantum group is a Hopf-von Neumann algebra (M, ∆) together with faithful normal semifinite weights ϕ, ψ which are left and right invariant, respectively. This means that ϕ(cid:0)(ω ⊗ ι)∆(x)(cid:1) = ϕ(x)h1, ωi, ψ(cid:0)(ι ⊗ ω)∆(y)(cid:1) = ψ(y)h1, ωi (ω ∈ M + + + ∗ , x ∈ m ϕ , y ∈ m ψ ). Using these weights, we can construct an antipode S, which will in general be unbounded. Then S has a decomposition S = Rτ−i/2, where R is the unitary antipode, and (τt) is the scaling group. The unitary antipode R is a normal anti-∗-automorphism of M, and ∆R = σ(R ⊗ R)∆, where σ : M⊗M → M⊗M is the tensor swap map. As R is normal, it drops to an isometric linear map R∗ : M∗ → M∗, which is anti-multiplicative. As usual, we make the canonical choice that ϕ = ψ ◦ R. Let H be the GNS space of ϕ, and let Λ : nϕ → H be the GNS map. There is a unitary W , the fundamental unitary, acting on H ⊗ H (the Hilbert space tensor product) such that ∆(x) = W ∗(1⊗x)W for x ∈ M. The left-regular representation of M∗ is the map λ : ω 7→ (ω ⊗ι)(W ). This is a homomorphism, and the σ-weak closure of λ(M∗) is a von Neumann algebra M . We define a coproduct ∆ on M by ∆(x) = W ∗(1 ⊗ x) W , where W = ΣW ∗Σ (here Σ : H ⊗ H → H ⊗ H is the swap map). Then we can find invariant weights to turn ( M , Λ) into a locally compact quantum group -- the dual group to M. We have the biduality theorem that M = M canonically. As is becoming common, we shall write G for the abstract "object" to be thought of as a locally compact quantum group. We then write L∞(G) for M, L1(G) for M∗, and L2(G) for H. In this paper, we shall often have two quantum groups G1 and G2. Then we shall denote by Si the antipode of Gi, for i = 1, 2, and similarly for Ri, ψi, and so forth. There is of course a parallel C∗-algebraic theory, but we shall introduce this below in Section 4. 2.1 Isomorphisms of quantum groups Definition 2.1. A quantum group isomorphism between G1 and G2 is a normal ∗-isomorphism θ : L∞(G1) → L∞(G2) which intertwines the coproducts. Suppose we have a ∗-isomorphism θ : L∞(G1) → L∞(G2) which intertwines the coproducts. Then, arguing as in [14, Proposition 5.45], θ must intertwine the antipode, the unitary antipode, and the scaling group. As the Haar weights are unique up to a constant, we may actually choose the weights to be intertwined by θ. Hence every object associated to G1 is transfered to G2 by θ. 3 Definition 2.2. A quantum group commutant isomorphism between G1 and G2 is a normal anti- ∗-isomorphism θ : L∞(G1) → L∞(G2) which intertwines the coproducts. The commutant von Neumann algebraic quantum group to G is G′, which has L∞(G′) = L∞(G)′, the commutant of L∞(G), and ∆′(x) = (J ⊗ J)∆(JxJ)(J ⊗ J), for x ∈ L∞(G)′. All the other objects (such as W ′, R′, ϕ′) associated to G′ can be related to those of G using the modular conjugation operator J. See [13, Section 4] for further details. Then, if θ : L∞(G1) → L∞(G2) is a commutant isomorphism, then θ′(x) = Jθ(x)∗J defines a quantum group isomorphism from G1 to G′ 2 = G2; thus we have avoided the terminology "quantum group anti-isomorphism", as this would be misleading in the motivating commutative situation. 2; this motivates our choice of terminology. Notice that if G2 is commutative, then G′ 3 Isometries of convolution algebras Throughout this section, fix two locally compact quantum groups G1 and G2, and let T∗ : L1(G2) → L1(G1) be a linear bijective isometry which is an algebra homomorphism (in short, T∗ is an isometric algebra isomorphism). Then T = (T∗)∗ : L∞(G1) → L∞(G2) is a bijective linear isometry between von Neumann algebras. Kadison studied such maps in [8] (see also [6, Section 5.4]) where it is shown that T (1) is a unitary in L∞(G2) and the map T1 : x 7→ T (x)T (1)∗ is a Jordan ∗-homomorphism. That is, T1(x)∗ = T1(x∗), T1(xy + yx) = T1(x)T1(y) + T1(y)T1(x) (x, y ∈ L∞(G1)). In our situation, we can say more about the unitary T (1). Definition 3.1. Let G = (M, ∆) be a Hopf-von Neumann algebra. The intrinsic group of G is the collection of unitaries u ∈ M with ∆(u) = u ⊗ u. Recall that a character on a Banach algebra is a non-zero multiplicative functional. The following is more than we need, but is of independent interest; it generalises [6, Theorem 3.6.10] (which again makes extensive use of a bounded antipode for a Kac algebra). Recall that M(C0(G)) is the multiplier algebra of C0(G); for further details see Section 4 below. Theorem 3.2. Let G = (M, ∆) be a Hopf-von Neumann algebra. For x ∈ M , the following are equivalent: 1. x is a character of the Banach algebra M∗; 2. x 6= 0 and ∆(x) = x ⊗ x. If G is a locally compact quantum group, then a character x ∈ L∞(G) is a unitary, and so auto- matically x is a member of the intrinsic group of G. Furthermore, x ∈ M(C0(G)) and x ∈ D(S) with S(x) = x∗. The maps L1(G) → L1(G); ω 7→ ωx , xω are isometric automorphisms of the algebra L1(G). Proof. The equivalence of (1) and (2) is an easy calculation. Suppose that G is a locally compact quantum group, and x 6= 0 is such that ∆(x) = x ⊗ x. Suppose also that x ≥ 0; we shall prove that x = 1. The von Neumann algebra which x generates 4 is abelian, and so isomorphic to L∞(K) for some measure space K. Let x be the image of x in L∞(K). We note that as kxk = k∆(x)k = kx ⊗ xk = kxk2, necessarily kxk = 1. Let r ∈ [0, 1], and using the Borel functional calculus, let p = χ[r,1](x). Thus p is the indicator function of the set {k ∈ K : x(k) ≥ r}. The von Neumann algebra generated by x ⊗ x embeds into L∞(K ×K) by sending x⊗x to x⊗ x, which is just the function (k, l) 7→ x(k)x(l). Then χ[r,1](x⊗ x) is the indicator function of the set {(k, l) ∈ K × K : x(k)x(l) ≥ r}. Thus, if χ[r,1](x ⊗ x)(k, l) = 1 then x(k)x(l) ≥ r so certainly x(k) ≥ r (as kxk = 1) and so (p ⊗ 1)(k, l) = 1. It follows that χ[r,1](x ⊗ x) ≤ p ⊗ 1. By the homomorphism property of the Borel functional calculus, ∆(p) = χ[r,1](∆(x)) = χ[r,1](x ⊗ x) ≤ p ⊗ 1. However, we can now appeal to [14, Lemma 6.4] to conclude that p = 0 or p = 1 (as an aside on notation, A as used in [14] is simply L∞(G), see [14, Page 874]). So, we have that χ[r,1](x) = 1 or 0 for every r ∈ [0, 1]. It follows that x = 1. Now let x ∈ L∞(G) be non-zero with ∆(x) = x ⊗ x. As ∆ is a ∗-homomorphism, it follows that ∆(x∗x) = x∗x ⊗ x∗x, and so from the previous paragraph, x∗x = 1. Similarly, xx∗ = 1, so x is a unitary, as required. Then 1 ⊗ x = (x∗ ⊗ 1)∆(x), 1 ⊗ x∗ = ∆(x∗)(x ⊗ 1), and so from (the von Neumann algebraic analogue of) [14, Proposition 5.33] we conclude that x ∈ D(S) with S(x) = x∗. To show that x is a multiplier of C0(G), we adapt an idea from [26, Section 4], which in turn is inspired by [1, Page 441]. We have that W ∈ M(C0(G) ⊗ K(L2(G))), where K(L2(G)) is the compact operators on L2(G), see [14, Section 3.4] or compare [26, Theorem 1.5]. Then x ⊗ 1 = (1 ⊗ x∗)∆(x) = (1 ⊗ x∗)W ∗(1 ⊗ x)W ∈ M(C0(G) ⊗ K(L2(G))), and so x ∈ M(C0(G)) as required. Finally, for ω, ω′ ∈ L1(G), we see that hy, (ωω′)xi = h(x ⊗ x)∆(y), ω ⊗ ω′i = hy, (ωx)(ω′x)i (y ∈ L∞(G)), so the map ω 7→ ωx is an algebra homomorphism, with inverse ω 7→ ωx∗. The case of ω 7→ xω is analogous. We remark that similar results to the above theorem have been obtained independently by Neufang and Kalantar, see Kalantar's thesis, [9, Theorem 3.2.11] and [10, Theorem 3.9]. We hence see that if T∗ : L1(G2) → L1(G1) is an isometric algebra isomorphism, then so is T1,∗ : ω 7→ T∗(T (1)∗ω). For the rest of this section, we shall just assume that actually T (1) = 1. Let p ∈ L∞(G2) be a central projection, and let Tp be the map x 7→ T (x)p. As in [6, Section 5.4], we define Ph = (cid:8)p a central projection in L∞(G2) with Tp an algebra homomorphism(cid:9), Pa = (cid:8)p a central projection in L∞(G2) with Tp an algebra anti-homomorphism (cid:9). Then [6, Lemma 5.4.5] shows that both Pa and Ph have greatest elements, say sa and sh. From [19, Theorem 3.3], there is some p ∈ Pa with 1 − p ∈ Ph, and so sa + sh ≥ 1. The following results are also shown in [6], but we give sketch proofs to verify that the results still hold for locally compact quantum groups. 5 Lemma 3.3. Let x ∈ L∞(G) be a central projection with ∆(x) ≥ x ⊗ x and R(x) = x. Then W (x ⊗ x) = (x ⊗ x)W and ∆(x)(x ⊗ 1) = ∆(x)(1 ⊗ x) = x ⊗ x. Proof. We have that x⊗x = (x⊗x)∆(x) = (x⊗x)W ∗(1⊗x)W , and so (x⊗x)W ∗(1⊗x) = (x⊗x)W ∗. Now we use that ( J ⊗ J)W ( J ⊗ J) = W ∗, see [13, Corollary 2.2]. Thus x ⊗ x = (x ⊗ x)( J ⊗ J)W ( J ⊗ J)(1 ⊗ x)( J ⊗ J)W ∗( J ⊗ J), but JxJ = x∗ = x as x is central and self-adjoint, and Jx J = R(x∗) = R(x) = x by assumption. So x ⊗ x = (x ⊗ x)W (1 ⊗ x)W ∗. Taking adjoints gives x ⊗ x = W (1 ⊗ x)W ∗(x ⊗ x). As W ∗ ∈ L∞(G)⊗L∞( G), we see that W ∗(x ⊗ x) = (x ⊗ 1)W ∗(1 ⊗ x), and so, from above, x ⊗ x = W (x ⊗ x)W ∗(1 ⊗ x) = W (x ⊗ x)W ∗. Thus W (x ⊗ x) = (x ⊗ x)W . Then, arguing similarly, ∆(x)(x ⊗ 1) = W ∗(1 ⊗ x)W (x ⊗ 1) = W ∗(x ⊗ x)W = x ⊗ x. The case of ∆(x)(1 ⊗ x) follows by applying the result to Gop (see [13, Section 4]). Corollary 3.4. Let p, q ∈ L∞(G) be central projections with ∆(p) ≥ p ⊗ p and ∆(q) ≥ q ⊗ q, with R(p) = p and R(q) = q, and with p + q ≥ 1. Then p = 1 or q = 1. Proof. By the lemma, ∆(p)((1 − p) ⊗ p) = ∆(p)(1 ⊗ p) − p ⊗ p = 0, and ∆(q)(q ⊗ (1 − q)) = 0. As 1 − q ≤ p and 1 − p ≤ q, it follows that ∆(p)((1 − p) ⊗ (1 − q)) = 0 and ∆(q)((1 − p) ⊗ (1 − q)) = 0. As ∆(p) + ∆(q) ≥ 1, it follows that (1 − p) ⊗ (1 − q) = 0, so p = 1 or q = 1. Proposition 3.5. Form Sa and Sh as above. Then: 1. (TSh ⊗ TSh)∆1(x) = ∆2(T (x))(Sh ⊗ Sh) for x ∈ L∞(G1); 2. (TSa ⊗ TSa)∆1(x) = ∆2(T (x))(Sa ⊗ Sa) for x ∈ L∞(G1); 3. ∆2(Sh) ≥ Sh ⊗ Sh; 4. ∆2(Sa) ≥ Sa ⊗ Sa. Proof. We prove claims for Sa; the proofs for Sh are easier. The preadjoint of TSa is the map ω 7→ T∗(Saω). Firstly, let ω, ω′ ∈ L1(G2), and calculate h(TSa ⊗ TSa)∆1(x), ω ⊗ ω′i = hx, T∗(Saω)T∗(Saω′)i = h∆2(T (x)), Saω ⊗ Saω′i, which shows (2). As Sa is central, we see that Sa ⊗ Sa ∈ L∞(G2)′⊗L∞(G2)′ ⊆ ∆2(L∞(G2))′. Let q ∈ L∞(G2) be such that ∆2(q) is the central support of Sa ⊗ Sa (so q is the smallest central projection with ∆2(q)(Sa ⊗ Sa) = Sa ⊗ Sa). Then Φ : ∆2(L∞(G2))(Sa ⊗ Sa) → ∆2(L∞(G2)q); ∆2(x)(Sa ⊗ Sa) 7→ ∆2(xq) = ∆2(x)∆2(q), is readily seen to be an isomorphism. Then, for x ∈ L∞(G1), ∆2(Tq(x)) = ∆2(T (x)q) = Φ(cid:0)∆2(T (x))(Sa ⊗ Sa)(cid:1) = Φ(cid:0)(TSa ⊗ TSa)∆1(x)(cid:1). So x 7→ ∆2(Tq(x)) is anti-multiplicative, and so q ∈ Pa. Thus q ≤ Sa, and so ∆2(Sa) ≥ ∆(q) ≥ Sa ⊗ Sa as required. 6 At this point, we can no longer follow [6]. We would like to show that T R1 = R2T (that is, T ′ as defined in the next proposition, is the identity map) but we have to proceed somewhat indirectly. Proposition 3.6. Suppose that the map T ′ = T −1R2T R1 : L∞(G1) → L∞(G1) is a homomor- phism. Then T is either a ∗-homomorphism or an anti-∗-homomorphism. Proof. As the unitary antipode R2 is an anti-∗-homomorphism, it is easy to see that R2(Sh) is a central projection. For x ∈ L∞(G1), TR2(Sh)(x) = T (x)R2(Sh) = R2(cid:0)R2(T (x))Sh(cid:1) = R2(cid:0)T (T ′(R1(x)))Sh(cid:1). As y 7→ T (y)Sh is a homomorphism, it follows that TR2(Sh) is a homomorphism, and so R2(Sh) ≤ Sh. As R2 preserves the order, also Sh ≤ R2(Sh), so Sh = R2(Sh). A similar argument establishes that R(Sa) = Sa. So, combining the previous proposition and corollary, we conclude that either Sh = 1, in which case T is a ∗-homomorphism, or Sh = 0, so Sa = 1, and T is an anti-∗-homomorphism. We are henceforth motivated to study the map T ′ = T −1R2T R1. Notice that this map is normal, and the preadjoint T ′ ∗ is an isometric algebra isomorphism from L1(G2) to itself. 3.1 Characterising the unitary antipode We now study the unitary antipode more closely. For us, an important characterisation of R is the following, given in [14, Proposition 5.20]: R(cid:0)(ψ ⊗ ι)((a∗ ⊗ 1)∆(b))(cid:1) = (ψ ⊗ ι)(cid:0)∆(σψ −i/2(a∗))(σψ −i/2(b) ⊗ 1)(cid:1), where a, b ∈ Tψ. (We shall shortly explain further exactly what this formula means). We are hence motivated to look at the right Haar weights, and how they interact with T . We shall then split L∞(G1) into a direct summand, with T acting as a homomorphism in the first component, and as an anti-homomorphism in the second. Then R1 and R2 will interact well with T on these components, but less well on the cross-terms. However, this "bad interaction" will cancel out if we consider T ′2, for T ′ as defined above. Lemma 3.7. The map L∞(G1)+ → [0, ∞]; x 7→ ψ2(T (x)) is a right-invariant, normal semi-finite faithful weight on L∞(G1), and is hence proportional to ψ1. Proof. As T is a Jordan homomorphism, it restricts to an order isomorphism L∞(G1)+ → L∞(G2)+. Thus we can define ψ = ψ2 ◦ T : L∞(G1)+ → [0, ∞], and it follows that ψ is a faithful weight, and + + ψ = T −1(m ψ2). Thus also mψ = T −1(mψ2). As T is σ-weakly continuous, it is now routine to m establish that ψ is semi-finite, and normal (as T is an order isomorphism on the positive cones). + It remains to check that ψ is right-invariant. For ω ∈ L1(G1)+ and y ∈ m ψ , a simple calculation + ∗ (ω) ≥ 0 and T (y) ∈ m ψ2, it follows ∗ (ω))∆2(T (y)). As T −1 shows that T ((ι ⊗ ω)∆1(y)) = (ι ⊗ T −1 that ψ(cid:0)(ι ⊗ ω)∆(y)(cid:1) = ψ2(cid:0)(ι ⊗ T −1 ∗ (ω))∆2(T (y))(cid:1) = h1, T −1 ∗ (ω)iψ2(T (y)). As T is unital, this shows that ψ is right-invariant. 7 Henceforth, we shall actually assume that that ψ1 = ψ2 ◦ T . Henceforth, using [19, Theorem 3.3], we fix a central projection p ∈ L∞(G2) such that Tp is a homomorphism, and T1−p is an anti-homomorphism. Note that we cannot necessarily assume that p = Sa and 1 − p = Sh. Let q = T −1(p). Lemma 3.8. With p, q as above, we have that q is a central projection in L∞(G1). Then L∞(G1) decomposes as qL∞(G1) ⊕ (1 − q)L∞(G1), L∞(G2) decomposes as pL∞(G2) ⊕ (1 − p)L∞(G2), and under these identifications, T decomposes as Tp ⊕ T1−p. Proof. Let x ∈ L∞(G1). Then T (xq)p = Tp(xq) = Tp(x)Tp(q) = T (x)pT (q)p = T (x)p = Tp(x), and similarly Tp(qx) = Tp(x), and T1−p(qx) = T1−p(xq) = 0. Thus T (xq − qx) = Tp(xq − qx) + T1−p(xq − qx) = Tp(xq) − Tp(qx) = Tp(x) − Tp(x) = 0. So q is central; it is easily seen to be a projection. The remaining claims now follow by simple calculation. This lemma means that, for example, given a ∈ qL∞(G1) and x ∈ L∞(G1), T (ax) = T (cid:0)axq + ax(1 − q)(cid:1) = Tp(a)Tp(x) = Tp(a)T (x) = T (a)Tp(x) = T (a)T (x). Thus we understand T quite well; what is unclear is how T interacts with the unitary antipodes R1 and R2. We can then restrict ψ1 = ψ2 ◦ T to qL∞(G1) and to (1 − q)L∞(G2), say giving ψq As Tp is a ∗-homomorphism, it is clear that Tp gives a bijection from nψq ∗-homomorphism, we have that x ∈ n for the modular automorphism groups, for t ∈ R, we shall let σ2,p similarly for ψ1. if and only if T (x∗) = T (x)∗ ∈ n ψp t = σ t ψ1−q 1 to nψp ψ1−p 2 1 1 and ψ1−q . 2 . As T1−p is an anti- . To ease notation 1 2 and σ2,1−p t ψ1−p = σ t 2 , and Lemma 3.9. The map T intertwines the modular automorphism groups in the following ways: Tp ◦ σ1,q t = σ2,p t ◦ Tp, T1−p ◦ σ1,1−q t = σ2,1−p −t ◦ T1−p (t ∈ R). Proof. As Tp is a ∗-isomorphism between L∞(G1)q and L∞(G2)p, it is standard that it intertwines the modular automorphism group, compare [20, Corollary 1.4, Chapter VIII]. As T1−p is an anti-∗- isomorphism, a variant of the standard argument will show that we get the sign change t 7→ −t. As in [12, Section 6] (see also the C∗-algebraic version in [14, Section 1.5]) we let ψ1 (ω ∈ L1(G1)+(cid:9). ψ1⊗ι = (cid:8)x ∈ (L∞(G1)⊗L∞(G1))+ : (ι ⊗ ω)(x) ∈ m m + + Then m by m + ψ1⊗ι is a hereditary cone in (L∞(G1)⊗L∞(G1))+. Let mψ1⊗ι be the ∗-subalgebra generated + + ψ1⊗ι; this agrees with the linear span of m ψ1⊗ι. There is a linear map (ψ1 ⊗ ι) : mψ1⊗ι → L∞(G1) with h(ψ1 ⊗ ι)(x), ωi = ψ1(cid:0)(ι ⊗ ω)x(cid:1). We then set + nψ1⊗ι = (cid:8)x ∈ L∞(G1)⊗L∞(G1) : x∗x ∈ m ψ1⊗ι(cid:9). This is a left ideal in L∞(G1)⊗L∞(G1), and mψ1⊗ι is the linear span of n∗ ψ1⊗ιnψ1⊗ι. As ψ1 is right-invariant, a simple calculation shows that for a, b ∈ nψ1, we have that ∆(b) ∈ nψ1⊗ι and that a ⊗ 1 ∈ nψ1⊗ι. Thus (a∗ ⊗ 1)∆(b) ∈ mψ1⊗ι, and similarly ∆(a∗)(b ⊗ 1) ∈ mψ1⊗ι. In particular, for a, b ∈ Tψ1, we can make sense of the formula R1(cid:0)(ψ1 ⊗ ι)((a ⊗ 1)∆1(b))(cid:1) = (ψ1 ⊗ ι)(cid:0)∆1(σψ1 −i/2(a))(σψ1 −i/2(b) ⊗ 1)(cid:1). 8 Lemma 3.10. Let a ∈ Tψ1 and b ∈ Tψ1q. Then T (cid:0)(ψ1 ⊗ ι)((b ⊗ 1)∆1(a))(cid:1) = (ψ2 ⊗ ι)(cid:0)(T (b) ⊗ 1)∆2(T (a))(cid:1), T (cid:0)(ψ1 ⊗ ι)(∆1(a)(b ⊗ 1))(cid:1) = (ψ2 ⊗ ι)(cid:0)∆2(T (a))(T (b) ⊗ 1)(cid:1) Proof. Let ω, ω′ ∈ L1(G2). Then, for x ∈ L∞(G1), hx, T∗(ω′)bi = hT (bx), ω′i = hT (b)T (x), ω′i = hx, T∗(ω′T (b))i, using that b ∈ L∞(G1)q. Thus hT (cid:0)(ι ⊗ T∗(ω))((b ⊗ 1)∆1(a))(cid:1), ω′i = h(b ⊗ 1)∆1(a), T∗(ω′) ⊗ T∗(ω)i = h∆1(a), T∗(ω′T (b)) ⊗ T∗(ω)i = hT (a), (ω′T (b))ωi = h(T (b) ⊗ 1)∆2(T (a)), ω′ ⊗ ωi. Hence finally hT (cid:0)(ψ1 ⊗ ι)((b ⊗ 1)∆1(a))(cid:1), ωi = ψ1(cid:0)(ι ⊗ T∗(ω))((b ⊗ 1)∆1(a))(cid:1) = ψ2(cid:0)(ι ⊗ ω)(T (b) ⊗ 1)∆2(T (a))(cid:1) = h(ψ2 ⊗ ι)(cid:0)(T (b) ⊗ 1)∆2(T (a))(cid:1), ωi, as required. Now, as ψ2 is a weight, we have that ψ2(x∗) = ψ2(x) for x ∈ mψ2. We can also verify that (ι⊗ω)(x∗) = (ι⊗ω∗)(x)∗ for x ∈ mψ2⊗ι and ω ∈ L1(G2)+. It follows that (ψ2 ⊗ι)(x∗) = (ψ2⊗ι)(x)∗. As Tψ1 is a ∗-algebra, and T respects the involution, applying this calculation to a∗ and b∗ and then taking the adjoint yields the second claimed equality. Lemma 3.11. Let a ∈ Tψ1 and b ∈ Tψ1(1 − q). Then T (cid:0)(ψ1 ⊗ ι)((b ⊗ 1)∆1(a))(cid:1) = (ψ2 ⊗ ι)(cid:0)∆2(T (a))(T (b) ⊗ 1)(cid:1), T (cid:0)(ψ1 ⊗ ι)(∆1(a)(b ⊗ 1))(cid:1) = (ψ2 ⊗ ι)(cid:0)(T (b) ⊗ 1)∆2(T (a))(cid:1). Proof. As in the previous proof, but now using that b ∈ L∞(G1)(1 − q), we check that for ω, ω′ ∈ L1(G2), we have that T∗(ω′)b = T∗(cid:0)T (b)ω′(cid:1), which leads to T (cid:0)(ι ⊗ T∗(ω))((b ⊗ 1)∆1(a))(cid:1) = (ι ⊗ ω)(cid:0)∆2(T (a))(T (b) ⊗ 1)(cid:1), which gives the first result. The second equality now follows by taking adjoints. Proposition 3.12. As before, let T ′ = T −1R2T R1. If a, b ∈ Tψ1q or a, b ∈ Tψ1(1 − q), we have that T ′(cid:0)(ψ1 ⊗ ι)((b ⊗ 1)∆1(a))(cid:1) = (ψ1 ⊗ ι)((b ⊗ 1)∆1(a)). Proof. Suppose that a, b ∈ Tψ1(1 − q), the other case being analogous. We have that −i/2 (a) ⊗ 1)(cid:1)(cid:1) −i/2 (b))(σ1,1−q T R1(cid:0)(ψ1 ⊗ ι)((b ⊗ 1)∆1(a))(cid:1) = T (cid:0)(ψ1 ⊗ ι)(cid:0)∆1(σ1,1−q −i/2 (b))(cid:1) i/2 T1−p(b))(cid:1) = (ψ2 ⊗ ι)(cid:0)(T1−pσ1,1−q = (ψ2 ⊗ ι)(cid:0)(σ2,1−p i/2 T1−p(a) ⊗ 1)∆2(σ2,1−p −i/2 (a) ⊗ 1)∆2(T1−pσ1,1−q using first Lemma 3.11 (applied to σ1,1−q −i/2 (a) ∈ Tψ1(1 − q)) and then Lemma 3.9. Thus, taking adjoints gives that T R1(cid:0)(ψ1 ⊗ ι)((b ⊗ 1)∆1(a))(cid:1) = (ψ2 ⊗ ι)(cid:0)∆2(σ2,1−p = R2(cid:0)(ψ2 ⊗ ι)(cid:0)(T1−p(b∗) ⊗ 1)∆2(T1−p(a∗))(cid:1)∗(cid:1) = R2T (cid:0)(ψ1 ⊗ ι)(cid:0)∆1(a∗)(b∗ ⊗ 1)(cid:1)∗(cid:1) = R2T (cid:0)(ψ1 ⊗ ι)(cid:0)(b ⊗ 1)∆1(a)(cid:1)(cid:1), −i/2 T1−p(b∗))(σ2,1−p −i/2 T1−p(a∗) ⊗ 1)(cid:1)∗ as required. 9 Proposition 3.13. As before, let T ′ = T −1R2T R1. If a ∈ Tψ1(1 − q) and b ∈ Tψ1q, or vice versa, we have that T ′(cid:0)(ψ1 ⊗ ι)((b ⊗ 1)∆1(a))(cid:1) = (ψ1 ⊗ ι)(∆1(a)(σψ1 T ′(cid:0)(ψ1 ⊗ ι)(∆1(a)(b ⊗ 1))(cid:1) = (ψ1 ⊗ ι)((σψ1 −i (b) ⊗ 1)), i (b) ⊗ 1)∆1(a)). Proof. Suppose that a ∈ Tψ1(1 − q) and b ∈ Tψ1q, so we can follow the previous proof through to get that T R1(cid:0)(ψ1 ⊗ ι)((b ⊗ 1)∆1(a))(cid:1) = (ψ2 ⊗ ι)(cid:0)(σ2,1−p i/2 T1−p(a) ⊗ 1)∆2(σ2,p −i/2Tp(b))(cid:1), where here we remember that b ∈ Tψ1q. Thus T R1(cid:0)(ψ1 ⊗ ι)((b ⊗ 1)∆1(a))(cid:1) = (ψ2 ⊗ ι)(cid:0)∆2(σ2,p i/2Tp(b∗))(σ2,1−p −i/2 T1−p(a∗) ⊗ ι)(cid:1)∗ = R2(cid:0)(ψ2 ⊗ ι)(cid:0)(σ2,p = R2T (cid:0)(ψ1 ⊗ ι)(cid:0)(σ1,q i (Tp(b∗)) ⊗ 1)∆2(T1−p(a∗))(cid:1)∗(cid:1) i (b∗) ⊗ 1)∆1(a∗)(cid:1)∗(cid:1) = R2T (cid:0)(ψ1 ⊗ ι)(cid:0)∆1(a)(σψ1 −i (b) ⊗ 1)(cid:1)(cid:1), as required, using Lemma 3.10. The case when a ∈ Tψ1q and b ∈ Tψ1(1−q) follows similarly. Again, taking adjoints (and remembering that σψ1 −i (b∗) gives the second claimed equality). i (b)∗ = σψ1 Corollary 3.14. We have that T ′2 = ι. Proof. By density, it is enough to verify these identities on elements of the form (ψ1 ⊗ ι)((b ⊗ 1)∆1(a)) for a, b ∈ Tψ1. By linearity, we may suppose that a, b ∈ Tψ1 q or a, b ∈ Tψ1(1 − q), in which case the result follows from Proposition 3.12, or that a ∈ Tψ1q, b ∈ Tψ1(1 − q) or vice versa, in which case the result follows from Proposition 3.13. Finally, we wish to show that T ′ commutes with the scaling group (τt). For this, recall from [14, Proposition 6.8] that ∆1σψ1 t = (σψ1 t ⊗ τ−t)∆1. Proposition 3.15. We have that τtT ′ = T ′τt for each t ∈ R. Proof. Let a ∈ Tψ1(1 − q) and b ∈ Tψ1 q. Then, from Proposition 3.13, τtT ′(cid:0)(ψ1 ⊗ ι)((b ⊗ 1)∆1(a))(cid:1) = τt(ψ1 ⊗ ι)(∆1(a)(σψ1 −i (b) ⊗ 1)). Now, for ω, ω′ ∈ L1(G1), h(ι ⊗ ω ◦ τt)(∆1(a)(σψ1 −i (b) ⊗ 1)), ω′i = h(ι ⊗ τt)∆1(a), σψ1 −i(b)ω′ ⊗ ωi = h(σψ1 = hσψ1 t ⊗ ι)∆1(σψ1 t (cid:0)(ι ⊗ ω)(cid:0)∆1(σψ1 −t(a)), σψ1 −i(b)ω′ ⊗ ωi = h(σψ1 −t−i(b) ⊗ 1)(cid:1)(cid:1), ω′i −t(a))(σψ1 t ⊗ ι)(cid:0)∆1(σψ1 −t(a))(σψ1 −t−i(b) ⊗ 1)(cid:1), ω′ ⊗ ωi Thus also hτt(ψ1 ⊗ ι)((b ⊗ 1)∆1(a)), ωi = ψ1(cid:0)(ι ⊗ ω ◦ τt)(∆1(a)(σψ1 −i (b) ⊗ 1))(cid:1) t (cid:0)(ι ⊗ ω)(cid:0)∆1(σψ1 = ψ1(cid:0)σψ1 = h(ψ1 ⊗ ι)(cid:0)∆1(σψ1 −t(a))(σψ1 −t(a))(σψ1 −t−i(b) ⊗ 1)(cid:1)(cid:1)(cid:1) −t−i(b) ⊗ 1)(cid:1), ωi, 10 and so we conclude that τtT ′(cid:0)(ψ1 ⊗ ι)((b ⊗ 1)∆1(a))(cid:1) = (ψ1 ⊗ ι)(cid:0)∆1(σψ1 −t(a))(σψ1 −t−i(b) ⊗ 1)(cid:1). Similarly, we find that h(ι ⊗ ω ◦ τt)((b ⊗ 1)∆1(a)), ω′i = h(ι ⊗ τt)∆1(a), ω′b ⊗ ωi t ⊗ ι)(cid:0)(σψ1 −t(a)), ω′b ⊗ ωi = h(σψ1 t ⊗ ι)∆1(σψ1 = h(σψ1 −t(b) ⊗ 1)∆1(σψ1 −t(a))(cid:1), ω′ ⊗ ωi. So arguing similarly, T ′τt(cid:0)(ψ1 ⊗ ι)((b ⊗ 1)∆1(a))(cid:1) = T ′(cid:0)(ψ1 ⊗ ι)((σψ1 −t(b) ⊗ 1)∆1(σψ1 −t(a)))(cid:1) = (ψ1 ⊗ ι)(cid:0)∆1(σψ1 −t(a))(σψ1 −t−i(b) ⊗ 1)(cid:1) = τtT ′(cid:0)(ψ1 ⊗ ι)((b ⊗ 1)∆1(a))(cid:1). The same argument works if a ∈ Tψ1q and b ∈ Tψ1(1 − q). Similarly, by using Proposition 3.12, a similar calculation works for a, b ∈ Tψ1 q or a, b ∈ Tψ1(1 − q). By linearity and density, the result follows. 3.2 The main result We are now in a position to state and prove our main result. Theorem 3.16. Let T∗ : L1(G2) → L1(G1) be an isometric algebra isomorphism. Then u = T (1) ∈ L∞(G2) is a member of the intrinsic group, and there is a quantum group isomorphism, or quantum group commutant isomorphism, θ : L∞(G1) → L∞(G2) such that T∗(ω) = θ∗(uω) (ω ∈ L∞(G2)). In particular, G1 is isomorphic to either G2 or G′ 2. Proof. Suppose that the result holds (with u = 1) when T = (T∗)∗ is unital. Then we apply this to T1 to find that T∗(T (1)∗ω) = T1,∗(ω) = θ∗(ω) (ω ∈ L1(G2)), from which the general case follows. So, we may suppose that T is unital. We wish to prove that T is either a ∗-homomorphism, or an anti-∗-homomorphism. Form T ′ = T −1R2T R1. By Corollary 3.14, T ′2 = ι, so R1T ′R1 = R1T −1R2T = T ′−1 = T ′; thus T ′ commutes with R1. Now, T ′(1) = 1 and T ′ it follows that T ′ is either a ∗-homomorphism, or an anti-∗-homomorphism. ∗ is an isometric algebra isomorphism. By Proposition 3.6, as T ′−1R1T ′R1 = If T ′ is a ∗- ι, homomorphism, then Proposition 3.6 now shows that T itself is either a ∗-homomorphism or an anti-∗-homomorphism, as required. If we reverse the roles of G1 and G2, and work with T −1, then the same arguments show that (T −1)′ = T R1T −1R2 is either a ∗-homomorphism or an anti-∗-homomorphism. If it is a ∗- homomorphism, then T −1 (and so T ) is either a ∗-homomorphism or an anti-∗-homomorphism, as required. So, the remaining case is when both T ′ and (T −1)′ are anti-∗-homomorphisms (and, to avoid special cases, by this we mean that T ′ and (T −1)′ are not also ∗-homomorphisms). Then we can consider the map Φ : L∞(G1) → L∞(G′ 1) = L∞(G1)′; x 7→ JT ′(x)∗J, which will be a ∗-isomorphism 11 which intertwines the coproducts. Thus Φ will also intertwine the antipode, the unitary antipode, and in particular the scaling group. The scaling group of L∞(G′ t(x) = Jτ−t(JxJ)J, see [13, Section 4]. So, for x ∈ L∞(G1), 1) is τ ′ JT ′(τt(x))∗J = Φ(τt(x)) = τ ′ t(Φ(x)) = Jτ−t(T ′(x)∗)J. Thus T ′τt = τ−tT ′. However, Proposition 3.15 shows that T ′τt = τtT ′; as T ′ bijects, it follows that τt = ι for all t. So the scaling group of G1 is trivial; arguing with (T −1)′ in place of T ′ shows that the same is true of G2. This does not quite show that G1 and G2 are Kac algebras (see [22, Page 7]) but it does give us enough that we can now easily follow the proof in the Kac algebra case, see [6, Section 5.5]. Indeed, let X = (Tp ⊗ ι)(W1) + (T1−pR1 ⊗ ι)(W ∗ 1 ) ∈ L∞(G2)⊗L∞( G1), where W1 is the fundamental unitary for G1. This makes sense, as both Tp and T1−pR1 are ∗-homomorphisms. Then X is unitary. This follows, as for x, y ∈ L∞(G1), Tp(x)T1−p(y) = T (x)T (y)p(1 − p) = 0. Thus X ∗X = (cid:0)(Tp ⊗ ι)(W ∗ 1 ) + (T1−pR1 ⊗ ι)(W1)(cid:1)(cid:0)(Tp ⊗ ι)(W1) + (T1−pR1 ⊗ ι)(W ∗ 1 )(cid:1) = (Tp ⊗ ι)(1) + (T1−pR1 ⊗ ι)(1) = 1, and similarly XX ∗ = 1. As the scaling group is trivial, the familiar formula for the antipode, [14, Proposition 8.3] or [13, Page 79], becomes R1(cid:0)(ι ⊗ ω)(W1)(cid:1) = (ι ⊗ ω)(W ∗ 1 ) (ω ∈ L1( G1)). Thus, for ω ∈ L1(G2) and ω′ ∈ L1( G1), as R2 1 = ι, h(ω ⊗ ι)(X), ω′i = hTp((ι ⊗ ω′)W1), ωi + hT1−pR1((ι ⊗ ω′)(W ∗ 1 )), ωi = hTp((ι ⊗ ω′)W1), ωi + hT1−p((ι ⊗ ω′)(W1)), ωi = hW1, T∗(ω) ⊗ ω′i = hλ1(T∗(ω)), ω′i. So the map L1(G2) → C0( G1); ω 7→ λ1(T∗(ω)) is a homomorphism, and now a simple calculation shows that (∆2 ⊗ ι)(X) = X13X23. Again, as S2 = R2, we can turn L1(G2) into a Banach ∗-algebra for the involution hx, ω♯i = hR2(x)∗, ωi (x ∈ L∞(G2), ω ∈ L1(G2)), compare with Section 4 below. As X is unitary, [11, Proposition 5.2] shows that λ1T∗ is a ∗- homomorphism. Hence T∗ is a ∗-homomorphism. So, for x ∈ L∞(G1) and ω ∈ L1(G2), hx, T∗(ω)♯i = hR1(x)∗, T∗(ω)i = hT (R1(x))∗, ωi = hT (x), ω♯i = hR2(T (x))∗, ωi, showing that R2T = T R1. Thus T ′ = ι, so T ′ is a ∗-homomorphisms, a contradiction, completing the proof. We remark that a corollary of the proof is that, actually, T ′ = ι all along! 12 3.3 Completely isometric homomorphisms It is increasingly common in non-abelian harmonic analysis to study objects in the category of operator spaces and completely bounded maps; see for example the survey [17]. It is well-known that the transpose mapping is the canonical example of an isometric, but not completely isometric, linear mapping. So we might suspect that a complete isometry cannot give rise to a anti-∗- homomorphism, and this is indeed the case -- this is well-known, but we include a sketch proof for completeness. Theorem 3.17. Let A and B be unital C∗-algebras, and let T : A → B be completely isometric bijection. Then T (1) is a unitary, and the map A → B; a 7→ T (a)T (1)∗ is a ∗-homomorphism. Proof. By Kadison, T (1) is unitary, and S : a 7→ T (a)T (1)∗ is a unital, completely isometric bijection. We can now follow [2, Section 1.3] to conclude that S is a ∗-homomorphism, as required. Indeed, S is unital and completely contractive, and so is completely positive. Then the Stinespring construction allows us to prove the Kadison-Schwarz inequality: S(a)∗S(a) ≤ S(a∗a). Applying this to S−1 as well, and using polarisation, yields the result. Theorem 3.18. Let T∗ : L1(G2) → L1(G1) be a completely isometric algebra isomorphism. Then u = T (1) ∈ L∞(G2) is a member of the intrinsic group, and there is a quantum group isomorphism θ : L∞(G1) → L∞(G2) such that T∗(ω) = θ∗(uω) (ω ∈ L∞(G2)). In particular, G1 is isomorphic to G2. Proof. The previous result shows that θ(x) = T (x)u∗ defines a ∗-homomorphism, and so the result is immediate. We remark that if T∗ : L1(G2) → L1(G1) is isometric, and completely contractive, then T∗ is induced by a quantum group isomorphism as above. Indeed, we only need to rule out the possibility that θ : x 7→ T (x)T (1)∗ is an anti-∗-isomorphism. As θ is still a complete contraction, the Kadison-Schwarz inequality would yield that θ(aa∗) = θ(a)∗θ(a) ≤ θ(a∗a), so applying θ−1 (which is an order-isomorphism) gives aa∗ ≤ a∗a, a contradiction (unless L∞(G1) is commutative, in which case θ is a homomorphism, as required). 4 Isometries between measure algebras In this section, we extend our results to isometric algebra isomorphisms between quantum measure algebras. We thus start with a survey of the C∗-algebraic theory of locally compact quantum groups. A morphism between two C∗-algebras A and B is a non-degenerate ∗-homomorphism φ : A → M(B) from A to the multiplier algebra of B. That φ is non-degenerate is equivalent to φ extending to a unital, strictly continuous ∗-homomorphism M(A) → M(B). Thus morphisms can be composed; for further details see [16, Appendix A]. Given G and its fundamental unitary W , the space {(ι ⊗ ω)(W ) : ω ∈ B(H)∗} is an algebra σ-weakly dense in L∞(G). However, the norm closure turns out to be a C∗-algebra, which we shall denote by C0(G). Then ∆ restricts to give a morphism C0(G) → M(C0(G) ⊗ C0(G)), and R, τt 13 and so forth all restrict to C0(G). It is possible to define locally compact quantum groups purely at the C∗-algebra level, although the necessary weight theory is more complicated; see [14]. As for L1(G), we use ∆ to turn C0(G)∗ = M(G) into a Banach algebra. In the commutative case, this is the measure algebra of a group, which justifies the notation. As C0(G) is σ-weakly dense in L∞(G), the embedding of L1(G) into M(G) is an isometry; clearly it is an algebra homomorphism, and actually L1(G) becomes an ideal in M(G), see [14, Page 914]. Actually, we work in a little generality, and introduce the following (non-standard) terminology. Definition 4.1. A quantum group above C0(G) is a triple (A, ∆A, π) where A is a C∗-algebra, ∆A : A → M(A ⊗ A) is a morphism, coassociative in the sense that (ι ⊗ ∆A)∆A = (∆A ⊗ ι)∆A, and π : A → C0(G) is a surjective ∗-homomorphism with ∆π = (π ⊗ π)∆A. Then π∗ : M(G) → A∗ is an algebra homomorphism, and we make the further requirement that π∗(L1(G)) is an essential ideal in A∗. Here essential means that if π∗(ω)µ = 0 for all ω ∈ L1(G), then µ = 0, and similarly with the orders reversed. For example, C0(G) itself is a quantum group above C0(G). In the cocommutative case, C0(G) = C ∗ r (G) the reduced group C∗-algebra of a locally compact group G, and so M(G) = Br(G), the reduced Fourier-Stieltjes algebra. We could alternatively study the full group C∗-algebra C ∗(G), whose dual is B(G) the Fourier-Stieltjes algebra. Then C ∗(G) is a quantum group above C ∗ r (G). It turns out that this example can be generalised to the quantum setting. We follow [11]. Let G be a locally compact quantum group, and let L1 ♯ ( G) be the collection of ω ∈ L1( G) such that there is w♯ ∈ L1( G) with hx, ω♯i = hS(x)∗, ωi (x ∈ D( S)). ♯ ( G), and let λu : L1 ♯ ( G) is a ∗-algebra for the involution ♯. Let Cu(G) be the universal enveloping C ∗-algebra Then L1 ♯ ( G) → Cu(G) be the canonical homomorphism. Then Cu(G) becomes a of L1 "quantum group" which is very similar to C0(G), the essential difference being that the left and right invariant weights are no longer faithful. For us, the important features are: • There is a non-degenerate ∗-homomorphism ∆u : Cu(G) → M(Cu(G) ⊗ Cu(G)) which is coassociative; • There is a surjective ∗-homomorphism π : Cu(G) → C0(G) with ∆π = (π ⊗ π)∆u. We note here that there are many differences in notation between [11] and that for Kac algebras used in [6]. We shall continue to follow [11]. It is shown in [4, Section 8] that L1(G) is an essential ideal in Cu(G)∗, and so Cu(G) is a quantum group above C0(G). In [15] examples of discrete groups G are given so that there is a compact quantum group r (G) and C ∗(G), in the sense that we have proper quotient maps (A, ∆A) which "sits between" C ∗ C ∗(G) → A → C ∗ r (G) which intertwine the coproducts. Then the inclusion maps Br(G) → A∗ → B(G) are isometric algebra homomorphisms. As the Fourier algebra A(G) is an essential ideal in B(G), it follows that A is a quantum group above C ∗ r (G). Indeed, this argument would work for any quantum group sitting between C0(G) and Cu(G) (but to our knowledge, [15] gives the first example of this phenomena). 4.1 Quantum group isomorphisms revisited Let θ : L∞(G1) → L∞(G2) be a quantum group isomorphism. Assuming we have normalised the Haar weights, θ will induce an isomorphism between the Hilbert spaces which intertwines the fundamental unitaries. Thus θ will restrict to give a ∗-isomorphism C0(G1) → C0(G2). 14 Similarly, θ induces a quantum group isomorphism θ : L∞( G2) → L∞( G1) which satisfies θλ2 = λ1θ∗. Then θ∗ will restrict to give a ∗-isomorphism between L1 ♯ (G2). This will induce a ∗-isomorphism θu : Cu(G1) → Cu(G2) which intertwines the coproducts, and satisfies π2θu = θπ1. ♯ (G1) and L1 For a quantum group commutant isomorphism θ we simply compose θ with the map x 7→ Jx∗J 2)= Gop to get a quantum group isomorphism from G1 to G′ 2 , where L∞( Gop 2 ) agrees with ♯ ( G2), but with the reversed product, and similarly Cu(G′ L1 2) is canonically equal to the opposite C∗- algebra to Cu(G2), but has the same coproduct. Thus, for example, θ lifts to an anti-∗-isomorphism θu : Cu(G1) → Cu(G2) which intertwines the coproduct (somewhat as we might hope). 2 ) = L∞( G2) with the opposite coproduct ∆op = σ ∆. Hence L1 2. In [13, Section 4] it is shown that (G′ ♯ ( Gop 4.2 Normal extensions Let B be a C∗-algebra non-degenerately represented on a Hilbert space H. Let M = B′′ be the von Neumann algebra generated by B. We can identify the multiplier algebra of B with M(B) = {x ∈ M : xa, ax ∈ B (a ∈ B)}. Let A be a C∗-algebra, and consider the enveloping C ∗-algebra A∗∗. Let φ : A → M(B) be a morphism. By the universal property of A∗∗, there is a unique normal ∗-homomorphism φ : A∗∗ → B′′ extending φ. As φ is non-degenerate, φ is unital. In the special case when B′′ = B∗∗ (say with B ⊆ B(H) the universal representation) the extension φ is nothing but the second adjoint φ∗∗. Now let (A, ∆A, π) be a quantum group above C0(G), and let A ⊆ B(H) be the universal representation, so that both A ⊗ A (the spacial C∗-tensor product) and A∗∗⊗A∗∗ are subalgebras of B(H ⊗ H). We can hence form the extension ∆A : A∗∗ → A∗∗⊗A∗∗. Notice that then (A∗∗, ∆A) becomes a Hopf-von Neumann algebra. Similarly, we form π : A∗∗ → L∞(G). The preadjoint of this map is simply the embedding π∗ : L1(G) → A∗, which is the composition of the isometry L1(G) → C0(G)∗ with the isometry π∗ : C0(G)∗ → A∗. Let supp π be the support projection of π, so supp π ∈ A∗∗ is the unique central projection with, for x ∈ A∗∗, x supp π = 0 if and only if π(x) = 0. Then π∗(L1(G))⊥ = {x ∈ A∗∗ : hx, π∗(ω)i = 0 (ω ∈ L1(G))} = ker π = (1 − supp π)A∗∗. It follows that π∗(L1(G)) = {µ ∈ A∗ : hx, µi = 0 (x ∈ (1 − supp π)A∗∗)} = {µ ∈ A∗ : (1 − supp π)µ = 0} = (supp π)A∗. Temporarily, let ∆0 be the coproduct on C0(G), and let ∆∞ be the coproduct on L∞(G). Identifying M(C0(G) ⊗ C0(G)) with a subalgebra of L∞(G)⊗L∞(G), we see that ∆∞ extends ∆0. It is easy to verify that as (π ⊗ π)∆A = ∆0π, also (π ⊗ π) ∆A = ∆∞π. We shall use this, and similar relations, without comment in the next section. We remark that we could work with a more general notion of a quantum group above C0(G). Indeed, suppose that (A, ∆A) is a C∗-bialgebra, and that π : A → L∞(G) is a ∗-homomorphism with σ-weak dense range. Then π ⊗ π is a ∗-homomorphism A ⊗ A → L∞(G) ⊗ L∞(G) ⊆ L∞(G)⊗L∞(G), and so, by taking a normal extension, we have a ∗-homomorphism π : M(A⊗A) → L∞(G)⊗L∞(G). We can thus make sense of the requirement that (π ⊗ π)∆A = ∆π. Then π∗ restricted to L1(G) gives a homomorphism L1(G) → A∗ (which is an isometry, as π has σ-weakly 15 dense range, and using Kaplansky Density). We again insist that π∗(L1(G)) is an essential ideal in A∗. A careful examination of the following proofs show that they would all work in this more general setting; but in the absence of any examples, we do not make this a formal definition. 4.3 Isometries of duals of quantum groups For i = 1, 2 let (Ai, ∆Ai, πi) be a quantum group above C0(Gi). Let T∗ : A∗ algebra isomorphism, and set T = (T∗)∗ : A∗∗ analogous way to the arguments in Section 3. 1 be an isometric 2 . The following is now proved in an entirely 1 → A∗∗ 2 → A∗ Lemma 4.2. T (1) is a unitary element of A∗∗ T1∗ : A∗∗ 1 → A∗∗ 2 ; ω 7→ T∗(T (1)∗ω) is an isometric algebra isomorphism. 2 which is a member of the intrinsic group. The map Again, we find that T1 = (T1∗)∗ is a Jordan homomorphism. We now show (in a similar, but more general, fashion to the arguments in [6, Section 5.6]) a link between π and the order properties of A∗∗, for (A, ∆A, π) a quantum group above C0(G). Proposition 4.3. Let G be a locally compact quantum group, and let (A, ∆A) a quantum group above C0(G). Let Q = {Q ∈ A∗∗ : Q is a projection, Q 6= 1, ∆A(Q) ≤ Q ⊗ Q}. Then Q has a maximal element, which is 1 − supp π. Proof. Let e = 1 − supp π, so as in Section 4.2 above, eA∗∗ = ker π and π∗(L1(G)) = (1 − e)A∗. Let µ, µ′ ∈ A∗, so there are ω, ω′ ∈ L1(G) with π∗(ω) = (1 − e)µ and π∗(ω′) = (1 − e)µ′. Then h ∆A(e)(e ⊗ e), µ ⊗ µ′i = he, (eµ)(eµ′)i = he, (µ − π∗(ω))(µ′ − π∗(ω′))i = he, µµ′i + he, π∗(ωω′) − µπ∗(ω′) − π∗(ω)µ′i = he, µµ′i = h ∆A(e), µ ⊗ µ′i, as π∗(L1(G)) is an ideal in A∗. It follows that ∆A(e)(e ⊗ e) = ∆A(e), and so ∆A(e) ≤ e ⊗ e. Thus e ∈ Q. Now let Q ∈ Q, so that ∆π(Q) = (∆π)(Q) = ((π ⊗ π)∆u)(Q) = (π ⊗ π) ∆u(Q) ≤ π(Q) ⊗ π(Q) ≤ 1 ⊗ π(Q). By [14, Lemma 6.4], this can only occur when π(Q) = 0 or 1. If π(Q) = 1, then Q ≥ supp π, and so Q + e ≥ 1. Thus ∆u(Q) + ∆u(e) ≥ 1 ⊗ 1, but as Q, e ∈ Q, it follows that Thus also 1 ⊗ 1 ≤ Q ⊗ Q + e ⊗ e. (1 − Q) ⊗ (1 − e) ≤ (cid:0)(1 − Q) ⊗ (1 − e)(cid:1)(cid:0)Q ⊗ Q + e ⊗ e(cid:1)(cid:0)(1 − Q) ⊗ (1 − e)(cid:1) = 0, and so Q = 1 or e = 1, a contradiction. Thus π(Q) = 0, showing that Q ≤ e as required. Proposition 4.4. With T∗, T, T1 as above, we have that: 1. T1(1 − supp π1) = 1 − supp π2. 2. T (ker π1) = ker π2. 16 3. T∗(π2,∗(L1(G2))) = π1,∗(L1(G1)). Proof. For i = 1, 2, form Qi for Ai as in Proposition 4.3. We claim that T1 gives a bijection Q1 to Q2. Let Q ∈ Q1, so as T1 is Jordan homomorphism, T1(Q) is a projection which is not equal to 1 (as T1(1) = 1 and T1 bijects). For µ, µ′ ∈ A∗ 2 positive, we see that h ∆A2(T1(Q)), µ ⊗ µ′i = h∆A1(Q), T1∗(µ) ⊗ T1∗(µ′)i ≤ hQ ⊗ Q, T1∗(µ) ⊗ T1∗(µ′)i = hT1(Q) ⊗ T1(Q), µ ⊗ µ′i. It follows that ∆A2(T1(Q)) ≤ T1(Q) ⊗ T1(Q), and so T1(Q1) ⊆ Q2. Applying the same argument to T −1 yields that T1(Q1) ⊇ Q2, giving the claim. As T1 preserves the order, and 1 − supp πi is the maximal element of Qi, it follows that T1(1 − supp π1) = 1 − supp π2 showing (1). 1 For i = 1, 2, we know that x ∈ ker πi if and only if x supp πi = 0. For x ∈ A∗∗ 1 , as T1 is a Jordan homomorphism, we see that 2T1(x) supp π2 = (supp π2)T1(x) + T1(x) supp π2 = T1(supp π1)T1(x) + T1(x)T1(supp π1) = T1((supp π2)x + x supp π2) = 2T1(x supp π2), using (1). Thus T1(ker π1) = ker π2. As ker π1 is an ideal, and T (1) is unitary, it follows that ker π1T (1) = ker π1, and so T (ker π1) = T1(ker π1T (1)) = ker π2 showing (2). As in Section 4.2 above, we have that πi,∗(L1(Gi)) = ⊥(ker πi), for i = 1, 2. Hence (3) follows immediately from (2). For i = 1, 2 we have that L1(Gi) ⊆ A∗ i isometrically, and so the restriction of T yields an isometric algebra homomorphism Tr : L1(G2) → L1(G1). We have already characterised such maps, and we next bootstrap this to determine the structure of T∗ on all of A∗ 2. For the moment, we restrict attention to the cases when Ai = C0(Gi) for i = 1, 2, or Ai = Cu(Gi), for i = 1, 2. In the next section we use quantum group duality to say something about the general case. Given a quantum group (commutant) isomorphism θ : L∞(G1) → L∞(G2), we recall from Section 4.1 that θ restricts to a (anti-) ∗-isomorphism θ : C0(G1) → C0(G2), and lifts to a (anti-) ∗-isomorphism θu : Cu(G1) → Cu(G2). In the following, we call such a map "associated". Theorem 4.5. Let G1 and G2 be locally compact quantum groups. Suppose that either A1 = C0(G1), A2 = C0(G2), or A1 = Cu(G1), A2 = Cu(G2). Let T∗ : A∗ 2 → A∗ 1 be a bijective isometric algebra homomorphism, and set T = (T∗)∗. Then v = T (1) and u = T ∗ r (1) are in the intrinsic groups of A∗∗ 2 and L∞(G2), respectively. There is either: 1. A quantum group isomorphism θ : L∞(G1) → L∞(G2) and associated ∗-isomorphism θ0 : A1 → A2 which intertwines the coproducts; or 2. A quantum group commutant isomorphism θ : L∞(G1) → L∞(G2) and associated anti-∗- isomorphism θ0 : A1 → A2 which intertwines the coproducts. In either case, for ω ∈ Mu(G2) and ω′ ∈ L1(G2), T∗(ω) = θ∗ 0(vω), Tr(ω′) = θ∗(uω′). Proof. By previous work, u = T ∗ map θ : L∞(G1) → L∞(G2); x 7→ T ∗ isomorphism, which in either case intertwines the coproduct. r (1) is a member of the intrinsic group of L∞(G2), and the r (x)u∗ is either a normal ∗-isomorphism, or a normal anti-∗- 17 Suppose we are in the first case, where θ is a ∗-isomorphism. Then we have an associated ∗-isomorphism θ0 : A1 → A2 which intertwines the coproducts, and which satisfies π2θ0 = θπ1. Taking adjoints gives that θ∗ 0 π2,∗ = π1,∗θ∗. For ω ∈ L1(G2), we have that θ∗(ω) = Tr(u∗ω). Also, Tr is constructed so that T∗π2,∗ = π1,∗Tr. Thus T∗(cid:0)π2,∗(u∗ω)(cid:1) = π1,∗(cid:0)Tr(u∗ω)(cid:1) = π1,∗(cid:0)θ∗(ω)(cid:1) = θ∗ (ω ∈ L1(G2)). Recall that v = T (1) ∈ A∗∗ 2 is also unitary. Then u = T ∗ simple calculation shows that vπ2,∗(ω) = π2,∗(uω) for ω ∈ L1(G2). r π1(1) = π2T (1) = π2(v). A 0(cid:0)π2,∗(ω)(cid:1) r (1) = T ∗ Let µ ∈ A∗ 2 and ω ∈ L1(G2), so we can find ω′ ∈ L1(G2) with π2,∗(ω′) = π2,∗(ω)µ. Then T∗(cid:0)π2,∗(ω)(cid:1)T∗(µ) = T∗(cid:0)π2,∗(ω′)(cid:1) = θ∗ 0(cid:0)π2,∗(uω′)(cid:1) = θ∗ 0(cid:0)vπ2,∗(ω′)(cid:1) = θ∗ = θ∗ 0(cid:0)(vπ2,∗(ω))(vµ)(cid:1) = θ∗ 0(vπ2,∗(ω))θ∗ 0(vµ) = T∗(cid:0)π2,∗(ω)(cid:1)θ∗ 0(cid:0)v(cid:0)π2,∗(ω)µ(cid:1)(cid:1) 0(vµ). Recall that, from the hypothesis, π1,∗(L1(G2)) is an essential ideal in A∗ to π1,∗(L1(G2)), we see that 1. As T∗ bijects π2,∗(L1(G2)) T∗(µ) = θ∗ 0(vµ) (µ ∈ A∗ 2), as claimed. The other case, when θ is a quantum group commutant isomorphism, is entirely analogous. The previous theorem needs a characterisation of the intrinsic group of A∗∗, for A a quantum group above C0(G). The following results show that it is enough to know the intrinsic group of L∞(G). Lemma 4.6. Let A be a Banach algebra, and let I ⊆ A be a closed ideal. Let ΦI be the character space I, and let X be the collection of characters on A which do not restrict to the zero functional on I. Then restriction of linear functionals gives a bijection from X to ΦI . Proof. Let f, g ∈ X induce the same (non-zero) character on I. Pick a0 ∈ I with f (a0) = g(a0) = 1. Then, for a ∈ A, we see that f (a) = f (a)f (a0) = f (aa0) = g(aa0) = g(a)g(a0) = g(a), using that aa0 ∈ I. Thus f = g, so the restriction map is injective. Now let u ∈ ΦI, and pick a0 ∈ I with u(a0) = 1. Define f ∈ A∗ by f (a) = u(aa0) for each a ∈ A. Then, for a, b ∈ A, f (ab) = u(aba0) = u(a0)u(aba0) = u(a0aba0) = u(a0a)u(ba0) = u(a0a)u(a0)f (b) = u(a0aa0)f (b) = u(a0)u(aa0)f (b) = f (a)f (b). So f is a character on A. For a ∈ I, also f (a) = u(aa0) = u(a)u(a0) = u(a), and so f ∈ X and f restricts to u. Thus the restriction map is a bijection. The following should be compared with [24, Theorem 1] where Walter shows this in the co- commutative case. Theorem 4.7. Let (A, ∆A, π) be a quantum group above C0(G). For a character u on A∗, the following are equivalent: 1. u is a member of the intrinsic group of A∗∗; 2. u is invertible in A∗∗; 18 3. π(u) 6= 0, that is, u does not induce the zero functional on π∗(L1(G)). Moreover, π : A∗∗ → L∞(G) restricts to a bijection between the intrinsic groups of A∗∗ and L∞(G). Proof. Let Y be the intrinsic group of L∞(G), which by Theorem 3.2 is the character space of L1(G). Let X1 be the intrinsic group of A∗∗, let X2 be the collection of invertible characters, and let X3 be the collection of characters not sent to zero by π. If u ∈ X2 then 1 = π(1) = π(uu−1) = π(u)π(u−1), showing that π(u) 6= 0 and hence u ∈ X3. Thus X1 ⊆ X2 ⊆ X3. By the lemma, π restricts to a bijection between X3 and Y . Let u ∈ X3, so by Theorem 3.2, ∆A(u) = u ⊗ u. As ∆A is a ∗-homomorphism, also u∗u is a character. As π(u) ∈ L∞(G) is a (non-zero) character, it is unitary, and so 1 = π(u)∗π(u) = π(u∗u). Thus u∗u ∈ X3, and as π injects on X3, and 1 ∈ X3, we conclude that u∗u = 1. Similarly, uu∗ = 1. Thus u is a member of the intrinsic group of A∗∗, that is, u ∈ X1. We hence have the required equalities X1 = X2 = X3. In special cases, we can say more. Proposition 4.8. The intrinsic group of Cu(G)∗∗, respectively C0(G)∗∗, is a subgroup of the unitary group of M(Cu(G)), respectively M(C0(G)). Proof. Let x ∈ Cu(G)∗∗ be a member of the intrinsic group, and set y = π(x) ∈ L∞(G). By Theorem 3.2, we have that y is unitary, and y ∈ M(C0(G)). Thus, in the language of [11, Proposition 6.6], y is a unitary corepresentation of C0(G) on C, and so there is x0 ∈ M(Cu(G)) with π(x0) = y and ∆u(x0) = x0 ⊗ x0. By uniqueness (from the previous theorem) we must have that x0 = x, treating M(Cu(G)) as a subalgebra of Cu(G)∗∗. Now let x ∈ C0(G)∗∗ be a member of the intrinsic group. Again, π(x) = y ∈ M(C0(G)), so let x0 be the image of y under the embedding M(C0(G)) → C0(G)∗∗. Thus x0 is a member of the intrinsic group of C0(G)∗∗ and π(x0) = π(x), so again by uniqueness, we conclude that x0 = x. 4.4 The picture under duality We now show that by using the duality theory of locally compact quantum groups, we can handle the more general situation; this also gives results more reminiscent of those for Kac algebras, see [6, Section 5.6]. Let (A, ∆A, π) be a quantum group above C0(G). As L1(G) is an essential ideal in A∗, each member of A∗ induces a (completely bounded) multiplier (or centraliser) of L1(G). Let us introduce the notation that given µ ∈ A∗, we have maps Lµ, Rµ : L1(G) → L1(G) with π∗Lµ(ω) = µπ∗(ω), π∗Rµ(ω) = π∗(ω)µ (ω ∈ L1(G)). Let us denote by Mcb(L1(G)) the algebra of completely bounded multipliers of L1(G). In [4, Theorem 8.9], a homomorphism Λ : Mcb(L1(G)) → C b( G) was constructed (and a more general construction, with one-sided multipliers, is given in [7]). We hence find a map, which we shall continue to denote by Λ, from A∗ to C b( G), which is uniquely determined by the properties that Λ(µ)λ(ω) = λ(Lµ(ω)), λ(ω)Λ(µ) = λ(Rµ(ω)) (µ ∈ A∗, ω ∈ L1(G)). An important link between multipliers and the antipode is established in [3]. In particular, given µ ∈ A∗, define an associated left multiplier L† µ by L† µ(ω) = Lµ(ω∗)∗ so π∗L† µ(ω) = (cid:0)µπ∗(ω∗)(cid:1)∗ = µ∗π∗(ω), 19 µ = Lµ∗. (Recall that ω∗ is the normal functional x 7→ hx∗, ωi so this calculation follows that is, L† immediately from the fact that ∆A and π are ∗-homomorphisms). Then [3, Theorem 5.9] shows that Λ(µ∗) ∈ D( S)∗ and Λ(µ) = S(Λ(µ∗)∗). For λ : L1(G) → C0(G) (which Λ extends) we can see this directly. Recall (see [14, Proposi- tion 8.3]) that S((ι ⊗ ω)( W )) = (ι ⊗ ω)( W ∗). As W = σW ∗σ, we see that λ(ω) = (ω ⊗ ι)(W ) = S((ω ⊗ ι)(W ∗)) = S(λ(ω∗)∗). Lemma 4.9. Let u ∈ L∞(G) be a member of the intrinsic group. For x ∈ L∞( G), let γu(x) = uxu∗. Then γu is a ∗-automorphism of L∞( G), which restricts to a ∗-automorphism of C0( G). Furthermore, γuλ(ω) = λ(uω) for ω ∈ L1(G). Proof. We have that ∆(u) = u ⊗ u, so W ∗(1 ⊗ u)W = u ⊗ u. Using that W and u are unitary, it follows that (1 ⊗ u)W (1 ⊗ u∗) = W (u ⊗ 1). Then, for ω ∈ L1(G), γuλ(ω) = u(ω ⊗ ι)(W )u∗ = (ω ⊗ ι)(cid:0)(1 ⊗ u)W (1 ⊗ u∗)(cid:1) = (ω ⊗ ι)(cid:0)W (u ⊗ 1)(cid:1) = λ(uω), as claimed. By density, it follows that γu is a self-map of C0( G), which clearly has the inverse γu∗. As γu is normal, it follows that γu is also an automorphism of L∞( G). For the construction of θ in the following, we again refer the reader to Section 4.1. Theorem 4.10. For i = 1, 2, let Gi be a locally compact quantum groups, and let (Ai, ∆Ai, πi) be a quantum group above C0(Gi). Let T∗ : A∗ 1 be a bijective isometric algebra homomorphism, and set T = (T∗)∗. Then v = T (1) and u = π2(v) are members of the intrinsic groups of A∗∗ 2 and L∞(G2), respectively. Then either: 2 → A∗ 1. There is a quantum group isomorphism θ : L∞(G1) → L∞(G2), leading to a quantum group isomorphism θ : L∞( G2) → L∞( G1), with Λ1T∗ = θγuΛ2. 2. There is a quantum group commutant isomorphism θ : L∞(G1) → L∞(G2), leading to a quantum group isomorphism θ : L∞( Gop 2 ) → L∞( G1), with Λ1T∗ = θ R2 S−1 2 γuΛ2. In particular, G1 is isomorphic to either G2 or G′ 2. Proof. In this more general situation, the proof of Theorem 4.5, and Theorem 4.7, still give the facts about v and u, and yields θ such that T∗(cid:0)π2,∗(u∗ω)(cid:1) = π1,∗(cid:0)θ∗(ω)(cid:1) (ω ∈ L1(G2)). Suppose first that θ is a quantum group isomorphism. Let θ : L∞( G2) → L∞( G1) be the quantum group isomorphism induced by θ, which satisfies λ1θ∗ = θλ2. Let ω ∈ L1(G2) and µ ∈ A∗ 2. There is ω′ ∈ L1(G2) with µπ2,∗(ω) = π2,∗(ω′). Then Λ1(T∗(µ))Λ1(T∗(π2,∗(ω))) = Λ1(T∗(π2,∗(ω′))) = λ1(θ∗(uω′)) = θ(λ2(uω′)) = θ(γu(λ2(ω′))) = θγu(cid:0)Λ2(µ)λ2(ω)(cid:1) = θγuΛ2(µ)Λ1(T∗(π2,∗(ω))), 20 using that, similarly, Λ1(T∗(π2,∗(ω))) = θ(γu(λ2(ω))). As the set {γu(λ2(ω)) : ω ∈ L1(G2)} is norm dense in C0( G2), working in M(C0( G2)), we conclude that Λ1T∗(µ) = θγuΛ2(µ) (µ ∈ A∗ 2), as required. 2); x 7→ Jx∗J, and set θ′ = Φθ : L∞(G1) → L∞(G′ In the case when θ is a quantum group commutant isomorphism, define Φ : L∞(G2) → 2), which is a quantum group isomor- 2)) → L∞( G1). As θ = σ(θ ⊗ θ) ∆2. L∞(G′ phism. As in Section 4.1 we find a quantum group isomorphism θ′ : L∞((G′ (G′ Then θλ′ 2) = ( G2)op, this gives a normal ∗-isomorphism θ : L∞( G2) → L∞( G1) with ∆1 2 = λ1θ′ We now calculate λ′ ∗ . Let ξ, η, α, β ∈ L2(G), so ∗ = λ1θ∗Φ∗. 2Φ−1 (cid:0)λ′ 2Φ−1 ∗ (ωξ,η)α(cid:12)(cid:12)β(cid:1) = (cid:0)λ′ 2(ωJη,Jξ)α(cid:12)(cid:12)β(cid:1) = (cid:0)W ′ = (cid:0)(J ⊗ J)W2(η ⊗ Jα)(cid:12) = (cid:0)(ωξ,η ⊗ ι)(W ∗ 2 )Jβ(cid:12) 2(Jη ⊗ α)(cid:12)(cid:12)Jξ ⊗ β(cid:1) (cid:12)Jξ ⊗ β(cid:1) = (cid:0)W ∗ 2 (ξ ⊗ Jβ)(cid:12) (cid:12)Jα(cid:1) = (cid:0)J((ωξ,η ⊗ ι)(W ∗ 2 ))∗Jα(cid:12) (cid:12)η ⊗ Jα(cid:1) (cid:12)β(cid:1), using that W ′ that 2 = (J ⊗ J)W2(J ⊗ J). With reference to the discussion before Lemma 4.9, we see 2Φ−1 λ′ ∗ (ω) = R2(cid:0)(ω ⊗ ι)(W ∗ 2 )(cid:1) = R2(cid:0)λ2(ω∗)∗(cid:1) = R2 S−1 2 λ2(ω) (ω ∈ L1(G2)). In particular, λ1θ∗ = θ R2 S−1 2 λ2. Finally, we follow the previous argument through. So let ω, ω′ ∈ L1(G2) and µ ∈ A∗ 2 with µπ2,∗(ω) = π2,∗(ω′). Then Λ1(T∗(µ))Λ1(T∗(π2,∗(ω))) = λ1(θ∗(uω′)) = θ R2 S−1 2 γuλ2(ω′) = θ R2 S−1 2 γu(cid:0)Λ2(µ)λ2(ω)(cid:1) = (cid:0)θ R2 S−1 2 γuΛ2(µ)(cid:1)Λ1(T∗(π2,∗(ω))), using that R2 S−1 2 is a homomorphism on D(S−1 2 ). This completes the proof. References [1] S. Baaj, G. Skandalis, "Unitaires multiplicatifs et dualit´e pour les produits crois´es de C*-alg`ebres", Ann. Sci. ´Ecole Norm. Sup. 26 (1993) 425 -- 488. [2] D. P. Blecher, C. Le Merdy, Operator algebras and their modules: an operator space approach. London Mathematical Society Monographs. New Series, 30. Oxford Science Publications. (The Clarendon Press, Oxford University Press, Oxford, 2004). [3] M. Daws, "Multipliers of locally compact quantum groups via Hilbert C∗-modules", J. Lond. Math. Soc. 84 (2011) 385 -- 407. [4] M. Daws, "Multipliers, Self-Induced and Dual Banach Algebras", Dissertationes Mathematicae 470 (2010) 62pp. 21 [5] J. De Canni`ere, M. Enock, J.-M. Schwartz, "Sur deux r´esultats d'analyse harmonique non- commutative: une application de la th´eorie des alg`ebres de Kac", J. Operator Theory 5 (1981) 171 -- 194. [6] M. Enock, J.-M. Schwartz, Kac Algebras and Duality of Locally Compact Groups (Springer-Verlag, Berlin, 1992) [7] M. Junge, M. Neufang, Z.-J. Ruan, "A representation theorem for locally compact quantum groups", Internat. J. Math. 20 (2009) 377 -- 400. [8] R. V. Kadison, "Isometries of operator algebras", Ann. Of Math. 54 (1951) 325 -- 338. [9] M. Kalantar, "Towards harmonic analysis on locally compact quantum groups", Ph.D. thesis, Car- leton University, Ottawa, 2010. [10] M. Kalantar, M. Neufang, "From quantum groups to groups", preprint, see arXiv:1110.5129v1. [11] J. Kustermans, "Locally compact quantum groups in the universal setting", Internat. J. Math. 12 (2001), no. 3, 289 -- 338. [12] J. Kustermans, "Locally compact quantum groups" in Quantum independent increment processes. I, Lecture Notes in Math. 1865, pp. 99 -- 180 (Springer, Berlin, 2005). [13] J. Kustermans, S. Vaes, "Locally compact quantum groups in the von Neumann algebraic setting", Math. Scand. 92 (2003) 68 -- 92. [14] J. Kustermans, S. Vaes, "Locally compact quantum groups", Ann. Sci. ´Ecole Norm. Sup. (4) 33 (2000) 837 -- 934. [15] D. Kyed, P. M. So ltan, "Property (T) and exotic quantum group norms", to appear in J. Noncom- mutative Geometry, see arXiv:1006.4044v1 [math.OA] [16] T. Masuda, Y. Nakagami, S. L. Woronowicz, "A C ∗-algebraic framework for quantum groups", In- ternat. J. Math. 14 (2003) 903 -- 1001. [17] V. Runde, "Applications of operator spaces to abstract harmonic analysis", Expo. Math. 22 (2004) 317 -- 363. [18] V. Runde, "Lectures on Amenability", Lecture Notes in Mathematics, 1774. (Springer-Verlag, Berlin, 2002). [19] E. Størmer, "On the Jordan structure of C ∗-algebras", Trans. Amer. Math. Soc. 120 (1965) 438 -- 447. [20] M. Takesaki, Theory of operator algebras. II. Encyclopaedia of Mathematical Sciences, 125. Operator Algebras and Non-commutative Geometry, 6. (Springer-Verlag, Berlin, 2003). [21] S. Vaes, "Locally compact quantum groups", PhD. thesis, Katholieke Universiteit Leuven, 2001. Available from http://wis.kuleuven.be/analyse/stefaan/ [22] S. Vaes, L. Vainerman, "Extensions of locally compact quantum groups and the bicrossed product construction", Adv. Math. 175 (2003) 1 -- 101. [23] A. Van Daele, "Locally compact quantum groups. A von Neumann algebra approach." See arXiv:math/0602212v1 [math.OA] 22 [24] M. E. Walter, "W ∗-algebras and nonabelian harmonic analysis", J. Functional Analysis 11 (1972) 17 -- 38. [25] J. G. Wendel, "On isometric isomorphism of group algebras", Pacific J. Math. 1 (1951) 305 -- 311. [26] S. L. Woronowicz, "From multiplicative unitaries to quantum groups", Internat. J. Math. 7 (1996) 127 -- 149. Matthew Daws School of Mathematics, University of Leeds, LEEDS LS2 9JT United Kingdom Email: [email protected] Hung Le Pham School of Mathematics, Statistics, and Operations Research, Victoria University of Wellington, Wellington 6012 New Zealand Email: [email protected] 23
1210.6067	1	1210	2012-10-22T20:55:47	Operator algebras and C*-correspondences: A survey	[ "math.OA" ]	In this paper we survey our recent work on C*-correspondences and their associated operator algebras; in particular, on adding tails, the Shift Equivalence Problem and Hilbert bimodules.	math.OA	math	OPERATOR ALGEBRAS AND C-CORRESPONDENCES: A SURVEY EVGENIOS T.A. KAKARIADIS AND ELIAS G. KATSOULIS Abstract. In this paper we survey our recent work on C∗- correspon- dences and their associated operator algebras; in particular, on adding tails, the Shift Equivalence Problem and Hilbert bimodules. Introduction Traditionally, the theory of C∗-correspondences has been used to gener- alize concrete results either from the theory of Cuntz-Krieger C∗-algebras or from the theory of crossed product C∗-algebras. Our goal in this project has been to discover a path in the opposite direction: to use the theory of C∗-correspondences in order to obtain results which are new even for the Cuntz-Krieger or the crossed product C∗-algebras. For instance in Theorem 2.1 we describe a general process for adding tail to a C∗-correspondence that has lead to new results in the theory of crossed product C∗-algebras and Cuntz's twisted crossed products. In the same spirit, our Theorem 3.8 on the concept of shift equivalence of C∗-correspondences leads to a new re- sult regarding the strong Morita equivalence of Cuntz-Krieger C∗-algebras (Theorem 3.10(4)). Therefore, even though the next few pages may seem rather abstract or encyclopedic, we believe that this abstraction will bene- fit even the reader who is interested only in the special classes of operator algebras mentioned above. 1. Preliminaries Let A be a C∗-algebra. An inner-product right A-module is a linear space X which is a right A-module together with a map such that (·, ·) 7→ h·, ·iX : X × X → A hξ, λy + ηiX = λ hx, yiX + hx, ηiX hξ, ηaiX = hξ, ηiX a hη, ξiX = hξ, ηi∗ hξ, ξiX ≥ 0; if hξ, ξiX = 0 then ξ = 0. X (ξ, y, η ∈ X, λ ∈ C), (ξ, η ∈ X, a ∈ A), (ξ, η ∈ X), 2010 Mathematics Subject Classification. 46L08, 47L55. Key words and phrases: C∗-correspondences, C∗-envelope, shift equivalence. 1 2 E.T.A. KAKARIADIS AND E.G. KATSOULIS A compatibility relation for the scalar multiplication is required, that is λ(ξa) = (λξ)a = ξ(λa), for all λ ∈ C, ξ ∈ X, a ∈ A. For ξ ∈ X we write kξkX := khξ, ξiAkA and one can deduce that k·kX is actually a norm. X equipped with that norm will be called right Hilbert A-module if it is complete and will be denoted as XA. For X, Y Hilbert A-modules we define the set L(X, Y ) of the adjointable maps that consists of all maps s : X → Y for which there is a map s∗ : Y → X such that hsξ, yiY = hξ, s∗yiY , (x ∈ X, y ∈ Y ). Every element of L(X, Y ) is automatically a bounded A-linear map. An element u ∈ L(X, Y ) is called unitary if it is onto Y and huξ, uζiY = hξ, ζiX, for all ξ, ζ ∈ X. In particular, for ξ ∈ X and y ∈ Y we define Θy,ξ : X → Y such that Θy,ξ(ζ) = y hξ, ζiX , for all ζ ∈ X. It is easy to check that Θy,ξ ∈ L(X, Y ) with Θ∗ y,ξ = Θξ,y. We denote by K(X, Y ) the closed linear space of L(X, Y ) spanned by {Θy,ξ : ξ ∈ X, y ∈ Y }. If X = Y then K(X, X) ≡ K(X) is a closed ideal of the C∗-algebra L(X, X) ≡ L(X). In a dual way we call X a left Hilbert A-module if it is complete with respect to the norm induced by an inner-product left A-module [·, ·]X . The term Hilbert module is reserved for the right Hilbert modules, whereas the left case will be clearly stated. Given a Hilbert A-module X over A, let X ∗ = {ξ∗ ∈ L(X, A) : ξ∗(ζ) = hξ, ζiX } be the dual left Hilbert A-module, with a · ξ∗ = (ξa∗)∗, [ξ∗, ζ ∗]X ∗ = hξ, ζiX , (ξ ∈ X, a ∈ A), (ξ, ζ ∈ X). A Hilbert A-module X is called self-dual when X ∗ coincides with the set of bounded (not necessarily adjointable) A-linear mappings from X to A, hence a Riesz-Fr´echet Theorem is valid. Example 1.1. A C∗-algebra A is a (trivial) Hilbert A-module, when it is viewed as a Banach space with a · b := ab and ha, biA := a∗b for all a, b ∈ A. It is a left Hilbert A-module when it is endowed with the left inner product [a, b]A := ab∗, for all a, b ∈ A. Finally K(A) ≃ A and L(AA) ≃ M(A), i.e. the multiplier algebra of A. 1.1. C∗-correspondences. Even though the class of C∗ - correspondences has been thoroughly investigated for the last 25 years, the terminology still differs from author to author. We therefore present the terminology that we will be using in this paper. Definition 1.2. An A-B-correspondence X is a right Hilbert B-module together with a ∗-homomorphism ϕX : A → L(X). We will denote this by AXB. When A = B we will simply refer to X as C∗-correspondence over A. A submodule Y of X is a subcorrespondence of AXB, if it is a C-D- correspondence for some C∗-subalgebras C and D of A and B, respectively. OPERATOR ALGEBRAS AND C-CORRESPONDENCES: A SURVEY 3 A C∗-correspondence over A is called non-degenerate if the closed linear span of ϕX (A)X is dense in X. Moreover, X is called full if hX, XiX is dense in A. Also, X is called regular if it is injective, i.e. ϕX is injective, and ϕX (A) ⊆ K(X). Two A-B-correspondences X and Y are called unitarily equivalent, if there is a unitary u ∈ L(X, Y ) such that u(ϕX (a)ξb) = ϕY (a)(u(ξ))b, for all a ∈ A, b ∈ B, ξ ∈ X. In that case we write X ≈ Y . We write X . Y when X ≈ Y0 for a subcorrespondence Y0 of Y . Example 1.3. Every Hilbert A-module X is a K(X)-A-correspondence when endowed with the left multiplication ϕX ≡ idK(X) : K(X) → L(X). A left inner product over K(X) can be defined by [ξ, η]X = Θξ,η, for all ξ, η ∈ X. Also X ∗ is an A-K(X)-correspondence, when endowed with the following operations hξ∗, η∗iX ∗ = [ξ, η]X = Θξ,η ξ∗ · k = (k∗ξ)∗ ϕX ∗ (a)ξ∗ = a · ξ∗ = (ξ · a∗)∗ (ξ∗, η∗ ∈ X ∗), (ξ∗ ∈ X ∗, k ∈ K(X)), (ξ∗ ∈ X ∗, a ∈ A). Example 1.4. For Hilbert A-modules X and Y , L(X, Y ) becomes L(Y )- L(X)-correspondence by defining hs, ti := s∗t, t · a := ta and b · t := bt, for every s, t ∈ L(X, Y ), a ∈ L(X) and b ∈ L(Y ). Trivially, K(X, Y ) is a K(Y )-K(X)-subcorrespondence of L(X, Y ). Note for Y , then K(Y ) acts that, when a c.a.i. faithfully on K(X, Y ). When X = Y this is automatically true. in hX, XiX is a right c.a.i. 1.2. Interior Tensor Product. The interior tensor product of two Hilbert modules plays the role of a generalized multiplication, stabilized by the elements of a common C∗-algebra (see [37] for more details). Let the C∗- correspondences AXB and BYC; the interior or stabilized tensor product, denoted by X ⊗B Y or simply by X ⊗ Y , is the quotient of the vector space tensor product X ⊗alg Y by the subspace generated by the elements of the form ξa ⊗ y − ξ ⊗ ϕ(a)y, for all ξ ∈ X, y ∈ Y, a ∈ A. It becomes a Hilbert B-module when equipped with (ξ ⊗ y)b := ξ ⊗ (yb), hξ1 ⊗ y1, ξ2 ⊗ y2iX⊗Y := hy1, ϕ(hξ1, ξ2iX)y2iY , (ξ ∈ X, y ∈ Y, b ∈ B), (ξ1, ξ2 ∈ X, y1, y2 ∈ Y ). For s ∈ L(X) we define s ⊗ idY ∈ L(X ⊗ Y ) as the mapping ξ ⊗ y 7→ s(ξ)⊗ y. Clearly, (s ⊗ idY )∗ = s∗ ⊗ idY ; hence X ⊗ Y becomes an A-C-correspondence by defining ϕX⊗Y (a) := ϕX(a) ⊗ idY . One can prove that the interior tensor product is associative, that is if Z is a C-D-correspondence, then (X ⊗B Y ) ⊗C Z = X ⊗B (Y ⊗C Z). Example 1.5. When a Hilbert A-module X is considered as the K(X)-A- correspondence, then X ⊗A X ∗ ≈ K(X) as C∗-correspondences over K(X), 4 E.T.A. KAKARIADIS AND E.G. KATSOULIS via the mapping u1 : X ⊗A X ∗ → K(X) : ξ ⊗ ζ ∗ 7→ Θξ,ζ, and X ∗ ⊗K(X) X ≈ hX, XiX, as C∗-correspondences over A via the mapping u2 : X ∗ ⊗K(X) X → hX, XiX : ξ∗ ⊗ ζ 7→ hξ, ζi In particular X ∗ ⊗K(X) X ≈ A, when X is full. 1.3. Hilbert Bimodules. There are AXB C∗-correspondences that are both left and right Hilbert modules. If a compatibility relation is satisfied between the two inner products then it is called Hilbert bimodule. Definition 1.6. A Hilbert A-B-bimodule AXB is a C∗-correspondence X together with a left inner product [·, ·]X : X × X → A, which satisfy: [ϕX(a)ξ, η]X = a [ξ, η]X , [ξ, η]X = [η, ξ]∗ [ξ, ξ]X ≥ 0; if [ξ, ξ]X = 0 then ξ = 0, ϕX ([ξ, η]X)ζ = ξ hη, ζiX , X , (ξ, η ∈ X, a ∈ A), (ξ, η ∈ X) (ξ, η, ζ ∈ X). The last equation implies that ϕX ([ξ, η]X) = ΘX ξ,η. It is clear that Hilbert bimodules are a special case of C∗-correspondences. Let IX be the ideal, IX = span{[ξ, η]X : ξ, η ∈ X}, in A. Using the very definitions, one can prove that a ∈ ker ϕX , if and only if a ∈ I ⊥ X. Hence, ϕX is ∗-injective, if and only if, the Hilbert A-bimodule X is essential, i.e. when the ideal IX is essential in A. Definition 1.7. An A-B-imprimitivity bimodule or equivalence bimodule is an A-B-bimodule M which is simultaneously a full left and a full right Hilbert A-module. That is [M, M ]M is dense in A and hM, M iM is dense in B. It is easy to see that when X is an A-B-imprimitivity bimodule then A ≃ϕ K(M ). Thus imprimitivity bimodules are automatically non-degenerate and regular. 1.4. Matrix C∗-correspondences. There is a number of ways of consid- ering a direct sum of C∗-correspondences. They are contained (as subcorre- spondences) in the notion of the matrix C∗-correspondence that is presented below. For the C∗-correspondences AEA, BFB, ARB and BSA the matrix C∗-correspondence X = (cid:20) E R of the linear space of the matrices (cid:20) e S F (cid:21) over A⊕B is the Hilbert (A⊕B)-module s f (cid:21), e ∈ E, r ∈ R, s ∈ S, f ∈ F , r OPERATOR ALGEBRAS AND C-CORRESPONDENCES: A SURVEY 5 with r s rb f (cid:21) · (a, b) = (cid:20) ea (cid:20) e (cid:28)(cid:20) e1 such that the ∗-homomorphism ϕ : A ⊕ B → L(cid:18)(cid:20) E R = (cid:0) he1, e2iE + hs1, s2iS , hr1, r2iR + hf1, f2iF (cid:1), S F (cid:21)(cid:19) is defined as sa f b (cid:21) , f1 (cid:21)(cid:29)X f1 (cid:21) ,(cid:20) e1 s1 r1 r1 s1 follows ϕ(a, b)(cid:20) e s f (cid:21) = (cid:20) ϕE(a)e ϕR(a)r ϕS (b)s ϕF (b)f (cid:21) . r The careful reader can see that this is exactly the exterior direct sum C∗- correspondence of the two interior direct sum C∗-correspondences A⊕B(E + S)A and A⊕B(R + F )B. Hence, the linear space of the matrices (cid:20) e s f (cid:21), r with e ∈ E, r ∈ R, s ∈ S, f ∈ F , is complete with respect to the induced norm, thus a Hilbert (A ⊕ B)-module. Moreover E, F, R, S imbed naturally as subcorrespondences in (cid:20) E R S F (cid:21). The following lemma reasons the ter- minology matrix C∗-correspondence, as tensoring comes by "multiplying" the matrices. Lemma 1.8. Let E, F, R, S be C∗-correspondences as above. Then S F (cid:21) ⊗A⊕B (cid:20) E R (cid:20) E R ≈ (cid:20) (E ⊗A E) + (R ⊗B S) (S ⊗A E) + (F ⊗B S) S F (cid:21) ≈ (E ⊗A R) + (R ⊗B F ) (S ⊗A R) + (F ⊗B F ) (cid:21) , (unitary equivalent) as C∗-correspondences. 1.5. Representations of C∗-correspondences. Let us make a brief pre- sentation on the representation theory of C∗-correspondences. Let X be a C∗-correspondence over A. A (Toeplitz) representation (π, t) of X into a C∗-algebra B, is a pair of a ∗-homomorphism π : A → B and a linear map t : X → B, such that (1) π(a)t(ξ) = t(ϕX (a)(ξ)), (2) t(ξ)∗t(η) = π(hξ, ηiX ), for a ∈ A and ξ, η ∈ X. An easy application of the C∗-identity shows that t(ξ)π(a) = t(ξa) is also valid. A representation (π, t) is said to be injective if π is injective; in that case t is an isometry. The C∗-algebra generated by a representation (π, t) equals the closed lin- ear span of tn(¯ξ)tm(¯η)∗, where for simplicity we used the notation ¯ξ ≡ ξ1 ⊗ · · ·⊗ξn ∈ X ⊗n and tn(¯ξ) ≡ t(ξ1) . . . t(ξn). For any representation (π, t) there exists a ∗-homomorphism ψt : K(X) → B, such that ψt(ΘX ξ,η) = t(ξ)t(η)∗. 6 E.T.A. KAKARIADIS AND E.G. KATSOULIS Let J be an ideal in ϕ−1 X (K(X)); we say that a representation (π, t) is J-coisometric if The representations (π, t) that are JX -coisometric, where ψt(ϕX (a)) = π(a), for any a ∈ J. JX = ker ϕ⊥ X ∩ ϕ−1 X (K(X)), are called covariant representations [32]. The ideal JX is the largest ideal on which the restriction of ϕX is injective. We define the Toeplitz-Cuntz-Pimsner algebra TX as the universal C∗- algebra for all Toeplitz representations of X. Similarly, the Cuntz-Pimsner algebra OX is the universal C∗-algebra for all covariant representations of X. A concrete presentation of both TX and OX can be given in terms of the generalized Fock space FX which we now describe. The Fock space FX over the correspondence X is the interior direct sum of the X ⊗n with the structure of a direct sum of C∗-correspondences over A, FX = A ⊕ X ⊕ X ⊗2 ⊕ . . . . Given ξ ∈ X, the (left) creation operator t∞(ξ) ∈ L(FX ) is defined as t∞(ξ)(ζ0, ζ1, ζ2, . . . ) = (0, ξζ0, ξ ⊗ ζ1, ξ ⊗ ζ2, . . . ), (Here X ⊗0 ≡ A, X ⊗1 ≡ X and where ζn ∈ X ⊗n, n ≥ 0 and ζ0 ∈ A. X ⊗n = X ⊗ X ⊗n−1, for n ≥ 2.) For any a ∈ A, we define π∞(a) ∈ L(FX ) to be the diagonal operator with ϕX (c) ⊗ idn−1 at its X ⊗n-th entry. It is easy to verify that (π∞, t∞) is a representation of X which is called the Fock representation of X. Fowler and Raeburn [19] (resp. Katsura [32]) have shown that the C∗-algebra C∗(π∞, t∞) (resp C∗(π∞, t∞)/K(FX )JX ) is ∗-isomorphic to TX (resp. OX ). Definition 1.9. The tensor algebra T + X of a C∗-correspondence AXA is the norm-closed subalgebra of TX generated by all elements of the form π∞(a), tn ∞(¯ξ), a ∈ A, ¯ξ ∈ X n, n ∈ N. The tensor algebras for C∗-correspondences were pioneered by Muhly and Solel in [40]. They form a broad class of non-selfadjoint operator algebras which includes as special cases Peters' semicrossed products [45], Popescu's non-commutative disc algebras [47], the tensor algebras of graphs (intro- duced in [40] and further studied in [29]) and the tensor algebras for mul- tivariable dynamics [13], to mention but a few. For more examples, see [31]. There is an important connection between T + X and OX given in the fol- lowing Theorem of Katsoulis and Kribs [30]. Recall that, for an opera- tor algebra A and a completely isometric representation ι : A → A, where A = C∗(ι(A)), the pair (A, ι) is called a C∗-cover for A. The C∗-envelope of the operator algebra A is the universal C∗-cover (A, ι) such that, if (B, ι′) is any other C∗-cover for A, then there exists a (unique) ∗-epimorphism OPERATOR ALGEBRAS AND C-CORRESPONDENCES: A SURVEY 7 Φ : B → A, such that Φ(ι′(a)) = ι(a), for any a ∈ A . For the existence of the C∗-envelope see [20, 16, 2, 6, 24] . Theorem 1.10. [30, Theorem 3.7] The C∗-envelope of the tensor algebra T + X is OX . As a consequence the Toeplitz-Cuntz-Pimsner algebra is the extension of the Cuntz-Pimsner algebra by the Silov ideal. (Any ideal J ⊆ C, for a C∗- cover (C, ι) of an operator algebra A, with the property that the restriction of the natural projection C → C/J on A is a complete isometry, is called a boundary ideal and the Silov ideal is the largest such ideal.) Now, let us see how we can generalize the previous facts in the case of arbi- trary C∗-correspondences by using the notion of matrix C∗-correspondences. A representation of an A-B-correspondence X should be a triplet (πA, πB, t) such that πA and πB are ∗-homomorphisms, t is a linear mapping of X and (1) πA(a)t(ξ)πB(b) = t(ϕX (a)(ξ)b), (2) t(ξ)∗t(η) = πB(hξ, ηiX), for all ξ ∈ X, a ∈ A, b ∈ B. If that is the case then one could define 0 t : X → B(H (2)) such that t(ξ) = (cid:20) 0 t(ξ) fines a representation of the (A ⊕ B)-correspondence (cid:20) 0 X if (π, t) is a representation of (cid:20) 0 X resentation of X. Hence, we can identify X with (cid:20) 0 X 0 (cid:21). Then (πA ⊕ πB, t) de- 0 (cid:21). Conversely, 0 (cid:21), then (πA, πB, t) defines a rep- 0 (cid:21) and define 0 0 0 the Toeplitz-Cuntz-Pimsner, the Cuntz-Pimsner and the tensor algebra of the A-B-correspondence X as the corresponding algebras of the (A ⊕ B)- correspondence (cid:20) 0 X 0 (cid:21). However, note that most of the results known for C∗-correspondences over the same C∗-algebra must be verified, basically because X ⊗n is absurd for all n ≥ 2. 0 Remark 1.11. We already gave a brief description of X ∗ of the Hilbert A-module X. When X is a correspondence over A this can be simplified. Let (πu, tu) be the universal representation of AXA; then X ∗ is the closed linear span of t(ξ)∗, ξ ∈ X with the left multiplication and inner product inherited by the trivial correspondence C∗(πu, tu). Via this identification one can produce a theory for the left analogue of C∗-correspondences (that means also, to start with left Hilbert modules), but in most of the cases it can be recovered. 1.6. Examples. One of the fundamental examples in the theory of C∗- correspondences are the C∗-algebras of directed graphs. (For more details see [48].) 8 E.T.A. KAKARIADIS AND E.G. KATSOULIS Let G be a countable directed graph with vertex set G(0), edge set G(1) and range and source maps r and s respectively. A family of partial isome- tries {Le}e∈G(1) and projections {Lp}p∈G(0) is said to obey the Cuntz-Krieger relations associated with G if and only if they satisfy (†)   (1) LpLq = 0 (2) L∗ eLf = 0 (3) L∗ eLe = Ls(e) (4) LeL∗ e ≤ Lr(e) (5) Pr(e)=p LeL∗ ∀ p, q ∈ G(0), p 6= q ∀ e, f ∈ G(1), e 6= f ∀ e ∈ G(1) ∀ e ∈ G(1) e = Lp ∀ p ∈ G(0) with r−1(p) 6= 0, ∞ The relations (†) have been refined in a series of papers by the Australian school and reached the above form in [5, 49]. All refinements involved con- dition (5) and as it stands now, condition (5) gives the equality requirement for projections Lp such that p is not a source and receives finitely many edges. (Indeed, otherwise condition (5) would not be a C∗-condition.) It can been shown that there exists a universal C∗-algebra, denoted as OG, associated with the relations (†). Indeed, one constructs a single family of partial isometries and projections obeying (†). Then, OG is the C∗-algebra generated by a 'maximal' direct sum of such families. It turns out that there OG is the Cuntz-Pimsner algebra of a certain C∗-correspondence [40]. The associated Cuntz-Pimsner -Toeplitz algebra is the universal algebra for the first four relations in (†) and is denoted as TG. Example 1.12. Let G, G′ be two graphs with adjacent matrices AG and AG′. If XG and XG′ are the corresponding C∗-correspondences, then XG ⊗ XG′ is unitarily equivalent to the C∗-correspondence that comes from the adjacent matrix AG · AG′. Let X be an imprimitivity bimodule that comes from a graph G = (G(0), G(1), r, s) (we follow the notation in [48]); then G is either a cycle or a double infinite path. Pick your favorite completely isometric represen- tation of T + X ; for us it is a Cuntz-Krieger family {Pv, Se : v ∈ G(0), e ∈ G(1)} (because of [29]). Apart from the usual relations, due to the form of the e = Pr(e), for all e ∈ G(1). There- graph we have the simplified relation SeS∗ e : v ∈ G(0), e ∈ G(1)} defines a Cuntz-Krieger family of the graph fore {Pv, S∗ G∗ = (G(0), G(1), r∗, s∗) where the arrows are reversed, i.e. r∗ = s and s∗ = r, thus X ∗ is the C∗-correspondence coming from this graph G∗. A second example comes from the class of dynamical systems. Let β : A → B be a ∗-homomorphism of C∗-algebras. The trivial Hilbert module BB becomes a A-B-correspondence, denoted by (Xβ, A), when endowed with the left action ϕB such that ϕB(a)b = β(a)b for all a ∈ A and b ∈ B. Example 1.13. If (Xα, C) is a C-A-correspondence via a ∗-homomorphism α : C → A, and (Xβ, A) is a A-B-correspondence via a ∗-homomorphism β : A → B, then A ⊗A B is unitarily equivalent to the C-B-correspondence (Xβ◦α, C) associated to the ∗-homomorphism β ◦ α : C → B. OPERATOR ALGEBRAS AND C-CORRESPONDENCES: A SURVEY 9 Moreover, (Xβ, A) is an imprimitivity bimodule if and only if β is a ∗- isomorphism. In this case (Xβ, A)∗ ≈ (Xα−1 , B). In particular, when A = B, then OXβ is the usual crossed product B ⋊β Z and the tensor algebra TXβ is Peter's semicrossed product [45] (for this and various types of semicrossed products see also [22, 23, 25]). 2. Adding Tails In [26] the authors developed a method of "adding tails" that extends the one developed by Muhly and Tomforde in [43]. Let G be a connected, directed graph with a distinguished sink p0 ∈ G(0) and no sources. We assume that G is contractible at p0, i.e. the subalgebra CLp0 is a full corner of the Cuntz-Krieger algebra OG. So there exists a unique infinite path w0 = e1e2e3 . . . ending at p0, i.e. r(w0) = p0. Let pn ≡ s(en), n ≥ 1. Let (Ap)p∈G(0) be a family of C∗-algebras parameterized by the vertices of G so that Ap0 = A. For each e ∈ G(1), we now consider a full, right Hilbert As(e) - module Xe and a ∗-homomorphism ϕe : Ar(e) −→ L(Xe) satisfying the following requirements. For e 6= e1, the homomorphism ϕe are required to be injective and map onto K(Xe), i.e. ϕe(Ar(e)) = K(Xe). Therefore, each Xe, e 6= e1, is an Ar(e)-As(e)-equivalence bimodule, in the language of Rieffel. For e = e1, we require K(Xe1 ) ⊆ ϕe1(A) and (1) JX ⊆ ker ϕe1 ⊆ (ker ϕX )⊥ . In addition, there is also a linking condition ϕ−1 e1 (K(Xe1 )) ⊆ ϕ−1 (2) X (K(X)) required between the maps ϕX and ϕe1. (0) − p∈G (0) − p∈G ) denote the c0-sum of the family (Ap) Let T0 = c0( (Ap) G(0) − ≡ G(0)\{p0}. Consider the set c00((Xe)e∈G(1)) ⊆ c0((Xe)e∈G(1)), consist- ing of sequences which are zero everywhere but on a finite number of entries. Equip c00((Xe)e∈G(1)) with a T0-valued inner product defined by , where hu, vi (p) = Xs(e)=p u∗ eve, p ∈ G(0) − , for any u, v ∈ c00((Xe)e∈G(1)). Let T1 be the completion of c00((Xe)e∈G(1)) with respect to that inner product. Equip now T1 with a right T0 - action, so that (ux)e = uexs(e), e ∈ G(1), for any x ∈ T0, so that the pair (T1, T0) becomes a right T0-Hilbert module. The pair (T0, T1) is the tail for AXA. 10 E.T.A. KAKARIADIS AND E.G. KATSOULIS To the C∗-correspondence AXA and the data τ ≡ (cid:16)G, (Xe)e∈G(1), (Ap)p∈G(0), (ϕe)e∈G(1)(cid:17), we now associate (3) Aτ ≡ A ⊕ T0 Xτ ≡ X ⊕ T1. Using the above, we view Xτ as a Aτ -Hilbert module with the standard right action and inner product for direct sums of Hilbert modules. We also define a left Aτ -action ϕτ : Aτ → L(Xτ ) on Xτ by setting where ϕτ (a, x )(ξ, u) = (ϕX (a)ξ, v), ve = (cid:26) ϕe1(a)(ue1), ϕe(xr(e))ue, if e = e1 otherwise for a ∈ A, ξ ∈ X, x ∈ T0 and u ∈ T1. Theorem 2.1. [26, Theorem 3.10] Let AXA be a non-injective C∗- corre- spondence and let Xτ be the graph C∗-correspondence over Aτ defined above. Then Xτ is an injective C∗-correspondence and the Cuntz-Pimsner algebra OX is a full corner of OXτ . Regarding Theorem 2.1 and the conditions imposed on the graph G and the maps (ϕe)e∈G(1), we have asked that the graph G be contractible. We cannot weaken this assumption to include more general graphs. Indeed, we want the tail associated with the data τ = (cid:16)G, (Xe)e∈G(1), (Ap)p∈G(0), (ϕe)e∈G(1)(cid:17), to work with any possible Cuntz-Pimsner algebra OX that can be "added on". This should apply in particular to the Cuntz-Krieger algebra OGp0 of the (trivial) graph Gp0 consisting only of one vertex p0. By taking τ to be the "usual" tail associated with G, i.e. Xe = Ae = CLp0 and ϕe left multiplication for all e, we see that OGp0 is a full corner of OXτ if and only if G is contractible at p0. Conditions (1) and (2) are also necessary, as the following result suggests. Proposition 2.2. Let AXA be a non-injective C∗-correspondence and let Xτ be the C∗-correspondence over Aτ associated with the data τ = (cid:16)G, (Xe)e∈G(1), (Ap)p∈G(0), (ϕe)e∈G(1)(cid:17), as defined at the beginning of the section. If Xτ is injective, (ϕ−1 1 (K(X1)) + JX) ⊕ 0 ⊆ JXτ , and the covariant representations of Xτ restrict to covariant representations of ϕX , then JX ⊆ ker ϕ1 ⊆ (ker ϕX )⊥ , OPERATOR ALGEBRAS AND C-CORRESPONDENCES: A SURVEY 11 and the linking condition ϕ−1 1 (K(X1)) ⊆ ϕ−1 X (K(X)) holds. Lets see now how the work of Muhly and Tomforde fits in our theory. Example 2.3 (The Muhly-Tomforde tail [43]). Given a (non-injective) cor- respondence (X, A, ϕX ), Muhly and Tomforde construct in [43] the tail that results from the previous construction, with respect to data τ = (cid:16)G, (Xe)e∈G(1), (Ap)p∈G(0), (ϕe)e∈G(1)(cid:17) defined as follows. The graph G is illustrated in the figure below. •p0 •p1 •p2 e3 •p3 e2 e1 • . . . Ap = Xe = ker ϕX , for all p ∈ G(0) − and e ∈ G(1). Finally, ϕe(a)ue = aue, for all e ∈ G(1), ue ∈ Xe and a ∈ Ar(e). 2.1. Applications. 2.1.1. Semicrossed Products. The tail of Muhly and Tomforde has had sig- nificant applications in the theory of C∗-correspondences, including a char- acterization for the C∗-envelope of the tensor algebra of a non-injective cor- respondence [30]. However, it also has its limitations, as we are about to see. Example 2.4. Let (Xα, A) be the C∗-correspondence canonically associated with a dynamical system (A, α) and let O(A,α) be the associated Cuntz- Pimsner C∗-algebra. If α is not injective, then by using the Muhly-Tomforde tail we obtain an injective C∗-correspondence (Y, B, ϕY ) so that O(A,α) is a full corner of OY . Remarkably, (Y, B, ϕY ) may not come from any dynamical system in general. Assume that ker α ⊆ A is an essential ideal of A; then the Muhly-Tomforde tail produces an injective correspondence (Y, B, ϕY ) with Y = A ⊕ c0(ker α), B = A ⊕ c0(ker α) and ϕY defined by 1, c1c′ 2, c2c′ 3, . . .(cid:1), ϕY(cid:0)a, (ci)i(cid:1)(cid:0)a′, (c′ where a, a′ ∈ A and (ci)i, (c′ i)i(cid:1) = (cid:0)α(a)a′, α(a)c′ i)i ∈ c0(ker α). If there was a ∗-homomorphism β satisfying ϕY (b)(b′) = β(b)b′, (4) o o o o o o o o o o 12 E.T.A. KAKARIADIS AND E.G. KATSOULIS then by equating second coordinates in the equation we would obtain, ϕY(cid:0)1, (ci)i(cid:1)(cid:0)a′, (c′ i)i(cid:1) = β(cid:0)1, (ci)i(cid:1)(cid:0)a′, (c′ i)i(cid:1) for all c′ = 1, a contradiction. 1 ∈ ker α. Since ker α is an essential ideal, we have ker α ∋ β(cid:0)1, (ci)i(cid:1)2 c′ 1 = β(cid:0)1, (ci)i(cid:1)2c′ 1, Therefore, the Muhly-Tomforde tail produces an injective correspondence but not necessarily an injective dynamical system. Nevertheless, there exists a tail that can be added to (Xα, A) and produce an injective correspondence that comes from a dynamical system. Example 2.5. If (Xα, A) is the C∗-correspondence canonically associated with a dynamical system (A, α), then the appropriate tail for (Xα, A) comes from the data τ = (cid:16)G, (Xe)e∈G(1), (Ap)p∈G(0), (ϕe)e∈G(1)(cid:17) where G is as in Example 2.4 but for any p ∈ G(0) − and e ∈ G(1), Ap = Xe = θ(A), where θ : A −→ M (ker α) is the map that extends the natural inclusion ker α ⊆ M (ker α)) in the multiplier algebra. Finally ϕe(a)ue = aue, for all e ∈ G(1), ue ∈ Xe and a ∈ Ar(e). Then the correspondence Xτ is canonically associated to the dynamical system (B, β), where B = A ⊕ c0(θ(A) and β(a, (ci)i) = (α(a), θ(a), (ci)i). 2.1.2. Multivariable Dynamical Systems. We now apply the method of adding tails to C∗-dynamics. Apart from their own merit, this application will also address the necessity of using more elaborate tails than that of Muhly and Tomforde in the process of adding tails to C∗-correspondences. This neces- sity has been already noted in the one-variable case. A multivariable C∗-dynamical system is a pair (A, α) consisting of a C∗- algebra A along with a tuple α = (α1, α2, . . . , αn), n ∈ N, of ∗- endomor- phisms of A. The dynamical system is called injective if ∩n i=1 ker αi = {0}. In the C∗-algebra literature, the algebras O(A,α) are denoted as A ×α On and go by the name "twisted tensor products by On". They were first intro- duced and studied by Cuntz [8] in 1981. In the non-selfadjoint literature, these algebras are much more recent. In [13] Davidson and the second named author introduced the tensor algebra T(A,α) for a multivariable dy- namical system (A, α). It turns out that T(A,α) is completely isometrically isomorphic to the tensor algebra for the C∗-correspondence (Xα, A, ϕα). As such, O(A,α) is the C∗-envelope of T(A,α). Therefore, O(A,α) provides a very OPERATOR ALGEBRAS AND C-CORRESPONDENCES: A SURVEY 13 important invariant for the study of isomorphisms between the tensor alge- bras T(A,α). To the multivariable system (A, α) we associate a C∗-correspondence i=1A be the usual right A-module. (Xα, A, ϕα) as follows. Let Xα = An = ⊕n That is (1) (a1, . . . , an) · a = (a1a, . . . , ana), (2) h(a1, . . . , an), (b1, . . . , bn)i = Pn Also, by defining the ∗-homomorphism i=1 hai, bii = Pn i=1 a∗ i bi. ϕα : A −→ L(Xα) : a 7−→ ⊕n i=1αi(a), Xα becomes a C∗-correspondence over A, with ker ϕα = ∩n i=1 ker αi and ϕ(A) ⊆ K(Xα). It is easy to check that in the case where A and all αi are unital, Xα is finitely generated as an A-module by the elements e1 := (1, 0, . . . , 0), e2 := (0, 1, . . . , 0), . . . , en := (0, 0, . . . , 1), In that case, (π, t) is a representation of this C∗- corre- where 1 ≡ 1A. spondence if and only if t(ξi) are isometries with pairwise orthogonal ranges and π(c)t(ξ) = t(ξ)π(αi(c)), i = 1, . . . , n. Definition 2.6. The Cuntz-Pimsner algebra O(A,α) of a multivariable C∗- dynamical system (A, α) is the Cuntz-Pimsner algebra of the C∗- correspon- dence (Xα, A, ϕα) constructed as above. The graph G that we associate with (Xα, A, ϕα) has no loop edges and a single sink p0. All vertices in G(0)\{p0} emit n edges, i.e. as many as the maps involved in the multivariable system, and receive exactly one. In the case where n = 2, the following figure illustrates G. (PPPPPPPPPPP v♥♥♥♥♥♥♥♥♥ (PPP . . . •q3 v♥♥♥♥ v♥♥♥♥ •p3 •q2 e3 (PPPPPPPPP v♥♥♥♥♥♥♥♥♥ (PPP . . . •p2 •p0 e2 (PPPPPPPPP v♥♥♥♥♥♥♥♥♥ e1 •p1 (PPPPPPPPP v♥♥♥♥♥♥♥♥♥♥ (PPPPP •q1 (PPPPPPPPPPP • v♥♥♥♥♥ Clearly, G is p0-accessible. There is also a unique infinite path w ending at p0 and so G is contractible at p0. ( ( ( v ( v ( v v v ( v ( v ( 14 E.T.A. KAKARIADIS AND E.G. KATSOULIS Let J ≡ ∩n i=1 ker αi and let M (J ) be the multiplier algebra of J . Let θ : A −→ M (J ) the map that extends the natural inclusion J ⊆ M (J )). Let Xe = As(e) = θ(A), for all e ∈ G(1), and consider (Xe, As(e)) with the natural structure that makes it into a right Hilbert module. For e ∈ G(1)\{e1} we define ϕe(a) as left multiplication by a. With that left action, clearly Xe becomes an Ar(e)-As(e)-equivalence bimodule. For e = e1, it is easy to see that ϕe1(a)(θ(b)) ≡ θ(ab), a, b ∈ A defines a left action on Xe1 = θ(A), which satisfies both (1) and (2). Theorem 2.7. [26, Theorem 4.2] If (A, α) is a non-injective multivari- able C∗-dynamical system, then there exists an injective multivariable C∗- dynamical system (B, β) so that the associated Cuntz-Pimsner algebras O(A,α) is a full corner of O(B,β). Moreover, if A belongs to a class C of C∗-algebras which is invariant under quotients and c0-sums, then B ∈ C as well. Fur- thermore, if (A, α) is non-degenerate, then so is (B, β). The reader familiar with the work of Davidson and Roydor may have noticed that the arguments in the proof of Theorem 2.7, when applied to multivariable systems over commutative C∗-algebras produce a tail which is different from that of Davidson and Roydor in [14, Theorem 4.1]. It turns out that the proof of [14, Theorem 4.1] contains an error and the technique of Davidson and Roydor does not produce a full corner, as claimed in [14]. Nevertheless, [14, Theorem 4.1] is valid as Theorem 2.7 demonstrates (a fact also mentioned in [14, Corrigendum]). We illustrate this by examining their arguments in the following simple case. Example 2.8. (The Davidson-Roydor tail [14]). Let X ≡ {u, v} and consider the maps σi : X → X , i = 1, 2, with σi(u) = v and σi(v) = v. Set σ ≡ (σ1, σ2) and let O(X ,σ) be the Cuntz-Pimsner algebra associated with the multivariable system (X , σ), which by [13] is the C∗-envelope of the associate tensor algebra. We now follow the arguments of [14]. In order to obtain O(X ,σ) as a full corner of an injective Cuntz-Pimsner algebra, Davidson and Roydor add a tail to the multivariable system. They define T = {(u, k) k < 0} and X T = X ∪ T. For each 1 ≤ i ≤ 2, they extend σi to a map σT i : X T → X T by σT (u, k) = (u, k + 1) for k < −1 and σT i (u, −1) = u. They then consider the new multivariable system (X T , σT ) and its associated Cuntz-Pimsner algebra O(X T , σT ). It is easy to see that the Cuntz-Pimsner algebra O(X ,σ) for the multivari- able system (X , σ) is the Cuntz-Krieger algebra OG of the graph G illustrated below, while the Cuntz-Pimsner algebra O(X T , σT ) is isomorphic to the OPERATOR ALGEBRAS AND C-CORRESPONDENCES: A SURVEY 15 Cuntz-Krieger algebra OGT of the following graph GT , where for simplicity we write uk instead of (u, k), k < 0, G : •v •u , GT : •v •u e f •u−1 •u−2 . . . In [14, page 344], it is claimed that the projection P associated with the characteristic function of X ⊆ X T satisfies P O(X T ,σT )P = O(X ,σ) and so O(X ,σ) is a corner of O(X T ,σT ). In our setting, this claim translates as follows: if P = Lu + Lv, then P OGT P = OG. However this is not true. For instance, P (Lf L∗ e)P = Lf L∗ e /∈ OG. 2.1.3. Multivariable Dynamical Systems and Crossed Products by endomor- phism. We can describe the Cuntz-Pimsner algebra of an injective and non- degenerate multivariable system as a crossed product of a C∗-algebra B by an endomorphism β. This idea appears first in [14] for a different crossed product than the one presented here. We start with the pertinent definitions. Definition 2.9. Let B be a (not necessary unital) C∗-algebra and let β be an injective endomorphism of B. A covariant representation (π, v) of the dynamical system (B, β) consists of a non-degenerate ∗-representation π of B and an isometry v satisfying (i) π(β(b)) = vπ(b)v∗, ∀ b ∈ B, i.e. v implements β, (ii) v∗π(B)v ⊆ π(B), i.e. v is normalizing for π(B), (iii) vk(v∗)kπ(B) ⊆ π(B), ∀ k ∈ N. The crossed product B ×β N is the universal C∗-algebra associated with this concept of a covariant representation for (B, β). Specifically, B ×β N is generated by B and BV , where V is an isometry satisfying (i), (ii) and (iii) in Definition 2.9 with π = id. Furthermore, for any covariant representation (π, v) of (B, β), there exists a ∗-homomorphism π : B ×β N → B(H) extending π and satisfying π(bV ) = π(b)v, for all b ∈ B. In the case where B is unital, condition (iii) is redundant and this version of a crossed product by an endomorphism was introduced by Paschke [44]; in the generality presented here, it is new. It has the advantage that for any covariant representation of (π, v) of (B, β) admitting a gauge action, the fixed point algebra of (π, v) equals π(B) (see Remark 2.10 below). This allows us to claim a gauge invariance uniqueness theorem for B ×β N: if (π, v) is a faithful covariant representation of (B, β) admitting a gauge action, then the C∗-algebra generated by π(B) and π(B)v is isomorphic to B ×β N. Remark 2.10. Observe that conditions (i), (ii) and (iii) in Definition 2.9 imply that π (B ×β N) is generated by polynomials of the form π(b0) + b0, bk, bl ∈ B. Indeed, item (i) implies that Pk π(bk)vk + Pl (v∗)lπ(bl), C C i i u u C C i i u u j j t t k k s s l l r r 16 E.T.A. KAKARIADIS AND E.G. KATSOULIS vπ(b) = π(β(b))v, thus π(b)v∗ = v∗π(β(b)), for all b ∈ B (since B is self- adjoint). With this in hand and item (ii) of the definition we can see that the product (v∗)kπ(bk)π(bl)vl can be written as a trigonometric polynomial of the above form, for all k, l ≥ 0. Item (iii) is used to show the same thing for the products π(bk)vk(v∗)lπ(bl), when k ≥ l. Finally, for k < l we use the fact that v is an isometry and item (iii) so that π(bk)vk(v∗)lπ(bl) = π(bk)(v∗)l−kvl(v∗)lπ(bl) = π(bk)(v∗)l−kπ(b′) = (v∗)l−kπ(b′′), which completes the argument. There is a related concept of a crossed product by an endomorphism which we now discuss. For a C∗-algebra B and an injective endomorphism β, Stacey [50] imposes on a covariant representation (π, v) of (B, β) only condition (i) from Definition 2.9. He then defines the crossed product B ⋊β N to be the universal C∗-algebra associated with his concept of a covariant representation for (B, β). Muhly and Solel have shown [41] that in the case where B is unital, Stacey's crossed product is the Cuntz-Pimsner algebra of a certain correspondence. Using a gauge invariance uniqueness theorem one can prove that if the isometry V in B ⋊β N satisfies condition (ii) in Definition 2.9, then B ⋊β N ≃ B ×β N. Theorem 2.11. [26, Theorem 4.6] If (A, α) is an injective multivariable system, then there exists a C∗-algebra B and an injective endomorphism β of B so that O(A,α) is isomorphic to the crossed product algebra B ×β N. Furthermore, if A belongs to a class C which is invariant under direct limits and tensoring by Mk(C), k ∈ N, then B also belongs to C. Combining Theorem 2.11 with Theorem 2.7 we obtain the following. Corollary 2.12. If (A, α) is a multivariable system, then there exists a C∗- algebra B and an injective endomorphism β of B so that O(A,α) is isomorphic to a full corner of the crossed product algebra B ×β N. Furthermore, if A belongs to a class C which is invariant under direct limits, quotients and tensoring by Mk(C), k ∈ N, then B also belongs to C. Finally, let us give a quick application of Theorem 2.11, that readily follows from Paschke's result [44] on the simplicity of B ×β N. Corollary 2.13. Let A be a UHF C∗-algebra and let α = (α1, . . . , αn) be a multivariable system with n ≥ 2. If αi(1) = 1, for all i = 1, 2, . . . , n, then O(A,α) is simple. 2.2. Graph C∗-correspondences. Graph C∗-correspondences can be used to summarize in a very delicate way properties and constructions of C∗- correspondences. Let G = (G(0), G(1), r, s) be a directed graph. Let (Ap)p∈G(0) be a family of C∗-algebras parameterized by the vertices of G and for each e ∈ G(1), we consider Xe be a Ar(e)-As(e)-correspondences Xe. OPERATOR ALGEBRAS AND C-CORRESPONDENCES: A SURVEY 17 Let AG = c0( (Ap)p∈G(0)) denote the c0-sum of the family (Ap)p∈G(0). Also, let Y0 = c00((Xe)e∈G(1)) which is equipped with a AG-valued inner product hu, vi (p) = Xs(e)=p hue, veiAp , p ∈ G(0), If XG is the completion of Y0 with respect to the inner product, then XG is a AG-Hilbert module when equipped with the right action (ux)e = uexs(e), e ∈ G(1). It becomes a C∗-correspondence, when equipped with the ∗-homomorphism ϕG : AG → L(XG), such that (ϕG (x)u)e = ϕe(xr(e))(ue), e ∈ G(1), for x ∈ AG and u ∈ XG. The C∗-correspondence XG over AG is called the graph C∗-correspondence with respect to the data (cid:8)G, {Ap}p∈G(0), {Xe}e∈G(1)(cid:9). In other words, a graph C∗-correspondence can be viewed as a graph G such that on every vertex sits a C∗-algebra, on each edge sits a C∗-correspondence and the actions and the inner product are defined via the information pro- vided by the graph (they "remember" the form of the graph) as in the following image ❖ ❊ ❴❤ ✾ •Ar(e) Xe ✆ •Bs(e) ♦ ② ❱❴ Example 2.14. Every C∗-correspondecne X over A can be visualized in the language of graph C∗-correspondences trivially as X •A Moreover, every AXB C∗-correpondence can be visualized as •A X •B The Fock space of a graph C∗-correspondence contains the path C∗- correspondences Xµ, for some path µ. That is if µ = xen . . . e1y is a path of the graph G then Xµ := Xen ⊗Ar(en) · · · ⊗Ar(e2) Xe1 is a Ax-Ay- correspondence. Definition 2.15. A graph C∗-correspondence is called commutative if for any two paths µ = xen . . . e1y and ν = xfk . . . f1y, that have the same range and source, the corresponding path C∗-correspondences Xen ⊗ · · · ⊗ Xe1 and Xfn ⊗ . . . Xf1 are unitarily equivalent. j j v v r r q q * 18 E.T.A. KAKARIADIS AND E.G. KATSOULIS For example if E ≈ R ⊗B S for some C∗-correspondences AEA, ARB and BSA, then one can form the commutative graph C∗-correspondence •A E •A S R •B Note that the graph C∗-correspondence that comes from this graph does not contain AEA as a subcorrespondence, though. A graph C∗-correspondence that has additionally that property is the following E •A * •B S R Example 2.16. Let Xτ be the correspondence that occurs from the adding- tail construction established in [26]. It can be visualized in the language of graph C∗-correspondences as follows . . . Xe1 Xe2 X •Ap0 . . . where Ap0 ≡ A. . . . •A1 . . . Xe4 Xe3 Xe5 •A2 . . . Example 2.17. (Muhly-Tomforde tail [43]) When X is visualized on a cycle graph, then Muhly-Tomforde tail produces the following graph C∗- correspondence X ker ϕX ker ϕX ker ϕX •A •ker ϕX •ker ϕX · · · Example 2.18. If (Xα, A) is the C∗-correspondence canonically associated with a dynamical system (A, α), then the tail produced by Muhly and Tom- forde for (Xα, A) comes from the graph X ker α ker α ker α •A •ker α •ker α · · · Example 2.19. If (Xα, A) is the C∗-correspondence canonically associated with a dynamical system (A, α), then the appropriate tail for (Xα, A) comes from the graph X θ(A) θ(A) θ(A) •A •θ(A) •θ(A) · · · , , ' ' F F * j j q q k k r r k k s s t t q q q q t t r r r r t t r r r r OPERATOR ALGEBRAS AND C-CORRESPONDENCES: A SURVEY 19 where θ : A −→ M (ker α) is the map that extends the natural inclusion ker α ⊆ M (ker α)) in the multiplier algebra. Therefore, the C∗- corre- spondence (Y, ϕY , B) is the C∗-correspondence canonically associated with a dynamical system (B, β), where B = A ⊕ c0(θ(A)) and β(a, (xn)) = (α(a), θ(a), (xn)). Example 2.20. If (Xα, A) is the C∗-correspondence canonically associated with a C∗-dynamical system (A, α1, α2), then the appropriate tail for (Xα, A) comes from the graph X •A θ(A) θ(A) . . . •θ(A) . . . θ(A) θ(A) θ(A) •θ(A) . . . where J = ker α1 ∩ ker α2 and θ : A −→ M (J ) is the map that extends the natural inclusion J ⊆ M (J ) in the multiplier algebra. 3. Shift Equivalence Problem In his pioneering paper [52] Williams studied certain notions of relations for the class of matrices with non-negative integer entries. We say that two such matrices E and F are elementary strong shift equivalent, and write s E ∼ F , if there are two matrices R and S such that E = RS and F = SR. Whereas this relation is symmetric, it may not be transitive. In [52, Example 2] Williams gives the following counterexample; let (cid:20) 10 2 1 (cid:21) s 2 ∼ (cid:20) 5 6 4 6 (cid:21) s ∼ (cid:20) 9 4 3 2 (cid:21) , 2 ∼, denoted by SSE 1 (cid:21) is not elementary strong shift equivalent to (cid:20) 9 4 but (cid:20) 10 2 transitive closure s ∼ F if there is a se- SSE ∼ Ti+1. quence of matrices Ti, i = 0, . . . , n, such that E = T0, F = Tn and Ti Williams also defined a third relation, which is proved to be transitive. We say that E is shift equivalent to F , and write E SE ∼ F , if there are R, S such that En = RS, F n = SR and ER = SF , F R = SE for some n ∈ N. 3 2 (cid:21). The ∼ , implies that ESSE ∼ and SE A purpose of [52] was to prove that the relations SSE ∼ are equiva- lent. Unfortunately, an error in [52] made invalid the proof of a key lemma, and this task remained unsolved for over than 20 years, known as Williams' Conjecture. The research interest in this area contributed to the growth of symbolic dynamics and to the search of (complete) invariants for both the equivalence relations. A major change was made by Kim and Roush in [34], where they proved that Williams' Conjecture was false for the class of non- negative integral matrices. Their work suggests that Williams' Conjecture SE can be renamed as Shift Equivalence Problem; i.e. ∼ SSE ∼ equivalent to is r r r r { { r r 20 E.T.A. KAKARIADIS AND E.G. KATSOULIS for a class S. This formulation is a little vague as one has to extend the definition of the relations described above to a class S. The notion of elementary and strong shift equivalence for C∗- correspon- dences was studied by Muhly, Tomforde and Pask [39]. In addition they prove that strong shift equivalence of C∗-correspondences implies the Morita equivalence of the associated Cuntz-Pimsner algebras, thus extending clas- sical results of Cuntz and Krieger [9], Bates [4] and Drinen and Sieben [15] for graph C∗-algebras. The concept of shift equivalence has been studied extensively from both the dynamical and the ring theoretic viewpoint. (See [51] for a comprehensive exposition.) In general, shift equivalence has been recognized to be a more manageable invariant than strong shift equivalence, as it is decidable over certain rings [33]. Unlike strong shift equivalence, the study of shift equivalence, from the viewpoint of C∗-correspondences, has been met with limited success [38]. (Other operator theoretic view- points however have been quite successful [36].) The concept of strong Morita equivalence for C∗-correspondences was first developed and studied by Abadie, Eilers and Exel [1] and Muhly and Solel [42], and plays the role of a generalized Conjugacy (see Example 3.2 below). Among others these authors show that if two C∗-correspondences are strong Morita equivalent then the associated Cuntz-Pimsner algebras OE and OF are (strong) Morita equivalent as well. Let us give the definitions. Definition 3.1. Let the C∗-correspondences AEA and BFB. Then we say that (1) E is Morita equivalent to F , and we write E SME ∼ F , if there is an imprimitivity bimodule AMB such that E ⊗A M = M ⊗B F . (2) E is elementary strong shift equivalent to F , and we write E s ∼ F , if there are ARB and BSA such that E = R ⊗A S and F = S ⊗B R, SSE ∼ F , if there are Ti, i = 0, . . . n, such that T0 = E, Tn = F and Ti (3) E is strong shift equivalent to F , and we write E s ∼ Ti+1, SE ∼ F , if there are ARB and BSB such that E⊗m = R ⊗B S, F ⊗m = S ⊗A R and E ⊗A R = R ⊗B F , S ⊗A E = F ⊗B S. (4) E is shift equivalent to F with lag m, and we write E Example 3.2. Let AEA and BFB be C∗-correspondences arising from two dynamical systems (A, α) and (B, β) and assume that E SME ∼ F via an imprimitivity bimodule M . If we wish M to arise in a similar way then it should be the C∗-correspondence associated to a ∗-isomorphism γ : A → B. As showed in Example 1.13, E ⊗A M is the C∗-correspondence associated to γ ◦α : A → B and M ⊗B B is the C∗-correspondence associated to β ◦γ : A → B. Therefore the unitary equivalence E ⊗A M ≈ M ⊗B B induces a unitary u ∈ L(B) = B, such that a · u(1B) = u(a · 1B), for all a ∈ A; equivalently that β ◦ γ(a)u = uγ ◦ α(a), for all a ∈ A, hence that the systems (A, α) and (B, β) are (outer) conjugate. OPERATOR ALGEBRAS AND C-CORRESPONDENCES: A SURVEY 21 In [27] we studied these relations and the interaction between them. Note ∼ are equivalence relations for non-degenerate C∗- corre- that SME spondences. One of our main result is the following. ∼ and SE ∼ , SSE Theorem 3.3. [27] The Shift Equivalence Problem Problem is true for the SE ∼ class of imprimitivity bimodules. In particular, the relations coincide. SME ∼ , s ∼, SSE ∼ , A weaker version of the previous Theorem holds in general. Theorem 3.4. [27] Let the C∗-correspondences AEA and BFB; then E SME ∼ F 3 E s ∼ F 3 E SSE ∼ F 3 E SE ∼ F . In [39] Muhly, Pask and Tomforde have provided a number of counterex- amples showing that Morita equivalence of C∗-correspondences differs from the elementary strong shift equivalence. The previous result shows that it is in fact stronger. One of the basic tools we use in [27] is the Pimsner dilation of an injective C∗-correspondence X to a Hilbert bimodule X∞. This construction was first introduced by Pimsner in [46]. In [26, Appendix A] we revisited this construction by using direct limits. Consider the directed system ρ0−→ L(X) ρ1−→ L(X ⊗2) ρ2−→ · · · , A where ρ0 = ϕX : A = L(A) −→ L(X), ρn : L(X ⊗n) −→ L(X ⊗n+1) : r 7−→ r ⊗ idX, n ≥ 1, (L(X ⊗n), ρn) that is generated and let A∞ be the C∗-subalgebra of B = lim −→ by the copies of K(X ⊗n), for n ∈ Z+. Consider also the directed system of Banach spaces where L(A, X) σ0−→ L(X, X ⊗2) σ1−→ · · · , σn : L(X ⊗n, X ⊗n+1) → L(X ⊗n+1, X ⊗n+2) : s 7→ s ⊗ idX, n ≥ 1, and let X∞ be the Banach subalgebra of Y = lim −→ erated by the copies of K(X ⊗n, X ⊗n+1), for n ∈ Z+. Note that the map (L(X ⊗n, X ⊗n+1), σn) gen- ∂ : X −→ L(A, X) : ξ 7−→ ∂ξ, where ∂ξ(a) = ξa, ξ ∈ X, maps a copy of X isometrically into K(A, X) ⊆ In particular, if ϕX (A) ⊆ K(X), then one can verify that X∞ = X∞. (K(X ⊗n), ρn). Thus, in this case, lim −→ X∞ is a full left Hilbert bimodule. (K(X ⊗n, X ⊗n+1), σn) and A∞ = lim −→ If r ∈ L(X ⊗n), s ∈ L(X ⊗n, X ⊗n+1) and [r], [s] are their equivalence classes in B and Y respectively, then we define [s] · [r] := [sr]. From this, it + + + 22 E.T.A. KAKARIADIS AND E.G. KATSOULIS is easy to define a right B-action on Y . Similarly, we may define a B-valued right inner product on Y by setting for s, s′ ∈ L(X ⊗n, X ⊗n+1), n ∈ N, and then extending to Y × Y . Finally we define a ∗-homomorphism ϕY : B → L(Y ) by setting (cid:10)[s′], [s](cid:11)Y ≡ [(s′)∗s] ∈ B. ϕY ([r])([s]) ≡ [rs], r ∈ L(X ⊗n), s ∈ L(X ⊗n−1, X ⊗n), n ≥ 0 and extending to all of B by continuity. We therefore have a left B-action on Y and thus Y becomes a C∗-correspondence over B. The following diagrams depict the above construction in a heuristic way: the right action is "defined" through the diagram A · ϕX ≡ρ0 L(X) · ρ2 L(X ⊗2) ρ3 · . . . · L(A, X) σ0 / / L(X, X ⊗2) σ1 / / L(X ⊗2, X ⊗3) σ3 / . . . B · / Y while the left action is "defined" through the diagram A ϕX ≡ρ0 / L(X) ρ2 L(X ⊗2) ρ3 id· xq q q q q q σ0 / L(A, X) id· w♥ ♥ ♥ ♥ ♥ ♥ id· s s / L(X ⊗2, X ⊗3) ys / L(X, X ⊗2) σ1 / . . . / B ϕY s s s / . . . / Y For a proof of the following Theorem see [46, Theorem 2.5] or [26, The- orem 6.6]. The main difference between the two approaches is that in [26] we have represented X ⊗A A∞ as the direct limit X∞, thus having no con- cern in checking the form of the tensor product. Moreover, it appears that "tensoring" X with A∞ is equivalent to multiplying X with A∞ in lim −→ Theorem 3.5. [46, Theorem 2.5] [26, Theorem 6.6] Let AXA be an injec- tive C∗-correspondence and let X∞ be the A∞-correspondence defined above. Then X∞ is an essential Hilbert bimodule and its Cuntz-Pimsner algebra OX∞ is ∗-isomorphic to OX . (L(X ⊗n, X ⊗n+1), σn). The idea of using direct limits for Pimsner dilation is in complete analogy to the direct limit process for dynamical systems (see [25]). Example 3.6. Let (A, α) denote a dynamical system where α is a ∗- injec- tive endomorphism of A. We can define the direct limit dynamical system (A∞, α∞) by A α / A α / A α / · · · / A∞ α α α α∞ A α / / A α / / A α / / · · · / A∞ / / ✤ ✤ ✤ / / ✤ ✤ ✤ / / ✤ ✤ ✤ / / ✤ ✤ ✤ ✤ ✤ ✤ ✤ / / / / / x / / w y / ✤ ✤ ✤ / / / / / / / OPERATOR ALGEBRAS AND C-CORRESPONDENCES: A SURVEY 23 The limit map α∞ is an automorphism of A∞ and extends α (note that A imbeds in A∞ since α is injective). Then the A∞-A∞-correspondence Xα∞ , is the Pimsner dilation of Xα. The main question imposed in [27] was the following: is it true that SE E∞ ∼ F∞ (where ∼ may be ∼) if E ∼ F (in the same way)? When E and F are non-degenerate and regular C∗-correspondences we get the following result, by making use of the generalized notion of dilations. SME ∼ , SSE ∼ , s ∼, Theorem 3.7. [27] Let AEA and BFB be non-degenerate and regular C∗- SE correspondences. If E ∼ F (where ∼ may be ∼ ), then E∞ ∼ F∞ (in the same way). SME ∼ , SSE ∼ , s ∼, For full right, non-degenerate, regular C∗-correspondences we have the following. Theorem 3.8. [27] Let AEA and BFB be full right, non-degenerate, regular C∗-correspondences. Then the above scheme holds E SME ∼ F 3 E s ∼ F 3 E SSE ∼ F 3 E SE ∼ F E∞ SME ∼ F∞ k 3 E∞ s ∼ F∞ k 3 E∞ SSE ∼ F∞ k 3 E∞ SE ∼ F∞ The vertical arrows in Theorem 3.8 are not equivalences in general. In- SE ∼ E∞. The follow- deed, if E∞ ∼ F∞ implied E ∼ F then, in particular E ing Theorem shows that this happens only trivially, i.e. when E = E∞. Theorem 3.9. [27] Let AEA be a full, non-degenerate and regular C∗- correspondence. If E SE ∼ E∞ then E is an imprimitivity bimodule. It would be interesting if we could prove the validity of the Shift Equiv- alence Problem for this class of C∗-correspondences (as, after imprimitivity bimodules, it is the next best thing). An obstacle that prevents the con- struction of counterexamples that would give a negative answer to the Shift Equivalence Problem for non-degenerate and regular correspondences is that the theory of invariants of correspondences is rather poor. The best results (to our opinion) obtained so far are those appearing in [42, 39] which we state for sake of completeness. Note that the additional item (3) below is an immediate consequence of item (2) and item (4) below is an immediate consequence of Theorem 3.5 and Theorem 3.8. Theorem 3.10. (1) [42, Theorem 3.2] Let AEA and BFB be non- de- SME generate, injective C∗-correspondences. ∼ F then the correspond- ing Toeplitz-Cuntz-Pimsner algebras and Cuntz-Pimsner algebras are strong Morita equivalent as C∗-algebras, and the corresponding tensor algebras are strong Morita equivalent in the sense of [7]. If E + + + s + s + s + 24 E.T.A. KAKARIADIS AND E.G. KATSOULIS (2) [39, Theorem 3.14] Let AEA and BFB be non-degenerate, regular C∗- correspondences. If E ∼ F , then OE s SME ∼ OF . (3) [27] Let AEA and BFB be non-degenerate, regular C∗- correspon- dences. If E ∼ F , then OE SSE SME ∼ OF . (4) [27] Let AEA and BFB be full right, non-degenerate, regular C∗- correspondences. If E SE ∼ F , then OE SME ∼ OF . In particular we obtain the following result for Cuntz-Krieger C∗-algebras. Corollary 3.11. Let G and G′ be finite graphs with no sinks or sources SE ∼ AG ′, in the sense of and let AG and AG ′ be their adjacent matrices. If AG Williams, then the Cuntz-Krieger C∗-algebras OG and OG ′ are strong Morita equivalent. There is also a direct application to unital injective dynamical systems. Corollary 3.12. Let (A, α) and (B, β) be unital injective dynamical sys- tems. If Xα and B ⋊β∞ ∼ Yβ∞ and the crossed products A∞ ⋊α∞ Z are strong Morita equivalent. ∼ Xβ, then Xα∞ SE SE Z s ∼, SME ∼ , Theorem 3.10 shows that Cuntz-Pimsner algebras is a rather coarse in- variant. After all, Cuntz-Pimsner algebras are not a complete invariant SE ∼ to subclasses of C∗- for the restriction of the relations correspondences. For example, let (A, α) be the dynamical system con- structed by Hoare and Parry in [21]. Then α is a ∗-isomorphism and α is not conjugate to its inverse. If E is the C∗-correspondence of (A, α) and F is the C∗-correspondence of (A, α−1), then there is not a C∗-correspondence M of a dynamical system (B, β) such that E ⊗ M ≈ M ⊗ F , because then the two dynamical systems would be conjugate. But OA = A ⋊α Z is always ∗-isomorphic to OB = A ⋊ SSE ∼ or α−1 Z. ∼ does not imply SME On the other hand the tensor algebras of C∗-correspondences may be more eligible. They were used in [39] to show that s ∼ and they provide a complete invariant for the conjugacy problem for dynamical systems in various cases, as shown by Davidson and Katsoulis [11, 12], and recently by Davidson and Kakariadis [10]. Moreover, for aperiodic C∗- SME ∼ is equivalent to strong correspondences, Muhly and Solel [42] prove that Morita equivalence of the tensor algebras in the sense of [7]. Along with the further investigation of tensor algebras, it is natural to suggest to work on the development of other invariants such as periodicity, existence of cycles, saturated and/or hereditary submodules etc. 4. Hilbert Bimodules Recall that if X is a Hilbert bimodule then IX = span{[ξ, η] : ξ, η ∈ X}. OPERATOR ALGEBRAS AND C-CORRESPONDENCES: A SURVEY 25 If a Hilbert A-bimodule X is considered as a C∗-correspondence over A, then JX = IX . Hence, for ξ, η ∈ X, the element [ξ, η] ∈ A is identified with the unique element a ∈ JX such that ϕX (a) = ΘX ξ,η. The converse is also true. The following result is well-known. Proposition 4.1. Let X be a C∗-correspondence over A. Then the following are equivalent (1) X is a bimodule, (2) K(X) ⊆ ϕX (A) and ϕ−1 (3) the restriction of ϕX to JX is a ∗-isomorphism onto K(X). X (K(X)) = ker ϕX ⊕ JX , In particular, if ϕX is injective then X is a Hilbert bimodule if and only if K(X) ⊆ ϕX(A). It turns out that the property of a C∗-correspondence being a Hilbert bimodule has an important non-selfadjoint operator algebra manifestation. We remind that an operator algebra A is called Dirichlet if A + A∗ is dense (via a completely isometric homomorphism) in C∗ env(A). Theorem 4.2. [28] Let X be a C∗-correspondence over A. Then the fol- lowing are equivalent: (1) X is a Hilbert bimodule, (2) ψt(K(X)) ⊆ π(A), for any injective covariant representation (π, t) that admits a gauge action, (3) the tensor algebra T + X has the Dirichlet property. The above result allows us to correct a misconception regarding semi- crossed products and Dirichlet algebras. Corollary 4.3. Let (A, α) be a dynamical system. Then the semicrossed product A ×α Z+ has the Dirichlet property if and only if α is surjective and ker α is orthocomplemented in A. In particular, when α is injective we deduce that the semi-crossed product has the Dirichlet property if and only if α is onto (thus a ∗-isomorphism). Thus [17, Proposition 3] is false. Nevertheless the main results of [17] are correct since they do not require that Proposition. It remains of interest though to determine whether a semi-crossed product has the unique ex- tension property, since this would allow us to extend Duncan's results to non-commutative dynamics. Recall that an operator algebra A is said to have the the unique extension env(A) to A is property if the restriction of every faithful representation of C∗ maximal, i.e. it has no non-trivial dilations. Theorem 4.4. [28] Let (A, α) be a unital injective dynamical system of a C∗-algebra. Then the semicrossed product A ×α Z+ has the unique extension property. 26 E.T.A. KAKARIADIS AND E.G. KATSOULIS The unique extension property of an operator algebra (or in general an operator space) implies the existence of the Choquet boundary in the sense of Arveson [3], i.e. the existence of sufficiently many irreducible represen- tations such that their restriction is maximal. Indeed, let P S(C∗ env(A)) be the set of the pure states of C∗ env(A))πτ be the free atomic representation of C∗ env(A), hence by the unique extension property its restriction to A is maximal. More- over, every πτ is maximal as a direct summand of a maximal representation. Hence, env(A), and let Π = ⊕τ ∈P S(C∗ env(A). Then Π is faithful on C∗ k[xij]k = k[Π(xij)]k = sup{k[πτ (xij)]k : τ ∈ P S(C∗ env(A))}, for all [xij] ∈ Mν (A) and ν ∈ N. The existence of the Choquet boundary for separable operator systems (or operator spaces) was proved by Arveson in [3] and it is still an open problem for the non-separable cases. Recently it was proved by Kleski in [35] that the supremum above can be replaced by a maximum, at least for the separable cases, where Arveson's Theorem applies. Note that the semicrossed products can give examples of non-separable operator algebras that have a Choquet boundary. Finally, we have a result that relates our adding of a tail process to the concept of a Hilbert bimodule. Theorem 4.5. [28] Let X be a non-injective C∗-correspondence. Then the graph C∗-correspondence Xτ , as defined in section 2, is an (essential) Hilbert bimodule if and only if X is a Hilbert bimodule and s−1(p) = r−1(p) = 1 for every p 6= p0. References [1] B. Abadie, S. Eilers and R. Exel, Morita equivalence for crossed products by Hilbert C∗- bimodules, Trans. Amer. Math. Soc. 350 (1998) 3043 -- 3054. [2] W. Arveson, Notes on the unique extension property, 2006, http://math.berkeley.edu/∼arveson/Dvi/unExt.pdf. [3] W. Arveson, The noncommutative Choquet boundary, J. Amer. Math. Soc. 21(4) (2008), 1065 -- 1084. [4] T. Bates, Applications of the Gauge-Invariant Uniqueness Theorem, Bulletin of the Australian Mathematical Society 65 (2002), 57 -- 67. [5] T. Bates, J. Hong, I. Raeburn, W. Szymanski, The ideal structure of the C∗-algebras of infinite graphs, Illinois J. Math. 46 (2002), 1159 -- 1176. [6] D. P. Blecher and C. Le Merdy, Operator algebras and their modules -- an operator space approach, volume 30 of London Mathematical Society Monographs, New Series, The Clarendon Press, Oxford University Press, Oxford, 2004. [7] D. Blecher, P. Muhly, V. Paulsen, Categories of operator modules -- Morita equivalence and projective modules, Mem. Amer. Math. Soc. 143 (2000), no 681. [8] J. Cuntz, K-theory for certain C∗-algebras II, J. Operator Theory 5 (1981), 101 -- 108. [9] J. Cuntz and W. Krieger, A class of C∗-algebras and topological Markov chains, Inventiones Math. 56 (1980), 251 -- 268. [10] K. R. Davidson, E. T.A. Kakariadis, Conjugate Dynamical Systems on C∗-algebras, Int. Mat. Res. Not., to appear. OPERATOR ALGEBRAS AND C*-CORRESPONDENCES: A SURVEY 27 [11] K. R. Davidson and E. G. Katsoulis, Isomorphisms between topological conjugacy algebras, J. Reine Angew. Math. (Crelle), 621 (2008), 29 -- 51. [12] K. R. Davidson and E. G. Katsoulis, Semicrossed Products of Simple C∗-algebras, Math. Ann. 342 (2008), 515 -- 525. [13] K. R. Davidson, E. G. Katsoulis, Operator algebras for multivariable dynamics, Mem. Amer. Math. Soc. 209, (2011), no 983. [14] K. R. Davidson, J. Roydor, C∗-envelopes of tensor algebras for multivariable dynam- ics, Proc. Edinb. Math. Soc. (2) 53 (2010), 333 -- 351; corrigendum Proc. Edinb. Math. Soc. (2) 54 (2011) 643 -- 644. [15] D. Drinen and N. Sieben, C∗-equivalences of graphs, J. Operator Theory 45 (2001), 209 -- 229. [16] M. A. Dritschel, S. A. McCullough, Boundary representations for families of represen- tations of operator algebras and spaces, J. Operator Theory, 53(1) (2005), 159 -- 167. [17] B. L. Duncan C∗-envelopes of universal free products and semicrossed products for multivariable dynamics, Indiana Univ. Math. J. 57(4) (2008), 1781 -- 1788. [18] N. Fowler, P. Muhly, I. Raeburn, Representations of Cuntz-Pimsner algebras Indiana Univ. Math. J. 52 (2003), 569 -- 605. [19] N. Fowler, I. Raeburn, The Toeplitz algebra of a Hilbert bimodule, Indiana Univ. Math. J. 48 (1999), 155 -- 181. [20] M. Hamana, Injective envelopes of operator systems, Publ. RIMS Kyoto Univ. 15(1979), 773 -- 785. [21] H. Hoare, W. Parry, Affine transformations with quasi-discrete spectrum. I, J. London Math. Soc. 41 (1966), 88 -- 96. [22] E. T.A. Kakariadis, Semicrossed products and reflexivity, J. Operator Theory 67(2) (2012), 379 -- 395. [23] E. T.A. Kakariadis, Semicrossed products of C∗-algebras and their C∗-envelopes, preprint (arXiv:1102.2252). [24] E. T.A. Kakariadis, The Silov Boundary for Operator Spaces, preprint (arXiv:1204.4495). [25] E. T.A. Kakariadis, E. G. Katsoulis Semicrossed products of operator algebras and their C∗-envelopes, J. Funct. Anal., 262(7) (2012), 3108 -- 3124. [26] E. T.A. Kakariadis, E. G. Katsoulis, Contributions to the theory of C∗- correspondences with applications to multivariable dynamics, Trans. Amer. Math. Soc. 364 (2012), 6605 -- 6630. [27] E. T.A. Kakariadis, E. G. Katsoulis, C∗-algebras and Equivalences for C∗- correspondences, preprint. [28] E. T.A. Kakariadis, E. G. Katsoulis, The Dirichlet Property for tensor algebras, preprint. [29] E. Katsoulis and D. Kribs, Isomorphisms of algebras associated with directed graphs, Math. Ann. 330, (2004), 709 -- 728. [30] E. G. Katsoulis, D. Kribs, Tensor algebras of C∗-correspondences and their C∗- envelopes, J. Funct. Anal. 234(1) (2006), 226 -- 233. [31] T. Katsura, On C ∗-algebras associated with C∗-correspondences, Contemp. Math. 335 (2003), 173 -- 182. [32] T. Katsura, On C ∗-algebras associated with C∗-correspondences, J. Funct. Anal. 217(2) (2004), 366 -- 401. [33] K. H. Kim, F. W. Roush, Decidability of shift equivalence, Dynamical systems (College Park, MD, 198687), 374-424, Lecture Notes in Math. 1342, Springer, Berlin, 1988. [34] K. H. Kim, F. W. Roush The Williams conjecture is false for irreducible subshifts, Ann. of Math 149(2) (1999), 545 -- 558. [35] C. Kleski, Boundary representations and pure completely positive maps, J. Operator Theory, to appear. 28 E.T.A. KAKARIADIS AND E.G. KATSOULIS [36] W. Krieger, On dimension functions and topological Markov chains, Invent. Math. 56 (1980), 239-250. [37] C. Lance, Hilbert C∗-modules. A toolkit for operator algebraists, London Mathemati- cal Society Lecture Note Series, 210 Cambridge University Press, Cambridge, 1995. x+130 pp. ISBN: 0-521-47910-X. [38] K. Matsumoto, Actions of symbolic dynamical systems on C∗-algebras, J. Reine Angew. Math. 605 (2007), 23-49. [39] P. S. Muhly, D. Pask, M. Tomforde, Strong Shift Equivalence of C∗-correspondences, Israel J. Math. 167 (2008), 315 -- 345. [40] P.S. Muhly, B. Solel, Tensor algebras over C∗-correspondences: representations, di- lations and C∗-envelopes J. Funct. Anal. 158 (1998), 389 -- 457. [41] P.S. Muhly, B. Solel, On the simplicity of some Cuntz-Pimsner algebras, Math. Scand. 83 (1998), 53 -- 73. [42] P.S. Muhly, B. Solel, On the Morita Equivalence of Tensor Algebras, Proc. London Math. Soc. 3 (2000), 113 -- 168. [43] P. S. Muhly, M. Tomforde, Adding tails to C∗-correspondences, Doc. Math. 9 (2004), 79 -- 106. [44] W. Paschke, The crossed product of a C∗-algebra by an endomorphism Proc. Amer. Math. Soc. 80 (1980), 113 -- 118. [45] J. Peters, Semicrossed products of C∗-algebras, J. Funct. Anal. 59 (1984), 498 -- 534. [46] M. Pimsner, A class of C ∗-algebras generalizing both Cuntz-Krieger algebras and crossed products by Z, Free probability theory (Waterloo, ON, 1995), 189 -- 212, Fields Inst. Commun., 12, Amer. Math. Soc., Providence, RI, 1997. [47] G. Popescu, Non-commutative disc algebras and their representations Proc. Amer. Math. Soc. 124, (1996), 2137 -- 2148. [48] I. Raeburn, Graph algebras, CBMS Regional Conference Series in Mathematics, 103, 2005. [49] I. Raeburn, W. Szymanski, Cuntz-Krieger algebras of infinite graphs and matrices, Trans. Amer. Math. Soc. 356 (2004) 39 -- 59. [50] P. Stacey, Crossed products of C∗-algebras by endomorphisms, J. Austr. Math. Soc., Ser. A 56 (1993), 204 -- 212. [51] J. B. Wagoner, Strong shift equivalence theory and the shift equivalence problem, Bull. Amer. Math. Soc. (N.S.) 36 (1999), 271 -- 296. [52] R. Williams, Classification of subshifts of finite type, Annals of Math. 98 (1973), 120 -- 153; erratum, Annals of Math. 99 (1974), 380 -- 381. Pure Mathematics Department, University of Waterloo, Ontario N2L-3G1, Canada E-mail address: [email protected] Department of Mathematics, University of Athens, 15784 Athens, GREECE Alternate address: Department of Mathematics, East Carolina University, Greenville, NC 27858, USA E-mail address: [email protected]
1209.4260	4	1209	2013-02-19T18:05:13	Limit theorems for monotonic convolution and the Chernoff product formula	[ "math.OA", "math.PR" ]	Bercovici and Pata showed that the correspondence between classically, freely, and Boolean infinitely divisible distributions holds on the level of limit theorems. We extend this correspondence also to distributions infinitely divisible with respect to the additive monotone convolution. Because of non-commutativity of this convolution, we use a new technique based on the Chernoff product formula. In fact, the correspondence between the Boolean and monotone limit theorems extends from probability measures to positive measures of total weight at most one. Finally, we study this correspondence for multiplicative monotone convolution, where the Bercovici-Pata bijection no longer holds.	math.OA	math	LIMIT THEOREMS FOR MONOTONIC CONVOLUTION AND THE CHERNOFF PRODUCT FORMULA MICHAEL ANSHELEVICH AND JOHN D. WILLIAMS ABSTRACT. Bercovici and Pata showed that the correspondence between classically, freely, and Boolean infinitely divisible distributions holds on the level of limit theorems. We extend this corre- spondence also to distributions infinitely divisible with respect to the additive monotone convolution. Because of non-commutativity of this convolution, we use a new technique based on the Chernoff product formula. In fact, the correspondence between the Boolean and monotone limit theorems ex- tends from probability measures to positive measures of total weight at most one. Finally, we study this correspondence for multiplicative monotone convolution, where the Bercovici-Pata bijection no longer holds. 3 1 0 2 b e F 9 1 ] . A O h t a m [ 4 v 0 6 2 4 . 9 0 2 1 : v i X r a 1. INTRODUCTION This article studies limit theorems for measures, but first we state a corollary which can be expressed purely in terms of analytic functions. Let A =(cid:26)F : C+ → C+ analytic, lim y↑∞ F (iy)/(iy) = 1(cid:27) . Note that A is closed under composition. We say that F is infinitely divisible if F ∈ A and for any n ∈ N, there exists gn ∈ A such that F = gn ◦ gn ◦ . . . ◦ gn . } Theorem 1.1. [Bel05, Proposition 3.8] F ∈ A is infinitely divisible if and only if there exist {Ft : t ≥ 0} ⊂ A which form a semigroup under composition, Ft ◦ Fs = Ft+s, F1 = F, and is continuous in the topology of uniform convergence on compact sets. In this case we write F ◦t = Ft; each F ◦t is uniquely defined. {z n ∂t = Φ(Ft). In terms of analytic functions, our main result is the following Moreover, according to Proposition 2.5 below, there exists a function Φ with Φ(z) + z ∈ A such that ∂Ft Corollary 1.2. Fix {gn : n ∈ N} , F ∈ A, F infinitely divisible, and a sequence of positive integers k1 < k2 < . . .. Then Date: October 16, 2018. 2010 Mathematics Subject Classification. Primary 46L53; Secondary 30D05, 46L54, 47D06, 60B10. The first author was supported in part by NSF grants DMS-0900935 and DMS-1160849. gn ◦ gn ◦ . . . ◦ gn → F } kn {z 1 2 MICHAEL ANSHELEVICH AND JOHN D. WILLIAMS uniformly on compact sets if and only if for Φ as above. kn (gn(z) − z) → Φ, The proof of this corollary requires us to work with limit theorems for measures rather than analytic functions, which we now explain. A fundamental result in free probability, due to Bercovici and Pata [BP99], is that limit theorems for sums of freely independent random variables are in a precise correspondence with limit theorems for independent random variables. More specifically, denoting by ⊞ the (additive) free convolution and by ∗ the usual convolution, a kn-fold convolution µn ⊞ µn ⊞ . . . ⊞ µn converges to a limit if and only if a kn-fold convolution µn ∗ µn ∗ . . . ∗ µn converges to a limit. The correspondence between the limit measures is known as the Bercovici-Pata bijection, which has a surprisingly concrete form based upon the L´evy-Hincin representations of the various infinitely divisible measures. In addition to this, the same authors proved that the same result also holds for the (additive) Boolean convolution ⊎. According to [Spe97, BGS02, Mur02, Mur03] in addition to the usual, free, and Boolean inde- pendence, the only other notion of non-commutative independence with a universal property is the monotonic independence of Muraki [Mur01]. He defined monotone convolution ⊲ for com- pactly supported measures, and this operation was extended to general probability measures on R in [Fra09]. In this article, we are interested in limit theorems with respect to the monotone convo- lution. Standard proofs of limit theorems for independent random variables use the method of characteristic functions, based on the observation that (the logarithm of) the Fourier transform is a linearizing transform for the convolution: Free and Boolean convolutions also have linearizing transforms: log Fµ∗ν(θ) = log Fµ(θ) + log Fν(θ). ϕµ⊞ν(z) = ϕµ(z) + ϕν(z); Eµ⊎ν(z) = Eµ(z) + Eν(z) (the notation will be defined in the following section), and their properties are used in the proof of the Bercovici and Pata results. However, the monotone convolution is not commutative, and as such cannot have a linearizing transform. As a result, a very different approach is necessary to incorporate monotone convolution into this bijection. In addition to the proof using characteristic functions, Feller (Section IX.7 of [Fel71]) gives an alternative proof of classical limit theorems using semigroups of operators and their generators. It is this approach, based on the Chernoff product formula, that works well for monotonic independence. Note that classical probability deals with commuting random variables, which demands a much simpler variant of these Chernoff style arguments than our non-commutative setting. The idea of using the full power of the Chernoff product formula in a probabilistic setting goes back to Goldstein [Gol76a, Gol76b] (see also [Pfe83]). Besides the central limit theorem and the Poisson limit theorem [Mur01, HS11], the only other limit theorems in the monotone case of which we are aware are Wang's results on the central limit theo- rem in the general (non-compactly supported) case [Wan11] and on the strict domains of attraction of strictly stable distributions [Wan12]. As far as we know, our result is new even in the general LIMIT THEOREMS FOR MONOTONIC CONVOLUTION 3 Poisson case. In addition to these existing works, Uwe Franz and Takahiro Hasebe have informed us that they are currently developing related results. Our main result is the addition of part (d) in the following theorem. Various measures νγ,σ are defined in the next section. Theorem 1.3. Fix a finite positive Borel measure σ on R, a real number γ, a sequence of probability measures {µn}n∈N, and a sequence of positive integers k1 < k2 < · · · The following assertions are equivalent: converges weakly to νγ,σ ; ∗ converges weakly to νγ,σ ⊞ ; converges weakly to νγ,σ ⊎ ; converges weakly to νγ,σ ⊲ ; kn (a) The sequence µn ∗ µn ∗ · · · ∗ µn } (b) The sequence µn ⊞ µn ⊞ · · · ⊞ µn } (c) The sequence µn ⊎ µn ⊎ · · · ⊎ µn } (d) The sequence µn ⊲ µn ⊲ · · · ⊲ µn } (e) The measures {z {z {z {z kn kn kn x2 kn x2 + 1 dµn(x) → σ weakly, and lim n↑∞ knZR x x2 + 1 dµn(x) = γ. The equivalence of items (a-c) and (e) is Theorem 6.3 in [BP99]. While the usual convolution is linear in each of its arguments, and so is defined for general positive or even signed measures, free convolution is naturally defined only for probability measures. We note that Boolean and monotone convolutions, defined in the complex-analytic setting rather than through the appropriate notion of independence, also can be treated as binary operations on positive measures. Our proof of the equivalence between parts (c) and (d) of Theorem 1.3 works, with very minor modifications, for the setting of positive measures of total weight at most one; we do not try to find the appropriate generalization of the statement in part (e) to this setting. In contrast with the additive case, multiplicative convolutions arise by taking products of indepen- dent random variables as opposed to sums. These forms of monotone convolution were defined and studied in [Ber05, Fra06]. A version of the Boolean-free Bercovici-Pata bijection for the multiplica- tive case was proven by Wang in [Wan08]. In Section 4, we investigate the analog of Theorem 1.3 for multiplicative convolution. We show that the direct analog of the theorem does not hold in gen- eral, but holds under additional conditions. Note that we only consider multiplicative convolution for measures on the unit circle, and not on the positive real line as for example in [BP00]. The paper is organized as follows. Section 1 consists of this introduction. Section 2 consists of preliminaries for additive non-commutative probability as well as the semigroup theory applicable to our proof of Theorem 1.3. In Section 3, we prove our main result -- the equivalence between 4 MICHAEL ANSHELEVICH AND JOHN D. WILLIAMS the Boolean and monotone limit theorems -- first for measures with finite variance (by functional- analytic methods) and then for general positive measures of total weight at most one (by complex- analytic methods); Theorem 1.3 follows as a corollary. Section 4 consists of the preliminaries for multiplicative convolutions as well as multiplicative analogues of our main results. Acknowledgements: The authors would like to thank J.C. Wang for reading this paper thoroughly and providing excellent advice during the revision process, and Serban Belinschi for some remarks. 2.1. Transforms and distributions. In what follows, we shall denote by 2. PRELIMINARIES C+ = {z ∈ C : ℑ(z) > 0}, C− = {z ∈ C : ℑ(z) < 0}, C+ r = {z ∈ C : ℑ(z) > r} for r ∈ R+. For α, β > 0 we shall refer to the truncated cone Γα,β = {z ∈ C+ as the Stolz angle associated to these real numbers. Let µ and ν denote finite (positive) non-zero Borel measures on the real line. The Cauchy transform associated to such a measure is the function β : ℑ(z) > αℜ(z)} Gµ(z) :=ZR 1 z − x dµ(x) : C+ → C− ∪ R We define the F-transform associated to this measure by letting Fµ(z) := 1/Gµ(z). Finally, there exist α, β > 0 such that Fµ is injective when restricted to Γα,β and we define the Voiculescu trans- form by setting ϕµ(z) := F −1 µ (z) − z : Γα′,β ′ → C− ∪ R where this function takes on real values if and only if the associated measure is a Dirac mass. The Voiculescu transform may be viewed as an analogue of the logarithm of the Fourier transform for free probability insofar as ϕµ⊞ν = ϕµ + ϕν. We define the E-transform of a finite, non-zero Borel measure µ as Eµ(z) := 1 µ(R) z − Fµ(z). For finite non-zero Borel measures µ and ν, we may define their Boolean convolution µ ⊎ ν by requiring that Eµ⊎ν = Eµ + Eν, (µ ⊎ ν)(R) = µ(R)ν(R); the existence of a positive measure µ ⊎ ν follows from Nevanlinna theory. Observe that, in contrast to the free case, the E-transform is well defined on all of C+. This fact may be used to prove that all Borel probability measures are infinitely divisible (in the sense defined below) with respect to Boolean convolution. Let {µi}i∈I denote a family of finite positive measures. We say that this family it tight if for every ǫ > 0 there exists an N ∈ N so that µi([−N, N]c) < ǫ for all i ∈ I. It is a basic result in measure theory that a family of measures is uniformly bounded and tight if and only if this family is sequentially precompact in the weak topology. All the measures below will have total weight at most one, so we will omit the uniform boundedness condition. convergence of the Cauchy transforms, or of the F transforms, on compact subsets of C+, and µn → µ vaguely ifR f dµn → R f dµ for every f ∈ C0(R). This is equivalent to the uniform implies that µ(R) ≤ lim inf n→∞ µn(R). µn → µ weakly ifR f dµn →R f dµ for every bounded LIMIT THEOREMS FOR MONOTONIC CONVOLUTION 5 continuous f, and is equivalent to vague convergence plus µn(R) → µ(R), or to this last condition plus the uniform convergence on compact sets of the E-transforms. Thus when µn and µ are all probability measures, the two modes of convergence are equivalent. We say that a measure µ is infinitely divisible with respect to the convolution operation ⊞ if, for all n ∈ N there exists a Borel probability measure µn such that µ = µn ⊞ µn ⊞ · · · ⊞ µn where the convolution is n-fold. Analogous definitions serve for all of the convolution operations discussed in this paper. It is known that a ⊞-infinitely divisible measure µ can be included as µ1 in a ⊞- convolution semigroup {µt : t ≥ 0} , µt ⊞ µs = µt+s, and this property also holds for the other convolution operations discussed in this paper. In what follows, we shall make liberal use of the following basic function theoretic facts. We refer to [BV93] for an excellent overview of the relevant machinery. Lemma 2.1. limy↑∞ Fµ(iy)/(iy) = 1 µ(R). (a) ℑ(Fµ(z)) ≥ 1 (b) An analytic function F : C+ → C+ ∪ R is the F -transform of a Borel measure µ, with µ(R)ℑ(z) with equality at any point z if and only if µ is a Dirac mass. (c) (cid:12)(cid:12)(cid:12)Fµi(z) − 1 µi(R)z(cid:12)(cid:12)(cid:12) = o(z) uniformly for z ∈ Γα,β and {µi}i∈I a uniformly bounded, tight (d) There exists a finite measure σ and a real number γ such that family of measures. Fµ(z) = −γ + 1 µ(R) z +ZR 1 + xz x − z dσ(x) The last of these refers to the Nevanlinna representation of certain complex analytic functions. The following lemma is a slight reformulation of the results from [Maa92]. Lemma 2.2. (a) Let ρ be a finite measure on R. The Cauchy transform Gρ(z) is a bounded function on C+ 1 . (b) If ρ has a finite first moment, then zGρ(z) and zG′ (c) Let µ be a finite measure with finite variance, which for a non-probability measure means 1 . Moreover, if ρ(z) are bounded functions on C+ 1 . µ(R) z − Fµ(z) is a bounded function on C+ µ(x2)µ(R) − µ(x)2 < ∞. Then 1 sup {Var[µn] : n ∈ N} < ∞, the bound is uniform in n. Proof. For finite ρ, z − x For ρ with a finite first moment and z ∈ C+ 1 , z 1 (cid:12)(cid:12)(cid:12)(cid:12)ZR (z − x)2 dρ(x)(cid:12)(cid:12)(cid:12)(cid:12) ≤(cid:12)(cid:12)(cid:12)(cid:12)ZR z (cid:12)(cid:12)(cid:12)(cid:12)ZR Finally, for µ a probability measure with finite variance, by Proposition 2.2 of [Maa92], ρ(R). 1 ℑ(z) dρ(x)(cid:12)(cid:12)(cid:12)(cid:12) ≤ dρ(x)(cid:12)(cid:12)(cid:12)(cid:12) ≤ ρ(R) + z − x 1 µ(R) z − Fµ(z) = µ(x) µ(R)2 + Gσ(z), 1 ℑ(z)ZR x dρ(x). 6 MICHAEL ANSHELEVICH AND JOHN D. WILLIAMS where σ is a finite measure. In fact, σ(R) = µ(x2)µ(R)−µ(x)2 Remark 2.3. The classical L´evy-Hincin formula provided an equivalent definition of infinite divis- ibility based on the class of characteristic functions associated to these measures. Related formulae were developed for free and Boolean independence. In defining these formulae, let γ ∈ R and σ ⊎ = ν1,γ,σ denote a finite Borel measure on R and define the measures νγ,σ ) in terms of the relevant transforms by letting , which implies the last result. ⊞ ; νm,γ,σ (resp. νγ,σ ; νγ,σ µ(R)2 (cid:3) ⊎ ⊎ ∗ (Fνγ,σ ∗ )(t) = exp(cid:18)iγt +ZR (z) = γ +ZR ϕνγ,σ ⊞ −Eνm,γ,σ ⊎ (z) = Fνm,γ,σ ⊎ (z) − (eitx − 1 − itx) dσ(x), 1 + xz z − x 1 m z = −γ +ZR x2 + 1 x2 dσ(x)(cid:19) , z ∈ C+, dσ(x), 1 + xz x − z t ∈ R, z ∈ C+. A Borel probability measure µ is infinitely divisible with respect to classical (resp. free ; Boolean) ⊞ ; µ = νγ,σ convolution if and only if there exists a γ and σ as above so that µ = νγ,σ ⊎ ) We shall define the class of monotone infinitely divisible measures, which we shall denote by νγ,σ ⊲ , below. Remark 2.4. The reader should note that the classes of infinitely divisible probability measures are all indexed by a real number γ and a finite measure σ. That this bijection is more than formal is the main content of Theorem 1.3. (resp. µ = νγ,σ ∗ 2.2. Monotonic independence and monotone convolution. The notion of monotonic indepen- dence is originally due to Muraki (see [Mur01] and references therein). In [Mur00], he defined the corresponding convolution operation ⊲ on compactly supported probability measures. This defini- tion was extended to general probability measures by Franz [Fra09]. Up to a change in notation, their definition amounts to requiring that Fµ⊲ν(z) = Fµ(Fν(z)). It follows immediately from Nevanlinna theory that the same definition works for general non-zero positive measures µ and ν, and (µ ⊲ ν)(R) = µ(R)ν(R). 2.3. Monotone convolution semigroups. See Theorem 4.5 of [Bia98], based on Section 5.2 of [MZ74], for the proof of the following result, and [BP78, Sis98, Has10] for related results. Proposition 2.5. Let {νt : t ≥ 0} form a monotone convolution semigroup (so that νt ⊲ νs = νt+s), which in particular is strongly continuous. Then, denoting Ft = Fνt, the family {Ft : t ≥ 0} form a semigroup of analytic transformations of C+, which extends to a local group of ana- lytic transformations of some Γα,β for some ε and −ε ≤ t ≤ ε. If m = ν1(R) ≤ 1, then ℑ(Ft(z)) ≥ 1 mtℑ(z) ≥ ℑ(z). Therefore there exists an analytic function Φ : C+ → C+ ∪ R such that (1) ∂Ft ∂t By Nevanlinna theory Φ has a representation = Φ(Ft). Φ(z) = −γ − log(m)z +ZR 1 + xz x − z dσ(x) LIMIT THEOREMS FOR MONOTONIC CONVOLUTION 7 for a finite measure σ and real numbers 0 < m ≤ 1 and γ. Conversely, given such Φ : C+ → C+, the equation (1) has a unique solution with initial condition F0(z) = z corresponding to a strongly continuous monotone convolution semigroup. Definition 2.6. For a finite measure σ and real numbers 0 < m ≤ 1 and γ, denote z ∈ C+. Φm,γ,σ(z) := −γ − log(m)z +ZR It should be noted that Φm,γ,σ = −Eνγ,σ semigroup it generates in the sense of the preceding proposition. Denote ⊎ − log(m)z. Let {νt : t ≥ 0} be the monotone convolution 1 + xz x − z dσ(x), νm,γ,σ ⊲ = ν1. Lemma 2.7. For i = 1, 2, let {Fi(z, t) : z ∈ C+, t ≥ 0} be two semigroups of analytic transforma- tions of C+, such that ∂Fi(z,t) ∂t = Φi(Fi(z, t)), Fi(z, 0) = z. Given ε > 0, suppose that for some compact K and C, we have Fi(K, t) ⊂ C for t ∈ [0, 1], and for z ∈ C, Φ1(z) − Φ2(z) < ε. Then for any z ∈ K, F1(z, 1) − F2(z, 1) ≤ cε, where c is a constant depending on Φ2 but not on Φ1. Proof. Note first that so in particular this second derivative exists. Then denoting ∂2Fi(z, t) ∂t2 = Φ′ i(Fi(z, t))Φi(Fi(z, t)), by Taylor's formula M2 = max i=1,2 sup z∈K ∂2Fi(z, t) ∂t2 sup t∈[0,1](cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12) , Fi(z, t) − Fi(z, t0) − (t − t0)Φi(Fi(z, t0)) ≤ M2 2 (t − t0)2. Thus ≤ ≤ Therefore for large N. So denoting M1 = supz∈C Φ′ F1(z, t0 + 1/N) − F2(t0 + 1/N) M2 2N 2 . Φi(Fi(z, t0))(cid:12)(cid:12)(cid:12)(cid:12) < 2(z)) F1(z, t0) − F2(z, t0)(cid:19) + M2 N 2 1 N (cid:12)(cid:12)(cid:12)(cid:12)Fi(z, t0 + 1/N) − Fi(z, t0) − N (cid:18)sup z∈C Φ1(z) − Φ2(z) + (N + sup M2 2(z), it follows that 1 z∈K Φ′ ε N + M1 N 2 +(cid:18)1 + F1(z, 1) − F2(1) ≤(cid:18) ε N (cid:19)F1(z, t0) − F2(z, t0) . N 2(cid:19) N −1Xk=0(cid:18)1 + M2 N + ≈ (eM1 − 1)(cid:18) ε M1 + M1 N (cid:19)k M1N(cid:19) ≈ M2 eM1 − 1 M1 ε (cid:3) 8 MICHAEL ANSHELEVICH AND JOHN D. WILLIAMS Remark 2.8. We note here that weak convergence of a uniformly bounded family of measures is equivalent to the convergence of their F -transforms uniformly on compact sets (see [BV93]). We will use this fact without reference throughout the paper. With this in mind, consider the monotone infinitely divisible measures {µn}n∈N and µ with associated semigroup generators {Φn}n∈N and Φ, respectively. If we assume that Φn → Φ uniformly on compact sets then, according to the previous lemma, µn → µ vaguely. This fact will play a key role in the proof of our main theorem. 2.4. Chernoff product formula. We will use the following version of the Chernoff Product For- mula. Proposition 2.9. Let (kn) be an increasing sequence of positive integers, and {Vn}n∈N a family of contractions on a Banach space X . Suppose B is an unbounded operator which generates a strongly continuous semigroup of contractions {T (t) : t ≥ 0}, D is a core for B, and for each x ∈ D, lim n↑∞ kn(Vn − I)x → Bx. Then for each x ∈ X , lim n↑∞ V kn n x = T (1)x. The proof is very similar to the continuous version, Theorem 5.2 of [EN00]. It is provided for completeness. tjkj Proof. Denote Bn = kn(Vn − I). Then Bnx → Bx for all x ∈ D. Moreover, ≤ 1, n kV j nk j! (cid:13)(cid:13)etBn(cid:13)(cid:13) ≤ e−tkn(cid:13)(cid:13)etknVn(cid:13)(cid:13) ≤ e−tkn (cid:13)(cid:13)eBnx − T (1)x(cid:13)(cid:13) → 0 so the semigroups(cid:8)etBn : t ≥ 0(cid:9) are contractive. Therefore by the first Trotter-Kato Approxima- tion Theorem (Theorem 4.8 in [EN00]), eBnt → T (t) strongly and, in particular, for each x ∈ X , as n ↑ ∞. On the other hand, by Lemma 5.1 from [EN00], 1 √kn kBnxk → 0. ∞Xj=0 (cid:13)(cid:13)eBnx − V kn n x(cid:13)(cid:13) =(cid:13)(cid:13)ekn(Vn−I)x − V kn n x(cid:13)(cid:13) ≤pkn kVnx − xk = The result follows. (cid:3) 3. LIMIT THEOREMS FOR MONOTONE CONVOLUTION ON R In this section we prove our main theorem. The Chernoff product formula is at the heart of the proof of the forward direction (a, b, c, e ⇒ d in the parlance of Theorem 1.3). The reader should note that, to satisfy the requisite hypotheses, we must make additional moment assumptions on the relevant random variables (these assumptions will later be discarded). Denote by M the sub-probability measures, that is, positive Borel measures with total weight at most one, by M2 ⊂ M the subset of measures with finite variance, and by M1 general finite positive Borel measures on R with finite first moment. LIMIT THEOREMS FOR MONOTONIC CONVOLUTION 9 Proposition 3.1. Let {µn}n∈N ⊂ M2 be a family of sub-probability measures on R. Suppose weakly, where {µ⊎kn n }∞ n=1 and νm,γ,σ ⊎ weakly. Proof. Let have uniformly bounded, finite variance. Then n → νm,γ,σ µ⊎kn ⊎ µ⊲kn n → νm,γ,σ ⊲ D = {Gρ : ρ ∈ M1} . By Nevanlinna theory, D is invariant under composition operators by functions Fµ for µ ∈ M2. Let A be the completion of D with respect to the uniform norm on C+ 1 , which we denote by k·k∞. Then each right composition operator by Fµ is a contraction on A (since Fµ(C+ We will utilize Chernoff's theorem (Proposition 2.9). Towards this end, we define operators on A by letting 1 ) ⊂ C+ 1 ). for h ∈ A. We further define a (possibly unbounded) operator Vn · h := h ◦ Fµn B · h := Φm,γ,σh′ using the notation from Definition 2.6. Note that the notation for the operators match those within the statement of Proposition 2.9. To invoke this result, we must show that D is a core for the operator B. This will follow once we show that D is in the domain of this operator. Towards this end, let h = Gρ ∈ D. We will show that Φm,γ,σh′ is a limit of elements in A, proving that Φm,γ,σh′ ∈ A. Indeed, kkn(h ◦ Fµn − h) − Φm,γ,σh′k∞ 1 Φm,γ,σ(z) 1 = sup z∈C+ = sup z∈C+ ≤ sup z∈C+ 1 ≤ sup z∈C+ + (kn(z − Fµn(z)) + Φm,γ,σ(z)) Gρ(z) + sup Fµn(z) − x − (Fµn(z) − x)(z − x) 1(cid:12)(cid:12)(cid:12)(cid:12)ZR(cid:18)kn(cid:18) z − x(cid:19) + 1(cid:12)(cid:12)(cid:12)(cid:12)ZR(cid:18)kn(z − Fµn(z)) + Φm,γ,σ(z) (z)(cid:17) Gρ(z)(cid:12)(cid:12)(cid:12) + sup 1(cid:12)(cid:12)(cid:12)(cid:16)F 1(cid:12)(cid:12)(Fµn(z) − z)Φm,γ,σ(z)G′ ρ(z)(cid:12)(cid:12) . (z) − Fνm,γ,σ (Fµn(z) − z)Φm,γ,σ(z) (z − x)2(cid:19) dρ(x)(cid:12)(cid:12)(cid:12)(cid:12) (Fµn(z) − x)(z − x)2(cid:19) dρ(x)(cid:12)(cid:12)(cid:12)(cid:12) ρ(z)(cid:12)(cid:12) 1(cid:12)(cid:12)(Fµn(z) − z)Φm,γ,σ(z)G′ 1(cid:12)(cid:12)(cid:12)(cid:12)(cid:18)kn(cid:18)1 − µn(R)(cid:19) − log(m)(cid:19) zGρ(z)(cid:12)(cid:12)(cid:12)(cid:12) z∈C+ z∈C+ ⊎kn n 1 µ ⊎ + sup z∈C+ The hypothesis implies that µn(R)kn → νm,γ,σ implies that µ⊎kn ⊎ n → νm,γ,σ ⊎ µ ⊎kn n (z) − Eνm,γ,σ (R) = m, and in particular µn(R) → 1. It also and µn → δ0 weakly, and so on any compact set (cid:16)E (z)(cid:17) , µn(R)(cid:19) = log µn(R)kn → log m. µn(R)(cid:19) ≈ −kn log(cid:18) 1 (Fµn(z) − z) 1 ⊎ kn(cid:18)1 − converge to zero uniformly, and in addition 10 MICHAEL ANSHELEVICH AND JOHN D. WILLIAMS On the other hand, the same convergence results imply that the variances of(cid:8)µ⊎kn n (cid:9)∞ are all uniformly bounded, and so by Lemma 2.2, the functions n=1 and {µn}∞ n=1 (cid:16)E ⊎kn n µ (z) − Eνm,γ,σ ⊎ (Fµn(z) − z)Φγ,σ(z) (z)(cid:17) , are bounded uniformly in n. Finally, by the same lemma zGρ(z) is bounded and Gρ(z) = o(1), G′ ρ(z) = o(1). Combining these results, we conclude that kkn(h ◦ Fµn − h) − Φm,γ,σh′k∞ → 0. We have just shown that kn(Vn − I)h → Bh for h ∈ D, so that D is indeed in the domain of B. According to [BP78, Sis98], the operator B is precisely the generator of the semigroup of composition operators corresponding to the semigroup of functions generated by Φm,γ,σ. This semigroup is strongly continuous. Since these composition operators preserve D, we may conclude that D is a core for B. It now follows from Proposition 2.9 that V kn n → T (1) strongly, where T (1)h = h ◦ Fν1 = h ◦ Fνm,γ,σ ⊲ . In particular, for h(z) = 1 z which implies that weakly. n − Gνm,γ,σ ⊲ (cid:13)(cid:13)(cid:13)Gµ⊲kn (cid:13)(cid:13)(cid:13)∞ → 0, n → νm,γ,σ µ⊲kn ⊲ (cid:3) The converse to the Chernoff product formula is, in general, false [Che74, Che76]. We implicitly prove a variation of this converse that is quite specific to this setting (although it seems plausible that this proof may be adapted for a robust result in the setting of complex composition operators). The following facts are necessary at several distinct steps of the proof of our main theorem so they are isolated for easy reference. Lemma 3.2. Let {µn : n ∈ N} ⊂ M satisfy {z (a) The family of measures {µ⊲j n : n ∈ N, j = 1, . . . , kn} is tight. (b) The family {knℑ(Fµn(z) − z)}n∈N is pointwise bounded, and for every ǫ > 0 there exist z(cid:19) m−1 − 1 − log(m) ≤ 2ℑ(cid:18)F ◦kn knℑ(cid:18)Fµn(z) − z(cid:19) ≤ ǫz weakly. The following are true: µn ⊲ µn · · · ⊲ µn α, β > 0 such that → νm,γ,σ ⊲ µn (z) − µn(R)kn } kn 1 1 µn(R) for large n ∈ N and z ∈ Γα,β. LIMIT THEOREMS FOR MONOTONIC CONVOLUTION 11 Proof. We begin with property (a) listed above. The family(cid:8)µ⊲kn n (cid:9)n∈N converges to νm,γ,σ by assumption, and therefore is tight. It follows that for any ε, there is a Stolz angle Γα,β, such that µn (z) − 1 µn(R)kn z) < ε z for z ∈ Γα,β. Since Fµn increases the imaginary part, it is also true ℑ(F ◦kn that ℑ(F ◦j µn(R)j z) < ε z for any 1 ≤ j ≤ kn. µn(z) − 1 ⊲ In order to show that the family {µ⊲j n : n ∈ N, 1 ≤ j ≤ kn} is tight, we assume that this does not hold and obtain a contradiction. Suppose that there exists a δ > 0 such that for any K ∈ N there are j(K), n(K) with µ⊲j(K) n(K) ([−K, K]c) > δ. We define ρK = µ⊲j(K) n(K) [−K,K], λK = µ⊲j(K) n(K) − ρK Note that ρK(R) ≤ µn(K)(R)j(K) − δ. It then follows from the Nevanlinna representation of FρK that ℑ(FρK (z)) ≥ (µn(K)(R)j(K) − δ)−1ℑ(z), µn(K)(R)j(K) z(cid:19) ≥ 1 δ (µn(K)(R)j(K) − δ)µn(K)(R)j(K)ℑ(z). ℑ(cid:18)FρK (z) − Also, for any fixed z, GλK (z) → 0 as K → ∞. Since Fµ ⊲j(K) n(K) = FρK 1 + FρK GλK it follows that for sufficiently large K, ℑ(cid:18)Fµ ⊲j(K) n(K) (z) − 1 µn(K)(R)j(K) z(cid:19) ≥ δ µn(K)(R)2j(K)ℑ(z). Taking z purely imaginary and ε = δ µn(K)(R)2j(K) , we obtain a contradiction. In order to address property (b), we first claim that for ǫ > 0 there exists a Stolz angle Γα,β so that ℑ(cid:18)F ◦j µn(z) − 1 µn(R) F ◦(j−1) µn (z)(cid:19) ≥ (1 − ǫ)ℑ(cid:18)Fµn(z) − 1 µn(R) z(cid:19) for all n ∈ N, j = 1, . . . , kn and z ∈ Γα,β. 12 MICHAEL ANSHELEVICH AND JOHN D. WILLIAMS Indeed, consider the following chain of equalities and inequalities: 1 + tF j−1 t − F ◦(j−1) µn (z) (z) µn dσn(t)# − 1 µn(R) F ◦(j−1) µn (z)! 1 µn(R) F ◦(j−1) µn 1 µn(R) F ◦(j−1) µn (z))(1 + t2) (z)(cid:19) (z) +ZR dσn(t) t − F ◦(j−1) (z)2 µn (z))(1 + t2) µn µn(z) − ℑ(cid:18)F ◦j = ℑ "−γn + =ZR ℑ(F ◦(j−1) =ZR ≥ZR t∈R( t∈R( ℑ(F ◦(j−1) ≥ inf t − z2 = inf µn µn t − z2 t − F ◦(j−1) t − z2 t − F ◦(j−1) µn t − z2 ℑ(z)(1 + t2) t − z2 t − F ◦(j−1) µn (z)2 dσn(t) dσn(t) t − z2 t − F ◦(j−1) µn (z)2 ℑ(z)(1 + t2) (z)2)ZR (z)2)ℑ(cid:18)Fµn(z) − t − z2 dσn(t) 1 µn(R) z(cid:19) The inequalities arise because the above integrands are non-negative. The first inequality is a result of the fact that F -transforms increase the imaginary part. Our claim will follow if we can show that the infimum above is arbitrarily close to 1 for all z in a n }n∈N, j=1,...,kn forms a tight family sufficiently small Stolz angle. Indeed, we have shown that {µ⊲j of measures. By Lemma 2.1(c), this implies that F ◦j µn(R)j z = o(z) uniformly over j and µn(z) − 1 n for z in a sufficiently small Stolz angle. The claim follows from simple geometric considerations. Note that tightness also implies that the infimum is finite for every fixed z (that is, z need not lie in the Stolz angle). Next, observe that we may utilize this claim to attain a bound for knℑ(Fµn(z) − 1 Stolz angle. Indeed, if we recall that ℑ(F ◦j telescoping argument implies that knℑ(cid:18)Fµn(z) − µn(R) z) on this (z)) ≥ 0 for j = 1, . . . , kn, a mild z(cid:19) 1 z(cid:19) knXj=1 µn(R)kn − 1 µn(R) − 1 1 µn(z) − 1 µn(R)F ◦(j−1) µn(R)kn−j µn(R) µn 1 1 1 1 µn(R) z(cid:19) m−1 − 1 − log(m) ≈ ℑ(cid:18)Fµn(z) − = ℑ(cid:18)Fµn(z) − ℑ(cid:18)F ◦j knXj=1 ≤ (1 − ǫ)−1 = (1 − ǫ)−1ℑ(cid:18)F ◦kn µn(R) µn (z) − µn(z) − 1 µn(R) F ◦(j−1) µn 1 µn(R)kn−j (z)(cid:19) 1 µn(R)kn z(cid:19) LIMIT THEOREMS FOR MONOTONIC CONVOLUTION 13 for ε < 1/2 and sufficiently large n, and where m−1−1 − log(m) = 1 for m = 1. Note that the right hand side of this inequality is uniformly o(z) for z ∈ Γα,β. This is a result of the fact that the F - transforms of a uniformly bounded, tight family of measures have this property (Lemma 2.1) where the tightness is a consequence of the fact that the family of measures converges. This proves the second statement in part (b). Finally, if z is fixed, pointwise finiteness of the infimum and the same argument imply the first statement in part (b). (cid:3) Remark 3.3. Part (a) utilizes an approach found in [Wil12]. Part (b) in the previous lemma provides an estimate for the finite measures arising from the Nevanlinna representations of the associated F - transforms. Indeed, given that Fµn(z) = −γn + 1 µn(R) z +ZR 1 + tz t − z dσn(t) we have that knσn(t > y) ≤ 2knZR 1 + t2 t2 + y2 dσn(t) = 2kn y ℑ(cid:18)Fµn(iy) − 1 µn(R) iy(cid:19), and the previous lemma provides us with a bound for the right hand side of the inequality (in the case where the monotone infinitesimal array converges). This estimate will be used in our proof of the main theorem. Proposition 3.4. Let {µn}n∈N be a family of sub-probability measures on R. Then weakly if and only if weakly. n → νm,γ,σ µ⊎kn ⊎ µ⊲kn n → νm,γ,σ ⊲ Proof. As a first step, we extend the result in Proposition 3.1 to full generality. Assume that {µn}n∈N satisfies µn ⊎ · · · ⊎ µn → νm,γ,σ where we have no moment assumptions on any of the relevant probability measures. ⊎ Consider the functions −Eµn(z) = Fµn(z) − 1 µn(R) z = −γn +Z ∞ −∞ 1 + tz t − z dσn(t) where the function on the right hand side is the Nevanlinna representation and recall that our hy- pothesis is equivalent to −knEµn(z) → −γ +Z ∞ −∞ 1 + tz t − z dσ(t) = −Eνm,γ,σ ⊎ (z) where the convergence is uniform on compact subsets of C+. It follows from this fact that knγn → γ and knσn → σ where the latter is with respect to the weak topology. We define a new family of measures {µn,N : n, N ∈ N} implicitly through the equation z = −γn +Z N +ǫ(n,N ) −Eµn,N (z) = Fµn,N (z) − 1 + tz t − z −(N +δ(n,N )) dσn(t) µn(R) 1 14 MICHAEL ANSHELEVICH AND JOHN D. WILLIAMS where the (small) numbers ǫ(n, N), δ(n, N) are chosen so that knσn[−N −δ(n,N ),N +ǫ(n,N )] → σ[−N,N ] =eσN (mass may converge to N so this slight correction is required). It follows that, for each N, uniformly on compact subsets of C+. Since also the functions knEµn,N converge to Eν µn,N (R)kn = µn(R)kn → m, this is equivalent to µn,N ⊎ · · · ⊎ µn,N → νm,γ,eσN in the weak topol- ogy. The reader should note that these measures have support contained in [−N − δ(N), N + ǫ(N)] where δ(N) = supn∈N δ(n, N) ǫ(N) = supn∈N ǫ(n, N) which we may assume is as close to 0 as we would like. Thus, the measures are compactly supported in a uniform sense so that the hypotheses of Proposition 3.1 are satisfied (as this implies the requisite uniform bound on the variance). Consider the following inequality: m,γ,eσN ⊎ ⊎ ≤ F ◦kn µn (z) − F ◦kn Fµ⊲kn µn,N (z) + Fµ⊲kn (z) − Fνm,γ,σ (z) − F m,γ,eσN ⊲ n,N ν ⊲ n (z) (z) + F m,γ,eσN ⊲ ν (z) − Fνm,γ,σ ⊲ (z) where we shall refer to the terms on the right hand side of the inequality as (1), (2) and (3), respectively. We claim that we may make each of these terms arbitrarily small with the proper choice of N and n large enough. We begin by bounding (3). Choose K ⊂ C+ 1 of the initial value problem 1 compact and ǫ > 0. Fνm,γ,σ (z) is the solution at time ⊲ ∂tFt(z) + Eνm,γ,σ (Ft(z)) + log(m)Ft(z) = 0 ; F0(z) = z ⊎ ∂t is the solution of the corresponding system involving Eν . By Lemma 2.7 m,γ,eσN ⊎ Invoking Proposition 3.1, for every N ∈ N, there exists n(N) ∈ N such that (2) < ǫ for n ≥ n(N). We may further assume that n(N) is chosen large enough so that µ⊲kn n,N ∈ UN (the Proposition implies that these measures converge to νm,γ,σN ∈ UN as n ↑ ∞). Since tightness is equivalent to sequential precompactness, we have that {µ⊲kn n,N : N ∈ N, n ≥ n(N)} forms a tight family (our neighborhoods UN were chosen for this purpose). By the same argument as in Lemma 3.2, we have that {µ⊲j n,N : N ∈ N, n ≥ n(N), j = 1, . . . , kn} forms a tight family, so that ⊲ C = [N ∈N [n≥n(N ) kn[j=1 F ◦j µn,N (K) has compact closure. Let M0 be the upper bound on the magnitude of the derivative of the family of functions {F ◦j µn : n ∈ N, j = 1, . . . , kn} on this set C and M1 denote the upper bound on the magnitude of the elements in C. The upper bound M0 exists since this family of F -transforms arises m,γ,eσN ⊲ in C+, while Fν there exists N0 ∈ N such that term (3) < ǫ on K for N ≥ N0. In order to control terms (1) and (2), note that νm,γ,σN Lemma 2.7). Choose a family {UN}N ∈N such that ⊲ ⊲ (a) UN is a weak neighborhood of νm,γ,σ (b) νm,γ,σN ∈ UN (c) UN +1 ⊂ UN (d) ∩∞ N =1UN = {νm,γ,σ } ⊲ ⊲ → νm,γ,σ ⊲ as N ↑ ∞ (this follows from LIMIT THEOREMS FOR MONOTONIC CONVOLUTION 15 from a uniformly bounded, tight family of measures. Since this implies sequential precompactness on compact sets, this family is normal, so that such an upper bound exists for any compact set C. Since M0 and M1 do not depend on our choice of N, we may refine our choice of N0 so that ℑ(Fνm,γ,σ ⊎ (iN) − 1 N miN) < ǫ 4M0M1 , sup z∈C,t>N 1 + tz t − z < (1 + ǫ)M1 for all N ≥ N0 (the t are real numbers). The first estimate follows from the asymptotics of the F -transform. These inequalities will play a role in bounding term (1) Now fix N ≥ N0 and assume that n ≥ n(N) so that terms (2) and (3) are both bounded by ǫ. For any z ∈ K, we have the following inequality involving term (1): (1) = F ◦kn µn (z) − F ◦kn µn,N (z) µn (Fµn(z)) − F ◦(kn−1) ≤ F ◦kn−1 ≤ M0Fµn(z) − Fµn,N (z) + F ◦(kn−2) µn µn (Fµn,N (z)) + F ◦(kn−1) µn (Fµn,N (z)) − F ◦(kn−1) µn,N (Fµn,N (z)) (Fµn(Fµn,N (z))) − F ◦(kn−2) µn (Fµn,N (Fµn,N (z))) + F ◦(kn−2) µn (Fµn,N (Fµn,N (z))) − F ◦(kn−2) µn,N (Fµn,N (Fµn,N (z))) Continuing in this way, we get the estimate (1) ≤ M0 kn−1Xj=0 Fµn,N ◦ F ◦(kn−j−1) µn,N (z) − Fµn ◦ F ◦(kn−j−1) µn,N (z) For the key step in the estimate, observe that, for z ∈ C (cid:12)(cid:12)Fµn,N (z) − Fµn(z)(cid:12)(cid:12) ≤(cid:12)(cid:12)(cid:12)(cid:12)ZR\[−N,N ] 1 + tz t − z z∈C, t>N(cid:12)(cid:12)(cid:12)(cid:12) dσn(t)(cid:12)(cid:12)(cid:12)(cid:12) ≤ sup 1 + tz t − z (cid:12)(cid:12)(cid:12)(cid:12) σn(t > N). and note that we have already shown that the supremum has a bound of (1 + ǫ)M1. Now, recalling Remark 3.3 and the fact that N ≥ N0, knσn(t > N) ≤ 2knℑ(Fµn(iN) − 1 µn(R) iN) N ⊎ = N ⊎kn n µ ≤ (iN) − 1 2ℑ(F µn(R)kn iN) 4ℑ(Fνm,γ,σ (iN) − 1 N (Note that the last of these inequalities simply follows from the fact that µ⊎kn and a fundamental fact about the asymptotics of the F-transforms of convergent families of measures found in [BV93]. As such, it may be necessary to choose our n larger, but this does not create any new dependence since we have fixed our N ≥ N0.) Thus, our estimate for (1) becomes 4 zj ℑ(Fνm,γ,σ Nkn n → νm,γ,σ ⊎ (iN) − 1 (iN) − 1 (1) ≤ M0 4ℑ(Fνγ,σ zj kn miN) miN) miN) = M0 N . ⊎ , ⊎ kn−1Xj=0 where these kn−1Xj=0 zj = F ◦(kn−j−1) µn,N (z) ∈ C, n ≥ n(N), N ≥ N0 16 MICHAEL ANSHELEVICH AND JOHN D. WILLIAMS Combining our estimates, we have that (1) ≤ (4M0(1 + ǫ)M1)(cid:18) ǫ 4M0M1(cid:19) = ǫ(1 + ǫ) Thus, n ⊲ Fµ⊲kn (z) < (3 + ǫ)ǫ (z) − Fνm,γ,σ for z ∈ C and n ≥ n(N). This implies that these functions converge uniformly on compact sets n → νm,γ,σ which is equivalent to the fact that µ⊲kn weakly. This completes the proof of the forward direction. We now assume that µn ⊲ µn ⊲· · · ⊲ µn → νm,γ,σ νm,γ,σ ⊎ we have and claim that this implies µn ⊎ µn ⊎· · ·⊎ µn → . To see this, note that Lemma 3.2 implies that, for fixed large y > 0 and sufficiently large n, ⊲ ⊲ knσn({t > y}) ≤ 2knZR = 2kn y ℑ(Fµn(iy) − 1 µn(R) 1 + t2 y2 + t2 dσn(t) iy) ≤ − log(m) m−1 − 1 4ℑ(F ◦kn µn (iy) − 1 µn(R)kn iy) y . Using Lemma 2.1(c) again, the right hand side of the inequality goes to 0 as y ↑ ∞ (since this is a tight family). Also, knσn(R) = knℑ(cid:18)Fµn(i) − 1 µn(R) i(cid:19) , so the first statement in Lemma 3.2(b) implies that knσn(R) is bounded over n. Therefore this estimate implies tightness of the family of measures knσn. Now, let σ′ denote a weak cluster point for this family of finite measures. We claim that knγn is bounded along the relevant subsequence. Indeed, note that ℜ(F ◦kn µn (i)) = −knγn + kn−1Xj=0 ZR (t − ℜ(F ◦j µn(i)))(1 + tℜ(F ◦j µn(i)2 t − F ◦j µn(i))) − ℑ(F ◦j µn(i))2 dσn(t) The left hand side of the equation is convergent by assumption. As we have shown that {µ⊲j n }n∈N,j=1,...,kn is a tight family under these assumptions, the magnitude of the integrands on the right hand side of the equation have a uniform upper bound of c. Thus, we have the inequality, (cid:12)(cid:12)ℜ(F ◦kn µn (i)) + knγn(cid:12)(cid:12) ≤ knσn(R)c which implies that knγn is a bounded sequence. Thus, we may additionally assume that knγn → γ′ along this subsequence. Consider the function F γ ′,σ′ (z) = 1 m z − γ′ +ZR 1 + tz t − z dσ′(t). µn(R) z) → F γ ′,σ′(z) − 1 Then, along an appropriate subsequence, we have that kn(Fµn(z) − 1 mz uniformly on compact subsets of C+. Along with µn(R)kn → m, this implies that µ⊎kn n → νm,γ ′,σ′ ⊎ along an appropriate subsequence. But we have just shown that this fact implies that µ⊲kn n → LIMIT THEOREMS FOR MONOTONIC CONVOLUTION 17 ⊲ along this subsequence. Since we are assuming that µ⊲kn νm,γ ′,σ′ γ = γ′ and σ = σ′. This completes our proof. n → νm,γ,σ ⊲ we may conclude that (cid:3) Proof of Theorem 1.3. Equivalence of parts (c) and (d) follows by applying Proposition 3.4 to prob- ability measures. The remaining equivalences were proved in [BP99]. (cid:3) 4. LIMIT THEOREMS FOR MULTIPLICATIVE MONOTONE CONVOLUTION ON T 4.1. Preliminaries. Let µ denote a probability measure on the unit circle T. We define the trans- forms Note that the mean of µ is (2) ψµ(z) =ZT ZT zζ 1 − zζ dµ(ζ), ηµ(z) = ψµ(z) 1 + ψµ(z) . ζdµ(ζ) = lim z→0 ηµ(z) z = η′(0). We will always assume that this quantity is non-zero, in which case η−1 of zero and we may define a new transform µ is defined in a neighborhood Σµ(z) = η−1 µ (z) z It is immediate from the definition of ηµ that it takes the unit disk D to itself and ηµ(0) = 0, so that in fact for z ∈ D, ηµ(z) ≤ z. This fact is necessary in what follows as we will treat these transforms as composition operators on certain spaces of functions on D1/2 =(cid:26)z ∈ C : z < 1 2(cid:27) . By taking products of random variables that are freely, Boolean and monotonically independent, we may develop multiplicative forms of convolution. We will forgo the operator algebraic definition of the convolution operations and refer to [Voi87], [Fra08] and [Ber05] for the theory relevant to free, Boolean and monotone convolution, respectively. Instead, given probability measures µ and ν, we define the free, Boolean and monotone multiplicative convolution operations (in symbols µ ⊠ ν, µ ×∪ ν and µ (cid:8) ν) through their transforms. That is, we define binary operations on the space of probability measures on T implicitly through their transforms as follows: Σµ⊠ν(z) = Σµ(z)Σν(z), ηµx∪ν(z) z = ηµ(z) ην(z) z z , ηµ(cid:8)ν(z) = ηµ ◦ ην(z) It follows from equation (2) that ZT ζd(µ ×∪ ν)(ζ) =ZT ζdµ(ζ)ZT ζdν(ζ) =ZT ζd(µ (cid:8) ν)(ζ). Note that there is an analogous version of multiplicative convolution for classical independence. This binary operation is represented by the symbol ⊛. We will not discuss this type of convolution directly although it figures in our results. Σνγ,σ ⊛ )(p) = γp exp(cid:18)ZT (F νγ,σ (z) = γ exp(cid:18)ZT (z) = γz exp(cid:18)−ZT Definition 4.1. For β ∈ R and σ as above, denote Aβ,σ(z) = z(cid:18)iβ −ZT ηνγ,σ ⊠ x∪ Note that if γ = eiβ, then 1 − ℜ(ζ) 1 + ζz 1 − ζz 1 + ζz 1 − ζz dσ(ζ)(cid:19) , dσ(ζ)(cid:19) , dσ(ζ)(cid:19) , 1 + ζz 1 − ζz z ∈ D, z ∈ D. z ∈ D. 18 MICHAEL ANSHELEVICH AND JOHN D. WILLIAMS According to [Ber05] (see also [BP78]), the η-transforms of a multiplicative monotone convolution semigroup satisfy an equation of the form dηµt(z) dt = A(ηµt(z)), where the generator A of the semigroup is a general function of the form A(z) = zB(z), where B is analytic in D and ℜ(B(z)) ≤ 0, in other words B(z) = iβ −ZT 1 + ζz 1 − ζz dσ(ζ). According to [Wan08], we may identify the classes of ⊛, ⊠ and ×∪ infinitely divisible Borel proba- bility measures on T with γ ∈ T and σ a finite Borel measure on T dσ(ζ)(cid:19) , ζ p − 1 − ipℑ(ζ) p ∈ Z, exp(cid:18) 1 z Aβ,σ(z)(cid:19) = 1 z ηνγ,σ x∪ (z). Let {νt : t ≥ 0} be the multiplicative monotone convolution semigroup Aβ,σ generates, and denote νβ,σ (cid:8) = ν1. Remark 4.2. According to [EGRS02] (see [CM95] for the background), any multiplicative mono- tone convolution semigroup has the form ηνt(z) = u−1(rteiθtu(z)) for 0 < r ≤ 1 and θ ∈ R. Here rteiθt is the mean of νt, and the generator of the semigroup is A(z) = (log r + iθ) u(z) u′(z) , which thus specifies the conditions on the function u. ν1 determines r uniquely, u up to a multi- plicative constant, and θ up to an additive multiple of 2π. It follows that if νβ1,σ (cid:8) = νβ2,σ (cid:8) , then 2πik u(z) u′(z) = zi(β1 − β2), and the two generators are constant multiples of z. Such functions generate monotone convolution semigroups of delta measures. In all other cases, νβ1,σ (cid:8) . In particular, the reasoning behind the non-uniqueness arguments in the last section of [Ber05] does not hold. Nevertheless, as Hari Bercovici has pointed out to us, Theorem 4.4 and Proposition 4.5 (and their proofs) in that section 6= νβ2,σ (cid:8) LIMIT THEOREMS FOR MONOTONIC CONVOLUTION 19 are correct: any (cid:8)-infinitely divisible distribution can be included in a (cid:8)-convolution semigroup, and specifying the means determines the (cid:8)-square root uniquely. Lemma 4.3. Both νγ,σ x∪ and νβ,σ (cid:8) , where γ = eiβ, have mean γe−σ(T). Proof. The first statement is clear directly from the formula. For the second, since dηνt(z) dt = Aβ,σ(ηνt(z)), we have dη′ νt(0) dt It follows that The result follows. = (Aβ,σ)′(ηνt(0))η′ νt(0) = (Aβ,σ)′(0)η′ νt(0) = (iβ − σ(T))η′ νt(0). νt(0) = exp((iβ − σ(T))t) = eiβte−σ(T)t. η′ (cid:3) 4.2. Main Results. The following theorem is a restriction of the results from [BW08] and [Wan08] from infinitesimal arrays to sequences of measures. See these references also for results concerning the classical multiplicative convolution ⊛, as well as the special case of convergence to the Haar measure. Theorem 4.4. Fix a finite positive Borel measure σ on T, a complex number γ ∈ T, a sequence of probability measures {µn}n∈N on T converging to δ1 weakly, and a sequence of positive integers k1 < k2 < · · · The following assertions are equivalent: (a) The sequence µn ⊠ µn ⊠ · · · ⊠ µn } (b) The sequence µn ×∪ · · · ×∪ µn } {z {z ×∪ µn (c) kn kn and converges weakly to νγ,σ ⊠ ; converges weakly to νγ,σ x∪ ; kn(1 − ℜ(ζ)) dµn(ζ) → σ exp(cid:18)iknZT ℑ(ζ) dµn(ζ)(cid:19) → γ. Remark 4.5. A straightforward inclusion of the multiplicative monotone convolution into the pre- ceding theorem does not hold. Indeed, let η(z) = u−1(reiθu(z)), where 0 < r < 1. Condi- tions on u for η to be an η-transform are known, see Remark 4.2 (for example, we could take u(z) = 1 − z−(k−1) and reiθ = eα(k−1)). Then η◦(1/kn)(z) = u−1(r1/kneiθ/kne2πiℓn/knu(z)). For large kn, and assuming that ℓn/kn → 0, Therefore (cid:18) 1 z η◦(1/kn)(z)(cid:19)kn η◦(1/kn)(z) ≈ z +(cid:0)r1/kneiθ/kne2πiℓn/kn − 1(cid:1) u(z) ≈ exp(cid:18)kn(cid:0)r1/kneiθ/kne2πiℓn/kn − 1(cid:1) u(z) u′(z) . zu′(z)(cid:19) ≈ exp(cid:18)(iθ + 2πiℓn) u(z) zu′(z)(cid:19) , 20 MICHAEL ANSHELEVICH AND JOHN D. WILLIAMS which depends on the choice of ℓn. So the kn'th Boolean power of the kn's monotone root need not have a limit. Proposition 4.6. Suppose the sequence µn ×∪ µn β ∈ R with γ = eiβ, there exist λn ∈ T, λkn converges weakly to νβ,σ µn (cid:8) µn (cid:8) · · · (cid:8) µn {z kn (cid:8) kn {z ×∪ · · · ×∪ µn } Proof. To prove this proposition, first note that our hypotheses imply that converges weakly to νγ,σ x∪ . Then for any n = 1 such that for µn = δλn (cid:8) µn, the sequence } (cid:18) 1 z z z z 2πiℓn 1 + ζz 1 z → 1 + ζz 1 − ζz dσ(ζ)(cid:19) . (cid:12)(cid:12)(cid:12)(cid:12)kn(cid:20)log(cid:18) 1 1 + ζz 1 − ζz(cid:19)(cid:12)(cid:12)(cid:12)(cid:12) → 0 With these considerations in mind, we introduce the following correction. Let λn = e2πiℓn/kn. Define µn = δλn (cid:8) µn. Observe that ηγ,σ(z) = γ exp(cid:18)−ZT ηµn(z)(cid:19)kn z ηµn(z) → 1 on D1/2. We may take the logarithm of both sides of the and µn → δ1 weakly so that 1 above limit, but there is ambiguity as to the branches of the logarithm. Fix β with eiβ = γ. We may conclude that there exists a sequence {ℓn}n↑∞ of integers so that ηµn(z)(cid:19) + (3) kn (cid:21) −(cid:18)iβ −ZT as n ↑ ∞ where we have fixed the branch of the logarithm with ℑ(log(1)) = 0. The fact that log(cid:0) 1 z ηµn(z)(cid:1) → 0 implies that the same must be true of ℓn/kn. 1 − ζz(cid:19)(cid:12)(cid:12)(cid:12)(cid:12) =(cid:12)(cid:12)(cid:12)(cid:12)kn log(cid:18) 1 (cid:12)(cid:12)(cid:12)(cid:12)kn log(cid:18) 1 η(z)(cid:19) + 2πℓn −(cid:18)iβ −ZT ηµn(z) − 1(cid:19)(cid:12)(cid:12)(cid:12)(cid:12) → 0 ηµn(z)(cid:19) −(cid:18) 1 = kn(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ηµn(z)(cid:19)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∞Xp=2 ∞Xj=1 Indeed, since 1 series expansion for the logarithm centered at 1 so that our quantity becomes (1 − 1 ηµn(z)(cid:19) −(cid:18)iβ −ZT kn(cid:12)(cid:12)(cid:12)(cid:12)log(cid:18) 1 ηµn(z))j−1/j ≥ kn(1 − for n large enough. Since the left hand side is bounded (it converges to iβ −RT paragraphs calculation), we have that kn(1 − 1 1−ζz by the previous z ηµn(z)) is bounded. These facts imply our claim. z ηµn(z) → 1 (this holds since this limit is true for µn and λn → 1) we may take a which converges to 0 uniformly over D1/2. We next claim that, as n ↑ ∞, z ηµn(z)) is bounded. Indeed, appealing to the series expansion of the Now, observe that kn(1 − 1 logarithm; 1 + ζz 1 − ζz(cid:19)(cid:12)(cid:12)(cid:12)(cid:12) kn(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∞Xp=1 ηµn(z)) = kn(1 − (1 − 1 z ηµn(z))p p z ηµn(z))p p −(cid:18)1 − 1 z kn log( 1 z 1 z ηµn(z))/2 1 z ηµn(z)) (1 − 1 z z z (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 1+ζz LIMIT THEOREMS FOR MONOTONIC CONVOLUTION 21 Thus, we may conclude that lim n↑∞ and finally Let kn(cid:18) 1 z ηµn(z) − 1(cid:19) = lim n↑∞ kn log(cid:18) 1 z ηµn(z)(cid:19) → iβ −ZT 1 + ζz 1 − ζz dσ(ζ) kn (ηµn(z) − z) → Aβ,σ(z). D = {ψρ : ρ finite Borel measure on T} . = sup = sup zζ D is invariant under composition operators by functions ηµ, and these operators are all contractions on it. Let A be the completion of D with respect to the uniform norm on D1/2, which we denote by k·k∞. Then each composition operator by ηµ as above is a contraction on A. For h(z) = ψρ ∈ D, (cid:13)(cid:13)kn(h ◦ ηµn − h) − Aβ,σh′(cid:13)(cid:13)∞ z∈D1/2(cid:12)(cid:12)(cid:12)(cid:12)ZT(cid:18)kn(cid:18) ηµn(z)ζ 1 − zζ(cid:19) − z∈D1/2(cid:12)(cid:12)(cid:12)(cid:12)ZT(cid:18) (kn(ηµn(z) − z) − Aβ,σ(z))ζ z∈D1/2(cid:12)(cid:12)kn(ηµn(z) − z) − Aβ,σ(z)(cid:12)(cid:12) + 8ρ(T) sup (1 − zζ)2(cid:19) dρ(ζ)(cid:12)(cid:12)(cid:12)(cid:12) (1 − ηµn(z)ζ)(1 − zζ)2(cid:19) dρ(ζ)(cid:12)(cid:12)(cid:12)(cid:12) z∈D1/2(cid:12)(cid:12)(ηµn(z) − z)Aβ,σ(z)(cid:12)(cid:12) . On D1/2, Aβ,σ is bounded, and (cid:0)kn(ηµn(z) − z) − Aβ,σ(z)(cid:1) and (ηµn(z) − z) converge to zero uniformly. It follows that(cid:13)(cid:13)kn(h ◦ ηµn − h) − Aβ,σh′(cid:13)(cid:13)∞ → 0. Denote by Vn, B the operators on A given by (ηµn(z) − z)Aβ,σ(z)ζ (1 − ηµn(z)ζ)(1 − zζ) 1 − ηµn(z)ζ − ≤ 4ρ(T) sup Aβ,σ(z)ζ + and Then we have just shown that Vnh = h ◦ ηµn, Bh = Aβ,σh′. kn(Vn − I)h → Bh for h ∈ D, so in particular D is in the domain of B. The rest of the argument proceeds as in Proposition 3.1, and implies (by taking h(z) = z = ηδ1) that We conclude that n − ηνγ,σ (cid:13)(cid:13)(cid:13)ηµ(cid:8)kn (cid:8) (cid:13)(cid:13)(cid:13)∞ → 0. n → νβ,σ µ(cid:8)kn (cid:8) weakly. Theorem 4.7. Fix a sequence of probability measures {µn}n∈N on T, and a sequence of positive integers k1 < k2 < · · · . Assume that (4) (cid:3) knZT ℑ(ζ) dµn(ζ) → β. Then for γ = eiβ, the following assertions are equivalent: 22 MICHAEL ANSHELEVICH AND JOHN D. WILLIAMS converges weakly to νγ,σ x∪ ; so that kn {z {z kn (a) The sequence µn ×∪ µn n → γe−σ(T), converges weakly to νβ,σ (cid:8) ; kn log an → −σ(T). ×∪ · · · ×∪ µn } (b) The sequence µn (cid:8) µn (cid:8) · · · (cid:8) µn } Proof. Denote an =RT ζ dµn(ζ). Either of the assumptions (a) or (b) implies that akn Since an ≤ 1, the series expansion of the logarithm immediately gives kn(an − 1) → −σ(T). We also know that knℑ(an) → β. Combining these we conclude that kn(ℜ(an) − 1) → −σ(T) and, after a little more work, kn(log(an) − (an − 1)) → 0, where we again use the principal branch of the logarithm. Note also that the assumption (4) implies that µn → δ1 weakly. Suppose that µx∪kn in the weak topology. Plugging z = 0 in equation (3), using the compu- tation above, and taking the imaginary part, we see that ℓn in Proposition 4.6 is zero, and it follows from that proposition that µ(cid:8)kn n → νβ,σ are supported on T, they form a tight Now suppose that µ(cid:8)kn n family. Fix a subsequence converging to a measure νγ ′,σ′ (the fact that the point is infinitely divisible may be found in [Wan08]). Moreover, by comparing the means as above, it follows that eiβ = γ′ = (cid:8) . We may therefore conclude that γ. Then by the reverse implication, it follows that µ(cid:8)kn σ = σ′, completing the proof. (cid:3) n → νβ,σ (cid:8) . Since the measures µx∪kn x∪ n → νβ,σ′ n → νγ,σ x∪ (cid:8) REFERENCES []Bel05 Serban Teodor Belinschi, Complex analysis methods in noncommutative probability, ProQuest LLC, Ann Arbor, MI, 2005, Thesis (Ph.D.) -- Indiana University. MR []BGS02 Anis Ben Ghorbal and Michael Schurmann, Non-commutative notions of stochastic independence, Math. []Ber05 []BP78 []BP99 []BP00 []BV93 Proc. Cambridge Philos. Soc. 133 (2002), no. 3, 531 -- 561. MR1919720 (2003k:46096) Hari Bercovici, Multiplicative monotonic convolution, Illinois J. Math. 49 (2005), no. 3, 929 -- 951 (elec- tronic). MR1919720 (2003k:46096) Earl Berkson and Horacio Porta, Semigroups of analytic functions and composition operators, Michigan Math. J. 25 (1978), no. 1, 101 -- 115. MR1919720 (2003k:46096) Hari Bercovici and Vittorino Pata, Stable laws and domains of attraction in free probability theory, Ann. of Math. (2) 149 (1999), no. 3, 1023 -- 1060, With an appendix by Philippe Biane. MR1919720 (2003k:46096) , Limit laws for products of free and independent random variables, Studia Math. 141 (2000), no. 1, 43 -- 52. MR1919720 (2003k:46096) Hari Bercovici and Dan Voiculescu, Free convolution of measures with unbounded support, Indiana Univ. Math. J. 42 (1993), no. 3, 733 -- 773. MR1254116 (95c:46109) []BW08 Hari Bercovici and Jiun-Chau Wang, Limit theorems for free multiplicative convolutions, Trans. Amer. []Bia98 []Che74 []Che76 []CM95 Math. Soc. 360 (2008), no. 11, 6089 -- 6102. MR1254116 (95c:46109) Philippe Biane, Processes with free increments, Math. Z. 227 (1998), no. 1, 143 -- 174. MR1254116 (95c:46109) Paul R. Chernoff, Product formulas, nonlinear semigroups, and addition of unbounded operators, American Mathematical Society, Providence, R. I., 1974, Memoirs of the American Mathematical Society, No. 140. MR1254116 (95c:46109) , On the converse of the Trotter product formula, Illinois J. Math. 20 (1976), no. 2, 348 -- 353. MR1254116 (95c:46109) Carl C. Cowen and Barbara D. MacCluer, Composition operators on spaces of analytic functions, Studies in Advanced Mathematics, CRC Press, Boca Raton, FL, 1995. LIMIT THEOREMS FOR MONOTONIC CONVOLUTION 23 []EGRS02 Mark Elin, Victor Goryainov, Simeon Reich, and David Shoikhet, Fractional iteration and functional equa- tions for functions analytic in the unit disk, Comput. Methods Funct. Theory 2 (2002), no. 2, 353 -- 366. MR1254116 (95c:46109) Klaus-Jochen Engel and Rainer Nagel, One-parameter semigroups for linear evolution equations, Gradu- ate Texts in Mathematics, vol. 194, Springer-Verlag, New York, 2000, With contributions by S. Brendle, M. Campiti, T. Hahn, G. Metafune, G. Nickel, D. Pallara, C. Perazzoli, A. Rhandi, S. Romanelli and R. Schnaubelt. MR1254116 (95c:46109) []EN00 []Fel71 William Feller, An introduction to probability theory and its applications. Vol. II., Second edition, John []Fra06 []Fra08 []Fra09 []Gol76a []Gol76b []Has10 []HS11 Wiley & Sons Inc., New York, 1971. MR1254116 (95c:46109) Uwe Franz, Multiplicative monotone convolutions, Quantum probability, Banach Center Publ., vol. 73, Polish Acad. Sci., Warsaw, 2006, pp. 153 -- 166. MR2423123 , Boolean convolution of probability measures on the unit circle, Analyse et probabilit´es. 16 (2008), 83 -- 94. MR2423123 , Monotone and Boolean convolutions for non-compactly supported probability measures, Indiana Univ. Math. J. 58 (2009), no. 3, 1151 -- 1185. MR2541362 Jerome A. Goldstein, Semigroup-theoretic proofs of the central limit theorem and other theorems of analy- sis, Semigroup Forum 12 (1976), no. 3, 189 -- 206. MR2541362 , Corrigendum on: "Semigroup-theoretic proofs of the central limit theorem and other theorems of analysis" (Semigroup Forum 12 (1976), no. 3, 189-206), Semigroup Forum 12 (1976), no. 4, 388. MR2541362 Takahiro Hasebe, Monotone convolution semigroups, Studia Math. 200 (2010), no. 2, 175 -- 199. MR2541362 Takahiro Hasebe and Hayato Saigo, The monotone cumulants, Ann. Inst. Henri Poincar´e Probab. Stat. 47 (2011), no. 4, 1160 -- 1170. []Maa92 Hans Maassen, Addition of freely independent random variables, J. Funct. Anal. 106 (1992), no. 2, 409 -- []MZ74 438. MR1165862 (94g:46069) Deane Montgomery and Leo Zippin, Topological transformation groups, Robert E. Krieger Publishing Co., Huntington, N.Y., 1974, Reprint of the 1955 original. MR1165862 (94g:46069) []Mur00 Naofumi Muraki, Monotonic convolution and monotonic L´evy-Hincin formula, preprint, 2000. []Mur01 , Monotonic independence, monotonic central limit theorem and monotonic law of small numbers, []Mur02 []Mur03 []Pfe83 []Sis98 []Spe97 []Voi87 []Wan08 []Wan11 []Wan12 []Wil12 Infin. Dimens. Anal. Quantum Probab. Relat. Top. 4 (2001), no. 1, 39 -- 58. MR1824472 (2002e:46076) , The five independences as quasi-universal products, Infin. Dimens. Anal. Quantum Probab. Relat. Top. 5 (2002), no. 1, 113 -- 134. MR1895232 (2003e:46113) , The five independences as natural products, Infin. Dimens. Anal. Quantum Probab. Relat. Top. 6 (2003), no. 3, 337 -- 371. MR2016316 (2005h:46093) D. Pfeifer, A semigroup-theoretic proof of Poisson's limit law, Semigroup Forum 26 (1983), no. 3-4, 379 -- 382. MR2016316 (2005h:46093) Aristomenis G. Siskakis, Semigroups of composition operators on spaces of analytic functions, a review, Studies on composition operators (Laramie, WY, 1996), Contemp. Math., vol. 213, Amer. Math. Soc., Providence, RI, 1998, pp. 229 -- 252. MR2016316 (2005h:46093) Roland Speicher, On universal products, Free probability theory (Waterloo, ON, 1995), Fields Inst. Com- mun., vol. 12, Amer. Math. Soc., Providence, RI, 1997, pp. 257 -- 266. MR1426844 (98c:46141) D.V. Voiculescu, Multiplication of certain noncommuting random variables, J. Operator Theory, 18 (1987), 223 -- 235. MR1426844 (98c:46141) Jiun-Chau Wang, Limit laws for Boolean convolutions, Pacific J. Math. 237 (2008), no. 2, 349 -- 371. MR1426844 (98c:46141) , The central limit theorem for monotone convolution, preprint, 2011. , Strict limit types for monotone convolution, J. Funct. Anal. 262 (2012), no. 1, 35 -- 58. MR1426844 (98c:46141) John D. Williams, A Khinthine Decomposition for Free Probability, Annals of Probability, Accepted for Publication. 24 MICHAEL ANSHELEVICH AND JOHN D. WILLIAMS DEPARTMENT OF MATHEMATICS, TEXAS A&M UNIVERSITY, COLLEGE STATION, TX 77843-3368 E-mail address: [email protected], [email protected]
1612.01050	2	1612	2017-08-01T12:03:38	Residual finite dimensionality and representations of amenable operator algebras	[ "math.OA", "math.FA" ]	We consider a version of a famous open problem formulated by Kadison, asking whether bounded representations of operator algebras are automatically completely bounded. We investigate this question in the context of amenable operator algebras, and we provide an affirmative answer for representations whose range is residually finite-dimensional. Furthermore, we show that weak-${}^$ closed, amenable, residually finite-dimensional operator algebras are similar to $C^$-algebras, and in particular have the property that all their bounded representations are completely bounded. We prove our results for operator algebras having the so-called total reduction property, which is known to be weaker than amenability.	math.OA	math	RESIDUAL FINITE DIMENSIONALITY AND REPRESENTATIONS OF AMENABLE OPERATOR ALGEBRAS RAPHAEL CLOU ATRE AND LAURENT W. MARCOUX Abstract. We consider a version of a famous open problem formulated by Kadison, ask- ing whether bounded representations of operator algebras are automatically completely bounded. We investigate this question in the context of amenable operator algebras, and we provide an affirmative answer for representations whose range is residually finite-dimen- sional. Furthermore, we show that weak-∗ closed, amenable, residually finite-dimensional operator algebras are similar to C ∗-algebras, and in particular have the property that all their bounded representations are completely bounded. We prove our results for operator algebras having the so-called total reduction property, which is known to be weaker than amenability. 1. Introduction Let B(H) denote the C ∗-algebra of bounded linear operators on some complex Hilbert space H. Given a Banach algebra A, we shall refer to a continuous algebra homomorphism θ : A → B(H) as a representation of A. In particular, representations of C ∗-algebras are not assumed to be self-adjoint. One of the most intriguing questions in the theory of C ∗-algebras is the following, which stems from a 1955 paper of R.V. Kadison [13]. Kadison's similarity problem. Let A be a C ∗-algebra and let θ : A → B(H) be a representation. Does there exist an invertible operator X ∈ B(H) so that the representation θX : A → B(H) defined by is a ∗-representation? θX(a) = X −1θ(a)X, a ∈ A We say that A has Kadison's similarity property if the problem above has an affirmative answer. It is still unknown to this day whether every C ∗-algebra has this property, although some deep partial results have emerged throughout the years. For instance, E. Christensen [6] has solved the problem in the affirmative for amenable C ∗-algebras (we note here that due to sophisticated results of A. Connes [8] and U. Haagerup [10], the classes of amenable and nuclear C ∗-algebras are known to coincide). It was shown by Haagerup [11] that Kadison's similarity problem has an affirmative answer for every C ∗-algebra, provided that one restricts one's attention to only those representations which admit a finite cyclic set of vectors. More- over, it is known [11] that a representation of a C ∗-algebra is similar to a ∗-representation if and only if it is completely bounded. This allows one to reformulate Kadison's similarity problem so that it makes sense in a non self-adjoint context. The following problem appeared in [17]. The generalised similarity problem. Let A be an operator algebra and let θ : A → B(H) be a representation. Is θ necessarily completely bounded? The first author was partially supported by an FQRNT postdoctoral fellowship and an NSERC Discovery Grant. The second author was partially supported by an NSERC Discovery grant. 1 2 R. Clouatre and L.W. Marcoux Following [17], we say that the A has the SP property if the problem above has an affirmative answer. In producing an example of a polynomially bounded operator which is not similar to a contraction -- thereby answering a long-standing open question (i.e. the Halmos Problem) as to the existence of such an operator -- Pisier demonstrated that the classical disc algebra is an example of an operator algebra without the SP property [16]. On the other hand, a straightforward verification shows that if A ⊂ B(H) has the SP property, then so does XAX −1 for any invertible operator X ∈ B(H). In particular, it follows from Christensen's result mentioned above that any operator algebra that is similar to an amenable C ∗-algebra has the SP property. In the 1980's there arose the question of determining which operator algebras are similar to amenable C ∗-algebras. Since amenability is preserved under Banach algebra isomorphisms -- of which similarity of operator algebras is an example -- an obvious restriction on such operator algebras is that they be amenable. It was conjectured by a number of people in the Banach algebra community that this was in fact the only obstruction, and that any amenable operator algebra was similar to a C ∗-algebra. This conjecture was verified in some special cases. Indeed, it was shown to hold for uniform algebras by M.V. Seınberg [20], for abelian amenable subalgebras of finite von Neumann algebras by Y. Choi [4], and very recently for arbitrary abelian amenable operator algebras by the second author and A. Popov [14]. However, other recent developments have shown the conjecture to be incorrect in general, and an example of a non-abelian, non-separable, amenable operator algebra which fails to be similar to a C ∗-algebra was constructed in [5] by Y. Choi, I. Farah and N. Ozawa. It is not currently known whether a separable, amenable operator algebra must always be similar to a C ∗-algebra. Nevertheless, this counterexample shows that a positive solution to the generalised similarity problem for amenable operator algebras cannot be achieved through this approach by relying on Christensen's result. Interestingly, the counterexample does have the SP property, as shown in the companion paper [7]. Thus, the question of whether every amenable operator algebra has the SP property remains unanswered and it motivated much of our efforts. In fact, we will be interested in a slightly more general question. An illuminating insight into Kadison's similarity problem and the generalised similarity problem was offered by J.A. Gifford in his PhD thesis [9]. Therein, he obtained a complete characterization of the C ∗-algebras having Kadison's similarity property as those having a remarkably well behaved lattice of invariant subspaces, a feature he called the total reduction property (the precise definition is given in Section 2). He also observed that the total reduction property is strictly weaker than amenability (we point out that the result of [14] mentioned above was in fact proved under this weaker assumption). We can now state the basic question which we wish to address in this paper. Question. Let A be a norm-closed operator algebra with the total reduction property. Is every representation θ : A → B(H) completely bounded? The reader will notice that this question is a special case of the generalised similarity problem. Based on the discussion above, we believe it to be a meaningful step towards a complete understanding of the latter. We now describe the organization of the paper, and state our main results. Section 2 deals with preliminaries: we give precise definitions of the concepts we require throughout, gather relevant results from the literature and prove some basic facts in preparation for later work. In Section 3, we explain how the general problem we are trying to solve can be reduced to one involving matrix algebras, at least in the amenable case. Consequently, in trying to determine whether every amenable operator algebra has the SP property, it is no RFD and representations of amenable operator algebras 3 loss of generality to focus on representations whose domains are residually finite-dimensional algebras. Motivated by this observation, in Section 4 we examine the structure of subalgebras of products of matrix algebras that possess the total reduction property. This information is then leveraged to prove our first main result (see Theorem 4.5 and Corollary 4.6). 1.1. Theorem. Let (k(λ))λ be a net of positive integers and suppose that A ⊂Qλ Mk(λ) is a subalgebra with the total reduction property. If A is closed in the weak-∗ topology, then A is similar to a C ∗-algebra and in particular it has the SP property. Although the condition of being weak-∗ closed restricts the range of applicability of the previous result, we view it as noteworthy partial step towards clarifying the general situation. Finally, Section 5 is devoted to the study of representations whose range is residually finite- dimensional, and our second main result is proved therein (see Corollary 5.5). Roughly speaking, it says that one has uniform control on the completely bounded norm of finite- dimensional representations in the presence of the total reduction property. 1.2. Theorem. Let (k(λ))λ be a net of positive integers, let A be an operator algebra k(λ) be a representation. Then, θ is completely bounded. with the total reduction property and let θ : A →Qλ The previous result fails without the total reduction property (Example 5.7). M 2. Preliminaries and background material 2.1. Operator algebras and completely bounded maps. Throughout, by an operator algebra we mean a subalgebra of some B(H) which is closed in the norm topology. If A ⊂ B(H) is an operator algebra, then for each integer n ≥ 1 the algebra Mn(A) inherits a norm when viewed as a subalgebra of the C ∗-algebra B(H(n)). If B is another operator algebra, then a linear map ϕ : A → B induces a sequence of maps ϕ(n) : Mn(A) → Mn(B), n ≥ 1 by setting ϕ(n)([ai,j]) = [ϕ(ai,j)] for all [ai,j] ∈ Mn(A). The map ϕ is completely bounded if the quantity kϕkcb = sup n≥1 kϕ(n)k is finite. The map ϕ is said to be completely contractive (respectively, completely isometric) if each ϕ(n) is contractive (respectively, isometric). We refer the reader to [15] for a thorough exposition of the theory of completely bounded maps and of operator algebras. 2.2. Amenability and the total reduction property. We recall the notion of amenabil- ity of a Banach algebra, which was introduced by Johnson in [12]. Let A be a Banach algebra and let X be a Banach A-bimodule. If X ∗ denotes the dual space of X, then X ∗ also carries the structure of an A-bimodule under the dual actions defined by (x∗a)(x) = x∗(ax) (ax∗)(x) = x∗(xa) and for all a ∈ A, x∗ ∈ X ∗ and x ∈ X. When it is equipped with this precise module action inherited from that of A on X, we say that X ∗ is a dual Banach A-bimodule. A derivation δ : A → X is a continuous linear map which satisfies δ(ab) = δ(a)b + aδ(b) 4 R. Clouatre and L.W. Marcoux for all a, b ∈ A. The derivation is inner if there exists a fixed element z ∈ X so that δ(a) = az − za for all a ∈ A. Finally, we say that A is amenable if every derivation of A into a dual Banach A-bimodule is inner. As mentioned in the introduction, we will be mostly concerned with a notion, introduced in the thesis of J.A. Gifford [9], that is weaker than amenability. An operator algebra A is said to have the total reduction property if, whenever θ : A → B(H) is a representation and M ⊂ H is a closed θ(A)-invariant subspace, there exists another closed θ(A)-invariant subspace N which is a topological complement of M , in the sense that H = M + N and M∩N = {0}. Equivalently, any closed θ(A)-invariant subspace is the range of some bounded idempotent lying in the commutant θ(A)′. More precise information about these idempotents is available [9, Proposition 2.2.13]. Indeed, for each t ≥ 0 there is a positive constant C(t) with the property that whenever θ : A → B(H) is a representation with kθk ≤ t and M ⊂ H is a closed θ(A)-invariant subspace, there is an idempotent E ∈ θ(A)′ with EH = M and kEk ≤ C(t). For each t ≥ 0, we denote by κA(t) the minimum of all positive constants C(t) satisfying this condition. Clearly, κA is an increasing function. We single out another one of its basic properties. 2.3. Lemma. Let A be an operator algebra with the total reduction property and let be a representation of A such that (A) is closed. Then, (A) has the total reduction property. Moreover, if θ is a representation of (A), then In particular, for every t > 0 we see that κ(A)(kθk) ≤ κA(kθ ◦ k). κ(A)(t) ≤ κA(tkk). Proof. The fact that (A) has the total reduction property is simply [9, Proposition 3.3.1]. Next, let θ : (A) → B(H) be a representation and let M ⊂ H be a closed θ((A))- invariant subspace. Since A has the total reduction property, there is an idempotent E ∈ θ((A))′ such that EH = M and kEk ≤ κA(kθ ◦ k). We conclude that κ(A)(kθk) ≤ κA(kθ ◦ k) ≤ κA(kθkkk). (cid:3) We mention in passing that if, in the above result, we do not assume that (A) is closed, then a similar argument shows that (A) has the total reduction property, and that κ(A)(t) ≤ κA(tkk) for all t ≥ 0. It is relevant to point out that amenability implies the total reduction property [9, Propo- sition 2.3.2]. However, the latter is strictly weaker than amenability: if H is an infinite- dimensional Hilbert space, then B(H) has the total reduction property [9, Corollary 2.4.7 ] but it is not amenable since it is not nuclear [22],[8]. Before proceeding, we gather here several facts about operator algebras with the total reduction property that we require numerous times in the sequel. First, we consider ideals. RFD and representations of amenable operator algebras 5 2.4. Theorem. Let A be an operator algebra with the total reduction property and let J ⊂ A be a closed two-sided ideal. Then, J has the total reduction property. Moreover, J admits a bounded approximate identity: there exists a bounded net (ei)i in J such that lim i kaei − ak = lim i keia − ak = 0 for every a ∈ J . Proof. This follows from [9, Propositions 3.2.7 and 3.3.3]. (cid:3) Next, we deal with a minor technical detail. Recall that an algebra A ⊂ B(H) acts non- degenerately if AH = H. While operator algebras with the total reduction property do not necessarily act non-degenerately, they nearly do so. 2.5. Lemma. Let A ⊂ B(H) be an operator algebra with the total reduction property. Then, there exists an invertible operator X ∈ B(H) with kXk = kX −1k ≤ 1 + κA(1) such that according to some orthogonal decomposition H = H0 ⊕ H⊥ non-degenerately acting subalgebra with the total reduction property. 0 . Moreover, A0 ⊂ B(H0) is a XAX −1 = A0 ⊕ {0} Proof. Let M = AH which is a closed A-invariant subspace. There is an idempotent E ∈ A′ such that EH = M and kEk ≤ κA(1). Then, it is well-known that there is an invertible operator X ∈ B(H) with kXk = kX −1k ≤ 1 + kEk ≤ 1 + κA(1) and such that XEX −1 is a self-adjoint projection. The space XM is then reducing for XAX −1. Put H0 = XM and A0 = (XAX −1)H0. We note that so that XAX −1 = A0⊕{0} according to the decomposition H = H0⊕H⊥ 0 . Thus, A0 has the total reduction property by Lemma 2.3. Moreover, A has a bounded approximate identity (ei)i by Theorem 2.4. Then, if ξ ∈ H and a ∈ A we have that XAX −1H ⊂ XM = H0 aξ = lim eiaξ ∈ AM i whence M = AM . Thus which shows that A0 acts non-degenerately on H0. A0H0 = XAM = XM = H0 (cid:3) We can now extract more information about ideals. 2.6. Theorem. Let A ⊂ B(H) be a weak-∗ closed operator algebra with the total reduction property and let J ⊂ A be a weak-∗ closed two-sided ideal. Then, there is a central idempotent e ∈ J ∩ A′ such that J = eA and kek ≤ κA(1). Proof. The existence of an idempotent e ∈ J ∩ A′ such that J = eA follows from Lemma 2.5 and [9, Proposition 3.2.2 and Corollary 3.1.5]. To get the announced norm estimate, we proceed as follows. Consider the closed subspace M = eH. Since e ∈ A′, we see that M is A-invariant. Hence, there is another idempotent f ∈ A′ with kfk ≤ κA(1) such that M = fH. Note that ef = f e because e ∈ J ⊆ A. Commuting idempotents with identical ranges must be equal, so that indeed kek ≤ κA(1). (cid:3) 6 R. Clouatre and L.W. Marcoux As a consequence of the previous theorem, we see that if A ⊂ B(H) is a weak-∗ closed operator algebra with the total reduction property, then A has a unit u with kuk ≤ κA(1). In particular, if J ⊂ A is a weak-∗ closed two-sided ideal, then we have a topological direct sum decomposition We now state the result mentioned in the introduction relating the total reduction property A = J + (u − e)A where J = eA. to Kadison's similarity property. 2.7. Theorem. Let A be a C ∗-algebra. (1) If A has Kadison's similarity property, then A has the total reduction property. (2) If A has the total reduction property, then A has Kadison's similarity property. More precisely, if θ : A → B(H) is a representation, then there is an invertible operator X ∈ B(H) such that is a ∗-homomorphism of A and a 7→ Xθ(a)X −1, a ∈ A In particular, we see that kXkkX −1k ≤ 128κA(kθk)2. kθkcb ≤ 128κA(kθk)2. Proof. This is a combination of Lemmas 2.4.1, 2.4.3 and Proposition 2.4.4 in [9]. (cid:3) We close this section with one of the main results of [9], which states that if an operator algebra consisting of compact operators has the total reduction property, then it is similar to a C ∗-algebra. We require a refined form of a special case of this theorem, which we prove below. 2.8. Theorem. Let H be a finite-dimensional Hilbert space and suppose that A ⊂ B(H) is an operator algebra with the total reduction property. Then, there exists an invertible operator X ∈ B(H) with the property that XAX −1 is a C ∗-algebra, and such that kXkkX −1k ≤ (1 + κA(1))2 128(1 + 2κA(1))κA(1 + 2κA(1))2. Proof. First, we note that by virtue of Lemma 2.5, we may assume without loss of general- ity that A acts non-degenerately upon conjugating with an invertible operator V ∈ B(H) satisfying It is this potential initial conjugation that accounts for the first term of (1 + κA(1))2 on the right-hand side of the inequality appearing in the statement. kV k = kV −1k ≤ 1 + κA(1). The proof then consists of a combination of results scattered throughout [9]. We see that A consists of compact operators on H, and so by [9, Lemma 4.3.12] there exist finitely many minimal idempotents E1, . . . , En ∈ A′′ ∩ A′ such that andPn that for each k the algebra (EkAEk)′′ = EkA′′Ek contains no proper central idempotent. k=1 Ek = I. Necessarily we have that these idempotents are pairwise orthogonal and A = E1AE1 + . . . + EnAEn RFD and representations of amenable operator algebras 7 By [9, Lemmas 1.0.3 and 3.2.3], we know that there exists an invertible operator Y ∈ B(H) such that Pk = Y EkY −1 is a self-adjoint projection for every 1 ≤ k ≤ n, and moreover Note that for each 1 ≤ k ≤ n, so we find kY kkY −1k ≤ 1 + 2κA(1). PkY AY −1Pk = Y EkAEkY −1 Y AY −1 = Y E1AE1Y −1 + . . . + Y EnAEnY −1 = Moreover, we see that nMk=1 PkY AY −1Pk. (PkY AY −1Pk)′′ = Y (EkAEk)′′Y −1 contains no proper central idempotent for each 1 ≤ k ≤ n. Using [9, Lemma 4.3.11], for each k we find an invertible operator Zk such that the algebra ZkPkY AY −1PkZ −1 is self-adjoint and k Now, by Lemma 2.3 we see that kZkk = kZ −1 k k ≤ √128κPkY AY −1Pk (1). κPkY AY −1Pk (1) ≤ κA(kY kkY −1k) ≤ κA(1 + 2κA(1)) so that kZkk = kZ −1 Since the orthogonal projections P1, . . . , Pn are pairwise orthogonal and satisfyPn if we set Z =Ln √128κA(1 + 2κA(1)). k=1 PkZkPk then Z is invertible with Z −1 =Ln kZkkZ −1k ≤ 128κA(1 + 2κA(1))2. k=1 PkZ −1 k k ≤ see that Finally, by setting X = ZY we obtain k=1 Pk = I, k Pk. Moreover, we XAX −1 = which is a C ∗-algebra, and nMk=1 PkZkPkY AY −1PkZ −1 k Pk kXkkX −1k ≤ (1 + 2κA(1))128κA(1 + 2κA(1))2. (cid:3) 3. A reduction to residually finite-dimensional operator algebras This section is meant as motivation for the rest of the paper. The goal here is to show that for amenable operator algebras, the generalized similarity problem can be transplanted to the concrete setting of products of matrix algebras without loss of generality. We will accomplish this by considering cones of operator algebras. Recall that if A is a Banach algebra, then the cone of A is the Banach algebra C(A) = {f : [0, 1] → A : f is continuous and f (0) = 0} where if f ∈ C(A) then kfk = sup t∈[0,1]kf (t)k. 8 R. Clouatre and L.W. Marcoux Alternatively, we see that C(A) = C0((0, 1]) ⊗ A, equipped with the injective tensor norm. Interestingly, taking the cone of an algebra preserves amenability [19, Exercise 2.3.6]. We need the following routine fact. 3.1. Proposition. Let A and B be operator algebras and θ : A → B be a representation. Then θ induces a representation Φθ : C(A) → C(B) defined by the formula (Φθf )(t) = θ(f (t)), t ∈ [0, 1]. Furthermore, the map θ is completely bounded if and only if Φθ is completely bounded, and we have kθkcb = kΦθkcb. Proof. Let n ∈ N. We see that if f ∈ Mn(C(A)), then k(Φ(n) θ f )(t)kMn(C(B)) ≤ kθ(n)kkf (t)kMn(A) for every t ∈ [0, 1], so that kΦ(n) Define fa ∈ Mn(C(A)) as θ k ≤ kθ(n)k. For the reverse inequality, let a ∈ Mn(A). Then, we see that kfakMn(C(A)) = kakMn(A) and fa(t) = ta, 0 ≤ t ≤ 1. Hence Φ(n) θ (fa) = fθ(n)(a). kθ(n)(a)kMn(B) = kfθ(n)(a)kMn(C(B)) = kΦ(n) θ kkfakMn(C(A) = kΦ(n) ≤ kΦ(n) θ (fa)kMn(C(B)) θ kkakMn(A) which shows that kθ(n)k ≤ kΦ(n) θ k. (cid:3) We thus see that to verify whether an operator algebra A has the SP property, it is sufficient to check that the cone C(A) has it. Before we proceed further, we introduce some notation which will be used throughout the remainder of the paper. Let Λ 6= ∅ be a set and let k : Λ → N be a function. We associate with k the following C ∗-algebra: Mk =Yλ M k(λ) = {(aλ)λ : aλ ∈ Mk(λ) for all λ ∈ Λ and sup λ kaλk < ∞}. Let us also define for each λ ∈ Λ the component map qk λ : Mk → Mk(λ) via When Λ is a directed set, we shall use the notation Lk instead of Mk in order to emphasize this distinction. The direction on Λ allows us to define the closed, two-sided ideal qk λ((aα)α) = aλ. Jk = {(aλ)λ ∈ Lk : lim λ kaλk = 0}. It is easily verified that Jk is nuclear, and hence amenable [10]. We may now construct the quotient C ∗-algebra We let Qk = Lk/Jk. πk : Lk → Qk RFD and representations of amenable operator algebras 9 denote the quotient map. The main observation of this section is the following, which is a combination of classical facts. It is a direct adaptation of the discussion found after [5, Theorem 1]. We present the proof here for the convenience of the reader. 3.2. Theorem. The following statements are equivalent. (i) Every amenable operator algebra has the SP property. (ii) Let k : Λ → N and k′ : Λ′ → N be nets, let D ⊂ Lk be an amenable operator algebra, and let θ : D → Qk′ be a representation. Then θ is completely bounded. Proof. We need only prove that (ii) implies (i). Assume therefore that (ii) holds and that A ⊂ B(H) is an amenable operator algebra. Let θ : A → B(Hθ) be a representation. We proceed to show that θ is completely bounded. For this purpose, let A = C ∗(A) ⊂ B(H). Then, C(A) ⊂ C(A). It is easy to see that C(A) is homotopic to zero, so that C(A) is quasidiagonal by [23, Theorem 5] (alternatively, see [3, Corollary 7.3.7] for the precise statement we need). In particular, by a straightforward adaptation of [3, Exercise 7.1.3] we may view C(A) as a C ∗-subalgebra of Qk for some net k : Λ → N. We conclude that C(A) ⊂ Qk. An identical argument shows that C(B) ⊂ Qk′ for some net k′ : Λ′ → N, where B = θ(A). By Proposition 3.1, there is a representation Φθ : C(A) → C(B) which is completely bounded if and only if θ is. completely bounded if and only if In turn, it is easily verified that Φθ is Φθ ◦ πk : π−1 k (C(A)) → C(B) ⊂ Qk′ is completely bounded. We know that the cone C(A) is amenable. If we let D = π−1 k (C(A)), then we may conclude from [19, Theorem 2.3.10] that D is an amenable subalgebra of Lk. Hence (ii) implies that Φθ ◦ πk, and thus θ, is completely bounded. (cid:3) We remark here that it is plausible that a version of this theorem holds for algebras which merely have the total reduction property. However, a direct adaptation of the proof would require some technology which is unavailable at present, and as such we postpone this interesting issue to future work. We also emphasize that Theorem 3.2 shows that from the point of view of attempting to solve the generalised similarity problem for amenable operator algebras, it is very meaningful to study amenable subalgebras of products of matrix algebras. We undertake this task for the larger class of operator algebras with the total reduction property, and accordingly we introduce the following convenient terminology. A subalgebra A ⊂ B(H) is said to be residually finite-dimensional if there exists a family of finite-dimensional Hilbert spaces Hλ and a family of completely contractive representations λ : A → B(Hλ) such that the map a 7→Mλ λ(a), a ∈ A is completely isometric. We mention that this definition is consistent with common usage of the term within the realm of C ∗-algebras. 10 R. Clouatre and L.W. Marcoux It is clear that for any function k : Λ → N, the algebra Mk considered above is residually finite-dimensional. Furthermore, subalgebras of residually finite-dimensional operator alge- bras are residually finite-dimensional as well. We now exhibit a less trivial example, which we will revisit later in the paper. 3.3. Example. Let H be a Hilbert space and let (Pλ)λ be a net of finite rank projections increasing strongly to I. Let T ⊂ B(H) denote the collection of operators which are trian- gular with respect to these projections, that is T ∈ T if and only if PλT Pλ = T Pλ for every λ. A standard verification shows that T is a weak-∗ closed algebra. For each λ, the map defined by λ : T → B(PλH) λ(T ) = PλT Pλ, T ∈ T (cid:3) (cid:3) is a completely contractive homomorphism. Moreover, since (Pλ)λ increases to I, it is easy pletely isometric weak-∗ homeomorphic algebra homomorphism. For future use, we also point that each λ is clearly weak-∗ continuous. By a standard Interestingly, residual finite dimensionality of an operator algebra A ⊂ B(H) is not equiv- to see thatLλ λ is completely isometric, so that T is residually finite-dimensional. application of the Krein-Smulian theorem [1, Theorem A.2.5], we see thatLλ λ is a com- alent to that of C ∗(A), as the following examples show. 3.4. Example. Let H be an infinite-dimensional, separable Hilbert space with orthonor- mal basis {em}∞ m=1. For each n ∈ N, denote by Pn the orthogonal projection of H onto span{e1, e2, . . . , en}. Let T ⊂ B(H) be the algebra of triangular operators with respect to (Pn)n. By Example 3.3, we see that T is residually finite-dimensional. On the other hand, it is easy to verify that the ideal of compact operators K(H) belongs to C ∗(T ). To show that C ∗(T ) is not residually finite-dimensional, it suffices to show that K(H) is not. But any completely contractive homomorphism of K(H) is a ∗-homomorphism, and there are no non-zero ∗-homomorphisms from K(H) into a finite-dimensional C ∗-algebra, in view of H being infinite-dimensional. We close this section by exhibiting a class of residually finite-dimensional C ∗-algebras which is of particular interest to us, in light of Theorem 3.2. We suspect that the following statement is well-known to experts, but we provide a proof for the reader's convenience. 3.5. Proposition. Let k : Λ → N be a bounded net. Then, the C ∗-algebra Qk is residually finite-dimensional. More precisely, there is another set Λ′ 6= ∅, a constant function r : Λ′ → N and a completely isometric ∗-homomorphism Γ : Qk → Mr. Proof. Let r ∈ N such that k(λ) ≤ r for every λ ∈ Λ. We note that if m ∈ N and m ≤ r, then there is a completely isometric ∗-homomorphism εm : Mm → Mr defined via εm(a) = a ⊕ 0r−m, a ∈ Mm. Next, given b = (bλ)λ + Jk ∈ Qk, it is easily verified that λ≥µ kbλk. sup kbk = inf µ∈Λ RFD and representations of amenable operator algebras 11 In particular, we see that kbk is a cluster point of {kbλk : λ ∈ Λ}. Thus, there exists a cofinal ultrafilter Fb on Λ for which kbk = limλ→Fb kbλk. Closed balls in Mr are compact, so that given (dλ)λ ∈ Lk the limit lim λ→Fb εk(λ)(dλ) exists in Mr. Note also that since Fb is cofinal, we have that lim λ→Fb εk(λ)(dλ) = 0 whenever (dλ)λ ∈ Jk. We may therefore define a map γb : Qk → Mr such that if d = (dλ)λ + Jk then γb(d) = lim λ→Fb εk(λ)(dλ). A routine verification establishes that γb is a ∗-homomorphism. For any b ∈ Qk we see that kγb(b)k = lim λ→Fb kεk(λ)(bλ)k = lim λ→Fb kbλk = kbk Finally, let Λ′ denote the unit sphere of Qk. Define r : Λ′ → N as r(b) = r for every by choice of the ultrafilter Fb. b ∈ Λ′. Then, the map defined by Γ : Qk → Mr Γ(d) = (γb(d))b∈Λ′ is an isometric ∗-homomorphism, and is thus completely isometric. (cid:3) 4. Residually finite-dimensional operator algebras with the total reduction property Motivated by Theorem 3.2, in this section we examine in detail the structure of subalgebras of Mk with the total reduction property, where Λ 6= ∅ is a set and k : Λ → N is a function. We establish one of our main results based partly on this detailed analysis. Our first goal is to show that if A ⊂ Mk has the total reduction property, then up to λ is surjective completely bounded isomorphism, we may assume that each component map qk on A. 4.1. Theorem. Let Λ 6= ∅ be a set and let k : Λ → N be a function. Suppose that A ⊂ Mk is a subalgebra with the total reduction property. Then, there exist a set Λ′, a function m : Λ′ → N and a subalgebra B ⊂ Mm which is completely boundedly isomorphic to A and such that for every α ∈ Λ′ we have that qm α (B) = Mm(α). Furthermore, B is weak-∗ closed if A is. Proof. For each λ ∈ Λ, let Aλ = qk λ(A). Note that Aλ ⊂ Mk(λ) so that Aλ is necessarily closed and hence has the total reduction property by Lemma 2.3. Moreover, we see from Lemma 2.3 that By Theorem 2.8, for each λ ∈ Λ there exists an invertible operator Xλ ∈ Mk(λ) such that XλAλX −1 λ ⊂ Mk(λ) is a C ∗-algebra and κAλ (1) ≤ κA(kqk λk) ≤ κA(1). kXλkkX −1 λ k ≤ ∆, 12 R. Clouatre and L.W. Marcoux where ∆ is a positive constant depending only on κA(1). Upon rescaling, we may assume that for each λ ∈ Λ. Then, the operator X =Lλ∈Λ Xλ ∈ Mk is bounded and invertible, and we have kXλk = kX −1 XAX −1 ⊂Mλ λ k ≤ ∆1/2 XλAλX −1 λ . For each λ, there exist a natural number rλ and non-negative integers d(λ, 0), d(λ, 1), . . . , d(λ, rλ) along with an orthogonal decomposition Ck(λ) = Cd(λ,0) ⊕ Cd(λ,1) ⊕ Cd(λ,2) ⊕ ··· ⊕ Cd(λ,rλ). With respect to this decomposition we must have XλAλX −1 λ ⊂ {0} ⊕ Md(λ,1) ⊕ Md(λ,2) ⊕ ··· ⊕ Md(λ,rλ) and for all 1 ≤ j ≤ rλ, where qd(λ,j) ◦ pλ(XλAλX −1 λ ) = M d(λ,j) pλ : {0} ⊕ M d(λ,1) ⊕ M d(λ,2) ⊕ ··· ⊕ M d(λ,rλ) → M d(λ,1) ⊕ M d(λ,2) ⊕ ··· ⊕ M d(λ,rλ) and qd(λ,j) : Md(λ,1) ⊕ Md(λ,2) ⊕ ··· ⊕ Md(λ,rλ) → Md(λ,j) denote the natural projections. Define p = ⊕λpλ which is completely isometric XAX −1 and weak-∗ homeomorphic on the weak-∗ closure of XAX −1. Put B = p(XAX −1). It remains to define the set Λ′ and the function m : Λ′ → N. For each λ ∈ Λ, we define Σλ = {(λ, j) : 1 ≤ j ≤ rλ, d(λ, j) 6= 0}. We put Λ′ = ∪λ∈ΛΣλ and m((λ, j)) = d(λ, j) for all (λ, j) ∈ Λ′. (cid:3) Next, we make an important observation: simple subalgebras of Mk with the total re- duction property are similar to finite-dimensional C ∗-algebras. 4.2. Corollary. Let Λ 6= ∅ be a set, let k : Λ → N be a function and let A ⊂ Mk be a simple subalgebra which has the total reduction property. Then, A is similar to a finite- dimensional C ∗-algebra. Proof. Invoking [11, Theorem 1.10], we see that it is sufficient to prove the statement for a completely boundedly isomorphic image of A. Hence, by virtue of Theorem 4.1 we may assume that qk λ(A) = Mk(λ) for every λ ∈ Λ. Fix λ0 ∈ Λ, and consider the surjective, contractive homomorphism qk λ0A : A → Mk(λ0). Since A is simple, we see that qk λ0A is invertible, and hence that A is boundedly isomorphic to the finite-dimensional C ∗-algebra Mk(λ0). We conclude that A must be similar to a ∗-isomorphic image of Mk(λ0) [24]. (cid:3) We mention in passing that Corollary 4.2 can be extended to cover the case where the algebra possesses finitely many ideals. We leave the details to the interested reader. Before proceeding with the main result of this section, we require two preliminary facts. We first establish a useful estimate. RFD and representations of amenable operator algebras 13 4.3. Lemma. Let Λ 6= ∅ be a set and let k : Λ → N be the constant function k(λ) = k, λ ∈ Λ for some fixed k ∈ N. Let (Xλ)λ∈Λ be a collection of invertible operators in Mk and let Suppose that A ⊂ Mk, that A has the total reduction property and that there exists λ0 ∈ Λ for which Xλ0 = I. Then A =(Mλ∈Λ λ T Xλ : T ∈ Mk) . X −1 sup λ kXλkkX −1 λ k ≤ 4κA(1)2. Proof. Upon rescaling if necessary, we may assume that kXλk = kX −1 λ k = kXλk1/2 kX −1 λ k1/2 for all λ ∈ Λ. For each ν ∈ Λ with ν 6= λ0, let Γν : A → M2k be defined by Γν Mλ∈Λ λ T Xλ! = T ⊕ X −1 X −1 ν T Xν for every T ∈ Mk. Observe that since Xλ0 = I, each such map Γν is a contractive homo- morphism. Therefore Γν(A) has the total reduction property by Lemma 2.3. Now consider the subspace Clearly, Wν is invariant for Γν(A), so there exists an idempotent Eν ∈ Γν(A)′ with kEνk ≤ κA(1) and whose range is Wν. Routine calculations show that Wν = {ξ ⊕ X −1 ν ξ : ξ ∈ Ck}. Γν(A)′ =(cid:26)(cid:20) αIk γX −1 ν βXν δIk(cid:21) : α, β, γ, δ ∈ C(cid:27) ⊂ M2k. Using the fact that the range of Eν is Wν, we infer that Eν =(cid:20) ανIk ανX −1 ν βνXν βνIk(cid:21) for an appropriate choice of αν, βν ∈ C. Moreover, since Eν is idempotent and dim Wν = k, we deduce that k = tr(Eν ). On the other hand, we have that tr(Eν ) = (αν + βν)k, so that αν + βν = 1. In particular whence max{αν,βν} ≥ 1/2 Thus kXνk ≤ 2kEνk ≤ 2κA(1). We conclude that kEνk ≥ max{kαν X −1 ν k,kβν Xνk} ≥ kXνk/2. kXνk kX −1 ν k = kXνk2 ≤ 4κA(1)2 for every ν ∈ Λ \ {λ0}. Since Xλ0 = Ik and κA(1) ≥ 1, we are done. (cid:3) We also need a property of central idempotents in weak-∗ closed algebras having the total reduction property. 14 R. Clouatre and L.W. Marcoux 4.4. Lemma. Let A ⊂ B(H) be a weak-∗ closed algebra with the total reduction property and with unit u ∈ A. Let (ei)i∈I be a family of central idempotents in A such that ∩i∈I (u − ei)A = {0}. Let S ⊂ B(H) denote the smallest weak-∗ closed subspace containing eiA for every i ∈ I. Then, A = S. Proof. We note that eiA ⊂ A for every i ∈ I, so that S ⊂ A since A is weak-∗ closed. It is readily checked that S is a weak-∗ closed two-sided ideal, and thus by Theorem 2.6 there is a central idempotent f ∈ A with fA = S. Note that ei = eiu ∈ eiA ⊂ S, so that eif = ei and ei(u − f ) = 0 for every i ∈ I . We claim that fA = A. Indeed, if a ∈ A then we see that for every i ∈ I, so that (u − f )a = (u − ei)(u − f )a (u − f )a ∈ ∩i∈I(u − ei)A = {0} whence f a = a. We conclude that A = fA = S. (cid:3) We can now prove one of the main results of the paper. 4.5. Theorem. Let Λ 6= ∅ be a set and let k : Λ → N be a function. Suppose that A ⊂ Mk is a subalgebra which has the total reduction property. If A is weak-∗ closed, then A is similar to a C ∗-algebra. Proof. We start by noting once again that it is sufficient to prove the statement for a com- pletely boundedly isomorphic image of A. Thus, by virtue of Theorem 4.1 we may assume that A is a weak-∗ closed subalgebra of Mk such that qk λ(A) = Mk(λ) for every λ ∈ Λ. For each λ ∈ Λ, consider the weak-∗ closed ideal Jλ = ker qk λ. Then A/Jλ ≃ qk λ(A) = Mk(λ). A = Jλ + (u − fλ)A A = Jλ + Kλ. Moreover, by Theorem 2.6 there exists a central idempotent fλ ∈ Jλ ∩ A′ with Jλ = fλA, kfλk ≤ κA(1) and such that we have a topological direct sum decomposition where u denotes the unit of A and satisfies kuk ≤ κA(1) (see the discussion following Theorem 2.6). Set now eλ = (u − fλ) for each λ ∈ Λ, and put Kλ = eλA so that By Lemma 2.3 we see that Kλ has the total reduction property and that κKλ(1) ≤ κA(keλk) ≤ κA(2κA(1)). Moreover, we note that Kλ ≃ A/Jλ ≃ Mk(λ) and consequently Kλ is simple and a minimal ideal of A. Hence, for each α ∈ Λ we must have that qk α(Kλ) is a two-sided ideal of Mk(α), and therefore is either {0} or Mk(α). Equivalently, α(eλ) = 0 or qk qk For each λ ∈ Λ let α(eλ) = I. Note that for every λ ∈ Λ we have qk λ(A) = Mk(λ). Recall that ∆λ = {α ∈ Λ : qk α(eλ) = I}. λ(u) = I since qk fλA = Jλ = ker qk λ RFD and representations of amenable operator algebras 15 thus λ(eλ) so that λ ∈ ∆λ for every λ ∈ Λ. In particular, we see that λ(fλ) = I − qk 0 = qk Λ = ∪λ∈Λ∆λ. If λ1, λ2 are distinct elements of Λ, then by minimality we must have either Kλ1 = Kλ2 or Kλ1 ∩ Kλ2 = {0}. Note also that Kλ1 ∩ Kλ2 = {0} if and only if eλ1 eλ2 = 0, and Kλ1 = Kλ2 if and only if eλ1 = eλ2. We conclude that if λ1, λ2 ∈ Λ, then either ∆λ1 = ∆λ2, or ∆λ1 ∩ ∆λ2 = ∅. The equivalence relation partitions Λ into a set Ω of disjoint subsets. For each ω ∈ Ω, choose λω ∈ ω. Then, we have the following disjoint union λ1 ∼ λ2 if and only if eλ1 = eλ2 ∪ω∈Ω∆λω = Λ. = 0. Let now a ∈ A and assume that eλω a = 0 for eλω2 If ω1, ω2 ∈ Ω are distinct, then eλω1 every ω ∈ Ω. Then, we see that qk α(a) = 0 for every α ∈ ∪ω∈Ω∆λω = Λ, so that a = 0. By Lemma 4.4, we conclude that A is the smallest weak-∗ closed subspace containing eλωA for = ∅. Hence, every ω ∈ Ω. On the other hand, if ω1, ω2 ∈ Ω are distinct, then ∆λω1 ∩ ∆λω2 we find that A =Mω∈Ω eλωA =Mω∈Ω Kλω . Now, for each λ ∈ Λ we have that qk α(Kλ) = Mk(α) for all α ∈ ∆λ. Since Kλ is simple, we see that qk αKλ is a bounded isomorphism between Kλ and Mk(α) for every α ∈ ∆λ. In particular, k is constant on ∆λ. In addition, if we let µλ be a fixed element of ∆λ, then we see that for each α ∈ ∆λ there is an invertible operator Xλ,α ∈ Mk(α) such that for every a ∈ Kλ. Since Xλ,αqk µλ(a)X −1 λ,α = qk α(a) α(a) : a ∈ Kλ) , qk Kλ =(Mα∈Λ {0} ⊕Mα∈∆λ Xλ,αbX −1 λ,α : b ∈ Mk(µλ) . we see that up to a (unitary) reordering of the components Kλ = Mα∈Λ\∆λ kX −1 By Lemmas 2.3 and 4.3, we find that λ,αk kXλ,αk ≤ 4κKλ(1)2 ≤ 4κA(2κA(1))2 for every α ∈ ∆λ. Upon rescaling we may assume that kXλ,αk = kX −1 λ,αk ≤ 2κA(2κA(1)). for every α ∈ ∆λ. Hence, the operator X =Mω∈Ω  Mα∈Λ\∆λω I ⊕ Mα∈∆λω Xλω,α 16 R. Clouatre and L.W. Marcoux is invertible with bounded inverse. Finally, we find X −1AX =X −1 Mω∈Ω =Mω∈Ω ≃Mω∈Ω Kλω! X {0} ⊕ Mα∈∆λω  Mα∈Λ\∆λω Mk(µλω ) b : b ∈ Mk(µλω )  (cid:3) which is a C ∗-algebra. The reader will notice that the reason we require the algebra above to be weak-∗ closed is to avail ourselves of Theorem 2.6. We now record a straightforward consequence of the previous result. 4.6. Corollary. Let Λ 6= ∅ be a set and let k : Λ → N be a function. Suppose that A ⊂ Mk is a subalgebra which has the total reduction property. If A is weak-∗ closed, then it has the SP property . Proof. We know that A is similar to a C ∗-algebra by Theorem 4.5. Next, recall that C ∗- algebras with the total reduction property have Kadison's similarity property by Theorem 2.7, so that A must have the SP property. (cid:3) Finally, we close this section with an application to triangular algebras (see Example 3.3 for the definition). 4.7. Corollary. Let A ⊂ B(H) be a weak-∗ closed triangular operator algebra with the total reduction property. Then, A is similar to a C ∗-algebra. Proof. By Example 3.3, we know that there is a completely isometric weak-∗ homeomorphic homomorphism Φ : A → Lk, for some net k : Λ → N. Thus, by Lemma 2.3 and Theorem 4.5 we see there is an invertible operator X such that XΦ(A)X −1 is a C ∗-algebra. The map XΦ(a)X −1 7→ a, a ∈ A is completely bounded on a C ∗-algebra, and thus is similar to a ∗-homomorphism [11]. Consequently, A is similar to the image of a C ∗-algebra under a ∗-homomorphism, and so is similar to a C ∗-algebra. (cid:3) 5. Representations with residually finite-dimensional range In the previous section, we proved that a weak-∗ closed residually finite-dimensional op- erator algebra with the total reduction property has the SP property: all its representations are automatically completely bounded. In other words, we restricted our attention to the case where the domains of the representations are contained in a product of matrix al- gebras. In this section, we shift our focus to the range and assume that it is residually finite-dimensional, while the domain is allowed to be an arbitrary operator algebra with the total reduction property. The driving force behind our efforts in this section is the following simple observation, which we record for ease of reference. RFD and representations of amenable operator algebras 17 5.1. Lemma. Let A and B be operator algebras. Assume that B is residually finite- dimensional and let (λ)λ be the corresponding family of completely contractive represen- tations of B such that Lλ λ is completely isometric. Let θ : A → B be a representation. Then θ is completely bounded if and only if supλ kλ ◦ θkcb < ∞. Although completely elementary, this lemma shows that to determine whether a repre- sentation with residually finite-dimensional range is completely bounded, we may restrict our attention to finite-dimensional representations. In particular, we have the following consequence which the reader may want to compare with Theorem 3.2. 5.2. Theorem. Let A be an operator algebra and let k : Λ → N be a bounded net. If θ : A → Qk is a bounded representation, then θ is completely bounded. Proof. By Proposition 3.5 we can find another set Λ′ 6= ∅ and a constant function r : Λ′ → N for which there exists a completely isometric ∗-homomorphism Γ : Qk → Mr. It suffices to show that Γ◦ θ is completely bounded. To see this, use Lemma 5.1 to conclude that θ is completely bounded if and only if λ kqk sup λ ◦ θkcb < ∞. Now, a classical result of R.R. Smith [21] shows that for any λ ∈ Λ we have kqk λ ◦ θkcb ≤ k(λ)kqk λ ◦ θk ≤(cid:18)sup λ∈Λ k(λ)(cid:19) kθk. The proof is complete. (cid:3) The remainder of this section is devoted to establishing another one of our main results dealing with representations of operator algebras with the total reduction property whose range are residually finite-dimensional. The key technical tool is the following observation, which generalizes the fact that finite-dimensional C ∗-algebras can be faithfully represented on finite-dimensional Hilbert spaces. We suspect it is well-known. The idea is reminiscent of that found in the proof of [18, Theorem 6.3], which apparently has its origins in the work of Dixon. We also note that the classical Blecher-Ruan-Sinclair theorem [2] may not produce a finite-dimensional Hilbert space and thus does not meet our specific needs. 5.3. Proposition. Let A be a finite-dimensional operator algebra, and fix ε > 0 and d ∈ N. Then, there exist a finite-dimensional Hilbert space H and a completely contractive homomorphism ϕ : A → B(H) with kϕ(n)(A)kMn(B(H)) ≥ (1 − ε)kAkMn(A) for every A ∈ Mn(A) and every 1 ≤ n ≤ d. Proof. By considering the unitization of A (which is also finite-dimensional) if necessary, we may assume that A is unital [1]. Since A is finite-dimensional, its closed unit ball is compact. Hence, we may choose α1, . . . , αN ∈ A such that kαkk ≤ 1 for every 1 ≤ k ≤ N , and with the property that for every a ∈ A with kak ≤ 1 there is 1 ≤ k ≤ N such that ka − αkk < ε/d. 18 R. Clouatre and L.W. Marcoux In particular, given A ∈ Mn(A) with kAkMn(A) = 1 where 1 ≤ n ≤ d, we can find A′ = (a′ ij)ij ∈ Mn(A) with the property that each a′ ij belongs to the set {α1, . . . , αN} and kA − A′kMn(A) < ε. Next, note that there is a finite-dimensional subspace H1 with the property that kP (n) H1 A∗AP (n) H1 k ≥ (1 − ε)kAk2 Mn(A) whenever A = (aij)ij ∈ Mn(A) for some 1 ≤ n ≤ d and each aij belongs to the set {α1, . . . , αN}. Define a unital completely positive map ω : C ∗(A) → B(H1) as ω(t) = PH1tH1 for every t ∈ C ∗(A). Next, we carefully analyze the Stinespring dilation of ω, and for that purpose we briefly recall the details of its construction. Define a positive semi-definite bilinear form on the vector space C ∗(A) ⊗ H1 as ΨXi sj ⊗ kj =Xi,j ti ⊗ hi,Xj Z = {v ∈ C ∗(A) ⊗ H1 : Ψ(v, v) = 0}, hω(s∗ j ti)hi, kjiH1. which is a subspace of C ∗(A)⊗H1. The quotient (C ∗(A)⊗H1)/Z is an inner product space, and we denote its completion by E. Given v ∈ C ∗(A) ⊗ H1 we denote its image in E by [v]. Define Let via π : C ∗(A) → B(E) π(t)Xj sj ⊗ hj =Xj tsj ⊗ hj for every t ∈ C ∗(A). The complete positivity of ω implies that π is well-defined. Further- more, it is readily verified that π is a unital ∗-homomorphism. Now, here is the key point: we denote by H the finite-dimensional space spanned by the elements of the form [a⊗ h] for a ∈ A and h ∈ H1. Since A is an algebra, the space H is invariant for π(A). Hence, we may define a unital completely contractive homomorphism ϕ : A → B(H) as ϕ(a) = π(a)H for every a ∈ A. It only remains to establish the announced lower bound. Notice that for h, k ∈ H1 we have Ψ(1 ⊗ h, 1 ⊗ k) = hω(1)h, kiH1 = hh, kiH1 and thus k[1 ⊗ h]kH = khkH1 for every h ∈ H1. In particular, if ξ = (ξ1, . . . , ξn)t ∈ H(n) satisfies kξkH(n) = 1 then the vector 1 1 Ξ = (1 ⊗ ξ1, . . . , 1 ⊗ ξn)t ∈ H(n) also satisfies kΞkH(n) = 1. Furthermore, if A ∈ Mn(A) then a routine calculation shows that hω(n)(A∗A)ξ, ξiH(n) 1 = kπ(n)(A)Ξk2 E (n) = kϕ(n)(A)Ξk2 H(n). Hence, hω(n)(A∗A)ξ, ξiH(n) 1 ≤ kϕ(n)(A)k2 Mn(B(H)) RFD and representations of amenable operator algebras 19 and we conclude that kϕ(n)(A)k2 Mn(B(H)) ≥ kω(n)(A∗A)kMn(B(H1)) = kP (n) H1 A∗AP (n) H1 k ≥ (1 − ε)kAk2 Mn(A) whenever A = (aij)ij ∈ Mn(A) where 1 ≤ n ≤ d and each aij belongs to the set {α1, . . . , αN}. As noted above, given a general element A ∈ Mn(A) with 1 ≤ n ≤ d and kAkMn(A) = 1, we can find A′ = (a′ ij belongs to the set {α1, . . . , αN} and ij)ij ∈ Mn(A) where each a′ kA − A′kMn(A) < ε. In particular, kA′kMn(A) ≥ 1 − ε. Since ϕ is completely contractive, we find kϕ(n)(A)kMn(B(H)) ≥ kϕ(n)(A′)kMn(B(H)) − ε ≥ (1 − ε)3/2 − ε and the proof is complete. (cid:3) This theorem allows us to prove a uniform estimate for certain representations of operator algebras with the total reduction property. 5.4. Theorem. Let A be an operator algebra with the total reduction property and let θ : A → B(H) be a representation. Assume that A/ ker θ is finite-dimensional. Then, there is a positive constant ∆ depending only on κA and kθk such that kθkcb ≤ ∆. Proof. Let bθ : A/ ker θ → B(H) be the representation induced by θ. Then, we have kθkcb = kbθkcb. By Proposition 5.3, there is a finite-dimensional Hilbert space H′, an operator algebra B ⊂ B(H′) and completely contractive isomorphism ϕ : A/ ker θ → B with kϕ−1k ≤ 2. Therefore, B has the total reduction property and κB(t) ≤ κA/ ker θ(kϕkt) ≤ κA(t) for every t ≥ 0, by Lemma 2.3. Then, an application of Theorem 2.8 yields an invertible operator X ∈ B(H′) such that XBX −1 is a C ∗-algebra and kXkkX −1k ≤ ∆0 for some positive constant ∆0 depending only on κA(1). We will use the following notation We see that AdX(T ) = XT X −1, T ∈ B(H′). X kcb ≤ ∆0. Then, XBX −1 = AdX (B) has the total reduction property and kAdXkcb ≤ ∆0, kAd−1 κXBX −1 (t) ≤ κB(kAdXkt) ≤ κA(∆0t) for every t ≥ 0, again by Lemma 2.3. Now, by virtue of Theorem 2.7, we see that the map bθ ◦ ϕ−1 ◦ Ad−1 X : XBX −1 → B(H) 20 R. Clouatre and L.W. Marcoux is completely bounded with kbθ ◦ ϕ−1 ◦ Ad−1 X kcb ≤ 128κXBX −1 (kbθ ◦ ϕ−1 ◦ Ad−1 X k)2 ≤ 128κA(∆0kbθ ◦ ϕ−1 ◦ Ad−1 X k)2 X kkθk)2 ≤ 128κA(∆0kϕ−1kkAd−1 ≤ 128κA(2∆2 bθ =bθ ◦ ϕ−1 ◦ Ad−1 0kθk)2 ≤ 128∆0κA(2∆2 kbθkcb ≤ kAdXkcbkϕkcbkbθ ◦ ϕ−1 ◦ Ad−1 X ◦ AdX ◦ ϕ. 0kθk)2. X kcb Finally, note that so that so we may take ∆ = 128∆0κA(2∆2 0kθk)2. (cid:3) We can now establish the main result of this section, which says that for operator algebras with the total reduction property, representations with residually finite-dimensional ranges are necessarily completely bounded. 5.5. Corollary. Let Λ 6= ∅ be a set and let k : Λ → N be a function. Let A be an operator algebra with the total reduction property and let θ : A → Mk be a representation. Then, θ is completely bounded. Proof. Combine Lemma 5.1 and Theorem 5.4. (cid:3) We give an application to triangular algebras (see Example 3.3 for the definition). 5.6. Corollary. Let A be an operator algebra with the total reduction property and let θ : A → B(H) be a representation such that θ(A) is a triangular algebra. Then, θ is completely bounded. Proof. By Example 3.3, we know that there is a completely isometric homomorphism Φ : θ(A) → Lk, for some net k : Λ → N. Thus, Φ ◦ θ is completely bounded by virtue of Corollary 5.5. Since Φ is completely isometric, we conclude that θ is completely bounded as well. (cid:3) In closing, we exhibit an example showing that the total reduction property cannot simply be removed from the assumptions of Corollary 5.5. 5.7. Example. For each n ∈ N, let tn : Mn → Mn denote the transpose map. Then, it is well-known that ktnk = 1 and Next, let An ⊂ M2n denote the unital subalgebra consisting of elements of the form 0 kt(n) n k = n. λIn(cid:21) (cid:20)λIn A λIn (cid:21) , tn(A) where λ ∈ C, A ∈ Mn. The map θn : An → M2n defined as θn(cid:18)(cid:20)λIn A λIn(cid:21)(cid:19) =(cid:20)λIn 0 0 λ ∈ C, A ∈ M2n RFD and representations of amenable operator algebras 21 is easily verified to be a unital homomorphism with the property that kθnk = 1 and n k ≥ kt(n) Consequently, we see that the map Θ :Qn An →Qn Θ(an)n = (θn(an))n, kθ(n) n k = n. M2n defined as (an)n ∈Yn An is a unital bounded representation which is not completely bounded. Finally, fix m ∈ N and note that the subspace Cm ⊕ {0} is invariant for Am yet it does not admit an invariant topological complement as a straightforward calculation establishes. Therefore, Am does not have the total reduction property. Since Am is a closed two-sided ideal ofQn An, we conclude from Theorem 2.4 thatQn An does not have the total reduction property either. (cid:3) References [1] David P. Blecher and Christian Le Merdy. Operator algebras and their modules -- an operator space approach, volume 30 of London Mathematical Society Monographs. New Series. The Clarendon Press, Oxford University Press, Oxford, 2004. Oxford Science Publications. [2] D.P. Blecher, Z.J. Ruan, and A.M. Sinclair. A characterization of operator algebras. J. Functional Analysis, 89:188 -- 201, 1990. [3] N.P Brown and N. Ozawa. C ∗-algebras and finite-dimensional approximations, volume 88 of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI, 2008. [4] Y. Choi. On commutative, operator amenable subalgebras of finite von Neumann algebras. J. Reine Angew. Math., 678:201 -- 222, 2013. [5] Y. Choi, I. Farah, and N. Ozawa. A nonseparable amenable operator algebra which is not isomorphic to a C ∗-algebra. Forum Math. Sigma, 2:12pp, 2014. [6] E. Christensen. On non self-adjoint representations of C ∗-algebras. Amer. J. Math., 103:817 -- 833, 1981. [7] R. Clouatre and L.W. Marcoux. Compact ideals and rigidity of representations for amenable operator algebras. preprint, 2016. [8] A. Connes. On the cohomology of operator algebras. J. Funct. Anal., 28:248 -- 253, 1978. [9] J.A. Gifford. Operator algebras with a reduction property. PhD thesis, Australian National University, 1997, arXiv:1311.3822. [10] U. Haagerup. All nuclear C ∗-algebras are amenable. J. Funct. Anal., 74:305 -- 319, 1983. [11] U. Haagerup. Solution of the similarity problem for cyclic representations of C ∗-algebras. Annals of Math., 118:215 -- 240, 1983. [12] B.E. Johnson. Cohomology in Banach algebras, volume 127. Mem. Amer. Math. Soc., Providence, Rhode Island, 1972. [13] R.V. Kadison. On the orthogonalization of operator representations. Amer. J. Math., 77:600 -- 620, 1955. [14] L.W. Marcoux and A.I. Popov. Abelian, amenable operator algebras are similar to C ∗-algebras. Duke J. Math., 165:2391 -- 2406, 2016. [15] V. Paulsen. Completely Bounded Maps and Operator Algebras, volume 78 of Cambridge Studies in Ad- vanced Mathematics. Cambridge University Press, Cambridge, 2002. [16] G. Pisier. A polynomially bounded operator on Hilbert space which is not similar to a contraction. J. Amer. Math. Soc., 10(2):351 -- 369, 1997. [17] G. Pisier. Similarity problems and length. International Conference on Mathematical Analysis and its Applications (Kaohsiung, 2000). Taiwanese Math. J., 5:1 -- 17, 2001. [18] Gilles Pisier. Introduction to operator space theory, volume 294 of London Mathematical Society Lecture Note Series. Cambridge University Press, Cambridge, 2003. [19] Volker Runde. Lectures on amenability, volume 1774 of Lecture Notes in Mathematics. Springer-Verlag, Berlin, 2002. [20] M.V. Seınberg. A characterization of the algebra C(ω) in terms of cohomology groups (Russian). Uspehi Mat. Nauk, 32:203 -- 204, 1977. [21] R.R. Smith. Completely bounded maps between C ∗-algebras. J. London Math. Soc., 27:157 -- 166, 1983. [22] A. Szankowski. B(H) does not have the approximation property. Acta Math., 147(1-2):89 -- 108, 1981. [23] D. Voiculescu. A note on quasi-diagonal C ∗-algebras and homotopy. Duke Math J., 62:267 -- 271, 1991. [24] S. Wright. Similarity orbits of approximately finite C ∗-algebras. Pacific J. Math., 87:223 -- 231, 1980. 22 R. Clouatre and L.W. Marcoux Department of Mathematics, University of Manitoba, Winnipeg, Manitoba, Canada R3T 2N2 E-mail address: [email protected] Department of Pure Mathematics, University of Waterloo, Waterloo, Ontario, Canada N2L 3G1 E-mail address: [email protected]
1711.04943	2	1711	2018-05-28T21:51:19	W$^*$-superrigidity for coinduced actions	[ "math.OA", "math.DS", "math.FA" ]	We prove W$^$-superrigidity for a large class of coinduced actions. We prove that if $\Sigma$ is an amenable almost-malnormal subgroup of an infinite conjugagy class (icc) property (T) countable group $\Gamma$, the coinduced action $\Gamma\curvearrowright X$ from an arbitrary probability measure preserving action $\Sigma\curvearrowright X_0$ is W$^$-superrigid. We also prove a similar statement if $\Gamma$ is an icc non-amenable group which is measure equivalent to a product of two infinite groups. In particular, we obtain that any Bernoulli action of such a group $\Gamma$ is W$^*$-superrigid.	math.OA	math	W∗-SUPERRIGIDITY FOR COINDUCED ACTIONS DANIEL DRIMBE Abstract. We prove W∗-superrigidity for a large class of coinduced actions. We prove that if Σ is an amenable almost-malnormal subgroup of an infinite conjugagy class (icc) property (T) countable group Γ, the coinduced action Γ y X from an arbitrary probability measure preserving action Σ y X0 is W∗-superrigid. We also prove a similar statement if Γ is an icc non-amenable group which is measure equivalent to a product of two infinite groups. In particular, we obtain that any Bernoulli action of such a group Γ is W∗-superrigid. 8 1 0 2 y a M 8 2 ] . A O h t a m [ 2 v 3 4 9 4 0 . 1 1 7 1 : v i X r a 1. Introduction and statement of the main results 1.1. Introduction. To every measure preserving action Γ y (X, µ) of a countable group Γ on a standard probability space (X, µ), one associates the group measure space von Neumann algebra L∞(X) ⋊ Γ [MvN36]. If the action Γ y X is free, ergodic and probability measure preserving (pmp), then L∞(X) ⋊ Γ is a II1 factor which contains L∞(X) as a Cartan subalgebra, i.e. a maximal abelian von Neumann algebra whose normalizer generates L∞(X) ⋊ Γ. The classification of group measure space II1 factors L∞(X) ⋊ Γ is a central problem in the theory of von Neumann algebras. Two free ergodic pmp actions Γ y (X, µ) and Λ y (Y, ν) on standard probability spaces (X, µ) and (Y, ν) are said to be W∗-equivalent if L∞(X) ⋊ Γ is isomorphic to L∞(Y ) ⋊ Λ. If the groups are amenable, the classification up to W∗-equivalency has been completed in the 1970s. More precisely, the celebrated theorem of Connes [Co76] asserts that all II1 factors arising from free ergodic pmp actions of countable amenable groups are isomorphic to the hyperfinite II1 factor. In contrast, the non-amenable case is much more challenging and it has led to a beautiful rigidity theory in the sense that one can deduce conjugacy from W∗-equivalence. A major breakthrough in the classification of II1 factors was made by Popa between 2001-2004 through the invention of deformation/rigidity theory (see [Po07, Va10a, Io12a] for surveys). In particular, he obtained the following W∗-rigidity result: let Γ y X be a free ergodic pmp action of an infinite conjugacy class (icc) countable group Γ which has an infinite normal subgroup with the relative property (T) and let Λ y Y := Y Λ 0 be a Bernoulli action of a countable group Λ. Popa proved that if the two actions are W∗-equivalent, then the actions are conjugate [Po03,Po04], i.e. there exist a group isomoprhism d : Γ → Λ and a measure space isomorphism θ : X → Y such that θ(gx) = d(g)θ(x) for all g ∈ Γ and almost everywhere (a.e.) x ∈ X. The most extreme form of rigidity for an action Γ y (X, µ) is W∗-superrigidity, i.e. whenever Λ y (Y, ν) is a free ergodic pmp action W∗-equivalent to Γ y (X, µ), then the two actions are conjugate. A few years ago, Peterson was able to show the existence of virtually W∗-superrigid actions [Pe09]. Soon after, Popa and Vaes discovered the first concrete families of W∗-superrigid actions [PV09]. Ioana then proved in [Io10] a general W∗-superrigidity result for Bernoulli actions. Theorem (Ioana, [Io10]). If Γ is an icc property (T) group and (X0, µ0) is a non-trivial standard probability space, then the Bernoulli action Γ y (X0, µ0)Γ is W∗-superrigid. The author was partially supported by NSF Career Grant DMS #1253402. 1 2 DANIEL DRIMBE The main ingredient of his proof was the discovery of a beautiful dichotomy result for abelian subalgebras of II1 factors coming from Bernoulli actions. Using a similar method, Ioana, Popa and Vaes were able to prove later that any Bernoulli action of an icc non-amenable group which is a product of two infinite groups is also W∗-superrigid [IPV10]. A few years ago Boutonnet extended these results to Gaussian actions in [Bo12]. Several other classes of W∗-superrigid actions have been found in [FV10, CP10, HPV10, Va10b, CS11, CSU11, PV11, PV12, CIK13, CK15, Dr15, GITD16]. 1.2. Statement of the main results. Our first theorem is a generalization of Ioana's W∗- superrigidity result [Io10, Theorem A] to coinduced actions. Before stating the theorem, we explain first the terminology that we use starting with the notion of coinduced actions (see e.g. [Io06b]). Definition 1.1. Let Γ be a countable group and let Σ be a subgroup. Let φ : Γ/Σ → Γ be a section. Define the cocycle c : Γ × Γ/Σ → Σ by the formula c(g, i) = φ−1(gi)gφ(i), for all g ∈ Γ and i ∈ Γ/Σ. Let Σ define an action Γ σ y X Γ/Σ 0 σ0 y (X0, µ0) be a pmp action, where (X0, µ0) is a non-trivial standard probability space. We , called the coinduced action of σ0, as follows: σg((xi)i∈Γ/Σ) = (x′ i)i∈Γ/Σ, where x′ i = c(g−1, i)−1xg−1i. Note the following remarks: • σ is a pmp action of Γ on the standard probability space X Γ/Σ • if we consider the trivial action of Λ = {e} on X0, then the coinduced action of Γ on 0 . X Γ/{e} 0 = X Γ 0 is the Bernoulli action. Recall that an inclusion Γ0 ⊂ Γ of countable groups has the relative property (T) if for every ǫ > 0, there exist δ > 0 and a finite subset F ⊂ Γ such that if π : Γ → U (K) is a unitary representation and ξ ∈ K is a unit vector satisfying kπ(g)ξ − ξk < δ, for all g ∈ F , then there exists ξ0 ∈ K such that kξ − ξ0k < ǫ and π(h)ξ0 = ξ0, for all h ∈ Γ0. The group Γ has the property (T) if the inclusion Γ ⊂ Γ has the relative property (T). To give some examples, note that Z2 ⊂ Z2 ⋊ SL2(Z) has the relative property (T) and SLn(Z), n ≥ 3, has the property (T) [Ka67, Ma82]. Finally, we say that a subgroup Σ of a countable group Γ is called n-almost malnormal if for any g1, g2, ..., gn ∈ Γ such that g−1 if finite. The subgroup Σ i is called almost malnormal if it is n-almost malnormal for some n ≥ 1. i gj /∈ Σ for all i 6= j, the group ∩n i=1giΣg−1 Theorem A. Let Γ be an icc group which admits an infinite normal subgroup Γ0 with relative property (T) and let Σ be an amenable almost malnormal subgroup of Γ. Let σ0 be a pmp action of Σ on a non-trivial standard probability space (X0, µ0) and denote by σ the coinduced action of Γ on X := X Γ/Σ σ y X is W ∗-superrigid. . Then Γ 0 Example 1.2. In particular, Theorem A can be applied for Γ = SL3(Z) and Σ = hAi, where [PV06, Section 7]. See [PV06] for more concrete examples of amenable almost A =   1 0 1 0 −1 0 0 −1 0   malnormal subgroups of P SLn(Z), n ≥ 3. See also [RS10, Theorem 1.1], a result which proves the existence of amenable almost malnormal subgroups of torsion-free uniform lattices in connected semisimple real algebraic groups with no compact factors. W∗-SUPERRIGIDITY FOR COINDUCED ACTIONS 3 We now generalize Ioana-Popa-Vaes' result [IPV10, Theorem 10.1] to coinduced actions. First, recall that two countable groups Γ and Λ are called measure equivalent in sense of Gromov if there exist two commuting free measure preserving actions of Γ and Λ on a standard measure space (Ω, m), such that the actions of Γ and Λ each admit a finite measure fundamental domain [Gr91]. Natural examples of measure equivalent groups are provided by pairs of lattices Γ, Λ in an unimodular locally compact second countable group. Theorem B. Let Γ be an icc non-amenable group which is measure equivalent to a product of two infinite groups. Let Σ be an amenable almost malnormal subgroup and let σ0 be a pmp action of Σ on a non-trivial standard probability space (X0, µ0) and denote by σ the coinduced action of Γ on X := X Γ/Σ 0 σ y X is W ∗-superrigid. Then Γ . See Theorem 6.3 for a more general statement in which it is assumed instead that Γ is measure equivalent to a group Λ0 whose group von Neumann algebra L(Λ0) is not prime. Note that Theorems A and B provide a complementary class of W∗-superrigid coinduced actions from the one found in [Dr15, Corollary 1.4]. Example 1.3. A more general statement of Theorem B can be appplied for Σ ⊂ Γ = ∆ ≀ Σ with ∆ non-amenable and Σ amenable (see Remark 6.4). The following remark shows that if Σ is not almost malnormal, the action Γ y X is not necessary W∗-superrigid. To put this in context, we recall first the notion of OE-superrigidity and Singer's result [Si55]. Two actions Γ y X and Λ y Y are orbit equivalent (OE) if there exists a measure space isomorphism θ : X → Y such that θ(Γx) = Λθ(x), for a.e. x ∈ X. A pmp action Γ y X is OE-superrigid if whenever Λ y Y is a free ergodic pmp action which is OE to Γ y X, then the two actions are conjugate. Singer proved in [Si55] that two free ergodic pmp actions Γ y X and Λ y Y are OE if and only if there exists an isomoprhism of the group measure space algebras L∞(X) ⋊ Γ and L∞(Y ) ⋊ Λ which preserves the Cartan algebras L∞(X) and L∞(Y ). In particular, W∗-superrigidity implies OE-superrigidity. Remark 1.4. If Σ is not almost malnormal, the action Γ y X may fail to be W∗-superrigid. Indeed, suppose Γ is an icc group which splits as a direct product Γ = Σ × ∆, with Σ amenable and ∆ a non-amenable group. Connes and Jones have found in [CJ82] a class of groups Σ and a class of σ0 y X0 for which the coinduced action Γ y X of σ0 is not W∗-superrigid. free ergodic pmp actions Σ Precisely, they have proven that M := L∞(X) ⋊ Γ is McDuff, i.e. M ≃ M ¯⊗R, where R is the hyperfinite II1 factor. However, [Dr15, Corollary 1.3] implies that Γ y X is OE-superrigid. Note that Theorem B extends the class of groups whose Bernoulli actions are W∗-superrigid. There- fore we record the following result. Corollary C. Let Γ be an icc non-amenable group which is measure equivalent to a product of two infinite groups. Let (X0, µ0) be a non-trivial standard probability space. Then the Bernoulli action Γ y X Γ 0 is W ∗-superrigid. We recall the well known theorem due to Borel which asserts that every connected non-compact semisimple Lie group contains a lattice (see [Bo63] and [Ra72, Theorem 14.1]). Using this, we obtain an immediate consequence of Corollary C. Corollary D. Let Γ be an icc lattice in a product G = G1 × · · · × Gn of n ≥ 2 connected non- compact semisimple Lie groups and let (X0, µ0) be a non-trivial standard probability space. Then the Bernoulli action Γ y X Γ 0 is W ∗-superrigid. 4 DANIEL DRIMBE Note that a combination of Popa's cocycle superrigidity theorem for product groups [Po06] and the results on uniqueness of Cartan subalgebras from [PV12] already proves Corollary D, but only in the case when each factor G1, . . . , Gn is of rank one. 1.3. Comments on the proof of Theorem B. For obtaining the proofs of Theorem A and Theorem B, we adapt the proofs used by Ioana [Io10] and Ioana-Popa-Vaes [IPV10] to the context of coinduced actions. We outline briefly and informally the proof of Theorem B since it has as a consequence Corollary C. To this end, let Γ be an icc group and let Σ be an almost malnormal subgroup. Assume Γ is measure equivalent to a product Λ0 = Λ1 × Λ2 of two countable groups. By [Fu99], Γ and Λ0 must have stably orbit equivalent actions. To simplify notation, assume there exist free ergodic pmp actions of Γ and Λ0 on a probability space (Y0, µ) whose orbits are equal, almost everywhere. Thus, L∞(Y0) ⋊ Γ = L∞(Y0) ⋊ Λ0. Suppose Σ y X0 is a pmp action on a non-trivial standard probability space and let Γ X Γ/Σ Assume that Λ y Y is an arbitrary free ergodic pmp action such that be the corresponding coinduced action. Our goal is to show that Γ 0 σ y X := σ y X is W∗-superrigid. M := L∞(X) ⋊ Γ = L∞(Y ) ⋊ Λ. First, we reduce the problem to showing that the Cartan subalgebras L∞(X) and L∞(Y ) are unitarily conjugated. We do this by proving in Section 4 a cocycle superrigidity theorem for Γ y X. Combined with [Po05, Theorem 5.6], we obtain that Γ y X is OE-superrigid. Therefore, by a result of Singer [Si55] it is enough to show that L∞(Y ) is unitarily cojugate to L∞(X) in M . We note that this is actually equivalent to L∞(Y ) ≺M L∞(X), by [Po06b, Theorem A.1]. See Section 2.2 for the definition of Popa's intertwining symbol "≺". As is [Io10], we make use of the decomposition M = L∞(Y ) ⋊ Λ via the comultiplication ∆ : M → M ¯⊗M defined by ∆(bvλ) = bvλ ⊗ λ , for all b ∈ L∞(Y ) and λ ∈ Λ, introduced in [PV09]. Here we denote by {vλ}λ∈Λ the canonical unitaries implementing the action of Λ on L∞(Y ). The next step is to prove that there exists a unitary u ∈ M ¯⊗M such that (1.1) u∆(L(Γ))u∗ ⊂ L(Γ) ¯⊗L(Γ). This is obtained in two steps. A main technical contribution of our paper is to use the rigidity of Γ inherited from the product structure of Λ0 through measure equivalence. We do this in Section 4 by introducing an "amplified" version of the comultiplication map ∆ which is defined on the larger von Neumann algebra (L∞(Y0) ¯⊗L∞(X)) ⋊ Γ. Combined with the spectral gap rigidity theorem for coinduced actions (Theorem 3.1) proved in Section 3, we obtain the conclusion (1.1). In Section 5, following Ioana's idea [Io10], we obtain a dichotomy theorem for certain abelian algebras. The result is a straightforward adaptation of [IPV10, Theorem 5.1] to coinduced actions and has two consequences. First, we obtain ∆(L∞(X))′ ∩ (M ¯⊗M ) ≺ L∞(X) ¯⊗L∞(X). Second, it implies a weaker version of Popa's conjugacy criterion adapted to coinduced actions. This will altogether prove Theorem B. 1.4. Acknowledgements. I am very grateful to my advisor Adrian Ioana for all the help given through valuable discussions. I would also like to thank R´emi Boutonnet for helpful comments about the paper. W∗-SUPERRIGIDITY FOR COINDUCED ACTIONS 5 2. Preliminaries 2.1. Terminology. A von Neumann algebra M is called tracial if it is equipped with a faithful normal tracial state τ. We denote by L2(M ) the completion of M with respect to the norm kxk2 = pτ (x∗x). For Q ⊂ M , a unital von Neumann subalgebra of M , we denote by eQ : L2(M ) → L2(Q) the orthogonal projection onto L2(Q). We denote by EQ : M → Q, the conditional expectation onto Q. Jones' basic construction of the inclusion Q ⊂ M is defined as the von Neumann subalgebra of B(L2(M )) generated by M and eQ. Denote by U (M ) the group of unitary elements of M and by NM (Q) = {u ∈ U (M )uQu∗ = Q} the normalizer of Q inside M . Denote also by Q′ ∩ M = {x ∈ M xq = qx, for all q ∈ Q} the relative commutant of Q in M and by Z(M ) = M ∩ M ′ the center of M . 2.2. Popa's intertwining by bimodules. We recall from [Po03, Theorem 2.1 and Corollary 2.3] Popa's intertwining by bimodules tehnique which is fundamental to deformation/rigidity theory. Theorem 2.1. [Po03] Let P and Q be (not necessarily unital) subalgebras of a tracial von Neumann algebra M . The following are equivalent: • There exist non-zero projections p ∈ P, q ∈ Q, a ∗-homomorphism ϕ : pP p → qQq and a non-zero partial isometry v ∈ pM q such that xv = vϕ(x), for all x ∈ pP p. • For any group U ⊂ U (P ) such that U ′′ = P there is no sequence {un} ⊂ U such that kEQ(xuny)k2 → 0, for all x, y ∈ M. If one of these conditions holds true, then we write P ≺M Q and we say that a corner of P embeds into Q. If P p′ ≺M Q for any non-zero projection p′ ∈ P ′ ∩ pM p, then we write P ≺s M Q. 2.3. Bimodules and weak containment. Let M, N be tracial von Neumann algebras. An M -N - bimodule M HN is a Hilbert space H equipped with two commuting normal unital ∗-homomorphisms M → B(H) and N op → B(H). An M − N -bimodule M HN is weakly contained in a M -N -bimodule M KN and we write M HN ⊂ M KN if for any ǫ > 0, finite subsets F ⊂ M, G ⊂ N and ξ ∈ H, weak there exist η1, . . . , ηn ∈ K such that hxξy, ξi − n X i=1 hxηiy, ηii ≤ ǫ, for all x ∈ F, y ∈ G. Given two bimodules M HN and N KP , one can define the Connes tensor product H⊗N K which is an M -P bimodule (see [Co94, V.Appendix B]). If M HN ⊂ M K ⊗N LP , weak for any N -P bimodule L. M KN , then M H ⊗N LP ⊂ weak 2.4. Relative amenability. Let (M, τ ) be a tracial von Neumann algebra. Let p ∈ M be a projection and P ⊂ pM p, Q ⊂ M be von Neumann subalgebras. Following [OP07, Definition 2.2], we say that P ⊂ pM p is amenable relative to Q inside M if there exists a positive linear functional ϕ : phM, eQi → C such that ϕpM p = τ and ϕ is P -central. We say that M is amenable if M is amenable relative to C1 inside M . By [OP07, Section 2.2], P is amenable relative to Q inside M if and only if M L2(M p)P is weakly contained in M L2(hM, eQip)P . A von Neumann subalgebra P ⊂ pM p is strongly non-amenable relative to Q if for all non-zero projections p1 ∈ P ′ ∩ pM p, the von Neumann algebra p1P is non-amenable relative to Q. For B ⊂ M a von Neumann subalgebra, we have L2(M ) ⊗B L2(M ) ∼= L2(hM, eBi) as M -M - M L2(M ) ⊗ L2(M )M . bimodules. Note that B is amenable if and only if M L2(M ) ⊗B L2(M )M ⊂ weak 6 DANIEL DRIMBE Recall that a countable group Γ is amenable if and only if every unitary representation of Γ is weakly contained in the left regular representation ( [BHV08, Theorem G.3.2]). The next lemma is the analogous statement for amenable von Neumann algebras. The result is likely well-known, but for a lack of reference, we include a proof. Lemma 2.2. Let A be a tracial von Neumann algebra. Then A is amenable if and only if every A-A-bimodule K is weakly contained in the coarse bimodule L2(A) ⊗ L2(A). Proof. Suppose A is amenable and let K be an A-A-bimodule. Then the trivial bimodule AL2(A)A is weakly contained in the coarse bimodule AL2(A) ⊗ L2(A)A. Since L2(A) ⊗A K identifies with K as A-A bimodules, we obtain that (2.1) AKA ⊂ weak AL2(A) ⊗ KA. Now, since any right module of A is contained in LN L2(A) as a right A-submodule, we have that (2.2) CL2(A)A. CKA ⊂ weak Thus, (2.1) and (2.2) implies that AKA is weakly contained in the coarse A-A-bimodule. The converse is clear by taking K = L2(A), the trivial A-A-bimodule. (cid:4) We end this subsection by recording an immediate corollary of [DHI16, Lemma 2.6]. We provide a proof for the reader's convenience. Lemma 2.3. [DHI16, Lemma 2.6] Let P and Q be two von Neumann subalgebras of a tracial von Neumann algebra (M, τ ). If P is non-amenable relative to Q, then there exists a non-zero projection z ∈ NM (P )′ ∩ M such that P z is strongly non-amenable relative to Q. Proof. Using Zorn's lemma and a maximality argument, we can find a projection z ∈ P ′ ∩ M such that P z is strongly non-amenable relative to Q and P (1−z) is amenable relative to Q. Using [DHI16, Lemma 2.6] there exists z1 ∈ NM (P )′ ∩ M such that 1 − z ≤ z1 and P z1 is amenable relative to Q. Therefore, P (z1 − (1 − z)) is amenable relative to Q, which implies that 1 − z = z1 ∈ NM (P )′ ∩ M . (cid:4) 3. Intertwining of rigid algebras 3.1. The free product deformation for coinduced actions. In [Io06a] Ioana introduced a malleable deformation for general Bernoulli actions, where the base is any tracial von Neumann algebra. This is a variant of the malleable deformation discovered by Popa [Po03] in the case of Bernoulli actions with abelian or hyperfinite base and it was used in the context of coinduced actions in [Dr15]. Coinduced actions for tracial von Neumann algebras are defined as in Section 1.1. More precisely, let Γ be a countable group and let Σ be a subgroup. Let φ : Γ/Σ → Γ be a section. Define the cocycle c : Γ × Γ/Σ → Σ by the formula c(g, i) = φ−1(gi)gφ(i), for all g ∈ Γ and i ∈ Γ/Σ. Let Σ We define an action Γ σ y AΓ/Σ 0 σ0 y (A0, τ0) be a trace preserving action, where (A0, τ0) is a tracial von Neumann algebra. , called the coinduced action of σ0, as follows: σg((ai)i∈Γ/Σ) = (a′ i)i∈Γ/Σ, where a′ i = (σ0)c(g−1,i)−1(ag−1i). W∗-SUPERRIGIDITY FOR COINDUCED ACTIONS 7 Note that σ is a trace preserving action of Γ on the tracial von Neumann algebra AΓ/Σ Consider the free product A0 ∗ L(Z) with respect to the natural traces. Extend canonically σ0 to an action on A0 ∗ L(Z). Denote by M = (A0 ∗ L(Z))Γ/Σ ⋊σ Γ the corresponding crossed product of the coinduced action Γ Take u ∈ L(Z) the canonical generating Haar unitary. Let h = h∗ ∈ L(Z) be such that u = exp(ih) and set ut = exp(ith) for all t ∈ R. Define the deformation (αt)t∈R by automorphisms of M by σ y (A0 ∗ L(Z))Γ/Σ of σ0. 0 . αt(ug) = ug and αt(⊗h∈Γ/Σah) = ⊗h∈Γ/Σ Ad(ut)(ah), for all g ∈ Γ, t ∈ R and ⊗h∈Γ/Σah ∈ (A0 ∗ L(Z))Γ/Σ an elementary tensor. 3.2. Spectral gap rigidity for coinduced actions. Theorem 3.1. Let Γ be an icc countable group and let Σ be an almost malnormal subgroup. Let σ0 be a pmp action of Σ on a non-trivial standard probability space (X0, µ0). Denote by M = L∞(X)⋊Γ σ y (X0, µ)Γ/Σ associated to the crossed-product von Neumann algebra of the coinduced action Γ σ0 y (X0, µ). Let N be an arbitrary tracial von Neumann algebra and suppose Q ⊂ p(M ¯⊗N )p is a Σ von Neumann subalgebra such that Q′ ∩ p(M ¯⊗N )p is strongly non-amenable relative to 1 ⊗ N . Then, sup k(αt ⊗ id)(b) − bk2 converges to 0 as t → 0. b∈U (Q) Theorem 3.1 and its proof are similar with other results from the literature [Po06, Lemma 5.1], [IPV10, Corollary 4.3] and especially with [BV12, Theorem 3.1] (where the generalized Bernoulli action might have non amenable stabilizers) and with [KV15, Theorem 2.6] (which is another version of this result for coinduced actions). Proof of Theorem 3.1. Put M := M ¯⊗N and M := M ¯⊗N. The proof of this theorem goes along the same lines as the proof of [BV12, Theorem 3.1]. Therefore, instead of working with the bimodule ML2( M ⊖ M)M, we use the following M-M-submodule K := sp  (⊗i∈F ai)ug ⊗ n(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) F ⊂ Γ/Σ with k ≤ F < ∞, n ∈ N and g ∈ Γ ai ∈ A0 ∗ L(Z) for all i ∈ F ai ∈ (A0 ∗ L(Z)) ⊖ A0 for at least k elements i ∈ F .   Claim 1. The M-M-bimodule K is weakly contained in the bimodule L2(M) ⊗1⊗N L2(M). Proof of Claim 1. Let A ⊂ A0 ⊖ C1 be an orthonormal basis of L2(A0) ⊖ C1 and denote by u the canonical Haar unitary of L(Z). Define the orthonormal set A ⊂ L2(A0 ∗ L(Z)) ⊖ L2(A0) by A := {un1a1un2a2 . . . unk−1ak−1unkk ≥ 1, nj ∈ Z \ {0}, aj ∈ A for all j} This gives us the following orthogonal decomposition of L2(A0 ∗ L(Z)) into A0-A0 submodules: (3.1) If we denote L2(A0 ∗ L(Z)) = L2(A0) ⊕ M a∈ A A0aA0. C := {(⊗i∈F ci) ⊗ 1F finite , k ≤ F < ∞, ci ∈ A, for all i ∈ F}, then the decomposition (3.1) implies that the bimodule K can be written as the linear span K = spc∈C McM. To finish the proof of this claim, note that it is enough to consider an element 8 DANIEL DRIMBE c ∈ C and prove that the M -M -bimodule sp M cM is weakly contained in the coarse bimodule L2(M ) ⊗ L2(M ). Let c = (⊗i∈F ci) ⊗ 1 ∈ C. We denote by Γ0 := {g ∈ Γgf = f, for all f ∈ F}, the stabilizer of F for the action Γ y Γ/Σ and by Γ1 := {g ∈ Γg · F = F}, the normalizer of F for the same action. Since Σ is k-almost malnormal and Γ0 is a finite index subgroup of Γ1, we obtain that Γ1 is a finite group. Denote P = A ⋊ Γ1. Since P is amenable, Lemma 2.2 implies that the P -P -bimodule sp M cM is weakly contained in the coarse bimodule L2(P ) ⊗ L2(P ). Thus, for each ǫ > 0, F ⊂ Γ1 and E ⊂ A finite subsets, there exist η1, η2, ..., ηn ∈ L2(P ) ⊗ L2(P ) such that (3.2) haugc(buh)∗, ci − n X i=1 haugηi(buh)∗, ηii ≤ ǫ, for all g, h ∈ F and a, b ∈ E. Using the canonical inclusion L2(P ) ⊂ L2(M ), we obtain that haugηi(buh)∗, ηii=0, for all (g, h) ∈ (Γ × Γ) \ (Γ1 × Γ1) and a, b ∈ A. Note that also haugc(buh)∗, ci = 0, for all (g, h) ∈ (Γ × Γ) \ (Γ1 × Γ1) and a, b ∈ A. Using these observations together with (3.2), we obtain that the M -M -bimodule sp M cM is weakly contained in the coarse bimodule L2(M ) ⊗ L2(M ). This finishes the proof of the claim. (cid:3) Denote by PK the orthogonal projection of L2( M) onto the closed subspace K. Claim 2. supb∈U (Q)kPK((αt⊗id)(b))k2 converges to 0 as t → 0. Proof of Claim 2. Suppose the claim is false. Then there exist δ > 0, a sequence of positive numbers tn → 0, as n → ∞, and a sequence of unitaries bn ∈ U (Q) such that kPK((αtn ⊗id)(bn))k2 ≥ δ, for all n ≥ 1. Define ξn = PK((αtn ⊗id)(bn)). For all x ∈ Q′ ∩ pMp, we have kξnx − xξnk → 0, as n → ∞. Note also that lim inf n→∞ kξnk2 ≥ δ and kxξnk2 ≤ kxk2, for all x ∈ M. Then, [Ho15, Lemma 2.3] implies that there exists a projection q ∈ Z(Q′ ∩ pMp) such that the M-(Q′ ∩ pMp)q bimodule L2(Mq) is weakly contained in K. Claim 1 implies now that the M-(Q′ ∩ pMp)q bimodule L2(Mq) is weakly contained in the bimodule L2(M) ⊗1⊗N L2(M). This implies that (Q′ ∩ pMp)q is amenable relative to 1 ⊗ N inside M, which contradicts the hypothesis. This proves the claim. (cid:3) In order to finish the proof of the theorem we need a variant of Popa's transversality property. In the proof of [BV12, Theorem 3.1] it is proven the following fact for generalized Bernoulli if supb∈U (Q)kPK((αt⊗id)(b))k2 converges to 0 as t → 0, then supb∈U (Q) k(αt ⊗ id)(b) − actions: bk2 converges to 0 as t → 0. With the same proof we obtain the same result for coinduced actions. Claim 2 completes now the proof of the theorem. (cid:4) For Q ⊂ M a von Neumann subalgebra, we define QNM (Q) ⊂ M to be the set of all elements x ∈ M for which there exist x1, . . . , xn, y1, . . . , yn satisfying xQ ⊂ Pn i=1 yiQ. The weak closure of QNM (Q) is called the quasi-normalizer of Q inside M and note that it is a von Neumann subalgebra of M which contains both Q and Q′ ∩ M. i=1 Qxi and Qx ⊂ Pn The proof of [IPV10, Theorem 4.2] carries over verbatim and gives us the following result. Theorem 3.2. Let Γ be an icc countable group and let Σ be an almost malnormal subgroup. Let σ0 be a pmp action of Σ on a non-trivial standard probability space (X0, µ0). Denote by M = L∞(X)⋊Γ σ y (X0, µ)Γ/Σ associated to the crossed-product von Neumann algebra of the coinduced action Γ σ0 y (X0, µ). Let N be a II1 factor and suppose Q ⊂ p(M ¯⊗N )p is a von Neumann subalgebra. Σ Denote by P the quasi-normalizer of Q in p(M ¯⊗N )p. W∗-SUPERRIGIDITY FOR COINDUCED ACTIONS 9 If there exist 0 < t < 1 and δ > 0 such that τ (b∗(αt ⊗ id)(b)) ≥ δ, for all b ∈ U (Q), then one of the following statements is true: • Q ≺ 1 ⊗ N , • P ≺ (A ⋊ Σ) ¯⊗N , • there exists a unitary u ∈ M ¯⊗N such that uP u∗ ⊂ L(Γ) ¯⊗N. 3.3. Controlling intertwiners and relative commutants. In the Appendix of his PhD thesis [Bo14], Boutonnet has presented a unified approach to the notion of mixing for von Neumann algebras. As a consequence, we obtain results which give us good control over intertwiners between certain subalgebras of von Neumann algebras arising from coinduced actions. Definition 3.3. Let A ⊂ N ⊂ M be an inclusion of finite von Neumann algebras. We say that the inclusion N ⊂ M is mixing relative to A if for any sequence of unitaries {xn} ⊂ U (N ) with kEA(yxnz)k2 → 0 for all y, z ∈ N , we have kEN (m1xnm2)k2 → 0 for all m1, m2 ∈ M ⊖ N. Proposition 3.4. [Bo14, Appendix A] Let A ⊂ N ⊂ M be an inclusion of finite von Neumann algebras such that N ⊂ M is mixing relative to A. Let Q ⊂ pM p be a subalgebra such that Q ⊀M A. Denote by P the quasi-normalizer of Q in pM p. (1) If Q ⊂ N , then P ⊂ N . (2) If Q ≺ N , then there exists a non-zero partial isometry v ∈ pM such that vv∗ ∈ P and v∗P v ⊂ N. (3) If N is a factor and if Q ≺s M N , then there exists a unitary u ∈ U (M ) such that uP u∗ ⊂ N. Lemma 3.5. Let Σ be a subgroup of a countable group Γ. Let Σ non-trivial von Neumann algebra A0 and let Γ Γ y C be another tracial action. Then C ⋊ Γ ⊂ (C ¯⊗A) ⋊ Γ is mixing relative to C ⋊ Σ. σ0 y A0 be a tracial action on a be the coinduced action of σ0. Let σ y A := AΓ/Σ 0 Proof. Denote M := (C ¯⊗A) ⋊ Γ and I := Γ/Σ. Let {xn} ⊂ U (C ⋊ Γ) be a sequence of unitaries such that kEC⋊Σ(yxnz)k2 → 0, for all y, z ∈ C ⋊ Γ. Let a, b ∈ M ⊖ (C ⋊ Γ). We have to show that kEC⋊Γ(axnb)k2 → 0. Since EC⋊Γ is C ⋊ Γ-bimodular, we can assume a, b ∈ A. Moreover, we can suppose that there exist a finite subset J ⊂ I and j0 ∈ J such that a, b = ⊗j∈J bj ∈ AJ 0 with i=1giΣg−1 bj0 ∈ A0 ⊖ C. If j0 = g0Σ and J = {g1Σ, . . . , gnΣ} note that Σ0 := {g ∈ Γgj0 ∈ J} = ∪n 0 . Now, since axnb = Pg∈Γ aEC(xnu∗ g)σg(b)ug, we have n kEC⋊Γ(axnb)k2 2 = X g∈Σ0 τ (aσg(b))2kEC (xnu∗ g)k2 2 ≤ kak2 2kbk2 2 kEC⋊Σ(u∗ gixnug0)k2 2, X i=1 which goes to zero because of the assumption. This proves the lemma. (cid:4) Proposition 3.4 together with Lemma 3.5 give the following result. σ0 y A0 be a tracial action on a Corollary 3.6. Let Σ be a subgroup of a countable group Γ. Let Σ σ y A := AΓ/Σ non-trivial tracial von Neumann algebra A0 and let Γ be the coinduced action of σ0. Let Γ y C be another tracial action and let N be an arbitrary factor. Define M := (C ¯⊗A) ⋊ Γ. Suppose Q ⊂ p(M ¯⊗N )p is a von Neumann subalgebra such that Q ⊀ (C ⋊ Σ) ¯⊗N . Denote by P the quasi-normalizer of Q inside p(M ¯⊗N )p. 0 (1) If Q ⊂ p((C ⋊ Γ) ¯⊗N )p, then P ⊂ p((C ⋊ Γ) ¯⊗N )p. 10 DANIEL DRIMBE (2) If Q ≺ (C ⋊ Γ) ¯⊗N , then there exists a non-zero partial isometry v ∈ p(M ¯⊗N ) such that vv∗ ∈ P and v∗P v ⊂ (C ⋊ Γ) ¯⊗N. (3) If Q ≺s M ¯⊗N (C ⋊ Γ) ¯⊗N , then there exists a unitary u ∈ U (M ¯⊗N ) such that uP u∗ ⊂ (C ⋊ Γ) ¯⊗N. The proof of the following proposition is similar to [Bo12a, Corollary 3.7] and we leave it to the reader. Proposition 3.7. Let Γ y C be a tracial action and denote M0 = C ⋊ Γ. Let Σ be an almost malnormal subgroup of Γ. Suppose Q ⊂ pM0p is a von Neumann subalgebra such that Q ≺ C ⋊ Σ and Q ⊀ C. Denote by P the quasi-normalizer of Q inside pM0p. Then P ≺ C ⋊ Σ. 4. Rigidity coming from measure equivalence In this section we establish some results needed in the proof of Theorem B. Throughout the section, we will work with coinduced actions satisfying the following: Assumption 4.1. Let Σ be a subgroup of a countable icc group Γ. Let σ0 be a pmp action of Σ on a non-trivial standard probability space (X0, µ0) and denote by σ the coinduced action of Γ on X := X Γ/Σ . Suppose: 0 • Γ is a non-amenable icc group which is measure equivalent to a group Λ0 for which the group von Neumann algebra L(Λ0) is not prime. • Σ is almost malnormal. Note that since Σ is almost malnormal in Γ, we have that [Γ : Σ] = ∞. Before stating the results of this section, we need to introduce some notation. Notation 4.2. The group von Neumann algebra L(Λ0) is not prime, therefore there exist von Neumann algebras R1 and R2, both not of type I, such that L(Λ0) = R1 ¯⊗R2. Since L(Λ0) is diffuse and non-amenable, there exists z0 ∈ Z(L(Λ0)) such that R1z0 and R2z0 are diffuse and L(Λ0)z0 is non-amenable. The group Γ is measure equivalent to Λ0. By [Fu99, Lemma 3.2], Γ and Λ0 admit stably orbit equivalent free ergodic pmp actions. Thus, we may find a free ergodic pmp action Γ y (Z0, ν) and ℓ ≥ 1, such that the following holds: consider the product action Γ × Z/ℓZ y (Z0 × Z/ℓZ, ν × c), where Z/ℓZ acts on itself by addition and c denotes the counting measure on Z/ℓZ. Then there exist a non-negligible measurable set Y0 ⊂ Z0 × Z/ℓZ and a free ergodic measure preserving action Λ0 y Y0 such that R(Λ0 y Y0) = R(Γ × Z/ℓZ y Z0 × Z/ℓZ) ∩ (Y0 × Y0). We put C0 = L∞(Y0), M0 = L∞(Z0 × Z/ℓZ) ⋊ (Γ × Z/ℓZ), p = 1Y0, and note that C0 ⋊ Λ0 = pM0p. We identify L∞(Z/ℓZ) ⋊ Z/ℓZ = Mℓ(C), and use this identification to write M0 = C ⋊ Γ, where C = L∞(Z0) ⊗ Mℓ(C) and Γ acts trivially on Mℓ(C). Denote A = L∞(X) and let {ug}g∈Γ ⊂ (C ¯⊗A) ⋊ Γ denote the canonical unitaries implementing the diagonal action of Γ on C ¯⊗A. Remark 4.3. Throughout this section we will use many times the following easy observation (see [Va08, Lemma 3.4]). Let P ⊂ pM p and Q ⊂ qM q be von Neumann subalgebras of a tracial von Neumann algebra (M, τ ). Then: • if p0P p0 ≺ Q for a non-zero projection p0 ∈ P , then P ≺ Q. W∗-SUPERRIGIDITY FOR COINDUCED ACTIONS 11 • if P p′ ≺ Q for a non-zero projection p′ ∈ P ′ ∩ pM p, then P ≺ Q. Lemma 4.4. Let w : Γ → U (A ¯⊗N ) be a cocycle for the action σ⊗id, where N is II1 factor. Define the ∗-homomorphism d : C ⋊ Γ → (A ⋊ Γ) ¯⊗N ¯⊗(C ⋊ Γ) by d(cug) = wgug ⊗ cug, g ∈ Γ, c ∈ C. Let Q ⊂ pM0p be a subalgebra and let Σ0 ⊂ Γ be a subgroup. The following hold: (1) If Q ⊀ C , then d(Q) ⊀ 1 ⊗ N ¯⊗(C ⋊ Γ). (2) If [Γ : Σ0] = ∞, then d(L(Λ0)) ⊀ (A ⋊ Σ0) ¯⊗N ¯⊗(C ⋊ Γ). (3) If Q is non-amenable, then d(Q) is non-amenable relative to 1 ⊗ N ¯⊗(C ⋊ Γ). Proof. Denote M = (A ⋊ Γ) ¯⊗N ¯⊗M0 and N = 1 ⊗ N ¯⊗M0. (1) Let {un}n≥1 ⊂ U (Q) be a sequence of unitaries such that kEC (unug)k2 → 0, for all g ∈ Γ. We claim that kE1⊗N ¯⊗M0(xd(un)y)k2 → 0, for all x, y ∈ M. Since EN is N -bimodular, by Kaplansky's density theorem we may assume x = aug ⊗ 1 ⊗ 1, y = buh ⊗ 1 ⊗ 1 for some a, b ∈ A and g, h ∈ Γ. Then for all n ≥ 1, we have xd(un)y = X k∈Γ aσg(wk)σgk(b)ugkh ⊗ EC(unu∗ k)uk. Therefore, kEN (xd(un)y)k2 ≤ kakkbkkEC (unu∗ (2) Assume d(L(Λ0)) ≺ (A ⋊ Σ0) ¯⊗N ¯⊗M0. Since d(C0) ⊂ 1 ⊗ 1 ⊗ C0, we obtain d(C0 ⋊ Λ0) ≺ (A ⋊ Σ0) ¯⊗N ¯⊗M0. Therefore d(L(Γ)) ≺ (A ⋊ Σ0) ¯⊗N ¯⊗M0, which implies L(Γ) ≺ L(Σ0). Indeed, suppose by contrary that L(Γ) ⊀ L(Σ0). Then there exists a sequence un ∈ U (L(Γ)) such that kEL(Σ0)(xvny)k2 → 0, for all x, y ∈ L(Γ). We would like to prove that g−1h−1)k2 → 0. (4.1) kE(A⋊Σ0) ¯⊗N ¯⊗M0(xd(un)y)k2 → 0, for all x, y ∈ (A ⋊ Γ) ¯⊗N ¯⊗M0. For proving (4.1), it is enough to consider x = ug ⊗ 1 ⊗ 1 and y = uh ⊗ 1 ⊗ 1, with g, h ∈ Γ. In this case one can check that kE(A⋊Σ0) ¯⊗N ¯⊗M0(xd(un)y)k2 = kEL(Σ0)(ugunuh)k2, which goes to 0. Therefore (4.1) is proven and we obtain that d(L(Γ)) ⊀ (A ⋊ Σ0) ¯⊗N ¯⊗M0, contradiction. Thus L(Γ) ≺ L(Σ0), which implies that Σ0 has finite index in Γ by [DHI16, Lemma 2.5]. (3) Suppose by contrary that d(Q) is amenable relative to N . Then there exists a positive linear functional φ : d(p)hM, eN id(p) → C such that φd(p)Md(p) = τ and φ is d(Q)-central. Define now ϕ : phM0, eCip → C by N N ϕ( X i=1 mieCni) = φ( X i=1 d(mi)eN d(ni)), Indeed, suppose PN i=1 d(mi)τ (ni) = 0. Since EN (d(m)) = τ (m), for all m ∈ M0, we obtain PN where N ≥ 1, mi, ni ∈ M0, i ∈ {1, . . . , N }. Note that ϕ is a well defined positive linear func- i=1 mieCni = 0, with mi, ni ∈ M0, for all 1 ≤ i ≤ N. This implies tional. i=1 d(mi)EN (d(ni)) = i=1 d(mi)eN d(ni) = 0. Therefore, ϕ is a positive linear functional which is (cid:4) PN 0, which implies PN Q-central and ϕpM0p = τ . We obtain Q is amenable, contradiction. Denote by Uf in the class of Polish groups which arise as closed subgroups of the unitary groups of II1 factors [Po05]. In particular, all countable discrete groups and all compact Polish groups belong to Uf in. 12 DANIEL DRIMBE Theorem 4.5. (Cocycle superrigidity.) Let Γ y X be as in Assumption 4.1. Then any cocycle w : Γ × X → Λ valued in a group Λ ∈ Uf in untwists, i.e. there exists a measurable map ϕ : X → Λ and a group homomorphism d : Γ → Λ such that w(g, x) = ϕ(gx)d(g)ϕ(x)−1 for all g ∈ Γ and a.e. x ∈ X. This result was proven in [PS09] for Bernoulli actions using deformations obtained from closable derivations. In our case, we will provide a direct proof for Theorem 4.5 which uses only the free product deformation αt defined in Section 3.1. Proof. Define A := L∞(X) and let N be a II1 factor such that Λ ⊂ U (N ). We associate to w : Γ×X → U (N ) the cocycle w : Γ → U (A ¯⊗N ), given by wg(x) = w(g, g−1x). Define Q = {wgug}′′ g∈Γ. Claim. We have sup k(αt ⊗ id)(b) − bk2 converges to 0 as t → 0. b∈U (Q) Proof of the Claim. As in Lemma 4.4 we define the ∗-homomorphism d : C⋊Γ → (A⋊Γ) ¯⊗N ¯⊗(C⋊Γ) by d(cug) = wgug ⊗ cug, g ∈ Γ, c ∈ C. Denote M = (A ⋊ Γ) ¯⊗N ¯⊗M0. Without loss of generality assume that R1z0 is non-amenable. Lemma 4.4 implies that d(R1z0) is non-amenable relative to 1 ⊗ N ¯⊗(C ⋊ Γ). By Lemma 2.3 there exists a non-zero projection z ∈ Nd(z0)Md(z0)d(R1z0)′ ∩ d(z0)Md(z0) such that d(R1)z is strongly non-amenable relative to 1 ⊗ N ¯⊗(C ⋊ Γ). Using Theorem 3.1 we obtain that sup k(αt ⊗ id ⊗ id)(b) − bk2 converges to 0 as t → 0. b∈U (d(R2)z) and therefore by Theorem 3.2 we obtain that one of the following hold: (1) d(R2)z ≺ 1 ⊗ N ¯⊗(C ⋊ Γ), (2) d(L(Λ0))z ≺ (A ⋊ Σ) ¯⊗N ¯⊗(C ⋊ Γ), (3) d(L(Λ0))z ≺ L(Γ) ¯⊗N ¯⊗(C ⋊ Γ). Note that (1) and (2) are not possible by Lemma 4.4 since R2z0 is diffuse and [Γ : Σ] = ∞. Therefore (3) is true. Now, together with the remark that d(C) ⊂ 1 ⊗ 1 ⊗ C we obtain that d(C ⋊ Γ) ≺ L(Γ) ¯⊗N ¯⊗(C ⋊ Γ). One can check directly this fact or use [BV12, Lemma 2.3]. Proceeding in the same way, we obtain actually d(C ⋊ Γ) ≺s M L(Γ) ¯⊗N ¯⊗(C ⋊ Γ). Lemma 4.4 implies that d(C ⋊ Γ) ⊀ L(Σ) ¯⊗N ¯⊗(C ⋊ Γ), so by Corollary 3.6 we obtain sup k(αt ⊗ id)(b) − bk2 converges to 0 as t → 0. b∈U (Q) (cid:3) Using a result which goes back to Popa [Po05], the claim implies that the cocycle w untwists (see [Dr15, Theorem 2.15], the proof of [Dr15, Proposition 3.2] and [Dr15, Remark 3.3]). (cid:4) Theorem 4.6. Let Γ y X be as in Assumption 4.1 and supppose that Σ is amenable. Let Λ y B be a tracial action on a non-trivial abelian von Neumann algebra B such that A ⋊ Γ = B ⋊ Λ. Denote by ∆ : B ⋊ Λ → (B ⋊ Λ) ¯⊗L(Λ) the comultiplication ∆(bvλ) = bvλ ⊗ vλ for all b ∈ B and λ ∈ Λ (we let {vλ}λ∈Λ ⊂ B ⋊ Λ denote the canonical unitaries implementing the action of Λ on B). Then there exists a unitary u ∈ U ((A ⋊ Γ) ¯⊗(A ⋊ Γ)) such that u∆(L(Γ))u∗ ⊂ L(Γ × Γ). W∗-SUPERRIGIDITY FOR COINDUCED ACTIONS 13 Define M := (C ¯⊗A) ⋊ Γ and θ : M → M ¯⊗M ¯⊗M by θ(caug) = cug ⊗ ∆(aug), for all c ∈ C, a ∈ A and g ∈ Γ. In the following lemma we record some properties of the unital ∗-homomorphism θ which are similar to the ones of [Io10, Lemma 10.2]. Lemma 4.7. Let Q ⊂ qM q. The following hold: (1) If Q is diffuse, then θ(Q) ⊀ M ¯⊗1 ¯⊗M . (2) If Q ⊀ B, then θ(Q) ⊀ M ¯⊗M ⊗ 1. (3) If Q has no amenable direct summand, then θ(Q) is strongly non-amenable relative to M ¯⊗M ⊗ 1 and M ¯⊗1 ¯⊗M. We continue now with the proof of Theorem 4.6 and we will give the proof of Lemma 4.7 at the end of this section. Proof of Theorem 4.6. Without loss of generality we can assume that R1z0 is non-amenable. Take z ∈ Z(R1z0) such that R1z has no amenable direct summand. Claim 1. We have supb∈U (∆(L(Γ)) k(id ⊗ αt)(b) − bk2 converges to 0 as t → 0. Proof of Claim 1. Note that θ(R1z) is strongly non-amenable relative to M ¯⊗M ⊗ 1 by Lemma 4.7. Therefore by Theorem 3.1, we obtain (4.2) sup k(id ⊗ id ⊗ αt)(b) − bk2 converges to 0 as t → 0. b∈U (θ(R2z)) Using Theorem 3.2 we obtain that one of the following three conditions holds: (1) θ(R2z) ≺ M ¯⊗M ⊗ 1, (2) θ(L(Λ0)z) ≺ M ¯⊗M ¯⊗(A ⋊ Σ), (3) there exists a unitary u ∈ M such that uθ(L(Λ0)z)u∗ ⊂ M ¯⊗M ¯⊗L(Γ). If (1) holds, Lemma 4.7 implies R2z ≺M B. By applying [Va08, Lemma 3.5], we obtain B ≺M zM z∩(R2z)′. Note that if R2z ≺M C⋊Σ, Proposition 3.7 implies that L(Λ0) ≺ C⋊Σ. Using [BV12, Lemma 2.3] we deduce that C ⋊ Γ ≺ C ⋊ Σ. This is a contradiction since [Γ : Σ] = ∞. Therefore R2z ⊀M C ⋊ Σ and Corollary 3.6 implies that zM z ∩ (R2z)′ ⊂ C ⋊ Γ, so B ≺M C ⋊ Γ. On the other hand, since B ⊂ A ⋊ Γ, we obtain B ≺A⋊Γ L(Γ). Proposition 3.7 implies that B ⊀A⋊Γ L(Σ). Finally, using Corollary 3.6 we obtain that A ⋊ Γ ≺A⋊Γ L(Γ), which is a contradiction. Now, if (2) holds, we obtain θ(L(Λ0)) ≺ M ¯⊗M ¯⊗(A⋊Σ). Together with θ(C) ⊂ C ⊗1⊗1, we obtain θ(M0) ≺ M ¯⊗M ¯⊗(A ⋊ Σ). Since Σ is amenable, it implies that θ(M0) is not strongly non-amenable relative to M ¯⊗M ⊗ 1. Now, M0 is a factor, so Lemma 4.7 gives that M0 is amenable, which is a contradiction. Thus, (3) holds. Since θ(C0) ⊂ C0 ⊗ 1 ⊗ 1, we obtain θ(M0) ≺ M ¯⊗M ¯⊗L(Γ) M ¯⊗M ¯⊗M M ¯⊗M ¯⊗L(Γ). With the same computation, we obtain θ(M0) ≺s Lemma 4.7 implies that θ(M0) ⊀ M ¯⊗M ¯⊗L(Σ) since Σ is amenable and M0 is a factor. By Corollary 3.6 we obtain that supb∈U (∆(L(Γ)) k(id ⊗ αt)(b) − bk2 converges to 0 as t → 0. (cid:3) Claim 2. We have supb∈U (∆(L(Γ))) k(αt ⊗ id)(b) − bk2 converges to 0 as t → 0. Proof of Claim 2. As in Claim 1, by applying Lemma 4.7, Theorem 3.1 and Theorem 3.2 we obtain that one of the following conditions hold: (1) θ(R2z) ≺ M ¯⊗1 ¯⊗M , (2) θ(L(Λ0)z) ≺ M ¯⊗(A ⋊ Σ) ¯⊗M , 14 DANIEL DRIMBE (3) there exists a unitary u ∈ M such that uθ(L(Λ0)z)u∗ ⊂ M ¯⊗L(Γ) ¯⊗M. Note that by Lemma 4.7, (1) is not possible since R2z is diffuse. As before, (2) is not possible, which implies (3) holds true and by reasoning as before we obtain the claim. (cid:3) Notice that ∆(L(Γ)) is a factor since Γ is icc. Using Claim 1 and 2 and by applying twice Theorem 3.2 and [IPV10, Lemma 10.2] we obtain the conclusion. (cid:4) Proof of Lemma 4.7. The proofs of (1) and (2) are similar to the proof of Lemma 4.4.1 (see also the proof of [IPV10, Lemma 10.2]). For proving (3), denote M := M ¯⊗M ¯⊗(A⋊Γ) and ψ : M → M ¯⊗M , by ψ(caug) = cug ⊗ aug for all c ∈ C, a ∈ A and g ∈ Γ. Claim 1. We have ML2(M) ⊗M ¯⊗M ⊗1 L2(M)θ(M ) ⊂ weak (here we consider that ψ(M ) ⊂ M ¯⊗M acts to the right on L2(M ) ⊗ L2(M ) ⊗ L2(M ) ⊗ L2(M ) on the first and fourth positions.) ML2(M) ⊗ L2(A ⋊ Γ)ψ(M )1,4 . Proof of the Claim 1. Note that we have the identification ML2(M) ⊗M ¯⊗M ⊗1 L2(M)θ(M ) ≃M1,2,3 L2(M ¯⊗M ¯⊗(A ⋊ Γ) ¯⊗(A ⋊ Γ))θ(M )1,2,4 as M-M -bimodules. Therefore, it is enough to show that M ¯⊗M ⊗1L2(M ¯⊗M ¯⊗(A ⋊ Γ))θ(M ) ⊂ weak M ¯⊗M ⊗1L2(M ¯⊗M ¯⊗(A ⋊ Γ))ψ(M )1,3 . Let B be an orthonormal basis for L2(B) and note that we have the following orthogonal decom- position into (M ¯⊗M )-M -bimodules: L2(M ¯⊗M ¯⊗(A ⋊ Γ)) = M b∈B sp (M ¯⊗M ⊗ 1)(1 ⊗ 1 ⊗ b) θ(M ) First, notice that for a fixed b ∈ B we have sp (M ¯⊗M ⊗ 1)(1 ⊗ 1 ⊗ b) θ(M ) ≃(M ¯⊗M )1,2 L2(M ) ⊗ L2(M ) ⊗B L2(A ⋊ Γ)ψ(M )1,3 as (M ¯⊗M )-M -bimodules. Indeed, let m1, m2, m3 ∈ M and let us prove that (4.3) h(m1 ⊗ m2 ⊗ 1)(1 ⊗ 1 ⊗ b)θ(m3), 1 ⊗ 1 ⊗ bi = h(m1 ⊗ m2 ⊗ 1)(1 ⊗ 1 ⊗B 1)ψ(m3), 1 ⊗ 1 ⊗B 1i We may assume m3 = caug for some c ∈ C, a ∈ A and g ∈ Γ. Write aug = Pl∈Λ blvl ∈ B ⋊ Λ, with bl ∈ B for all l ∈ Λ. Therefore, the LHS of (4.3) equals to τ ((m1 ⊗ m2 ⊗ b∗b)θ(m3)) = τ (m1cug ⊗ ((m2 ⊗ b∗b)∆(aug))) = τ (m1cug)τ (m2be). On the other hand, the RHS of (4.3) equals to τ ((m1 ⊗ EB(m2))ψ(m3)) = τ (m1cug ⊗ EB(m2)aug) = τ (m1cug)τ (m2be), which proves (4.3). Now since B is amenable, we obtain that (M ¯⊗M )1,2L2(M ) ⊗ L2(M ) ⊗B L2(A ⋊ Γ)ψ(M )1,3 ⊂ weak This finishes the proof of the claim. (M ¯⊗M )1,2L2(M ) ⊗ L2(M ) ⊗ L2(A ⋊ Γ)ψ(M )1,3 . (cid:3) Claim 2. We have ML2(M) ⊗ L2(M )ψ(M )1,4 ⊂ weak ML2(M) ⊗ L2(M )M . Proof of the Claim 2. First, note that it is enough to prove M L2(M ) ⊗ L2(M )ψ(M ) ⊂ weak M L2(M ) ⊗ L2(M )M . W∗-SUPERRIGIDITY FOR COINDUCED ACTIONS 15 Let C be an orthonormal basis for L2(C) and note that we have the following orthogonal decom- position into M -M -bimodules: L2(M ) ⊗ L2(M ) = M c∈C sp M (1 ⊗ c) d(M ). Note that sp M (1 ⊗ c) d(M ) ∼= L2(M ) ⊗C L2(M ) as M -M -bimodules. Indeed, let us take m1 = c1a1ug1, m2 = c2a2ug2, and note that hm1(1 ⊗ c)ψ(m2), 1 ⊗ ci = hc1a1ug1c2ug2 ⊗ ca2ug2, 1 ⊗ ci = δg1,eδg2,eτ (c1c2)τ (a1)τ (a2) and hm1eCm2, eC i = τ (EC(c1a1ug1)c2a2ug2) = δg1,eδg2,eτ (c1c2)τ (a1)τ (a2). This implies that sp M (1 ⊗ c) ψ(M ) ∼= L2(M ) ⊗C L2(M ) as M -M -bimodules. Since C is amenable, the claim is proven. Now, assume that θ(Q) is not strongly non-amenable relative to M ¯⊗M ⊗ 1. Then there exists a non-zero projection p ∈ θ(Q)′ ∩ θ(q)Mθ(q) such that (cid:3) ML2(Mp)θ(Q) ⊂ weak ML2(M) ⊗M ¯⊗M ⊗1 L2(M)θ(Q). Using Claim 1 and 2, we obtain now that ML2(Mp)θ(Q) ⊂ weak ML2(M) ⊗ L2(Q)Q. Take z ∈ Q such that θ(z) is the support projection of Eθ(Q)(p). Note that z is a non-zero central projection in Q and that θ embeds the trivial Qz-Qz-bimodule into θ(Qz)L2(θ(Qz))θ(Qz). There- QzL2(Qz) ⊗ fore, QzL2(Qz)Qz ⊂ weak L2(Qz)Qz, which means that Qz is amenable, contradiction. In a similar way, one can prove that θ(Q) is strongly non-amenable relative to M ¯⊗1 ¯⊗M. This ends the proof. (cid:4) θ(Qz)L2(M) ⊗ L2(Qz)Qz. Finally, we obtain QzL2(Qz)Qz ⊂ weak 5. Intertwining of abelian subalgebras Throughout this section we will use the following notation. Let Γ be a countable group. Let Σ be an almost malnormal subgroup and let σ0 be a tracial action of Σ on a non-trivial abelian von Neumann algebra A0. Denote by σ the coinduced action of Γ on A := AΓ/Σ . Finally, denote M = A ⋊ Γ. 0 The next result is a localization theorem for coinduced actions which goes back to [Io10, Theorem 6.1]. The form presented in this paper is very similar to [IPV10, Theorem 5.1], but written with coinduced actions instead of generalized Bernoulli ones. Theorem 5.1. Assume that D ⊂ M ¯⊗M is an abelian von Neumann subalgebra which is normalized by a group of unitaries (γ(s))s∈Λ that belong to L(Γ) ¯⊗L(Γ). Denote by P the quasi-normalizer of D inside M ¯⊗M. We make the following assumptions: (1) D ⊀ M ⊗ 1 and D ⊀ 1 ⊗ M , (2) P ⊀ M ¯⊗(A ⋊ Σ) and P ⊀ (A ⋊ Σ) ¯⊗M , (3) P ⊀ M ¯⊗L(Γ) and P ⊀ L(Γ) ¯⊗M , (4) γ(Λ)′′ ⊀ L(Γ) ¯⊗L(Σ) and γ(Λ)′′ ⊀ L(Σ) ¯⊗L(Γ). Define C := D′ ∩ (M ¯⊗M ). Then for every non-zero projection q ∈ Z(C) we have Cq ≺ A ¯⊗A. 16 DANIEL DRIMBE The proof is identically with the one of [IPV10, Theorem 5.1], since essentially the same computa- tions still hold once we replace generalized Bernoulli actions by coinduced ones. Next, we obtain a similar statement if one considers an abelian von Neumann algebra in M and not in M ¯⊗M . Theorem 5.2. Assume that D ⊂ M is an abelian von Neumann subalgebra which is normalized by a group of unitaries (γ(s))s∈Λ that belong to L(Γ). Denote by P the quasi-normalizer of D inside M. We make the following assumptions: (1) D is diffuse, (2) P ⊀ A ⋊ Σ, (3) P ⊀ L(Γ), (4) γ(Λ)′′ ⊀ L(Σ). Define C := D′ ∩ M. Then for every non-zero projection q ∈ Z(C) we have Cq ≺ A. As noticed in [Io10], we obtain as a corollary a weaker version of Popa's conjugacy criterion adapted in this case to coinduced actions. Theorem 5.3. Suppose Γ is icc and Σ is amenable. Let Λ y B be another tracial action of a countable group Λ on a non-trivial abelian von Neumann algebra B such that M = A ⋊ Γ = B ⋊ Λ and L(Λ) ⊂ L(Γ). Then B ≺ A. Proof. The proof is a direct application of Theorem 5.2. Note that the quasi-normalizer of the abelian algebra B is M . Now, notice that if M ≺ A ⋊ Σ, by [DHI16, Lemma 2.5.1] we obtain that [Γ : Σ] < ∞. This is not possible since Σ is almost malnormal in Γ. Also L(Λ) ⊀ L(Σ) since Σ is amenable and therefore we obtain B ≺ A. (cid:4) 6. Proof of the main results In [Io10], Ioana has proven that any Bernoulli action of an arbitrary icc property (T) group is W∗-superrigid. The strategy of his proof was successfully applied also in [IPV10] and [Bo12]. 6.1. A general method for obtaining W∗-superrigidity. Using Ioana's proof, we identify a couple of steps for proving that a certain free ergodic pmp action Γ y X is W∗- superrigid (see also the introduction of [Bo12]). Consider an arbitrary free ergodic pmp action Λ y Y such that M := A ⋊ Γ = B ⋊ Λ, where A = L∞(X) and B = L∞(Y ). Define the comultiplication ∆ : M → M ¯⊗L(Λ) by ∆(bvλ) = bvλ ⊗ vλ, for all b ∈ B, λ ∈ Λ, where we denote by vλ, λ ∈ Λ, the canonical unitaries corresponding to the action of Λ. Step 1. One has to show that Γ y X is OE superrigid. From now on, using Singer's result [Si55], it is enough to assume that B is not unitarily conjugated to A in M , which is equivalent to B ⊀M A [Po06b, Theorem A.1]. Step 2. One can also assume that there exists a non-zero projection s0 ∈ L(Λ)′ ∩ M such that L(Λ)s0 ⊀ L(Γ). Step 3. One shows that there exists a unitary u ∈ U (M ¯⊗M ) such that u∆(L(Γ))u∗ ⊂ L(Γ × Γ). W∗-SUPERRIGIDITY FOR COINDUCED ACTIONS 17 Step 4. Next, one proves that the algebra C := ∆(A)′ ∩ (M ¯⊗M ) satisfies Cq ≺M ¯⊗M A ¯⊗A for all q ∈ Z(C). Step 5. Using the previous steps together with a generalization of [Po04, Theorem 5.2], one essen- tially obtains that there exist a unitary v ∈ U (M ¯⊗M ), a group homomorphism δ : Γ → Γ × Γ and a character ω : Γ → C such that vCv∗ = A ¯⊗A and v∆(ug)v∗ = ω(g)uδ(g), for all g ∈ Γ (the precise statement is the Step 3 of the proof [IPV10, Theorem 10.1]). Step 6. Using Step 5, one proves that for every sequence (xn)n in M for which the Fourier co- efficient (w.r.t. the decomposition M = A ⋊ Γ) converges to 0 pointwise in k · k2, then the Fourrier coefficient of ∆(xn) (w.r.t. the decomposition M ¯⊗M = (M ¯⊗A) ⋊ Γ) also converges to 0 pointwise in k · k2. This shows B ≺ A and Step 1 implies that Γ y X is W∗-superrigid. 6.2. Proof of Theorem A. We record first the following observation. Remark 6.1. Since Σ is almost malnormal in Γ, using [Dr15, Lemma 5.3], the action Γ y X is free (see also [Io06b, Lemma 2.1]). Proof of Theorem A. Assume that Λ y (Y, ν) is an arbitrary free ergodic pmp action such that M := L∞(X) ⋊ Γ = L∞(Y ) ⋊ Λ. We put A = L∞(X), B = L∞(Y ). Define ∆ : M → M ¯⊗M by ∆(bvs) = bvs ⊗ vs, for all b ∈ B and s ∈ Λ, where we denote by vs, s ∈ Λ, the canonical unitaries corresponding to the action of Λ. Since the action Γ y X is OE superrigid (using [Dr15, Theorem A] and [Po05, Theorem 5.6]), Step 1 is completed. To prove Step (2), suppose L(Λ)q ≺ L(Γ) for all q ∈ L(Λ)′∩M . Since Σ is amenable, L(Λ) ⊀ L(Σ), so by Corollary 3.6, there exists a unitary u ∈ U (M ) such that uL(Λ)u∗ ⊂ L(Γ). Based on Step 1, Theorem 5.3 proves that Γ y X is W∗-superrigid. This completes Step 2. Therefore, we take a non-zero projection q0 ∈ L(Λ)′ ∩ M such that L(Λ)q0 ⊀ L(Γ). Step (3) is obtained by combining Theorem 3.2 and [IPV10, Lemma 10.2.5]. Proof of Step (4). Note that Theorem 5.1 proves this step by considering the abelian subalgebra D0 := ∆(A)(1 ⊗ q0). For showing this, denote C0 = D′ 0 ∩ (M ¯⊗q0M q0), C = ∆(A)′ ∩ (M ¯⊗M ) and note that C0 = C(1 ⊗ q0). Since L(Λ)q0 ⊀ L(Γ), [Io10, Lemma 9.2.4] implies that ∆(M )(1 ⊗ q0) ⊀ M ¯⊗L(Γ). Using [IPV10, Lemma 10.2], we see that all the conditions of Theorem 5.1 are satisfied. Therefore, we obtain that C0q ≺ A ¯⊗A, for all q ∈ Z(C0) = Z(C)(1 ⊗ q0). (cid:3) Proof of Step (5). For proving Step (3) of the proof of [IPV10, Theorem 10.1], one only needs to show: • If H is a subgroup of Γ × Γ such that H acts non-ergodically on A ¯⊗A, then ∆(L(Γ)) ⊀ (A ¯⊗A) ⋊ H. Suppose by contrary that ∆(L(Γ)) ⊀ (A ¯⊗A) ⋊ H. It is easy to prove that there exists a finite set T ⊂ Γ such that H ⊂ (∪t∈T tΣ) × Γ or H ⊂ Γ × (∪t∈T tΣ). This implies that ∆(L(Γ)) ≺ (A ⋊ Σ) ¯⊗M or ∆(L(Γ)) ≺ M ¯⊗(A ⋊ Σ). By applying [IPV10, Lemma 10.2.5], we obtain a contradiction. (cid:3) Step (6) works in general once the other steps are proven. This finishes the proof of the theorem. (cid:4) 18 DANIEL DRIMBE 6.3. Proof of Theorem B. In this subsection we will prove a more general statement of Theorem B. Assumption 6.2. Let Σ be a subgroup of a countable icc group Γ. Let σ0 be a pmp action of Σ on a non-trivial standard probability space (X0, µ0) and denote by σ the coinduced action of Γ on X := X Γ/Σ . Suppose: 0 • Γ is a non-amenable icc group which is measure equivalent to a group Λ0 for which the group von Neumann algebra L(Λ0) is not prime. • Σ is amenable and almost malnormal. Theorem 6.3. Let Γ y X be as in Assumption 6.2. Then Γ y X is W∗-superrigid. Proof. The proof of this theorem goes along the same lines as the proof of Theorem A. We point out only the differences. The action Γ y X is OE superrigid using Theorem 4.5 and [Po05, Theorem 5.6]. Step (3) follows by Theorem 4.6. All the other steps follow as in the proof of Theorem A, which finishes the proof. (cid:4) Remark 6.4. A careful handling of Thorem 4.6 shows that Assumption 6.2 can be improved by supposing the weaker assumption that L(Λ0) contains a commuting pair of diffuse subalgebras P1 and P2 such that P2 is non-amenable and NL(Λ0)(P1 ∨ P2)′′ = L(Λ0) (see also Step 1 of the proof of [IPV10, Theorem 8.2]). Corollary 6.5. Let Γ be an icc non-amenable group which is measure equivalent to a group Λ0 for which L(Λ0) is not prime. Then the Bernoulli action Γ y (X, µ)Γ is W ∗-superrigid, where (X, µ) is a non-trivial standard probability space. References [BHV08] B.Bekka, P. de la Harpe, A.Valette: Kazhdan's Property (T), (New Mathematical Monographs, 11), Cam- bridge University Press, Cambridge, 2008. [Bo63] A. Borel: Compact Clifford-Klein forms of symmetric spaces, Topology 2 (1963), 111-122. [Bo12a] R. Bouttonet: On solid ergodicity for Gaussian actions, J. Funct. Anal.,263 (2012) 1040-1063. [Bo12] [Bo14] [BV12] M. Berbec, S. Vaes: W∗-superrigidity for group von Neumann algebras or left-right wreath products, Pro- R. Boutonnet: W∗-superrigidity of mixing Gaussian actions of rigid groups, Adv. Math. 244 (2013) R. Bouttonet: Plusieurs aspects de rigidit´e des alg`ebres de von Neumann, PhD thesis (2014). [CI08] [CI17] [CIK13] ceedings of the London Mathematical Society 108 (2014), 1116-1152. I. Chifan, A. Ioana: Ergodic subequivalence relations induced by a Bernoulli action, Geometric and Func- tional Analysis, 20(1): 53-67, 2010. I. Chifan, A. Ioana: Amalgamated free product rigidity for group von Neumann algebras, Preprint, arxiv:1705.07350. I. Chifan, A. Ioana, Y. Kida: W∗-superrigidity for arbitrary actions of central quotients of braid groups, Math. Ann. 361 (2015), 925-959. [CJ82] A. Connes, V.F.R. Jones: A II1 factor with two non-conjugate Cartan subalgebras, Bull. Amer. Math. Soc. [CK15] 6 (1982), 211-212. I. Chifan, Y. Kida: OE and W* superrigidity results for actions by surface braid groups, Proceedings of the London Mathematical Society, in press. [Co76] A. Connes: Classification of Injective Factors Cases II1, II∞, IIIλ, λ 6= 1, Ann. of Math., 104 (1976), 73-115. [Co94] A. Connes: Noncommutative Geometry, Academic Press, 1994. [CP10] I. Chifan, J. Peterson: Some unique group-measure space decomposition results, Duke Math. J. 162 (2013), no. 11, 1923-1966. I. Chifan, T. Sinclair: On the structural theory of II1 factors of negatively curved groups, Ann. Sci. c. Norm. Supr. (4) 46 (2013), 1-33. [CS11] W∗-SUPERRIGIDITY FOR COINDUCED ACTIONS 19 [CSU11] I. Chifan, T. Sinclair, B. Udrea: On the structural theory of II1 factors of negatively curved groups, II. Actions by product groups, Adv. Math. 245 (2013), 208-236. [DHI16] D. Drimbe, D. Hoff, A. Ioana:Prime II1 factors arising from irreducible lattices in products of rank one simple Lie groups to apear in J. Reine. Angew. Math. [dHW14] P. de la Harpe, C. Weber: Malnormal subgroups and Frobenius groups: basics and examples, with an [Dr15] appendix by Denis Osin, Confluentes Math. 6 (2014), no. 1, 6576 D. Drimbe: Cocycle and orbit equivalence superrigidity for coinduced actions, to appear in Ergodic Theory Dynam. Systems. A. Furman: Orbit equivalence rigidity, Ann. of Math. (2) 150 (1999), no. 3, 1083-1108. [Fu99] [FV10] P. Fima, S. Vaes: HNN extensions and unique group measure space decomposition of II1 factors Trans. Amer. Math. Soc. 354 (2012) 2601-2617. [Ho15] D. Hoff: Von Neumann algebras of equivalence relations with nontrivial one-cohomology, J. Funct. Anal. 270 (2016), no. 4, 1501-1536. [GITD16] D. Gaboriau, A. Ioana, R. Tucker-Drob: Cocycle superrigidity for translation actions of product groups, summited, arXiv 1603.07616 (2016). [Gr91] M. Gromov: Asymptotic invariants of infinite groups, Geometric group theory, Vol. 2 (Sussex, 1991), London Math. Soc. Lecture Note Ser., vol. 182, Cambridge Univ. Press, Cambridge, 1993, pp. 1-295. [HPV10] C. Houdayer, S. Popa, S. Vaes: A class of groups for which every action is W∗-superrigid, Groups Geom. Dyn. 7 (2013), 577-590. [Io06a] A. Ioana: Rigidity results for wreath product of II1 factors, J. Funct. Anal. 252 (2007), 763-791. [Io06b] A. Ioana: Orbit inequivalent actions for groups containing a copy of F2, Invent. Math. 185 (2011), 55-73. A. Ioana: W∗-superrigidity for Bernoulli actions of property (T) groups J. Amer. Math. Soc. 24 (2011), [Io10] 1175-1226. [Io12a] A. Ioana: Classification and rigidity for von Neumann algebras, European Congress of Mathematics, EMS (2013), 601-625. [IPV10] A. Ioana, S. Popa, S. Vaes: A Class of superrigid group von Neumann algebras, Ann. of Math. (2) 178 (2013), 231-286 [Ka67] D. Kazhdan: On the connection of the dual space of a group with the structure of its closed subgroups, Funct. Anal. and its Appl. 1(1967), 63-65. [KV15] A. Krogager, S. Vaes: A class of II1 factors with exactly two crossed product decompositions, preprint arXiv:1512.06677. [Ma82] G. Margulis: Finitely-additive invariant measures on Euclidian spaces, Ergodic Theory Dynam. Systems 2(1982), 383-396. [MvN36] F. J. Murray, J. von Neumann, On rings of operators. Ann. of Math. 37 (1936). 116-229. [OP07] N. Ozawa, S. Popa: On a class of II1 factors with at most one Cartan subalgebra, Ann. of Math. (2) 172 [Pe09] [Po01] [Po03] [Po04] [Po05] (2010), no. 1, 713-749. J. Peterson: Examples of group actions which are virtually W∗-superrigid, Preprint. arXiv:1002.1745. S. Popa: On a class of type II1 factors with Betti numbers invariants, Ann. of Math. 163 (2006), 809-899. S. Popa: Strong rigidity of II1 factors arising from malleable actions of w-rigid groups. I., Invent. Math. 165 (2006), 369-408. S. Popa: Strong rigidity of II1 factors arising from malleable actions of w-rigid groups. II., Invent. Math. 165(2) (2006), 409-451. S. Popa: Cocycle and orbit equivalence superrigidity for malleable actions of w-rigid groups, Invent. Math. 170 (2007), 243-295. [Po06b] S. Popa: On a class of type II1 factors with Betti numbers invariants, Ann. of Math. 163 (2006), 809-889. S. Popa: On the superrigidity of malleable actions with spectral gap, J. Amer. Math. Soc. 21 (2008), 981- [Po06] 1000. S. Popa: Deformation and rigidity for group actions and von Neumann algebras, In Proceedings of the ICM (Madrid, 2006), Vol. I, European Mathematical Society Publishining House, 2007, 445-477. J. Peterson, T. Sinclair: On cocycle superrigidity for Gaussian actions Erg. Th. and Dyn. Sys.32 (2012), no. 1, 249-272. S. Popa, S. Vaes: Strong rigidity of generalized Bernoulli actions and computations of their symmetry groups, Adv. Math. 217 (2008), 833-872. S.Popa, S. Vaes: Group measure space decomposition of II1 factors and W -superrigidity, Invent. Math. 182 (2010), no. 2, 371-417. S. Popa, S. Vaes: Unique Cartan decomposition for II1 factors arising from arbitrary actions of free groups, Acta Mathematica 212 (2014), 141-198. [PV06] [PV09] [Po07] [PS09] [PV11] 20 DANIEL DRIMBE [PV12] S. Popa, S. Vaes: Unique Cartan decomposition for II1 factors arising from arbitrary actions of hyperbolic groups, Journal fur die reine und angewandte Mathematik, 690 2014, 433-458. [Ra72] M. S. Raghunathan: Discrete subgroups of Lie groups, Springer-Verlag, New York, 1972, Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 68. [RS10] G. Robertson,Tim Steger: Malnormal subgroups of lattices and the Pukanszky invariant in group factors, [Si55] [Va08] J. Funct. Anal. 258-8 (2010), 2708-2713. I. M. Singer: Automorphism of finite factors, Amer. J. Math. 77 (1955), 117-133. S. Vaes: Explicit computations of all finite index bimodules for a family of II1 factors, Ann. Sci. ´Ec. Norm. Sup´er. (4) 41 (2008), no. 5, 743-788. [Va10a] S. Vaes: Rigidity for von Neumann algebras and their invariants, Proceedings of the ICM (Hyderabad, India, 2010), Vol. III, Hindustan Book Agency (2010), 1624-1650. [Va10b] S Vaes: One-cohomology and the uniqueness of the group measure space decomposition of a II1 factor, Math. Ann. 355 (2013), 661-696. Mathematics Department; University of California, San Diego, CA 90095-1555 (United States). E-mail address: [email protected]
1807.08067	1	1807	2018-07-21T01:35:41	The suspension of a graph, and associated C*-algebras	[ "math.OA", "math.DS" ]	Given a directed graph E, we construct for each real number l a quiver whose vertex space is the topological realisation of E, and whose edges are directed paths of length l in the vertex space. These quivers are not topological graphs in the sense of Katsura, nor topological quivers in the sense of Muhly and Tomforde. We prove that when l = 1 and E is finite, the infinite-path space of the associated quiver is homeomorphic to the suspension of the one-sided shift of E. We call this quiver the suspension of E. We associate both a Toeplitz algebra and a Cuntz-Krieger algebra to each of the quivers we have constructed, and show that when l = 1 the Cuntz-Krieger algebra admits a natural faithful representation on the \ell^2-space of the suspension of the one-sided shift of E. For graphs E in which sufficiently many vertices both emit and receive at least two edges, and for rational values of l, we show that the Toeplitz algebra and the Cuntz-Krieger algebra of the associated quiver are homotopy equivalent to the Toeplitz algebra and Cuntz-Krieger algebra respectively of a graph that can be regarded as encoding the l-th higher shift associated to the one-sided shift space of E.	math.OA	math	THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS AIDAN SIMS Abstract. Given a directed graph E, we construct for each real number l a quiver whose vertex space is the topological realisation of E, and whose edges are directed paths of length l in the vertex space. These quivers are not topological graphs in the sense of Katsura, nor topological quivers in the sense of Muhly and Tomforde. We prove that when l = 1 and E is finite, the infinite-path space of the associated quiver is homeomorphic to the suspension of the one-sided shift of E. We call this quiver the suspension of E. We associate both a Toeplitz algebra and a Cuntz -- Krieger algebra to each of the quivers we have constructed, and show that when l = 1 the Cuntz -- Krieger algebra admits a natural faithful representation on the ℓ2-space of the suspension of the one-sided shift of E. For graphs E in which sufficiently many vertices both emit and receive at least two edges, and for rational values of l, we show that the Toeplitz algebra and the Cuntz -- Krieger algebra of the associated quiver are homotopy equivalent to the Toeplitz algebra and Cuntz -- Krieger algebra respectively of a graph that can be regarded as encoding the lth higher shift associated to the one-sided shift space of E. 1. Introduction Each finite directed graph E determines a corresponding 1-sided shift space (XE, σ). Cuntz -- Krieger algebras [5], and more generally graph C ∗-algebras [8, 18, 19], encode the dynamics (XE, σ) C ∗-algebraically, and there has been intense interest in these C ∗- algebras ever since their introduction -- see, for example [3, 4, 6, 7, 10, 30] and the bibli- ography of [25]. In addition to the shift space (XE, σ) itself, a directed graph E determines a family of dynamical systems indexed by the rational numbers. Given a rational number m/n with n > 0, we first form the directed graph Dn(E) obtained by inserting n − 1 new vertices along every edge of E -- so that each edge of E corresponds to a path of length n in Dn(E) -- and then consider the mth higher-power graph Dn(E)(0, m) of Dn(E), whose vertices are those of Dn(E) and whose edges are paths of length m in Dn(E). The shift space associated to this graph can be regarded as encoding the m/nth higher shift of E in the sense that the system (XDn(E)(0,m), σn Dn(E)(0,m)) decomposes into n disjoint copies of (XE, σm E ). All of these fractional higher shifts associated to a finite directed graph E are naturally encoded in the suspension of the base shift XE. The suspension of XE is the quotient M(σ) := (XE × [0, ∞))/∼ by the equivalence relation in which (x, t + m) ∼ (σm(x), t) for any positive integer m. For l ∈ [0, ∞), the map (x, t) 7→ (x, t + l) induces a continuous map ltl on M(σ). For positive rationals m/n, the restriction of ltm/n to {[x, t] : t ∈ 1 Z} n Date: June 7, 2021. 1991 Mathematics Subject Classification. Primary 46L05. Key words and phrases. Graph algebra, Cuntz -- Krieger algebra, symbolic dynamics, suspension flow. This research was supported by the Australian Research Council. 1 2 AIDAN SIMS is a copy of (XDn(E)(0,m), σDn(E)(0,m)). By analogy, we can regard the dynamics ltl for arbitrary l as corresponding to the lth higher shift of E. In this paper we construct from each locally finite directed graph E with no sources a family of C ∗-algebras parameterised by l ∈ R, and we prove that for l = m/n rational, and for graphs E in which enough vertices both emit and receive at least two edges, the corresponding C ∗-algebra is isomorphic to C([0, 1], C ∗(Dn(E)(0, m))). Our approach is to associate to each directed graph E a family of quivers S lE, one for each real parameter l. By a quiver here we mean a quadruple (Q0, Q1, r, s) where Q0 and Q1 are topological spaces, and r, s : Q1 → Q0 are continuous maps. The vertex space S lE0 of the quiver S lE is the topological realisation of E: it is the quotient of E1 × [0, 1] by the equivalence relation under which (e, 1) is glued to (f, 0) if s(e) = r(f ). The edge space S lE1 is, roughly speaking, the collection of directed paths of length l in the vertex space. We link this construction to suspension flows by showing that if E is finite, then the infinite-path space of S 1E is homeomorphic to the suspension flow M(σ) of the one-sided shift associated to E. Based on this, we call S 1E the suspension of E, and denote it SE. The quivers S lE, for l 6= 0, are typically not topological quivers in the sense of Muhly and Tomforde [23] because neither r nor s is typically an open map: if l > 0, then s is open map if and only if every vertex of E receives exactly one edge and r is open if and only if each vertex emits exactly one edge; if l < 0 then the roles of emitters and receivers are reversed. In particular, our quivers are not topological graphs in the sense of Katsura, and Cc(S lE1) does not admit a natural C0(S lE0)-valued inner-product. Thus Katsura's modification of Pimsner theory does not help us to construct a C ∗-algebra from S lE. The source map for the natural groupoid associated to the shift map on the infinite-path space of S lE is also typically not open, so this groupoid does not admit a Haar system, and we cannot employ Renault's theory of groupoid C ∗-algebras to construct a C ∗-algebra from S lE. Instead, we associate a Toeplitz algebra and a Cuntz -- Krieger algebra to each S lE by considering natural actions by bounded linear operators of C0(S lE0) and of Cc(S lE1) on the (nonseparable) Hilbert space ℓ2(S lE∗) with basis indexed by all finite paths in S lE. We define the Toeplitz algebra T C ∗(S lE) of S lE to be the C ∗-algebra generated by all of these operators. For each vertex ω of S lE, the subspace ℓ2(S lE∗ω), with ba- is invariant for T C ∗(S lE). Thus, for each sis indexed by paths whose source is ω, ω, we obtain a homomorphism from T C ∗(S lE) to the Calkin algebra Q(ℓ2(S lE∗ω)) = B(ℓ2(S lE∗ω))/K(ℓ2(S lE∗ω)). We define the Cuntz -- Krieger algebra C ∗(S lE) to be the im- age of T C ∗(S lE) inLω Q(ℓ2(S lE∗ω)) under the direct sum of these homomorphisms. If E has just one vertex v and one edge e (so that r(e) = s(e) = v), then S lE is the topological graph associated to the rotation-by-l homeomorphism of the circle R/Z, and C ∗(S lE) is isomorphic to the rotation algebra Al (see Example 6.8). Our main results give a complete description of each of T C ∗(S m/nE) and C ∗(S m/nE) for integers m ∈ Z and n ≥ 1 provided that enough vertices of E emit and receive at least two edges (the exact technical hypothesis is complicated, but a simple sufficient condition is that every vertex both emits and receives at least two edges). We prove that T C ∗(S m/nE) is homotopy equivalent to T C ∗(Dn(E)(0, m)), and that C ∗(S m/nE) is isomorphic to C([0, 1], C ∗(Dn(E)(0, m))). This suggests that the algebras C ∗(S lE) for irrational l are potential candidates for the role of the C ∗-algebras of the lth higher-power shifts of XE. THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 3 We obtain our description of T C ∗(S m/nE) and C ∗(S m/nE) in stages. We first reduce the problem to that of describing T C ∗(S mE) and C ∗(S mE) for nonnegative integers m. To do this, we first eliminate the case m = 0 by showing that T C ∗(S 0E) ∼= C0(SE0) ⊗ T , where T is the classical Toeplitz algebra, and that this isomorphism descends to an isomorphism C ∗(S 0E) ∼= C0(SE0) ⊗ C(T). We then show that T C ∗(S lE) ∼= T C ∗(S lEop) and C ∗(S lE) ∼= C ∗(S lEop) for negative l. We then show that for m > 0, we have T C ∗(S m/nE) ∼= T C ∗(S mDn(E)) and C ∗(S m/nE) ∼= C ∗(S mDn(E)). Thus for any graph E′ and any rational m/n, we have T C ∗(S m/nE′) ∼= T C ∗(S mE) and C ∗(S m/nE′) ∼= C ∗(S mE) for a suitable graph E. 0 := limtր0 at and a+ The bulk of the technical work in the paper goes into analysing T C ∗(S mE) and then C ∗(S mE) for a locally finite directed graph E with no sinks or sources. We do this by showing that they are both C(S)-algebras, where S is the circle R/Z, and analysing their fibres. We make use of the higher dual graphs E(1, m + 1) and E(0, m) studied by Bates [1]: E(1, m + 1) is the graph with vertices E1 and edges Em+1, and with range and source maps given by µ1 · · · µm+1 7→ µ1 and µ1 · · · µm+1 7→ µm+1 respectively; and E(0, m) is the graph with vertices E0 and edges Em and the usual range and source maps. It is relatively straightforward to show that the fibre of T C ∗(S mE) over each t ∈ S \ {0} is canonically isomorphic to T C ∗(E(1, m + 1)) and indeed that the ideal of T C ∗(S mE) corresponding to 0 ∈ S is isomorphic to C0((0, 1), T C ∗(E(1, m + 1)). The fibre over 0 ∈ S is more t for the space of paths in S mE whose source lies in the image complicated. Writing S mE∗ of E1 × {t} in SE0, we describe natural unitary transformations Ut : ℓ2(E(1, m + 1)∗) → ℓ2(S mE∗ t ) for t 6= 0. For each a ∈ T C ∗(S mE) and t 6= 0, this allows us to view the image of at in the fibre of T C ∗(S mE) over t as an operator on ℓ2(E(1, m + 1)∗). We prove that a− 0 := limtց0 at exist in B(ℓ2(E(1, m + 1))) and belong to the image of T C ∗(E(1, m + 1)), and that the map a 7→ a+ 0 descends to an injective homomorphism η from the fibre of T C ∗(S mE) over 0 to T C ∗(E(1, m+1))⊕T C ∗(E(1, m+ 1)). Following the arguments of [1], we show that there is a canonical injection 1,m+1 : T C ∗(E(0, m)) → T C ∗(E(1, m + 1)) that descends to the isomorphism C ∗(E(1, m + 1)) ∼= C ∗(E(0, m)) of [1, Theorem 3.1]. We then show that if the set of vertices of E that emit at least two edges has hereditary closure E0 in E(0, m), then the image of T C ∗(S mE)0 under η is precisely T C ∗(E(1, m + 1)) ⊕ 1,m+1(T C ∗(E(0, m))). It is then straightforward to obtain our description of T C ∗(S mE), and we prove that it is homotopy equivalent to T C ∗(SE(0, m)), allowing us to compute its K-theory. We prove that the inclusion T C ∗(S mE) ֒→ C([0, 1], T C ∗(E(1, m + 1))) takes T C ∗(S mE) ∩ ⊕ω∈SmE0K(ℓ2(S mE∗ω)) into C([0, 1], K(ℓ2(E(1, m + 1)∗))). We then deduce that C ∗(S mE) ∼= C([0, 1], C ∗(E(1, m + 1)). By showing that C ∗(Dn(E)(1, m + 1)) is Morita equivalent to C ∗(E(0, m)) when m, n are coprime, we describe K∗(C ∗(S lE)) for rational l provided that E is locally finite and enough vertices of E admit and receive at least two edges. A sufficient condition is that E is a finite, strongly connected graph that is not a simple cycle and has period 1 (in the sense of Perron -- Frobenius theory). 0 ⊕ a− The paper is organised as follows. We present the background that we will need on directed graphs and their C ∗-algebras, on topological graphs, and on C(X)-algebras in Section 2. We then present the definition of SE in Section 3. In Section 4 we describe the path space and the infinite-path space of SE, and prove that the latter coincides with M(σ) if E is finite. We define the C ∗-algebras of SE in Section 5. In Section 6, we describe our general construction of S lE and its C ∗-algebras, and prove that the case l = 1 4 AIDAN SIMS coincides with our previous definitions of SE and its C ∗-algebras, and also that l = −1 corresponds to the suspension of the opposite graph of E. This is not the most efficient order of presentation: we could have simply defined S lE and its C ∗-algebras immediately after Section 2, and then defined SE := S 1E. But we feel that SE, which is the key point of contact with suspension flows, is important enough to warrant separate discussion, and also that the later definitions of S lE and its C ∗-algebras are better motivated by first discussing SE and its C ∗-algebras. Section 6 also contains our reduction of the analysis of the C ∗-algebras of S lE for rational l to the analysis of the C ∗-algebras of S mE for positive integers m. Our analyses of T C ∗(S mE) and C ∗(S mE) occupy Sections 7 and 8 respectively. 2. Background We recall some background about directed graphs and their C ∗-algebras, on topological graphs, and on C(X)-algebras. 2.1. Graphs. We take our conventions for graph C ∗-algebras from [25]. Throughout this paper, all of the graphs that we consider are directed graphs in the sense that the edges have an orientation. We omit the adjective throughout. A graph is a quadruple E = (E0, E1, r, s) consisting of finite or countably infinite sets E0 and E1, and functions r, s : E1 → E0. We think of the elements of E0 as vertices, and draw them as dots, and we regard the elements of E1 as directed edges connecting vertices, and draw each as an arrow from the vertex s(e) the vertex r(e). It is convenient to think of E0 as the objects and E1 as the indecomposable morphisms in the countable category E∗ of finite paths in E as follows. For n ≥ 2, we define En := {µ1µ2 . . . µn : µi ∈ E1 and s(µi) = r(µi+1) for all i}, and refer to the elements of En as paths of length n in E. We think of vertices as paths of length 0 and edges as paths n=0 En for the path space of E. For v ∈ E0, we write r(v) = s(v) = v, and for n ≥ 1 and µ ∈ En we write r(µ) = r(µ1) and s(µ) = s(µn). We can concatenate µ, ν ∈ E∗ \ E0 to form the path µν if r(ν) = s(µ). For v ∈ E0 and µ ∈ E∗, the concatenation vµ is defined if r(µ) = v, in which case, we have vµ = µ, and similarly the concatenation µv is defined if v = s(µ), in which case µv = µ. If µ ∈ En, we write µ = n. of length 1, and then write E∗ :=S∞ Given U, V ⊆ E∗, we define UV := {µν : µ ∈ U, ν ∈ V, and s(µ) = r(ν)}. When U is a singleton U = {µ}, we write µV rather than {µ}V for the set {µν : ν ∈ V and r(ν) = s(µ)}. In particular, for v ∈ E0 we have vE1 = {e ∈ E1 : r(e) = v} and E1v = {e ∈ E1 : s(e) = v}. We extend this notational convention in the obvious ways, so that for example if v, w ∈ E0 then vE1w = vE1 ∩ E1w. The adjacency matrix of the graph E is the integer matrix AE given by AE(v, w) = vE1w. We then have An E(v, w) = vEnw for all v, w, n. We say that E is finite if E0 and E1 are both finite sets. We say that E is locally finite if each E1v ∪ vE1 is a finite set; that is, if the row-sums and column sums of the matrix AE are finite. A sink in E is a vertex v such that E1v = ∅, and a source is a vertex v such that vE1 = ∅. In this paper, we are concerned exclusively with graphs that are locally-finite and have no sources. THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 5 A cycle in E is a path µ ∈ E∗ \ E0 such that r(µ) = s(µ). We say that µ has an entrance if there exists i ≤ µ such that r(µi)E1 ≥ 2. 2.2. Infinite paths. An infinite path in E is a string x = x1x2x3 · · · of edges of E such that xi ∈ E1r(xi+1) for all i. We write E∞ for the set of all infinite paths in E and call it the infinite-path space of E. For x ∈ E∞, we define r(x) = r(x1) ∈ E0; and for µ ∈ E∗ and x ∈ E∞ with r(x) = s(µ) we write µx for the infinite path µ1 · · · µnx1x2 · · · . We write µE∞ := {µx : x ∈ E∞}. We endow E∞ with the topology that it inherits as a subspace ofQ∞ i=1 E∞, a basis for which is the collection {µE∞ : µ ∈ E∗}. The set µE∞ is called the cylinder set of µ and is often denoted Z(µ) elsewhere in the literature. When E is locally finite, the sets µE∞ are compact open sets in E∞, and the topology is a locally compact Hausdorff totally disconnected topology. The shift map σ : E∞ → E∞ is defined by σ(x)i = xi+1, so σ(x1x2x3 · · · ) = x2x3 · · · , and σ(ex) = x for all x ∈ E∞ and e ∈ E1 with s(e) = r(x). This σ is a local homeomor- phism, as it restricts to a homeomorphism eE∞ → s(e)E∞ for each e ∈ E1. 2.3. Graph C ∗-algebras. Let E be a locally finite graph with no sources. A Toeplitz -- Cuntz -- Krieger family for E in a C ∗-algebra A consists of a map t : E1 → A, written e 7→ te and a map q : E0 → A, written v 7→ qv such that the elements qv are mutually orthogonal projections in A, and such that (TCK1) t∗ ete = qs(e) for all e ∈ E1, and (TCK2) qv ≥Pe∈vE1 tet∗ (CK) qv =Pe∈vE1 tet∗ e for all v ∈ E0. e for all v ∈ E0. A Cuntz -- Krieger family for E is a Toeplitz -- Cuntz -- Krieger family (t, q) for E such that Relations (TCK1) and (TCK2) imply that for each µ ∈ En the element tµ = tµ1tµ2 . . . tµn is a partial isometry. As a notational convenience, we write tv := qv for v ∈ E0. With this notation, the C ∗-algebra generated by the elements te and the elements qv is equal to the closed linear span C ∗(t, q) = span{tµt∗ ν : µ, ν ∈ E∗, s(µ) = s(ν)}, and we have tµtν = δs(µ),r(ν)tµν. An induction shows that t∗ we have µtµ = ts(µ) for all µ ∈ E∗, and tµt∗ νtηt∗ ζ = tµt∗ ζν′ tµη′t∗ ζ 0 if ν = ην′ if η = νη′ otherwise. There is a C ∗-algebra T C ∗(E) generated by a Toeplitz -- Cuntz -- Krieger family (T, Q) that is universal in the sense that given any other Toeplitz -- Cuntz -- Krieger family (t, q) in a C ∗-algebra A, there is a homomorphism πt,q : T C ∗(E) → A such that πt,q(Te) = te and πt,q(Qv) = qv. The universal property ensures that there is an action γ : T → Aut(T C ∗(E)) called the gauge action such that γz(Te) = zTe and γz(Qv) = Qv for all e ∈ E1 and v ∈ E0. There is a faithful representation π : T C ∗(E) → B(ℓ2(E∗)) called the path-space rep- resentation, and determined by π(Te)hµ = δs(e),r(µ)heµ and π(Qv)hµ = δv,r(µ)hµ. (The existence of π follows from the universal property, and injectivity follows from an appli- cation of [9, Theorem 4.1].) 6 AIDAN SIMS There is also a C ∗-algebra C ∗(E) generated by a Cuntz -- Krieger E-family (s, p) that is universal in the sense that given any other Cuntz -- Krieger family (s′, p′) there is a homomorphism of C ∗(E) taking each se to s′ v. This C ∗(E) is isomorphic to the quotient of T C ∗(E) by the ideal IE generated by the projections ∆v := Qv − e and each pv to p′ Pe∈vE1 TeT ∗ e indexed by v ∈ E0. Let E be a locally finite graph with no sources and let (t, q) be a Cuntz -- Krieger E- family. Since the projections ∆v are fixed by the gauge action on T C ∗(E), it descends to an action, also called the gauge action and denoted γ, on C ∗(E). The gauge-invariant uniqueness theorem [11, 25] states that if there is an action β : T → Aut(C ∗(t, q)) such that βz(te) = zte for all e ∈ E1, then πq,t : C ∗(E) → C ∗(t, q) is injective if and only if each qv is nonzero. The Cuntz -- Krieger uniqueness theorem [5, 25] says that if every cycle in E has an entrance, then πq,t is injective if and only if each qv is nonzero. With a little work, one can check that the path-space representation of T C ∗(E) carries the ideal IE to π(T C ∗(E)) ∩ K(ℓ2(E∗)). It follows that there is a homomorphism from C ∗(E) to the Calkin algebra Q(ℓ2(E∗)) := B(ℓ2(E∗))/K(ℓ2(E∗)) given by pv 7→ Qv + K(ℓ2(E∗)) and se 7→ Te+K(ℓ2(E∗)). We call this homomorphism the Calkin representation of C ∗(E). An argument using the gauge-invariant uniqueness theorem shows that the Calkin representation of C ∗(E) is injective. The subspaces ℓ2(E∗v) ⊆ ℓ2(E∗) indexed by A set H ⊆ E0 is called hereditary if s(HE1) ⊆ H, or equivalently HE∗ ⊆ E∗H. the Calkin representation can be regarded as an injective homomorphism of C ∗(E) into v ∈ E0 are invariant for π, and we have π(T C ∗(E)) ∩ K(ℓ2(E∗)) =Lv∈E0 K(ℓ2(E∗v)). So Lv∈E0 Q(ℓ2(E∗v)) given by se 7→Lv(Teℓ2(E∗v) + K(ℓ2(E∗v))). 2.4. Dual graphs. Given a locally finite graph E with no sources, the dual graph bE is the graph bE = (E1, E2, r, s) where r(ef ) = e and s(ef ) = f for all ef ∈ E2. The from E to bE and vice versa. There is a homeomorphism E∞ ∼= bE∞ that carries the infinite path x = x1x2x3 · · · of E to the infinite path x = (x1x2)(x2x3)(x3x4) · · · of bE. Corollary 2.5 of [2] shows that there is an isomorphism C ∗(bE) ∼= C ∗(E) satisfying e for all ef ∈ bE1 and all e ∈ bE0. properties of being row-finite or locally finite and of having no sinks or no sources pass More generally (see [1]), we can construct from any pair of integers 0 < p < q a new graph E(p, q) from E. We define E(p, q) to be the graph with f and pe 7→ ses∗ sef 7→ sesf s∗ E(p, q)0 := Ep and E(p, q)1 := Eq, with range and source maps given by and r(e1) if p ≥ 1 if p = 0 sp,q(e1 · · · eq) :=(eq−p+1 · · · eq Bates shows in [1, Theorem 3.1] that C ∗(E(p + 1, q + 1)) ∼= C ∗(E(p, q)) for all 0 < p < q, and hence, by induction, that C ∗(E(p, q)) ∼= C ∗(E(0, q −p)) for all 0 < p < q, generalising rp,q(e1 · · · eq) :=(e1 · · · ep So E(0, 1) ∼= E, E(1, 2) ∼= bE, and E(0, p) is the pth higher-power graph (E0, Ep, r, s). the usual isomorphism C ∗(bE) ∼= C ∗(E) of [2, Corollary 2.5]. We will need the following jections ∆v := Qv −Pe∈vE1 TeT ∗ Recall that for any graph E, we denote by IE the ideal of T C ∗(E) generated by the pro- e indexed by v ∈ E0, and that then C ∗(E) = T C ∗(E)/IE. analogue of this result for Toeplitz algebras. if p ≥ 1 if p = 0. s(eq) THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 7 Lemma 2.1. Let E be a row-finite graph with no sources and fix integers 0 < p < q. For v ∈ E0 let qv := Pµ∈vEp Qµ ∈ T C ∗(E(p, q)), and for µ ∈ Eq−p = E(0, q − p)1, let tµ :=Pν∈s(µ)Ep Tµν ∈ T C ∗(E(p, q)). Then there is an injective homomorphism p,q : T C ∗(E(0, q − p)) ֒→ T C ∗(E(p, q)) such that p,q(Qv) = qv for v ∈ E0, and p,q(Te) = te for e ∈ E1. We have (2.1) ∆µ for all v ∈ E0, p,q(∆v) = Xµ∈vEp and p,q(IE(0,q−p)) ⊆ IE(p,q). There is an isomorphism p,q : C ∗(E(0, q − p)) → C ∗(E(p, q)) such that p,q(a + IE(0,q−p)) = p,q(a) + IE(p,q) for all a ∈ T C ∗(E(0, q − p)). Proof. The elements qv are mutually orthogonal projections in T C ∗(E(p, q)) because {Qµ : µ ∈ E(p, q)0} is a collection of mutually orthogonal projections. For µ, ν ∈ Eq−p, we have t∗ µtν = Xα∈s(µ)Ep,β∈s(ν)Ep T ∗ µαTνβ = Xα∈s(µ)Ep,β∈s(ν)Eq δµα,νβQα = δµ,ν Xα∈s(µ)Ep Qα = δµ,νqs(µ). Hence (t, q) satisfies (TCK1), and the partial isometries {tµ : µ ∈ Eq−p} have mutually orthogonal range projections. For v ∈ E0 and µ ∈ vEq−p, we have qvtµ = Xν∈vEq−p Qν Xα∈s(µ)Ep Tµα = Xν∈vEq−p,α∈s(µ)Ep δµ,νTµα = Xα∈s(µ)Ep Tµα = tµ. µ ≤ qr(µ), and since the tµt∗ In particular, each tµt∗ µ are mutually orthogonal, it follows that (q, t) satisfies (TCK2). So the universal property of T C ∗(E(0, q − p)) gives a homomor- phism p,q : T C ∗(E(0, q − p)) → T C ∗(E(p, q)) such that p,q(Qv) = qv for v ∈ E0, and p,q(Te) = te for e ∈ E1. To see that this homomorphism is injective, observe that for v ∈ E0, we have tνt∗ (2.2) qv − Xν∈vEq−p ν = Xµ∈vEp = Xµ∈vEp Qµ − Xν∈vEq−p Xα∈s(ν)Ep Qµ − Xλ∈vEq µ ∈ vEp, we have π(Qµ −Pα∈s(µ)Eq−p TµαT ∗ π(Qµ −Pα∈s(µ)Eq−p) 6= 0. Hence qv −Pµ∈vEq−p tµt∗ shows that p,q is injective. TλT ∗ Let π : T C ∗(E(p, q)) → B(ℓ2(E(p, q)∗)) be the path-space representation. For each µα) = θhµ,hµ ∈ K(ℓ2(E(p, q)∗)), and hence µ 6= 0. Theorem 4.1 of [9] therefore TναT ∗ να λ = Xµ∈vEp(cid:16)Qµ − Xβ∈s(µ)Eq−p TµβT ∗ µβ(cid:17). The calculation (2.2) establishes (2.1), and since p,q(IE) is an ideal of (T C ∗(E(0, q − p))), it follows that it is contained in the ideal of T C ∗(E(p, q)) generated by the ∆µ, which is IE(p,q) by definition. So p,q descends to a homomorphism p,q : C ∗(E(0, q − p)) → C ∗(E(p, q)) satisfying p,q(pv) = p,q(Qv + IE(0,q−p)) = p,q(Qv) + IE(p,q) = Xµ∈vEp p,q(sν) = p,q(Tν + IE(0,q−p)) = p,q(Tν) + IE(p,q) = Xα∈s(ν)Ep pµ, and sνα. This is an isomorphism because it agrees on generators with that obtained from [1, Corol- lary 3.3]. (cid:3) 8 AIDAN SIMS 2.5. Topological graphs and their C ∗-algebras. We present here a very brief intro- duction to those parts of Katsura's theory of topological graphs and their C ∗-algebras that we will need later. For a more comprehensive overview, see [25, Chapter 8]; for full details, see [14, 15, 16, 17]. As defined by Katsura [14], a topological graph is a quadruple E = (E0, E1, r, s) where E0 and E1 are locally compact Hausdorff spaces, r : E1 → E0 is a continuous map, and s : E1 → E0 is a local homeomorphism. The associated graph bimodule is defined as follows. The space Cc(E1) is a C0(E0)- bimodule with respect to the actions(cid:0)a · ξ(cid:1)(e) = a(r(e))ξ(e) and(cid:0)ξ · a(cid:1)(e) = ξ(e)a(s(e)) for a ∈ C0(E0) and ξ ∈ Cc(E1). The formula hξ ηiC0(E0)(v) :=Ps(e)=v ξ(e)η(e) defines a C0(E0)-valued inner-product on Cc(E1), and the completion X(E) of Cc(E1) in the norm kξk = khξ, ξiAk1/2 is a Hilbert C0(E0)-bimodule called the graph correspondence of E. Katsura proves that X(E) is equal to the space Cd(E1) of functions (2.3) Cd(E1) :=nξ ∈ C0(E1) :(cid:16)v 7→ Xs(e)=v ξ(e)2(cid:17) ∈ C0(E)0o. The left action of C0(E0) on X(E) determines a homomorphism φ : C0(E0) → L(X(E)) by φ(a)ξ = a·ξ. If r : E1 → E0 is a proper map in the sense that the preimages of compact sets are compact, then φ takes values in K(X(E)). A representation of X(E) in a C ∗-algebra A is a pair (π, ψ) such that π : C0(E0) → A is a homomorphism, ψ : X(E) → A is a linear map, and we have π(a)ψ(ξ) = ψ(a · ξ), ψ(ξ)π(a) = ψ(ξ · a) and ψ(ξ)∗ψ(η) = π(hξ, ηiC0(E0)) for all ξ, η ∈ X(E). The pair (π, ψ) induces a homomorphism ψ(1) : K(X(E)) → A such that ψ(1)(θξ,η) = ψ(ξ)ψ(η)∗ for all ξ, η ∈ X(E) (see [24, Lemma 3.2]). The representation (ψ, π) is Cuntz -- Pimsner covariant if we have ψ(1)(φ(a)) = π(a) whenever φ(a) ∈ K(X(E)); in particular, if r : E1 → E0 is proper, then (ψ, π) is Cuntz -- Pimsner covariant if ψ(1) ◦ φ = π. The topological graph C ∗-algebra, denoted here by C ∗(E) is the C ∗-algebra generated by a universal Cuntz -- Pimsner covariant representation of E. Its Toeplitz algebra T C ∗(E) is the C ∗-algebra generated by a universal representation of E. If E is a graph and Y is a locally compact Hausdorff space then the topological graph E × Y is defined by (E × Y )0 := E0 × Y , (E × Y )1 := E1 × Y and r(e, y) := (r(e), y) and s(e, y) := (s(e), y). If E is locally finite, then r : (E × Y )1 → (E × Y )0 is a proper map. The universal properties of C ∗(E × Y ) and of the tensor product C ∗(E) ⊗ C0(Y ) imply (see [15, Proposition 7.7]) that there is an isomorphism C ∗(E) ⊗ C0(Y ) ∼= C ∗(E × Y ) that carries se ⊗ h to ψ(cid:0)(f, y) 7→ δe,f h(y)(cid:1) and carries pv ⊗ h to π(cid:0)(w, y) 7→ δv,wh(y)(cid:1) for all e ∈ E1, v ∈ E0 and h ∈ C0(Y ). 2.6. C(X)-algebras. We need only the bare bones of the theory of C(X)-algebras here. For details, see [32, Appendix C]. Let X be a locally compact Hausdorff space. A C ∗-algebra A is called a C(X)-algebra if there exists a nondegenerate homomorphism ι : C(X) → ZM(A) of C(X) into the centre of the multiplier algebra of A. For each x ∈ X, the maximal ideal Jx = {f ∈ C(X) : f (x) = 0} of C(X) generates an ideal Ix = ι(Jx)A of A. We define Ax := A/Ix to be the corresponding quotient. For a ∈ A, the map x 7→ ka+ Ixk is upper semicontinuous. There is a unique topology on A :=Fx∈X Ax under which the functions x 7→ a + Ix are THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 9 all continuous. In this topology, (2.4) lim n→∞ ka + Ixnk = 0 =⇒ lim n→∞ a + Ixn = 0. So A is an upper-semicontinuous bundle of C ∗-algebras over X, and each a ∈ A determines a section γa : x 7→ a + Ix of A that vanishes at infinity in the sense that for each ε > 0 there exists a compact set K ⊆ X such that kγa(x)k < ε for all x 6∈ K. The map a 7→ γa is an isomorphism of A onto the algebra Γ0(X, A) of continuous sections of A that vanish at infinity. So every C(X)-algebra is the algebra of sections of an upper-semicontinuous bundle of C ∗-algebras over X. Conversely, if A is an upper-semicontinuous bundle over X then there is a nondegenerate homomorphism ι : C(X) → ZM(Γ0(X, A)) characterised by (ι(f )γ)(x) = f (x)γ(x). 3. The suspension of a graph In this section we define the suspension SE of a graph E and describe its basic prop- erties. Our motivation is the relationship between the infinite-path space of SE and the suspension flow of the shift-space associated to E, which we establish in Section 4. The constructions in this section will be subsumed by the more-general construction of the quivers S lE parameterised by l ∈ R in Section 6. Let E be a locally finite graph with no sources. Let ∼ denote the smallest equivalence relation on ((E∗ \ E0) × [0, 1]) ⊔ E∗ such that (3.1) (µf, 0) ∼ µ and (eµ, 1) ∼ µ for all µ ∈ E∗, e ∈ E1r(µ) and f ∈ s(µ)E1. Observe that ∼ restricts to an equivalence relation on (En+1 × [0, 1]) ∪ En for each n ≥ 0. For µ ∈ E∗ \ E0 and t ∈ [0, 1], we write [µ, t] for the equivalence-class of (µ, t) under ∼, and for µ ∈ E∗ we write [µ] for the equivalence class of µ under ∼. Putting µ = v ∈ E0 in (3.1), we see that [e, 1] ∼ [f, 0] whenever s(e) = r(f ). For n ∈ N = {0, 1, 2, . . . }, we write It is straightforward to check that there are well-defined maps r, s : SE∗ → SE0 such that r([eµ, t]) := [e, t] and s([µf, t]) := [f, t] for µ ∈ E∗ and e ∈ E1r(µ) and f ∈ s(µ)E1. These maps satisfy r([µ]) = [r(µ)] and s([µ]) = [s(µ)] for µ ∈ E∗. Definition 3.1. Let E be a locally finite graph with no sources. We call the quadruple SE := (SE0, SE1, r, s) the suspension of E. The following lemma will make it easier to work with the suspension of a graph. Lemma 3.2. Let E be a locally finite graph with no sources. For µ, ν ∈ E∗ and s, t ∈ [0, 1], we have (µ, s) ∼ (ν, t) if and only if one of the following holds: (1) µ = ν and s = t; (2) s = t = 0 and µ1 · · · µµ−1 = ν1 · · · νν−1, or s = t = 1 and µ2 · · · µµ = ν2 · · · νν; or (3) s = 1, t = 0 and µ2 · · · µµ = ν1 · · · νν−1, or s = 0, t = 1 and µ1 · · · µµ−1 = ν2 · · · νν. and we define SEn :=(cid:0)(En+1 × [0, 1]) ⊔ En(cid:1)/∼, SE∗ :=(cid:0)((E∗ \ E0) × [0, 1]) ⊔ E∗(cid:1)/∼ =Gn SEn. 10 AIDAN SIMS Each α ∈ SE∗ has a representative of the form (µ, t) where µ ∈ E∗ and t ∈ [0, 1). If α 6∈ {[µ] : µ ∈ E∗} then this representative is unique. The map µ 7→ [µ] is an injection of E∗ into SE∗. Proof. Consider the relation R0 on (E∗ \ E0) × [0, 1] given by (1) -- (3). Then R0 ⊆ ∼. It is clear that R0 is reflexive and symmetric, and a quick case-by-case check of the possible combinations of (2) and (3) shows that it is transitive. If (µ, s) ∼ (ν, t), then there is a sequence (µ, s) = (µ0, s0) ∼ ν1 ∼ (µ1, s1) ∼ ν2 ∼ · · · ∼ νk ∼ (µk, sk) = (ν, t) where each of the equivalences in the chain is one of the forms appearing in (3.1). It then follows that each equivalence (µi, si) ∼ (µi+1, si+1) is of one of the forms appearing in (2) or (3). Thus ∼ is contained in R0, and the two are equal. If α ∈ SE∗ then by definition we have either α = [µ] for some µ ∈ E∗ or α = [ν, t] for some (ν, t) ∈ (E∗ \ E0) × [0, 1]. If α = [µ] then, since E has no sources, we can find e ∈ E1 with r(e) = s(µ), and then α = [µe, 0]. Likewise if α = [ν, 1] then we can find f ∈ E1 with r(f ) = s(ν), and then α = [νf, 0]. Otherwise α = [ν, t] already has the desired form. If α 6∈ {[µ] : µ ∈ E∗}, then α = [ν, t] for some t ∈ (0, 1) and ν ∈ E∗ \ E0, and then if we also have α = [ν′, s] then in particular (ν, t) ∼ (ν′, s) with t 6= {0, 1}. In particular, this equivalence does not appear in (2) or (3), and we deduce that it is of the form (1), so ν = ν′ and t = s. Suppose that µ, ν ∈ E∗ satisfy µ ∼ ν. We can write [µ] = [µe, 0] and [ν] = [νf, 0] for any e ∈ s(µ)E1 and f ∈ s(ν)E1. We then have (µe, 0) ∼ (νf, 0) which means that the equivalence is of the form (1) or the first of the forms appearing in (2), each of which forces µ = ν. So µ 7→ [µ] is an injection. (cid:3) We call elements of SE∗ paths in SE, and elements of paths of length n in SE. SEn :=(cid:0)(En+1 × [0, 1]) ⊔ En(cid:1)/∼ Notation 3.3. Using Lemma 3.2, we regard E∗ as a subset of SE∗. In particular, for n ∈ N, we write SEn \ En = {[µ, t] : µ ∈ En+1 and t ∈ (0, 1)}. There is a partially defined composition map on SE∗ as follows: the pair ([µ, t], [ν, s]), is composable if and only if s([µ, t]) = r([µ, s]), which in particular forces t = s. If s([µ, t]) = r([ν, t]), and if these representatives have been chosen with t 6= 1 as above, then the composition is defined as [µ, t][ν, t] = [µ1 · · · µµ−1ν1 · · · νν, t]. It then follows that for µ, ν ∈ E∗ with s(µ) = r(ν), we have [µ][ν] = [µν]. We also have r([µ, t][ν, t]) = r([µ, t]) and s([µ, t][ν, t]) = s([ν, t]) whenever [µ, t][ν, t] makes sense. Under this concatenation, SE∗ is a small category with objects SE0. Given α = [µ, t] ∈ SE∗ with t ∈ [0, 1), we can write µ = µ1 · · · µn+1 with each µi ∈ E1. Defining αi := [µiµi+1, t] ∈ SE1 for i ≤ n, we obtain a factorisation α = α1 · · · αn. This is the unique factorisation of α as a composition of edges of SE. As with directed graphs, given subsets U, V ⊆ SE∗, we write UV := {αβ : α ∈ U, β ∈ V and s(α) = r(β)}. If U is a singleton U = {α}, then we write αV and V α in place of {α}V and V {α}. In particular, for ω ∈ SE0 and a subset U ⊆ SE∗ we have ωU = U ∩ r−1(ω) and Uω = U ∩ s−1(ω). THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 11 Throughout the paper, we write S for the circle R/Z. Each element of S has a unique representative in [0, 1). We often abuse notation slightly and regard elements of [0, 1] as elements of S (so 1 and 0 are equal as elements of S). Lemma 3.4. Let E be a locally finite graph with no sources. There is a continuous map : SE∗ → S such that ([µ, t]) = t for µ ∈ E∗ \ E0 and t ∈ [0, 1). Proof. Lemma 3.2 shows that [µ, t] 7→ t is well-defined from SE∗ to S. To see that it is continuous, let q :(cid:0)(E∗ \ E0) × [0, 1]) → SE0 be the quotient map. For 0 < a < b < 1, we have q−1(cid:0)−1((a, b))(cid:1) = (E∗ \ E0) × (a, b), which is open. So −1((a, b)) is open. Likewise, for ε < 1/2, we have q−1(cid:0)−1((−ε, ε))(cid:1) = (E∗ \ E0) × ([0, ε) ∪ (1 − ε, 1]), which again is open. So −1((−ε, ε)) is open. (cid:3) Observe that the map of Lemma 3.4 satisfies (α) = (s(α)) = (r(α)) for all α ∈ SE∗. Notation 3.5. For t ∈ S, we define SE∗ t = −1(t) = {[µ, t] : µ ∈ E∗}, and for n ∈ N, we define SEn t We then have SEn SE∗ t = (SEn)(SE0 := SEn ∩ SE∗ t ) = (SE0 t = {[µ, t] : µ ∈ En+1}. t )(SEn) for all n, t. With this notation, SE∗ 1 = 0 = {[µ] : µ ∈ E∗}. We aim to construct a C ∗-algebra from SE. The following lemma shows that we can- not employ Katsura's theory of topological-graph C ∗-algebras, or Muhly and Tomforde's theory of topological-quiver C ∗-algebras: to get off the ground, both theories require at least that the source map s is an open map. Lemma 3.6. Let E be a finite graph with no sources. The maps s, r : SE1 → SE0 are continuous maps, and restrict to local homeomorphisms from SE1 \ E1 to SE0 \ E0. For e ∈ E1, the map s is open at [e] ∈ SE1 if and only if E1s(e) = 1, and the map r is open at [e] if and only if r(E)E1 = 1. In particular, s is an open map if and only if E1v = 1 for all v ∈ E0, and r is an open map if and only if vE1 = 1 for all v ∈ E0. Proof. The range and source maps are continuous by construction. If 0 < t < 1 and ef ∈ E2, then for any ε such that (t − ε, t + ε) ⊆ (0, 1), the restrictions of s, r to {[ef, s] : t − s < ε} are homeomorphisms onto open sets. Fix e ∈ E1. To see that s is open at [e] if and only if E1s(e) = 1, first suppose that E1s(e) = 1, so E1s(e) = {e}. Then the sets Uε := {[f e, t] : f ∈ E1r(e), t > 1 − ε} ∪ {[ef, t] : f ∈ s(e)E1, t < ε} indexed by ε ∈ (0, 1/2) form a neighbourhood base at [e] and we have s(Uε) = {[e, t] : t > 1 − ε} ∪ {[f, t] : f ∈ s(e)E1, t < ε}. Since E1s(e) = {e}, we deduce that, writing v = s(e), s(Uε) = {[f, t] : f ∈ E1v, t > 1 − ε} ∪ {[f, t] : f ∈ vE1, t < ε}, which is a basic open neighbourhood of [v]. So s is open at [e]. Now suppose that E1s(e) ≥ 2, say f ∈ E1s(e) \ {e}, and write v := s(e). Consider the set U := {[he, t] : h ∈ E1r(e), 1/2 < t ≤ 1} ∪ {[eh, t] : h ∈ vE1, 0 ≤ t < 1/2}. 12 AIDAN SIMS This set is open in the quotient topology. We have s(U) = {[e, t] : 1/2 < t ≤ 1} ∪ {[h, t] : h ∈ vE1, 0 ≤ t < 1/2}. In particular, we have [v] = [e, 1] ∈ s(U), but the sequence ([f, (n − 1)/n])∞ in the complement of s(U) and converges to [v]. So s is not open at [e]. n=1 is contained This completes the proof that s is open if and only if each E1v = 1. A symmetric argument (or the same argument applied to the opposite graph) shows that r is open at [e] if and only if r(e)E1 is a singleton. The first statement implies in particular that r, s are open at α for every α ∈ SE1 \ E1, (cid:3) so the final statement follows immediately. Example 3.7. Consider the simplest example of a finite graph with no sources: E0 = {v} and E1 = {e}, so r(e) = s(e) = v. Then the map of Lemma 3.4 restricts to a homeomorphism : [e, t] 7→ t from SE0 to S. For each w ∈ SE0 there is a unique fw ∈ SE1 with r(fw) = s(fw) = w, and then SE1 = {fw : w ∈ SE0} ∼= S, and s and r are homeomorphisms. Example 3.8. Now consider the finite graph such that E0 = {v} and E1 = {e, f }, so r(e) = s(e) = r(f ) = s(f ) = v. So SE0 is equal to the union {e, f } × S of two circles glued at a point by gluing (e, 0) to (f, 0). For g ∈ E1 and t ∈ S, consider the vertex [g, t] ∈ SE0. We have SE1[g, t] = {[eg, t], [f g, t]}, and the ranges of these edges are [e, t] and [f, t] respectively. So as a set, we have SE1 ∼= {e, f }×SE0. The sequences ([ee, 1−1/n])∞ n=1 and n=1 converge in SE1 to [ee, 1] = [f e, 1], and s([ee, 1 − 1/n]) = [e, 1 − 1/n] = ([f e, 1 − 1/n])∞ s([f e, 1 − 1/n]) for all n, so s is not a local homeomorphism at [ee, 1]. 4. The infinite-path space of the suspension of a graph In this section we describe the infinite-path space of the suspension of E, and we show that if E is finite, then SE∞ is homeomorphic to the one-sided suspension flow of the shift space of the graph. In Section 5 we will define the Toeplitz algebra T C ∗(SE) and the Cuntz -- Krieger algebra C ∗(SE) of the suspension of a graph E using analogues of the path-space representation and Calkin representation of a graph C ∗-algebra. We will then link this to symbolic dynamics by showing that C ∗(SE) has a natural representation on ℓ2(SE∞). An infinite path in SE is a sequence α1α2α3 · · · of edges αi ∈ SE1 such that r(αi+1) = s(αi) for all i. We write SE∞ for the set of all infinite paths in SE. Lemma 4.1. Let E be a locally finite graph with no sources. There is a bijection θ∞ : E∞ × [0, 1) → SE∞ such that θ∞(e1e2e3 · · · , t) = [e1e2, t][e2e3, t][e3e4, t] · · · for all e1e2 · · · ∈ E∞ and t ∈ [0, 1). Proof. To see that θ∞ is surjective, fix ξ = α1α2 · · · in SE∞. Write each αi = (eifi, ti) with ti ∈ [0, 1) as in Lemma 3.2. Then ti = t1 for all i. Let t := t1. If t 6= 0, then s(αi) = r(αi+1) forces [fi, t] = [ei+1, t], and hence Lemma 3.2 gives fi = ei+1, for all i. Thus x = e1e2e3 · · · ∈ E∞ and we have ξ = θ∞(x, t). If t = 0, then s(αi) = r(αi+1) forces s(ei) = r(ei+1) for all i, and then since t = 0 we have αi = [eiei+1, 0] for each i. So again, x = e1e2 · · · belongs to E∞ and ξ = θ∞(x, t). For injectivity, suppose that θ∞(x, s) = θ∞(y, t) = α1α2 · · · . Put write x = e1e2 · · · and y = f1f2 · · · . Then [e1e2, s] = α1 = [f1f2, t], and since s, t 6= 1, Lemma 3.2 forces s = t. The definition of ≈ THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 13 shows that if s, t 6= 1 and [ef, s] = [gh, t] then e = g. Since [eiei+1, s] = αi = [fifi+1, s] for all i, we deduce that ei = fi for all i, and so θ∞ is injective. (cid:3) We next describe a natural topology on SE∞. Lemma 4.2. Let E be a locally finite graph with no sources. For µ ∈ E∗ and 0 < a < b < 1, let Z(µ, (a, b)) := {θ∞(x, t) : x ∈ µE∞ and t ∈ (a, b)}, and for µ ∈ E∗ and 0 < ε < 1 2 , let Z(µ, ε) := {θ∞(eµx, t) : e ∈ E1r(µ), x ∈ s(µ)E∞, and t ∈ (1 − ε, 1)} ∪ {θ∞(µx, t) : x ∈ s(µ)E∞ and t ∈ [0, ε)} Then there is a second-countable Hausdorff topology on SE∞ with basis B = {Z(µ, (a, b)) : µ ∈ E∗ and 0 < a < b < 1} ∪ {Z(µ, ε) : µ ∈ E∗ and 0 < ε < 1 2 }. Proof. To see that B is a basis, first observe that for t 6= 0 any element of the form θ∞(x, t) belongs to Z(r(x), (a, b)) for any 0 < a < t < b < 1; and any element of the form θ∞(x, 0) belongs to Z(x1, 1 2) for any e ∈ E1 with s(e) = r(x), soS B = SE∞. Given µ, ν ∈ E∗, we define µ ∨ ν := if µ = νµ′ µ if ν = µν′ ν ∞ otherwise, where ∞ is used here purely as a formal symbol. As a notational convenience, we define Z(∞, (a, b)) = ∅ = Z(∞, ε) for any a, b, ε. We then have Z(µ, (a, b)) ∩ Z(ν, (c, d)) = Z(µ ∨ ν, (max{a, c}, min{b, d})), Z(µ, ε) ∩ Z(ν, δ) = Z(µ ∨ ν, min{ε, δ}), and Z(µ, (a, b)) ∩ Z(ν, ε) =(cid:16) [e∈E1r(ν) Z(cid:0)µ ∨ eν, (a, b) ∩ (1 − ε, 1)(cid:1)(cid:17) ∪ Z(cid:0)µ ∨ ν, (a, b) ∩ [0, ε)(cid:1). So B is a base for a topology on SE∞. This topology is second countable because restricting the values of a, b, and ε to rational values in the definition of B yields a countable base for the same topology. To see that this topology is Hausdorff, fix distinct elements θ∞(x, s) and θ∞(y, t) of SE∞. First suppose that s = t. Then x 6= y so we can find µ, ν ∈ E∗ \ E0 such that x ∈ µE∞, y ∈ νE∞ and µ ∨ ν = ∞. If s 6= 0 then for any 0 < a < s < b < t, the sets Z(µ, (a, b)) and Z(ν, (a, b)) are disjoint neighbourhoods of θ∞(x, s) and θ∞(y, t). If s = 0, then that µ, ν 6∈ E0 implies that eµ ∨ f ν = ∞ for any e ∈ E1r(µ) and f ∈ E1r(ν). We already have µ∨ν = ∞, so we deduce that Z(µ, 1 2) are disjoint neighbourhoods of θ∞(x, s) and θ∞(y, t). Now suppose that s 6= t; without loss of generality, s 6= 0. Fix µ with x ∈ µE∞ and y ∈ νE∞. Choose 0 < a < s < b < 1 such that t 6∈ (a, b). If t = 0, then for any ε < min{a, 1 − b} and any f ∈ E1 with s(f ) = r(ν), the sets Z(µ, (a, b)) and Z(f ν, ε) are disjoint neighbourhoods of θ∞(x, s) and θ∞(y, t); and if t 6= 0 then for any 2 ) and Z(ν, 1 14 AIDAN SIMS 0 < c < t < d < 1 such that (a, b) ∩ (c, d) = ∅, the sets Z(µ, (a, b)) and Z(ν, (c, d)) are disjoint neighbourhoods of θ∞(x, s) and θ∞(y, t). (cid:3) Let ∼σ be the equivalence relation on E∞ × [0, 1] defined by (x, s) ∼σ (y, t) if and only if either x = y and s = t or y = σ(x), s = 1 and t = 0 or x = σ(y), s = 0 and t = 1; that is, the smallest equivalence relation such that (x, 1) ∼σ (σ(x), 0) for all x ∈ E∞. The suspension of (E∞, σ) is the topological quotient space For t ∈ [0, 1], we write M(σ) := (E∞ × [0, 1])/∼σ. M(σ)t := {[x, t] : x ∈ E∞} ⊆ M(σ). ∼= E∞. So M(σ)1 = M(σ)0 Remark 4.3. We can identify M(σ) with the quotient space (E∞ × [0, ∞))/∼σ where (x, s) ∼σ (y, t) if and only if s − ⌊s⌋ = t − ⌊t⌋ and σ⌊s⌋(x) = σ⌊t⌋(y). The identification sends [x, t] ∈ Mσ to the corresponding class Jx, tK in (E∞ × R)/∼σ; the inverse sends Jx, tK to [σ⌊t⌋(x), t − ⌊t⌋]. Proposition 4.4. Let E be a finite graph with no sources. The suspension M(σ) is a compact Hausdorff space, and the map θ∞ : E∞ × [0, 1) → SE∞ described in Lemma 4.1 induces a homeomorphism of M(σ) onto SE∞. Proof. The equivalence classes for ∼σ in E∞ × [0, 1] are finite: if t 6∈ {0, 1} then the equivalence class of (x, t) is a singleton; if t = 0 then the equivalence class of (x, t) is {(x, 0)} ∪ {(ex, 1) : e ∈ E1r(x)}, and if t = 1 then the equivalence class of (x, t) is {(σ(x), 0)} ∪ {(eσ(x), 1) : e ∈ E1r(σ(x))}. In particular, the ∼σ-equivalence classes in E∞ × [0, 1] are discrete, and so the quotient topology on M(σ) is Hausdorff. Since E∞ × [0, 1] is compact, so is M(σ). The quotient map q from E∞ ×[0, 1] to M(σ) restricts to a bijection from E∞ ×[0, 1) to M(σ), and so θ∞ induces a bijection τ from M(σ) to SE∞ such that τ (q(x, t)) = θ∞(x, t) for x ∈ E∞ and t ∈ [0, 1). Since SE∞ is Hausdorff and M(σ) is compact, to see that τ is a homeomorphism, it suffices to show that it is continuous. For this, observe that for µ ∈ E∗ and 0 < a < b < 1, we have q−1(cid:0)τ −1(cid:0)Z(µ, (a, b))(cid:1)(cid:1) = µE∞ × (a, b), and so τ −1(cid:0)Z(µ, (a, b))(cid:1) is open by definition of the quotient topology. Similarly, q−1(cid:0)τ −1(Z(eµ, ε))(cid:1) =(cid:0)eµE∞ × (1 − ε, 1](cid:1) ∪(cid:0)µE∞ × [0, ε)(cid:1), which again is open in E∞ × [0, 1]; so again by definition of the quotient topology, τ −1(Z(eµ, ε)) is open, and hence τ is continuous. (cid:3) 5. The C ∗-algebras of the suspension of a graph In this section we define two C ∗-algebras associated to the suspension of a graph E. We define the first of these algebras in terms of a concrete representation on a non-separable Hilbert space, for which we use the following notational convention. Notation 5.1. Throughout the rest of the paper, given any set X, we write ℓ2(X) := {f : X → C : f −1(C \ {0}) is countable, and Xx∈X v(x)2 < ∞}, which is a Hilbert space with inner product given by hf, gi =Px∈X v(x)w(x). We denote the canonical basis elements of ℓ2(X) by {hx : x ∈ X}; so hx(y) = δx,y for x, y ∈ X. THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 15 Remark 5.2. Using Notation 3.5, the set SEn of En in SE∗. In particular, there is a unitary U0 : ℓ2(E∗) → ℓ2(SE∗ 0 = SEn 1 is the canonical copy {[µ] : µ ∈ En} 0) satisfying U0hµ = h[µ]. For t ∈ (0, 1), and for each n ∈ N, there is a bijection SEn t → bEn given by [µ, t] 7→ (µ1µ2)(µ2µ3) · · · (µnµn+1). In particular, for each t ∈ (0, 1) there is a unitary Ut : ℓ2(bE∗) → Uth(µ1µ2)(µ2µ3)···(µµ−1µµ) = h[µ,t]. t ) given by ℓ2(SE∗ Lemma 5.3. Let E be a locally finite graph with no sources. There is an injective non- degenerate representation ρ : C0(SE0) → B(ℓ2(SE∗) such that ρ(a)hα = a(r(α))hα for all a ∈ C0(SE0) and α ∈ SE∗. There is a linear map ψ : Cc(SE1) → B(ℓ2(SE∗)) such that (5.1) ξ(β)hβα for all ξ ∈ Cc(SE1) and α ∈ SE∗. ψ(ξ)hα = Xβ∈SE1r(α) We have kψ(ξ)k ≤ kξk∞ · {ef ∈ E2 : ξ([ef, t]) 6= 0 for some t ∈ [0, 1]}. Proof. The representation ρ is the direct-sum of the representations a 7→ a(ω) Idℓ2(ωSE∗) indexed by ω ∈ SE0. It is nondegenerate because for each α ∈ SE∗ and any a ∈ C0(SE0) with a(r(α)) = 1 we have ρ(a)hα = hα. To see that it is injective, note that kρ(a)k ≥ supω∈SE0 kρ(a)hωk = kak∞. To see that there is a linear map ψ as claimed, let π be the path-space representation of T C ∗(E). For t ∈ (0, 1), the operator At := Xef ∈ bE1 ξ([ef, t])π(tetf t∗ f ) ∈ π(T C ∗(E)) ⊆ B(ℓ2(E∗)) satisfies kAtk ≤Pef ∈ bE1 ξ([ef, t]) ≤ ξ∞ · {ef ∈ E2 : ξ([ef, t]) 6= 0 for some t} because f ) are partial isometries. Similarly, the operator the π(tetf t∗ A0 := Xe∈E1 ξ([e])π(te) Ut : ℓ2(SE∗ satisfies kA0k ≤ Pe∈E1 ξ([e]) ≤ ξ∞ · {ef ∈ E2 : ξ([ef, t]) 6= 0 for some t}. Let Remark 5.2. Let U :=L Ut : ℓ2(SE∗) →Lt∈[0,1) ℓ2(E∞). Then At(cid:17)U ∈ B(ℓ2(SE∗)) t ) → ℓ2(bE∗), 0 < 1 < t, and U0 : ℓ2(SE∗ ψ(ξ) := U ∗(cid:16) Mt∈[0,1) 0 ) → ℓ2(E∗) be the unitaries of satisfies (5.1). The map ψ∞ thus defined is clearly linear, and satisfies the desired norm estimate because the At all do. (cid:3) We are now ready to define the Toeplitz algebra of SE. Definition 5.4. Let E be a locally finite graph with no sources. We define T C ∗(SE) to be the C ∗-subalgebra of B(ℓ2(SE∗)) generated by ρ(C0(SE0)) and ψ(Cc(SE1)). To define C ∗(SE) we first need to observe that the operators ρ(a) and ψ(ξ) above respect the fibration of ℓ2(SE∗) over S. 16 AIDAN SIMS Lemma 5.5. Let E be a locally finite graph with no sources. For each ω ∈ SE0, the subspace ℓ2(SE∗ω) ⊆ ℓ2(SE∗) is invariant for T C ∗(SE). In particular, each ℓ2(SE∗ t ) is t , we have T C ∗(SE)ℓ2(SE∗ω) ∩ invariant for T C ∗(SE). For each t ∈ S and each ω ∈ SE0 K(ℓ2(SE∗ t )) = K(ℓ2(SE∗ω)). Proof. For a ∈ C0(SE0), ξ ∈ Cc(SE1) and α ∈ SE∗, the element ρ(a)hα is a scalar multiple of hα. We have ψ(ξ)hα = Pβ∈SE1r(α) ξ(β)hβα, and a quick calculation using inner-products shows that ψ(ξ)∗hα =(ξ(α1)hα2···αα 0 if α ≥ 1 otherwise. t ) =Lω∈SE0 t In particular, ρ(a)hα, ψ(ξ)hα and ψ(ξ)∗hα all belong to ℓ2(SE∗s(α)). Since the elements ρ(a) and ψ(ξ) generate T C ∗(SE), it follows that each ℓ2(SE∗ω) is invariant. Consequently each ℓ2(SE∗ ω) is invariant as well. ℓ2(SE∗ To prove the final statement, we first show that θω,ω ∈ T C ∗(SE)ℓ2(SE∗ t . For this, fix ω ∈ SE0. Fix a ∈ C0(SE0) such that aSE0 t ) for each t ∈ S and ω ∈ SE0 = δω (this is t is discrete in SE1, t is a discrete subset of SE0). Likewise, the set SE1 possible since SE0 so for each α ∈ ωSE1 we can choose ξα ∈ Cc(SE1) such that ξαSE1 = δα. Using the formulas described in the preceding paragraph for the actions of ρ(a) and the ψ(ξα) and their adjoints on basis elements, we see that t t θhω,hω =(cid:16)ρ(a) − Xα∈ωSE1 ψ(ξα)ψ(ξα)∗(cid:17)(cid:12)(cid:12)(cid:12)ℓ2(SE∗ t ) ∈ T C ∗(SE)ℓ2(SE∗ t ). Now fix t ∈ S, ω ∈ SE0 t is discrete, we choose ξ1, . . . , ξm ∈ Cc(SE1) such that ξiSE1 t , and α ∈ SE∗ω \ {ω}. Factor α = α1 · · · αm where each αi ∈ SE1. Using that SE1 = δαi. Again calculating with basis vectors, we see that θhα,hω = ψ(ξ1) · · · ψ(ξm)θhω,hω ∈ T C ∗(SE)ℓ2(SE∗ t ). Since each ℓ2(SE∗ω) is invariant for T C ∗(SE), we deduce that θhω,hα = t ). Now for arbitrary α, β ∈ SE∗ω we have θhω,hω(ψ(ξ1) · · · ψ(ξm))∗ ∈ T C ∗(SE)ℓ2(SE∗ θhα,hβ = θhα,hωθhω,hβ ∈ T C ∗(SE)ℓ2(SE∗ t ). t It follows that K(ℓ2(SE∗ ω)) ⊆ T C ∗(SE)ℓ2(SE∗ Since we also know that each ℓ2(SE∗ω) ⊆ ℓ2(SE∗ reverse containment as well. t ) ∩K(ℓ2(SE∗ t )) for each t ∈ S and ω ∈ SE0 t . t ) is invariant for T C ∗(SE), we have the (cid:3) Given a Hilbert space H, we write Q(H) for the Calkin algebra B(H)/K(H). Definition 5.6. Let E be a locally finite graph with no sources. We define ρ : C0(SE0) → Lt∈S Q(ℓ2(SE∗ t )) by and for ξ ∈ Cc(SE1), we define ρ(a) =Mt∈S ψ(ξ) =Mt∈S ρ(a)ℓ2(SE∗)t + K(ℓ2(SE∗ t )), ψ(ξ)ℓ2(SE∗)t + K(ℓ2(SE∗ t )). We define C ∗(SE) to be the C ∗-subalgebra ofLt∈S Q(ℓ2(SE∗ and ψ(Cc(SE1)). t )) generated by ρ(C0(SE0)) THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 17 Remark 5.7. For each t ∈ S and each ω ∈ SE0 t ) is invariant for T C ∗(SE). It follows that there is an injective homomorphism from C ∗(SE) t , the subspace ℓ2(SE∗ω) ⊆ ℓ2(SE∗ toLω∈SE0 Q(ℓ2(SE∗ω)) that carriesLt∈S(cid:0)aℓ2(SE∗ K(ℓ2(SE∗ω))(cid:1) for all a ∈ T C ∗(SE). t ) +K(ℓ2(SE∗ t ))(cid:1) toLω∈SE0(cid:0)aℓ2(SE∗ω) + We link our definition of C ∗(SE) to the suspension of one-sided shift of E using the infinite-path space of SE described in the preceding section. Proposition 5.8. Let E be a locally finite graph with no sources, and suppose that every cycle in E has an entrance. Then there is a faithful representation Θ : C ∗(SE) → ℓ2(M(σ)) such that for x ∈ E∞ and t ∈ [0, 1) Θ(ρ(a))h[x,t] = a([x1, t])h[x,t] for all a ∈ C0(SE0) and Θ( ψ(ξ))h[x,t] = Xe∈E1r(x) ξ([e, t])h[ex,t] for all ξ ∈ Cc(SE1). Proof. Since every cycle in E has an entrance, the Cuntz -- Krieger uniqueness theorem [19, Theorem 3.7] shows that the infinite-path-space representations π∞ : C ∗(E) → B(ℓ2(E∞)) Hence θ := π∞ ◦ π−1 Q : πQ(C ∗(E)) → π∞(C ∗(E)) is an isomorphism that carries Qv + K(ℓ2(E∗)) to Pv := projℓ2(vE∞) and carries Te + K(ℓ2(E∗)) to Se : hx 7→ δs(e),r(x)hex. Similarly, θ := π∞ ◦ π−1 and π∞ : C ∗(bE) → B(ℓ2(bE∞)) are both faithful. As discussed in Section 2.3, the Calkin representations πQ : C ∗(E) → Q(ℓ2(E∗)) and πQ : C ∗(bE) → Q(ℓ2(bE∗)) are also faithful. Q : πQ(C ∗(bE)) → π∞(C ∗(bE)) is an isomorphism carrying Qe + K(ℓ2(bE∗)) to Pe := projℓ2(e bE∞) and carrying Tef + K(ℓ2(bE∗)) to Sef : hx 7→ δf,r(x)h(ef )x. θ) : C ∗(SE) → ℓ2(E∞) ⊕(cid:0)Lt∈S\{0} ℓ2(bE∞)(cid:1) is a faithful It follows that θ ⊕ (Lt∈S\{0} t ) → ℓ2(bE∗), 0 < 1 < t, and U0 : ℓ2(SE∗ ℓ2(E∗) be the unitaries of Remark 5.2, and let U :=L Ut : ℓ2(SE∗) →Lt∈[0,1) ℓ2(E∞). Then Θ(x) := U ∗(cid:0)θ ⊕ (Lt∈S\{0}bθ)(cid:1)(x)U defines a faithful representation of C ∗(SE) on Lt∈S ℓ2(SE∞ t ) = ℓ2(SE∞). Direct calculation using the definitions of ρ, ψ, θ and θ shows (cid:3) As in the proof of Lemma 5.3, let Ut : ℓ2(SE∗ that Θ satisfies the prescribed formulae. representation. 0) → 6. Fractional higher-power graphs and their C ∗-algebras In this section, given a graph E, we generalise the construction of Section 3 by con- structing, for each real number l a quiver S lE in such a way that S 1E = SE and S −1E = SEop. To each S lE we associate a Toeplitz algebra T C ∗(S lE) and a C ∗-algebra C ∗(S lE) so that T C ∗(S 1E) ∼= T C ∗(SE) and T C ∗(S −1E) ∼= T C ∗(SEop), and similarly at the level of Cuntz -- Krieger algebras. We show that T C ∗(S 0E) ∼= C0(SE0) ⊗ T and C ∗(S 0E) ∼= C0(SE0) ⊗ C(T). We then show that if l = m n is rational, then there is a graph F , closely related to E, such that T C ∗(S lE) ∼= T C ∗(S mF ), and this isomorphism descends to an isomorphism C ∗(S lE) ∼= C ∗(S mF ). This motivates the next section, in which we give a concrete description of each of T C ∗(S mE) and C ∗(S mE) for a large class of graphs E and all positive integers m. Combining this result with those of the current section yields an 18 AIDAN SIMS explicit description of T C ∗(S lE) and C ∗(S lE) for all rational l and for a large class of graphs E. We use the following notation: given integers 0 ≤ m ≤ n ≤ L, if µ = µ1µ2 . . . µL ∈ EL, then µ(m, n) =(µm+1 · · · µn r(µm+1) if n > m if n = m. It will be convenient in this section to use the following alternative description of SE0: it is homeomorphic to the quotient ofFn≥0 En × [0, n] by the equivalence relation (µ, s) ∼ (ν, t) if s − ⌊s⌋ = t − ⌊t⌋ and µ(⌊s⌋, ⌈s⌉) = ν(⌊t⌋, ⌈t⌉) (see [12]). Definition 6.1. Let E be a locally finite graph with no sources. Fix l ∈ R. Let ≈l be the equivalence relation on {(µ, t) : µ ∈ E∗ and t, t + l ∈ [0, µ]} given by (µ, s) ≈l (ν, t) if and only if s − ⌊s⌋ = t − ⌊t⌋ and (µ(⌊s⌋, ⌈s + l⌉) = ν(⌊t⌋, ⌈t + l⌉) µ(⌊s + l⌋, ⌈s⌉) = ν(⌊t + l⌋, ⌈t⌉) if l ≥ 0, if l ≤ 0. We define S lE1 := {(µ, t) : µ ∈ E∗ and t, t + l ∈ [0, µ]}/≈l. We write [µ, t]l for the equivalence class of (µ, t) under ≈l. Define rl, sl : SEl → SE0 by rl([µ, t]l) := [µ, t]l and sl([µ, t]l) = [µ, t + l]l. The quadruple (SE0, S lE1, r, s) is denoted S lE. Remark 6.2. The special case l = 1 coincides with the space SE1 and range and source maps r, s : SE1 → SE0 of Section 3. When l = −1 there is a homeomorphism from the space S −1E1 to the space (SEop)1 associated to the opposite graph of E satisfying [ef, t] 7→ [f opeop, 1 − t], and this homeomorphism intertwines r−1 with r : (SEop)1 → (SEop)0 and intertwines s−1 with s : (SEop)1 → (SEop)0. When l = 0, we see that S 0E1 is a copy of the vertex space, and r0, s0 are both just the identity map SE0 → SE0. Notation 6.3. We will frequently just write r, s in place of rl, sl, and [µ, t] in place of [µ, t]l, when the parameter l is clear from context. As with SE, the quivers S lE are not usually topological graphs, except if l = 0. Lemma 6.4. Let E be a locally finite graph with no sources. The range and source maps s0 and r0 on S 0E are homeomorphisms. Fix l ∈ R \ {0}. Then rl and sl are continuous maps. If l > 0, then the map sl is an open map if and only if each E1v = 1 and rl is an open map if and only if each vE1 = 1. If l < 0, then the map sl is open if and only if each vE1 = 1, and rl is open if and only if each E1v = 1. Proof. The final statement of Remark 6.2 shows that s0 and r0 are homeomorphisms. The proof of the remaining statements is very similar to that of Lemma 3.6. (cid:3) sl([µi, ti]) = rl([µi+1, ti+1]) for all 1 ≤ i < k. We write S lE∗ for the space F∞ A path in S lE is a sequence [µ1, t1][µ2, t2][µ3, t3] · · · [µk, tk] of elements of S lE1 such that i=0 S lEi of all paths in S lE, including the vertices, which are regarded as paths of length 0. So S lE0 = SE0. We define r(v) = s(v) = v for each v ∈ SE0, and for α = [µ1, t1][µ2, t2][µ3, t3] · · · [µk, tk] ∈ S lEk we define r(α) = r([µ1, t1]) and s(α) = s([µk, tk]). THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 19 t ) = SE0 Remark 6.5. For each t ∈ S we write S lEn t ). t We have r(S lEn t−l, where addition on the subscript is modulo Z. If l = m ∈ Z is an integer, then as in Remark 5.2, the space S mEn 1 is the canonical copy {[µ] : µ ∈ Enm} of Enm in S mE∗. In particular there is a unitary U0 : ℓ2(E(0, m)∗) → ℓ2(S mE∗ 0 ) satisfying for {α ∈ S lEn : (s(α)) = t} = (S lEn)(SE0 0 = S mEn U0hµ = h[µ]. For m ≥ 0, t ∈ (0, 1), and each n ∈ N, there is a bijection S mEn t → E(1, m + 1)n such that [µ, t] 7→ (µ1 · · · µm+1)(µm · · · µ2m+1) · · · (µ(n−1)m · · · µnm+1). Thus for each t ∈ (0, 1) there is a unitary Ut : ℓ2(E(1, m + 1)∗) → ℓ2(S mE∗ t ) given by Uth(µ1···µm+1)(µm···µ2m+1)···(µ(n−1)m···µnm+1) = h[µ,t] The space S lE1 is endowed with the quotient topology inherited from SE⌈l⌉+1 × [0, 1]. Lemma 6.6. Let E be a locally finite graph with no sources and fix l ∈ R. Let {hα : α ∈ S lE∗} denote the canonical orthonormal basis for ℓ2(S lE∗). There is an injective nondegenerate representation ρl : C0(SE0) → B(ℓ2(S lE∗) such that ρl(a)hα = a(r(α))hα for all a ∈ C0(SE0) and α ∈ S lE∗, and there is a linear map ψl : Cc(S lE1) → B(ℓ2(S lE∗)) such that ψl(ξ)hα = Xβ∈S lE1r(α) We have ξ(β)hβα for all ξ ∈ Cc(S lE1) and α ∈ S lE∗. (6.1) kψl(ξ)k ≤ kξk∞ · {µ ∈ E⌈l⌉+1 : ξ([µ, t]) 6= 0 for some t ∈ [0, 1]}. For each ω ∈ SE0, the subspace ℓ2(S lE∗ω) ⊆ ℓ2(S lE∗) is invariant for the C ∗-subalgebra of B(ℓ2(S lE∗)) generated by the images of ρl and ψl. Proof. The first two statements follow from an argument almost identical to that of Lemma 5.3. The final statement follows from an argument nearly identical to the proof of the first statement of Lemma 5.5. (cid:3) Definition 6.7. Let E be a locally finite graph with no sources. (1) We define T C ∗(S lE) := C ∗(cid:0)ρl(C0(SE0)) ∪ ψl(Cc(S lE1))(cid:1), the C ∗-subalgebra of B(ℓ2(S lE∗)) generated by the images of ρl and ψl. (2) We define ρl : C0(SE0) → ⊕ω∈SE0Q(ℓ2(S lE∗ω)) by and we define ψl : Cc(S lE1) → ⊕ω∈SE0Q(ℓ2(S lE∗ω)) by ρl(a) = Mω∈SE0(cid:16)ρ(a)ℓ2(S lE∗ω) + K(ℓ2(S lE∗ω))(cid:17) ψl(ξ) = Mω∈SE0(cid:16)ψ(ξ)ℓ2(S lE∗ω) + K(ℓ2(S lE∗ω))(cid:17). (3) We define C ∗(S lE) to be the subalgebra of Lω∈SE0 Q(ℓ2(S lE∗ω)) generated by ρl(C0(SE0)) and ψl(Cc(S lE1)). 20 AIDAN SIMS Example 6.8. If E consists of a single vertex v and a single edge e, then each S lE is a copy of the topological graph Fl := (S, S, t 7→ t−l, id) determined by the rotation homeomorphism t 7→ t − l of S. We have Cc(S lE1) = C(S lE1) = C(F 1 l ) = X(Fl), the topological-graph bimodule of Fl. It is routine to verify that (ρl, ψl) is a representation of X(Fl) in the sense of Section 2.5. So there is a surjective homomorphism ψl ×ρl : T C ∗(Fl) → T C ∗(S lE) that carries iX(Fl)(ξ) to ψl(ξ) and carries iC(S)(a) to ρl(a). The space S lE∗ can be identified with S × N by the map that sends α ∈ S lEn to (s(α), n). Under this identification, the map a 7→ ρl(a)(1 − ψl(1)ψl(1)∗) is the canonical faithful representation of C(S) on ℓ2(S × {1}) ∼= ℓ2(S), so the uniqueness theorem [9, Theorem 2.1] shows that ψl × ρl is injective. Therefore T C ∗(S lE) is isomorphic to the Toeplitz algebra T C ∗(Fl). Since 1 − ψl(1)ψl(1)∗ belongs to Lt∈S K(S lE∗t), we see that ψl(1) ψl(1)∗ = 1 in C ∗(S lE). For any a ∈ C(S lE0) the left action of a on Fl is given by a · ξ = θa,1(ξ), and so we see that if φ : C(S lE0) → L(Fl) denotes the homomorphism implementing the left action, we have ρl(a) − ψ(1) l (φ(a)) = ρl(a) − ψl(a) ψl(1)∗ = ρl(a)(ρl(1) − ψl(1) ψl(1)∗) = 0. Hence ( ψl, ρl) is a covariant representation of Fl, and therefore induces a homomorphism ψl × ρl : C ∗(Fl) → C ∗(S lE). The action of T on B(ℓ2(S lE∗)) determined by conjuga- tion by the unitaries {Wz : z ∈ T} given by Wz(ht,n) = znht,n induces an action on Lt∈S Q(ℓ2(S lE∗t)), and it is routine to check that ψl × ρl is equivariant for this action : a 7→ Lω a(ω)1Q(ℓ2(S lE∗ω)) is injective, it and the gauge action on C ∗(Fl). Since ρl follows from the gauge-invariant uniqueness theorem [14, Theorem 4.5] that ψ × ρ is injective. So C ∗(S lE) ∼= C ∗(Fl). By [15, Proposition 10.5] there is an isomorphism of the rotation algebra Al -- the universal C ∗-algebra generated by unitaries U, V such that UV = e2πilV U -- onto C ∗(Fl) that carries U to iC(S)(t 7→ e2πit) and carries V to iX(F1)(1). ∼= C ∗(S lE) that carries U to ρl(t 7→ e2πit) So we deduce that there is an isomorphism Al and carries V to ψl(1S lE1). Remark 6.9. More generally, if E is a row-finite graph with no sources and satisfying E1v = 1 for all v, then for each l > 0 the quadruple F := (SE0, SEl, r, s) is a topological graph in the sense of Katsura, and an analysis like that of Example 6.8 shows that T C ∗(S lE) and C ∗(S lE) coincide with the topological-graph C ∗-algebras T C ∗(F ) and C ∗(F ) respectively. In the next few sections we will give a recipe for describing both T C ∗(S lE) and C ∗(S lE) for rational values of l provided that sufficiently many vertices of E both emit and receive at least two edges. We do not yet have a concrete description of C ∗(S lE) for arbitrary l ∈ R and an arbitrary graph E. The following theorem relates the constructions described in this section with those of Section 3 and Section 5. For the following result, we denote by T the classical Toeplitz algebra generated by a non-unitary isometry S. Theorem 6.10. Let E be a locally finite graph with no sources. (1) There is an isomorphism T C ∗(S 1E) ∼= T C ∗(SE) that carries ρ1(a) to ρ(a) for a ∈ C0(SE0) and carries ψ1(ξ) to ψ(ξ) for ξ ∈ Cc(S 1E1). This isomorphism descends to an isomorphism C ∗(S 1E) ∼= C ∗(SE). (2) There is an isomorphism T C ∗(S 0E) ∼= C0(SE0)⊗T that carries ρ0(a) to a⊗1 and 0 ) ⊗ S for ξ ∈ Cc(S 0E1) = Cc(SE0), and this isomorphism carries ψ0(ξ) to (ξ ◦ r−1 descends to an isomorphism C ∗(SE0) ∼= C0(SE0) ⊗ C(T). THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 21 (3) There is an isomorphism T C ∗(S −1E) ∼= T C ∗(SEop) that carries ρ−1(a) to the el- ement ρ([eop, t] 7→ a([e, 1 − t])) for a ∈ C0(SE0) and carries ψ−1(ξ) to ψ(cid:0)[ef, t] 7→ ξ([f opeop, (1 − t)])(cid:1) for ξ ∈ Cc(S −1E1). This isomorphism descends to an isomor- phism C ∗(S −1E) ∼= C ∗(SEop). Proof. The proofs of the first and third statements are almost identical. For the first statement, observe that the identification S 1E1 ∼= SE1 of Remark 6.2 intertwines r, s with r1 and s1, and so induces a homeomorphism S 1E∗ ∼= SE∗, which induces a unitary U1 : ℓ2(S 1E∗) ∼= ℓ2(SE∗). This unitary intertwines ρ and ρ1 and intertwines ψ and ψ1, and so AdU1 restricts to the desired isomorphism T C ∗(S 1E) ∼= T C ∗(SE). Since U1 carries each ℓ2(S 1E∗ω) to ℓ2(SE∗ω), the map AdU1 carries each K(ℓ2(S 1E∗ω)) to K(ℓ2(SE∗ω)), and so descends to an isomorphism C ∗(S 1E) ∼= C ∗(SE) as claimed. For the third statement, we argue exactly the same way, using the homeomorphism S −1E1 ∼= (SEop)1 of Remark 6.2 to induce a unitary U−1 : ℓ2(S −1E∗) ∼= ℓ2((SEop)∗). 0 = π(ξ ◦ r−1 For the second statement, first identify ℓ2(S 0E∗) with ℓ2(SE0)⊗ℓ2(N) by the unitary U0 that carries h[µ,t] to h[µ1,t] ⊗hµ−1. Let π : C0(SE0) → B(ℓ2(SE0)) be the canonical faithful representation π(a)hω = a(ω)hω. Direct calculation shows that U0ρ0(a)U ∗ 0 = π(a) ⊗ id, and U0ψ0(ξ)U ∗ 0 ) ⊗ S for a ∈ C0(SE0) and ξ ∈ Cc(S 0E1). So AdU0 carries T C ∗(S 0E) onto the subalgebra of B(ℓ2(SE0) ⊗ ℓ2(N)) generated by products of the form (π(a) ⊗ id)(id ⊗S). This is precisely the tensor product C0(SE0) ⊗ T . Moreover, AdU0 carries each K(ℓ2(S 0E∗ω)) to K(Chω ⊗ ℓ2(N)), and so it carries the kernel of the quotient map T C ∗(S 0E) → C ∗(S 0E) to C0(SE0) ⊗ K(ℓ2(N)) ⊳ C0(SE0) ⊗ T . It therefore descends to an isomorphism C ∗(S 0E) ∼= (C0(SE0) ⊗ T )/(C0(SE0) ⊗ K) ∼= C0(SE0) ⊗ (T /K) ∼= C0(SE0) ⊗ C(T). (cid:3) The remainder of the section is devoted to reducing the study of the C ∗-algebras T C ∗(S lE) and C ∗(S lE) for rational values of l to the study of the C ∗-algebras T C ∗(S mF ) and C ∗(S mF ) for nonnegative integers m and appropriate graphs F . We will analyse these latter in the next two sections. Our first step is to show that we need only consider l ≥ 0 by showing that T C ∗(S lE) ∼= T C ∗(S lEop) for l ∈ (−∞, 0). Lemma 6.11. Let E be a locally finite graph with no sources. Fix l ∈ (−∞, 0). There is an isomorphism T C ∗(S lE) ∼= T C ∗(S lEop) that carries ρl(a) to ρl(cid:0)[eop, t] 7→ a([e, 1 − t])(cid:1) for a ∈ C0(SE0) and carries ψl(ξ) to ψl(cid:0)[µop, t] 7→ ξ([µ, µ − t])(cid:1) for ξ ∈ Cc(S lE1). This isomorphism descends to an isomorphism C ∗(S lE) ∼= C ∗(S lEop). Proof. This follows the argument of Theorem 6.10(3): The map [µ, t] 7→ [µop, µ − t] defines homeomorphism τl of S lE∗ onto (S lEop)∗ that intertwines the range and source maps. This homeomorphism determines a unitary U : ℓ2(S lE∗) → ℓ2(S lEop, conjugation by which implements an isomorphism T C ∗(S lE) ∼= T C ∗(S lEop) that satisfies the desired formulae. Since U(ℓ2(S lE∗ω)) = ℓ2(S lEopτl(ω)) for each ω ∈ SE0, we have UK(ℓ2(S lE∗ω))U ∗ = K(ℓ2(S lEopτl(ω))) for each ω. Hence AdU descends to the desired isomorphism C ∗(S lE) ∼= C ∗(S lEop). (cid:3) In the remainder of the section we must show that if m ∈ N and n ∈ N\{0}, then there is nE) ∼= T C ∗(S mF ) and similarly at the level of Cuntz -- Krieger a graph F such that T C ∗(S m algebras. 22 AIDAN SIMS Given a graph E and an integer n ≥ 1, the nth delay of E is the graph Dn(E) described as follows. We set Dn(E)0 := E0 ⊔ {we,j : e ∈ E1, 1 ≤ j ≤ n − 1} Dn(E)1 := {fe,j : e ∈ E1, 1 ≤ j ≤ n}. and The range and source maps are given by r(fe,j) =(we,j−1 r(e) if j ≥ 2 if j = 1 and s(fe,j) =(we,j s(e) if j < n if j = n. In words, Dn(E) is the graph obtained by inserting n − 1 new vertices along each edge of E. The example below pictures a graph E on the left and the delayed graph D3(E) on the right. wg,2 fg,3 fe,3 we,2 g v E e fg,2 v wg,1 fg,1 fe,1 D3(E) fe,2 we,1 We will prove that for m ≥ 0 and n > 0, the graph S m Observe that there is a range and source preserving map D∗ nE is isomorphic to S m(Dn(E)). k=0 Dn(E)kn given by n : E∗ → S∞ D∗ n(e1 . . . ek) = fe1,1 . . . fe1,n fe2,1 · · · fe2,n · · · fek,1 · · · fek,n. Lemma 6.12. Let E be a locally finite graph with no sources, and fix integers n ≥ 1 and m ≥ 0. There are homeomorphisms SD0 nE1 → S mDn(E)1 such that for j ∈ {1, . . . , n} and t ∈ [0, 1], nE0 → S mDn(E)0 and SD1 n : S m n : S m ≤ µ. We have SD0 n(s(α)) = s(SD1 n(α)) and for all µ ∈ E∗ such that 0 ≤ j−1+t SD0 n(α)) for all α ∈ S mDn(E)1. n(r(α)) = r(SD1 0 , j−1+m+t n n Proof. Define SDn v ∈ E0, and SDn : E0 ⊔ (E1 × [0, 1]) → SDn(E)0 = S mDn(E)0 by SDn n (cid:17)(cid:17) = [fe,j, t] for e ∈ E1, 1 ≤ j ≤ n and t ∈ [0, 1]. Then 0(cid:16)(cid:16)e, j−1+t ((e, 0)) = [fe,0, 0] = [r(fe,0)] = [r(e)] = SDn (r(e)), 0 0 0 (v) := [v] for ((e, 1)) = SDn and similarly, SDn S mDn(E)0. Since SDn that SD0 If ω = [e, j−1+t nE0 → n. It is routine using Lemma 3.2 to see n is injective, and it is clearly surjective. To see that it is open, fix ω ∈ SE0. ] : s ∈ (−δ, δ)} indexed by ] with t ∈ (0, 1), then the sets {[e, j−1+t+s is continuous, so is SD0 descends to a map SD0 (s(e)), so SDn 0 n : S m 0 0 n n SDn 0 for all e ∈ E1, and SD0 n(cid:16)he, j − 1 + t n i(cid:17) = [fe,j, t] SD1 n(cid:16)hµ, j − 1 + t n i(cid:17) = [D∗ n(µ), j − 1 + t] THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 23 n ] for some 1 < j ≤ n, then the sets {[e, j−1+s sufficiently small δ form a neighbourhood basis at ω and are carried to the open sets {[fe,j, t + s] : s ∈ (−δ, δ)}. If ω = [e, j−1 ] : s ∈ (−δ, δ)} indexed by δ ∈ (0, 1 2) form a neighbourhood base at ω and are carried to the open sets {[fe,j−1, t] : t > 1 − δ} ∪ {[fe,j, t] : t < δ}. Finally, if ω = [v] for v ∈ E0, then the sets {[e, s] : e ∈ E1v : s > 1 − δ} ∪ {[e, s] : e ∈ vE1 : s < δ} indexed by δ < 1 n are a neighbourhood base at E, and are carried to the open sets {[fe,n, t] : e ∈ E1v, t > 1 − nδ} ∪ {[fe,1, t] : e ∈ vE1, t < nδ}. Therefore SD0 n is an open map, and therefore a homeomorphism. A similar argument shows that SD1 n n exists and is a homeomorphism, and a simple n and SD1 n are compatible with the range and (cid:3) comparison of formulas shows that SD0 source maps as claimed. Corollary 6.13. Let E be a locally finite graph with no sources. Fix integers n ≥ 1 and m ≥ 0. There is a unitary Um,n : ℓ2(S m nE∗) → ℓ2(S mDn(E)∗) such that Um,nρm/n(a)U ∗ m,n = ρm(a ◦ SD0 n) for a ∈ C0(S mDn(E)0) and Um,nψm/n(ξ)U ∗ m,n = ψm(ξ ◦ SD0 n) for ξ ∈ Cc(S mDn(E)1). Conjugation by Um,n restricts to an isomorphism Θm,n : T C ∗(S m nE) → T C ∗(S mDn(E)). and This Θm,n induces an isomorphism eΘm,n : C ∗(S m eΘm,n(ρm/n(a)) = ρm(a ◦ SD0 eΘm,n( ψm/n(ξ)) = ψm(ξ ◦ SD0 Proof. For k ≥ 1 there is a bijection SDk SD1 obtain a length-preserving bijection SD∗ n(α1) · · · SD1 n : S m n(αk). Combining these bijections SDk n : S m nE∗) → ℓ2(S mDn(E)∗). m Um,n : ℓ2(S nE) → C ∗(S mDn(E)) such that n) for a ∈ C0(S mDn(E)0) n) for ξ ∈ Cc(S mDn(E)1). nEk → S mDn(E)k given by SDk n(α1 · · · αk) = n we nE∗ → S mDn(E)∗, which induces a unitary n for k > 0 with the bijection SD0 This Um,n intertwines ρm/n and a 7→ ρm(a ◦ SD0 SD1 ℓ2(S mDn(E)∗SD0 n(ω)) for each ω ∈ S m nE0, we have n). This proves the first statement. Since the unitary Um,n carries ℓ2(S m n) and intertwines ψm,n with ξ 7→ ψm(ξ ◦ nE∗ω) to AdUm,n(cid:16) Mω∈S m n E0 K(ℓ2(S m nE∗ω)(cid:17) = Mω∈SmDn(E)0 K(ℓ2(S mDn(E)∗ω). Hence Θm,n carries the kernel of the quotient map T C ∗(S m nE) to the kernel of the quotient map T C ∗(S mDn(E)) → C ∗(S mDn(E)). It follows that Θm,n descends to (cid:3) nE) → C ∗(S m the desired isomorphism eΘm,n. 7. Analysis of T C ∗(S mE) We now analyse the C ∗-algebras T C ∗(S mE) and C ∗(S mE) for integers m ≥ 1 (both T C ∗(S 0E) and C ∗(S 0E) are described by part (2) of Theorem 6.10). This will complete our analysis of the C ∗-algebras T C ∗(S lE) and C ∗(S lE) for rational l. To analyse T C ∗(S mE) we will first establish that the ideal of T C ∗(S mE) generated by the image of C0(SE0 \ E0) is isomorphic to T C ∗(E(1, m + 1)) ⊗ C0((0, 1)), and show that T C ∗(S mE) itself is a C(S)-algebra. We begin with some preliminary structural results. 24 AIDAN SIMS Given a locally compact Hausdorff space X, we write Cb(X) for the algebra of bounded continuous complex-valued functions on X. Lemma 7.1. Let E be a locally finite graph with no sources and fix m ∈ N \ {0}. The space Cc(S mE1) is a Cb(SE0)-bimodule with respect to the actions (a · ξ)(α) = a(r(α))ξ(α) and (ξ · a)(α) = ξ(α)a(s(α)). For each ξ ∈ Cc(S mE1) there exists a ∈ Cc(SE0) such that a · ξ = ξ = ξ · a. Proof. For a ∈ Cb(SE0) and ξ ∈ Cc(S mE1), the function a · ξ is the pointwise product of a ◦ r and ξ and therefore a continuous function. Since its support is contained in that of ξ it belongs to Cc(S mE1). Similarly ξ · a ∈ Cc(S mE1). It is routine that these actions make Cc(S mE1) into a Cb(SE0)-bimodule. For the final assertion, fix ξ ∈ Cc(S mE1). Since r, s : S mE1 → SE0 are continuous, K := r(supp(ξ)) ∪ s(supp(ξ)) ⊆ SE0 is compact, so Tietze's theorem yields a function a ∈ Cc(SE0) such that aK ≡ 1. We then have a · ξ = ξ = ξ · a by definition of the actions of C0(SE0) on Cc(S mE1). (cid:3) To analyse the ideal of T C ∗(S mE) generated by C0(SE0 \ E0), we first observe that the subgraph of S mE with vertex set SE0 \ E0 is a topological graph in the sense of Katsura. Lemma 7.2. Let E be a locally finite graph with no sources and fix m ∈ N \ {0}. Then (SE0 \ E0, S mE1 \ E1, r, s) is a topological graph isomorphic to the product E(1, m + 1) × (0, 1). Proof. Lemma 3.2 shows that the quotient maps from E1 × [0, 1] to SE0 and from Em+1 × [0, 1] to S mE1 restrict range and source preserving homeomorphisms from E(1, m + 1)0 × (0, 1) to SE0 \ E0 and from E(1, m + 1)1 × (0, 1) to S mE1 \ E1. (cid:3) We can now describe the ideal of T C ∗(S mE) generated by C0(SE0\E0). In the following proof, given e ∈ E1 and g ∈ C0((0, 1)), we denote by 1e × g the element of C0(SE0 \ E0) ⊆ C0(SE0) given by (1e × g)([f, t]) := δe,f g(t) for all f ∈ E1 and t ∈ (0, 1), and likewise for µ ∈ Em+1 and g ∈ C0((0, 1)), we write 1µ ×g for the element of C0(S mE1 \ Em) ⊆ C0(S mE1) given by (1µ × g)([ν, t]) := δµ,νg(t) for all ν ∈ Em+1 and t ∈ (0, 1). Lemma 7.3. Let E be a row-finite graph with no sources and fix m ∈ N \ {0}. Let J be the ideal of T C ∗(S mE) generated by ρm(C0(SE0)). There is an isomorphism κ0 : C0((0, 1)) ⊗ T C ∗(E(1, m + 1)) → J such that and κ0(g ⊗ Tµ) = ψ(1µ × g) κ0(g ⊗ Qe) = π(1e × g) for all g ∈ C0((0, 1)), all e ∈ E(1, m + 1)1 = E1 and all µ ∈ E(1, m + 1)1 = Em+1. Proof. By Lemma 7.2, we have (SE0 \ E0, S mE1 \ Em, r, s) ∼= E(1, m + 1) × (0, 1) as topological graphs. Recall from [22] that an s-section in a topological graph F is an open set U ⊆ F 1 such that s : U → s(U) is a homeomorphism onto an open subset of F 0. The collection B :=(cid:8){[ν, t] : t ∈ (a, b)} : ν ∈ E(1, m)1 and 0 < a < b < 1(cid:9) is a basis of open s-sections for the topology on S mE1 \ Em. Let F := {1ν × g : ν ∈ E(1, m + 1)1 and g ∈ C0((0, 1))}. Then this F and B satisfy [22, Equation (4.4)]. Di- rect computation with basis elements on each ℓ2(S mEt) show that (ρm, ψmspan F ) satisfy THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 25 [22, Equation (4.5)] and [22, Equation (4.6)]. Thus [22, Proposition 4.12] shows that (ρmC0(SE0\E0), ψmCc(SmE1\E1)) is a representation of the topological graph E(1, m + 1) × (0, 1). The range of ρmC0(SE0\E0) belongs to J by definition. The image of ψmCc(SmE1\Em) belongs to J by the argument of Lemma 7.1. These elements generate J because C0(SE0 \ E0) · Cc(S mE1) is contained in the space Cd(S mE1 \ Em) described at (2.3). Thus [22, Theorem 2.4] shows that there is a surjective homomorphism (ρ × ψ) : T C ∗(E(1, m + 1) × (0, 1)) → J such that (ρ × ψ) ◦ iE(1,m+1)0×(0,1) = ρmC0(SE0\E0) and (ρ × ψ) ◦ iE(1,m+1)1×(0,1) = ψmCc(SmE1\Em). To see that this homomorphism is injective, recall that J is a subalgebra of B(ℓ2(S mE∗ \ E(0, m)∗)), and observe that ℓ2(SE0 \ E0) ⊆ {ψm(ξ)v : ξ ∈ Cc(S mE1 \ Em), v ∈ ℓ2(S mE∗)} ⊥ . For a ∈ C0(SE0 \ E0), the restriction of ρm(a) to ℓ2(SE0 \ E0) is given by ρm(a)hω = a(ω)hω, and so the reduction of ρm to this subspace is faithful. Hence [9, Theorem 2.1] shows that (ρ × ψ) is injective. The argument of [15, Proposition 7.7] shows that T C ∗(E(1, m+1)×(0, 1)) ∼= C0((0, 1))⊗ T C ∗(E(1, m+1)), so we obtain a surjective representation of C0((0, 1))⊗T C ∗(E(1, m+1)) in J that carries g ⊗ Qe to π(1e × g) and g ⊗ Tν to ψ(1ν × g). (cid:3) We observe next that the actions of Cb(SE0) on Cc(S mE1) induce a central action of C(S) via the surjection SE0 → S of Lemma 3.4. Corollary 7.4. Let E be a locally finite graph with no sources and fix m ∈ N \ 0. There are left and right actions of C(S) on Cc(S mE1) given by (g · ξ)([µ, t]) = g(t)ξ([µ, t]) = (ξ · g)([µ, t]) for g ∈ C(S) and ξ ∈ Cc(S mE1). Proof. The surjection : SE0 → S of Lemma 3.4 induces an injection ∗ : C(S) → Cb(SE0) given by ∗(g)([e, t]) = g(([e, t])) = g(t). So g · ξ := ∗(g) · ξ and ξ · g := ξ · ∗(g) satisfy the formulae given for the desired action. The definition of shows that g · ξ = ξ · g. (cid:3) Proposition 7.5. Let E be a locally finite graph with no sources and fix m ∈ N \ 0. The pair (ρm, ψm) is a bimodule homomorphism in the sense that ρm(a)ψm(ξ) = ψm(a · ξ) and ψm(ξ)ρm(a) = ψm(ξ · a) for all a ∈ C0(SE0) and ξ ∈ Cc(S mE1). Writing ¯ρm for the extension of ρm to Cb(SE0) = M(C0(SE0)) and ∗ : C(S) → Cb(SE0) for the homomorphism induced by the map : SE0 → S of Lemma 3.4, and writing ιm := ¯ρm◦∗, the action of C(S) on Cc(SE1) of Corollary 7.4 satisfies ιm(g)ψm(ξ) = ψm(g · ξ) = ψm(ξ)ιm(g) for all g ∈ C(S) and ξ ∈ Cc(S mE1). In particular, ιm is an injective unital inclusion of C(S) in ZM(T C ∗(S mE)). Proof. We just calculate with basis vectors: for a, b ∈ C0(SE0) and ξ ∈ Cc(S mE1), and for α ∈ S mE∗, we have ρm(a)ψm(ξ)ρm(b)hα = Xβ∈SmE1r(α) = Xβ∈SmE1r(α) a(r(β))ξ(β)b(r(α))hβα (a · ξ · b)(β)hβα = ψm(a · ξ · b)hα. 26 AIDAN SIMS Taking a such that a · ξ = ξ as in Lemma 7.1 gives ψm(ξ)ρm(b) = ψm(ξ · b). Likewise, taking b such that ξ · b = ξ gives ρm(a)ψm(ξ) = ψm(a · ξ). Now fix ξ ∈ Cc(S mE1) and g ∈ C(S). Choose a ∈ Cc(SE0) such that a · ξ = ξ = ξ · a. Then by definition of ¯ρ, ιm(g)ψm(ξ) = ιm(g)ψm(a · ξ) = ¯ρm(∗(g))ρm(a)ψm(ξ) = ρm(∗(g)a)ψm(ξ) = ψm((∗(g)a) · ξ) = ψm(g · (a · ξ)) = ψm(g · ξ). Likewise ψm(ξ)ιm(g) = ψm(ξ · g). Corollary 7.4 shows that g · ξ = ξ · g, so we obtain ιm(g)ψm(ξ) = ψm(g · ξ) = ψm(ξ)ιm(g) as claimed. Since ψm(Cc(SE1) and ρm(C0(SE0) generate T C ∗(S mE), we deduce that ιm(C(S))T C ∗(S mE) ⊆ T C ∗(S mE), and taking adjoints gives T C ∗(S mE)ιm(C(S)) ⊆ T C ∗(S mE) as well. So we can regard ιm as a homomorphism of C(S) into M(T C ∗(S mE)). Since C(SE0) is abelian, the elements of ιm(C(S)) commute with the elements ρm(a), and we have just established that they commute with the elements of ψm(Cc(S mE1). Again, since the ρm(a) and the ψm(ξ) generate T C ∗(S mE) we see that ι takes values in ZM(T C ∗(S mE)). Finally, the identity function 1 ∈ C(S)) satisfies 1 · ξ = ξ = ξ · 1 for all ξ ∈ Cc(S mE1). So we obtain ιm(1)ψm(ξ) = ψm(1 · ξ) = ψm(ξ) for all ξ, and clearly ιm(1)ρm(a) = ρm(a) for all a by definition of ιm and ∗. Hence ιm is unital. (cid:3) The general theory of C(X)-algebras (see Section 2.6) now implies that T C ∗(S mE) is isomorphic to the algebra of continuous sections of an upper-semicontinuous bundle of C ∗-algebras over S. Notation 7.6. Let E be a locally finite graph with no sources and fix m ∈ N \ {0}. For each t ∈ S we write Jt for the ideal of T C ∗(S mE) generated by ι({g ∈ C(S) : g(t) = 0}). Following the standard conventions for C(X)-algebras, we then write T C ∗(S mE)t for the quotient T C ∗(S mE)/Jt. For each a ∈ T C ∗(S mE), we write γa : S →Ft∈S T C ∗(S mE)t for the section given by γa(t) := a + Jt. We first show that for t ∈ (0, 1), the fibre T C ∗(S mE)t is a copy of T C ∗(E(1, m)), and describe standard representatives in T C ∗(S mE) of its canonical generators. The following notation will be helpful for the next few results. Notation 7.7. Let E be a locally finite graph with no sources. By Remark 6.5 the space 0 can be identified with E(0, m)∗ via the map [µ] 7→ µ. Let π0 be the representation of S mE∗ T C ∗(E(0, m)) on ℓ2(S mE∗ 0 ) obtained from the path-space representation of T C ∗(E(0, m)) and this identification; so π0(Qv)h[ν] = δv,r(ν)h[ν] and π0(Tµ)h[ν] = δs(µ),r(ν)h[µν]. t can be identified with E(1, m + 1)∗ via the map S mEk For t ∈ S \ {0}, the set S mE∗ t ∋ [µ, t] 7→ (µ1 · · · µm+1)(µm+1 · · · µ2m+1) · · · (µ(k−1)mµkm+1) ∈ E(1, m + 1)k for k ∈ N, µ ∈ Ekm+1 and t ∈ (0, 1). We write πt for the representation of T C ∗(E(1, m+ 1)) on ℓ2(S mE∗ t ) obtained from this identification and the path-space representation of T C ∗(E(1, m + 1)). So πt(Qe)h[µ,t] = δe,µ1h[µ,t] and πt(Tµ)h[ν,t] = δµµ,ν1h[µν2···νν,t]. THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 27 Lemma 7.8. Let E be a locally finite graph with no sources, fix m ∈ N \ {0}, and take , then πm(a) + Jt = πm(a′) + Jt = a′SE0 t ∈ [0, 1). in T C ∗(S mE)t. If ξ, ξ′ ∈ Cc(S mE1) satisfy ξCc(SmE1 t ), then ψm(ξ) + Jt = ψm(ξ′) + Jt in T C ∗(S mE)t. If a, a′ ∈ C0(SE0) satisfy aSE0 t ) = ξ′Cc(SmE1 t t Proof. Let d be the quotient metric on S induced by the usual metric on R. For each n, fix a function fn ∈ C0(S \ {t}) such that 0 ≤ fn ≤ 1 and fn(s) = 1 whenever d(s, t) ≥ 1/n. For the first statement, note that C0(SE0 \ {[e, t] : e ∈ E1}) belongs to the ideal t )) ⊆ Jt. Since a − a′ ∈ C0(SE0 \ SE0 t ) generated by the ∗(fn), and so ρ(C0(SE0 \ SE0 this proves the first statement. For the second statement, let N := {µ ∈ E(1, m + 1)1 : max{ξ([µ, s]), ξ′([µ, s])} > 0 for some s ∈ [0, 1)}. Since ξ and ξ′ have compact support, N is finite. Fix ε > 0. The set Xε := r({α ∈ S mE1 : (ξ − ξ′)(α) ≥ ε/N}) is a compact subset of SE0 \ SE0 t and so there exists n > 0 such that fnXε ≡ 1. For this n, we have and supp(cid:0)(ξ − ξ′) − fn · (ξ − ξ′)(cid:1) ⊆ supp(ξ) ∪ supp(ξ′). For any η ∈ Cc(S mE1), we have, using the representations πt of Notation 7.7, kψm(η)k = sup s∈S kψm(η)ℓ2(SmE∗ (cid:13)(cid:13)(ξ − ξ′) − fn · (ξ − ξ′)k∞ ≤ ε/N, η([µ, s])πs(Tµ)(cid:13)(cid:13)(cid:13),(cid:13)(cid:13)(cid:13) Xν∈Em s∈(0,1)(cid:13)(cid:13)(cid:13) Xµ∈Em+1 s )k η(α). ≤ maxn sup s∈S Xα∈SmE1 ≤ sup s η([ν])π0(Tν)(cid:13)(cid:13)(cid:13)o Applying this to η = (ξ − ξ′) − fn · (ξ − ξ′) and using the definition of N, we deduce that (cid:13)(cid:13)ψm(cid:0)(ξ − ξ′) − fn · (ξ − ξ′)(cid:1)(cid:13)(cid:13) ≤ ε. Since the fn all vanish at t, it follows that ξ − ξ′ ∈ Jt We can now prove that for each t ∈ (0, 1), the corresponding fibre T C ∗(S mE)t is as claimed. (cid:3) isomorphic to T C ∗(E(1, m + 1)). Proposition 7.9. Let E be a locally finite graph with no sources and fix m ∈ N\{0}. Take t ∈ (0, 1). For each e ∈ E1, fix a function ae,t ∈ C0(SE0, [0, 1]) such that supp(ae,t) ⊆ {[e, s] : 0 < s < 1} and ae,t([e, t]) = 1. For each µ ∈ Em+1, fix a function ξµ,t ∈ Cc(S mE1) such that supp(ξµ,t) ⊆ {[µ, s] : 0 < s < 1} and ξµ,t([µ, t]) = 1. There is an isomorphism θt : T C ∗(E(1, m + 1)) → T C ∗(S mE)t such that θt(Qe) = ρm(ae,t) + Jt for all e ∈ E(1, m + 1)0 = E1, and such that θt(Tµ) = ψm(ξµ,t) + Jt for all µ ∈ E(1, m + 1)1 = Em+1. Proof. Lemma 7.8 shows that the elements ρm(ae,t) + Jt and ψm(ξef,t) + Jt generate T C ∗(S mE)t, so it suffices to construct an injective homomorphism θt satisfying the given formulae. For this, define qe := ρm(ae,t) + Jt for each e ∈ E1 and tµ := ψm(ξµ,t) + Jt for each µ ∈ Em+1. We will show that (q, t) is a Toeplitz -- Cuntz -- Krieger E(1, m + 1)-family. 28 AIDAN SIMS Since a2 e,t([f, t]) = ae,t([f, t]) = ae,t([f, t]) for all f , Lemma 7.8 shows that the qe are projections. We have ae,taf,t = 0 in C0(SE0) for e 6= f , so the qe are mutually orthogonal. Fix µ ∈ Em+1. Define a′ µ,t : SE0 → C by µ,t([e, s]) =(ξµ,t(s)2 0 a′ if µm+1 = e otherwise. Since supp(ξµ,t) is a compact subset of {[µ, s] : 0 < s < 1}, we have a′ by construction, a′ that qµm+1 = ρm(a′ µ,t ∈ Cc(SE0), and µ,t and aµm+1,t agree at [g, t] for every g ∈ E1. So Lemma 7.8 implies µ,t) + Jt. Using the representations πt of Notation 7.7, we have ψm(ξµ,t)∗ψ(ξµ,t) =(cid:16) Xα,β∈Em ξµ,t([α])ξµ,t([β])π0(T ∗ αTβ)(cid:17) ⊕ M0<s<1 Xη,ζ∈Em+1 = 0 ⊕ M0<s<1 ξµ,t([η, s])ξµ,t([ζ, s])πs(T ∗ η Tζ) ξµ,t([µ, s])2π∞(Qµm+1) = ρm(a′ µ,t). Thus t∗ µtµ = ρ(a′ µ,t) + Jt = ρ(aµm+1,t) + Jt = qs1,m+1(µ). Now fix e ∈ E1. For each µ ∈ s(e)Em, define ξ′ ν ∈ Em+1 and s ∈ [0, 1). Each ξ′ s < 1}. Lemma 7.8 shows that teµ = ψm(ξ′ above, we see that eµ,t ∈ Cc(S mE1) because ae,t is supported on {[e, s] : 0 < eµ,t) for each µ ∈ E(1, m + 1)1. Arguing as eµ,t([ν, s]) = δeµ,νpae,t([e, s]) for eµ,t by ξ′ eµ,t)ψ(ξ′ eµ,t)∗ Xeµ∈eE(1,m)1 ψ(ξ′ teµt∗ eµ = Xµ∈s(e)Em = 0 ⊕ M0<s<1 Xµ∈s(e)Em = 0 ⊕ M0<s<1 ρm(ae,t) = 0 ⊕ M0<s<1 ξ′ eµ,t(eµ, s)2πs(TeµT ∗ eµ) ae,t(e, s)1span{h[µ,s]:µ∈eE(1,m+1)∗\{e}}. ae,t(e, s)1span{h[µ,s]:µ∈eE(1,m+1)∗}. (7.1) Also, (7.2) eµ,t)ψm(ξ′ eµ,t)∗. In particular, in the quo- We deduce that ρm(ae,t) >Peµ∈eE(1,m+1)1 ψm(ξ′ tient, qe ≥Peµ∈eE(1,m+1)1 teµt∗ eµ. So (q, t) is a Toeplitz -- Cuntz -- Krieger E(1, m + 1)-family as claimed. The universal property of T C ∗(E(1, m + 1)) therefore yields a homomorphism θt : T C ∗(E(1, m + 1)) → T C ∗(S mE)t such that θt(Qe) = ρm(ae,t) + Jt and θt(Tµ) = ψm(ξµ,t) + Jt. It remains to prove that θt is injective. Since Jt is contained in the kernel of the ef used t ) on T C ∗(S mE)t, we see that (with the functions ξ′ restriction map x 7→ xℓ2(SmE∗ in the calculation (7.1) above), each kqe − Xeµ∈eE(1,m+1)1 teµt∗ eµk ≥(cid:13)(cid:13)(cid:13)(cid:16)ρm(ae,t) − Xeµ∈eE(1,m+1)1 ψm(ξ′ eµ,t)ψm(ξ′ t )(cid:13)(cid:13)(cid:13). eµ,t)(cid:17)(cid:12)(cid:12)(cid:12)ℓ2(SmE∗ THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 29 The calculations (7.1) and (7.2) therefore show that (cid:13)(cid:13)(cid:13)qe − Xeµ∈eE(1,m+1)1 teµt∗ eµ(cid:13)(cid:13)(cid:13) ≥ k1Chek = 1. eµ 6= 0, and the uniqueness theorem [9, Theorem 4.1] shows (cid:3) So each qe −Peµ∈eE(1,m+1)1 teµt∗ that θt is injective. Corollary 7.10. Let E be a locally finite graph with no sources and fix m ∈ N \ {0}. Take t ∈ (0, 1). Let Ut : ℓ2(E(1, m + 1)∗) → ℓ2(S mE∗ t ) be the unitary of Remark 6.5. Let π : T C ∗(E(1, m + 1)) → B(ℓ2(E(1, m + 1)∗)) be the path-space representation. Then the t Ut) is an isomorphism of T C ∗(S mE)t onto T C ∗(E(1, m+1)). map a+Jt 7→ π−1(U ∗ of the isomorphism described in Proposition 7.9. It is Proof. Consider the inverse θ−1 straightforward to check that for a ∈ C0(SE0) and ξ ∈ Cc(S mE1), we have t aℓ2SmE∗ t and θ−1 t (ρm(a) + Jt) = Xe∈E1 t (ψm(ξ) + Jt) = Xµ∈E(1,m+1)1 θ−1 a([e, t])Qe = π−1(U ∗ t ρm(a)ℓ2(SmE∗ t )Ut) ξ([µ, t])Tef = π−1(U ∗ t ψm(ξ)ℓ2(SE∗ t )Ut). Since the elements ρm(a) + Jt and ψm(ξ) + Jt generate T C ∗(S mE)t, it follows that x 7→ π−1(U ∗ (cid:3) , so is an isomorphism as claimed. t )Ut) agrees with θ−1 t xℓ2(SmE∗ t t aℓ2(SmE∗ We must now describe the fibre T C ∗(S mE)0. The idea is that for a ∈ T C ∗(S mE0), using the unitaries Ut : ℓ2(E(1, m + 1)∗) → ℓ2(S mE∗ t ) of Remark 6.5, the function t 7→ U ∗ t )Ut from (0, 1) to B(ℓ2(E(1, m + 1)∗)) converges in norm as t → 0 and as t → 1, and the limits ε0(a) and ε1(a) belong to the image of T C ∗(E(1, m + 1)) in its path-space representation. We use these limits to construct an injective homomorphism of T C ∗(S mE)0 into T C ∗(E(1, m + 1)) ⊕ T C ∗(E(1, m + 1)). Lemma 7.11. Let E be a locally finite graph with no sources and fix m ∈ N \ {0}. For t ∈ (0, 1), let Ut t ) be the unitary of Remark 6.5. Let π : T C ∗(E(1, m + 1)) → B(ℓ2(E(1, m + 1)∗)) be the path-space representation. For a ∈ C0(SE0) we have : ℓ2(E(1, m + 1)∗) → ℓ2(S mE∗ For ξ ∈ Cc(S mE1), we have U ∗ lim tց0 t ψm(ξ)ℓ2(SmE∗ U ∗ t ρm(a)ℓ2(SmE∗ lim tց0 a([r(e)])π(Qe). t )Ut = Xe∈E1 t )Ut = Xµ∈Em,e∈s(µ)E1 ξ[µ]π(Tµe). For a ∈ T C ∗(S mE), the limit limtց0 U ∗ 1))), and ε0 : a 7→ π−1(limtց0 U ∗ T C ∗(E(1, m + 1)). t aℓ2(SmE∗ t )Ut exists and belongs to π(T C ∗(E(1, m + t )Ut) is a homomorphism from T C ∗(S mE) to t aℓ2(SmE∗ Proof. Fix a ∈ C0(SE0). Fix ε > 0, and let F := {e ∈ E1 : a([e, t]) ≥ ε/2 for some t ∈ [0, 1]}. Then F is finite, and for each e ∈ F there exists δe > 0 such that 0 < t < δe =⇒ a([e, t]) − a([e, 0]) < ε. Let δ = mine∈F δe. Then for e ∈ E1 and 0 < t < δ, if 30 AIDAN SIMS e ∈ F then a([e, t]) − a([e, 0]) < ε by choice of δ and if e 6∈ F , then a([e, t]) − a([e, 0]) ≤ a([e, t])+a([e, 0]) < ε by choice of F . By definition of ρm, we have U ∗ t )Ut = Pe∈E1 a([e, t])π(Qe). Since [e, 0] = [r(e)] for each e and since each kQek = 1 it follows t ρm(a)ℓ2(SmE∗ that (7.3) U ∗ t ρm(a)ℓ2(SmE∗ lim tց0 a([r(e)])π(Qe). t )Ut = Xe∈E1 Now fix ξ ∈ Cc(S mE1). Let F = {µ ∈ Em+1 : ξ([µ, t]) 6= 0 for some t ∈ [0, 1]}. Fix ε > 0. For each µ ∈ F there exists δµ > 0 such that 0 < t < δµ implies ξ([µ, t]) − ξ([µ, 0]) < ε/F . Let δ := minµ∈F δµ. Fix t ∈ (0, δ). We have t ψm(ξ)ℓ2(SmE∗ (cid:13)(cid:13)(cid:13)U ∗ ξ[µ]π(Tµe)(cid:13)(cid:13)(cid:13) t )Ut − Xµ∈Em,e∈s(µ)E1 =(cid:13)(cid:13)(cid:13)Xµe∈F (ξ[µe, t] − ξ([µ]))π(Tµe)(cid:13)(cid:13)(cid:13) ≤ Xµe∈F ξ[µe, t] − ξ([µ])kπ(Tµ)k < ε since each kπ(Tµ)k = 1. Hence (7.4) lim tց0 U ∗ t ψm(ξ)ℓ2(SmE∗ t )Ut = Xµ∈Em,e∈s(µ)E1 ξ[µ]π(Tµe). t αi,jℓ2(SmE∗ where each αi,j ∈ ρm(C0(SE0))∪ψm(Cc(S mE1))∪ψm(S mE1)∗. By the first two statements, t )Ut = π(βi,j) for some βi,j ∈ T C ∗(E(1, m + 1)), for each i, j we have limtց0 U ∗ and it follows that limtց0 U ∗ For the final statement, first consider a finite linear combination x =Pi αi,1αi,2 · · · αi,ki t )Ut = π(cid:0)Pi βi,1 . . . βi,kj(cid:1) ∈ π(T C ∗(E(1, m + 1))). Now fix x ∈ T C ∗(S mE). Fix ε > 0. Fix a linear combination a =Pi αi,1αi,2 · · · αi,ki where each αi,j ∈ ρm(C0(SE0)) ∪ ψm(Cc(S mE1)) ∪ ψm(S mE1)∗ such that ka − xk < ε/4. Then in particular, k(a − x)ℓ2(SmE∗ t )k < ε/4 for all t ∈ (0, 1). By the preceding paragraph, a0 := limtց0 U ∗ t )Ut exists and belongs to T C (E(1, m + 1)), so there exists δ > 0 such that kU ∗ t )Ut − a0k < ε/4 for all t < δ. In particular, there exists δ > 0 such that 0 < s, t < δ implies t xℓ2(SmE∗ t aℓ2(SmE∗ t aℓ2(SmE∗ s )Us)k + kU ∗ t )Utk + kU ∗ s aℓ2(SmE∗ t (aℓ2(SmE∗ s )Us − a0k t ) − xℓ2(SmE∗ t ))Utk < ε. kU ∗ s xℓ2(SmE∗ s )Us − U ∗ s (xℓ2(SmE∗ ≤ kU ∗ t )Utk t xℓ2(SmE∗ s ) − aℓ2(SmE∗ t aℓ2(SmE∗ Hence (cid:0)U ∗ x0 ∈ π(T C ∗(E(1, m + 1))). We show that limtց0 U ∗ + ka0 − U ∗ 1/n)U1/n(cid:1)∞ 1/nxℓ2(SmE∗ n=1 is a Cauchy sequence, and therefore converges to some t xℓ2(SmE∗ t )Ut = x0. Fix ε > 0. Using that U ∗ 1/nxℓ2(SmE∗ x0 and also the preceding paragraph, we can choose δ > 0 such that kU ∗ 1/nxℓ2(SmE∗ x0k < ε/2 whenever n > δ−1, and such that kU ∗ s )Us − U ∗ t xℓ2(SmE∗ whenever s, t < δ. In particular, for t < δ, and any choice of n > δ−1, we have s xℓ2(SmE∗ 1/n)U1/n → 1/n)U1/n− t )Utk < ε/2 kU ∗ t xℓ2(SmE∗ ≤ kU ∗ t )Ut − x0k t xℓ2(SmE∗ t )Ut − U ∗ 1/nxℓ2(SmE∗ 1/n)U1/nk + kU ∗ 1/nxℓ2(SmE∗ 1/n)U1/n − x0k < ε. THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 31 Hence U ∗ t xℓ2(SmE∗ t )Ut → x0 ∈ π(T C ∗(E(1, m + 1))) as claimed. Since π is injective, we deduce that the map ε0 exists. It is a homomorphism because t ) is a homomorphism and the algebraic operations in T C ∗(S mE) are (cid:3) each a 7→ aℓ2(SmE∗ continuous. Lemma 7.12. Let E be a locally finite graph with no sources and fix m ∈ N \ {0}. For t ∈ (0, 1), let Ut t ) be the unitary of Remark 6.5. Let π : T C ∗(E(1, m + 1)) → B(ℓ2(E(1, m + 1)∗)) be the path-space representation. For a ∈ C0(SE0) we have : ℓ2(E(1, m + 1)∗) → ℓ2(S mE∗ U ∗ t ρm(a)ℓ2(SmE∗ lim tր1 a([s(e)])π(Qe). t )Ut = Xe∈E1 t )Ut = Xµ∈Em,e∈E1r(µ) For ξ ∈ Cc(S mE1), we have U ∗ t ψm(ξ)ℓ2(SmE∗ lim tր1 ξ[µ]π(Teµ). For a ∈ T C ∗(S mE), the limit limtր1 U ∗ 1))), and ε1 : a 7→ π−1(limtր1 U ∗ T C ∗(E(1, m + 1)). t aℓ2(SmE∗ t )Ut exists and belongs to π(T C ∗(E(1, m + t )Ut) is a homomorphism from T C ∗(S mE) to t aℓ2(SmE∗ Proof. The proof is essentially identical to that of Lemma 7.11. (cid:3) Proposition 7.13. Let E be a locally finite graph with no sources and fix m ∈ N \ {0}. There is an injective homomorphism η : T C ∗(S mE)0 → T C ∗(E(1, m+1))⊕T C ∗(E(1, m+ 1)) such that, for any a ∈ C0(SE0) we have and such that for any ξ ∈ Cc(S mE1) we have η(ρm(a)0) = Xv∈E0 η(ψm(ξ)0) = Xµ∈Em a([v])(cid:16)(cid:16) Xe∈E1v ξ([µ])(cid:16)(cid:16) Xe∈E1r(µ) Qf(cid:17)(cid:17), Qe(cid:17) ⊕(cid:16) Xf ∈vE1 Tµf(cid:17)(cid:17). Teµ(cid:17) ⊕(cid:16) Xf ∈s(µ)E1 Proof. Let ε0, ε1 : T C ∗(S mE) → T C ∗(E(1, m + 1)) be the homomorphisms of Lemmas 7.11 and 7.12. Since ε0, ε1 vanish on ρm(C0(SE0 \ E0)), they descend to homomorphisms ε0, ε1 : T C ∗(S mE)0 → T C ∗(E(1, m + 1)). The homomorphism η := ε0 ⊕ ε1 satisfies the formulae above, so it suffices to show that this homomorphism η is injective. For this, fix x ∈ T C ∗(S mE) such that η(x0) = 0; so ε0(x) = ε1(x) = 0. We must show that x0 = 0. We have (7.5) 0 = kε0(x)k = lim tց0 kxℓ2(SmE∗ t )k, and 0 = kε1(x)k = lim tր1 kxℓ2(SmE∗ t )k. t )k = kxtk for x ∈ T C ∗(S mE) and t ∈ (0, 1), and Corollary 7.10 implies that kxℓ2(SmE∗ so (7.5) implies that limt→0 kxtk = 0. It now follows from the properties of upper semi- continuous C ∗-bundles -- see equation 2.4 -- that xt → 0 ∈ C ∗(SE)0. Since t 7→ xt is a continuous section, we deduce that x0 = 0. (cid:3) We will show that the image of η is isomorphic to T C ∗(E(1, m + 1)) ⊕ T C ∗(E(0, m)) provided that enough vertices in E admit at least two edges. 32 AIDAN SIMS Proposition 7.14. Let E be a locally finite graph with no sources and fix m ∈ N \ {0}. Suppose that for every v ∈ E0 there exist n ≥ 1 and µ ∈ Enmv such that E1r(µ) ≥ 2. Let η : T C ∗(S mE)0 → T C ∗(E(1, m + 1)) ⊕ T C ∗(E(1, m + 1)) be the homomorphism of Proposition 7.13. Then the range of η is T C ∗(E(1, m + 1)) ⊕ 1,m+1(T C ∗(E(0, m))), and (id ⊕−1 1,m+1)◦η is an isomorphism of T C ∗(S mE)0 onto T C ∗(E(1, m+1))⊕T C ∗(E(0, m)). Proof. By Proposition 7.13, we just need to show that the range of η is T C ∗(E(1, m + 1)) ⊕ (T C ∗(E(0, m))). For each v ∈ E0, let wv :=Pe∈E1v Qe ∈ T C ∗(E(1, m + 1)), and let xv :=Pf ∈vE1 Qf ∈ T C ∗(E(1, m + 1)). For each µ ∈ Em, let yµ :=Pe∈E1r(µ) Teµ ∈ T C ∗(E(1, m + 1)) and let zµ :=Pf ∈s(µ)E1 Tµf ∈ T C ∗(E(1, m + 1)). We first show that the elements wv and yµ generate T C ∗(E(1, m + 1)). For this, let A := C ∗({wv : v ∈ E0} ∪ {yµ : µ ∈ Em}) ⊆ T C ∗(E(1, m + 1)). We must show that T C ∗(E(1, m + 1)) ⊆ A. Fix µ ∈ Em. For e, f ∈ E1r(µ), we have T ∗ eµTf µ = δe,f Qµm. Therefore (7.6) T ∗ eµTf µ = E1r(µ)Qµm. y∗ µyµ = Xe,f ∈E1r(µ) Since E has no sinks, E1r(µ) 6= 0, so Qµm ∈ A. Again since E has no sinks, for each e ∈ E1, the set Em−1r(e) is nonempty, so for any e ∈ E1, we have Qe = y∗ λeyλe ∈ A for any λ ∈ Em−1r(e). Since the Qe are mutually orthogonal projections, for e ∈ E1 and µ ∈ s(e)Em we have Teµ = QeTeµ = Qe Xf ∈E1r(µ) Tf µ = Qeyµ ∈ A. We have now established that all the generators of T C ∗(E(1, m + 1)) belong to A, and so T C ∗(E(1, m + 1)) ⊆ A. Next note that we have xv = 1,m+1(Qv) for v ∈ E0 and zµ = 1,m+1(Tµ) for µ ∈ Em, so (x, z) is a Toeplitz -- Cuntz -- Krieger E(0, m)-family, and C ∗({xv : v ∈ E0} ∪ {zµ : µ ∈ Em}) = 1,m+1(T C ∗(E(0, m))). We show next that (0, xv) ∈ η(T C ∗(SE)0) for each v ∈ E0. First, fix v ∈ E0 and µ ∈ Em satisfying E1r(µ) ≥ 2. Equation (7.6) and that (x, z) is a Toeplitz -- Cuntz -- Krieger family show that η(T C ∗(SE)0) ∋ ((yµ, zµ)∗(yµ, zµ))2 − (yµ, zµ)∗(yµ, zµ) = (E1r(µ)2Qµm, xv) − (E1r(µ)Qµm, xv) =(cid:0)(E1r(µ)2 − E1r(µ))Qµm, 0(cid:1). Since E1r(µ) ≥ 2, we have E1r(µ)2 − E1r(µ) > 0, and so (Qµm, 0) ∈ η(T C ∗(SE)0). We then obtain η(T C ∗(SE)0) ∋ (ye, ze)∗(ye, ze) − E1r(µ)(Qµm, 0) = (0, xv). Now take any v ∈ E0. By hypothesis, there exists n ∈ N and µ ∈ Enmr(e) such that E1r(µ) ≥ 2. Write µ = µ1 · · · µn where each µi ∈ Em. We have (0, xs(µ)) = (yµn, zµn)∗ · · · (yµ2, zµ2)∗(0, xs(µ1))(yµ2, zµ2) · · · (yµn, zµn) ∈ η(T C ∗(SE)0). This shows that (0, xv) ∈ B for every v ∈ E0 as claimed. THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 33 It follows that (0, ze) = (0, xr(e))(ye, ze) ∈ η(T C ∗(SE)0) for each e ∈ E1. Since each xv = (Qv) and each ze = (Te), we deduce that 0 ⊕ (T C ∗(E(0, m))) ⊆ η(T C ∗(SE)0). It now suffices to show that T C ∗(E(1, m + 1)) ⊕ 0 ⊆ η(T C ∗(SE)0) as well. Since we have already proved that 0 ⊕ 1,m(T C ∗(E(0, m))) ⊆ η(T C ∗(SE)0), we know that each (wv, 0) = (wv, xv) − (0, xv) and each (yµ, 0) = (yµ, zµ) − (0, zµ) belongs to η(T C ∗(SE)0). We saw above that the elements wv and yµ generate T C ∗(E(1, m + 1)). This completes the proof. (cid:3) If x ∈ T C ∗(S mE) belongs to the ideal generated by C0(S \ {0}), then xℓ2(SmE∗ 0 0(cid:13)(cid:13) ≤ for x ∈ T C ∗(S mE). It therefore follows from Proposition 7.13 that(cid:13)(cid:13)xℓ2SmE∗ Consequently, there is a homomorphism T C ∗(S mE)0 → B(ℓ2(S mE∗ xℓ2SmE∗ kη(x0)k for all x ∈ T C ∗(S mE). The following result gives a direct proof of this by showing that in fact we can use the injection 1,m+1 to see that the x 7→ xℓ2(SmE∗ 0 ) can be identified with the map obtained by following η with the second-coordinate projection T C ∗(E(1, m + 1)) ⊕ T C ∗(E(1, m + 1)) → 0 ⊕ T C ∗(E(1, m + 1)). We will also make use of this identification in our analysis of C ∗(S mE) in Section 8. 0 ) = 0. 0)) such that x0 7→ 0 ) and, for 0 < t < 1, Ut : ℓ2(E(1, m + 1)∗) → ℓ2(S mE∗ Lemma 7.15. Let E be a locally finite graph with no sources. Let U0 : ℓ2(E(0, m)∗) → ℓ2(S mE∗ t ) be the unitaries of Re- mark 6.5. Let πE(0,m) : T C ∗(E(0, m))) → B(ℓ2(E(0, m)∗)) and πE(1,m+1) : T C ∗(E(1, m + 1))) → B(ℓ2(E(1, m + 1)∗)) be the path-space representations. For x ∈ T C ∗(SE), we have lim tց0 U ∗ t xℓ2(SmE∗ t )Ut = πE(1,m+1)(1,m+1(π−1 E(0,m)(U ∗ 0 xℓ2(SmE∗ 0 )U0))). Proof. For x ∈ ρm(C0(SE0)), this follows from Equation 7.3 in the proof of Lemma 7.11, and for x ∈ ψm(Cc(S mE1)), it follows from Equation 7.4 in the same proof. Since T C ∗(S mE) is generated by ρm(C0(SE0)) ∪ ψm(Cc(S mE1)), the result follows. (cid:3) We are now able to give an explicit description of T C ∗(S mE) provided that enough vertices of E emit at least two edges. Theorem 7.16. Let E be a locally finite graph with no sources and fix m ∈ N \ {0}. Suppose that for every v ∈ E0 there exist n ≥ 1 and µ ∈ Enmv such that E1r(µ) ≥ 2. Let 1,m+1 : T C ∗(E(0, m)) ֒→ T C ∗(E(1, m+1)) be the injective homomorphism of Lemma 2.1. There is an isomorphism κm : T C ∗(S mE) → {f ∈ C([0, 1], T C ∗(E(1, m + 1))) : f (0) ∈ 1,m+1(T C ∗(E(0, m)))} such that for a ∈ C0(SE0) and ξ ∈ Cc(S mE1), we have Proof. For t ∈ S let qt : T C ∗(S mE) → T C ∗(S mE)t be the quotient map a 7→ a + Jt. For t ∈ (0, 1), let θt : T C ∗(E(1, m+ 1)) → T C ∗(S mE)t be the isomorphism of Proposition 7.9, and κm(ρm(a))(t) = κm(ψm(ξ))(t) = Pv∈E0 a([v])1,m+1(Qv) Pe∈E1 a([e, t])Qe Pv∈E0 a([v])Pe∈E1v Qe Pµ∈Em ξ([µ])(Tµ) Pν∈Em+1 ξ([ν, t])Tν Pµ∈Em ξ([µ])Pe∈E1r(µ) Teµ if t = 0 if t ∈ (0, 1) if t = 1 if t = 0 if t ∈ [0, 1) if t = 1. 34 AIDAN SIMS t and define εt := θ−1 ◦ qt : T C ∗(S mE) → T C ∗(E(1, m + 1)). Let ε1 : T C ∗(S mE) → T C ∗(E(1, m+1)) be the map of Lemma 7.12, and let ε0 : T C ∗(S mE) → T C ∗(E(1, m+1)) be the homomorphism of Lemma 7.11. Proposition 7.14 shows that ε1 is surjective and that the range of ε0 is 1,m+1(T C ∗(E(0, m))). Fix a ∈ C0(SE0) and ξ ∈ Cc(S mE1). Lemma 7.3 implies that for any a ∈ T C ∗(S mE) the function t 7→ εt(a) is continuous at each t ∈ (0, 1); Lemmas 7.11 and 7.12 show that it is continuous at 0 and 1 as well. Hence there is a homomorphism κm : T C ∗(S mE) → C([0, 1], T C ∗(E(1, m + 1))) given by κm(a)(t) = εt(a) for all a ∈ T C ∗(S mE) and t ∈ [0, 1]. To see that κ is injective, suppose that κ(a) = 0. We must show that a = 0. Propo- sition C.10(c) of [32] shows that, kak = supt∈S kqt(a)k, so it suffices to show that each qt(a) = 0. Since κ(a) = 0, we have εt(a) = 0 for all t. Since θt is an isomorphism for t ∈ (0, 1), we deduce that qt(a) = 0 for t 6= 0, and Proposition 7.14 shows that kq0(a)k = max{kε0(a)k, kε1(a)k} = 0. It remains to show that (7.7) κ(T C ∗(S mE)) = {f ∈ C([0, 1], T C ∗(E(1, m + 1))) : f (0) ∈ (T C ∗(E(0, m)))}. The containment ⊆ follows from Proposition 7.14. For the reverse containment, fix an element f of the right-hand side of (7.7). Proposition 7.14 shows that there ex- ists a ∈ T C ∗(S mE) such that η0(a) = f (0) and η1(a) = f (1). Hence f − κ(a) ∈ C0((0, 1), T C ∗(E(1, m + 1))). Consequently Lemma 7.3 shows that there exists b ∈ T C ∗(S mE) such that κ(b) = f − κ(a). Hence f = κ(a + b) ∈ κ(T C ∗(S mE)). (cid:3) We deduce that under the hypotheses of the preceding theorem, T C ∗(S mE) is homotopy equivalent to T C ∗(E(0, m)), and hence compute its K-theory. For this, recall from [28, Definition 3.2.5] that C ∗-homomorphisms ϕ0, ϕ1 : A → B are homotopic if there is a homomorphism ϕ : A → C([0, 1], B) such that ϕ(a)(0) = ϕ0(a) and ϕ(a)(1) = ϕ1(a) for all a ∈ A. Also recall that C ∗-algebras A and B are homotopy equivalent if there are homomorphisms ϕ : A → B and ψ : B → A such that ϕ ◦ ψ is homotopic to idB and ψ ◦ ϕ is homotopic to idA. We use the following elementary lemma. Lemma 7.17. Let A and B be C ∗-algebras, and let ι : B → A be an injective homomor- phism. Then the C ∗-algebra Cι := {f ∈ C([0, 1], A) : f (0) ∈ ι(B)} is homotopy equivalent to B. Proof. Define ϕ : Cι → B by ψ(f ) = ι−1(f (0)) and define ψ : B → Cι by ψ(b)(t) = ι(b) for all t ∈ [0, 1]. Then ϕ ◦ ψ is equal, and in particular homotopic, to idB. Define ρ : Cι → C([0, 1], Cι) by (cid:0)ρ(f )(s)(cid:1)(t) =(f (0) f (t − s) if t ≤ s if t > s. Then ρ is a homomorphism, and ρ(f )(0) = f = idCι(f ) and ρ(f )(1) = ψ(ϕ(f )) for all f ∈ Cι. So ψ ◦ ϕ is homotopic to idCι. (cid:3) Corollary 7.18. Let E be a locally finite graph with no sources and fix m ∈ N \ {0}. Suppose that for every v ∈ E0 there exist n ≥ 1 and µ ∈ Enmv such that E1r(µ) ≥ 2. Then T C ∗(S mE) is homotopy equivalent to T C ∗(E(0, m)), and we have K0(T C ∗(S mE)) ∼= ZE0, and K1(T C ∗(S mE)) = 0. THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 35 Proof. By Theorem 7.16, for the first statement we just have to show that the algebra A := {f ∈ C([0, 1], T C ∗(E(1, m + 1))) : f (0) ∈ (T C ∗(E(0, m)))} is homotopy equivalent to T C ∗(E(0, m)). This follows from Lemma 7.17 applied to ι = 1,m+1. Now [28, Proposition 3.2.6] shows that K0(A) ∼= K0(T C ∗(E(0, m))), and [28, Propo- sition 8.2.2(vi)] shows that K1(A) ∼= K1(T C ∗(E(0, m))). By [9, Theorem 4.1], the al- gebra T C ∗(E(0, m)) is isomorphic to the Toeplitz algebra of a Hilbert bimodule over C0(E(0, m)0) = C0(E0). Theorem 4.4 of [24] therefore implies that T C ∗(E(0, m)) is KK-equivalent to C0(E0), and hence K∗(T C ∗(E(0, m))) ∼= K∗(C0(E0)) ∼= (ZE0, 0). (cid:3) 8. Analysis of C ∗(S mE) In this section we analyse the quotient C ∗(S mE), using the analysis of T C ∗(S mE) from the previous section. The quotient map T C ∗(S mE) → C ∗(S mE) induces a homomor- phism πt : T C ∗(S mE)t → C ∗(S mE)t on each fibre. We show that under the isomor- ∼= T C ∗(E(1, m + phisms T C ∗(S mE)t 1)) ⊕ T C ∗(E(0, m)) described in the preceding section, these homomorphisms πt become the canonical quotient maps T C ∗(E(1, m + 1)) → C ∗(E(1, m + 1)) (for t 6= 0), and T C ∗(E(0, m)) ⊕ T C ∗(E(1, m + 1)) → C ∗(E(0, m)) ⊕ C ∗(E(1, m + 1)) (for t = 0). ∼= T C ∗(E(0, m + 1)) (for t 6= 0) and T C ∗(S mE)0 projections ∆v := Qv −Pe∈vE1 TeT ∗ Recall that given any graph E, we denote by IE the ideal of T C ∗(E) generated by the e indexed by v ∈ E0. The path-space representation πE : T C ∗(E) → B(ℓ2(E∗)) restricts to an isomorphism of IE onto ⊕v∈E0K(ℓ2(E∗v)) ⊆ K(ℓ2(E∗)), and in particular satisfies π(Tµ∆s(µ)T ∗ ν ) = θµ,ν ∈ K(ℓ2(E∗s(µ))) for all µ, ν ∈ E∗ with s(µ) = s(ν). Lemma 8.1. Let E be a locally finite graph with no sources and fix m ∈ N \ {0}. Let K ⊳ T C ∗(S mE) be the ideal K := {x ∈ T C ∗(S mE) : xℓ2(SmE∗ t ) ∈ K(ℓ2(S mE∗ t )) for all t ∈ S}. Let κm : T C ∗(S mE) → {f ∈ C([0, 1], T C ∗(E(1, m + 1))) : f (0) ∈ 1,m+1(T C ∗(E(0, m)))} be the isomorphism of Theorem 7.16. Then κm(K) = {f ∈ κm(T C ∗(S mE)) : f (t) ∈ IE(1,m+1) for all t}. Proof. Fix x ∈ K. Corollary 7.10 shows that for t 6∈ {0, 1}, we have κ−1 m (x)t = π−1 E(1,m+1)(U ∗ t xℓ2(SmE∗ t )Ut). So the discussion preceding this lemma shows that κm(x)t ∈ K(ℓ2(E(1, m + 1))) for t 6∈ {0, 1}. Since K(ℓ2(E(1, m + 1))) is closed, we deduce from the definitions of ε0 and ε1 that κ−1 m (x)1 belong to K(ℓ2(E(1, m + 1))) as well. So κm(K) ⊆ {f ∈ κm(T C ∗(S mE)) : f (t) ∈ IE(1,m+1) for all t}. m (x)0 and κ−1 For the reverse inclusion, suppose that f ∈ κm(T C ∗(S mE)) and that f (t) ∈ IE(1,m+1) t ) ∈ m (f ). Corollary 7.10 shows that for t 6= 0 we have xℓ2(SmE∗ for all t. Let x := κ−1 K(ℓ2(S mE∗ t ), and Lemma 7.15 shows that πE(1,m+1)(1,m+1(π−1 E(0,m)(U ∗ 0 xℓ2(SmE∗ 0 )U0))) = lim tց0 U ∗ t xℓ2(SmE∗ t )Ut. Since each U ∗ t xℓ2(SmE∗ t )Ut ∈ K(ℓ2(E(1, m + 1)∗)), we deduce that (8.1) πE(1,m+1)(1,m+1(π−1 E(0,m)(U ∗ 0 xℓ2(SmE∗ 0 )U0))) ∈ K(ℓ2(E(1, m + 1)∗)). 36 AIDAN SIMS Lemma 2.1 shows that −1 K(ℓ2(E(1, m + 1)∗)) and πE(0,m) carries IE(0,m) to K(ℓ2(E(0, m)∗)), we deduce that 1,m+1(IE(1,m+1)) = IE(0,m). Since πE(1,m+1) carries IE(1,m+1) into πE(0,m)(−1 1,m+1(π−1 E(1,m+1)(K(ℓ2(E(1, m + 1)∗))) ⊆ K(ℓ2(E(0, m)∗)). So (8.1) gives U ∗ we have xℓ2(SmE∗ 0 xℓ2(SmE∗ t ) ∈ K(ℓ2(S mE∗ 0 )U0 ∈ K(ℓ2(E(0, m)∗)) and hence xℓ2(SmE∗ 0 ) ∈ K(ℓ2(S mE∗ t )) for all t ∈ S, and therefore f = κm(x) ∈ κm(K). 0)). So (cid:3) Theorem 8.2. Let E be a locally finite graph with no sources and fix m ∈ N\{0}. Suppose that for every v ∈ E0 there exist n ≥ 1 and µ ∈ Enmv such that E1r(µ) ≥ 2. There is an isomorphism κm : C ∗(S mE) → C([0, 1], C ∗(E(1, m + 1))) such that for a ∈ C0(SE0) and ξ ∈ Cc(S mE1), we have κm(ρm(a))(t) = κm( ψm(ξ))(t) = Pv∈E0 a([v])Pe∈vE1 pe Pe∈E1 a([e, t])pe Pv∈E0 a([v])Pe∈E1v pe Pµ∈Em ξ([µ])Pe∈s(µ)E1 sµe Pν∈Em+1 ξ([ν, t])sν Pµ∈Em ξ([µ])Pe∈E1r(µ) seµ if t = 0 if t ∈ (0, 1) if t = 1 if t = 0 if t ∈ [0, 1) if t = 1. and Proof. Let A := {f ∈ C([0, 1], T C ∗(E(1, m + 1))) : f (0) ∈ 1,m+1(T C ∗(E(0, m)))}, I := {f ∈ A : f (t) ∈ IE(1,m+1) for all t}. and Lemma 8.1 and the definition of C ∗(S mE) show that the isomorphism κm : T C ∗(S mE) → {f ∈ C([0, 1], T C ∗(E(1, m + 1))) : f (0) ∈ (T C ∗(E(0, m)))} of Theorem 7.16 descends to an isomorphism κ′ : C ∗(S mE) → A/I. The last statement of Lemma 8.1 shows that I = {f : f (t) ∈ IE(1,m+1) for t 6= 0 and f (0) ∈ IE(0,m)}. It follows that there is an injective homomorphism κ′′ : A/I → C([0, 1], C ∗(E(1, m + 1))) that carries x + I to the function t 7→(x(t) + IE(1,m+1) −1 1,m+1(x(0)) + IE(0,m) if t 6= 0 if t = 0. We claim that κ′′ is surjective. For each t ∈ (0, 1], we have {κ′′(x)(t) : x ∈ A/I} = C ∗(E(1, m + 1)) because {x(t) : x ∈ A} = T C ∗(E(1, m + 1)). At t = 0 we have κ′′({x(0) : x ∈ A/I}) = 1,m+1(C ∗(E)) = C ∗(E(1, m + 1)) because {x(0) : x ∈ A} = 1,m+1(T C ∗(E(0, m))). The extension eκ′′ of κ′′ to M(A/I) carries the canonical copy of C([0, 1]) in M(A/I) to the canonical copy of C([0, 1]) ∈ M(C([0, 1], C ∗(E(1, m + 1)))). Thus for f ∈ C([0, 1]) and x ∈ A we have f ·κ′′(a+I) = κ′′((f ·a)+I) ∈ κ′′(A/I). It there- fore follows from [32, Proposition C.24] that κ′′(A/I) is dense in C([0, 1], C ∗(E(1, m+1))), and therefore all of C([0, 1], C ∗(E(1, m + 1))) because the range of a C ∗-homomorphism is closed. (cid:3) To finish this section, we use our earlier results to describe, up to Morita equivalence, the C ∗-algebras C ∗(S lE) for rational values of l and for locally finite graphs E with no sources or sinks such that for every v and every m there exist p, q ≥ 1, µ ∈ vEpm and ν ∈ Eqmv THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 37 such that s(µ)E1 ≥ 2 and E1r(ν) ≥ 2. In particular, we show that this applies to any finite, strongly connected graph with period 1 (in the sense of Perron -- Frobenius theory). We first need the following elementary result about the Cuntz -- Krieger algebras of the delayed graphs associated to a graph E. Lemma 8.3. Let E be a locally finite graph with no sinks or sources, and let m, n be coprime positive integers. Then C ∗(Dn(E)(1, m+1)) is Morita equivalent to C ∗(E(0, m)). Proof. By [1, Theorem 3.1], we have C ∗(Dn(E)(1, m + 1)) ∼= C ∗(Dn(E)(0, m)), so it suffices to show that the latter is Morita equivalent to C ∗(E(0, m)). To see this, observe that, by [2, Lemma 1.1], the seriesPv∈E0 pv converges to a multi- We claim that P is full. For this, first partition Dn(E)0 as Dn(E)0 = Sj∈Z/nZ Vj by plier projection P ∈ MC ∗(Dn(E)(0, m)). setting Vj :=(E0 {we,j : e ∈ E1} if j = 0 if j 6= 0. Then for α ∈ Dn(E)1, we have r(α) ∈ Vj if and only if s(α) ∈ Vj+1, and it follows that for λ ∈ Dn(E)∗, we have r(λ) ∈ Vj if and only if s(λ) ∈ Vj+[λ]n. Now fix u ∈ Dn(E)0, say u ∈ Vj. Since m, n are coprime, there exists k ∈ N such that km ≡ j (mod n). Since E has no sinks, there exists λ ∈ Ekmu, and since s(λ) ∈ Vj, it follows that r(λ) ∈ Vj−[km] = V0, and therefore pr(λ) ≤ P . Hence pu = s∗ λpr(λ)sλ belongs to the ideal generated by P . Now for α ∈ Dn(E)1, the generator sα = sαps(α) also belongs to the ideal generated by P , and it follows that P is full. To complete the proof, it suffices to show that P C ∗(Dn(E)(0, m))P ∼= C ∗(E(0, m)). We begin by constructing a Cuntz -- Krieger E(0, m)-family in P C ∗(Dn(E)(0, m))P . First, for µ ∈ Em, we define α(µ) ∈ Dn(E)(0, m)n by α(µ) = fµ1,1 · · · fµ1,nfµ2,1 · · · fµ2,n · · · fµm,1 · · · fµm,n. For v ∈ E(0, m)0 = E0, we define Pv := pv ∈ P C ∗(Dn(E)(0, m))P , and for µ ∈ E(0, m)1 = Em, we define Sµ := sα(µ) ∈ P C ∗(Dn(E)(0, m))P . It is routine to see that (P, S) is a Cuntz -- Krieger E(0, m)-family, so the universal property of C ∗(E(0, m)) implies that there is a homomorphism π : C ∗(E(0, m)) → P C ∗(Dn(E)(0, m))P such that π(pv) = Pv for all v ∈ E0 and π(sµ) = Sµ for all µ ∈ E(0, m)1. We have α(Em) = (Dn(E)(0, m)n)E0 ⊆ Dn(E)(0, m)n. The universal property of C ∗(Dn(E)(0, m)) shows that there is an action β of T on C ∗(Dn(E)(0, m)) such that βz(sν) = sν for all ν ∈ (Dn(E)(0, m)1) \ (Dn(E)(0, m)1)E0, and such that βz(sν) = zsν for all ν ∈ (Dn(E)(0, m)1)E0. Since gcd(m, n) = 0, for each µ ∈ Dn(E)(0, m)n, if we factor α(µ) = α1 · · · αn with each αi ∈ Dn(E)(0, m)1, we have s(αn) ∈ E0 and s(αi) 6∈ E0 for i < n. Consequently βz(sα(µ)) = sα1sα2 · · · sαn−1(zsαn) = zsα(µ). Hence, writing γ for the gauge action on C ∗(E(0, m)) we have π ◦ γz = βz ◦ π. The gauge-invariant uniqueness theorem [2, Theorem 2.1] therefore implies that π is injective. It now suffices to show that the range of π is P C ∗(Dn(E)(0, m))P . We have ζ : η, ζ ∈ V0Dn(E)(0, m)∗, s(η) = s(ζ)}. P C ∗(Dn(E)(0, m))P = span{sηs∗ Fix η, ζ ∈ V0(Dn(E)(0, m)∗) such that s(η) = s(ζ), say η ∈ Dn(E)(0, m)k = Dn(E)km and ζ ∈ Dn(E)(0, m)l = Dn(E)lm. Then s(η) ∈ V[km] and s(ζ) ∈ V[lm] forcing km ≡ 38 AIDAN SIMS lm (mod n). Since gcd(m, n) = 1, we deduce that k ≡ l (mod n). Fix p such that k + p ∈ nZ. Then l + p ∈ nZ too, and the Cuntz -- Krieger relation forces sηs∗ ζ = Xξ∈s(η)Dn(E)pm sηξs∗ ζξ = Xξ∈s(η)Dn(E)(0,m)p sηξs∗ ζξ By construction, each ηξ has the form α(µ1) · · · α(µk+p) for some µi ∈ E(0, m)1, and so each sηξ ∈ π(C ∗(E(0, m))), and similarly for each sζξ. So sηs∗ ζ ∈ π(C ∗(E(0, m))). Thus π is an isomorphism of C ∗(E(0, m)) onto P C ∗(Dn(E)(0, m))P as required. (cid:3) Corollary 8.4. Let E be a locally finite graph with no sinks or sources. Suppose that for every v ∈ E0 and every m ∈ Z \ {0}, there exist n ≥ 1 and µ ∈ Enmv such that E1r(µ) ≥ 2. For n ∈ N \ {0} and m ∈ Z such that gcd(m, n) = 1, (8.2) C ∗(S C ∗(S C ∗(S m m nE) ∼Me C([0, 1], C ∗(E(0, m))) nE) ∼= C0(SE0) ⊗ C(T) nE) ∼Me C([0, 1], C ∗(Eop(0, −m))) m if m > 0, if m = 0, and if m < 0. In particular, K∗(C ∗(S m nE)) ∼= E)m), ker(1 − (At (cid:0) coker(1 − (At (cid:0)K 0(SE0) ⊕ K 1(SE0), K 0(SE0) ⊕ K 1(SE0)(cid:1) (cid:0) coker(1 − Am E ), ker(1 − Am E)m(cid:1) E )(cid:1) if m > 0, if m = 0, and if m < 0. Proof. For m > 0, Theorem 8.2 combined with Lemma 8.3 shows that C ∗(S m n) ∼Me C([0, 1], C ∗(E(0, m))), and then for m < 0, it follows from Lemma 6.11 that C ∗(S m n) ∼Me C([0, 1], C ∗(Eop(0, −m))). Combining this with Theorem 6.10(2) proves the first state- ment. For the second statement, since K∗(C([0, 1], C ∗(E(0, m)))) ∼= K∗(C ∗(E(0, m))) (for example, apply Lemma 7.17 to ι = idC∗(E(0,m)), and then use [28, Proposition 3.2.6 and Theorem 8.2.2(vi)]), the computation [26, Theorem 3.2] of K-theory for graph C ∗-algebras gives A similar argument gives K∗(C([0, 1], C ∗(E(0, m)))) ∼=(cid:0) coker(1 − (At K∗(C([0, 1], C ∗(E(0, m)))) ∼=(cid:0) coker(1 − Am E)m), ker(1 − At E ), ker(1 − Am E)m(cid:1) E )(cid:1) for m > 0. for m < 0 Eop = AE and our hypotheses are symmetrical in E and Eop. Finally, the because At Kunneth theorem and that operator K-theory agrees with topological K-theory for com- mutative C ∗algebras implies that Ki(C0(SE0) ⊗ C(T)) ∼= K 0(C0(SE0)) ⊕ K 1(C0(SE0)) for i = 1, 2. (cid:3) We finish by applying Corollary 8.4 to strongly connected finite graphs with period 1. We say that a graph E is strongly connected if for all v, w ∈ E0 the set vE∗w \ E0 is nonempty (we take the convention that a graph consisting of a single vertex and no edges is not strongly connected). If E is a strongly connected finite graph, then the period of E is defined as P (E) := gcd{µ : µ ∈ E∗ \ E0, r(µ) = s(µ)}. Also recall that if E is a graph, then there is a map ∂ : ZE0 → ZE1 given by ∂(a)(e) = a(r(e)) − a(s(e)). The 0th and 1st homology groups of E are defined by H1(E) = ZE1/∂1(ZE0), and H0(E) = ker(∂1). The higher homology groups Hn(E), THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 39 n ≥ 2 are trivial (see, for example, [20, Remark 3.6]). The group H0(E) is isomorphic to the free abelian group generated by the connected components of E. We say that a finite graph E is a simple cycle if, putting n = E0, there are bijections i 7→ vi and i 7→ ei of Z/nZ onto E0 and E1 respectively such that r(ei) = vi and s(ei) = vi+1 for all i. Corollary 8.5. Let E be a finite strongly connected graph that is not a simple cycle, and suppose that P (E) = 1. Then for every v ∈ E0 and every m ∈ Z \ {0}, there exist n ≥ 1 and µ ∈ Enmv such that E1r(µ) ≥ 2. For n > 0 and m ∈ Z with m, n coprime, C ∗(S C ∗(S In particular, m nE) ∼Me C([0, 1], C ∗(E(0, m))) nE) ∼= C0(SE0) ⊗ C(T) m K∗(C ∗(S m nE)) ∼=((cid:0) coker(1 − Am (cid:0)Z ⊕ H1(E), Z ⊕ H1(E)(cid:1) E ), ker(1 − Am E )(cid:1) if m 6= 0, and if m = 0. if m 6= 0 if m = 0. Proof. First observe that the opposite graph Eop is also strongly connected with P (Eop) = 1, and is also not a simple cycle. To prove the first statement, it therefore suffices to consider m > 0 (the case m = 0 follows from the second line of (8.2)). So fix m ≥ 1 and v ∈ E0. Since E is strongly connected, we have E1v ≥ 1 for all v, and since E is not a simple cycle, a counting argument shows that there exists w ∈ E0 such that E1w ≥ 2. Since E is strongly connected, the set wE∗v is nonempty, say α ∈ wE∗v. It is standard (see for example [21, Lemma 6.1] applied with k = 1) that P (E) = {λ − ν : λ, ν ∈ vE∗v}. In particular, there are cycles λ, ν ∈ vE∗v such that λ − ν = m − α. It follows that αλνm−1 = α + (λ − ν) + mν = m(ν + 1). So n := ν + 1 and µ := αλνm−1 satisfy µ ∈ Enmv and E1r(µ) = E1w ≥ 2. By Corollary 8.4, it now suffices to show that C([0, 1], C ∗(Eop(0, m))) ∼Me C([0, 1], C ∗(E(0, m))) for all m ≥ 1, and that K∗(C(SE0 × S)) ∼=(cid:0)Z ⊕ H1(E), Z ⊕ H1(E)(cid:1). strongly connected finite graphs that are not simple cycles. Since Fix m > 0. Since E has period 1, it is easy to see that E(0, m) and Eop(0, m) are ker(At) ∼= ker(A) and coker(At) ∼= coker(A), (see, for example [13, Section 6.2]), we have K∗(C ∗(Eop(0, m))) ∼= K∗(C ∗(E(0, m))), and so [27, Theorem 6.1] shows that C ∗(Eop(0, m)) ∼Me C ∗(E(0, m)). Thus the function algebras C([0, 1], C ∗(Eop(0, m))) and C([0, 1], C ∗(E(0, m))) are Morita equivalent as well. Now take m = 0. We have K ∗(S) = (Z, Z). Thus, by the Kunneth theorem in K- theory, it suffices to show that K 0(SE0) ∼= Z and K 1(SE0) ∼= H1(E). Since SE0 and its suspension are finite CW-complexes of dimension at most 2, [31, Theorem 1] shows that K 0(SE0) ∼= Ln≥0 H2n(SE0) = H0(SE0) and that K 1(SE0) is isomorphic to the direct sum of the even homology groups of the suspension of SE0, and hence to the direct sum of the odd homology groups of SE0. Theorem 6.3 of [20] gives H∗(SE0) ∼= H∗(E), and we have H0(E) ∼= Z because E is connected. (cid:3) 40 AIDAN SIMS References [1] T. Bates, Applications of the gauge-invariant uniqueness theorem for graph algebras, Bull. Austral. Math. Soc. 66 (2002), 57 -- 67. [2] T. Bates, D. Pask, I. Raeburn and W. Szyma´nski, The C ∗-algebras of row-finite graphs, New York J. Math. 6 (2000), 307 -- 324. [3] N. Brownlowe, T.M. Carlsen and M.F. Whittaker, Graph algebras and orbit equivalence, Ergodic Theory Dynam. Systems 37 (2017), 389 -- 417. [4] A.L. Carey, J. Phillips and A. Rennie, Semifinite spectral triples associated with graph C ∗-algebras, Aspects Math., E38, Traces in number theory, geometry and quantum fields, 35 -- 56, Friedr. Vieweg, Wiesbaden, 2008. [5] J. Cuntz and W. Krieger, A class of C ∗-algebras and topological Markov chains, Invent. Math. 56 (1980), 251 -- 268. [6] D. Drinen and M. Tomforde, Computing K-theory and Ext for graph C ∗-algebras, Illinois J. Math. 46 (2002), 81 -- 91. [7] S. Eilers and M. Tomforde, On the classification of nonsimple graph C ∗-algebras, Math. Ann. 346 (2010), 393 -- 418. [8] M. Enomoto and Y. Watatani, A graph theory for C ∗-algebras, Math. Japon. 25 (1980), 435 -- 442. [9] N.J. Fowler and I. Raeburn, The Toeplitz algebra of a Hilbert bimodule, Indiana Univ. Math. J. 48 (1999), 155 -- 181. [10] J.H. Hong and W. Szyma´nski, Quantum lens spaces and graph algebras, Pacific J. Math. 211 (2003), 249 -- 263. [11] A. an Huef and I. Raeburn, The ideal structure of Cuntz -- Krieger algebras, Ergodic Theory Dynam. Systems 17 (1997), 611 -- 624. [12] S. Kaliszewski, A. Kumjian, J. Quigg and A. Sims, Topological realizations and fundamental groups of higher-rank graphs, Proc. Edinb. Math. Soc. (2) 59 (2016), 143 -- 168. [13] J. Kaminker andI. Putnam, K-theoretic duality for shifts of finite type, Commun. Math. Phys. 187 (1997), 509 -- 522. [14] T. Katsura, A class of C ∗-algebras generalizing both graph algebras and homeomorphism C ∗-algebras I. Fundamental results, Trans. Amer. Math. Soc. 356 (2004), 4287 -- 4322. [15] T. Katsura, A class of C ∗-algebras generalizing both graph algebras and homeomorphism C ∗-algebras II. Examples, Internat. J. Math. 17 (2006), 791 -- 833. [16] T. Katsura, A class of C ∗-algebras generalizing both graph algebras and homeomorphism C ∗-algebras III. Ideal structures, Ergodic Theory Dynam. Systems 26 (2006), 1805 -- 1854. [17] T. Katsura, A class of C ∗-algebras generalizing both graph algebras and homeomorphism C ∗-algebras. IV. Pure infiniteness, J. Funct. Anal. 254 (2008), 1161 -- 1187. [18] A. Kumjian, D. Pask, I. Raeburn and J. Renault, Graphs, groupoids, and Cuntz -- Krieger algebras, J. Funct. Anal. 144 (1997), 505 -- 541. [19] A. Kumjian, D. Pask and I. Raeburn, Cuntz -- Krieger algebras of directed graphs, Pacific J. Math. 184 (1998), 161 -- 174. [20] A. Kumjian, D. Pask and A. Sims, Homology for higher-rank graphs and twisted C ∗-algebras, J. Funct. Anal. 263 (2012), 1539 -- 1574. [21] M. Laca, N.S. Larsen, S. Neshveyev, A. Sims and S.B.G. Webster, Von Neumann algebras of strongly connected higher-rank graphs, Math. Ann. 363 (2015), 657 -- 678. [22] H. Li, D. Pask and A. Sims, An elementary approach to C ∗-algebras associated to topological graphs, New York J. Math. 20 (2014), 447 -- 469. [23] P.S. Muhly and M. Tomforde, Topological quivers, Internat. J. Math. 16 (2005), 693 -- 755. [24] M.V. Pimsner, A class of C ∗-algebras generalizing both Cuntz -- Krieger algebras and crossed products by Z, Fields Inst. Commun., 12, Free probability theory (Waterloo, ON, 1995), 189 -- 212, Amer. Math. Soc., Providence, RI, 1997. [25] I. Raeburn, Graph algebras, Published for the Conference Board of the Mathematical Sciences, Washington, DC, 2005, vi+113. [26] I. Raeburn and W. Szyma´nski, Cuntz -- Krieger algebras of infinite graphs and matrices, Trans. Amer. Math. Soc. 356 (2004), 39 -- 59. [27] M. Rørdam, Classification of Cuntz -- Krieger algebras, K-Theory 9 (1995), 31 -- 58. THE SUSPENSION OF A GRAPH, AND ASSOCIATED C ∗-ALGEBRAS 41 [28] M. Rørdam, F. Larsen and N. Laustsen, An introduction to K-theory for C ∗-algebras, Cambridge University Press, Cambridge, 2000, xii+242. [29] A. Sims, The co-universal C ∗-algebra of a row-finite graph, New York J. Math. 16 (2010), 507 -- 524. [30] A.P.W. Sørensen, Geometric classification of simple graph algebras, Ergodic Theory Dynam. Systems 33 (2013), 1199 -- 1220. [31] A. Thomas, A relation between K-theory and cohomology, Trans. Amer. Math. Soc. 193 (1974), 133 -- 142. [32] D.P. Williams, Crossed products of C ∗-algebras, American Mathematical Society, Providence, RI, 2007, xvi+528. E-mail address: [email protected] School of Mathematics and Applied Statistics, University of Wollongong, NSW 2522, AUSTRALIA
1911.01689	1	1911	2019-11-05T09:56:04	Approximately order zero maps between C*-algebras	[ "math.OA", "math.FA" ]	We investigate linear operators between C$^\ast$-algebras which approximately preserve involution and orthogonality, the latter meaning that for some $\varepsilon>0$ we have $\\|\phi(x)\phi(y)\\|\leq\varepsilon\\|x\\|\\|y\\|$ for all positive $x,y$ with $xy=0$. We establish some structural properties of such maps concerning approximate Jordan-like equations and almost commutation relations. In some situations (e.g. when the codomain is finite-dimensional), we show that $\phi$ can be approximated by an approximate Jordan $^\ast$-homomorphism, with both errors depending only on $\\|\phi\\|$ and $\varepsilon$.	math.OA	math	APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS TOMASZ KOCHANEK Abstract. We investigate linear operators between C∗-algebras which approximately preserve involution and orthogonality, the latter meaning that for some ε > 0 we have kφ(x)φ(y)k 6 εkxkkyk for all positive x, y with xy = 0. We establish some structural properties of such maps concerning approximate Jordan-like equations and almost com- mutation relations. In some situations (e.g. when the codomain is finite-dimensional), we show that φ can be approximated by an approximate Jordan ∗-homomorphism, with both errors depending only on kφk and ε. 9 1 0 2 v o N 5 ] . A O h t a m [ 1 v 9 8 6 1 0 . 1 1 9 1 : v i X r a Contents 1. Introduction 2. Preliminaries 3. An extension result and approximate Jordan equations 4. The second adjoint on bounded Borel functions 5. Almost commutation relations 6. The range of an approximately order zero map 7. Decomposition -- reducing to the unital case 8. Examples References 1 5 8 14 18 26 30 40 42 1. Introduction There is a widely developed theory concerning the question to what extent the zero-product structure determines the whole structure of a given Banach algebra, or to what extent the action on zero-product elements characterizes homomorphisms, derivations etc. For exam- ple, [2] contains a series of results characterizing zero-product-preserving maps and dealing with the question whether such maps must be automatically weighted homomorphisms. In [1], it is shown that a certain generalized zero-product-preserving property force the map in question to be a homomorphism. In the setting of C∗-algebras, an important role is played by linear operators which preserve orthogonality or, equivalently, preserve zero products of self-adjoint elements. Such maps are usually assumed to be completely positive, as well-behaved amplifications 2010 Mathematics Subject Classification. Primary 46L05, Secondary 46L85. Key words and phrases. Order zero map, disjointness preserving, almost Jordan ∗-homomorphism. This work was supported by the GA CR project 18-00960Y. 1 2 TOMASZ KOCHANEK to matrix algebras are quite useful in noncommutative topology. Winter and Zacharias [25] called those maps order zero and they exhibited their importance as 'building blocks' of noncommutative partitions of unity. Consequently, order zero maps proved to be the key ingredient to define nuclear dimension of C∗-algebras, a noncommutative analogue of the covering dimension (see [26] and the references therein). In this paper, we deal with approximately order zero operators between C∗-algebras, a notion somewhat analogous to the notion of approximately multiplicative maps which were profoundly investigated by B.E. Johnson in a series of his paper (see, e.g., [17] and [18]). Favoring the ∗-algebra structure over the order structure, we do not assume complete positivity, instead we deal with operators which simultaneously preserve orthogonality and involution. Recall that the usual relation of orthogonality x ⊥ y, for x, y from a given C∗-algebra, is defined by the condition xy = yx = x∗y = xy∗ = 0. Note that for self-adjoint x, y the condition x ⊥ y is equivalent to xy = 0. For a C∗-algebra A, we denote by Asa and A+ the sets of all self-adjoint and all positive elements of A, respectively. Definition 1.1. Let A, B be C∗-algebras, φ : A → B a bounded linear operator and ε > 0. We say that φ is an ε-order zero map (ε-o.z. for short) if it satisfies the condition (1.1) x, y ∈ A+, x ⊥ y ==⇒ kφ(x)φ(y)k 6 εkxkkyk. We say φ is ε-self-adjoint (ε-s.a. for short) if it satisfies kφ(x∗) − φ(x)∗k 6 εkxk for every x ∈ A. Finally, we call φ an ε-disjointness preserving map (ε-d.p. for short), provided it is both ε-o.z. and ε-s.a. Order zero maps, self-adjoints maps and disjointness preserving maps are defined as above with ε = 0. It is worth mentioning that the stability problem for almost disjointness preserving operators between C(X)-spaces was first considered by Dolinar [12] and then completely solved in a series of papers by Araujo and Font ([5], [6], [7]). Recently, almost disjointness preserving operators on Banach lattices were studied by Oikhberg and Tradacete [22]. Below, we recall two important results which characterize operators preserving orthog- onality on C∗-algebras. The first one says, roughly, that self-adjoint maps of this type are compressions of Jordan ∗-homomorphisms, whereas the second one says that completely positive maps of this type are compressions of ∗-homomorphisms. For more results char- acterizing disjointness preserving maps, the reader may consult [10] and [13]. Wolff [27] defined a bounded linear operator T : A → B to be disjointness preserving, provided that it is self-adjoint, i.e. T (x∗) = T (x)∗ for every x ∈ A, and T (x)T (y) = 0 for all x, y ∈ Asa with xy = 0. Winter and Zacharias [25] defined a c.p. map ϕ : A → B to have order zero, provided that ϕ(x) ⊥ ϕ(y) for all x, y ∈ A+ with xy = 0. In fact, such a map must preserve orthogonality of all elements of A (see [25, Remark 2.4]). Theorem 1.2 ([27, Thm. 2.3]). Let A and B be C∗-algebras with A unital. Let T : A → B be a disjointness preserving operator with h := T (1A). Then we have: (a) T (A) ⊆ C := h{h}′; APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 3 M (C) such that S(1A) = 1M (C) and T (x) = hS(x) for every x ∈ A. (b) there exists a Jordan ∗-homomorphism S mapping A into the multiplier algebra Theorem 1.3 ([25, Thm. 3.3]). Let A and B be C∗-algebras and ϕ : A → B a c.p. order zero map with C := C∗(ϕ(A)). Then, there exists a positive element h ∈ M (C) ∩ C′ with khk = kϕk and a ∗-homomorphism π : A → M (C)∩{h}′ such that ϕ(x) = hπ(x) for every x ∈ A. Moreover, if A is unital, then h = ϕ(1A). The study of almost zero-product-preserving, or almost orthogonality preserving maps can be regarded as a part of the general Ulam's stability problem [24]. One deep theorem in this context was given by Alaminos, Extremera and Villena [3] who showed that in many cases an almost zero-product-preserving operator T must be close to a compression of a homomorphism. However, a crucial assumption was that said map is surjective, in which case one can consider its openness index defined by op(T ) = inf(cid:8)M > 0 : for every x ∈ X there is y ∈ Y with T (x) = y and kxk 6 Mkyk(cid:9). (By the Open Mapping Theorem, such a constant is finite whenever T maps a Banach space onto a Banach space.) Theorem 1.4 ([3, Thm. 4.7]). Let A be either the group algebra L1(G) for a locally compact group G or a C∗-algebra and let B be a Banach algebra. Assume both A and B are amenable and that for some Banach B-bimodule X the multiplier algebra M (B) is isomorphic as a Banach B-bimodule to X ∗. For all ε, K, M > 0 there exists δ = δ(ε, K, M) > 0 such that for every surjective operator T ∈ L (A,B) satisfying: (i) kTk 6 K, (ii) op(T ) 6 M, (iii) kT (x)T (y)k 6 δkxkkyk for all x, y ∈ A with xy = 0, there exist an invertible element ν in the cener Z(M(B)) and a continuous epimorphism Φ : A → B with kT − νΦk 6 ε. At this point, let us stress that there is an essential difference between assuming that the product of values vanishes for all pairs with zero product or merely for those which are orthogonal in the C∗-algebraical sense. Indeed, it follows from another result by Alaminos, Extremera and Villena [3] (Theorem 2.1 below) that zero-product-preserving maps must satisfy a multiplicativity-like property -- in the unital case they must be simply multiplica- tive. On the other hand, by Wolff's Theorem 1.2, for (unital) d.p. maps we cannot go further than Jordan ∗-homomorphisms. In view of Theorem 1.2, given an ε-d.p. map φ we should expect that it can be ap- proximated by a compression of a Jordan ∗-homomorphism. Hence, a natural question is whether φ itself is 'almost' a compressed Jordan ∗-homomorphism, and a positive answer is provided by Proposition 3.2. A large part of the present paper is, however, devoted to the problem of approximating φ by genuine (unital) almost Jordan ∗-homomorphisms. Definition 1.5. Let A, B be C∗-algebras, φ : A → B a bounded linear operator and ε > 0. We say that φ is an ε-Jordan ∗-homomorphism (ε-J.h. for short) if it is ε-s.a. and satisfies kφ(x)2 − φ(x2)k 6 εkxk2 for every x ∈ A. 4 TOMASZ KOCHANEK Our main goal is to show that, in some natural situations, every ε-d.p. map can be approx- imated to within δ(ε) by a map which behaves like an η(ε)-J.h., where both δ(ε) and η(ε) converge to zero as ε → 0. Therefore, the stability problem for ε-d.p. maps gets reduced to the stability problem for δ-J.h. maps. We believe the latter one can be attacked by similar cohomological methods as those introduced by B.E. Johnson ([15], [16]) and applied to almost multiplicative maps ([17], [18]). We use standard notation. By L (A,B) we denote the space of all bounded linear operators between given C∗-algebras A and B. In the second dual A∗∗ we consider the usual (Arens) multiplication which extends the multiplication in A and makes A∗∗ a von Neumann algebra (see, e.g., [9, §III.5.2]). In Theorems A and B below we summarize main results of this paper. The first one is more of a structural nature; it follows from Lemma 3.1, Proposition 3.2, Lemma 6.4 and a combination of Proposition 5.2 with Corollary 6.5. Theorem B is a part of Theorem 7.2 and Corollary 8.2. Theorem A. Let A and B be C∗-algebras and φ ∈ L (A,B) be an ε-o.z. map. If A is nonunital and π is a nondegenerate representation of B on a Hilbert space, then φ can be extended to an ε-o.z. map φ† ∈ L (A†, π(B)′′) defined on the unitization A† of A. Assuming that A is unital and h := φ(1A), the following assertions hold true: (a) We have kφ(x)2 − hφ(x2)k 6 108εkxk for every x ∈ A. (b) There is an absolute constant K < ∞ such that if φ is self-adjoint, then its range or the range of φ lies close to the commutant {h}′ in the sense that for every complex polynomial P ∈ C[z] with P (0) = 0 and each x ∈ A we have z=khkP (z) · kxk, sup where C > 0 depends only on the degree of algebraicity of h. k[P (h), φ(x)]k 6 Cε khk Theorem B. Let A, B be C∗-algebras with A unital and let π be a nondegenerate repre- sentation of B on a Hilbert space H. Let also φ ∈ L (A,B) be a self-adjoint ε-o.z. map with some ε ∈ (0, 1] and with h := ψ(1A) being an algebraic element of B. Then, there exists a decomposition φ = φs + φr, where the operators φs, φr ∈ L (A, π(B)′′) satisfy the following conditions: (i) kφsk 6 (6K + 7)kφk4/5ε1/16; (ii) φr takes values in a corner subalgebra C of hπ(B)′′h; lies close to the hereditary subalgebra hBh in the sense that dist(cid:0)φ(x), hBh(cid:1) 6 Kkφk3/5ε1/5kxk for every x ∈ A. (c) If φ is self-adjoint and h 6= 0 is an algebraic element of B, then either kφk 6p(K + 2)5ε, APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 5 (iii) either φr = 0 or φr(1A) is invertible in C, in which case φr(1A)−1φr(· ) is a unital δ-J.h. map with where, again, C > 0 depends only on the degree of algebraicity of h. δ = 24(cid:0)C 2(K + 2)5 + 10C + 17(cid:1)kφkε1/16, In particular, if φ : A → Mn(C) is a positive ε-o.z. map, then there exists a corner subal- gebra C of Mn(C) and an operator Φ ∈ L (A,C) satisfying kφ − Φk 6 37kφk4/5ε1/16 and such that either Φ = 0 or Φ(1A) is invertible in C, in which case the operator Φ(1A)−1Φ(· ) is δ-J.h. with δ = O(cid:0)256n(cid:1)kφkε1/16 2. Preliminaries as n → ∞. In this section, we record a few simple technical observations which will be useful in the sequel. However, we should start with quoting a deep result by Alaminos, Extremera and Villena [3] which says, roughly speaking, that almost zero-product-preserving maps on C∗-algebras must satisfy an approximate version of a multiplicativity-like property. They introduced an error function defined by the formula where ξ(s) = A(s) + B(s) + Γ(s) with: 0 8π ξ ζ(s) = 2π(cid:12)(cid:12)2 sin(s) + s(1 − cos(s))(cid:12)(cid:12) + (cid:12)(cid:12)(cid:12) Xk∈Z,k6=0,1(cid:12)(cid:12)1 − eiks(cid:12)(cid:12) B(s) =(cid:12)(cid:12)(cid:12) 3 + 1)s−1/4 − 1(cid:1) − 1 if s > 0 q3(cid:0)( 172 π(cid:12)(cid:12)(cid:12)s + 2(cid:16)1 − cos(s) , Γ(s) = Xk∈Z,k6=0,1 1 − eis A(s) = πk2 1 1 s if s = 0, s (cid:17) cos(s)(cid:12)(cid:12)(cid:12), . sin(1 − k)s πk(k − 1) Theorem 2.1 ([3, Thm. 3.5]). Let A be a C∗-algebra, X a Banach space and Φ : A×A → X a bounded bilinear map satisfying x, y ∈ A, xy = 0 ==⇒ kΦ(x, y)k 6 εkxkkyk with some ε > 0. Let also K be any number satisfying K > kΦk and K > ε. Then kΦ(xy, z)−Φ(x, yz)k 6h(cid:16)172 3 + 1(cid:17)2 K 1/2ε1/2(cid:16)2 + ζ(cid:16) ε K(cid:17)(cid:17) + Kζ(cid:16) ε K(cid:17)ikxkkykkzk for all x, y, z ∈ A. 6 TOMASZ KOCHANEK Putting ε = 0 and Φ(x, y) = φ(x)φ(y), for a given map φ ∈ L (A,B), where B is a Banach algebra, we see that Theorem 2.1 yields a characterization of zero-product-preserving maps on C∗-algebras. In general, it reduces the study of almost zero-product-preserving maps to the study of stability of the equation φ(xy)φ(z) = φ(x)φ(yz). In contrast, the main goal of this paper is to reduce the study of almost order zero (almost disjointness preserving) maps to the study of almost Jordan homomorphisms, that is, stability of the equation φ(x)2 = φ(x2). It will be quite helpful for us to know the asymptotic behavior of the error function ζ. Lemma 2.2. We have ζ(s) = O(s1/16) as s → 0+. Proof. By elementary trigonometry, we have A(s) = O(s) as s → 0+. We shall prove that B(s) = O(s1/2) and Γ(s) = O(s1/2). The result will then follow, since ζ can be written as a composition ξ ◦ α, where For a moment, fix any M > 0 and 0 < s 6 π/M. Notice that for each k ∈ Z with C1√C2 C1 α(s) s1/8 = k 6 M we have with suitable constants C1, C2 > 0. pC2 − 3s1/4 − s1/8 −−−−→s→0+ (cid:12)(cid:12)1 − eiks(cid:12)(cid:12) 6(cid:12)(cid:12)1 − eiM s(cid:12)(cid:12) =p2(1 − cos Ms) = Msr2(cid:16) 1 + Xk6M,k6=0,1 = Xk>M ∞Xk=M +1 Xk∈Z,k6=0,1(cid:12)(cid:12)1 − eiks(cid:12)(cid:12) π Xk6M,k6=0 1 k2 + Therefore, 2! − < 4 π πk2 Ms (Ms)2 4! + (Ms)4 6! − . . .(cid:17) < Ms. 1 k2 < 4 πM + πMs 3 . Thus, putting M := s−1/2 with s → 0+ we see that B(s) = O(s1/2). (1 − k)s 6 Ms. Hence, with some absolute constant C > 0, we have Again, fix any M > 0 and note that for k ∈ Z, k < M we have sin(1 − k)s 6 Γ(s) = Xk∈Z,k6=0,1 ∞Xk=M 2 π < sin(1 − k)s πk(k − 1) 1 = Xk>M + Xk<M,k6=0,1 + CMs = (k − 1)k 2 π(M − 1) + CMs. Putting M := s−1/2 as above we obtain Γ(s) = O(s1/2). (cid:3) Lemma 2.3. Let φ : A → B be an δ-s.a. map between C∗-algebras A and B. Then there exists a self-adjoint map ψ : A → B such that kφ − ψk 6 1 2δ. Moreover, if φ is ε-o.z., then ψ can be picked to be (ε + 1 2δkφk)-o.z. APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 7 Proof. Consider the standard involution φ 7→ φ∗ in L (A,B) given by φ∗(x) = φ(x∗)∗ and define ψ = 1 2(φ + φ∗). Plainly, ψ is self-adjoint and satisfies the desired inequality. Now, if φ is ε-o.z., then for all x, y ∈ A+ with xy = 0 we have kφ(x)φ(y)∗k 6 kφ(x)φ(y)k + kφ(x)(φ(y)∗ − φ(y))k 6 (ε + δkφk)kxkkyk and, of course, the same estimate is valid for kφ(x)∗φ(y)k. Hence, 1 4kφ(x)φ(y) + φ(x)φ(y)∗ + φ(x)∗φ(y) + φ(x)∗φ(y)∗k 6 kψ(x)ψ(y)k = as desired. Lemma 2.4. Let φ : A → B be an ε-o.z. map between C∗-algebras A and B. Then: (4ε + 2δkφk)kxkkyk, (cid:3) 1 4 (a) for all x, y ∈ Asa with x ⊥ y we have kφ(x)φ(y)k 6 4εkxkkyk; (b) for all x, y ∈ A with x ⊥ y we have kφ(x)φ(y)k 6 16εkxkkyk; (c) if φ is a completely positive contraction, then kφ(x)φ(y)k 6 ε1/2kxkkyk for all x, y ∈ A with x ⊥ y. Proof. (a) Fix x, y ∈ Asa, x ⊥ y and let x = x1 − x2, y = y1 − y2 be the Jordan decompo- sitions of x and y, that is, x1, x2, y1, y2 ∈ A+, x1x2 = 0 and y1y2 = 0. These elements are defined by functional calculus on C∗(x, y), namely, x1 = f (x), x2 = g(x), y1 = f (y) and y2 = g(y), where f (t) = max{t, 0} and g(t) = − min{t, 0}. Since xy = 0, we have xiyj = 0 and hence kφ(xi)φ(yj)k 6 εkxikkyjk for all 1 6 i, j 6 2. Therefore, kφ(x)φ(y)k = k(φ(x1) − φ(x2))(φ(y1) − φ(y2))k 6 2Xi,j=1 kφ(xi)φ(yj)k 6 ε kxikkyjk 6 4εkxkkyk. 2Xi,j=1 2Xi,j=1 (b) We simply decompose x and y into real and imaginary parts: x = x1 + ix2, y = y1 + iy2. where x1 = 1 2i (y − y∗). By the assumption that x ⊥ y, we have xiyj = 0 for all 1 6 i, j 6 2. Of course, kxikkyjk 6 kxkkyk and hence assertion (a) yields 2i (x − x∗), y1 = 1 2 (x + x∗), x2 = 1 2(y + y∗), y2 = 1 kφ(x)φ(y)k = kφ(x1)φ(y1) − φ(x2)φ(y2) + iφ(x1)φ(y2) + iφ(x2)φ(y1)k 6 2Xi,j=1 kφ(xi)φ(yj)k 6 4ε kxikkyjk 6 16εkxkkyk. (c) First, recall that for any x, y ∈ A we have x ⊥ y if and only if the following four orthogonality conditions hold: x∗x ⊥ y∗y, x∗x ⊥ yy∗, xx∗ ⊥ y∗y and xx∗ ⊥ yy∗. Fix any elements x, y ∈ A with x ⊥ y and kxk,kyk 6 1. Due to the remark above, we have kφ(y∗y)φ(x∗x)k 6 ε and, in view of Kadison's inequality (see [9, §II.6.9.14]), φ(x)∗φ(x) 6 φ(x∗x) and φ(y)φ(y)∗ 6 φ(y∗y). Therefore, 8 TOMASZ KOCHANEK kφ(x)φ(y)k4 = kφ(y)∗φ(x)∗φ(x)φ(y)k2 6 kφ(y)∗φ(x∗x)φ(y)k2 = kφ(x∗x)1/2φ(y)φ(y)∗φ(x∗x)1/2k2 6 kφ(x∗x)1/2φ(y∗y)φ(x∗x)1/2k2 = kφ(x∗x)1/2φ(y∗y)φ(x∗x)φ(y∗y)φ(x∗x)1/2k, where z := φ(x∗x)1/2φ(y∗y)φ(x∗x)φ(y∗y)φ(x∗x)1/2 ∈ Bsa. Observe that for each n ∈ N, we have The norm of z, being equal to its spectral radius, is then estimated by kznk 6 kφ(x∗x)k · kφ(y∗y)φ(x∗x)k2n−1 6 ε2n−1. kzk 6 lim n→∞ ε(2n−1)/n = ε2. (cid:3) Remark. The above argument is very similar to the one used in the proof of [26, Prop. 3.1]. Of course, one can obtain the same estimates for φ(y)φ(x), φ(x)∗φ(y) and φ(x)φ(y)∗. Hence, ε-o.z. c.p.c. maps send orthogonal elements to 'ε1/2-orthogonal' ones. 3. An extension result and approximate Jordan equations For a nonunital C∗-algebra A we denote by A† its one-point unitization, i.e. A† = A ⊕ C as a vector space, where A forms a closed ideal of A† of codimension one. Recall that A† is equipped with the operator norm (regarding elements of A ⊕ C as left multiplication operators on A) defined by k(x, α)kop = sup(cid:8)kxy + αyk : y ∈ A, kyk 6 1(cid:9) (x ∈ A, α ∈ C) and which satisfies the C∗-condition. The ℓ1-norm on A ⊕ C, although not being a C∗- norm, happens to be equivalent to the operator norm. Indeed, as was shown by Gaur and Kov´ar´ık [14], we have and the constant 3 is sharp. kxk + α 6 3k(x, α)kop for all x ∈ Asa, α ∈ C Recall that, by the von Neumann Bicommutant Theorem, if M is a ∗-algebra acting nondegenerately on a Hilbert space H, then M′′ coincides with the closure of M with respect to the weak (equivalently, strong) operator topology (see, e.g., [23, §II.3]). Note also that on bounded sets the weak topology coincides with the σ-weak topology which is the same as the weak∗ topology on L (H) generated by its canonical predual, the space of trace-class operators (see [23, Lemma II.2.5]). Lemma 3.1. Let A, B be C∗-algebras, A be nonunital, and let π be a nondegenerate representation of B on a Hilbert space H. Then, for every ε-o.z. operator φ : A → B there exists an ε-o.z. operator φ† : A† → π(B)′′ which extends φ so that the following diagram commutes: φ π / / π(B) ❣ ❣ ❣ ❣ ❣ B φ† ❣ ❣ ❣ ❣ ❣ A A† / π(B)′′ ⊆ L (H) 3❣ ❣ ❣ / / _ / 3 APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 9 i.e. φ†(x) = π(φ(x)) for every x ∈ A. Moreover, if φ is δ-s.a., then we can pick φ† to be 6δ-s.a. If φ is completely positive, then φ† can be completely positive as well. Proof. Fix a bounded approximate unit (uλ)λ∈Λ of A. By passing to a subnet and using the Banach -- Alaoglu theorem (or the w.o.t.-compactness of the unit ball of L (H)) we may assume that there exists a limit z0 := w.o.t. lim λ π ◦ φ(uλ). Define φ† : A† → π(B)′′ by the formula φ†(x + α · 1A†) = π ◦ φ(x) + αz0 (x ∈ A, α ∈ C). w.o.t. Note that φ† does take values in π(B)′′ = π(B) Plainly, φ† is a bounded linear operator. We shall prove that it is ε-o.z. according to the Bicommutant Theorem. Fix two elements x + α · 1A†, y + β · 1A† ∈ A† + such that x + α · 1A† ⊥ y + β · 1A†. Since orthogonality passes to quotient algebras, we have α = 0 or β = 0. With no loss of generality assume that β = 0 and hence y ∈ A+. Notice that x ∈ Asa and α > 0, as positivity is preserved by quotient algebras as well. For any fixed λ ∈ Λ we have x + α · 1A† = (x + α · 1A†)1/2(1A† − uλ)(x + α · 1A†)1/2 + (x + α · 1A†)1/2uλ(x + α · 1A†)1/2. Observe that the latter summand belongs to A and is dominated by x + α · 1A†, therefore y ⊥ (x + α · 1A†)1/2uλ(x + α · 1A†)1/2. Since φ is ε-o.z., we have (cid:13)(cid:13)φ†(y)φ†(cid:0)(x + α · 1A†)1/2uλ(x + α · 1A†)1/2(cid:1)(cid:13)(cid:13) 6 ε(cid:13)(cid:13)y(cid:13)(cid:13)(cid:13)(cid:13)(x + α · 1A†)1/2uλ(x + α · 1A†)1/2(cid:13)(cid:13) 6 ε(cid:13)(cid:13)y(cid:13)(cid:13)(cid:13)(cid:13)x + α · 1A†(cid:13)(cid:13). z = φ†(y)φ†(x + α · 1A†); note that z = z1,λ + z2,λ, where (3.1) Set and z1,λ = φ†(cid:0)y(cid:1)φ†(cid:0)(x + α · 1A†)1/2uλ(x + α · 1A†)1/2(cid:1) z2,λ = φ†(cid:0)y(cid:1)φ†(cid:0)(x + α · 1A†)1/2(1A† − uλ)(x + α · 1A†)1/2(cid:1). Under this notation we have (3.2) kz1,λk 6 εkykkx + α · 1A†k and z2,λ w.o.t.−−−−→λ 0. 10 TOMASZ KOCHANEK In fact, the former statement is just a rewriting of (3.1). For the latter one observe that since (uλ)λ∈Λ is approximately central in A, we have w.o.t. lim λ = w.o.t. lim λ = w.o.t. lim λ φ†(cid:0)(x + α · 1A†)1/2(1A† − uλ)(x + α · 1A†)1/2(cid:1) φ†(cid:0)(1A† − uλ)(x + α · 1A†)(cid:1) φ†(cid:0)α(1A† − uλ)(cid:1) φ†(uλ)(cid:1) = 0. = α(cid:0)z0 − w.o.t. lim λ Now, the desired inequality kzk 6 εkykkx + α · 1A†k follows easily. Indeed, otherwise we could pick ξ, η ∈ H with kξk = kηk = 1 and such that hz(ξ), ηi = hz1,λ(ξ), ηi + hz2,λ(ξ), ηi > kykkx + α · 1A†k. Passing to limit over λ ∈ Λ we obtain a contradiction with (3.2). since the involution is weakly continuous, we have We shall now prove that φ† is 6δ-s.a. provided that φ is δ-s.a.. To this end, notice that z0 − z∗ 0 = w.o.t. lim λ π(φ(uλ) − φ(uλ)∗), thus kz0 − z∗ for any x ∈ Asa and α ∈ C we have 0k 6 δ, in view of the fact that kπk 6 1 and kuλk 6 1 for each λ ∈ Λ. Now, φ†((x + α · 1A†)∗) − (φ†(x + α · 1A†))∗ = π ◦ φ(x) − (π ◦ φ(x))∗ + αz0 − αz∗ 0, hence the norm of the left-hand side is at most kφ(x) − φ(x)∗k + αkz0 − z∗ 0k 6 δ(kxk + α) 6 3δkx + α · 1A†k, where the last estimate follows from the above mentioned Gaur -- Kov´ar´ık inequality. We have thus shown that kφ†(z) − φ†(z)∗k 6 3δkzk for every z ∈ (A†)sa. From this it immediately follows that φ† is 6δ-s.a. by splitting any element of A† into its real and imaginary parts. The assertion that φ† is completely positive whenever φ is can be proved by appealing to the Stinespring's theorem in the same way as in the proof of [25, Prop. 3.2]. Note that in this case the weak limit defining φ†(1A†) can be replaced by the strong limit after picking an increasing net (uλ)λ∈Λ, since then the net (φ(uλ))λ∈Λ is bounded and monotone increasing. (cid:3) Proposition 3.2. Let A and B be C∗-algebras and assume A is unital. If φ ∈ L (A,B) is an ε-o.z. operator with h := φ(1A), then kφ(x)2 − hφ(x2)k 6 108εkxk2 for every x ∈ A. Proof. First, we shall prove that (3.3) kφ(x)2 − hφ(x2)k 6 8εkxk2 for every x ∈ A+. APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 11 By homogeneity, it is enough to consider any x ∈ Asa such that 0 6 x 6 1A. In view of the Gelfand -- Naimark theorem, we have an isomorphism C∗(x, 1A) ∼= C(σ(x)) between the C∗-subalgebra of A generated by {x, 1A} and the algebra of complex-valued continuous functions on the spectrum σ(x) ⊆ [0, 1], where x corresponds to the identity function idσ(x). By this identification we can regard φ as an operator defined on C(σ(x)). Its second adjoint φ∗∗ is then defined on C(σ(x))∗∗ which contains the space of all bounded Borel functions on σ(x). For any n ∈ N, we consider a partition of σ(x) given by X0,n =h0, 1 ni ∩ σ(x), X1,n =(cid:16) 1 n , 2 ni ∩ σ(x), . . . , Xn−1,n =(cid:16) n − 1 n , 1i ∩ σ(x) and we pick arbitrary points xk,n ∈ Xk,n for 0 6 k < n (if some Xk,n = ∅, we ignore the symbol xk,n in all computations below). Consider any f ∈ C(σ(x)) regarded canonically as an element of C(σ(x))∗∗. Define fn = n−1Xk=0 f (xk,n)1Xk,n for n ∈ N and observe that fn w∗ −−→ f . Therefore, φ(f ) = φ∗∗(f ) = lim n→∞ φ∗∗(fn) = lim n→∞ f (xk,n)φ∗∗(1Xk,n) n−1Xk=0 (3.4) (3.5) and, consequently, φ(f )2 − hφ(f 2) = lim n→∞( X06j6=k<n + X06j<n f (xj,n)f (xk,n)φ∗∗(1Xj,n)φ∗∗(1Xk,n) f (xj,n)2hφ∗∗(1Xj,n)2 − φ∗∗(1σ(x))φ∗∗(1Xj,n)i). We are going to estimate the norms of both the above sums separately. To this end, we start with the following observation: Given any two sets A and B of the form A = a, b] ∩ σ(x), B = (c, d] ∩ σ(x), where 0 6 a < b 6 c < d 6 1, we have kφ∗∗(1A)φ∗∗(1B)k 6 ε. Indeed, let us define, for each n ∈ N, piecewise linear maps een,egn : [0, 1] → [0, 1] by een(t) = 2n+1 or b + 2−n 6 t 6 1 0 if 0 6 t 6 a + b−a 1 if a + b−a 2n 6 t 6 b continuous and linear elsewhere, andegn by a similar formula, where the endpoints a and b are replaced by c and d, respec- w∗−−→ 1A tively. Set also en =eenσ(x) and gn =egnσ(x). By Lebesgue's theorem, we have en w∗−−→ 1B. Observe also that engm = 0 for every m ∈ N and n sufficiently large. By and gn 12 TOMASZ KOCHANEK the assumption, for all such pairs (m, n) we have kφ(en)φ(gm)k 6 ε. Fixing m ∈ N and passing with n to infinity we obtain kφ∗∗(1A)φ(gm)k 6 ε, as multiplication is separately continuous with respect to the weak∗ topology on C(σ(x))∗∗ and φ∗∗ is weak∗-to-weak∗ continuous. Next, passing with m to infinity we obtain the announced inequality. Note that a similar reasoning, with suitably modified en's and gn's, applies in the case where A and B are disjoint finite unions of intervals intersected with σ(x). Moreover, it is easily seen that the argument goes through if we multiply each term of the form φ∗∗(1I∩σ(x)), where I is an interval, by any weight of modulus at most one. Hence, for every function f ∈ C(σ(x)) with kfk∞ 6 1 and for any disjoint sets M, N ⊂ {0, 1, . . . , n − 1} we have (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:16)Xj∈M f (xj,n)φ∗∗(1Xj,n)(cid:17)(cid:16)Xk∈N f (xk,n)φ∗∗(1Xk,n)(cid:17)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 6 ε. In what follows, we still assume that f ∈ C(σ(x)) and kfk∞ 6 1. Denote by Π the collection of all nontrivial ordered partitions of {0, 1, . . . , n−1}, that is, Π consists of all pairs (M, N) with M 6= ∅ 6= N, M∩N = ∅ and M ∪N = {0, 1, . . . , n−1}. Obviously, we have Π = 2n − 2 and hence f (xj,n)f (xk,n)φ∗∗(1Xj,n)φ∗∗(1Xk,n)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 6 (2n − 2)ε. Notice that for any fixed integers 0 6 j 6= k < n, the number of partitions from (M, N) ∈ Π that separate j and k and satisfy j ∈ M equals 2n−2. Therefore, by the triangle inequality, we obtain X{M,N }∈Π(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xj∈M, k∈N (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) X06j6=k<n (3.6) In order to estimate the norm of the second sum in formula (3.5) we consider only even integers n. Each summand can be written in the form 2n − 2 2n−2 ε −−−→n→∞ 4ε. 6 f (xj,n)f (xk,n)φ∗∗(1Xj,n)φ∗∗(1Xk,n)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) sj := f (xj,n)2hφ∗∗(1Xj,n)2 − φ∗∗(1σ(x))φ∗∗(1Xj,n)i = −f (xj,n)2φ∗∗(1Xj,n)φ∗∗(1σ(x)\Xj,n) = −f (xj,n)2φ∗∗(1Xj,n)Xk6=j φ∗∗(1Xk,n). Let Π′ be the collection of all ordered partitions (M, N) of {0, 1, . . . , n − 1} such that M = N = n/2. Note that Π′ =(cid:0) n n/2(cid:1) and that for every (M, N) ∈ Π′ we have (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:16)Xj∈M f (xj,n)2φ∗∗(1Xj,n)(cid:17)(cid:16)Xk∈N φ∗∗(1Xk,n)(cid:17)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 6 ε. APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 13 Summing up all these inequalities we obtain Notice that for any fixed j ∈ {0, 1, . . . , n − 1} we have (3.7) (3.8) n/2(cid:19)ε. 6(cid:18) n (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) X(M,N )∈Π′Xj∈M φ∗∗(1Xk,n)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) f (xj,n)2φ∗∗(1Xj,n)Xk∈N n/2(cid:19) (cid:12)(cid:12){(M, N) ∈ Π′ : j ∈ M}(cid:12)(cid:12) =(cid:18)n − 1 2(cid:0)n−1 n/2(cid:1) 1 (n − 1)(n − 2) · . . . · (n − n 2 ) 1 · 2 · . . . · n n 2 · = 2 and every partition as above gives rise to an expression f (xj,n)2φ∗∗(1Xj,n)Pk∈N φ∗∗(1Xk,n), where the last sum has n/2 summands. By symmetry, each sj is realized under the norm sign in (3.7) exactly n n−1 times. It is indeed an integer which can be written in the form Therefore, the sum under the norm sign in (3.7) equals 1 n − 1 · (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)X06j<n 2 − 1) n 1 · 2 · . . . · ( n (n − 2)(n − 3) · . . . · (n − n 2 ) n/2(cid:19). n/2 − 1(cid:19) = 4(n − 1)(cid:18) n =(cid:18) n − 2 4(n−1)(cid:0) n n/2(cid:1)P06j<n sj and hence f (xj,n)2hφ∗∗(1Xj,n)2 − φ∗∗(1σ(x))φ∗∗(1Xj,n)i(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 4(n − 1) ε −−−→n→∞ 4ε. 6 n n Combining (3.6) and (3.8) with formula (3.5) applied to the function f = idσ(x), we obtain the announced inequality (3.3). The result now follows by splitting an arbitrary x ∈ A into its real and imaginary parts, and applying the Jordan decomposition to each of them. Indeed, observe that if y ∈ Asa, y = y1 − y2, where y1, y2 ∈ A+ and y1y2 = 0, then φ(y2) = φ(y2 1 + y2 2) = φ(y2 1) + φ(y2 2) and Hence, making use of (3.3), we get φ(y)2 = φ(y1)2 + φ(y2)2 − φ(y1)φ(y2) − φ(y2)φ(y1). kφ(y)2 − hφ(y2)k 6 kφ(y1)φ(y2)k + kφ(y2)φ(y1)k + 8ε(ky1k2 + ky2k2) 6 2εky1kky2k + 8ε(ky1k2 + ky2k2) 6 18εkyk2. Now, if x = y + iz with y, z ∈ Asa, then we have φ(x2) = φ(y2) − φ(z2) + iφ(yz + zy) = φ(y2) − φ(z2) + 1 2 i(cid:0)φ((y + z)2) − φ((y − z)2)(cid:1) 14 and Therefore, TOMASZ KOCHANEK φ(x)2 = φ(y)2 − φ(z)2 + i(φ(y)φ(z) + φ(z)φ(y)) i(cid:0)φ(y + z)2 − φ(y − z)2(cid:1). = φ(y)2 − φ(z)2 + 1 2 kφ(x)2 − hφ(x2)k 6 18ε(kyk2 + kzk2) + 9ε(ky + zk2 + ky − zk2) 6 108εkxk2. (cid:3) 4. The second adjoint on bounded Borel functions For future use, we shall isolate a part of the proof of Proposition 3.2 (the one about disjointness preserving properties of the second adjoint operator) and give it a somewhat stronger form. Before doing it, note that if A0 ⊆ A is a commutative C∗-subalgebra of A, then A∗∗ contains a C∗-subalgebra isomorphic to the algebra B(σ(A0)) of bounded Borel functions on the spectrum of A0 (see [9, §III.5.13]). Thus, for any φ ∈ L (A,B) it makes sense to speak about the restriction of φ∗∗ to B(σ(A0)). Proposition 4.1. Let A and B be C∗-algebras and φ ∈ L (A,B) be an ε-o.z. map. Then, for every commutative separable C∗-subalgebra A0 of A, the operator φ∗∗ ↾B(σ(A0)) is also ε-o.z. Before proving this assertion let us collect essential tools from descriptive set theory. As usual, we denote by N = NN the Baire space of all countably infinite sequences of natural numbers. For any metric space X we write B(X) for the σ-algebra of Borel subsets of X, and we use the standard notation: Σ0 1(X) = {U ⊆ X : U is open}, Π0 ξ(X) = {X \ D : D ∈ Σ0 ξ(X)}, Σ0 ξ(X) =nSn∈N Dn : Dn ∈ Π0 ξn(X), ξn < ξ, n ∈ No ξ(X) = Sξ<ω1Π0 for any ordinal 1 < ξ < ω1. Plainly, B(X) = Sξ<ω1Σ0 ξ(X) (see, e.g., [19, §II.11]). We also use standard notation for function spaces: C0(X) for continuous and vanishing at infinity functions, Cb(X) for continuous bounded functions, and B(X) for Borel bounded functions (all complex-valued and defined on X). Recall that for ordinals 1 6 ξ < ω1 the Baire classes Bξ(X) are defined as follows: A function f : X → C is of Baire class 1 if f −1(U) is an Fσ-set for every open set U ⊆ C. As we consider complex-valued functions, it is equivalent to saying that f is the pointwise limit of a sequence of continuous functions (cf. [19, Thm. 24.10]). Next, for 1 < ξ < ω1, we say that f is of Baire class ξ provided that f is the pointwise limit of a sequence of functions fn : X → C, where each fn is of Baire class ξn with some ξn < ξ. In a similar fashion we define classes Bξ(X, Y ) consisting of Baire class ξ functions from X to Y , where Y is any separable metric space. (One difference is that, in general, Baire class 1 functions may not be pointwise limits of sequences of continuous functions.) According to the theorem of Lebesgue, Hausdorff and Banach ([19, Thm. 24.3]), the unionS16ξ<ω1Bξ(X, Y ) is the whole class of Borel functions mapping X into Y . APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 15 Theorem 4.2 (Lusin -- Souslin; see [19, Thm. 13.7]). Let X be a Polish space and A ⊆ X a Borel set. Then, there exists a closed set F ⊆ N and a continuous bijection Φ : F → A. We will also use the following classical result which plays a key role in the proof of the above quoted theorem of Lusin and Souslin. Theorem 4.3 ([19, Thm. 13.1]). Let (X,T ) be a Polish space and A ⊆ X a Borel set. Then, there exists a Polish topology TA ⊇ T such that B(TA) = B(T ) and A is clopen with respect to TA. Lemma 4.4. Let X be a metric space. Every nonnegative bounded function f ∈ B1(X) is the pointwise limit of a sequence (fn)∞ n=1 of simple functions fn : X → [0,∞) such that for every n ∈ N we have f −1 2(X) for any t > 0. n (0,∞) ⊆ f −1(0,∞), kfnk∞ 6 kfk∞ and f −1 n ({t}) ∈ Σ0 Proof. Obviously, for each n ∈ N we can pick a sequence 0 = tn,0 < tn,1 < tn,2 < . . . < tn,kn such that: • tn,kn−1 < kfk∞ < tn,kn for each n ∈ N; • tn,i − tn,i−1 < 1/n for all n ∈ N, 1 6 i 6 kn; • {tm,i : 1 6 m < n, 1 6 i 6 km} ∩ {tn,i : 1 6 i 6 kn} = ∅ for each n ∈ N. Set Dn,i = {x ∈ X : tn,i−1 < f (x) < tn,i} for all n ∈ N, 1 6 i 6 kn, and define a simple function fn : X → [0,∞) by fn = tn,i−11Dn,i. knXi=1 2(X). Also, fn vanishes whenever f Since f is of Baire class 1, all the sets Dn,i are in Σ0 does and obviously we have kfnk∞ 6 kfk∞ for every n ∈ N. Finally, note that fn −→ f pointwise. Indeed, fix any x ∈ X with f (x) > 0 and observe that f (x) = tm,i for at most one pair (m, i) with m ∈ N, 1 6 i 6 km. Then, for every n > m (or every n ∈ N if that m does not exist) we have fn(x) − f (x) < 1 n . (cid:3) Proof of Proposition 4.1. Note that since A0 ∼= C0(σ(A0)) is separable, the spectrum σ(A0) is metrizable (and, as always, locally compact). Moreover, separability of A0 implies that A0 is σ-unital, i.e. has a countable approximate unit, and hence σ(A0) is also σ-compact. This implies that σ(A0) is a Polish space (see, e.g., [19, Thm. 5.3]). Let be a complete metric compatible with its topology. Denote by S the collection of all nonnegative simple functions in B(σ(A0)). Each f ∈ S i=1 αi1Di with k > 0, αi > 0 and mutually disjoint nonempty sets Di ∈ B(σ(A0)) (1 6 i 6 k). We then denote D(f ) = {D1, . . . , Dk}. Define D to be the largest class contained in B(σ(A0)) with the following property: For all f1, f2 ∈ S satisfying f1f2 = 0 and D(f1) ∪ D(f2) ⊂ D, we have (4.1) can be written uniquely as f = Pk kφ∗∗(f1)φ∗∗(f2)k 6 εkf1k∞kf2k∞. 16 TOMASZ KOCHANEK We are to prove that D is the whole of B(σ(A0)). Claim 1. Σ0 2(σ(A0)) ⊆ D. We first prove that D contains all the closed sets. Fix any f1, f2 ∈ S with D(f1)∪ D(f2) ⊂ Π0 1(σ(A0)) and such that f1f2 = 0 which means nothing butS D(f1) ∩S D(f2) = ∅. For now, assume additionally that both f1 and f2 are compactly supported. For i = 1, 2 and every n ∈ N, define closed sets and the standard (continuous) "cut-off" functions 1 no Fn,i :=nλ ∈ σ(A0) : dist(cid:16)λ,[ D(fi)(cid:17) > distρ(λ, Fn,i) + distρ(cid:0)λ,S D(fi)(cid:1) . distρ(λ, Fn,i) fn,i(λ) := large enough. then (λ, λj) < 2rj for some 1 6 j 6 m which means that for each n > (minj rj)−1 the set Notice that fn,i vanishes outside the set Un,i := {λ : dist(λ,S D(fi)) < 1/n}. Since σ(A0) is locally compact andS D(fi) has compact closure, the set Un,i has compact closure for n Indeed, there are finitely many points λ1, . . . , λm ∈ S D(fi) and positive numbers r1, . . . , rm such that the open balls B(λj, rj) cover the whole of S D(fi) and B(λj, 2rj) have compact closures (1 6 j 6 m). Hence, if dist(λ,S D(fi)) < minj rj, Un,i is contained in the relatively compact set SjB(λj, rj). Since the closed setsS D(f1) andS D(f2) are at positive distance, we have Un,1∩Un,2 = ∅ for sufficiently large n. Hence, we may pick n0 ∈ N so that for each n > n0 we have fn,1fn,2 = 0 and fn,1, fn,2 have compact supports. For i = 1, 2 write fi =PD∈D(fi) αi(D)1D. By appealing to Urysohn's lemma, we pick continuous functions g1, g2 > 0 on σ(A0) such that gi↾D= αi(D) and kgik∞ = kfik∞ for i = 1, 2 and D ∈ D(fi). Then (gifn,i)n>n0 is a sequence of compactly supported nonnegative continuous functions converging pointwise, and hence weak∗, to fi (i = 1, 2). Since φ is ε-o.z., we have kφ(g1fn,1)φ(g2fn,2)k 6 εkg1fn,1k∞kg2fn,2k∞ = εkf1k∞kf2k∞ for each n > n0. Using the fact that multiplication is separately weak∗ continuous and φ∗∗ is weak∗-to-weak∗ continuous (as in the proof of Proposition 3.2) we obtain inequality (4.1). If we drop the assumption that f1 and f2 are compactly supported, using σ-compactness n=1 ⊂ S (i = 1, 2) of compactly supported functions such of σ(A0) we pick sequences (fn,i)∞ that: • fn,i −→ fi pointwise for i = 1, 2; • D(fn,i) ⊂ Π0 • fn,1fn,2 = 0 for every n ∈ N. Inequality (4.1) then follows from the previous part by passing to limit as above. This concludes the argument for Π0 1(σ(A0)) for all n ∈ N, i = 1, 2; The next step is to observe that D is closed under countable unions. Indeed, let f1, f2 ∈ S satisfy f1f2 = 0, D(f1) = {D1, . . . , Dk} and D(f2) = {E1, . . . , El}, where for all 1 6 i 6 k, 1(X) ⊆ D. APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 17 m=1Di,m and Ej = S∞ 1 6 j 6 l we have Di = S∞ to D. For m ∈ N, define 1Dr,s(cid:17)·f1 f1,m =(cid:16) kXr=1 mXs=1 and f2,m =(cid:16) lXr=1 mXs=1 1Er,s(cid:17)·f2, m=1Ej,m with all the Di,m, Ej,m belonging and notice that fi,m −→ fi pointwise for i = 1, 2, whereas by the definition of D, we have kφ∗∗(f1,m)φ∗∗(f2,m)k 6 εkf1,mk∞kf2,mk∞ 6 εkf1k∞kf2k∞ for each m ∈ N. Again, by passing to limit we obtain inequality (4.1). Claim 1 has thus been established. Now, suppose F = {A1, . . . , Am} is an arbitrary finite family of mutually disjoint sets in B(σ(A0)). According to Theorem 4.3, there is a sequence TA1 ⊆ TA1A2 ⊆ . . . ⊆ TA1A2...Am of Polish topologies on σ(A0), all finer than the original one, for which B(σ(A0)) = B(TA1A2...Am) and such that each of A1, . . . , Am is clopen in TA1A2...Am. Theorem 4.2 pro- duces a closed set F ⊆ N and a continuous bijection Φ : F → (σ(A0),TA1A2...Am) (which is, of course, continuous also with respect to ). Plainly, {Φ−1(Ai) : 1 6 i 6 m} is then a collection of mutually disjoint closed subsets of N . Claim 2. The operator ϕ ∈ L (Cb(F ),B∗∗) defined by ϕ(f ) = φ∗∗(f ◦ Φ−1) is ε-o.z. First of all, note that ϕ is well-defined because Φ−1 : σ(A0) → F is a Borel map (with respect to each of the topologies on σ(A0) considered above). Indeed, it follows from an- other Lusin -- Souslin theorem (see [19, Thm. 15.1]) that since Φ is continuous and injective, Φ(U) is Borel for every open set U ⊆ F . This means that Φ−1 is Borel. Fix any nonnegative functions f1, f2 ∈ Cb(F ) with f1f2 = 0. Take an increasing approx- n=1 for C0(σ(A0)) such that 0 6 hn 6 1 for each n ∈ N. Since fi ◦ Φ−1 imate unit (hn)∞ belongs to B(σ(A0)), a subalgebra of C0(σ(A0))∗∗, we have hn · (fi ◦ Φ−1) ր fi ◦ Φ−1 pointwise as n → ∞ (for i = 1, 2).1 Now, by transfinite induction, we argue that for every Borel function Ψ : σ(A0) → F and each n ∈ N we have (4.2) kφ∗∗(hn·(f1 ◦ Ψ))φ∗∗(hn·(f2 ◦ Ψ))k 6 εkf1k∞kf2k∞. Indeed, by the Lebesgue -- Hausdorff -- Banach theorem mentioned above, we infer that there is an ordinal 1 6 ξ < ω1 for which Ψ ∈ Bξ(σ(A0), F ). If ξ = 1, we appeal to Lemma 4.4 to produce sequences (gm,i)∞ m=1 of nonnegative simple functions assuming each positive value on a Σ0 m,i(0,∞) ⊆ (fi ◦ Ψ)−1(0,∞) and kgm,ik∞ 6 kfik∞ for m ∈ N, i = 1, 2. In view of Claim 1, for every m ∈ N we have 2(σ(A0))-set and such that: gm,i −→ hn · (fi ◦ Ψ) pointwise, g−1 kφ∗∗(gm,1)φ∗∗(gm,2)k 6 εkgm,1k∞kgm,2k 6 εkf1k∞kf2k∞, whence inequality (4.2) follows by passing to limit as m → ∞. 1If (uλ) is an increasing approximate unit of a C∗-algebra A, then it converges σ-strongly to the identity in the enveloping von Neumann algebra A′′ of A (see [9, III.5.2.11]). 18 TOMASZ KOCHANEK Next, assume that 1 < ξ < ω1 and inequality (4.2) holds true whenever Ψ is of Baire class η with η < ξ. Any Ψ ∈ Bξ(σ(A0), F ) is the pointwise limit of a sequence (Ψm)∞ for some Ψm ∈ Bξm(σ(A0), F ) with ξm < ξ (m ∈ N). By induction hypothesis, estimate (4.2) is valid with Ψm in the place of Ψ, for any m, n ∈ N. Once again, passing to limit as m → ∞ we obtain the announced assertion. Having proved inequality (4.2), we pass to limit once more, this time as n → ∞, and conclude that ϕ is ε-o.z.. This completes the proof of Claim 2. Claim 3. Assume g1, g2 ∈ Cb(F ) are nonnegative simple functions such that g1g2 = 0 and g−1 i ({t}) ∈ Π0 We argue as in the first part of the proof of Claim 1, where it was shown that D contains all the closed sets. Replacing σ(A0) by F the argument goes through mutatis mutandis, except the fact that the present case is even easier, as we do not need to care about compact supports. 1(F ) for all t > 0, i = 1, 2. Then kϕ∗∗(g1)ϕ∗∗(g2)k 6 εkg1k∞kg2k∞. m=1 Finally, we are prepared to complete the proof. Fix any f1, f2 ∈ S with f1f2 = 0. Recall that the continuous bijection Φ : F → σ(A0) was chosen according to an arbitrarily fixed disjoint finite family F ⊂ B(σ(A0)). Now, we take F := D(f1) ∪ D(f2) and choose Φ correspondingly. Define gi = fi ◦ Φ for i = 1, 2. These are Borel step functions on F . By definition, ϕ = φ∗∗ ◦ Γ, where Γ : Cb(F ) → B(σ(A0)) is the composition operator Γf = f ◦ Φ−1. Recall that, due to the Lusin -- Souslin theorem, {Φ−1(A) : A ∈ F} is a collection of mutually disjoint closed subsets of F . This means that g−1 1(F ) for i = 1, 2 and every t > 0. As we have seen in the proof of Claim 1, such functions can be approximated pointwise by (uniformly bounded) sequences of continuous functions. Therefore, by the weak∗-to-weak∗ continuity of Γ∗∗, we have Γ∗∗gi = gi◦Φ−1 = fi (i = 1, 2). Hence, i ({t}) ∈ Π0 In view of Claim 3, we thus have ϕ∗∗(gi) = φ∗∗∗∗ ◦ Γ∗∗(gi) = φ∗∗∗∗(fi) = φ∗∗(fi) for i = 1, 2. kφ∗∗(f1)φ∗∗(f2)k = kϕ∗∗(g1)ϕ∗∗(g2)k 6 εkg1k∞kg2k∞ = εkf1k∞kf2k∞. Consequently, we have shown that D = B(σ(A0)). It remains to notice that every nonnegative bounded Borel function on σ(A0) is the point- wise limit of an increasing sequence of Borel simple funtions. Therefore, in a similar fashion as several times above, we conclude that φ∗∗ is ε-o.z. on B(σ(A0)). (cid:3) 5. Almost commutation relations It follows from Theorems 1.2 and 1.3 that the range of both a disjointness preserving and an order zero operator ϕ on a unital C∗-algebra A is contained in the commutant of ϕ(1A). As we will see, this has some approximate counterparts, despite two main disadvantages. The first one is the well-known fact that, in general, two almost commuting operators on a Hilbert space may not be 'sufficiently' close to commuting operators. In fact, this can happen for matrices -- Choi [11] constructed, for each n ∈ N, matrices A, B ∈ Mn(C) satisfying kAk = 1− 1 n for every pair A′, B′ ∈ Mn(C) with [A′, B′] = 0. On a positive side, the famous Lin's theorem [20] n , yet kA− A′k +kB − B′k > 1− 1 n , kBk 6 1, k[A, B]k 6 2 APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 19 says that for any δ > 0 there is ε > 0 such that for any self-adjoint matrices A, B ∈ Mn(C) with kAk,kBk 6 1 and k[A, B]k < δ there exists a pair of commuting self-adjoint matrices A′, B′ ∈ Mn(C) such that kA − A′k + kB − B′k < ε. However, we cannot guarantee that one can take e.g. A′ = A, so an almost commutation relation between A and B does not automatically imply any similar relation between P (A) and B for P being a polynomial. Another disadvantage is, therefore, that given an ε-o.z. map φ on a unital C∗-algebra A, an upper bound on the norm of the commutator [P (φ(1A)), φ(x)] grow quite rapidly when deg P → ∞ (see Proposition 5.1 below). Nonetheless, there are some situations where we can obtain a 'uniform' estimate up to a supremum norm of P (see Propositions 5.2 and 5.4) and get almost commutation relations with all spectral projections of φ(1A) (as in Lemma 5.5) which will be important later on. We denote by τ the operator on the space C[z] of complex polynomials acting as the left-shift on coefficients, i.e. τ P (z) = z−1(P (z)−P (0)). Given any ε-o.z. map φ on a unital C∗-algebra A, we define for every P ∈ C[z] with degP = N > 0 a nonnegative number ΘN (P ) recursively as follows:  (5.1) where Θ1(az + b) = 8aε ΘN (P ) = 8kφk+ΘN −1(τ P ) + 8kτ P (φ(1A))kε for N > 2, kφk+ := sup{kφ(x)k : x ∈ A+, kxk 6 1}. Obviously, the Jordan decomposition yields kφk 6 4kφk+, however, we prefer to be as precise as possible in estimate (5.2) below, in order to eventually obtain better constants in Lemma 6.6 and, ultimately, in Corollary 8.2. Proposition 5.1. Let A and B be C∗-algebras and assume A is unital. Let φ ∈ L (A,B) be an ε-o.z. map with h := φ(1A) and P ∈ C[z] be a complex polynomial with degP = N > 0. Then we have (5.2) kP (h)φ(x) − φ(x)P (h)k 6 ΘN (P )kxk for every x ∈ A+. Proof. To start the induction we show the inequality (5.3) khφ(x) − φ(x)hk 6 8εkxk for every x ∈ A+. By homogeneity, we can assume that 0 6 x 6 1A. We use the notation from the proof of Proposition 3.2. Observe that in the algebra C∗(x, 1A) ∼= C(σ(x)), for each n ∈ N, we have 1σ(x) = n−1Xk=0 1Xk,n and h = φ∗∗(1Xk,n). n−1Xk=0 20 TOMASZ KOCHANEK Therefore, for any f ∈ C(σ(x)) with kfk∞ 6 1 we have h · n−1Xk=0 f (xk,n)φ∗∗(1Xk,n) = n−1Xk=0 f (xk,n)(φ∗∗(1Xk,n))2 + X06j6=k<n f (xk,n)φ∗∗(1Xj,n)φ∗∗(1Xk,n). As it was proved above (cf. inequality (3.6)), the norm of the last summand is at most 4ε. The reversed product can be represented similarly, so subtracting these equations we get (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) h ·(cid:16)n−1Xk=0 In view of formula (3.4), we obtain inequality (5.3), that is, our assertion for degP = 1. f (xk,n)φ∗∗(1Xk,n)(cid:17) −(cid:16)n−1Xk=0 f (xk,n)φ∗∗(1Xk,n)(cid:17) · h(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) N. Fix any x ∈ Asa such that 0 6 x 6 1A and write P (z) =PN Now, fix any polynomial P ∈ C[z] with degP = N > 2 and assume that the desired inequality holds true for all ε-o.z. maps and all complex polynomials of degree smaller than j=0 ajzj. Again, we use the notation from the proof of Proposition 3.2. 6 8ε. For any f ∈ C(σ(x)), define f (xl,n)φ∗∗(1Xl,n)! βn(f ) := P (h) n−1Xl=0 f (xl,n) n−1Xk=0 n−1Xk=0 aj X06k0,k1,...,kj<n NXj=0 NXj=0 aj = = φ∗∗(1Xl,n) φ∗∗(1Xk,n)!j f (xkj,n)φ∗∗(1Xk0,n)φ∗∗(1Xk1,n) · . . . · φ∗∗(1Xkj ,n). Similarly, we calculate γn(f ) := n−1Xk=0 NXj=0 f (xk,n)φ∗∗(1Xk,n)!P (h) aj X06k0,k1,...,kj<n = f (xk0,n)φ∗∗(1Xk0,n)φ∗∗(1Xk1,n) · . . . · φ∗∗(1Xkj ,n). Both these expressions βn(f ) and γn(f ) have the same part corresponding to (j + 1)- tuples (k0, k1, . . . , kj) for which k0 = kj. After reducing this common part we obtain that APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 21 βn(f ) − γn(f ) = β′ n(f ) − γ′ n(f ), where β′ n(f ) is defined as NXj=1 aj X06k06=kj<n f (xkj,n)φ∗∗(1Xk0,n) n−1Xk1,...,kj−1=0 f (xl,n)φ∗∗(1Xk,n) NXj=1 = X06k6=l<n = X06k6=l<n f (xl,n)φ∗∗(1Xk,n) τ P (h)φ∗∗(1Xl,n) φ∗∗(1Xk1,n) · . . . · φ∗∗(1Xkj−1,n)!φ∗∗(1Xkj ,n) ajhj−1!φ∗∗(1Xl,n) n(f ) is defined by an analogous formula with f (xl,n) replaced by f (xk,n). (Notice and γ′ that for j = 1 the term in brackets is meant to be the identity in B∗∗.) We claim that for all n ∈ N and f ∈ C(σ(x)) with kfk∞ 6 1 we have (5.4) kβ′ n(f )k, kγ′ n(f )k 6 (4 − 23−n)(kφk+ΘN −1(τ P ) + kτ P (h)k)ε. We consider the inequality for β′ n(f ); the other one just requires changing one index. By virtue of Proposition 4.1, the operator φ∗∗ ↾B(σ(x)) is ε-o.z. Therefore, we can apply our induction hypothesis to φ∗∗ and the polynomial τ P of degree N−1. As before, Π stands for the collection of nontrivial ordered partitions of {0, 1, . . . , n − 1}. Given any (K, L) ∈ Π, we thus obtain (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 1Xk,n(cid:17) τ P (h)φ∗∗(cid:16)Xl∈L φ∗∗(cid:16)Xk∈K 6(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 1Xk,n(cid:17)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ·(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) φ∗∗(cid:16)Xk∈K f (xl,n)1Xl,n(cid:17)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) τ P (h)φ∗∗(cid:16)Xl∈L f (xl,n)1Xl,n(cid:17) − φ∗∗(cid:16)Xl∈L +(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 1Xl,n(cid:17)φ∗∗(cid:16)Xl∈L φ∗∗(cid:16)Xk∈K f (xl,n)1Xl,n(cid:17)τ P (h)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) f (xl,n)1Xl,n(cid:17)τ P (h)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 6 kφk+ΘN −1(τ P ) + kτ P (h)kε. (Notice that the estimate by kφk+ follows from the fact that φ∗∗ is weak∗-to-weak∗ contin- 1Xk,n is a pointwise limit of continuous positive functions.) Summing up over all partitions we get uous andPk∈K X(K,L)∈Π(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:16)Xk∈K f (xk,n)φ∗∗(1Xk,n)(cid:17) τ P (h)(cid:16)Xl∈L φ∗∗(1Xl,n)(cid:17)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 6 (2n − 2)(kφk+ΘN −1(τ P ) + kτ P (h)kε). For any fixed pair (k, l) with 0 6 k 6= l < n, the number of those partitions (K, L) ∈ Π for which k ∈ K and l ∈ L equals 2n−2. Consequently, (5.4) follows from the triangle inequality. 22 TOMASZ KOCHANEK By Lebesgue's theorem, we have βn(idσ(x)) w∗−−→ P (h)φ(x) and γn(idσ(x)) w∗−−→ φ(x)P (h). Hence, inequality (5.4) yields kP (h)φ(x) − φ(x)P (h)k 6 lim inf = lim inf 6 8(kφk+ΘN −1(τ P ) + kτ P (h)kε) = ΘN (P ). n→∞ (cid:13)(cid:13)βn(idσ(x)) − γn(idσ(x))(cid:13)(cid:13) n(idσ(x))(cid:13)(cid:13) n→∞ (cid:13)(cid:13)β′ n(idσ(x)) − γ′ This completes the induction. (cid:3) Obviously, the estimate given in Proposition 5.1 is meaningless if ΘN (P ) is too large, e.g. greater than 2kφk+kP (h)k. However, in some cases inequality (5.2) can be transformed so that we obtain 'almost commutation' relations between φ(x) and spectral projections of h. In what follows, we shall see two such cases. The first one happens when h is an algebraic element, i.e. P (h) = 0 for some monic polynomial P ∈ C[z]. The second one is more subtle and relies on a deep result by Alaminos, Extremera and Villena [3] mentioned before. Proposition 5.2. For all N ∈ N and M > 0 there exists C = C(N, M) < ∞ with the following property: If A and B are C∗-algebras, A is unital, φ ∈ L (A,B) is a nonzero ε-o.z. map and h := φ(1A) is a self-adjoint algebraic element of order at most N such that kφk+ 6 Mkhk, then for every P ∈ C[z] with P (0) = 0 and x ∈ A+ we have kP (h)φ(x) − φ(x)P (h)k 6 Cε khk z=khkP (z) · kxk. sup Proof. First, observe that if h ∈ B is an algebraic element of order at most N, then the commutator [P (h), φ(x)] is the same as [Q(h), φ(x)] with some Q ∈ C[z] satisfying deg Q < N. Therefore, we can assume that P satisfies P (0) = 0 and deg P < N. For each 1 6 j < N define a function eΘj on the set of complex polynomials of degree j by eΘj(P )ε = Θj(P ), where Θj are defined by formulas (5.1). Notice that eΘ1(z) = 8 = 8 khk z∈σ(h)z, sup hence our inequality holds true for deg P = 1 in view of Proposition 5.1. Fix any 1 < j < N and any P ∈ C[z] with deg P = j. Let also φ and h be as above. Assume that for every polynomial Q ∈ C[z] with deg Q = j − 1 we have Plainly, formulas (5.1) yield eΘj−1(Q) 6 Cj−1 sup z=khkQ(z). eΘj(P ) = 8kφk+eΘj−1(τ P ) + 8 sup z∈σ(h)τ P (z) APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 23 and hence our inductive hypothesis implies that eΘj(P ) 6 8(Cj−1kφk+ + 1) sup z=khkτ P (z) 1 Cj−1 + Consequently, if we define a sequence (C1, . . . , CN −1) recursively by = 8(cid:16)kφk+ khk C1 = 8 khk z=khkP (z) 6 8(cid:16)MCj−1 + khk(cid:17) sup and Cj = 8(cid:16)MCj−1 + 1 khk(cid:17) sup khk(cid:17) for 1 < j < N, 1 z=khkP (z). then in view of Proposition 5.1 we have kP (h)φ(x) − φ(x)P (h)k 6 Cj ε sup z=khkP (z) · kxk for every x ∈ A+. It is easily seen that Cj = khk−1Pj−1 C =PN −2 i=0 8i+1M i we obtain the desired estimate. i=0 8i+1M i for each 1 6 j < N. Hence, putting (cid:3) Remark. We will later show that the inequality kφk+ 6 Mkhk, with some absolute constant M < ∞, can be safely assumed for φ being self-adjoint, as otherwise the norm of φ is small (see Corollaries 6.5 and 6.6). This observation will be used several times in the proof of Theorem 7.2. Lemma 5.3. Let A and B be unital C∗-algebras and φ ∈ L (A,B) be an ε-o.z. map with h := φ(1A) and satisfying kφk > ε1/2. Assume also that the C∗-subalgebra of B generated by h lies in the range of the center of A under φ, that is, C∗(h) ⊆ φ(Z(A)). Then, there exists a constant M > 0 such that for all x, y ∈ A+ with xy = 0 and every polynomial P ∈ C[z] we have (5.5) Moreover, in the case where A is commutative and φ is surjective, one can take M to be the openness index of φ. kφ(x)P (h)φ(y)k 6 Mkφk(cid:0)kφk15/8O(ε1/16) + 24ε(cid:1)kP (h)kkxkkyk. Proof. Similarly as in the proof of the Open Mapping Theorem, we note that there exists r > 0 such that (5.6) where BX stands for the closed unit ball of X. (Recall that Z(A) is a closed subspace of A.) Indeed, since rBB ∩ C∗(h) ⊂ φ(BZ(A)), φ(cid:16)1 2 BZ(A)(cid:17) − φ(cid:16) 1 2 BZ(A)(cid:17) ⊆ φ(cid:16) 1 2 BZ(A)(cid:17) − φ(cid:16)1 2 BZ(A)(cid:17) ⊆ φ(BZ(A)), 24 TOMASZ KOCHANEK it is enough to show that φ( 1 2BZ(A)) has nonempty interior relatively to C∗(h). As we have C∗(h) ⊆ φ(Z(A)) = ∞[k=1 kφ(cid:16)1 2 BZ(A)(cid:17), it follows from the Baire Category Theorem that for some k ∈ N the set kφ( 1 not nowhere dense relatively to C∗(h). The same is then true for φ( 1 follows. 2BZ(A)) is 2BZ(A)) and our claim Now, by (5.6), we infer that there exists M > 0 such that for every P ∈ C[z] and any δ > 0 there exists v ∈ Z(A) satisfying kφ(v) − P (h)k 6 δ and kvk 6 MkP (h)k. At this point, note that if Z(A) = A and φ is surjective, the Open Mapping Theorem guarantees that we can take δ = 0 and M = op(φ). Fix x and y as above and consider the commutative C∗-subalgebra C∗(x, y, 1A) of A generated by {x, y, 1A}. Regarding any its elements u, v as continuous functions, we easily see that the condition uv = 0 is equivalent to u⊥ v. Therefore, by Lemma 2.4(b), we have u, v ∈ C∗(x, y, 1A), uv = 0 ==⇒ kφ(u)φ(v)k 6 16εkxkkyk. Consider the bilinear map Φ(u, v) = φ(u)φ(v) and put K := 16kφk2 > max{kΦk, 16ε} (recall that kφk > ε1/2). In view of Theorem 2.1, we have (5.7) kφ(uv)φ(w) − φ(u)φ(vw)k 6 ηkukkvkkwk for all u, v, w ∈ C∗(x, y, 1A), where (5.8) By appealing to Lemma 2.2, it can be easily verified that η, defined by formula (5.8) as a function of ε, satisfies the estimate 3 + 1(cid:17)2 K 1/2ε1/2(cid:16)2 + ζ(cid:16)16ε η = 4(cid:16)172 K (cid:17)(cid:17) + Kζ(cid:16)16ε K (cid:17). η(ε) 6 M(cid:0)kφkε1/2 + kφk7/8ε9/16 + kφk15/8ε1/16(cid:1), with some absolute constant M < ∞. Since kφk > ε1/2, we have kφkε1/2 < kφk15/8ε1/16 and kφk7/8ε9/16 < kφk15/8ε1/16, whence (5.9) η(ε) = kφk15/8O(ε1/16). Fix P ∈ C[z] and consider any v ∈ Z(A) with kvk 6 MkP (h)k. For simplicity, assume that kxk,kyk 6 1. Notice that xvy = 0 = yxv, as well as (xv)∗y = v∗x∗y = 0 = xvy∗, i.e. xv ⊥ y. Hence, Lemma 2.4(b) yields kφ(xv)φ(y)k 6 16εkvk. In view of (5.7), (5.9) and Proposition 5.1, we have kφ(x)φ(v)φ(y)k 6 kφ(x)φ(v) − φ(xv)hkkφk + kφ(xv)hφ(y)k 6 η(ε)kφkkvk + kφ(xv)φ(y)hk + kφkkvkkhφ(y) − φ(y)hk 6 kφkkvk(cid:0)η(ε) + 16ε + 8ε(cid:1) 6 Mkφk(cid:0)kφk15/8O(ε1/16) + 24ε(cid:1)kP (h)k. Since v ∈ Z(A) can be chosen so that φ(v) is arbitrarily close to P (h), we obtain inequal- ity (5.5). (cid:3) APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 25 Proposition 5.4. Let A and B be C∗-algebras and assume A is unital. Let φ ∈ L (A,B) be an ε-o.z. map with 0 6= h := φ(1A) and satisfying C∗(h) ⊆ φ(Z(A)). Then, there exists a constant M > 0 such that for every polynomial P ∈ C[z] with P (0) = 0 and every x ∈ A+ we have khk(cid:0)kφk15/8O(ε1/16) + 24ε(cid:1)) sup kP (h)φ(x) − φ(x)P (h)k 6 max(2ε1/2, 8M kφk z=khkP (z)·kxk. Moreover, in the case where A is commutative (i.e. Z(A) = A) and φ is surjective, one can take M to be the openness index of φ. Proof. First, observe that if kφk 6 ε1/2, then the above inequality is trivial. So, suppose that kφk > ε1/2 and take any x ∈ A+ with 0 6 x 6 1A. As we have seen in the proof of Proposition 5.1, the commutator [P (h), φ(x)] is the weak∗ limit of βn(idσ(x))− γn(idσ(x)) = β′ n(idσ(x)) − γ′ n(idσ(x)), where β′ n(idσ(x)) = X06k6=l<n xl,nφ∗∗(1Xk,n)τ P (h)φ∗∗(1Xl,n) n(idσ(x)) is defined analogously with xk,n instead of xl,n. Applying Lemma 5.3 to the and γ′ ε-o.z. map φ∗∗ we infer that for every partition (K, L) ∈ Π we have The last sentence of our assertion follows from the fact that M is the same constant as the one stemming from Lemma 5.3. (cid:3) The following simple lemma which guarantees 'almost commutation' relations with spec- tral projections will be used in our decomposition result in Section 7. Lemma 5.5. Let S, T ∈ L (H) be operators on a complex Hilbert space H with S being self-adjoint. Suppose δ, R > 0 are such that for every complex polynomial P ∈ C[z] with P (0) = 0 we have kP (S)T − T P (S)k 6 δ sup z=RP (z). Then for every spectral projection V of S we have kV T − T V k 6 δ. Summing over all partitions, using the triangle inequality and Lebesgue's theorem as in the proof of Proposition 5.1, we obtain xl,n1Xl,n(cid:17)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 1Xk,n(cid:17) τ P (h)φ∗∗(cid:16)Xl∈L (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) φ∗∗(cid:16)Xk∈K 6 Mkφk(cid:0)kφk15/8O(ε1/16) + 24ε(cid:1)kτ P (h)k. kP (h)φ(x) − φ(x)P (h)k 6 8Mkφk(cid:0)kφk15/8O(ε1/16) + 24ε(cid:1)kτ P (h)k. kτ P (h)k = sup z∈σ(h) τ P (z) 6 sup z=khkτ P (z) = z=khkP (z). sup 1 khk It remains to observe that 26 TOMASZ KOCHANEK Proof. Let E be the spectral measure of S. For any ξ, η ∈ H we denote by Eξ,η the complex Borel measure on σ(S) given by Eξ,η(ω) = hE(ω)ξ, ηi. Fix arbitrarily ξ, η ∈ H with kξk = kηk = 1, denote µ = T ∗η and notice that for every P ∈ C[z] (with P (0) = 0) we have and Therefore, by our assumption, hP (S)T ξ, ηi =Zσ(S) P (z) dET ξ,η(z) P (z) dEξ,µ(z). hT P (S)ξ, ηi = hP (S)ξ, µi =Zσ(S) (cid:12)(cid:12)(cid:12)Zσ(S) P (z) dET ξ,η(z) −Zσ(S) P (z) dEξ,µ(z)(cid:12)(cid:12)(cid:12) 6 δ sup z=RP (z). Define ν to be the complex Borel measure that extends the measure ET ξ,η − Eξ,µ to the whole disc {z 6 R}, that is, ν(ω) = (ET ξ,η − Eξ,µ)(ω ∩ σ(S)) for every Borel subset ω of {z 6 R}. Then we have (cid:12)(cid:12)(cid:12)Z{z6R} P (z) dν(z)(cid:12)(cid:12)(cid:12) 6 δ sup z=RP (z) and, of course, the same inequality holds true if we replace P (z) by P (z) as the measure ν is supported only on σ(S) ⊂ R. The Stone -- Weierstrass theorem and the Maximum Modulus Principle thus imply that ν defines a linear functional of norm at most δ on the Banach space of complex continuous functions on {z 6 R} vanishing at 0 (equipped with the supremum norm). Hence, ν(σ(S)) 6 δ and in particular for every Borel set ω ⊆ σ(S) we have ν(ω) 6 δ which means that hE(ω)T ξ, ηi − hT E(ω)ξ, ηi 6 δ. Since ξ and η were arbitrary unit vectors, we obtain kE(ω)T − T E(ω)k 6 δ, as desired. (cid:3) 6. The range of an approximately order zero map In this section, we seek for approximate counterparts of the fact that the range of any dis- jointness preserving or order zero operator ϕ on a unital C∗-algebra A has a range contained in the closure of ϕ(1A){ϕ(1A)}′, as stated in Theorems 1.2 and 1.3. It turns out that the range of any self-adjoint ε-o.z. map ψ : A → B lies close to the hereditary C∗-subalgebra ψ(1A)Bψ(1A) of B (it is indeed hereditary as ψ(1A) = ψ(1A)∗; see [9, Prop. II.3.4.2]). To show this, we need a deep result by Aleksandrov and Peller [4] which says that any α-Holder function on R, with 0 < α < 1, is also operator Holder: Theorem 6.1 ([4, Thm. 4.1]). Let 0 < α < 1. There exists a constant c > 0 depending only on α such that for every α-Holder function f : R → R and any self-adjoint operators A, B on a Hilbert space, we have kf (A) − f (B)k 6 ckfkΛα(R)kA − Bkα, APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 27 where kfkΛα(R) = sup x6=y f (x) − f (y) x − yα . Our next lemma actually gives a more precise information about the range of ψ than just that it lies close to ψ(1A)Bψ(1A). This requires a use of some well-known results on (weak) polar decompositions in C∗-algebras. There are the "left-handed","right-handed" and "two-sided" versions which we quote below. Proposition 6.2 (see [9, II.3.2.1 and II.3.2.2]). Let A be a C∗-algebra, x ∈ A and a ∈ A+. 2 there exists u ∈ Aa satisfying x = uaα and 2 there exists v ∈ aA satisfying x = aαv and (ii) If xx∗ 6 a, then for every 0 < α < 1 (i) If x∗x 6 a, then for every 0 < α < 1 kuk 6 ka1/2−αk. kvk 6 ka1/2−αk. Proposition 6.3 (see [21, Lemma 2.2.4]). Let A be a C∗-algebra, x ∈ A and a ∈ A+. If x∗x 6 a and xx∗ 6 a, then for every 0 < α < 1 4 there exists d ∈ A satisfying x = aαdaα and kdk 6 kxk1/2−2α. Lemma 6.4. There exists an absolute constant K < ∞ such that the following holds. Let A,B be C∗-algebras, with A unital, let ψ : A → B be a self-adjoint ε-o.z. map and define h := ψ(1A). Then we have (6.1) Moreover: dist(cid:0)ψ(x), hBh(cid:1) 6 Kkψk3/5ε1/5kxk (a) for all 0 < α < 1 for every x ∈ A. (b) for all 0 < α < 1 (c) for all 0 < α < 1 and and and 10 , x ∈ A there exists u(x) ∈ B such that ku(x)k 6 2kψk1−2αkxk kψ(x) − u(x)(h2)αk 6 Kkψk3/5ε1/5kxk; 10, x ∈ A there exists v(x) ∈ B such that kv(x)k 6 2kψk1−2αkxk kψ(x) − (h2)αv(x)k 6 Kkψk3/5ε1/5kxk; 20, x ∈ A there exists d(x) ∈ B satisfying kd(x)k 6 2kψk1/2+6αkxk1/2+10α kψ(x) − (h2)αd(x)(h2)αk 6 Kkψk3/5ε1/5kxk. Proof. According to Proposition 3.2, for any x ∈ A we have kψ(x)3 − hψ(x2)ψ(x)k 6 kψ(x)2 − hψ(x2)k·kψ(x)k 6 108kψkεkxk3. Consider an arbitrary x ∈ Asa and define ω(x) := hψ(x2)ψ(x)ψ(x2)h ∈ hBhsa. Obviously, we have kψ(x)3ψ(x2)h − ω(x)k 6 kψ(x)3 − hψ(x2)ψ(x)k·kψ(x2)hk 6 108kψk3εkxk5 28 and TOMASZ KOCHANEK kψ(x)3ψ(x2)h − ψ(x)5k 6 kψ(x)3k·k(ψ(x)2 − hψ(x))∗k 6 108kψk3εkxk5. Therefore, Now, an elementary verification shows that the map R ∋ t 7−→ t1/5 is Holder of order 1 (with constant 1). Hence, appealing to the Aleksandrov -- Peller Theorem 6.1 we get 5 kψ(x)5 − ω(x)k 6 216kψk3εkxk5. kψ(x) − ω(x)1/5k 6 Ckψk3/5ε1/5kxk for every x ∈ Asa, (6.2) with an absolute constant C < ∞. Of course, ω(x)1/5 ∈ hBh, so the desired estimate (6.1) has been proven for any self-adjoint x ∈ A. For general x ∈ A we simply consider its real and imaginary parts and we obtain (6.1) with K = 2C. In order to prove the assertions (a) -- (c), note that for every x ∈ Asa we have ψ(x2)ψ(x)ψ(x2)h2ψ(x2)ψ(x)ψ(x2) 6 kψk8kxk10·1B†, whence (ω(x)1/5)∗ω(x)1/5 = ω(x)2/5 6 kψk8/5kxk2(h2)1/5. ω(x)∗ω(x) = hψ(x2)ψ(x)ψ(x2)h2ψ(x2)ψ(x)ψ(x2)h 6 kψk8kxk10h2. 2 and β = 2 By Lowner's theorem (see [9, Prop. II.3.1.10]), the map [0,∞) ∋ t 7−→ tβ is operator monotone for each β ∈ [0, 1]. Using this fact successively for β = 1 5 we obtain (6.3) Now, Proposition 6.2(i) applied to ω(x)1/5 and any α ∈ (0, 1 B such that ω(x)1/5 = u(x)(h2)α and (6.4) This, jointly with inequality (6.2), proves the assertion (a) for x ∈ Asa with constant C instead of K. For an arbitrary x ∈ A we take the usual decomposition x = x1 + ix2 with x1, x2 ∈ Asa, kx1k,kx2k 6 kxk, and we define u(x) = u(x1) + iu(x2). Then, obviously, ku(x)k 6 2kψk1−2αkxk and the assertion (a) follows with K = 2C. ku(x)k 6 kψk8αkxk10αk(kψk8/5kxk2(h2)1/5)1/2−5αk 6 kψk1−2αkxk. 10) produces an element u(x) ∈ The clause (b) is proved along the same lines by using Proposition 6.2(ii). Finally, we apply Proposition 6.3 to ω(x)1/5, for any x ∈ Asa. In view of (6.3) and the 20 ) there is d(x) ∈ B such that fact that ω(x) is self-adjoint, we infer that for each α ∈ (0, 1 and ω(x)1/5 = (h2)αd(x)(h2)α kd(x)k 6 kψk16αkxk20αkω(x)1/5k1/2−10α. Since kω(x)k 6 kψk5kxk5, we have (6.5) Again, appealing to inequality (6.2) and splitting any x ∈ Asa into its real and imaginary parts, we obtain the assertion (c). (cid:3) kd(x)k 6 kψk1/2+6αkxk1/2+10α. APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 29 Corollary 6.5. The absolute constant K from Lemma 6.4 has the following property. For any C∗-algebras A and B with A unital, and any self-adjoint ε-o.z. map ψ ∈ L (A,B) with h := ψ(1A), at least one of the following two inequalities holds true: (a) kψk 6p(K + 2)5ε, (b) kψk 6 (K + 2)5khk. Proof. Take any α ∈ (0, 1 10 ). By Lemma 6.4(a), for each x ∈ A with kxk 6 1 there exists u(x) ∈ B such that ku(x)k 6 2kψk1−2α and kψ(x)k 6 Kkψk3/5ε1/5 + 2kψk1−2αk(h2)αk. Since h is self-adjoint, we have k(h2)αk = khk2α. Passing to the limit as α → 1 10 we thus obtain (6.6) For h = 0 our assertion is trivial. If h 6= 0, we rewrite the last inequality as (6.7) kψk2/5 6 Kε1/5 + 2kψk1/5khk1/5. (cid:16)kψk khk(cid:17)1/5 kψkkhk(cid:17)1/5 6 K(cid:16) + 2. ε Hence, if kψkkhk 6 ε, then (6.6) yields inequality (a). Otherwise, (6.7) implies (b). (cid:3) It is hard to calculate the value of K arising from the Aleksandrov -- Peller inequality applied to the Holder exponent α = 1 5. Birman, Koplienko and Solomyak [8], however, proved that for all positive self-adjoint operators A, B on a Hilbert space and for any 0 < α < 1, we have kAα − Bαk 6 kA − Bkα. Therefore, if in Lemma 6.4 we additionally assume that ψ is positive, we can apply the above inequality to the operators ψ(x)5 and ω(x), for any fixed x ∈ A+. Hence, repeating our reasoning, we conclude that for every x ∈ A+ inequality (6.1) and all assertions (a) -- (c) hold true with constant 1 instead of K, kψk+ instead of kψk and with estimates (6.4) (both for u(x) and v(x)) and (6.5). This leads to the following conclusion. Remark. Suppose all the assumptions of Lemma 6.4 are satisfied and that ψ is positive. Then, by first considering positive elements and then using the Jordan decomposition, we infer that (6.1) and assertions (a) -- (c) hold true with constant 4 instead of constants K and 2 and with kψk+ in the place of kψk. Consequently, considering positive elements of A we obtain the following version of Corollary 6.5. Corollary 6.6. For any C∗-algebras A and B with A unital, and any positive ε-o.z. map ψ ∈ L (A,B) with h := ψ(1A), at least one of the following two inequalities holds true: (a') kψk+ 6 166√ε, (b') kψk+ 6 2khk. Proof. As indicated in the remark above, we have kψk2/5 + 6 ε1/5 + kψk1/5 + khk1/5. 30 TOMASZ KOCHANEK Take R = 13764 and consider two cases: If kψk+khk 6 Rε, then kψk+ 6p(1 + R1/5)5ε which yields (a'). Otherwise kψk+ 6 (1 + R−1/5)5khk which implies (b'). (cid:3) Recall that in the result by Alaminos, Extremera and Villena (Theorem 2.1), an im- portant assumption was that the given operator T is surjective and the obtained estimate depended on the openness index of T . In Lemma 5.3 and Proposition 5.4, the obtained estimates depend on a constant M arising from the condition C∗(h) ⊆ φ(Z(A)). More precisely, any M such that every element of the unit ball of C∗(h) can be approximated with arbitrarily small error by some φ(v) with kvk 6 M, does the job. Such a constant was produced by inclusion (5.6), therefore, we define an openness index of the restriction of φ to the center of A relative to C∗(h) by (6.8) oph(φ ↾Z(A)) := 1 sup(cid:8)r > 0 : rBC∗(h) ⊂ φ(BZ(A))(cid:9). As we have already noted, if Z(A) = A and φ happens to be surjective, then oph(φ ↾Z(A)) 6 op(φ). 7. Decomposition -- reducing to the unital case We shall now collect our knowledge about approximate Jordan-like equations, almost com- mutation relations and ranges of ε-o.z. maps in order to prove a decomposition result. Namely, we show that under certain conditions, a self-adjoint ε-o.z. map ψ can be decom- posed into a 'small' part ψs and a 'regular' part ψr which is a unital δ-J.h. map, with δ → 0 as ε → 0. Therefore, the study of almost order zero maps may be sometimes reduced to the study of almost Jordan ∗-homomorphisms, and this will be the topic of another paper. Lemma 7.1. Let S ∈ L (H) be a normal operator on a Hilbert space H and E0 be its spectral measure. Let also δ > 0 and T ∈ L (H) be the operator defined via functional calculus as T = SSδ. Denote by E1 the spectral measure of T . Then, for every ε > 0, we have E0({λ ∈ σ(S) : λ 6 ε}) = E1({λ ∈ σ(T ) : λ 6 ε1+δ}). Proof. For any ξ, η ∈ H we define complex Borel measures E0,ξ,η and E1,ξ,η on σ(S) and σ(T ), respectively, by the formulas E0,ξ,η = hE0(·)ξ, ηi and E1,ξ,η = hE1(·)ξ, ηi. By the spec- tral theorem for C∗(S) and C∗(T ), we have (7.1) Zσ(S) f (λλδ) dE0,ξ,η = hf (T )ξ, ηi = Zσ(T ) f (λ) dE1,ξ,η(λ) for all ξ, η ∈ H and f ∈ C(σ(T )). The first integral can be transformed by substitution as follows. Consider the map σ(S) ∋ λ 7−−→ Φ(λ) := λλδ ∈ σ(T ). APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 31 Then, Φ−1(σ(T )) = σ(S). Hence, for every f ∈ C(σ(T )), we have (7.2) f (λ) dνξ,η, Zσ(S) f (λλδ) dE0,ξ,η = Zσ(T ) where νξ,η is the image measure of E0,ξ,η, that is, νξ,η(A) = E0,ξ,η(Φ−1(A)) for each Borel set A ⊆ σ(T ). The right-hand sides of (7.1) and (7.2) are equal for each f ∈ C(σ(T )) and therefore E1,ξ,η = E0,ξ,η ◦ Φ−1. Since ξ, η ∈ H were arbitrary, and since the scalar measures Ei,ξ,η (ξ, η ∈ H) uniquely determine the spectral measure Ei (i = 1, 2), we have E1 = E0 ◦ Φ−1. Hence, E0({λ ∈ σ(S) : λ 6 ε}) = E0({λ ∈ σ(S) : λ1+δ 6 ε1+δ}) = E0 ◦ Φ−1({λ ∈ σ(T ) : λ 6 ε1+δ}) = E1({λ ∈ σ(T ) : λ 6 ε1+δ}). (cid:3) Below, K < ∞ stands for the absolute constant from Lemma 6.4. To avoid any irrelevant technical difficulties, we restrict ourselves to parameters ε ∈ (0, 1]. Theorem 7.2. Let A, B be C∗-algebras with A unital and let π be a nondegenerate rep- resentation of B on a Hilbert space H. Let also ψ ∈ L (A,B) be a self-adjoint ε-o.z. map with some ε ∈ (0, 1] and with h := ψ(1A) which satisfies at least one of the following two conditions: (H1) h is an algebraic element of B; (H2) C∗(h) ⊆ ψ(Z(A)). Then, there exists a decomposition ψ = ψs + ψr, where the operators ψs, ψr ∈ L (A, π(B)′′) satisfy the following conditions: (A1) kψsk 6( (6K + 7)kψk4/5ε1/16 under (H1) (6K + 7)kψk4/5ε0.0003 under (H2); (A2) ψr takes values in a corner subalgebra C of hπ(B)′′h; (A3) either ψr = 0 or ψr(1A) is invertible in C, in which case ψr(1A)−1ψr(· ) is a unital δ-J.h. map with δ =( 24(cid:0)C 2(K + 2)5 + 10C + 17(cid:1)kψkε1/16 under (H1) D(cid:0)kψk391/128 + kψk31/8(cid:1)O(ε0.0003) under (H2), where C depends only on the degree of algebraicity of h as in Proposition 5.2, whereas D depends only on oph(ψ ↾Z(A)). Moreover, if ψ is assumed to be positive, then under hypothesis (H1) we have δ = 24(2C 2 + 10C + 17)kψkε1/16. kψsk 6 37kψk4/5ε1/16 and 32 TOMASZ KOCHANEK Proof. First, observe that in the case h = 0 our assertion follows easily from Proposition 3.2. Indeed, for every x ∈ A we have kψ(x)2k 6 108εkxk2 and since ψ is self-adjoint, we have kψ(y)k 6 √108εkyk for each y ∈ Asa. It follows that kψ(x)k 6 2√108εkxk for x ∈ A. So, in this case we simply set ψs = ψ and ψr = 0. In the rest of the proof we thus assume that h 6= 0. Secondly, observe that our assertion is also valid with ψs = ψ and ψr = 0 in the case where inequality (a) from Corollary 6.5 holds true. Henceforth, we can thus assume that ψ satisfies the other inequality: (7.3) Consequently, there is nothing to prove if khk 6 ε1/2 (once more, direct verification shows that the assertion holds true with ψr = 0), so we can additionally assume from this point on that khk > ε1/2. For the sake of readability, we will divide the rest of the proof into several steps. kψk 6 (K + 2)5khk. Step 1: Corner decomposition with parameter We claim that for any θ > 0 there is a sufficiently small β > 0 such that kψ(x) − (h2)βψ(x)(h2)βk 6 (3Kkψk3/5ε1/5 + θ)kxk for every x ∈ A. 20, let d(x) be the element of B produced by (7.4) Indeed, for arbitrary x ∈ A and 0 < α < 1 Lemma 6.4(c). Then, for any β > 0, we have k(h2)βψ(x)(h2)β−(h2)α+βd(x)(h2)α+βk 6 kψk4β·kψ(x) − (h2)αd(x)(h2)αk 6 Kkψk3/5+4βε1/5kxk. Therefore, (7.5) kψ(x)−(h2)βψ(x)(h2)βk 6 k(h2)αd(x)(h2)α − (h2)α+βd(x)(h2)α+βk + Kkψk3/5(1 + kψk4β)ε1/5kxk. Define χα,β = k(h2)α − (h2)α+βk and estimate the first summand in (7.5) as follows: k(h2)αd(x)(h2)α−(h2)α+βd(x)(h2)α+βk 6 k(h2)αk·kd(x)(h2)α − d(x)(h2)α+βk + kd(x)(h2)α+βk·χα,β 6 kψk2α(1 + kψk2β)kd(x)k·χα,β 6 2kψk1/2+8α(1 + kψk2β)kxk1/2+10α·χα,β. By the Gelfand -- Naimark theorem for C∗(h), it is easily seen that limβ→0+ χα,β = 0 uni- formly for α's from any bounded set, in particular, for α ∈ (0, 1 20). Consequently, for any fixed θ > 0, one can pick β > 0 so that the right-hand side of the last inequality is smaller than θkxk1/2+10α, for each α ∈ (0, 1 20 ). Of course, we can also assume that kψk4β < 2. Passing to the limit as α → 1 20 from the left and using (7.5) we obtain the announced inequality (7.4). APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 33 Fix any θ > 0. Having established the claim, we define an operator eψ : A → hBh by where β is chosen to be any positive number for which inequality (7.4) holds true. Observe that, in fact, the so-defined operator takes values in the hereditary C∗-subalgebra hBh. Indeed, eψ(x) = (h2)βψ(x)(h2)β, (h2)β ∈ C∗((h2)3) ⊆ h2Bh2 ⊆ hBh true for eψ(x), as multiplication is jointly norm continuous. and therefore (h2)β can be approximated in norm by elements from hBh, so the same is Henceforth, for simplicity of notation, we assume that B acts nondegenerately on H, that is, π is the identity (however, we still use the symbol π(B)′′ rather than B′′ to avoid any confusion). We will use the following convention: We write a . b provided that for any η > 0, the inequality a 6 b + η is true after possibly decreasing the parameter β, while not violating any other statements along the proof. Let us also point out that working under assumption (H1) we keep track on actual constants of approximation, whereas under (H2) we just care about whether they depend only on the openness index relative to C∗(h). Define g := eψ(1A) = h·h4β ∈ Bsa; note that g 6= 0 because h 6= 0. Let E0 and E1 be the spectral measures of h and g, respectively. For any parameter γ > 0 satisfying εγ < khk (equivalently: εγ(1+4β) < kgk), we consider the spectral projection in π(B)′′ given by (7.6) pγ := E1({λ ∈ σ(g) : λ 6 εγ(1+4β)}). In fact, pγ does not depend on β, since from Lemma 7.1 it follows that (7.7) pγ = E0({λ ∈ σ(h) : λ 6 εγ}). Therefore, for every δ > 0, we have pγ(h2)δ = (h2)δpγ = Z [−εγ,εγ]∩σ(h) λ2δ dE0(λ) and hence (7.8) Similarly, (7.9) kpγ(h2)δk = k(h2)δpγk = sup [−εγ,εγ]∩σ(h) λ2δ 6 ε2γδ. k(1H − pγ)(h2)δk = k(h2)δ(1H − pγ)k 6 sup σ(h) λ2δ 6 khk2δ 6 kψk2δ. Consider the corner decomposition For each γ > 0 with εγ < khk, we may thus write ψ = ψs,γ + ψr,γ, where (7.10) eψ(x) = pγeψ(x)pγ + pγeψ(x)(1H − pγ) + (1H − pγ)eψ(x)pγ + (1H − pγ)eψ(x)(1H − pγ). ψs,γ(x) = (ψ(x) −eψ(x)) + pγeψ(x)pγ + pγeψ(x)(1H − pγ) + (1H − pγ)eψ(x)pγ 34 and (7.11) TOMASZ KOCHANEK ψr,γ(x) = (1H − pγ)eψ(x)(1H − pγ). In what follows, we shall prove that regardless of which of the hypotheses (H1) and (H2) holds true, there is an appropriate value of γ for which the operators ψs = ψs,γ and ψr = ψr,γ enjoy all the desired properties. Step 2: Upper bound for kψs,γk and x ∈ A, let u(x) ∈ B be as in that lemma. Notice that in view of inequality (7.8) and First, in order to estimate the norm of pγeψpγ, we apply Lemma 6.4(a). For any 0 < α < 1 the definition of eψ, we have 10 kpγeψ(x)pγ − pγ(h2)βu(x)(h2)α+βpγk 6 kpγ(h2)βk·kψ(x) − u(x)(h2)αk·k(h2)βpγk 6 Kkψk3/5ε1/5+4βγkxk. The estimate for ku(x)k given in Lemma 6.4(a) jointly with inequality (7.8) yields kpγ(h2)βu(x)(h2)α+βpγk 6 2kψk1−2αε2(α+2β)γkxk. Hence, for every 0 < α < 1 10 , we have (7.12) kpγeψpγk 6 Kkψk3/5ε1/5+4βγ + 2kψk1−2αε2(α+2β)γ. any 0 < α < 1 (7.9), we have 10 and x ∈ A, let v(x) ∈ B be as in that lemma. By inequalities (7.8) and Now, in order to estimate the norm of pγeψ(1H − pγ), we appeal to Lemma 6.4(b). For kpγeψ(x)(1H − pγ) − pγ(h2)α+βv(x)(h2)β(1H − pγ)k 6 kpγ(h2)βk·kψ(x) − (h2)αv(x)k·k(h2)β(1H − pγ)k 6 Kkψk3/5+2βε1/5+2βγkxk. Using the estimate for kv(x)k, along with (7.8) and (7.9), we obtain kpγ(h2)α+βv(x)(h2)β(1H − pγ)k 6 2kψk1−2α+2βε2(α+β)γkxk. Hence, (7.13) (7.14) By a similar argument, and appealing to the "left-handed" assertion (a) from Lemma 6.4 instead of the "right-handed" assertion (b), we obtain kpγeψ(1H − pγ)k 6 Kkψk3/5+2βε1/5+2βγ + 2kψk1−2α+2βε2(α+β)γ. k(1H − pγ)eψpγk 6 Kkψk3/5+2βε1/5+2βγ + 2kψk1−2α+2βε2(α+β)γ. APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 35 Combining inequalities (7.4), (7.12), (7.13) and (7.14), and recalling formula (7.10), we obtain kψs,γk 6 3Kkψk3/5ε1/5 + θ + Kkψk3/5ε1/5+4βγ + 2kψk1−2αε2(α+2β)γ + 2Kkψk3/5+2βε1/5+2βγ + 4kψk1−2α+2βε2(α+β)γ. Taking a suitable θ > 0 (depending on ε, γ and kψk), picking β > 0 small enough and passing to the limit as α → 1 (7.15) 10−, we can guarantee that kψs,γk 6 6Kkψk3/5ε1/5 + 7kψk4/5εγ/5. Step 3: Properties of the regular part Now, we claim that ψr,γ has the following properties: (i) ψr,γ : A −→ Cγ := (1H − pγ)hπ(B)′′h(1H − pγ), (ii) kψr,γk 6 kψk1+4β, (iii) ψr,γ is self-adjoint, (iv) ψr,γ(1A) is invertible in Cγ and kψr,γ(1A)−1k 6 ε−γ(1+4β). The property (i) follows from formula (7.11) and the fact that eψ takes values in hBh. The clauses (ii) and (iii) follow from the very definition. To see that (iv) holds true, observe that ψr,γ(1A) = (1H − pγ)g and that, in view of (7.6), we have σ((1H − pγ)g) ⊆ [−kgk,−εγ(1+4β)] ∪ [εγ(1+4β),kgk], the spectrum taken in the corner algebra Cγ. in view of formula (7.7), we have We shall now verify that ψr,γ is an almost order zero map. To this end, first observe that (1H − pγ)(1H − (h2)2β) = (1 − λ4β) dE(λ). {λ∈σ(h) : λ>εγ} Z Therefore, (7.16) κβ := k(1H − pγ)(1H − (h2)2β)k 6 sup εγ<t6khk1 − t4β −−−−→β→0+ 0. Fix any x, y ∈ A+ with xy = 0. By the definition (7.11) of ψr,γ, and the fact that pγ and h commute, we have (7.17) ψr,γ(x)ψr,γ(y) = (1H − pγ)(h2)βψ(x)(1H − pγ)(h2)2βψ(y)(1H − pγ)(h2)β. Plainly, we have kψ(x)(1H − pγ)(h2)2β − ψ(x)(1H − pγ)k 6 κβkψkkxk and hence, in view of (7.17), we get (7.18) kψr,γ(x)ψr,γ(y)k 6 k(1H − pγ)(h2)βk2(cid:16)κβkψk2kxkkyk + kψ(x)(1H − pγ)ψ(y)k(cid:17). 36 TOMASZ KOCHANEK Recalling that khk > εγ and that we have assumed (7.3), we infer that under hypothesis (H1), Proposition 5.2 yields kP (h)ψ(x) − ψ(x)P (h)k 6 Cε1−γ sup z=khkP (z) · kxk for every P ∈ C[z] with P (0) = 0, where C depends only on the degree of algebraicity of h. On the other hand, it follows from Proposition 5.4 that under hypothesis (H2), we have kP (h)ψ(x) − ψ(x)P (h)k 6 8(K + 2)5oph(ψ ↾Z(A))(cid:0)kψk15/8O(ε1/16) + 24ε(cid:1) sup z=khkP (z)·kxk, where in the role of M we took the openness index relative to C∗(h) and, once again, we used inequality (7.3). Note also that we have omitted the term 2ε1/2 under the maximum sign, as in our case kψk > khk > ε1/2 (see the beginning of the proof of Proposition 5.4). Since kψk15/8 > ε15/16, the term 24ε is majorized by kψk15/8O(ε1/16) and we can rewrite the above inequality in a simpler form: kP (h)ψ(x) − ψ(x)P (h)k 6 Dkψk15/8O(ε1/16) sup z=khkP (z) · kxk, where D depends only on the openness index of ψ ↾Z(A) relative to C∗(h). In each case, we can apply Lemma 5.5 to the spectral projection V = 1H − pγ and the operator T = ψ(x). In this way, we obtain an estimate on the norm of the commutator: k[1H − pγ, ψ(x)]k 6 Cε1−γkxk under (H1) Dkψk15/8O(ε1/16)kxk under (H2). (7.19) (7.20) (7.21) Therefore, by the fact that ψ is ε-o.z., we obtain kψ(x)(1H − pγ)ψ(y)k 6 k(1H − pγ)ψ(x)ψ(y)k + k[1H − pγ, ψ(x)]kkψ(y)k Ckψkε1−γ + ε under (H1) Dkψk23/8O(ε1/16) + ε under (H2). 6 kxkkyk · In view of (7.16), by decreasing β if necessary, we can assure that κβ is arbitrarily small and kψk4β close to 1. Hence, combining (7.18) and (7.20), we obtain an estimate on kψr,γ(x)ψr,γ(y)k which means that the operator ψr,γ is ξ-o.z., where ξ = Ckψkε1−γ + 2ε under (H1) Dkψk23/8O(ε1/16) under (H2). APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 37 Note that in the second row there is a new constant D which still depends only on oph(ψ ↾Z(A)), while the value of C did not change and it comes from Proposition 5.2 applied to N being the degree of algebraicity of h and M = (K + 2)5. Step 4: Estimating the norm of the commutator [ψr,γ(1A)−1, ψr,γ(x)] Fix any x ∈ A+ and denote hγ := ψr,γ(1A). Since εγ < khk, the projection 1H − pγ does not affect the largest in absolute value elements of σ(h) and hence khγk = khk1+4β. By our assumption (7.3) and inequality (ii), we thus obtain (7.22) kψr,γk 6 (K + 2)5(1+4β)khγk. First, assume (H1) holds true. Notice that the spectrum of hγ is finite and has no more elements than σ(h). Hence, hγ is also algebraic of degree not larger than the degree of h. By virtue of Proposition 5.2, for every P ∈ C[z] with P (0) = 0 we have (7.23) z=khγ kP (z) · kxk. sup kP (hγ)ψr,γ(x) − ψr,γ(x)P (hγ)k . Cξ khγk By (7.3), (7.21) and the fact that khk > εγ, we have khk4β + 6 C(K + 2)5 ε1−γ ξ khγk Ckψkε1−γ + 2ε = khk1+4β 2ε εγ(1+4β) . (C(K + 2)5 + 2)ε1−γ. Therefore, (7.23) yields (7.24) kP (hγ)ψr,γ(x) − ψr,γ(x)P (hγ)k . (C 2(K + 2)5 + 2C)ε1−γ z=khγ kP (z) · kxk. sup Consider a linear operator defined on the space of complex continuous functions on the ring {εγ(1+4β) 6 z 6 khγk} ⊇ σ(hγ) by the formula f 7→ f (hγ)ψr,γ(x) − ψr,γ(x)f (hγ). The Stone -- Weierstrass theorem and estimate (7.24) imply that, by decreasing β, the norm of such an operator can be estimated by any number larger than (C 2(K + 2)5 + 2C)ε1−γ. Taking the inverse function z 7→ z−1, whose supremum norm on the ring equals ε−γ(1+4β), we obtain (7.25) kh−1 γ ψr,γ(x) − ψr,γ(x)h−1 γ k . (C 2(K + 2)5 + 2C)ε1−2γkxk. Now, assume (H2) holds true and notice that C∗(h) = (h2)βC∗(h)(h2)β ⊆ (h2)βψ(Z(A))(h2)β = eψ(Z(A)). (For the first equality one can use Proposition 6.3, or simpler, apply the Gelfand -- Naimark theorem to C∗(h).) Consequently, C∗(hγ) ⊆ (1H − pγ)C∗(h) ⊆ ψr,γ(Z(A)), which means that ψr,γ satisfies the condition analogous to (H2). More precisely, any w ∈ C∗(hγ) can be written as w = (1H − pγ)(h2)2βw′ with w′ ∈ C∗(h) satisfying kw′k = khk−4βkwk. Hence, appealing to formula (6.8), we easily get that ophγ (ψr,γ ↾Z(A)) . oph(ψ ↾Z(A)). 38 TOMASZ KOCHANEK By Proposition 5.4 and inequalities (7.22), (ii), for every P ∈ C[z] with P (0) = 0, we have kP (hγ)ψr,γ(x) − ψr,γ(x)P (hγ)k . 8(K + 2)5oph(ψ ↾Z(A))(cid:0)kψk15/8O(ξ1/16) + 24ξ(cid:1) sup z=khγkP (z) · kxk. Hence, in view of (7.21), kP (hγ)ψr,γ(x) − ψr,γ(x)P (hγ)k 6 Dnkψk263/128O(ε1/256) + kψk23/8O(ε1/16)o sup z=khγkP (z) · kxk, where D is a new constant depending only on oph(ψ ↾Z(A)). Arguing as before we conclude that, for a possibly new value of D, we have (7.26) kh−1 γ k 6 Dε−γnkψk263/128O(ε1/256) + kψk23/8O(ε1/16)o · kxk. γ ψr,γ(x) − ψr,γ(x)h−1 Step 5: Picking the right parameter We keep x ∈ A+ fixed. As we have proved that ψr,γ is ξ-o.z., it follows from Proposition 3.2 (see the beginning of the proof) that kψr,γ(x)2 − hγψr,γ(x2)k 6 8ξkxk2. Taking the inverse of h2 γ in the corner algebra Cγ we get kh−2 γ ψr,γ(x)2 − h−1 (7.27) Define an operator Ξγ : A → Cγ by Ξγ(x) = h−1 by (iii) we have γ ψr,γ(x2)k 6 8ξkh−1 γ k2kxk2 . 8ξε−2γkxk2. γ ψr,γ(x). Note that Ξγ is unital and since (7.28) k Ξγ(y∗) − Ξγ(y)∗k = kh−1 γ ψr,γ(y) − ψr,γ(y)h−1 γ k (y ∈ A), it is also δ-s.a., where δ can be estimated with the aid of either (7.25) or (7.26) depending on whether (H1) or (H2) holds true (here, we use the Jordan decomposition for the real and imaginary parts of y and hence the error terms given by (7.25) and (7.26) should be multiplied by 4). By inequalities (ii), (iv), (7.25) and (7.26), we also have kh−2 γ ψr,γ(x)2 − Ξγ(x)2k = kh−2 γ ψr,γ(x)2 − h−1 γ ψr,γ(x)h−1 γ ψr,γ(x)k (7.29) 6 kh−1 γ k · kh−1 . kxk2 · γ k · kψr,γ(x)k γ ψr,γ(x) − ψr,γ(x)h−1 (C 2(K + 2)5 + 2C)kψkε1−3γ Dε−2γnkψk391/128O(ε1/256) + kψk31/8O(ε1/16)o under (H2). under (H1) Now, it is convenient to assume that kψk > ε15/46 (otherwise the assertion holds with ψs = ψ and ψr = 0), as it implies that kψk23/8O(ε1/16) is majorized by kψk391/128O(ε1/256). Combining (7.27) and (7.29), and recalling formula (7.21), we obtain (7.30) k Ξγ(x)2 − Ξγ(x2)k . ∆(ε,kψk)kxk2, APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 39 where (7.31) ∆(s, t) := (C 2(K + 2)5 + 10C)ts1−3γ + 16s1−2γ under (H1) Here, once again, we have possibly changed the value of D, whereas C remained the same. Ds−2γ(cid:8)t391/128O(s1/256) + t31/8O(s1/16)(cid:9) under (H2). γ ψr,γ(x) − ψr,γ(x)h−1 Observe also that for any x, y ∈ A+ with xy = 0, we have k Ξγ(x)Ξγ(y)k 6 kh−1 γ k2·kψr,γ(x)ψr,γ(y)k, where the first summand can be estimated as in (7.29) replacing kxk2 by kxkkyk, whereas the second summand satisfies γ k·kψr,γ(y)k + kh−1 γ k·kh−1 kh−1 γ k2·kψr,γ(x)ψr,γ(y)k . kxkkyk · Ckψkε1−3γ + 2ε1−2γ under (H1) Dε−2γkψk23/8O(ε1/16) under (H2), due to inequality (iv) and formula (7.21). Hence, recalling that kψk23/8O(ε1/16) is majorized by kψk391/128O(ε1/256), for all x, y as above, we obtain (7.32) Notice that the right-hand side of (7.28) is also majorized by ∆(ε,kψk). Using the Jordan decomposition exactly in the same way as at the end of the proof of Proposition 3.2, we conclude from (7.30) and (7.32) that given any η > 0, we can decrease β so that k Ξγ(x)Ξγ(y)k . ∆(ε,kψk)kxkkyk. (7.33) Ξγ is η + 24∆(ε,kψk)-J.h. Now, we will optimize our choice of the parameter γ to make all the relevant error estimates as good as possible in terms of their behavior with respect to ε. To this end, recall that γ was supposed to satisfy εγ < khk and if this is not true, then (7.3) yields kψk 6 (K + 2)5εγ; in this case we set ψs = ψ, ψr = 0. So, as indicated several lines above and at the very beginning of the proof, the possible estimates on kψs,γk are of order ε15/46 and εγ, whereas (7.15) gives an estimate of order εγ/5 and, in general, this is the largest term. Assuming (H1), we see from (7.33) that Ξγ is δ(ε)-J.h. with an error δ(ε) of order ε1−3γ. Thus, we want to maximize 1 5γ and 1 − 3γ at the same time, hence we pick γ so that 1 5γ = 1 − 3γ. Similarly, under hypothesis (H2), Ξγ is δ(ε)-J.h. with δ(ε) of order ε1/256−2γ and thus we solve the equation 1 256 − 2γ. Therefore, we put 5γ = 1 γ = under (H1) 5 16 5 2816 under (H2). Notice that, indeed, our choice is compatible with the required estimate on kψsk, as the inequality kψk 6 (K + 2)5εγ implies that (A1) works with ψs = ψ (we have either 1 5γ = 1 or 1 5γ = 1 Having specified γ, we define ψs = ψs,γ, ψr = ψr,γ and C = Cγ. Notice that (A1) follows from (7.15), while (A2) is just condition (i). Assertion (A3) follows from (iv) and, for 2816 > 0.0003). 16 40 TOMASZ KOCHANEK a suitable β > 0, from formula (7.31) and condition (7.33). Note that under (H1), the term 16ε1−2γ was majorized by 16kψkε1−3γ and slightly increased due to the appearance of η > 0 in (7.33), whereas under hypothesis (H2) we simply majorized both ε1/256−2γ and ε1/16−2γ by ε0.0003. Finally, assuming that ψ is positive we can apply Corollary 6.6. If inequality (a') holds true, then our assertion is trivial with ψs = ψ. Otherwise, instead of (7.3) we can assume that (b') is valid. Notice that (7.4) remains true with 4 in the place of K, whereas in (7.12) -- (7.14) we replace the pair (K, 2) by (4, 4), due to the remark after Corollary 6.5. Consequently, instead of (7.15) we obtain kψs,γk 6 24kψk3/5ε1/5 + 13kψk4/5εγ/5 which can be assumed to be at most 37kψk4/5ε1/16 (after picking γ = 5 16). Furthermore, in formula (7.21) we can replace kψk by kψk+ and then in (7.22), (7.24), (7.25), (7.29) and (7.31) we substitute constant 2 for (K + 2)5. (cid:3) Remark. Having at disposal a stability result for almost Jordan ∗-homomorphisms, it is possible that another choice of γ would be better. For example, suppose that any ε-J.h. map between given C∗-algebras can be approximated by a Jordan ∗-homomorphism to within O(εω). Then, under hypothesis (H1), such a homomorphism would lie at distance of order εω(1−3γ) from Ξγ and we should pick γ so that 1 5γ = ω(1 − 3γ). A similar modification would take place under hypothesis (H2). This is actually the main reason (besides trying to make the whole proof more readable) for which we have been working with the parameter γ all the way through and picked its right value just at the end of the proof. 8. Examples Assumption (H1) of Theorem 7.2 is automatically satisfied if the codomain algebra is a matrix algebra Mn(C), as every n × n matrix has a minimal polynomial of degree at most n. From the proof of Proposition 5.2, it follows that the constant C appearing in assertion (A3) can be then defined by (8M)n−1 − 1 , C = M − 1 8 where M 6 (K +2)5, as can be seen from inequality (7.3). If the map in question is positive we have M 6 2, as then we work under assumption (b') from Lemma 6.6. Therefore, the parameter δ from assertion (A3) satisfies as n → ∞, (8.1) δ = O(cid:0)C 2(K + 2)5(cid:1)kψkε1/16 = O(cid:0)(64(K + 2)10)n(cid:1)kψkε1/16 or, in the positive case, δ = O(256n)kψkε1/16 as n → ∞. A similar observation applies in the case where the codomain B is an arbitrary finite- dimensional C∗-algebra, that is, for some n1, . . . , nk ∈ N it can be represented as (8.2) Here, every element of B is algebraic of order at most N = n1 · . . .· nk, hence estimate (8.1) is valid with N instead of n. B ∼= Mn1(C) ⊕ Mn2(C) ⊕ . . . ⊕ Mnk (C). APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 41 Corollary 8.1. Let A be any C∗-algebra and B be a finite-dimensional C∗-algebra given by (8.2) with N = n1 · . . . · nk. Assume that φ ∈ L (A,B) is an ε-d.p. map, with some ε ∈ (0, 1]. Then, there exists a corner C∗-subalgebra C of B and an operator Φ ∈ L (A†,C) satisfying and such that either Φ = 0 or Φ(1A†) is invertible in C, in which case the operator Φ(1A†)−1Φ(· ) is δ-J.h. with 1 kφ − Φk 6 (6K + 7)(cid:16)kφk4/5 + 32kφk9/5(cid:17)ε1/16 + δ = O(cid:0)(64(K + 2)10)N(cid:1) max{1,kφk2}ε1/16. 1 2 ε Proof. Appealing to Lemma 2.3 we obtain a self-adjoint (ε + 1 such that kφ − ψk 6 1 which is still self-adjoint and (ε + 1 a suitable decomposition ψ† = ψs + ψr and observe that the operator Φ := ψr satisfies 2εkφk)-o.z. map ψ ∈ L (A,B) 2ε. Using Lemma 3.1 we may extend ψ to a map ψ† ∈ L (A†,B) 2 εkφk)-o.z. It remains to apply Theorem 7.2 to obtain kφ − Φk 6 kφ − ψk + kψsk 6 (6K + 7)kψk4/5(cid:16)1 + 32kφk(cid:17)ε1/16 + 6 (6K + 7)kφk4/5(cid:16)1 + 1 2 1 ε. 1 2kφk(cid:17)1/16 ε1/16 + 1 2 ε (The last estimate follows from Bernoulli's inequality and the fact that kψk 6 kφk.) From Theorem 7.2 and the discussion above it follows that either Φ = 0 or Φ(1A†) is invertible in the range algebra, in which case the operator Φ(1A†)−1Φ(· ) is δ-J.h. with δ 6 O(cid:0)(64(K + 2)10)N(cid:1)kφk(cid:16)1 + 1 32kφk(cid:17)ε1/16. (cid:3) If the map in question is positive, there is, of course, no need of applying Lemma 2.3 and then Theorem 7.2 yields the following result. Corollary 8.2. Let A be any C∗-algebra and B be a finite-dimensional C∗-algebra given by (8.2) with N = n1 · . . . · nk. Assume that φ ∈ L (A,B) is a positive ε-o.z. map, with some ε ∈ (0, 1]. Then, there exists a corner C∗-subalgebra C of B and an operator Φ ∈ L (A†,C) satisfying kφ − Φk 6 37kφk4/5ε1/16 δ = O(cid:0)256N(cid:1)kφkε1/16. and such that either Φ = 0 or Φ(1A†) is invertible in C, in which case the operator Φ(1A†)−1Φ(· ) is δ-J.h. with Assumption (H2) of Theorem 7.2 is automatically satisfied if the domain is a commutative C∗-algebra and the map considered is surjective, which leads to another corollary. Corollary 8.3. Let X be a locally compact Hausdorff space and B be a C∗-algebra acting nondegenerately on a Hilbert space H. Assume that ψ ∈ L (C0(X),B) is a surjective self- adjoint ε-o.z. map, with some ε ∈ (0, 1]. Then, there exists a corner C∗-subalgebra C of B′′ and an operator Ψ ∈ L (C0(X)†,C) satisfying kψ − Ψk 6 (6K + 7)kψk4/5O(ε0.0003) 42 TOMASZ KOCHANEK and such that either Ψ = 0 or Ψ(1C0(X)†) is invertible in C, in which case the operator Ψ(1C0(X)†)−1Ψ(· ) is δ-J.h. with where D depends only on the openness index op(ψ). δ = D(cid:0)kψk391/128 + kψk31/8(cid:1)O(ε0.0003), References 1. J. Alaminos, M. Bresar, J. Extremera, A.R. Villena, Characterizing homomorphisms and derivations on C∗-algebras, Proc. Royal Soc. Edinburgh 137A (2007), 1 -- 7. 2. J. Alaminos, M. Bresar, J. Extremera, A.R. Villena, Maps preserving zero products, Studia Math. 193 (2009), 131 -- 159. 3. J. Alaminos, J. Extremera, A.R. Villena, Approximately zero-product-preserving maps, Israel J. Math. 178 (2010), 1 -- 28. 4. A.B. Aleksandrov, V.V. Peller, Operator Holder -- Zygmund functions, Adv. Math. 224 (2010), 910 -- 966. 5. J. Araujo, J.J. Font, Stability of weighted composition operators between spaces of continuous functions, J. London Math. Soc. 79 (2009), 363 -- 376. 6. J. Araujo, J.J. Font, On the stability index for weighted composition operators, J. Approx. Theory 162 (2010), 2136 -- 2148. 7. J. Araujo, J.J. Font, Stability of weighted point evaluation functionals, Proc. Amer. Math. Soc. 138 (2010), 3163 -- 3170. 8. M.S. Birman, L.S. Koplienko, M.Z. Solomyak, Estimates of the spectrum of a difference of fractional powers of selfadjoint operators, Izv. Vyssh. Uchebn. Zaved. Mat. 3 (1975), 3 -- 10. English transl.: Soviet Math. (Iz. VUZ) 19 (1975), 1 -- 6. 9. B. Blackadar, Operator algebras. Theory of C∗-algebras and von Neumann algebras, Encyclopaedia of Mathematical Sciences, vol. 122, Springer-Verlag, Berlin Heidelberg 2006. 10. M.A. Chebotar, W.-F. Ke, P.-H. Lee, N.-C. Wong, Mappings preserving zero products, Studia Math. 155 (2003), 77 -- 94. 11. M.-D. Choi, Almost commuting matrices need not be nearly commuting, Proc. Amer. Math. Soc. 102 (1988), 529 -- 533. 12. G. Dolinar, Stability of disjointness preserving mappings, Proc. Amer. Math. Soc. 130 (2001), 129 -- 138. 13. J.J. Font, Disjointness preserving mappings between Fourier algebras, Colloq. Math. 77 (1998), 179 -- 187. 14. A.K. Gaur, Z.V. Kov´ar´ık, Norms on unitizations of Banach algebras, Proc. Amer. Math. Soc. 117 (1993), 111 -- 113. 15. B.E. Johnson, Cohomology in Banach algebras, Memoirs of the American Mathematical Society 127, American Mathematical Society, Providence, R.I. 1972. 16. B.E. Johnson, Approximate diagonals and cohomology of certain annihilator Banach algebras, Amer. J. Math. 94 (1972), 685 -- 698. 17. B.E. Johnson, Approximately multiplicative functionals, J. London Math. Soc. 34 (1986), 489 -- 510. 18. B.E. Johnson, Approximately multiplicative maps between Banach algebras, J. London Math. Soc. 37 (1988), 294 -- 316. 19. A.S. Kechris, Classical descriptive set theory, Graduate Texts in Mathematics 156, Springer-Verlag, New York 1995. 20. H. Lin, Almost commuting selfadjoint matrices and applications. Operator algebras and their applica- tions (Waterloo, ON, 1994/1995), 193 -- 233, Fields Inst. Commun. 13, Amer. Math. Soc., Providence, RI 1997. 21. T.A. Loring, Lifting solutions to perturbing problems in C∗-algebras, Fields Institute Monographs, vol. 8, Providence, R.I., Amer. Math. Soc. 1997. APPROXIMATELY ORDER ZERO MAPS BETWEEN C∗-ALGEBRAS 43 22. T. Oikhberg, P. Tradacete, Almost disjointness preservers, Canadian J. Math. 69 (2017), 650 -- 686. 23. M. Takesaki, Theory of operator algebras I, Springer-Verlag, New York 1979. 24. S.M. Ulam, A collection of mathematical problems, Wiley -- Interscience 1960. 25. W. Winter, J. Zacharias, Completely positive maps of order zero, Munster J. Math. 2 (2009), 311 -- 324. 26. W. Winter, J. Zacharias, The nuclear dimension of C∗-algebras, Adv. Math. 224 (2010), 461 -- 498. 27. M. Wolff, Disjointness preserving operators on C∗-algebras, Arch. Math. 62 (1994), 248 -- 253. Institute of Mathematics, Polish Academy of Sciences, ´Sniadeckich 8, 00-656 Warsaw, Poland and Institute of Mathematics, University of Warsaw, Banacha 2, 02-097 Warsaw, Poland E-mail address: [email protected]
1012.5586	2	1012	2011-06-17T14:08:12	Freeness of Linear and Quadratic Forms in von Neumann Algebras	[ "math.OA" ]	We characterize semicircular distribution by the freeness of linear and quadratic forms in noncommutative random variables from a tracial $W^*$-probability space with relaxed moment conditions.	math.OA	math	FREENESS OF LINEAR AND QUADRATIC FORMS IN VON NEUMANN ALGEBRAS. G. P. CHISTYAKOV1 , 2 , 4, F. G OTZE1 , 4, AND F. LEHNER3 , 4 Abstract. We characterize the semicircular distribution by freeness of lin- ear and quadratic forms in noncommutative random variables from tracial W ∗- probability spaces with relaxed moment conditions. 1. Introduction The intensive research in the asymptotic theory of random matrices has mo- tivated increased research on infinitely dimensional limiting models. Free convo- lution of probability measures, introduced by D. Voiculescu, may be regarded as such a model [20], [21]. The key concept of this definition is the notion of freeness, which can be interpreted as a kind of independence for noncommutative random variables. As in classical probability the concept of independence gives rise to clas- sical convolution, the concept of freeness leads to a binary operation on probability measures on the real line which is called free convolution. Many classical results in the theory of addition of independent random variables have their counterpart in this theory, such as the law of large numbers, the central limit theorem, the L´evy- Khintchine formula and others. We refer to Voiculescu, Dykema and Nica [22], Hiai and Petz [7], and Nica and Speicher [16] for an introduction to these topics. The central limit theorem for free random variables holds with limit distribution equal to a semicircle law. Semicircle laws play in many respects the role of Gaussian laws, when independence is replaced by freeness in a noncommutative probability space. In usual probability theory various characterizations of the Gaussian law have been obtained, for instance see [9]. In particular, there is the well-known fact that the independence of the sample mean and the simple variance of independent identically distributed random variables characterizes the Gaussian laws, see [19] and [10]. Date: November 16, 2018. 1991 Mathematics Subject Classification. Primary 46L50, 60E07; Secondary 60E10. Key words and phrases. Free random variables, free convolutions, a characterization of the semicircle law. 1) Faculty of Mathematics, University of Bielefeld, Germany. 2) Institute for Low Temperature Physics and Engineering, Kharkov, Ukraine. 3) Institut fur Mathematische Strukturtheorie, Technische Universitat Graz. 4) Research supported by SFB 701. 1 2 G. P. CHISTYAKOV, F. G OTZE, AND F. LEHNER Hiwatashi, Nagisa and Yoshida [8] established the characterization of the semi- circle law by freeness of a certain pair of a linear and a quadratic form in free identically distributed bounded noncommutative random variables, which covers the free analogue of the previous result in usual probability theory. In this paper we generalize the Hiwatashi, Nagisa and Yoshida result to the case of not necessarily bounded identically distributed noncommutative random variables requiring only finiteness of the second moment. Unbounded operators affiliated to a von Neumann algebra play the role of un- bounded measurable random variables in noncommutative probability. A general theory of such operators has been developed already by Murray and Neumann [15]. In free probability unbounded random variables have so far only been considered by Maassen [14] from the analytic point of view and by Bercovici and Voiculescu [4] in great detail. The plan of the present paper is as follows. In Section 2 we formulate our results. In Section 3 we give auxiliary results on measurable operators. In Section 4 we prove auxiliary analytic results. Finally in Section 5 we prove our main result by carefully adapting classical moment estimates to the noncommutative situation. Acknowledgements. The authors thank the anonymous referee for numerous remarks, in particular pointing out an error in Proposition 4.3. 2. Results Assume that A is a finite von Neumann algebra with normal faithful trace state τ acting on a Hilbert space H. The pair (A, τ ) will be called a tracial W ∗- probability space. We will denote by A the set of all operators on H which are affiliated with A and by Asa its real subspace of selfadjoint operators. Recall that a (generally unbounded) selfadjoint operator X on H is affiliated with A if all the spectral projections of X belong to A. The elements of Asa will be regarded as (possibly) unbounded random variables. The set A is actually an algebra, as shown by Murray and von Neumann [15], and the usual problems concerning domains of definition are settled once for all. The distribution µT of an element T ∈ Asa is the unique probability measure on R satisfying the equality τ (u(T )) = Z R u(λ) µT (dλ) for every bounded Borel function u on R. A family (Tj)j∈I of elements of T ∈ Asa is said to be free if for all bounded continuous functions u1, u2, . . . , un on R we have τ (u1(Tj1)u2(Tj2) . . . un(Tjn)) = 0 whenever τ (ul(Tjl)) = 0, l = 1, . . . , n, for every choice of alternating indices j1, j2, . . . , jn. FREENESS OF LINEAR AND QUADRATIC FORMS IN VON NEUMANN ALGEBRAS. 3 Denote by µwm,r the semicircle distribution with density 2 where m ∈ R, r ∈ R+ and a+ := max{a, 0} for a ∈ R. This distribution plays the role of Gaussian one, when independence is replaced by freeness. πr2p(r2 − (x − m)2)+, The main aim of this note is to prove the following characterization theorem. Theorem 2.1. Let T1, T2, . . . , Tn be free identically distributed random variables with zero expectations, τ (Tj) = 0, and τ (T 2 j ) < ∞ in W ∗-probability space (A, τ ). Let A = (aij) ∈ Mn(R) be an n×n symmetric real matrix and b = t( b1, b2, . . . , bn) ∈ Rn be an n-dimensional vector satisfying the conditions Ab = 0 and n Xj=1 bm j ajj 6= 0 for m ∈ N. (2.1) Then the linear form L = Pn free if and only if T1 has semicircle distribution. j=1 bjTj and the quadratic form Q = Pn j,k ajkTjTk are Corollary 2.2. Let T1, T2, . . . , Tn be free identically distributed random variables j ) < ∞ in W ∗-probability space (A, τ ). with zero expectations, τ (Tj) = 0, and τ (T 2 Then the sample mean T = 1 j=1(Tj − T )2 are free if and only if T1 has semicircle distribution. j=1 Tj and the sample variance V = 1 n Pn n Pn Theorem 2.1 and Corollary 2.2 for bounded free identically distributed random variables under the assumptions that A is non-negative definite and b is non- negative was proved by Hiwatashi, Nagisa and Yoshida [8]. A more general ver- sion of Theorem 2.1 for bounded free identically distributed random variables was proved by the last author in [12]. Therefore we only need to prove the "only if" part of Theorem 2.1. In order to do this we establish that the freeness of L and Q im- plies that the distribution of T1 has moments of all order, i.e., τ (T1k) < ∞, k ∈ N, where T = (T ∗T )1/2. Namely, we prove the following result. Theorem 2.3. Let T1, T2, . . . , Tn be free identically distributed random variables in W ∗-probability space (A, τ ) such that τ (T 2 j ) < ∞. We consider the linear form j,k ajkTjTk with real coefficients bj j=1 bjTj and the quadratic form Q = Pn L = Pn and ajk such that bjajj 6= 0 for some j ∈ {1, 2, . . . , n}. (2.2) If the forms L and Q are free, then τ (T1k) < ∞, k = 1, 2, . . . . In particular, we infer from this result that under very weak assumptions free- ness of linear and quadratic forms in noncommutative random variables from a tra- cial W ∗-probability space automatically implies finiteness of all moments. 4 G. P. CHISTYAKOV, F. G OTZE, AND F. LEHNER 3. Auxiliary results. Measurable operators and integral for a trace We fix a faithful finite normal trace τ on a finite von Neumann algebra A. By A we denote the completion of A with respect to τ -measure topology. We denote A+ = {a∗a : a ∈ A} as well. The function τ on A+ enjoys the following properties (see [18], p. 176): τ (a + b) = τ (a) + τ (b), a, b ∈ A+, τ (λa) = λ τ (a), λ ≥ 0; τ (x∗x) = τ (xx∗), x ∈ A. For 1 ≤ p < ∞, set kxkp = τ (xp)1/p, x ∈ A; Lp(A, τ ) = {x ∈ A : kxkp < ∞}. Then Lp(A, τ ) is a Banach space in which A ∩ Lp(A, τ ) is dense. Furthermore, Lp(A, τ ) is a two-sided operator ideal and kaxkp ≤ kak kxkp , kxakp ≤ kak kxkp (3.1) for each a ∈ A, x ∈ Lp(A, τ ). If 1/p1 + · · · + 1/pn = 1 and pj > 1, j = 1, . . . , n, then the product of Lp1(A, τ ), . . . , Lpn(A, τ ) coincides with L1(A, τ ) and we have the Holder inequal- ity: τ (x1x2 · · · xn) ≤ kx1kp1 xn ∈ Lpn(A, τ ). (3.2) Since x1x2 · · · xn admits a representation x1x2 · · · xn = u x1x2 · · · xn, where x1 ∈ Lp1(A, τ ), . . . , · · · kxnkpn , kx2kp2 u ∈ A is a partial isometry, we have, using (3.1) and (3.2), kx1x2 · · · xnk1 = τ (x1x2 · · · xn) = τ (u∗x1x2 · · · xn) ≤ ku∗x1kp1 kx2kp2 · · · kxnkpn = kx1kp1 ≤ ku∗k∞ kx1kp1 kx2kp2 · · · kxnkpn kx2kp2 · · · kxnkpn . (3.3) In later reference we state the noncommutative Minkowski inequality kx1 + · · · + xnkp ≤ kx1kp + · · · + kxnkp (3.4) for 1 ≤ p < ∞. 4. Auxiliary analytic results Denote by M the family of all Borel probability measures on the real line R. Let T1 and T2 are free random variables with distributions µ1 and µ2 from M, respectively. Following Bercovici and Voiculescu [4] we define the additive free convolution µ1 ⊞ µ2 as the distribution of T1 + T2. Let M+ be the set of probability measures µ on R+ = [0, +∞) such that µ({0}) < 1. Fix probability measures µ1, µ2 ∈ M+ and fix random variables Tj such that . 2 T1T 1/2 their disributions µTj = µj. Following [4] we set µ1 ⊠µ2 = µT 1/2 = µT 1/2 1 T2T 1/2 1 2 FREENESS OF LINEAR AND QUADRATIC FORMS IN VON NEUMANN ALGEBRAS. 5 Define, following Voiculescu [21], the ψµ-function of a probability measure µ ∈ M+, by ψµ(z) = Z zξ 1 − zξ µ(dξ) R+ (4.1) for z ∈ C \ R+. The measure µ is completely determined by ψµ. Note that ψµ : C\R+ → C is an analytic function such that ψµ(¯z) = ψµ(z), and z(ψµ(z)+1) ∈ C+ for z ∈ C+. Consider the function Kµ(z) := ψµ(z)/(1 + ψµ(z)), z ∈ C \ R+. (4.2) It is easy to see that Kµ(z) ∈ K, where K is the subclass of N of functions f such that f (z) is analytic and nonpositive on the negative real axis, and f (−x) → 0 as x ↓ 0. This subclass of N was described by M. Krein [11], therefore we denote it by K. Theorem 4.1. There exist two uniquely determined functions Z1(z) and Z2(z) in the Krein class K such that Z1(z) Z2(z) = zKµ1(Z1(z)) and Kµ1(Z1(z)) = Kµ2(Z2(z)), z ∈ C+. (4.3) Moreover Kµ1⊠µ2 = Kµ1(Z1(z)). This result was proved by Biane [5]. Belinschi and Bercovici [2] and Chistyakov and Gotze [6] proved this theorem by purely analytic methods. For a probability measure µ ∈ M, define its absolute moment of order α and for µ ∈ M+, define ρα(µ) := Z R xα µ(dx) mα(µ) := Z R+ xα µ(dx), where α ≥ 0. We now characterize existence of moments in terms of Taylor expansions of the Krein function. A similar result for the R-transform was obtained by Benaych- Georges [3] and applied to additive free infinite divisibility. Proposition 4.2. Let µ ∈ M+. In order that mp(µ) < ∞ for some p ∈ N it is necessary and sufficient that the Krein function (4.2) admits the expansion 1 x Kµ(−x) = −r1(µ)+r2(µ)x+· · ·+(−1)prp(µ)xp−1+o(xp−1) for x > 0 and x ↓ 0, (4.4) with some real coefficients r1(µ), r2(µ), . . . , rp(µ). 6 G. P. CHISTYAKOV, F. G OTZE, AND F. LEHNER The coefficients r1(µ), r2(µ), . . . coincide with the so-called boolean cumulants, see Speicher and Woroudi [17]. Note that rk(µ) depends on m1(µ), m2(µ), . . . , mk(µ) only. Proof. Necessity. Assume that mp(µ) < ∞. Then we see that, for x > 0, ψµ(−x) + 1 = = where 1 R+ µ(du) 1 x + u x Z x(cid:16)x − m1(µ)x2 + · · · + (−1)pmp(µ)xp+1 + (−1)p+1xp+1Z 1 R+ up+1µ(du) x + u (cid:17), 1 up+1 µ(du) 1 x + u Z R+ → 0 as x → 0. (4.5) (4.6) By (4.5), we have the relation, for the same x, Kµ(−x) = ψµ(−x) ψµ(−x) + 1 = ψµ(−x) − ψ2 µ(−x) + · · · + (−1)p−1ψp µ(−x) + O(xp+1) = −r1(µ)x + r2(µ)x2 + · · · + (−1)prp(µ)xp + (−1)p+1xpZ R+ up+1µ(du) 1 x + u + O(xp+1). (4.7) Now (4.6) and (4.7) imply the necessity of the assumptions of Proposition 4.2. Sufficiency. Note that, for positive sufficiently small 0 < x ≤ x0, 1 x ψµ(−x) + 1 Z ψµ(−x) + 1 Z u µ(du) 1 + ux Kµ(−x) = µ(du) ≥ u 2 ≥ 1 1 − R+ [0,1/x) 1 2 Z [0,1/x) u µ(du). By (4.4), we conclude that m1(µ) < ∞. Assume that the inequality mk(µ) < ∞ holds for any k ≤ p − 1. From (4.5) we obtain the formula ψµ(−x) = −m1(µ)x + · · · + (−1)kmk(µ)xk + (−1)k+1xkZ R+ uk+1 µ(du) 1 x + u , x > 0. Using this formula and (4.7) with p = k we note that, for small x > 0, (−1)k+1(cid:16)Kµ(−x) + r1(µ)x − r2(µ)x2 − · · · − (−1)krk(µ)xk(cid:17) = xk Z R+ uk+1 µ(du) 1 x + u + O(xk+1) ≥ 1 2 xk+1 Z [0,1/x) uk+1 µ(du) + O(xk+1). (4.8) FREENESS OF LINEAR AND QUADRATIC FORMS IN VON NEUMANN ALGEBRAS. 7 On the other hand, by (4.4) with p = k + 1, we have, for small x > 0, Kµ(−x) + r1(µ)x − r2(µ)x2 − · · · − (−1)krk(µ)xk = (−1)k+1rk+1(µ)xk+1 + o(xk+1). Therefore we easily conclude from (4.8) that mk+1(µ) < ∞. Thus induction may be used and the sufficiency of the assumptions of Proposition 4.2 is also proved. (cid:3) Speicher and Woroudi [17] indicated a universal formula for calculation of boolean cumulants rk(µ). For example r1(µ) = m1(µ), r2(µ) = m2(µ) − m2 r4(µ) = m4(µ) − m2 2(µ) − 2m1(µ)m3(µ) + 3m1(µ)2m2(µ) − m4 1(µ). 1(µ), r3(µ) = m3(µ) − 2m1(µ)m2(µ) + m3 1(µ), (4.9) Proposition 4.3. Let µ ∈ M+ and α ∈ (0, 1). Then 1 2(cid:16)mα(µ) − Z (0,1) uα µ(du)(cid:17) ≤ −(1 − α) Z (0,1] Kµ(−x) dx x1+α ≤ c(µ)α−1mα(µ), (4.10) where c(µ) := 1/RR+ µ(du) 1+u . Moreover, mα(µ) < ∞ with α ∈ (0, 1) if and only if − Z (0,1] Kµ(−x) dx x1+α < ∞. (4.11) Proof. In the first step we shall prove the right-hand side of (4.10). Without loss of generality we assume that mα(µ) < ∞. Since 1 + ψµ(−x) ≥ 1 c(µ) for x ∈ (0, 1], we have − Kµ(−x) ≤ −c(µ)ψµ(−x) ≤ c(µ)(cid:16)x Z [0,1/x) u µ(du) + µ([1/x, ∞))(cid:17), x ∈ (0, 1]. Taking into account that mα(µ) = α RR+ x−α Z c(µ) Z Kµ(−x) dx ≤ Z x1+α 1 − xα−1µ([x, ∞)) dx, we finally obtain u µ(du) dx + Z x−1−αµ([1/x, ∞)) dx (0,1] (0,1] [0,1/x) (0,1] x−α dx µ(du) + 1 1 − α Z [0,1) u µ(du) + mα(µ) α ≤ Z u Z [1,∞) (0,1/u] ≤ mα(µ) α(1 − α) . (4.12) 8 G. P. CHISTYAKOV, F. G OTZE, AND F. LEHNER Let us prove the left-hand side of (4.10), assuming without loss of generality that Kµ(−x) dx x1+α < ∞. Since, for x > 0, R(0,1] − Kµ(−x) ≥ −ψµ(−x) ≥ 1 2 x Z [0,1/x) u µ(du), we have the lower bound 1 Kµ(−x) dx − Z ≥ x1+α (0,1] = u µ(du) dx ≥ x−α Z 2 Z 2(1 − α)(cid:16)mα(µ) − Z (0,1] 1 [0,1/x) uα µ(du)(cid:17). (0,1) 1 2 Z u Z [1,∞) (0,1/u] x−α dx µ(du) The inequalities (4.10) follow from (4.12) and (4.13). Finally, statement (4.11) is a direct consequence of (4.10). (4.13) (cid:3) Lemma 4.4. Let µ1 and µ2 be probability measures from M+ such that mp(µ1) < ∞ and mp(µ2) < ∞ for some p ∈ N. Then mp(µ1 ⊠ µ2) < ∞. Proof. By Theorem 4.1, there exist Z1(z) and Z2(z) from the class K such that (4.3) holds. By Proposition 4.2, Kµj (−x) = −r1(µj)x+r2(µj)x2+· · ·+(−1)prp(µj)xp+o(xp) for x > 0 and x ↓ 0, where r1(µ), r2(µ), . . . , rp(µ) are the boolean cumulants. Hence j (−x) + · · · + rp(µj)Z p Kµj (Zj(−x)) = r1(µj)Zj(−x) + r2(µj)Z 2 j (−x)) (4.15) for x > 0, x ↓ 0 and j = 1, 2. From the first relation of (4.3) we conclude that, for the same x, j (−x) + o(Z p (4.14) Zj(−x) = −r1(µk)x + o(x), Let us assume that there exist real numbers t(j) such that j, k = 1, 2, 1 , t(j) 2 , . . . , t(j) j 6= k. m , j = 1, 2, m ≤ p − 1, Zj(−x) = t(j) 1 x + t(j) 2 x2 + · · · + t(j) m xm + o(xm) for x > 0 and x ↓ 0. (4.16) Then from the first relation of (4.3) and from (4.15), (4.16) we conclude that Zj(−x) = −r1(µk)x + r2(µk)xZk(−x) + · · · + (−1)prp(µk)xZ p−1 (−x) + o(xZ p−1 k (−x)) j = t(j) 1 x + · · · + t(j) m+1xm+1 + o(xm+1) (4.17) for real numbers t(j) m+1, j = 1, 2, and for x > 0, x ↓ 0. Thus, induc- tion may be used and (4.17) holds for m = p. Since Kµj (Zj(−x)) = Kµ1⊠µ2(−x), x > 2 , . . . , t(j) 1 , t(j) m xm + t(j) m , t(j) FREENESS OF LINEAR AND QUADRATIC FORMS IN VON NEUMANN ALGEBRAS. 9 0, we easily obtain the assertion of the lemma from (4.15), (4.17) with m = p and from Proposition 4.2. (cid:3) Lemma 4.5. Let µ1 and µ2 be probability measures from M+ such that mα(µ1) < ∞ and mβ(µ2) < ∞, where 0 < α, β ≤ 1. Then mαβ(µ1 ⊠ µ2) < ∞. Proof. If the assumptions of the lemma hold with α = β = 1 the assertion of the lemma follows from Lemma 4.4. Consider the case, where the assumptions of the lemma hold with 0 < α < β and β = 1. By Theorem 4.1, there exist Z1(z) and Z2(z) from the class K such that (4.3) holds. By Proposition 4.2 and (4.9), Kµ1(−x) = −m1(µ1)x(1 + o(1)) for positive x such that x ↓ 0. Hence Kµ1(Z1(−x)) = m1(µ1)Z1(−x)(1 + o(1)) for the same x and, by (4.3), we have From this relation and Proposition 4.3 we conclude that Z2(−x) = −m1(µ1)x(1 + o(1)). − Z x−1−αKµ2(Z2(−x)) dx ≤ − Z (0,x0] (0,x0] x−1−αKµ2(−2m1(µ)x) dx < ∞, where x0 is a sufficiently small positive constant. Since Kµ2(Z2(−x)) = Kµ1⊠µ2(−x), by Proposition 4.3, we arrive at the assertion of the lemma for α ∈ (0, 1) and β = 1. Consider the case, where the assumptions of the lemma hold with 0 < α, β < 1. As above, by Theorem 4.1, there exist Z1(z) and Z2(z) from the class K such that (4.3) holds. By Proposition 4.3, we have x−1−αKµ1(−x) dx ≤ −Z (0,1] c(µ1)mα(µ1) α(1 − α) and −Z (0,1] x−1−βKµ2(−x) dx ≤ c(µ2)mβ(µ2) β(1 − β) , (4.18) where c(µj), j = 1, 2, are constants defined in Proposition 4.3. We obtain from (4.3) the relation, for x > 0, Kµ1(Z1(−x)) = Kµ2(Z2(−x)). (4.19) Recalling (4.2) we deduce from (4.19) that, for x ∈ (0, x0] with sufficiently small x0 > 0, − 1 2 ψµ1(Z1(−x)) ≤ −ψµ2(Z2(−x)) ≤ −Z2(−x) Z u µ2(du) + µ2([−1/Z2(−x), ∞)) ≤ 2mβ(µ2)(−Z2(−x))β. (4.20) [0,−1/Z2(−x)) 10 G. P. CHISTYAKOV, F. G OTZE, AND F. LEHNER Since, by (4.3), − Z2(−x) = xKµ1(Z1(−x))/Z1(−x)) ≤ 2xψµ1(Z1(−x))/Z1(−x)), x ∈ (0, x0], we get from (4.20) the bound − 1 2 (ψµ1(Z1(−x))/Z1(−x))1−βZ1(−x) ≤ 21+βmβ(µ2)xβ, x ∈ (0, x0]. (4.21) On the other hand f (x) := ψµ1(Z1(−x))/Z1(−x) is a positive strictly monotone function such that limx→0 f (x) is not equal to 0. Hence we obtain from (4.21) that − Z1(−x) ≤ c(µ1, µ2)mβ(µ2)xβ, x ∈ (0, x0], (4.22) where c(µ1, µ2) is a positive constant depending on µ1 and µ2 only. It remains to note, using (4.18), that − Z x−1−αβKµ1(Z1(−x)) dx ≤ − Z (0,x0] (0,x0] ≤ c(µ1, µ2, α, β)(mβ(µ2))α < ∞, x−1−αβKµ1(−c(µ1, µ2)mβ(µ2)xβ) dx where c(µ1, µ2, α, β) is a positive constant depending on µ1, µ2, α, and β only. By Proposition 4.3, the lemma is proved. (cid:3) Proposition 4.6. Let T and S be free random variables such that mp/2(µT 2 ⊠ µS2) < ∞ with some p > 0. Then T S ∈ Lp(A, τ ) and τ (T Sp) = mp/2(µT 2 ⊠ µS2) < ∞. Proof. Since the distribution of T S2 T is µT 2 ⊠ µS2, we have τ ((T S2 T )p/2) = mp/2(µT 2 ⊠ µS2) < ∞. Using the polar decomposition T = u T , where u ∈ A is a unitary element, we obtain τ (T Sp) = τ ((u T S2 T u∗)p/2) = τ (u(T S2 T )p/2u∗) = τ ((T S2 T )p/2) = mp/2(µT 2 ⊠ µS2) < ∞. The proposition is proved. (cid:3) Lemma 4.7. Let µ1, µ2, . . . , µn be probability measures from M such that ρ1(µ1 ⊞ µ2 ⊞ · · · ⊞ µn) < ∞. Then ρ1(µ1) < ∞, . . . , ρ1(µn) < ∞. Proof. It suffices to prove the case n = 2, the general case follows by induction. Let T1, T2 be free random variables with distributions µ1, µ2, respectively, such that τ (T1 + T2) < ∞. Without loss of generality we may assume that the dis- tributions µ1, µ2 are not point masses. For, if T is any measurable operator and λ is any constant, by a simple application of the Minkowski inequality we have kT k1 < ∞ if and only if kT + λIk1 < ∞. By the same argument we can ensure that the spectra of each T1 and T2 are not contained in either the positive or the negative real axis. FREENESS OF LINEAR AND QUADRATIC FORMS IN VON NEUMANN ALGEBRAS. 11 Consider the projections p(t) T2 = eT2([0, t]), where t > 0 and eT1, eT2 are an A-valued spectral measures on R, which are countably additive in the weak∗ topology on A. By our assumptions, these projections are nonzero for sufficiently large t. Set T (t) T2 for j = 1, 2 and note that by (3.1), T1 = eT1([0, t]) and p(t) j = p(t) T1 Tjp(t) 2 (cid:12)(cid:12)) ≤ τ (T1 + T2) < ∞. 1 + T (t) τ ((cid:12)(cid:12)T (t) (4.23) On the other hand, since the random variables p(t) using freeness of the corresponding random variables, we have τ (T (t) τ (p(t) T2 ) and we obtain from (4.23) that T1 T1 and T2p(t) T1 T1) τ (p(t) T1 ) τ (T2p(t) T2 are bounded, 2 ) = 1 + T (t) T2 ) + τ (p(t) T2 ) Z τ (p(t) u µ1(du) + τ (p(t) T1 ) Z u µ2(du) ≤ τ (T1 + T2) < ∞ [0,t] [0,t] and in the limit t → ∞ this implies both u µ1(du) < ∞ Z [0,∞) Z [0,∞) u µ2(du) < ∞. In the same way we prove that Z (−∞,0) u µ1(du) < ∞ Z (−∞,0) u µ2(du) < ∞. u µj(du) < ∞ for j = 1, 2. (cid:3) Thus we have proved that ρ1(µj) = R(−∞,∞) Proposition 4.8. Let {Tj}k j=1 be a family of free elements in Asa such that τ (Tjs) < ∞ for all s ∈ N and j = 1, 2, . . . , k. Then τ (Tj1Tj2 · · · Tjn) = 0 whenever τ (Tjl) = 0, l = 1, 2, . . . , n, and all alternating sequences j1, j2, . . . , jn of 1's, 2's, and k's, i.e., j1 6= j2 6= · · · 6= jn. This proposition is well-known. In particular one can obtain a proof using arguments of the paper by Bercovici and Voiculescu [4]. 5. Proofs of the main results In order to prove Theorems 2.1 and 2.3 we need the following lemma. Lemma 5.1. Let (Tj)j∈I be free random variables in W ∗-probability space (A, τ ) such that τ (Tjd) < ∞ for some d ∈ {2, 3, . . . } and any j ∈ I. Then for any choice of indices k1 6= k2 6= · · · 6= ks, s ≥ 2 and any choice of strictly positive integers n1, n2, . . . , ns such that n1 + n2 + · · · + ns = d + 1. k1 T n2 k2 · · · T ns τ ((cid:12)(cid:12)T n1 ks (cid:12)(cid:12)) < ∞ (5.1) 12 G. P. CHISTYAKOV, F. G OTZE, AND F. LEHNER Proof. We may assume that the distributions of Ti are not point masses, otherwise the concerning operator is bounded and the conclusion is trivial. Let d = 2p, p ∈ N. Then we write T n1 k2 · · · T ns ks . By Lemma 4.4 and by Proposition 4.6, we have (Tk1Tk2)T n2−1 ks = T n1−1 k1 T n2 · · · T ns k1 k2 τ (Tk1Tk2d) = mp(µT 2 k1 ⊠ µT 2 k2 ) < ∞. (5.2) Applying the Holder inequality (3.3), we easily obtain, using (5.2), k1 τ ((cid:12)(cid:12)T n1−1 (Tk1Tk2)T n2−1 k2 · · · T ns ks (cid:12)(cid:12)) ≤ (τ (Tk1d))(n1−1)/d(τ (Tk1Tk2d))1/d(τ (Tk2d))(n2−1)/d · · · (τ (Tksd))ns/d < ∞. (5.3) Let d = 2p + 1, p ∈ N. Consider first the case s = 2, i.e., terms of the form T n1 k1 T n2 k2 with k1 6= k2 and n1 + n2 = d + 1. By the assumptions of the lemma, we see that md/(2n1)(µT 2n1 k1 ) < ∞ and md/(2n2)(µT ) < ∞. 2n2 k2 If n1 = n2 = p + 1, then d 2n1 = d 2n2 Lemma 4.5, we conclude that m1/2(µT 2n1 k1 tion 4.6, we have = 1 − 1 ⊠ µT 2n2 k2 2p+2 and (cid:16)1 − 1 2. By ) < ∞ and then, by Proposi- 2p+2(cid:17)2 > 1 k1 T n2 τ ((cid:12)(cid:12)T n1 k2 (cid:12)(cid:12)) = m1/2(µT 2n1 k1 ⊠ µT 2n2 k2 ) < ∞. (5.4) If 1 ≤ n1 < p + 1 and p + 1 < n2 ≤ 2p + 1, then d 2n1 < ) < ∞ and, by Proposition 4.6, we have > 1 and 1 2 ≤ d 2n2 1. By Lemma 4.5, m1/2(µT 2n1 k1 the relation (5.4) again. ⊠ µT 2n2 k2 Consider now terms of the form T n1 n3 = d. Let for definiteness n1 ≤ p and n3 ≥ p + 1. Since m1(µT 2 k2 ⊠ µT 2n3 md/(2n3)(µT 2n3 k3 k3 ) < ∞. Then, ) < ∞, we get, by Lemma 4.5, that md/(2n3)(µT 2 k2 ⊠ µT k3 with k1 6= k2, k3 6= k2 and n1 + ) < ∞ and ) < ∞. By k1 Tk2T n3 2n3 k3 using the Holder inequality (3.3), we obtain d/n3) = md/(2n3)(µT 2 Proposition 4.6, we see that τ ((cid:12)(cid:12)Tk2T n3 k3 (cid:12)(cid:12) d/n1))n1/d(τ ((cid:12)(cid:12)Tk2T n3 k3 (cid:12)(cid:12)) ≤ (τ ((cid:12)(cid:12)T n1 k1 (cid:12)(cid:12) k3 (cid:12)(cid:12) k1 T n2 Now consider a term of the form T n1 τ ((cid:12)(cid:12)T n1 k1 Tk2T n3 k2 T n3 d + 1, n1 ≥ 1, n2 ≥ 2, n3 ≥ 1. Rewrite it in the form k2 k3 with k1 6= k2 6= k3 and n1+n2+n3 = d/n3))n3/d < ∞. (5.5) T n1 k1 T n2 k2 T n3 k3 = T n1−1 k1 (Tk1Tk2)T n2−2 k2 (Tk2Tk3)T n3−1 k3 and note that as in the proof of (5.2) we have τ (Tk1Tk2d−1) = τ (Tk1Tk22p) = mp(µT 2 k1 ⊠ µT 2 k2 ) < ∞ and similarly τ (Tk2Tk3d−1) < ∞. (5.6) (5.7) FREENESS OF LINEAR AND QUADRATIC FORMS IN VON NEUMANN ALGEBRAS. 13 Now in view of (5.6) and (5.7), we deduce with the help of the Holder inequality (3.3) k1 T n2 k2 T n3 τ ((cid:12)(cid:12)T n1 n1−1 k3 (cid:12)(cid:12)) ≤ (τ (Tk1d−1)) × (τ (Tk2d−1)) d−1 (τ (Tk1Tk2d−1)) d−1 (τ (Tk2Tk3d−1)) n2−2 1 d−1 1 d−1 (τ (T3d−1)) n3−1 d−1 < ∞. (5.8) Now for any positive integers k1 6= k2 6= · · · 6= ks, s ≥ 4, and any positive integers n1, n2, . . . , ns such that n1 + n2 + · · · + ns = d + 1 we can write k2 T n3 k1 T n2 T n1 T n3−1 k3 Repeating the previous arguments, we easily obtain (Tk1Tk2)T n2−1 ks = T n1−1 k4 · · · T ns k3 T n4 k1 k2 (Tk3Tk4)T n4−1 k4 · · · T ns ks . k1 T n2 k2 T n3 k3 T n4 k4 · · · T ns τ ((cid:12)(cid:12)T n1 n1−1 ks (cid:12)(cid:12)) ≤ (τ (Tk1d−1)) × (τ (Tk3d−1)) × (τ (Tk5d−1)) n3−1 d−1 (τ (Tk1Tk2d−1)) d−1 (τ (Tk3Tk4d−1)) d−1 · · · (τ (Tksd−1)) d−1 (τ (Tk2d−1)) d−1 (τ (Tk4d−1)) ns d−1 < ∞. n5 1 1 n2−1 d−1 n4−1 d−1 (5.9) The assertion of the lemma follows from (5.3) -- (5.5), (5.8) and (5.9). (cid:3) Proof of Theorem 2.3. We need to prove that under the assumptions of Theorem 2.3 if the forms L and Q are free, then τ (T1s) < ∞ for all s ∈ N. Consider the free elements L and Q of the probability space (A, τ ). In the first step we shall prove that τ (T13) < ∞). Write the relation QL = Xj k ) + Xj6=k, k6=l j Tk + ajkbkTjT 2 j +Xj6=k ajkblTjTkTl. (ajjbkT 2 ajjbjT 3 (5.10) By the Minkowski inequality (3.4), we see that (τ (L2))1/2 ≤ X bj τ (Tj2)1/2 < ∞. Since, by (3.3), (τ (TjTk))2 ≤ τ (T 2 the Minkowski inequality (3.4) again, j )τ (T 2 k ) < ∞, j, k = 1, 2, . . . , n, we have, by τ (Q) ≤ n Xj,k ajk τ (TjTk) ≤ n Xj ajj τ (Tj2)+ n Xj6=k ajk (τ (Tj2))1/2 (τ (Tk2))1/2 < ∞. This means that L has finite second moment and Q has finite first moment. Since QL2 = QL2Q, we note that µQL2 = µQ2 ⊠µL2 and τ (QL) = m1/2(µQ2 ⊠ µL2). Noting that, m1/2(µQ2) < ∞ and m1(µL2) < ∞, by Lemma 4.5, we arrive at the inequality m1/2(µQL2) < ∞. Hence, by Proposition 4.6, τ (QL) < ∞. By Lemma 5.1, we have the following bounds τ ((cid:12)(cid:12)TkT 2 j (cid:12)(cid:12)) < ∞, τ ((cid:12)(cid:12)T 2 k Tj(cid:12)(cid:12)) < ∞, j 6= k, and τ (TjTkTl) < ∞, j 6= k 6= l. (5.11) 14 G. P. CHISTYAKOV, F. G OTZE, AND F. LEHNER Return to (5.10). Using the Minkowski inequality (3.4) and (5.11) we obtain from (5.10) that ajjbjT 3 τ ((cid:12)(cid:12)(cid:12)Xj j (cid:12)(cid:12)(cid:12) ) ≤ τ (QL) +Xj6=k + Xj6=k, k6=l bk(cid:0)ajj τ ((cid:12)(cid:12)T 2 j Tk(cid:12)(cid:12)) + ajk τ ((cid:12)(cid:12)TjT 2 k(cid:12)(cid:12))(cid:1) ajkbl τ (TjTkTl) < ∞. (5.12) By Lemma 4.7, we conclude from this bound that τ (T13) < ∞ as was to be proved. Now assume that τ (Tjd) < ∞ for d ≥ 3. We have, by the Minkowski inequality (3.4) that (τ (Ld))1/d ≤ Xj bj (τ (Tjd))1/d < ∞. In addition, for p = 3, 4, we have, by Lemma 4.5 and Proposition 4.6, τ (TjTkp/2) = mp/4(µT 2 j ⊠ µT 2 k ) < ∞. Therefore, for p = 3, 4, τ (Qp/2) ≤ n Xj ajj (τ (Tjp))2/p + n Xj6=k ajk (τ (TjTkp/2))2/p < ∞, if τ (Tjp) < ∞ for p = 3, 4, respectively. Let d = 3. In view of the inequalities m3/4(µQ2) < ∞ and m3/4(µL4) < ∞, by Lemma 4.5, we arrive at the inequality m9/16(µQL22) = m9/16(µQ2 ⊠ µL4) < ∞. Therefore, by Proposition 4.6, τ (QL2) < ∞. Let d ≥ 4. Since m1(µQ2) < ∞ and m1/2(µL2(d−1)) < ∞, by Lemma 4.5, we arrive at the inequality m1/2(µQ2 ⊠ µL2(d−1)) < ∞. Hence, by Proposition 4.6, τ ((cid:12)(cid:12)QLd−1(cid:12)(cid:12)) = m1/2(µQ2 ⊠ µL2(d−1)) < ∞. Consider the relation d+1 QLd−1 = Xj ajjbd−1 j T d+1 j + Xs=2 X αk1k2···ksT n1 k1 T n2 k2 · · · T ns ks , (5.13) where the summation in sum of the second summand on the right-hand side of (5.13) is taken over all positive integers k1 6= k2 6= · · · 6= ks such that kj = 1, 2, . . . , n, and any positive integers n1, n2, . . . , ns such that n1 + n2 + · · · + ns = d + 1, and αk1k2···ks are real coefficients. By Lemma 5.1, we see that, for the considered values of kj and nj, k1 T n2 k2 · · · T ns τ ((cid:12)(cid:12)T n1 ks (cid:12)(cid:12)) < ∞. (5.14) FREENESS OF LINEAR AND QUADRATIC FORMS IN VON NEUMANN ALGEBRAS. 15 Using the Minkowski inequality (3.4) and (5.14) we obtain from (5.13) that ajjbd−1 j T d+1 j τ ((cid:12)(cid:12)(cid:12)Xj d+1 (cid:12)(cid:12)(cid:12) ) ≤ τ ((cid:12)(cid:12)QLd−1(cid:12)(cid:12) + Xs=2 X αk1k2···ks τ ((cid:12)(cid:12)T n1 k1 T n2 k2 · · · T ns ks (cid:12)(cid:12)) < ∞. Now, by Lemma 4.7, we conclude that τ (T1d+1) < ∞. Thus, induction may be used and the theorem is proved. (cid:3) Proof of Theorem 2.1. Let the free random variables T1, T2, . . . , Tn satisfy the assumptions of Theorem 2.1. Then, as it is easy to see, the free random variables T1, T2, . . . , Tn satisfy the assumptions of Theorem 2.3 as well. By this theorem τ (Tjk) < ∞, k ∈ N, j = 1, 2, . . . , n. Noting that the arguments of the paper [12] hold for free identically distributed random variables with finite moments of all order, we obtain the desired result repeating step by step these arguments (see [12], p. 416 -- 418). (cid:3) References [1] Akhiezer, N. I. The classical moment problem and some related questions in analysis. Hafner, New York (1965). [2] Belinschi, S. T. and Bercovici, H. A new approach to subordination results in free probability. J. Anal. Math. 101, 357 -- 365 (2007). [3] Benaych-Georges, F., Taylor expansions of R-transforms: application to supports and mo- ments, Indiana Univ. Math. J. 55 (2006), no. 2, 465 -- 481. [4] Bercovici, H., and Voiculescu, D. Free convolution of measures with unbounded support. Indiana Univ. Math. J., 42, 733 -- 773 (1993). [5] Biane, Ph. Processes with free increments. Math. Z., 227, 143 -- 174 (1998). [6] Chistyakov, G. P. and Gotze, F. The arithmetic of distributions in free probability theory. ArXiv: math/0508245. [7] Hiai, F. and Petz, D. The semicircle law, free random variables and entropy. American Mathematical Society, (2000). [8] Hiwatashi, O., Nagisa, M. and Yoshida, H. The characterizations of a semicircle law by the certain freeness in a C ∗-probability space. Probab. Theory Relat. Fields, 113, 115 -- 133 (1999). [9] Kagan, A. A., Linnik, Yu. V. and Rao, C. R. Characterization problems in mathematical statistics. John Wiley & Sons, New York, London, Sydney, Toronto (1973). [10] Kawata, T. and Sakamoto, H. On the characterization of the independence of the sample mean and the sample variance. J. Math. Soc. Japan. 1, 111-115 (1949). [11] Krein, M. G. and Nudel'man, A. A. The Markov moment problem and extremal problem. Amer. Math. Soc., Providence, Rhode Island (1977). [12] Lehner, F., Cumulants in noncommutative probability theory II. Probab. Theory Relat. Fields, 127, 407 -- 422 (2003). [13] Lehner, F., Cumulants in noncommutative probability theory I. Noncommutative exchange- ability systems. Probab. Theory Relat. Fields, 248, 67 -- 100 (2004). [14] Maassen, H., Addition of freely independent random variables. J. Funct. Anal. 106, no. 2, 409 -- 438 (1992). 16 G. P. CHISTYAKOV, F. G OTZE, AND F. LEHNER [15] Murray, H. and von Neumann, J. On rings of operators. Ann. of Math. (2) 37, no. 1, 116 -- 229 (1936). [16] Nica, A. and Speicher, R. Lectures on the combinatorics of free probability. Cambridge Uni- versity Press, (2006). [17] Speicher, R. and Woroudi, R. Boolean convolutions. Free probability theory (Waterloo, ON, 1995), 267 -- 279, Fields Inst. Commun., 12, Amer. Math. Soc., Providence, RI, 1997. [18] Takesaki, M. Theory of operator algebras II. Springer-Verlag Berlin Heidelberg New York (2003). [19] Zinger, A. A. On independent samples from a normal population. Uspekhi Matem. Nauk, 6, 172 (1951). [20] Voiculescu, D. V. Addition of certain noncommuting random variables. J. Funct. Anal., 66, 323 -- 346 (1986). [21] Voiculescu, D. V. Multiplication of certain noncommuting random variables. J. Operator Theory, 18, 223 -- 235 (1987). [22] Voiculesku, D. V., Dykema, K., and Nica, A. Free random variables. CRM Monograph Series, No 1, A.M.S., Providence, RI (1992). Gennadii Chistyakov Fakultat fur Mathematik Universitat Bielefeld Postfach 100131 33501 Bielefeld Germany E-mail address: [email protected], [email protected] Friedrich Gotze Fakultat fur Mathematik Universitat Bielefeld Postfach 100131 33501 Bielefeld Germany E-mail address: [email protected] Franz Lehner Institut fur Mathematische Strukturtheorie Technische Universitat Graz Steyrergasse 30, A-8010 Graz Austria E-mail address: [email protected]
1906.01889	2	1906	2019-06-26T09:19:47	Quantization of subgroups of the affine group	[ "math.OA", "math.QA" ]	Consider a locally compact group $G=Q\ltimes V$ such that $V$ is abelian and the action of $Q$ on the dual abelian group $\hat V$ has a free orbit of full measure. We show that such a group $G$ can be quantized in three equivalent ways: (1) by reflecting across the Galois object defined by the canonical irreducible representation of $G$ on $L^2(V)$; (2) by twisting the coproduct on the group von Neumann algebra of $G$ by a dual $2$-cocycle obtained from the $G$-equivariant Kohn-Nirenberg quantization of $V\times\hat V$; (3) by considering the bicrossed product defined by a matched pair of subgroups of $Q\ltimes\hat V$ both isomorphic to $Q$. In the simplest case of the $ax+b$ group over the reals, the dual cocycle in (2) is an analytic analogue of the Jordanian twist. It was first found by Stachura using different ideas. The equivalence of approaches (2) and (3) in this case implies that the quantum $ax+b$ group of Baaj-Skandalis is isomorphic to the quantum group defined by Stachura. Along the way we prove a number of results for arbitrary locally compact groups $G$. Using recent results of De Commer we show that a class of $G$-Galois objects is parametrized by certain cohomology classes in $H^2(G;\mathbb T)$. This extends results of Wassermann and Davydov in the finite group case. A new phenomenon is that already the unit class in $H^2(G;\mathbb T)$ can correspond to a nontrivial Galois object. Specifically, we show that any nontrivial locally compact group $G$ with group von Neumann algebra a factor of type I admits a canonical cohomology class of dual $2$-cocycles such that the corresponding quantization of $G$ is neither commutative nor cocommutative.	math.OA	math	QUANTIZATION OF SUBGROUPS OF THE AFFINE GROUP P. BIELIAVSKY, V. GAYRAL, S. NESHVEYEV, AND L. TUSET Abstract. Consider a locally compact group G = Q ⋉ V such that V is abelian and the action of Q on the dual abelian group V has a free orbit of full measure. We show that such a group G can be quantized in three equivalent ways: (1) by reflecting across the Galois object defined by the canonical irreducible representation of G on L2(V ); (2) by twisting the coproduct on the group von Neumann algebra of G by a dual 2-cocycle obtained from the G-equivariant Kohn -- Nirenberg quantization of V × V ; (3) by considering the bicrossed product defined by a matched pair of subgroups of Q ⋉ V both isomorphic to Q. In the simplest case of the ax + b group over the reals, the dual cocycle in (2) is an analytic It was first found by Stachura using different ideas. The analogue of the Jordanian twist. equivalence of approaches (2) and (3) in this case implies that the quantum ax + b group of Baaj -- Skandalis is isomorphic to the quantum group defined by Stachura. Along the way we prove a number of results for arbitrary locally compact groups G. Using recent results of De Commer we show that a class of G-Galois objects is parametrized by certain cohomology classes in H 2(G; T). This extends results of Wassermann and Davydov in the finite group case. A new phenomenon is that already the unit class in H 2(G; T) can correspond to a nontrivial Galois object. Specifically, we show that any nontrivial locally compact group G with group von Neumann algebra a factor of type I admits a canonical cohomology class of dual 2-cocycles such that the corresponding quantization of G is neither commutative nor cocommutative. 9 1 0 2 n u J 6 2 ] . A O h t a m [ 2 v 9 8 8 1 0 . 6 0 9 1 : v i X r a Contents Introduction 1. Preliminaries 2. Galois objects and dual cocycles 2.1. Projective representations and Galois objects 2.2. Dual cocycles 2.3. Dual cocycles defined by genuine representations 2.4. Examples: subgroups of the affine group 3. Kohn -- Nirenberg quantization 3.1. Kohn -- Nirenberg quantization of V × V 3.2. Dual cocycles on subgroups of Aff(V ) 3.3. Identification of the Galois objects 3.4. Deformation of the trivial cocycle 3.5. Multiplicative unitaries 3.6. Stachura's dual cocycle 4. Bicrossed product construction References 1 3 4 4 7 11 12 15 15 17 23 26 27 31 32 34 Although the problem of quantization of Lie bialgebras was solved in full generality more than 20 years ago by Etingof and Kazhdan [15], the list of noncompact Poisson -- Lie groups Introduction Date: June 5, 2019; minor changes June 25, 2019. 1 admitting nonformal (analytic) quantizations is still quite short. The difficulty lies not only in making sense of certain formal constructions, but in that there exist real obstacles in doing so. A famous example is the group SU (1, 1). At the Hopf ∗-algebraic level its quantization is well understood, but by a "no go" result of Woronowicz there is no way of making sense of it at the operator algebraic level [36]. As was first realized by Korogodsky [19] and then completed by Koelink and Kustermans [18], the right group to quantize in this case is the nonconnected group SU (1, 1) ⋊ Z/2Z, the normalizer of SU (1, 1) in SL(2, C). The present paper is motivated by an even easier example, the ax + b group G over the reals. Its Lie algebra g is generated by two elements x, y such that [x, y] = y. Consider the Lie bialgebra (g, δ), with the cobracket δ defined by the triangular r-matrix It can be explicitly quantized using the Jordanian twist r := x ⊗ y − y ⊗ x. Ω := exp{x ⊗ log(1 + hy)} ∈ (U g ⊗ U g)[[h]] found independently by Coll -- Gerstenhaber -- Giaquinto [8] and Ogievetsky [27].∗ This twist and its generalizations have been extensively studied, see, e.g., [20], [21]. It is particularly popular in physics literature, as it can be used to construct the κ-Minkowski space [6], which by [22] can also be obtained from a bicrossed product construction. If we want to make sense of Ω as a unitary operator on L2(G × G), since the elements x and y are skew-adjoint, our best bet is to take h ∈ iR, but then we still have a problem with the logarithm, as the spectrum of y is the entire line iR. A correct analytic analogue of the Jordanian twist was found by Stachura [30], see formula (3.31) below, but it turns out that, similarly to the case of SU (1, 1), it is important to work with the entire nonconnected ax + b group. What to do in the connected case remains an open problem. In fact, in an earlier paper [4] the first two authors found a universal deformation formula for the actions of the connected component of the ax + b group (and, more generally, of Kahlerian Lie groups) on C∗-algebras. Unfortunately, despite the claims in [26] and [5], this formula turned out to define a coisometric but nonunitary dual 2-cocycle, which is the term we prefer to use in the analytical setting instead of the "twist". See Remark 2.12 below and erratum to [5] for further discussion. If we do consider the nonconnected ax + b group, then even earlier, Baaj and Skandalis constructed its quantization as a bicrossed product of two copies of R∗ [29]. One disadvantage of this construction is that from the outset it is not clear how justifiable it is to call their quantum group a quantization of the ax + b group. Some justification was given later by Vaes and Vainerman [33]. The present work grew out of the natural question how the constructions in [29], [4] and [30] are related. As we already said, we found out that [4] does not lead to a unitary cocycle and therefore cannot actually be used to quantize the ax + b group. But the constructions in [29] and [30] turned out to be equivalent, as was conjectured by Stachura. Furthermore, we found an interpretation of the Jordanian twist/Stachura cocycle in terms of the Kohn -- Nirenberg quantization, which allowed us to construct quantum analogues of a class of semidirect products Q ⋊ V . We also realized that these constructions have a very natural description within De Commer's analytic version of the Hopf -- Galois theory [11, 12]. In more detail, the main results and organization of the paper are as follows. After a short pre- liminary section, we begin by discussing G-Galois objects for general locally compact groups G in Section 2. These are von Neumann algebras equipped with actions of G that are in an ap- propriate sense free and transitive. For compact groups such actions are known in the operator algebra literature as full multiplicity ergodic actions. Using recent results of De Commer [12] we show that the G-Galois objects with underlying algebras factors of type I are classified by ∗In [8], the Jordanian twist does not appear in exponential form. To our knowledge, it was first observed in [16] that Coll -- Gerstenhaber -- Giaquinto's twist can be put in the exponential form and therefore it coincides with Ogievetsky's twist. 2 certain second cohomology classes on G (Theorem 2.4). For finite groups such a result quickly leads to a complete classification of G-Galois objects obtained by Wassermann [34] (although the result is not very explicit there, see [25]) and Davydov [10]. For infinite groups the situa- tion is of course more complicated, as in general there exist Galois objects built on non-type-I algebras. We show next that under extra assumptions a G-Galois object of the form (B(H), Ad π), where π is a projective representation of G on H, defines a dual unitary 2-cocycle (Proposi- tion 2.9). We do not know whether these assumptions are always automatically satisfied, but we show that they are if π is a genuine representation (Theorem 2.13). This implies that if the group von Neumann algebra W ∗(G) of a nontrivial group G is a type I factor, then there exists a canonical nontrivial cohomology class of dual cocycles on G. This gives probably the shortest explanation why a quantization of, for example, the ax + b group exists at the operator algebraic level. At the Lie (bi)algebra level this is related to Drinfeld's result on quantization of Frobenius Lie algebras [13]. The construction of the dual cocycle in Section 2 is, however, rather inexplicit and in Sec- tion 3 we find a formula for such a cocycle for the semidirect products G = Q ⋉ V such that V is abelian and the action of Q on the dual abelian group V has a free orbit of full measure (Assumption 2.15). It is well-known that producing a dual 2-cocycle/twist is essentially equiv- alent to finding a G-equivariant deformation of an appropriate algebra of functions on G. Our assumptions on G = Q ⋉ V imply that we can identify L2(G) with L2(V × V ) in a G-equivariant way. The Kohh -- Nirenberg quantization of V × V provides then a deformation of L2(G) and gives rise to a dual unitary 2-cocycle Ω (Theorem 3.12). The cohomology class of this cocycle is exactly the one we defined in Section 2 (Theorem 3.18). In fact, there are two versions of the Kohn -- Nirenberg quantization, so we get two cohomolo- gous dual cocycles. In the case of the ax + b group we show that one of these cocycles coincides with Stachura's cocycle (Proposition 3.28). Finally, in Section 4 we consider the bicrossed product defined by a matched pair of two copies of Q in Q ⋊ V . We show that this quantum group is self-dual and isomorphic to (W ∗(G), Ω ∆(·)Ω∗) (Theorem 4.1 and Corollary 4.2). This is achieved by showing that the multiplicative unitary of the twisted quantum group (W ∗(G), Ω ∆(·)Ω∗) is given by a pentago- nal transformation on Q × V (Theorem 3.26) and by applying the Baaj -- Skandalis procedure of reconstructing a matched pair of groups from such a transformation [2]. Let G be a locally compact group. We fix a left invariant Haar measure dg on G and denote by Lp(G), p ∈ [1, ∞], the associated function spaces. 1. Preliminaries The modular function ∆ = ∆G is defined by the relation ZG f (hg) dh = ∆(g)−1ZG f (h) dh for f ∈ Cc(G). Then ∆(g)−1dg is a right invariant Haar measure on G. In a similar way, if q ∈ Aut(G), then the modulus q = qG of q is defined by the identity ZG f (q(h)) dh = q−1ZG f (h) dh for f ∈ Cc(G). We let λ and ρ be the left and right regular (unitary) representations of G on L2(G): (λgf )(h) = f (g−1h) and (ρgf )(h) = ∆(g)1/2f (hg). For a function f on G we define a function f by f (g) := f (g−1). 3 We also let J = JG and J = JG be the modular conjugations of L∞(G) and W ∗(G) := λ(G)′′: Jf := ¯f and Jf := ∆−1/2 ¯f, and we use the shorthand notation J := J J = J J, so that J f = ∆−1/2 f . (1.1) The multiplicative unitary W = WG : L2(G) ⊗ L2(G) → L2(G) ⊗ L2(G) of G is defined by (W f )(g, h) = f (g, g−1h). The multiplicative unitary W = WG of the dual quantum group is defined by W = W ∗ 21, so that ( W f )(g, h) = f (hg, h). The coproduct ∆ : W ∗(G) → W ∗(G) ¯⊗W ∗(G) on the group von Neumann algebra W ∗(G) is defined by ∆(λg) = λg ⊗ λg. We then have ∆(x) = W ∗(1 ⊗ x) W for x ∈ W ∗(G). Let now V be a locally compact Abelian group and V be its Pontryagin dual. Elements of V will be denoted by the Latin letters v, vj, v′ . . . while elements of V will be denoted by the Greek letters ξ, ξj, ξ′ . . . . We will use additive notation both on V and on V . The duality paring V × V → T will be denoted by eihξ,vi. This is just a notation, we do not claim that there is an exponential function here. We also let We fix a Haar measure dv on V and we normalize the Haar measure dξ of V so that the e−ihξ,vi := eihξ,vi = eih−ξ,vi = eihξ,−vi. Fourier transform FV defined by (FV f )(ξ) :=ZV e−ihξ,vif (v) dv becomes unitary from L2(V ) to L2( V ). For functions in several variables only one of which is in V , we use the same symbol FV to denote the partial Fourier transform in that variable. 2. Galois objects and dual cocycles 2.1. Projective representations and Galois objects. Hopf -- Galois objects is a well-studied topic in Hopf algebra theory. An adaption of this notion to locally compact quantum groups has been developed by De Commer [11]. Let us recall the main definitions. We will do this for genuine groups, as this is mainly the case we are interested in, but it will be important for us that the theory is developed at least for locally compact groups and their duals. Let G be a locally compact group and β be an action of G on a von Neumann algebra N . Such an action is called integrable if the operator-valued weight P : N → N β, N+ ∋ a 7→ZG βg(a) dg, is semifinite. If β is in addition ergodic, then we get a normal semifinite faithful weight ϕ on N such that P (a) = ϕ(a)1. Note that ϕ(βg(a)) = ∆(g)−1 ϕ(a) for all a ∈ N+. (2.1) We can then define an isometric map G : L2(N , ϕ) ⊗ L2(N , ϕ) → L2(G; L2(N , ϕ)), G(cid:0) Λ(a) ⊗ Λ(b)(cid:1)(g) = Λ(βg(a)b), where Λ : N ϕ → L2(N , ϕ) denotes the GNS-map. The pair (N , β) consisting of a von Neumann algebra N and an ergodic integrable action β of G on N is called a G-Galois object if the Galois map G is unitary. The following characterization of Galois objects is often easier to use. 4 Proposition 2.1. A pair (N , β) consisting of a von Neumann algebra N and an ergodic inte- grable action β of a locally compact group G on N is a G-Galois object if and only if N ⋊ G is a factor, which is then necessarily of type I. Indeed, the integrability Proof. This is a simple consequence of results in [11, Section 2]. assumption implies that N ⋊ G has a canonical representation η on L2(N , ϕ), and then by [11, Theorem 2.1] the pair (N , β) is a G-Galois object if and only if η is faithful. If N ⋊ G is a factor, then η is faithful, so we get one implication in the lemma. Next, the ergodicity of the action β implies that the action on N ′ by the automorphisms Ad η(λg) is ergodic as well, that is, N ′ ∩ η(W ∗(G))′ = C1. It follows that η(N ⋊ G) = B(L2(N , ϕ)). Hence, if (N , β) is a Galois object, then N ⋊ G is a factor canonically isomorphic to B(L2(N , ϕ)). (cid:3) We will be interested in G-Galois objects that are themselves factors of type I, in which case we say, following [12], that (N , β) is a I-factorial G-Galois object. Identifying N with B(H) for a Hilbert space H, we then get a projective unitary representation π : G → P U (H) such that βg = Ad π(g). Note that the equivalence class of π is uniquely determined by (N , β). Remark 2.2. Ergodicity of β = Ad π is equivalent to irreducibility of π. Assuming ergod- icity, integrability of the action Ad π is equivalent to square-integrability of the irreducible projective representation π, meaning that there are nonzero vectors ξ, ζ for which the function g 7→ (π(g)ξ, ζ) is square-integrable. This observation goes back to [9, Example 2.8, Chap- ter III], but let us give some details. Assume first that the action Ad π is integrable. Then the domain of definition of the weight ϕ must contain a nonzero rank-one operator θξ,ξ. Then, for every ζ ∈ H, the function g 7→ ((Ad π(g))(θξ,ξ)ζ, ζ) = (π(g)ξ, ζ)2 is integrable, so π is square-integrable. Conversely, assume π is square-integrable. Then by [14] (for genuine representations) and by [1] (for projective representations) there exists a unique positive, possibly unbounded, non- singular operator K on H, called the Duflo -- Moore formal degree operator, such that ZG (π(g)ξ, ζ)2dg = kK 1/2ξk2kζk2 for all ξ ∈ Dom(K 1/2) and ζ ∈ H. This implies that θξ,ξ is in the domain of definition of the weight ϕ and ϕ(θξ,ξ) = kK 1/2ξk2. It follows that the action Ad π is integrable and Note for a future use that property (2.1) translates into ϕ = Tr(K 1/2 · K 1/2). (Ad π(g))(K) = ∆(g)K. (2.2) Since the action Ad π is ergodic, this determines K uniquely up to a scalar factor. In particular, as was already observed in [14], if G is unimodular then K is scalar, and otherwise K is unbounded. Remark 2.3. By [11], given a G-Galois object, we get a locally compact quantum group G′ obtained by reflecting G across the Galois object. If G is abelian, then G′ = G. But if G is a nonabelian genuine locally compact group and our Galois object has the form (B(H), Ad π) for a projective representation π of G, then G′ is a genuine quantum group. Indeed, assume G′ is a group. By the general theory we know that B(H) is a G′-Galois object with respect to an action β′ of G′ commuting with the action of G, see [11]. There exist scalars χg(g′) ∈ T such that g′(π(g)) = χg(g′)π(g) for all g ∈ G, g′ ∈ G′, β′ where π(g) is any lift of π(g) to U (H). Then χg is a character of G′. Furthermore, by ergodicity of the action β′, if χg1 = χg2 for some g1, g2 ∈ G, then π(g1) and π(g2) coincide up to a scalar factor, which by surjectivity of the Galois map for (B(H), Ad π) is possible only when g1 = g2. 5 Therefore the map g 7→ χg is an injective homomorphism from G into the group of characters of G′. Hence G is abelian, which contradicts our assumption. By a recent duality result of De Commer [12], for any locally compact quantum group G, there is a bijection between the isomorphism classes of I-factorial G-Galois object and I-factorial G-Galois objects. This bijection is constructed as follows. Suppose we are given a I-factorial G-Galois object (N , β). Then N ′ ∩ (N ⋊ G), equipped with the dual action, becomes a I- factorial Galois object for G. (More precisely, we rather get a Galois object for the opposite comultiplication on L∞( G) and then an additional application of modular conjugations is needed to really get a Galois object for G, but this is unnecessary in our setting of genuine groups and their duals.) There is a simple class of G-Galois objects constructed as follows. Assume from now on that G is second countable. Let ω be a T-valued Borel 2-cocycle on G. Consider the ω-twisted left regular representation of G on λω : G → B(L2(G)) defined by (λω g f )(h) = ω(g, g−1h)f (g−1h), g λω gh, and let W ∗(G; ω) := λω(G)′′ ⊂ B(L2(G)). Then W ∗(G; ω) satisfying λω h = ω(g, h)λω g ⊗ λg of G (or in other words, the action of G) is a G-Galois equipped with the coaction λω g 7→ λω object, see [11, Section 5] for this statement in the setting of locally compact quantum groups. In fact, this covers all possible I-factorial G-Galois objects and by duality we get a description of the I-factorial G-Galois objects: Theorem 2.4. For any second countable locally compact group G, there is a bijection be- tween the isomorphism classes of I-factorial G-Galois objects and the cohomology classes [ω] ∈ H 2(G; T) such that the twisted group von Neumann algebra W ∗(G; ω) is a type I factor. Here H 2(G; T) denotes the Moore cohomology of G, which is based on Borel cochains [23]. Explicitly, the bijection in the theorem is defined as follows. To a I-factorial G-Galois object (B(H), Ad π) we associate the cohomology class [ωπ] ∈ H 2(G; T) defined by the projective representation π of G. Recall that this means that we lift π to a Borel map π : G → U (H) and define a Borel T-valued 2-cocycle ωπ on G by the identity π(g)π(h) = ωπ(g, h)π(gh). The inverse map associates to [ω] ∈ H 2(G; T) (such that W ∗(G; ω) is a type I factor) the isomorphism class of the G-Galois object (W ∗(G; ω), Ad λω). For finite groups G, this theorem is essentially due to Wassermann [34] (see also [25]), as well as to Davydov [10] in the purely algebraic setting. We divide the proof of Theorem 2.4 into a couple of lemmas. Let (B(H), Ad π) be a I-factorial G-Galois object and ω be the cocycle defined by a lift π of π. Lemma 2.5. The G-Galois object associated with the I-factorial G-Galois object (B(H), Ad π) is isomorphic to (W ∗(G; ¯ω), α), where the coaction α of G is defined by α(λ¯ω g ) = λ¯ω g ⊗ λg. (2.3) Proof. We have an isomorphism B(H) ⋊ G ∼= B(H) ¯⊗W ∗(G; ¯ω), B(H) ∋ T 7→ T ⊗ 1, λg 7→ π(g) ⊗ λ¯ω g . Explicitly, the crossed product B(H) ⋊ G is the von Neumann subalgebra of B(L2(G; H)) generated by the operators λg considered as operators on L2(G; H) and the operators T for T ∈ B(H) defined by ( T ξ)(g) = π(g)∗T π(g)ξ(g). Then the required isomorphism is given by Ad U , where U : L2(G; H) → L2(G; H) is defined as This gives the result. (U ξ)(g) = π(g)ξ(g). 6 (cid:3) In particular, it follows that W ∗(G; ¯ω) is a type I factor. As JW ∗(G; ω)J = W ∗(G; ¯ω), the von Neumann algebra W ∗(G; ω) is a type I factor as well. By De Commer's duality result [12] we conclude that the map associating [ωπ] to the isomorphism class of (B(H), Ad π) is injective and its image is contained in the set of cohomology classes [ω] such that W ∗(G; ω) is a type I factor. Consider now an arbitrary cohomology class [ω] ∈ H 2(G; T) such that W ∗(G; ω) is a type I factor. To finish the proof it suffices to establish the following. Lemma 2.6. The pair (W ∗(G, ω), Ad λω) is a G-Galois object. Proof. Let us start with the I-factorial G-Galois object (W ∗(G; ¯ω), α), with the coaction α given by (2.3). By definition, the crossed product W ∗(G, ¯ω) ⋊α G is the von Neumann subalgebra of B(L2(G) ⊗ L2(G)) generated by α(W ∗(G; ¯ω)) and 1 ⊗ L∞(G). The unitary ω21 W commutes with 1 ⊗ L∞(G) and satisfies ω21 W (λ¯ω g ⊗ λg) = (1 ⊗ λ¯ω g )ω21 W . Therefore the conjugation by this unitary defines an isomorphism of W ∗(G, ω) ⋊α G onto the algebra 1 ⊗ B(L2(G)). This isomorphism maps α(W ∗(G; ¯ω)) onto 1 ⊗ W ∗(G; ¯ω). In order to understand what happens with the dual action, initially defined by the automor- phisms Ad(1 ⊗ ρg), it is convenient to assume that the cocycle ω satisfies ω(g, e) = ω(e, g) = ω(g, g−1) = 1 for all g ∈ G, which is always possible to achieve by replacing ω by a cohomologous cocycle. Then J is the J, we then modular conjugation of W ∗( G; ¯ω), so that W ∗( G; ¯ω)′ = JW ∗( G; ¯ω) J . Put ρω have g = Jλ¯ω g (ρ¯ω g f )(h) = ∆(g)1/2ω(g, g−1h−1)f (hg) = ∆(g)1/2ω(h, g)f (hg). It is then not difficult to check that ω21 W (1 ⊗ ρg) W ∗ ¯ω21 = (λ¯ω g ⊗ ρω g ). We thus see that the von Neumann algebra α(W ∗(G; ¯ω))′ ∩ (W ∗(G; ¯ω) ⋊α G), together with the restriction of the dual action, is isomorphic to ρω(G)′′ = JW ∗(G; ¯ω) J with the action given by the automorphisms Ad ρω g . As g = J λ¯ω ρω g J = J Jλω g J J , we conclude that the G-Galois object associated with the G-Galois object (W ∗(G; ¯ω), α) is isomorphic to (W ∗(G; ω), Ad λω). (cid:3) 2.2. Dual cocycles. An important class of G-Galois objects arises from dual 2-cocycles. By a dual unitary 2-cocycle on G we mean a unitary element Ω ∈ W ∗(G) ¯⊗W ∗(G) such that (Ω ⊗ 1)( ∆ ⊗ ι)(Ω) = (1 ⊗ Ω)(ι ⊗ ∆)(Ω). (2.4) Similarly to the G-Galois objects W ∗(G; ω) considered above, such cocycles lead to G-Galois ob- jects W ∗( G; Ω) (which for notational consistency with W ∗(G; ω) we should have rather denoted by W ∗( G; Ω∗)). Following [26, Section 4], they can be described as follows. Identify as usual the Fourier algebra A(G) with the predual of W ∗(G). Given a dual unitary 2-cocycle Ω ∈ W ∗(G) ¯⊗W ∗(G), the von Neumann algebra N := W ∗( G; Ω) ⊂ B(L2(G)) is generated by the operators Define an action β of G on N by πΩ(f ) := (f ⊗ ι)( W Ω∗), f ∈ A(G). βg(x) := (Ad ρg)(x), x ∈ N , 7 where we remind that ρg = J λg J is the right regular representation. The map πΩ is the representation of the algebra A(G) equipped with the new product The representation πΩ has the equivariance property (f1 ⋆Ω f2)(g) := (f1 ⊗ f2)(cid:0) ∆(λg)Ω∗(cid:1). βg(cid:0)πΩ(f )(cid:1) = πΩ(λgf ). (2.5) (2.6) By [33, Section 1.3], the canonical weight ϕ on N has the following description. Its GNS-space can be identified with L2(G), with the GNS-map Λ : N ϕ → L2(G) uniquely determined by Λ(πΩ(f )) = f for f ∈ A(G) such that f ∈ L2(G), where we remind that f (g) = f (g−1). In particular, for f as above we have ϕ(πΩ(f )∗πΩ(f )) = k f k2 2. (2.7) (2.8) Two dual unitary 2-cocycles Ω, Ω′ are called cohomologous if there exists a unitary u ∈ W ∗(G) such that Ω′ = (u ⊗ u)Ω ∆(u)∗. The cohomology classes form a set H 2( G; T). structure. In general this set does not have any extra Proposition 2.7. Two dual unitary 2-cocycles Ω, Ω′ on a locally compact group G are coho- mologous if and only if they define isomorphic G-Galois objects. Proof. As ∆(u) = W ∗(1 ⊗ u) W , it is easy to see that if Ω′ = (u ⊗ u)Ω ∆(u)∗ then Ad u defines a G-equivariant isomorphism of W ∗( G; Ω) onto W ∗( G; Ω′). Conversely, assume we have a G-equivariant isomorphism θ : W ∗( G; Ω) → W ∗( G; Ω′). Denote by Λ and Λ′ the GNS-maps for these objects as described above. Then the isomorphism θ is implemented by the unitary u defined by uΛ(x) = Λ′(θ(x)). Since by (2.6) and (2.7) the actions of G are implemented in a similar way by the unitaries ρg: ρg Λ(x) = ∆(g)1/2 Λ(βg(x)) and ρg Λ′(x′) = ∆(g)1/2 Λ′(β′ g(x′)), we conclude that uρg = ρgu, hence u ∈ W ∗(G). Denote by G and G′ the corresponding Galois maps. Then by definition we have (1 ⊗ u)G = G′(u ⊗ u). On the other hand, by [11, Proposition 5.1] we have G = W Ω∗ and G′ = W Ω′∗. Hence (1 ⊗ u) W Ω∗ = W Ω′∗(u ⊗ u). Using again that ∆(u) = W ∗(1 ⊗ u) W , we conclude that Ω′ = (u ⊗ u)Ω ∆(u)∗. (cid:3) Combined with Theorem 2.4 this proposition allows one to describe a part of H 2( G; T) in I ( G; T) the subset of H 2( G; T) formed by the terms of cohomology of G. Namely, denote by H 2 classes [Ω] such that W ∗( G; Ω) is a type I factor. Given such an Ω, we can identify W ∗( G; Ω) with B(H) for a Hilbert space H. Then the action β of G on W ∗( G; Ω) is given by a projective representation π of G on H, and we denote by cΩ the corresponding 2-cocycle ωπ on G. Similarly, denote by H 2 I (G; T) the subset of H 2(G; T) formed by the classes [ω] such that W ∗(G; ω) is a type I factor. Corollary 2.8. For any second countable locally compact group G, the map Ω 7→ [cΩ] defines an embedding of H 2 I ( G; T) into H 2 I (G; T). A natural question is whether this embedding is onto. We do not know the answer, but as a step towards a solution of this problem let us explain how dual cocycles arise from Galois maps under extra assumptions. 8 It will be useful to go beyond Galois objects. Assume we are given a square-integrable irreducible projective representation π of G on H. Assume also that we are given a unitary map Op : L2(G) → HS(H) such that Op(λgf ) = (Ad π(g))(Op(f )), (2.9) which we will call a quantization map. Here HS(H) denotes the Hilbert space of Hilbert -- Schmidt operators on H. As in Section 2.1, consider the Duflo -- Moore operator K on H and the weight ϕ = Tr(K 1/2 · K 1/2). As the GNS-space for ϕ we could take HS(H), with the GNS-map Λ uniquely determined by Λ(T K −1/2) := T for T ∈ HS(H) such that T K −1/2 is a bounded operator. But using the unitary Op we can transport everything to L2(G). Thus we take L2(G) as the GNS-space, with the GNS-map uniquely determined by Λ(Op(f )K −1/2) := f for f ∈ L2(G) such that Op(f )K −1/2 ∈ B(H). (2.10) Consider the corresponding Galois map G, so G : L2(G) ⊗ L2(G) → L2(G) ⊗ L2(G), G(f1 ⊗ f2)(g, h) = Λ(cid:16)(cid:0) Ad π(g)(cid:1)(Op(f1)K −1/2) Op(f2)K −1/2(cid:17)(h). Finally, define where we remind that J = J J. Ω := (J ⊗ J )G∗(1 ⊗ J ) W , (2.11) (2.12) Proposition 2.9. With the above setup and notation, the operator Ω is coisometric. It lies in the algebra W ∗(G) ¯⊗W ∗(G) and satisfies the cocycle identity (2.4). In particular, Ω is a dual unitary 2-cocycle on G if and only if (B(H), Ad π) is a G-Galois object. Moreover, if Ω is indeed unitary, then (B(H), Ad π) is isomorphic to the G-Galois object (W ∗( G; Ω), β) defined by Ω. Proof. Since the Galois maps are always isometric, it is clear that Ω is coisometric. Next, a straightforward application of definition (2.11) together with scaling property (2.2) yield the following identities for G, cf. [11, Lemma 3.2]: G(λg ⊗ 1) = (ρg ⊗ 1)G, G(1 ⊗ λg) = (λg ⊗ λg)G. Together with the identities W (ρg ⊗ 1) = (ρg ⊗ 1) W , W (1 ⊗ ρg) = (λg ⊗ ρg) W , J ρg = λgJ this implies that Ω commutes with the operators ρg ⊗ 1 and 1 ⊗ ρg. Hence Ω ∈ W ∗(G) ¯⊗W ∗(G). Turning to the cocycle identity, by [11, Proposition 3.5] the Galois map G (denoted by G in op. cit.) satisfies the following hybrid pentagon relation: W12G13G23 = G23G12. (More precisely, the result in [11] is formulated only for the Galois objects, but an inspection of the proof shows that it remains valid for arbitrary integrable ergodic actions.) Plugging in G = (1 ⊗ J ) W Ω∗(J ⊗ J ) we get W12 W13Ω∗ 13 W23Ω∗ 23 = W23Ω∗ 23 W12Ω∗ 12, and using ∆(x) = W ∗(1 ⊗ x) W and the pentagon relation W12 W13 W23 = W23 W12 we obtain the required cocycle identity. Finally, by the definition of Ω, the Galois map G is unitary if and only if Ω is unitary. Assuming that Ω and G are unitary, by [11, Proposition 3.6(1)] the elements (ω ⊗ ι)(G) for ω ∈ B(L2(G))∗ span a σ-weakly dense subspace of π ϕ(B(H)). Recalling the definition of W ∗( G; Ω) we conclude that π ϕ(B(H)) = J W ∗( G; Ω)J . The action of G on B(H) is implemented on the GNS-space by the unitaries λg, while that on W ∗( G; Ω) by the unitaries ρg. Since J ρgJ = λg, we see that the Galois objects (B(H), Ad π) and (W ∗(G; Ω), β) are indeed isomorphic. (cid:3) 9 Remark 2.10. Although [11, Proposition 3.6(1)] is formulated only for the Galois objects, its proof remains valid for any integrable ergodic action. Therefore we see that starting from a square-integrable irreducible projective representation π of G on H and a unitary quantization map Op : L2(G) → HS(H), we can define a dual coisometric cocycle Ω by (2.12) and then the set of elements (ω ⊗ ι)( W Ω∗) for ω ∈ B(L2(G))∗ span a σ-weakly dense subspace of J π ϕ(B(H))J . Therefore Ω contains a complete information about (B(H), Ad π) independently of whether we deal with a Galois object or not. Remark 2.11. The quantization map Op defines a product ⋆ on L2(G) by In general the product ⋆Ω defined by Ω is not the same as ⋆ on A(G) ∩ L2(G). Indeed, the GNS-map ΛΩ for W ∗( G; Ω) defined by (2.7) is related to the GNS-map (2.10) by Op(f1 ⋆ f2) = Op(f1) Op(f2). ΛΩ(πΩ(f )) = f = J Λ(Op(∆−1/2f )K −1/2). This identity should be taken with a grain of salt, since it is not clear why Op(∆−1/2f )K −1/2 ∈ B(H) for sufficiently many elements f ∈ A(G)∩∆1/2L2(G). Nevertheless, if we proceed formally, we see that ⋆Ω is related to the quantization map Op′ defined by Op′(f ) = Op(∆−1/2f )K −1/2, so that Op′(f1 ⋆Ω f2) = Op′(f1) Op′(f2). In good cases these two products coincide, or in other words, the map f 7→ Op−1(Op(∆−1/2f )K −1/2) is an automorphism with respect to ⋆. To understand better when this might happen, observe that since ∆ is the only positive measurable function F on G such that λgF = ∆(g)−1F , any reasonable extension of Op to a class of functions including ∆s should satisfy Op(∆1/2) = cK −1/2 for a constant c > 0. Then Op−1(Op(∆−1/2f )K −1/2) = c−1(∆−1/2f ) ⋆ ∆1/2. From this we see that for the map f 7→ Op−1(Op(∆−1/2f )K −1/2) to be an automorphism it suffices to have the identities ∆s ⋆ ∆t = ∆s+t, ∆−1/2f = c∆−α ⋆ f ⋆ ∆α−1/2 (2.13) for some α ∈ R. For the examples studied in this paper we will indeed have such identities, with c = 1 and α = 1/2. On other hand, for the example studied in [5] (which is not a Galois object) we had c = 1 and α = 1/4. We thus see that the problem of describing H 2 I ( G; T) reduces to the following question: it is true that for any I-factorial Galois object (B(H), Ad π) there is a unitary quantization map (2.9)? This can be reformulated as a representation-theoretic problem as follows. Assume we are given a 2-cocycle ω on G such that W ∗(G; ω) is a type I factor. We identify W ∗(G; ω) with B(H) and put πω(g) := λω ω(g) is a well-defined unitary ω(g)¯ξ := πω(g)ξ. Is this representation equivalent to the representation of G on H ⊗ ¯H, where πc regular representation? g ∈ B(H). Then g 7→ πω(g) ⊗ πc The answer is known to be "yes" for finite groups [34, 24]. Indeed, the Galois map gives a unitary equivalence πω ⊗ πc ω ⊗ εH ⊗ ε ¯H ∼ ρ ⊗ εH ⊗ ε ¯H , where εL denotes the trivial representation of G on the Hilbert space L. This implies the required equivalence πω ⊗ πc ω ∼ λ for finite groups G, but falls short of what we need for general G. Remark 2.12. In order to stress that it can be dangerous to rely too much on analogies with the finite group case, note that for any square-integrable irreducible projective representation π of G on H, the Galois map always defines an embedding of the representation Ad π ⊗ εHS(H) on HS(H) ⊗ HS(H) into ρ ⊗ εHS(H). It follows that for finite groups existence of a unitary quantization map (2.9) is equivalent to (B(H), Ad π) being a Galois object. This is certainly (but until recently unexpectedly!) not the case for general groups. For example, for the con- nected component G of the ax + b group over R there are two inequivalent infinite dimensional 10 irreducible unitary representations. They are both square-integrable and both admit unitary quantization maps [4]. But G has no I-factorial Galois objects, since H 2(G; T) is trivial and W ∗(G) is the sum of two type I factors. 2.3. Dual cocycles defined by genuine representations. We now turn to Galois objects (B(H), Ad π) defined by genuine representations. Theorem 2.13. For any nontrivial second countable locally compact group G, a (square- integrable, irreducible) unitary representation π : G → B(H) such that (B(H), Ad π) is a G- Galois object exists if and only if W ∗(G) is a type I factor. Moreover, if π exists, then (i) π is unique up to unitary equivalence; explicitly, by identifying W ∗(G) with B(H) we can take π(g) = λg; (ii) the Galois object (B(H), Ad π) is defined by a dual unitary 2-cocycle Ω on G. Recall that by Proposition 2.7 the cohomology class of Ω is determined by the isomorphism class of the corresponding Galois object. Therefore the above theorem shows that if W ∗(G) is a type I factor, then we get a canonical class [Ω] ∈ H 2( G; T). In terms of Corollary 2.8, this class corresponds to the unit of H 2(G; T). Note also that the condition that W ∗(G) is a type I factor implies that G is neither compact nor discrete, so the situation described in the theorem is a purely analytical phenomenon. Proof of Theorem 2.13. The first statement is an immediate consequence of Theorem 2.4 ap- plied to genuine representations and, correspondingly, to the trivial 2-cocycle ω = 1 on G. Furthermore, that theorem implies that π is unique up to equivalence as a projective represen- tation. Therefore to prove part (i) we only have to show a slightly stronger statement that π is also unique up to equivalence as a genuine representation. Thus, we identify W ∗(G) with B(H), take π(g) = λg, and we have to show that for any character η : G → T the represen- tations ηπ and π are equivalent. The representations ηλ and λ are unitarily equivalent, e.g., by Fell's absorption principle. It follows that there exists an automorphism θ of W ∗(G) such that θ(λg) = η(g)λg for all g. As θ is an automorphism of B(H) = W ∗(G), it is unitarily implemented, which means exactly that ηπ and π are unitarily equivalent. In order to prove part (ii), by our results in Section 2.2 it suffices to show that the represen- tation π ⊗ πc is equivalent to the regular representation. Since we can identify W ∗(G) with B(H) in such a way that π(g) = λg, the representation λ is a multiple of π, so we can write λ ∼ π ⊗ εL, where εL is the trivial representation on a separable Hilbert space L. Since G is nontrivial, the Hilbert space H must be infinite dimensional. But then the multiplicity of the square-integrable representation π in λ must be infinite as well [14], so the Hilbert space L is infinite dimensional. By passing to the conjugate representations we get πc ⊗ εL ∼ λc ∼ λ ∼ π ⊗ εL. This implies that the irreducible representations πc and π are equivalent. Next, using Fell's absorption principle we get π ⊗ π ⊗ εL ∼ π ⊗ λ ∼ λ ⊗ εH ∼ π ⊗ εL ⊗ εH. From this we see that the representation π ⊗ π is a multiple of π, and in order to conclude that π ⊗π ∼ λ it suffices to show that the multiplicity of π in π ⊗π is infinite. In other words, we have to check that the commutant of (π ⊗ π)(G) in B(H ⊗ H) is infinite dimensional. Equivalently, that the commutant of (λ ⊗ λ)(G) in W ∗(G) ¯⊗W ∗(G) is infinite dimensional. More generally, let us show that if G1 is a closed nonopen subgroup of a second countable locally compact group G2 such that W ∗(G1) is a type I factor, then the relative commutant W ∗(G1)′ ∩ W ∗(G2) is infinite dimensional. Assume W ∗(G1)′∩W ∗(G2) is finite dimensional. Denote by ∆i the modular function of Gi, by µi the Haar measure on Gi and by ϕi the standard Haar weight on W ∗(Gi). The modular group of ϕ2 is given by σt(λg) = ∆2(g)itλg. It preserves W ∗(G1), and since W ∗(G1) is a type I factor, 11 1 on W ∗(G1) with the same modular group. there exists a normal semifinite faithful weight ϕ′ Consider the unique normal semifinite operator-valued weight P : W ∗(G2) → W ∗(G1) such that 1P = ϕ2. Since W ∗(G1) is a type I factor and W ∗(G1)′ ∩ W ∗(G2) is finite dimensional, it ϕ′ follows, e.g., from [32, Corollary 12.12] applied to M = W ∗(G1) that such an operator-valued weight must be bounded, hence it is a scalar multiple of a conditional expectation. In particular, the weight ϕ2W ∗(G1) is semifinite. This, in turn, implies, that ϕ2W ∗(G1) is a Haar weight on W ∗(G1), hence ϕ2W ∗(G1) = c ϕ1 for a constant c > 0. We can also conclude that ∆2G1 = ∆1. Denote by E the conditional expectation obtained by rescaling P , so that c ϕ1E = ϕ2. Using the identity ϕ2(xy) = c ϕ1(E(x)y) for appropriate elements x ∈ W ∗(G2) and y ∈ W ∗(G1), it is F (g)λg dµ2(g), with not difficult to compute E on a dense set of elements. Namely, if x = RG2 F = f ∗ f ′ (convolution in L1(G2)) for some f, f ′ ∈ Cc(G2), then we must have E(x) = c−1ZG1 F (g)λg dµ1(g). Applying this to functions f ′ ≥ 0 supported in arbitrarily small neighbourhoods of the identity and normalized by kf ′k1 = 1, which form an approximate unit in L1(G2), and using the assumed continuity of E, we conclude that E(cid:16)ZG2 f (g)λg dµ2(g)(cid:17) = c−1ZG1 f (g)λg dµ1(g) for all f ∈ Cc(G2). But this formula certainly defines an unbounded map, so we get a contradiction. Indeed, take any nonzero function f ∈ Cc(G1). Since G1 has zero measure in G2, we can extend f to a function f ∈ Cc(G2) with the same supremum-norm but supported in a set of arbitrarily small Haar measure, so that the L1-norm of f can be made arbitrarily small. Since the operator norm f (g)λg dµ1(g) under E has (cid:3) is dominated by the L1-norm, we thus see that the preimage of RG1 elements of arbitrarily small norm. As we have already observed, the class [Ω] ∈ H 2( G; T) corresponds to the unit of H 2(G; T), so one might wonder whether Ω is also a coboundary, that is, cohomologous to 1. But this is surely not the case, since W ∗( G; Ω) ∼= B(H), while W ∗( G) = L∞(G). In fact, the following stronger nontriviality property holds. Proposition 2.14. If G is a nontrivial second countable locally compact group with group von Neumann algebra a factor of type I, and Ω is the dual unitary 2-cocycle given by Theorem 2.13, then the twisted locally compact quantum group (W ∗(G), Ω ∆(·)Ω∗) is neither commutative nor cocommutative. Proof. The algebra W ∗(G), being a nontrivial type I factor, is clearly noncommutative. This implies that the group G is noncommutative. By Remark 2.3 it follows that the quantum group GΩ obtained by reflecting G across the Galois object (W ∗( G; Ω), β) is a genuine quantum group, that is, the coproduct ∆Ω on W ∗(GΩ) is not cocommutative. But (W ∗(GΩ), ∆Ω) is exactly (W ∗(G), Ω ∆(·)Ω∗), see [11]. (cid:3) 2.4. Examples: subgroups of the affine group. We now introduce the main class of exam- ples studied in this paper. Let V be a nontrivial second countable locally compact abelian group, Q be a second count- able locally compact group of continuous automorphisms of V and call G the semidirect product Q ⋉ V ⊂ Aff(V ) := Aut(V ) ⋉ V . It has the group law (q, v)(q′, v′) = (qq′, v + qv′), q, q′ ∈ Q, v, v′ ∈ V. The unit element is (id, 0) and the inverse is (q, v)−1 = (q−1, −q−1v). Whenever convenient we identify q ∈ Q with (q, 0) ∈ Q ⋉ V and v ∈ V with (id, v) ∈ Q ⋉ V . We will also usually write 1 instead of id for the unit of Q. We denote by q♭ the dual action of Q on V defined by the identity eihq♭ξ,vi = eihξ,q−1vi, 12 where we remind that (ξ, v) 7→ eihξ,vi simply denotes the duality paring V × V → T. We will study the groups Q ⋉ V satisfying the following property. Assumption 2.15. There is an element ξ0 ∈ V such that the map φ : Q → V , q 7→ q♭ξ0, is a measure class isomorphism. condition. In this case, we say that (V, Q) satisfies the dual orbit Remark 2.16. If (V, Q) satisfies the dual orbit condition, then V cannot be compact, since the neutral element of V cannot belong to the Q-orbit of ξ0 and therefore V cannot be discrete. Note also that the stabilizer of ξ0 in Q must be trivial, so that the map φ is injective. In order to give some concrete examples, let K be a nondiscrete locally compact field. If K is commutative then K is a local field. This means that K is isomorphic either to R, C, a finite degree extension of the field of p-adic numbers Qp or to a field Fq((X)) of Laurent series with coefficients in a finite field Fq. If K is a skew-field, then K is isomorphic to a finite dimensional division algebra over a local field k. As an abelian group K is self-dual, with a pairing K×K → T given by (a, b) 7→ χ(TrK/k(ab)) for a nontrivial character χ of k. Denote by χK the character χ ◦ TrK/k. Example 2.17. For n ≥ 1, let V = Matn(K) and Q = GLn(K), with the action given by left matrix multiplication. Under the isomorphism V ≃ V associ- ated with the pairing (A, B) 7→ χK(Tr(tAB)), the dual action is given by (A, M ) 7→ tA−1M . Therefore, both (V, Q) and ( V , Q) satisfy the dual orbit condition. Example 2.18. Let τ be any order-two ring automorphism of Matn(K). Consider the quater- nionic type group H± n (K, τ ) given by the subgroup of GL2n(K) of elements of the form (cid:18) A ±τ (B) B τ (A)(cid:19) , A, B ∈ Matn(K). Set V = Matn(K) ⊕ Matn(K) and Q = H± Here also, both (V, Q) and ( V , Q) satisfy the dual orbit condition. n (K, τ ). Example 2.19. For n ≥ 1 and m ≥ 2, let V = Matn(K) ⊕ · · · ⊕ Matn(K) and Q =   1 · · · ... . . . 0 · · · 0 · · · ... 0 Matn(K) ... 1 Matn(K) 0 GLn(K)   ⊂ GLnm(K). } m {z Then, the dual pair (V, Q) satisfies the dual orbit condition but the pair ( V , Q) does not satisfy Indeed, we have for M1, · · · , Mm, B1, · · · , Bm−1 ∈ Matn(K) and the dual orbit condition. A ∈ GLn(K):   1 · · · ... . . . 0 · · · 0 · · · B1 ... 0 ... 1 Bm−1 0 A     M1 ... Mm−1 Mm   =   −Pm−1 p=1 M1 ... Mm−1 tA−1Mp + tA−1Mm tBp   . Therefore, there is no orbit of full Haar measure in V . 13 Example 2.20. Let A be a nondiscrete second countable locally compact (unital, but not nec- essarily commutative) ring, such that the set A× of invertible elements is of full Haar measure. Then, the pair ( A, A×) satisfies the dual orbit condition. See [3, Section 4] for such explicit examples. Note that some of these examples show that the map φ : Q → V is not necessarily open. Let us now say a few words about the case when Q is a real or complex Lie group, V is a finite dimensional vector space and the action of Q on V is given by a representation ρ. In this case we can identify V with V ∗, and the action of Q on V ∗ is given by the contragredient representation ρc. If (Q, V ) satisfies the dual orbit condition, then Q has the same dimension as V and the map φ : Q → V ∗ is open, so (V ∗, ρc(Q)) is a prehomogeneous vector space. Remark 2.21. In the complex case, assuming dim Q = dim V , the dual orbit condition is satisfied for ξ0 ∈ V ∗ as long as ξ0 has trivial stabilizer. Indeed, consider the set Ω of vectors ξ ∈ V ∗ such that the map q → V ∗, X 7→ (dρc)(X)ξ, is a linear isomorphism. This set is nonempty, as ξ0 ∈ Ω, and Zariski open in V ∗. It follows that it is a dense, connected, open subset of V ∗ in the usual topology. Since the Q-orbit of every element of Ω is open, it follows that Ω consists entirely of one orbit, so the dual orbit condition is satisfied. As a byproduct we see that Q must be connected. Note that these arguments do not apply in the real case, as a Zariski open subset of Rn can have finitely many connected components. By a repeated use of the following simple lemma we can construct more and more complicated examples. Lemma 2.22. Assume (Q, V ) satisfies the dual orbit condition, with Q a complex Lie group and the action of Q given by a representation ρ : Q → GL(V ). Then the adjoint action of the Lie group G := Q ⋉ V on its Lie algebra g also satisfies the dual orbit condition. Proof. By Remark 2.21 it suffices to find a vector η0 ∈ g∗ with trivial stabilizer. By assumption there exists ξ0 ∈ V ∗ with trivial stabilizer in Q. Identifying g with q×V we let η0(X, w) := ξ0(w). The adjoint action is given by (cid:0) Ad(q, v)(cid:1)(X, w) =(cid:16)(Ad q)(X), ρ(q)w − (dρ)(cid:0)(Ad q)(X)(cid:1)v(cid:17). Assume now that (q, v) ∈ G lies in the stabilizer of η0, that is, ξ0(ρ(q)w) − ξ0(cid:16)(dρ)(cid:0)(Ad q)(X)(cid:1)v(cid:17) = ξ0(w) for all X ∈ q and w ∈ V. Taking X = 0, we see that q stabilizes ξ0, hence q = e. Since the map q → V ∗, X 7→ (dρc)(X)ξ0, is an isomorphism, we then conclude that v = 0. (cid:3) This lemma and its proof show, both in real and complex cases, that if (Q, V ) satisfies the dual orbit condition, then the Lie algebra of G := Q ⋉ V is Frobenius, meaning that G has an open coadjoint orbit, or equivalently, there exists η0 ∈ g∗ such that the bilinear form η0([X, Y ]) on g is nondegenerate. This has already been observed in [28]. Moreover, the converse is almost true: by [28, Theorem 4.1], if dim Q = dim V and the Lie algebra of Q ⋉ V is Frobenius, then there exists ξ0 ∈ V ∗ such that the map q → V ∗, X 7→ (dρc)(X)ξ0, is a linear isomorphism. This is a bit less than what we need since the stabilizer of ξ0 can still be a nontrivial discrete subgroup of Q and, in the real case, the open set ρc(Q)ξ0 ⊂ V ∗ can be nondense. In addition to Lemma 2.22, another way of producing new examples is to start with a pair (Q, V ) satisfying the dual orbit condition and multiply the representation of Q on V by a quasi-character of Q. This can destroy the dual orbit condition, but not necessarily. Example 2.23. Consider Q = C∗, V = C, with Q acting on V by multiplication. Obviously, the dual orbit condition is satisfied. By Lemma 2.22 the adjoint action of the ax + b group G := Q ⋉ V satisfies the dual orbit condition as well. This is basically Example 2.19 for n = 1 14 and m = 2. If we multiply the adjoint representation of G by the quasi-character (q, v) 7→ qk, k ∈ Z, we get the representation ρk : G → GL2(C), ρk(q, v) =(cid:18) qk −qkv qk+1(cid:19) . 0 It defines an action satisfying the dual orbit condition if and only if k 6= −1. We remark in passing that for k = 1 the group G ⋉ρ1 C2 is an extension of C∗ by the Heisenberg group. It is used in the construction of an extended Jordanian twist in [20]. Note also that we can use the same formulas in the real case. Then we get a pair satisfying the dual orbit condition if and only if k is even. Returning to the general case, we have the following result. Proposition 2.24. If a pair (V, Q) satisfies the dual orbit condition, then the group von Neu- mann algebra of G = Q ⋉ V is a type I factor. Proof. By conjugating by the partial Fourier transform FV we get W ∗(G) ∼= L∞( V ) ⋊ Q. But by the dual orbit condition the space V , considered as a measure space equipped with an action of Q, can be identified with Q equipped with the action of Q by left translations. Hence W ∗(G) ∼= B(L2(Q)). (cid:3) By Theorem 2.13 it follows that we get a canonical class [Ω] ∈ H 2( G; T) defined by the Galois object (W ∗(G), Ad λ). In order to get an explicit representative of this class we need to fix a (unique up to equivalence) representation π as in that theorem and choose a unitary quantization map (2.9). This is the task we are going to undertake in the next section. 3. Kohn -- Nirenberg quantization 3.1. Kohn -- Nirenberg quantization of V × V . Quantization of the abelian self-dual group V × V has been considered long ago [31, 35]. Here we consider the Kohn -- Nirenberg quantization. It has some benefit compared with the Weyl type quantization for which one has to assume the map V → V , v 7→ 2v, to be a homeomorphism. The Kohn -- Nirenberg quantization can be initially defined as the continuous injective linear map OpKN : S ′(V × V ) −→ L(cid:0)S (V ), S ′(V )(cid:1), from tempered Bruhat distributions on V × V to continuous linear operators from the Bruhat- Schwartz space on V to tempered Bruhat distributions on V (see [7, Section 9] for a precise definition). It is defined by the formula OpKN(F )ϕ(v) :=ZV × V eihξ,v−v′i F (v′, ξ) ϕ(v′) dv′dξ, F ∈ S ′(V × V ), ϕ ∈ S (V ). Remark 3.1. This quantization map should rather be called anti-Kohn -- Nirenberg. The Kohn -- Nirenberg one is given by the formula ZV × V eihξ,v−v′i F (v, ξ) ϕ(v′) dv′dξ. The two quantization maps are unitarily equivalent and for our purposes it is easier to work with anti-Kohn -- Nirenberg. The distributional kernel of the operator OpKN(F ) is therefore given by (v, v′) 7→(cid:0)(1 ⊗ F ∗ V )F(cid:1)(v′, v − v′). (3.1) Since an operator on L2(V ) with kernel K is Hilbert -- Schmidt if and only if K ∈ L2(V × V ), we immediately deduce: Lemma 3.2. The Kohn -- Nirenberg quantization map OpKN extends to a unitary isomorphism from L2(V × V ) onto HS(L2(V )). 15 Hence, the Hilbert space L2(V × V ) can be endowed with an associative product f1 ⋆0 f2 := Op∗ KN(cid:0) OpKN(f1) OpKN(f2)(cid:1). (3.2) Very important here are the symmetries of the Kohn -- Nirenberg product ⋆0. We have an action of Aff(V ) on V × V : Aff(V ) × (V × V ) → V × V , (q, v).(w, ξ) = (qw + v, q♭ξ). (3.3) With · V the modulus function of Aut(V ) and · V the one of Aut( V ), we observe that We will usually use only the modulus qV in various formulas and write it simply as q. The action (3.3) gives rise to a unitary representation πV × V of Aff(V ) on L2(V × V ) defined by q♭ V = q−1 V . πV × V (q, v)ϕ(w, ξ) = ϕ(q−1(w − v), q−1♭ ξ). We can also define a unitary representation πV of Aff(V ) on L2(V ) by (πV (q, v)ϕ)(w) = q−1/2ϕ(q−1(w − v)). Lemma 3.3. For all f ∈ L2(V × V ) and (q, v) ∈ Aff(V ), we have OpKN(πV × V (q, v)f ) = πV (q, v) OpKN(f )πV (q, v)∗. (3.4) (3.5) In particular, the operators of the representation πV × V of Aff(V ) act on L2(V × V ) by algebra automorphisms for ⋆0. Proof. Take ϕ ∈ L2(V ). Then (cid:0) OpKN(πV × V (q, v)(f ))πV (q, v)ϕ(cid:1)(w) =ZV × V = q−1/2ZV × V eihξ,w−w′i(πV × V (q, v)f )(w, ξ)(πV (q, v)ϕ)(w′) dw′dξ eihξ,w−w′if (q−1(w′ − v), q−1♭ ξ)ϕ(q−1(w′ − v)) dw′dξ. (3.6) On the other hand, (cid:0)πV (q, v) OpKN(f )ϕ(cid:1)(w) = q−1/2(cid:0) OpKN(f )ϕ(cid:1)(q−1(w − v)) = q−1/2ZV × V eihξ,q−1(w−v)−w′if (w′, ξ)ϕ(w′) dw′dξ. Using that eihξ,q−1(w−v)−w′i = eihq♭ξ,w−vieihq♭ξ,−qw′i and that the Haar measure dw′dξ of V × V is invariant under the transformations (w′, ξ) 7→ (qw′, q♭ξ), we can write the last integral as q−1/2ZV × V eihξ,w−vieihξ,−w′if (q−1w′, q−1♭ ξ)ϕ(q−1w′) dw′dξ, and this is equal to (3.6) by translation invariance of the Haar measure dw′ on V . (cid:3) 16 3.2. Dual cocycles on subgroups of Aff(V ). Let Q ⋉ V ⊂ Aff(V ) be a subgroup satisfying the dual orbit Assumption 2.15. Let dQ(q) be a Haar measure on Q and ∆Q the modular function. Routine computations show that a left invariant Haar measure and modular function on G are respectively given by d(q, v) = dQ(q) dv q and ∆(q, v) = ∆Q(q) q . The measure class isomorphism: G → V × V : (q, v) 7→ (q, v).(0, ξ0) = (v, q♭ξ0), (3.7) intertwines the left action of G on itself with the action (3.3) on V × V . Since the Haar measure on V × V is invariant under the affine action of G given in (3.3), the pull-back by the map (3.7) of the Haar measure on V × V defines a nonzero left invariant Radon measure on G and thus is a scalar multiple of the Haar measure of G. We therefore may assume that the Haar measure on Q is normalized so that the G-equivariant linear operator Uφ : L2(V × V ) → L2(G) defined by ( Uφf )(q, v) := f(cid:0)v, φ(q)(cid:1), (3.8) where φ(q) = q♭ξ0, is unitary. Equivalently, our normalization is such that the Haar measure dξ on V is the push-forward under φ of the measure q−1dQ(q) on Q, that is, ZQ f (φ(q)) dQ(q) q =Z V f (ξ) dξ for all f ∈ L1( V ). (3.9) Note in passing that this uniquely determines the Haar measure on G: if we multiply dv by a scalar, then (3.9) and unitarity of FV force us to divide dQ(q) by the same scalar. Therefore we get a unitary quantization map (3.10) and if we denote by π the canonical representation of G ⊂ Aff(V ) on L2(V ), that is, π = πV G, then by Lemma 3.3 we have Op : L2(G) → HS(L2(V )), Op(f ) = OpKN( U ∗ φf ), Op(λgf ) = π(g) Op(f )π(g)∗. Lemma 3.4. The representation π of G on L2(V ) is irreducible, square-integrable and the associated Duflo -- Moore formal degree operator is K = F ∗ V M (∆−1 ◦ φ−1) FV . Since ∆ is a V -invariant function on G, it is naturally viewed as a function on Q. Therefore by M (∆−1 ◦ φ−1) we mean the operator of multiplication by the function ξ 7→ ∆(φ−1(ξ))−1 = φ−1(ξ)/∆Q(φ−1(ξ)). Proof. It is more convenient to work on L2(Q) with the equivalent representation π given by π(q, v) := V π(q, v) V ∗, where V : L2(V ) → L2(Q), (Vϕ)(q) := q−1/2(FV ϕ)(q♭ξ0). A quick computation shows that (π(q, v)ϕ)(q′) = e−ihφ(q′),vi ϕ(q−1q′). The restriction of the representation π to V is simply the regular representation. It follows that any operator in B(L2(V )) commuting with π(G) must belong to π(V )′′. Passing to the equivalent representation π, this means that any operator in B(L2(Q)) commuting with π(G) must be an operator of multiplication by a function in L∞(Q). In addition this function must be invariant under the left translations on Q, hence it is constant. Thus π is irreducible. Turning to square-integrability, for ϕ1 ∈ Cc(Q) and ϕ2 ∈ L2(Q), we have (π(q, v)ϕ1, ϕ2) =ZQ =Z V e−ihφ(q′),vi ϕ1(q−1q′)ϕ2(q′) dQ(q′) e−ihξ,vi ϕ1(cid:0)q−1φ−1(ξ)(cid:1) ϕ2(cid:0)φ−1(ξ)(cid:1) φ−1(ξ) dξ, 17 with the second equality following from (3.9). If we set fq(ξ) := ϕ1(cid:0)q−1φ−1(ξ)(cid:1) ϕ2(cid:0)φ−1(ξ)(cid:1) φ−1(ξ), then, for every fixed q ∈ Q, the function fq has compact essential support and is bounded. Hence fq ∈ L2( V ) and (π(q, v)ϕ1, ϕ2) = (F V fq)(v). The Plancherel formula gives then ZG (π(q, v)ϕ1, ϕ2)2 dQ(q)dv q =ZQ(cid:16)ZV =ZQ(cid:16)Z V =ZQ×Q =ZQ×Q =ZQ q (F V fq)(v)2 dv(cid:17) dQ(q) fq(ξ)2 dξ(cid:17) dQ(q) ϕ1(cid:0)q−1φ−1(ξ)(cid:1)2 ϕ2(cid:0)φ−1(ξ)(cid:1)2 φ−1(ξ)2 dQ(q) dξ q q ϕ1(q−1q′)2 ϕ2(q′)2 q−1q′ dQ(q) dQ(q′) ϕ1(q)2 q ∆Q(q) dQ(q)ZQ ϕ2(q′)2 dQ(q′). This shows that π is square-integrable and the Duflo -- Moore operator is the operator of multi- plication by the function q 7→ q/∆Q(q). This gives the result. (cid:3) The next natural question is whether (B(L2(V )), Ad π) is a G-Galois object, or equivalently, by Theorem 2.13, whether π is quasi-equivalent to the regular representation. If it is, then we can construct a dual unitary 2-cocycle on G by the procedure described in Section 2.2. Instead of exactly following that procedure, however, we will construct this cocycle directly from the product ⋆ on L2(G) defined by f1 ⋆ f2 := Op∗(Op(f1) Op(f2)) = Uφ(cid:0) U ∗ φf1 ⋆0 U ∗ φf2(cid:1). (3.11) According to Remark 2.11 this approach should not necessarily work, but if it does, it has some technical advantages. Let us start by observing that by definition the algebra (L2(G), ⋆) is unitarily isomorphic to the algebra HS(L2(V )) of Hilbert -- Schmidt operators. Hence (3.12) We will need explicit formulas for the product ⋆ on dense subspaces of (L2(G), ⋆). First, we kf1 ⋆ f2k2 ≤ kf1k2kf2k2 for all f1, f2 ∈ L2(G). introduce an auxiliary space: Definition 3.5. Let E(G) be the Banach space completion of Cc(G) with respect to the norm Lemma 3.6. For any f1, f2 ∈ E(G) and a.a. (q, v) ∈ G, we have f (q, v)dv +ZQ sup q∈Q f (q, v) sup v∈V dQ(q) q . kf kE := kf k1 + kf k2 +ZV (f1 ⋆ f2)(q, v) =ZG (cid:0)FV (f1 ⋆ f2)(cid:1)(q, ξ) =Z V and eihq′♭ξ0−ξ0,v′if1(cid:0)(q, v)(1, v′)(cid:1) f2(cid:0)(q, v)(q′, 0)(cid:1) d(q′, v′), (FV f1)(q, ξ′) (FV f2)(cid:0)φ−1(φ(q) − ξ′), ξ − ξ′(cid:1) dξ′. (3.13) Here, following our conventions, FV : L2(G) = L2(Q ⋉ V ) → L2(Q × V ) is the partial Fourier transform in V -variables. Note that for this map to be unitary we have to equip Q × V with the measure q−1dQ(q) dξ (which in general is not the Haar measure of the semidirect product Q ⋉ V for the dual action). 18 Proof. For f ∈ L2(G), we let Kf ∈ L2(V × V ) be the kernel of the Hilbert -- Schmidt operator Op(f ). From (3.1) and (3.10) we get Kf (v, v′) =(cid:0)(1 ⊗ F ∗ V ) U ∗ φf(cid:1)(v′, v − v′) for v, v′ ∈ V. If f1, f2 ∈ E(G), then, since Op(f1 ⋆ f2) = Op(f1)Op(f2), the product formula for operator kernels gives Kf1⋆f2(v, v′) =ZV Kf1(v, w)Kf2 (w, v′) dw =ZV (cid:0)(1 ⊗ F ∗ =ZV (cid:16)Z V =ZV × V × V V ) U ∗ φf1(cid:1)(w, v − w)(cid:0)(1 ⊗ F ∗ eihξ1,v−wif1(φ−1(ξ1), w) dξ1(cid:17)(cid:16)Z V V ) U ∗ φf2(cid:1)(v′, w − v′) dw eihξ2,w−v′if2(φ−1(ξ2), v′) dξ2(cid:17)dw eihξ1,v−wieihξ2,w−v′if1(φ−1(ξ1), w)f2(φ−1(ξ2), v′) dw dξ1 dξ2, where the last step is justified by Fubini's theorem: note that for any (v, v′) ∈ V × V the function V × V × V → C, (w, ξ1, ξ2) 7→ eihξ1,v−wieihξ2,w−v′if1(φ−1(ξ1), w)f2(φ−1(ξ2), v′), belongs to L1(V × V × V ) and its L1-norm is not greater than kf1k1kf2kE . Next, still for f1, f2 ∈ E(G), we put f (q′, v) :=ZG =ZG =ZG eihq♭ξ0−ξ0,wif1(cid:0)(q′, v)(1, w)(cid:1) f2(cid:0)(q′, v)(q, 0)(cid:1) d(q, w) eihq♭ξ0−ξ0,wif1(q′, q′w + v) f2(q′q, v) d(q, w) eihq♭ξ0−q′♭ξ0,w−vif1(q′, w) f2(q, v) d(q, w), (3.14) with absolutely converging integrals. It is easy to see that f ∈ L2(G) and kf k2 2 ≤ kf1k1kf2k1ZV f1(q, v)dvZQ sup q∈Q sup v∈V f2(q, v) dQ(q) q ≤ kf1k2 E kf2k2 E . In particular, we can compute Kf , the kernel of the operator Op(f ): Kf (v, v′) =(cid:0)(1 ⊗ F ∗ V ) U ∗ φf(cid:1)(v′, v − v′) =Z V eihξ,v−v′i f (φ−1(ξ), v′) dξ =Z V eihξ,v−v′i(cid:16)ZG eihq♭ξ0−ξ,w−v′if1(φ−1(ξ), w) f2(q, v′) d(q, w)(cid:17) dξ, which by Fubini (and a simplification of the phases) becomes Kf (v, v′) =Z V ×G =Z V ×V × V eihξ,v−wieihq♭ξ0,w−v′if1(φ−1(ξ), w) f2(q, v′) d(q, w) dξ eihξ,v−wieihξ2,w−v′if1(φ−1(ξ), w) f2(φ−1(ξ2), v′) dξ dw dξ2. Hence Kf1⋆f2 = Kf , and (3.13) follows by injectivity of the map L2(G) → L2(V × V ), f 7→ Kf . 19 To get the second formula in the formulation of the lemma, we apply the partial Fourier transform to (3.14) and using Fubini's theorem one more time obtain (cid:0)FV (f1 ⋆ f2)(cid:1)(q′, ξ) =ZG =ZQ =Z V This concludes the proof. eihφ(q)−φ(q′),wif1(q′, w) (FV f2)(q, ξ + φ(q) − φ(q′)) d(q, w) (FV f1)(q′, −φ(q) + φ(q′)) (FV f2)(q, ξ + φ(q) − φ(q′)) dQ(q) q (FV f1)(q′, ξ′) (FV f2)(cid:0)φ−1(φ(q′) − ξ′), ξ − ξ′(cid:1) dξ′. (cid:3) Corollary 3.7. Cc(G) is a subalgebra of (L2(G), ⋆). Proof. Since Cc(G) ⊂ E(G), the result follows from formula (3.14) which clearly entails that when f1, f2 ∈ Cc(G), then also f1 ⋆ f2 ∈ Cc(G). (cid:3) Next we consider a space of functions with good behavior in the partial Fourier space: Definition 3.8. For a measure space (X, µ), we let L(X, µ) be the subspace of L∞(X, µ) consisting of functions that are (essentially) zero outside a set of finite measure. We then let FL(G) be the subspace of L2(G) consisting of functions of the form F ∗ V f with f ∈ L(Q × V , q−1dQ(q) dξ). Lemma 3.9. FL(G) is a subalgebra of (L2(G), ⋆). Moreover, for any f1, f2 ∈ FL(G) and a.a. (q, ξ) ∈ Q × V , we have (cid:0)FV (f1 ⋆ f2)(cid:1)(q, ξ) =Z V (FV f1)(q, ξ′) (FV f2)(cid:0)φ−1(φ(q) − ξ′), ξ − ξ′(cid:1) dξ′. (3.15) Proof. Let Kj ⊂ Q and Kj ⊂ V (j = 1, 2) be Borel sets of finite measure such that fj is (essentially) zero outside Kj × Kj. Then the function defined by the right hand side of (3.15) is zero for (q, ξ) outside K1 × ( K1 + K2). Therefore if (3.15) holds, then f1 ⋆ f2 ∈ FL(G). Turning to the proof of (3.15), we already know from Lemma 3.6 that this identity holds for f if hn → f in the L2-norm, FV hn → FV f a.e., and fj ∈ E(G) ∩ FL(G). Let us write hn the sequence (FV hn)n is dominated by a function in L1(Q × V ) ∩ L∞(Q × V ). Assume that we can find functions hj,n ∈ FL(G) such that hj,n fj and τ−→ n τ−→ n (cid:0)FV (h1,n ⋆ h2,n)(cid:1)(q, ξ) =Z V (FV h1,n)(q, ξ′) (FV h2,n)(cid:0)φ−1(φ(q) − ξ′), ξ − ξ′(cid:1) dξ′ (3.16) for almost all (q, ξ). By (3.12) we have h1,n ⋆ h2,n → f1 ⋆ f2 in L2(G) as n → ∞, hence the left hand side of (3.16) (considered as a function in (q, ξ)) converges to FV (f1 ⋆ f2) in the L2- norm. On the other hand, the right hand side converges to the right hand side of (3.15) by the dominated convergence theorem. Therefore to finish the proof it suffices to show that for every f ∈ FL(G) there exists a sequence of functions hn ∈ E(G) ∩ FL(G) such that hn f . First for all, if (Kn)n is an increasing sequence of compact subsets of Q× V with union Q× V , f . Therefore it suffices to consider f ∈ FL(G) V (f 1Kn) ∈ FL(G) and F ∗ then F ∗ such that FV f is compactly supported. V (f 1Kn) τ−→ n Let K be any compact such that its interior contains the support of FV f . By Lusin's theorem, we can find a uniformly bounded sequence of continuous functions gn supported in K such that gn → FV f a.e. Then F ∗ f . Therefore it suffices to consider f such that FV f ∈ Cc(Q × V ). V (gn) ∈ FL(G) and F ∗ V (gn) In a similar fashion, by approximating functions in Cc(Q × V ) by elements of the algebraic V h for some tensor product Cc(Q) ⊗ Cc( V ), we may assume that f = F ∗ V (g ⊗ h) = g ⊗ F ∗ τ−→ n 20 τ−→ n g ∈ Cc(Q) and h ∈ Cc( V ). Finally, by approximating h by the convolution of two functions, we V (h1 ∗ h2) for g ∈ Cc(Q) and hi ∈ Cc( V ). But such a function is may assume that f = g ⊗ F ∗ already in E(G). (cid:3) It follows from (3.13) that a candidate for the dual cocycle on G defining the product ⋆ by formula (2.5) is given by Ω :=ZG e−ihq♭ξ0−ξ0,viλ(1,v)−1 ⊗ λ(q,0)−1 d(q, v). (3.17) For the moment this is just a formal expression, but it at least makes sense as a sesquilinear form Ω on Cc(G × G): Ω(ϕ1, ϕ2) :=ZG e−ihq♭ξ0−ξ0,vi(cid:0)(λ(1,v)−1 ⊗ λ(q,0)−1 )ϕ1, ϕ2(cid:1) d(q, v) for ϕ1, ϕ2 ∈ Cc(G × G). Our first goal is to prove that Ω makes sense as a unitary operator on L2(G × G). This will be proven by showing that Ω factorizes as a product of three unitaries. Consider the following almost everywhere defined measurable transformation: Ξ : Q × V × G → Q × V × G, (q, ξ, g) 7→(cid:0)q, ξ, φ−1(ξ0 + ξ)g(cid:1). The operator UΞ on L2(cid:0)Q × V × G, q−1dQ(q) dξ dg(cid:1) mapping f into f ◦ Ξ is unitary. We then have: Lemma 3.10. The convolution operator Ω factorizes as follows: Ω = (F ∗ V ⊗ 1) UΞ (FV ⊗ 1). Proof. For ϕ1, ϕ2 ∈ Cc(G × G), the function (q, v; g1, g2) 7→ ϕ1(cid:0)(1, v)g1; (q, 0)g2(cid:1)ϕ2(cid:0)g1, g2(cid:1), belongs to L1(G3). Hence, we may use Fubini's theorem to write Ω(ϕ1, ϕ2) as follows: ZG3 e−ihq♭ξ0−ξ0,viϕ1(cid:0)(1, v)(q1, v1); (q, 0)(q2, v2)(cid:1) ϕ2(cid:0)q1, v1; q2, v2(cid:1) d(q, v) d(q1, v1) d(q2, v2) =ZG3 e−ihq♭ξ0−ξ0,viϕ1(cid:0)q1, v + v1; qq2, qv2(cid:1)ϕ2(cid:0)q1, v1; q2, v2(cid:1) d(q, v) d(q1, v1) d(q2, v2) =ZG3 e−ihq♭ξ0−ξ0,v−v1iϕ1(cid:0)q1, v; qq2, qv2(cid:1)ϕ2(cid:0)q1, v1; q2, v2(cid:1) d(q, v) d(q1, v1) d(q2, v2) =ZQ3×V (cid:0)(FV ⊗ 1)ϕ1(cid:1)(cid:0)q1, q♭ξ0 − ξ0; qq2, qv2(cid:1)(cid:0)(FV ⊗ 1)ϕ2(cid:1)(cid:0)q1, q♭ξ0 − ξ0; q2, v2(cid:1) dQ(q1) dQ(q2) dv2 dQ(q) × q1 q2 q =ZQ× V ×Q×V (cid:0)(FV ⊗ 1)ϕ1(cid:1)(cid:0)q1, ξ; φ−1(ξ0 + ξ)q2, φ−1(ξ0 + ξ)v(cid:1)(cid:0)(FV ⊗ 1)ϕ2(cid:1)(cid:0)q1, ξ; q2, v(cid:1) which completes the proof. × dQ(q1) dξ dQ(q2) dv q1 q2 , (cid:3) Next, by the definition of Ω it is clear that Ω commutes with the operators ρg ⊗ 1 and 1 ⊗ ρg. Hence Ω ∈ W ∗(G) ¯⊗W ∗(G). Lemma 3.11. For all f1, f2 ∈ A(G) ∩ L2(G), we have f1 ⋆ f2 = (f1 ⊗ f2)( ∆(·)Ω∗). 21 Proof. Recall that A(G) consists of functions of the form ϕ1 ∗ ϕ2, with ϕi ∈ L2(G), which correspond to the linear functionals (· ϕ2, ¯ϕ1) on W ∗(G). Under the identification A(G) ≃ W ∗(G)∗, Lemma 3.6 says that if f1, f2 ∈ E(G) ∩ A(G), then (f1 ⋆ f2)(g) =ZG eihq′♭ξ0−ξ0,v′i(f1 ⊗ f2)(cid:0)λg(1,v′) ⊗ λg(q′,0)(cid:1) d(q′, v′). Now, for fj = ϕj ∗ ϕ′ j, with ϕj, ϕ′ j ∈ Cc(G), we have (f1 ⊗ f2)( ∆(g)Ω∗) = ( ∆(g)Ω∗(ϕ′ 1 ⊗ ϕ′ 2), ¯ϕ1 ⊗ ¯ϕ2). Using the initial definition of Ω as a bilinear form on Cc(G × G), we get (f1 ⊗ f2)( ∆(g)Ω∗) =ZG =ZG eihq′♭ξ0−ξ0,v′i(cid:0)(λg(1,v′) ⊗ λg(q′,0))(ϕ′ eihq′♭ξ0−ξ0,v′i(f1 ⊗ f2)(cid:0)λg(1,v′) ⊗ λg(q′,0)(cid:1) d(q′, v′). 1 ⊗ ϕ′ 2), ¯ϕ1 ⊗ ¯ϕ2(cid:1) d(q′, v′) Hence the equality in the formulation of the lemma holds for all f1, f2 of the form ϕ1 ∗ ϕ2, with ϕi ∈ Cc(G). Therefore in order to prove the lemma it suffices to show that every function f ∈ A(G) ∩ L2(G) can be approximated by functions of the form ϕ1 ∗ ϕ2, with ϕi ∈ Cc(G), simultaneously in the norms on A(G) and L2(G). Consider first a function of the form f = f1 ∗ f2, with f1 ∈ L2(G) and f2 ∈ Cc(G). If ϕn → f1 in L2(G), ϕn ∈ Cc(G), then ϕn ∗ f2 → f1 ∗ f2 both in A(G) and L2(G). Consider now an arbitrary f ∈ A(G) ∩ L2(G). By the previous case in order to finish the proof it suffices to show that f can be approximated simultaneously in A(G) and L2(G) by functions of the form f ∗ ϕ, with ϕ ∈ Cc(G). Take a standard approximate unit (ϕn)n in L1(G) consisting of functions ϕn ∈ Cc(G) such that ϕn ≥ 0, RG ϕn dg = 1, with the supports of ϕn eventually contained in arbitrarily small neighbourhoods of the unit. Then f ∗ ϕn → f in L2(G). At the same time, if we write f as f1 ∗ f2 for some fi ∈ L2(G) and use that f ∗ ϕn = f1 ∗ (ϕn ∗ f2) and ϕn ∗ f2 → f2 in L2(G), we also see that f ∗ ϕn → f in A(G). (cid:3) We thus see that A(G) ∩ L2(G) is a subalgebra of (L2(G), ⋆). Since A(G) ∩ L2(G) is dense in A(G), the associativity of the product ⋆ on this subalgebra implies that Ω satisfies the cocycle identity (2.4). To summarize, we have proved the following result. Theorem 3.12. For any second countable locally compact group G = Q ⋉ V satisfying the dual orbit Assumption 2.15, formula (3.17) defines a dual unitary 2-cocycle Ω on G. The corre- sponding product ⋆Ω on A(G) coincides on A(G) ∩ L2(G) with the product ⋆ defined by (3.11). Remark 3.13. If we started from the opposite Kohn -- Nirenberg quantization (see Remark 3.1) we would have obtained the following dual 2-cocycle: Ω :=ZG eihq♭ξ0−ξ0,viλ(q,0)−1 ⊗ λ(1,v)−1 d(q, v), (3.18) which differs from Ω by the inversion of legs and by the sign of the phase: Ω = (J ⊗ J)Ω21(J ⊗ J) = ( R ⊗ R)(Ω∗ 21), where R is the unitary antipode of W ∗(G) given on the generators by R(λg) = λg−1. By [11, Proposition 6.3, iii)], the dual cocycles Ω and Ω are cohomologous, with the unitary opera- tor implementing the cohomological relation equal to J J, which we will explicitly compute in Section 3.5. 22 3.3. Identification of the Galois objects. To complete the picture it remains to check that the Galois object defined by the dual cocycle Ω is exactly the pair (B(L2(V )), Ad π). For f ∈ L2(G), consider the operator L⋆(f ) on L2(G) defined by By (3.12) we have L⋆(f )f ′ = f ⋆ f ′. kL⋆(f )k ≤ kf k2. Furthermore, under the identification of (L2(G), ⋆) with HS(L2(V )) via Op, the map L⋆ is simply the left regular representation of HS(L2(V )) on itself. This representation is a multiple of the canonical representation of HS(L2(V )) on L2(V ). It follows that there is a unique isomorphism B(L2(V )) ∼= L⋆(L2(G))′′ such that Op(f ) 7→ L⋆(f ). (3.19) Therefore in order to find an isomorphism W ∗( G; Ω) ∼= B(L2(V )) it suffices to find an isomor- phism W ∗( G; Ω) ∼= L⋆(L2(G))′′. From formula (2.7) for the GNS-map on W ∗( G; Ω), for every f ∈ A(G) we have the following equality: πΩ(f )ϕ = S L⋆(f ) Sϕ for all ϕ ∈ A(G)∩ L2(G) such that the right hand side is well-defined, where S is the unbounded operator defined by Sϕ = ϕ. In other words, using the unitary operator J defined in (1.1), J = J J = JJ = M (∆−1/2)S = SM (∆1/2), we have πΩ(f )ϕ = J M (∆−1/2) L⋆(f ) M (∆1/2) J ϕ (3.20) for all ϕ ∈ A(G) ∩ L2(G) such that the right hand side is well-defined. We thus see that we need to understand a connection between the operators L⋆(f ) and M (∆s). For this we will first get another useful formula for L⋆. First, we denote by Uφ the variant of the unitary operator Uφ, defined in (3.8), without the permutations of variables: Uφ : L2( V × V ) → L2(G), (Uφ)f (q, v) := f(cid:0)φ(q), v(cid:1). Then we denote by γ the unitary representation of V on L2(G) given by γ(ξ) = UφF ∗ V (τξ ⊗ τξ)FV U ∗ φ, (3.21) (3.22) where τξ : L2( V ) → L2( V ) is the left regular representation of V given by (τξϕ)(ξ′) = ϕ(ξ′ − ξ). Next, for fixed ξ ∈ V and f ∈ FL(G), we denote by (FV f )(•, ξ) the V -invariant function on G given by (cid:2)(q, v) 7→ (FV f )(q, ξ)(cid:3) and by M(cid:0)(FV f )(•, ξ)(cid:1) the bounded operator on L2(G) of multiplication by the function (FV f )(•, ξ). Lemma 3.9 then leads to the following result. Lemma 3.14. For any f ∈ FL(G), we have the absolutely convergent (in the operator norm) integral formula L⋆(f ) =Z V M(cid:0)(FV f )(•, ξ)(cid:1) γ(ξ) dξ, (3.23) Finally, we introduce a family (Tz)z∈C of operators on functions on G by (Tzf )(q, v) =Z V ∆(φ−1(ξ0 − q−1♭ ξ))−z eihξ,vi (FV f )(q, ξ) dξ, (3.24) where we remind that ∆(q′) = ∆Q(q′)/q′. We need a dense subspace of FL(G) preserved by these operators: 23 Definition 3.15. For compact subsets K, L ⊂ Q, let LK,L(Q × V ) be the subspace of L(Q × V ) consisting of functions supported on the compact set (cid:8)(q, ξ) ∈ Q × V q ∈ K, ξ0 − q−1♭ (3.25) We denote by L0(Q × V ) the union of the spaces LK,L(Q × V ) and by FL0(G) (respectively, by FLK,L(G)) the subspace of FL(G) consisting of functions f ∈ FL(G) such that FV f belongs to L0(Q × V ) (respectively, to LK,L(Q × V )). Lemma 3.16. The space FL0(G) is dense in L2(G) and A(G) ∩ FL0(G) is dense in A(G). Moreover, FL0(G) is stable under Tz and Tz(A(G) ∩ FL0(G)) is dense in L2(G) for any z ∈ C. ξ ∈ φ(L)(cid:9). Proof. By definition (3.24), the operator Tz conjugated by the partial Fourier transform is the operator of multiplication by the function (q, ξ) 7→ ∆(cid:0)φ−1(ξ0 − q−1♭ ξ)(cid:1)−z. This immediately shows that FLK,L(G) is stable under Tz as the modular function ∆ is bounded, as well as bounded away from zero, on any compact subset of Q. Hence FL0(G) is also stable under Tz. Since FV is unitary, to prove density of FL0(G) in L2(G) it suffices to show that L0(Q × V ) is dense in L2(Q × V , q−1dQ(q) dξ). For this it suffices to show that for every compact set K ⊂ Q the union of the sets (3.25) over the compact sets L ⊂ Q is a subset of K × V of full measure. But this is clear, since this union is and by assumption φ(Q) is a subset of V of full measure. (cid:8)(q, ξ) ∈ Q × V q ∈ K, ξ ∈ q♭ξ0 − φ(Q)(cid:9) Since the map f 7→ f is bounded on L2(K × V, q−1dQ(q) dξ) ⊂ L2(G), with image L2(K −1 × V, q−1dQ(q) dξ), the functions f for f ∈ FL0(G) form a dense subspace of L2(G) as well. Hence the functions ϕ∗f for ϕ ∈ Cc(G) and f ∈ FL0(G) are dense in A(G). An easy calculation shows that for any g = (q, v) ∈ G and compacts K, L ⊂ Q, we have λg(FLK,L(G)) ⊂ FLqK,L(G). Hence, if f ∈ FLK,L(G) and ϕ ∈ Cc(G) has support contained in U × V for a compact set U ⊂ Q, then ϕ ∗ f ∈ FLU K,L(G). Therefore ϕ ∗ f ∈ A(G) ∩ FL0(G) for all ϕ ∈ Cc(G) and f ∈ FL0(G). Thus A(G) ∩ FL0(G) is dense in A(G). Taking a standard approximate unit in L1(G) for ϕ, we see also that functions of the form ϕ ∗ f , for ϕ ∈ Cc(G) and f ∈ FL0(G), are dense in FL0(G), hence in L2(G). Moreover, since the operators Tz are bounded on the spaces FLK,L(G), we may also conclude that the functions Tz(ϕ ∗ f ) are dense in L2(G) for all z. In particular, Tz(A(G) ∩ FL0(G)) is dense in L2(G) for all z. (cid:3) Proposition 3.17. The operator M (∆) is affiliated with the von Neumann algebra L⋆(L2(G))′′. Moreover, for all f ∈ FL0(G) and z ∈ C, we have M (∆z)L⋆(f ) = L⋆(∆zf ) on L2(G), M (∆z)L⋆(f )M (∆−z) = L⋆(Tzf ) on Cc(G). Proof. Since ∆ depends only on the coordinate Q, the operators M (∆it), t ∈ R, commute with the partial Fourier transform FV . From formula (3.15) we see then that M (∆it)L⋆(f ) = L⋆(∆itf ) for all f ∈ FL(G). As L⋆(FL0(G)) is σ-weakly dense in L⋆(L2(G))′′, this implies that M (∆) is affiliated with L⋆(L2(G))′′. The same formula also shows that since ∆zf ∈ FL0(G) for f ∈ FL0(G), we have M (∆z)L⋆(f ) = L⋆(∆zf ) on L2(G). Next, formula (3.23) and definition (3.24) of Tz give, for f ∈ FL0(G), the absolutely conver- gent integral L⋆(Tzf ) =Z V M(cid:16)∆(cid:0)φ−1(ξ0 − •−1♭ 24 ξ)(cid:1)−z(FV f )(•, ξ)(cid:17) γ(ξ) dξ. (3.26) On the other hand, using again (3.23), we have on ∆ℜ(z)L2(G): L⋆(f )M (∆−z) =Z V M(cid:0)(FV f )(•, ξ)(cid:1) γ(ξ)M (∆−z) dξ. Now, for ϕ ∈ ∆ℜ(z)L2(G), we have (γ(ξ)M (∆−z)ϕ)(q, v) =(cid:0)UφF ∗ V (τξ ⊗ τξ)FV U ∗ φ(∆−zϕ)(cid:1)(q, v). As FV commutes with operators of multiplication by V -invariant functions, from this we easily deduce: Hence we have (cid:0)γ(ξ)M (∆−z)ϕ(cid:1)(q, v) = ∆(φ−1(q♭ξ0 − ξ))−z(γ(ξ)ϕ)(q, v) = ∆(q)−z∆(cid:0)φ−1(ξ0 − q−1♭ γ(ξ)M (∆−z) = M (∆−z)M(cid:0)∆(φ−1(ξ0 − •−1♭ ξ)−z)(cid:1) γ(ξ), ξ))−z(γ(ξ)ϕ)(q, v). which by closedness of M (∆−z) finally gives L⋆(f )M (∆−z) = M (∆−z)Z V M(cid:16)∆(cid:0)φ−1(ξ0 − •−1♭ ξ(cid:1)−z(cid:17)(FV f )(•, ξ)(cid:19) γ(ξ) dξ on L2(G) ∩ ∆ℜ(z)L2(G). Together with (3.26) this shows that the identity holds on L2(G) ∩ ∆ℜ(z)L2(G), in particular, on Cc(G). L⋆(f )M (∆−z) = M (∆−z)L⋆(Tzf ) This proposition and identity (3.20) imply that for any f ∈ A(G) ∩ FL0(G) we have πΩ(f ) = J L⋆(T−1/2f )J (cid:3) (3.27) on Cc(G) ∗ Cc(G) ⊂ A(G) ∩ Cc(G), hence on L2(G), as both sides of the identity are bounded operators. Since A(G) ∩ FL0(G) is dense in A(G) and T−1/2(A(G) ∩ FL0(G)) is dense in L2(G) by Lemma 3.16, it follows that W ∗( G; Ω) = J L⋆(L2(G))′′J . Recalling also that the action of G on W ∗( G; Ω) is given by the automorphisms Ad ρg, we see that this action transforms under the isomorphism Ad J of W ∗( G; Ω) onto L⋆(L2(G))′′ into the action given by the automorphisms Ad λg. Using the isomorphism (3.19) the latter action transforms, in turn, into the action Ad π on B(L2(V )). We therefore get the required isomorphism of the Galois objects: Theorem 3.18. For any second countable locally compact group G = Q ⋉ V satisfying the dual orbit Assumption 2.15, the G-Galois object (W ∗( G; Ω), β) defined by the dual cocycle (3.17) is isomorphic to (B(L2(V )), Ad π); explicitly, the isomorphism maps πΩ(f ), f ∈ A(G) ∩ FL0(G), into Op(T−1/2f ). As a byproduct we see that (B(L2(V )), Ad π) is indeed a Galois object. We remind that by Theorem 2.13 this is therefore the unique up to isomorphism I-factorial Galois object defined by a genuine representation of G. Remark 3.19. Formula (3.23) shows that it is natural to extend the representation L⋆ to a larger class of functions including all measurable functions on Q (viewed as functions on G) by letting L⋆(f ) = M (f ) for such functions. With this definition we have ∆s ⋆ ∆t = ∆s+t and, by Proposition 3.17, ∆s ⋆ f = ∆sf , f ⋆ ∆s = ∆sT−sf for f ∈ FL0(G). We are therefore exactly in the situation discussed in Remark 2.11, with identities (2.13) satisfied for c = 1 and α = 1/2. This "explains" why we were able to construct the dual cocycle Ω using the quantization map Op instead of the modified quantization map Op′. 25 3.4. Deformation of the trivial cocycle. We continue to consider a second countable locally compact group G = Q⋉V satisfying the dual orbit Assumption 2.15 and a dual unitary 2-cocycle Ω on G defined by (3.17). By replacing the distinguished point ξ0 ∈ V by ξq 0 := q♭ξ0 we in fact get a family of such dual cocycles Ωq indexed by the elements q ∈ Q. We already know that all these cocycles are cohomologous, since they correspond to the unique Galois object defined by a genuine representation. This is also easy to see as follows. Proposition 3.20. We have Ωq = (λq ⊗ λq)Ω ∆(λq)∗ for all q ∈ Q. In particular, the map q 7→ Ωq is continuous in the so-topology. Proof. Considering as before the partial Fourier transform FV as a map L2(G) → L2(Q × V , q−1dQ(q) dξ), we have (FV λqF ∗ V f )(q′, ξ′) = qf (q−1q′, q−1♭ ξ′). From this and Lemma 3.10 we get (λq ⊗ λq)Ω ∆(λq)∗ = (F ∗ where Ξq : Q × V × G → Q × V × G is defined by V ⊗ 1)UΞq (FV ⊗ 1), Ξq(q′, ξ′, g) := (q′, ξ′, qφ−1(ξ0 + q−1♭ ξ′)q−1g). On the other hand, consider the map φq : Q → V defined by ξq 0 = q♭ξ0, so that φq(q′) := q′ξq 0 = φ(q′q). Then qφ−1(ξ0 + q−1♭ ξ′)q−1 = φ−1(ξq 0 + ξ′)q−1 = φ−1 q (ξq 0 + ξ′), which shows that Ξq is exactly the map from the expression for Ωq in Lemma 3.10 (with ξ0 replaced by ξq (cid:3) 0). This proves the proposition. Although the dual cocycles Ωq are all cohomologous, under quite general assumptions they can be used to construct a continuous deformation of the trivial cocycle. Namely, we have the following result. Proposition 3.21. Assume that G = Q ⋉ V in addition to satisfying the dual orbit Assump- tion 2.15 is such that the map φ : Q → V is open. Assume also that there exists a sequence (zn)n of elements of the center Z(Q) of Q such that z♭ → 1 in the so-topology. n → 0 in End( V ) pointwise. Then Ωz−1 n → id a.e., or equivalently, φ−1(ξ0 + z♭ Proof. Using the notation from the proof of the previous proposition it suffices to show that nξ) → e for a.e. ξ ∈ V . But this is clear, since by Ξz−1 assumption the image of φ contains a neighbourhood of ξ0 and φ is a homeomorphism of Q onto its image. (cid:3) n Example 3.22. Consider the ax + b group G over the reals, so that Q = R∗, V = R, and Q acts on V by multiplication. In other words, G is the group of matrices (cid:26)(cid:18)a b 0 1(cid:19) a ∈ R∗, b ∈ R(cid:27) . We identify R with R via the pairing eixy. Then s♭t = s−1t for s ∈ Q and t ∈ V . Take −1 as ξ0. We then get a continuous family of cohomologous dual unitary 2-cocycles Ωθ, θ ∈ R∗, on G such that Ωθ = (F ∗ R ⊗ 1)UΞθ (FR ⊗ 1), where Ξθ : R∗ × R × G → R∗ × R × G is defined by Ξθ(s, t, g) :=(cid:0)s, t,(cid:18)(1 − θt)−1 0 1(cid:19) g(cid:1). 0 26 In this case the pointwise convergence θ♭ → 0 in End( V ) means that θ−1 → 0 in R. So by the above proposition we have Ωθ → 1 as θ → 0, which is obviously the case. 3.5. Multiplicative unitaries. Our next goal is to find an explicit formula for the multiplica- tive unitary of the twisted quantum group (W ∗(G), Ω ∆(·)Ω∗) for the dual cocycle Ω defined by (3.17). The main issue is to determine the modular conjugation J for the canonical weight ϕ on W ∗( G; Ω). It is more convenient to work with the isomorphic Galois object (L⋆(L2(G))′′, Ad λ). The algebra (L2(G), ⋆) becomes a ∗-algebra if we transport the ∗-structure on HS(L2(V )) to L2(G). Since the ∗-structure on the algebra of Hilbert -- Schmidt operators is isometric, the corresponding ∗-structure on L2(G) must have the form f 7→ U Jf for a unitary operator U on L2(G). In other words, the operator U is defined by the identity Op(f )∗ = Op(U Jf ). Lemma 3.23. The operator U is given by where Uφ is the operator defined by (3.21). U = UφF ∗ V W ∗ V FV U ∗ φ, Proof. Consider the unitary fOp : L2(V × V ) → HS(L2(V )) defined by Then by definition fOp(f ) is the integral operator with kernel (v, v′) 7→ f (v′, v − v′), hence its fOp(f ) = OpKN((1 ⊗ FV )f ). adjoint has the kernel (v, v′) 7→ f (v, v′ − v). In other words, where f #(v, v′) = f (v + v′, −v′). Since fOp(f )∗ = fOp(f #), f # = (JV ⊗ JV ) W ∗ V f, we conclude that Using that we arrive at U J = Uφ(1 ⊗ FV )(JV ⊗ JV ) W ∗ V (1 ⊗ F ∗ V ) U ∗ φ. (JV ⊗ JV ) W ∗ V = WV (JV ⊗ JV ) and JV F ∗ V = F ∗ V J V , Finally, using that Uφ = UφΣ and U J = Uφ(1 ⊗ FV ) WV (1 ⊗ F ∗ V ) U ∗ φJ. WV = Σ(F ∗ V ⊗ F ∗ V ) W ∗ V (FV ⊗ FV )Σ, we get the desired formula. (cid:3) Proposition 3.24. The modular conjugation for the canonical weight ϕ on the Galois object W ∗( G; Ω) (with respect to the GNS-map (2.7)) is J = J U JJ . Proof. The proof is similar to that of [5, Proposition 2.8]. Let us start with the canonical weight ϕL for the Galois object (L⋆(L2(G))′′, Ad λ). Note that since M (∆) is affiliated with L⋆(L2(G))′′ and the function ∆ is, up to a scalar factor, the only positive measurable function F on G such that λgF = ∆(g)−1F , the isomorphism L⋆(L2(G))′′ ∼= B(L2(V )), L⋆(f ) 7→ Op(f ), must map M (∆it) (t ∈ R) to citK −it for some c > 0, where K is the Duflo -- Moore operator of formal degree (explicitly given by Lemma 3.4). We have densely defined operators on HS(L2(V )) of multiplication on the right by czK −z (z ∈ C). Correspondingly, we have densely defined operators on L2(G), which we suggestively denote by f 7→ f ⋆ ∆z. Thus, by definition, L⋆(f ⋆ ∆z) = L⋆(f )M (∆z) 27 for f in a dense subspace of L2(G). Explicitly, by Proposition 3.17 we have f ⋆ ∆z = ∆zT−zf for f ∈ FL0(G). Recalling the description of the GNS-representation for (B(L2(V )), Ad π) in Section 2.2, and formula (2.10) in particular, we see that as the GNS-space for ϕL we can take L2(G), with the GNS-map ΛL : N ϕL → L2(G) uniquely determined by ΛL(L⋆(f )) = c1/2f ⋆ ∆−1/2 for f ∈ L2(G) such that the right hand side is well-defined. The corresponding modular conju- gation JL is simply given by the involution on L2(G) ∼= HS(L2(V )), so JL = U J. Now, using the isomorphism Ad J between W ∗( G; Ω) and L⋆(L2(G))′′, we can consider the space L2(G) as the GNS-space for ϕ using the map N ϕ ∋ x 7→ J ΛL(J xJ ). (3.28) In this picture the modular conjugation for ϕ is J U JJ . Therefore we only have to check that the above GNS-map is exactly the map Λ used to define J . Recall that Λ is given by for all f ∈ A(G) ∩ ∆1/2L2(G). Since by (3.27) we have Λ(πΩ(f )) = f = J ∆−1/2f J πΩ(f )J = L⋆(T−1/2f ) = L⋆((∆−1/2f ) ⋆ ∆1/2) for f ∈ A(G) ∩ FL0(G), we see that J ΛL(J πΩ(f )J ) = c1/2 Λ(πΩ(f )) for all such f . By comparing the norms of both sides we can already conclude that c = 1. Then, since any two GNS-representations associated with ϕ are unitary conjugate and the vectors Λ(πΩ(f )) = f , with f ∈ A(G) ∩ FL0(G), form a dense subspace of L2(G), it follows that the maps Λ and (3.28) are equal. (cid:3) As was shown in [11], the unitary operator J J must belong to W ∗(G). The following makes this explicit. Lemma 3.25. For any ϕ ∈ Cc(G) and g ∈ G we have: (J UJ ϕ)(g) =ZG eihq♭ξ0−ξ0,vi (λ(q,v)ϕ)(g) ∆(q)−1/2 d(q, v). Proof. We have, with absolutely convergent integrals: (U ϕ)(q, v) = (UφF ∗ V W ∗ V U ∗ φFV ϕ)(q, v) =Z V =Z V =Z V ×V =ZQ×V =ZG =ZG eihξ,vi ( W ∗ V φFV ϕ)(q♭ξ0, ξ) dξ U ∗ eihξ,vi (U ∗ φFV ϕ)(q♭ξ0 − ξ, ξ) dξ eihξ,v−v′i ϕ(cid:0)φ−1(q♭ξ0 − ξ), v′(cid:1) dξ dv′ e−ihq′♭ξ0−q♭ξ0,v−v′i ϕ(q′, v′) dQ(q′) dv′ q′ eih(qq′)♭ξ0−q♭ξ0,qv′i ϕ(cid:0)(q, v)(q′, v′)(cid:1) d(q′, v′) eihq′♭ξ0−ξ0,v′i ∆(q′)−1/2 (ρ(q′,v′)ϕ)(q, v) d(q′, v′). As J ρgJ = λg, applying this to J ϕ instead of ϕ we get the announced formula. (cid:3) 28 By [11, Proposition 5.4], we deduce that the multiplicative unitary WΩ for the deformed quantum group (W ∗(G), Ω ∆(·)Ω∗) is given by the formula WΩ = (J U JJ ⊗ J) Ω W ∗ (J ⊗ J) Ω∗. (3.29) Conjugating by the partial Fourier transform, we get a more explicit formula: Theorem 3.26. Let G = Q ⋉ V be a second countable locally compact group satisfying the dual orbit Assumption 2.15, and Ω be the dual unitary 2-cocycle defined by (3.17). Then for the multiplicative unitary WΩ of the deformed quantum group (W ∗(G), Ω ∆(·)Ω∗) and any f ∈ L2(G × G) we have V )f(cid:1)(q1, ξ1; q2, ξ2) = φ−1(φ(q−1 V ⊗ F ∗ 2ξ1; φ−1(φ(q−1 2 ) + ξ1)−1φ−1(ξ0 + ξ1), φ−1(φ(q−1 2 ) + ξ1) 2 ) + ξ1)−1♭ ♭ (q−1 2 ξ2 − ξ1)(cid:1). Proof. We know by Lemma 3.10 that for f ∈ L2(Q × V × G, q1−1dQ(q1) dξ1 dg), we have From this we obtain, for f ∈ L2(G × G): V ⊗ 1)f )(q1, ξ1; q2, v2) = f(cid:0)q1, ξ1; φ−1(ξ0 + ξ1)q2, φ−1(ξ0 + ξ1)v2(cid:1). V )f(cid:1)(q1, ξ1; q2, ξ2) = φ−1(ξ0+ξ1)−1f(cid:0)q1, ξ1; φ−1(ξ0+ξ1)q2, φ−1(ξ0+ξ1)♭ξ2(cid:1). (cid:0)(FV ⊗ FV ) WΩ(F ∗ × f(cid:0)q2q1, q♭ ((FV ⊗ 1)Ω(F ∗ V ⊗F ∗ It follows that (cid:0)(FV ⊗FV )Ω(F ∗ (cid:0)(FV ⊗ FV )Ω∗(F ∗ V ⊗ F ∗ V )f(cid:1)(q1, ξ1; q2, ξ2) On the other hand, with the help of Lemma 3.25 we can perform calculations similar to those = φ−1(ξ0 + ξ1)f(cid:0)q1, ξ1; φ−1(ξ0 + ξ1)−1q2, φ−1(ξ0 + ξ1)−1♭ ξ2(cid:1). of Lemma 3.10 to get, for f ∈ L2(Q × V , q−1dq(q) dξ), that (FV J UJ F ∗ V f )(q, ξ) = Moreover, one easily finds that φ−1(ξ0 + ξ)3/2 ∆Q(cid:0)φ−1(ξ0 + ξ)(cid:1)1/2 f(cid:0)φ−1(ξ0 + ξ)−1q, φ−1(ξ0 + ξ)−1♭ ξ(cid:1). (FV JF ∗ V f )(q, ξ) = f (q, −ξ), (FV J F ∗ V f )(q, ξ) = and (cid:0)(FV ⊗ FV ) W ∗(F ∗ V ⊗ F ∗ V )f(cid:1)(q1, ξ1; q2, ξ2) = q2f(cid:0)q−1 2 q1, q−1 2 Hence we get (cid:0)(FV ⊗ FV ) WΩ(F ∗ V ⊗ F ∗ V )f(cid:1)(q1, ξ1; q2, ξ2) q3/2 ∆Q(q)1/2 f(cid:0)q−1, q−1♭ξ(cid:1), ξ1; q2, ξ1 + ξ2(cid:1). ♭ (3.30) = V ⊗ F ∗ φ−1(ξ0 + ξ1)3/2 =(cid:0)(FV ⊗ FV )(J UJ ⊗ 1)(J ⊗ J) Ω W ∗ (J ⊗ J) Ω∗(F ∗ V )f(cid:1)(q1, ξ1; q2, ξ2) ∆Q(cid:0)φ−1(ξ0 + ξ1)(cid:1)1/2(cid:0)(FV ⊗ FV )(J ⊗ J ) Ω W ∗ (J ⊗ J) Ω∗(F ∗ V )f(cid:1) ξ1; q2, ξ2(cid:1) (cid:0)φ−1(ξ0 + ξ1)−1q1, φ−1(ξ0 + ξ1)−1♭ ∆Q(cid:0)φ−1(ξ0 + ξ1)(cid:1)1/2∆Q(q2)1/2(cid:0)(FV ⊗ FV )Ω W ∗ (J ⊗ J ) Ω∗(F ∗ (cid:0)φ−1(ξ0 + ξ1)−1q1, −φ−1(ξ0 + ξ1)−1♭ φ−1(ξ0 + ξ1)3/2q23/2 V ) ¯f(cid:1) ξ2(cid:1) ξ1; q−1 2 , q−1 2 V ⊗ F ∗ V ⊗ F ∗ = ♭ Using that φ−1(ξ0 − φ−1(ξ0 + ξ1)−1♭ ξ1) = φ−1(ξ0 + ξ1)−1, 29 the expression above becomes V ⊗ F ∗ V ) ¯f ) = φ−1(ξ0 + ξ1)5/2q23/2 φ−1(ξ0 + ξ1)3/2q21/2 (cid:0)φ−1(ξ0 + ξ1)−1q1, −φ−1(ξ0 + ξ1)−1♭ ∆Q(cid:0)φ−1(ξ0 + ξ1)(cid:1)1/2∆Q(q2)1/2(cid:0)(FV ⊗ FV ) W ∗ (J ⊗ J) Ω∗(F ∗ ∆Q(cid:0)φ−1(ξ0 + ξ1)(cid:1)1/2∆Q(q2)1/2(cid:0)(FV ⊗ FV )(J ⊗ J ) Ω∗(F ∗ q2(cid:0)(FV ⊗ FV )Ω∗(F ∗ V )f(cid:1)(cid:0)q2q1, q♭ 2ξ1; φ−1(ξ0 + ξ1)−1q−1 2 , φ−1(ξ0 + ξ1)−1♭ (cid:0)q2q1, −q♭ V ⊗ F ∗ (q−1 = 1 ♭ 2 ξ1; φ−1(ξ0 + ξ1)−1q−1 2ξ1; q2φ−1(ξ0 + ξ1), ξ2 − q♭ 2ξ1; φ−1(φ(q−1 2 ) + ξ1)−1φ−1(ξ0 + ξ1), φ−1(φ(q−1 ♭ q−1 2 ξ2(cid:1) V ⊗ F ∗ 2 , φ−1(ξ0 + ξ1)−1♭ V ) ¯f(cid:1) ξ2 − ξ1)(cid:1) 2ξ1(cid:1) 2 ) + ξ1)−1♭ (q−1 2 ♭ ξ2 − ξ1)(cid:1), (cid:3) = φ−1(φ(q−1 2 ) + ξ1) f(cid:0)q2q1, q♭ which is what we need. Recall that in Section 3.4 we considered a continuous family of cohomologous dual unitary 2-cocycles Ωq, q ∈ Q, defined by replacing ξ0 by ξq 0 = q♭ξ0. Corollary 3.27. We have: (i) the map Q ∋ q 7→ WΩq is so-continuous; (ii) if φ : Q → V is open and z♭ zn ∈ Z(Q), then WΩ z −1 n → W in the so-topology. n → 0 in End( V ) pointwise for a sequence of elements Proof. Part (i) follows already from formula (3.29). Indeed, the map q 7→ Ωq is continuous by Proposition 3.20. On the other hand, the unitary Uq in formula (3.29) for the dual cocycle Ωq is given by Lemma 3.23, with ξ0 replaced by ξq 0. To be more precise, that lemma is formulated under the assumption that the Haar measure on Q is normalized by (3.9). If we replace ξ0 by ξq 0 = q♭ξ0, and, correspondingly, the map φ by φq(q′) = φ(q′q), but want to keep the same measure on Q, then the map Uφq : L2( V × V ) → L2(G), Uφq f (q′, v′) := f(cid:0)φq(q′), v′(cid:1), is unitary only up to a scalar factor. But this means that for Lemma 3.23 to remain true we just have to state it as the equality Uq = Uφq F ∗ V W ∗ V FV U −1 φq . As the map q 7→ Uφq is obviously continuous in the so-topology, we conclude that the map q 7→ Uq is continuous as well, hence so is the map q 7→ WΩq . In order to prove (ii), recall that in the proof of Proposition 3.21 we already showed that if φ is open, then φ−1(ξ0 + z♭ nξ) → e as z♭ n → 0. It follows that, for all q ∈ Q, (q−1♭ φ−1 z−1 n ξz−1 0 + ξ) = q−1φ−1(ξ0 + z♭ n nξ) → q−1. By Theorem 3.26 we then conclude that (FV ⊗FV ) WΩ is given by z −1 n (F ∗ V ⊗F ∗ V ) → Y , where the operator Y (Y f )(q1, ξ1; q2, ξ2) = q2−1f (q2q1, q♭ 2ξ1; q2, ξ2 − q♭ 2ξ1). Since by (3.30) we have Y = (FV ⊗ FV ) W (F ∗ V ), this proves the result. V ⊗ F ∗ 30 (cid:3) 3.6. Stachura's dual cocycle. In this section we consider the simplest example of our setup, the ax + b group G over the reals. In this case Stachura [30] already defined a dual cocycle on G. We refer the reader to his paper for a motivation of the construction and just present an explicit form of the cocycle. Consider the following operators affiliated with W ∗(G): X := i d dt λ(et,0)(cid:12)(cid:12)(cid:12)t=0 , Y := i d dt λ(1,t)(cid:12)(cid:12)(cid:12)t=0 , I := λ(−1,0). Then the dual unitary cocycle on G found by Stachura, see [30, Lemma 5.6], is defined by ΩS := exp(cid:8)iX ⊗ log 1 + Y (cid:9) Ch(cid:0)1 ⊗ sgn(1 + Y ), I ⊗ 1(cid:1), where Ch : {−1, 1} × {−1, 1} → {−1, 1} is the unique nontrivial bicharacter.† (3.31) Proposition 3.28. The dual cocycle ΩS coincides with the dual cocycle Ω defined by (3.18) for ξ0 = −1. In particular, ΩS is cohomologous to the dual cocycle Ω defined by (3.17). Proof. Observe that Hence, using that I ⊗ 1 commutes with X ⊗ 1 and 1 ⊗ Y , we get 2Ch(ε1, ε2) = 1 + ε1 + ε2 − ε1ε2. 2ΩS = exp(cid:8)iX ⊗ log 1 + Y (cid:9)(cid:0)1 + 1 ⊗ sgn(1 + Y )(cid:1) +(cid:0)I ⊗ 1(cid:1) exp(cid:8)iX ⊗ log 1 + Y (cid:9)(cid:0)1 − 1 ⊗ sgn(1 + Y )(cid:1). In terms of the functions Fε : R2 → T, ε ∈ {0, 1}, defined by we therefore have Fε(x, y) := exp{ix ln 1 + y} sgn(1 + y)ε, 2ΩS =(cid:0)F0(X ⊗ 1, 1 ⊗ Y ) + F1(X ⊗ 1, 1 ⊗ Y )(cid:1)+(cid:0)I ⊗ 1(cid:1)(cid:0)F0(X ⊗ 1, 1 ⊗ Y ) − F1(X ⊗ 1, 1 ⊗ Y )(cid:1). Normalizing the 2-dimensional Fourier transform by with the integrals understood in the distributional sense. We now need to compute the inverse Fourier transforms (in the sense of tempered distribu- tions): (F ∗ R2 Fε)(s, t) = 1 2πZR2 ei(xs+yt) exp{ix ln 1 + y} sgn(1 + y)ε dx dy. †In fact, Stachura works in a representation of G equivalent to the regular representation. His operators X, Y, I are the operators on L2(R∗ × R, q−2dq dξ) given by (Xf )(q, ξ) = i(q∂q + ξ∂ξ)f (q, ξ), (Y f )(q, ξ) = ξf (q, ξ), (If )(q, ξ) = f (−q, −ξ). The equivalence is implemented by the unitary (2π)−1/2FRU : L2(G) → L2(R∗ × R, q−2dq dξ), where U : L2(G) → L2(G) is defined by (U f )(q, v) = qf (q−1, v). 31 we then obtain ΩS = = (FR2f )(s, t) = 1 e−i(xs+yt) f (x, y) dx dy, 2π ZR2 R2F1)(s, t)(cid:1) e−isX ⊗ e−itY ds dt R2 F0)(s, t) + (F ∗ 1 4πZR2(cid:0)(F ∗ 4πZR2(cid:0)(F ∗ + 1 + 1 4π ZR2(cid:0)(F ∗ 4π ZR2(cid:0)(F ∗ 1 R2F0)(s, t) − (F ∗ R2 F1)(s, t)(cid:1) Ie−isX ⊗ e−itY ds dt. R2 F0)(s, t) + (F ∗ R2F1)(s, t)(cid:1) λ(es,0) ⊗ λ(1,t) ds dt R2F0)(s, t) − (F ∗ R2 F1)(s, t)(cid:1) λ(−es,0) ⊗ λ(1,t) ds dt, Note first that 1 2πZR eix(s+ln 1+y) dx = δ0(cid:0)s + ln 1 + y(cid:1). Consider now, for fixed s ∈ R, the function ϕ(y) := s + ln 1 + y. It has two simple zeros located at y± = ±e−s − 1 and it is continuously differentiable (away from −1) with ϕ′(y±) = ±es. Therefore δ0(cid:0)s + ln 1 + y(cid:1) = ϕ′(y+)−1δy+(y) + ϕ′(y−)−1δy−(y) = e−s(cid:0)δ−1+e−s(y) + δ−1−e−s(y)(cid:1). Hence we get (F ∗ R2Fε)(s, t) = e−sZR eiyt (cid:0)δ−1+e−s(y) + δ−1−e−s(y)(cid:1) sgn(1 + y)ε dy = e−s−it(cid:0)eie−st + (−1)εe−ie−st(cid:1). From this we obtain (F ∗ R2F0)(s, t)+(F ∗ and therefore R2 F1)(s, t) = 2e−s−iteie−st and (F ∗ R2 F0)(s, t)−(F ∗ R2 F1)(s, t) = 2e−s−ite−ie−st, Setting now q = es and v = −t we get ΩS = 1 2π ZR×R 2π ZR+×R 2π ZR∗×R 1 1 ΩS = = e−s−it(cid:0)eie−st λ(es,0) + e−ie−st λ(−es,0)(cid:1) ⊗ λ(1,t) ds dt. q−1eiv(cid:0)e−iq−1v λ(q−1,0) + eiq−1v λ(−q−1,0)(cid:1) ⊗ λ(1,−v) q−1dq dv ei(v−q−1v) λ(q−1,0) ⊗ λ(1,−v) dq dv . q2 Setting ξ0 = −1 and remembering that we have here q♭ξ = q−1ξ and d(q, v) = (2π)−1q−2dq dv, we finally get ΩS =ZG eihq♭ξ0−ξ0,vi λ(q,0)−1 ⊗ λ(1,v)−1 d(q, v), which is exactly the dual cocycle Ω defined by (3.18) in Remark 3.13. As we already observed there, Ω is cohomologous to Ω. (cid:3) As was suggested by Stachura [30], the quantum group (W ∗(G), ΩS ∆(·)Ω∗ S ) is isomorphic to the quantum ax + b group of Baaj and Skandalis [29] (see also [33, Section 5.3]), but his arguments fall a bit short of proving that this is indeed the case. The above proposition together with Theorem 4.1 below complete his work. 4. Bicrossed product construction Recall that a pair (G1, G2) of closed subgroups a locally compact second countable group G is called a matched pair if G1 ∩ G2 = {e} and G1G2 is a subset of G of full measure [3]. Given such a pair, we have almost everywhere defined measurable left actions α of G1 and β of G2 on the measure spaces G2 and G1, resp., such that gs−1 = αg(s)−1βs(g) for g ∈ G1, s ∈ G2. We can then define a bicrossed product G1 ⊲◭ G2. This is a locally compact quantum group with the function algebra L∞( G1 ⊲◭ G2) := G1 ⋉α L∞(G2). The coproduct on L∞( G1 ⊲◭ G2) is a bit more difficult to describe, but we will not need to know the exact definition and refer the reader for that to [3] or [33].‡ Then by [33, Propostion 2.9 and ‡To be more precise, we are considering the quantum group (M, ∆) from [33, Section 4.2], with i : G1 → G and j : G2 → G defined by i(g) = g, j(s) = s−1. This is the same as the quantum group ( M , ∆) from [3, Section 3], see the discussion following [3, Definition 3.3]. 32 Theorem 2.13] the dual quantum group is G1 ◮⊳ G2 = G2 ⊲◭ G1, the bicrossed product defined by the matched pair (G2, G1) of subgroups of G. Theorem 4.1. Let G = Q ⋉ V be a second countable locally compact group satisfying the dual orbit Assumption 2.15 and Ω be the dual unitary 2-cocycle Ω on G defined by (3.17). Then the quantum group (W ∗(G), Ω ∆(·)Ω∗) is isomorphic to the bicrossed product quantum group defined by the matched pair (Q, ξ0Qξ−1 0 ) of subgroups of Q ⋉ V . Here by ξ−1 0 we of course mean the element (id, −ξ0) ∈ Aff( V ). Proof. Consider the measure space X = Q × V , with the measure class defined by the product of Haar measures. By Theorem 3.26 the multiplicative unitary WΩ is unitarily conjugate to the unitary associated with the measurable (almost everywhere defined) transformation v : X ×X → X × X given by v(q1, ξ1; q2, ξ2) =(cid:0)q2q1, q♭ 2ξ1; φ−1(q−1 2 ♭ ξ0 + ξ1)−1φ−1(ξ0 + ξ1), φ−1(q−1 2 ♭ ξ0 + ξ1)−1 ♭ ♭ (q−1 2 Hence this transformation is pentagonal [2]. By a result of Baaj and Skandalis [2], see also [3, Proposition 5.1] for a correction, under mild technical assumptions the pentagonal transforma- tions arise from matched pairs of groups. Let us follow the proof in [2] and see which pair we get. Following [2] we write the transformation v as v(x, y) = (x • y, x#y). The maps (x, y) 7→ (x•y, y) and (x, y) 7→ (x, x#y) are measure class isomorphisms, so the assumptions of [3, Propo- sition 5.1] are satisfied and therefore the pentagonal transformation v and the multiplicative unitary WΩ indeed come from a matched pair of groups. It is not difficult to check that the inverse map v−1, which we will write as v−1(x, y) = (x ⋄ y, x ∗ y), is given by ξ2 − ξ1)(cid:1). v−1(q1, ξ1; q2, ξ2) =(cid:0)φ−1(φ−1(ξ0 + ξ1)♭q♭ φ−1(φ−1(ξ0 + ξ1)♭q♭ 2ξ0 − ξ1)−1q1, φ−1(φ−1(ξ0 + ξ1)♭q♭ 2ξ0 − ξ1)−1♭ ξ1; 2ξ0 − ξ1), ξ1 + φ−1(ξ0 + ξ1)♭ξ2(cid:1). By [2, Lemma 2.1], there exists a second countable locally compact group G1, a right action of G1 on X and an equivariant measurable map f1 : X → G1 such that for almost all pairs (x, y) ∈ X × X we have x • y = xf1(y). Although this is not explicitly stated in [2], it is not difficult to see that the group G1, the action of G1 on X and the map f1 are uniquely determined by these properties up to an isomorphism. In our case it is easy to see what we get: G1 = Q, (q, ξ)q1 = (q−1 1 q, q−1 1 ♭ ξ), f1(q, ξ) = q−1. In a similar way there exist a (unique up to isomorphism) second countable locally compact group G2, a right action of G2 on X and an equivariant measurable map f2 : X → G2 such that for almost all pairs (x, y) ∈ X × X we have In our case we get x ∗ y = yf2(x). G2 = Q, f2(q, ξ) = φ−1(ξ + ξ0)−1, (ξ + ξ0) − ξ0). According to [2, 3] we then get a locally compact group G′ and embeddings hi : Gi → G of the groups Gi as closed subgroups such that the map G1 × G2 → G′, (s, g) 7→ h1(s)h2(g) is injective and the complement of its image is a set of measure zero. By [2, Lemma 3.5(b)], for almost all (x, y) ∈ X × X we have (q♭ξ0 − ξ0) + ξ0), q−1 2 (q, ξ)q2 = (φ−1(q−1 2 ♭ ♭ h2(f2(x))h1(f1(y)) = h1(f1(b))h2(f2(a)), where (a, b) = v(x, y). 33 Again, it is not difficult to see that these properties completely determine the locally compact group G′ up to isomorphism. In our case the above identity reads h2(φ−1(ξ + ξ0)−1)h1(q−1) = h1(φ−1(ξ + ξ0)−1φ−1(q−1♭ ξ0 + ξ))h2(φ−1(q♭ξ + ξ0)−1). Letting q1 = q and q2 = φ−1(ξ + ξ0) we equivalently get h2(q−1 2 )h1(q−1) = h1(q−1 2 φ−1(q−1 1 ♭ or in other words, ξ0 + q♭ 2ξ0 − ξ0))h2(φ−1(q♭ 1(q♭ 2ξ0 − ξ0) + ξ0)−1), h1(q1)h2(q2) = h2(φ−1(q♭ 1q♭ 2ξ0 − q♭ 1ξ0 + ξ0))h1(φ−1(q♭ 1q♭ 2ξ0 − q♭ 1ξ0 + ξ0)−1q1q2). We then see that these properties are satisfied by the group G′ := Q ⋉ V and the embeddings h1(q) := q = (q, 0), h2(q) := ξ0qξ−1 0 = (q, ξ0 − q♭ξ0). Therefore we conclude that WΩ is unitarily conjugate to the unitary W from [3] defined by the 0 ) of subgroups of Q ⋉ V , or equivalently, to the unitary W from [33] de- matched pair (Q, ξ0Qξ−1 fined by the matched pair (ξ0Qξ−1 0 , Q), see the discussion following [3, Definition 3.3]. Therefore (W ∗(G), Ω ∆(·)Ω∗) is isomorphic to the dual of the bicrossed product defined by (ξ0Qξ−1 0 , Q), hence to the bicrossed product defined by (Q, ξ0Qξ−1 (cid:3) Corollary 4.2. The locally compact quantum group (W ∗(G), Ω ∆(·)Ω∗) is self-dual. If G is nontrivial, then this quantum group is noncompact and nondiscrete, and if G is nonunimodular (that is, ∆Q 6= · V ), then the quantum group is also nonunimodular, with nontrivial scaling group and scaling constant 1. Proof. In order to prove self-duality it suffices to show that the matched pairs (Q, ξ0Qξ−1 0 ) and 0 , Q) are isomorphic. The conjugation by the element (− id, ξ0) ∈ Aff( V ) gives such an (ξ0Qξ−1 isomorphism. Next, the quantum group (W ∗(G), Ω ∆(·)Ω∗) cannot be discrete, since W ∗(G) is a factor. By self-duality it then cannot be compact either. The rest follows by [33, Proposi- tions 4.15, 4.16]. (cid:3) 0 ). References [1] P. Aniello, "Square integrable projective representations and square integrable representations modulo a relatively central subgroup", Int. J. Geom. Methods Mod. Phys. 3 (2006), no. 2, 233 -- 267. [2] S. Baaj and G. Skandalis, "Transformations pentagonales". C.R. Acad. Sci., Paris, S´er. I 327 (1998), 623 -- 628. [3] S. Baaj, G. Skandalis and S. Vaes, "Non-semi-regular quantum groups coming from number theory", Comm. Math. Phys. 235 (2003), 139 -- 167. [4] P. Bieliavsky and V. Gayral, "Deformation quantization for actions of Kahlerian Lie groups", Mem. Amer. Math. Soc. 236 (2015), no. 1115; available with erratum at arXiv:1109.3419v7 [math. OA]. [5] P. Bieliavsky, V. Gayral, S. Neshveyev and L. Tuset, "On deformation of C∗-algebras by actions of Kahlerian Lie groups", Int. J. Math. 27, 1650023 (2016); available with erratum at arXiv:1508.07762v3 [math. OA]. [6] A. Borowiec and A. Pachol, "kappa-Minkowski spacetime as the result of Jordanian twist deformation", Phys. Rev. D 79, 045012 (2009). [7] F. Bruhat, "Distributions sur un groupe localement compact et applications `a l'´etude des repr´esentations des groupes p-adiques", Bull. Soc. Math. France 89 (1961), 43 -- 75. [8] V. Coll, M. Gerstenhaber and A. Giaquinto, "An explicit deformation formula with non-commuting deriva- tions", Israel Math. Conf. Proc. Ring Theory, vol. 1, Weizmann Science Press, New York, 1989, pp. 396 -- 403. [9] A. Connes and M. Takesaki, "The flow of weights on factors of type III", Tohoku Math. J. (2) 29 (1977), no. 4, 473 -- 575. [10] A.A. Davydov, "Galois algebras and monoidal functors between categories of representations of finite groups", J. Algebra 244 (2001), no. 1, 273 -- 301. [11] K. De Commer, "Galois objects and cocycle twisting for locally compact quantum groups", J. Operator Theory 66 (2011), no. 1, 59 -- 106. [12] K. De Commer, "I-factorial quantum torsors and Heisenberg algebras of quantized universal enveloping type", preprint arXiv:1702.08191v1 [math.OA]. [13] V.G. Drinfeld, "Constant quasiclassical solutions of the Yang-Baxter quantum equation" (Russian), Dokl. Akad. Nauk SSSR 273 (1983), no. 3, 531 -- 535. 34 [14] M. Duflo and C.C. Moore, "On the regular representation of a nonunimodular locally compact group", J. Funct. Anal. 21 (1976), no. 2, 209 -- 243. [15] P. Etingof and D. Kazhdan, "Quantization of Lie bialgebras. I", Selecta Math. (N.S.) 2 (1996), no. 1, 1 -- 41. [16] H. Grabe and A. Vlassov, "On a formula of Coll-Gerstenhaber-Giaquinto", J. Geom. Phys. 28 (1998), no. 1-2, 129 -- 142. [17] D. Jondreville, "A locally compact quantum group arising from quantization of the affine group of a local field", Lett. Math. Phys. 109 (2019), 781 -- 797. [18] E. Koelink and J. Kustermans, "A locally compact quantum group analogue of the normalizer of SU (1, 1) in SL(2, C)", Comm. Math. Phys. 233 (2003), no. 2, 231 -- 296. [19] L.I. Korogodsky, "Quantum group SU (1, 1) ⋊ Z2 and "super-tensor" products", Comm. Math. Phys. 163 (1994), no. 3, 433 -- 460. [20] P.P. Kulish, V.D. Lyakhovsky and A.I. Mudrov, "Extended Jordanian twists for Lie algebras", J. Math. Phys. 40 (1999), no. 9, 4569 -- 4586. [21] P.P. Kulish, V.D. Lyakhovsky and A. Stolin, "Chains of Frobenius subalgebras of so(M ) and the correspond- ing twists", J. Math. Phys. 42 (2001), no. 10, 5006 -- 5019. [22] S. Majid and H. Ruegg, "Bicrossproduct structure of κ-Poincar´e group and non-commutative geometry", Phys. Lett. B 334 (1994), no. 3 -- 4, 348 -- 354. [23] C.C. Moore, "Group extensions and cohomology for locally compact groups. III", Trans. Amer. Math. Soc. 221 (1976), no. 1, 1 -- 33. [24] M.V. Movshev, "Twisting in group algebras of finite groups", Funct. Anal. Appl. 27 (1993), no. 4, 240 -- 244 (1994). [25] S. Neshveyev and L. Tuset, "On second cohomology of duals of compact groups", Internat. J. Math. 22 (2011), no. 9, 1231 -- 1260. [26] S. Neshveyev and L. Tuset, "Deformation of C∗-algebras by cocycles on locally compact quantum groups", Adv. Math. 254 (2014), 454 -- 496. [27] O. Ogievetsky, "Hopf structures on the Borel subalgebra of sl(2)", in: Proceedings of Winter School in Geometry and Physics, Zdikov, January 1993, Supplemento ai Rendiconti del Circolo Matematico di Palermo, Serie II, 37 (1994), 185 -- 199. [28] A.I. Ooms, "On Frobenius Lie algebras", Comm. Algebra 8 (1980), no. 1, 13 -- 52. [29] G. Skandalis, "Duality for locally compact Quantum groups" (joint work with Baaj). Mathematisches Forschungsinstitut Oberwolfach, Tagungsbericht 46/1991, C ∗-algebren, (1991), p. 20. [30] P. Stachura, "On the quantum 'ax+b' group", Journal of Geometry and Physics 73 (2013), 125 -- 149. [31] I. Segal, "Transforms for operators and symplectic automorphisms over a locally compact Abelian group", Math. Scand. 13 (1963), 31 -- 43. [32] S. Stratila, Modular theory in operator algebras. Abacus Press, Tunbridge Wells, (1981). [33] S. Vaes and L. Vainerman, "Extensions of locally compact quantum groups and the bicrossed product construction", Adv. Math. 175 (2003), no. 1, 1 -- 101. [34] A. Wassermann, "Ergodic actions of compact groups on operator algebras. II. Classification of full multi- plicity ergodic actions", Canad. J. Math. 40 (1988), no. 6, 1482 -- 1527. [35] A. Weil, "Sur certain groupes d'op´erateurs unitaires", Acta Math. 111 (1964), 143 -- 211. [36] S. L. Woronowicz, "Unbounded elements affiliated with C ∗-algebras and noncompact quantum groups", Comm. Math. Phys. 136 (1991), no. 2, 399 -- 432. E-mail address: [email protected] Institut de Recherche en Math´ematique et Physique, Universit´e Catholique de Louvain, Chemin du Cyclotron, 2, 1348 Louvain-la-Neuve, Belgium E-mail address: [email protected] Laboratoire de Math´ematiques, CNRS FRE 2011, Universit´e de Reims Champagne-Ardenne, Moulin de la Housse - BP 1039, 51687 Reims, France E-mail address: [email protected] Department of Mathematics, University of Oslo, P.O. Box 1053 Blindern, NO-0316 Oslo, Nor- way E-mail address: [email protected] Department of Computer Science, OsloMet - storbyuniversitetet, P.O. Box 4 St. Olavs plass, NO-0130 Oslo, Norway 35
1010.1432	1	1010	2010-10-07T13:48:28	Minimal and Maximal Operator Spaces and Operator Systems in Entanglement Theory	[ "math.OA", "math.FA", "quant-ph" ]	We examine k-minimal and k-maximal operator spaces and operator systems, and investigate their relationships with the separability problem in quantum information theory. We show that the matrix norms that define the k-minimal operator spaces are equal to a family of norms that have been studied independently as a tool for detecting k-positive linear maps and bound entanglement. Similarly, we investigate the k-super minimal and k-super maximal operator systems that were recently introduced and show that their cones of positive elements are exactly the cones of k-block positive operators and (unnormalized) states with Schmidt number no greater than k, respectively. We characterize a class of norms on the k-super minimal operator systems and show that the completely bounded versions of these norms provide a criterion for testing the Schmidt number of a quantum state that generalizes the recently-developed separability criterion based on trace-contractive maps.	math.OA	math	MINIMAL AND MAXIMAL OPERATOR SPACES AND OPERATOR SYSTEMS IN ENTANGLEMENT THEORY NATHANIEL JOHNSTON, DAVID W. KRIBS, VERN I. PAULSEN, AND RAJESH PEREIRA Abstract. We examine k-minimal and k-maximal operator spaces and operator systems, and investigate their relationships with the separability problem in quantum information theory. We show that the matrix norms that define the k-minimal operator spaces are equal to a family of norms that have been studied independently as a tool for detecting k-positive linear maps and bound entanglement. Similarly, we investigate the k-super minimal and k-super maximal operator systems that were recently introduced and show that their cones of positive elements are exactly the cones of k-block positive operators and (unnormalized) states with Schmidt number no greater than k, respectively. We characterize a class of norms on the k-super minimal operator systems and show that the completely bounded versions of these norms provide a criterion for testing the Schmidt number of a quantum state that generalizes the recently-developed separability criterion based on trace-contractive maps. Keywords: operator space, operator system, quantum information theory, entanglement 1. Introduction A primary goal of this paper is to formally link central areas of study in operator theory and quantum information theory. More specifically, we connect recent investigations in operator space and operator system theory [1, 2] on the one hand and the theory of entan- glement [3, 4] on the other. As benefits of this combined perspective, we obtain new results and new elementary proofs in both areas. We give further details below before proceeding. Given a (classical description of a) quantum state ρ, one of the most basic open questions in quantum information theory asks for an operational criterion for determining whether ρ is separable or entangled. Much progress has been made on this front over the past two decades. For instance, a revealing connection between the separability problem and operator theory was established in [5], where it was shown that ρ is separable if and only if it remains positive under the application of any positive map to one half of the state. Another more recent approach characterizes separability via maps that are contractive in the trace norm on Hermitian operators [6]. In this work we show that these two approaches to the separability problem can be seen as arising from the theory of minimal and maximal operator systems and operator spaces, respectively. Additionally, this work can be seen as demonstrating how to rephrase certain positivity questions that are relevant in quantum information theory in terms of norms that are relevant in operator theory instead. For example, instead of using positive maps to detect separability of quantum states, we can construct a natural operator system into which positive maps are completely positive. Then the completely bounded norm on that operator system serves as a tool for detecting separability of quantum states as well. 1 2 N. JOHNSTON, D. W. KRIBS, V. I. PAULSEN, AND R. PEREIRA A natural generalization of the characterization of separable states in terms of positive linear maps was implicit in [7] and proved in [8] -- a state has Schmidt number no greater than k if and only if it remains positive under the application of any k-positive map to one half of the state. Recently, a further connection was made between operator theory and quantum information: a map is completely positive on what is known as the maximal (resp. minimal) operator system on Mn, the space of n × n complex matrices, if and only if it is a positive (resp. entanglement-breaking [9]) map [10]. Thus, the maps that serve to detect quantum entanglement are the completely positive maps on the maximal operator system on Mn. Similarly, completely positive maps on "k-super maximal" and "k-super minimal" oper- ator systems on Mn [11] have been studied and shown to be the same as k-positive and k-partially entanglement breaking maps [12], respectively. We will reprove these state- ments via an elementary proof that shows that the cones of positive elements that define the k-super maximal (resp. k-super minimal) operator systems are exactly the cones of (unnormalized) states with Schmidt number at most k (resp. k-block positive operators). Analogous to the minimal and maximal operator systems, there are minimal and maximal operator spaces (and appropriate k-minimal and k-maximal generalizations). We will show the norms that define the k-minimal operator spaces on Mn coincide with a family of norms that have recently been studied in quantum information theory [13, 14, 15, 16, 17, 18, 19] for their applications to the problems of detecting k-positive linear maps and NPPT bound entangled states. Furthermore, we will connect the dual of a version of the completely bounded minimal operator space norm to the separability problem and extend recent results about how trace-contractive maps can be used to detect entanglement. We will see that the maps that serve to detect quantum entanglement via norms are roughly the completely contractive maps on the minimal operator space on Mn. The natural generalization to norms that detect states with Schmidt number k is proved via a stabilization result for the completely bounded norm from Mr to the k-minimal operator space (or system) of Mn. In Section 2 we introduce the reader to the various relevant notions from quantum infor- mation theory such as separability and Schmidt rank. In Section 3 we introduce (abstract) operator spaces and the k-minimal and k-maximal operator space structures, and investi- gate their relationship with norms that have been used in quantum information theory. In Section 4 we give a similar treatment to abstract operator systems and the k-super minimal and k-super maximal operator system structures. We then investigate some norms on the k-super minimal operator system structures in Section 5. We close in Section 6 by consid- ering the completely bounded version of some of the norms that have been presented and establish a relationship with the Schmidt number of quantum states. 2. Quantum Information Theory Preliminaries Given a vector space V , we will use Mm,p(V ) to denote the space of m× p matrices with elements from V . For brevity we will write Mm(V ) := Mm,m(V ) and Mm := Mm(C). It will occasionally be convenient to use tensor product notation and identify Mm ⊗ V ∼= Mm(V ) in the standard way, especially when V = Mn. We will make use of bra-ket notation from quantum mechanics as follows: we will use "kets" vi ∈ Cn to represent unit (column) vectors and "bras" hv := vi∗ to represent the dual (row) vectors, where (·)∗ represents OPERATOR SPACES AND OPERATOR SYSTEMS IN ENTANGLEMENT THEORY 3 the conjugate transpose. Unit vectors represent pure quantum states (or more specifically, the projection vihv onto the vector vi represents a pure quantum state) and thus we will sometimes refer to unit vectors as states. Mixed quantum states are represented by density operators ρ ∈ Mm ⊗ Mn that are positive semidefinite with Tr(ρ) = 1. A state vi ∈ Cm ⊗ Cn is called separable if there exist v1i ∈ Cm, v2i ∈ Cn such that vi = v1i ⊗ v2i; otherwise it is said to be entangled. The Schmidt rank of a state vi, denoted SR(vi), is the least number separable states vii needed to write vi = Pi αivii, where αi are some (real) coefficients. The analogue of Schmidt rank for a bipartite mixed state ρ ∈ Mm ⊗ Mn is Schmidt number [7], denoted SN (ρ), which is defined to be the least integer k such that ρ can be written in the form ρ =Pi piviihvi with {pi} forming a probability distribution and SR(vii) ≤ k for all i. An operator X = X∗ ∈ Mm ⊗ Mn is said to be k-block positive (or a k-entanglement witness) if hvXvi ≥ 0 for all vectors vi with SR(vi) ≤ k. In the extreme case when k = min{m, n}, we see that the k-block positive operators are exactly the positive semidefinite operators (since SR(vi) ≤ min{m, n} for all vectors vi), and for smaller k the set of k-block positive operators is strictly larger. In [13, 14], a family of operator norms that have several connections in quantum infor- mation theory was investigated. Arising from the Schmidt rank of bipartite pure states, they are defined for operators X ∈ Mm ⊗ Mn as follows: (1) (cid:13)(cid:13)X(cid:13)(cid:13)S(k) = sup vi,wi(cid:8)hvXwi : SR(vi), SR(wi) ≤ k(cid:9). These norms were shown to be useful for determining whether or not an operator is k-block positive, and also have applications to the problem of determining whether or not there exist bound entangled non-positive partial transpose states [13, 16]. The problem of computing these norms was investigated in [14]. The completely bounded norm of a linear map Φ : Mr → Mn is defined to be (cid:13)(cid:13)Φ(cid:13)(cid:13)cb := sup It was shown by Smith [20] (and independently later by Kitaev [21] from the dual perspec- m≥1n(cid:13)(cid:13)(idm ⊗ Φ)(X)(cid:13)(cid:13) : X ∈ Mm(Mr) with (cid:13)(cid:13)X(cid:13)(cid:13) ≤ 1o. tive) that it suffices to fix m = n so that (cid:13)(cid:13)Φ(cid:13)(cid:13)cb = (cid:13)(cid:13)idn ⊗ Φ(cid:13)(cid:13). We will see in Section 6 a connection between the norms (cid:13)(cid:13)idk ⊗ Φ(cid:13)(cid:13) for 1 ≤ k ≤ n and the norms (1). 3. k-Minimal and k-Maximal Operator Spaces We will now present (abstract) operator spaces and the k-minimal and k-maximal op- erator space structures. An operator space is a vector space V together with a family of L∞ matrix norms k · kMm(V ) on Mm(V ) that make V into a matrix normed space. That is, we require that if A = (aij), B = (bij) ∈ Mp,m and X = (xij) ∈ Mm(V ) then (cid:13)(cid:13)A · X · B∗(cid:13)(cid:13)Mp(V ) ≤(cid:13)(cid:13)A(cid:13)(cid:13)(cid:13)(cid:13)X(cid:13)(cid:13)Mm(V )(cid:13)(cid:13)B(cid:13)(cid:13), where Xk,ℓ=1 aikxkℓbjℓ(cid:17) ∈ Mp(V ) A · X · B∗ :=(cid:16) m 4 N. JOHNSTON, D. W. KRIBS, V. I. PAULSEN, AND R. PEREIRA and (cid:13)(cid:13)A(cid:13)(cid:13),(cid:13)(cid:13)B(cid:13)(cid:13) represent the operator norm on Mp,m. The L∞ requirement is that (cid:13)(cid:13)X ⊕ Y(cid:13)(cid:13)Mm+p(V ) = max(cid:8)kXkMm(V ),kY kMp(V )(cid:9) for all X ∈ Mm(V ), Y ∈ Mp(V ). When the particular operator space structure (i.e., the family of L∞ matrix norms) on V is not important, we will denote the operator space simply by V . We will use Mn itself to denote the "standard" operator space structure on Mn that is obtained by associating Mm(Mn) with Mmn in the natural way and using the operator norm. For a more detailed introduction to abstract operator spaces, the interested reader is directed to [1, Chapter 13]. Given an operator space V and a natural number k, one can define a new family of norms on Mm(V ) that coincide with the matrix norms on Mm(V ) for 1 ≤ m ≤ k and are minimal (or maximal) for m > k. We will use M IN k(V ) and M AX k(V ) to denote what are called the k-minimal operator space of V and the k-maximal operator space of V , respectively. that define the k-minimal and k-maximal operator spaces of V , respectively. For X ∈ Mm(V ) we will use (cid:13)(cid:13)X(cid:13)(cid:13)Mm(M IN k(V )) and (cid:13)(cid:13)X(cid:13)(cid:13)Mm(M AX k(V )) to denote the norms For X ∈ Mm(V ) one can define the k-minimal and k-maximal operator space norms via (2) (3) (cid:13)(cid:13)X(cid:13)(cid:13)Mm(M IN k(V )) := supn(cid:13)(cid:13)(Φ(Xij))(cid:13)(cid:13) : Φ : V → Mk with (cid:13)(cid:13)Φ(cid:13)(cid:13)cb ≤ 1o and (cid:13)(cid:13)X(cid:13)(cid:13)Mm(M AX k(V )) := supn(cid:13)(cid:13)(Φ(Xij))(cid:13)(cid:13) : Φ : V → B(H) with (cid:13)(cid:13)idk ⊗ Φ(cid:13)(cid:13) ≤ 1o. Indeed, the names of these operator space structures come from the facts that if O(V ) is any operator space structure on V such that k · kMm(V ) = k · kMm(O(V )) for 1 ≤ m ≤ k then k · kMm(M IN k(V )) ≤ ·Mm(O(V )) ≤ k · kMm(M AX k(V )) for all m > k. In the k = 1 case, these operator spaces are exactly the minimal and maximal operator space structures that are fundamental in operator space theory [1, Chapter 14]. The interested reader is directed to [22] and the references therein for further properties of M IN k(V ) and M AX k(V ) when k ≥ 2. One of the primary reasons for our interest in the k-minimal operator spaces is the following result, which says that the k-minimal norm on Mm(Mn) is exactly equal to the S(k)-norm (1) from quantum information theory. Theorem 1. Let X ∈ Mm(Mn). Then (cid:13)(cid:13)X(cid:13)(cid:13)Mm(M IN k(Mn)) =(cid:13)(cid:13)X(cid:13)(cid:13)S(k). Proof. A fundamental result about completely bounded maps says (see [23, Theorem 19], for example) that any completely bounded map Φ : Mn → Mk has a representation of the form (4) Φ(Y ) = nk Xi=1 nk Xi=1 BiB∗i(cid:13)(cid:13)(cid:13) 2 cb. =(cid:13)(cid:13)Φ(cid:13)(cid:13) AiY B∗i with Ai, Bi ∈ Mk,n and (cid:13)(cid:13)(cid:13) nk Xi=1 AiA∗i(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) By using the fact that Φ is completely contractive (so (cid:13)(cid:13)Φ(cid:13)(cid:13)cb = 1) and a rescaling of the operators (cid:8)Ai(cid:9) and (cid:8)Bi(cid:9) we have BiB∗i(cid:13)(cid:13) = 1o, (cid:13)(cid:13)X(cid:13)(cid:13)Mm(M IN k(Mn)) = supn(cid:13)(cid:13) Xi=1 (Im ⊗ Ai)X(Im ⊗ B∗i )(cid:13)(cid:13) :(cid:13)(cid:13) AiA∗i(cid:13)(cid:13) =(cid:13)(cid:13) Xi=1 nk nk Xi=1 nk OPERATOR SPACES AND OPERATOR SYSTEMS IN ENTANGLEMENT THEORY 5 where the supremum is taken over all families of operators (cid:8)Ai(cid:9),(cid:8)Bi(cid:9) ⊂ Mk,n satisfying the normalization condition. Now define αijaiji := A∗i ji and βijbiji := B∗i ji, and let vi =Pk j=1 δjdji ⊗ ji ∈ Cm ⊗ Ck be arbitrary unit vectors. Then simple algebra reveals j=1 γjcji ⊗ ji,wi =Pk k νivii := (Im ⊗ A∗i )vi = µiwii := (Im ⊗ B∗i )wi = Xj=1 Xj=1 k αijγjcji ⊗ aiji and βijδjdji ⊗ biji. In particular, SR(vii), SR(wii) ≤ k for all i. Furthermore, by the normalization condition on (cid:8)Ai(cid:9) and (cid:8)Bi(cid:9) we have that Xi=1 AiA∗i )vi = ν2 i ≤ 1 and hw(Im ⊗ BiB∗i )wi = hv(Im ⊗ µ2 i ≤ 1. Xi=1 (5) nk nk nk nk Thus we can write nk Xi=1 hv(Im ⊗ Ai)(X)(Im ⊗ B∗i )wi(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Xi=1 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (6) Xi=1 νiµihviXwii(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) nk =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Xi=1 ≤ nk Xi=1 νiµi(cid:12)(cid:12)hviXwii(cid:12)(cid:12). The normalization condition (5) and the Cauchy-Schwarz inequality tell us that there is a particular i′ such that the sum (6) ≤ hvi′Xwi′i. Taking the supremum over all vectors vi and wi gives the "≤" inequality. The "≥" inequality can be seen by noting that if we have two vectors in their Schmidt decompositions vi = Pk i=1 βidii ⊗ bii, then we can define operators A, B ∈ Mk,n by setting their ith row in the standard basis to be hai and hbi, respectively. Because the rows of A and B form orthonormal sets, (cid:13)(cid:13)A(cid:13)(cid:13) = (cid:13)(cid:13)B(cid:13)(cid:13) = 1. Additionally, if we define v′i =Pk−1 Taking the supremum over all vectors vi,wi with SR(vi), SR(wi) ≤ k gives the result. (cid:13)(cid:13)(Im ⊗ A)(X)(Im ⊗ B∗)(cid:13)(cid:13) ≥(cid:12)(cid:12)hv′(Im ⊗ A)(X)(Im ⊗ B∗)w′i(cid:12)(cid:12) =(cid:12)(cid:12)hvXwi(cid:12)(cid:12). i=0 αicii ⊗ ii and w′i =Pk−1 i=1 αicii ⊗ aii and wi = Pk i=0 βidii ⊗ ii, then (cid:3) Remark 2. When working with an operator system (instead of an operator space) V , it is more natural to define the norm (2) by taking the supremum over all completely positive unital maps Φ : V → Mk rather than all complete contractions (similarly, to define the norm (3) one would take the supremum over all k-positive unital maps rather than k- contractive maps). In this case, the k-minimal norm no longer coincides with the S(k)-norm on Mm(Mn) but rather has the following slightly different form: (7) (cid:13)(cid:13)X(cid:13)(cid:13)Mm(OM IN k(Mn)) = sup vi,win(cid:12)(cid:12)hvXwi(cid:12)(cid:12) : SR(vi), SR(wi) ≤ k and ∃ P ∈ Mm s.t. (P ⊗ In)vi = wio, 6 N. JOHNSTON, D. W. KRIBS, V. I. PAULSEN, AND R. PEREIRA where the notation OM IN k(Mn) refers to a new operator system structure that is being assigned to Mn, which we discuss in detail in the next section. Intuitively, this norm has the same interpretation as the norm (1) except with the added restriction that the vectors vi and wi look the same on the second subsystems. We will In particular, we will see in Theorem 8 examine this norm in more detail in Section 5. that the norm (7) is a natural norm on the k-super minimal operator system structure (to be defined in Section 4), which plays an analogous role to the k-minimal operator space structure. Now that we have characterized the k-minimal norm in a fairly concrete way, we turn our attention to the k-maximal norm. The following result is a direct generalization of a corresponding known characterization of the M AX(V ) norm [1, Theorem 14.2]. Theorem 3. Let V be an operator space and let X ∈ Mm(V ). Then (cid:13)(cid:13)X(cid:13)(cid:13)Mm(M AX k(V )) = infn(cid:13)(cid:13)A(cid:13)(cid:13)(cid:13)(cid:13)B(cid:13)(cid:13) : A, B ∈ Mm,rk, xi ∈ Mk(V ),kxikMk(V ) ≤ 1 with X = A · diag(x1, . . . , xr) · B∗o, where diag(x1, . . . , xr) ∈ Mrk(V ) is the r × r block diagonal matrix with entries x1, . . . , xr down its diagonal, and the infimum is taken over all such decompositions of X. Proof. The "≤" inequality follows simply from the axioms of an operator space: (Xij) = A · diag(x1, . . . , xr) · B∗ ∈ Mm(V ) then (Φ(Xij)) = A · diag((idk ⊗ Φ)(x1), . . . , (idk ⊗ Φ)(xr)) · B∗. if X = Thus (cid:13)(cid:13)(Φ(Xij))(cid:13)(cid:13) ≤(cid:13)(cid:13)A(cid:13)(cid:13)(cid:13)(cid:13)B(cid:13)(cid:13) max(cid:8)k(idk ⊗ Φ)(x1)k, . . . ,k(idk ⊗ Φ)(xr)k(cid:9). By taking the supremum over maps Φ with kidk ⊗ Φk ≤ 1, the "≤" inequality follows. We will now show that the infimum on the right is an L∞ matrix norm that coincides with k · kMm(V ) for 1 ≤ m ≤ k. The "≥" inequality will then follow from the fact that k · kMm(M AX k(V )) is the maximal such norm. First, denote the infimum on the right by (cid:13)(cid:13)X(cid:13)(cid:13)m,inf and fix some 1 ≤ m ≤ k. Then the inequality (cid:13)(cid:13)X(cid:13)(cid:13)Mm(V ) ≤ (cid:13)(cid:13)X(cid:13)(cid:13)m,inf follows immediately by picking any particular decom- position X = A · diag(x1, . . . , xr) · B∗ and using the axioms of an operator space to see that (cid:13)(cid:13)A · diag(x1, . . . , xr) · B∗(cid:13)(cid:13)Mm(V ) ≤ (cid:13)(cid:13)A(cid:13)(cid:13)(cid:13)(cid:13)B(cid:13)(cid:13) max(cid:8)kx1k, . . . ,kxrk(cid:9) ≤ (cid:13)(cid:13)A(cid:13)(cid:13)(cid:13)(cid:13)B(cid:13)(cid:13) ≤ (cid:13)(cid:13)X(cid:13)(cid:13)m,inf . The fact that equality is attained by some decomposition of X comes simply from writing X =(cid:0)(cid:13)(cid:13)X(cid:13)(cid:13)Mm(V )I(cid:1) · (X ⊕ 0k−m) · I. It follows that k · kMm(V ) = k · km,inf for 1 ≤ m ≤ k. All that remains to be proved is that k · km,inf is an L∞ matrix norm, which we omit as it is directly analogous to the proof of [1, Theorem 14.2]. (cid:3) OPERATOR SPACES AND OPERATOR SYSTEMS IN ENTANGLEMENT THEORY 7 As one final note, observe that we can obtain lower bounds of the k-minimal and k- maximal operator space norms simply by choosing particular maps Φ that satisfy the nor- malization condition of their definition. Upper bounds of the k-maximal norms can be obtained from Theorem 3. The problem of computing upper bounds for the k-minimal norms was investigated in [14]. 4. k-Super Minimal and k-Super Maximal Operator Systems We will now introduce (abstract) operator systems, and in particular the minimal and maximal operator systems that were explored in [10] and the k-super minimal and k-super maximal operator systems that were explored in [11]. Our introduction to general operator systems will be brief, and the interested reader is directed to [1, Chapter 13] for a more thorough treatment. Let V be a complex (not necessarily normed) vector space as before, with a conjugate linear involution that will be denoted by ∗ (such a space is called a ∗-vector space). Define Vh := {v ∈ V : v = v∗} to be the set of Hermitian elements of V . We will say that (V, V +) is an ordered ∗-vector space if V + ⊆ Vh is a convex cone satisfying V + ∩ −V + = {0}. Here V + plays the role of the "positive" elements of V -- in the most familiar ordering on square matrices, V + is the set of positive semidefinite matrices. Much as was the case with operator spaces, an operator system is constructed by con- sidering the spaces Mm(V ), but instead of considering various norms on these spaces that behave well with the norm on V , we will consider various cones of positive elements on Mm(V ) that behave well with the cone of positive elements V+. To this end, given a ∗- vector space V we let Mm(V )h denote the set of Hermitian elements in Mm(V ). It is said that a family of cones Cm ⊆ Mm(V )h (m ≥ 1) is a matrix ordering on V if C1 = V + and they satisfy the following three properties: • each Cm is a cone in Mm(V )h; • Cm ∩ −Cm = {0} for each m; and • for each n, m ∈ N and X ∈ Mm,n we have X∗CmX ⊆ Cn. A final technical restriction on V is that we will require an element e ∈ Vh such that, for any v ∈ V , there exists r > 0 such that re − v ∈ V + (such an element e is called an order unit). It is said that e is an Archimedean order unit if re + v ∈ V + for all r > 0 implies that v ∈ V +. A triple (V, C1, e), where (V, C1) is an ordered ∗-vector space and e is an Archimedean order unit, will be referred to as an Archimedean ordered ∗-vector space or an AOU space for short. Furthermore, if e ∈ Vh is an Archimedean order unit then we say that it is an Archimedean matrix order unit if the operator em := Im ⊗ e ∈ Mm(V ) is an Archimedean order unit in Cm for all m ≥ 1. We are now able to define abstract operator systems: Definition 4. An (abstract) operator system is a triple (V,{Cm}∞m=1, e), where V is a ∗-vector space, {Cm}∞m=1 is a matrix ordering on V , and e ∈ Vh is an Archimedean matrix order unit. For brevity, we may simply say that V is an operator system, with the understanding that there is an associated matrix ordering {Cm}∞m=1 and Archimedean matrix order unit e. Recall from [10] that for any AOU space (V, V +, e) there exists minimal and maximal 8 N. JOHNSTON, D. W. KRIBS, V. I. PAULSEN, AND R. PEREIRA m }∞m=1 and {C max operator system structure OM IN (V ) and OM AX(V ) -- that is, there exist particular m }∞m=1 such that if {Dm}∞m=1 is any other matrix families of cones {C min ordering on (V, V +, e) then C max m for all m ≥ 1. In [11] a generalization of these operator system structures, analogous to the k-minimal and k-maximal operator spaces presented in Section 3, was introduced. Given an operator system V , the k-super minimal operator system of V and the k-super maximal operator system of V , denoted OM IN k(V ) and OM AX k(V ) respectively, are defined via the following families of cones: m ⊆ Dm ⊆ C min m C min,k C max,k m :=(cid:8)(Xij) ∈ Mm(V ) : (Φ(Xij)) ∈ M + :=(cid:8)A · D · A∗ ∈ Mm(V ) : A ∈ Mm,rk, D = diag(D1, . . . , Dr), m ∀ unital CP maps Φ : V → Mk(cid:9), Dℓ ∈ Mk(V )+ ∀ ℓ, r ∈ N(cid:9). If V is infinite-dimensional then the cones C max,k need not define an operator system due to Im ⊗ e perhaps not always being an Archimedean order unit, though it was shown in [11] how to Archimedeanize the space to correct this problem. We will avoid this technicality by working explicitly in the V = Mn case from now on. m Observe that the interpretation of the k-super minimal and k-super maximal operator systems is completely analogous to the interpretation of k-minimal and k-maximal operator spaces. The families of positive cones C min,k coincide with the families of positive cones of V for 1 ≤ m ≤ k, and out of all operator system structures on V with this property they are the largest (smallest, respectively) for m > k. ⊆ Mm ⊗ Mn are exactly the cones of k-block positive operators, and the cones C max,k ⊆ Mm ⊗ Mn are exactly the cones of (unnormalized) density operators ρ with SN (ρ) ≤ k. These facts have appeared implicitly in the past, but their importance merits making the details explicit: In terms of quantum information theory, the cones C min,k and C max,k m m m m m Theorem 5. Let X, ρ ∈ Mm ⊗ Mn. Then (a) X ∈ C min,k (b) ρ ∈ C max,k Proof. To see (a), we will use techniques similar to those used in the proof of Theorem 1. Use the Choi-Kraus representation of completely positive maps so that X ∈ C min,k if and only if if and only if X is k-block positive; and if and only if SN (ρ) ≤ k. m m nk unit vector. Then some algebra reveals Xi=1 (Im ⊗ Ai)X(Im ⊗ A∗i ) ∈ (Mm ⊗ Mk)+ for all (cid:8)Ai(cid:9) ⊂ Mk,n with Now define αijaiji := A∗i ji and let vi = Pk−1 Xj=0 νivii := (Im ⊗ A∗i )vi = αijγjcji ⊗ aiji. k−1 AiA∗i = Ik. nk Xi=1 j=0 γjcji ⊗ ji ∈ Cm ⊗ Ck be an arbitrary OPERATOR SPACES AND OPERATOR SYSTEMS IN ENTANGLEMENT THEORY 9 In particular, SR(vii) ≤ k for all i. Thus we can write Xi=1 Xi=1 hv(Im ⊗ Ai)(X)(Im ⊗ A∗i )vi = (8) nk nk ν2 i hviXvii ≥ 0. m Part (a) follows by noting that we can choose vi and a CP map with one Kraus operator A1 so that (Im⊗ A∗1)vi is any particular vector of our choosing with Schmidt rank no larger than k. To see the "only if" implication of (b), we could invoke various known duality results from operator theory and quantum information theory so that the result would follow from (a), but for completeness we will instead prove it using elementary means. To this end, suppose ρ ∈ C max,k . Thus we can write ρ = A · D · A∗ for some A ∈ Mm,rk and D = diag(D1, . . . , Dr) =Pr ℓ=1 ℓihℓ⊗ Dℓ with Dℓ ∈ Mk(Mn)+ for all ℓ. Furthermore, write Dℓ = Ph dℓ,hvℓ,hihvℓ,h where vℓ,hi = Pk i=1 ii ⊗ dℓ,h,ii. Then if we define αℓ,iaℓ, ii := A(ℓi ⊗ ii), we have Xij=1 Xij=1 A(ℓihℓ ⊗ iihj)A∗ ⊗ dℓ,h,iihdℓ,h,j αℓ,iαℓ,jaℓ,iihaℓ,j ⊗ dℓ,h,iihdℓ,h,j A · D · A∗ = dℓ,h dℓ,h = kn k k r r Xℓ=1 Xℓ=1 Xℓ=1 r kn Xh=1 Xh=1 Xh=1 kn = where dℓ,hwℓ,hihwℓ,h, k wℓ,hi := αℓ,iaℓ,ii ⊗ dℓ,h,ii. Xi=1 Since SR(wℓ,hi) ≤ k for all ℓ, h, it follows that SN (ρ) ≤ k as well. For the "if" implication, we note that the above argument can easily be reversed. (cid:3) One of the useful consequences of Theorem 5 is that we can now easily characterize com- pletely positive maps between these various operator system structures. Recall that a map Φ between operator systems (V,{Cm}∞m=1, e) and (V,{Dm}∞m=1, e) is said to be completely positive if (Φ(Xij)) ∈ Dm whenever (Xij) ∈ Cm. We then have the following result that characterizes k-positive maps, entanglement-breaking maps, and k-partially entanglement breaking maps as completely positive maps between these k-super minimal and k-super maximal operator systems. Corollary 6. Let Φ : Mn → Mn and let k ≤ n. Then (a) Φ : OM IN k(Mn) → Mn is completely positive if and only if Φ is k-partially entangle- (b) Φ : Mn → OM AX k(Mn) is completely positive if and only if Φ is k-partially entangle- (c) Φ : OM AX k(Mn) → Mn is completely positive if and only if Φ is k-positive; ment breaking; ment breaking; 10 N. JOHNSTON, D. W. KRIBS, V. I. PAULSEN, AND R. PEREIRA (d) Φ : Mn → OM IN k(Mn) is completely positive if and only if Φ is k-positive. Proof. Fact (a) follows from [12, Theorem 2] and fact (b) follows from the fact that Φ is k-partially entanglement breaking (by definition) if and only if SN ((idm ⊗ Φ)(ρ)) ≤ k for all m ≥ 1. Fact (c) follows from [7, Theorem 1] and (d) follows from (c) and the fact that the cone of unnormalized states with Schmidt number at most k and the cone of k-block positive operators are dual to each other [24]. (cid:3) Remark 7. Corollary 6 was originally proved in the k = 1 case in [10] and for arbitrary k in [11]. Both of those proofs prove the result directly, without characterizing the cones C min,k as in Theorem 5. and C max,k m m 5. Norms on Operator Systems Given an operator system V , the matrix norm induced by the matrix order (cid:8)Cm(cid:9)∞ defined for X ∈ Mm(V ) to be (9) m=1 is (cid:13)(cid:13)X(cid:13)(cid:13)Mm(V ) := inf(cid:26)r :(cid:18)rem X X∗ rem(cid:19) ∈ C2m(cid:27) . In the particular case of X ∈ Mm(OM IN k(V )) or X ∈ Mm(OM AX k(V )), we will denote the norm (9) by (cid:13)(cid:13)X(cid:13)(cid:13)Mm(OM IN k(V )) and (cid:13)(cid:13)X(cid:13)(cid:13)Mm(OM AX k(V )), respectively. Our first result characterizes (cid:13)(cid:13)X(cid:13)(cid:13)Mm(OM IN k(Mn)) in terms of the Schmidt rank of pure states, much like Theorem 1 characterized (cid:13)(cid:13)X(cid:13)(cid:13)Mm(M IN k(Mn)). Theorem 8. Let X ∈ Mm(OM IN k(Mn)). Then (cid:13)(cid:13)X(cid:13)(cid:13)Mm(OM IN k(Mn)) = sup vi,wi(cid:8)hvXwi : SR(vi), SR(wi) ≤ k and ∃ P ∈ Mm s.t. (P ⊗ In)vi = wi(cid:9). Proof. Given X ∈ Mm(OM IN k(Mn)), consider the operator X :=(cid:18)rIn X rIn(cid:19) ∈ (M2 ⊗ Mm) ⊗ Mn ∼= M2m ⊗ Mn. X∗ Then X ∈ C min,k If we multiply on the left and the right by a Schmidt-rank k vector vi :=Pk where aii = αi11i ⊗ ai1i + αi22i ⊗ ai2i ∈ C2 ⊗ Cm and bii ∈ Cn, we get 2m (Mn) if and only if hv Xvi ≥ 0 for all vi ∈ C2m ⊗ Cn with SR(vi) ≤ k. i=1 βiaii⊗bii, k i α2 i1 + β2 i α2 i2(cid:1) + Xij=1 2αi1αj2βiβjRe(cid:0)(hai1 ⊗ hbi)X(aj2i ⊗ bji)(cid:1) 2αi1αj2βiβjRe(cid:0)(hai1 ⊗ hbi)X(aj2i ⊗ bji)(cid:1) k hv Xvi = Xi=1 = r + k r(cid:0)β2 Xij=1 = r + 2c1c2Re(cid:0)hv1Xv2i(cid:1), where c1v1i := Pk i=1 αi1βiai1i ⊗ bii, c2v2i := Pk that the normalization of the Schmidt coefficients tells us that c2 i=1 αi2βiai2i ⊗ bii ∈ Cm ⊗ Cn. Notice 2 = 1. Also notice 1 + c2 OPERATOR SPACES AND OPERATOR SYSTEMS IN ENTANGLEMENT THEORY 11 that v1i and v2i can be written in this way using the same vectors bii on the second subsystem if and only if there exists P ∈ Mm such that (P ⊗ In)v1i = v2i. Now taking the infimum over r and requiring that the result be non-negative tells us that the quantity we are interested in is 1 + c2 (cid:13)(cid:13)X(cid:13)(cid:13)Mm(OM IN k(Mn)) = supn2c1c2Re(cid:0)hv1Xv2i(cid:1) : SR(v1i), SR(v2i) ≤ k, c2 ∃ P ∈ Mm s.t. (P ⊗ In)v1i = v2io ∃ P ∈ Mm s.t. (P ⊗ In)v1i = v2io, where the final equality comes applying a complex phase to v1i so that Re(hv1Xv2i) = hv1Xv2i, and from Holder's inequality telling us that the supremum is attained when c1 = c2 = 1/√2. = supn(cid:12)(cid:12)hv1Xv2i(cid:12)(cid:12) : SR(v1i), SR(v2i) ≤ k and 2 = 1, (cid:3) Of course, the matrix norm induced by the matrix order is not the only way to define a norm on the various levels of the operator system V . What is referred to as the order norm of v ∈ Vh [25] is defined via (10) kvkor := inf{t ∈ R : −te ≤ v ≤ te}. It is not difficult to see that for a Hermitian element X = X∗ ∈ Mm(V ), the matrix norm induced by the matrix order (9) coincides with the order norm (10). It was shown in [25] how the order norm on Mm(V )h can be extended (non-uniquely) to a norm on all of Mm(V ). Furthermore, there exists a minimal order norm k · km and a maximal order norm k · kM satisfying k · km ≤ k · kM ≤ 2k · km. We will now examine properties of these two norms as well as some other norms (all of which coincide with the order norm on Hermitian elements) on the k-super minimal operator system structures. We will consider an operator X ∈ Mm(OM IN k(Mn)), where recall by this we mean X ∈ Mm(Mn), where the operator system structure on the space is OM IN k(Mn). Then we recall the minimal order norm, decomposition norm k · kdec, and maximal order norm from [25]: r r r : X = (cid:13)(cid:13)X(cid:13)(cid:13)m := sup(cid:8)f (X) : f : Mm(OM IN k(Mn)) → C a pos. linear functional s.t. f (I) = 1(cid:9), (cid:13)(cid:13)X(cid:13)(cid:13)dec := inf((cid:13)(cid:13)(cid:13)(cid:13) λiPi(cid:13)(cid:13)(cid:13)(cid:13)or Xi=1 (cid:13)(cid:13)X(cid:13)(cid:13)M := inf( r Xi=1 λi(cid:13)(cid:13)Hi(cid:13)(cid:13)or : X = λiPi with Pi ∈ C min,k λiHi with Hi = H∗i and λi ∈ C) . (Mn) and λi ∈ C) , Our next result shows that the minimal order norm can be thought of in terms of vectors with Schmidt rank no greater than k, much like the norms k· kS(k) and k· kMm(OM IN k(Mn)) introduced earlier. Xi=1 Xi=1 m 12 N. JOHNSTON, D. W. KRIBS, V. I. PAULSEN, AND R. PEREIRA Theorem 9. Let X ∈ Mm(OM IN k(Mn)). Then (cid:13)(cid:13)X(cid:13)(cid:13)m = sup vi n(cid:12)(cid:12)hvXvi(cid:12)(cid:12) : SR(vi) ≤ ko. m Proof. Note that if we define a linear functional f : Mm(OM IN k(Mn)) → C by f (X) = hvXvi for some fixed vi with SR(vi) ≤ k then it is clear that f (X) ≥ 0 whenever X ∈ C min,k (Mn) (by definition of k-block positivity) and f (I) = 1. The "≥" inequality follows immediately. To see the other inequality, note that if X = X∗ = (Xij ) and vi,wi can be written vi =Pk r=1 γrcri ⊗ bri, then r=1 αrari ⊗ bri and wi =Pk hvXwi = αrγshar(hbrXijbsi)ijcsi k Xrs=1 = (α1ha1,··· , αkhak)  (hb1Xijb1i)ij ... hbkXijb1i)ij ··· . . . ··· hb1Xijbki)ij ... hbkXijbki)ij    γ1c1i...  γkcki   . Because X is Hermitian, so is the operator in the last line above, so if we take the supremum over all vi,wi of this form, we may choose αiaii = γicii for all i. It follows that supnhvXvi : SR(vi) ≤ ko = supn(cid:12)(cid:12)hvXwi(cid:12)(cid:12) : SR(vi), SR(wi) ≤ k and ∃ P ∈ Mm s.t. (P ⊗ In)vi = wio. The "≤" inequality follows from Theorem 8, the fact that k·kMm(OM IN k(Mn)) is an order norm, and the minimality of k · km among order norms. (cid:3) The characterization of k · km given by Theorem 9 can be thought of as in the same vein as [25, Proposition 5.8], where it was shown that for a unital C∗-algebra, k · km coincides with the numerical radius. In our setting, k·km can be thought of as a bipartite analogue of the numerical radius, which has been studied in quantum information theory in the k = 1 case [26]. In the case where X is not Hermitian, equality need not hold between any of the order norms that have been introduced. We now briefly investigate how they compare to each other in general. Proposition 10. Let X ∈ Mm(OM IN k(Mn)). Then (cid:13)(cid:13)X(cid:13)(cid:13)m ≤(cid:13)(cid:13)X(cid:13)(cid:13)Mm(OM IN k(Mn)) ≤(cid:13)(cid:13)X(cid:13)(cid:13)dec ≤(cid:13)(cid:13)X(cid:13)(cid:13)M . Proof. The first and last inequalities follow from the fact that k · km and k · kM are the minimal and maximal order norms, respectively. Thus, all that needs to be shown is that (cid:13)(cid:13)X(cid:13)(cid:13)Mm(OM IN k(Mn)) ≤(cid:13)(cid:13)X(cid:13)(cid:13)dec. To this end, let vi,wi ∈ Cm⊗Cn with SR(vi), SR(wi) ≤ k be such that there exists some P ∈ Mm such that (P ⊗ In)vi = wi. Then for any decomposition X = Pr (Mn) and λi ∈ C we can use a similar i=1 λiPi with Pi ∈ C min,k m 1 √n n−1 Xi=0 φi := ii ⊗ ii It is easily verified that if vi =Pk n−1 k n−1 1 √n ψi := Xi=0 i=1 αiaii ⊗ bii then ii ⊗ i + 1(mod n)i. OPERATOR SPACES AND OPERATOR SYSTEMS IN ENTANGLEMENT THEORY 13 because each Pi is k-block positive (by Theorem 5). Thus r argument to that used in the proof of Theorem 9 to see that hvPivi ≥ (cid:12)(cid:12)hvPiwi(cid:12)(cid:12) ≥ 0 λiPi(cid:13)(cid:13)(cid:13)(cid:13)or (cid:12)(cid:12)hvXwi(cid:12)(cid:12) =(cid:12)(cid:12)(cid:12)(cid:12) Taking the supremum over all such vectors vi and wi and the infimum over all such decompositions of X gives the result. λihvPivi ≤(cid:13)(cid:13)(cid:13)(cid:13) λi(cid:12)(cid:12)hvPiwi(cid:12)(cid:12) ≤ λihvPiwi(cid:12)(cid:12)(cid:12)(cid:12) Xi=1 Xi=1 Xi=1 Xi=1 ≤ (cid:3) . r r r We know in general that k · km and k · kM can differ by at most a factor of two. We now present an example some of these norms and to demonstrate that in fact even k · km and k · kMm(OM IN k(Mn)) can differ by a factor of two. Example 11. Consider the rank-1 operator X := φihψ ∈ OM IN k n (Mn), where k k 1 1 = n−1 n(cid:12)(cid:12)(cid:12)(cid:12) n(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)hvXvi(cid:12)(cid:12) = αrαshariihbriihjasihj + 1(mod n)bsi(cid:12)(cid:12)(cid:12)(cid:12) Xrs=1 Xij=0 αrarihbr(cid:17) · Tr(cid:16) Xr=1 hj(cid:16) Xr=1 Xj=0 αrarihbr(cid:17)j + 1(mod n)i(cid:12)(cid:12)(cid:12)(cid:12) In the final line above we have the trace of an operator with rank at most k, multiplied by the sum of the elements on the superdiagonal of the same operator, subject to the constraint that the Frobenius norm of that operator is equal to 1. It follows that (cid:12)(cid:12)hvXvi(cid:12)(cid:12) ≤ k 2n and 2n (equality can be seen by taking vi =Pk−1 1√2kii ⊗ (ii + i + 1(mod n)i)). so kXkm = k i=0 ii⊗ii and wi = 1√kPk−1 To see that kXkop is twice as large, consider vi = 1√kPk−1 i=0 ii⊗ i + 1(mod n)i. Then it is easily verified that hvXwi = k n . Moreover, if P ∈ Mm is the cyclic permutation matrix such that Pii = i − 1(mod n)i for all i then (P ⊗ In)vi = wi, showing that kXkop ≥ k n . i=0 . 6. Contractive Maps as Separability Criteria We now investigate the completely bounded version of the k-minimal operator space norms and k-super minimal operator system norms that have been introduced. We will see that these completely bounded norms can be used to provide a characterization of Schmidt number analogous to its characterization in terms of k-positive maps. Given operator spaces V and W , the completely bounded (CB) norm from V to W is defined by (cid:13)(cid:13)Φ(cid:13)(cid:13)CB(V,W ) := sup m≥1n(cid:13)(cid:13)(idm ⊗ Φ)(X)(cid:13)(cid:13)Mm(W ) : X ∈ Mm(V ) with (cid:13)(cid:13)X(cid:13)(cid:13)Mm(V ) ≤ 1o. Clearly this reduces to the standard completely bounded norm of Φ in the case when V = Mr and W = Mn. We will now characterize this norm in the case when V = Mr and 14 N. JOHNSTON, D. W. KRIBS, V. I. PAULSEN, AND R. PEREIRA W = M IN k(Mn). In particular, we will see that the k-minimal completely bounded norm of Φ is equal to the perhaps more familiar operator norm(cid:13)(cid:13)idk ⊗ Φ(cid:13)(cid:13) -- that is, the CB norm in this case stabilizes in much the same way that the standard CB norm stabilizes (indeed, in the k = n case we get exactly the standard CB norm). This result was originally proved in [22], but we prove it here using elementary means for completeness and clarity, and also because we will subsequently need the operator system version of the result, which can be proved in the same way. Theorem 12. Let Φ : Mr → Mn be a linear map and let 1 ≤ k ≤ n. Then (cid:13)(cid:13)idk ⊗ Φ(cid:13)(cid:13) =(cid:13)(cid:13)Φ(cid:13)(cid:13)CB(Mr ,M IN k(Mn)). Proof. To see the "≤" inequality, simply notice that (cid:13)(cid:13)Y(cid:13)(cid:13)Mk(M IN k(Mn)) = (cid:13)(cid:13)Y(cid:13)(cid:13)Mk(Mn) for all Y ∈ Mk(Mn). We thus just need to show the "≥" inequality, which we do in much the same manner as Smith's original proof that the standard CB norm stabilizes. First, use Theorem 1 to write (11) (cid:13)(cid:13)Φ(cid:13)(cid:13)CB(Mr ,M IN k(Mn)) = sup m≥1n(cid:13)(cid:13)(idm ⊗ Φ)(X)(cid:13)(cid:13)S(k) :(cid:13)(cid:13)X(cid:13)(cid:13) ≤ 1o. Now fix m ≥ k and a pure state vi ∈ Cm⊗ Cn with SR(vi) ≤ k. We begin by showing that there exists an isometry V : Ck → Cm and a state vi ∈ Ck⊗ Cn such that (V ⊗ In)vi = vi. To this end, write vi in its Schmidt Decomposition vi =Pk i=1 αiaii⊗bii. Because k ≤ m, we may define an isometry V : Ck → Cm by V ii = aii for 1 ≤ i ≤ k. If we define vi :=Pk i=1 αiii ⊗ bii then (V ⊗ In)vi = vi, as desired. Now choose X ∈ Mm(Mr) such that (cid:13)(cid:13) X(cid:13)(cid:13) ≤ 1 and the supremum (11) (holding m fixed) is attained by X. Then choose vectors vi,wi ∈ Cm ⊗ Cn with SR(vi), SR(wi) ≤ k such that As we saw earlier, there exist isometries V, W : Ck → Cm and unit vectors vi, wi ∈ Ck⊗Cn such that (V ⊗ In)vi = vi and (W ⊗ In) wi = wi. Thus (cid:13)(cid:13)(idm ⊗ Φ)( X)(cid:13)(cid:13)S(k) =(cid:12)(cid:12)hv(idm ⊗ Φ)( X)wi(cid:12)(cid:12). (cid:13)(cid:13)(idm ⊗ Φ)( X)(cid:13)(cid:13)S(k) =(cid:12)(cid:12)hv(V ∗ ⊗ In)(idm ⊗ Φ)( X)(W ⊗ In) wi(cid:12)(cid:12) =(cid:12)(cid:12)hv(idk ⊗ Φ)((V ∗ ⊗ Ir) X(W ⊗ Ir)) wi(cid:12)(cid:12) ≤(cid:13)(cid:13)(idk ⊗ Φ)((V ∗ ⊗ Ir) X(W ⊗ Ir))(cid:13)(cid:13) ≤ supn(cid:13)(cid:13)(idk ⊗ Φ)(X)(cid:13)(cid:13) : X ∈ Mk(Mr) with (cid:13)(cid:13)X(cid:13)(cid:13) ≤ 1o, where the final inequality comes from the fact that(cid:13)(cid:13)(V ∗ ⊗ Ir) X(W ⊗ Ir)(cid:13)(cid:13) ≤ 1. The desired inequality follows, completing the proof. We will now show that the operator system versions of these norms have applications to testing separability of quantum states. To this end, notice that if we instead consider the completely bounded norm from Mr to the k-super minimal operator systems on Mn, then a statement that is analogous to Theorem 12 holds. Its proof can be trivially modified to (cid:3) OPERATOR SPACES AND OPERATOR SYSTEMS IN ENTANGLEMENT THEORY 15 show that if Φ : Mr → Mn and 1 ≤ k ≤ n then (12) = sup supn(cid:12)(cid:12)hv(idk ⊗ Φ)(X)vi(cid:12)(cid:12) :(cid:13)(cid:13)X(cid:13)(cid:13) ≤ 1, X = X∗o m≥1n(cid:12)(cid:12)hv(idm ⊗ Φ)(X)vi(cid:12)(cid:12) :(cid:13)(cid:13)X(cid:13)(cid:13) ≤ 1, X = X∗, SR(vi) ≤ ko. Equation (12) can be thought of as a stabilization result for the completely bounded version of the norm described by Theorem 9. We could also have picked one of the other order norms on the k-super minimal operator systems to work with, but from now on we will be working exclusively with Hermiticity-preserving maps Φ. So by the fact that all of the operator system order norms are equal on Hermitian operators, it follows that these versions of their completely bounded norms are all equal as well. Before proceeding, we will need to define some more notation. If Φ : Mn → Mr is a linear map, then we define a Hermitian version of the induced trace norm of Φ: H tr := supn(cid:13)(cid:13)Φ(X)(cid:13)(cid:13)tr :(cid:13)(cid:13)X(cid:13)(cid:13)tr ≤ 1, X = X∗o. (cid:13)(cid:13)Φ(cid:13)(cid:13) Because of convexity of the trace norm, it is clear that the above norm is unchanged if instead of being restricted to Hermitian operators, the supremum is restricted to positive operators or even just projections. Now by taking the dual of the left and right norms described by Equation (12), and using the fact that the operator norm is dual to the trace norm, we arrive at the following corollary: Corollary 13. Let Φ : Mn → Mr be a Hermiticity-preserving linear map and let 1 ≤ k ≤ n. Then (cid:13)(cid:13)idk ⊗ Φ(cid:13)(cid:13) H tr = sup m≥1n(cid:13)(cid:13)(idm ⊗ Φ)(ρ)(cid:13)(cid:13)tr : ρ ∈ Mm ⊗ Mn with SN (ρ) ≤ ko. We will now characterize the Schmidt number of a state ρ in terms of maps that are contractive in the norm described by Corollary 13. Our result generalizes the separability test of [6]. We begin with a simple lemma that will get us most of the way to the linear contraction characterization of Schmidt number. The k = 1 version of this lemma appeared as [6, Lemma 1], though our proof is more straightforward. Lemma 14. Let ρ ∈ Mm ⊗ Mn be a density operator. Then SN (ρ) ≤ k if and only if (idm ⊗ Φ)(ρ) ≥ 0 for all trace-preserving k-positive maps Φ : Mn → M2n. Proof. The "only if" implication of the proof is clear, so we only need to establish that if SN (ρ) > k then there is a trace-preserving k-positive map Φ : Mn → M2n such that (idm ⊗ Φ)(ρ) 6≥ 0. To this end, let Ψ : Mn → Mn be a k-positive map such that (idm ⊗ Ψ)(ρ) 6≥ 0 (which we know exists by [7, 8]). Without loss of generality, Ψ can be scaled n . Then if Ω : Mn → Mn is the completely depolarizing channel defined by Ω(ρ) = 1 n In for all ρ ∈ Mn, it follows that (Ω − Ψ)(ρ) ≥ 0 for all ρ ≥ 0 and so the map Φ := Ψ ⊕ (Ω − Ψ) : Mn → M2n is k-positive (and easily seen to be trace-preserving). Because (idm ⊗ Ψ)(ρ) 6≥ 0, we have (idm ⊗ Φ)(ρ) 6≥ 0 as well, completing the proof. so that (cid:13)(cid:13)Ψ(cid:13)(cid:13)tr ≤ 1 (cid:3) 16 N. JOHNSTON, D. W. KRIBS, V. I. PAULSEN, AND R. PEREIRA We are now in a position to prove the main result of this section. Note that in the k = 1 case of the following theorem it is not necessary to restrict attention to Hermiticity- preserving linear maps Φ (and indeed this restriction was not made in [6]), but our proof for arbitrary k does make use of Hermiticity-preservation. Theorem 15. Let ρ ∈ Mm ⊗ Mn be a density operator. Then SN (ρ) ≤ k if and only Proof. To see the "only if" implication, simply use Corollary 13 with r = 2n. if (cid:13)(cid:13)(idm ⊗ Φ)(ρ)(cid:13)(cid:13)tr ≤ 1 for all Hermiticity-preserving linear maps Φ : Mn → M2n with (cid:13)(cid:13)idk ⊗ Φ(cid:13)(cid:13) Hermiticity-preserving and has (cid:13)(cid:13)Ψ(cid:13)(cid:13) positive trace-preserving map Φ has(cid:13)(cid:13)idk⊗Φ(cid:13)(cid:13) linear maps Φ with (cid:13)(cid:13)idk ⊗ Φ(cid:13)(cid:13) For the "if" implication, observe that any positive trace-preserving map Ψ is necessarily tr ≤ 1. Letting Ψ = idk ⊗ Φ then shows that any k- tr ≤ 1. Thus the set of Hermiticity-preserving tr ≤ 1 contains the set of k-positive trace-preserving maps, so the "if" implication follows from Lemma 14. (cid:3) H tr ≤ 1. H H H Acknowledgements. Thanks are extended to Marius Junge for drawing our attention to the k-minimal and k-maximal operator space structures. N.J. was supported by an NSERC Canada Graduate Scholarship and the University of Guelph Brock Scholarship. D.W.K. was supported by Ontario Early Researcher Award 048142, NSERC Discovery Grant 400160 and NSERC Discovery Accelerator Supplement 400233. R.P. was supported by NSERC Discovery Grant 400096. References [1] V. I. Paulsen, Completely bounded maps and operator algebras. Cambridge University Press, Cambridge (2003). [2] G. Pisier, Introduction to operator space theory. Cambridge University Press, Cambridge (2003). [3] R. Horodecki, P. Horodecki, M. Horodecki, K. Horodecki, Quantum entanglement. Rev. Mod. Phys. 81, 865 -- 942 (2009). [4] I. Bengtsson, K. Zyczkowski, Geometry of quantum states: an introduction to quantum entanglement. Cambridge University Press, Cambridge (2006). [5] M. Horodecki, P. Horodecki, R. Horodecki, Separability of mixed states: necessary and sufficient condi- tions. Physics Letters A 223, 1 -- 8 (1996). [6] M. Horodecki, P. Horodecki, R. Horodecki, Separability of mixed quantum states: linear contractions approach. Open Syst. Inf. Dyn. 13, 103 (2006). [7] B. M. Terhal, P. Horodecki, Schmidt number for density matrices. Phys. Rev. A 61, 040301R (2000). [8] K. S. Ranade, M. Ali, The Jamio lkowski isomorphism and a simplified proof for the correspondence between vectors having Schmidt number k and k-positive maps. Open Syst. Inf. Dyn. 14, 371 -- 378 (2007). [9] M. Horodecki, P. W. Shor, M. B. Ruskai, General entanglement breaking channels. Rev. Math. Phys 15, 629 -- 641 (2003). [10] V. Paulsen, I. Todorov, M. Tomforde, Operator system structures on ordered spaces. Proc. of the LMS (to appear). [11] B. Xhabli, Universal operator system structures on ordered spaces and their applications. PhD Thesis (2009). [12] D. Chruscinski, A. Kossakowski, On partially entanglement breaking channels. Open Sys. Information Dyn. 13, 17 -- 26 (2006). [13] N. Johnston, D. W. Kribs, A family of norms with applications in quantum information theory. To appear in J. Math. Phys. (2010). arXiv:0909.3907v3 [quant-ph] OPERATOR SPACES AND OPERATOR SYSTEMS IN ENTANGLEMENT THEORY 17 [14] N. Johnston, D. W. Kribs, A family of norms with applications in quantum information theory II. Preprint (2010). arXiv:1006.0898v1 [quant-ph] [15] N. Johnston, Characterizing operations preserving separability measures via linear preserver problems. Preprint (2010). arXiv:1008.3633v1 [quant-ph] [16] L. Pankowski, M. Piani, M. Horodecki, P. Horodecki, A few steps more towards NPT bound entangle- ment. To appear in IEEE Trans. Inf. Theory (2010). arXiv:0711.2613v2 [quant-ph] [17] D. Chru´sci´nski, A. Kossakowski, Spectral conditions for positive maps. Commun. Math. Phys. 290, 10511064 (2009) [18] D. Chru´sci´nski, A. Kossakowski, G. Sarbicki, Spectral conditions for entanglement witnesses vs. bound entanglement. Preprint (2009). arXiv:0908.1846v1 [quant-ph] [19] D. P. DiVincenzo, P. W. Shor, J. A. Smolin, B. M. Terhal, A. V. Thapliyal, Evidence for Bound Entan- gled States with Negative Partial Transpose, Phys. Rev. A 61, 062312 (2000). arXiv:quant-ph/9910026v3 [20] R. R. Smith, Completely bounded maps between C∗-algebras. J. London Math. Soc. 27, 157-166 (1983). [21] A.Yu. Kitaev, Quantum computations: algorithms and error correction, Russian Math. Surveys 52 (1997), 1191-1249. [22] T. Oikhburg, E. Ricard, Operator spaces with few completely bounded maps. Math. Ann. 328, 229-259 (2004). [23] N. Johnston, D. W. Kribs, and V. Paulsen, Computing Stabilized Norms for Quantum Operations. [24] Quantum Information & Computation 9 1 & 2, 16 -- 35 (2009). L. Skowronek, E. Størmer, and K. Math. Phys. 50, 062106 (2009). Zyczkowski, Cones of positive maps and their duality relations. J. [25] V. Paulsen, M. Tomforde, Vector spaces with an order unit. Indiana Univ. Math. J (to appear). arXiv:0712.2613v4 [math.OA] [26] P. Gawron, Z. Puchala, J. A. Miszczak, L. Skowronek, M.-D. Choi, K. Zyczkowski, Local numerical range: a versatile tool in the theory of quantum information. E-print: arXiv:0905.3646v1 [quant-ph] Department of Mathematics and Statistics, University of Guelph, Guelph, Ontario N1G 2W1, Canada E-mail address: [email protected] Department of Mathematics and Statistics, University of Guelph, Guelph, Ontario N1G 2W1, Canada and Institute for Quantum Computing, University of Waterloo, Waterloo, Ontario N2L 3G1, Canada E-mail address: [email protected] Department of Mathematics, University of Houston, Houston, Texas 77204-3476, U.S.A. E-mail address: [email protected] Department of Mathematics and Statistics, University of Guelph, Guelph, Ontario N1G 2W1, Canada E-mail address: [email protected]
1611.06225	2	1611	2017-06-30T11:20:05	Hopf images in locally compact quantum groups	[ "math.OA", "math.FA", "math.QA" ]	This manuscript is devoted to the study of the concept of a generating subset (a.k.a. Hopf image of a morphism) in the setting of locally compact quantum groups. The aim of this paper is to provide an accurate description of the Hopf image of a given morphism. We extend and unify the previously existing approaches for compact and discrete quantum groups and present some results that can shed light on some local perspective in the theory of quantum groups. In particular, we provide a characterization of fullness of Hopf image in the language of partial actions as well as in representation-theoretic terms in the spirit of representation $C^*$-categories, extending some known results not only to the broader setting of non-compact quantum groups, but also encompassing a broader setting of generating subsets.	math.OA	math	HOPF IMAGES IN LOCALLY COMPACT QUANTUM GROUPS PAWE L J ´OZIAK, PAWE L KASPRZAK, AND PIOTR M. SO LTAN Abstract. This manuscript is devoted to the study of the concept of a generating subset (a.k.a. Hopf image of a morphism) in the setting of locally compact quantum groups. The aim of this paper is to provide an accurate description of the Hopf image of a given morphism. We extend and unify the previously existing approaches for compact and discrete quantum groups and present some results that can shed light on some local perspective in the theory of quantum groups. In particular, we provide a characterization of fullness of Hopf image in the language of partial actions as well as in representation-theoretic terms in the spirit of representation C ∗-categories, extending some known results not only to the broader setting of non-compact quantum groups, but also encompassing a broader setting of generating subsets. Introduction The concept of a subgroup is central to understanding locally compact quantum groups as group-theoretic objects and accordingly it has been receiving an increasing interest in recent years, see e.g. [BB09, BY14, BCV17, DKSS12, KKS16, KSS16, KS14a, KS15]. It arises naturally when studying quantum homogeneous spaces, as the natural examples of homogeneous spaces are of quotient type, see [KS14b], or when studying the behavior of quantum groups with respect to some natural constructions, see e.g. [KSS16, Ver04, Wan95, CHK]. The starting point for this project was to improve the understanding of the notion of closed subgroup in the realm of locally compact quantum groups. Classically, in order for a non-empty closed subset X ⊂ G of a locally compact group G, to be a group, it has to satisfy X 2 ⊂ X and, if G is non-compact, also X −1 = X. Clearly, generically these conditions do not hold, and the way to cope with this is to consider the subgroup generated by the given subset X, i.e. the smallest closed subgroup containing it. One immediately realizes that this smallest subgroup is given by Sn∈Z X n (and in the compact case Z may be replaced by N). This viewpoint was already utilized in [BB10] in the case of Hopf ∗-algebras and in [SS16, BCV17, Chi15] in the case of compact quantum groups, where analytic issues are minor, and it was implicit in various writings on discrete quantum groups, and formulated explicitly e.g. in [Izu02, Ver05, Ver07]. Once the notion of generation was established in a satisfactory way in this restricted setting, the question of moving to a broader, and technically more involved, class of locally compact quantum groups arises naturally. In this manuscript we study the following concept: let G be a locally compact quantum group (in the sense of Kustermans-Vaes). Let B be a C∗-algebra and let β ∈ Mor(Cu 0 (G), B) be a morphism of C∗-algebras (in the sense of [Wor95]). We think of it as the Gelfand dual of a map bβ: X → G from a quantum space into a quantum group and ask what is the closed quantum subgroup (in the sense of Vaes, see [DKSS12]) of G generated by bβ(X) ⊂ G. Formally speaking, we consider the following category, which we denote by Cβ. Objects of Cβ are triples (π, H, β) consisting of: a closed quantum subgroup H of G such that π ∈ 0 (H)) is the associated morphism intertwining the coproducts and β ◦ π = β (as Mor(Cu morphisms of C∗-algebras), i.e. the map β factors through the C∗-algebra of functions on the 0 (H) (where H is embedded into G using π) and β ◦ π = β is the factorization. For subgroup Cu two objects h = (π, H, β), k = (π′, K, β′) ∈ Ob(Cβ), a morphism ϕ ∈ MorCβ (h, k) is a C∗-algebra morphism ϕ ∈ Mor(Cu 0 (H)) (note that the direction of arrows in Cβ is same as the direction 0 (G), Cu 0 (K), Cu 2010 Mathematics Subject Classification. Primary: 46L89 Secondary: 46L85, 46L52. Key words and phrases. Hopf image, locally compact quantum groups, quantum subgroups. 1 2 PAWE L J ´OZIAK, PAWE L KASPRZAK, AND PIOTR M. SO LTAN of maps on the level of quantum groups, which is opposite to the ones on the level of C∗-algebras of functions), which intertwines the respective coproducts and such that the following diagram commutes: Cu 0 (G) β B π β π′ Cu 0 (H) ϕ β′ Cu 0 (K) Diagram 1: Morphisms in Cβ The object we are interested in is the initial object of the category Cβ. Our study of Hopf images begins in Section 2 by showing, among others, that Theorem A. Given a locally compact quantum group G and a morphism β as above, there always exists the initial object of the category Cβ. This initial object is then called the Hopf image of β, following the situation studied first in the algebraic setting by T. Banica and J. Bichon in [BB10]. When this initial object happens to be (id, G, β), we then say that β is a generating morphism (this was earlier known as inner faithful morphism and as morphism faithful in the discrete quantum group sense in the context of Hopf ∗-algebras, but the latter term is long and does not extend nicely beyond the compact quantum group setting, and the former might unnecessarily suggest some connections to the adjoint action). The prominent examples of induction from subgroups of G to G of the following properties: Connes' embeddability of L∞(G), given in [BCV17], and residual fininite dimensionality of O(G), given in [Chi15], puts the setting of quantum group generated by a family of its subgroups in the center of the study of Hopf images. We analyze an analogous situation in Section 3. Intriguingly, the closed quantum subgroups H ⊂ G are characterized by the so-called Baaj- Vaes subalgebras of L∞(bG) (the precise definitions are given in Section 1.4). It follows from our discussion (cf. Theorem 2.6) that the most natural examples of invariant subalgebras of L∞(bG) -- the ones coming from representations of G -- satisfy the following phenomenon: if they are τ - preserved (which is the case e.g. for quantum groups that are compact, discrete or of Kac type), they are automatically Baaj-Vaes. This means that in many examples part of the assumptions of the Baaj-Vaes theorem are actually implied by sole invariance. It is not known to the authors whether this phenomenon holds in full generality. The Hopf image construction highlights a certain local perspective in the theory of quantum groups. This rather vague idea can be forged into concrete statements. For instance, classically a homomorphism G → K is uniquely determined by its values on the generating set. A similar statement can be proved in the quantum case: Theorem B. With the notation as above, assume that β is a generating morphism and consider two homomorphisms G → K described by ϕ, ϕ ∈ Mor(Cu 0 (G)). If β◦ϕ = β◦ ϕ, then ϕ = ϕ. 0 (K), Cu For G discrete the converse is also true in the following sense: if a quantum subset has the property that homomorphisms are uniquely determined by the values they attain on this set, it is generating. We address the problem of describing the Hopf image of a given morphism by means of other objects related to it. Associated to the map β, philosophically being the Gelfand dual of a map X → G, there are: restrictions of representations of G to X; partial action by right shifts X y G HOPF IMAGES IN LOCALLY COMPACT QUANTUM GROUPS 3 and a representation of the dual quantum group X ∈ L∞(bG) ¯⊗B(H). We have the following result expressing generation in these terms: Theorem C. With the notation as above, consider the following four statements: (i) The morphism β is generating. (ii) {(id ⊗ω)X ω ∈ B∗}′′ = L∞(bG). (iii) The partial action X y G is ergodic. (iv) any two distinct representations of G, when restricted to X, remain distinct. We have (i) =⇒ (ii) ⇐⇒ (iii) ⇐⇒ (iv). Moreover, (i) =⇒ (ii) provided that G is compact or discrete or τ u t (ker(β)) ⊆ ker(β) for all t ∈ R (in particular, if G is of Kac type). The ingredients of Theorem C are described in detail in course of constructing the Hopf image. We later move to the study of generation-type questions in the language of representation category of the quantum group in question. Assume {Hi}i∈I is a family of closed subgroups of G and denote by U Hi the restriction of a representation U of G to Hi. Recall that HomG(U, U ) is the set of intertwiners between U and U as representations of G. We have Theorem D. G is generated by (Hi)i∈I if and only if for all pairs of representations U, U of G we have that HomG(U, U ) = \i∈I HomHi (U Hi , U Hi) The manuscript is organized as follows: Section 1 is devoted to establishing notation and con- ventions. We collect also from various places in the literature some ingredients of the theory of locally compact quantum groups that are most relevant to our constructions. In Section 2 we deal with Theorem A and reveal the structure of objects useful in the later study of Hopf images. In Section 3 we analyze the case of the generating set coming from a family of subgroups. Section 4 is devoted to the proof of Theorem B and its converse. In Section 5 we introduce the concept of restricting a representation to a subset and prove a reformulation of Theorem C. In Section 6 we study the concept of restriction of a representation to a subset from the point of view of inter- twiners. In particular, we derive Theorem D. Section 7 discusses some examples and applications of the Hopf image construction. 1. Preliminaries 1.1. C∗-algebras, von Neumann algebras etc. The symbol σ denotes the flip, i.e. unique extension of the map A ⊗ B ∋ a ⊗ b 7→ b ⊗ a ∈ B ⊗ A. We use the leg numbering notation, which is now commonly understood: for T ∈ M (A ⊗ C), we denote by T13 ∈ M (A ⊗ B ⊗ C) the operator given by (σ ⊗ id)(1 ⊗ T ). Similarly, if t ∈ M (B), then 1 ⊗ t ⊗ 1 ∈ M (A ⊗ B ⊗ C) will be denoted by t2, and so on. For two vectors ξ, η ∈ H, by the functional ωξ,η ∈ B(H)∗ we mean a map T 7→ hξT ηi (note that the inner product is linear in the right variable). We will use the following lemma a few times: Lemma 1.1. Let A ⊆ B(H), B ⊆ B(K) be concrete C∗-algebras and let A = A′′ and B = B′′. Let T ∈ M (A ⊗ B) ⊆ A ¯⊗B and let (ωi)i∈I ⊆ B(H)∗ be a net of normal functionals such that i∈I−−→ ω ∈ B(H)∗ in the weak∗-topology. Then xi = (ωi ⊗ id)T i∈I−−→ (ω ⊗ id)T = x in σ-WOT of ωi B. Proof. Recall that σ-WOT is the weak∗-topology in B. Pick µ ∈ B∗. We have (1.1) Since t = (id ⊗µ)T ∈ A and ωi µ(xi) = (ωi ⊗ µ)(T ) = ωi((id ⊗µ)T ). i∈I−−→ ω ∈ A∗ weak∗, we have ωi(t) i∈I−−→ ω(t). This is equivalent to µ(xi) i∈I−−→ ω(t) = (ω ⊗ µ)(T ) = µ((ω ⊗ id)T ) which finishes the proof. (cid:3) 4 PAWE L J ´OZIAK, PAWE L KASPRZAK, AND PIOTR M. SO LTAN 1.2. Locally compact quantum groups. We work in the setting of locally compact quantum groups as defined by Kustermans and Vaes ( [KV00, Vae01]), but sometimes we use the results proven in the setting of multiplicative unitaries ( [BS93,SW07,Wor96]), as the latter axiomatization covers the former. Hence we use the following objects to study a quantum group G: (1) The von Neumann algebra L∞(G), endowed with a coproduct ∆G and n.s.f. weights ϕG, ψG satisfying the left- and right-invariance conditions (called the left and right Haar weights, respectively); (2) L1(G), the predual of L∞(G); (3) The reduced C∗-algebra C0(G), endowed with the same structure as above; (4) The universal C∗-algebra Cu 0 (G), its coproduct will be denoted by ∆u quotient map will be denoted by ΛG: Cu 0 (G) → C0(G); G and the canonical (5) The Kac-Takesaki operator WG ∈ M (C0(bG) ⊗ C0(G)) and its universal companions: 0 (G)) and VVG ∈ M (Cu WG ∈ M (Cu 0 (G)); The Kac-Takesaki operator WG implements the coproduct in C0(G) (and also in L∞(G)) in the following way: 0 (bG) ⊗ C0(G)), WG ∈ M (C0(bG) ⊗ Cu 0 (bG) ⊗ Cu (1.2) ∆G(x) = WG(x ⊗ 1)(WG)∗ The semi-universal incarnations of the Kac-Takesaki operator are linked to the Kac-Takesaki operator by means of the reducing morphism: (id ⊗ΛG)W G = WG = (Λ bG ⊗ id)WG. The universal version of the Kac-Takesaki operator is linked to the semi-universal companions in a similar manner: (id ⊗ΛG)VVG = WG and (Λ bG ⊗ id)VVG = WG. These operators obey the following pentagonal-like equation (see [MRW12, Proposition 4.4]): (1.3) VVG 13 = (WG 12)∗W G 23 WG 12(W G 23)∗ ∈ M (C0(bG) ⊗ K(L2(G)) ⊗ C0(G)) bG = σ(WG)∗. Then The dual quantum group bG is governed by its own Kac-Takesaki operator W the coproduct in L∞(bG) takes the form: ∆bG(x) = σ(WG)∗(x ⊗ 1)σ(WG) = σ(cid:0)(WG)∗(1 ⊗ x)WG(cid:1) . (1.4) When discussing a single locally compact quantum group, we will often drop the G and G decora- tions of the coproduct, Kac-Takesaki operators etc. Then the structure of the dual group will be decorated only with the hat decoration, e.g. b∆ will be the coproduct in L∞(bG) etc. The study of G and bG is supplemented with (6) the unitary antipode R: C0(G) → C0(G), living also on the von Neumann algebra level: R: L∞(G) → L∞(G); (7) its universal lift: Ru: Cu (8) the scaling group: Neumann level: τ G t : L∞(G) → L∞(G); 0 (G) → Cu 0 (G); for every t ∈ R there is τ G t : C0(G) → C0(G), extending to the von (9) the universal counterpart: the group of transformations τ u t : Cu 0 (G) → Cu 0 (G). (10) analogous structure for the dual: bRu, bR, bτ u (11) the analytic continuation of τt to the upper half-plane C+ yields the analytic generator of the group of transformations τt, denoted τi/2. It appears in the polar decomposition of the antipode: S = R◦τi/2. In case τt = id, one has τi/2 = id and S = R is bounded. In such cases we say that G is of Kac type. t and bτt. The scaling groups and unitary antipodes are compatible with the reducing morphisms in the following sense: (1.5) τt ◦Λ = Λ◦τ u t and R◦Λ = Λ◦Ru and similarly for bG. Moreover, the scaling groups and unitary antipodes are compatible with the universal version of the Kac-Takesaki operator in the following sense (cf. [SW07, Proposition 39, HOPF IMAGES IN LOCALLY COMPACT QUANTUM GROUPS 5 Lemma 40 and Proposition 42]): (1.6) t ⊗ τ u t )VV = VV and (bτ u (bRu ⊗ Ru)VV = VV 1.3. Representation theory. By a representation of a locally compact quantum group G we will always mean a unitary element U ∈ M (K(HU ) ⊗ C0(G)) satisfying: (1.7) Recall that M (K(H) ⊗ C0(G)) ⊆ B(HU ) ¯⊗L∞(G), we often just write U ∈ B(HU ) ¯⊗L∞(G). It is known that any U ∈ B(HU ) ¯⊗L∞(G) satisfying (1.7) actually satisfies U ∈ M (K(HU ) ⊗ C0(G)), see e.g. [Wor96, Theorem 1.6(2)]. (id ⊗∆G)U = U12U13 Out of two representations U ∈ M (K(HU ) ⊗ C0(G)) and U ∈ M (K(H U ) ⊗ C0(G)) one can form two new representations: the direct sum and tensor product, we will need a precise formula of the former later on, so we recall it. The direct sum is obtained by using the canonical inclusion maps ι: B(HU ) ֒→ B(HU ⊕ H U )) and ι: B(H U ) ֒→ B(HU ⊕ H U )) induced by the spatial maps HU , H U ֒→ HU ⊕ H U . Then the direct sum is nothing but (1.8) U ⊕ V : = (ι ⊗ id)(U ) + (ι ⊗ id) U ∈ M (K(HU ⊕ H U ) ⊗ C0(G)). The viewpoint U ∈ B(HU ) ¯⊗L∞(G) enables us to make sense of the following crucial observation, which is contained in [Wor96, Theorem 1.6]. Let ω ∈ B(HU )∗ be a normal functional. Then (1.9) (ω ⊗ id)U ∈ D(S G) and S G((ω ⊗ id)U ) = (ω ⊗ id)(U ∗). The representations of G can always be realized by means of a ∗-homomorphism from Cu We note the following result: 0 (bG). Theorem 1.2 ( [Kus01, Proposition 5.2]). Let U ∈ M (K(HU ) ⊗ C0(G)) be a representation. Then there exists a unique φU ∈ Mor(Cu 0 (bG), K(HU )) such that (φU ⊗ id)WG = U. Conversely, given a C∗-algebra B, a representation in a Hilbert space ρ ∈ Mor(B, K(H)) and morphism φ ∈ Mor(Cu representation of G. 0 (bG)), B), the unitary Uφ,ρ = (ρ ◦ φ ⊗ id)WG ∈ M (K(H) ⊗ C0(G)) is a 1.4. Homomorphisms and subgroups. An in-depth description of homomorphisms between quantum groups was given in [MRW12], let us recall the main points. Fix two locally compact quantum groups G and H. A homomorphism of quantum groups H → G can be equivalently described by three objects: • Hopf ∗-homomorphisms: ϕ ∈ Mor(Cu 0 (G), Cu (ϕ ⊗ ϕ)◦∆u 0 (H)), which intertwine the coproducts: G = ∆u H ◦ ϕ • Bicharacters: unitary elements V ∈ M (C0(bG) ⊗ C0(H)), which are (anti)representations on both legs: (1.10) (∆bG ⊗ id)V = V23V13 and (id ⊗∆H)V = V12V13 • Right quantum group homomorphisms: morphisms ρ ∈ Mor(C0(G), C0(G) ⊗ C0(H)) sat- Moreover, they satisfy (bRG ⊗ RH)V = V and (τ isfying bG t ⊗ τ H t )V = V . (1.11) (∆G ⊗ id) ◦ ρ = (id ⊗ρ) ◦ ∆G and (id ⊗∆H) ◦ ρ = (ρ ⊗ id) ◦ ρ Moreover, they satisfy (id ⊗ρ)WG = WG 12V13, where V is the corresponding bicharacter. as follows. If ϕ ∈ Mor(Cu Furthermore, each homomorphism H → G has its dual homomorphism bG → bH. It can be described bϕ ∈ Mor(Cu 0 (bG)), these maps are linked via 0 (H)) is a Hopf ∗-homomorphism, then there exists a unique 0 (bH), Cu 0 (G), Cu (1.12) (id ⊗ϕ)VVH = (bϕ ⊗ id)VVG. 6 PAWE L J ´OZIAK, PAWE L KASPRZAK, AND PIOTR M. SO LTAN Let us also stress that bicharacters and right quantum group homomorphisms are equally well Equivalently, if V ∈ M (C0(bG) ⊗ C0(H)) is a bicharacter representing a homomorphism H → G, then bV = σ(V ∗) ∈ M (C0(H) ⊗ C0(bG)) is a bicharacter representing its dual homomorphism bG → bH. studied in the von Neumann algebraic context, so that a unitary V ∈ L∞(bG) ¯⊗L∞(H) satisfying (1.10) and a normal ∗-homomorphism ρ: L∞(G) → L∞(G) ¯⊗L∞(H) satisfying (1.11) also describe a homomorphism of quantum groups H → G. Right quantum group homomorphisms in the von Neumann algebraic context are in fact normal extensions of the respective maps in the C∗-algebraic context: they are implemented by V by the formula ρ(x) = V (x ⊗ 1)V ∗. Let us note that second condition in (1.11) corresponds to ρ being a right action of H on L∞(G). These are analogues of the natural actions by right shifts. A thorough treatment of the notion of subgroup was given in [DKSS12], we recall some of the main points of that article. Let H → G be a homomorphism of quantum groups (described by a Hopf ∗-homomorphism ϕ ∈ Mor(Cu right quantum group homomorphism ρ ∈ Mor(C0(G), C0(G) ⊗ C0(H))). We say that H is a closed 0 (G)), a bicharacter V ∈ M (C0(bG) ⊗ C0(H)) and a quantum subgroup1 of G if there exists a normal injective ∗-homomorphism γ: L∞(bH) → L∞(bG) such that V = (γ ⊗ id)WH. This map γ is nothing but the incarnation on the von Neumann algebra level of the reduced version of π, namely 0 (H), Cu (1.13) and hence, by (1.12), in particular we have ΛG ◦bϕ = γ ◦ΛH (1.14) (id ⊗ϕ)W G = (γ ⊗ id)WH Let M ⊆ L∞(G) be a von Neumann subalgebra. Recall from [TT72] that M is called invariant if ∆(M) ⊆ M ¯⊗M. We moreover say that M is a Baaj-Vaes subalgebra if it is invariant, R(M) = M and τt(M) = M for all t ∈ R. The Baaj-Vaes theorem [BV05, Proposition A.5] states that Baaj- Vaes subalgebras of L∞(G) are in one to one correspondence with closed quantum subgroups of 0 (bG), Cu bG. This means, in particular, that M can be endowed with Haar weights, and that the inclusion 0 (bH)), the bicharacter and the right quantum group homomorphism. Moreover, ι ι: M ֒→ L∞(G) induces the full data of a quantum group homomorphism: the morphism π ∈ Mor(Cu is the von Neumann incarnation of the reduced version of π. Let us comment on the assumptions of Baaj-Vaes theorem. It turns out that in the case G is compact, discrete, classical and dual to classical the conditions R(M) = M and τt(M) = M for all t ∈ R actually follow from some more general principles. In case G is compact, this follows from restriction of Haar state to M, in case G is discrete this is [NY14, Theorem 3.1], after applying co-duality techniques of [KS14b] (an elementary proof is also available, see [J´oz16, Theorem 1.35]). The case of classical groups and dual to classical groups was covered in [TT72] by Takesaki and Tatsuuma. It turns out that the von Neumann algebras constructed in Section 2 are automatically invariant, and once made τt-invariant for all t ∈ R, they are also R-invariant. This covers an abundance of invariant subalgebras, especially in the Kac case, and it is not known to the authors whether there exists an invariant von Neumann subalgebra that is not a Baaj-Vaes subalgebra. 2. Construction of Hopf image 2.1. First steps towards the construction. The goal of this part is to construct a quantum group H which will later be shown to satisfy the defining properties of Hopf image. So let us fix a morphism β ∈ Mor(Cu 0 (G), B), where B is some C∗-algebra, as in the introduction. Application of Λ bG ⊗ id ⊗ id to both sides of (1.3) yields: (2.1) W23W12W∗ 23 = W12W13. 1or sometimes we call them Vaes-closed quantum subgroups, as there is a competing definition of Woronowicz- closed quantum subgroup and they agree in case G is compact, discrete, classical or dual to classical, see [DKSS12, KKS17] HOPF IMAGES IN LOCALLY COMPACT QUANTUM GROUPS 7 Let us denote X = (id ⊗β)W ∈ M (C0(bG) ⊗ B). Computing the value of (id ⊗ id ⊗β) at both sides of the equality (2.1) results in: (2.2) or, equivalently, (2.3) X23W12X ∗ 23 = W12X13 (∆bG ⊗ id)X = X23X13 Applying (ω ⊗ id ⊗ id) for ω ∈ L1(bG) to both sides of (2.2) we obtain: X(a ⊗ 1)X ∗ = (ω ⊗ id ⊗ id)(W12X13), (2.4) where a = (ω ⊗ id)W. As (2.5) W12 ∈ M (C0(bG) ⊗ C0(G) ⊗ B) ⊆ M (K(L2(G)) ⊗ C0(G) ⊗ B) X13 ∈ M (C0(bG) ⊗ C0(G) ⊗ B) ⊆ M (K(L2(G)) ⊗ C0(G) ⊗ B) we get that X(a ⊗ 1)X ∗ ∈ M (C0(G) ⊗ B) and hence the map θ: C0(G) → M (C0(G) ⊗ B) can be defined by (2.6) C0(G) ∋ a θ7−−→ X(a ⊗ 1)X ∗ ∈ M (C0(G) ⊗ B). Let us assume that B is (faithfully, nondegenerately) represented on a Hilbert space H: B ⊆ B(H). Then we can view θ as a representation θ ∈ Rep(C0(G), L2(G) ⊗ H). One has: (2.7) (id ⊗ θ)(W) = W12X13 ∈ M (C0(bG) ⊗ C0(G) ⊗ B) ⊆ M (C0(bG) ⊗ K(L2(G) ⊗ H)). As W ∈ M (C0(bG) ⊗ C0(G)) generates C0(G) in the sense of [Wor95], we conclude that: Proposition 2.1. θ ∈ Mor(C0(G), C0(G) ⊗ B). Now we are in position to state the main construction. Let M = {(id ⊗ω)X ω ∈ B∗} ⊆ L∞(bG) As X ∈ M (C0(bG) ⊗ B) one sees that M ⊆ M (C0(bG)) ⊆ L∞(bG). Denote by M1 the ∗-algebra generated by M and by M1 its norm-closure. Proposition 2.2. M′ 1 = M ′ 1 is a von Neumann algebra Proof. Indeed, as we have T ∈ M′ 1 ⇐⇒ X(T ⊗ 1) = (T ⊗ 1)X ⇐⇒ (T ⊗ 1)X ∗ = X ∗(T ⊗ 1) ⇐⇒ ⇐⇒ X(T ∗ ⊗ 1) = (T ∗ ⊗ 1)X ⇐⇒ T ∗ ∈ M′ 1 Thus also M1 = M′′ = M -- weak∗ 1 is a von Neumann algebra. Let MBV be the smallest Baaj-Vaes subalgebra containing M1 (so in particular containing M1). The existence of such a von Neumann algebra follows from standard argument: it is the intersection (cid:3) itself is such an algebra. Later on we will see that it can be constructed more explicitly. Thanks of all Baaj-Vaes subalgebras of L∞(bG) containing M1: this collection is non-empty because L∞(bG) to Baaj-Vaes theorem, there exists H ⊂ G such that L∞(bH) = MBV , in particular, we have a map MBV ⊆ L∞(bG) via (1.14). 0 (H)) coming from [DKSS12, Theorem 3.5], which is linked to the embedding π ∈ Mor(Cu 0 (G), Cu 8 PAWE L J ´OZIAK, PAWE L KASPRZAK, AND PIOTR M. SO LTAN 2.2. Properties of the algebra M1. Lemma 2.3. Let β ∈ Mor(Cu a Hilbert space H: B ⊆ B(H). Let C = β(Cu 0 (G)) ⊆ M (B). Denote: 0 (G), B) and assume B is faithfully, nondegenerately represented on • M1 = {(id ⊗ω)(X) ω ∈ B∗}′′ • M2 = {(id ⊗ω)(X) ω ∈ C∗}′′ • M3 =(cid:8)(id ⊗ω)(X) ω ∈ B(H)∗(cid:9)′′ • M4 = {(id ⊗ω)(X) ω ∈ B(H)∗}′′ • M5 = {(id ⊗ωξ,η)(X) ξ, η ∈ H}′′ Then • β ∈ Mor(Cu • M1 = M2 = M3 = M4 = M5 0 (G), C) Remark 2.4. By first part of Lemma 2.3 we see that we can restrict our attention to maps β: Cu 0 (G) → B that are surjective (philosophically speaking, the maps that are Gelfand duals of embeddings bβ: X ֒→ G of a quantum space X as a closed quantum subset of G, where B = C0(X)). Proof. The first statement follows from standard reasoning, so we omit it. It is obvious that M5 ⊆ M4 ⊆ M3. To show that M3 ⊆ M4, let us fix an element ω ∈ B(H)∗ and consider x = (id ⊗ω)X. Recall that B(H)∗ ⊆ B(H)∗ is weak∗-dense, so pick a net (ωi)i∈I ⊆ B(H)∗ i∈I−−→ ω in weak∗-topology. Then xi = (id ⊗ωi)X ∈ M4 and by Lemma 1.1 we have such that ωn that xi → x in σ-WOT. Hence x ∈ M3 by σ-WOT closedness of the latter and we are done. Now as every functional in B(H)∗ restricts to B and C we have that M3 ⊆ M1, M2. But as B, C ⊆ B(H) are closed, any continuous functional from B∗ and C∗ extends to a continuous functional in B(H)∗ by Hahn-Banach theorem, so B(H)∗ ։ C∗, B∗. In particular, this means M1, M2 ⊆ M3. For the equality M4 = M5 recall that the linear span of vector functionals ωξ,η is norm dense i∈I−−→ ω in in B(H)∗ (see, e.g. [Bla06, III.2.1.4]). Now by standard calculation we show that if ωi norm, then (id ⊗ωi)X i∈I−−→ (id ⊗ω)X in WOT. Pick then ξ, η ∈ H, we have: hξ(id ⊗(ω − ωn))Xηi = (ωξ,η ⊗ (ω − ωn))X ≤ kωξ,η ⊗ (ω − ωn)kkXk= hξηikω − ωnkkXk→ 0. By WOT-closedness of M5 any element of the generating set of M4 is in fact in M5, so we conclude by von Neumann's bicommutant Theorem. (cid:3) Proposition 2.5. The algebra M1 is invariant, i.e. b∆(M1) ⊆ M1 ¯⊗M1. If moreover τ u ker β (in particular if G is Kac type), then M1 is is preserved by bτt for each individual t ∈ R. that b∆(x) ∈ M5 ¯⊗M5. Pick an orthonormal basis (ej)j∈J of H and recall that 1 =Pj∈J ejihej is Proof. For the invariance, let us first pick x = (id ⊗ωξ,η)X ∈ M5 for some ξ, η ∈ H, we will show a WOT-convergent resolution of identity into rank one projections. We compute: t (ker β) ⊆ b∆(x) = (b∆ ⊗ ωξ,η)(X) = (id ⊗ id ⊗ωξ,η)(X23X13) = ejihej )X13) = (id ⊗ id ⊗ωξ,η)(X23(1 ⊗ 1 ⊗ ejihej )X13) = = (id ⊗ id ⊗ωξ,η)(X23(1 ⊗ 1 ⊗Xj∈J =Xj∈J =Xj∈J =Xj∈J (id ⊗ id ⊗ωξ,ej )(X23)(id ⊗ id ⊗ωej ,ηi)(X13) = (id ⊗ωξ,ej )(X) ⊗ (id ⊗ωej ,η)(X) ∈ M5 ¯⊗M5 We conclude by normality of b∆ and equality M1 = M5 obtained in Lemma 2.3. HOPF IMAGES IN LOCALLY COMPACT QUANTUM GROUPS 9 Let t ∈ R, assume ker(β) is τ u t unique) functional ωt ∈ B∗ such that ω◦β ◦τ u theory: there exists an isometry s making the following diagram commute: invariant and let ω ∈ B∗. Then there exists a (necessarily t = ωt ◦β. This follows from a general Banach space Cu 0 (G) β C q s Cu 0 (G)(cid:14)ker β Now, as τt(ker β) ⊆ ker β, using s we can conclude the existence of a map τ C t = τ C t ◦β. Then β ◦τ u t : C → C such that {(id ⊗ω ◦β)W ω ∈ C∗}′′ ={(id ⊗ω ◦β)(τt ⊗ τ u t )W ω ∈ C∗}′′ = t )W ω ∈ C∗}′′(cid:19) = t ◦β)W ω ∈ C∗}′′(cid:19) = =τt(cid:18){(id ⊗ω ◦β ◦τ u =τt(cid:18){(id ⊗ω ◦τ C =τt(cid:18){(id ⊗ω ◦β)W ω ∈ C∗}′′(cid:19) Because τ C t : C → C is a bijection, also (τ C t )∗: C∗ → C∗ is a bijection. (cid:3) Theorem 2.6. The minimal Baaj-Vaes subalgebra of L∞(bG) containing M1 is given by MBV = ([t∈Rbτt(M1))′′. In particular, MBV = M1 if G is compact or discrete or if τ u t (ker(β)) ⊆ ker(β) for all t ∈ R. Proof. For the purpose of the proof, let us denote by MRHS the von Neumann algebra appearing on [MNW03, Theorem 1.9(2)] we conclude that MRHS is again invariant. To see that MRHS = MBV , {xω,t ω ∈ B(H)∗}. In turn, MRHS is generated by {xω,t ω ∈ B(H)∗, t ∈ R}. Indeed, from the above description we see that all elements xω,t ∈ MRHS. The converse inclusion follows easily from von Neumann's bicommutant Theorem: if y commutes with all xω,t for all ω and t, then the right hand side of Theorem 2.6. This algebra is clearlybτ -invariant, and as M1 is invariant, using we need to show that it is preserved by the unitary antipode bR. First, for t ∈ R and ω ∈ B(H)∗ let us denote xω,t = (bτt ⊗ ω)X and observe that M1 = M4 is generated by xω,0 for all ω ∈ B(H)∗. Furthermore, for t ∈ R fixed, bτt(M1) is generated by y ∈Tt∈Rbτt(M1)′ ⊆ (St∈Rbτt(M1))′. ω∗,0 ∈ D(bS) = D(bτi/2) (recall that in fact X is antirepresentation, so σ(X ∗) is a representation Now we use M1 = M4 from Lemma 2.3. Let us pick ω ∈ B(H)∗. Then ω∗ defined as ω∗(a) = ω(a∗) is again a normal functional. Then by (1.9) we have that (id ⊗ω)(X ∗) = [(id ⊗ω∗)(X)]∗ = x∗ of G), see [Wor96]. Hence bR((id ⊗ω)(X ∗)) = (bτ−i/2 ◦bS)((id ⊗ω)X ∗) =bτ−i/2((id ⊗ω)X) But as MRHS is bτ -invariant, it is also preserved by its analytic generator, as it is defined uniquely: the analytic generator of bτ−t ↾MRHS is precisely (bτ−i/2) ↾MRHS (this is in fact the uniqueness of analytic continuation of a function). Hence bR(x∗ bR(xω,0) ∈ MRHS for all ω ∈ B(H)∗. Next, thanks to the relation bτt ◦bR = bR ◦bτt, it follows that bR(xω,t) = bτt(bR(xω,0)) ∈ bτt(MRHS) = MRHS. Together with the description of the generating set in the first step, this finishes the proof of the main assertion. ω∗,0) ∈ MRHS for all ω ∈ B(H)∗ and in turn (2.8) 10 PAWE L J ´OZIAK, PAWE L KASPRZAK, AND PIOTR M. SO LTAN The last part of the Theorem follows from the observation that in all these cases the algebra M1 is automatically bτ -invariant: in case of G compact or discrete this was discussed at the end t (ker(β)) ⊆ ker(β) for all t ∈ R follows from the discussion in (cid:3) of Section 1.4 and the case τ u Proposition 2.5. 2.3. Verification of defining properties. In this part we show that the quantum subgroup H constructed in Section 2.1 indeed satisfies the defining properties of the Hopf image, i.e. firstly, there exists β ∈ Mor(Cu 0 (H), B) as described in Diagram 1, that is showing that (π, H, β) ∈ Cβ, and secondly, that it is an initial object of the category Cβ. Lemma 2.7. Let K be a closed quantum subgroup of G and denote the associated Hopf ∗-homomor- phism and normal embedding by πK ∈ Mor(Cu 0 (G), Cu Then (πK, K, β) ∈ Cβ, i.e. there exists β ∈ Mor(Cu 0 (K))) and γ: L∞(bK) → L∞(bG), respectively. 0 (K), B) such that β = β ◦ πK if and only if M1 ⊆ γ(L∞(bK)). Proof. Assume that (πK, K, β) ∈ Cβ and pick ω ∈ B∗. Using (1.13) and (1.14), we have that (id ⊗ω)(id ⊗β)(W G) = (id ⊗ω)(id ⊗ β ◦πK)(W G) = γ(cid:18)(id ⊗ω ◦ β)WK(cid:19) Hence the algebra M1 constructed for β and the corresponding algebra constructed for β, seen for β. This shows the necessity. as subalgebras of L∞(bG), coincide, hence so do their C∗-envelopes M1 and the corresponding one Assume that M1 ⊆ γ(L∞(bK)). In order to get a morphism β ∈ Mor(Cu show that σ(X ∗) is a representation of bK by Theorem 1.2. Now observe that [DKSS12, Lemma 1.4] shows that X ∈ M (M1 ⊗ B) ⊆ M (C0(bH) ⊗ B). But as ∆bK and ∆bG ↾γ(L∞( bK)) coincide and from (2.3), we have that X23X13 = (∆bK ⊗ id)X, so σ(X ∗) satisfies hypothesis of Theorem 1.2. (cid:3) 0 (K), B) it is enough to bG Proof of Theorem A. From Lemma 2.7 we get that H constructed at the end of Section 2.1 can be endowed with the morphism β completing the desired factorization, i.e. (π, H, β) ∈ Cβ. Let bG- now k = (πK, K, β′) ∈ Cβ. From Lemma 2.7 we have that M0 ⊆ L∞(bK) and L∞(bK) is a R t - and ∆bG-invariant subalgebra of L∞(bG). As L∞(bH) is chosen to be a minimal von Neumann subalgebra with this property, we necessarily have L∞(bH) ⊆ L∞(bK). In particular the inclusion map satisfies the defining property of H being a closed quantum subgroup of K, so we conclude by [DKSS12, Theorem 3.5] and [KSS16, Lemma 2.5]. (cid:3) ,τ 2.4. More on X ∈ M (C0(bG) ⊗ B) and θ ∈ Mor(C0(G), C0(G) ⊗ B). In this section we inves- tigate the mutual relation between the objects describing the embedding X ֒→ G as phrased in Remark 2.4, i.e. the morphism β ∈ Mor(Cu and the morphism θ ∈ Mor(C0(G), C0(G) ⊗ B) in the spirit of [MRW12]. 0 (G), B), the unitary antirepresentation X ∈ M (C0(bG) ⊗ B) From the discussion in Section 2.1 it is clear that out of β one can canonically construct the morphism β ∈ Mor(Cu unitary X, which is an antirepresentation of bG. But Theorem 1.2 (applied to σ(X ∗) as in the proof of Lemma 2.7) shows that to a unitary X ∈ M (C0(bG) ⊗ B) there corresponds a unique antirepresentation of bG one can uniquely construct the morphism θ ∈ Mor(C0(G), C0(G) ⊗ B). Observe that this morphism satisfies the following condition: (∆ ⊗ id)◦θ = (id ⊗ θ)◦∆. Indeed, for a ∈ C0(G) we have that Again, from the discussion in Section 2.1 it is clear that out of a unitary X, which is an 0 (G), B). (2.9) (∆ ⊗ id)◦θ(a) = (∆ ⊗ id)(X(a ⊗ 1)X ∗) = W12X13(a ⊗ 1 ⊗ 1)X ∗ 13W∗ 12. HOPF IMAGES IN LOCALLY COMPACT QUANTUM GROUPS 11 Now using (2.2), we can continue calculations from (2.9) and get: (2.10) W12X13(a ⊗ 1 ⊗ 1)X ∗ 13W∗ 12 =X23W12X ∗ 23(a ⊗ 1 ⊗ 1)X23W∗ 23 = 12X ∗ 12X ∗ 23 = =X23W12(a ⊗ 1 ⊗ 1)W∗ =X23(cid:18)W(a ⊗ 1)W∗(cid:19)12 X23 = =X23(∆(a))12X ∗ 23 = (id ⊗ θ)◦∆(a) Proposition 2.8. Assume θ ∈ Mor(C0(G), C0(G) ⊗ B) is such that (∆ ⊗ id)◦ θ = (id ⊗ θ)◦ ∆. Then there is a unique unitary X ∈ M (C0(bG) ⊗ B) such that θ(a) = X(a ⊗ 1)X ∗ and X is an antirepresentation of bG. Proof. The proof is essentially the same as first part of the proof of [MRW12, Theorem 5.3], but we repeat it for later use. Denote X = W∗ 12 ((id ⊗ θ)(W)) ∈ M (C0(bG) ⊗ C0(G) ⊗ B). We will show that W23 X124W∗ X134, then using [MRW12, Theorem 2.6] we conclude that X ∈ M (C0(bG) ⊗ C1 ⊗ B), so in fact there exists X ∈ M (C0(bG) ⊗ B) with X = X13. We compute 23 = W23 X124W∗ 23 = W23W∗ 23 = W∗ 12W∗ 13W∗ 23W23(cid:18)(id ⊗ θ)W(cid:19)124 12(id ⊗∆ ⊗ id)(cid:18)(id ⊗ θ)W(cid:19) = 12(cid:18)(id ⊗ id ⊗ θ)(id ⊗∆)W(cid:19) = 12(cid:18)(id ⊗ id ⊗ θ)W12W23(cid:19) = 13(cid:18)(id ⊗ id ⊗ θ)W23(cid:19) = X134 13W∗ 13W∗ = W∗ = W∗ = W∗ = W∗ We now check that θ(a) = X(a⊗1)X ∗. This is equivalent to showing that θ(a)13 = X(a⊗1⊗1) X ∗. We compute: X(a ⊗ 1 ⊗ 1) X ∗ = W∗ 12(cid:18)(id ⊗ θ)W(cid:19)(a ⊗ 1 ⊗ 1)(cid:18)(id ⊗ θ)W∗(cid:19)W12 = 12(cid:18)(id ⊗ θ)(W(a ⊗ 1)W∗)(cid:19)W12 = 12(cid:18)(id ⊗ θ)∆(a)(cid:19)W12 = 12(cid:18)(∆ ⊗ id)θ(a)(cid:19)W12 = = W∗ = W∗ = W∗ = W∗ 12(W12θ(a)13W∗ 12)W12 = θ(a)13 Showing that X is antirepresentation amounts to showing that (b∆ ⊗ id ⊗ id) X = X234 X134. Using (1.4) and the definition of X we get that: (b∆ ⊗ id ⊗ id) X =(cid:18)((b∆ ⊗ id)W∗) ⊗ 1(cid:19)(b∆ ⊗ θ)W) X234(b∆ ⊗ θ)W) = X234 X134 =W∗ =W∗ 13 13W∗ 23((id ⊗ id ⊗ θ)W23)((id ⊗ id ⊗ θ)W13) = where in the last equality we used the fact that X234 = X24 commutes with W∗ 13. To prove uniqueness, assume Y ∈ M (C0(G) ⊗ B) is another such unitary. Because slices of W are dense in 12 PAWE L J ´OZIAK, PAWE L KASPRZAK, AND PIOTR M. SO LTAN C0(G), we have that X(a ⊗ 1)X ∗ = Y (a ⊗ 1)Y ∗ for all a ∈ C0(G) is equivalent to saying that X23W12X ∗ 23 = Y23W12Y ∗ 23 Rearranging terms, this is equivalent to W∗ 12(Y ∗ 23X23)W12 = Y ∗ 23X23 which, in turn, is equivalent to (b∆ ⊗ id)(Y ∗X) = Y ∗ Applying (b∆ ⊗ id) to both sides of this equality we get X23X13(1 ⊗ 1 ⊗ u) = (b∆ ⊗ id)(X(u ⊗ 1)) = (b∆ ⊗ id)Y 13X13 = Y23Y13 = X23(1 ⊗ 1 ⊗ u)X13(1 ⊗ 1 ⊗ u) and hence u = 1, which finishes the proof. (cid:3) Summarizing, there are three equivalent ways of studying an embedding X ֒→ G of a locally compact quantum space into a locally compact quantum group (we recall that B = C0(X)), these are as follows: (1) the morphism β ∈ Mor(Cu 0 (G), B); (2) the unitary X ∈ M (C0(bG) ⊗ B), which is an antirepresentation of bG and (3) the morphism θ ∈ Mor(C0(G), C0(G) ⊗ B) satisfying (∆ ⊗ id)◦θ = (id ⊗ θ)◦∆, which corresponds to the partial action X y G by right shifts. Fixing a non-degenerate representation of B ⊆ B(H) and denoting by B = B′′ the WOT-closure of B in the WOT-topology induced by this embedding, we can (similarly as in the case of ho- momorphisms), study the above objects in the von Neumann algebraic context. Indeed, we have X ∈ L∞(bG) ¯⊗B ⊆ L∞(bG) ¯⊗B(H) and θ: L∞(G) → L∞(G) ¯⊗B satisfying (∆ ⊗ id)◦θ = (id ⊗ θ)◦∆, because θ is obtained by conjugating with a unitary and as such extends to the WOT-closure of C0(G). We will switch between these viewpoints freely later on. The passage from von Neumann level to C∗-level is pretty much the same as in Section 1.3. 3. The case of two subgroups The goal of this section is to discuss the notion of compact quantum group generated by two closed quantum subgroups in the sense of [BCV17], as well as the notion of joint fullness of a family of (corestriction functors induced by) Hopf quotients in the sense of [Chi15], in the context of Hopf image and to extend it to the non-compact case. Let then G be a locally compact quantum group and let H1, H2 be its two closed subgroups (for i = 1, 2, denote by πi: Cu 0 (Hi) the corresponding Hopf surjection, by γi: L∞(bHi) → L∞(bG) the corresponding inclusions and by V Hi ∈ L∞(bG) ¯⊗L∞(Hi) the corresponding bicharac- ters). Consider the two ideals: C0(G \ (H1 ∪ H2)): = ker(π1) ∩ ker(π2) and C0(G \ (H1 · H2)): = ker((π1 ⊗ π2)◦∆u G) and the two quotients 0 (G) → Cu We conclude from [MRW12, Corollary 2.9] that in fact X(1 ⊗ u) = Y for some unitary u ∈ M (B). (3.1) and (3.2) q∪: Cu 0 (G) → C0(H1 ∪ H2) = Cu q•: Cu 0 (G) → C0(H1 · H2) = Cu 0 (G).C0(G \ (H1 ∪ H2)) 0 (G).C0(G \ (H1 · H2)) Proposition 3.1. The following von Neumann subalgebras of L∞(bG) are equal: HOPF IMAGES IN LOCALLY COMPACT QUANTUM GROUPS 13 • M1,2, the smallest von Neumann algebra containing both γ1(L∞(bH1)) and γ2(L∞(bH2)); • M• = {(id ⊗ω ◦q•)W: ω ∈ C0(H1 · H2)∗}′′; • M∪ = {(id ⊗ω ◦q∪)W: ω ∈ C0(H1 ∪ H2)∗}′′. • MV,1,2 = {(id ⊗ω1 ⊗ ω2)V H1 12 V H2 13 : ωi ∈ L1(Hi)}′′ Proof. M1,2 = M∪. Observe that ker(q∪) = ker(π1 ⊕ π2), hence arguing as in the proof of the first part of Lemma 2.3 we may replace q∪ with π1 ⊕ π2 in the definition of M∪ (and the functionals are 0 (H2)∗, hence 0 (H1)∗ ⊕ Cu on a different C∗-algebra then). Recall that (Cu 0 (Hi)∗. Hence every ω appearing in the definition of M∪ can be written as ω = ω1 ⊕ ω2 for ωi ∈ Cu 0 (H2))∗ = Cu 0 (H1) ⊕ Cu M∪ =(cid:18)γ1(L∞(bH1)) + γ2(L∞(bH2))(cid:19)′′ = M1,2 as desired. M1,2 = M•. Recall that the linear span of the functionals of the form ω1 ⊗ ω2 on A⊗ B is weak∗- dense in (A⊗B)∗ for any C∗-algebras A, B. Further, as (id ⊗q•)W = ((id ⊗π1)W)12((id ⊗π2)W)13, to compute M• it is enough to understand the von Neumann algebra generated by operators of the form ((id ⊗ω1◦π1)W)((id ⊗ω2◦π2)W) thanks to Lemma 1.1. But it is then clear that as desired. MV,1,2 = M1,2. Similarly as in the previous step, it is immediate to see that = M1,2 M∪ =(cid:18)γ1(L∞(bH1)) · γ2(L∞(bH2))(cid:19)′′ MV,1,2 =(cid:18)γ1(L∞(bH1)) · γ2(L∞(bH2))(cid:19)′′ = M1,2 (cid:3) Definition 3.2. We call the Hopf image of either of the maps q• and q∪ the closed quantum subgroup generated by H1 and H2 and denote it by hH1, H2i. More generally, if Hi ⊂ G is a family of subgroups (i ∈ I), then we denote by hSi∈I Hii the subgroup generated by this family: the subgroup corresponding to the Baaj-Vaes subalgebra (Si∈I γi(L∞(bHi)))′′. 4. Separation of homomorphisms G → K be described by a Hopf ∗-homomorphism ϕ ∈ Mor(Cu Let G, K be locally compact quantum groups and let a homomorphism of quantum groups 0 (G)) and a bicharacter V ∈ 0 (G), B), the corresponding unitary X ∈ M (C0(bK) ⊗ C0(G)). Let B be a C∗-algebra, β ∈ Mor(Cu M (C0(bG) ⊗ B) is given by X = (id ⊗β)W G. Let us describe in detail the unitary corresponding to β ◦ϕ. It is given by Y = (id ⊗β ◦ϕ)W K. 0 (K), Cu Lemma 4.1. The unitaries X, Y, V obey the following equation: Y13 = V ∗ 12X23V12X ∗ 23 ∈ M (C0(bK) ⊗ K(L2(G)) ⊗ B) 0 (bG), Cu fixed homomorphism G → K. Application of Λ bK ⊗ ΛG to both sides of (1.12) yields 0 (bK)) be the Hopf ∗-homomorphism linked to ϕ by (1.12), that is, Proof. Let bϕ ∈ Mor(Cu this Hopf ∗-homomorphism describes the homomorphism of quantum groups bK → bG dual to the V = (id ⊗ΛG◦ϕ)WK = (Λ bK ◦bϕ ⊗ id)WG Application of Λ bK ◦bϕ ⊗ id ⊗β to both sides of (1.3) gives: (LHS) =(Λ bK◦bϕ ⊗ id ⊗β)VVG =(id ⊗ id ⊗β)(cid:18)Λ bK ⊗ id ⊗ϕ)VVG 13(cid:19) = 13 = (id ⊗ id ⊗β)(cid:18)Λ bK ◦bϕ ⊗ id ⊗ id)VVG 13(cid:19) =(cid:18)(id ⊗β ◦ϕ)WG(cid:19)13 = Y13 PAWE L J ´OZIAK, PAWE L KASPRZAK, AND PIOTR M. SO LTAN 14 and (RHS) =(Λ bK ◦bϕ ⊗ id ⊗β)(cid:18)(WG =(cid:18)(Λ bK ◦bϕ ⊗ id)WG(cid:19)∗ =(cid:18)(id ⊗ΛG ◦ϕ)(WK)∗(cid:19)12 12)∗W G 23 WG 12(WG 23)∗(cid:19) = 12(cid:18)(id ⊗β)W G(cid:19)23(cid:18)(Λ bK ◦bϕ ⊗ id)WG(cid:19)12(cid:18)(id ⊗β)WG(cid:19)∗ X23(cid:18)(id ⊗ΛG◦ϕ)(W K)(cid:19)12 12X23V12X ∗ 23 23 = V ∗ X ∗ 23 = (cid:3) Recall the vague interpretation B = C0(X) and β: X ֒→ G. With the notation as above, by the restriction of a homomorphism G → K to the subset X ⊂ G we mean the map β ◦ ϕ ∈ Mor(Cu 0 (K), B). Theorem 4.2 (Theorem B). Assume β is generating. Consider two quantum group homomor- phisms G → K described by ϕ, ϕ ∈ Mor(Cu 0 (G)). If their restrictions to X coincide, then the homomorphisms coincide on the whole of G: ϕ = ϕ. 0 (K), Cu bicharacters and right quantum group homomorphisms corresponding to ϕ, ϕ, respectively. Proof. Let us denote by V, V ∈ M (C0(bK) ⊗ C0(G)) and by ρ, ρ ∈ Mor(C0(K), C0(G) ⊗ C0(K)) the Now, if β◦ϕ = β◦ ϕ, then the corresponding unitaries coincide: Y = Y ∈ M (C0(bK) ⊗ B). Using Lemma 4.1 one may rewrite this as 12X23 V12X ∗ V ∗ 12X23V12X ∗ 23 23 = V ∗ or equivalently 12X23 V12 = V ∗ V ∗ 12X23V12. By slicing with ω ∈ B∗ on the third leg we see that the right quantum group homomorphisms ρ and ρ agree on the M1 (they are normal ∗-homomorphisms). But applying (τ t ⊗ id) and t ⊗ τ G)(V ) = V (and likewise for V ) we see that ρ and ρ have the same remembering that (τ values on bτt(M1) and once again, by normality of ρ, ρ, they coincide on (St∈Rbτt(M1))′′ = MBV (by Theorem 2.6). By assumption, MBV = L∞(bG), hence ρ = ρ. This ends the proof, as ρ and ρ determine the homomorphisms G → K uniquely (see [MRW12, Theorem 5.3]). bK t ⊗ τ G (cid:3) bK Theorem 4.2 says that a homomorphism between quantum groups is uniquely determined by its values on a quantum generating set. The rest of Section 4 is devoted to reversing the implication in case G is discrete: if a subset has the property that homomorphisms between quantum groups are uniquely determined by their values on this set, then the set in question is generating. In other words, we will manufacture two different homomorphisms that agree on a given subset, provided that it is not generating. To this end, let β ∈ Mor(c0(G), B) be a morphism with Hopf image (π, H, β) such that H ( G. Denote by K = G ∗H G. Using [Wan95, Theorem 3.4 & Corollary 3.5], one can describe K as follows. Consider first the free product K′ = G∗G, it is given by the C∗-algebra Cu(cK′) = Cu(bG)∗Cu(bG) (amalgamated over C1). Denote by i1, i2 the maps Cu(bG) → Cu(bG) ∗ Cu(bG) = Cu(cK′) putting the copy of Cu(bG) in the first and second spot, respectively, these maps are Hopf morphisms. Denote by bπ: Cu(bH) → Cu(bG) the homomorphism dual to π. Lemma 4.3. bπ is not surjective. Proof. Assume it is. Using (1.13) we then have that γ: L∞(bH) → L∞(bG) is surjective, which Denote by I ⊆ Cu(cK′) the closed ideal generated by {i1◦bπ(x) − i2 ◦bπ(x) : x ∈ Cu(bH)}. Lemma 4.4. There exists y ∈ Cu(G) such that i1(y) − i2(y) /∈ I. contradicts our assumption H 6= G. (cid:3) HOPF IMAGES IN LOCALLY COMPACT QUANTUM GROUPS 15 Proof. Consider the smallest C∗-subalgebra of Cu(cK′) generated by i1(Cu(bG)) and i2(bπ(Cu(bH)) inside Cu(cK′): let it be called A. It follows from the concrete description of the free product (see, e.g. [BO08, §4.7]) that A 6= Cu(cK′) (this of course needs Lemma 4.3). Let then ω ∈ Cu(cK′)∗ be a non-zero functional such that ω ↾A= 0. Then in particular ω ↾I= 0. Let y ∈ Cu(bG) \bπ(Cu(bH)). Then ω(i1(y) − i2(y)) = −ω(i2(y)) It is now clear that for a given y as above one could have chosen ω such that ω(i2(y)) 6= 0 and ω ↾A= 0. Indeed, first pick non-zero ω ∈ Cu(bG) such that ω ↾bπ(C u( bH))= 0 and ω(y) 6= 0. Then ω = ω ◦i2 ∈ Cu(cK′)∗ is the required functional. As the final step of constructing K = G ∗H G, one considers q: C(cK′) → C(bK) = C(cK′) /I . Consider the morphisms ϕj = dq◦ij: c0(K) → c0(G) for j = 1, 2. Then one has Proposition 4.5. The Hopf ∗-homomorphisms ϕj satisfy β ◦ϕ1 = β ◦ϕ2 and are distinct. (cid:3) Proof. Observe that it is enough to show that ϕj coincide on H, i.e. (4.1) because β = β ◦π. The equality (4.1) is equivalent to the equality π◦ϕ1 = π◦ϕ2 (4.2) [π◦ϕ1 = [π◦ϕ2 (4.3) But composition of morphisms satisfies dϕ◦φ = bφ◦bϕ, hence (4.2) is equivalent to which holds in the quotient C(bK) = C(cK′) /I , as desired. Now using y ∈ Cu(bG) from Lemma 4.4 we can see that ϕ1 6= ϕ2. Indeed, q◦i1◦bπ = cϕ1 ◦bπ = cϕ2 ◦bπ = q◦i2◦bπ bϕ1(y) − bϕ2(y) = q((i1 − i2)(y)) 6= 0 from the definition of y and hence ϕ1 6= ϕ2. (cid:3) Let us summarize these ideas in the following Theorem 4.6. Let G be a discrete quantum group and let β ∈ Mor(c0(G), B) be a morphism. Then β is generating if and only if for any quantum group K and any pair of homomorphisms G → K described by ϕ, ϕ ∈ Mor(c0(K), c0(G)) one has β ◦ϕ = β ◦ ϕ ⇐⇒ ϕ = ϕ . 5. β-restriction and generating morphisms Let A be a C∗-algebra and let U ∈ M (A ⊗ C0(G)) be a representation of G on A. Reasoning similarly as in the proof of Proposition 2.8 (with certain W's replaced by U ), one can prove the following proposition: Proposition 5.1. Let θ ∈ Mor(C0(G), C0(G) ⊗ B) be a morphism satisfying (∆G ⊗ id)◦θ = (id ⊗ θ)◦∆G. There exists a unique unitary element Y ∈ M (A ⊗ B) such that (id ⊗ θ)U = U12Y13. Definition 5.2. Let β ∈ Mor(Cu 0 (G), B) be a morphism, θ ∈ Mor(C0(G), C0(G) ⊗ B) be the morphism assigned to β via (2.6), let U ∈ M (A ⊗ C0(G)) be a representation of G in A. Then Y ∈ M (A ⊗ B) obtained by Proposition 5.1 will be denoted U β and called the β-restriction of U . Remark 5.3. Observe that (2.7) may be interpreted as Wβ = X = (id ⊗β)W. Theorem 5.4 (Theorem C). With the notation as above (i) The morphism β is generating. (ii) {(id ⊗β∗ω)W ω ∈ B∗}′′ = L∞(bG). (iii) {x ∈ L∞(G) θ(x) = x ⊗ 1} = C1. 16 PAWE L J ´OZIAK, PAWE L KASPRZAK, AND PIOTR M. SO LTAN (iv) β-restriction is an injective assignment: β-restrictions of any two distinct representations of G are distinct. We have (i) =⇒ (ii) ⇐⇒ (iii) ⇐⇒ (iv). Moreover, (i) =⇒ (ii) provided that G is compact or discrete or τ u t (ker(β)) ⊆ ker(β) for all t ∈ R (in particular, if G is of Kac type). Proof. (ii) =⇒ (i). This is obvious in view of Theorem 2.6. (ii) =⇒ (iii). This follows by applying co-duality to both sides of the equation in (ii), see [KS14b, Section 3]. An elementary proof is given in [J´oz16, Theorem 3.15]. (iii) =⇒ (iv). Let U, V ∈ M (K(H) ⊗ C0(G)) be two representations of G in the same Hilbert space H. Assume that U β = V β ∈ M (K(H) ⊗ B). We have: 13)∗V ∗ (id ⊗ θ)(U V ∗) = U12U β 13(V β 12 = U12V ∗ 12. Thus condition (ii) ensures us that there exists a unitary element u ∈ B(H) such that U = (u⊗1)V . Applying (id ⊗∆) to both sides of this equality we get that u = 1 as in the last step of the proof of Proposition 2.8. ′ ( M′ (iv) =⇒ (ii). Assume that (ii) does not hold, i.e. we have that M1 ( L∞(bG). Then we have L∞(bG) (it exists, as von Neumann algebras are spanned by their unitary elements). Consider U = (u ⊗ 1)W(u∗ ⊗ 1) (it is obvious that U ∈ Rep(G)). From the definition of u it is clear that U 6= W. But on the other hand we have that 1 ⊂ B(L2(G)), so pick a unitary u ∈ M′ 1 \ L∞(bG) ′ U β 13 =U ∗ 12((id ⊗ θ)U ) = 12(cid:18)(id ⊗ θ)(cid:18)(u ⊗ 1)W(u∗ ⊗ 1)(cid:19)(cid:19) = =U ∗ =(u ⊗ 1 ⊗ 1)W∗ 12(u∗ ⊗ 1 ⊗ 1)(u ⊗ 1 ⊗ 1)(cid:18)(id ⊗ θ)W(cid:19)(u∗ ⊗ 1 ⊗ 1) = =(u ⊗ 1 ⊗ 1)Wβ =(u ⊗ 1 ⊗ 1)X13(u∗ ⊗ 1 ⊗ 1) = X = Wβ 13 13(u∗ ⊗ 1 ⊗ 1) = Where the equalities in the last line follow from the fact that Wβ = X (cf. Remark 5.3) and the fact that X ∈ M1 ¯⊗B(H) (for a fixed non-degenerate representation π ∈ Mor(B, K(H))) and hence the first leg of X commutes with u. (i) =⇒ (ii) under additional assumptions. This was discussed in Proposition 2.5 and Theorem 2.6 (cid:3) and boils down to M1 being automatically τ -invariant. Remark 5.5. In fact Theorem 4.2 can be deduced from Theorem 5.4 in the case of compact, discrete and Kac type quantum groups: it relies on the implication (iv) =⇒ (iii) for representations coming from bicharacters describing homomorphisms. Its converse, Theorem 4.6, valid for discrete quantum groups, for which (iv) =⇒ (iii) holds automatically, is stronger than just this implication: one can detect injectivity of the β-restriction map not only in the class of all representations, but also in the class of bicharacters coming from homomorphisms. 6. Promotion of intertwiners Let U ∈ M (K(HU ) ⊗ C0(G)) ⊆ B(HU ) ¯⊗L∞(G) be a representation of G. Let us call B the C∗-subalgebra of B(HU ) generated by slices of U , then U ∈ M (B ⊗ C0(G)) ⊆ B ¯⊗L∞(G), where the bicommutant B = B′′ is taken inside B(HU ). Let ϕ ∈ Mor(Cu such that U = (ϕ ⊗ id)W (given by Theorem 1.2). 0 (bG), B) be the unique morphism We would like to interpret X = σ(U )∗ ∈ M (C0(G) ⊗ B) (which is now an antirepresentation) as a quantum subset X ⊂ bG. Let then θ: L∞(bG) → L∞(bG) ¯⊗B be the map given by (2.6). Also, for any representation U ∈ M (A ⊗ C0(bG)) of bG there exists a unique U ϕ ∈ M (A ⊗ B) -- the restriction of the family of unitaries U to the quantum subset X ⊂ bG (given by Proposition 5.1). The precise formula is (6.1) U ϕ 13 = U ∗ 12(id ⊗ θ)( U ). HOPF IMAGES IN LOCALLY COMPACT QUANTUM GROUPS 17 Let us use the following notation: whenever π ∈ Mor(Cu 0 (H)) is a Hopf ∗-homomor- phism, we denote by πr = ΛH◦π ∈ Mor(Cu 0 (G), C0(H)) the reduction of this morphism. We begin with computing the restriction of U ∈ M (B ⊗ C0(G)) to the subset H ⊂ G, where H is a closed quantum subgroup (via π: Cu 0 (G), Cu 0 (H), normal embedding γ: L∞(bH) → L∞(bG) and with the aid of the bicharacter V ∈ M (C0(bH) ⊗ C0(G))). 0 (G) → Cu U πr 13 = U ∗ 12V23U ∗ 12V ∗ 23 (6.2) = (ϕ ⊗ id ⊗Λ◦π)(W∗ 12 = ((ϕ ⊗ Λ◦π)VV)13 = ((ϕ◦ π ⊗ id)WH)13 W23W12W∗ 23) In the first equality we used (6.1) and (1.3) in the third equality. In fact the above computation works also for Woronowicz-closed quantum subgroups. For later use, let us compute (θ ⊗ id)V . (θ ⊗ id)V = X12V13X ∗ 12 = (σ ⊗ id)(U ∗ 12V23U12) (6.3) = (σ ⊗ id)((ϕ ⊗ id ⊗Λ◦π)W∗ 12 = (σ ⊗ id)((ϕ ⊗ id ⊗Λ◦π)(VV13W23) = ((ϕ ⊗ Λ◦π)VV)23V13 = U πr 23 V13 W23W12) In particular if H = G and V = W, then (6.3) simply says that (θ ⊗ id)W = U23W13 (cf. (2.7)). Let us denote by θ the map given by (2.6), associated to the U πr , a representation of H. Lemma 6.1. The maps (γ ⊗ id)◦ θ and θ◦γ coincide on L∞(bH). Proof. Using (6.3) we have (6.4) ((θ◦γ) ⊗ id)(WH) = (θ ⊗ id)V = U πr 23 V13 On the other hand using (6.2) we obtain: (σ ⊗ id)◦(θ ⊗ id)(WH) = (U πr )∗ 12WH 23U πr 12 (6.5) = (ϕ◦ π ⊗ id ⊗ id)((WH = (ϕ◦ π ⊗ id ⊗ id)((WH WH 12)∗WH 23 13)WH 12) 23) = U πr 13 WH 23, where in the third equality we used (1.3) (after reducing the third leg) . Application of (σ ⊗ id) to both sides of (6.5) yields: (6.6) (θ ⊗ id)(WH) = U πr 23 WH 13, Applying (γ ⊗ id) to both sides of (6.6) and comparing it to (6.4) we obtain: (6.7) ((θ◦γ) ⊗ id)(WH) = (γ ⊗ id)◦(θ ⊗ id)(WH) and the conclusion follows from WOT-density of slices of WH in L∞(bH). Theorem 6.2. Let H1, H2 ⊂ G be closed quantum subgroups of a locally compact quantum group G identified via πi ∈ Mor(Cu 0 (Hi)). Then the following conditions are equivalent: 0 (G), Cu (cid:3) (i) G = hH1, H2i (in the sense of Section 3); (ii) for all representations U ∈ B(HU ) ¯⊗L∞(G) of G we have that (6.8) {(id ⊗ω)(U ) : ω ∈ L1(G)}′′ = {(id ⊗ω1 ⊗ ω2)(U πr 1 12 U πr 2 23 ) : ωi ∈ L1(Hi)}′′; (iii) for the right regular representation W ∈ B(L2(G)) ¯⊗L∞(G) we have that (6.9) {(id ⊗ω)(W) : ω ∈ L1(G)}′′ = {(id ⊗ω1 ⊗ ω2)(Wπr 1 12 Wπr 2 23 ) : ωi ∈ L1(Hi)}′′. Proof. (i) =⇒ (ii). Assume that G = hH1, H2i, where the embedding Hi ⊂ G is described by means of a Hopf ∗-homomorphism πi ∈ Mor(Cu From Proposition 3.1 this is to say that 0 (Hi)) and a bicharacter V Hi ∈ L∞(bG) ¯⊗L∞(Hi). 0 (G), Cu (6.10) L∞(bG) = {(id ⊗ω)W : ω ∈ L1(G)}′′ = {(id ⊗ω1 ⊗ ω2)V H1 12 V H2 13 : ωi ∈ L1(Hi)}′′ 18 PAWE L J ´OZIAK, PAWE L KASPRZAK, AND PIOTR M. SO LTAN Let us fix a representation U ∈ B ¯⊗L∞(G) of G interpret it as a quantum subset X ⊂ bG as in the introduction. Let then θ: L∞(bG) → L∞(bG) ¯⊗B be the corresponding morphism. Let us apply the map θ to middle and right hand side of (6.10). The right hand side is then (6.11) {(θ ⊗ ω1 ⊗ ω2)(V H1 12 V H2 13 ) : ωi ∈ L1(Hi)}′′ = {(id ⊗ id ⊗ω1 ⊗ ω2)(U πr = {(id ⊗ id ⊗ω1 ⊗ ω2)(U πr 23 V H1 23 U πr 13 U πr 24 V H1 24 V H2 13 V H2 14 ) : ωi ∈ L1(Hi)}′′ 14 ) : ωi ∈ L1(Hi)}′′ 1 1 2 2 whereas the left (middle) hand side is (6.12) {(θ ⊗ ω)(W) : ω ∈ L1(G)}′′ = {(id ⊗ id ⊗ω)(U23W13) : ω ∈ L1(G)}′′ Applying (η ⊗ id) for η ∈ L1(bG) to all elements appearing in (6.12) and (6.11), by normality, yields: {(η ⊗ id ⊗ω1 ⊗ ω2)(U πr 23 U πr 2 1 14 ) : ωi ∈ L1(Hi)}′′ = {(η ⊗ id ⊗ω)(U23W13) : ω ∈ L1(G)}′′ 13 V H2 24 V H1 Now letting η run through the whole set L1(bG), we obtain {(η ⊗ id ⊗ω1 ⊗ ω2)(U πr 23 U πr 2 (6.13) 1 24 V H1 13 V H2 14 ) : ωi ∈ L1(Hi), η ∈ L1(bG)}′′ = {(η ⊗ id ⊗ω)(U23W13) : ω ∈ L1(G), η ∈ L1(bG)}′′ Remembering that (η ⊗ id)W generate C0(G) and that C0(G) ⊆ B(L2(G)) is nondegenerate, observe that the natural action of C0(G) on B(L2(G))∗ is non-degenerate (cf. [DKSS12, eq. (1.2)]), hence the right-hand side of (6.13) reads as: (6.14) {(η ⊗ id ⊗ω)(U23W13) : ω ∈ L1(G), η ∈ L1(bG)}′′ = {(η ⊗ id ⊗ω)(U23) : ω ∈ L1(G), η ∈ L1(bG)}′′ = {(id ⊗ω)(U ) : ω ∈ L1(G)}′′ and similarly the left-hand side of (6.13) turns into (6.15) {(η ⊗ id ⊗ω1 ⊗ ω2)(U πr 24 V H1 = {(η ⊗ id ⊗ω1 ⊗ ω2)(U πr 23 U πr 13 V H2 23 U πr 1 2 1 2 14 ) : ωi ∈ L1(Hi), η ∈ L1(bG)}′′ 24 ) : ωi ∈ L1(Hi), η ∈ L1(bG)}′′ 23 ) : ωi ∈ L1(Hi)}′′ 12 U πr 2 = {(id ⊗ω1 ⊗ ω2)(U πr 1 since V Hi generates C0(Hi) (see [DKSS12, Definition 3.2, Theorem 3.5 & Theorem 3.4]). Com- bining (6.13) with (6.15) and (6.14), we obtain (6.8). (ii) =⇒ (iii) is obvious, as one specializes U = W, whereas (iii) =⇒ (i) was already shown in Proposition 3.1 (and used as the starting point of the implication (i) =⇒ (ii)). (cid:3) Recall that for two representations U ∈ B(HU ) ¯⊗L∞(G) and U ∈ B(H U ) ¯⊗L∞(G) of a locally compact quantum group G we denote by HomG(U, U ) = {t ∈ B(HU , H U ) : (t ⊗ 1)U = U (t ⊗ 1)} the set of intertwiners between U and U . Then one obviously has HomG(U, U ) ⊆ HomH(U πr , U πr ) for every closed quantum subgroup H ⊂ G. Lemma 6.3. With the notation as above, let t ∈ B(HU ). Then t ∈ HomG(U, U ) if and only if (1 ⊗ t)θ(a) = θ(a)(1 ⊗ t) for all a ∈ L∞(bG). Proof. ( =⇒ ) This follows from the following simple computation, with a ∈ L∞(bG) and t ∈ σ(cid:18)(1 ⊗ t)θ(a)(cid:19) = (t ⊗ 1)U (1 ⊗ a)U ∗ = U (1 ⊗ a)(t ⊗ 1)U ∗ = U (1 ⊗ a)U ∗(t ⊗ 1) = σ(cid:18)θ(a)(1 ⊗ t)(cid:19) MorG(U, U ): HOPF IMAGES IN LOCALLY COMPACT QUANTUM GROUPS 19 ( =⇒ ) It is clear (and follows from WOT-density of slices of WG = W in L∞(bG)) that the assumption is equivalent to: (6.16) (1 ⊗ t ⊗ 1)(θ ⊗ id)(W) = (θ ⊗ id)(W)(1 ⊗ t ⊗ 1) and using (2.6) one develops (6.16) to obtain: (6.17) which can be rewritten as t1U ∗ 12W23U12 = U ∗ 12W23U12t1 (6.18) Observe that the right hand side of (6.18) is nothing but (id ⊗∆)(U12t1U ∗ of coproduct (see, e.g. [MRW12, Theorem 2.6]) we have that 12 = W23U12t1U ∗ U12t1U ∗ 12W∗ 23 12), so using ergodicity U12t1U ∗ 12 ∈ B(HU ) ¯⊗C1 ¯⊗C1 Let then m ∈ B be such that: (6.19) and write this as (6.20) U (t ⊗ 1)U ∗ = m ⊗ 1 U (t ⊗ 1) = (m ⊗ 1)U Applying (id ⊗∆) to both sides of (6.20) and rearranging the terms one arrives at 12 ∈ B(HU ) ¯⊗L∞(G) ¯⊗C1 B(HU ) ¯⊗C1 ¯⊗L∞(G) ∋ U13t1U ∗ 13 = U12m1U ∗ (6.21) Hence it follows (see, e.g. [KR92, 12.4.36]) that (6.22) m1 = U13t1U ∗ 13 = U12m1U ∗ 12 and consequently m ⊗ 1 commutes with U and hence t = m, which finishes the proof in view of (6.19). (cid:3) Theorem 6.4 (Theorem D). Let (Hi)i∈I ⊂ G be a family of closed subgroups. Then G is generated by (Hi)i∈I if and only if for all representations U ∈ B(HU ) ¯⊗L∞(G) and U ∈ B(H U ) ¯⊗L∞(G) we have that HomG(U, U ) = \i∈I HomG(U, U ) ⊆ \i∈I HomHi (U πr i , U πr i ) HomHi(U πr i , U πr i ) (6.23) Proof. It is clear (and noted before Lemma 6.3) that i , U πr hence the genuine statement is to obtain the converse containment, under assumption that G is generated by (Hi)i∈I . Assume first that U = U and t ∈Ti∈I HomHi(U πr for all i ∈ I and a ∈ L∞(bHi), where θi is as Lemma 6.1 for Hi ⊂ G. But Lemma 6.1, together with the assumption L∞(bG) = (Si∈I γi(L∞(bHi)))′′ and normality of all the maps involved gives for all and a ∈ L∞(bG), hence we conclude by Lemma 6.3 that t ∈ HomG(U, U ) If now U and U are arbitrary, one can consider U ⊕ U . Observe that then t ∈ HomG(U, U ) if (t ⊗ 1)θi(a) = θi(a)(t ⊗ 1) (t ⊗ 1)θ(a) = θ(a)(t ⊗ 1) i ). Then us that and only if T =(cid:18)0 t 0 0(cid:19) ∈ HomG(U ⊕ U , U ⊕ U ) To conclude (6.23), we apply the first part of the proof to U ⊕ U and T . The only things that one needs to verify is that (U ⊕ U )πr i (which is clear in view of (1.8)) and that the block form of T remains after we restrict U ⊕ U to any Hi, which is again obvious. i = U πr i ⊕ U πr 20 PAWE L J ´OZIAK, PAWE L KASPRZAK, AND PIOTR M. SO LTAN To prove the other implication, let us pick 1 ∈ I and consider H1 ⊂ G and H2 = hSi∈I\{1} Hji. It is now enough to take (6.23) with U = U = W to arrive at (the commutant of) (6.9). We are done thanks to Theorem 6.2. (cid:3) Let us note the following Corollary 6.5. Assume that G is generated by (Hi)i∈I and let U ∈ B(HU ) ¯⊗L∞(G) be a repre- sentation of G. If K ⊆ H is preserved by each Hi, then it is preserved by G. Proof. K ⊆ H is preserved by each Hi if and only if pK ∈ MorHi(U πr to Theorem 6.4. i ). We are done thanks (cid:3) i , U πr Now let us remark the following: Theorem 6.6. Let G be a compact quantum group. (1) The quantum group G is topologically generated by the quantum subgroups H1 and H2 (in the sense of [BCV17]) if and only if G = hH1, H2i (in the sense of Definition 3.2). (2) The family of Hopf quotients Cu(G) → Cu(Hi) forms a jointly full family (in the sense of [Chi15]) if and only if G = hSi∈I Hii (in the sense of Definition 3.2). Proof. (1) The proof was outlined in [BCV17, Remark 3] and relies on Theorem 6.4 which extends [BB10, Corollary 8.2] outside the compact realm. (2) The way joint fullness is phrased in [Chi15, Definition 2.15] is precisely the conclusion of Theorem 6.4. (cid:3) 7. Examples and applications 7.1. Classical subsets in quantum groups. Fix a morphism β ∈ Mor(Cu 0 (G), B) and assume that the C∗-algebra B is commutative. Then from general theory of C∗-algebras it follows that β factors through the abelianization of Cu 0 (G). The latter was identified in [Daw16, KN13] as a closed quantum subgroup of G: the group of characters of G, or the maximal classical subgroup of G, which we denote by G. Thus the Hopf image of the morphism β is not bigger than G, and it is precisely the subgroup of G generated by the image of the spectrum of B under the Gelfand dual map of β, as one can easily check ( [J´oz16, Theorem 3.5]). 7.2. Quantum "az + b" groups. The quantum "az + b" groups were introduced in [Wor01] and [So l05] are quantum deformations of the group of affine transformations of C. The construction of a quantum "az + b" group begins with a choice of a complex deformation parameter q from a certain set (see [So l05]). The parameter determines a multiplicative subgroup Γ of C \ {0} and we let Γ be the closure of Γ in C which is Γ ∪ {0}. The C∗-algebra C0(G) is then isomorphic to the crossed product C0(Γ) ⋊ Γ with the action of Γ on Γ given by multiplication of complex numbers. Let b be the image under the natural morphism C0(Γ) → C0(Γ) ⋊ Γ = C0(G) of the element z affiliated with C0(Γ) given by z(γ) = γ for all γ ∈ Γ. Then b is normal and its spectrum is equal to Γ. Furthermore let {Uγ}γ∈Γ be the family of unitaries in M (C0(G)) implementing the action of Γ. Then for any γ ∈ Γ we have UγbU ∗ γ = γb. The quantum group G is coamenable ( [So l05, Section 6.2]), so C0(G) = Cu 0 (G) and so any morphism β ∈ Mor(Cu 0 (G), B) is determined by a covariant representation of the dynamical system (Γ, Γ). Using this one can show that β is either injective, or it has commutative image. In the latter case the Hopf image is a classical group, so there is no generating morphism β which is not an isomorphism. In the dual language we can formulate this by saying that the quantum "az + b" group does not contain a proper subset which generates it. 7.3. Double groups. The procedure of constructing so called "double groups" in the C∗-algebraic framework goes back to the paper [PW90] where a quantum deformation of the group SL(2, C) was constructed via the double group construction applied to the quantum SU(2) group. The construction which we will very briefly recall below yields always a non-compact locally compact quantum group and various examples have been presented in literature (see e.g. [Pus02]). Let us HOPF IMAGES IN LOCALLY COMPACT QUANTUM GROUPS 21 note that in [MNW03] the authors use an alternative name for the double group construction, namely "quantum codouble". The very rough view of the construction of the double group built over a locally compact quan- tum group K is as follows: the C∗-algebra C0(G) is defined as C0(K)⊗C0(bK) with comultiplication ∆G defined on simple tensors by ∆G(a ⊗ b) = WK 23∆K(a)13∆bK(b)24WK 23 ∗. It turns out that G defined by (C0(G), ∆G) is a locally compact quantum group. Moreover K and bK are both closed quantum subgroups of G. If we let γK and γ bK be the corresponding embeddings L∞(bK) → L∞(bG) and L∞(K) → L∞(bG) then by [Yam00, Section 4] (see in particular [Yam00, Corollary 4.12 and Formula (QD7)]) L∞(bG) is generated by the images of γK and γ bK. It follows from Proposition 3.1 that G is generated by K and bK. closed subgroup Γ of a locally compact group G and a 2-cocycle Ψ on bΓ. The resulting quantum 7.4. Rieffel deformations. Rieffel deformation is a procedure which allows one to deform a lo- cally compact group into a quantum group. In order to perform it one needs to fix an abelian group is denoted by GΨ, see [Kas11, Rie93, Rie95, EVn96] for further informations on Rieffel de- formations. As noted in [Kas11, Section 3], if Γ ⊂ H ⊂ G is a chain of closed subgroups, then HΨ ⊂ GΨ. Now suppose that hH1, H2i = G and that Γ ⊂ H1 ∩ H2. It follows from [Kas09, Theorem 4.12] 2 i = GΨ. A particular example of this construction is obtained by and Proposition 3.1 that hHΨ taking G = SL(2, C), H1 -- the group of upper-triangular matrices and H2 are the lower-triangular matrices; Γ = H1 ∩ H2 ∼= C× is the group of diagonal matrices. For the 2-cocycle Ψ one can take a cocycle described in [Kas09, Section 5]. 1 , HΨ Acknowledgement PJ, PK and PMS were partially supported by the NCN (National Center of Science) grant no. 2015/17/B/ST1/00085. PJ would like to express his gratitude to Alex Chirvasitu for asking questions that led to the answers contained in Section 6, as well as polishing the presentation of these results compared to the one given in [J´oz16, Section 3.2.3]. References [BB09] [BB10] Teodor Banica and Julien Bichon. Quantum groups acting on 4 points. J. Reine Angew. Math., 626:75 -- 114, 2009. Teodor Banica and Julien Bichon. Hopf images and inner faithful representations. Glasg. Math. J., 52(3):677 -- 703, 2010. [BCV17] Michael Brannan, Benoı t Collins, and Roland Vergnioux. The Connes embedding property for quantum [Bla06] [BO08] [BS93] [BV05] [BY14] [Chi15] [CHK] group von Neumann algebras. Trans. Amer. Math. Soc., 369(6):3799 -- 3819, 2017. B. Blackadar. Operator algebras, volume 122 of Encyclopaedia of Mathematical Sciences. Springer- Verlag, Berlin, 2006. Theory of C ∗-algebras and von Neumann algebras, Operator Algebras and Non- commutative Geometry, III. Nathanial P. Brown and Narutaka Ozawa. C ∗-algebras and finite-dimensional approximations, volume 88 of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI, 2008. Saad Baaj and Georges Skandalis. Unitaires multiplicatifs et dualit´e pour les produits crois´es de C ∗- alg`ebres. Ann. Sci. ´Ecole Norm. Sup. (4), 26(4):425 -- 488, 1993. Saad Baaj and Stefaan Vaes. Double crossed products of locally compact quantum groups. J. Inst. Math. Jussieu, 4(1):135 -- 173, 2005. Julien Bichon and Robert Yuncken. Quantum subgroups of the compact quantum group SU−1(3). Bull. Lond. Math. Soc., 46(2):315 -- 328, 2014. Alexandru Chirvasitu. Residually finite quantum group algebras. J. Funct. Anal., 268(11):3508 -- 3533, 2015. A. Chirvasitu, S. O. Hoche, and P. Kasprzak. Fundamental isomorphism theorems for quantum groups. Expo. Math. in press. [Daw16] Matthew Daws. Categorical aspects of quantum groups: multipliers and intrinsic groups. Canad. J. Math., 68(2):309 -- 333, 2016. [DKSS12] Matthew Daws, Pawe l Kasprzak, Adam Skalski, and Piotr M. So ltan. Closed quantum subgroups of locally compact quantum groups. Adv. Math., 231(6):3473 -- 3501, 2012. 22 PAWE L J ´OZIAK, PAWE L KASPRZAK, AND PIOTR M. SO LTAN [EVn96] Michel Enock and Leonid Vaı nerman. Deformation of a Kac algebra by an abelian subgroup. Comm. [Izu02] Math. Phys., 178(3):571 -- 596, 1996. Masaki Izumi. Non-commutative Poisson boundaries and compact quantum group actions. Adv. Math., 169(1):1 -- 57, 2002. Pawe l J´oziak. Hopf images in Locally compact quantum groups. PhD thesis, IMPAN, 2016. P. Kasprzak. Rieffel deformation via crossed products. J. Funct. Anal., 257(5):1288 -- 1332, 2009. P. Kasprzak. Rieffel deformation of homogeneous spaces. J. Funct. Anal., 260(1):146 -- 163, 2011. [J´oz16] [Kas09] [Kas11] [KKS16] Mehrdad Kalantar, Pawe l Kasprzak, and Adam Skalski. Open quantum subgroups of locally compact quantum groups. Adv. Math., 303:322 -- 359, 2016. [KKS17] P. Kasprzak, F. Khosravi, and P. M. So ltan. Integrable actions and quantum subgroups. Int. Math. Res. Not., Published online (2017), 2017. [KN13] Mehrdad Kalantar and Matthias Neufang. From quantum groups to groups. Canad. J. Math., [KR92] [KS14a] 65(5):1073 -- 1094, 2013. Richard V. Kadison and John R. Ringrose. Fundamentals of the theory of operator algebras. Vol. IV. Birkhauser Boston, Inc., Boston, MA, 1992. Special topics, Advanced theory -- an exercise approach. P. Kasprzak and P. M. So ltan. Quantum groups with projection and extensions of locally compact quantum groups. ArXiv e-prints, December 2014. [KS14b] Pawe l Kasprzak and Piotr M. So ltan. Embeddable quantum homogeneous spaces. J. Math. Anal. Appl., [KS15] 411(2):574 -- 591, 2014. Pawe l Kasprzak and Piotr M. So ltan. Quantum groups with projection on von Neumann algebra level. J. Math. Anal. Appl., 427(1):289 -- 306, 2015. [KSS16] Pawe l Kasprzak, Adam Skalski, and Piotr M. So ltan. The canonical central exact sequence for locally [Kus01] [KV00] compact quantum groups. Math. Nachr., Published online (2016), 2016. Johan Kustermans. Locally compact quantum groups in the universal setting. Internat. J. Math., 12(3):289 -- 338, 2001. Johan Kustermans and Stefaan Vaes. Locally compact quantum groups. Ann. Sci. ´Ecole Norm. Sup. (4), 33(6):837 -- 934, 2000. [MNW03] Tetsuya Masuda, Yoshiomi Nakagami, and Stanis law Lech Woronowicz. A C ∗-algebraic framework for quantum groups. Internat. J. Math., 14(9):903 -- 1001, 2003. [MRW12] Ralf Meyer, Sutanu Roy, and Stanis law Lech Woronowicz. Homomorphisms of quantum groups. Munster [NY14] J. Math., 5:1 -- 24, 2012. Sergey Neshveyev and Makoto Yamashita. Categorical duality for Yetter-Drinfeld algebras. Doc. Math., 19:1105 -- 1139, 2014. [Pus02] W. Pusz. Quantum GL(2, C) group as double group over "az + b" quantum group. Rep. Math. Phys., [PW90] 49(1):113 -- 122, 2002. P. Podle´s and S. L. Woronowicz. Quantum deformation of Lorentz group. Comm. Math. Phys., 130(2):381 -- 431, 1990. [Rie93] Marc A. Rieffel. Deformation quantization for actions of Rd. Mem. Amer. Math. Soc., 106(506):x+93, 1993. [Rie95] Marc A. Rieffel. Non-compact quantum groups associated with abelian subgroups. Comm. Math. Phys., [So l05] [SS16] [SW07] 171(1):181 -- 201, 1995. Piotr Miko laj So ltan. New quantum "az + b" groups. Rev. Math. Phys., 17(3):313 -- 364, 2005. Adam Skalski and Piotr M. So ltan. Quantum families of invertible maps and related problems. Canad. J. Math., 68(3):698 -- 720, 2016. Piotr M. So ltan and Stanis law Lech Woronowicz. From multiplicative unitaries to quantum groups. II. J. Funct. Anal., 252(1):42 -- 67, 2007. [TT72] Masamichi Takesaki and Nobuhiko Tatsuuma. Duality and subgroups. II. J. Functional Analysis, 11:184 -- [Vae01] [Ver04] 190, 1972. Stefaan Vaes. Locally compact quantum groups. PhD thesis, Katholieke Universiteit Leuven, 2001. Roland Vergnioux. K-amenability for amalgamated free products of amenable discrete quantum groups. J. Funct. Anal., 212(1):206 -- 221, 2004. Roland Vergnioux. Orientation of quantum Cayley trees and applications. J. Reine Angew. Math., 580:101 -- 138, 2005. Roland Vergnioux. The property of rapid decay for discrete quantum groups. J. Operator Theory, 57(2):303 -- 324, 2007. Shuzhou Wang. Free products of compact quantum groups. Comm. Math. Phys., 167(3):671 -- 692, 1995. Stanis law Lech Woronowicz. C ∗-algebras generated by unbounded elements. Rev. Math. Phys., 7(3):481 -- 521, 1995. Stanis law Lech Woronowicz. From multiplicative unitaries to quantum groups. Internat. J. Math., 7(1):127 -- 149, 1996. [Wor01] S. L. Woronowicz. Quantum "az + b" group on complex plane. Internat. J. Math., 12(4):461 -- 503, 2001. [Yam00] Takehiko Yamanouchi. Double group construction of quantum groups in the von Neumann algebra [Wan95] [Wor95] [Ver05] [Ver07] [Wor96] framework. J. Math. Soc. Japan, 52(4):807 -- 834, 2000. HOPF IMAGES IN LOCALLY COMPACT QUANTUM GROUPS 23 Institute of Mathematics of the Polish Academy of Sciences and Institute of mathematics, Univer- sity of Wroc law, Poland E-mail address: [email protected] Department of Mathematical Methods in Physics, Faculty of Physics, University of Warsaw, Poland E-mail address: [email protected] Department of Mathematical Methods in Physics, Faculty of Physics, University of Warsaw, Poland E-mail address: [email protected]
1303.2151	2	1303	2013-09-05T11:37:14	Haagerup property for quantum reflection groups	[ "math.OA", "math.QA" ]	In this paper we prove that the duals of the quantum reflection groups have the Haagerup property for all $N\ge4$ and $s\in[1,\infty)$. We use the canonical arrow onto the quantum permutation groups, and we describe how the characters of the quantum reflection groups behave with respect to this canonical morphism thanks to the description of the fusion rules binding irreducible corepresentations of $H_N^{s+}$. This allows us to construct states on the central algebra of $H_{N}^{s+}$ and to use a fundamental theorem proved by M.Brannan giving a method to construct nets of trace-preserving, normal, unital and completely positive maps on the von Neumann algebra of a compact quantum group of Kac type.	math.OA	math	HAAGERUP APPROXIMATION PROPERTY FOR QUANTUM REFLECTION GROUPS FRANC¸ OIS LEMEUX s+ N ) → C(S+ Abstract. In this paper we prove that the duals of the quantum reflection s+ N have the Haagerup property for all N ≥ 4 and s ∈ [1, ∞). We groups H N ) onto the quantum permutation use the canonical arrow π : C(H s+ N ) behave with respect groups, and we describe how the characters of C(H to this morphism π thanks to the description of the fusion rules binding irre- s+ ducible corepresentations of C(H N ) ([4]). This allows us to construct states s+ s+ on the central C ∗-algebra C(H N ) N )0 generated by the characters of C(H and to use a fundamental theorem proved by M.Brannan giving a method to construct nets of trace-preserving, normal, unital and completely positive maps on the von Neumann algebra of a compact quantum group G of Kac type ([6]). Introduction A (classical) discrete group Γ has the Haagerup property if (and only if) there is a net (ϕi) of normalized positive definite functions in C0(Γ) converging pointwise to the constant function 1. There are lots of examples of discrete groups with the Haagerup property: all amenable groups have this property. The free groups FN are examples of discrete groups with Haagerup property (see [13]) but which are not amenable. Thus, one says that the Haagerup property is a weak form of amenability. This property is also known as a "strong negation" of Kazhdan's property (T ): the only (classical) discrete groups with both properties are finite. Another weak form of amenability is the weak amenability, see below for examples in the quantum setting. One can find more examples and a more complete approach to the problems and questions related to the Haagerup property, also called "a-T - amenability", in [9]. The Haagerup property has many interests in various fields of mathematics such as geometry of groups or functional analysis. We can mention e.g. groups with wall space structures (see [10] and [8]) as illustrations of the interest in the Haagerup property with respect to the theory of geometry of groups. In functional analy- sis, the Haagerup property appears e.g. in questions related to the Baum-Connes conjecture (see [14]) or in Popa's deformation/rigidity techniques (see [16]). In [6], a natural definition of the Haagerup property for compact quantum groups G of Kac type is proposed: G has the Haagerup approximation property if and only if its associated (finite) von Neumann algebra L∞(G) has the Haagerup property (see also below, Definition 1.1). We use this definition with the slight modification: the dual bG of G has the Haagerup property if L∞(G) has the Haagerup property, so that this definition is closer to the classical case where bG is a classical discrete 2010 Mathematics Subject Classification. Primary : 46L54, 16T20. Secondary : 46L65, 20G42. 1 2 FRANC¸ OIS LEMEUX group. The author of [12] proposes another definition for the Haagerup property of discrete quantum groups: bG has the Haagerup property if there exists a net (ai) in c0(bG) which converges to 1 pointwise and such that the associated multipliers mai are unital and completely positive. These approaches are equivalent in the unimodular case. N and S+ In fact, in [7], O+ erty. phism group of a finite dimensional C∗-algebra has the Haagerup property. In [6] and [7], the author shows that the duals of the compact quantum groups N , U + N , introduced by Wang (see [19] and [20]) have the Haagerup prop- it is proved that any trace-preserving quantum automor- In N has the Haagerup property (precisely that some completely positive multipliers can be N is weakly amenable (in fact, it is also proved N is weakly amenable, and an argument of monoidal N is weakly amenable too). In [11], a definition of property (T ) for discrete quantum groups and some classical proper- ties for discrete groups are generalized, for instance: discrete quantum groups with property (T ) are finitely generated and unimodular. [12], using some block decompositions and Brannan's proof of the fact thatdO+ found), the author proves that dO+ in [12] that the dU + equivalence allows to prove, in particular, that cS+ N ⊂ Z ∗dO+ The aim of this paper is to prove that the duals of quantum reflection groups H s+ N , introduced in [1], have the Haagerup property. It is a natural generalization of the case s = 1 treated in [7] (since H 1+ N ). However, this generalization is not immediate: as a matter of fact, the sub C∗-algebra generated by the characters is not commutative so that the strategy used in [6] and [7] does not work anymore. However, a fundamental tool of the proof of the main result of our paper is [6, Theorem 3.7]. N = S+ a discrete group which has the Haagerup property thenbΓ ≀w S+ What motivates our paper is also the fact that quantum reflection groups are free wreath products between Zs and S+ N (see [5] and Theorem 1.17 below) and the result proved in our paper naturally leads to the following question: is it true that if Γ is N has the Haagerup property ? One can notice the similarity with the result in [10] concerning (classical) wreath products of discrete groups: If Γ, Γ′ are countable discrete groups with the Haagerup property then Γ ≀ Γ′ also has the Haagerup property. This similarity is however formal: in our paper we are considering (free) wreath products of groups whose duals have the Haagerup property. Our proof of the fact dH s+ N have the Haagerup property relies on the knowledge of the fusion rules of the associated compact quantum group H s+ N , determined in [4]. Indeed, there is no general result about fusion rules for free wreath products of compact quantum groups yet. The rest of the paper is organized as follows. In the section 1, we recall the definition of the Haagerup property for compact quantum groups of Kac type and we give the result of Brannan concerning the construction of normal, unital, completely positive and trace-preserving maps on L∞(G) (see Theorem 1.2). We also give a positive answer to a question asked in [22], in the discrete and Kac setting case, concerning symmetric tensors with respect to the coproduct. Then we collect some results on Tchebyshev polynomials: some are already mentioned and used in [6], but we give suitably adapted statements and proofs for our purpose. Thereafter, we recall the definition of quantum reflection groups H s+ N , and we describe their irreducible corepresentations and the fusion rules binding them. We also recall that HAAGERUP APPROXIMATION PROPERTY FOR QUANTUM REFLECTION GROUPS 3 N . In section 2, we identify N ) by the canonical morphism onto N ). In the section 3, we prove that the duals of the quantum reflection groups at s = 1, we get the quantum permutation groups S+ the images of the irreducible characters of C(H s+ C(S+ H s+ N have the Haagerup approximation property for all N ≥ 4. 1. Preliminaries Let us first fix some notations. One can refer to [6], [18], [15] and [24] for more details. In this paper, G = (C(G), ∆) will denote a compact quantum group, where C(G) is a full Woronowicz C∗-algebra. Furthermore, every compact quantum group G considered in this paper is of Kac type (or equivalentely, its dual bG is unimodular) that is: the unique Haar state h on C(G), is tracial. (We recall that L∞(G) is defined by L∞(G) = Cr(G)′′ = πh(C(G))′′, where (L2(G), πh) is the GNS construction associated to h). 1.1. Haagerup property for compact quantum groups of Kac type. Definition 1.1. The dual bG of a compact quantum group G = (C(G), ∆) of Kac type has the Haagerup approximation property if the finite von Neumann algebra (L∞(G), h) has the Haagerup approximation property i.e. if there exists a net (φx) of trace preserving, normal, unital and completely positive maps on L∞(G) such that their unique extensions to L2(G) are compact operators and (φx) converges to idL∞(G) pointwise in L2-norm. One essential tool to construct nets of normal, unital, completely positive and trace preserving maps (we will say NUCP trace preserving maps) is the next the- orem proved in [6]. We will denote by Irr(G) the set indexing the equivalence classes of irreducible corepresentations of a compact quantum group G and by P ol(G) the linear space spanned by the matrix coefficients of such corepresenta- α(G) ⊂ L2(G) be the subspace spanned tions uα, α ∈ Irr(G). If α ∈ Irr(G), let L2 by the GNS images of matrix coefficients uα ij, i, j ∈ {1, . . . , dα} of the irreducible ij)) and pα : L2(G) → L2 unitary corepresentation uα (dα = dim(uα α(G) be the α(G). We de- note by C(G)0 ⊂ C(G) the C∗-algebra generated by the irreducible characters ii of a compact quantum group G and χα the character of the associ- ated conjugate corepresentation uα. associated orthogonal projection. Then L2(G) = l2 −Lα∈Irr(G) L2 χα =Pdα i=1 uα Theorem 1.2. [6, Theorem 3.7] Let G = (C(G), ∆) be a compact quantum group of Kac type. Then for any state ψ ∈ C(G)∗0, the map pα ψ(χα) Tψ = Xα∈Irr(G) dα is a unital contraction on L2(G) and the restriction of Tψ to L∞(G) defines a NUCP h-preserving map still denoted Tψ. The averaging methods used to prove this theorem allow us to answer, in a restricted setting, a question asked in [22]. Let G = (C(G), ∆) be a compact quantum group. Then consider the C∗- subalgebra C(G)central := {a ∈ C(G) : ∆(a) = Σ ◦ ∆(a)} i.e. the C∗-subalgebra of the symmetric tensors in C(G) ⊗ C(G) with respect to ∆ (Σ denotes the usual flip map Σ : C(G) ⊗ C(G) → C(G) ⊗ C(G), a ⊗ b 7→ b ⊗ a). In [22], the author also 4 FRANC¸ OIS LEMEUX defines P ol(G)central := {a ∈ P ol(G) : ∆(a) = Σ ◦ ∆(a)}. We recall the question asked by Woronowicz (see [22] thereafter Proposition 5.11): Question 1.3. Is P ol(G)central dense in C(G)central (for the norm of C(G)) ? Then the answer is yes, at least in the Kac and discrete setting. We simply denote by . the norm on C(G). It is clear, and proved in [22], that P ol(G)central = span{χα : α ∈ Irr(G)} where χα =Pdα i=1 uα ii denotes the character of an irreducible finite dimentional corepresentation (uα ij ). So the problem reduces to prove that C(G)central ⊂ span.{χα : α ∈ Irr(G)}, the other inclusion being clear. Theorem 1.4. Let Gr = (C(Gr), ∆r) be a compact quantum group of Kac type with faithful Haar state. Then P ol(Gr)central . = C(Gr)central. Proof. We first note that ∆r preserves the trace in the sense that (h⊗ h)◦ ∆r = h. As a result the Hilbertian adjoint ∆∗r, of the L2-extension of ∆r, is well-defined and we have ∆∗r(x) ≤ x for x ∈ C(Gr)⊗ C(Gr) with respect to the operator norms (note that this is particular to the tracial situation). Since ∆∗r clearly maps the subspace P ol(Gr) ⊗ P ol(Gr) of L2(Gr) ⊗ L2(Gr) to P ol(Gr), it also restricts to a contractive map from C(Gr)⊗ C(Gr) to C(Gr), still denoted ∆∗r. Now we put E = ∆∗r ◦ Σ ◦ ∆r : C(Gr) → C(Gr). We have E ≤ 1, and for a ∈ C(Gr)central, E(a) = ∆∗r ◦ ∆r(a) = a so that C(Gr)central ⊂ E(C(Gr)). But, on the other hand, for any matrix coefficient of a finite dimensional unitary corepresentations (uα ij), we have E(uα ij) = ∆∗r ◦ Σ ◦ ∆r(uα We compute ∆∗r(cid:16)Pk uα pq, ∆∗r Xk *uβ uα kj ⊗ uα product coming from the Haar state h: let β ∈ Irr(Gr), then for all 1 ≤ p, q ≤ dβ: uα ik ⊗ uα ik! . kj ⊗ uα uα kj! = ∆∗r Xk ij) = ∆∗r ◦ Σ Xk ik(cid:17) using the duality pairing induced by the inner ik!+h ikEh kj ⊗ uα δαβδijδpq lq, uα d2 α = kj ⊗ uα =Xl,k Duβ =(cid:28)uβ pq, pl ⊗ uβ χα(cid:29)h δij dα . Then summarizing, we have E(uα E = 1 and EP ol(Gr )central = id. Thus we obtain a conditional expectation E : C(Gr) → . = span.{χα : α ∈ Irr(Gr)}. But we have C(Gr)central ⊂ P ol(Gr)central E(C(Gr)) and the result follows. χα ∈ Pol(Gr)central, ij) = δij dα (cid:3) Notation 1.5. We will denote by P ol(G)0 and C(G)0 the central ∗-algebras and C∗-algebras generated by the irreducible characters of a compact quantum group G. HAAGERUP APPROXIMATION PROPERTY FOR QUANTUM REFLECTION GROUPS 5 1.2. Tchebyshev polynomials. Definition 1.6. We define a family of polynomials (At)t∈N as follows: A0 = 1, A1 = X and for all t ≥ 1 (1.1) A1At = At+1 + At−1. We call them the dilated Tchebyshev polynomials of second kind. We will use the following results on Tchebyshev polynomials At. The second one is based upon a result proved in [6, Proposition 4.4], but suitably adapted to our purpose. Proposition 1.7. for all t, s ≥ 1 we have: AtAs = At+s + At−1As−1 Proof. This result is easily proved by induction on t ≥ 1. Proposition 1.8. Let N ≥ 2. For all x ∈ (2, N ), there exists a constant c ∈ (0, 1) such that for all integers t ≥ 1 we have At(x) (cid:3) 0 < N(cid:17)ct At(N ) ≤(cid:16) x . Proof. First, we follow the proof of [6, Proposition 4.4] and introduce the function q(x) = x+√x2−4 , for x > 2. Then an induction and the recursion formula (1.1) for the polynomials At show that for all t ≥ 0, we have 2 At(x) = q(x)t+1 − q(x)−t−1 q(x) − q(x)−1 Then using the same tricks as in [6], we get that for all fixed x ∈ (2, N ) and all t ≥ 1 At(x) At(N ) = q(x)t+1 − q(x)−t−1 q(N )t+1 − q(N )−t−1 q(N ) − q(N )−1 q(x) − q(x)−1 N(cid:17)t =(cid:16) x x2 1 +q1 − 4 1 +q1 − 4 N 2 t  1 − q(x)−2t−2 1 − q(N )−2N−2 1 − q(N )−2 1 − q(x)−2 . Now notice that the factor Furthermore, we have 1−q(x)−2t−2 1−q(N )−2N −2 is less than 1 because q is increasing. since the last factor does not depend on t and t x2 N 2  1 +q1 − 4  1 +q1 − 4 N(cid:17)t At(N ) ≤ (cid:16) x At(N ) ≤ (cid:0) x N(cid:17)ct0 N(cid:17)ct ≤ (cid:16) x At(x) At(x) 1 − q(N )−2 1 − q(x)−2 −→t→∞ 1+q1− 4 1+q1− 4 N 2 x2 0 < 1. Hence, there ex- for all t ≥ t0. It remains to show that there N(cid:1)ct0 for all t = 1, . . . , t0 − 1, since for . To prove that such a c exists, we notice ists t0 such that exists c ∈ (0, 1) such that all 0 < t < t0, (cid:16) x 6 FRANC¸ OIS LEMEUX At(N ) : t = 1, . . . , t0 − 1o := D < 1 since the Tchebyshev polynomials are increasing on (2, +∞). Hence, it is clear that we can find c > 0 such that (cid:3) that maxn At(x) N(cid:17)ct0 (cid:16) x ≥ D. Remark 1.9. (1) In [6, Proposition 6.4], the exponent is better (there is no constant c) but there is a constant multiplying (cid:16) x N(cid:17)t an easy proof of Proposition 3.3 below. . Our version allows (2) The previous proposition gives information on the behavior of the dilated At(x) At(N ) Tchebyshev polynomials on (2, +∞): the quotient has an exponen- tial decay with respect to t ≥ 1. We will also need some informations on this quotient when x ∈ (0, 2) and N = 2. That is the aim of the next paragraph. The polynomials At are linked to the Tchebyshev polynomials of second kind Ut by the following formula: ∀t ∈ N, x ∈ [0, 1], At(2x) = Ut(x). Indeed, we recall (see [17] for more details) that the Tchebyshev polynomials of second kind Ut are defined for all x ∈ [−1, 1] by (1.2) Ut(x) = sin((t + 1) arccos(x)) √1 − x2 = sin((t + 1)θ) sin(θ) , with x = cos(θ). In particular, U0 = 1, U1(x) = 2x and for all t ∈ N∗, Ut(1) = t + 1. Then one At(x) can check that for all t ∈ N and x ∈ [0, 1]: 2xUt(x) = Ut+1(x) + Ut−1(x). Proposition 1.10. Let x ∈ (0, 2). Then for any integer t ≥ 1 At(2)(cid:12)(cid:12)(cid:12)(cid:12) = (cid:12)(cid:12)(cid:12)(cid:12) , with x = cos(θ). In particular, there exists a positive constant D < 1 such that ∀t ≥ 1,(cid:12)(cid:12)(cid:12)(cid:12) At(x) At(2) −→ 0 as t → ∞. t + 1(cid:12)(cid:12)(cid:12)(cid:12) sin((t + 1)θ) sin(θ) (cid:12)(cid:12)(cid:12)(cid:12) and 1 At(x) At(2)(cid:12)(cid:12)(cid:12)(cid:12) ≤ D Proof. First, by what we recalled above, we can write At(x) At(2) = Ut( x 2 ) Ut(1) = Ut( x 2 ) t + 1 . Thus, if x = 2 cos(θ), we have by the relation (1.2) above (cid:12)(cid:12)(cid:12)(cid:12) At(x) At(2)(cid:12)(cid:12)(cid:12)(cid:12) = 1 t + 1(cid:12)(cid:12)(cid:12)(cid:12) sin((t + 1)θ) sin(θ) 0. (cid:12)(cid:12)(cid:12)(cid:12) −→t→∞ On the other hand, on [0, 1], the polynomials Ut, t ≥ 1 have t + 1 as a maximum, only attained in 1. Then, it is clear that for all t ≥ 1 and x ∈ (0, 2): 0 < At(x) At(2) = Ut( x 2 ) t + 1 < 1. So the existence of the announced constant D is clear. (cid:3) HAAGERUP APPROXIMATION PROPERTY FOR QUANTUM REFLECTION GROUPS 7 N . We also recall that C(H s+ 1.3. Quantum reflection groups. In this subsection, we recall the definition of the quantum reflection groups H s+ N and the particular case of the quantum permutation groups S+ N ) is the free wreath product of two quantum permutation algebras. In the end of this subsection, we recall the description of the irreducible corepresentations of C(H s+ N ) together with the fusion rules binding them. Definition 1.11. [1, Definition 11.3] Let N ≥ 2, s ≥ 1 be integers. The quantum reflection group H s+ N ), ∆) composed of the universal C∗-algebra generated by N 2 normal elements Uij satisfying the following relations N is the pair (C(H s+ (1) U = (Uij) is unitary, (2) tU = (Uji) is unitary, (3) pij = UijU∗ij is a projection, (4) U s ij = pij together with the coproduct ∆ : C(H s+ N ) → C(H s+ N ) ⊗ C(H s+ N ) given by ∆(Uij ) =Xk Uik ⊗ Ukj . Remark 1.12. (1) For s = 1 we get the quantum permutation group S+ thus may be summed up as follows (see also [20]): S+ where (a) C(S+ N . The definition of S+ N N is the pair (C(S+ N ), ∆) N ) is the universal C∗-algebra generated by N 2 elements vij such ij (i.e. v is a that the matrix v = (vij ) is unitary and vij = v∗ij = v2 magic unitary). (b) The coproduct is given by the usual relations making of v a corepre- sentation (the fundamental one) of C(S+ N ). (2) For s = 2, we find the hyperoctahedral quantum group, i.e. the easy quan- tum group H + N studied e.g. in [21]. (3) There is a morphism C(H s+ only has to check that the generators vij of C(S+ described in Definition 1.11, which is clear. N ) → C(S+ N ) of compact quantum groups: one N ), satisfy the relations Notation 1.13. We will denote by π : C(H s+ mentioned in the remark above. N ) → C(S+ N ) the canonical arrow Here are the results concerning the irreducible corepresentations of C(S+ N ): Theorem 1.14. [3, Theorem 4.1] There is a maximal family(cid:0)v(t)(cid:1)t∈N of pairwise inequivalent irreducible finite dimensional unitary representations of S+ N such that: (1) v(0) = 1 and v is equivalent to 1 ⊕ v(1). (2) The conjugate of any v(t) is equivalent to itself that is v(t) ≃ v(t), ∀t ∈ N. (3) The fusion rules are the same as for SO(3): v(s) ⊗ v(t) ≃ 2 min(s,t)Mk=0 v(s+t−k) We denote by χk =Pdk i=1 v(k) ii We will need the following proposition, proved in [7]: the character associated to v(k). 8 FRANC¸ OIS LEMEUX Proposition 1.15. Let χ be the character associated to the fundamental corep- resentation v of C(S+ N ). Then, χ∗ = χ and there is a ∗-isomorphism C∗(χ) = C(S+ N )0 = C∗(χt : t ∈ N) ≃ C([0, N ]) identifying χt to the polynomial defined Πt by Π0 = 1, Π1 = X − 1 and ∀t ≥ 1, Π1Πt = Πt+1 + Πt + Πt−1. Remark 1.16. (1) The recursion formula defining the polynomials Πt is the (2) The polynomials At and Πt are linked by the formula: Πt(x) = A2t(√x). Before describing the fusion rules of C(H s+ N ), we recall that these compact quan- one satisfied by the irreducible characters χt. tum groups are free wreath products: Theorem 1.17. [4, Theorem 3.4] Let N ≥ 2, then we have the following isomor- phisms of compact quantum groups: N ) ≃ C(Zs)∗w C(S+ N )/ < [zi, vij] = 0 > where zi is the generator of the i-th copy Zs in the free product Z∗N N ) = C∗(Z∗N s )∗ C(S+ - C(H s+ . s - In particular C(H s+ 2 ) ≃ C(Zs) ∗w C(Z2), C(H s+ 3 ) ≃ C(Zs) ∗w C(S3). Let us now give the description of the irreducible corepresentations of C(H s+ N ). Theorem 1.18. [4, Theorem 4.3, Corollary 6.4] C(H s+ N ) has a unique family of N -dimensional corepresentations (called basic corepresentations) {Uk : k ∈ Z}, satisfying the following conditions: (1) Uk = (U k ij ) for any k > 0. (2) Uk = Uk+s for any k ∈ Z. (3) U k = U−k for any k ∈ Z. (4) U1, . . . , Us−1 are irreducible. (5) U0 = 1 ⊕ ρ0, ρ0 irreducible. (6) ρ0, U1, . . . , Us−1 are inequiva- lent corepresentations. Notation 1.19. We will denote the basic irreducible corepresentations of C(H s+ N ) by ρt, t ∈ {0, . . . , s − 1}, with ρt = Ut ∀t ∈ {1, . . . , s − 1} and ρ0 = U0 ⊖ 1 (where U0 = (Uij U∗ij)). The proof of the first three assertions follows from the definitions of corepresen- tations of compact quantum groups and of the definition of C(H s+ N ). The proof of the last three assertions is based upon Woronowicz's Tannaka-Krein duality (see [23]) and methods inspired by [2], [3] and [1]. Now, we can give the description of the fusion rules: Theorem 1.20. [4, Theorem 8.2] Let M be the monoid M = ha, z : zs = 1i with involution a∗ = a, z∗ = z−1, and the fusion rules obtained by recursion from the formulae (1.3) vazi ⊗ zjaw = vazi+jaw ⊕ δi+j,0 (v ⊗ w) Then the irreducible corepresentations rα of C(H s+ N ) can be indexed by the ele- ments α of the submonoid S generated by the elements azia, i = 0, . . . , s − 1, with involution and fusion rules above. Remark 1.21. (1) S is composed of elements aL1zJ1 . . . zJK−1aLK with - Ji, Li > 0 integers. - L1, LK odd integers and all the Li's, i ∈ {2, . . . , K − 1} even integers. - Except if K = 1, then LK is an even integer. HAAGERUP APPROXIMATION PROPERTY FOR QUANTUM REFLECTION GROUPS 9 (2) With this description, we can identify the basic corepresentations introduced above: the corepresentation ra2 is the corepresentation ρ0 = (Uij U∗ij) ⊖ 1 and for t 6= 0, razta is the corepresentation ρt = (U t ij). (3) In Proposition 2.1, we will use the suggestive notation vazi+jaw = (vazi ⊗ zjaw) ⊖ δi+j,0(v ⊗ w), which simply means that we have the relation (1.3) in the monoid S. (4) If α = aL1zJ1 . . . zJK−1aLK ∈ S, then the conjugate corepresentation of rα is indexed by α = aLK z−JK−1 . . . z−J1aL1 We end this subsection by the following proposition which summarizes the results above: Proposition 1.22. The canonical morphism π : C(H s+ corepresentations Ut, t ∈ Z onto the fundamental corepresentation v of C(S+ other words, it maps all ρt = razta, t 6= 0 onto v and ρ0 = ra2 onto v(1). N ) → C(S+ N ) maps all the N ) ; in 2. Characters of quantum reflection groups and quantum permutation groups As announced in the introduction, we find the images of the irreducible characters of C(H S+ N ) under the canonical morphism π : C(H s+ N ) → C(S+ N ). Proposition 2.1. Let χα be the character of an irreducible corepresentation rα of C(H s+ N )0 with C([0, N ]), the image of χα, say Pα, satisfies: N ). Write α = al1zj1 . . . zjk−1alk . Then, identifying C(S+ Pα(X 2) = π(χα)(X 2) = Ali(X). kYi=1 i=1 li us- ing the description of the fusion rules given by Theorem 1.20, the recursion formula satisfied by the Tchebyshev polynomials, Proposition 1.7 and Proposition 1.22. Proof. We shall prove this proposition by induction on the even integerPk Let HR(λ) be the following statement: π(χα)(X 2) =Qk al1zj1 . . . zjk−1alk such that 2 ≤Pi li ≤ λ. Let us begin by studying simple examples (and initializing the induction). Consider the element aza. Then, the irreducible corepresentation raza (written ρ1 in Notation 1.19) is sent by π onto v = 1 ⊕ v(1) by Proposition 1.22. Thus, in term of characters, we obtain by Proposition 1.15 i=1 Ali (X) for any α = i.e. π(χaza)(X) = 1 + (X − 1) = X = A1(X) Paza(X 2) = X 2 = A1(X)A1(X). Actually, this holds for all elements α = azja, j ∈ {1, . . . , s − 1} (since every irreducible corepresentation razj a is sent by π onto 1 ⊕ v(1), as is raza). Consider the element a2. Then, the irreducible corepresentation ra2 (written ρ0 in Notation 1.19) is sent by π onto v(1). Thus π(χa2 )(X) = X − 1. i.e. Pa2 (X 2) = X 2 − 1 = A2(X). 10 FRANC¸ OIS LEMEUX To prove HR(2) one has to show that π(χa2 )(X 2) = A2(X) and π(χazj a)(X 2) = i=1 Ali (X) for any β = al1 zj1 . . . zjk−1alk A1A1(X) for all j ∈ {1, . . . , s − 1}, what we have just done above. Now assume HR(λ) holds: π(χβ)(X 2) =Qk such that 2 ≤Pi li ≤ λ. We now show HR(λ + 2). Let α = aL1zJ1 . . . aLK , with Pi Li = λ + 2. In order to use HR(λ), we must "break" α using the fusion rules as in the examples above. Then, essentially, one has to distinguish the cases LK = 1, LK = 3 and LK ≥ 5 (in the case LK ≥ 5 we can "break α at aLK " but in the other cases we must use aLK−1 or aLK−2 if they exist, that is if there are enough factors aLi). So first, we deal with two special cases below, in order to have "enough" factors aL in α in the sequel. We use the fusion rules described in Theorem 1.20 (and the notations described after, see Remark 1.21). - If K = 1 i.e. LK = λ + 2, Ji = 0 ∀i, write: α = aλ+2 = (aλ ⊗ a2) ⊖ (aλ−1 ⊗ a) = (aλ ⊗ a2) ⊖ aλ ⊖ aλ−2. Then using the hypothesis of induction and Proposition 1.7, we get π(χα)(X 2) = AλA2(X) − Aλ(X) − Aλ−2(X) = AλA2(X) − (Aλ(X) + Aλ−2(X)) = AλA2(X) − Aλ−1A1(X) = Aλ+2(X). (Notice that if λ = 2 one has λ − 2 = 0 and a4 = (a2 ⊗ a2) ⊖ (a ⊗ a) = (a2 ⊗ a2) ⊖ a2 ⊖ 1 so that the result we want to prove then is still true.) - If K = 2, J := J1 6= 0, write α = aL1zJaL2. We have L1 + L2 = λ + 2 ≥ 4 and L1, L2 are odd hence L1 or L2 ≥ 3, say L1 ≥ 3. Write aL1zJ aL2 = (a2 ⊗ aL1−2zJ aL2) ⊖ (a ⊗ aL1−3zJ aL2). If L1 = 3 then the tensor product a ⊗ aL1−3zJ aL2 is equal to azJaL2 hence α = a3zJ aL2 satisfies π(χα)(X 2) = A2A1AL2(X) − A1AL2(X) = A3(X)AL2(X). If L1 > 3 (i.e. L1 ≥ 5), then the tensor product a ⊗ aL1−3zJ aL2 is equal to aL1−2zJaL2 ⊕ aL1−4zJaL2. We get π(χα)(X 2) = A2AL1−2AL2 (X) − AL1−2AL2 (X) − AL1−4AL2 (X) = AL1 (X)AL2(X). - From now on, we suppose that there are more than three factors aLi in α i.e. K ≥ 3. We will have to distinguish three cases: LK = 1, LK = 3 and LK ≥ 5. If 5 ≤ LK <Pi Li, write LK = mK + 2. Then we have mK ≥ 3, so aL1zJ1 . . . aLK = aL1zJ1 . . . amK +2 = (aL1 zJ1 . . . amK ⊗ a2) ⊖ (aL1zJ1 . . . amK−1 ⊗ a) = (aL1 zJ1 . . . amK ⊗ a2) ⊖ aL1zJ1 . . . amK ⊖ aL1zJ1 . . . amK−2. HAAGERUP APPROXIMATION PROPERTY FOR QUANTUM REFLECTION GROUPS 11 Then π(χα)(X 2) = AL1 . . . ALK−1Amk A2(X) − AL1 . . . AmK (X) − AL1 . . . AmK−2(X) = AL1 . . . ALK−1ALK (X). If mK = 1, i.e. LK = 3, we proceed in the same way using aL1zJ1 . . . zJK−1a3 = (aL1zJ1 . . . a ⊗ a2) ⊖ aL1zJ1 . . . zJK−1a. To conclude the induction, one has to deal with the case LK = 1. We have to distinguish the following cases: If LK−1 ≥ 4. We have aL1zJ1 . . . aLK−1zJK−1a = (aL1zJ1 . . . aLK−1−1 ⊗ azJK−1a) ⊖ (aL1zJ1 . . . aLK−1−2 ⊗ zJK−1a) = (aL1zJ1 . . . aLK−1−1 ⊗ azJK−1a) ⊖ aL1zJ1 . . . aLK−1−2zJK−1a. Then π(χα)(X 2) = AL1 . . . ALK−1−1A1A1(X) − AL1 . . . ALK−1−2A1(X) = AL1 . . . ALK−1A1(X). If LK−1 = 2 and JK−1 + JK−2 = 0 mod s, we can proceed in the same way using aL1zJ1 . . .aLK−2zJK−2a2zJK−1a = (aL1zJ1 . . . aLK−2zJK−2a ⊗ azJK−1a) ⊖ aL1zJ1 . . . aLK−2+1 ⊖ aL1zJ1 . . . aLK−2−1. The last case to deal with is LK−1 = 2 and JK−1 + JK−2 6= 0 mod s, and again we can conclude thanks to aL1zJ1 . . . zJK−2a2zJK−1a = (aL1 . . . aLK−2zJK−2a ⊗ azJK−1a) ⊖ aL1zJ1 . . . aLK−2zJK−2+JK−1a. As a corollary, we can get the result also proved in [4] (see Theorem 9.3): (cid:3) Corollary 2.2. Let rα be an irreducible corepresentation of C(H s+ al1zj1 . . . alk . Then N ) with α = dim(rα) = Ali(√N ). kYi=1 N ) ◦ π(χα) since π is a morphism of Proof. We have dim(rα) = ǫC(Hs+ Hopf algebras. But the counit on C(S+ N )0 is given by the evaluation in N . Indeed, an immediate corollary of Theorem 1.14 and Proposition 1.15, is ǫ(Πt) = Πt(N ) for all polynomials Πt, which form a basis of R[X]. Now by the previous proposition (cid:3) N )(χα) = ǫC(S+ i=1 Ali (√x), then ǫC(S+ i=1 Ali (√N ). π(χα)(x) =Qk N ) ◦ π(χα) =Qk 12 FRANC¸ OIS LEMEUX 3. Haagerup property for quantum reflection groups N ) = In this section we show that duals of the quantum reflection groups C(H s+ N ) on characters of C(S+ N ), s ≥ 1 have the Haagerup property for N ≥ 4. We still denote by π the canonical surjection π : C(H s+ N ) → C(S+ N )0 ≃ C([0, N ]) used to show that C(S+ C(Zs) ∗w C(S+ N ) and by ψx = evx the states on C(S+ N ) have the Haagerup property (see [7]). Essentially, we are going to use both morphisms π, ψx in this way: we can define states φx composing these maps, ψx ◦ π, where π sends characters of C(H s+ N ). Thus, we obtain states on the central algebra C(H s+ N )0 and, after checking that these states have some decreasing properties, we can use the Theorem 1.2 and conclude. Lemma 3.1. Let ψx, x ∈ [0, N ] be the states given by the evaluation in x on the central C∗-algebra C(S+ N )0. Then for all x ∈ [0, N ], φx = ψx ◦ π is a state on C(H s+ Proof. One just has to note that π is Hopf ∗-homomorphism and hence sends C(H s+ N )0. The rest is clear. (cid:3) N )0. Then ψx ◦ π is indeed a functional on C(H s+ N )0 to C(S+ N )0. Notation 3.2. We introduce a proper function on the monoid S (see Theorem i=1 li for α = al1zj1 . . . alkα . Notice that for 1.20). Let L be defined by L(α) =Pkα all R > 0 the set BR = nα = al1zj1 . . . alkα : L(α) =Pkα Thus we get that a net (fα)α∈S belongs to c0(S) ⇐⇒ ∀ǫ > 0 ∃R > 0 : ∀α ∈ S, (L(α) > R ⇒ fα < ǫ). We say that a net (fα)α converges to 0 as α → ∞ if (fα)α ∈ c0(S). Proposition 3.3. Let N ≥ 5 and let χα be an irreducible character of C(H s+ N ) associated to the irreducible corepresentation rα with α = al1zj1 al2 . . . alkα . Then for all x ∈ [0, N ] i=1 li ≤ Ro ⊂ S is finite. Cα(x) := φx(χα) dim(rα) = ψx ◦ π(χα)(X) dim(rα) = Ali(√x) Ali (√N ) . kαYi=1 Moreover Cα(x) converges to 0 as α → ∞ for all x ∈ [4, N ). Proof. Let α = al1zj1 . . . alkα . We obtain the first assertion using Proposition 2.1 and Corollary 2.2: π(χα)(X) = Al1 . . . Alkα (√X), Ali (√N ). dα := dim(rα) = kαYi=1 By Proposition 1.8, for any fixed x ∈ (4, N ), there exists a constant 0 < c < 1 such that Al(√x) √N(cid:19)cl Al(√N ) ≤(cid:18) √x kαYi=1 φx(χα) dim(rα) = Cα(x) = for all l ≥ 1. Then Ali(√x) N(cid:17) c Ali(√N ) ≤(cid:16) x 2 Pi li 2 L(α) N(cid:17) c =(cid:16) x 0. −→α→∞ (cid:3) HAAGERUP APPROXIMATION PROPERTY FOR QUANTUM REFLECTION GROUPS 13 Proposition 3.4. (Case N = 4) Let χα be an irreducible character of C(H s+ 4 ) associated to the irreducible corepresentation rα with α = al1zj1 al2 . . . alkα . Then for all x ∈ [0, 4] Cα(x) := φx(χα) dim(rα) = ψx ◦ π(χα)(X) dim(rα) = Ali(√x) Ali (2) . kαYi=1 Moreover Cα(x) converges to 0 as α → ∞ for all x ∈ (0, 4). Proof. The proof of the first assertion is similar to the one of the previous proposi- tion. For the second assertion, we use Proposition 1.10. We recall that we proved in that proposition that there exists a constant D < 1 such that for all x ∈ (0, 4) and all l ≥ 1 (3.1) Al(√x) Al(2) ≤ D. Let ǫ > 0 and x ∈ (0, 4). We want to prove that, (3.2) < ǫ for α large enough. Ali (√x) Ali (2) kαYi=1 By (3.1), there exists a K > 0 such that Qkα i=1 Ali (√x) Ali (2) < ǫ for all α ∈ S with < ǫ for Al(√x) Al(2) kα ≥ K. But by Proposition 1.10 there is also an L > 0 such that all l ≥ L, since this quotient converges to 0. exists i0 ∈ {1, . . . , kα} such that li0 ≥ L. In both case we can get (3.2) since Now let α = al1 zj1 . . . alkα ∈ S, with L(α) ≥ LK. Then either kα ≥ K, or there Ali(√x) Ali (2) ≤ (√x) (2) Ali0 Ali0 kαYi=1 , (cid:3) each factor of the product being less that one. Then we can prove the theorem: Theorem 3.5. The dual of H s+ N has the Haagerup property for all N ≥ 4. N . We prove that the dual of H s+ Proof. We follow the proof in [6] for O+ N has the Haagerup approximation property for all N ≥ 4 using both previous propositions. Consider the net (Tφx)x∈IN with IN = (4, N ) if N ≥ 5, IN = (0, 4) if N = 4 and Tφx = Xα∈Irr(Hs+ N ) φx(χα) dα pα N ), and their restrictions to L∞(H s+ The φx are states on C(H s+ N )0 so, by Theorem 1.2, the Tφx are a unital contrac- tions of L2(H s+ N ) are NUCP h-preserving maps. Moreover, Proposition 3.4 in the case N = 4 and Proposition 3.3 in the cases N ≥ 5, together with the fact that the pα are finite rank operators, show that for each x ∈ IN , the operator Tφx is compact. To conclude one has to show that for all x ∈ IN , (3.3) 0 Tφxa − aL2 −→x→N 14 FRANC¸ OIS LEMEUX N ) ֒→ L2(H s+ for all a ∈ L∞(H s+ N ) (via a ∈ L∞(H s+ N )). First let us prove that it is true for any element a ∈ P ol(H s+ N ) i.e. any linear combination of matrix coefficients U α N ) (by linearity, we can ij). Notice that if α = al1 zj1 . . . zjkα −1alkα then do that only for the elements U α α = alkα z−jkα−1 . . . z−j1 al1 = alkα zs−jkα −1 . . . zs−j1al1 . Thus by Proposition 2.1 φx(χα) = ψx ◦ π(χα) = ψx ◦ π(χα) = φx(χα). Hence, ij of irreducible corepresentations of C(H s+ TφxU α ij − U α ijL2 = U α ijL2 1 − kαYi=1 Ali (√x) Ali (√N )! , so let x → N and the assertion (3.3) holds for all these matrix coefficient. Now by L2-density of P ol(H s+ N ) and the fact that all Tφx, x ∈ IN , are unital contractions (and thus are uniformly bounded), we obtain that (3.3) is true for any a ∈ L2(H s+ N ). (cid:3) Remark 3.6. In [5], it is proved that there is a ∗-Hopf algebras isomorphism between C(H s+ 2 ) and C∗(Zs ∗ Zs × Z2) (see Example 2.5 and thereafter in that paper). Furthermore, the Haar state on C∗(Zs) ∗w C(S+ 2 ) is given by h = h1 ⊗ h2 where h2 is the Haar state on C(S+ 2 ) and h1 is the free product of the Haar states on C∗(Zs)). Then, it is clear that H s+ has the Haagerup property by the stability properties of the Haagerup property on groups (see e.g. [9]) 2 The algebra C(H s+ 3 ) is more complicated and does not reduce to a more com- prehensive tensor product as for the case N = 2. We are unable at the moment to prove that H s+ has the Haagerup property. 3 Acknowledgements I am very grateful to my advisors Uwe Franz and Roland Vergnioux for the time they spent discussing the arguments of this paper. I would also like to thank Pierre Fima for discussions on various topics on quantum groups, Amaury Freslon for discussions on averaging methods and approximation properties on quantum groups and Mikael de la Salle for very useful suggestions and commentaries on some special cases of the theorem proved in this paper. References 1. T. Banica, S. T. Belinschi, M. Capitaine, and B. Collins, Free Bessel laws, Canad. J. Math. 63 (2011), no. 1, 3 -- 37. MR 2779129 (2011m:46121) 2. Teodor Banica, Th´eorie des repr´esentations du groupe quantique compact libre O(n), C. R. Acad. Sci. Paris S´er. I Math. 322 (1996), no. 3, 241 -- 244. MR 1378260 (97a:46108) 3. , Symmetries of a generic coaction, Math. Ann. 314 (1999), no. 4, 763 -- 780. MR 1709109 (2001g:46146) 4. Teodor Banica and Roland Vergnioux, Fusion rules for quantum reflection groups, J. Non- commut. Geom. 3 (2009), no. 3, 327 -- 359. MR 2511633 (2010i:46109) 5. Julien Bichon, Free wreath product by the quantum permutation group, Algebr. Represent. Theory 7 (2004), no. 4, 343 -- 362. MR 2096666 (2005j:46043) 6. Michael Brannan, Approximation properties for free orthogonal and free unitary quantum groups, Journal fur die reine und angewandte Mathematik Volume 2012 (2012), 223 -- 251. 7. , Reduced operator algebras of trace-preserving quantum automorphism groups, preprint arXiv:1202.5020, 2012. 8. Indira Chatterji, Cornelia Drut¸u, and Fr´ed´eric Haglund, Kazhdan and Haagerup proper- ties from the median viewpoint, Adv. Math. 225 (2010), no. 2, 882 -- 921. MR 2671183 (2011g:20059) HAAGERUP APPROXIMATION PROPERTY FOR QUANTUM REFLECTION GROUPS 15 9. Cherix, Cowling, Jolissaint, Julg, and Valette, Groups with the Haagerup property, Progress in Mathematics, vol. 197, Birkhauser Verlag, Basel, 2001, Gromov's a-T-menability. 10. Yves Cornulier, Yves Stalder, and Alain Valette, Proper actions of wreath products and gen- eralizations, Trans. Amer. Math. Soc. 364 (2012), no. 6, 3159 -- 3184. MR 2888241 11. Pierre Fima, Kazhdan's property T for discrete quantum groups, Internat. J. Math. 21 (2010), no. 1, 47 -- 65. MR 2642986 (2011f:46089) 12. Amaury Freslon, Examples of weakly amenable discrete quantum groups, Journal of Functional Analysis (2013). 13. Uffe Haagerup, An example of a nonnuclear C ∗-algebra, which has the metric approximation property, Invent. Math. 50 (1978/79), no. 3, 279 -- 293. MR 520930 (80j:46094) 14. Nigel Higson and Gennadi Kasparov, E-theory and KK-theory for groups which act properly and isometrically on Hilbert space, Invent. Math. 144 (2001), no. 1, 23 -- 74. MR 1821144 (2002k:19005) 15. Johan Kustermans and Stefaan Vaes, Locally compact quantum groups, Ann. Sci. ´Ecole Norm. Sup. (4) 33 (2000), no. 6, 837 -- 934. MR 1832993 (2002f:46108) 16. Sorin Popa, On a class of type II1 factors with Betti numbers invariants, Ann. of Math. (2) 163 (2006), no. 3, 809 -- 899. MR 2215135 (2006k:46097) 17. Theodore J. Rivlin, Chebyshev polynomials, second ed., Pure and Applied Mathematics (New York), John Wiley & Sons Inc., New York, 1990, From approximation theory to algebra and number theory. MR 1060735 (92a:41016) 18. Stefaan Vaes and Roland Vergnioux, The boundary of universal discrete quantum groups, ex- actness, and factoriality, Duke Math. J. 140 (2007), no. 1, 35 -- 84. MR 2355067 (2010a:46166) 19. Shuzhou Wang, Free products of compact quantum groups,, Comm. Math. Phys. 167 (1995), no. 3, 671 -- 692. 20. , Quantum symmetry groups of finite spaces, Comm. Math. Phys. 195 (1998), no. 1, 195 -- 211. MR 1637425 (99h:58014) 21. Moritz Weber, On the classification of easy quantum groups, Advances in Mathematics 245 (2013), 500 -- 533. 22. S. L. Woronowicz, Compact matrix pseudogroups, Comm. Math. Phys. 111 (1987), no. 4, 613 -- 665. MR 901157 (88m:46079) 23. 24. , Tannaka-Kreın duality for compact matrix pseudogroups. Twisted SU(N ) groups, Invent. Math. 93 (1988), no. 1, 35 -- 76. MR 943923 (90e:22033) , Compact quantum groups, Sym´etries quantiques (Les Houches, 1995), North-Holland, Amsterdam, 1998, pp. 845 -- 884. MR 1616348 (99m:46164) Franc¸ois Lemeux, Laboratoire de math´ematiques de Besanc¸on, UFR Sciences et Tech- niques, Universit´e de Franche-Comt´e,16 route de Gray, 25000 Besanc¸on, France E-mail address: [email protected]
1503.07937	3	1503	2015-10-22T19:11:32	Quantum expanders and growth of group representations	[ "math.OA", "math.RT" ]	Let $\pi$ be a finite dimensional unitary representation of a group $G$ with a generating symmetric $n$-element set $S\subset G$. Fix $\vp>0$. Assume that the spectrum of $\|S\|^{-1}\sum_{s\in S} \pi(s) \otimes \overline{\pi(s)}$ is included in $ [-1, 1-\vp]$ (so there is a spectral gap $\ge \vp$). Let $r'_N(\pi)$ be the number of distinct irreducible representations of dimension $\le N$ that appear in $\pi$. Then let $R_{n,\vp}'(N)=\sup r'_N(\pi)$ where the supremum runs over all $\pi$ with ${n,\vp}$ fixed. We prove that there are positive constants $\delta_\vp$ and $c_\vp$ such that, for all sufficiently large integer $n$ (i.e. $n\ge n_0$ with $n_0$ depending on $\vp$) and for all $N\ge 1$, we have $\exp{\delta_\vp nN^2} \le R'_{n,\vp}(N)\le \exp{c_\vp nN^2}$. The same bounds hold if, in $r'_N(\pi)$, we count only the number of distinct irreducible representations of dimension exactly $= N$.	math.OA	math	Quantum expanders and growth of group representations by Gilles Pisier∗ Texas A&M University College Station, TX 77843, U. S. A. and Universit´e Paris VI Inst. Math. Jussieu, ´Equipe d'Analyse Fonctionnelle, Case 186, 75252 Paris Cedex 05, France September 1, 2018 5 1 0 2 t c O 2 2 ] . A O h t a m [ 3 v 7 3 9 7 0 . 3 0 5 1 : v i X r a Abstract Let π be a finite dimensional unitary representation of a group G with a generating symmetric n-element set S ⊂ G. Fix ε > 0. Assume that the spectrum of S−1Ps∈S π(s)⊗π(s) is included in [−1, 1 − ε] (so there is a spectral gap ≥ ε). Let r′ N (π) be the number of distinct irreducible representations of dimension ≤ N that appear in π. Then let R′ N (π) where the supremum runs over all π with n, ε fixed. We prove that there are positive constants δε and cε such that, for all sufficiently large integer n (i.e. n ≥ n0 with n0 depending on ε) and for all n,ε(N ) ≤ exp cεnN 2. The same bounds hold if, in r′ N ≥ 1, we have exp δεnN 2 ≤ R′ N (π), we count only the number of distinct irreducible representations of dimension exactly = N . n,ε(N ) = sup r′ 1 Introduction We wish to formulate and answer a natural extension of a question raised explicitly by Wigderson in several lectures (see e.g. [23, p.59]) and also implicitly in [18]. Although the variant that we answer seems to be much easier, it may shed some light on the original question. Wigderson's question concerns the growth of the number rN (G) of distinct irreducible representations of dimension ≤ N that may appear on a finite group G when the order of G is arbitrarily large and all that one knows is that G admits a generating set S of n elements for which the Cayley graph forms an expander with a fixed spectral gap ε > 0. The problem is to find the best bound of the form rN (G) ≤ R(N ) with R(N ) independent of the order of G (but depending on n, ε). We consider a more general framework: the finite group G is replaced by a finite dimensional representation π (playing the role of the regular representation λG for finite groups) such that the representation π ⊗ ¯π admits a spectral gap, meaning that the trivial representation is isolated with a gap ≥ ε from the other irreducible components of π ⊗ ¯π. When π = λG we recover the previous notion of spectral gap. Let then r′ N (π) be the number of distinct irreducible representations of dimension ≤ N appearing in π (note that rN (G) = r′ N (π) ≤ R′(N ) when the only restriction on π is that n, ε remain fixed (but the dimension of π is arbitrary). We observe that the previously known bound for R(N ) namely R(N ) = eO(nN 2) is also valid for R′(N ) N (λG)), and let R′(N ) denote the least upper bound r′ ∗Partially supported by ANR-2011-BS01-008-01. 1 and also that R(N ) ≤ R′(N ). Our main result, which follows from the metric entropy estimate for quantum expanders in [20], is that this bound for R′(N ) is sharp: there is δ > 0 such that for all n large enough (i.e. ∀n ≥ n0(ε)) we have R′(N ) ≥ eδnN 2 also [7, 8]). The term "quantum expander" was coined in [9, 2, 3] to which we refer for background (see for all N . 2 Main result Let G be any group with a finite generating set S ⊂ G with S = n. For any unitary representation π : G → Hπ we set λ(π, S) = n−1 sup{ℜ hXs∈S π(s)ξ, ξi ξ ∈ H inv π ⊥ ,kξkHπ = 1}. where H inv π ⊂ Hπ denotes the subspace of all π-invariant vectors. When S is symmetric,Ps∈S π(s) being selfadjoint, the real part sign ℜ can be omitted. We then set ε(π, S) = 1 − λ(π, S). It will be useful to record here the elementary observation that if π is unitarily equivalent to the direct sum ⊕i∈I πi of a family of unitary representations, then λ(π, S) = supi∈I λ(πi, S) and hence (2.1) ε(πi, S). ε(π, S) = inf i∈I In particular, if π1 is contained in π2, then ε(π1, S) ≥ ε(π2, S). We denote ε(G, S) = inf{ε(π, S)} where the infimum runs over all unitary representations π : G → Hπ. Thus the condition ε(G, S) > 0 means that G has Kazhdan's "property (T)", (or in otherwords is a Kazhdan-group), see [1] for more background. We start by the following result somewhat implicitly due to S. Wassermann [22] and explicitly proved in detail in [6]. Proposition 2.1 ([22, 6]). For any ε > 0 there is a constant cε such that for any n, any group G and any S ⊂ G with S = n such that ε(G, S) ≥ ε, the number rN (G) of distinct irreducible unitary representations σ : G → B(Hσ) with dim(Hσ) ≤ N is majorized as follows: (2.2) rN (G) ≤ exp (cεnN 2). Of course, here distinct means up to unitary equivalence. Remark 2.2. Note that it suffices to prove a bound of the same form for the number of distinct Indeed, if the latter irreducible unitary representations σ : G → B(Hσ) with dim(Hσ) = N . number is denoted by sN (G), we have rN (G) =PN d=1 sd(G), so that it suffices to have a bound of the form sd(G) ≤ exp (c′ See [14, 15] for some examples of estimates of the growth of rN (G). εnd2) to obtain (2.2). 2 We note that it was originally proved by Wang [21] that for any Kazhdan-group G this number rN (G) is finite for any N . There is an indication of proof of (2.2) in [22], and detailed proofs appear in [6] (see also [18]). We will prove a simple extension of this bound below. Recall that a sequence (Gk, Sk) of finite groups equipped with generating sets Sk ⊂ Gk such that sup k Sk < ∞, Gk → ∞ and inf k ε(Gk, Sk) > 0 is called an expander or an expanding family. This corresponds to the usual notion among Cayley graphs to which we restrict the entire discussion. Let G denote as usual the (finite) set of all irreducible unitary representations of a finite group G (up to unitary equivalence). We note in passing that it is well known (and this also can be derived from Proposition 2.1) that any expander satisfies (2.3) lim k→∞ max{dim(Hσ) σ ∈ Gk} = ∞. We refer the reader to the surveys [10, 17] for more information on expanders. The question raised by Wigderson in this context can be formulated as follows: Let Rn,ε(N ) = sup{rN (G)} where the supremum runs over all finite groups G admitting a subset S with S = n such that ε(G, S) ≥ ε. Actually the question is just as interesting for arbitrary (Kazhdan) groups G, but it is more natural to restrict to finite groups, because there are infinite Kazhdan groups without any (nontrivial) finite dimensional representations. Moreover, since, for a finite group G, all representations are weakly contained in the left regular representation λG, we have clearly by (2.1) (2.4) ε(G, S) = ε(λG, S). By (2.2), we have (2.5) Rn,ε(N ) ≤ exp (cεnN 2). and a fortiori simply Rn,ε(N ) = exp O(N 2). Wigderson asked whether this upper bound can be improved. More explicitly, what is the precise order of growth of log Rn,ε(N ) when N → ∞. Does it grow like N rather than like N 2 ? The motivation for this question can be summarized like this: In [18, Th. 1.4] an exponential bound exp O(N ) is proved for a special class of groups G (namely monomial groups), admitting a fixed spectral gap with generating sets of very slowly growing size (but not bounded) and it is asked whether the same exponential bound holds in general for Rn,ε(N ). Moreover, in a remark following the proof of [18, Th. 1.4], Meshulam and Wigderson observe that for any prime number p > 2, there is a group Gp with a generating set of (unbounded) size log p admitting a fixed spectral gap and such that rp(G) ≈ 2p/p. Remark 2.3. By classical results, originating in the works of Kazhdan and Margulis (see e.g. [16] or [17, Cor. 2.4]), for any fixed m ≥ 3, the family {SLm(Zp) p prime} is an expander, so that we have (for suitable ℓ, δ) Rℓ,δ(N ) ≥ supp rN (SLm(Zp)). 3 Similarly, let Gk denote the symmetric group of all permutations of a k element set. Kassabov [11] proved that the family {Gk k ≥ 1} forms an expanding family with respect to subsets Sk ⊂ Gk of a fixed size ℓ and a fixed spectral gap δ > 0. Thus we find a lower bound Rℓ,δ(N ) ≥ supk rN (Gk). Quite remarkably, it is proved in [13] that the family itself of all non-commutative finite simple groups forms an expander (for some suitable n, ε). Remark 2.4. However, it seems the resulting lower bounds are still far from being exponential in N . Actually, in many important cases (see e.g. [4]), the proof that certain finite groups G give rise to expanders uses the fact that the smallest dimension of a (non-trivial) irreducible representation on G is ≥ cGa for some a > 0. Then since G = Pπ∈ G dim(π)2 the cardinal of G is bounded above by G1−2a/c2. Therefore, for any N ≥ cGa we have rN (G) ≤ G1−2a/c2 ≤ c′N (1/a)−2, so that the resulting growth implied for Rn,ε(N ) is at most polynomial in N . (I am grateful to N. Ozawa for drawing my attention to this point). Nevertheless, we have: Remark 2.5. (Communicated by Martin Kassabov). For suitable n, ε the numbers Rn,ε(N ) grow faster than any power of N . In fact, we will prove the Claim : There is an expanding family of Cayley graphs (Gk) of groups generated by 3 elements with a positive spectral gap ε and such that for Nk = 23k − 2, Gk admits 2k2 distinct irreducible representations of dimension Nk. From this claim follows that R3,ε(Nk) ≥ 2k2 , say for all k large enough, and hence ≥ 2(log(Nk))2 Rn,ε(N ) ≥ 2(log(N ))2 for infinitely many N 's. 2 k of Rk = 2k2 To prove the claim we use the ideas from [12]. Let Rk denote the (finite) ring Mk(F2) of k × k matrices with entries in the field with 2 elements. It is known that the cartesian product Πk = R2k copies of Rk is generated by 3 Indeed, Rk itself is generated as a ring by two elements, e.g. a = e12 and the shift elements. b = e12 + e23 + ··· + ek−1k + ek1, then Πk is generated as a ring by {A, B, C} where A (resp. B) is the element with all coordinates equal to a (resp. b), and C is such that its coordinates are in one to one correspondence with the elements of Rk. To check this, let R ⊂ Πk be the ring generated by {A, B, C}. Note, by the choice of C, the following easy observation: for any coordinate i, there is x ∈ R such that xi = 0 but xj 6= 0 for all j 6= i. For any subset I of the index set let pI : R → RI k be the coordinate projection. One can then prove by induction on m = I that pI(R) = RI k for all I. Indeed, assume the fact established for m − 1. For any I with I = m we pick i ∈ I and we consider the set I = {y ∈ RI\i (0, y) ∈ pI (R)}. By the induction hypothesis, I is an ideal in RI\i k , but, since Rk is simple, the above observation implies that I = RI\i k , and since a, b generate Rk we have p{i}(R) = Rk, so we obtain pI (R) = RI k. This implies that the free associative ring Zhx, y, zi (in 3 non-commutative variables) can be mapped onto the product Πk. Consider now the group EL3(Zhx, y, zi) generated by the elementary matrices in GL3(Zhx, y, zi). This is a noncommutative universal lattice in the terminology of [12, 5]. First observe that EL3(Zhx, y, zi) is generated by 3 elements. Indeed, let α, β generate k SL3(Z). Then α, β, γ will generate EL3(Zhx, y, zi) where γ = 1 x y 0 1 z 0 0 1 EL3(Zhx, y, zi) has Kazhdan's property T. It follows that the groups . Moreover, by [5, Th.1.1] Gk = EL3(Πk) 4 have expanding generating sets with 3 elements. But it turns out that Gk can be identified with the product SL3k(F2)2k 2 . 2 2 k ) ≃ EL3(Rk)2k Indeed, firstly one easily checks the natural isomorphism EL3(R2k , secondly it is well known that, since F2 is a field, ELn(F2) = SLn(F2) for any n, and hence (taking n = 3k) we have a natural isomorphism EL3(Rk) = SL3k(F2); this yields the identification Gk = SL3k(F2)2k . To conclude, we will use the fact that SL3k(F2) admits a nontrivial irreducible representation π with dimension Nk = 23k − 2. (Just consider its action by permutation on the projective space, which has 23k − 1 elements; the action is transitive and doubly transitive, therefore the associated Koopman representation π is irreducible and of dimension 23k − 2). This immediately produces 2k2 distinct irreducible representations of dimension Nk on SL3k(F2)2k . Indeed, it is an elementary fact that if Γ = Γ1×···×Γm is a product group, and if π1,··· πm are arbitrary nontrivial irreducible representations on the factor groups Γ1,··· , Γm, then the representationseπj defined on Γ byeπj(g) = πj(gj ) are distinct (meaning not unitarily equivalent), irreducible on Γ and dim(eπj) = dim(πj) for any j. So taking all Γj's equal to SL3k(F2), with πj = π and m = 2k2 claim. , we obtain the announced 2 2 In any case, the problem of finding the correct behaviour of log Rn,ε(N ) (or of Rn,ε(N ) itself) when N → ∞ appears to be still wide open. and show that for this (much easier) modified version, N 2 is the correct order of growth. In this paper we consider a modified version of this question involving "quantum expanders" The term "quantum expander" was introduced in [9] and [2, 3], independently, to designate a sort of non-commutative, or matricial, analogue of expanders, as follows. Fix an integer n. Consider an n tuple of N × N unitary matrices, say u = (uj) ∈ U (N )n. We view each of them uj as a linear operator on the N -dimensional Hilbert space H. Then uj ⊗ uj is naturally viewed as a linear operator on the (Hilbert space sense) tensor product H ⊗ ¯H. Using the (canonical) identification H ∗ ≃ ¯H, the tensor product H ⊗ ¯H can be isometrically identified with the space of linear operators from H to H equipped with the Hilbert-Schmidt norm denoted by k k2 (sometimes called the Frobenius norm in the present finite dimensional context). Then, the identity operator IdH : H → H defines a distinguished element of H ⊗ ¯H that we denote by I. We set nX1 λ(u) = n−1 sup{ℜh( and uj ⊗ ¯uj)ξ, ξi ξ ∈ H ⊗ ¯H, ξ ⊥ I, kξkH⊗ ¯H = 1}, In other words, with the preceding identifications, the condition ε(u) ≥ ε means that for any x ∈ MN with tr(x) = 0 we have where kxk2 = (tr(x∗x))1/2. When T = Pn 1 uj ⊗ ¯uj is self adjoint (in particular when the set {u1,··· , un} is selfadjoint) the real part ℜ can be omitted in the two preceding lines. In group theoretic language, if π : Fn → U (N ) is the group representation on the free group Fn, equipped with a set of n free generators S = {g1,··· , gn}, such that π(gj) = uj (1 ≤ j ≤ n), then we have ε(u) = 1 − λ(u). ℜX tr(ujxu∗ j x∗) ≤ (1 − ε)kxk2, ε(u) = ε(π ⊗ π, S). 5 Definition 2.6. A sequence {u(k) k ∈ N} with each u(k) ∈ U (Nk)n such that Nk → ∞ (with n remaining fixed) and inf k{ε(u(k))} > 0 is called a quantum expander. We say that n is its degree and infk{ε(u(k))} > 0 its spectral gap. Remark 2.7. The existence of quantum expanders can be deduced as follows from that of expanders. Recalling (2.4), assume given a finite group G and S ⊂ G as before such that ε(G, S) = ε(λG, S) ≥ ε > 0. Recall that each σ ∈ G is contained in λG. Let π ∈ G. Since any representation on G without invariant vectors, being a direct sum of non trivial irreps, is weakly contained in λG, the ⊥ is weakly contained in the non trivial part of λG. In representation ρ = π ⊗ π restricted to H inv particular, we have by (2.1) ρ Therefore, we have λ(ρ, S) ≤ λ(λG, S). ε(π ⊗ π, S) ≥ ε(λG, S) ≥ ε. Thus if we are given an expander (Gk, Sk) as above, say with Sk = {s1(k),··· , sn(k)}, we can choose by (2.3) σk ∈ Gk such that dim(Hσk ) → ∞, and if we set uj(k) = σk(sj(k)) (1 ≤ j ≤ n), then u(k) = {u1(k),··· , un(k)} forms a quantum expander. The next statement is a simple generalization of Proposition 2.1 Proposition 2.8. For any 0 < ε < 1 there is a constant c′ ε > 0 for which the following holds. Let G be any group and let π : G → B(H) be any unitary representation on a finite dimensional Hilbert space H. Let us assume that there is an n-element subset S ⊂ G and ε > 0 such that ε(π ⊗ π, S) ≥ ε. In other words, π satisfies the following spectral gap condition: (2.6) λ(π ⊗ π, S) ≤ 1 − ε Let π = ⊕t∈T πt be the decomposition into distinct irreducibles (where each πt has multiplicity dt ≥ 1), then (2.7) {t ∈ T dim(πt) ≤ N} ≤ exp c′ εnN 2. Proof. Let σ = ⊕t∈T πt be the direct sum where each component is included with multiplicity equal to 1. We may clearly view σ as a subpresentation of π, acting on a subspace K ⊂ H so that the orthogonal projection Q : H → K is intertwining, i.e. satisfies Qπ = σQ. Then we also have (Q ⊗ ¯Q)(π ⊗ ¯π) = (σ ⊗ ¯σ)(Q ⊗ ¯Q), from which it is easy to derive that if we denote Vπ = H inv π⊗¯π, we have (Q ⊗ ¯Q)Vπ = Vσ and (Q ⊗ ¯Q)V ⊥ σ . This implies π = V ⊥ λ(σ ⊗ σ, S) ≤ λ(π ⊗ π, S) ≤ 1 − ε. Thus, replacing π by σ, we may as well assume that the multiplicities dt are all equal to 1. Let H = ⊕t∈T Ht denote the decomposition corresponding to π = ⊕t∈T πt. We have π ⊗ ¯π = ⊕t,r∈T πt ⊗ πr, with associated decomposition H ⊗ ¯H = ⊕t,r∈T Ht ⊗ Hr. From this follows that the subspace Vπ ⊂ H ⊗ ¯H of π ⊗ ¯π-invariant vectors is equal to ⊕t,r∈T Vt,r where Vt,r ⊂ Ht ⊗ Hr is the subspace of invariant vectors of πt ⊗ πr. Since for any t 6= r ∈ T , πt 6≃ πr, by Schur's lemma Vt,r = {0}, and hence Vπ ⊂ ⊕t∈T Vt,t. In particular, this shows that ∀t 6= r ∈ T Ht ⊗ Hr ⊂ V ⊥ π . 6 Let T ′ = {t ∈ T dim(πt) = N}. It suffices to show an estimate of the form (2.8) T ′ ≤ exp cεnN 2. Let H be the Hilbert space obtained by equipping M n N with the norm H = N −1n−1 kxk2 tr(x∗ j xj). nX1 Let S = {s1,··· , sn}. For any t ∈ T ′ we define x(t) ∈ M n N by x(t)j = πt(sj) 1 ≤ j ≤ n. Note that, by our normalization, kx(t)kH = 1 for any t ∈ T ′. Moreover, since for any t 6= r ∈ T πt 6≃ πr, by Schur's lemma the representation πt ⊗ πr has no invariant vector, and hence lies inside (π ⊗ π)V ⊥ π . Therefore, by (2.1) λ(πt ⊗ πr, S) ≤ λ(π ⊗ π, S), and hence for any unit vector ξ ∈ Hπt ⊗ Hπr we have n−1ℜ(Xs∈S (πt ⊗ πr)ξ, ξi) ≤ 1 − ε. In particular, if t 6= r ∈ T ′, we may realize πt, πr as representations on the same N -dimensional space, so that taking ξ = N −1/2I we find ℜhx(t), x(r)iH = (nN )−1ℜ Xs∈S which implies tr(πt(s)∗πr(s))! ≤ 1 − ε, √2ε. Thus we have T ′ points in the unit sphere of H that are √2ε-separated. Since dim(H) = nN 2, (2.8) follows immediately by a well known elementary volume argument (see e.g. [19, p. 57]). kx(t) − x(r)kH ≥ Remark 2.9. To derive Proposition 2.1 from the preceding statement, consider, in the situation of Proposition 2.1, a finite set {σt t ∈ T} of distinct finite dimensional irreducible representations of G, let π be their direct sum and let ρ = π ⊗ π. By the assumption in Proposition 2.1, we know εnN 2. Applying this to π = λG, this shows that ε(ρ, S) ≥ ε, and hence (2.7) implies T ≤ exp c′ Proposition 2.8 contains Proposition 2.1. For any finite dimensional unitary representation π : G → B(H) on an arbitrary group, let us N (π) the number of distinct irreducible representations appearing in the decomposition denote by r′ of π of dimension at most N . Let then R′ n,ε(N ) = sup r′ N (π) where the sup runs over all π's and G's admitting an n-element generating set S ⊂ G such that ε(π ⊗ ¯π, S) ≥ ε. 7 Note that r′ N (λG) = rN (G) and hence With this notation (2.7) means that Rn,ε(N ) ≤ R′ n,ε(N ). R′ n,ε(N ) ≤ exp c′ εnN 2. While it seems very difficult to give a good lower bound for Rn,ε(N ), we can answer the analogous n,ε(N ): Indeed, the main result of [20] (see [20, Th. 1.3]), which follows, implies the question for R′ desired lower bound when reformulated in terms of representations. Theorem 2.10 ([20]). For each 0 < ε < 1, there is a constant βε > 0 such that and for all sufficiently large integer n (i.e. n ≥ n0 with n0 depending on ε) and for all N ≥ 1, there is a subset T ⊂ U (N )n with T ≥ exp βεnN 2 such that ∀u 6= v ∈ T kXn 1 uj ⊗ vjk ≤ n(1 − ε) (we call these "ε − separated"), and ε(u) ≥ ε for all u ∈ T (we call these "ε-quantum expanders"). More precisely, for all u ∈ T we have k(X uj ⊗ uj)I ⊥k ≤ n(1 − ε). Theorem 2.11. The estimate in Proposition 2.8 is best possible in the sense that for any 0 < ε < 1 there is a constant βε > 0 such that for any n large enough (i.e. n ≥ n0(ε)), for any N ≥ 1 there is a group G and a finite dimensional representation π on G satisfying (2.6) and admitting a decomposition π = ⊕t∈T πt, with distinct irreducibles πt each with multiplicity 1 (or any specified value ≥ 1) and acting on an N -dimensional space, with T ≥ exp βεnN 2. Proof. Fix N > 1. Let T ⊂ U (N )n be the subset appearing in Theorem 2.10, i.e. T is such that T ≥ exp βεnN 2 and ∀t 6= r ∈ T we have (2.9) and also (2.10) kX tj ⊗ ¯rjk ≤ n(1 − ε), k(X tj ⊗ ¯tj)I ⊥k ≤ n(1 − ε). Let sj = ⊕t∈T tj ∈ U (m) with m = TN , and let G ⊂ U (m) be the subgroup generated by S = {s1,··· , sn}. Note that π(G) ⊂ ⊕t∈T MN . Let π : G → U (m) be the inclusion map viewed as a representation on G. Let Pt : ⊕t∈T MN → MN be the ∗-homomorphism corresponding to the projection onto the coordinate of index t. For any t ∈ T , let πt : G → U (N ) be the representation defined by πt = Pt(π). Then, by definition, we have π = ⊕t∈T πt. By the spectral gap condition (2.10) the commutant of πt(S) (which is but the commutant of {t1,··· , tn}) is reduced to the scalars, so πt is irreducible, and by (2.9) for any t 6= r ∈ T the representations πt and πr are not unitarily equivalent. Remark 2.12. In particular, this means that ∀n ≥ n0(ε) and ∀N n,ε(N ) ≥ exp βεnN 2. R′ Acknowledgement I am indebted to Martin Kassabov for letting me include Remark 2.5 and very grateful for his patient explanations on its contents. 8 References [1] B. Bekka, P. de la Harpe and A. Valette, Kazhdan's property (T). Cambridge University Press, Cambridge, 2008. [2] A. Ben-Aroya and A. Ta-Shma, Quantum expanders and the quantum entropy difference problem, 2007, arXiv:quant-ph/0702129. [3] A. Ben-Aroya, O. Schwartz, and A. Ta-Shma, Quantum expanders: motivation and construc- tions. Theory Comput. 6 (2010), 47-79. [4] J. Bourgain and A. Gamburd, Uniform expansion bounds for Cayley graphs of SL2(Fp), Ann. Math., 167 (2008), 625 -- 642. [5] M. Ershov and A. Jaikin-Zapirain, Property (T) for noncommutative universal lattices, In- vent. Math. 179 (2010), no. 2, 303 -- 347. [6] P. de la Harpe, A.G. Robertson and A. Valette, On the spectrum of the sum of generators for a finitely generated group. Israel J. Math. 81 (1993), 6596. [7] A. Harrow, Quantum expanders from any classical Cayley graph expander. Quantum Inf. Comput. 8 (2008), 715 -- 721. [8] M. Hastings and A. Harrow, Classical and quantum tensor product expanders. Quantum Inf. Comput. 9 (2009), 336 -- 360. [9] M. Hastings, Random unitaries give quantum expanders, Phys. Rev. A (3) 76 (2007), no. 3, 032315, 11 pp. [10] S. Hoory, N. Linial, and A. Wigderson, Expander graphs and their applications. Bull. Amer. Math. Soc. 43 (2006), 439-561. [11] M. Kassabov, Symmetric groups and expander graphs. Invent. Math. 170 (2007), no. 2, 327354. [12] M. Kassabov, Universal lattices and unbounded rank expanders, Invent. Math. 170 (2007), no. 2, 297 -- 326. [13] M. Kassabov, A. Lubotzky and N. Nikolov, Finite simple groups as expanders. Proc. Natl. Acad. Sci. USA 103 (2006), 6116 -- 6119. [14] M. Kassabov and N. Nikolov, Cartesian products as profinite completions. Int. Math. Res. Not. 2006, Art. ID 72947, 17 pp. [15] M. Larsen and A. Lubotzky, Representation growth of linear groups, J. Eur. Math. Soc. 10 (2008), 351 -- 390. [16] A. Lubotzky. Discrete groups, expanding graphs and invariant measures. Progress in Math. 125. Birkhauser, 1994. [17] A. Lubotzky. Expander graphs in pure and applied mathematics, Bull. Amer. Math. Soc. 49 (2012) 113 -- 162. 9 [18] R. Meshulam and A. Wigderson, Expanders in group algebras. Combinatorica 24 (2004), 659-680. [19] G. Pisier, The volume of Convex Bodies and Banach Space Geometry . (Book) Cambridge University Press.1989. [20] G. Pisier, Quantum Expanders and Geometry of Operator Spaces, J. Eur. Math. Soc. (JEMS) 16 (2014), 1183 -- 1219. [21] P. S. Wang, On isolated points in the dual spaces of locally compact groups. Math. Ann. 218 (1975), 19-34. [22] S. Wassermann, C ∗-algebras associated with groups with Kazhdan's property T. Ann. of Math. 134 (1991), 423-431. [23] A. Wigderson, lecture notes for the 22nd mcgill invitational workshop on computational complexity, Bellairs Institute Holetown, Barbados Lecturers: Ben Green and AviWigderson. 10
1005.4850	1	1005	2010-05-26T15:43:50	Lie Group-Lie Algebra Correspondences of Unitary Groups in Finite von Neumann Algebras	[ "math.OA", "math.FA" ]	We give an affirmative answer to the question whether there exist Lie algebras for suitable closed subgroups of the unitary group $U(\mathcal{H})$ in a Hilbert space $\mathcal{H}$ with $U(\mathcal{H})$ equipped with the strong operator topology. More precisely, for any strongly closed subgroup $G$ of the unitary group $U(\mathfrak{M})$ in a finite von Neumann algebra $\mathfrak{M}$, we show that the set of all generators of strongly continuous one-parameter subgroups of $G$ forms a complete topological Lie algebra with respect to the strong resolvent topology. We also characterize the algebra $\mathfrak{M}$ of all densely defined closed operators affiliated with $\mathfrak{M}$ from the viewpoint of a tensor category.	math.OA	math	Lie Group-Lie Algebra Correspondences of Unitary Groups in Finite von Neumann Algebras Research Institute for Mathematical Sciences, Kyoto University Hiroshi Ando Kyoto, 606-8502, Japan E-mail: [email protected] Yasumichi Matsuzawa1,2, ∗ 1 Mathematisches Institut, Universitat Leipzig Johannisgasse 26, 04103, Leipzig, Germany 2 Department of Mathematics, Hokkaido University Kita 10, Nishi 8, Kita-ku, Sapporo, 060-0810, Japan E-mail: [email protected] November 18, 2018 Abstract We give an affirmative answer to the question whether there exist Lie algebras for suitable closed subgroups of the unitary group U (H) in a Hilbert space H with U (H) equipped with the strong operator topology. More precisely, for any strongly closed subgroup G of the unitary group U (M) in a finite von Neumann algebra M, we show that the set of all generators of strongly continuous one-parameter subgroups of G forms a complete topological Lie algebra with respect to the strong resolvent topology. We also characterize the algebra M of all densely defined closed operators affiliated with M from the viewpoint of a tensor category. Keywords. finite von Neumann algebra, unitary group, affiliated operator, measurable operator, strong resolvent topology, tensor category, infinite dimen- sional Lie group, infinite dimensional Lie algebra. Mathematics Subject Classification (2000). 22E65, 46L51. ∗Supported by Research fellowships of the Japan Society for the Promotion of Science for Young Scientists (Grant No. 2100165000). 1 Contents 1 Introduction and Main Theorem 2 Preliminaries 2.1 von Neumann Algebras 2.2 Murray-von Neumann's Result 2.3 Converse of Murray-von Neumann's Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Topological Structures of M 3.1 Strong Resolvent Topology . . . . . . . . . . . . . . . . . . . . . 3.2 Strong Exponential Topology . . . . . . . . . . . . . . . . . . . . 3.3 τ -Measure Topology . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Almost Everywhere Convergence . . . . . . . . . . . . . . . . . . 3.5 Proof of Lemma 3.12 . . . . . . . . . . . . . . . . . . . . . . . . . 3.6 Direct Sums of Algebras of Unbounded Operators . . . . . . . . 3.7 Proof of Theorem 3.10 . . . . . . . . . . . . . . . . . . . . . . . . 3.8 Local Convexity . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Lie Group-Lie Algebra Correspondences 4.1 Existence of Lie Algebra . . . . . . . . . . . . . . . . . . . . . . . 4.2 Closed Subalgebras of M . . . . . . . . . . . . . . . . . . . . . . 5 Categorical Characterization of M 5.1 5.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . fvN and fRng as Tensor Categories . . . . . . . . . . . . . . . . A Direct Sums of Operators B Fundamental Results of SRT C Tensor Categories 2 5 5 7 14 15 16 19 20 21 22 25 30 30 32 32 35 37 37 38 49 51 52 1 Introduction and Main Theorem Lie groups played important roles in mathematics because of its close relations with the notion of symmetries. They appear in almost all branches of mathe- matics and have many applications. While Lie groups are usually understood as finite dimensional ones, many infinite dimensional symmetries appear in natural ways: for instance, loop groups C∞(S1, G) [18], current groups C∞ c (M, G) [1], diffeomorphism groups Diff ∞(M ) of manifolds [3] and Hilbert-Schmidt groups [5] are among well-known cases. They have been extensively investigated in several concrete ways. In this context, it would be meaningful to consider a general theory of infi- nite dimensional Lie groups. One of the most fundamental infinite dimensional groups are Banach-Lie groups. They are modeled on Banach spaces and many 2 theorems in finite dimensional cases are also applicable to them. Since it has been shown that a Banach-Lie group cannot act transitively and effectively on a compact manifold as a transformation group [17], however, Banach-Lie groups are not sufficient for treating infinite dimensional symmetries. After the birth of Banach-Lie group theory, more general notions of infinite dimensional Lie groups have been scrutinized to date: locally convex Lie groups [13], ILB-Lie groups [16], pro-Lie groups [6, 7] and so on. While there are many interesting and im- portant results about them, we note that not all theorems in finite dimensional cases remain valid in these categories and their treatments are complicated. For example, the exponential map might not be a local homeomorphism and the Baker-Campbell-Hausdorff formula may no longer be true [11]. We understand that the one of the most fundamental class of finite dimen- sional Lie groups are the unitary groups U (n) in such a sense that any compact Lie group can be realized as a closed subgroup of them. From this viewpoint, it would be important to study the infinite dimensional analogue of it; that is, we like to explicate the Lie theory for the unitary group U (H) of an infi- nite dimensional Hilbert space H. One of the most fundamental question is whether Lie(G) defined as the set of all generators of continuous one-parameter subgroups of a closed subgroup G of U (H) forms a Lie algebra or not. For the infinite dimensional Hilbert space H, there are at least two topologies on U (H), (a) the norm topology and (b) the strong operator topology. We discuss the above topologies separately. In the case (a), U (H) is a Banach-Lie group and for each closed subgroup the set Lie(G) forms a Lie algebra. But it is well known that there are not many "nice" continuous unitary representations of groups in H, and hence, U (H) with the norm topology is very narrow. On the other hand, U (H) with the strong operator topology (b) is important, because there are many "nice" continuous unitary representations of groups in H -- say, diffeomorphism groups of compact manifolds, etc. However, the answer is neg- ative to the question whether there exists a corresponding Lie algebra or not. Indeed, by the Stone theorem, the Lie algebra of U (H) coincides with the set of all (possibly unbounded) skew-adjoint operators on H, but we cannot define naturally a Lie algebra structure with addition and Lie bracket operations on it. This arises from the problem of the domains of unbounded operators. For two skew-adjoint operators A, B on H, dom(A + B) = dom(A) ∩ dom(B) is not always dense. Even worse, it can be {0} (see Remark 2.17). Because of this, the Lie theory for U (H) has not been successful, although the group itself is a very natural object. On the other hand, it is possible that even though the whole group U (H) does not have a Lie algebra, some suitable class of closed subgroups of it have ones. Indeed their Lie algebras Lie(G) are smaller than Lie(U (H)). We give an affirmative answer to the last question. Furthermore we prove that for a suitable subgroup G, Lie(G) is a complete topological Lie algebra with respect to some natural topology. We outline below the essence of our detailed discussions in the text. First, a group G to be studied in this paper is a closed subgroup of the unitary group U (M) of some finite von Neumann algebra M acting on a Hilbert space H. Clearly it is also a closed subgroup of U (H). The key proposition is 3 the following result of Murray-von Neumann (cf. Theorem 2.18): Theorem 1.1 (Murray-von Neumann). The set M of all densely defined closed operators affiliated with a finite von Neumann algebra M on H, M :=(cid:26)A ; A is a densely defined closed operator on H such that uAu∗ = A for all u ∈ U (M′). (cid:27) , constitutes a -algebra under the sum A + B, the scalar multiplication αA (α ∈ C), the product AB and the involution A∗, where X denotes the closure of a closable operator X. The inclusion G ⊂ U (M) implies Lie(G) ⊂ M and hence, for arbitrary two elements A, B ∈ Lie(G), the sum A + B, the scalar multiplication αA, the Lie bracket [A, B] := AB − BA are determined as elements of M. We can prove that they are again elements of Lie(G), which is not trivial. Therefore Lie(G) indeed forms a Lie algebra which is infinite dimensional in general. Thus if we do not introduce a topology, it is difficult to investigate it. Then, what is the natural topology on Lie(G)? Since Lie(G) is a Lie algebra, it should be a vector space topology. Furthermore, in view of the correspondences between Lie groups and Lie algebras it is natural to require the continuity of the mapping exp : Lie(G) ∋ A 7−→ eA ∈ G, where G is equipped with the strong operator topology and eA is defined by the spectral theorem. Under these assumptions, a necessary condition for a sequence {An}∞ n=1 ⊂ Lie(G) to converge to A ∈ Lie(G) is given by s- lim n→∞ etAn = etA, for all t ∈ R. This condition is equivalent to s- lim n→∞ (An + 1)−1 = (A + 1)−1. The latter convergence is well known in the field of (unbounded) operator the- ory as the convergence with respect to the strong resolvent topology. Therefore it seems natural to consider the strong resolvent topology for Lie(G). How- ever, there arises, unfortunately, another troublesome question as to whether the vector space operations and the Lie bracket operation are continuous with respect to the strong resolvent topology of Lie(G). For example, even if se- quences {An}∞ n=1 of skew-adjoint operators converge, respectively to skew-adjoint operators A, B with respect to the strong resolvent topology, the sequences {An + Bn}∞ n=1 are not guaranteed to converge to A + B (see Remark 3.3). We can solve this difficulty by applying the noncommutative integration theory and proving that the Lie algebraic operations are continuous with respect to the strong resolvent topology and that Lie(G) is complete as a uniform space. Hence Lie(G) forms a complete topological Lie algebra. Finally, let us remark one point: remarkably, Lie(G) is not locally convex in general. Most of infinite n=1, {Bn}∞ 4 dimensional Lie theories assume the local convexity explicitly, but as soon as we consider such groups as natural infinite dimensional analogues of classical Lie groups, there appear non-locally convex examples. We shall explain the contents of the paper. §2 is a preliminary section. We recall the basic facts about closed operators affiliated with a finite von Neumann algebra and explain the generalization of the Murray-von Neumann theorem for a non-countably decomposable case. In §3, we introduce three topologies on the set M of all densely defined closed operators affiliated with a finite von Neumann algebra M. The first topology originates from (unbounded) operator theory, the second one is Lie theoretical and the last one derives from the noncommutative integration theory. We discuss their topological properties and show that they do coincide on M. The main result of this section is Theorem 3.10 which states that M forms a complete topological -algebra with respect to the strong resolvent topology. In §4 constituting the main contents of the paper, we show that Lie(G) is a complete topological Lie algebra and discuss some aspects of it. The main result is given in Theorem 4.6. In §5, applying the results of §3, we consider the following problem: What kind of unbounded operator algebras can they be represented in the form of M? We give their characterization from the viewpoint of a tensor category. We show that R can be represented as M if and only if it is an object of the category fRng (cf. Definition 5.2). The main result is Theorem 5.5, which says that the category fRng is isomorphic to the category fvN of finite von Neumann algebras as a tensor category. In Appendix, we list up some fundamental definitions and results of the direct sums of operators, the strong resolvent convergence and the categories. 2 Preliminaries In this section we review some basic facts about operator algebras and un- bounded operators. For the details, see [19, 24]. See also Appendix A for the direct sums of operators. 2.1 von Neumann Algebras Let H be a Hilbert space with an inner product hξ, ηi, which is linear with respect to η. We denote the algebra of all bounded operators on H by B(H). Let M be a von Neumann algebra acting on H. The set M′ := {x ∈ B(H) ; xy = yx, for all y ∈ M} is called the commutant of M. The group of all unitary operators in M is denoted by U (M). The lattice of all projections in M is denoted by P (M). The orthogonal projection onto the closed subspace K ⊂ H is denoted by PK. For a projection p in M, we denote 1 − p as p⊥. Definition 2.1. Let M be a von Neumann algebra acting on a Hilbert space H. 5 (1) A von Neumann algebra with no non-unitary isometry is called finite. (2) A von Neumann algebra is called countably decomposable if it admits at most countably many non-zero orthogonal projections. (3) A subset D of H is called separating for M if xξ = 0, x ∈ M for all ξ ∈ D implies x = 0. It is known that a von Neumann algebra M acting on a Hilbert space H is countably decomposable if and only if there exists a countable separating subset of H for M. Definition 2.2. Let M be a von Neumann algebra. (1) A state τ on M is called tracial if for all x, y ∈ M, τ (xy) = τ (yx) holds. (2) A tracial state τ is called faithful if τ (x∗x) = 0 (x ∈ M) implies x = 0. (3) A tracial state τ is called normal if it is σ-weakly continuous. It is known that a von Neumann algebra is countably decomposable and finite if and only if there exists a faithful normal tracial state on it. For more informations about tracial states, see [24]. Let M be a von Neumann algebra and p ∈ M ∪ M′ be a projection. Define the set Mp of bounded operators on the Hilbert space ran(p) as (cid:8)pxran(p) ; x ∈ M(cid:9) , then Mp forms a von Neumann algebra acting on the Hilbert space ran(p) and (Mp)′ = (M′)p holds. If (M,H) and (N,K) are von Neumann algebras and if there exists a unitary operator U of H onto K such that U MU ∗ = N, then (M,H) and (N,K) are said to be spatially isomorphic. The map π of M onto N defined by in called a spatial isomorphism. The next Lemma is useful. ϕ(x) = U xU ∗, x ∈ M, Lemma 2.3. Let (M,H) be a finite von Neumann algebra. Then there exists a family of countably decomposable finite von Neumann algebras {(Mα,Hα)}α such that (M,H) is spatially isomorphic to the direct sum(cid:16)Lb α Mα,Lα Hα(cid:17). 6 A von Neumann algebra M is called atomic if each non-zero projection in M majorizes a non-zero minimal projection. It is known that a finite von Neumann algebra is atomic if and only if it is spatially isomorphic to the direct sum of finite dimensional von Neumann algebras Mn(C) (n ∈ N), where Mn(C) is the algebra of all n × n complex matrices. A von Neumann algebra with no non-zero minimal projection is called dif- fuse. It is known that every von Neumann algebra is spatially isomorphic to the direct sum of some atomic von Neumann algebra Matomic and diffuse von Neumann algebra Mdiffuse. These von Neumann algebras Matomic and Mdiffuse are unique up to spatial isomorphism. We call Matomic and Mdiffuse the atomic part and the diffuse part of M, respectively. 2.2 Murray-von Neumann's Result The domain of a linear operator T on H is written as dom(T ) and the range of it is written as ran(T ). If T is a closable operator, we write T for the closure of T . Definition 2.4. A densely defined closable operator T on H is said to be affiliated with a von Neumann algebra M if for any u ∈ U (M′), uT u∗ = T holds. If T is affiliated with M, so is T . The set of all densely defined closed operators affiliated with M is denoted by M. Each element in M is called a affiliated operator. Note that T is affiliated with M if and only if xT ⊂ T x for all x ∈ M′. Next, we define algebraic structures of unbounded operators in the style of Murray-von Neumann [12]. Let x1, y1, x2, y2,··· be (finite or countable infinite number of) indetermi- nants. A non-commutative monomial with indeterminants {xi, yi}i is a formal product z1z2 ··· zn, where all zk equal to xi or yi. If n = 0, we write this mono- mial as 1. A non-commutative polynomial p(x1, y1,··· ) is a formal sum of finite number of monomials. p(x1, y1,··· ) has the following form: Here, aρ ∈ C and we allow such a term as 0 · z1z2 ··· zn in this expression. If there is a term with coefficient 0, it cannot be omitted in the representation. Hence x1 is different from x1 + 0 · y1 as non-commutative polynomials . If there are two such terms as a · z1 ··· zn, b · z1 ··· zn, we identify the sum of them with the term (a + b) · z1 ··· zn. The sum, the scalar multiplication and the multiplication of non-commutative polynomials are defined naturally, where we do not ignore the terms with 0 coefficients. Once a non-commutative polynomial p(x1, y1,··· ) is given we obtain a new polynomial p(r)(x1, y1,··· ) by omitting terms with coefficient aρ = 0 in the 7 p(x1, y1,··· ) = qXρ=1 0 aρ · z(ρ) 1 ··· z(ρ) nρ (q = 1, 2,··· ), (q = 0). representation of p. We call p(r)(x1, y1,··· ) the reduced polynomial of p. We also define the adjoint element by x+ := xi. We also define the i conjugate polynomial of p by := yi, y+ i p(x1, y1,··· )+ := qXρ=1 0 aρ · (z(ρ) nρ )+ ··· (z(ρ) 1 )+ (q = 1, 2,··· ), (q = 0). Suppose there is a corresponding sequence {Xi}i of densely defined closed operators on H. For all i, we assume (xi, yi) corresponds to the pair of the closed operators (Xi, X ∗ 1 ,··· ) obtained by substituting each {xi, yi} in the representation of p(x1, y1,··· ) of the pairs (Xi, X ∗ 1 ,··· ) is defined according to the following rules: i ). In this case we define a new operator p(X1, X ∗ i ). More precisely, the domain of p(X1, X ∗ (1) dom(0) = dom(1) = H, 0ξ := 0, 1ξ := ξ, for all ξ ∈ H, (2) dom(aX) := dom(X), (aX)ξ := a(Xξ), for all ξ ∈ dom(aX), (3) dom(X + Y ) := dom(X) ∩ dom(Y ), (X + Y )ξ := Xξ + Y ξ, for all ξ ∈ dom(X + Y ), (4) dom(XY ) := {ξ ∈ dom(Y ) ; Y ξ ∈ dom(X)}, (XY )ξ := X(Y ξ), for all ξ ∈ dom(XY ), where X and Y are densely defined closed operators on H and a ∈ C. In general, M is not a -algebra under these operations. This is the reason for the difficulty of constructing Lie theory in infinite dimensions. However, Murray and von Neumann proved, in the pioneering paper [12], that for a finite von Neumann algebra M, M does constitute a -algebra of unbounded operators, which we will explain more precisely in the sequel. Murray-von Neumann proved the following results for a countably decompos- able case. Since we need to apply these results for a general finite von Neumann algebra case, we shall offer the generalization of their proofs. First of all, we recall the notion of complete density, which is important for later discussions. Definition 2.5. A subspace D ⊂ H is said to be completely dense for a finite von Neumann algebra M if there exists an increasing net {pα}α ⊂ P (M) of projections in M such that (1) pα ր 1 (strongly). (2) pαH ⊂ D for any α. It is clear that a completely dense subspace is dense in H. We often omit the phrase "for M" when the von Neumann algebra in consideration is obvious from the context. 8 Remark 2.6. In [12], Murray and von Neumann used the term "strongly dense". However, this terminology is somewhat confusing. Therefore we tenta- tively use the term "completely dense". Lemma 2.7. Let M be a countably decomposable, finite von Neumann algebra on a Hilbert space H, τ be a faithful normal tracial state on M. For a subspace D ⊂ H, the following are equivalent. (1) D is completely dense. (2) There exists an increasing sequence {pn}∞ pn ր 1 (strongly), ran(pn) ⊂ D. (3) For every ε > 0, there exists p ∈ P (M) such that τ (p⊥) < ε, pH ⊂ D. n=1 ⊂ P (M) such that Proof . It is clear that (2)⇒(1)⇒(3) holds. We shall prove (3)⇒(2). By as- n ) < 1/2n and sumption, for all n ∈ N, there exists pn ∈ P (M) such that τ (p⊥ pnH ⊂ D. Put qn := pk ∈ P (M). ∞^k=n Since qn ≤ qn+1, the strong limit s-limn→∞ qn =: q ∈ P (M) exists. It holds that τ (q⊥) = lim n→∞ τ (q⊥ n ) = lim n→∞ k! p⊥ τ ∞_k=n ∞Xk=n τ (p⊥ k ) ≤ lim n→∞ 1 2k = 0. ≤ lim n→∞ Therefore we have q = 1. ∞Xk=n Lemma 2.8. Let {(Mλ,Hλ)}λ∈Λ be a family of countably decomposable, finite von Neumann algebras. Let For each λ ∈ Λ, let Dλ ⊂ Hλ be a completely dense subspace for Mλ. Then M := bMλ∈Λ Mλ, H :=Mλ∈Λ cLλ∈ΛDλ ⊂ H is a completely dense subspace for M. Hλ. Proof . By Lemma 2.7, for each λ ∈ Λ, there exists an increasing sequence {pλ,n}∞ n=1 ⊂ P (Mλ) such that pλ,n ր 1 (strongly) and ran(pλ,n) ⊂ Dλ. For a finite set F ⊂ Λ, define pF,n := ⊕λp(λ) F,n, F,n :=(pλ,n p(λ) 0 (λ ∈ F ), (λ /∈ F ). 9 Then we have pF,n ∈ P (M) and {pF,n}(F,n) is an increasing net of projections. Here, we define (F, n) ≤ (F ′, n′) by F ⊂ F ′ and n ≤ n′. It is clear that pF,n ր 1 (strongly) and ran(pF,n) ⊂cLλ∈ΛDλ. Hence cLλ∈ΛDλ is completely dense. Remark 2.9. Lemma 2.7 does not hold if M is not countably decomposable. We will show a counterexample. Let H :=Mt∈R ℓ2(N), M := bMt∈R Mt, D := ℓ2(N) [Mt∈R Here, all Mt are isomorphic copies of some finite von Neumann algebra on ℓ2(N). By Lemma 2.8, D is completely dense for M. Suppose (2) of Lemma 2.7 holds. Then there exists pn ∈ P (M) such that ran(pn) ⊂ D and pn ր 1 (strongly). Represent pn as ⊕tpt,n (pt,n ∈ P (Mt)). Then we have Mt∈R ran(pt,n) = ran(pn) ⊂ D = ℓ2(N). [Mt∈R Therefore for each n ∈ N, there exists a finite set Fn ⊂ R such that pt,n = 0 n=1 Fn ⊂ R is at most countable, there exists some t0 /∈ F . Choose ξ(t0) ∈ ℓ2(N) to be a unit vector and ξ(t) := 0 (t 6= t0). Then for t /∈ Fn. Since F := S∞ for ξ =(cid:8)ξ(t)(cid:9)t∈R ∈ H, it follows that pnξ − ξ2 =Xt∈R pt,nξ(t) − ξ(t)2 = pt0,n{z}=0 = ξ(t0)2 = 1. ξ(t0) − ξ(t0)2 On the other hand, we have pnξ − ξ2 → 0, which is a contradiction. Proposition 2.10 (Murray-von Neumann [12]). Let M be a finite von Neu- mann algebra on a Hilbert space H. Let {Di}∞ i=1 ⊂ H be a sequence of com- Di is also completely pletely dense subspaces for M. Then the intersection ∞\i=1 dense. The proof requires some lemmata. Lemma 2.11. Proposition 2.10 holds if M is countably decomposable. Proof . Let τ be a faithful normal tracial state on M. By Lemma 2.7, for each i ) < ε/2i and piH ⊂ Di. ε > 0 and i ∈ N, there exists pi ∈ P (M) such that τ (p⊥ Put p := ∞^i=1 pi ∈ P (M). 10 Then we have p⊥ τ (p⊥) = τ ∞_i=1 ∞\i=1 i ! ≤ ∞Xi=1 ∞\i=1 Hence by Lemma 2.7, the intersectionT∞ (piH) ⊂ pH = Di. τ (p⊥ i ) ≤ ε 2i = ε, ∞Xi=1 i=1 Di is completely dense. Lemma 2.12. Let {(Mλ,Hλ)}λ∈Λ be a family of countably decomposable, finite von Neumann algebras. Put M := bMλ∈Λ Mλ, H :=Mλ∈Λ Hλ. Let D ⊂ H be a completely dense subspace for M. Then for each λ ∈ Λ, there exists some completely dense subspace Dλ ⊂ Hλ for Mλ such that [Mλ∈Λ Dλ ⊂ D. Proof . By the definition, there exists an increasing net {pα}α∈A ⊂ P (M) such that pα ր 1 (strongly) and ran(pα) ⊂ D. Let pα =: ⊕λpλ,α (pλ,α ∈ P (Mλ)). Then it holds that pλ,α ր 1 (strongly). Put Dλ := [α∈A ran(pλ,α) ⊂ Hλ. We see that Dλ is completely dense for Mλ. holds. It is clear that cLλ∈ΛDλ ⊂ D i=1 D′ subspaces Dλ,i ⊂ Hλ for Mλ such that D′ Proof of Proposition 2.10. Since M is finite, there exists a family of count- ably decomposable, finite von Neumann algebras {(Mλ,Hλ)}λ∈Λ and a unitary i := UDi. i is completely dense operator U : H →Lλ∈Λ Hλ such that U MU ∗ =Lλ∈Λ Mλ. Put D′ To prove the proposition, it suffices to prove thatT∞ for Lλ∈Λ Mλ. By Lemma 2.12, for each i ∈ N, there exist completely dense i ⊃cLλ∈ΛDλ,i. Then it follows that [Mλ∈Λ ∞\i=1 ∞\i=1 D′ i ⊃ By Lemma 2.11, T∞ i=1 Dλ,i) is completely dense forLλ∈Λ Mλ, which impliesT∞ 2.8,cLλ∈Λ (T∞ is also completely dense forLλ∈Λ Mλ. i=1 Dλ,i is completely dense for Mλ. Therefore by Lemma i=1 D′ ∞\i=1 [Mλ∈Λ Dλ,i! = Dλ,i! . i 11 Proposition 2.13 (Murray-von Neumann [12]). Let M be a finite von Neu- mann algebra. Then for each X ∈ M and a completely dense subspace D for M, the subspace {ξ ∈ dom(X) ; Xξ ∈ D} In particular, dom(X) is completely dense for all is also completely dense. X ∈ M. Proof . See [12]. Proposition 2.14 (Murray-von Neumann [12]). Let M be a finite von Neu- mann algebra. (1) Every closed symmetric operator in M is self-adjoint. (2) There are no proper closed extensions of operators in M. Namely, if X, Y ∈ M satisfy X ⊂ Y , then X = Y . (3) Let {Xi}i be a (finite or infinite) sequence in M. The intersection of domains DP := \p∈P dom(p(X1, X ∗ 1 , X2, X ∗ 2 ,··· )) of all unbounded operators obtained by substituting {Xi}i into the non- commutative polynomial p(x1, y1,··· ) is completely dense for M, where P is the set of all non-commutative polynomials with indefinite elements {xi, yi}i. Proof . See [12]. Remark 2.15. Murray-von Neumann proved (1) of Proposition 2.14 using Cay- ley transform, but there is a simpler proof. We record it here. Proof . Let A ∈ M be a symmetric operator. It is easy to see that A + i is injective. Let A + i = uA + i be its polar decomposition. From the injectivity, u∗u = Pker(A+i)⊥ = 1H. Since M is finite and uu∗ = Pran(A+i), we see that ran(A + i) = H. On the other hand, since A is closed and symmetric, ran(A + i) is closed. Therefore we obtain ran(A + i) = H. By the same way, it holds that ran(A − i) = H, which means A is a self-adjoint operator. Similarly, we see that for X ∈ M the injectivity of X is equivalent to the density of ran(X). Lemma 2.16 (Murray-von Neumann [12]). Let M be a finite von Neumann algebra and {Xi}i be a (finite or infinite) sequence in M. Let p(x1, y1, x2, y2,··· ), q(x1, y1, x2, y2,··· ), be non-commutative polynomials and p(X1, X ∗ tained by substituting (xi, yi) by (Xi, X ∗ i ). r(x1, y1, x2, y2,··· ) 2 ,··· ) be an operator ob- 1 , X2, X ∗ 12 (1) p(X1, X ∗ 1 , X2, X ∗ 2 ,··· ) is a densely defined closable operator on H, and p(X1, X ∗ 1 , X2, X ∗ 2 ,··· ) ∈ M. (2) If p(r)(x1, y1,··· ) = q(r)(x1, y1,··· ), then p(X1, X ∗ 1 ,··· ) = q(X1, X ∗ 1 ,··· ). Namely, the closure of the substitution of operators depends on a reduced polynomial only. (3) If p(x1, y1,··· )+ = q(x1, y1,··· ), then np(X1, X ∗ 1 ,··· )o∗ = q(X1, X ∗ 1 ,··· ). (4) If αp(x1, y1,··· ) = q(x1, y1,··· ) (α ∈ C), then α ·np(X1, X ∗ 1 ,··· )o = q(X1, X ∗ 1 ,··· ). p(X1, X ∗ (5) If p(x1, y1,··· ) + q(x1, y1,··· ) = r(x1, y1,··· ), then 1 ,··· ) = r(X1, X ∗ (6) If p(x1, y1,··· ) · q(x1, y1,··· ) = r(x1, y1,··· ), then 1 ,··· ) = r(X1, X ∗ 1 ,··· ) + q(X1, X ∗ 1 ,··· ) · q(X1, X ∗ p(X1, X ∗ 1 ,··· ). 1 ,··· ). Proof . See [12]. Remark 2.17. Lemma 2.16 (1) is not trivial. Indeed one can construct a pair of densely defined closed operators whose intersection of domains is {0}. See, e.g., [9, 20]. In summary, we have the following theorem. Theorem 2.18 (Murray-von Neumann [12]). For an arbitrary finite von Neu- mann algebra M, the set M forms a -algebra of unbounded operators, where the algebraic operations are defined by1 (X, Y ) 7→ X + Y , (X, Y ) 7→ XY , (α, X) 7→ αX, X 7→ X ∗. To conclude these preliminaries, we shall show a simple but useful lemma. 1 αX equals αX when α 6= 0. However, dom(0 · X) = H 6= dom(X). 13 Lemma 2.19. Let M be a finite von Neumann algebra, A be an operator in M. If D is a completely dense subspace of H contained in dom(A), then it is a core of A. That is, AD = A. Proof . From the complete density of D, there exists an increasing net of closed subspaces {Mα}α of H with Pα := PMα ∈ M such that D0 :=[α Mα ⊂ D is dense in H. Define A0 := AD0 . Take an arbitrary u ∈ U (M′). Let ξ ∈ D0 = dom(A0), so that there is some α such that ξ ∈ Mα. Then we have uA0ξ = uAξ = Auξ = AuPαξ = APαuξ = A0Pαuξ = A0uξ. Therefore uA0 ⊂ A0u holds. Since u ∈ U (M′) is arbitrary, we have uA0u∗ = A0. Taking the closure of both sides, we see that A0 = uA0u∗. This means A0 ∈ M. Therefore, it follows that A0 = AD ⊂ A = A Therefore by Proposition 2.14, we have A0 = A. 2.3 Converse of Murray-von Neumann's Result The converse of Theorem 2.18 is also true. We shall give a proof here. Lemma 2.20. Let M be a von Neumann algebra acting on a Hilbert space H. Assume that, for all A, B ∈ M, the domains dom(A + B) and dom(AB) are dense in H. Then A + B and AB are densely defined closable operators on H and the closures A + B and AB are affiliated with M for all A, B ∈ M. Proof . By the assumption, A + B is densely defined and (A + B)∗ ⊃ A∗ + B∗. Since the right hand side is densely defined, A + B is closable. As same as the above, we see that AB is closable. Affiliation property is easy. Remark 2.21. Let M be a von Neumann algebra. It is easy to check that αA (α ∈ C, A ∈ M) is always densely defined closable and its closure αA is affiliated with M. Moreover M is closed with respect to the involution A 7→ A∗. Theorem 2.22. Let M be a von Neumann algebra acting on a Hilbert space H. Assume that, for all A, B ∈ M, the domains dom(A + B) and dom(AB) are dense in H. If the set M forms a -algebra with respect to the sum A + B, the scalar multiplication αA (α ∈ C), the multiplication AB and the involution A∗, then M is a finite von Neumann algebra. 14 B := C := 1 2(cid:0)A + A∗(cid:1) , 1 2i(cid:0)A − A∗(cid:1) , Proof . Step 1. We first show that all closed symmetric operators affiliated with M are automatically self-adjoint. Let A be a closed symmetric operator affiliated with M. Define operators B ∈ M and C ∈ M as then B and C are self-adjoint and A = B + iC holds because M is a -algebra. Since A is symmetric, we see that C ⊃ 1 2i (A − A∗) ⊃ 1 2i (A − A) = 0dom(A). By taking the closure, we obtain C = 0. Hence A = B is self-adjoint. Step 2. We shall prove that M is finite. Let v be an arbitrary isometry in M. By the Wold decomposition, there exists a unique projection p ∈ M such that ran(p) reduces v, s := vran(p) ∈ Mp is a unilateral shift operator and u := vran(p⊥) ∈ Mp⊥ is unitary. It is easy to see that ker(1 − s) = {0}, ker(1 − s∗) = {0}, so that we can define the closed symmetric operator T on ran(p) as follows: We immediately see that T is affiliated with the von Neumann algebra Mp. T := i(1 + s)(1 − s)−1. Define the operator A on H = ran(p)L ran(p⊥) by A := T ⊕ 0ran(p⊥), then A is a closed symmetric operator and it is affiliated with M. From Step 1., A is self-adjoint, so that T is also self-adjoint. Since the Cayley transform of a self-adjoint operator is always unitary and the Cayley transform of T is s, s is unitary. This implies p = 0 because a unilateral shift operator admits no non-zero reducing closed subspace on which it is unitary. Hence v = u is unitary. 3 Topological Structures of M In this section we investigate topological properties of M. We need these results in the next section. We first endow M with two topologies, called the strong resolvent topology and the strong exponential topology. The former is (un- bounded) operator theoretic and the latter is Lie theoretic. To show that these two topologies do coincide and M forms a complete topological -algebra with respect to them, we introduce another topology, called the τ -measure topology which originates from the noncommutative integration theory. They seem quite different to each other, but in fact they also coincide. The main topic of the present section is to study correlations between them. 15 3.1 Strong Resolvent Topology First of all, we define the topology called the strong resolvent topology on the suitable subset of densely defined closed operators. Let H be a Hilbert space. We call a densely defined closed operator A on H belongs to the resolvent class RC (H) if A satisfies the following two conditions: (RC.1) there exist self-adjoint operators X and Y on H such that the intersec- tion dom(X) ∩ dom(Y ) is a core of X and Y , (RC.2) A = X + iY , A∗ = X − iY . Note that (RC.1) implies dom(X)∩ dom(Y ) is dense, so X + iY and X − iY are closable. Thus X + iY and X − iY are always defined. Furthermore, we have 1 2 (A + A∗) = 1 2 (X + iY + X − iY ) ⊃ Xdom(X)∩dom(Y ). Since A + A∗ is closable and by (RC.1), we get 1 2 A + A∗ ⊃ X. As X is self-adjoint, X has no non-trivial symmetric extension, we have 1 2 A + A∗ = X. Therefore, X is uniquely determined. As same as the above, Y is also unique and 1 2i A − A∗ = Y. We denote Re(A) := X = 1 2 A + A∗, Im(A) := Y = 1 2i A − A∗. Also note that bounded operators and (possibility unbounded) normal operators belong to RC (H). the weakest topology for which the following mappings Now we endow RC (H) with the strong resolvent topology (SRT for short), RC (H) ∋ A 7−→ {Re(A) − i}−1 ∈ (B(H), SOT ) and RC (H) ∋ A 7−→ {Im(A) − i}−1 ∈ (B(H), SOT ) are continuous. Thus a net {Aα}α in RC (H) converges to A ∈ RC (H) with respect to the strong resolvent topology if and only if {Re(Aα) − i}−1ξ → {Re(A) − i}−1ξ, {Im(Aα) − i}−1ξ → {Im(A) − i}−1ξ, 16 for each ξ ∈ H. This topology is well-studied in the field of unbounded operator theory and suitable for the operator theoretical study. We denote the system of open sets of the strong resolvent topology by OSRT. Let M be a finite von Neumann algebra on a Hilbert space H. We shall show that M is a closed subset of the resolvent class RC (H). This fact follows from Proposition 2.10, Theorem 2.18, Lemma 2.19 and the following lemmata. Lemma 3.1. Let M be a finite von Neumann algebra on a Hilbert apace H, A be in M. Then there exist unique self-adjoint operators B and C in M such that Proof . Put A = B + iC. B := 1 2 A + A∗, C := 1 2i A − A∗. Applying Proposition 2.10, dom(B) and dom(C) are dense in H. Hence B and C are closed symmetric operators affiliated with M. By Proposition 2.14, in fact, B and C are self-adjoint. As M is a -algebra, we have A = B + iC. Lemma 3.2. Let M be a finite von Neumann algebra. Then M is closed with respect to the strong resolvent topology. Proof . Let {Aα}α ⊂ M be a net converging to A ∈ RC (H) with respect to the strong resolvent topology. Then, for all u ∈ U (M′), we have {uRe(A)u∗ − i}−1 = u{Re(A) − i}−1u∗ = s- lim α u{Re(Aα) − i}−1u∗ α {uRe(Aα)u∗ − i}−1 = s- lim α {Re(Aα) − i}−1 = s- lim = {Re(A) − i}−1. This implies Re(A) belongs to M. As same as the above, we obtain Im(A) ∈ M. Thus so is A = Re(A) + Im(A). Remark 3.3. In general, the strong resolvent topology is not linear. Indeed, there exists sequences {An}∞ n=1 of self-adjoint operators and self- adjoint operators A, B such that the following conditions hold: n=1, {Bn}∞ n=1 converge to A and B in the strong resolvent n=1 and {Bn}∞ (1) {An}∞ topology, respectively. (2) An + Bn is essentially self-adjoint for each n ∈ N. (3) A + B is essentially self-adjoint. (4) (cid:8)An + Bn(cid:9)∞ resolvent topology, but C 6= A + B. n=1 converges to some self-adjoint operator C in the strong 17 For the details, see [22]. However, as we see in the sequel, the strong resolvent topology is linear on M. The next lemma is important in our discussion. Lemma 3.4. Let M be a finite von Neumann algebra acting on a Hilbert space H. Then the following are equivalent: (1) M is countably decomposable, (2) (M, SRT ) is metrizable as a topological space, (3) (M, SRT ) satisfies the first countability axiom. Proof . (1) ⇒ (2). Let {ξk}k be a countable separating family of unit vectors in H for M. For each A, B ∈ M, we define d(A, B) :=Xk 1 2k k{Re(A) − i}−1ξk − {Re(B) − i}−1ξkk +Xk 1 2k k{Im(A) − i}−1ξk − {Im(B) − i}−1ξkk. It is easy to see that the above d is a distance function on the space M, and the topology induced by the distance function d coincide with the strong resolvent topology on M. (2) ⇒ (3) is trivial. (3) ⇒ (1). Let S ⊂ P (M) be a family of mutually orthogonal nonzero projections in M. Since (M, SRT ) satisfies the first countability axiom, the origin 0 ∈ M has a countable fundamental system of neighborhoods {Vk}k. Put Sk := {p ∈ S ; p /∈ Vk}, then S = Sk Sk. This follows from the Hausdorff property of the strong re- solvent topology. Next we show that each Sk is a finite set. Suppose Sk is an infinite set, then we can take a countably infinite subset {pn ; n ∈ N} of Sk. Define For every ξ ∈ H we see that kpnξk = k nXi=1 nXi=1 ≤ k −→ 0. p := s- lim N→∞ pn. NXn=1 pnξ − n−1Xi=1 pnξk pnξ − pξk + kpξ − n−1Xi=1 pnξk 18 Thus pn converges strongly to 0. By Lemma B.1, this implies pn converges to 0 with respect to the strong resolvent topology. Hence there exists a number n ∈ N such that pn ∈ Vk. This is a contradiction to pn ∈ Sk. Therefore Sk is a countable. finite set. From the above arguments, we conclude that S =Sk Sk is at most Remark 3.5. As we see in the sequel, (M, SRT ) is a Hausdorff topological lin- ear space. Thus in the case that M satisfies conditions (1), (2) or (3) of Lemma 3.4, (M, SRT ) is metrizable with a translation invariant distance function. In particular, it is also metrizable as a uniform space. Finally, we state one lemma. Lemma 3.6. Let M be a finite von Neumann algebra acting on a Hilbert space H. Then the strong resolvent topology and the strong operator topology coincide on the closed unit ball M1. Proof . Note that if a von Neumann algebra is finite, then the involution is strongly continuous on the closed unit ball. The lemma follows immediately from this fact, Lemma B.1 and Lemma B.5. See Appendix B for more informations of the strong resolvent topology. 3.2 Strong Exponential Topology Next we introduce a Lie theoretic topology on M. Let H be a Hilbert space. For each A ∈ RC (H), each SOT-neighborhood V at 1 ∈ B(H) and each compact set K of R, we define W (A; V, K) the subset of RC (H) by W (A; V, K) :=(cid:26)B ∈ RC (H) ; e−itRe(A)eitRe(B) ∈ V, e−itIm(A)eitIm(B) ∈ V, ∀t ∈ K. (cid:27) , then {W (A; V, K)}A,V,K is a fundamental system of neighborhoods on RC (H). We denote the system of open sets of the topology induced by this fundamental system of neighborhoods by OSET, and call this topology the strong exponential topology (SET for short). Note that a net {Aλ}λ∈Λ in RC (H) converges to A ∈ RC (H) in the strong exponential topology if and only if eitRe(Aλ)ξ −→ eitRe(A)ξ, eitIm(Aλ)ξ −→ eitIm(A)ξ, for each ξ ∈ H, uniformly for t in any finite interval. This topology is important from the viewpoint of Lie theory. Indeed it can be defined by the unitary group U (H) only. Before stating the main theorem in this section, we study relations between the strong resolvent topology and the strong exponential topology. Lemma 3.7. Let M be a countably decomposable finite von Neumann algebra acting on a Hilbert space H. Then (M, SET ) is metrizable as a topological space. 19 Proof . Let {ξn}n be a countable separating family of unit vectors in H for M. For each A, B ∈ M we define 1 ∞Xm=1 d(A, B) :=Xn ∞Xm=1 +Xn 2n+m sup t∈[−m,m]keitRe(A)ξn − eitRe(B)ξnk 1 2n+m sup t∈[−m,m]keitIm(A)ξn − eitIm(B)ξnk. It is easy to see that the above d is a distance function on the space M, and the topology induced by the distance function d coincide with the strong exponential topology on M. Lemma 3.8. Let M be a countably decomposable finite von Neumann algebra. Then the strong resolvent topology and the strong exponential topology coincide on M. Proof . This follows immediately from Lemma 3.4, Lemma 3.7 and Lemma B.2. Remark 3.9. Similar to the above argument, one can prove that the strong resolvent topology and the strong exponential topology coincide on RC (H) if the Hilbert space H is separable. But the authors do not know whether this is true or not if H is not separable. However we can show the following theorem. The next is the main theorem in this section. Theorem 3.10. Let M be a finite von Neumann algebra acting on a Hilbert space H. Then M is a complete topological -algebra with respect to the strong resolvent topology. Moreover the strong resolvent topology and the strong expo- nential topology coincide on M. Throughout this section, we prove the above theorem. 3.3 τ -Measure Topology We first prove Theorem 3.10 in a countably decomposable von Neumann algebra case. In this case, we can use the nonmmutative integration theory thanks to a faithful normal tracial state. We shall introduce the τ -measure topology. Let M be a countably decomposable finite von Neumann algebra acting on a Hilbert space H. Fix a faithful normal tracial state τ on M. The τ -measure topology (MT for short) on M is the linear topology whose fundamental system of neighborhoods at 0 is given by N (ε, δ) :=(cid:26)A ∈ M ; there exists a projection p ∈ M such that kApk < ε, τ (p⊥) < δ (cid:27) , where ε and δ run over all strictly positive real numbers. It is known that M is a complete topological -algebra with respect to this topology [15]. We denote the system of open sets with respect to the τ -measure topology by Oτ . Note that the τ -measure topology satisfies the first countability axiom. 20 Remark 3.11. In this context, the operators in M are sometimes called τ - measurable operators [4]. Thus there are two topologies on M, the strong resolvent topology and the It seems that these two topologies are quite different. τ -measure topology. However, in fact, they coincide on M, i.e., Lemma 3.12. Let M be a countably decomposable finite von Neumann algebra acting on a Hilbert space H. Then the strong resolvent topology and the τ - measure topology coincide on M. In particular, M forms a complete topological -algebra with respect to the strong resolvent topology. Moreover the τ -measure topology is independent of the choice of a faithful normal tracial state τ . This lemma is the first step to our goal. 3.4 Almost Everywhere Convergence To prove Lemma 3.12, we define almost everywhere convergence. Let M be a countably decomposable finite von Neumann algebra on a Hilbert space H. Definition 3.13. A sequence {An}∞ n=1 ⊂ M converges almost everywhere (with respect to M) to A ∈ M if there exists a completely dense subspace D such that (i) D ⊂T∞ n=1 dom(An) ∩ dom(A), (ii) Anξ converges to Aξ for each ξ ∈ D. We shall investigate the relations between the almost everywhere conver- gence and the other topologies. Lemma 3.14. Let {An}∞ to A in the τ -measure topology, then there exists a subsequence {Ank}∞ {An}∞ Proof . For all j ∈ N, we can take nj ∈ N and pj ∈ P (M) which satisfy the following conditions: n=1 ⊂ M be a sequence, A ∈ M. Suppose An converges k=1 of n=1 such that Ank converges almost everywhere to A. Put p :=W∞ hand, j ) < 1/2j, nj < nj+1. τ (p⊥ k(Anj − A)pjk < 1/2j, k=l pk ∈ P (M), then ran(p) =S∞ l=1V∞ k! ≤ lim τ (p⊥) = lim l→∞ p⊥ l=1T∞ ∞Xk=l l→∞ τ (p⊥ k ) τ ∞_k=l ∞Xk=l ≤ lim l→∞ 1 2k = 0. k=l ran(pk). On the other 21 Therefore, H = ran(p) =S∞ l=1T∞ k=l ran(pk). This implies D0 := ∞[l=1 ∞\k=l ran(pk) is completely dense. Let D1 be the intersection of the domains of all non- commutative polynomials of operators {Ank , A, pk}∞ k=1, where we do not take closure for each non-commutative polynomial of operators. Then D1 is also completely dense and so is D := D0 ∩ D1. Take ξ ∈ D, then there exists k0 ∈ N k=k0 ran(pk). Consequently, for all k ≥ k0, we get such that ξ ∈T∞ ξ = pkξ, and pkξ ∈ dom(A) ∩ dom(Ak), ξ ∈ dom(A) ∩ dom(Ak), k(Ank − A)ξk = k(Ank − A)pjξk ≤ k(Anj − A)pjk · kξk ≤ 1 2k · kξk −→ 0. n=1 converges to A in the strong resolvent topology. n=1 be a sequence in M converging almost everywhere ∗}∞ n=1 also converges almost everywhere to A∗, then Thus Ank converges almost everywhere to A. Lemma 3.15. Let {An}∞ to A ∈ M. Suppose {An {An}∞ Proof . It is easy to check that Re(An) and Im(An) converge almost every where to Re(A) and Im(A), respectively. Applying Lemma B.1 and Lemma 2.19 to Re(An) and Im(An), we see that Re(An) and Im(An) converge to Re(A) and Im(A) in the strong resolvent topology, respectively. This implies {An}∞ converges to A in the strong resolvent topology. n=1 The following is well-known: Lemma 3.16. Let X be a metric space, {xn}∞ Suppose for each subsequence {xnk}∞ k=1 of {xn}∞ of {xnk}∞ k=1 which converges to x, then xn converges to x. n=1 ⊂ X be a sequence, x ∈ X. n=1 has a subsequence {xnkl}∞ l=1 3.5 Proof of Lemma 3.12 We shall start to prove Lemma 3.12. We prove that the system of open sets of the strong resolvent topology OSRT and the system of open sets of the τ -measure topology Oτ coincide on M. Let {An}∞ OSRT ⊂ Oτ : Suppose that {An}∞ ogy. Let {Ank}∞ 3.14, there exists a subsequence {Ankl}∞ n=1 converges to A in the τ -measure topol- n=1. By Proposition k=1 such that {Ankl}∞ l=1 k=1 be an arbitrary subsequence of {An}∞ n=1 ⊂ M be a sequence, A ∈ M. l=1 of {Ank}∞ 22 ∗}∞ l=1 converge almost everywhere to A and A∗, respectively. Apply- and {Ankl converges to A in the strong resolvent topology. This ing Lemma 3.15, Ankl implies An converges to A in the strong resolvent topology, by Lemma 3.16. Thus we get OSRT ⊂ Oτ . Oτ ⊂ OSRT: Suppose that {An}∞ n=1 converges to A with respect to the strong resolvent topology. First we consider the case that An and A are self-adjoint. 0 λdE(λ) be spectral resolutions of An and A, respectively. Fix an arbitrary positive number ε > 0. It is clear that s- limλ→∞ E([0, λ)) = 1, so there exists a positive number Λ > 0 such that τ (E([0, Λ))⊥) < ε, where we can take Λ > 0 which is not a point spectrum of A. Indeed, self-adjoint operators have at most countable point spectra, as M is countably decomposable. Next we define a continuous function φ on R as follows: 0 λdEn(λ) and A =:R ∞ Let An =:R ∞ φ(λ) :=  if λ ≤ −2Λ, 0 −λ − 2Λ if −2Λ ≤ λ ≤ −Λ, if −Λ ≤ λ ≤ Λ, λ −λ + 2Λ if Λ ≤ λ ≤ 2Λ, if 2Λ ≤ λ. 0 Let φ(An)− φ(A) =:R ∞ 0 λdFn(λ) be a spectral resolution of φ(An)− φ(A), e be a spectral measure of A. Note that E([0, Λ)) = e((−Λ, Λ)). For each ξ ∈ H, hξ, AE([0, Λ))ξi =Z(−Λ,Λ) =Z(−Λ,Λ) =ZR λdhξ, e(λ)ξi φ(λ)dhξ, e(λ)ξi φ(λ)dhξ, e(λ)E([0, Λ))ξi = hξ, φ(A)E([0, Λ))ξi. Thus we have AE([0, Λ))ξ = φ(A)E([0, Λ))ξ. Similar to the above argument, we get AnEn([0, Λ))ξ = φ(An)En([0, Λ))ξ. Therefore, for all ξ ∈ H, we see that k(An − A){En([0, Λ)) ∧ E([0, Λ)) ∧ Fn([0, ε))}ξk2 = k{φ(An) − φ(A)}{En([0, Λ)) ∧ E([0, Λ)) ∧ Fn([0, ε))}ξk2 = kφ(An) − φ(A){En([0, Λ)) ∧ E([0, Λ)) ∧ Fn([0, ε))}ξk2 =Z[0,ε) λ2dkFn(λ){En([0, Λ)) ∧ E([0, Λ)) ∧ Fn([0, ε))}ξk2 ≤ ε2kξk2. This implies k(An − A){En([0, Λ)) ∧ E([0, Λ)) ∧ Fn([0, ε))}k ≤ ε. 23 On the other hand, τ ({En([0, Λ)) ∧ E([0, Λ)) ∧ Fn([0, ε))}⊥) ≤ τ (En([0, Λ))⊥) + τ (E([0, Λ))⊥) + τ (Fn([0, ε))⊥) ≤ τ (En([0, Λ))⊥) + ε + τ (Fn([0, ε))⊥). By Lemma B.4, An converges to A in the strong resolvent topology, as the function is bounded continuous. By Lemma B.3, R ∋ λ 7−→ (λ − i)−1 ∈ C En([0, Λ)) = En((−1, Λ)) SOT−−−→ E((−1, Λ)) = E([0, Λ)). Thus for all sufficiently large number n ∈ N, τ (En([0, Λ))⊥) = τ (E([0, Λ))⊥) + τ (E([0, Λ)) − En([0, Λ))) ≤ ε + ε = 2ε. Furthermore, by Lemma B.4, φ(An) converges strongly to φ(A). We obtain that for each ξ ∈ H, kφ(An) − φ(A)ξk = k{φ(An) − φ(A)}ξk −→ 0. Applying Lemma B.1 and Lemma B.3 to φ(An) − φ(A), we see that Fn([0, ε)) = Fn((−1, ε)) SOT−−−→ 1. Hence, for all sufficiently large numbers n ∈ N, τ (Fn([0, ε))⊥) < ε. Thus, for all sufficiently large numbers n ∈ N, we have τ ({En([0, Λ)) ∧ E([0, Λ)) ∧ Fn([0, ε))}⊥) ≤ 4ε. From the above argument, we conclude that An converges to A in the τ -measure topology. In a general case, self-adjoint operators Re(An) and Im(An) converge to Re(A) and Im(A) in the strong resolvent topology, respectively. By the above argument, we see that Re(An) and Im(An) converge to Re(A) and Im(A) in the τ -measure topology, respectively. Since the addition is continuous with respect to the τ -measure topology, An converges to A in the τ -measure topology. This implies Oτ ⊂ OSRT. Hence the proof of Lemma 3.12 is complete. Remark 3.17. We referred to the proof of Theorem 5.5 of the paper [23] to prove the inclusion Oτ ⊂ OSRT. 24 3.6 Direct Sums of Algebras of Unbounded Operators To prove Theorem 3.10 in a general case, we show some facts about the di- rect sums of unbounded operators. See Appendix A for the definition of the direct sums of unbounded operators. The next lemma follows immediately from Lemma A.3. Lemma 3.18. Let Hα be a Hilbert space, H be the direct sum Hilbert space of {Hα}α. For each α, we consider a net {Aα,λ}λ∈Λ of self-adjoint operators on Hα and self-adjoint operator Aα on Hα. Set and on the Hilbert space H. Aλ := ⊕αAα,λ, A := ⊕αAα, (1) Aλ converges to A in the strong resolvent topology if and only if each {Aα,λ}λ∈Λ converges to Aα in the strong resolvent topology. (2) Aλ converges to A in the strong exponential topology if and only if each {Aα,λ}λ∈Λ converges to Aα in the strong exponential topology. Proof . (1) By Lemma A.3, we have (Aλ − i)−1 = ⊕α(Aα,λ − i)−1, (A − i)−1 = ⊕α(Aα − i)−1. The necessary condition is trivial. On the other hand, it is easy to see that {(Aλ − i)−1}λ∈Λ converges to (A − i)−1 on ⊕αHα. Since cLαHα is dense in Lα Hα and {(Aλ − i)−1}λ∈Λ is uniformly bounded, the sufficient condition follows. (2) Similar to the proof (1). The next lemma is the key to prove Theorem 3.10. Lemma 3.19. Let Mα be a finite von Neumann algebra acting on Hα, and put Then Mα. Mα M := bMα M =Mα holds. The sum, the scalar multiplication, the multiplication and the involution are given by (⊕αAα) + (⊕αBα) = ⊕α(cid:0)Aα + Bα(cid:1) , λ (⊕αAα) = ⊕α(cid:0)λAα(cid:1) , (⊕αAα) (⊕αBα) = ⊕α(cid:0)AαBα(cid:1) , (⊕αAα)∗ = ⊕α (Aα ∗) . f or all λ ∈ C, 25 In addition, if each Mα is countably decomposable, then M is a complete topo- logical -algebra with respect to the strong resolvent topology, and the strong resolvent topology coincides with the strong exponential topology on M. Proof . we shall prove this lemma step by step. Step 1. We first show thatLα Mα ⊂ M. Indeed let ⊕αAα ∈Lα Mα. By Lemma A.1 and Lemma A.4, each unitary operator u ∈ U (M′) can be written as u = ⊕αuα, where uα ∈ U (M′ α). Thus we have u (⊕αAα) = ⊕α (uαAα) ⊂ ⊕α (Aαuα) = (⊕αAα) u. This implies ⊕αAα ∈ M. Step 2. We show that the converse inclusion M ⊂Lα Mα. For each β, we put qβ := ⊕α (δαβ1Hα) ∈ M′, where δαβ is the Kronecker delta and 1Hα is the identity operator on Hα. From Lemma A.1, qβ is a projection and ran(qβ ) =Mα (δαβHα) =: Hβ. Let A ∈ M be a self-adjoint operator. We would like to prove A ∈ Lα Mα. Since qβ ∈ M′, we have qβA ⊂ Aqβ for all β. This implies that each Hβ reduces A. We denote reduced part of A to Hβ by A Hβ is obviously self-adjoint. For each β, we consider natural unitary operator vβ : Hβ −→ Hβ. Then the operator Aβ := v∗ vβ is again self-adjoint. To prove A = ⊕αAα, we take an ) ⊂ dom(A), we see that arbitrary ξ ∈cLαdom(Aα). Since vαξ(α) ∈ dom(A Hα . A Hβ βA Hβ ξ = Xα finite sum vαξ(α) ∈ dom(A). Therefore we obtain Avαξ(α) αAvαξ(α)oα (⊕αAα) ξ =nv∗ = A Xα = Xα vαξ(α) = Aξ. finite sum finite sum Hence (⊕αAα)cLαdom(Aα) ⊂ A. By Lemma A.2, we have ⊕αAα ⊂ A. On the other hand, both of ⊕αAα and A are self-adjoint and self-adjoint operators have no non-trivial self-adjoint extension. These facts implies ⊕αAα = A. Next we 26 show that each Aα is in Mα. Taking arbitrary unitary operators uα ∈ U (M′ α) and putting u := ⊕αuα, then by Lemma A.4, u is a unitary operator in M′. Since A ∈ M, we see that (⊕αuα) (⊕αAα) = uA ⊂ Au = (⊕αAα) (⊕αuα) . Lemma A.2. We observe that Thus for all α, uαAα ⊂ Aαuα holds. This implies Aα ∈ Mα for all α. Hence Next we consider an arbitrary element A ∈ M. Putting B := Re(A), C := Im(A), then A = B + iC. Since B and C are self-adjoint, by the above argument, there exist operators Bα ∈ Mα and Cα ∈ Mα such that A ∈Lα Mα. B = ⊕αBα and C = ⊕αCα holds. Set D :=cLα (dom(Bα) ∩ dom(Cα)). Since dom(Bα) ∩ dom(Cα) is a core of Bα + iCα, D is a core of ⊕α(cid:0)Bα + iCα(cid:1) by so that A ⊃ ⊕α(cid:0)Bα + iCα(cid:1) follows. Now we use Step 1., then we see that ⊕α(cid:0)Bα + iCα(cid:1) ∈ M because Bα + iCα ∈ Mα for all α. Since M is a finite von Neumann algebra, Lemma 2.14 means that A = ⊕α(cid:0)Bα + iCα(cid:1) ∈Lα Mα. Hence M =Lα Mα follows. A = B + iC ⊃(cid:0)B + iC(cid:1)D =(cid:8)⊕α(cid:0)Bα + iCα(cid:1)(cid:9) D, Step 3. We shall show the formulae with respect to the sum, the scalar product, the product and the involution. The formulae with respect to the scalar product and the involution are trivial. We first prove the formula of the sum. Let ⊕αAα, ⊕αBα ∈ Lα Mα = M. Put D+ := cLα (dom(Aα) ∩ dom(Bα)). Since dom(Aα)∩ dom(Bα) is a core of Aα + Bα, D+ is a core of ⊕α(cid:0)Aα + Bα(cid:1). We observe that follows. Since both sides are elements in M, we have (⊕αAα) + (⊕αBα) ⊃(cid:8)⊕α(cid:0)Aα + Bα(cid:1)(cid:9)D+ , (⊕αAα) + (⊕αBα) ⊃ ⊕α(cid:0)Aα + Bα(cid:1) (⊕αAα) + (⊕αBα) = ⊕α(cid:0)Aα + Bα(cid:1) . Next, to show the formula of the product, we put D× := cLαdom(AαBα). Since dom(AαBα) is a core of AαBα, D× is a core of ⊕α(cid:0)AαBα(cid:1). We observe that so that so that follows. Since both sides are elements in M, we have (⊕αAα) (⊕αBα) ⊃(cid:8)⊕α(cid:0)AαBα(cid:1)(cid:9)D+ , (⊕αAα) (⊕αBα) ⊃ ⊕α(cid:0)AαBα(cid:1) (⊕αAα) (⊕αBα) = ⊕α(cid:0)AαBα(cid:1) . 27 Hence the proof of Step 3. is complete. In the sequel, we assume that each Mα is countably decomposable. Note that, by Lemma 3.12, each Mα is a complete topological -algebra with respect to the strong resolvent topology. Step 4. Let {Aλ}λ∈Λ be a net in M, A be an element of M. Corresponding to M = ⊕αMα, we can write them as follows: Aλ = ⊕αAα,λ, A = ⊕αAα, Aα,λ ∈ M, Aα ∈ M. We shall show that Aλ converges to A with respect to the strong resolvent topology if and only if each {Aα,λ}λ∈Λ converges to Aα with respect to the strong resolvent topology. From Step 3., we obtain Re(Aλ) = ⊕αRe(Aα,λ), Re(A) = ⊕αRe(Aα), Im(Aλ) = ⊕αIm(Aα,λ), Im(A) = ⊕αIm(Aα), so that, by Lemma 3.18, the above equivalence of convergence follows. By Step 3. and Step 4., we see that M forms a topological -algebra with respect to the strong resolvent topology. Next, to prove the completeness, we prepare some facts. Step 5. Fix an arbitrary α0 and let V (α0) be an arbitrary SOT-open set in α V (α). Then V is a SOT-open set in M. Indeed, since for any x = ⊕αxα ∈ V , we have xα0 ∈ V (α0), there exists a positive number ε > 0 and finitely many vectors ξ(α0) ∈ Hα0 such that Mα0 . Set V (α) := Mα (α 6= α0) and V :=Lb n\k=1ny ∈ Mα0 ; k(y − xα0 )ξ(α0) k < εo ⊂ V (α0). ∈ Hα0 ,· · ·, ξ(α0) n k 1 Set ξ(α) k := 0 (α 6= α0), then we get ξk ∈Lα Hα and x ∈ {y ∈ M ; k(y − x)ξkk < ε} ⊂ V. n\k=1 open set. k=1 {y ∈ M ; k(y − x)ξkk < ε} is a SOT-open set in M, V is a SOT- Since Tn at 0 ∈ Mα0 . Set W (α) := Mα (α 6= α0) and W := Lα W (α). Then W is a neighborhood at 0 ∈ M. Indeed, 0 ∈ W is trivial. On the other hand, since 0 ∈ Mα0 , there exists finitely many SOT-open sets V (α0) in Mα0 such Step 6. Fix an arbitrary α0. Let W (α0) be an arbitrary SRT-neighborhood , ···, V (α0) n 1 28 that 0 ∈ o ⊂ W (α0). k α V (α) n\k=1nA ∈ Mα0 ; {Re(A) − i}−1 , {Im(A) − i}−1 ∈ V (α0) := Mα (α 6= α0) and Vk :=Lb n\k=1nA ∈ M ; {Re(A) − i}−1 , {Im(A) − i}−1 ∈ Vko ⊂ W. n\k=1nA ∈ M ; {Re(A) − i}−1 , {Im(A) − i}−1 ∈ Vko 0 ∈ Put V (α) SOT-open set in M and k , then, by Step 5., each Vk is a k By the definition of the strong resolvent topology, is a open set, so that W is a SRT-neighborhood at 0 ∈ M. Cauchy net in M. For each λ ∈ Λ we can write as Aλ = ⊕αAα,λ ∈Lα Mα. Fix := Mα (α 6= α0) and W := Lα W (α), then, by Step 6., W is Aλ − Aµ ∈ W for all λ, µ ≥ λ0. Since Aλ − Aµ = ⊕α(cid:0)Aα,λ − Aα,µ(cid:1), this Step 7. We shall give a proof of the completeness of M. Let {Aλ}λ∈Λ be a an arbitrary α0 and let W (α0) be an arbitrary SRT-neighborhood at 0 ∈ Mα0 . Set W (α) a SRT-neighborhood at 0 ∈ M. Therefore there exists λ0 ∈ Λ such that implies that Aα0,λ − Aα0,µ ∈ W (α0) for all λ, µ ≥ λ0. Hence {Aα0,λ}λ∈Λ is a Cauchy net in Mα0 . We now use the completeness of Mα0 , then there exists an element Aα0 ∈ Mα0 such that Aα0,λ → Aα0. Since α0 is arbitrary, so that this means that for each α, there exists an element Aα ∈ Mα such that Aα,λ → Aα. Put A := ⊕αAα, then, by Step 4., we conclude that Aλ → A. Thus M is com- plete. Step 8. The strong resolvent topology coincides with the strong exponential topology on M. This fact follows from Lemma 3.8 and Lemma 3.18. Lemma 3.20. Let (M,H) and (N,K) be spatially isomorphic finite von Neu- mann algebras. If a unitary operator U of H onto K induces the spatial isomor- phism, then the map Φ : M → N, X 7→ U XU ∗ is a -isomorphism. Moreover Φ is a homeomorphism with respect to the strong resolvent topology and the strong exponential topology. Proof . It is easy to see that Φ(X) ∈ N for all X ∈ M. Thanks to Proposition 2.14 (2), it is not difficult to show that Φ is a unital -homomorphism: U(cid:0)X + Y(cid:1) U ∗ = U XU ∗ + U Y U ∗ U(cid:0)XY(cid:1) U ∗ = U XU ∗U Y U ∗ U X ∗U ∗ = (U XU ∗)∗. 29 Furthermore, it is straightforward to verify that Φ is invertible, the inverse of which is given by N ∋ Y 7→ U ∗Y U ∈ M. Topological property is trivial. 3.7 Proof of Theorem 3.10 We shall give a proof of Theorem 3.10. By Lemma 2.3, there exists a fam- ily of countably decomposable finite von Neumann algebras {Mα}α such that α Mα. From Lemma 3.20, there exists a M is spatially isomorphic onto Lb -isomorphism of M ontoLα Mα which is homeomorphic with respect to the Lα Mα is complete topological -algebra, so that so is M. Hence the proof of strong resolvent topology and the strong exponential topology. By Lemma 3.19, Theorem 3.10 is complete. 3.8 Local Convexity We study the local convexity of(cid:0)M, SRT(cid:1) here. Proposition 3.21. Let M be a finite von Neumann algebra. Then the following are equivalent. (1) (M, SRT ) is locally convex. (2) M is atomic. We need some lemmata to prove the above proposition. Lemma 3.22. Let M be an atomic finite von Neumann algebra, then(cid:0)M, SRT(cid:1) is locally convex. Proof . Every atomic finite von Neumann algebra is spatially isomorphic to the direct sum of matrix algebras {Mnλ(C)}λ∈Λ, where each Mnλ(C) is the algebra of all nλ × nλ complex matrices. Thus we should only prove this lemma in the λ∈Λ Mnλ(C). Note that Mnλ(C) =Mλ∈Λ Mnλ(C) =Mλ∈Λ x = ⊕λ∈Λxλ ∈ M =Mλ∈Λ Mnλ(C). Mnλ(C). case that M is equal toLb bMλ∈Λ M = Let pλ be a semi-norm on M defined by pλ(x) := kxλk, Then the strong resolvent topology on M coincides with the locally convex topology induced by the semi-norms {pλ}λ∈Λ because there is only one Haus- dorff linear topology on a finite dimensional linear space. Hence the proof is complete. Lemma 3.23. Let M be a diffuse finite von Neumann algebra, then there exists no non-zero SRT-continuous linear functional on M. 30 Proof . Suppose there exists a non-zero SRT-continuous linear functional f on M and we shall show a contradiction. Since, by Lemma 3.6, the restriction of f onto M is a σ-strongly continuous linear functional on M and M is SRT-dense in M, there exists a projection e0 in M such that f (e0) 6= 0. Step 1. For any orthogonal family of non-zero projections {en}∞ f (en) = 0 except at most finitely many n ∈ N. Indeed, put n=1 of M, A := ∞Xn=1 anen ∈ M, an :=(cid:26) 1 f (en) 0 if f (en) 6= 0, if f (en) = 0, where convergence of A is in the strong resolvent topology. Then we have f (A) = ∞Xn=1 anf (en) = Xf (en)6=0 1 < ∞, so that f (en) = 0 except at most finitely many n ∈ N. Step 2. For any e ∈ P (M) with f (e) 6= 0, there exists e′ ∈ P (M) such that 0 6= e′ ≤ e and f (e′) = 0. Indeed, since M is diffuse, there exists an orthogonal family of non-zero projections {en}∞ 2., J := {n ∈ N ; f (en) 6= 0} is a finite set. In particular, n=1 in M such that e =Pn≥1 en. By Step e′ := e −Xn∈J en 6= 0 satisfies f (e′) = 0. Step 3. We shall get a contradiction. By Step 2., we can take a maximal orthogonal family of non-zero projections {eα}α∈A in M such that eα ≤ e0 and e = e0. Thus we have f (eα) = 0. Let e :=Pα∈A eα. The maximality of {eα}α∈A and Step 2. implies 0 6= f (e0) = Xα∈A f (eα) = 0, which is a contradiction. Hence there exists no non-zero SRT-continuous linear functional on M. as and(cid:0)M(cid:1)∗ Lemma 3.24. Let Ma be an atomic finite von Neumann algebra, Md be a diffuse finite von Neumann algebra and M := MaLb Md be the direct sum von Neumann algebra. Denote the conjugate spaces of (cid:0)Ma, SRT(cid:1) and (cid:0)M, SRT(cid:1) by(cid:0)Ma(cid:1)∗ , we define I(f ) ∈(cid:0)M(cid:1)∗ then I is a bijection between(cid:0)Ma(cid:1)∗ A ⊕ B ∈ M = MaM Md, onto(cid:0)M(cid:1)∗ respectively. For each f ∈(cid:0)Ma(cid:1)∗ I(f )(A ⊕ B) := f (A), . 31 Proof . This follows immediately from Lemma 3.23. Proof of Proposition 3.21. We have only to prove (1)⇒(2). Since M is spatially isomorphic to the direct sum of an atomic von Neumann algebra Matomic and a diffuse von Neumann algebra Mdiffuse, it is enough to show that Mdiffuse = {0}. Suppose Mdiffuse 6= {0} and take y ∈ Mdiffuse\{0}. Then, by local convexity of M, there exists a SRT-continuous linear functional f on M such that f (0 ⊕ y) 6= 0. However this is a contradiction by Lemma 3.24. Similarly one can prove the following proposition. Proposition 3.25. Let M be a finite von Neumann algebra. Then the following are equivalent. (1) There exists no non-zero SRT-continuous linear functional on M. (2) M is diffuse. 4 Lie Group-Lie Algebra Correspondences In this section we state and prove the main result of this paper. As explained in the introduction, Lie theory for U (H) is a difficult issue. What one has to resolve for discussing the Lie group-Lie algebra correspondence is a domain problem of the generators of one parameter subgroups of G ⊂ U (H). The second to be discussed is a continuity of the Lie algebraic operations. However we can show that, for any strongly closed subgroup G of unitary group U (M) of some finite von Neumann algebra M, there exists canonically a complete topological Lie algebra. Since there are continuously many non-isomorphic finite von Neumann algebras on H, there are also varieties of such groups. We hope that the "Lie Groups-Lie Algebras Correspondences" will play some important roles in the infinite dimensional Lie theory. We study the SRT-closed subalgebra of M, too. 4.1 Existence of Lie Algebra Let M be a finite von Neumann algebra acting on a Hilbert space H. Recall that a densely defined closable operator A is called a skew-adjoint operator if A∗ = −A, and A is called essentially skew-adjoint if A is skew-adjoint. Remark 4.1. In general, the strong limit of unitary operators is not necessarily unitary. It is known that U (M) is strongly closed in B(H) if and only if M is a finite von Neumann algebra. Definition 4.2. For a strongly closed subgroup G of U (M), the set g = Lie(G) := {A ; A∗ = −A on H, etA ∈ G, for all t ∈ R} is called the Lie algebra of G. The complexification gC of g is defined by If G = U (M), we sometimes write g as u(M). gC :=(cid:8)A + iB ; A, B ∈ g(cid:9) . 32 At first sight, it is not clear whether we can define algebraic operations on g. However, Lemma 4.3. Under the above notations, g ⊂ M holds. Proof . Let u ∈ U (M′) and A ∈ g. By definition, we have etAu = uetA. Taking the strong derivative on each side, we have uA ⊂ Au. Since u is arbitrary we obtain uA = Au, which implies A ∈ M. Therefore the sum A + B and the Lie bracket AB − BA are well-defined operations in M, but it is not clear whether they belong to g again. The following Lemma 4.5 guarantees the validity of the name "Lie algebra". The former part of the proof is based on the two lemmata established by Trotter- Kato and E. Nelson, which are of importance in their own. Lemma 4.4 (Trotter-Kato, Nelson [14]). Let A, B be skew-adjoint operators on a Hilbert space H. (1) If A + B is essentially skew-adjoint on dom(A)∩ dom(B), then it holds that et(A+B) = s- lim n→∞(cid:16)etA/netB/n(cid:17)n , for all t ∈ R. (2) If (AB − BA) is essentially skew-adjoint on then it holds that dom(A2) ∩ dom(AB) ∩ dom(BA) ∩ dom(B2), n B(cid:17)n2 n→∞(cid:16)e−√ t n Ae−√ t n Be√ t n Ae√ t et[A,B] = s- lim , for all t > 0, where [A, B] := AB − BA. Lemma 4.5. Let G be a strongly closed subgroup of U (M). Then g is a real Lie algebra with the Lie bracket [X, Y ] := XY − Y X. Proof . Let A, B ∈ g. It suffices to prove that A + B and AB − BA belong to g. Since dom(A) ∩ dom(B) is completely dense, A + B is essentially skew-adjoint. Therefore by Lemma 4.4 (1), we have et(A+B) ∈ G = G for all t ∈ R. This implies A + B ∈ g. It is clear that λA ∈ g for all λ ∈ R. On the other hand, as AB − BA is essentially skew-adjoint on s D := dom(AB) ∩ dom(BA) ∩ dom(A2) ∩ dom(B2), since D is completely dense by Proposition 2.10 and AB − BA ∈ M. Therefore by Proposition 4.4 (2), we have et(AB−BA) ∈ G for all t > 0. Thanks to the unitarity, this equality is also valid for t < 0. Thus we obtain [A, B] ∈ g. The associativity of the algebraic operations follows from the Murray-von Neumann's Theorem 2.18. 33 Now we state the main result of this paper, whose proof is almost completed in the previous arguments. Theorem 4.6. Let G be a strongly closed subgroup of the unitary group U (M) of a finite von Neumann algebra M. Then g is a complete topological real Lie algebra with respect to the strong resolvent topology. Moreover, gC is a complete topological Lie ∗-algebra. Proof . The Lie algebraic properties are proved in Lemma 4.5. By Lemma 3.10, we see that g and gC are SRT-closed Lie subalgebras of M. As the algebraic operations (X, Y ) 7→ X + Y , [X, Y ] are continuous with respect to the strong resolvent topology , so that the topological properties follow. Remark 4.7. It is easy to see that for G = U (M), its Lie algebra u(M) is equal to {A ∈ M; A∗ = −A} and the exponential map exp : u(M) → U (M) is continuous by Lemma B.2 and surjective by the spectral theorem. Proposition 4.8. Let M1, M2 be finite von Neumann algebras on Hilbert spaces H1, H2 respectively. Let Gi be a strongly closed subgroup of U (Mi) (i = 1, 2). For any strongly continuous group homomorphism ϕ : G1 → G2, there exists a unique SRT-continuous Lie algebra homomorphism Φ : Lie(G1) → Lie(G2) such that ϕ(eA) = eΦ(A) for all A ∈ Lie(G1). In particular, if G1 is iso- morphic to G2 as a topological group, then Lie(G1) and Lie(G2) are isomorphic as a topological Lie algebra. Proof . Let X be an element in Lie(G1). From the strong continuity of ϕ, t 7→ ϕ(etX ) is a strongly continuous one-parameter unitary group. Therefore by Stone theorem, there exists uniquely a skew-adjoint operator Φ(X) on H2 such that ϕ(etX ) = etΦ(X). This equality implies Φ(X) ∈ Lie(G2). Since ϕ is strongly continuous, thanks to Lemma 4.4, we see that etΦ([X,Y ]) = ϕ(et·[X,Y ]) = s- lim = s- lim = ϕ(cid:16)s- lim n→∞he−√ t n→∞hϕ(cid:16)e−√ t n→∞he−√ t = et·[Φ(X),Φ(Y )], n X e−√ t n Y e√ t n X(cid:17) ϕ(cid:16)e−√ t n Φ(Y )e√ t n Φ(X)e−√ t n Yin(cid:17) n X e√ t n X(cid:17) ϕ(cid:16)e√ t n Y(cid:17) ϕ(cid:16)e√ t n Φ(Y )in n Φ(X)e√ t n Y(cid:17)in for all t > 0. Taking the inverse of unitary operators, the equality etΦ([X,Y ]) = et[Φ(X),Φ(Y )] is also valid for all t < 0. Therefore from the uniqueness of a generator of one-parameter group, we have Φ([X, Y ]) = [Φ(X), Φ(Y )]. Similarly, we can prove that Φ is linear. Thus, Φ is a Lie algebra homomorphism. The SRT-continuity of Φ follows from the continuity of ϕ and the definition of the strong exponential topology. 34 As above, G has finite dimensional characters. On the other hand, it also has an infinite dimensional character. Proposition 4.9. Let M be a finite von Neumann algebra, then the following are equivalent. (1) The exponential map exp : u(M) ∋ X 7→ exp(X) ∈ U (M) is locally injective. Namely, the restriction of the map onto some SRT-neighborhood of 0 ∈ M is injective. (2) M is finite dimensional. Proof . (2)⇒(1) is trivial. We should only prove that (1)⇒(2). Step 1. For each orthogonal family of non-zero projections in M, its car- dinal number is finite. Indeed if there exists a orthogonal family of non-zero projections in M whose cardinal number is infinite, we can take a countably infinite subset of it and write it as {pn}∞ n=1. Since pn converges strongly to 0, it also converges to 0 in the strong resolvent topology. Define xn := 2πipn 6= 0. Since the spectral set of pn is {0, 1}, we have exn = 1 for all n ∈ N, while xn converges to 0 in the strong resolvent topology. This implies that the map exp(·) is not locally injective, which is a contradiction. Step 2. M is atomic. Indeed if M is not atomic, the diffuse part of it is not {0}. Thus we can take an infinite sequence of non-zero mutually orthogonal projections in M. But this is a contradiction to Step 1.. Step 3. We shall show that M is finite dimensional. By Step 2., M is spatially isomorphic to the direct sum of a family {Mnλ(C)}λ∈Λ (nλ ∈ N), where Mnλ(C) is the algebra of all nλ × nλ complex matrices. By Step 1., the cardinal number of Λ is finite. Hence M is finite dimensional. Remark 4.10. Lie(G) is not always locally convex, whereas most of infinite dimensional Lie theories, by contrast, assume local convexity. Indeed, by Propo- sition 3.21, u(M) is locally convex if and only if M is atomic. 4.2 Closed Subalgebras of M Next, we characterize closed -subalgebras of M. Proposition 4.11. Let M be a finite non Neumann algebra on a Hilbert space H, R be a SRT-closed -subalgebra of M with 1H. Then there exists a unique von Neumann subalgebra N of M such that R = N. Remark 4.12. A von Neumann subalgebra of a finite von Neumann algebra is also finite. 35 Proof of Proposition 4.11. Put N := {x ∈ R ; x is bounded}. Since 0, 1 ∈ N, N is not empty. We first show that N is a von Neumann algebra. It is clear that N is a subalgebra of M. Thus it is enough to check that N is closed with respect to the strong operator topology. Let {xα} be a net in N converging to x ∈ N with respect to the strong* operator topology. So we have with respect to the strong* operator topology. By Lemma B.1, Re(xα) −→ Re(x), Im(xα) −→ Im(x) Re(xα) −→ Re(x), Im(xα) −→ Im(x) with respect to the strong resolvent topology. As Re(xα) ∈ R, Im(xα) ∈ R and R is SRT-closed, we see that Re(x) ∈ R and Im(x) ∈ R. Therefore x = Re(x) + iIm(x) ∈ R. Since x is bounded, x belongs to N. Thus N is a von Neumann algebra. Next, we show that R ⊂ N. Let A be an element of R. It is enough to consider the case that A is self-adjoint. Put where EA(·) is the spectral measure of A. CA is completely dense and all ele- ments of CA are entire analytic vectors for A. Thus we have for all ξ ∈ CA, eitA = lim J→∞ Therefore the sequence CA := ran(EA([−n, n])), ∞[n=1 (itA)j j! ξ. ⊂ R ⊂ M JXj=1 j!  ∞ J=1 (itA)j  JXj=1 converges almost everywhere to eitA. By Proposition 3.15, it converges to eitA with respect to the strong resolvent topology. Since R is SRT-closed and eitA is bounded, we get eitA ∈ N. This implies A belongs to N. of N, we see that N ⊂ R. Thus we conclude that N = R. subalgebra of M satisfying L = R. Then, we have On the other hand, by the definition of N, N ⊂ R. Since N is a SRT-closure Finally, we show that the uniqueness for N. Let L be a von Neumann Thus N is unique. N = {x ∈ N ; x is bounded} = {x ∈ R ; x is bounded} = {x ∈ L ; x is bounded} = L. 36 Corollary 4.13. Let M be a finite non Neumann algebra on a Hilbert space H, g be a real SRT-closed Lie subalgebra of u(M). Then the following are equivalent: (1) there exists a von Neumann subalgebra N of M such that g = u(N), (2) 1H ∈ g and for all A, B ∈ g, i(cid:0)AB + BA(cid:1) ∈ g. In the above case, N is unique. Proof . First of all, we shall show (1) ⇒ (2). Since u(N) ⊂ N, we have i(cid:0)AB + BA(cid:1) ∈ N. On the other hand, Thus i(cid:0)AB + BA(cid:1) ∈ u(N) = g. Next we shall show (2) ⇒ (1). By direct u(N) = {X ∈ N ; X ∗ = −X}. computations, we see that R :=(cid:8)X + iY ∈ M ; X, Y ∈ g(cid:9) is a SRT-closed -subalgebra of M. Thus, by Proposition 4.11, there is a von Neumann subalgebra N of M such that R = N. Then we see that g = {X ∈ R ; X ∗ = −X} =(cid:8)X ∈ N ; X ∗ = −X(cid:9) = u(N). Finally, we show the uniqueness for N. Let L be a von Neumann subalgebra of M satisfying u(L) = g. Then, we have N =(cid:8)X + iY ; X, Y ∈ u(N)(cid:9) =(cid:8)X + iY ; X, Y ∈ u(L)(cid:9) = L By the uniqueness of Proposition 4.11, we get N = L. 5 Categorical Characterization of M 5.1 Introduction In this last section we turn the point of view and consider some categorical aspects of the -algebra M. Especially, we determine when a -algebra R of unbounded operators on a Hilbert space H turns out to be of the form M, without any reference to von Neumann algebraic structure in advance. As is well known, there are many examples of -algebra of unbounded operators that is not of the form M. For example, many O∗-algebras [21] are not related to any affiliated operator algebra. Therefore, the appropriate condition to single out suitable class of -algebras of unbounded operators are necessary. For this 37 purpose, we define the category fRng of unbounded operator algebras and com- pare this category with the category fvN of finite von Neumann algebras and show that both of them have natural tensor category structures (cf. Appendix C). Furthermore, we will see that they are isomorphic as a tensor category, in spite of the fact that the object in fRng is not locally convex in general while the one in fvN is a Banach space. However, the algebraic structures of M and M are very similar and in fact they constitute isomorphic categories. To begin with, let us introduce the structure of tensor category into fRng. 5.2 fvN and fRng as Tensor Categories Now we turn to the question of characterizing the category fRng of -algebras of unbounded operators which are realized as M, where M is a von Neumann algebra acting on a Hilbert space. It is well known that the usual tensor product (M1, M2) 7→ M1⊗M2 of von Neumann algebras and the tensor product of σ- weakly continuous homomorphisms (φ1, φ2) 7→ φ1 ⊗ φ2 makes the category of finite von Neumann algebras a tensor category. Therefore we define: Definition 5.1. The category fvN is a category whose objects are pairs (M,H) of a finite von Neumann algebra M acting on a Hilbert space H and whose morphisms are σ-weakly continuous unital -homomorphisms. The unit object is (C1C, C). The tensor functor is the usual tensor product functor of von Neumann algebras. The definition of left and right unit constraint functors might be obvious. If we are to characterize the objects in fRng, we must settle some subtleties due to the fact that we cannot use von Neumann algebraic structure from the outset. However, this difficulty can be overcome thanks to the the notion of the strong resolvent topology and the resolvent class whose definitions are indepen- dent of von Neumann algebras (See §3). We define fRng as follows. Definition 5.2. The category fRng is a category whose objects (R,H) consist of a SRT-closed subset R of the resolvent class RC (H) on a Hilbert space H with the following properties: (1) X + Y and XY are closable for all X, Y ∈ R. (2) X + Y , αX, XY and X ∗ again belong to R for all X, Y ∈ R and α ∈ C. (3) R forms a -algebra with respect to the sum X + Y , the scalar multi- plication αX, the multiplication XY and the involution X ∗. (4) 1H ∈ R. The morphism set between (R1,H1) and (R2,H2) consists of SRT-continuous unital -homomorphisms from R1 to R2. 38 Remark 5.3. From the definition of fRng, it is not clear whether, for each objects in fRng, its algebraic operations are continuous or not. However, the next lemma shows that R is a complete topological -algebra. Lemma 5.4. Let (R,H) be an object in fRng. Then there exists a unique finite von Neumann algebra M on H such that R = M. Furthermore, M = R∩ B(H) holds. Proof . Define M := R ∩ B(H). Then one can prove that M is von Neumann algebra by the same way as in Proposition 4.11. We next show that R ⊂ M. Let A ∈ R be a self-adjoint operator. Define the dense subspace CA according to the spectral decomposition of A: where CA := ran(EA([−n, n])), ∞[n=1 A =ZR λdEA(λ) is the spectral decomposition of A. Since all ξ ∈ CA is an entire analytic vector for A, we have eitAξ = lim n→∞ (itA)k k! ξ, nXk=0 for all t ∈ R. Let MA be a von Neumann algebra generated by {EA(J) ; J ∈ B(R)}, where B(R) is the one dimensional Borel field. Since MA is abelian, it is a finite von Neumann algebra. It is also clear that Bn := (itA)k k! nXk=0 ∈ (MA) ∩ R and eitA ∈ MA. Since CA is completely dense for MA, Bn converges almost everywhere to eitA in (MA). As MA is finite, we see that Bn converges to eitA in the strong resolvent topology. On the other hand, R is SRT-closed and therefore eitA ∈ R∩B(H) = M, for all t ∈ R. This implies A ∈ M. For a general operator B ∈ R, using real-imaginary part decomposition B = Re(B) + iIm(B), we have B ∈ M. We shall show that M ⊂ R. Let A ∈ M and A = UA be its polar 0 λdEA(λ) be decomposition, then U ∈ M ⊂ R and A ∈ M. Let A =: R ∞ the spectral decomposition of A. Put xn :=Z n 0 λdEA(λ) ∈ M ⊂ R, then xn converges to A in the strong resolvent topology. Thus A ∈ R. Therefore A = UA ∈ R. The finiteness of M follows immediately from Theorem 2.22. 39 Note that for each finite von Neumann algebra M on a Hilbert space H, The main result of this section is the next theorem. (M,H) is an object in fRng. Theorem 5.5. The category fRng is a tensor category. Moreover, fRng and fvN are isomorphic as a tensor category. To prove this theorem, we need many lemmata. The proof is divided into several steps. Next, we will define the tensor product R1⊗R2 of objects Ri (i = 1, 2) in fRng (cf. Definition 5.9). For this purpose, let us review the notion of the tensor product of closed operators. Let A, B be densely defined closed operators on Hilbert spaces H, K, respectively. Let A ⊗0 B be an operator defined by dom(A ⊗0 B) := dom(A) ⊗alg dom(B), (A ⊗0 B)(ξ ⊗ η) := Aξ ⊗ Bη, ξ ∈ dom(A), η ∈ dom(B). It is easy to see that A ⊗0 B is closable and denote its closure by A ⊗ B. Lemma 5.6. Let M1, M2 be finite von Neumann algebras acting on Hilbert spaces H1, H2, respectively. Let A ∈ M1 and B ∈ M2. Then we have A ⊗ B ∈ M1⊗M2. Proof . Let xi ∈ M′ (x1 ⊗ x2)(ξ ⊗ η) ∈ dom(A ⊗0 B) and i (i = 1, 2). For any ξ ∈ dom(A) and η ∈ dom(B), we have {(x1 ⊗ x2)(A ⊗0 B)}(ξ ⊗ η) = {(A ⊗0 B)(x1 ⊗ x2)}(ξ ⊗ η). Therefore, by the linearity, we have (x1 ⊗ x2)(A ⊗0 B) ⊂ (A ⊗0 B)(x1 ⊗ x2). Since (M1⊗M2)′ = M′ 2 is the strong closure of M′ 2, we have 1⊗M′ 1 ⊗alg M′ y(A ⊗0 B) ⊂ (A ⊗ B)y, for all y ∈ (M1⊗M2)′. Therefore by the limiting argument, we have y(A⊗B) ⊂ (A⊗B)y, which implies A ⊗ B is affiliated with M1⊗M2. Lemma 5.7. Let A, B be densely defined closed operators on Hilbert spaces H, K with cores DA,DB respectively. Then D := DA ⊗alg DB is a core of A ⊗ B. Proof . From the definition of A ⊗ B, for any ζ ∈ dom(A ⊗ B) and for any ε > 0, there exists some ζε =Pn ζ − ζε < ε, i=1 ξi ⊗ ηi ∈ dom(A) ⊗alg dom(B) such that (A ⊗ B)ζ − (A ⊗ B)ζε < ε. Put Since DB is a core of B, there exists ηε C := max 1≤i≤n{ξi, Aξi} + 1 > 0. i ∈ DB such that Bηi − Bηε i < ε nC , ε nC . ηi − ηε i < 40 Put C′ := max 1≤i≤n{ηε i , Bηε i } + 1 > 0. Similarly, since DA is a core of A, there exists ξε Aξi − Aξε ξi − ξε ε nC′ , i ∈ DA such that i < ε nC′ . i < Define ζε := nXi=1 ξε i ⊗ ηε i ∈ D. Then we have ζ − ζε ≤ ζ − ζε + ζε − ζε ξi ⊗ ηi − ξε ≤ ε + i ⊗ ηε i nXi=1 nXi=1 nXi=1 nXi=1 i + ξi ⊗ ηi − ξi ⊗ ηε nXi=1 ε nC′ · C′ ξiηi − ηε nXi=1 i + ε nC + ≤ ε + ≤ ε + C · ≤ ε + = 3ε. ··· (∗) ξi ⊗ ηε i − ξε i ⊗ ηε i nXi=1 ξi − ξε i ηε i Furthermore, (A ⊗ B)ζ − (A ⊗ B)ζε ≤ (A ⊗ B)ζ − (A ⊗ B)ζε + (A ⊗ B)ζε − (A ⊗ B)ζε ≤ ε + Aξi ⊗ Bηi − Aξε i ⊗ Bηε i nXi=1 nXi=1 nXi=1 nXi=1 Aξi ⊗ Bηi − Aξi ⊗ Bηε nXi=1 AξiBηi − Bηε ε nC′ · C′ i + ε nC + nXi=1 ≤ ε + ≤ ε + C · ≤ ε + = 3ε ··· (∗∗) Aξi ⊗ Bηε i − Aξε i ⊗ Bηε i i + nXi=1 Aξi − Aξε i Bηε i (∗) and (∗∗) implies D is a core of A ⊗ B. Next lemma says that the tensor product of algebras of affiliated operators has a natural -algebraic structures. 41 Lemma 5.8. Let M, N be finite von Neumann algebras acting on Hilbert spaces H, K respectively. Let A, C ∈ M, B, D ∈ N. Then we have (1) (A ⊗ B)(C ⊗ D) = AC ⊗ BD. (2) (A ⊗ B)∗ = A∗ ⊗ B∗. (3) A + C ⊗ B + D = A ⊗ B + A ⊗ D + C ⊗ B + C ⊗ D. (4) λ(A ⊗ B) = λA ⊗ B = A ⊗ λB (λ ∈ C). Proof . (1) From Proposition 2.13, D1 := {ξ ∈ dom(C); Cξ ∈ dom(A)} is a core of AC and D2 := {η ∈ dom(D); Dη ∈ dom(B)} is a core of BD. Define D := D1 ⊗alg D2, which is a core of AC ⊗ BD. Since dom((A ⊗ B)(C ⊗ D)) ⊃ dom((A ⊗ B)(C ⊗ D)) ⊃ D, it holds that for any ζ = ξi ⊗ ηi ∈ D, we have (A ⊗ B)(C ⊗ D)ζ = ACξi ⊗ BDηi = (AC ⊗ BD) nXi=1 ξi ⊗ ηi nXi=1 nXi=1 = (AC ⊗ BD)ζ. Therefore (A ⊗ B)(C ⊗ D) ⊃ (AC ⊗ BD)D. Since D is a core of AC ⊗ BD, we have (by taking the closure) (A ⊗ B)(C ⊗ D) ⊃ AC ⊗ BD. Since both operators belong to M⊗N by Lemma 5.6, we have (A ⊗ B)(C ⊗ D) = AC ⊗ BD. by Proposition 2.14(2). (2) It is easy to see that (A⊗B)∗ ⊃ A∗⊗B∗. Since (A⊗B)∗ and A∗⊗B∗ are closed operators belonging to M⊗N, we have (A⊗B)∗ = A∗⊗B∗ by Proposition 2.14 (2). (3) and (4) can be easily shown in a similar manner as in (1). Now we shall define the tensor product R1⊗R2 of (R1,H1) and (R2,H2) in Obj(fRng). Let Mi be finite von Neumann algebras on Hi such that Ri = Mi (i = 1, 2), respectively (cf. Lemma 5.4). From Lemma 5.8, the linear space R1 ⊗alg R2 spanned by {A1 ⊗ A2 ; Ai ∈ Ri, i = 1, 2} is a -algebra. Since R1 ⊗alg R2 is a subset of M1⊗M2, it belongs to RC (H1 ⊗ H2). Therefore: Definition 5.9. Under the above notations, we define R1⊗R2 to be the SRT- closure (for H1 ⊗ H2) of R1 ⊗alg R2. 42 Lemma 5.10. Let Ri (i = 1, 2) be as above. Then R1⊗R2 is also an object in fRng. More precisely, if Ri = Mi, where Mi is a finite von Neumann algebra (i = 1, 2), then M1⊗M2 = M1⊗M2. Proof . We first show that M1⊗M2 ⊂ M1⊗M2. Let Ti ∈ Mi (i = 1, 2). Then we can show that T1 ⊗ T2 ∈ M1⊗M2 by Lemma 5.6. Therefore by the linearity, we obtain M1 ⊗alg M2 ⊂ M1⊗M2. As the left hand side is SRT-closed in RC (H1 ⊗ H2), we have M1⊗M2 ⊂ M1⊗M2. Next we prove that M1⊗M2 ⊂ M1⊗M2. It is clear that M1 ⊗alg M2 ⊂ M1⊗M2. By the Kaplansky density theorem and Lemma 3.6, we have M1⊗M2 ⊂ M1⊗M2. By taking the SRT-closure, we obtain M1⊗M2 ⊂ M1⊗M2. The above Lemma says that (R1⊗R2,H1 ⊗H2) is again an object in fRng. Next, we discuss the extension of morphisms in fvN to ones in fRng. It requires some steps. Lemma 5.11. Let (M1,H1), (M2,H2) be finite von Neumann algebras. Then the mapping (M1, SRT ) × (M2, SRT ) −→ (M1⊗M2, SRT ), (A, B) 7−→ A ⊗ B, is continuous. Proof . Let {Aα}α ⊂ M1, {Bα}α ⊂ M2 be SRT-converging nets and A ∈ M1, B ∈ M2 be their limits respectively. We should only show that the net {Aα ⊗ Bα}α converges to A ⊗ B in the strong resolvent topology. Step 1. The above claim is true if all Aα, Bα, A and B are self-adjoint. Indeed, since eit(Aα⊗1) = eitAα ⊗ 1, eit(A⊗1) = eitA ⊗ 1 hold, we easily see that Aα ⊗ 1 converges to A ⊗ 1 in the strong exponential topology. Thus, by Theorem 3.10, the SRT-convergence of Aα ⊗ 1 to A ⊗ 1 follows. Similarly 1 ⊗ Bα converges to 1 ⊗ B in the strong resolvent topology. Therefore, by Lemma 5.8 and the SRT-continuity of the multiplication, we have Aα ⊗ Bα = (Aα ⊗ 1) (1 ⊗ Bα) → (A ⊗ 1) (1 ⊗ B) = A ⊗ B. 43 Step 2. In a general case, by Lemma 5.8, we obtain Aα ⊗ Bα =(cid:16)Re(Aα) + iIm(Aα)(cid:17) ⊗(cid:16)Re(Bα) + iIm(Bα)(cid:17) = Re(Aα) ⊗ Re(Bα) + iRe(Aα) ⊗ Im(Bα) +iIm(Aα) ⊗ Re(Bα) − Im(Aα) ⊗ Im(Bα) → Re(A) ⊗ Re(B) + iRe(A) ⊗ Im(B) +iIm(A) ⊗ Re(B) − Im(A) ⊗ Im(B) = A ⊗ B. Hence the proof of Lemma 5.11 is complete. Lemma 5.12. Let M be a finite von Neumann algebra on a Hilbert space H and e is a projection in M′, then Me is also finite. Proof . Well-known. Lemma 5.13. Let A be a densely defined closed operator on a Hilbert space H, K be a closed subspace of K such that PKA ⊂ APK. Then the operator B := Adom(A)∩K is a densely defined closed operator on K. Proof . This is a straightforward verification. The next proposition guarantees the existence and the uniqueness of the extension of morphisms in fvN to the morphisms in fRng. Note that the claim is not trivial, because many σ-weakly continuous linear mappings between finite von Neumann algebras cannot be extended SRT-continuously to the algebra of affiliated operators. Indeed, we can not extend any σ-weakly continuous state on a finite von Neumann algebra M SRT-continuously onto M if M is diffuse. Proposition 5.14. Let M1, M2 be finite von Neumann algebras on Hilbert spaces H1, H2 respectively. (1) For each SRT-continuous unital -homomorphism Φ : M1 → M2, the restriction ϕ of Φ onto M1 is a σ-weakly continuous unital -homomorphism from M1 to M2. (2) Conversely, for each σ-weakly continuous unital -homomorphism ϕ : M1 → M2, there exists a unique SRT-continuous unital -homomorphism Φ : M1 → M2 such that ΦM1 = ϕ. Proof . (1) We have to prove that Φ maps all bounded operators to bounded operators. For any u ∈ U (M1) and ξ ∈ dom(Φ(u)∗Φ(u)), we have Φ(u)ξ2 = hξ, Φ(u)∗Φ(u)ξi = hξ, Φ(u∗u)ξi = hξ, Φ(1)ξi = ξ2. Since dom(Φ(u)∗Φ(u)) is a (completely) dense subspace, Φ(u) ∈ M2 and Φ(u) is an isometry. Therefore the finiteness of M2 implies Φ(u) ∈ U (M2). Thus, we 44 see that Φ(U (M1)) ⊂ U (M2). Since any element in M1 is a linear combination of U (M1), Φ maps M1 into M2. To show that ϕ is σ-weakly continuous, it is sufficient to prove the (σ-) strong continuity on the unit ball, because it is a homomorphism. Since the strong resolvent topology coincides with the strong operator topology on the closed unit ball by Lemma 3.6, ϕ is strongly continuous on the closed unit ball. Therefore ϕ is a σ-weakly continuous homomorphism. (2) Regard ϕ as a composition of a surjection ϕ′ : M1 → ϕ(M1) and the inclusion map ι : ϕ(M1) ֒→ M2. Note that the σ-weak continuity of ϕ implies ϕ(M1) is a von Neumann algebra. Since ϕ′ is surjective, from Theorem IV.5.5 of [24], there exists a Hilbert space K, a projection e′ ∈ P (M′ 1⊗B(K)) and a unitary operator U : e′(H1 ⊗ K) ∼→ H2 such that ϕ′(x) = U (x ⊗ 1K)e′ U ∗ for all x ∈ M1. Now we would like to define the extension Φ′ of ϕ′ to M1 → ϕ(M1). Then we define Φ′ as follows: Φ′(X) = U (X ⊗ 1K)e′ U ∗, X ∈ M1. /_________ Φ′ M1 ϕ(M1) ·⊗1 (cid:9) U·U ∗ M1 ⊗ C1K reduction by e′ / (M1 ⊗ C1K)e′ More precisely, we define Z = (X ⊗ 1)e′ := e′(X ⊗ 1)ran(e′)∩dom(X⊗1), Φ′(X) := U ZU ∗. We have Z ∈ (M1 ⊗ C1K)e′ . Indeed, since e′ commutes with M ⊗ C1K, it reduces the operator X ⊗ 1 and therefore by Lemma 5.13, (X ⊗ 1)e′ is a densely defined closed operator on ran(e′). Since (Nf )′ = (N′)f for each von Neumann algebra N and f ∈ P (N′), the affiliation property is manifest. In addition, by Lemma 5.12, (M ⊗ C1K)e′ is a finite von Neumann algebra. Next, we prove the map M ∋ X 7→ (X ⊗ 1)e′ ∈ (M1 ⊗ C1K)e′ is a SRT-continuous unital - homomorphism. The continuity follows from Lemma 5.11. To prove that it is a -homomorphism, we have to show that for X, Y ∈ M, ((X + Y ) ⊗ 1)e′ = (X ⊗ 1)e′ + (Y ⊗ 1)e′ , To prove the first equality, by Lemma 5.8, we see that (XY ⊗ 1))e′ = (X ⊗ 1)e′ (Y ⊗ 1)e′, ((X ⊗ 1)e′ )∗ = (X ∗ ⊗ 1)e′ . (cid:0)(X + Y ) ⊗ 1(cid:1)e′ =(cid:0)X ⊗ 1 + Y ⊗ 1(cid:1)e′ ⊃ (X ⊗ 1)e′ + (Y ⊗ 1)e′ . 45 / / O O Taking the closure, by Lemma 2.14, we have (cid:0)(X + Y ) ⊗ 1(cid:1)e′ = (X ⊗ 1)e′ + (Y ⊗ 1)e′ . The others are proved in a similar manner. Next, by Lemma 3.20, the correspon- dence M1 ∋ X 7→ U (X ⊗ 1K)e′ U ∗ ∈ ϕ(M1) ⊂ M2 defines a SRT-continuous unital -homomorphism Φ′ which is clearly an extension of ϕ′. Therefore by considering Φ := ι′ ◦ Φ′ : M1 → M2 is the desired extension of ϕ, where ι′ : Φ′(M1) ֒→ M2 is the mere inclusion. Finally, we prove the uniqueness of the extension. Let Ψ be another SRT-continuous unital -homomorphism such that ΨM1 = ϕ. Let X ∈ M1. Then from the SRT-density of M1 in M1, there exists a net {xα} ⊂ M1 such that limα xα = X in the strong resolvent topology. Therefore we have Ψ(X) = lim α = lim α Ψ(xα) = lim α ϕ(xα) Φ(xα) = Φ(X). The next lemmata, together with Lemma 5.10, implies that fRng is a tensor category. Lemma 5.15. Let Ri, Si (i = 1, 2) be objects in Obj(fRng). If Ψ1 : R1 → S1, Ψ2 : R2 → S2 are SRT-continuous unital -homomorphisms, then there exists a unique SRT-continuous unital -homomorphism Ψ : R1⊗R2 → S1⊗S2 such that Ψ(A ⊗ B) = Ψ1(A) ⊗ Ψ2(B), for all A ∈ R1 and B ∈ R2. We define Ψ1 ⊗ Ψ2 to be the map Ψ. Proof . Let ψi be the restrictions of Ψi onto Mi (i = 1, 2). Then ψi is a σ-weakly continuous unital -homomorphism from Mi to Ni, where Ni = Si. Thus there exists a σ-weakly continuous unital -homomorphism ψ from M1⊗M2 to N1⊗N2 such that ψ(x ⊗ y) = ψ1(x) ⊗ ψ2(y), x ∈ M1, y ∈ M2. By Proposition 5.14, there exists a SRT-continuous unital -homomorphism Ψ from R1⊗R2 to S1⊗S2 whose restriction to M1⊗M2 is equal to ψ. For all A ∈ R1, B ∈ R2, we can take nets {xα}α ⊂ M1, {yα}α ⊂ M2 converging to A, B in the strong resolvent topology, respectively. Therefore, by Proposition 5.11, we have Ψ(A ⊗ B) = lim α = lim α Ψ(xα ⊗ yα) = lim ψ1(xα) ⊗ ψ2(yα) Ψ1(xα) ⊗ Ψ2(yα) = Ψ1(A) ⊗ Ψ2(B). α 46 Lemma 5.16. Let (Ri,Hi) (i = 1, 2, 3) be objects in fRng. Then we have a unique -isomorphism which is homeomorphic with respect to the strong resol- vent topology: (R1⊗R2)⊗R3 ∼= R1⊗(R2⊗R3) (X1 ⊗ X2) ⊗ X3 7→ X1 ⊗ (X2 ⊗ X3), for all Xi ∈ Ri We denote the map as αR1,R2,R3. Proof . Let Mi be a finite von Neumann algebra such that Ri = Mi (i = 1, 2, 3). Let α0 be the -isomorphism from (M1⊗M2)⊗M3 onto M1⊗(M2⊗M3) defined by (x1 ⊗ x2) ⊗ x3 7→ x1 ⊗ (x2 ⊗ x3). By Lemma 5.10, both (M1⊗M2)⊗M3 and M1⊗(M2⊗M3) are generated by (M1⊗M2)⊗M3 and M1⊗(M2⊗M3), re- spectively. Therefore by Proposition 5.14, α0 can be extended to the desired -isomorphism αR1,R2,R3. Proposition 5.17. fRng is a tensor category. Proof . We define the tensor product ⊗ : fRng × fRng → fRng by (R1,H1) ⊗ (R2,H2) := (R1⊗R2,H1 ⊗ H2) and for two morphisms Ψi : (Ri,Hi) → (Si,Ki) (i = 1, 2), define Ψ1 ⊗ Ψ2 according to Lemma 5.15. The unit object is I := (C1C, C). The associative constraint αR1,R2,R3 is the map defined in Lemma 5.16. The naturality of αR1,R2,R3 follows from Proposition 5.14. The definition of left (resp. right) constraint λ· (resp. ρ·) might be clear. Now it is a routine task to verify that the data (fRng,⊗, I, α, λ, ρ) constitutes a tensor category. Now we will prove that fvN is isomorphic to fRng as a tensor category. Define two functors E : fvN→ fRng, F : fRng→fvN. Definition 5.18. Define two correspondences E, F as follows: (1) For each object (M,H) in fvN, E(M,H) := (M,H), which is an object in fRng. For each morphism ϕ : M1 → M2 in fvN, E(ϕ) : M1 → M2 is the unique SRT-continuous extension of ϕ to M1, so that E(ϕ) is a morphism in fRng by Proposition 5.14. (2) For each object (R,H) in fRng, F (R,H) := (R ∩ B(H),H). For each morphism Φ : R1 → R2 in fRng, F (Φ) := ΦR1∩B(H), which is a morphism in fvN by Proposition 5.14. Lemma 5.19. E and F are tensor functors. 47 Proof . We define the tensor functor (E, h1, h2), where h1 : (C1C, C) id−→ (C1C, C) = E((C1C, C)), h2((M1,H1), (M2,H2)) : M1⊗M2 id−→ M1⊗M2, can be taken to be identity morphisms thanks to Lemma 5.15. It is clear that E(1M) = 1M, where 1M and 1M are identity map of M and M, respectively. ϕ2−→ M3 be a sequence of morphisms in fvN. Let x ∈ M1. It Let M1 holds that ϕ1−→ M2 E(ϕ2 ◦ ϕ1)(x) = (ϕ2 ◦ ϕ1)(x) = E(ϕ2)(ϕ1(x)) = E(ϕ2)(E(ϕ1)(x)) = {E(ϕ2) ◦ E(ϕ1)} (x). By Proposition 5.14 (2), we have E(ϕ2 ◦ ϕ1) = E(ϕ2) ◦ E(ϕ1). Therefore E is a functor. The conditions for (E, h1, h2) to be a tensor functor are described as the following three diagrams, the commutativity of which are almost obvious by Proposition 5.14 and "∼" symbols are followed from Lemma 5.16. (M1⊗M2)⊗M3 ∼ / / M1⊗(M2⊗M3) id id (M1⊗M2)⊗M3 (cid:9) M1⊗(M2⊗M3) id id (M1⊗M2)⊗M3 ∼ / / M1⊗(M2⊗M3) 1⊗X7→X / M C⊗M id (cid:9) X⊗17→X M⊗C / M id (cid:9) / M⊗C Thus, (E, h1, h2) is a tensor functor. The proof that (F , h′ functor, including the definitions of h′ / C⊗M M⊗C C⊗M 1, h′ 2 are easier. id id / 1, h′ 2) is a tensor Now we are able to prove the main theorem easily. Proof of Theorem 5.5. We will show that E and F are the inverse tensor functor of each other. By Lemma 5.19, they are tensor functors. Let (Mi,Hi) (i = 1, 2) be in Obj(fvN). Let ϕ : M1 → M2 be a morphism in fvN. Proposition 5.14 implies ϕ = (F ◦ E)(ϕ). By Proposition 5.4, we have (Mi,Hi) = (Mi ∩ B(Hi),Hi) = (F ◦ E)(Mi,Hi), therefore F ◦ E = idfvN. 48 / / / O O O O Let (Ri,Hi) (i = 1, 2) be objects in fRng, Φ : (R1,H1) → (R2,H2) be a morphism in fRng. By Proposition 5.4, we have Ri = Mi for a unique (Mi,Hi) in Obj(fvN). Similarly, we can prove that (Ri,Hi) = (E ◦ F )(Ri,Hi), (E ◦ F )(Φ) = Φ, hence E ◦ F = idfRng. Finally, we remark the correspondence of factors in fvN and ones in fRng. Recall that, for a -algebra A , its center Z(A ) is defined by Z(A ) := {x ∈ A ; xy = yx, for all y ∈ A } . Z(A ) is also a -algebra. Proposition 5.20. Let M be a finite von Neumann algebra on H. The following conditions are equivalent. (1) The center Z(M) of M is trivial. I.e., Z(M) = C1H. (2) The center Z(M) of M is trivial. Proof . (1) ⇒ (2) is evident. (2) ⇒ (1). Let A ∈ M be a self-adjoint element of the center Z(M). For any u ∈ U (M′), we have uAu∗ = A. Therefore from the unitary covariance of the functional calculus, it holds that u(A − i)−1u∗ = (A − i)−1 and (A − i)−1 ∈ M∩M′ = C1. Hence (A−i)−1 = α1 for some α ∈ C. By operating A−i on both sides, we see that A ∈ C1. For a general closed operator A ∈ Z(M), we know that there is a canonical decomposition A = Re(A) + i Im(A). Since A belongs to Z(M), Re(A), Im(A) also belong to Z(M) = C1. Therefore A ∈ C1. A Direct Sums of Operators We recall the theory of direct sums of operators and show some facts. We do not give proofs for well-known facts. See e.g., [2]. Let {Hα}α be a family of Hilbert spaces and H =Lα Hα be the direct sum Hilbert space of {Hα}α, i.e., H :=(ξ = {ξ(α)}α ; ξ(α) ∈ Hα, Xα kξαk2 < ∞.) . For a subspace Dα of Hα, we set dMαDα :=nξ = {ξ(α)}α ∈ H ; ξ(α) ∈ Dα, ξ(α) = 0 except finitely many α.o . It is known that cLαDα is dense in H whenever each Dα is dense in Hα. 49 Next we recall the direct sum of unbounded operators. Let Aα be a (possibly unbounded) linear operator on Hα. We define the liner operator A = ⊕αAα on H as follows: dom(A) :=(ξ = {ξ(α)}α ∈ H ; ξ(α) ∈ dom(Aα), Xα kAαξαk2 < ∞.) , (Aξ)(α) := Aαξ(α), ξ ∈ dom(A). A is said to be the direct sum of {Aα}α. It is easy to see that if each Aα is a densely defined closed operator then so is A. In this case, A∗ = ⊕αAα holds. The following lemmata are well-known. ∗ Lemma A.1. Assume the above notations. (1) A ∈ B(H) if and only if each Aα is in B(Hα) and supα kAαk < ∞. In this case, holds. kAk = sup α kAαk (2) A is unitary if and only if each Aα is unitary. (3) A is projection if and only if each Aα is projection. In this case, ran(A) =Mα ran(Aα) holds. Lemma A.2. Assume that each Aα is closed. Let Dα be a core of Aα. Then cLαDα is a core of A. Lemma A.3. Assume that each Aα is (possibly unbounded) self-adjoint. (1) A is self-adjoint. (2) For any complex valued Borel function f on R, holds. f (A) = ⊕αf (Aα) Finally, we study the direct sum of algebras of operators. Let Sα be a set of densely defined closed operators on Hα. Put Mα Sα := {⊕αAα ; Aα ∈ Sα} . 50 Note that each element inLα Lα Hα. If each Sα consists only of bounded operators, we also define Sα is a densely defined closed operator on H = Sα :=(cid:26)⊕αxα ; xα ∈ Sα, sup α kxαk < ∞.(cid:27) . By Lemma A.1, each element in Lb α well-known. Lemma A.4. Let Mα be a von Neumann algebra acting on Hα, and put Sα is bounded. The following is also bMα M := Mα. bMα Then M is a von Neumann algebra acting on H =Lα Hα. The sum, the scalar multiplication, the multiplication and the involution are given by (⊕αxα) + (⊕αyα) = ⊕α (xα + yα) , λ (⊕αxα) = ⊕α (λxα) , (⊕αxα) (⊕αyα) = ⊕α (xαyα) , ∗) . (⊕αxα)∗ = ⊕α (xα f or all λ ∈ C, Furthermore the followings hold. α M′ α. (1) M′ =Lb We callLb (2) M is a finite von Neumann algebra if and only if each Mα is a finite von Neumann algebra. α Mα the direct sum von Neumann algebra of {Mα}α. B Fundamental Results of SRT A and D ⊂Tλ∈Λ dom(Aλ)∩ dom(A). Suppose for all ξ ∈ D, limλ∈Λ Aλξ = Aξ, Let H be a Hilbert space. The following lemmata are well-known [19]: Lemma B.1. Let {Aλ}λ∈Λ be a net of self-adjoint operators on H, A be a self-adjoint operator on H, and D be a dense subspace of H which is a core of then Aλ converges to A in the strong resolvent topology. Lemma B.2. Let {An}∞ n=1 be a sequence of self-adjoint operators on H, A be a self-adjoint operator on H. Then An converges to A in the strong resolvent topology if and only if eitAn converges strongly to eitA for all t ∈ R. In this case, the strong convergence of eitAn to eitA is uniform on every finite interval of t. Lemma B.3. Let {An}∞ n=1 be a sequence of self-adjoint operators on H, A be a self-adjoint operator on H. Suppose An converges to A in the strong resolvent topology, then EAn ((a, b)) converges strongly to EA((a, b)) for each a, b ∈ R with a < b and a, b /∈ σp(A), where σp(A) is the set of point spectra of A. 51 Lemma B.4. Let {An}∞ n=1 be a sequence of self-adjoint operators on H, A be a self-adjoint operator on H. Suppose An converges to A in the strong resolvent topology, then for all complex valued bounded continuous function f on R, f (An) converges strongly to f (A). Lemma B.5. Let {xλ}λ∈Λ be a net of bounded self-adjoint operators on H, x be a bounded self-adjoint operator on H. Suppose that sup λ∈Λkxλk < ∞, and xλ converges to x in the strong resolvent topology, then xλ converges strongly to x. C Tensor Categories We briefly review the definition of tensor categories. For more details about category theory, see MacLane [10] (we follow the style in Kassel [8], Chapter XI). Definition C.1. Let C , C ′ be categories, F ,G be functors from C to C ′. A natural transformation θ : F → G is a function which assigns to each object A in C a morphism θ(A) : F (A) → G(A) of C ′ in such a way that for every morphism f : A → B in C , the following diagram commutes: F (A) F (f ) F (B) θ(A) (cid:8) θ(B) / G(A) G(f ) / G(B) If θ(A) is an invertible morphism for every A, we call θ a natural isomorphism. Definition C.2. A tensor category (C ,⊗, I, α, λ, ρ) is a category C equipped with (1) a bifunctor ⊗ : C × C → C called a tensor product 2, (2) an object I in C called a unit object, (3) a natural isomorphism α : ⊗(⊗ × 1C ) 3 → ⊗(1C × ⊗) called an asso- ciativity constraint. 2 This implies (f ′ ⊗ g′) ◦ (f ⊗ g) = (f ′ ◦ f ) ⊗ (g′ ◦ g) for all morphisms in C , and 1A ⊗ 1B = 1A⊗B for all objects in C . 3 ⊗(⊗ × 1C ) is the composition of the functors ⊗ × 1C : (C × C ) × C → C × C and ⊗ : C × C → C . 52 / / (3) means for any objects A, B, C in C , there is an isomorphism αA,B,C : (A ⊗ B) ⊗ C → A ⊗ (B ⊗ C) such that the diagram (A ⊗ B) ⊗ C αA,B,C / A ⊗ (B ⊗ C) (f ⊗g)⊗h (cid:8) f ⊗(g⊗h) (A′ ⊗ B′) ⊗ C′ / A′ ⊗ (B′ ⊗ C′) αA′ ,B′ ,C′ commutes for all morphisms f, g, h in C . (4) a natural isomorphism λ : ⊗(I×1C ) 4→ 1C (resp. ρ : ⊗(1C ×I) → 1C ) called a left (resp. right) unit constraint with respect to I. (4) means for any object A in C , there is an isomorphism λA : I ⊗ A → A (resp. ρA : A ⊗ I → A) such that the following two diagrams commute: λA (cid:8) λA′ I ⊗ A 1I ⊗f I ⊗ A′ / A A ⊗ I f f ⊗1I / A′ A′ ⊗ I ρA (cid:8) ρA′ / A f / A′ for each morphism f : A → A′ in C . These functors and natural isomorphisms satisfy the Pentagon Axiom and the Triangle Axiom. Namely, for all objects A, B, C and D, the following diagrams commute: ((A ⊗ B) ⊗ C) ⊗ D αA,B,C ⊗1D / (A ⊗ (B ⊗ C)) ⊗ D αA,B⊗C,D αA⊗B,C,D (cid:9) A ⊗ ((B ⊗ C) ⊗ D) 1A⊗αB,C,D (A ⊗ B) ⊗ (C ⊗ D) / A ⊗ (B ⊗ (C ⊗ D)) αA,B,C⊗D αA,I,B (A ⊗ I) ⊗ B / A ⊗ (I ⊗ B) &MMMMMMMMMM ρA⊗1B (cid:8) xqqqqqqqqqq 1A⊗λB A ⊗ B Definition C.3. Let (C ,⊗, I, α, λ, ρ), (C ′,⊗, I ′, α′, λ′, ρ′) be tensor categories. (1) A triple (F , h1, h2) is called a tensor functor from C to C ′ if F : C → C ′ is a functor, h1 is an isomorphism I ′ ∼→ F (I) and h2 is a natural 4 I × 1C is the functor from C to C × C given by A 7→ (I, A) for all objects in C and f 7→ (1I , f ) for all morphisms in C . 53 / / / / / / / / & / x isomorphism ⊗(F × F ) 5 ∼→ F⊗, and they satisfy (F (A) ⊗ F (B)) ⊗ F (C) h2(A,B)⊗1F(C) αF(A),F(B),F(C) F (A) ⊗ (F (B) ⊗ F (C)) 1F(A)⊗h2(B,C) F (A ⊗ B) ⊗ F (C) (cid:9) F (A) ⊗ F (B ⊗ C) h2(A⊗B,C) h2(A,B⊗C) F ((A ⊗ B) ⊗ C) F (αA,B,C ) / F (A ⊗ (B ⊗ C)) λ′ F(A) I ′ ⊗ F (A) / F (A) ρ′ F(A) F (A) ⊗ I ′ / F (A) h1⊗1F(A) (cid:9) F (λA) 1F(A)⊗h1 (cid:9) F (ρA) F (I) ⊗ F (A) h2(I,A) / F (I ⊗ A) F (A) ⊗ F (I) h2(A,I) / F (A ⊗ I) for all objects A, B, C in C . (2) A natural tensor transformation η : (F , h1, h2) → (F ′, h′ 2) between tensor functors from C to C ′ is a natural transformation F → F ′ such that the following diagrams commute: 1, h′ F (I) h2(A,B) F (A) ⊗ F (B) / F (A ⊗ B) (cid:9) η(I) η(A)⊗η(B) (cid:9) η(A⊗B) I h1 ={{{{{{{{ !CCCCCCCC h′ 1 F ′(I) F ′(A) ⊗ F ′(B) h′ 2(A,B) / F ′(A ⊗ B) for all objects A, B in C . If η is also a natural isomorphism, it is called a natural tensor isomorphism. (3) A tensor equivalence between tensor categories C , C ′ is a tensor functor F : C → C ′ such that there exists a tensor functor F ′ : C ′ → C and ∼→ F ◦ F ′ and θ : F ′ ◦ F ∼→ 1C . If natural tensor isomorphisms η : 1C ′ η and θ can be taken to be identity transformations, then we say C is isomorphic to C ′ as a tensor category. Acknowledgement The authors would like to express their sincere thanks to Professor Asao Arai at Hokkaido University, Professor Izumi Ojima at Kyoto University for the fruitful 5 ⊗(F × F ) is a functor C × C → C which assigns F (A) ⊗ F (B) for each object (A, B) in C × C and F (f ) ⊗ F (g) for each morphism (f, g) in C × C 54 / / / / / / O O / O O / = ! / discussions, insightful comments and encouragements. H.A. also thanks to his colleagues: Mr. Ryo Harada, Mr. Takahiro Hasebe, Mr. Kazuya Okamura and Mr. Hayato Saigo for their useful comments and discussions during the seminar. Y.M. also thanks to Professor Konrad Schmudgen at Universitat Leipzig for in- forming us of the paper [20] and Mr. Yutaka Shikano at MIT for his professional advice about LaTeX. Finally, the authors thank again to Professor Izumi Ojima for his careful proofreading and suggestions. References [1] S. Albeverio, R. Høegh-Krohn, J. Marion, D. Testard and B. Torresani, Noncommutative Distributions. Unitary Representation of Gauge Groups and Algebras, Monogr. Textbooks Pure Appl. Math., 175, Marcel Dekker, Inc., New York, 1993. [2] A. Arai, Fock Spaces and Quantum Fields, Nippon-Hyouronsha, Tokyo, 2000 (in Japanese). [3] A. Banyaga, The Structure of Classical Diffeomorphism Groups, Math. Appl., 400, Kluwer Academic Publishers, Dordrecht, 1997. [4] T. Fack and H. Kosaki, Generalized s-numbers of τ -measurable operators, Pacific J. Math., 123 (1986), 260 -- 300. [5] M. Gordina, Hilbert-Schmidt groups as infinite-dimensional Lie groups and their Riemannian geometry, J. Funct. Anal. 227 (2005), 245 -- 272. [6] K. Hofmann and S. Morris, The Lie Theory of Connected Pro-Lie Groups, Europ. Math. Soc. Publ. House, 2007. [7] K. Hofmann and K.-H. Neeb, Pro-Lie groups which are infinite-dimensional Lie groups, Math. Proc. Cambridge Philos. Soc. 146 (2009), 351 -- 378. [8] C. Kassel, Quantum Groups, Grad. Texts in Math., 155, Springer-Verlag, New York, 1995. [9] H. Kosaki, On intersections of domains of unbounded positive operators, Kyushu J. Math. 60 (2006), 3 -- 25. [10] S. MacLane, Categories for the Working Mathematician, Grad. Texts in Math., 5, Springer-Verlag, New York, 1998. [11] J. Milnor, Remarks on infinite-dimensional Lie groups, In: B. DeWitt and R. Stora (Eds.), Relativity, Groups and Topology II (Les Houches, 1983), North-Holland, Amsterdam, 1984, 1007 -- 1057. [12] F. Murray and J. von Neumann, On Rings of Operators, Ann. Math. 37 (1936), 116 -- 229. 55 [13] K.-H. Neeb, Towards a Lie theory of locally convex groups, Jpn. J. Math, 1 (2006), 291 -- 468. [14] E. Nelson, Topics in Dynamics I, Princeton University Press, Princeton, 1969. [15] E. Nelson, Notes on non-commutative integration, J. Funct. Anal. 15 (1974), 103 -- 116. [16] H. Omori, Infinite-Dimensional Lie Groups, Transl. Math. Monogr., 158, Amer. Math. Soc., 1997. [17] H. Omori, On Banach-Lie groups acting on finite dimensional manifolds, Tohoku Math. J. 30 (1978), 223 -- 250. [18] A. Pressley and G. Segal, Loop Groups, Clarendon Press, Oxford, 1986. [19] M. Reed and B. Simon, Methods of Modern Mathematical Physics I, Aca- demic Press, New York, 1972. [20] K. Schmudgen, On domains of powers of closed symmetric operators, J. Operator Theory. 9 (1983), 53 -- 75. [21] K. Schmudgen, Unbounded Operator Algebras and Representation Theory, Birkhauser Verlag, Basel, 1990. [22] B. Simon, Quadratic forms and Klauder's phenomenon: a remark on very singular perturbations, J. Funct. Anal. 14 (1973), 295 -- 298. [23] W. Stinespring, Integration theorems for gages and duality for unimodular groups. Trans. Amer. Math. Soc. 90 (1959), 15 -- 56. [24] M. Takesaki, Theory of Operator Algebras I, Springer-Verlag, Berlin, 1979. 56
1002.2180	2	1002	2010-03-08T18:26:41	Nuclear dimension and n-comparison	[ "math.OA" ]	It is shown that if a C-algebra has nuclear dimension $n$ then its Cuntz semigroup has the property of $n$-comparison. It then follows from results by Ortega, Perera, and Rordam that $\sigma$-unital C-algebras of finite nuclear dimension (and even of nuclear dimension at most $\omega$) are stable if and only if they have no non-zero unital quotients and no non-zero bounded traces.	math.OA	math	NUCLEAR DIMENSION AND n-COMPARISON LEONEL ROBERT Abstract. It is shown that if a C∗-algebra has nuclear dimension n then its Cuntz semigroup has the property of n-comparison. It then follows from results by Ortega, Perera, and Rørdam that σ-unital C∗-algebras of finite nuclear dimension (and even of nuclear dimension at most ω) are stable if and only if they have no non-zero unital quotients and no non-zero bounded traces. 1. Introduction In [7], Winter and Zacharias define nuclear dimension for C∗-algebras. This is a form of noncommutative dimension which directly generalizes the covering dimension of topological spaces. Finite nuclear dimension is specially relevant to the classification of C∗-algebras. The simple C∗-algebras of finite nuclear di- mension have been proposed as a likely class of C∗-algebras for which Elliott's classification in terms of K-theory and traces holds true. In the main result of this paper it is shown that the Cuntz semigroup of a C∗-algebra of finite nuclear dimension n satisfies the n-comparison property. For n = 0, this property is the same as almost unperforation in the Cuntz semigroup. For arbitrary n, it is reminiscent of the comparability between vector bundles whose fibrewise dimensions differ sufficiently relative to the dimension of the base space. The n-comparison property for the Cuntz semigroup was first considered by Toms and Winter (see [6, Lemma 6.1]). They showed that n-comparison holds under the more restrictive assumption that the C∗-algebra is simple unital of decomposition rank n (the decomposition rank bounds the nuclear dimension, and, unlike nuclear dimension, is infinite for UCT Kirchberg algebras). The n- comparison property was subsequently studied, and more precisely defined, by Ortega, Perera and Rørdam in [5]. These authors obtained a simple criterion of stability for σ-unital C∗-algebras with n-comparison in their Cuntz semigroup: the C∗-algebra is stable if and only if it has no non-zero unital quotients and no non-zero bounded 2-quasitraces. By Theorem 1 below, this stability criterion applies to all C∗-algebras of finite nuclear dimension. It then follows that σ-unital C∗-algebras of finite nuclear dimension have the corona factorization property. Let us recall the definition of nuclear dimension given in [7]. Definition 1. The C∗-algebra A has nuclear dimension n if n is the smallest nat- ural number for which there exist nets of completely positive contractions (hence- forth abbreviated as c.p.c.) ψi λ : A → F i λ and φi λ : F i λ → A, 1 with i = 0, 1, . . . , n, λ ∈ Λ, and F i such that λ finite dimensional C∗-algebras for all i and λ, λ is an order 0 map (i.e., preserves orthogonality) for all i and λ, (i) φi (ii) limλ Pn If no such n exists then A has infinite nuclear dimension. λ(a) = a for all a ∈ A. i=1 φi λψi Let us recall the definition given in [5] of the n-comparison property of an ordered semigroup. For x, y elements of an ordered semigroup S, let us write x 6s y if (k + 1)x 6 ky for some k ∈ N. Definition 2. The ordered semigroup S has the n-comparison property if x 6s yi for x, yi ∈ S and i = 0, 1, . . . , n, implies x 6 Pn i=0 yi. Let Cu(A) denote the stabilized Cuntz semigroup of the C∗-algebra A (i.e., the semigroup W (A ⊗ K); see [2]). It is shown in Lemma 1 below that for Cu(A) the n-comparison property can be reformulated as follows: if [a], [bi] ∈ Cu(A), with i = 0, 1, . . . , n, satisfy that for each i there is εi > 0 such that dτ (a) 6 (1 − εi)dτ (bi) for all dimension functions dτ induced by lower semicontinuous 2-quasitraces, then [a] 6 Pn i=0[bi]. It is this formulation of the n-comparison property that is used by Toms and Winter in [6], and that may potentially have the most applications. Theorem 1. If A has nuclear dimension n then Cu(A) has the n-comparison property. The following section is dedicated to the proof of Theorem 1. The last section discusses the application of Theorem 1, and of a variation on it that relates to ω-comparison, to establishing the stability of C∗-algebras of finite (or at most ω) nuclear dimension. 2. Proof of Theorem 1 Let us start by proving that the property of n-comparison may be formulated using comparison by lower semicontinuous 2-quasitraces instead of the relation 6s. This result, however, will not be needed in the proof of Theorem 1. For [a], [b] ∈ Cu(A), elements of the Cuntz semigroup of A, let us write [a] <τ [b] if there is ε > 0 such that dτ (a) 6 (1 − ε)dτ (b) for all dimension functions dτ induced by lower semicontinuous 2-quasitraces τ : A+ → [0, ∞] (see [3, Section 4]). We do not assume that the 2-quasitraces are necessarily finite on a dense subset of A+. i=0[bi]. Lemma 1. The ordered semigroup Cu(A) has the property of n-comparison if and only if for [a], [bi] ∈ Cu(A), with i = 0, 1, . . . , n, [a] <τ [bi] for all i implies that [a] 6 Pn Proof. It is clear that if [a] 6s [b] then [a] <τ [b]. Thus, n-comparison is implied by the property stated in the lemma. Suppose that we have n-comparison in Cu(A). Let [a], [bi] ∈ Cu(A), with i = 0, 1, . . . , n, be such that [a] <τ [bi] for all i. Let us show that [(a − ε)+] 6s [bi] for all ε > 0 and all i. Let λ : Cu(A) → [0, ∞] be additive and order preserving and let us define λ : Cu(A) → [0, ∞] by 2 λ([c]) = supδ>0 λ([(c − δ)+]). It is known that there is a lower semicontinuous 2-quasitrace τ such that λ([a]) = dτ (a) for all a ∈ (A ⊗ K)+ (see [3, Proposition 4.2, Lemma 4.7]). We have λ([(a − ε)+]) 6 λ([a]) 6 (1 − εi)λ([bi]) 6 (1 − εi)λ([bi]) for all i. By [5, Proposition 2.1], this implies that [(a − ε)+] 6s [bi] for all i. Since Cu(A) has n-comparison, [(a − ε)+] 6 Pn i=0[bi]. Taking supremum over ε > 0, we get that [a] 6 Pn (cid:3) i=0[bi]. Throughout the rest of this section Λ denotes the index set in the definition of nuclear dimension. This set may be chosen to be the pairs (F, ε) where F ⊆ A is finite and ε > 0. Let us denote by AΛ the algebra Qλ A/ Lλ A. Let ι : A → AΛ denote the diagonal embedding of A into AΛ. Notation convention. For a family of C∗-algebras (Aλ) we use the notation Lλ Aλ to refer to the C∗-algebra of nets (xλ) such that kxλk → 0, while Qλ Aλ denotes the nets of uniformly bounded norm. In [7, Proposition 3.2] Winter and Zacharias show that if A has nuclear di- mension n then the maps ψi λ in the definition of nuclear dimension may be cho- sen asymptotically of order 0. That is, such that the induced maps ψi : A → Qλ F i λ, for i = 0, 1, . . . , n, have order 0 . We get the following proposi- tion as a result of this. λ/ Lλ F i Proposition 1. If A has nuclear dimension n then for i = 0, 1, . . . , n there are c.p.c. order 0 maps ψi : A → Qλ F i λ → AΛ such that λ and φi : Qλ F i λ/ Lλ F i λ/ Lλ F i (1) ι = n X i=0 φiψi, where ι : A → AΛ is the diagonal embedding of A into AΛ. Proof. As pointed out in the previous paragraph, by [7, Proposition 3.2] the maps ψi λ in the definition of nuclear dimension may be chosen so that the induced maps ψi are of order 0. λ) : Qλ F i The equation (1) is a consequence of Definition 1 (ii). Let us show that the maps φi : Qλ F i λ/ Lλ F i λ, also have order 0. It is clear that (φi λ → AΛ, induced by the order 0 maps φi λ) : Qλ F i λ → Qλ A has order 0. Hence, α ◦ (φi λ → AΛ, where α is the quotient onto AΛ, has order 0. We will be done once we show that if φ : C → D is a c.p.c. map of order 0 and φI = 0 for some closed two sided ideal I, then the induced map φ : C/I → D has order 0. By [8, Corollary 3.1], there is a -homomorphisms π : C ⊗ C0(0, 1] → D such that φ(c) = π(c ⊗ t) for all c ∈ C. From π(I ⊗ t) = 0 we get that π(I ⊗ C0(0, 1]) = 0. Thus, π induces a -homomorphism π : C/I ⊗C0(0, 1] → D. Since φ(c) = π(c⊗t) for all c ∈ C/I, φ has order 0. (cid:3) An ordered semigroup S is said unperforated if kx 6 ky for x, y ∈ S and k ∈ N implies x 6 y. 3 Lemma 2. (i) If Cu(A) is unperforated then so is Cu(A/I) for any closed two- sided ideal I. (ii) If (Aλ)λ are C∗-algebras such that Cu(Aλ) is unperforated for all λ then so are Cu(Qλ Aλ) and Cu(Qλ Aλ/ Lλ Aλ). Proof. (i) Let [a], [b] ∈ Cu(A/I) be such that k[a] 6 k[b] for some k ∈ N. Then for [a], [b] ∈ Cu(A) lifts of [a] and [b] we have k[a] 6 k[b] + [i] 6 k([b] + [i]), for some i ∈ (I ⊗ K)+ (by [1, Proposition 1]). Since Cu(A) is unperforated, we have [a] 6 [b] + [i], and passing to Cu(A/I) we get that [a] 6 [b]. (ii) Let (aλ)λ, (bλ)λ ∈ (Qλ Aλ) ⊗ K ⊆ Qλ(Aλ ⊗ K) be positive elements of norm at most 1 such that k[(aλ)λ] 6 k[(bλ)λ] for some k. Let ε > 0. Then there is δ > 0 such that k[((aλ − ε)+)λ] 6 k[((bλ − δ)+)λ]. Hence, k[(aλ − ε)+] 6 k[(bλ − δ)+] for all λ. Since Cu(Aλ) is unperforated, [(aλ − ε)+] 6 [(bλ − δ)+]. Let xλ ∈ Aλ ⊗ K be such that (aλ − 2ε)+ = x∗ λxλ and xλx∗ λ ∈ her((bλ − δ)+). Since kxλk2 6 kaλk 6 1, we have (xλ)λ ∈ Qλ(Aλ ⊗ K). We also have xλx∗ whence λ 6 1 δ bλ, ((aλ − 2ε)+)λ = (xλ)∗ λ(xλ)λ and (xλ)λ(xλ)∗ λ ∈ her((bλ)λ). But (Qλ Aλ) ⊗ K sits as a hereditary subalgebra of Qλ(Aλ ⊗ K). Therefore, (xλ)λ ∈ (Qλ Aλ) ⊗ K and so [((aλ − 2ε)+)λ] 6 [(bλ)λ]. Since ε may be arbitrarily small, we have [(aλ)λ] 6 [(bλ)λ] as desired. (cid:3) Lemma 3. Let a, b ∈ (A ⊗ K)+. If [ι(a)] 6 [ι(b)] then [a] 6 [b]. Proof. Let ε > 0. Since [ι(a)] 6 [ι(b)], there is d ∈ AΛ ⊗ K such that d∗ι(b)d = ι(a − ε)+. That is, there are dλ ∈ Aλ ⊗ K such that d∗ λbdλ → (a − ε)+. Thus, [(a − ε)+] 6 [b]. Since ε > 0 may be arbitrarily small, we have [a] 6 [b]. (cid:3) Remark 2. In proving Theorem 1, a stronger property than n-comparison will be shown to hold for C∗-algebras of nuclear dimension n: if x, yi ∈ Cu(A), with i = 0, 1, . . . , n, satisfy that kix 6 kiyi for some ki ∈ N and all i, then x 6 Pn i=0 yi. This property, unlike n-comparison, does not seem to have a formulation in terms of comparison by lower semicontinuous 2-quasitraces. Proof of Theorem 1. Suppose that there are ki ∈ N such that ki[a] 6 ki[bi] for i = 0, 1, . . . , n. Since c.p.c. order 0 maps preserve Cuntz comparison (by [8, Corollary 3.5]), we have that ki[ψi(a)] 6 ki[ψi(bi)] for all i. Since the Cuntz semigroup of finite dimensional algebras is unperforated, we have by Lemma 2 that the Cuntz semigroup of Qλ F i λ is unperforated. Thus, [ψi(a)] 6 [ψi(bi)]. The maps φi preserve Cuntz equivalence (since they are c.p.c. of order 0), whence λ/ Lλ F i [φiψi(a)] 6 [φiψi(bi)] 6 [ X φjψj(bi)] = [ι(bi)]. n j=0 4 So, [ι(a)] = [ n X i=0 By Lemma 3, this implies that [a] 6 Pn i=0 i=0[bi]. φiψi(a)] 6 n X [φiψi(a)] 6 n X i=0 [ι(bi)]. (cid:3) 3. Stability of C∗-algebras A stable C∗-algebra has no non-zero unital quotients and no non-zero bounded 2-quasitraces (see [5, Proposition 4.6]). In [5, Proposition 4.8] Ortega, Perera and Rørdam show that the converse is true provided that the C∗-algebra is σ-unital and its Cuntz semigroup has the n-comparison property. This, combined with Theorem 1 and the fact that for exact C∗-algebras bounded 2-quasitraces are traces, implies that a σ-unital C∗-algebra of finite nuclear dimension is stable if an only if it has no non-zero unital quotients and no nonzero bounded traces. Ortega, Perera, and Rørdam also show that ω-comparison, a weakening of n- comparison, suffices to obtain the same stability criterion. Definition 3. (c.f. [5, Definition 2.9]) Let S be an ordered semigroup closed un- der the suprema of increasing sequences. Then S has the ω-comparison property if x 6s yi, for x, yi ∈ S and i = 0, 1, . . . , implies x 6 P∞ Remark 3. The definition of ω-comparison given above differs slightly from the definition given in [5]. Nevertheless, both definitions agree for ordered semigroups in the category Cu introduced in [2], and therefore, also for ordered semigroups arising as Cuntz semigroups of C∗-algebras. i=0 yi. A notion of nuclear dimension at most ω may be modelled after the statement of Proposition 1. Definition 4. Let us say that a C∗-algebra A has nuclear dimension at most ω if for i = 0, 1, 2 . . . there are nets of c.p.c maps ψi λ → A, with F i λ : A → F i λ and φi λ : F i λ finite dimensional C∗-algebras, such that λ/ Lλ F i (ii) the induced maps ψi : A → Qλ F i λ and φi : Qλ F i λ/ Lλ F i λ → AΛ i=0 φiψi(a) for all a ∈ A (the sum on the right side is norm are c.p.c. order 0, (iii) ι(a) = P∞ convergent). For example, if the C∗-algebras (Ai)∞ i=0 all have finite nuclear dimension, then L∞ i=0 Ai has nuclear dimension at most ω. It is not clear whether the assumption that the maps ψi λ be asymptotically order 0 may be dropped in Definition 4 (and then proved), or if the other results on finite nuclear dimension proved in [7] also hold for nuclear dimension at most ω. The proof of Theorem 1 goes through, mutatis mutandis, for nuclear dimension at most ω. We thus have, Theorem 4. If A has nuclear dimension at most ω then Cu(A) has the ω- comparison property. Combined with the results of [5], Theorem 4 yields the following corollary, which improves on [4, Theorem 0.1], [5, Corollary 4.10] and [5, Corollary 5.13]. 5 Corollary 1. Let A be a C∗-algebra of nuclear dimension at most ω and let B ⊆ A ⊗ K be hereditary and σ-unital. Then B is stable if and only if it has no non-zero unital quotients and no non-zero bounded traces. B has the corona factorization property. Proof. By Theorem 4, Cu(A) has the ω-comparison property. Since Cu(B) is an ordered semigroup ideal of Cu(A), Cu(B) has the ω-comparison property too. Hence, by [5, Proposition 4.8], B is stable if and only if it has no non-zero unital quotients and no non-zero bounded traces. Let C ⊆ B ⊗K be full and hereditary, and suppose that Mn(C) is stable. Then Mn(C), and consequently C, cannot have non-zero unital quotients or bounded traces. Thus C is stable. This shows that B has the corona factorization property. (cid:3) Remark 5. In the hypotheses of [5, Proposition 4.8] the C∗-algebra A is assumed separable. A closer look into the proof of this result reveals that it suffices to assume that the hereditary subalgebra B ⊆ A ⊗ K be σ-unital. This justifies the application of [5, Proposition 4.8] in the above proof. Acknowledgment. I am grateful to Wilhelm Winter for pointing out a cor- rection to the proof of Lemma 2. References [1] A. Ciuperca, L. Robert, and L Santiago, The Cuntz semigroup of ideals and quotients and a generalized Kasparov Stabilization Theorem, J. Operator Theory, to appear. [2] K. T. Coward, G. A. Elliott, and C. Ivanescu, The Cuntz semigroup as an invariant for C ∗-algebras, J. Reine Angew. Math. 623 (2008), 161 -- 193. [3] G. A. Elliott, L. Robert, and L. Santiago, The cone of lower semicontinuous traces on a C∗-algebra (2008), available at http://arxiv.org/abs/0805.3122. [4] P. W. Ng and W. Winter, Nuclear dimension and the corona factorization property (2009), available at http://arxiv.org/abs/0904.0716. [5] E. Ortega, F. Perera, and M. Rørdam, The Corona Factorization property, Stability, and the Cuntz semigroup of a C∗-algebra (2009), available at http://arxiv.org/abs/0903.2917. [6] A. S. Toms and W. Winter, The Elliott conjecture for Villadsen algebras of the first type, J. Funct. Anal. 256 (2009), no. 5, 1311 -- 1340. [7] W. Winter and J. Zacharias, The nuclear dimension of C∗-algebras (2009), available at http://arxiv.org/abs/0903.4914. [8] , Completely positive maps of order zero, Munster J. of Math. 2 (2009), 311 -- 324. Department of Mathematics and Statistics, York University, Toronto, Canada M3J 1P3 E-mail address: [email protected] 6
1606.00661	2	1606	2016-06-14T08:54:34	Quantum metrics on noncommutative spaces	[ "math.OA", "math.FA" ]	We introduce two new formulations for the notion of "quantum metric on noncommutative space". For a compact noncommutative space associated to a unital C-algebra, our quantum metrics are elements of the spatial tensor product of the C-algebra with itself. We consider some basic properties of these new objects, and state some connections with the Rieffel theory of compact quantum metric spaces. The main gap in our work is the lack of a nonclassical example even in the case of C*-algebras of matrices.	math.OA	math	QUANTUM METRICS ON NONCOMMUTATIVE SPACES MAYSAM MAYSAMI SADR Abstract. We introduce two new formulations for the notion of "quantum metric on noncommutative space". For a compact noncommutative space associated to a unital C- algebra, our quantum metrics are elements of the spatial tensor product of the C-algebra with itself. We consider some basic properties of these new objects, and state some connec- tions with the Rieffel theory of compact quantum metric spaces. The main gap in our work is the lack of a nonclassical example even in the case of C-algebras of matrices. There are at least two mathematically rigorous formulations for the notion of "quantum (noncommutative) metric space" in the literatures. The famous one is due to Rieffel, and the other has been recently introduced by G. Kuperberg and N. Weaver. Following some ideas from Connes [1, Chapter VI] in noncommutative Riemannian geometry [2], Rieffel has introduced the notions of "compact quantum metric space" and "quantum Hausdorff- Gromov distance" [7],[5],[6]. In his theory, a compact quantum metric space qA is identified with the state space of a unital C-algebra A (or more generally, with the state space of an order unit space) together with a weak-compatible metric which must be induced by a "Lipschitz seminorm" on A via Monge-Kantorovich's formula. Thus in the Rieffel theory the role of quantum metrics is played by Lipschitz seminorms. In Kuperberg-Weaver theory [4] the noncommutative space is distinguished by a von Neumann algebra M ⊆ L(H) and the role of quantum metric is played by a specific one-parameter family {Vt}t≥0 of weak closed operator systems in L(H) such that V0 = M′. This construction also can be characterized by a specific "quantum distance function" between projections of the von Neumann algebra M ¯⊗L(ℓ2). In this note, we introduce two new models for "quantum metrics on noncommutative spaces". Our formulations are natural translations of the concept of "(ordinary) metric" into noncommutative geometric language. For preliminaries on C and von Neumann algebras we refer the reader to [3] or [8]. Let X be a compact metrizable space. A function ρ is a compatible metric on X if and only if ρ ∈ C(X 2) = C(X ) ⊗C(X ) and the following five conditions are satisfied for x, y, z ∈ X . (i) ρ(x, y) ≥ 0. (ii) ρ(x, x) = 0. (iii) if ρ(x, y) = 0 then x = y. (iv) ρ(x, y) = ρ(y, x). (v) ρ(x, y) ≤ ρ(x, z) + ρ(z, y). Let A be a unital C-algebra. Suppose that an element ρ ∈ A ⊗A deserves to be called a compatible metric on qA. Then ρ must be satisfied the following analogues of (i),(iv),(v). (i)′ ρ ∈ (A ⊗A)+. (iv)′ F(ρ) = ρ, where F : A ⊗A → A ⊗A denotes the fillip. (v)′ M(ρ) ≤ ρ ⊗ 1 + 1 ⊗ ρ, where M : A ⊗A → A ⊗A ⊗A denotes the -morphism that puts 1 in the mid position, i.e. M(a ⊗ b) := a ⊗ 1 ⊗ b (a, b ∈ A). 2010 Mathematics Subject Classification. 46L52; 46L85; 81R60. Key words and phrases. Compact quantum metric space; C-algebra; enveloping von Neumann algebra. 1 2 M. M. SADR There are many ways to state the noncommutative analogues of (ii) and (iii). But it seems that the most effective and applicable way is as follows. Let π : A → L(H) denote a representation for A by bounded operators on a Hilbert space H. We say that π is an atomic representation if there is a family {πi : A → L(Hi)} of pairwise inequivalent irreducible representations of A such that π = ⊕πi. (Note that our atomic representations are special cases of the atomic representations defined in [8].) Then it follows from [3, Corollary 10.3.9] that the enveloping von Neumann algebra π(A)′′ is equal to ⊕iL(Hi). Let π : A → L(H) be a faithful atomic representation of A. We consider A as a subalgebra of L(H) and write A′′ for π(A)′′. The characteristic function of the diagonal (w.r.t. π) of qA × qA is denoted by Pδ and defined to be the supremum of the family of all projections of the form p ⊗ p in A′′ ¯⊗A′′ = (A ⊗A)′′ ⊆ L(H ¯⊗H) such that p ∈ A′′ is a minimal projection. (In the classical case that A = C(X ), if we choose π to be the reduced atomic representation then π(C(X ))′′ is isomorphic to ℓ∞(X ) and Pδ is identified with the usual characteristic function of the diagonal of X × X .) The analogues of (ii) and (iii) are as follows. (ii)′ ρPδ = Pδρ = 0. (iii)′ Let Hδ denote the the image of the projection Pδ in H ¯⊗H. Then, 0 is not an eigenvalue of the operator ρH⊥ δ ∈ L(H⊥ δ ). Definition 1. Let A be a unital C-algebra and let π : A → L(H) be a faithful atomic representation. A (compatible) quantum metric w.r.t. π on qA is an element ρ ∈ A ⊗A satisfying (i)′-(v)′. In this case we call (A, ρ, π) a compact quantum metric space. Let (A, ρ, π) be a compact quantum metric space. Comparing with the classical case, it is natural that we consider the value kρk as the diameter of ρ. It is clear that if (X , ρ) is an ordinary compact metric space then (C(X ), ρ, π) is a compact quantum metric space where π is an arbitrary atomic representation of C(X ). (Indeed, it is easily checked that for any of such representation π(C(X ))′′ is isomorphic to ℓ∞(X0) where X0 is a dense subspace of X .) We show that there is no quantum metric on the two-point noncommutative space qM2. Let A = M2 be the algebra of complex 2 × 2 matrices. Then π = id is an atomic repre- sentation of A on H = C2. Let {e1, e2} (resp. {f1, · · · , f4}) denote the Euclidean basis of C2 (resp. C4). We identify C2 ⊗ C2 with C4 via e1 ⊗ e1 7→ f1, e2 ⊗ e2 7→ f4, e1 ⊗ e2 7→ f2, e2 ⊗ e1 7→ f3. Then M2 ⊗ M2 is identified with M4 via Pi,j,k,ℓ λijkℓ1ij ⊗ 1kℓ 7→   λ1111 λ1112 λ1211 λ1212 λ1121 λ1122 λ1221 λ1222 λ2111 λ2112 λ2211 λ2212 λ2121 λ2122 λ2221 λ2222   . With these identifications, Pδ is the projection onto the linear subspace generated by f1, f4, f2+ f3, and hence Pδ =   0 1 0 0 1/2 1/2 0 0 1/2 1/2 0 0 1 0 0 0   . QUANTUM METRICS ON NONCOMMUTATIVE SPACES 3 Suppose that ρ ∈ M4 satisfies (i)′,(ii)′,(iii)′,(iv)′. Then ρ must be of the form ρ =   0 0 0 0 −λ 0 0 0 0 λ −λ 0 0 0 λ 0   , for some real number λ > 0. The 8 × 8 matrix M = ρ ⊗ 1 + 1 ⊗ ρ − M(ρ) is equal to M = λ   0 0 0 0 −1 0 2 0 −1 0 0 0 1 −1 0 0 0 0 0 0 0 0 0 0 1 0 −1 0 0 0 −1 1 0 0 0 0 0 0 0 0 0 0 0 −1 0 0 0 2 −1 0 0 0 0 −1 0 0 0 0 0 0 1 0 0 0   . For any vector X = (x1, · · · , x8) ∈ R8 we have λ−1hMX, Xi = (x3 − x2 − x5)2 + (x2 3 − x2 2 − x2 5) + (x6 − x4 − x7)2 + (x2 6 − x2 4 − x2 7). Thus M is not positive and hence ρ dose not satisfy (v)′. Although we just mentioned a negative result on the existence of quantum metrics but it seems that there must be a huge class of quantum metrics on qMn for n ≥ 3. Indeed we invite and request the readers of this note, specially who are familiar with computational matrix-softwares, to find some explicit examples of quantum metrics on qMn. Similar to the case of ordinary metric spaces we have the following three theorems. Theorem 2. Let ρ1 and ρ2 be quantum metrics on qA w.r.t. the same representation π of A. Then, for every positive real number r, ρ1 + rρ2 is a quantum metric on qA w.r.t. π. Proof. Straightforward. (cid:3) Theorem 3. Let (A1, ρ1, π1) and (A2, ρ2, π2) be compact quantum metric spaces and let r be a real number not less than 2−1 max(kρ1k, kρ2k). Then (A1 ⊕ A2, ρ, π1 ⊕ π2) is a compact quantum metric space where ρ = ρ1 + ρ2 + r1A1 ⊗A2 + r1A2 ⊗A1. Proof. The conditions (i)′ and (iv)′ are trivial for ρ. (ii)′ and (iii)′ follows from the fact that any minimal projection in (π1 ⊕ π2)(A1 ⊕ A2)′′ = π1(A1)′′ ⊕ π2(A2)′′ is a minimal projection of π1(A1)′′ or of π2(A2)′′. Let M,M1,M2 denote the corresponding morphisms as in (v)′ respectively for A1 ⊕ A2,A1,A2. With the notation 1ijk := 1Ai ⊗Ak we have, ⊗Aj M(ρ1) ≤ M1(ρ1) + 2r1121, M(ρ2) ≤ M2(ρ2) + 2r1212. 4 It follows that M. M. SADR M(ρ) = M(ρ1) + M(ρ2) + r1112 + r1122 + r1211 + r1221 ≤ M1(ρ1) + M2(ρ2) + 2r1121 + 2r1212 + r1112 + r1122 + r1211 + r1221 ≤ ρ1 ⊗ 11 + 11 ⊗ ρ1 + ρ2 ⊗ 12 + 12 ⊗ ρ2 + 2r1121 + 2r1212 + r1112 + r1122 + r1211 + r1221 ≤ ρ1 ⊗ 11 + ρ1 ⊗ 12 + ρ2 ⊗ 11 + ρ2 ⊗ 12 + r1121 + r1122 + r1211 + r1212 + 11 ⊗ ρ1 + 12 ⊗ ρ1 + 11 ⊗ ρ2 + 12 ⊗ ρ2 + r1112 + r1212 + r1121 + r1221 = ρ ⊗ 1 + 1 ⊗ ρ. Theorem 4. Let (A1, ρ1, π1) and (A2, ρ2, π2) be compact quantum metric spaces. Then (A1 ⊗A2, ρ, π1 ⊗ π2) is a compact quantum metric space where ρ = (ρ1 ⊗ 1A2 ⊗A2 + 1A1 ⊗A1 ⊗ ρ2) ∈ (A1 ⊗A1) ⊗(A2 ⊗A2) ∼= (A1 ⊗A2) ⊗(A1 ⊗A2). Proof. Straightforward. (cid:3) (cid:3) Let us now consider some relations between our model of "compact quantum metric space" and the other model introduced by Rieffel [6]. Let (A, ρ, π) be a compact quantum metric space. We are able to define a new seminorm on A which generalizes the Lipschitz seminorm for continuous functions on an ordinary metric space. Let H denote the Hilbert space of π and let Hδ be as in (iii)′. Let ρ−1 denote the inverse of the operator ρH⊥ δ ). For any a ∈ A, the Lipschitz seminorm kakLip are defined to be the (possibly infinite) value k(a ⊗ 1 − 1 ⊗ a)ρ−1k, that is the operator norm of (a ⊗ 1 − 1 ⊗ a)ρ−1 as an operator from the image of ρH⊥ into H ¯⊗H. For a, b ∈ A with ab = ba the Leibnitz rule is satisfied: δ ∈ L(H⊥ δ kabkLip = k(ab ⊗ 1 − 1 ⊗ ab)ρ−1k = k(ab ⊗ 1 − a ⊗ b + a ⊗ b − 1 ⊗ ab)ρ−1k = k[(a ⊗ 1)(b ⊗ 1 − 1 ⊗ b) + (a ⊗ 1 − 1 ⊗ a)(1 ⊗ b)]ρ−1k ≤ k(a ⊗ 1)(b ⊗ 1 − 1 ⊗ b)ρ−1k + k(1 ⊗ b)(a ⊗ 1 − 1 ⊗ a)ρ−1k ≤ kakkbkLip + kakLipkbk Also it is clear that for any normal element a we have kakLip = ka∗kLip. The seminorm k·kLip gives rise to a semimetric on the state space S(A) of A via Monge-Kantorovich formula: d(φ, ψ) := sup hφ − ψ, ai (φ, ψ ∈ S(A)). a∗=a,kakLip≤1 We give an upper bound for d(φ, ψ) in the case that φ and ψ are some special pure states of A: Let π be the direct sum of {πi : A → L(Hi)}. Suppose that i 6= j and let v and w be two unit vectors respectively in Hi and Hj. Let φ and ψ be pure states on A defined respectively by a 7→ hπi(a)v, vi and a 7→ hπj(a)w, wi. Let a be a self-adjoint element of A QUANTUM METRICS ON NONCOMMUTATIVE SPACES 5 with kakLip ≤ 1. Since v ⊗ w ∈ H⊥ δ , we have ka(v) ⊗ w − v ⊗ a(w)k ≤ kρ(v ⊗ w)k. Thus, hφ − ψ, ai2 = ha(v), vi2 + ha(w), wi2 − 2ha(v), viha(w), wi ≤ ha(v), a(v)i + ha(w), a(w)i − 2ha(v), viha(w), wi = ha(v), a(v)ihw, wi + ha(w), a(w)ihv, vi − 2ha(v), viha(w), wi = ka(v) ⊗ w − v ⊗ a(w)k2 ≤ kρ(v ⊗ w)k2. This shows that d(φ, ψ) ≤ kρ(v ⊗ w)k. As we saw above the most problematic part of the definition of a quantum metric is the translation of conditions (ii) and (iii). We now translate these conditions in another way where there is no using of enveloping von Neumann algebras. Let A be a unital spatially continuous multiplication C-algebra, that means the multiplication of A, m : a ⊗ b 7→ ab, is continuous w.r.t. spatial tensor norm (e.g. A is abelian or finite dimensional). For ρ ∈ A ⊗A satisfying (i)′, consider the following conditions. (ii)′′ m(ρ) = 0. (iii)′′ For every positive element ν ∈ A ⊗A with m(ν) = 1 and F(ν) = ν, the element ρ + ν is invertible in A ⊗A. In the case that A = C(X ) it is easily checked that these conditions coincide with (ii),(iii). Definition 5. Let A be a unital spatially continuous multiplication C-algebra. An element ρ ∈ A ⊗A which satisfies (i)′,(ii)′′, (iii)′′,(iv)′,(v)′ is called an algebraic (compatible) quantum metric on qA. In this case (A, ρ) is called an algebraic compact quantum metric space. Theorem 6. Let (A1, ρ1) and (A2, ρ2) be algebraic compact quantum metric spaces. Then (A1 ⊕ A2, ρ) is an algebraic compact quantum metric space where ρ is as in Theorem 3. Proof. We only show that ρ satisfies (iii)′′. The other conditions are easily checked. Let A := A1 ⊕ A2. Let m, m1, m2 denote respectively the multiplications of A, A1, A2 and let F, F1, F2 denote the corresponding fillips as in (iv)′. We have A ⊗A = ⊕i,j=1,2Ai ⊗Aj. Let ν be a positive element of A ⊗A with m(ν) = 1 and F(ν) = ν. Let νij ∈ Ai ⊗Aj be such that ν = P νij. Then νij is positive and we have mi(νii) = 1A1 and Fi(νii) = νii. It follows that ρi + νii is invertible in Ai ⊗Ai, and r1Ai ⊗ 1Aj + νij is invertible in Ai ⊗Aj for i 6= j. Thus ρ + ν is invertible in A ⊗A. (cid:3) Theorem 7. Let (A1, ρ1) and (A2, ρ2) be algebraic compact quantum metric spaces such that A1 is commutative. Then (A1 ⊗A2, ρ) is an algebraic compact quantum metric space where ρ is as in Theorem 4. Proof. We only show that ρ satisfies (iii)′. The other conditions are easily checked. Let m, m2 denote respectively the multiplications of A, A2 and let F, F2 denote the corresponding fillips as in (iv)′. Let X denote the Gelfand spectrum of A1. Thus A1 ∼= C(X ) and A1 ⊗A2 ∼= C(X , A2), the algebra of A2 valued continuous functions on X . Let ν ∈ C(X × X , A2 ⊗A2) be a positive element with m(ν) = 1 and F(ν) = ν. Then for every x, y ∈ X , ν(x, y) is positive, m2(ν(x, y)) = 1A2, and F2(ν(x, y)) = ν(x, y). Thus ρ2 + ν(x, y) is invertible in A2 ⊗A2. It follows that 1A1 ⊗A1 ⊗ ρ2 + ν (which is equal to the function (x, y) 7→ ρ2 + ν(x, y)) is invertible. Thus ρ + ν is also invertible. (cid:3) 6 M. M. SADR The main gap in our work is the lack of a nonclassical example: Problem 8. Give an example of a nonclassical (algebraic) quantum metric. References 1. A. Connes, Noncommutative geometry, Academic Press, 1994. 2. A. Connes, On the spectral characterization of manifolds, J. Noncommutative Geometry 7, no. 1 (2013): 1 -- 82. (arXiv:0810.2088 [math.OA]) 3. R.V. Kadison, J.R. Ringrose, Fundamentals of the theory of operator algebras: Advanced theory, Vol. 2, American Mathematical Soc., 1997. 4. G. Kuperberg, N. Weaver, A von Neumann algebra approach to quantum metrics/quantum relations, Vol. 215, no. 1010. American Mathematical Society, 2012. (arXiv:1005.0353 [math.OA]) 5. M.A. Rieffel, Metrics on state spaces, Doc. Math. 4 (1999): 559-600. (arXiv:math/9906151 [math.OA]) 6. M.A. Rieffel, Group C*-algebras as compact quantum metric spaces, Doc. Math 7 (2002): 605 -- 651. (arXiv:math/0205195 [math.OA]) 7. M.A. Rieffel, Gromov-Hausdorff distance for quantum metric spaces/Matrix algebras converge to the sphere for quantum Gromov-Hausdorff distance, Vol. 168, no. 796. American Mathematical Soc., 2004. (arXiv:math/0011063 [math.OA]) (arXiv:math/0108005 [math.OA]) 8. M. Takesaki, Theory of operator algebras I, Reprint of the first (1979) edition, Encyclopaedia of Mathe- matical Sciences, 124, Operator Algebras and Noncommutative Geometry, 5. (2002). Department of Mathematics, Institute for Advanced Studies in Basic Sciences, P.O. Box 45195-1159, Zanjan 45137-66731, Iran E-mail address: [email protected]
1807.10424	1	1807	2018-07-27T03:45:59	Inductive limits of C*-algebras and compact quantum metrics spaces	[ "math.OA", "math.FA" ]	Given a unital inductive limit of C-algebras for which each C-algebra of the inductive sequence comes equipped with a compact quantum metric of Rieffel, we produce sufficient conditions to build a compact quantum metric on the inductive limit from the quantum metrics on the inductive sequence by utilizing the completeness of the dual Gromov-Hausdorff propinquity of Latremoliere on compact quantum metric spaces. This allows us to place new quantum metrics on all unital AF algebras that extend our previous work with Latremoliere on unital AF algebras with faithful tracial state. As a consequence, we produce a continuous image of the entire Fell topology on the ideal space of any unital AF algebra in the dual Gromov-Hausdorff propinquity topology.	math.OA	math	INDUCTIVE LIMITS OF C-ALGEBRAS AND COMPACT QUANTUM METRICS SPACES KONRAD AGUILAR ABSTRACT. Given a unital inductive limit of C-algebras for which each C-algebra of the inductive sequence comes equipped with a compact quantum metric of Ri- effel, we produce sufficient conditions to build a compact quantum metric on the inductive limit from the quantum metrics on the inductive sequence by utilizing the completeness of the dual Gromov-Hausdorff propinquity of Latrémolière on compact quantum metric spaces. This allows us to place new quantum metrics on all unital AF algebras that extend our previous work with Latrémolière on unital AF algebras with faithful tracial state. As a consequence, we produce a continuous image of the entire Fell topology on the ideal space of any unital AF algebra in the dual Gromov-Hausdorff propinquity topology. CONTENTS Introduction and Background 1. 2. Quantum metrics on inductive limits by Cauchy sequences 3. All unital AF algebras are propinquity approximable References 1 5 14 23 1. INTRODUCTION AND BACKGROUND Noncommutative Metric Geometry introduced by Rieffel [33,40] and motivated by work of Connes [11, 12] provides a solid framework for producing continu- ous families of quantum metric spaces, which are, in part, C-algebras. This was done by producing a distance on the class of quantum metric spaces [40], which is analogous to the Gromov-Hausdorff distance between compact metric spaces. The first example of a continuous family in this metric was quantum tori with respect to their standard parametric space due to Rieffel [40]. Recently, Latré- molière developed a distance on certain classes of quantum metric spaces, called the Gromov-Hausdorff propinquity [25, 28], which allows one to capture the C- algebraic structure as well as the quantum metric structure while also establishing continuity results about Hilbert C-modules, group actions, and vector bundles associated to quantum metric spaces [20 -- 23, 38, 39]. Date: July 30, 2018. 2010 Mathematics Subject Classification. Primary: 46L89, 46L30, 58B34. Key words and phrases. Noncommutative metric geometry, Monge-Kantorovich distance, Quantum Metric Spaces, Lip-norms, inductive limits, AF algebras. 1 2 KONRAD AGUILAR This article focuses on the categorical notion of convergence of C-algebras given by inductive sequences of C-algebras and their inductive limits, which is a crucial topic of research in C-algebras due, in part, to the Elliott classification program that began in [16] where Elliott addressed the classification of inductive limits whose inductive sequences comprised of finite-dimensional C-algebras. Due to the inception of Noncommutative Metric Geometry, the question of when this categorical notion of convergence passes to a metric convergence arose nat- urally. But, first, to discuss convergence of C-algebras, one must first equip the C-algebras with the quantum metrics. There are many C-algebras that have been equipped with quantum metrics including: certain Group C-algebras [10, 32, 35], quantum tori and fuzzy tori [24, 33], noncommutative solenoids [29], quantum Podle´s sphere with D abrowski-Sitarz spectral triple [3, 13], and many more in- cluding untial AF algebras [1, 6]. Now, in the case of unital AF algebras equipped with faithful tracial states, in [5], quantum metrics were placed in such a way that any inductive sequence that formed the given unital AF algebra converges to the AF algebra in the metric sense of propinquity so that the authors were able to also establish convergence of classes of AF algebras such as UHF algebras [19] and Effros-Shen algebras [15]. However, the types of quantum metrics built in the faithful tracial state case to acheive such convergence results have not been extended to all unital AF algebras until this article (see Theorem 3.4 and Propo- sition 3.5) even though there are quantum metrics on all unital AF algebras as mentioned above. And, we apply these findings to capture the entire Fell topol- ogy on the ideal space on any given unital AF algebra in the propinquity topology by way of a continuous map in Theorem 3.9, which was not possible before since unital AF algebras may have ideals without faithful tracial states. This shows that the propinquity topology is diverse enough to capture this vital topology. However, our approach to produce these new quantum metrics on unital AF algebras is inspired by answering a more general question about quantum met- rics on inductive limits that need not necessarily be AF. In every case of quantum metrics on inductive limits, the quantum metric is built directly on the inductive limit, which indirectly builds quantum metrics on the terms of the inductive se- quence. In our experience and as seen in this article, it seems that in order to push results about quantum metrics on inductive limits to new avenues, one must have the ability to build quantum metrics on the inductive limit from suitable quan- tum metrics on the C-algebras of the inductive sequence just as the C-norm on an inductive limit of C-algebras is constructed from the the C-norms on the C- algebras of the inductive sequence. Thus, in this paper, we accomplish this task under suitable sufficient conditions to obtain our new quantum metrics on AF algebras. The main fact we use is that the Dual Gromov-Hausdorff Propinqiuity [25] is complete on certain classes of compact quantum metric spaces, and thus be- stows a method for forming limits of quantum metric spaces. Of course, this limit may not be an inductive limit of quantum metric spaces in any categorical sense, but the idea is to combine the categorical notion of inductive limit of C-algebras with the metric limit formed by completeness with respect to propinquity. This is done in Theorem 2.15 by using Cauchy sequences from inductive sequences of INDUCTIVE LIMITS AND QUANTUM METRICS 3 C-algebras in dual propinquity that exploit the C-inductive limit structure. We now introduce some background before we move onto our main results. Definition 1.1 ([33,34,36]). A compact quantum metric space (A, L) is an ordered pair where A is a unital C-algebra with unit 1A and L is a seminorm over R defined on sa (A) whose domain dom (L) = {a ∈ sa (A) : L(a) < ∞} is a unital dense subspace of sa (A) over R such that: (1) {a ∈ sa (A) : L(a) = 0} = R1A, (2) the Monge-Kantorovich metric defined, for all two states ϕ, ψ ∈ S (A), by: mkL(ϕ, ψ) = sup {ϕ(a) − ψ(a) : a ∈ dom (L), L(a) 6 1} metrizes the weak topology of S (A), and (3) the seminorm L is lower semi-continuous with respect to k · kA. If (A, L) is a compact quantum metric space, then we call the seminorm L a Lip- norm and we denote the diameter of the compact metric space (S (A), mkL) by diam (A, L). Definition 1.2 ([27]). A (C, D)-quasi-Leibniz compact quantum metric space (A, L), for some C ∈ R, C > 1 and D ∈ R, D > 0, is compact quantum metric space such that the seminorm L is (C, D)-quasi-Leibniz, i.e. for all a, b ∈ dom (L): max {L (a ◦ b) , L ({a, b})} 6 C (kakAL(b) + kbkAL(a)) + DL(a)L(b), where a ◦ b = ab+ba is the Jordan product and {a, b} = ab−ba 2 is the Lie product. 2i When C = 1, D = 0, we call L a Leibniz Lip-norm. When we do not specify C and D, we call (A, L) a quasi-Leibniz compact quantum metric space. Next, we define the notion of isomorphism for compact quantum metric spaces, formerly called an isometric isomorphism. Definition 1.3 ([25, Definition 2.20]). Let (A, LA), (B, LB) be two compact quan- tum metric spaces. A -isomorphism π : A → B is a quantum isometry if LB ◦ π = LA. Before we state the theorem that Latrémolière's dual propinquity is a complete metric up to quantum isometry, we first define a crucial object used to provide estimates for the dual propinquity. Definition 1.4 ([27, Definition 2.25]). Fix C, > 1, D > 0. Let (A1, L1), (A2, L2) be (C, D)-quasi-Leibniz compact quantum metric spaces. A (C, D)-tunnel τ from (A1, L1) to (A2, L2) is a 4-tuple τ = (D, LD, π1, π2), where (D, LD) is a (C, D)- quasi-Leibniz compact quantum metric space and for all j ∈ {1, 2}, the map πj : D → Aj is a unital -epimorphism such that for all a ∈ sa(cid:0)Aj(cid:1) : Lj(a) = inf{LD(d) : d ∈ sa (D) and πj(d) = a}. We take the following theorem from [27] since it adapts the dual propinquity from [25] to the quasi-Leibniz case. Theorem 1.5 ([27, Definition 2.23, Theorem 2.28]). Fix C > 1, D > 0. There exists a pseudo-metric on the class of (C, D)-quasi-Leibniz compact quantum metric spaces called the dual Gromov-Hausdorff propinquity, denoted by Λ∗, such that for any two (C, D)- quasi-Leibniz compact quantum metric spaces (A, LA), (B, LB) 4 KONRAD AGUILAR (1) Λ∗((A, LA), (B, LB)) = 0 if and only if (A, LA) and (B, LB) are quantum iso- metric, and thus Λ∗ is a metric up to the equivalence relation of quantum isometry, (2) Λ∗((A, LA), (B, LB)) 6 2λ(τ), where τ is any (C, D)-tunnel from (A, LA) to (B, LB) and λ(τ) is the length of the tunnel (see [27, Definition 2.25]), and furthermore Λ∗ is a complete metric on the class of (C, D)-quasi-Leibniz compact quantum metric spaces up to quantum isometry. We note futher that for any fixed (C, D), the dual propinquity is a noncommu- tative analogue of the Gromov-Hausdorff distance on the class of compact metric spaces up to isometry since one can embed this class homeomorphically into the class of (C, D)-quasi-Leibniz compact quantum metric up to quantum isometry with respect to the topology induced by Λ∗ (see [27, Theorem 2.28]). Although we do not provide the definition of the length of a tunnel in this background, we provide a useful tool for bestowing estimates of the lengths of certain tunnels that arise in this paper. This tool was key in defining the quantum Gromov-Hausdorff propinquity, which is another metric on the class of (C, D)- quasi-Leibniz compact quantum metric up to quantum isometry that is likely not complete, but still has many applications, which includes finding estimates for the dual propinquity. Definition 1.6 ([28, Definition 3.6, Lemma 3.4]). Let A, B be two unital C-algebras. A bridge γ from A to B is a 4-tuple γ = (D, ω, πA, πB) such that (1) D is a unital C-algebra and ω ∈ D, (2) the set S1(ω) = {ψ ∈ S (D) : ∀d ∈ D, ψ(d) = ψ(ωd) = ψ(dω)} is non-empty, in which case ω is called the pivot, and (3) πA : A → D and πB : B → D are unital -monomorphisms. The next lemma produces a characterization of lengths of the types of bridges that appear in this article, which follows immediately from definition. But, first, we introduce a definition for the types of bridges that appear in this article. Definition 1.7. Let A be a unital C-algebra, and let B ⊆ A be a unital C-subalgebra of A. We call the 4-tuple (A, 1A, ι, idA) the evident bridge from B to A, where ι : B → A is the inclusion mapping and idA : A → A is the identity map. Lemma 1.8 ([28, Definition 3.17]). Let (A, LA), (B, LB) be two compact quantum met- ric spaces. If a bridge γ from A to B is of the form γ = (D, 1D, πA, πB), then the length of the bridge is λ(γLA, LB) = max supb∈sa(B),LB(b)61ninfa∈sa(A),LA(a)61 {kπA(a) − πB(b)kD}o  supa∈sa(A),LA(a)61ninfb∈sa(B),LB(b)61 {kπA(a) − πB(b)kD}o , In particular, this holds for evident bridges. Next, we see how lengths of bridges can be used to estimate lengths of certain tunnels. We note that the length of any bridge between two compact quantum metric spaces is finite (see the discussion preceding [28, Definition 3.14]). INDUCTIVE LIMITS AND QUANTUM METRICS 5 Theorem 1.9 ([26, Theorem 3.48]). Fix C > 1, D > 0. Let (A, LA), (B, LB) be two (C, D)-quasi-Leibniz compact quantum metric spaces. Let γ = (D, ω, πA, πB) be a bridge from A to B. Fix any r > λ(γLA, LB), where λ(γLA, LB) is the length of the bridge γ. If we define for all (a, b) ∈ sa (A ⊕ B) Lr γLA,LB (a, b) = max(cid:26)LA(a), LB(b), kπA(a)ω − ωπB(b)kD r (cid:27) and we let pA : (a, b) ∈ A ⊕ B → a ∈ A and pB : (a, b) ∈ A ⊕ B → b ∈ B denote the canonical surjections, then τ = (A ⊕ B, Lr , pA, pB) is a (C, D)-tunnel from (A, LA) to (B, LB) with length λ(τ) 6 r, and γLA,LB Λ∗ ((A, LA) , (B, LB)) 6 2r. This allows us to define: Definition 1.10. Fix C > 1, D > 0. Let (A, LA), (B, LB) be two (C, D)-quasi- Leibniz compact quantum metric spaces. Let γ = (D, ω, πA, πB) be a bridge from A to B. We call the (C, D)-tunnel (A ⊕ B, Lr , pA, pB) from (A, LA) to (B, LB) of Theorem 1.9 the (r, γLA, LB)-evident tunnel associated to the bridge γ, Lip-norms LA, LB, and r > λ(γLA, LB). γLA,LB 2. QUANTUM METRICS ON INDUCTIVE LIMITS BY CAUCHY SEQUENCES We now give a name to when a Cauchy sequence of quasi-Leibniz compact quantum metric spaces coming from an inductive sequence of C-algebras does produce the inductive limit in the propinquity limit. In this section, we provide natural sufficient conditions for producing such a result in Theorem 2.15 by har- nessing the indcutive sequence structure. The following definition aims to de- scribe when a Cauchy sequence of quasi-Leibniz compact quantum metric spaces is compatible with C-algebraic structure of an inductive limt. First, we provide a convention for inductive sequences and limits of C-algebras. k·kA is a unital inductive limit of Convention 2.1. A unital C-algebra A = ∪n∈NAn C-algebras if (An)n∈N is a non-decreasing sequence of unital C-subalgebras of A. The above convention captures all unital inductive limits of C-algebras up to -isomorphism, so there is no loss of generality by making such a convention [31, Section 6.1]. k·kA be a unital inductive limit Definition 2.2. Fix C > 1, D > 0. Let A = ∪n∈NAn If ((An, LAn ))n∈N is a Cauchy sequence in dual propinquity of of C-algebras. (C, D)-quasi-Leibniz compact quantum metric spaces with limit (B, LB), then we call the sequence ((An, LAn ))n∈N an A−C-convergent sequence if A is -isomorphic to B. From this definition, we see that the inductive limit A itself will be a (C, D)- quasi-Leibniz compact quantum metric space and a limit to the given Cauchy se- qeunce in dual propinquity, and summarize this in: 6 KONRAD AGUILAR k·kA be a unital inductive limit Proposition 2.3. Fix C > 1, D > 0. Let A = ∪n∈NAn of C-algebras such that ((An, LAn))n∈N is a Cauchy sequence of (C, D)-quasi-Leibniz compact quantum metric spaces with dual propinquity limit (B, LB). If ((An, LAn ))n∈N is a A−C-convergent for some -isomorphism π : A → B, then LB A := LB ◦ π is a (C, D)-quasi-Leibniz Lip-norm on A such that Λ∗(cid:16)(B, LB) ,(cid:16)A, LB that Λ∗(cid:0)(B, LB) ,(cid:0)A, LB Λ∗(cid:16)(An, LAn ) ,(cid:16)A, LB A(cid:17)(cid:17) = 0 and lim A(cid:1)(cid:1) = 0 since π is a quantum isometry by construction. A is a (C, D)-quasi-Leibniz Lip-norm on A such A(cid:17)(cid:17) = 0. The convergence result follows from the triangle inequality. Proof. It is routine to check that LB n→∞ (cid:3) A key observation is that: it can be the case that an inductive limit of an inductive se- quence of C-algebras equipped with quantum metrics, which happen to produce a Cauchy sequence in dual propinquity, need not be -isomorphic to the C-algebra of the limit in the dual propinquity. Let us now provide an example of non-C-convergent sequence built from the CAR algebra [14, Example III.5.4] to motivate sufficient conditions that produce C-convergent sequences. We note that the following example is mo- tivated by work in [4]. Example 2.4 (A non-C-convergent sequence). Consider the inductive limit, the k·kM2∞ [14, Example III.5.4], where (M2∞)n ∼= CAR algebra M2∞ = ∪n∈N(M2∞ )n M2n(C) for all n ∈ N. For each n ∈ N, define L (M2∞ )n(a) = dim((M2∞)n) · ka − τ(a)1M2∞ kM2∞ for all a ∈ (M2∞)n, where τ is the unique faithful tracial state on M2∞ . By [5, Theorem 3.5], the pair(cid:16)(M2∞)n, L space. (M2∞ )n(cid:17) is a (2, 0)-quasi Leibniz compact quantum metric Now, consider σn : λ ∈ C 7→ λ1(M2∞ )n ∈ (M2∞)n and σn(C) with its unique Lip-norm LC, the 0-seminorm, which is also (2, 0)-quasi-Leibniz. Clearly σn(C) 6∼= M2∞. Fix n ∈ N, consider the evident bridge from σn(C) to (M2∞)n (Definition 1.7). Next, let λ ∈ σn(C) (so LC(λ) = 0 6 1), then kλ − λkM2∞ = 0 where L(M2∞ )n(λ) = 0. Now, let a ∈ (M2∞)n such that L(M2∞ )n(a) 6 1, then kσn(τ(a)) − akM2∞ 6 1/ dim((M2∞)n) where LC(σn(τ(a)) = 0 6 1. Thus, by Lemma 1.8, the length of the bridge is less than or equal to 1/ dim((M2∞)n). Hence, by and Theorem 1.9, we have that Λ∗(cid:16)(C, LC) ,(cid:16)(M2∞)n, L and thus limn→∞ Λ∗(cid:16)(C, LC) ,(cid:16)(M2∞)n, L quence(cid:16)(cid:16)(M2∞)n, L (M2∞ )n(cid:17)(cid:17) is not M2∞ −C-convergent. (M2∞ )n(cid:17)(cid:17) 6 (M2∞ )n(cid:17)(cid:17) = 0. Thus, the Cauchy se- dim((M2∞)n) The above example is interesting on its own as it reflects the fact that matri- ces can approximate any unital commutative C-algebra in the Gromov-Hausdorff propinquity [4]. However, this is not suitable for our current pursuits. Now, back to our main goal, it should be noted that there do exist Lip-norms L′ n on (M2∞)n such that the sequence (((M2∞)n, L′ n))n∈N is Cauchy in dual propinquity and is M2∞−C-convergent by [5, Theorem 3.5]. We would now like to provide some 4 INDUCTIVE LIMITS AND QUANTUM METRICS 7 sufficient conditions to build C-convergent sequences. The above example can motivate such conditions. Indeed, consider the inclusion mapping ιn : (M2∞)n → (M2∞)n+1 and let a ∈ (M2∞ )n. We have L (M2∞ )n+1 (ιn(a)) = dim((M2∞)n+1) · ka − τ(a)1M2∞ kM2∞ = dim((M2∞)n+1) dim((M2∞)n) L(M2∞ )n(a). (M2∞ )n+1 (ιn(a)) 66 L (M2∞ )n(a). Thus, the canonical embeddings are not Therefore L contractive with respect to the Lip-norms, which is not very compatible with the notion of an inductive limit. For instance, the embeddings for an inductive se- quence of C-algebras are contractive with respect to the C-norms. Thus, requir- ing contractivity seems to be a desirable condition along with the fact that the Lip- norms on AF algebras in [1, 5, 6] are all contractive. Furthermore, the Lip-norms in [5] allowed for explicit upper bounds on the lengths of the evident bridges of Definition 1.7 in dual propinquity. From this, the authors showed convergence of AF algebras and not just convergence of the inductive sequences that formed the AF algebras. However, in [5], these methods only worked for AF algebras with faithful tracial states. Thus, as a consequence of this paper and Section 3 and the following definition, we will have similar convergence results for all AF algebras with or without faithful tracial state by building Lip-norms on inductive limits from the Lip-norms on the terms of the inductive sequence. Of course, the Lip- norms on the inductive limits of [5] were built explicitly on the inductive limit, so our Lip-norms in this paper would be redundant in the faithful tracial state case and up to passing to a subsequence these inductive sequences satisfy the fol- lowing definition automatically, and of course C-convergent with respect to the given inductive limits. Thus, another main purpose of this paper is bestow a method to construct Lip-norms on inductive limits from Lip-norms on certain inductive sequences without knowledge of any quantum metric structure on the inductive limit itself. k·kA be a unital inductive limit Definition 2.5. Fix C > 1, D > 0. Let A = ∪n∈NAn of C-algebras. Let ((An, LAn))n∈N be a sequence of (C, D)-quasi-Leibniz com- pact quantum metric spaces. We the call the inductive limit A an ((An, LAn ))n∈N- propinquity approximable inductive limit if the following hold for each n ∈ N: (1) there exists a (C, D)-quasi-Leibniz Lip-norm LAn for An such that LAn is defined on An and {a ∈ An : LAn (a) < ∞} is a dense -subalgebra of An, (2) it holds that if a ∈ An, then LAn+1(a) 6 LAn (a), and (3) there exists a sequence (β(j))j∈N ⊂ (0, ∞) such that ∑∞ j=0 β(j) < ∞ and the length of evident bridge γn,n+1 = (An+1, 1A, ιn,n+1, idn+1) of Definition (1.7), satisfies λ(γn,n+1LAn, LAn+1) 6 β(n), and we denote the associated (2β(n), γn,n+1LAn, LAn+1)−evident tunnel of Definition 1.10 by τn,n+1. We note that the above definition satisifies the notion of an inductive sequence in a certain category of compact quantum metric spaces defined in [23, Definition 1.9], but so does the sequence in Example 2.4. Thus, the above definition is an at- tempt to provide further criteria to allow the inductive sequence to form some sort 8 KONRAD AGUILAR of limit even though it is not an inductive limit. In particular, the purpose of this definition is to focus on a situation that is capable of allowing for convergence of inductive limits themselves by providing explicit estimates from the inductive se- quences as seen in Theorem 2.16. Approximability occured in and was motivated by the work in [5]. However, we recall that this only occured in the case for AF al- gebras equipped with faithful tracial states. Approximable quantum metrics have not yet been provided for AF algebras outside the faithful tracial state case even by the Lip-norms on all AF algebras built from quotient norms in [1] and spectral triples in [6]. The fact that the inductive sequences converge to the inductive limit in these cases came from a compactness argument, which did not provide explicit estimates assocaited to evident bridges (see [1, Theorem 4.10 and Remark 4.11]). Yet, there are still advantages to these quantum metrics as they preserve more al- gebraic structure than the ones of this paper and [5] since they are strongly Leibniz of [37, Definition 2.1] and have domains that preserve taking inverses; hence, these Lip-norms of [1, 6] still require more investigation, which lies outside the context of this paper. We now begin the journey to show that Definition 2.5 gifts C-convergent se- quences. We begin with: Proposition 2.6. Fix C > 1, D > 0. Let A = ∪n∈NAn C-algebras. k·kA be a unital inductive limit of If A is ((An, LAn))n∈N-propinquity approximable for some sequence of (C, D)-quasi- Leibniz compact quantum metric spaces and summable (β(j))j∈N ⊂ (0, ∞), then: (1) the sequence ((An, LAn ))n∈N is Cauchy in dual propinquity, where for n ∈ N Λ∗(cid:0)(An, LAn ) ,(cid:0)An+1, LAn+1(cid:1)(cid:1) 6 2λ(τn,n+1) 6 4β(n), and (2) given any (C, D)-quasi-Leibniz limit (B, LB) up to quantum isometry of the Cauchy sequence ((An, LAn ))n∈N, it holds that Λ∗ ((An, LAn) , (B, LB)) 6 4 ∞ ∑ j=n β(j) for all n ∈ N. Proof. The inequalities of (1) follow immediately from Theorem 1.9. The fact that ((An, LAn ))n∈N is Cauchy follows from the fact that (β(j))j∈N is summable. Con- clusion (2) follows from the triangle inequality. (cid:3) Thus, if we find a limit (B, LB) of ((An, LAn))n∈N such that A ∼= B, then ((An, LAn ))n∈N will be A−C-convergent. Thankfully, Latrémolière provided a succinct construction of a limit of certain Cauchy sequences in dual propinquity in [25, Section 6], which Latrémolière used to prove that the dual propinquity is complete. Hence, we now provide a summary of the construction. Notation 2.7 ([25, Hypothesis 6.2, Lemma 6.19, and Lemma 6.20]). Fix C > 1, D > 0. Let ((An, LAn ))n∈N be sequence of (C, D)-quasi-Leibniz compact quantum met- ric spaces. For each n ∈ N, let τAn = (Dn, Ln, πn, ωn) be a (C, D)-tunnel from (An, LAn) to(cid:0)An+1, LAn+1(cid:1) and denote τAN = (τAn)n∈N. We consider D := ∏n∈N Dn as the unital C-algebra of bounded sequences. INDUCTIVE LIMITS AND QUANTUM METRICS 9 Define the seminorm S0 : (dn)n∈N ∈ sa ∏ n∈N Dn! 7−→ sup{Ln(dn) : n ∈ N} and K0 := {d = (dn)n∈N ∈ sa (D) (∀n ∈ N)πn+1(dn+1) = ωn(dn)} L0 := {d = (dn)n∈N ∈ K0 S0(d) < ∞} We denote Re(d) = (d + d∗)/2 and Im(d) = (d − d∗)/(2i) for all d ∈ ∏n∈N Dn. Define S0 := {d = (dn)n∈N ∈ D Re(d), Im(d) ∈ L0} and G0 := S0 k·kD. Next, define the subspace I0 :=n(dn)n∈N ∈ G0 lim n→∞ kdnkDn = 0o , and Now, let q : G0 → FτAN be the quotient map and define for all a ∈ sa(cid:0)FτAN(cid:1) (a) := inf{S0(d) : d ∈ sa (G0) and q(d) = a}. LτAN FτAN := G0/I0. With this, we state the key result of [25] that leads immediately to the proof of completeness of the dual propinquity. Theorem 2.8 ([25, Section 6, Proposition 6.26], [27, Theorem 2.28]). Fix C > 1, D > 0. Let ((An, LAn ))n∈N be sequence of (C, D)-quasi-Leibniz compact quantum metric spaces. For each n ∈ N, let τAn = (Dn, Ln, πn, ωn) be a (C, D)-tunnel from (An, LAn ) n=0 λ(τAn) < ∞, then: Using Notation 2.7, if ∑∞ (1) G0 is a unital C-subalgebra of D, (2) I0 is a closed two-sided ideal of G0, and thus FτAN is a unital C-algebra, to(cid:0)An+1, LAn+1(cid:1). (3) (cid:16)FτAN , LτAN(cid:17) is a (C, D)-quasi-Leibniz compact quantum metric space, (4) the sequence ((An, LAn ))n∈N is Cauchy in dual propinquity, and lim n→∞ Λ∗(cid:16)(An, LAn ) ,(cid:16)FτAN , LτAN(cid:17)(cid:17) = 0. Next, we gather more detail about the objects of Notation 2.7 in the setting of approximable inductive limits. But, first we state some hypotheses we will use in the next few theorems and definitions. k·kA be a unital inductive Hypthesis 2.9. Fix C > 1, D > 0. Let A = ∪n∈NAn limit of C-algebras. Let A be ((An, LAn))n∈N-propinquity approximable for some sequence of (C, D)-quasi-Leibniz compact quantum metric spaces and summable (β(j))j∈N ⊂ (0, ∞). Let τAn = τn,n+1 of Notation 2.7 be the (2β(n), γn,n+1LAn, LAn+1)-evident tun- nel from (An, LAn) to (An+1, LAn+1) of Definition 2.5 for each n ∈ N. Proposition 2.10. If we assume Hypothesis 2.9, then from Notation 2.7 n+1))n∈N ∈ sa (∏n∈N An ⊕ An+1) (∀n ∈ N)an n, an (1) K0 =n((an n+1 = an+1 n+1o, 10 KONRAD AGUILAR (2) 1D = 1G0 = ((1A, 1A))n∈N and for all d = ((an n+1))n∈N ∈ K0, we have n, an n − an+1 2β(n) n+1(cid:13)(cid:13)(cid:13)A LAn (an S0(d) = sup max  (3) and limn→∞ Λ∗(cid:16)(An, LAn ) ,(cid:16)FτAN , LτAN(cid:17)(cid:17) = 0, where for n ∈ N, n) ,(cid:13)(cid:13)(cid:13)an n∈N Λ∗(cid:16)(An, LAn ) ,(cid:16)FτAN , LτAN(cid:17)(cid:17) 6 4  ∞ ∑ j=n β(j). , Proof. Apply Definition 2.5 to Notation 2.7 to obtain conclusions (1) and (2). Con- clusion (3) follows immediately from Proposition 2.6 and Theorem 2.8. (cid:3) The above proposition shows us that FτAN is beginning to look like an induc- tive limit of C-algebras itself given Definition 2.5. Thus, next, we begin building our -isomorphism from A onto FτAN . To do this, we follow the standard method for providing -isomorphisms from inductive limits by universality [31, 6.1.2 The- orem]. Definition 2.11. Assuming Hypothesis 2.9, define ψ0 : A0 → D by and for n ∈ N \ {0}, define ψn : An → D by ψ0(a0) = ((a0, a0))n∈N, ψn(an) = ((0, 0), . . . , (0, 0), (0, an), (an, an), (an, an), . . .), where (0, an) ∈ Dn−1 = An−1 ⊕ An. Lemma 2.12. Assuming Hypothesis 2.9, the map ψn : An → D is a -monomorphism such that ψn(An) ⊆ G0 for all n ∈ N. Proof. Fix n ∈ N. The fact that ψn is a -monomorphism is clear. Let a ∈ An such that LAn (a) < ∞. If a ∈ C1A, then S0(ψn(a)) = 0 < ∞. So, assume a 6∈ C1A. Thus LAn (Re(a)) < ∞, LAn(Im(a)) < ∞ by (1) of Definition 2.5. In particular, since ψn(An) ⊆ K0 by Proposition 2.10 , we have S0(Re(ψn(a))) = max{LAn (Re(a)), kRe(a)kA/(2β(n − 1))} < ∞ and similarly S0(Im(ψn(a))) < ∞ by Proposition 2.10 and (2) of Definition 2.5. Therefore, by (1) of Definition 2.5 ψn(An) = ψn(cid:18){a ∈ An : LAn (a) < ∞} k·kA(cid:19) k·kD ⊆ ψn ({a ∈ An : LAn (a) < ∞}) ⊆ S0 k·kD = G0. by continuity. This lemma allows us to define: (cid:3) Definition 2.13. Assuming Hypothesis 2.9, for each n ∈ N, by Lemma 2.12 we may define ψ(n) : An → FτAN by where q : G0 → FτAN is the quotient map. ψ(n) := q ◦ ψn INDUCTIVE LIMITS AND QUANTUM METRICS 11 Lemma 2.14. Assuming Hypothesis 2.9, the map ψ(n) : An → FτAN is a unital -mono- morphism for each n ∈ N. Furthermore ψ(n) = ψ(n+1) on An for all n ∈ N. Proof. Fix n ∈ N. The map ψ(n) is clearly a -homomorphism. For unital, we note that ψn(1A) − 1D ∈ I0, and thus ψ(n)(1A) = 1F τAN . For injectivity, assume a, b ∈ An and ψ(n)(a) = ψ(n)(b). Thus ψn(a) − ψn(b) ∈ I0. Hence 0 = limn→∞ ka − bkA = ka − bkA which implies a = b. Finally, let a ∈ An ⊆ An+1. Then, again we have ψn(a) − ψn+1(a) ∈ I0, and thus ψ(n)(a) = ψ(n+1)(a). (cid:3) Next, we prove the main theorem of this section, where we see all the notions introduced thus far come together. Theorem 2.15. Fix C > 1, D > 0. Let A = ∪n∈NAn C-algebras. k·kA be a unital inductive limit of If A is ((An, LAn))n∈N-propinquity approximable for some sequence of (C, D)-quasi- Leibniz compact quantum metric spaces and summable (β(j))j∈N ⊂ (0, ∞), then the sequence ((An, LAn))n∈N is A−C-convergent with respect to(cid:16)FτAN , LτAN(cid:17), where: (1) τAN = (τn,n+1)n∈N are the evident tunnels from (3) of Definition 2.5, (2) there exists a unital -isomorphism ψ : A → FτAN such that ψ = ψ(n) on An for all n ∈ N of Definition 2.13, and (3) if we define LA := LτAN ◦ ψ, then (A, LA) is a (C, D)-quasi-Leibniz compact quantum metric space such that ∪n∈Ndom (LAn ) ⊆ dom (LA) with Λ∗ ((An, Ln) , (A, LA)) 6 4 ∞ ∑ j=n β(j) and lim n→∞ Λ∗ ((An, Ln) , (A, LA)) = 0. Proof. The fact that there exists a unital -monomorphism ψ : A → FτAN such that ψ = ψ(n) on An for all n ∈ N follows from [31, 6.1.2 Theorem] and Lemma 2.14. Next, we show ψ(A) = FτAN . Let a + I0 ∈ FτAN . Let ε > 0. There exists b = (bn)n∈N = ((bn n+1))n∈N ∈ S0 such that ka + I0 − b + I0kF < ε/2 by density. Thus Re(b), Im(b) ∈ L0. Hence, there exists r ∈ R, r > 0 such that S0(Re(b)) 6 r, S0(Im(b)) 6 r. There exists N ∈ N, N > 1 such that 2r · ∑∞ j=N β(j) < ε/4. n, bn Next, note that Re(b) = ((Re(bn n), Re(bn n+1)))n∈N. Define c = ((cn n, cn n+1))n∈N ∈ D in the following way: (cn n, cn n+1) = (0, 0) (0, Re(bN−1 (Re(bn N )) n), Re(bn n+1)) : 0 6 n 6 N − 2 : n = N − 1 : n > N. Therefore c ∈ L0 and c − Re(b) ∈ I0, which implies that c + I0 = Re(b) + I0 ∈ FτAN . Now, consider ψN(Re(bN−1 N )) and recall that Re(bN−1 N ) = Re(bN N ) and that Re(bn n+1) = Re(bn+1 n+1) for all n ∈ N by Proposition 2.10. Therefore (cid:13)(cid:13)(cid:13)ψN(Re(bN−1 N )) − c(cid:13)(cid:13)(cid:13)D = sup k∈N(cid:13)(cid:13)(cid:13)Re(bN N ) − Re(bN+k+1 . N+k+1)(cid:13)(cid:13)(cid:13)A 12 KONRAD AGUILAR Since S0(Re(b)) 6 r < ∞, we have that kRe(bn n ∈ N by Proposition 2.10. Hence for all k ∈ N n) − Re(bn+1 n+1)kA 6 r · 2β(n) for all (cid:13)(cid:13)(cid:13)Re(bN N ) − Re(bN+k+1 N+k+1)(cid:13)(cid:13)(cid:13)A 6 2r · N+k ∑ j=N β(j) 6 2r · ∞ ∑ j=N β(j) < ε/4. Thus kψN(Re(bN−1 I0 = d + I0 such that kψN(Im(bN−1 b + I0 = (c + id) + I0 and that kψN(bN−1 ψ(bN−1 N )) − ckD 6 ε/4. Similarly, we may find d ∈ L0 with Im(b) + N )) − dkD 6 ε/4. Note that c + id ∈ S0 with N ) − (c + id)kD 6 ε/2. Therefore, since ∈ AN, we gather N )) as bN−1 N N ) = ψ(N)(bN−1 ka + I0 − ψ(bN−1 N )kF τAN τAN + kψ(bN−1 N ) − b + I0kF τAN 6 ka + I0 − b + I0kF < ε/2 + kψ(bN−1 6 ε/2 + kψN (bN−1 6 ε/2 + ε/2 = ε N ) − (c + id) + I0kF τAN N ) − (c + id)kD by definition of quotient norm. In particular, the set ψ(A) is dense in FτAN . As ψ is an isometry and A is complete, it must be the case that ψ(A) = FτAN . Now, assume that a ∈ ∪n∈Ndom (LAn ), then there exists N ∈ N, N > 1 such that a ∈ sa (AN) and LAN (a) < ∞. Thus by Proposition 2.10, LA(a) = L τAN ◦ ψ(a) = L F F τAN ◦ ψ(N)(a) 6 S0(ψN(a)) = max{LAN (a), kakA/(2β(N − 1))} < ∞. The remaining follows from Proposition 2.3 and Proposition 2.6. (cid:3) Next, we see how our approximable inductive limits are well-suited for provid- ing convergent sequences of inductive limits by reducing the problem of showing convergence to the terms of the inductive sequence, which will be applied in The- orem 3.9. Theorem 2.16. Fix C > 1, D > 0. For each k ∈ N ∪ {∞}, let Ak = ∪n∈NAk n be a unital inductive limit of C-algebras such that Ak is(cid:16)(cid:16)Ak and summable (βk(j))j∈N ⊂ (0, ∞), and let(cid:16)Ak, L n(cid:17)(cid:17)n∈N Ak(cid:17) be the limit of(cid:16)(cid:16)Ak in dual propinquity given by (3) of Theorem 2.15. approximable for some sequence of (C, D)-quasi-Leibniz compact quantum metric spaces n(cid:17)(cid:17)n∈N -propinquity n, L n, L Ak Ak k·kA If: (1) there exists a summable (β(j))j∈N ⊂ (0, ∞) such that βk(j) 6 β(j) for all k ∈ N ∪ {0}, j ∈ N, and (2) for every n ∈ N, it holds thats then lim k→∞ lim k→∞ n, L Λ∗(cid:16)(cid:16)Ak Λ∗(cid:16)(cid:16)Ak, L Ak n , LA∞ n(cid:1)(cid:17) = 0, n(cid:17) ,(cid:0)A∞ Ak(cid:17) , (A∞, LA∞)(cid:17) = 0. N, L Ak N, LA∞ N, L Ak N, LA∞ N(cid:17) ,(cid:16)A∞ Λ∗(cid:16)(cid:16)Ak, L 6 Λ∗(cid:16)(cid:16)Ak, L + Λ∗(cid:16)(cid:16)A∞ Ak(cid:17) , (A∞, LA∞)(cid:17) Ak(cid:17) ,(cid:16)Ak N(cid:17)(cid:17) + Λ∗(cid:16)(cid:16)Ak N(cid:17) , (A∞, LA∞ )(cid:17) βk(j) + Λ∗(cid:16)(cid:16)Ak N(cid:17) ,(cid:16)A∞ N(cid:17) ,(cid:16)A∞ < (2ε/3) + Λ∗(cid:16)(cid:16)Ak Ak(cid:17) , (A∞, LA∞)(cid:17) 6 (2ε/3) < ε, N(cid:17)(cid:17) + 4 N(cid:17)(cid:17) . Λ∗(cid:16)(cid:16)Ak, L ∞ ∑ j=N ∞ ∑ j=N lim sup k→∞ N, L Ak N, LA∞ N, L Ak N, LA∞ 6 4 N(cid:17)(cid:17) β∞(j) Therefore INDUCTIVE LIMITS AND QUANTUM METRICS 13 Proof. Let ε > 0. There exists N ∈ N such that 4 ∑∞ rem 2.15 and the triangle inequality, for each k ∈ N, j=N β(j) < ε/3. Thus by Theo- which completes the proof as ε > 0 was arbitrary. (cid:3) In this process of building a Lip-norm on the inductive limit, we also created new Lip-norms on the terms on the inductive sequence itself. We close this section with some comparisons in the next proposition that also serves to explain further the structure of these new Lip-norms. We also consider the case when the induc- tive limit already comes equipped with a certain kind of Lip-norm. k·kA be a unital inductive limit Proposition 2.17. Fix C > 1, D > 0. Let A = ∪n∈NAn of C-algebras. Let A be ((An, LAn))n∈N-propinquity approximable for some sequence of (C, D)-quasi-Leibniz compact quantum metric spaces and summable (β(j))j∈N ⊂ (0, ∞), and let β(−1) := ∞ with the convention that (1/∞) = 0. If(cid:16)FτAN , LτAN(cid:17), ψ, and (A, LA) are as in Theorem 2.15, then (1) for each n ∈ N, the pair (An, LτAN quantum metric space such that ◦ ψ(n)) is a (C, D)-quasi-Leibniz compact LA(a) = LτAN for all a ∈ sa (An), and ◦ ψ(n)(a) 6 max{1, diam (An, LAn)/β(n − 1)} · LAn (a) (2) if there exists a (C, D)-quasi-Leibniz Lip-norm L′ A on A defined and lower semi- continuous on all of A such that L′ A = LAn on An for each n ∈ N, then LAn (a) 6 LτAN ◦ ψ(n)(a) 6 max{1, diam (An, LAn )/β(n − 1)} · LAn (a) for all a ∈ sa (An), equivalently A(a) 6 LA(a) 6 max{1, diam (An, LAn)/β(n − 1)} · L′ L′ A(a) for all a ∈ sa (An). Proof. Fix n ∈ N. Let a ∈ dom (LAn ). If n = 0, then the conclusions are clear since the only Lip-norm on C is the 0-seminorm. So, assume that n > 1. Then by Proposition 2.10, we have that LτAN ◦ ψ(n)(a) 6 S0(ψn(a)) = max{LAn (a), kakA/β(n − 1)} < ∞. 14 KONRAD AGUILAR Therefore, the domain of LτAN ◦ ψ(n) is dense in An. Now, fix a state µ ∈ S (FτAN ). (d) 6 1 and µ(d) = 0} is compact by [32, We have that {d ∈ sa(cid:0)FτAN(cid:1) LτAN Proposition 1.3]. Thus the set {d ∈ sa(cid:16)ψ(n)(An)(cid:17) LτAN (d) 6 1 and µ(d) = 0} is ◦ ψ(n) is lower semi-continuous on An and vanishes compact. Therefore, as LτAN only on scalars as ψ(n) is a unital -monomorphism by Lemma 2.14, we have that ◦ ψ(n)) is a (C, D)-quasi-Leibniz compact quantum metric space by [32, (An, LτAN Proposition 1.3] since ψ(n) is a -homomorphism. Next, fix a ∈ dom (LAn ). As Lip-norms vanish on scalars, we have for any λ ∈ R by Proposition 2.10 and since ψ(n) is unital, LτAN ◦ ψ(n)(a) = LτAN ◦ ψ(n)(a − λ1A) 6 S0(ψn(a − λ1A)) = max{LAn (a − λ1A), ka − λ1AkA/β(n − 1)} = max{LAn (a), ka − λ1AkA/β(n − 1)}. Thus, as λ ∈ R was arbitrary, we have LτAN ◦ ψ(n)(a) 6 max{LAn (a), inf λ∈R {ka − λ1AkA}/β(n − 1)} 6 max{LAn (a), (diam (An, LAn ))/β(n − 1))LAn(a)} = max{1, (diam (An, LAn))/β(n − 1))}LAn(a) by [33, 1.6 Proposition]. This completes (1). For (2), fix a ∈ An. Next, let d = ((dk k)k∈N ⊂ ∪l∈NAl converges to 0. And, the sequence (a − dk+n k+1))k∈N ∈ I0. Thus, the sequence k+n)k∈N ⊂ ∪l∈NAl k, dk A is lower semi-continuous on A, we have (dk converges to a. Therefore, since L′ A(a − dk+n L′ L′ A(a) 6 lim inf k→∞ k+n) = lim inf k→∞ LAk+n(a − dk+n k+n) 6 S0(ψn(a) − d) by Proposition 2.10. Thus L′ A(a) 6 LτAN ◦ ψ(n)(a) as d ∈ I0 was arbitrary. (cid:3) 3. ALL UNITAL AF ALGEBRAS ARE PROPINQUITY APPROXIMABLE In this section, we use our results in the previous section to show that any uni- tal AF algebra A is propinquity approximable with regard to any non-decreasing sequence of unital finite-dimensional C-subaglebras (An)n∈N such that, we have k·kA. This was already shown to be the case in [5] only for unital A = ∪n∈NAn AF algebras equipped with faithful tracial states. This allowed for investigation of UHF algebras [19] and Effros-Shen algebras [15], but left out critical examples of AF algebras like the unitalization of the compact operators on a separable Hilbert space [14, Example III.2.3] and the Boca-Mundici AF algebra [8, 30]. Now, all uni- tal AF algebras have been shown to be quasi-Leibniz compact quantum metric spaces [1, 6], yet outside the faithful tracial state case, these quantum metrics do not provide approximability. And, approximability is desirable since it allows one to prove theorems about convergence of sequences of unital AF algebras. We dis- play this directly in this section by showing that convergence of ideals of unital AF algebras in the Fell topology provides convergence of their unitizations in the dual propinquity topology in Theorem 3.9 with the new quasi-Leibniz quantum metrics INDUCTIVE LIMITS AND QUANTUM METRICS 15 on all unital AF algebras introduced in this section in Theorem 3.4. To build these and quantum metrics, we simply use the techniques of the previous section and [5] along with the observation that all finite-dimensional C-algebras have faithful tracial states, and we find that our construction generalizes the Lip-norms of [5] in Proposition 3.5. k·kA, a unital Convention 3.1. We call a unital C-algebra of the form A = ∪n∈NAn AF algebra if (An)n∈N is a non-decreasing sequence of unital finite-dimensional C-subalgebras such that A0 = C1A. We provide the definition of conditional expectations for convenience. Definition 3.2 ([9, Definition 1.5.9]). Let A be a unital C-algebra and let B ⊆ A be a unital C-subalgebra. A linear map E : A → B such that E(A) = B is a conditional expectation if (1) E(b) = b for all b ∈ B, (2) E is contractive completely positive [9, Definition 1.5.1], and (3) E(bxb′) = bE(x)b′ for all x ∈ A, b, b′ ∈ B. Next, we give notation for the conditional expectations for the finite-dimen- sional C-algebras of a given AF algebra. k·kA be a unital AF algebra. For each n ∈ N, let Notation 3.3. Let A = ∪n∈NAn En+1,n : An+1 → An be a condition expectation, which always exists by [9, Lemma 1.5.11] since finite-dimensional C-algebras have faithful tracial states and are von Neumann algebras. If n > m > 0, then denote En,m := Em+1,m ◦ · · · ◦ En,n−1 : An → Am, and if n = m, then En,m := idn, the identity map on An. k·kA be a unital AF algebra. Denote U = (An)n∈N. Theorem 3.4. Let A = ∪n∈NAn Let (β(j))j∈N ⊂ (0, ∞) be summable. For each n ∈ N, let En+1,n : An+1 → An be a contitional expectation onto An. If for each n ∈ N \ {0}, we define for all a ∈ An, Lβ An,E(a) := max m∈{0,...,n−1}(cid:26) ka − En,m(a)kA β(m) (cid:27) using Notation 3.3 and Lβ A0,E is the 0-seminorm on A0, then (1) for each n ∈ N, the pair (An, Lβ An,E) is a (2, 0)-quasi-Leibniz comact quantum (2) A is ((An, Lβ (3) the (2, 0)-quasi-Leibniz Lip-norm Lβ metric space such that diam(cid:16)An, Lβ isfies sa (∪n∈NAn) ⊆ dom(cid:16)Lβ Λ∗(cid:16)(cid:16)An, Lβ An,E(cid:17) 6 2β(0), U ,E(cid:17), where An,E(cid:17) ,(cid:16)A, Lβ lim n→∞ U ,E(cid:17)(cid:17) = 0 An,E))n∈N-approximable with respect to (β(j))j∈N ⊂ (0, ∞), and U ,E on A given by (2) and Theorem 2.15 sat- 16 KONRAD AGUILAR and Λ∗(cid:16)(cid:16)An, Lβ An,E(cid:17) ,(cid:16)A, Lβ U ,E(cid:17)(cid:17) 6 4 ∞ ∑ j=n β(j) for each n ∈ N, where for each n ∈ N, it holds that Lβ U ,E(a) 6 max{1, 2β(0)/β(n − 1)} · Lβ An,E(a) for all a ∈ sa (An), where β(−1) := ∞ with the convention that (1/∞) = 0. In particular, every unital AF algebra A is propinquity approximable, and A is propin- quity approximable by any increasing sequence (An)n∈N of finite-dimensional unital C- k·kA, and if k ∈ (1, ∞), for each subalgebras such that A0 = C1A and A = ∪n∈NAn j ∈ N, then we may choose β(j) 6 1/ dim(Aj)k. Proof. Fix n ∈ N \ {0}. Note that En,0(An) ⊆ A0 = C1A and A0 ⊆ Am for all m ∈ N. Thus Lβ An,E(a) = 0 if and only if a ∈ C1A. Furthermore {a ∈ An LAn (a) < ∞} = An and Lβ An,E is lower semi-continuous. Also Lβ An,E is (2, 0)-quasi-Leibniz by [5, Lemma 3.2]. Denote σ : λ1A ∈ A0 7→ λ ∈ C and consider µn := σ ◦ En,0, which is a state on An. Let a ∈ An such that Lβ An,E(a) 6 1 and µn(a) = 0. Thus En,0(a) = 0, and so kakA = ka − En,0(a)kA 6 β(0). Hence the set {a ∈ An Lβ An,E(a) 6 1 and µn(a) = 0} is totally bounded by finite-dimensionality. Therefore (An, Lβ An,E) is a (2,0)- quasi-Leibniz compact quantum metric space by [32, Proposition 1.3]. To estimate the diameter, we note that for any a ∈ An with Lβ An,E(a) 6 1 and any µ, ν ∈ S (An), µ(a) − ν(a) = µ(a − µn(a)1A) − (ν(a − µn(a)1A) = (µ − ν)(a − µn(a)1A) 6 2ka − µn(a)1AkA = 2ka − En,0(a)k 6 2β(0). Thus (1) is proven. Now, Fix n ∈ N. Assume a ∈ An. Then En+1,n(a) = a and so En+1,m(a) = En,m(a) for all m ∈ {0, . . . , n}. Therefore (3.1) An+1,E(a) = Lβ Lβ An,E(a) for all a ∈ An. An,E(a) 6 1. Then Lβ Next, we estimate lengths of the evident bridges. Fix n ∈ N. Let a ∈ An such An,E(a) 6 1 by Equation (3.1) and ka − An+1,E(a) 6 1. Fix m ∈ {0, . . . , n − 1}. Note that Lβ akA = 0. Now, let a ∈ An+1 such that Lβ that En+1,n(a) ∈ An and consider kEn+1,n(a) − En,m(En+1,n(a))kA = kEn+1,n(a) − En+1,m(a)kA An+1,E(a) = Lβ = kEn+1,n(a) − En+1,n(En+1,m(a))kA = kEn+1,n(a − En+1,m(a))kA 6 ka − En+1,m(a)kA, where we used the fact that Am ⊆ An in line 2. Thus Lβ An,E(En+1,n(a)) 6 Lβ An+1,E(a) 6 1. Furthermore, the assumption that Lβ An+1,E(a) 6 1 implies that ka − En+1,n(a)kA 6 β(n). Finally, note that if a ∈ sa (An+1), then En+1,n(a) ∈ sa (An) by positivity of INDUCTIVE LIMITS AND QUANTUM METRICS 17 conditional expectations. Therefore by Lemma 1.8, we have that the length of the evident bridge from Definition 2.5 satisfies An,E, Lβ An+1,E) 6 β(n). λ(γn,n+1Lβ Thus (2) is proven, and (3) follows immediately from Theorem 2.15 and Propo- sition 2.17. The remaining statement follows from the fact that there exist condi- tional expectations on any finite-dimensional C-algebra onto any C-subalgebra by [9, Lemma 1.5.11], and the fact that for any increasing sequence (An)n∈N such that A0 = C1A, it holds that dim(Aj) > j for all j ∈ N. (cid:3) We note that for certain AF algebras, the beta sequence can be given as β(n) = 1 dim(An) . For example, this is the case of UHF algebras, Effros-Shen algebras, and the unitalization of the compact operators as long as the increasing sequence An is chosen appropriately. However, for the C-algebra of continuous functions on the one-point compactification on the natural numbers C(N), one could take the increasing sequence to be An ∼= Cn+1, in which case taking the dimension to a power greater than 1 is necessary to produce a summable sequence. Next, we see that in the case of a unital AF algebra with a faithful tracial state, the Lip-norms of Theorem 3.4 can recover the Lip-norms of [5, Theorem 3.5] up to quantum isometry. Thus, we provide a generalization of the results of [5] in the case that (β(j))j∈N is summable, and as a consequence, we achieve convergence in the quantum Gromov-Hausdorff propinquity of Latrémolière [28] in the faithful tracial state case. Proposition 3.5. Let A = ∪n∈NAn tracial state µ. Denote U = (An)n∈N, and let (β(j))j∈N ⊂ (0, ∞) be summable. Let k·kA be a unital AF algebra equipped with a faithful Lβ U ,µ(a) = sup n∈N ka − En(a)kA β(n) be the (2,0)-quasi-Leibniz Lip-norm on A of [5, Theorem 3.5], where En : A → An is the unique µ-preserving conditional expectation onto An. If for each n ∈ N, we define En+1,n := EnAn+1 : An+1 → An and let Lβ U ,E be the associated (2,0)-quasi-Leibniz Lip-norms on An, A, respectively, from Theorem 3.4, then: An,E, Lβ (1) for each n ∈ N, we have Lβ (2) for each n ∈ N and all a ∈ sa (An), it holds that U ,µ(a) = Lβ An,E(a) for all a ∈ An, Lβ U ,µ(a) 6 Lβ U ,E(a) 6 max{1, 2β(0)/β(n − 1)}Lβ where β(−1) := ∞ with the convention that (1/∞) = 0, U ,µ(a), (3) there exists a quantum isometry π : A → A and so U ,µ(cid:17) ,(cid:16)A, Lβ Λ∗(cid:16)(cid:16)A, Lβ U ,E(cid:17)(cid:17) 6 2Λ(cid:16)(cid:16)An, Lβ U ,E(cid:17)(cid:17) = 0, An,E(cid:17) ,(cid:16)A, Lβ U ,E(cid:17)(cid:17) 6 2β(n), (4) and for each n ∈ N, Λ∗(cid:16)(cid:16)An, Lβ An,E(cid:17) ,(cid:16)A, Lβ 18 KONRAD AGUILAR and thus limn→∞ Λ(cid:16)(cid:16)An, Lβ Gromov-Hausdorff propinquity of [28]. An,E(cid:17) ,(cid:16)A, Lβ U ,E(cid:17)(cid:17) = 0, where Λ is the quantum Proof. Fix n ∈ N \ {0}. By the proof of [5, Theorem 3.5], it holds tht En ◦ En−1 = En−1. Therefore En,m = Em for all m ∈ {0, . . . , n − 1}. Thus (1) is proven. Conclu- sions (2) and (3) follow from Theorem 3.4 and [5, Theorem 3.5] and the fact that limits are unique up to quantum isometry in the dual propinquity. Conclusion (4) follows from [25, Theorem 5.5] and [5, Theorem 3.5]. (cid:3) We can now apply our findings to discover new continuous families of unital AF algebras since we now do not require faithful tracial states. We will form a con- tinuous map from the ideal space of any unital AF algebra into the dual propin- quity space by mapping an ideal to its unitization with carefully chosen Lip-norms from Theorem 3.4. So, although we cannot discuss convergence of ideals them- selves since ideals may not contain a unit, we can now in the very least capture the Fell topology by considering the unitization of the ideals. This was not pos- sible before due to the fact that not every unital AF algebra has a faithful tracial state. However, since an ideal may or may not contain unit, we have to be careful by what we mean by the "unitization" as these seem to differ when considering a unital versus a non-unital C-algebra. Yet, thankfully, there does exist a unitization process of C-algebras that does not distinguish between the unital and non-unital case, which allows us to proceed much more smoothly in our construction of Lip- norms. We present this unitization now. The only known reference that we know of this is from https://mathoverflow.net/questions/210025/unitization-process-of-unital-and-non-unital-c-algebras due to user UwF. Thus, we provide details for the proof found in the above link. Theorem-Definition 3.6. Let A be a C-algebra with or without unit. The unitization of A, denotedeA, is the vector space A ⊕ C equipped with (1) adjoint defined by (a, λ)∗ := (a∗, λ) for all a ∈ A, λ ∈ C, (2) multiplication defined by (a, λ)(b, µ) := (ab + µa + λb, λµ) for all a, b ∈ A, λ ∈ C, (3) and norm defined by k(a, λ)keA := max( sup b∈A,kbkA61 for all a ∈ A, λ ∈ C, {kab + λbkA} , λ) , Furthermore: whereeA equipped with these operations and norm is a unital C-algebra with unit 1eA = (0, 1) and kakA = k(a, 0)keA for all a ∈ A and A is -isomorphic to the maximal ideal of co-dimension one ofeA given by {(a, λ) ∈eA a ∈ A, λ = 0}. (4) if B is a C-subalgebra of A, then eB is the unital C-subalgebra of eA given by {(b, λ) ∈ eA b ∈ B, λ ∈ C}, (5) if A is non-unital, then k(a, λ)keA = supb∈A,kbkA61 {kab + λbkA} for all a ∈ (6) if A is unital, then k(a, λ)keA = max {ka + λ1AkA, λ} for all a ∈ A, λ ∈ C, in which case the direct sum C-algebra A ⊕ C given by coordinate-wise operations A, λ ∈ C, and INDUCTIVE LIMITS AND QUANTUM METRICS 19 and max norm is -isomorphic toeA by the -isomorphism (a, λ) ∈ A ⊕ C 7−→ (a − λ1A, λ) ∈eA. Proof. It is routine to check that eA is a -algebra with respect to the defined op- erations. Let B(A ⊕ C) be the unital Banach algebra of bounded operators from the direct sum C-algebra A ⊕ C (given by coordinate-wise operations and max norm) to itself. Denote the unit of B(A ⊕ C) by 1B(A⊕C). For each (a, λ) ∈ A ⊕ C, let L(a,λ) ∈ B(A ⊕ C) denote the left multiplication operator, that is L(a,λ)(b, µ) = (ab, λµ) for all (b, µ) ∈ A ⊕ C. By construction, the map (a, λ) ∈ eA 7−→ L(a,λ) + λ1B(A⊕C) ∈ B(A ⊕ C) is a linear multiplicative function. Now, we show it is injective. Assume that (0, λ) =⇒ (0, 0) = (0, λ) =⇒ λ = 0. Hence (0, 0) = (L(a,λ) + λ1B(A⊕C))(a∗, 0) = (a, λ) ∈ eA such that L(a,λ) + λ1B(A⊕C) = 0. Thus (L(a,λ) + λ1B(A⊕C))(0, 1) = L(a,0)(a∗, 0) = (aa∗, 0) =⇒ aa∗ = 0 =⇒ a = 0. For all (a, λ) ∈ eA, we gather that (cid:13)(cid:13)(cid:13)L(a,λ) + λ1B(A⊕C)(cid:13)(cid:13)(cid:13)B(A⊕C) (b,µ)∈A⊕C,k(b,µ)kA⊕C61(cid:26)(cid:13)(cid:13)(cid:13)(cid:16)L(a,λ) + λ1B(A⊕C)(cid:17) (b, µ)(cid:13)(cid:13)(cid:13)A⊕C(cid:27) b∈A,µ∈C,kbkA61,µ61(cid:8)k(ab + λb, λµ)kA⊕C(cid:9) sup sup = sup {max {kab + λbkA, λµ}} = = b∈A,µ∈C,kbkA61,µ61 = sup b∈A,kbkA61 {max {kab + λbkA, λ}} = k(a, λ)keA. satisfies the C-identity follows the same argument as the proof of the standard unitization [14, Proposition I.1.3]. Therefore k · keA defines a Banach algebra norm on eA. The fact that this norm For (4), cleary the set {(b, λ) ∈ eA b ∈ B, λ ∈ C} is a unital C-subalgebra of eA that equals eB. Since they have the same algebraic operations and adjoint, their fines a C-norm oneA, when A is non-unital by [14, Proposition I.1.3]. Conclusion (5) follows similiary since the expression on the right-hand side de- Finally, for (6), when A is unital, we have that for b ∈ A, kbkA 6 1 norms also agree since C-norms are unique. kab + λbkA = k(a + λ1A)bkA 6 ka + λ1AkA, and since 1A ∈ A, k1AkA = 1, we have the desired equality for the norm. It is routine to show that the map in (6) is a -isomorphism. (cid:3) Although we do not define the Fell topology on the ideals of a C-algebra, we provide some results about the topology. For a more detailed summary see [2, Section 3]. Theorem-Definition 3.7 ([17, Theorem 2.2]). Let A be a C-algebra. Let Ideal(A) de- note the set of norm-closed two-sided ideals of A including {0}, A. The Fell topology on Ideal(A) is a compact Hausdorff topology such that a net (Iµ)µ∈∆ ⊆ Ideal(A) converges 20 KONRAD AGUILAR to I ∈ Ideal(A) if and only if for each a ∈ A the net (ka + IµkA/Iµ )µ∈∆ ⊆ R converges to ka + IkA/I ∈ R with respect to the usual topology on R. On primitive ideals, the Fell topology is stronger than the Jacobson topology [2, Proposition 3.7], and is in fact built from the closed subsets of the primitive ideals in the Jacobson topology. The Fell topology is a general construction of a topology on closed subsets of a given topology introduced by Fell in [18]. The next summarizes a topology introduced in [2] on inductive limits of C-algebras built as an inverse limit topology from the Fellt topologies on the ideals of the C- algebras of the inductive seqeunce, and this topology happens to agree with the Fell topology in the case of AF algebras and is in general stronger than the Fell topology. k·kA be Proposition 3.8 ([2, Proposition 3.15 and Theorem 3.22]). Let A = ∪n∈NAn an inductive limit of C-algebras (not necessarily unital). Denote U = (An)n∈N. The following hold. (1) For each n ∈ N, the map ιn+1i : J ∈ Ideal(An+1) 7→ J ∩ An ∈ Ideal(An) is continuous with respect to the associated Fell topologies on Ideal(An+1) and Ideal(An). (2) The inverse limit Ideal(A)∞ ⊆ ∏n∈N Ideal(An) with its topology given by the inverse limit sequence (Ideal(An), Fell, ιn+1i)n∈N ([41, 29.9 Definition]) pro- vides a topology on Ideal(A) denoted Felli(U ) by the initial topology of the injec- tion J ∈ Ideal(A) 7→ (J ∩ An)n∈N ∈ Ideal(A)∞. (3) The topology Felli(U ) is stronger than Fell on Ideal(A), and if A is AF, then Felli(U ) = Fell and is metrized by the metric defined for each I, J ∈ Ideal(A) by i(U )(I, J) =(0 2− min{n∈N:I∩An6=J∩An} m : I ∩ An = J ∩ An for all n ∈ N : otherwise. Now, we are ready to prove our result about ideal convergence. 1,0 :eI → C1eI. Fix a summable sequence (β(j))j∈N ⊂ (0, ∞). k·kA be a unital AF algebra. Denote U = (An)n∈N. Theorem 3.9. Let A = ∪n∈NAn Using Theorem-Definition 3.6, for each n ∈ N \ {0} and each J ∈ Ideal(An+1), fix a conditional expectation EJ tional expectation EJ n+1,n : eJ → ^J ∩ An, and for each J ∈ Ideal(A1), fix a condi- Next, for every I ∈ Ideal(A), set UeI = (eIn)n∈N, where for n ∈ N \ {0}, we set k·keI , where UeI is an non-decreasing eIn = ^I ∩ An andeI0 = C1eI, and note thateI = ∪n∈NeIn sequence of finite-dimensional unital Csubalgebras ofeI by Theorem-Definition 3.6. For each, n ∈ N \ {0}, define for all (a, λ) ∈eIn, m∈{0,...,n−1}  eIn,EeI (a, λ) := max eI0,EeI be the zero seminorm oneI0. k(a, λ) − EI∩Am+1 m+1,m ◦ · · · ◦ EI∩An and let Lβ Lβ β(m) n,n−1(a, λ)keI , INDUCTIVE LIMITS AND QUANTUM METRICS 21 If for every I ∈ Ideal(A), we define Lβ UeI ,EeI as in Theorem 3.4, then, the map FA : I ∈ Ideal(A) 7−→(cid:18)eI, Lβ UeI,EeI(cid:19) ∈ qLcqms(2,0) is continuous, where Ideal(A) is equipped with the Fell topology and qLcqms(2,0) is the class of (2, 0)-quasi-Leibniz compact quantum metric spaces equipped with the topology induced by the dual propinquity Λ∗. Furthermore, if β(j) 6 2−j−5 for all j ∈ N, then the map FA is Lipschitz with respect to the metric on ideals in (3) of Proposition 3.8 and the dual propinquity. Proof. Let (I(n))n∈N∪{∞} ⊆ Ideal(A) be a sequence that converges to I(∞) with re- spect to the Fell topology of Theorem-Definition 3.7. We will use Theorem 2.16. Fix M ∈ N \ {0}. By Proposition 3.8, there exists N ∈ N such that m i(U )(I(n), I(∞)) < 2−(M+1) for all n > N. Therefore, for all n > N, it holds that I(n) ∩ Ak = I(∞) ∩ Ak I(∞)k for all k ∈ {0, . . . , M}. Therefore, by construction, we have that = = Lβ max (a, λ) if n > N, then and gI(n)k = ] gI(n)M,EgI(n) m∈{0,...,M−1} m∈{0,...,M−1} m∈{0,...,M−1} m∈{0,...,M−1} = Lβ max max max = = ] I(∞)M,E ] I(∞) (a, λ) k(a, λ) − EI(n)∩Am+1 m+1,m k(a, λ) − EI(n)∩Am+1 m+1,m k(a, λ) − EI(∞)∩Am+1 m+1,m k(a, λ) − EI(∞)∩Am+1 m+1,m ◦ · · · ◦ EI(n)∩AM β(m) ◦ · · · ◦ EI(n)∩AM β(m) β(m) ◦ · · · ◦ EI(∞)∩AM I(∞)M M,M−1 (a, λ)k]  M,M−1 (a, λ)kgI(n)  M,M−1 (a, λ)kgI(n)M   gI(n)M,EgI(n)(cid:19) and (a, λ)k] I(∞) In particu- M,M−1 ◦ · · · ◦ EI(∞)∩AM β(m) ] ] I(∞)M,E I(∞)M, Lβ I(∞)M by (4) of Theorem-Definition 3.6. for all (a, λ) ∈ gI(n)M = ] lar, the identity map is a quantum isometry between (cid:18)gI(n)M, Lβ (cid:18)] I(∞)(cid:19)(cid:19) = 0. I(∞)!! = 0 I(∞)(cid:19) for all n > N. Hence gI(n)M,EgI(n)(cid:19) ,(cid:18)] Λ∗(cid:18)(cid:18)gI(n)M, Lβ Λ∗ gI(n), Lβ ,EgI(n)! , ] UgI(n) I(∞)M, Lβ The same result is immediate for M = 0. Thus, as M ∈ N was arbitrary, we have that I(∞), Lβ ] I(∞)M,E lim n→∞ lim n→∞ U] I(∞) ] ] ,E 22 KONRAD AGUILAR by Theorem 2.16. Hence, the desired map is continuous. Next, assume that β(j) 6 2−j−5 for all j ∈ N. Fix I, J ∈ Ideal(A) such that I 6= J. i(U )(I, J) = 2−n. Assume that n > 1. By Therefore, there exists n ∈ N such that m the same argument as above and Theorem 3.4, we have Λ∗(cid:18)(cid:18)eI, Lβ 6 Λ∗(cid:18)(cid:18)eI, Lβ + Λ∗(cid:18)(cid:18)eJ, Lβ UeJ,EeJ(cid:19)(cid:19) UeI,EeI(cid:19) ,(cid:18)eJ, Lβ UeI ,EeI(cid:19) ,(cid:18)eIn−1, Lβ UeJ,EeJ(cid:19) ,(cid:18)eJn−1, Lβ 2−j−5 + 0 + 4 6 4 eIn−1,EeI(cid:19)(cid:19) + Λ∗(cid:18)(cid:18)eIn−1, Lβ eJn−1,EeJ(cid:19)(cid:19) eIn−1,EeI(cid:19) ,(cid:18)eJn−1, Lβ eJn−1,EeJ(cid:19)(cid:19) ∞ ∑ j=n−1 ∞ ∑ j=n−1 2−j−5 = (1/4)(4 · 2−n) = m i(U )(I, J). The case n = 0 follows similarly. (cid:3) We now make a remark about a certain choice we made in this section. Given a k·kA with a non-decreasing sequence of unital finite- unital AF algebra A = ∪n∈NAn dimensional C-subalgebras (An)n∈N of A, we could have used the conditional expectations from An+1 onto An of Theorem 3.4 to build conditional expectations from A onto An that behave much like the ones in the faithful tracial state case of Proposition 3.5. This would build Lip-norms on A that allow for propinquity ap- proximability with respect to quantum propinquity and not just dual propinquity in the case of any unital AF algebra, which is not that much of an advantage since the dual propinquity is preferred anyway as it's complete. However, to extend the conditional expectations in such a way to accomplish this, one would essentailly be using the injectivity of finite-dimensional C-algebras [7, Proposition IV.2.1.4, II.6.9.13, and Arveson's Extension Theorem II.6.9.12]. Although the main example of this paper is unital AF algebras, the goal of this paper is to provide a more gen- eral framework for future work on providing Lip-norms on inductive limits that allow for not only convergence of inductive sequences to inductive limits but also convergence of sequences of inductive limits, and injectivity of C-algebras is cer- tainly not a common property especially among the standard classes of inductive limits studied. Hence, we chose to present our work in this section without using the injectivity of finite-dimensional C-algebras to set the stage for future work. We conclude this paper by noting that the results in [5] are all true in this more general setting by Proposition 3.5. One just simply needs replace the convergence conditions on the sequence (β(n))n∈N with summability conditions when work- ing in the non-faithful tracial state case. For instance, the compactness results of [5, Section 6] and the convergence of families of Lip-norms on a fixed unital AF algebra result of [5, Section 8] are true wih this minor change, and we note that as a result [5, Section 8] provides continuous families of Lip-norms on the uniti- zation of the compact operators by varying the sequence (β(n))n∈N in a suitable topology. INDUCTIVE LIMITS AND QUANTUM METRICS 23 REFERENCES [1] K. Aguilar, AF algebras in the quantum Gromov-Hausdorff propinquity space. 32 pages, submitted (2016), ArXiv: 1612.02404. [2] , Convergence of quotients of AF algebras in quantum propinquity by convergence of ideals. 40 pages, submitted (2016), ArXiv: 1608.07016. [3] K. Aguilar and J. Kaad, The Podle´s sphere as a spectral metric space. (Accepted 2018) to appear in Jour- nal of Geometry and Physics https://doi.org/10.1016/j.geomphys.2018.07.015 (Available online July 25, 2018), 23 pages, arXiv: 1803.03027. [4] K. Aguilar and F. Latrémolière, Some applications of conditional expectations to convergence for the quantum Gromov-Hausdorff propinquity. submitted (2018), 12 pages, ArXiv: 1708.00595. [5] , Quantum ultrametrics on AF algebras and the Gromov-Hausdorff propinquity, Studia Mathe- matica 231 (2015), no. 2, 149 -- 193. ArXiv: 1511.07114. [6] C. Antonescu and E. Christensen, Spectral triples for AF C∗-algebras and metrics on the cantor set, J. Oper. Theory 56 (2006), no. 1, 17 -- 46. ArXiv: 0309044. [7] B. Blackadar, Operator algebras, Encyclopaedia of Mathematical Sciences, vol. 122, Springer-Verlag, Berlin, 2006. Theory of C∗-algebras and von Neumann algebras, Operator Algebras and Non- commutative Geometry, III. MR2188261 [8] F. Boca, An algebra associated with the Farey tessellation, Canad. J. Math. 60 (2008), no. 5, 975 -- 1000. ArXiv: math/0511505. [9] N. P. Brown and N. Ozawa, C-algebras and finite-dimensional approximations, Graudate Studies in Mathematics, vol. 88, American Mathematical Society, 2008. [10] M. Christ and M. A. Rieffel, Nilpotent group C∗-algebras-algebras as compact quantum metric spaces, Submitted (2015), 22 pages. ArXiv: 1508.00980. [11] A. Connes, Compact metric spaces, Fredholm modules and hyperfiniteness, Ergodic Theory and Dy- namical Systems 9 (1989), no. 2, 207 -- 220. [12] [13] L. Dabrowski and A. Sitarz., Dirac operator on the standard Podle´s quantum sphere, Noncommutative , Noncommutative geometry, Academic Press, San Diego, 1994. geometry and quantum groups (Warsaw, 2001), 2003, pp. 49 -- 58. ArXiv: math/0209048. [14] K. R. Davidson, C* -- algebras by example, Fields Institute Monographs, American Mathematical So- ciety, 1996. [15] E. G. Effros and C. L. Shen, Approximately finite C∗-algebras and continued fractions, Indiana Univ. Math. J. 29 (1980), no. 2, 191 -- 204. [16] G. A. Elliott, On the classification of inductive limits of sequences of semisimple finite-dimensional alge- bras, J. Algebra 38 (1976), no. 1, 29 -- 44. [17] J. M. G. Fell, The structure of algebras of operator fields, Acta. Math. 106 (1961), 233 -- 280. [18] J. M. G. Fell, A Hausdorff topology for the closed subsets of a locally compact non-Hausdorff space, Proc. Amer. Math. Soc. 13 (1962), 472 -- 476. MR0139135 [19] J. Glimm, On a certain class of operator algebras, Trans. Amer. Math. Soc. 95 (1960), 318 -- 340. [20] F. Latrémolière, Convergence of Cauchy sequences for the covariant Gromov-Hausdorff propinquity. Sub- [21] [22] [23] [24] [25] [26] [27] [28] mitted (2018), 25 pages, ArXiv: 1806.04721. , The covariant Gromov-Hausdorff propinquity. Submitted (2018), 29 pages, ArXiv: 1805.11229. , The modular Gromov-Hausdorff propinquity. Submitted (2016), 67 pages, ArXiv: 1608.04881. , Semigroupoid, groupoid and group actions on limits for the Gromov-Hausdorff propinquity. Sub- mitted (2017), 42 pages, ArXiv: 1708.01973. , Convergence of fuzzy tori and quantum tori for the quantum Gromov-Hausdorff propinquity: an explicit approach, Münster J. Math. 8 (2015), no. 1. ArXiv: math/1312.0069. , The dual Gromov -- Hausdorff Propinquity, Journal de Mathématiques Pures et Appliquées 103 (2015), no. 2, 303 -- 351. ArXiv: 1311.0104. , Quantum metric spaces and the Gromov-Hausdorff propinquity, Noncommutative geometry and optimal transport, 2016, pp. 47 -- 133. ArXiv: 1506.04341. , A compactness theorem for the dual Gromov-Hausdorff propinquity, Indiana Univ. Math. J. 66 (2017), no. 5, 1707 -- 1753. MR3718439 , The quantum Gromov-Hausdorff Propinquity, Trans. Amer. Math. Soc. (electronically pub- lished on May 22, 2015), : 49 Pages, http://dx.doi.org/10.1090/tran/6334, to appear in print. ArXiv: 1302.4058. 24 KONRAD AGUILAR [29] F Latrémolière and J Packer, Noncommutative solenoids and the Gromov-Hausdorff propinquity, Proc. Amer. Math. Soc. 145 (2017), no. 5, 2043 -- 2057. MR3611319 [30] D Mundici, Farey stellar subdivisions, ultrasimplicial groups, and K0 of AF C∗-algebras, Adv. in Math. 68 (1988), no. 1, 23 -- 39. MR931170 [31] G. J. Murphy, C∗-algebras and operator theory, Academic Press, San Diego, 1990. [32] N. Ozawa and M. A. Rieffel, Hyperbolic group C∗-algebras and free products C∗-algebras as compact quantum metric spaces, Canad. J. Math. 57 (2005), 1056 -- 1079. ArXiv: math/0302310. [33] M. A. Rieffel, Metrics on states from actions of compact groups, Documenta Mathematica 3 (1998), 215 -- 229. math.OA/9807084. , Metrics on state spaces, Documenta Math. 4 (1999), 559 -- 600. math.OA/9906151. , Group C∗-algebras as compact quantum metric spaces, Documenta Mathematica 7 (2002), 605 -- 651. ArXiv: math/0205195. , Compact quantum metric spaces, Operator algebras, quantization, and noncommutative ge- ometry, 2005, pp. 315 -- 330. ArXiv: 0308207. , Vector bundles and Gromov-Hausdorff distance, Journal of K-theory 5 (2010), 39 -- 103. ArXiv: math/0608266. , Matricial bridges for "matrix algebras converge to the sphere", Operator algebras and their applications, 2016, pp. 209 -- 233. ArXiv: 1502.00329. , Vector bundles for "matrix algebras converge to the sphere", Journal of Geometry and Physics 132 (2018), 181 -- 204. ArXiv: 1711.04054. , Gromov-Hausdorff distance for quantum metric spaces, Mem. Amer. Math. Soc. 168 (March [34] [35] [36] [37] [38] [39] [40] 2004), no. 796. math.OA/0011063. [41] S. Willard, General topology, Dover Publications, Inc., 2004. SCHOOL OF MATHEMATICAL AND STATISTICAL SCIENCES, ARIZONA STATE UNIVERSITY, 901 S. PALM WALK, TEMPE, AZ 85287-1804 E-mail address: [email protected]
1902.02266	2	1902	2019-02-22T07:56:49	Finite dimensional semigroups of unitary endomorphisms of standard subspaces	[ "math.OA" ]	Let $V$ be a standard subspace in the complex Hilbert space $H$ and $G$ be a finite dimensional Lie group of unitary and antiunitary operators on $H$ containing the modular group $(\Delta_V^{it})_{t \in R}$ of $V$ and the corresponding modular conjugation~$J_V$. We study the semigroup \[ S_V = \{ g\in G \cap U(H) : gV \subseteq V\} \] and determine its Lie wedge $L(S_V) = \{ x \in L(G) : exp(R_+ x) \subseteq S_V\}$, i.e., the generators of its one-parameter subsemigroups in the Lie algebra $L(G)$ of~$G$. The semigroup $S_V$ is analyzed in terms of antiunitary representations and their analytic extension to semigroups of the form $G exp(iC)$, where $C \subseteq L(G)$ is an $Ad(G)$-invariant closed convex cone. Our main results assert that the Lie wedge $L(S_V)$ spans a $3$-graded Lie subalgebra in which it can be described explicitly in terms of the involution $\tau$ of $L(G)$ induced by $J_V$, the generator $h \in L(G)^\tau$ of the modular group, and the positive cone of the corresponding representation. We also derive some global information on the semigroup $S_V$ itself	math.OA	math	Finite dimensional semigroups of unitary endomorphisms of standard subspaces Karl-H. Neeb∗ † February 25, 2019 Abstract Let V be a standard subspace in the complex Hilbert space H and G be a finite dimensional V )t∈R of Lie group of unitary and antiunitary operators on H containing the modular group (∆it V and the corresponding modular conjugation JV . We study the semigroup SV = {g ∈ G ∩ U(H) : gv ⊆ V } and determine its Lie wedge L(SV ) = {x ∈ g : exp(R+x) ⊆ SV }, i.e., the generators of its one- parameter subsemigroups in the Lie algebra g of G. The semigroup SV is analyzed in terms of antiunitary representations and their analytic extension to semigroups of the form G exp(iC), where C ⊆ g is an Ad(G)-invariant closed convex cone. Our main results assert that the Lie wedge L(SV ) spans a 3-graded Lie subalgebra in which it can be described explicitly in terms of the involution τ of g induced by JV , the generator h ∈ gτ of the modular group, and the positive cone of the corresponding representation. We also derive some global information on the semigroup SV itself. MSC 2010: Primary 22E45; Secondary 81R05, 81T05. Contents 1 Introduction 2 Endomorphisms of standard subspaces 2.1 The case where all unitary endomorphisms are invertible . . . . . . . . . . . . . 2.2 The algebra AV of V -real operators . . . . . . . . . . . . . . . . . . . . . . . . 2.3 One-parameter semigroups in AV . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Wick rotations of tubes and Olshanski semigroups 3.1 Wick rotations of tubes 3.2 Extensions of unitary representations to semigroups 3.3 Wick rotations of Olshanski semigroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 The subsemigroups SV in finite dimensional groups 4.1 The Lie wedge of SV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 The unit group GV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Some examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 6 6 7 10 12 12 14 16 19 19 23 23 ∗Department Mathematik, [email protected] FAU Erlangen-Nurnberg, Cauerstrasse 11, 91058-Erlangen, Germany; †Supported by DFG-grant NE 413/9-1. 1 5 Perspectives 5.1 Relations to von Neumann algebras . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Pairs of standard pairs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Classification problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4 Global information on the semigroup SV . . . . . . . . . . . . . . . . . . . . . . 5.5 Extensions to infinite dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5.1 Wick rotations for non-uniformly continuous actions . . . . . . . . . . . 5.5.2 The subsemigroup SV ⊆ U(H) . . . . . . . . . . . . . . . . . . . . . . . A Conjugation with unbounded operators B Some facts on Lie groups 25 25 27 28 28 31 31 31 32 34 1 Introduction Let H be a complex Hilbert space and M ⊆ B(H) be a von Neumann algebra. Further, let Ω ∈ H be a unit vector which is cyclic for M (MΩ is dense in H) and separating (the map M → H, M 7→ M Ω is injective). By the Tomita -- Takesaki Theorem ([BR87, Thm. 2.5.14]), the closed real subspace V := VM := {M Ω : M = M∗ ∈ M} is standard, i.e., V ∩ iV = {0} and H = V + iV (1) (cf. [Lo08] for the basic theory of standard subspaces). To the standard subspace V , we can associate a pair of modular objects (∆V , JV ), i.e., ∆V > 0 is a positive selfadjoint operator, JV is a conjugation (an antiunitary involution), and these two operators satisfy the modular relation JV ∆V JV = ∆−1 of the closed operator V . The pair (∆V , JV ) is obtained by the polar decomposition σV = JV ∆1/2 V with V = Fix(σV ). The main assertion of the Tomita -- Takesaki Theorem is that σV : D(σV ) := V + iV → H, x + iy 7→ x − iy JV MJV = M′ and ∆it V M∆−it V = M for t ∈ R. So we obtain a one-parameter group of automorphisms of M (the modular group) and a symmetry between M and its commutant M′, implemented by JV . Motivated by the Haag -- Kastler theory of local observables in Quantum Field Theory (QFT) ([Ha96], [BS93], [BDFS00]), we are interested in finite dimensional Lie groups G ⊆ U(H) of unitary operators fixing Ω, containing the corresponding modular group (∆it V )t∈R and invari- ant under conjugation with the modular conjugation JV . In this context, we would like to understand the subsemigroup SM := {g ∈ G : gMg−1 ⊆ M} of those elements of G acting by endomorphisms on M ([BLS11, LL15, Le15]). As gΩ = Ω for g ∈ G, we have VgMg−1 = gVM, so that gMg−1 ⊆ M implies gVM ⊆ VM. For V = VM, we therefore have (2) It follows in particular that, if SM has interior points, then so does the semigroup SV . In the present paper we determine its Lie wedge 1 SM ⊆ SV := {g ∈ G : gV ⊆ V }. L(SV ) = {x ∈ g : exp(R+x) ⊆ SV }, 1In the theory of Lie semigroups ([HHL89, HN93]) Lie wedges are the semigroup analogs of the Lie algebras of closed subgroups. A Lie wedge is a closed convex cone W in a Lie algebra g such that ead xW = W for x ∈ W ∩ −W . In particular, linear subspaces are Lie wedges if and only if they are Lie subalgebras. 2 i.e., the set of generators of its one-parameter subsemigroups in the Lie algebra g of G ([HHL89, HN93]). The current interest in standard subspaces arose in the 1990s from the work of Borchers and Wiesbrock ([Bo92, Wi93]). This in turn led to the concept of modular localization in Quantum Field Theory introduced by Brunetti, Guido and Longo in [BGL02, BGL94, BGL93]. We refer to Subsection 5.1 for more on the relation to von Neumann algebras. Compared to the rather inaccessible object SM, the semigroup SV can be analyzed in terms of antiunitary representations of graded Lie groups: A graded Lie group is a pair (G, εG), where εG : G → {±1} is a homomorphism and we write G± = ε−1 G (±1), so that G+ E G is a normal subgroup of index 2 and G− = G \ G+. An important example is the group AU(H) of unitary or antiunitary operators on a complex Hilbert space with AU(H)+ = U(H). A morphism of graded groups U : G → AU(H) is called an antiunitary representation. Then U (G+) ⊆ U(H) and U (G−) consist of antiunitary operators. We write Stand(H) for the set of standard subspaces of H. We have already seen that every standard subspace V determines a pair (∆V , JV ) of modular objects and that V can be recovered from this pair by V = Fix(JV ∆1/2 V ). This observation can be used to obtain a representation theoretic parametrization of Stand(H): each standard subspace V specifies a homomorphism U V : R× → AU(H) by U V (et) := ∆−it/2π V , U V (−1) := JV . (3) We thus obtain a bijection between Stand(H) and antiunitary representations of the graded Lie group R× with ε(r) = sgn(r) ([N ´O17]). For a given antiunitary representation (U, H) of a graded Lie group (G, εG), we thus obtain a natural map, the Brunetti -- Guido -- Longo map BGL : Homgr(R×, G) → Stand(H), γ 7→ Vγ with U Vγ = U ◦ γ (4) ([BGL02], [N ´O17]). Note that γ ∈ Homgr(R×, G) is completely determined by h := γ′(1) ∈ g and σ := γ(−1). As σ2 = e, it defines an involution τG(g) := σgσ on G, an involution τ = Ad(σ) on g with τ (h) = h, and G ∼= G+ ⋊ {idG, τG}. We thus arrive at the problem to determine for an injective antiunitary representation (U, H) of a graded Lie group (G, εG) and a standard subspace V = Vγ ⊆ H obtained by the BGL construction from a pair (τ, h), consisting of an involutive automorphism τ of g and an element h ∈ g with τ (h) = h, the semigroup SV := {g ∈ G+ : U (g)V ⊆ V }. A crucial piece of information on SV is contained in its Lie wedge L(SV ). To formulate our main results, for λ ∈ R and an ad h-invariant subspace F ⊆ g, write Fλ(h) := ker(ad h − λ idg) ∩ F for the corresponding eigenspace. We also put h := ker(τ − id) and q := ker(τ + id) and write CU := {x ∈ g : −i∂U (x) ≥ 0} for the positive cone of U . The Structure Theorem (Theorem 4.4) asserts that L(SV ) = C− ⊕ h0(h) ⊕ C+ (5) for the two pointed closed convex cones C± := L(SV ) ∩ q±1(h) = ±CU ∩ q±1(h). Further, L(SV ) spans a 3-graded Lie subalgebra gred, and the cones C± are abelian subsets of g. So we obtain an explicit description of the Lie wedge L(SV ) in terms of the positive cone CU of the representation (U, H), the involution τ of g induced by JV , and the generator h ∈ gτ of the modular group. It shows in particular that the most interesting situations are those 3 where g is 3-graded by ad h, i.e., g = g−1(h) ⊕ g0(h) ⊕ g1(h), and τ = eπi ad h. In this context, the representation U should be such that the cones CU ∩ g±1(h) generate g±1(h). We refer to Subsection 5.3 for more comments on classification problems. One of our key tools is a characterization of the operators contained in the algebra AV := {A ∈ B(H) : AV ⊆ V } of V -real operators in terms of the orbit maps αA(t) := αt(A) := ∆−it/2π V A∆it/2π V . The Araki -- Zsid´o Theorem ([AZ05]) asserts that, for A ∈ B(H), A ∈ AV is equivalent to the existence of an analytic continuation of αA from R to the closure of the strip Sπ = {z ∈ C : 0 < Im z < π} satisfying αA(πi) = JV AJV . It follows in particular, that AV is invariant under the involution A♯ := JV A∗JV , and that we obtain for every z ∈ Sπ an injective representation αz : AV → B(H), A 7→ αA(z) with kαzk ≤ 1. For z = πi commuting with JV . 2 we even obtain a ∗-representation α πi 2 : AV → B(HJV ), A 7→ bA by operators On the Lie group side, we mimic the Araki -- Zsid´o Theorem as follows. For a unitary representation U : G → U (H) of a Lie group G, assumed with discrete kernel, we can extend U to a representation of a semigroup SU = G exp(iCU ), where CU is the positive cone of U , and the polar map G × CU → SU , (g, x) 7→ g exp(ix) is a homeomorphism.2 Then U (g exp(ix)) = U (g)ei∂U (x) provides an extension of U to SU ([Ne00, §XI.2]). To bring modular conjugations into the picture, we also consider an involution τG ∈ Aut(G) (inducing an involution τ on g), for which U extends to an antiunitary representation of the graded Lie group G ⋊ {idG, τG}. Then J := U (τG) is a conjugation satisfying U (τG(g)) = JU (g)J for g ∈ G. For h ∈ h = gτ we can now consider the standard subspace V determined by JV = J and ∆V = e2πi∂U (h), so that ∆−it/2π = U (exp th). Now the role of the semigroup AV ∩ U(H) in the Araki -- Zsid´o Theorem is played by the subsemigroup SU inv ⊆ G, consisting of all elements s ∈ G for which the orbit map βs(t) = exp(th)s exp(−th), defined on R, extends analytically to a map from the closure of the strip Sπ to SU , such that βπi(s) = τG(s). Our second main result is the Inclusion Theorem (Theorem 3.11), asserting that SU It is used to obtain one inclusion in the Structure Theorem mentioned above. It has a partial converse in the Germ Theorem (Theorem 4.1) which shows that both subsemigroup have the same germ, i.e., that there exists an e-neighborhood U ⊆ G with SU inv ⊆ SV . inv ∩ U = SV ∩ U. V The content of this paper is as follows. In Section 2 we study for a standard subspace V ⊆ H the semigroup SV = {g ∈ U(H) : gV ⊆ V } of all unitary endomorphisms of V . First we observe that SV is a group if and only if ∆V is bounded, so that the situation is only interesting if ∆V is unbounded (Lemma 2.1). We also state the Araki -- Zsid´o Theorem (a complete proof is provided in Appendix A) and develop its consequences. In Section 3 we prepare the ground for our analysis of the subsemigroup SU inv ⊆ G ⊆ SU which provides a Lie theoretic framework for verifying the Araki -- Zsid´o condition. The main result in Section 3 is the Inclusion Theorem SU inv ⊆ SV (Theorem 3.11). Since both semigroups SU inv and SV are hard to describe globally, an important consequence of the Inclusion Theorem is the inclusion L(SU inv) ⊆ L(SV ). To use this inclusion to prove the Structure Theorem, 2Such semigroups are called Olshanski semigroups. They first appear in Olshanski's paper [Ol82] and an exposition of their theory can be found in [Ne00]. The refinements needed for representations with non-discrete kernel have recently been worked out in [Oeh18]. 4 we derive an explicit description of the wedge L(SU L(SU )inv = (g + iCU )inv in the abelian context. inv) by interpreting it as a similar object This motivates our independent discussion of the case where G is a real Banach space E, endowed with an involution τ and an operator h ∈ B(E), and W ⊆ E is a pointed closed convex cone invariant under −τ and the one-parameter group eRh (Subsection 3.1). In this simple situation the semigroup (E + iW )inv can be determined very explicitly by elementary means and provides an important prototype for the more general non-abelian situation: (E + iW )inv = (W ∩ E−1 (h)) ⊕ E+ 0 (h) ⊕ (−W ∩ E− −1(h)), where E± = ker(τ ∓ 1). In Subsection 3.2 we then recall the basic facts on Olshanski semigroups ΓG(W ) = G exp(iW ) for invariant cones W ⊆ g. They are non-abelian generalizations of the tubes E + iW . We prove the Inclusion Theorem in Subsection 3.3 and by applying it to the corresponding Lie wedges, we already obtain one inclusion of the Structure Theorem. The proof of the Structure Theorem (Theorem 4.4) is completed in Subsection 4.1, where we also prove the Germ Theorem. In Subsection 4.2 we describe the unit group GV of SV , and in Subsection 4.3 we discuss some classes of examples. We conclude this paper with Section 5 on perspectives and open problems. Some results that we did not find in the appropriate form in the literature are stated and proved in appendices. Notation • For a Lie group G, we write g for its Lie algebra, Ad: G → Aut(g) for the adjoint action of G on g, induced by the conjugation action of G on G, and ad x(y) = [x, y] for the adjoint action of g on itself. • (G, εG) denotes a graded group, where εG : G → {±1} is a homomorphism; G± = ε−1 G (±1). An important example is the group AU(H) of unitary or antiunitary opera- tors on a complex Hilbert space H with AU(H)+ = U(H). A morphism of graded groups U : G → AU(H) is called an antiunitary representation. If G is a topological group, then antiunitary representations are assumed to be continuous with respect to the strong operator topology on AU(H). • For a graded homomorphism γ : R× → G, we write σ := γ(−1), τ = Ad(σ), and h := γ′(1) ∈ gτ . Then g = h ⊕ q for the τ -eigenspaces h = ker(τ − idg) and q = ker(τ + idg). We further write τG(g) := σgσ for the corresponding involution on G. • For a real standard subspace V ⊆ H, we write (∆V , JV ) for the corresponding pair of modular objects with V = Fix(JV ∆1/2 V ). • Horizontal strips in the complex plane are denoted Sα,β := {z ∈ C : α < Im z < β} and we also abbreviate Sβ := S0,β for β > 0. • For a unitary representation U : G → U(H) of a finite dimensional Lie group G, we write H∞ for the dense subspace of smooth vectors ξ, for which the orbit maps U ξ : G → H, g 7→ Ugξ is smooth. We also have the dense subspace Hω ⊆ H∞ of analytic vectors for which the orbit map U ξ is analytic. On H∞ we have a representation dU of the complex Lie algebra gC given on x ∈ g by dU (x)ξ = d the unitary one-parameter group (U (exp tx))t∈R is denoted ∂U (x). It coincides with the closure of the operator dU (x). The closed convex Ad(G)-invariant cone dt(cid:12)(cid:12)t=0U (exp tx)ξ. The infinitesimal generator of is called the positive cone of the representation U . CU := {x ∈ g : − i∂U (x) ≥ 0} 5 2 Endomorphisms of standard subspaces For a standard subspace V ⊆ H, we are interested in the closed subsemigroup SV = {U ∈ U(H) : U V ⊆ V } of the unitary group. In the forthcoming sections, we shall study this semigroup by intersecting with finite dimensional subgroups of U(H). In the present section we discuss it on the general level to develop and present some tools that we shall use below. In Subsection 2.1 we show that SV is a group if and only if ∆V is bounded, so that the situation is only interesting if ∆V is unbounded. To understand the semigroup SV , it is natural to consider the full algebra AV := {A ∈ B(H) : AV ⊆ V } of V -real operators, which contains SV as AV ∩ U(H). In Subsection 2.2 we recall an important characterization of the elements of AV in terms of the orbit maps αA(t) := ∆−it/2π defined by the unitary group generated by ∆V : By results of Araki and Zsid´o [AZ05], A ∈ AV is equivalent to the existence of an analytic continuation of αA from R to the closed strip Sπ satisfying αA(πi) = JV AJV . We thus obtain for every z ∈ Sπ an injective representation αz on H, and for z = πi 2 we even obtain a ∗-representation α πi Subsection 2.3, where we take a brief look at one-parameter semigroups of contractions in AV . : AV → B(HJV ), A 7→ bA by operators commuting with JV . We conclude this section with V A∆it/2π V 2 2.1 The case where all unitary endomorphisms are invertible To understand the subsemigroups SV ⊆ U(H), one needs to understand when they are trivial in the sense that they are groups. This case is characterized in the following lemma which shows that standard subspaces with bounded modular operators ∆V are too rigid to have non-trivial unitary endomorphisms. In the proof we shall need the "complementary" standard subspace V ′ := (iV )⊥R = {ξ ∈ H : (∀v ∈ iV ) Rehv, ξi = 0} = {ξ ∈ H : (∀v ∈ V ) Imhv, ξi = 0}. Then ∆V ′ = ∆−1 V , JV = JV ′ and JV V = V ′ (6) ([Lo08, Prop. 3.2], [N ´O17, Lemma 3.7]). Lemma 2.1. For V ∈ Stand(H), the following are equivalent: (a) ∆V is bounded. (b) V + iV = H. (c) If H ⊇ V is standard, then H = V . (d) If H ⊆ V is standard, then H = V . (e) The closed subsemigroup SV ⊆ U(H) is a group. These conditions are in particular satisfied if H is finite dimensional. Here (e) corresponds to the well-known fact that every closed subsemigroup of the compact group Un(C) is a group; cf. also Proposition 5.8. Proof. The equivalence of (a) and (b) follows from V + iV = D(∆1/2 (b) ⇒ (c): If H = V + iV ∼= V ⊕ iV and H ⊇ V is standard, then H = H ⊕ iH implies V = H. (c) ⇔ (d): Follows from H ⊆ V if and only if V ′ ⊆ H′ and ∆V ′ = ∆−1 V . (d) ⇒ (e): For U ∈ SV the relation U V ⊆ V implies U V = V by (d) because U V is also standard. Then U−1V = V as well, so that U−1 ∈ SV . This shows that SV is a group. (e) ⇒ (d): We show that, if H ⊆ V is a proper standard subspace, then SV is not a group. In fact, the unitary operator U := JH JV satisfies U V = JH JV V = JH V ′ ⊆ JH H′ = H ⊆ V . V ). 6 Therefore U ∈ SV , and since U V is a proper subset of V , the inverse U−1 is not contained in SV . (d) ⇒ (a): We show that, if ∆V is unbounded, then V contains a proper standard subspace V1. Step 1: First we show that D(∆1/2 ). If this is not the case, then V ) 6⊆ D(∆−1/2 V JV D(∆−1/2 V ) = D(∆1/2 V ) ⊆ D(∆−1/2 V ) ) is JV -invariant. Since JV is an involution, this leads to D(∆−1/2 ) = V V implies that D(∆−1/2 JV D(∆1/2 V ) = D(∆1/2 Step 2: By Step 1, there exists a non-zero v0 ∈ V \ D(∆−1/2 consider the closed real hyperplane V ), contradicting the unboundedness of ∆V . V ) because D(∆1/2 V ) = V + iV . We V1 := {w ∈ V : Rehw, v0i = 0} = v⊥R 0 ∩ V ⊆ V. Then V1 ∩ iV1 ⊆ V ∩ iV = {0}. Further, the subspace V ⊥R form of V ′ + iV ′ + Cv0 = D(∆−1/2 implies that V1 + iV1 is dense in H. V ) ⊕ Cv0, so that (V1 + iV1)⊥R = V ⊥R 1 = V ⊥R ⊕ Rv0 = iV ′ ⊕ Rv0 is a real 1 = {0}, and this 1 ∩ iV ⊥R Since every standard subspace V is uniquely determined by the pair (∆V , JV ), we have: Lemma 2.2. For V ∈ Stand(H), the stabilizer in the unitary group coincides with the central- izer of the pair (∆V , JV ): U(H)V := {U ∈ U(H) : U V = V } = {U ∈ U(H) : U JV U−1 = JV , U ∆V U−1 = ∆V }. 2.2 The algebra AV of V -real operators Although we are primarily interested in the subsemigroup SV ⊆ U(H), it is of some advantage to consider also the closed real subalgebra AV = {A ∈ B(H) : AV ⊆ V } of V -real operators. The following characterization of the elements of A in terms of analytic continuation of orbits maps ([Lo08, Thm. 3.18], [AZ05, Thm. 2.12]) will be a central tool in the following. A proof can be found in Appendix A (Theorem A.3). Theorem 2.3. (Araki -- Zsid´o Theorem on V -real operators) For A ∈ B(H), the following are equivalent: (i) A ∈ AV , i.e., AV ⊆ V . (ii) A♯ := JV A∗JV ∈ AV . (iii) ∆1/2 (iv) The map αA : R → B(H), αt(A) := αA(t) := ∆−it/2π is defined on D(∆−1/2 V A∆−1/2 V V ) and coincides there with JV AJV . extends to a strongly contin- uous function on the closed strip Sπ = {z ∈ C : 0 ≤ Im z ≤ π} such that αA is holomorphic on Sπ and αA(πi) = JV AJV . V V A∆it/2π If these conditions are satisfied, then (a) kαA(z)k ≤ kAk for z ∈ Sπ (b) αA(z + t) = αt(αA(z)) = ∆−it/2π (c) αA(z + πi) = JV αA(z)JV for z ∈ Sπ. (d) αA(t)V ⊆ V and αA(t + πi)V ′ ⊆ V ′ for all t ∈ R. αA(z)∆it/2π V V for z ∈ Sπ, t ∈ R. Based on the Araki -- Zsid´o Theorem, we obtain the following remarkable fact, which charac- terizes in particular invertible elements in SV as those commuting either with JV or with ∆V . 7 Corollary 2.4. For a standard subspace V ∈ Stand(H) and A ∈ B(H), the following are equivalent: (i) AV ⊆ V and A commutes with (∆it V )t∈R. (ii) AV ⊆ V and A commutes with JV . (iii) A commutes with JV and (∆it V )t∈R. It follows in particular that U(H)V = {g ∈ SV : g∆V g−1 = ∆V } = {g ∈ SV : gJV g−1 = JV }. (7) Proof. (i) ⇒ (ii): If (i) is satisfied, then the function αA is constant. Hence A = αA(iπ) = JV αA(0)JV = JV AJV implies (ii). (ii) ⇒ (iii): If (ii) holds, then αA(iπ) = JV αA(0)JV = JV AJV = A = αA(0), so that Theo- rem 2.3(b) implies that αA(t + πi) = αA(t) for all t ∈ R. Therefore αA extends to a πi-periodic bounded holomorphic function on all of C. Now Liouville's Theorem implies that αA is con- stant, so that A commutes with (∆it (iii) ⇒ (i): Condition (iii) implies that the constant map αA(z) = A satisfies all requirements of Theorem 2.3(iv), so that A ∈ AV . V )t∈R. Finally, (7) follows from the equivalence of (i) and (ii) and Lemma 2.2. Corollary 2.5. The semigroup SV is invariant under the involution ♯, so that (SV , ♯) is an involutive semigroup. Its unitary group U(SV , ♯) := {s ∈ SV : s♯s = ss♯ = 1} = {s ∈ SV : s♯ = s−1} = {s ∈ SV : JV sJV = s} coincides with its unit group U(H)V = SV ∩ S−1 V . Proof. That SV is ♯-invariant follows from Theorem 2.3(ii). Clearly, U(SV , ♯) consists of units of SV . Conversely, any U ∈ SV ∩ S−1 satisfies U V = V , so that U JV U−1 = JV and thus U ♯ = U−1. V The following proposition is a key tool in the following. It provides an analytic interpolation between the representation U of SV on V by isometries and an an involutive ∗-representation by contractions on the real subspace HJV . Proposition 2.6. For every z ∈ Sπ, the map αz : (AV , ♯) → (B(H), ♯), A 7→ αA(z) is an injective contractive morphism of real involutive unital Banach algebras with the following properties: (i) The restriction of αz to the closed unit ball BV := {A ∈ AV : kAk ≤ 1} is continuous with respect to the strong operator topology on BV and B(H). (ii) For z = πi/2, we have α πi (A)∗ = α πi (A♯) and α πi (A)JV = JV α πi (A), so that α πi 2 2 defines a ∗-representation of AV on the real Hilbert space HJV . 2 2 2 Proof. Clearly, αt is multiplicative for every t ∈ R, so that αAB(t) = αA(t)αB(t) for t ∈ R, A, B ∈ AV . For ξ, η ∈ H, the maps z 7→ hξ, αAB(z)ηi and z 7→ hξ, αA(z)αB(z)ηi 8 are continuous because αA and αB are strongly continuous and bounded. As both functions are holomorphic on Sπ and coincide on R, they coincide on Sπ for all ξ, η ∈ H. This implies that αAB(z) = αA(z)αB(z) for all z ∈ Sπ. For A ∈ AV , we have αA♯ (t) = ∆−it/2π V JV A∗JV ∆it/2π V = JV ∆−it/2π V A∗∆it/2π V JV = JV αA(t)∗JV = αA(t)♯, and therefore αA♯ (z) = αA(z)♯ for z ∈ Sπ. Now we show that αz is injective. If αz(A) = αA(z) = 0, then αA(z + t) = 0 for all t ∈ R, and by analytic continuation we get αA = 0. In particular, A = αA(0) = 0. (i) We have to show that, for ξ ∈ H, the map γξ : BV → H, γξ(A) := αz(A)ξ is continuous with respect to the strong operator topology on AV . As the linear map H → ℓ∞(BV , H), ξ 7→ γξ satisfies kγξk∞ ≤ kξk (Theorem 2.3(a)), it suffices to assume that ξ has finite spectral support with respect to the selfadjoint operator log(∆V ). Then αz(A)ξ = ∆−iz/2π V A∆−1/2 V η for η := ∆i(z−πi)/2π V ξ. By Corollary A.2, the continuity of γξ on BV follows from the continuity of the maps BV → H, A 7→ ∆1/2 V A∆−1/2 V η = JV AJV η and A 7→ A∆−1/2 V η (Theorem 2.3(iii)). (ii) For z = πi/2, αz(A) commutes with JV (Theorem 2.3(c)), and thus αz(A)∗ = αz(A)♯ = αz(A♯). On B(H) we consider the W ∗-dynamical system defined by αt(A) = αA(t) = ∆−it/2π V A∆it/2π V for A ∈ B(H), t ∈ R. Then, for each A ∈ AV , the operators αA(z), z ∈ Sπ, belong to the space B(H)ω of α-analytic Lemma 2.7. For an operator B ∈ B(H) commuting with JV , there exists a (unique) A ∈ holomorphic function on the strip S−π/2,π/2 which extends to a strongly continuous function on the closure. Then A = αB(−πi/2). vectors. In particular, bA ∈ B(HJV ). Conversely, we have: AV with bA = B if and only if B is an α-analytic vector whose orbit map αB extends to a Proof. If B = bA = αA(πi/2), then αB(z) := αA(z + πi/2) defines a holomorphic function on S−π/2,π/2 which is strongly continuous on the closure and extends the orbit map of B. Suppose, conversely, that such a function αB exists on S−π/2,π/2. Then the relation JV BJV = B implies that JV αB(z)JV = αB(z) for Im z ≤ π 2 , so that αA(z) := αB(z − πi/2) defines a holomorphic function on Sπ, strongly continuous on the closure, extending the orbit map of A, and which satisfies αA(πi) = αB(πi/2) = JV αB(−πi/2)JV = JV αA(0)JV = JV AJV . In the following we shall mainly work with the characterization of elements A ∈ SV in terms of the analytic continuation of αA to Sπ, but the preceding lemma provides a second perspective: We may also get information on the contraction semigroup cSV ⊆ B(HJV ) and then obtain elements of SV by extending for B ∈ cSV the orbit map αB to − πi 2 . For a contraction B on HJV , the regularity condition of being injective with dense range comes naturally into play. In this regard, we record the following lemma. 9 Lemma 2.8. Let H be a real or complex Hilbert space. Then the subset B(H)reg ⊆ B(H) of injective operators with dense range is a multiplicative ∗-subsemigroup of B(H). It consists of those operators C : H → H for which the partial isometry U in the polar decomposition C = U eB, B = B∗ bounded from above, is unitary. Proof. First we observe that the injective operators and the operators with dense range are multiplicative subsemigroups of B(H). Hence their intersection B(H)reg also is a subsemigroup. As C(H)⊥ = ker(C∗) and C∗(H)⊥ = ker(C), this semigroup is ∗-invariant. If C = U P is the polar decomposition of C, then P = P ∗ ≥ 0, and U is a partial isometry from ker(C)⊥ onto C(H). Therefore the operator C is injective with dense range if and only if U is unitary. Then the positive bounded operator P is injective with dense range, so that it can be written as P = eB for the operator B := log P which is bounded from above. Although all strongly continuous one-parameter semigroups of B(H) which are either sym- metric or unitary are contained in B(H)reg, this is not true in general, as the following simple example shows: Example 2.9. (cf. [EN00, Ex. II.4.31]) On the Hilbert space H = L2([0, 1]) we obtain by (Utf )(x) :=(f (x + t) 0 for x + t ≤ 1, for x + t > 1, a strongly continuous contraction semigroup for which all operators Ut, t > 0, are nilpotent. For N t > 0 we have (Ut)N = UtN = 0. Problems 2.10. Let V ⊆ H be a standard subspace. or find an example where this is not the case. (b) Let B = B∗ = e−H ∈ B(HJV ) be a regular positive contraction for which a unitary A ∈ SV (a) Show that every one-parameter semigroup (Ut)t≥0 of SV satisfiescUt ∈ B(HJV )reg for t ≥ 0 with bA = B exists. Then the same is true for all powers Bn = cAn, n ∈ N, but what about the other operators Bt = e−tH for t ≥ 0? Are they also contained in cSV ? See also Example 5.7 for related problems. 2.3 One-parameter semigroups in AV Classically, bounded strongly continuous one-parameter semigroups on Banach spaces are stud- ied through their infinitesimal generators and their resolvents. We start our analysis in this subsection by recalling some key facts on one-parameter semigroups from [EN00]. This provides some tools used below for one-parameter subsemigroups of finite dimensional semigroups. Remark 2.11. (a) If (Ut)t≥0 is a strongly continuous one-parameter semigroup of contractions on the Banach space X and A : D(A) → X its infinitesimal generator, then we have for every λ ∈ C with Re λ > 0 an integral formula for the resolvent: R(λ, A) := (λ1 − A)−1 =Z ∞ 0 e−tλUt dt and kR(λ, A)k ≤ 1 Re λ (8) ([EN00, Thm. II.1.10]) (b) If, conversely, A : D(A) → X is a closed, densely defined operator on X such that, for λ > 0, the operators λ1 − A : D(A) → X have bounded inverses R(λ, A) satisfying kR(λ, A)k ≤ λ−1, then A is the infinitesimal generator of a uniquely determined semigroup of contractions ([EN00, Thm. II.3.5]). That this semigroup can actually be obtained as the strong limit for t > 0 (9) Ut = lim n→∞(cid:16)1 − t n A(cid:17)−n 10 follows from the discussion in [HP57, §12.3] (see also [Ch68] for related results). Note that our assumption on A implies that (cid:13)(cid:13)(cid:13)(cid:16)1 − t n A(cid:17)−1(cid:13)(cid:13)(cid:13) = n t(cid:13)(cid:13)(cid:13)(cid:16) n t 1 − A(cid:17)−1(cid:13)(cid:13)(cid:13) ≤ 1, so that the right hand side of (9) is a contraction whenever the limit exists. (c) If X is a Hilbert space and A a normal operator, then the assumption on A implies that Spec(A) ⊆ Cℓ := {z ∈ C : Re z ≤ 0}. Then, for any z ∈ Cℓ, we have (1 − tz/n)−n → etz for t ≥ 0, as a pointwise limit of bounded functions on Cℓ. Therefore (9) is an immediate consequence of the measurable spectral calculus and a normal operator A generates a one- parameter semigroup of contractions if and only if Spec(A) ⊆ Cℓ. (d) A linear operator A : D(A) → X on a Banach space is said to be dissipative if k(λ1 − A)ξk ≥ λkξk for λ > 0, ξ ∈ D(A), which is equivalent to k(1 − hA)ξk ≥ kξk for h > 0, ξ ∈ D(A). (10) According to the Lumer -- Phillips Theorem ([EN00, Thm. II.3.15]), a closed densely defined operator A generates a contraction semigroup if and only if it is dissipative and λ1 − A has dense range for some (hence for all) λ > 0. If X is a Hilbert space, then (10) implies that A is dissipative if and only if Rehξ, Aξi ≤ 0 for all ξ ∈ D(A) ([EN00, Prop. II.3.23]). For a standard subspace V ∈ Stand(H), we recall the subalgebra AV ⊆ B(H) from The- orem 2.3. We are mainly interested in the semigroup SV = AV ∩ U(H), and for that the representations αz of AV will be extremely helpful. As AV is strongly closed, (8) and (9) in Remark 2.11 lead to: Proposition 2.12. (One-parameter semigroups of contractions in AV ) Let (Ut)t≥0 be a strongly continuous one-parameter semigroup of contractions on H with infinitesimal generator A. Then (∀t > 0) UtV ⊆ V ⇐⇒ (∀λ > 0) (λ1 − A)−1V ⊆ V. Corollary 2.13. Let (Ut)t≥0 be a strongly continuous one-parameter semigroup of contractions in AV with infinitesimal generator A. Then, for every z ∈ Sπ, (αz(Ut))t≥0 is a strongly con- tinuous one-parameter semigroup of contractions on H. Its infinitesimal generator Az satisfies (λ1 − Az)−1 = αz((λ1 − A)−1) for λ > 0. Proof. The first assertion follows immediately from Proposition 2.6. By Proposition 2.12, (λ1 − A)−1 ∈ AV for every λ > 0, so that αz((λ1 − A)−1) is defined for z ∈ Sπ. For the second assertion we now use (8) in Remark 2.11 and the continuity of the representations αz. In the following, Corollary 2.13 is of particular interest for z = πi/2. If A♯ = A, then it leads to the infinitesimal generator bA = Aπi/2 of a symmetric contraction semigroup on HJV , showing that bA ≤ 0. This will be important in the proof of the Germ Theorem (Theorem 4.1). The following observation will not be used below, but we record it here because it adds interesting information on certain results obtained in [Ne18], where we have seen that Stand(H) carries the structure of a reflection space, specified by (−1)V (W ) = JV W ′. Accordingly, a curve γ : R → Stand(H) is called a geodesics if it is a morphism of reflection spaces, where R carries the canonical reflection structure given by the point reflections (−1)x(y) = 2y − x. By [Ne18, Prop. 2.9], geodesics γ : R → Stand(H) with γ(0) = V for which the curve (Jγ(t))t∈R is strongly continuous, are the curves of the form γ(t) = UtV, where (Ut)t∈R is a unitary one-parameter group satisfying JV UtJV = U−t for t ∈ R. 11 Proposition 2.14. Assigning to the generator A = A♯ = −A∗ of a strongly continuous ♯- symmetric unitary one-parameter semigroup in AV the curve (etAV )t∈R in Stand(H), we obtain a bijection onto the set of decreasing geodesics γ : R → Stand(H) with γ(0) = V . Proof. The relation A♯ = A is equivalent to A∗ = JV AJV . If, in addition, A∗ = −A, then JV AJV = −A. Then the curve γ(t) := etAV defines a geodesic in Stand(H) which is decreasing because t < s implies that γ(s) = etAe(s−t)AV ⊆ etAV = γ(t). That all decreasing geodesics are of this form follows from [Ne18, Prop. 2.9]. 3 Wick rotations of tubes and Olshanski semigroups To apply tools from finite dimensional Lie theory, we consider subsemigroups of B(H) that arise by analytic continuation of a unitary representation U : G → U (H) of a Lie group G to a semigroup SU = G exp(iCU ), where CU := {x ∈ g : − i∂U (x) ≥ 0} is the positive cone of U . Assuming that U has discrete kernel, the semigroup SU always exists and U (g exp(ix)) = U (g)ei∂U (x) provides an extension of U to SU . To implement JV as well, we also consider an involution τG ∈ Aut(G) (inducing an involution τ on g), for which U extends by to an antiunitary representation of the graded Lie group G ⋊ {idG, τG}. Then J := U (τG) is a conjugation satisfying U (τG(g)) = JU (g)J for g ∈ G. For h ∈ gτ we now consider the standard subspace V determined by JV = J and ∆V = e2πi∂U (h), so that ∆−it/2π V = U (exp th). By the Araki -- Zsid´o Theorem, we are now led to the problem to determine the subsemigroup SU inv of those elements s ∈ G for which the orbit map βs(t) = exp(th)s exp(−th) extends holomorphically to the closure of Sπ, in such a way that βπi(s) = τG(s). In Theorem 3.11, we show that SU inv ⊆ SV = {g ∈ G : U (g)V ⊆ V }. To prepare this theorem, we start in Subsection 3.1 with a discussion of the "abelian case", where G is simply a real Banach space E, endowed with an involution τ and an endomorphism h, and W ⊆ E is a pointed closed convex cone invariant under −τ and the one-parameter group eRh. In this simple situation the semigroup (E + iW )inv is a closed convex cone in E that can be determined very explicitly by elementary means. It provides an important blueprint for the more general non-abelian situation. In Subsection 3.2 we then recall the basic facts on Olshanski semigroups ΓG(W ) for invariant cones W ⊆ g. They are the non-abelian generalizations of the tube E + iW . Finally, we verify the inclusion SU inv ⊆ SV in Subsection 3.3. 3.1 Wick rotations of tubes In this section we develop another tool that we shall use below in the context of Lie algebras. This subsection represents some key geometric features that can already be formulated in the abelian context. Let E be a real Banach space endowed with the following data: • A continuous involution τ ∈ GL(E); we write E = E+ ⊕ E−, E± := ker(τ ∓ 1) for the τ -eigenspace decomposition. • An endomorphism h ∈ B(E), commuting with τ . • A closed convex cone W ⊆ E which is pointed, i.e., W ∩ −W = {0}, and invariant under −τ and the one-parameter group eRh. We consider the closed convex cone TW := E + iW ⊆ EC, 12 which is obviously invariant under eRh and −τ , where we use the same notation for the complex linear extensions to EC. We do not assume that the cone W has interior points, so that W − W may be a proper subspace of E. If σc : EC → EC is the antilinear involution with fixed point set Ec := E+ + iE−, then σc acts on iE as −τ , so that σc(TW ) = TW , and T σc W = TW ∩ Ec = E+ + i(W ∩ E−) is the closed convex cone of σc-fixed points in TW . We are interested in the closed convex cone TW,inv := {x ∈ E : eyihx ∈ TW for y ∈ [0, π]; eπihx = τ (x)}. Lemma 3.1. (a) For x ∈ E, the condition eπihx = τ (x) is equivalent to e (b) TW,inv = {x ∈ E : ezhx ∈ TW for z ∈ Sπ; eπihx = τ (x)}. πi 2 hx ∈ Ec. Proof. (a) As above, let σc : EC → EC denote the antilinear extension of τ , so that Fix(σc) = 2 hx, the condition xc ∈ Ec is equivalent to σc(xc) = xc, which is Ec. For x ∈ E and xc := e equivalent to πi πi 2 ad hx = σc(e e πi 2 ad hx) = e− πi 2 ad hσc(x) = e− πi 2 ad hτ (x). This in turn is equivalent to eπi ad hx = τ (x). (b) follows from the fact that TW is invariant under eRh. For an h-invariant real subspace F ⊆ EC, we write Fλ = Fλ(h) := F ∩ ker(h − λ1) for the h-eigenspaces in F . Lemma 3.2. For x ∈ E, the condition eπihx = τ (x) is equivalent to the existence of finitely many elements xn ∈ En(h) with x =Pn∈Z xn and τ (xn) = (−1)nxn. Proof. We write x = x+ + x− with x± ∈ E±. Then eπihx = τ (x) is equivalent to eπihx+ = x+ and eπihx− = −x−. (11) Combining both, we see that e2πihx = x. The space Efix C of fixed points of the automorphism e2πih ∈ GL(EC) carries a norm continuous action of the circle T ∼= R/Z, defined by βt(y) := e2πithy. As h is bounded, Efix C is a direct sum of finitely many h-eigenspaces EC,n(h), n ∈ Z. Accordingly, we write As khk < ∞, only finitely many summands are non-zero. The antilinear involution σ of EC, whose fixed point set is E, commutes with h. Therefore the h-eigenspaces are σ-invariant, and thus σ(x) = x implies σ(xn) = xn for every n ∈ Z, i.e., xn ∈ E. Now eπihxn = (−1)nxn and (11) imply that x+ is the sum of the xn with n even, and x− is the sum of the xn with n odd. As τ (x±) = ±x±, this in turn shows that τ (xn) = (−1)nxn for n ∈ Z. If, conversely, x =Pn∈Z xn with xn ∈ E satisfying hxn = nxn and τ (xn) = (−1)nxn, then the relation eπihx = τ (x) is obvious. The following proposition is a key geometric ingredient of the proof of our Structure Theorem (Theorem 4.4). Proposition 3.3. The cone TW,inv has the simple form TW,inv = (W ∩ E−1 (h)) ⊕ E+ 0 (h) ⊕ (−W ∩ E− −1(h)). (12) In particular, it is determined by the two cones W ∩ E− ±1(h). 13 x =Xn∈Z xn with hxn = nxn. Proof. First we note that the cone TW,inv is closed and invariant under eRh because TW is invariant under eRh, the operators eyih commute with eRh, and so does τ . Let x ∈ TW,inv. Then Lemma 3.2 implies that x =Pn∈Z xn is a finite sum with xn ∈ En(h) and τ (xn) = (−1)nxn. We claim that xn = 0 for n > 1. Suppose first that there exists an n > 1 with xn 6= 0 and assume that n is maximal with this property. Then the invariance of TW,inv under eRh and its closedness imply that xn = lim t→∞ e−ntXm∈Z emtxm = lim t→∞ e−ntethx ∈ TW,inv. (13) As n > 1, we now obtain e[0,π]ihxn = e[0,nπ]ixn ∋ ±ixn. This leads to ±xn ∈ W , and since W is pointed, we arrive at the contradiction xn = 0. An analogous argument shows that xn = 0 for n < −1. This shows that x = x1 + x0 + x−1 with x0 ∈ E+ 2 hx±1 ∈ TW = E + iW , hence x±1 ∈ ±W ∩ E− ±W ∩ E− 0 (h), and x±1 ∈ TW,inv, obtained from (13), implies that ±ix±1 = e Conversely, every element of the form x = x−1 + x0 + x1 with x0 ∈ E+ ±1(h) is contained in TW,inv because eπihx = −x−1 + x0 − x1 = τ (x), and 0 (h) and x±1 ∈ ±1(h). πi eyihx = x0{z}∈E+ ∈E− {z } because sin(y)(x1 − x−1) ∈ W for y ∈ [0, 2π]. + cos(y)(x−1 + x1) + i sin(y)(x1 − x−1) ∈ E + iW, iE− {z } From Proposition 3.3 we immediately obtain: Corollary 3.4. Let g be a finite dimensional real Lie algebra, endowed with an involution τ ∈ Aut(g) with eigenspace decomposition g = h ⊕ q, h = ker(τ − 1) and q := ker(τ + 1), an element h ∈ h, and a pointed closed convex cone W ⊆ g, invariant under −τ and eR ad h. For TW := g + iW , the cone TW,inv then has the simple form TW,inv = (W ∩ q1(h)) ⊕ h0(h) ⊕ (−W ∩ q−1(h)). (14) In particular, it is determined by the two pointed cones W ∩ q±1(h). 3.2 Extensions of unitary representations to semigroups Below we shall need non-abelian analogs of the tubes TW = E + iW , where E is replaced by a finite dimensional simply connected Lie group G and W ⊆ g is an Ad(G)-invariant closed convex cone. Definition 3.5. (Olshanski semigroups) Let G be a 1-connected Lie group with Lie algebra g and W ⊆ g be a pointed Ad(G)-invariant closed convex cone. 3 The corresponding Olshanski semigroup ΓG(W ) is defined as follows. Let GC be the 1-connected Lie group with Lie algebra gC and let ηG : G → GC be the canonical morphism of Lie groups for which L(η) : g ֒→ gC is the inclusion. 4 As GC is simply connected, the complex conjugation on gC integrates to an antiholomorphic involution σ : GC → GC with σ ◦ ηG = ηG, and this implies that ηG(G) coincides with the subgroup (GC)σ of σ-fixed points in GC. 5 3Then W is weakly elliptic in the sense that Spec(ad x) ⊆ iR holds for every x ∈ W . In fact, by [Ne00, Prop. VII.3.4(b)] W is weakly elliptic in the ideal W − W , and since [x, g] ⊆ W − W holds for any x ∈ W , we have Spec(ad x) ⊆ {0} ∪ Spec(ad xW−W ) ⊆ iR. 4 In general the map ηG is not injective, as the example G = fSL2(R) with GC = SL2(C) shows. 5Since GC is simply connected, this subgroup is connected by Corollary B.3. 14 As W is weakly elliptic, Lawson's Theorem ([Ne00, Thm. XIII.5.6]) asserts that Γ′G(W ) := Γ(GC )σ (W ) := (GC)σ exp(iW ) ⊆ GC is a closed subsemigroup of GC stable under the antiholomorphic involution s∗ := σ(s)−1, and that the polar map (GC)σ × W → Γ′G(W ), (g, x) 7→ g exp(ix) is a homeomorphism. Next we observe that ker ηG is a discrete subgroup of G and define ΓG(W ) as the simply connected covering of Γ′G(W ) ([Ne00, Def. XI.1.11]). Basic covering theory implies that ΓG(W ) inherits an involution given by (g exp(ix))∗ = exp(ix)g−1 = g−1 exp(Ad(g)ix) and a homeomorphic polar map G × W → ΓG(W ), (g, x) 7→ g exp(ix). We write exp : g + iW → ΓG(W ) for the canonical lift of the exponential function exp: L(Γ′G(W )) = g + iW → Γ′G(W ) ⊆ GC. For every x ∈ g + iW , the curve γx(t) := exp(tx) is a continuous one-parameter semigroup of ΓG(W ). If W has interior points, then the polar map restricts to a diffeomorphism from (GC)σ × W 0 onto the interior Γ′G(W 0) of Γ′G(W ). Further, ΓG(W 0) = G exp(iW 0) is a complex mani- fold with a holomorphic multiplication and the exponential function g + iW 0 → ΓG(W 0) is holomorphic, whereas the involution ∗ is antiholomorphic ([Ne00, Thm. XI.1.12]). We now turn to the analytic continuation of unitary representations of G to Olshanski semigroups ΓG(W ). Proposition 3.6. (Holomorphic extension of unitary representations) Let (U, H) be a uni- tary representation of G with discrete kernel and consider the ideal nU := CU − CU and the corresponding normal integral subgroup NU E G.6 Then the following assertions hold: (i) U extends by U (g exp(ix)) = U (g)ei∂U (x) to a strongly continuous contraction representa- tion of the Olshanski semigroup SU := G exp(iCU ) which is holomorphic on the complex manifold NU exp(iC 0 U ). (ii) If J : H → H is a conjugation and τG ∈ Aut(G) an involution with derivative τ ∈ Aut(g), satisfying JU (g)J = U (τG(g)) for g ∈ G, then the involutive automorphism of SU given by τS(g exp(ix)) = τG(g) exp(−iτ (x)) satisfies JU (s)J = U (τS(s)) for s ∈ SU . Proof. (i) The assumption that ker(U ) is discrete implies that CU is pointed. That U (g exp(ix)) = U (g)ei∂U (x) defines a representation which is holomorphic and non-degenerate on ΓNU (C 0 U ) = NU exp(iC 0 U ) follows from [Ne00, Thm XI.2.5]. Now [Ne00, Cor. IV.1.18, Prop. IV.1.28] imply that U is strongly continuous on ΓNU (CU ) because U (ΓNU (CU )) is bounded. The continuity on SU now follows from SU = GΓNU (CU ) = G exp(iCU ), the fact that the polar map is a homeomorphism, and the strong continuity of the multiplication on the operator ball. (ii) The relation JU (g)J = U (τG(g)) implies J∂U (x)J = ∂U (τ (x)) for x ∈ g, and therefore Ji∂U (x)J = −i∂U (τ (x)) implies that −τ (CU ) = CU . Therefore the involution τS(g exp(ix)) = τG(g) exp(−iτ (x)) on SU is defined. As it is the unique continuous lift of an automorphism of Γ′G(CU ) ⊆ GC, preserving the base point e ∈ SU , it defines an automorphism of SU = ΓG(CU ). For s = g exp(ix) we have JU (s)J = JU (g)JJei∂U (x)J = U (τG(g))e−i∂U (τ x) = U (τG(g) exp −τ (x)) = U (τS(s)). 6Normal integral subgroups of 1-connected Lie groups are always closed and 1-connected by [HN12, Thm. 11.1.21]. 15 Remark 3.7. (a) Let (U, H) be an antiunitary representation of the graded Lie group (G, εG) and σ ∈ G− be an involution. We write τG(g) = σgσ for the corresponding involutive auto- morphism of G and τ = Ad(σ) ∈ Aut(g) for the corresponding involution of the Lie algebra. Then the positive cone CU of U is a closed convex cone satisfying Ad(g)CU = εG(g)CU for g ∈ G. In particular, it is invariant under −τ (cf. Proposition 3.6(ii)). (b) The fixed point set of the involution τS on SU is the subsemigroup SU G := GτG exp(i(q ∩ CU )) ⊆ SU because s = g exp(ix) is τS-invariant if and only if τG(g) = g and τ (x) = −x, i.e., x ∈ q ∩ CU . 3.3 Wick rotations of Olshanski semigroups In this subsection we describe how holomorphic extensions of unitary representations of complex Olshanski semigroups can be used to obtain non-trivial endomorphism semigroups SV ⊆ G for certain standard subspaces. Let G be a 1-connected Lie group and W ⊆ g be a pointed invariant closed convex cone, so that we have the Olshanski semigroup ΓG(W ) = G exp(iW ) which is the simply connected covering of the semigroup Γ′G(W ) ⊆ GC. We write qS : ΓG(W ) → Γ′G(W ) ⊆ GC for the universal covering map (Definition 3.5). We further assume that τG ∈ Aut(G) is an involution and that the corresponding automorphism τ ∈ Aut(g) satisfies τ (W ) = −W . For an element h ∈ h = gτ , we consider the R-action on ΓG(W ), given by βt(s) := βs(t) := exp(th)s exp(−th) for s ∈ ΓG(W ), t ∈ R and note that the corresponding R-action on GC extends to a holomorphic C-action by βz(g) := βg(z) := exp(zh)g exp(−zh) for z ∈ C, g ∈ GC. (15) Definition 3.8. If M is a complex manifold, then we call a continuous map f : M → ΓG(W ) holomorphic if the composition qS ◦ f : M → GC is holomorphic. Holomorphic extensions of the orbits maps βs : R → ΓG(W ) have to be understood in this sense. For z ∈ C, we say that βs(z) exists if there exists a closed strip Sa,b ⊆ C containing R and z, and an extension of βs to a continuous map Sa,b → ΓG(W ) which is holomorphic on Sa,b. Then we write βs(z) = βz(s) for the value of this analytic continuation in z, which does not depend on the choice of a and b as long as a ≤ Im z ≤ b. Lemma 3.9. For z ∈ C, let ΓG(W )z ⊆ ΓG(W ) be the set of all elements s ∈ ΓG(W ) for which βs(z) exists. Then the following assertions hold: (i) ΓG(W )z = q−1 S (Γ′G(W )z) and qS ◦ βz = βz ◦ qS on ΓG(W )z. (ii) ΓG(W )z is a closed subsemigroup of ΓG(W ) and βz : ΓG(W )z → ΓG(W ) is a continuous homomorphism. (iii) The closed subsemigroup ΓG(W )τS := GτG exp(i(q ∩ W )) is the set of fixed points of the involutive automorphism τS of ΓG(W ), defined by τS(g exp(ix)) = τG(g) exp(−iτ (x)). (iv) ΓG(W )inv := {g ∈ G ∩ ΓG(W )πi : βπi(g) = τG(g)} is a closed subsemigroup of ΓG(W ) with Lie wedge L(ΓG(W )inv) = L(ΓG(W ))inv = (g + iW )inv. We recall from Corollary 3.4 the explicit description of (g + iW )inv as (g + iW )inv = (W ∩ q1(h)) ⊕ h0(h) ⊕ (−W ∩ q−1(h)). 16 Proof. (i) Since qS ◦ βt = βt ◦ qS holds for t ∈ R, the uniqueness of analytic continuation implies that qS(ΓG(W )z) ⊆ Γ′G(W )z with qS ◦ βz = βz ◦ qS : ΓG(W )z → GC. If s ∈ ΓG(W ) is such that qS(s) ∈ Γ′G(W )z, we fix an analytic continuation βqS (s) : Sa,b → Γ′G(W ) ⊆ GC of the orbit map βqS (s) : R → Γ′G(W ). As the closed strip Sa,b is simply connected, there exists a unique continuous lift eβs : Sa,b → ΓG(W ) with qS ◦eβs = βqS (s) ◦ qS and eβs(0) = s. (ii) In GC we have for Im z > 0 that Then the uniqueness of continuous lifts to coverings implies that eβs(t) = βs(t) for t ∈ R and, by construction, eβs is holomorphic on Sa,b. This implies that s ∈ ΓG(W )z with βs(z) = eβs(z). Γ′G(W )z = {s ∈ Γ′G(W ) : 0 ≤ Im w ≤ Im z ⇒ βw(s) ∈ Γ′G(W )} = \0≤Im w≤Im z β−1 w (Γ′G(W )), where βw ∈ Aut(GC) is the unique automorphism from (15) with L(βw) = ew ad h. Since Γ′G(W ) is a closed subset of GC, the subset Γ′G(W )z of Γ′G(W ) is closed. Now (i) implies that ΓG(W )z = q−1 S (Γ′G(W )z) is also closed. Next we show that ΓG(W )z is a subsemigroup on which βz is multiplicative. Let s1, s2 ∈ ΓG(W )z and consider the minimal strip Sa,b ⊆ C with R ∪ {z} ⊆ Sa,b. Then the map βs1 · βs2 : Sa,b → ΓG(W ), z 7→ βs1 (z)βs2 (z), is continuous and holomorphic on Sa,b. For t ∈ R, we have (βs1 βs2 )(t) = βt(s1)βt(s2) = βt(s1s2) because βt is an automorphism of ΓG(W ). Uniqueness of analytic continuation there- fore implies that s1s2 ∈ ΓG(W )z with βz(s1s2) = βs1 (z)βs2 (z) = βz(s1)βz(s2). (iii) On GC we have a unique antiholomorphic involution σc inducing on gC the antilinear extension of τ , so that its group of fixed points has the Lie algebra gc. It acts on s = g exp(ix) by σc(s) = τG(g) exp(−iτ (x)). By uniqueness of lifts to coverings, this implies that τS defines an involutive automorphism of ΓG(W ), and the assertion follows immediately from the formula for τS. (iv) follows from the trivial observation that, for a family (Sj )j∈J of closed subsemigroups of ΓG(W ), we have L(cid:0)Tj Sj(cid:1) =Tj L(Sj). The following lemma provides an interesting tool that permits us to work effectively with holomorphic maps with values in ΓG(W ), which neither is a manifold nor "complex". Lemma 3.10. Let W ⊆ g be a closed pointed convex invariant cone and U : ΓG(W ) → B(H), U (g exp(ix)) = U (g)ei∂U (x) be a ∗-representation obtained from a unitary representation U of G. Then, for every holo- morphic map f : M → ΓG(W ), M a finite dimensional complex manifold, the composition U ◦ f : M → B(H) is holomorphic. Proof. As the assertion is local with respect to M , we may w.l.o.g. assume that M is connected and that f (M ) has compact closure in ΓG(W ). Let f : M → ΓG(W ) be a holomorphic map. Then f′ := qS ◦ f : M → Γ′G(W ) = GC is holomorphic by definition. We consider the ideal n := W − W E g and the corresponding normal integral subgroup N E G. As N is closed and 1-connected by [HN12, Thm. 11.1.21], we obtain a quotient group Q := G/N . We likewise have a closed normal subgroup NC E GC and QC := GC/NC. Let r : GC → QC denote the quotient map. Then r(Γ′G(W )) = r(G exp(iW )) = r(G) ⊆ (QC)σ ⊆ QC 17 is contained in the totally real submanifold (QC)σ of fixed points of the antiholomorphic invo- lution σ of QC corresponding to the complex conjugation of qC with respect to q. We conclude that the holomorphic map r ◦ f′ : M → r(G) ⊆ QC is constant, hence equal to r(g′) for some g′ ∈ qS(G) ∩ f′(M ). This implies the existence of a holomorphic map h : M → NC with f′(m) = g′h(m) for m ∈ M . Lifting to the covering space ΓG(W ), we conclude that there exists a g ∈ G with f (m) = gh(m) for all m ∈ M , where h : M → ΓN (W ) is a holomorphic map. Pick x ∈ W 0 (the interior of W with respect to n) and put sn := exp(n−1x) ∈ ΓG(W 0). We thus obtain a sequence U (sn)∗ = U (sn) of hermitian operators converging strongly to 1. Further, snh(M ) is contained in the complex manifold ΓN (W 0) and the map hn : M → ΓN (W 0), m 7→ snh(m) is holomorphic. Therefore the maps Hn := U ◦ hn : M → B(H) are holomorphic and we want to show that H := U ◦ h is also holomorphic. For ξ, η ∈ H, we have lim n→∞ hξ, Hn(m)ηi = lim n→∞ hU (sn)ξ, H(m)ηi = hξ, H(m)ηi, and the boundedness of H(M )η implies that the convergence is uniform on M . This shows that the bounded function H : M → B(H) is weakly holomorphic, hence holomorphic by [Ne00, Cor. A.III.5]. Finally, the relation U (f (m)) = U (g)U (h(m)) = U (g)H(m) implies that U ◦ f is holomorphic. The following theorem is the main result of this section. It provides a mechanism to con- inv ⊆ SV . It implements the analytic struct unitary endomorphisms of V by the inclusion SU continuation process from Theorem 2.3 inside the Olshanski semigroup SU . Theorem 3.11. (Inclusion Theorem) Let G be a 1-connected Lie group with the involution τG and τ ∈ Aut(g) the induced automorphism. Further, let (U, H) be a continuous antiunitary representation of G ⋊ {idG, τG} with discrete kernel and consider the standard subspace V ⊆ H specified by JV = U (τG) and ∆V = e2πi∂U (h) for some h ∈ gτ . Then SU inv ⊆ SV := {g ∈ G : U (g)V ⊆ V } and (g + iCU )inv ⊆ L(SV ). Proof. We write U : SU → B(H), g exp(ix) 7→ U (g)ei∂U (x) for the canonical extension of the unitary representation U to SU (Proposition 3.6). For s ∈ SU inv, we consider the bounded function which is defined because βz(s) ∈ SU for z ∈ Sπ. We have F : Sπ → B(H), F (z) := U (βz(s)) F (z + t) = U (exp th)F (z)U (exp(−th)) = ∆−it/2π V F (z)∆it/2π V for t ∈ R, z ∈ Sπ, and F is strongly continuous (Proposition 3.6). That it is holomorphic on Sπ follows from Lemma 3.10 and the holomorphy of the map Sπ → ΓG(CU ), z 7→ βz(s) in the sense of Defini- tion 3.8. We further note that F (0) = U (s) ∈ U (G) is unitary and that JV F (0)JV = JV U (s)JV = U (τG(s)) = U (βπi(s)) = F (πi). Now Theorem 2.3(iv) implies that U (s) = F (0) ∈ AV , and thus s ∈ SV . The inclusion of the Lie wedges is an immediate consequence of L(SU inv) = (g + iCU )inv (Lemma 3.9(iv)). Problem 3.12. Show that we actually have the equality SU inv = SV . In the Germ Theorem (Theorem 4.1 below) we shall see that both subsemigroup do at least have the same germ, i.e., that there exists an e-neighborhood U in G with SU inv ∩ U = SV ∩ U. 18 Remark 3.13. (a) The construction in the preceding proof shows that, for s ∈ SU element sc := β πi (s) = h exp(ix) ∈ SU satisfies inv, the 2 (U (s)) = U (sc) = U (h)ei∂U (x). bU (s) = α πi 2 Therefore [U (s) is injective with dense range (cf. Lemma 2.8). If SV coincides with SU inv, this implies that cSV ⊆ B(H)reg. (b) On the level of the Lie wedge L(SV ), we know that (g + iCU )inv = (CU ∩ q1) ⊕ h0 ⊕ (−CU ∩ q−1) is a cone with a rather simple structure and completely determined by the pair (τ, h) and the cone CU . In Section 4, we study SV from a different perspective and we shall see that this cone actually generates a 3-graded Lie subalgebra (Theorem 4.4 and Corollary 4.5). The global structure of the semigroup SU inv is hard to analyze in the non-abelian context. In Subsection 5.4 we explain how to reduce the determination of this semigroup to the case where e2πi ad h = idgC , i.e., ad h is diagonalizable with integral eigenvalues. 4 The subsemigroups SV in finite dimensional groups As in Section 3.3, we start in this section with an antiunitary representation (U, H) of a finite dimensional graded Lie group G ⋊{idG, τG}, where G is 1-connected, and consider the standard subspace V = Vγ determined by JV = U (τG) and ∆−it/2π = U (exp th) with h ∈ h as explained in the introduction. Under the assumption that U has discrete kernel, we determine the Lie wedge of the closed subsemigroup V SV = {g ∈ G : U (g)V ⊆ V } with the unit group GV := {g ∈ G : U (g)V = V }. Our main result on L(SV ) (Theorem 4.4) asserts that L(SV ) = (−CU ∩ q−1(h)) ⊕ h0(h) ⊕ (CU ∩ q1(h)) and that L(SV ) spans a 3-graded Lie subalgebra of g. This result is based on the Germ Theorem (Theorem 4.1), asserting the existence of an e-neighborhood U in G with SU inv ∩ U = SV ∩ U, which implies in particular that L(SV ) = L(SU inv). In Subsection 4.2 we discuss the unit group GV of SV , and we discuss some examples in Subsection 4.3. 4.1 The Lie wedge of SV The following theorem shows that the subsemigroups SV and SU i.e., identical intersection with some e-neighborhood. inv of G have the same germ, Theorem 4.1. (The Germ Theorem) If ker(U ) is discrete, then there exists an e-neighborhood U ⊆ G such that SV ∩ U = Sinv ∩ U. inv ⊆ SV . Therefore it suffices to find inv. We write Hω ⊆ H for the subspace of analytic vectors of the Proof. In Theorem 3.11 we have already seen that SU U such that SV ∩ U ⊆ SU representation U of G (cf. [Nel59, Thm. 4]). Step 1: By [HN93, Lemma 9.16], there exists a dense subspace D ⊆ Hω which is equianalytic in the sense that there exists an open convex circular 0-neighborhood W ⊆ gC, such that the series U η(x) := 1 n! (dU (x))nη ∞Xn=0 19 converges for η ∈ D and x ∈ W and defines a holomorphic function U η : W → H. It satisfies U η(x) = U (exp x)η for x ∈ W ∩ g (see [Ne11, §6] for more on analytic vectors). Step 2: We claim that U η(ix) = ei∂U (x)η for x ∈ W, (16) (17) where the right hand side has to be understood in terms of the measurable functional calculus for the selfadjoint operator i∂U (x). First we observe that the function F : {z ∈ C : zx ∈ W} → H, F (z) = U η(zx) is holomorphic and coincides for z ∈ [−1, 1] ⊆ R with U (exp zx)η. Let ε > 0 be such that Ω := {a + ib : a < ε, −ε < b < 1 + ε} satisfies Ωx ⊆ W. Then the continuous function F (t) := U (exp tx)η on (−ε, ε) has an analytic continuation to Ω, and this implies that it even extends to the strip Ω′ := R + (−ε, 1 + ε)i ⊆ C, so that [N ´O18, Lemma A.2.5] shows that η ∈ D(eti∂U (x)) for −ε < t < 1 + ε. As the function Ω′ → H, z 7→ ez∂U (x) is also holomorphic, we obtain U η(ix) = ei∂U (x) by analytic continuation. Step 3: Let W′ ⊆ W ⊆ gC be an open convex 0-neighborhood such that the Baker -- Campbell Hausdorff product x ∗ y is defined by the convergent series for x, y ∈ W′ and defines a holo- morphic function W′ × W′ → W ⊆ gC ([HN12, §9.2.5]). We claim that U η(x1 ∗ x2) = U (exp x1)U η(x2) for x1 ∈ W′ ∩ g, x2 ∈ W′ (18) if W′ is chosen small enough. As U (exp x1) is unitary, both sides define H-valued functions holomorphic in x2. Fixing x1 ∈ g ∩ W′, both sides are equal by [Ne11, Lemma 6.7] if we choose W′ small enough. Here we use that the invariant subspace U (gC)η generated by η under the derived representation is equianalytic ([Ne11, Prop. 6.6]). Step 4: For x1, x2 ∈ g ∩ W′ we now obtain with (17) and (18) U η(x1 ∗ ix2) = U (exp x1)ei∂U (x2)η. (19) Shrinking W′ if necessary, we may further assume that the map (W′ ∩ g) × (W′ ∩ g) → W, (x, y) 7→ x ∗ iy is a diffeomorphism onto an open subset W′′ ⊆ W. As ker(U ) is discrete, we may further assume that U ◦ exp is injective on W′ ∩ g. Step 5: Now let fW :=T0≤y≤π e−yi ad h(W′ ∩ W′′) and observe that, by the compactness of [0, π], this is an open convex 0-neighborhood in gC. For x ∈ fW ∩ g with exp x ∈ SV , we then find an ε > 0 such that ez ad hx ∈ W′ ∩ W for z ∈ Ω′′ := {w = a + ib ∈ C : a < ε, −ε < b < 1 + ε}. Then Ω′′ → H, z 7→ U η(ez ad hx) is a holomorphic function, and so is the continuous function Sπ → H, z 7→ αz(U (exp x))η on the open strip Sπ (Theorem 2.3). Since both functions coincide on the interval (−ε, ε) ⊆ R, we obtain by analytic continuation αti(U (exp x))η = U η(eti ad hx) for 0 ≤ t ≤ π. (20) As eit ad hx ∈ W′′ for 0 ≤ t ≤ π, we can write this element uniquely as xt∗iyt with xt, yt ∈ g∩W′. We now obtain with (19) kU η(xt ∗ iyt)k = kU (exp xt)ei∂U (yt)ηk = kei∂U (yt)ηk 20 and kU η(xt ∗ iyt)k = kU η(eit ad hx)k = kαit(U (exp x))ηk ≤ kηk because kαit(U (exp x))k ≤ kU (exp x)k = 1 by Theorem 2.3. Comparing both terms, we see that kei∂U (yt)ηk ≤ kηk for η ∈ D. As the operator ei∂U (yt) is selfadjoint, it is in particular closed. The above estimate shows that the closure of the restriction ei∂U (yt)D is a bounded operator on D = H. We conclude that kei∂U (yt)k ≤ 1, and hence that i∂U (yt) ≤ 0. This implies that yt ∈ CU for 0 ≤ t ≤ π. This in turn shows that exp(eti ad hx) = exp(xt ∗ iyt) = exp(xt) exp(iyt) ∈ SU and therefore βit(exp x) ∈ SU exists for t ∈ [0, π]. To see that exp x ∈ SU that inv, it remains to show βπi(exp x) = τG(exp x) = exp(τ (x)). (21) From Theorem 2.3 we recall that απi(U (exp x)) = JV U (exp x)JV = U (exp τ (x)). We thus obtain for t = π and η ∈ D: kηk = kαπi(U (exp x))ηk = kei∂U (yπ )ηk, and since i∂U (yπ) ≤ 0, this leads to ∂U (yπ)η = 0. As ker(U ) is discrete, it follows that yπ = 0, so that eπi ad hx = xπ ∈ g. Now (20) yields U (exp τ (x))η = απi(U (exp x))η = U η(eπi ad hx) = U (exp(eπi ad hx))η for η ∈ D, which in turn leads to U (exp(eπi ad hx)) = U (exp τ (x)). As U ◦ exp is injective on W′ ∩ g, we obtain βπi(exp x) = exp(eπi ad hx) = τG(exp x), and this finally proves that exp x ∈ SU inv for x ∈ fW ∩ g with exp x ∈ SV . The following corollary is a converse to Theorem 3.11 on the level of infinitesimal genera- tors. It follows immediately from the Germ Theorem (Theorem 4.1), Lemma 3.9(iv), and the observation that subsemigroups with identical germs have identical Lie wedges. Corollary 4.2. If ker(U ) is discrete, then L(SV ) = (g + iCU )inv := {x ∈ g + iCU : e[0,π]i ad hx ⊆ (g + iCU ); eπi ad hx = τ (x)}. Before we can prove the Structure Theorem, we need one more ingredient. We recall that a standard pair (U, V ) consists of a standard subspace V ⊆ H and a unitary one-parameter group (Ut)t∈R satisfying UtV ⊆ V for t ≥ 0. Proposition 4.3. Let (G, εG) be a finite dimensional graded Lie group and (U, H) be an antiunitary representation of G. Suppose that (V, U j ), j = 1, 2, are standard pairs for which there exists a graded homomorphism γ : R× → G and x1, x2 ∈ g such that JV = U (γ(−1)), ∆−it/2π V = U (γ(et)), and U j (t) = U (exp txj), t ∈ R, j = 1, 2. Then the unitary one-parameter groups U 1 and U 2 commute. In Subsection 5.2 we describe an example showing that, without assuming that they come from a finite dimensional Lie group G, the two one-parameter groups U 1 and U 2 need not commute. 21 Proof. The positive cone CU ⊆ g of the representation U is a closed convex Ad(G)-invariant cone. As we may w.l.o.g. assume that U is injective, the cone CU is pointed. Writing ∆−it/2π = U (exp th) and U j V t = U (exp txj) with h, x1, x2 ∈ g, we have [h, xj] = xj for j = 1, 2 and x1, x2 ∈ CU . If gλ(h) = ker(ad h − λ1) is the λ-eigenspace of ad h in g, then [gλ(h), gµ(h)] ⊆ gλ+µ(h), so that g+ := Pλ>0 gλ(h) is a nilpotent Lie algebra. Therefore n := (CU ∩ g+) − (CU ∩ g+) is a nilpotent Lie algebra generated by the pointed invariant cone CU ∩ g+. By [Ne00, Ex. VII.3.21], n is abelian. Finally xj ∈ CU ∩ g1(h) ⊆ n implies that [x1, x2] = 0. The following theorem not only provides an explicit description of the Lie wedge L(SV ), we also show that L(SV ) spans a 3-graded Lie subalgebra gred of g. Theorem 4.4. (Structure Theorem for L(SV )) If ker(U ) is discrete, then L(SV ) = C− ⊕ h0(h) ⊕ C+, where C± = ±CU ∩ q±1(h). (22) If q± := C± − C± are the linear subspaces generated by C±, then L(SV ) spans the 3-graded Lie subalgebra gred := q− ⊕ h0(h) ⊕ q+. Proof. From Corollary 4.2 we know that (g + iCU )inv = L(SV ). Further, Corollary 3.4 implies that (g + iCU )inv = (CU ∩ q1(h)) ⊕ h0(h) ⊕ (−CU ∩ q−1(h)). This proves (22). It follows in particular that q± = C± − C± ⊆ g±1(h). Proposition 4.3 shows that the two subspaces q± of g are abelian. Further, q± ⊆ g±1(h) implies that [q−, q+] ⊆ h ∩ g0(h) = h0(h), and from Corollary 2.5 we know that h0(h) = L(SV ) ∩ − L(SV ). As the cone L(SV ) is a Lie wedge, the operators ead x, x ∈ L(SV ) ∩ h on q preserve the cone L(SV ) ∩ q. This shows that [[q+, q−], q±] ⊆ q±, which implies that gred is a Lie subalgebra of g. It is clearly 3-graded by ad h, and the restriction of τ to gred coincides with the restriction of eπi ad h. Corollary 4.5. (The wedge L(SV ) in the 3-graded case) Suppose that U has discrete kernel, and that g = g−1(h) ⊕ g0(h) ⊕ g1(h) so that h = g0(h) and q = g1(h) ⊕ g−1(h). Then with τ = eπi ad h, L(SV ) = C− ⊕ g0(h) ⊕ C+, where C± = g±1(h) ∩ ±CU . Independence of SV from JV Proposition 4.6. (Independence of SV from JV ) Let (U j , H)j=1,2 be antiunitary representa- tions of the graded Lie group (G, εG) which coincide on G+. Then, for every graded homomor- phism γ : R× → G, and the corresponding standard subspaces V 1 γ and V 2 γ , we have SV 1 γ = SV 2 γ . Proof. By [N ´O17, Thm. 2.11] the antiunitary representations U 1 and U 2 are equivalent because their restrictions to G+ coincide. Hence there exists a unitary operator Φ ∈ U(H) with Φ ◦ U 1(g) = U 2(g) ◦ Φ for all g ∈ G. This implies in particular that Φ(V 1 γ , and since Φ commutes with U 1(G+) = U 2(G+), it follows that SV 1 γ ) = V 2 = SV 2 γ . γ 22 4.2 The unit group GV In the following we denote the centralizer of x ∈ g in G by CG(x) := {g ∈ G : Ad(g)x = x}. Lemma 4.7. Suppose that ker(U ) is discrete. The groups GV = SV ∩ CG+ (h) ⊇ CG+ (h)τG have the same Lie algebra gV = L(SV ) ∩ g0(h) = h0(h). They coincide if U is injective. Proof. If g ∈ G+ satisfies Ad(g)h = h, i.e., g ∈ CG+ (h), then the unitary operator U (g) com- mutes with ∆V . Therefore Corollary 2.4 implies that U (g)V ⊆ V is equivalent to U (g)V = V . If g ∈ CG+ (h)τG , then U (g) commutes with JV and ∆V , so that U (g)V = V (Lemma 2.2). This shows that CG+ (h)τG ⊆ GV = SV ∩ CG+ (h). If, in addition, U is injective, then U (g) ∈ GV implies that U (g) commutes with U (τG) = JV , and therefore τG(g) = g. For the Lie algebras of these groups, we obtain gV = L(SV ) ∩ − L(SV ) = L(SV ) ∩ g0(h) ⊇ h0(h). Since the kernel of U is discrete and the derived representation is injective, the fact that every x ∈ gV generates a unitary one-parameter group commuting with JV = U (τG) (Lemma 2.2) implies that τ (x) = x, i.e., x ∈ h. We conclude that gV = h0(h). This proves the assertion on the Lie algebras. is Z(G+) ∼= Z ∼= π1(PSL2(R)). Here the fundamental group of PSL2(R) it is generated by the loop obtained from the inclusion PSO2(R) ֒→ PSL2(R). Let τG ∈ Aut(G+) be the involution Example 4.8. (Inequality in Lemma 4.7) We consider the group G+ = fSL2(R) whose center d(cid:19), and observe that it induces the map given on the Lie algebra level by τ(cid:18)a c b d(cid:19) =(cid:18) a −b −c τG(z) = z−1 on Z(G+). Now consider an antiunitary representation (U, H) of PGL2(R), so that the corresponding representation of G := G+ ⋊ {id, τG} has kernel Z(G+). We therefore have Z(G+) = ker(U ) ⊆ GV for every V ∈ Stand(H). On the other hand, Z(G+) ⊆ CG+ (h), but Z(G+) it is not pointwise fixed by τG. We therefore have a proper inclusion CG+ (h)τG ֒→ GV in Lemma 4.7. Remark 4.9. (Degenerate cases) (a) If q = {0} and G+ is connected, then τG = idG, so that Lemma 4.7 and Corollary 2.5 imply that SV = GV = CG+ (h). (b) If L(SV ) ∩ q = {0}, then the Structure Theorem 4.4 implies that L(SV ) = h0(h) is a Lie subalgebra of h. Now [HHL89, Thm. IV.2.11] implies that the group GV is "isolated" in SV , i.e., there exists an e-neighborhood U ⊆ G with U ∩ SV ⊆ GV . Here we may w.l.o.g. assume that U = UGV is a tubular neighborhood of GV . 4.3 Some examples Example 4.10. (a) (The affine group) For the graded group G := Aff(R) = R ⋊ R× and representations of the form U (b, a) := eibP U V (a), V ∈ Stand(H), we can use Theorem 4.4 to calculate the semigroup SV . Since SV contains all positive dilations (0, a), a > 0, this closed subsemigroup of R ⋊ R×+ is of the form SV = C ⋊ R×+, where C = (R × {1}) ∩ SV is a closed additive subsemigroup of R invariant under multiplication with positive scalars. This leaves only the possibilities C = {0}, [0, ∞) or (−∞, 0]. Comparing with the Structure Theorem 4.4, we obtain SV = C ⋊ R×, where C := {x ∈ R : xP ≥ 0}. (23) This case can also be derived from the standard subspace version of the Borchers -- Wiesbrock Theorem ([Lo08, §3.2] and [N ´O17, Thms. 3.13, 3.15]). (b) (Higher dimensional dilation groups) More generally, we consider a group of the form G = E ⋊α R×, where the homomorphism α : R× → GL(E) satisfies α(r) = r1 for r > 0. Then 23 τE := α(−1) is an involution and we write E = E+ ⊕ E− for the corresponding eigenspace decomposition. Let (U, H) be an antiunitary representation of G, and consider the standard subspace V ∈ Stand(H) with U V (r) = U (0, r) for r ∈ R×. Then we also have SV = C ⋊ R×+, where C ⊆ E is a closed subsemigroup containing 0 which is invariant under multiplication with positive scalars, hence a closed convex cone. As h0(h) = {0} × R and g1(h) = E, our Structure Theorem implies that C = CU ∩ E−. On the other hand, [Lo08, Thm. 3.15] implies that E ∩ CU ⊆ E−, so that C = CU ⊆ E−. Note that we cannot apply (a) directly to the one-dimensional subspaces of E because we did not assume that α(−1) = − idE. Example 4.11. (More general R×-actions) We consider a group of the form G = E ⋊α R×, so that τE := α(−1) is an involutive automorphism of E. Accordingly, we write E = E+ ⊕ E− with E± = ker(τE ∓ 1) for the τE-eigenspace decomposition. We then have g = E ⋊ Rh with q = E− and h = E+ ⋊ Rh. As SV contains {0} ⋊ R×, we have SV = (SV ∩ E) ⋊α R×+, (24) where SV ∩E is a closed subsemigroup of E, invariant under α(R×+). We know from Theorem 4.4 that L(SV ) = (E−1 (h) ∩ CU ) ⊕ (−E− −1(h) ∩ CU ) ⊕ (E+ 0 (h) ⊕ Rh), where −1(h) ∩ CU ) is a pointed convex cone determined by the positive cone CU of U , and L(SV ) ∩ E− = (E−1 (h) ∩ CU ) ⊕ (−E− L(SV ) ∩ − L(SV ) = E+ 0 (h) ⊕ Rh is a Lie subalgebra. Here we can even use Subsection 3.1 to determine the subsemigroup SU inv of SV . From SU inv ∩ E) ⋊ R×+ and inv = (SU SU inv ∩ E = (E + i(CU ∩ E))inv = L(SV ) ∩ E, we obtain SU inv = (L(SV ) ∩ E) ⋊ R×+, (25) so that SU inv is the maximal infinitesimal generated subsemigroup of SV . Presently we do not know if we always have SU inv = SV for this class of groups, but this is work in progress ([Ne19]). Example 4.12. Suppose that g is a simple real Lie algebra, that G = G+ ⋊ {id, τG}, where the group G+ is connected, and that (U, H) is an antiunitary representation with non-zero positive cone CU and discrete kernel. This already implies that g is quite special, it has to be a hermitian Lie algebra (see [Ne00] for details and a classification). As J := U (τG) is antiunitary, −τ (CU ) = CU by Remark 3.7. We pick h ∈ h = gτ and consider the corresponding semigroup SV . If q = {0}, then τ = idg, so that U (G+) ⊆ U(H)J implies that SV is a group, namely the centralizer of ∆V in G+ (Remark 4.9(a)). We may therefore exclude this case and assume that τ 6= idg. The Structure Theorem (Theorem 4.4) shows that L(SV ) = C− ⊕ h0(h) ⊕ C+, where C± = ±CU ∩ q±1(h). In general, this cone may be rather small, but we know from Theorem 4.4 that it spans a 3-graded Lie algebra gred. If g = gred, then g itself is 3-graded, hence a hermitian Lie algebra of 24 tube type, i.e., the conformal Lie algebra of a euclidean Jordan algebra (see [Ne18, §3] for more details). In this case and for the centerless group G with Lie algebra g, we have determined the semigroup SV in [Ne18, Thms. 3.8, 3.13]: It coincides with the product set SV = exp(C+)(GτG + ) exp(C−). 5 Perspectives 5.1 Relations to von Neumann algebras We already mentioned in the introduction that the interest in the semigroups SV of endomor- phisms of standard subspaces stems to some extent from their correspondence to endomor- phisms of von Neumann algebras in the context of the theory of local observables ([Ha96]). We now provide some more details on these applications. We recall the notion of a Haag -- Kastler net of C∗-subalgebras A(O) of a C∗-algebra A, asso- ciated to regions O in d-dimensional Minkowski space R1,d−1. The algebra A(O) is interpreted as observables that can be measured in the "laboratory" O. Accordingly, one requires isotony, i.e., that O1 ⊆ O2 implies A(O1) ⊆ A(O2) and that the A(O) generate A. Causality enters by the locality assumption that A(O1) and A(O2) commute if O1 and O2 are space-like separated, i.e., cannot correspond with each other. Finally one assumes an action σ : P (d)↑+ → Aut(A) of the connected Poincar´e group such that σg(A(O)) = A(gO). Every Poincar´e invariant state ω of the algebra A now leads by the GNS construction to a covariant representation (πω, Hω, Ω) of A, and hence to a net M(O) := πω(A(O))′′ of von Neumann algebras on Hω. Whenever Ω is cyclic and separating for M(O), we obtain modular objects (∆O, JO). This connection be- tween the Araki -- Haag -- Kastler theory of local observables and modular theory leads naturally to antiunitary group representations (cf. [N ´O17, §5] and the introduction). Let, more generally, (G, εG) be a graded group of spacetime symmetries, where εG(g) = 1 means that g preserves time orientation and εG(g) = −1 that it reverses time orientation; a typical example is the Poincar´e group P (d) with P (d)+ = P ↑(d). Then covariant representa- tions of Haag -- Kastler nets lead to families M(O) of von Neumann algebras and antiunitary representations U : G → AU(H) satisfying U (g)M(O)U (g)−1 = M(gO). If the vacuum vector Ω ∈ H is fixed by U (G) and Ω is cyclic and separating for the von Neumann algebra M(O), and U (G) contains the corresponding modular conjugation J and the one-parameter group (∆it O)t∈R, then we are in the situation mentioned in the introduction, and we obtain information on the subsemigroup SVO ⊇ SO := {g ∈ G+ : U (g)M(O)U (g)−1 ⊆ M(O)}. Theorem 4.4 implies that the Lie wedge L(SV ) spans a 3-graded Lie subalgebra gred such that the corresponding 3-graded subgroup Gred ⊆ G has the property that SV ∩Gred has interior points and that the modular conjugation and the modular group also come from U (Gred). Example 5.1. 7 (a) It is important to observe that, in the situation described in the intro- duction, where Ω is a cyclic separating unit vector for the von Neumann algebra M and V = {M Ω : M = M∗ ∈ M}, the inclusion may be proper. SM = {g ∈ G : gMg−1 ⊆ M} ⊆ SV = {g ∈ G : gV ⊆ V } 7We thank Yoh Tanimoto for the discussion that led to this example. 25 To see an example, we consider the Hilbert space H := B2(Cn) of Hilbert -- Schmidt operators on Cn with the scalar product hA, Bi := tr(A∗B). By matrix multiplications from the left, we obtain a von Neumann subalgebra M ⊆ B(H), isomorphic to Mn(C), and its commutant M′ consists of right multiplications. The unit vector Ω := 1√n 1n is cyclic and separating, and the corresponding standard subspaces for M and M′ coincide with the space VM = VM′ = Hermn(C) of hermitian matrices. Now θ(A) := A⊤ defines a unitary operator on H preserving VM = VM′ and satisfying θMθ−1 = M′. For G = U(H), we therefore obtain SV 6= SM. (b) In the situation above, when M is given, the G-orbit of M in the space of von Neumann subalgebras of B(H) can be identified with the homogeneous space G/GM, and similarly, G/GV ֒→ Stand(H), gGV 7→ gV is an embedding. The discrepancy between both spaces comes from the potential non-triviality of the action of the stabilizer group GV on the von Neumann algebra M. Related questions have been analyzed by Y. Tanimoto in [Ta10]. He refines the picture by considering the closed convex cone which leads to the inclusions V + M = {M Ω : 0 ≤ M = M∗ ∈ M} ⊆ VM, SM ֒→ SV + M = {g ∈ G : gV + M ⊆ V + M } ⊆ SVM . Here the semigroup SV + Thm. 2.10]. In this context it is also interesting to note that the map appears to be much closer to SM than SV ; see in particular [Ta10, M V + M → M+ ∗ , ξ 7→ ωξ, ωξ(M ) = hξ, M ξi is a homeomorphism by [Ko80, Thm. 1.2]. Accordingly, every element g ∈ SV + continuous map on M+ ∗ . Example 5.2. In many situations arising in QFT, the group G is the Poincar´e group P (d) ∼= R1,d−1 ⋊ O1,d−1(R) acting by affine isometries on d-dimensional Minkowski space R1,d−1. We define a grading on P (d) by time reversal, i.e., εG(v, g) = ε(g) and g(V+) = ε(g)V+ for the upper open light cone V+ := {(x0, x) ∈ R1,d−1 : x0 > 0, x2 induces a M 0 > x2}. The generator h ∈ so1,d−1(R) of the Lorentz boost on the (x0, x1)-plane h(x0, x1, . . . , xd−1) = (x1, x0, 0, . . . , 0). It satisfies e2πi ad h = 1, and τ := eπi ad h defines an involution on the Poincar´e -- Lie algebra p(d), acting on R1,d−1 by τ (x0, x1, . . . , xd−1) = (−x0, −x1, x2, . . . , xd−1) on R1,d−1. For any positive energy representation of P (d) with discrete kernel, we then have CU = V+, because this is, up to sign, the only non-zero pointed invariant cone in the Lie algebra p(d) (for d > 2). Therefore the Lie wedge of the corresponding semigroup SV associated to the standard subspace determined by the triple (U, τ, h) is given by L(SV ) = h0(h) ⊕ (q1(h) ∩ CU ) ⊕ (q−1(h) ∩ −CU ) (Theorem 4.4). Here h0(h) = g0(h) is the centralizer of the Lorentz boost: g0(h) = ({(0, 0)} × Rd−2) ⋊ (so1,1(R) ⊕ sod−2(R)) ∼= (Rd−2 ⋊ sod−2(R)) ⊕ Rh, 26 and, for qj := qj(h): q1 ∩CU = R(e0 +e1)∩V+ = R+(e1 +e0) and q−1 ∩(−CU ) = R(e0 −e1)∩−V+ = R+(e1 −e0). Therefore L(SV ) coincides with the Lie wedge of the semigroup SWR := {g ∈ P (d)+ : gWR ⊆ WR}, where WR := {x ∈ R1,d−1 : x1 > x0} is the open right wedge (see also [N ´O17, Lemma 4.12]). The starting point for the development that led to fruitful applications of modular the- ory in QFT was the Bisognano -- Wichmann Theorem, asserting that the modular automor- phisms αt(M ) = ∆−it/2πM ∆it/2π associated to the algebra M(WR) of observables corre- sponding to the right wedge WR in Minkowski space are implemented by the unitary ac- tion of a one-parameter group of Lorentz boosts preserving WR. This geometric implemen- tation of modular automorphisms in terms of Poincar´e transformations was an important first step in a rich development based on the work of Borchers and Wiesbrock in the 1990s [Bo92, Bo95, Bo97, Wi92, Wi93, Wi93c]. They managed to distill the abstract essence from the Bisognano -- Wichmann Theorem which led to a better understanding of the basic configurations of von Neumann algebras in terms of half-sided modular inclusions and modular intersections. In his survey [Bo00], Borchers described how these concepts have revolutionized quantum field theory. Subsequent developments can be found in [Ar99, BGL02, Lo08, LW11, JM18, Mo17]. 5.2 Pairs of standard pairs For V ∈ Stand(H), one may expect that one-parameter groups U 1 and U 2, for which (V, U j ) form a standard pair, commute. By Proposition 4.3 this is true if they both come from an antiunitary representation of a finite dimensional Lie group. The following example shows that this is not true in general, not even if the two one-parameter groups are conjugate under the stabilizer group U(H)V . Example 5.3. On L2(R) we consider the selfadjoint operators (Qf )(x) = xf (x) and (P f )(x) = if′(x), satisfying the canonical commutation relations [P, Q] = i1. For both operators, the Schwartz space S(R) is a core. Actually it is the space of smooth vectors for the representation of the 3-dimensional Heisenberg group generated by the corresponding unitary one-parameter groups (eitQf )(x) = eitxf (x) and (eitP f )(x) = f (x − t). Since eix3 is a smooth function for which all derivatives grow at most polynomially, it defines a continuous linear operator on S(R) ([Tr67, Thm. 25.5]). Therefore the unitary operator T := eiQ3 maps S(R) continuous onto itself, and is a selfadjoint operator for which S(R) is a core. For f ∈ S(R), we obtain eP := T P T ∗ = eiQ3 P e−iQ3 (eP f )(x) = ieix3 d dx e−ix3 f (x) = i(−i3x2f (x) + f′(x)), so that eP = P + 3Q2. The two selfadjoint operators Q and eP are the infinitesimal generators of the irreducible antiunitary representation of Aff(R) = R ⋊ R×, given by U (b, et) = eibeP eitQ and (U (0, −1)f )(x) = f (−x). 27 Accordingly, the pair (∆, J) with ∆ = e−2πQ and J = U (0, −1) specifies a standard subspace V which combines with U 1 to an irreducible standard pair (V, U 1). The unitary operator T commutes with ∆ and with J because JQJ = −Q, so that T (V ) = V . Therefore the unitary one-parameter group U 2 also defines a standard pair (V, U 2). These two one-parameter groups do not commute because otherwise the selfadjoint operators P and P + 3Q2 would commute in the strong sense, hence in particular on their core S(R). t := eiQ3 t e−iQ3 U 1 eP = eite t := eiteP 5.3 Classification problems In the light of our results on the structure of the Lie wedges L(SV ), one would like to classify all situations, where these cones generate the Lie algebra g. As this requires g to be 3-graded by ad h with τ = eπi ad h, we have to consider Lie groups G+ with Lie algebra g and ad- diagonalizable elements h ∈ g with Spec(ad h) ⊆ {−1, 0, 1}. Then we have to study unitary representations of G+ extending to antiunitary representations of G = G+ ⋊ {idG, τG} in such a way that the two cones g±1(h) ∩ CU generate g±1(h). Then the ideal g1 := [q, q] ⊕ q generated by q is contained in CU − CU , so that the cone CU 1 for the restriction U 1 := U G1 is generating. Since we expect the semigroup SV to be adapted to any direct integral decomposition into irreducible representations, the main point is to understand the irreducible representations. For the normal subgroup G1 we thus have to study irreducible antiunitary representations U 1 for which the cone CU 1 is pointed and generating. Up to the extendability question from G1 + to G1 (cf. [N ´O17, Thm. 2.11(c),(d)]), we are then dealing with unitary highest weight modules, whose classification theory can be found in [Ne00, §X.4]. So the first steps in a classification should start with a faithful unitary highest weight representation (Uλ, H) of a one-connected + and a derivation D ∈ der(g1) satisfying D3 = D, such that CU ∩g±1(D) generates Lie group G1 g±1(D). Then g = g1 ⋊D R is a Lie algebra to which our results apply. 5.4 Global information on the semigroup SV We recall the context of Theorem 3.11 with the semigroup SU = G exp(iCU ) on which the analytic extension of the unitary representation (U, H) of the 1-connected Lie group G lives, and the subsemigroup SU inv ⊆ SV which has the same germ as SV (Theorem 4.1). Therefore the picture is very clear for the Lie wedges, but the global semigroup SV and SU inv may be more complicated and not even generated by their one-parameter subsemigroups. It would be interesting to understand the structure of the subsemigroup SU inv ⊆ G better, but this problem is quite intricate as well. However, below we shall see that it reduces to the situation where e2πi ad h = 1, which is a non-abelian analog of Lemma 3.2. Lemma 5.4. Consider the 1-connected complex Lie group GC with Lie algebra gC and the two connected Lie subgroups G := Gσ C , where σ and σc are the two antiholomorphic involutions of GC for which the derivative in e is complex conjugation with respect to g and gc, respectively. Then the following assertions hold: C and Gc := Gσc (i) τGC = σσc = σcσ is the holomorphic involution integrating the complex linear extension of τ to gC. (ii) For ζG := βπi/2 ∈ Aut(GC) and g ∈ G, we have ζG(g) ∈ Gc if and only if βπi(g) = τG(g), and this implies that ζ 4 G(g) = g, so that ζ−1 G (Gc) ∩ G ⊆ Fix(ζ 4 G). (iii) For elements of the form gc = h exp(x) ∈ Gc with h ∈ H c := (Gc)τG and x ∈ iq with Spec(ad x) ⊆ R, we have gc ∈ ζG(G) if and only if h ∈ ζG(G) and x ∈ ζ(g). If this is the case, then eπi ad hx = −x and βπi(h) = h. 28 Proof. (i) follows by inspection of the differentials. (ii) For the automorphisms βz ∈ Aut(GC) with differential ez ad h, we have σ ◦ βz = βz ◦ σ and σc ◦ βz = βz ◦ σc for z ∈ C. For z = πi/2, we obtain in particular σ ◦ ζG = ζ−1 G ◦ σ and σc ◦ ζG = ζ−1 G ◦ σc. Now let g ∈ G. The condition ζG(g) ∈ Gc is by σc(g) = σcσ(g) = τG(g) equivalent to ζG(g) !=σc(ζG(g)) = ζ−1 G (σc(g)) = ζ−1 G (τG(g)), hence to τG(g) = ζ 2 G(g) = βπi(g). If this condition is satisfied, then g = τG(τG(g)) = τG(ζ 2 G(g)) = ζ 2 G(τG(g)) = ζ 4 G(g). (iii) If h ∈ ζG(G) and x ∈ ζ(g), then we clearly have h exp x ∈ ζG(G). Suppose, conversely, that gc = h exp x with h ∈ H c and x ∈ iq with Spec(ad x) ⊆ R satisfies g ∈ ζG(G). As ζG commutes with τG and the group G is invariant under τG, the group ζG(G) is also τG-invariant. Hence gc ∈ ζG(G) implies g♯ ∈ ζ(G) and thus also g♯g = exp 2x ∈ ζG(G). The latter condition can be written as exp(2σζ−1(x)) = exp(2ζ−1(x)). Since ad x has real spectrum and GC is simply connected, we obtain with Lemma B.1 that σζ−1(x) = ζ−1(x), i.e., x ∈ ζ(g). This in turn implies that h ∈ ζG(G). From (ii) we now obtain ζ(x) = ζ 2(ζ−1(x)) = τ (ζ−1(x)) = ζ−1(τ (x)) = −ζ−1(x), hence G (h), and G (h)) = τG(ζ−1 G (τG(h)) = ζ−1 G (h)) = ζ−1 G(ζ−1 ζ 2(x) = −x. We likewise get ζG(h) = ζ 2 therefore ζ 2 G(h) = h. The following proposition reduces the determination of SU inv to the case where ζ 4 = 1, i.e., where ad h is diagonalizable with integral eigenvalues. By Lemma 5.4(ii) we may even assume that τ ζ 2 = idgC , so that ζ 2 = τ and therefore ζ(g) = gc. Proposition 5.5. Let qS : SU = ΓG(CU ) → Γ′G(CU ) ⊆ GC be the universal covering map of Γ′G(CU ), where GC and G are the 1-connected Lie groups with Lie algebra gC and g, respectively. Then qS(SU inv) is contained in the connected subgroup Fix(τGβπi) ⊆ Fix(ζ 4 G) of GC. Proof. To apply Lemma 5.4(ii), we simply have to observe that qS(G) = (GC)σ is called G in Lemma 5.4 and that βπi(qS(s)) = qS(βπi(s)) = qS(τG(s)) = τG(qS(s)) for s ∈ SU inv. The subgroup Fix(ζ 4 G) = (GC)ζ4 G is connected by Theorem B.2. Remark 5.6. (a) One can even go one step further than the preceding proposition by using the same trick as in the proof of Lemma 3.10: Let g ∈ SU inv ⊆ G and consider the corresponding analytic extension βg : Sπ → SU , z 7→ βz(g) of the orbit map of g. Then the argument in the proof of Lemma 3.10 shows that βg(Sπ) ⊆ gΓN (CU ) for n = CU − CU , so that we obtain in particular n ∋ We conclude that d dt(cid:12)(cid:12)(cid:12)t=0 g−1βg(t) = Ad(g)−1h − h. Ad(g)h ∈ h + n for g ∈ SU inv. 29 Therefore SU min is contained in a Lie subgroup B ⊆ G satisfying Ad(B)h − h ⊆ n and ηG(B) ⊆ Fix(τGβπi) ⊆ Fix(ζ 4 G). For the Lie algebra b of B this implies that [b, h] ⊆ n, so that the semisimplicity of ad h yields where the last equality follows from the equality of eπi ad h and τ on b. b ⊆ n + b0(h) ⊆ n + h0(h), inv) and the corresponding integral subgroup H0(h) ⊆ G is contained As h0 ⊆ L(SV ) = L(SU inv, we have in SU SU inv ∩ B ⊆ (SU inv ∩ N )H0(h), so that the main point is to understand the subsemigroup SU inv ∩ N. (b) The subgroup Γ := eR ad h ⋊ {1, τ } ⊆ Aut(g) is abelian and ad h is diagonalizable over R. Its Zariski closure is generated by the single element γ := ead hτ because γ2 = e2 ad h generates a Zariski dense subgroup of eR ad h. Hence Theorem B.2 implies that the subgroup (GC)Γ is connected. Its Lie algebra is gΓ C = hC,0(h) and contains h0(h) as a real form. Each automorphism βz ∈ Aut(GC) commutes with the holomorphic involution τ , and hence with the holomorphic antiinvolution g♯ = τ (g)−1. As G and SU are invariant under ♯ because CU is invariant under −τ , it follows that SU inv is ♯-invariant as well. Therefore g = h exp(x) ∈ inv) ⊆ (SU )τS implies that exp(2x) = g♯g ∈ ζG(Sinv). But it is not clear if this implies ζG(SU that exp(x) ∈ ζG(SU The following question is of a similar nature. Let x ∈ g and suppose that z := eyi ad hx ∈ gC satisfies exp 2z ∈ Γ′G(W ) ⊆ GC. Does this imply that exp(z) ∈ Γ′G(W )? It seems that such questions are hard to answer, as the following example shows. inv). Example 5.7. Consider the subsemigroup S := {g ∈ GLn(C) : kgk ≤ 1} = Un(C) exp(C), where C = {X ∈ Hermn(C) : X ≤ 0}. We consider matrices of the form s := kgk−1g for g =(cid:18)ε 0 1 ε(cid:19) , ε > 0. Then ksk ≤ 1, so that s ∈ SU . Moreover, ε−1g is unipotent with Then 0 ε−1 X := log g = (log ε)1 +(cid:18)0 0 (cid:19) . Y := log s = X − (log kgk)1 = log(εkgk−1)1 +(cid:18)0 0(cid:19) 2ε(cid:18)0 2 (Y + Y ∗) = log(εkgk−1)1 + 0 ≥ 1 1 0 1 1 ε−1 0 (cid:19) satisfies s = eY ∈ S. That etY ∈ S holds for all t ≥ 0 is equivalent to Y being dissipative, i.e., to 2ε ≤ 0. For ε → 0, we have kgk → 1, 2ε > − log(ε) if ε is sufficiently small. For any such ε, we then have Y 6∈ L(S), although (Remark 2.11(d)), which is equivalent to log(ε)−log(kgk)+ 1 and 1 eY ∈ S. 30 5.5 Extensions to infinite dimensions 5.5.1 Wick rotations for non-uniformly continuous actions It would be very interesting to understand to which extent Section 3 can be generalized beyond uniformly continuous actions on Banach spaces, including W ∗-dynamical systems. A natural setting would be that E is a Banach space, endowed with the following data: • A continuous involution τ ∈ GL(E); we write E = E+ ⊕ E−, E± := ker(τ ∓ 1) for the τ -eigenspace decomposition. • A subspace E∗ ⊆ E′ of the topological dual space which is norm-determining in the sense that kvk = sup{α(v) : α ∈ E∗, kαk ≤ 1}. • An R-action α : R → GL(E) commuting with τ such that, for every λ ∈ E∗ and for every v ∈ E, the functions t 7→ λ(αt(v)) are continuous. We say that α is E∗-weakly continuous. • A pointed closed convex cone W ⊆ Ec := E+ + iE− ⊆ EC, invariant under the complex linear extension of −τ and the one-parameter group (αt)t∈R. We say that, for v ∈ EC and z0 ∈ C, the element αv(z0) ∈ EC exists, if the orbit map αv(t) := αt(v) extends analytically to an E∗-weakly continuous map on a closed strip Sa,b containing z0. We expect a natural analog of Lemma 3.2 to hold. If x ∈ E is such that απi(x) exists and equals τ (x), then we should have an E∗-weakly convergent expansion x =Pn∈Z xn with αt(xn) = etnxn for t ∈ R and τ (xn) = (−1)nxn. This reduces the interesting situations to the case where ζ := απi/2 exists on a dense subspace and satisfies ζ 4 = 1. As we cannot expect the expansion of x to be finite, the arguments in the proof of Proposition 3.3 fail. Presently, we are not aware of examples, where the conclusion of Proposition 3.3 fails. As Olshanski semigroups and the extension of unitary representations also works to some extent for Banach -- Lie groups [MN12], one may expect that large portions of our results can be generalized to Banach -- Lie groups endowed with a suitably continuous action of R×, encoding the modular objects. 5.5.2 The subsemigroup SV ⊆ U(H) It would be nice to find suitable regularity properties of V that guarantee that the subsemigroup SV = {g ∈ U(H) : gV ⊆ V } in the full unitary group is large in some sense. Of course, one could assume that it has interior points, but that this never leads to proper subsemigroups is easy to see: Proposition 5.8. Let O ⊆ U(H) be an open subset. Then there exists an N ∈ N such that ON = {g1 · · · gN : gj ∈ O} = U(H). In particular, every subsemigroup S ⊆ U(H) with interior points coincides with U(H). Proof. Since the exponential function exp : u(H) = {X ∈ B(H) : X∗ = −X} → U(H) is surjective, the open subset exp−1(O) is non-empty. Using spectral calculus, we find an n ∈ N and an element X ∈ exp−1(O) such that Spec(X) ⊆ 2πi Z. Then g := eX ∈ O is of finite n order n. Hence 1 ∈ On. Let Br ⊆ u(H) be the open operator ball of radius r with center 0. Pick m ∈ N such that exp(B≤π/m) ⊆ On. Then (On)m ⊇ exp(B≤π) = U(H). 31 A Conjugation with unbounded operators The following proposition provides a direct path to the main ingredients of the Araki -- Zsid´o Theorem (Theorem 2.3), namely the implication (iii) ⇒ (iv). We need its corollary in the proof of Proposition 2.6. For the sake of completeness, we also include a proof of the Araki -- Zsid´o Theorem in this appendix. Proposition A.1. Let H = H∗ be a selfadjoint operator and Ut = eitH denote the corre- sponding unitary one-parameter group. Fix β > 0. If A ∈ B(H) is such that AD(e−βH) = AR(eβH) ⊆ D(e−βH) and the operator Aβ := e−βHAeβH on D(e−βH) extends to a bounded operator on H, then the following assertions hold: (i) The map αA : R → B(H), αA(t) := UtAU∗t extends to a bounded strongly continuous function on the closed strip Sβ = {z ∈ C : 0 ≤ Im z ≤ β} which is holomorphic on Sβ and satisfies αA(βi) = Aβ. (ii) kαA(z)k ≤ max(kAk, kAβk) for z ∈ Sβ (iii) αA(z + t) = UtαA(z)U∗t for z ∈ Sβ, t ∈ R. Proof. Let Hfin ⊆ H denote the dense subspace of vectors contained in spectral subspaces for H corresponding to bounded intervals. Let ξ, η ∈ Hfin, so that both are entire vectors of exponential growth for (Ut)t∈R. Then is an entire function with αξ,η(t) = hξ, αA(t)ηi and αξ,η(z) := he−izH ξ, Ae−izHηi αξ,η(t + βi) = he−βHU−tξ, AeβHU−tηi = hU−tξ, AβU−tηi for t ∈ R. (26) From [Ru86, Thm. 12.9] we now derive that kαξ,η(z)k ≤ max(kAk, kAβk) · kξkkηk for z ∈ Sπ (27) because this estimate holds on ∂Sβ = R ∪ (βi + R). The map Hfin × Hfin → O(Sπ), (ξ, η) 7→ αξ,η is sesquilinear, and continuous with respect to the sup-norm on O(Sπ) by (27), hence it extends to a continuous map on H × H because Hfin is dense in H. From the one-to-one isometric correspondence between bounded operators and continuous sesquilinear maps on H via αξ,η(z) = hξ, αA(z)ηi for ξ, η ∈ H, (28) we thus obtain a weakly continuous bounded map αA : Sβ → B(H) which is weakly holomorphic on Sβ. That the function αA : Sβ → B(H) is holomorphic follows from [Ne00, Cor. A.III.5]. It remains to show that it is strongly continuous on Sβ, which is done below. (ii) follows from (27). (iii) follows by analytic continuation because it holds for z ∈ R. (i) (continued) For η ∈ H, we consider the functions αA,η : Sβ → H, z 7→ αA(z)η. By (27), we have kαA,ηk∞ ≤ max(kAk, kAβk)kηk, so that the map H → ℓ∞(Sβ, H), η 7→ αA,η is linear and continuous. Hence it suffices to verify the continuity of αA,η for η ∈ Hfin. For z = x + iy ∈ Sβ, we have 0 ≤ y ≤ β, so that Ae−izHη ∈ AD(e−βH) ⊆ D(e−βH) ⊆ D(e−yH) = D(eizH) 32 (cf. [N ´O18, Lemma A.2.5] for the next to last inclusion). We therefore have αA,η(z) = eizH Ae−izHη for z ∈ Sβ. As the multiplication of operators is strongly continuous on bounded subsets of B(H), (iii) shows that it suffices to verify the continuity of αA,η on the line segment {yi : 0 ≤ y ≤ β}. For 0 ≤ y, y0 ≤ β, we have αA,η(yi) = e−yHAeyHη = e−yHA(eyHη − ey0H η) + e−yHAey0H η. (29) Let E denote the spectral measure of H, so that H = RR x dE(x). For ξ ∈ H we obtain the positive finite measure Eξ := hξ, E(·)ξi. Now, for ξ ∈ D(e−βH), the function Sβ → H, z 7→ eizHξ is continuous, because the kernel (z, w) 7→ heiwH ξ, eizHξi =ZR ei(z−w)t dEξ(t) is continuous on S2β by the Dominated Convergence Theorem ([N ´O18, Lemma A.2.5]). We conclude that the second summand in (29) is a continuous function of y. We further have keizH ξk2 =ZR e−2(Im z)t dEξ(t) ≤ max(kξk2, ke−βHξk2) by the convexity of the Laplace transform of the measure Eξ ([Ne00, Prop. V.4.3]). This implies that ke−yHξk ≤ max(kξk, ke−βHξk), (30) and thus ke−yH A(eyHη − ey0H η)k ≤ kAkkeyHη − ey0Hηk + ke−βHA(eyHη − ey0H η)k = kAkkeyHη − ey0Hηk + kAβkke−(β−y)Hη − e−(β−y0)Hηk. This estimate implies the continuity in y0 of the first summand in (29), and this completes the proof of (i). The estimate (30) has an interesting consequence: Corollary A.2. Let X be a topological space and f : X → D(e−βH) be a function. If the two maps f : X → H and e−βH ◦ f : X → H are continuous, then the composition eizH ◦ f : X → H is continuous for every z ∈ Sβ. Theorem A.3. (Characterization of V -real operators) For A ∈ B(H), the following are equiv- alent: (i) A ∈ AV , i.e., AV ⊆ V . (ii) A∗V ′ ⊆ V ′. (iii) JV A∗JV ∈ AV . (iv) JV AJV ∆1/2 (v) ∆1/2 V A∆−1/2 (vi) The map αA : R → B(H), αA(t) = ∆−it/2π is defined on D(∆−1/2 V ⊆ ∆1/2 V A. V ) and coincides there with JV AJV . V extends to a bounded strongly contin- uous function αA on the closed strip Sπ = {z ∈ C : 0 ≤ Im z ≤ π} which is holomorphic on Sπ and satisfies αA(πi) = JV AJV . V V A∆it/2π If these conditions are satisfied, then (a) kαA(z)k ≤ kAk for z ∈ Sπ 33 (b) αA(z + t) = ∆−it/2π (c) αA(z + πi) = JV αA(z)JV for z ∈ Sπ. (d) αA(t)V ⊆ V and αA(t + πi)V ′ ⊆ V ′ for all t ∈ R. for z ∈ Sπ, t ∈ R. αA(z)∆it/2π V V Proof. (i) ⇒ (ii): If AV ⊆ V and v ∈ V , w ∈ V ′, then ImhA∗w, vi = Imhw, Avi = 0 shows that A∗V ′ ⊆ V ′. (ii) ⇒ (i) follows by apply applying the implication "(i) ⇒ (ii)" to V ′ and A∗ and using that A = (A∗)∗ and V = (V ′)′. (ii) ⇔ (iii): From V ′ = JV V , it follows that A∗V ′ ⊆ V ′ is equivalent to A∗JV V ⊆ JV V , which is (iii). (i) ⇔ (iv): For the antilinear involution σV = JV ∆1/2 σV A, i.e., to V , condition (iv) is equivalent to AσV ⊆ AD(σV ) = A(V + iV ) ⊆ V + iV = D(σV ) and AσV = σV A on V. V ) = JV D(∆−1/2 This is equivalent to (i). (i) ⇔ (v): Conjugating with JV , we see that (v) is equivalent to σV Aσ−1 defined on D(∆1/2 to (i). (v) ⇒ (vi) follows from Proposition A.1 with H = − 1 (vi) ⇒ (v): If (vi) is satisfied, then (26) in the proof of Proposition A.1 yields for ξ, η ∈ Hfin the relation V = σV AσV being ) and that it equals A on this space. This in turn is equivalent 2β log(∆V ) and ∆1/2 V = e−βH. V hA∆−1/2 ξ, ∆1/2 V V ηi = hJV AJV ξ, ηi. (31) V , the equality (31) holds for ξ ∈ D(∆−1/2 and ∆1/2 ) V As the dense subspace Hfin is a core of ∆−1/2 and η ∈ D(∆1/2 V ). It follows that V ∆1/2 V A∆−1/2 V ξ = JV AJV ξ for ξ ∈ D(∆−1/2 V ), which is (v). Now we assume that the equivalent conditions (i)-(vi) are satisfied. From (ii) and (iii) in Proposition A.1, we get (a) and (b). For z ∈ R, we derive (c) from (vi) and (b), and for general z ∈ Sπ, it follows by analytic continuation. Finally, (d) follows from the invariance of V under ∆it V for t ∈ R and JV V = V ′. B Some facts on Lie groups Lemma B.1. Let G be a finite dimensional Lie group with Lie algebra g and x, y ∈ g with exp x = exp y. If exp is not singular in x, then [x, y] = 0 and exp(x − y) = e. If, in addition, G is simply connected and ad x and ad y have real spectrum, then x = y. Proof. The first assertion follows from [HHL89, V.6.7]. If ad x and ad y have real spectrum, then exp is regular in x, so that [x, y] = 0 and z := x − y satisfies exp(z) = e. The latter condition implies ead z = 1, so that ad z is semisimple with purely imaginary spectrum. On the other hand, [ad x, ad y] = ad[x, y] = 0 implies that Spec(ad z) ⊆ Spec(ad x) − Spec(ad y) ⊆ R (there exists a common generalized eigenspace decomposition). Combining both facts, we see that ad z = 0, i.e., z ∈ z(g). If G is simply connected, then exp z(g) is injective because Z(G)0 = exp(z(g)) is simply connected ([HN12, Thm. 11.1.21]). This implies z = 0. Theorem B.2. Let G be a 1-connected Lie group and let Γ ⊆ Aut(G) be a subgroup such that the Lie algebra g is a semisimple Γ-module. Then the following assertions hold: 34 (i) There exists a Γ-invariant Levi decomposition G ∼= R ⋊ S, so that the subgroup of Γ-fixed points is GΓ ∼= RΓ ⋊ SΓ. (ii) The group RΓ is connected. (iii) If the action of Γ on the Lie algebra s of S has a relatively compact image in Aut(s) ∼= Aut(S) which contains a dense cyclic subgroup, then SΓ is connected. 8 (iv) If ηS : S → SC is the universal complexification, then the Γ-action on S induces an action on SC. If the image of Γ in the algebraic group Aut(s) is generated by a single semisimple automorphism in the Zariski topology, then (SC)Γ is connected.9 Further ηS(SΓ) is an open subgroup in the group ηS(S)Γ = (SC)Γ,σ, where σ is the complex conjugation on SC with fixed point set η(S) = (SC)σ. Proof. (i) With [KN96, Prop. I.2] we find a Γ-invariant Levi decomposition g = r ⋊ s, so that we obtain a Levi decomposition G ∼= R ⋊ S, where R is solvable, S is semisimple and both are 1-connected and Γ-invariant. This proves (i). (ii) We argue by induction on the dimension of R. If R is abelian, then this 1-connected group is isomorphic to some Rn and Γ acts by linear maps. This implies that RΓ is a linear subspace, hence connected. If R is not abelian, then its commutator subgroup R′ = (R, R) has smaller dimension and its Lie algebra r′ = [r, r] is a proper Γ-invariant ideal of r. Let n ⊇ r′ be a maximal proper Γ-invariant ideal of r and let N E R be the corresponding normal integral subgroup. Since R is 1-connected, N is closed and 1-connected and the abelian quotient group Q := R/N is also 1-connected ([HN12, Thm. 11.1.21]). As N is 1-connected, our induction hypothesis implies that N Γ is connected. As N is Γ-invariant, Q inherits a natural Γ-action and since Q is abelian, the above argument shows that the fixed point group QΓ is connected. Clearly, q(RΓ) ⊆ QΓ, and we claim that we actually have equality. Two cases may occur. If QΓ = {e}, then RΓ = N Γ is connected. If QΓ 6= {e}, then it is a connected subgroup of positive dimension. As the action of Γ on r is semisimple, there exists a Γ-invariant linear subspace e ⊆ r complementing n. Then L(q) : e → q is a linear Γ-equivariant isomorphism, and since expQ : (q, +) → Q also is an isomorphism of Lie groups, it follows that expQ ◦ L(q) = q ◦ expR : eΓ → QΓ is a bijection. Although e may not be a Lie subalgebra of r, the preceding argument shows that RΓ/N Γ ∼= q(RΓ) = QΓ. As N Γ and QΓ are connected, we conclude that the group RΓ is connected as well. (iii) Replacing Γ, considered as a subgroup of Aut(s) ∼= Aut(S), by its compact closure does not change the subgroup of fixed points because the action of Aut(s) ∼= Aut(S) on S is smooth ([HN12, Thm. 11.3.5]). So we may w.l.o.g. assume that Γ is compact. It therefore is contained in a maximal compact subgroup C ⊆ Aut(s) because Aut(s) is an algebraic group, hence has only finitely many connected components ([HN12, §12.4]). Now C ∩ Aut(s)0 = C ∩ Ad(S) is maximal compact in the identity component, and therefore K := {s ∈ S : Ad(s) ∈ C} is maximal compactly embedded in S. We conclude that K is 1- connected and therefore that K ∼= Z(K) × (K, K), where Z(K) is a vector space and (K, K) is 1-connected compact, a maximal compact subgroup of S ([HN12, Thm. 12.1.18]). As K is invariant under the action of C on S, it is in particular invariant under Γ. Since Γ acts by automorphisms on K, it preserves its center Z(K) and its commutator subgroup (K, K). Let p ⊆ s be the orthogonal complement of the Lie algebra k of K with respect to the Killing form. 8For any element γ ∈ Γ for which γ Z is dense in Γ we then have the same group of fixed points. Note also that this assumption is satisfied if Γ is a product of a torus and a finite cyclic group. 9In Borel's book [Bor91] one finds in particular that centralizers of complex tori are connected ([Bor91, Cor. 11.12]). Since every torus contains a single element with the same centralizer ([Bor91, Prop. 8.18]) this follows from the present statement of (iv). 35 Then the polar map K × p → S, (k, x) 7→ k exp x is a Γ-equivariant diffeomorphism. We thus obtain SΓ = K Γ exp(pΓ) ∼= (K′)Γ × Z(K)Γ × pΓ. As Z(K) is a vector space, the group Z(K)Γ is a linear subspace, hence connected. The same is true for pΓ. To verify the connectedness of (K′)Γ, we recall that there exists a single element γ ∈ Γ for which the cyclic subgroup γ Z is dense in Γ, considered as a subgroup of Aut(s). As Aut(s) ∼= Aut(S) acts smoothly on S ([HN12, Thm. 11.3.5]), it follows that Γ and γ have the same fixed points. Now the 1-connectedness of the compact group K′ implies that (K′)Γ = (K′)γ is connected ([HN12, Thm. 12.4.26]). This shows that SΓ is connected. (iv) Let γ ∈ Γ ⊆ Aut(s) be a semisimple element for which Γ is contained in the Zariski closure of the cyclic subgroup γ Z. Since the action of the algebraic group Aut(s) on the algebraic group SC is algebraic, γ and Γ have the same fixed point group. As the group SC is 1-connected, the connectedness of Sγ C now follows from [OV90, Thm. 4.4.9, p. 214]. The remaining assertions are clear. C = SΓ From Theorem B.2(i)-(iii), we obtain in particular: Corollary B.3. Let G be a 1-connected Lie group and ϕ ∈ Aut(G) an automorphism of finite order. Then the subgroup Gϕ = {g ∈ G : ϕ(g) = g} of fixed points is connected. Acknowledgments We thank Yoh Tanimoto and Roberto Longo for an invitation to a research visit in Rome and for many discussions with them, Vincenzo Morinelli and Yoshimichi Ueda on standard subspaces and modular theory of von Neumann algebras. In particular, we thank Yoh Tanimoto for pointing out an inaccuracy in an earlier version of this paper. Last, but not least, we also thank Daniel Oeh and Jan Frahm for reading earlier versions of this manuscript. References [Ar99] [AZ05] [Bo92] [Bo95] [Bo97] [Bo00] [Bor91] [BR87] Araki, H., "Mathematical Theory of Quantum Fields," Int. Series of Monographs on Physics, Oxford Univ. Press, Oxford, 1999 Araki, H., and L. Zsid´o, Extension of the structure theorem of Borchers and its application to half-sided modular inclusions, Rev. Math. Phys. 17:5 (2005), 491 -- 543 Borchers, H.-J., The CPT-Theorem in two-dimensional theories of local observables, Comm. Math. Phys. 143 (1992), 315 -- 332 -- , On the use of modular groups in quantum field theory, Ann. Ins. H. Poincar´e 63:4 (1995), 331 -- 382 -- , On the lattice of subalgebras associated with the principle of half-sided modular inclusion, Lett. Math. Phys. 40:4 (1997), 371 -- 390 -- , On revolutionizing quantum field theory with Tomita's modular theory, J. Math. Phys. 41 (2000), 3604 -- 3673; extended version with complete proofs available under ftp://ftp.theorie.physik.uni-goettingen.de/pub/papers/lqp/99/04/esi773.ps.gz Borel, A., "Linear Algebraic Groups," Graduate Texts in Math. 126, Springer Verlag, New York, Berlin, 1991 Bratteli, O., and D. W. Robinson, "Operator Algebras and Quantum Statistical Mechanics I," 2nd ed., Texts and Monographs in Physics, Springer-Verlag, 1987 36 [BGL93] Brunetti, R., Guido, D., and R. Longo, Modular structure and duality in conformal quantum field theory, Comm. Math. Phys. 156 (1993), 210 -- 219 [BGL94] -- , Group cohomology, modular theory and space-time symmetries, Rev. Math. Phys. 7 (1994), 57 -- 71 [BGL02] -- , Modular localization and Wigner particles, Rev. Math. Phys. 14 (2002), 759 -- 785 [BDFS00] Buchholz, D., O. Dreyer, M. Florig, and S. J. Summers, Geometric modular action and spacetime symmetry groups, Rev. Math. Phys. 12:4 (2000), 475 -- 560 [BLS11] Buchholz, D., Lechner, G., and S. J. Summers, Warped convolutions, Rieffel defor- mations and the construction of quantum field theories, Comm. Math. Phys. 304:1 (2011), 95 -- 123 [BS93] [Ch68] [EN00] [Ha96] Buchholz, D., and S. J. Summers, An algebraic characterization of vacuum states in Minkowski space, Comm. Math. Phys. 155:3 (1993), 449 -- 458 Chernoff, P. E., Note on product formulas for operator semigroups, J. Funct. Anal- ysis 2 (1968), 238 -- 242 Engel, K.-J., and R. Nagel, "One-parameter Semigroups for Linear Evolution Equa- tions," Graduate Texts in Mathematics 194, Springer-Verlag, New York, 2000 Haag, R., "Local Quantum Physics. Fields, Particles, Algebras," Second edition, Texts and Monographs in Physics, Springer-Verlag, Berlin, 1996 [HHL89] Hilgert, J., K.H. Hofmann, and J.D. Lawson, "Lie Groups, Convex Cones, and Semigroups," Oxford University Press, 1989 [HN93] [HN12] [HP57] [JM18] [Ko80] [KN96] [Le15] [LL15] [Lo08] [LW11] [MN12] [Mo17] Hilgert, J., and K.-H. Neeb, "Lie Semigroups and Their Applications," Lecture Notes in Math. 1552, Springer Verlag, Berlin, Heidelberg, New York, 1993 -- , "Structure and Geometry of Lie Groups," Springer Monographs in Mathematics, Springer, New York, 2012 Hille, E., and R.S. Phillips, "Functional Analysis and Semigroups," Amer. Math. Soc., 1957 Jakel, C., and J. Mund, The Haag -- Kastler axioms for the P (ϕ)2 model on the de Sitter space, Ann. Henri Poincar´e 19:3 (2018), 959 -- 977 Kosaki, H., Positive cones associated with a von Neumann algebra, Math. Scand. 47 (1980), 295 -- 307 Krotz, B., and K.-H. Neeb, On hyperbolic cones and mixed symmetric spaces, J. Lie Theory 6:1 (1996), 69 -- 146 Integrable Models Lechner, G., Algebraic Constructive Quantum Field Theory: and Deformation Techniques, in "Advances in Algebraic Quantum Field The- ory," Eds: Brunetti, R. et al; 397 -- 449, Math. Phys. Stud., Springer, 2015; arXiv:math.ph:1503.03822 Lechner, G., and R. Longo, Localization in Nets of Standard Spaces, Comm. Math. Phys. 336:1 (2015), 27 -- 61 Longo, R., Real Hilbert subspaces, modular theory, SL(2, R) and CFT in "Von Neumann Algebras in Sibiu", 33-91, Theta Ser. Adv. Math. 10, Theta, Bucharest Longo, R., and E. Witten, An algebraic construction of boundary quantum field theory, Comm. Math. Phys. 303:1 (2011), 213 -- 232 Merigon, S., and K.-H. Neeb, Analytic extension techniques for unitary representa- tions of Banach-Lie groups, Int. Math. Res. Notices, 18 (2012), 4260-4300 Morinelli, V., The Bisognano -- Wichmann property on nets of standard subspaces, some sufficient conditions, Ann. Henri Poincar´e, Online First, 2017 37 [Ne00] [Ne11] [Ne18] [Ne19] [N ´O17] [N ´O18] Neeb, K.-H., "Holomorphy and Convexity in Lie Theory," Expositions in Mathe- matics 28, de Gruyter Verlag, Berlin, 2000 -- , On analytic vectors for unitary representations of infinite dimensional Lie groups, Annales de l'Inst. Fourier 61:5 (2011), 1441 -- 1476 -- , On the geometry of standard subspaces, "Representation Theory, Symmet- ric Spaces, and Integral Geometry," Eds: Jens Gerlach Christensen, Susanna Dann, and Matthew Dawson; Contemporary Mathematics 714 (2018), 199 -- 223; arXiv:math.OA: 1707.05506 -- , The order on flat spaces of standard subspaces, in preparation Neeb, K.-H., and G. ´Olafsson, Antiunitary representations and modular theory, in "50th Sophus Lie Seminar", Eds. K. Grabowska et al, J. Grabowski, A. Fialowski and K.-H. Neeb; Banach Center Publications 113 (2017), 291 -- 362; arXiv:math- RT:1704.01336 -- , "Reflection Positivity. A Representation Theoretic Perspective," Springer Briefs in Mathematical Physics 32, 2018 [Nel59] Nelson, E., Analytic vectors, Annals of Math. 70:3 (1959), 572 -- 615 [Oeh18] [Ol82] [OV90] [Ru86] [Ta10] [Tr67] [Wi92] [Wi93] Oeh, D., Analytic extensions of representations of -semigroups without polar de- composition, arXiv:math-RT:1812.10751 Olshanski, G.I., Invariant cones in Lie algebras, Lie semigroups, and the holomor- phic discrete series, Funct. Anal. Appl. 15 (1982), 275 -- 285 Onishchick, A. L. and E. B. Vinberg, "Lie Groups and Algebraic Groups," Springer, 1990 Rudin, W., "Real and Complex Analysis," McGraw Hill, 1986 Tanimoto, Y., Inclusions and positive cones of von Neumann algebras, J. Operator Theory 64 (2010), 435 -- 452 Treves, F., "Topological Vector Spaces, Distributions, and Kernels," Academic Press, New York, 1967 Wiesbrock, H.-W., A comment on a recent work of Borchers, Lett. Math. Phys. 25 (1992), 157 -- 159 -- , Half-sided modular inclusions of von Neumann algebras, Commun. Math. Phys. 157 (1993), 83 -- 92 [Wi93c] -- , Symmetries and half-sided modular inclusions of von Neumann algebras, Lett. Math. Phys. 28:2 (1993), 107 -- 114 38
1308.0429	1	1308	2013-08-02T08:02:22	Finite group actions on certain stably projectionless C*-algebras with the Rohlin property	[ "math.OA" ]	We introduce the Rohlin property and the approximate representability for finite group actions on stably projectionless C-algebras and study their basic properties. We give some examples of finite group actions on the Razak-Jacelon algebra and show some classification results of these actions. This study is based on the work of Izumi, Robert's classification theorem and Kirchberg's central sequence C-algebras.	math.OA	math	FINITE GROUP ACTIONS ON CERTAIN STABLY PROJECTIONLESS C∗-ALGEBRAS WITH THE ROHLIN PROPERTY NORIO NAWATA Abstract. We introduce the Rohlin property and the approximate repre- sentability for finite group actions on stably projectionless C∗-algebras and study their basic properties. We give some examples of finite group actions on the Razak-Jacelon algebra W2 and show some classification results of these actions. This study is based on the work of Izumi, Robert's classification theorem and Kirchberg's central sequence C∗-algebras. 1. Introduction A C∗-algebra A is said to be stably projectionless if A ⊗ K has no non-zero pro- jections where K is the C∗-algebra of compact operators on an infinite-dimensional separable Hilbert space. Note that every stably projectionless C∗-algebra does not have a unit. In this paper we study finite group actions on stably projectionless C∗-algebras. In the theory of operator algebras, the study of group actions on operator al- gebras is one of the most fundamental subjects and has a long history. We refer the reader to [14] and references given there for this subject. Connes considered the Rohlin property for automorphisms of von Neumann algebras and classified automorphisms of the AFD factor of type II1 in [5] and [6]. It is important to consider the Rohlin property for classifying group actions on operator algebras. Using a suitable formulation of the Rohlin property of automorphisms of unital C∗-algebras, Kishimoto classified a large class of automorphisms on UHF algebras and certain AT algebras in [23] and [24]. Nakamura showed that every aperiodic automorphism of a Kirchberg algebra has the Rohlin property and classified them in [32]. Izumi, Katsura and Matui showed the classification results of large classes of ZN -actions (see [15], [18], [28] and [29]). Izumi defined the Rohlin property for finite group actions on unital C∗-algebras (see also [21] and [11]) and classified a large class of finite group actions on unital C∗-algebras in [12] and [13]. In the above formulations of the Rohlin property, a partition of unity consisting of projections (in the central sequence C∗-algebra) is used. Hence it is not clear how to define the Rohlin property for group actions on stably projectionless C∗-algebras. Although Evans and Kishimoto defined the Rohlin property for automorphisms of non-unital C∗-algebras and classified trace scaling automorphisms of certain AF algebras in [10], their definition requires that C∗-algebras have non-zero projections. Sato defined the weak Rohlin property for automorphisms of unital projectionless C∗-algebras and classified a large class of automorphisms of the Jiang-Su algebra Z in [43]. Matui and Sato defined the weak Rohlin property for countable amenable group actions and classified large classes of actions of Z2 and the Klein bottle group 2010 Mathematics Subject Classification. Primary 46L55, Secondary 46L35; 46L40. Key words and phrases. Stably projectionless C∗-algebra; Kirchberg's central sequence C∗- algebra; Rohlin property. The author is a Research Fellow of the Japan Society for the Promotion of Science. 1 2 NORIO NAWATA on Z in [30] and [31]. We do not need a partition of unity consisting of projections in the definition of the weak Rohlin property. But the weak Rohlin property is too weak from the viewpoint of the classification of finite group actions. Moreover their arguments are based on analyses of the actions on UHF algebras with the usual Rohlin property. Therefore we need to consider a suitable formulation of the Rohlin property for actions on stably projectionless C∗-algebras. We shall define the Rohlin property for finite group actions on σ-unital C∗- algebras by using Kirchberg's central sequence C∗-algebras defined in [19]. Kirch- berg's central sequence algebra F (A) of a C∗-algebra A is defined as the quotient C∗-algebra of the central sequence C∗-algebra A∞ by the annihilator of A. Kirch- berg's central sequence C∗-algebras are useful for proving Z-stability results for non-unital C∗-algebras (see [33] and [45]). One of the purposes of this paper is showing that Kirchberg's central sequence C∗-algebras are also useful for the clas- sification of group actions on certain stably projectionless C∗-algebras. Let W2 be the Razak-Jacelon algebra studied in [16], which is a certain simple nuclear stably projectionless C∗-algebra having trivial K-groups and a unique tra- cial state and no unbounded traces. The Razak-Jacelon algebra W2 is regarded as a stably finite analogue of the Cuntz algebra O2. Moreover Robert showed that W2 ⊗ K is isomorphic to O2 ⋊α R for some one parameter automorphism group α in [37] (see also [26], [27] and [8]). Hence it is natural and interesting to consider whether W2 has the similar properties of O2. In this paper we study mainly finite group actions on W2. This is based on Izumi's study of finite group actions on O2 in [12]. This paper is organized as follows: In Section 2, we collect notations and some results. In Section 3, we define the Rohlin property for finite group actions on σ-unital C∗-algebras and study their basic properties. In particular, we show that any two actions of a finite group on W2 with the Rohlin property are conjugate (Corollary 3.7). We also show that there exist strong K-theoretical constraints for the Rohlin property as in the unital case. In Section 4, we introduce the approximate representability for finite group actions on σ-unital C∗-algebras, which is the "dual" notion of the Rohlin property. We show a classification result of certain approximately representable actions (Corollary 4.5). Using this classification result, we give interesting examples of finite group actions on W2 with the Rohlin property (Corollary 4.11). In Section 5, we classify certain "locally representable" actions of Z2 on W2 up to conjugacy (Theorem 5.2) and up to cocycle conjugacy (Theorem 5.5). We also construct a locally representable action α of Z2 on W2 such that K0(W2 ⋊α Z2) = Z[ 1 2 ] and K1(W2 ⋊α Z2) = 0. 2. Preliminaries In this section we shall collect notations and some results. We refer the reader to [3] and [34] for basic facts of operator algebras. 1 2.1. Notations. We say that a C∗-algebra A is σ-unital if A has a countable ap- proximate unit. In particular, if A is σ-unital, then there exists a positive element n }n∈N is an approximate unit. Such a positive element s is s ∈ A such that {s called strict positive in A. If A is separable, then A is σ-unital. We denote by A the unitization algebra of A. A multiplier algebra, denoted by M (A), of A is the largest unital C∗-algebra that contains A as an essential ideal. It is unique up to isomorphism over A. If α is an automorphism of A, then α extends uniquely to an automorphism ¯α of M (A). A homomorphism φ from A to B is said to be nondegenerate if φ(A)B is dense in B. A nondegenerate homomorphism from A to B can be uniquely extended to a homomorphism ¯φ from M (A) to M (B). Note that if A ⊂ B is an nondegenerate inclusion, then every approximate unit for A is an approximate unit for B. FINITE GROUP ACTIONS ON CERTAIN STABLY PROJECTIONLESS C∗-ALGEBRAS 3 For a unitary element u in M (A), define an automorphism Ad(u) of A by Ad(u)(a) = uau∗ for a ∈ A, and such an automorphism is called an inner au- tomorphism. Let Aut(A) denote the automorphism group of A, which is equipped with the topology of pointwise norm convergence. An automorphism α is said to be approximately inner if α is in the closure of the inner automorphism group. We say that two automorphisms α and β are approximately unitarily equivalent if α ◦ β−1 is approximately inner. For a locally compact group G, an action α of G on A is a continuous homomor- phism from G to Aut(A). We say that α is outer if αg is not an inner automorphism for any g ∈ G \ {ι} where ι is the identity of G. An α-cocycle is a continuous map u from G to the unitary group U (M (A)) of M (A) such that u(gh) = u(g)αg(u(h)) If u is an α-cocycle, we define the action αu of G on A by for any g, h ∈ G. αu g (a) = Ad(u(g)) ◦ αg(a) for any a ∈ A and g ∈ G. For two G-actions α on A and β on B, we say that α and β are conjugate, written (A, α) ∼= (B, β), if there exists an isomorphism θ from A onto B such that θ ◦ αg = βg ◦ θ for any g ∈ G. They are said to be cocycle conjugate if there exists an α-cocycle u such that αu is conjugate to β. Let T (A) be the set of densely defined lower semicontinuous traces on A and T1(A) the set of tracial states on A, and put T0(A) := {τ ∈ T (A) kτ k ≤ 1}. If A is unital, then T1(A) is a Choquet simplex (see [42, Theorem 3.1.18]). There exists a natural one to one correspondence between T0(A) and T1( A), and hence T0(A) is a Choquet simplex. Every tracial state τ on A extends uniquely to a tracial state ¯τ on M (A) (see, for example, [45, Proposition 2.9]). For τ ∈ T1(A), consider the Gelfand-Naimark-Segal (GNS) construction (πτ , Hτ , ξ). Then τ extends uniquely to a normal tracial state τ on πτ (A) . Note that if τ is an extremal tracial state, is a finite factor. Let α be an automorphism of A such that τ ◦ α = τ . then πτ (A) Then α extends uniquely to an automorphism α of πτ (A) . Moreover if α is an action of G on A such that τ ◦ αg = τ for any g ∈ G, then α extends uniquely to a von Neumann algebraic action α on πτ (A) . We say that an action α of G on a C∗-algebra A with a unique tracial state τ is strongly outer if αg is not inner in πτ (A) for any g ∈ G \ {ι}. ′′ ′′ ′′ ′′ ′′ We denote by K(H), K and Mn∞ for n ∈ N the C∗-algebra of compact oper- ators on a Hilbert space H, the C∗-algebra of compact operators on an infinite- dimensional separable Hilbert space and the uniformly hyperfinite (UHF) algebra of type n∞, respectively. Let Zn denote the cyclic group of order n. For a Zn-action α, α1 is denoted by α for simplicity. If G is a finite group, we denote by G the order of G. 2.2. Crossed products and fixed point algebras. For an action α of G on A, we denote by A ⋊α G and Aα the reduced crossed product C∗-algebra and the fixed point algebra, respectively. If G is a compact group, then Aα is isomorphic to a conner algebra of A ⋊α G [40]. Indeed, define a projection eα in M (A ⋊α G) by eα :=ZG λgdg where λg is the implementing unitary of αg in M (A ⋊α G) and dg stands for the normalized Haar measure. Then Aα is isomorphic to eα(A ⋊α G)eα. Note that we have M (Aα) = M (A) ¯α. When G is a locally compact abelian group, let α denote the dual action of α on A ⋊α G. The Takesaki-Takai duality theorem [44] shows that there exists an isomorphism Φ from A⋊αG⋊ α G onto A⊗K(ℓ2(G)) such that Φ◦ α = (α⊗Ad(ρ))◦Φ where ρ is the right regular representation of G. If G is a finite abelian group, then we see that ¯Φ(eα) = 1M(A) ⊗ e where e ∈ K(ℓ2(G)) is the projection onto the 4 NORIO NAWATA constant functions and may regard Φ(a) as a diagonal matrix   a αg1 (a) . . . αgG−1 (a)   where a ∈ A and G = {ι, g1, ..., gG−1}. Assume that τ is a tracial state on A such that τ ◦ αg = τ for any g ∈ G. Then , α, G). If we can construct a von Neumann algebraic crossed product W ∗(πτ (A) A is simple, then we may consider ′′ A ⋊α G ⊂ M (A ⋊α G) ⊂ W ∗(πτ (A) ′′ , α, G) and A ⋊α G is weakly dense in W ∗(πτ (A) tation of A. The following proposition is well-known (see, for example, [1]). , α, G) because πτ is a faithful represen- ′′ Proposition 2.1. Let A be a simple C∗-algebra with a unique tracial state τ , and let α be a strongly outer action of a discrete group G on A. Then A ⋊α G has a unique tracial state E ◦ τ where E is the canonical conditional expectation from A ⋊α G onto A. Hiroki Matui told us the following lemmas. Lemma 2.2. Let A ⊂ B be a nondegenerate inclusion of separable C∗-algebras. If T1(A) is compact, then T1(B) is compact. Proof. On the contrary, suppose that T1(B) were not compact. Then there exist a sequence {τn}n∈N in T1(B) and an element τ in T0(B) such that τn converges to τ (in the weak-∗ topology) and kτ k < 1. Since T1(A) is compact, the restriction of τ on A is a tracial state on A. This is a contradiction because A ⊂ B is a nondegenerate inclusion. (cid:3) Lemma 2.3. Let B be a C∗-algebra, and let β be an action of a finite group G on B. Assume that T1(B) is compact and there exists a unique tracial state τ on B such that τ ◦ βg = τ for any g ∈ G. Then B has at most G extremal tracial states. Proof. By the Krein-Milman theorem, T1(B) is the closed convex hull of its extreme points ex(T1(B)). Let τ1 be an extremal tracial state on B. Then τ1 ◦ βg is an extremal tracial state for any g ∈ G. Since τ is the unique tracial state on B such that τ ◦ βg = τ for any g ∈ G, we have τ = 1 contrary, suppose that ex(T1(B)) contained at least G + 1 points. Then there exists an extremal tracial τ2 such that τ2 6= τ1 ◦ βg for any g ∈ G, and we have τ = 1 GPg∈G τ1 ◦ βg. On the GPg∈G τ2 ◦ βg. This is a contradiction because T0(B) is a Choquet simplex. (cid:3) Proposition 2.4. Let A be a separable C∗-algebra with a unique tracial state τ , and let α be an action of a finite abelian group on A. Then A ⋊α G has at most G extremal tracial states. Proof. Lemma 2.2 shows that T1(A ⋊α G) is compact. Since A ⋊α G ⋊ α G is isomorphic to A ⊗ MG(C), A ⋊α G ⋊ α G has a unique tracial state. Hence we see that the fixed point of α in T1(A ⋊α G) is only one point. Therefore A ⋊α G has at most G extremal tracial states by Lemma 2.3. (cid:3) Proposition 2.5. Let A be a simple separable C∗-algebra with a unique tracial state τ , and let α be an action of Z2 on A. Then A ⋊α Z2 has exactly two extremal tracial states if and only if there exists a unitary element Uα in πτ (A) such that α = Ad(Uα) and U 2 ′′ α = 1. FINITE GROUP ACTIONS ON CERTAIN STABLY PROJECTIONLESS C∗-ALGEBRAS 5 ′′ ′′ α = 1. Then W ∗(πτ (A) , α, Z2) is isomorphic to πτ (A) such that α = Proof. Assume that there exists a unitary element Uα in πτ (A) Ad(Uα) and U 2 . Hence we see that W ∗(πτ (A) , α, Z2) has two extremal normal tracial states σ1 and σ2 such that σ1(m0 + m1λ) = τ (m0) + τ (m1Uα) and σ2(m0 + m1λ) = τ (m0) − τ (m1Uα) where λ is the implementing unitary of α and m0, m1 ∈ πτ (A) . Since A ⋊α Z2 is weakly dense in W ∗(πτ (A) is a factor, we see that the restrictions of σ1 and σ2 on A ⋊α Z2 are distinct extremal tracial states on A ⋊α Z2. By the proposition above, A ⋊α Z2 has exactly two extremal tracial states. , α, Z2) and πτ (A) ⊕ πτ (A) ′′ ′′ ′′ ′′ ′′ ′′ Conversely, assume that A ⋊α Z2 has exactly two extremal tracial states. Then such that α = Ad(V ) by Proposition is a factor, there exists a real number t such that V 2 = e2πit1, (cid:3) there exists a unitary element V in πτ (A) 2.1. Since πτ (A) and put Uα := e−πitV . Then Uα has the desired property. ′′ ′′ ′′ ′′ α = 1. such that α = Ad(Uα) and Remark 2.6. Choose a unitary element Uα in πτ (A) U 2 such that α = Ad(V ). Then there exists (i) Let V be a unitary element in πτ (A) a real number t such that V = e2πitUα because πτ (A) is a factor. Hence the value τ (V ) does not depend on the choice of such a unitary element V and is determined by α. We have τ (V ) ∈ [0, 1] and τ (V ) = 1 if and only if α = id. Hence if α is an outer action, then τ (V ) ∈ [0, 1). (ii) The proof above shows that two extremal tracial states ω1 and ω2 on A ⋊α Z2 are given by the restriction of σ1 and σ2. Therefore we have ′′ ¯ω1(x0 + x1λ) = ¯τ (x0) + τ (x1Uα) and ¯ω2(x0 + x1λ) = ¯τ (x0) − τ (x1Uα) where λ is the implementing unitary of α and x0, x1 ∈ M (A). Note that if W is a such that α = Ad(W ) and W 2 = 1, then W = Uα or unitary element in πτ (A) W = −Uα. ′′ We shall define a conjugacy invariant for outer actions of Z2 on a simple C∗- algebra with a unique tracial state. By Proposition 2.5 and Remark 2.6, we see that the following definition is well-defined. Definition 2.7. Let α be an outer action of Z2 on a simple C∗-algebra A with a unique tracial state τ . Define an invariant ǫ(α) ∈ [0, 1] of α by ǫ(α) = 1 if A ⋊α Z2 has a unique tracial state and ǫ(α) = τ (V ) if A ⋊α Z2 has exactly two extremal tracial states where V is a unitary element in πτ (A) such that α = Ad(V ). ′′ 2.3. Kirchberg's central sequence C∗-algebras. We shall recall some proper- ties of Kirchberg's central sequence C∗-algebras in [19]. See also [33, Section 5]. For a σ-unital C∗-algebra A, set c0(A) := {(an)n∈N ∈ ℓ∞(N, A) lim n→∞ kank = 0}, A∞ := ℓ∞(N, A)/c0(A). Let B be a C∗-subalgebra of A. We identify A and B with the C∗-subalgebras of A∞ consisting of equivalence classes of constant sequences. Put A∞ := A∞ ∩ A′, Ann(B, A∞) := {(an)n ∈ A∞ ∩ B′ (an)nb = 0 for any b ∈ B}. Then Ann(B, A∞) is an closed two-sided ideal of A∞ ∩ B′, and define F (A) := A∞/Ann(A, A∞). We call F (A) the central sequence C∗-algebra of A. A sequence (an)n is said to be central if limn→∞ kana − aank = 0 for all a ∈ A. A central sequence is a represen- tative of an element in A∞. Since A is σ-unital, A has a countable approximate unit {hn}n∈N. It is easy to see that [(hn)n] is a unit in F (A). If A is unital, then F (A) = A∞. Note that F (A) is isomorphic to M (A)∞ ∩ A′/Ann(A, M (A)∞) and 6 NORIO NAWATA F (A ⊗ K). If α is an automorphism of A, α induces natural automorphisms of A∞, A∞ and F (A). We denote them by the same symbol α for simplicity. Note that if α is an inner automorphism of A, then the induced automorphism of α on F (A) is the identity map. 2.4. Strict comparison and almost stable rank one. For positive elements a, b ∈ A, we say that a is Cuntz smaller than b, written a - b, if there exists a sequence {xn}n∈N of A such that kx∗ nbxn − ak → 0. Positive elements a and b are said to be Cuntz equivalent, written a ∼ b, if a - b and b - a. For τ ∈ T (A), n ) for h ∈ (A ⊗ K)+. Then dτ is a dimension put dτ (h) = limn→∞ τ ⊗ Tr(h function. In this paper we say that A has strict comparison if a, b ∈ (A ⊗ K)+ with dτ (a) < dτ (b) < ∞ for any τ ∈ T (A), then a - b. A C∗-algebra A is said to have almost stable rank one if for every σ-unital hereditary subalgebra B ⊆ A ⊗ K we 1 have B ⊆ GL(eB). If A is unital, then A has stable rank one if and only if A has almost stable rank one. We have the following proposition. See, for example, [33, Corollary 3.3 and Proposition 3.4]. Proposition 2.8. Let A be a simple exact σ-unital stably projectionless C∗- algebra, and let a and b be positive elements in A. Assume that A has strict comparison and almost stable rank one. If dτ (a) = dτ (b) < ∞ for any τ ∈ T (A), then aA is isomorphic to bA as right Hilbert A-modules. We shall consider the comparison theory for projections in multiplier algebras of certain stably projectionless C∗-algebras. Proposition 2.9. Let A be a simple exact σ-unital stably projectionless C∗- algebra, and let p and q be projections in M (A). Assume that A has strict com- parison and almost stable rank one. If ¯τ (p) = ¯τ (q) < ∞ for any τ ∈ T (A), then p is Murray-von Neumann equivalent to q in M (A). Proof. Consider right Hilbert A-modules pA and qA. Since pAp and qAq are σ- unital, there exist positive elements a and b in A such that aA = pA and bA = qA. Note that {a n }n∈N are approximate units for pAp and qAq, respectively. Hence we have ¯τ (p) = dτ (a) and ¯τ (q) = dτ (b) for any τ ∈ T (A). By Proposition 2.8, we see that pA is isomorphic to qA as right Hilbert A-modules. Therefore p is Murray-von Neumann equivalent to q in M (A). (cid:3) n }n∈N and {b 1 1 We denote by Z the Jiang-Su algebra constructed in [17]. The Jiang-Su algebra Z is a unital separable simple infinite-dimensional nuclear C∗-algebra whose K- theoretic invariant is isomorphic to that of complex numbers. A C∗-algebra A is said to be Z-stable if A is isomorphic to A ⊗ Z. If A is a simple exact Z-stable C∗- algebra with traces, then A has strict comparison (see [39, Theorem 4.5]). Robert showed that if A is a simple σ-unital Z-stable stably projectionless C∗-algebra, then A has almost stable rank one. Therefore we have the following corollary. Corollary 2.10. Let A be a simple exact σ-unital stably projectionless Z-stable C∗-algebra, and let p and q be projections in M (A). If ¯τ (p) = ¯τ (q) < ∞ for any τ ∈ T (A), then p is Murray-von Neumann equivalent to q in M (A). 2.5. Razak-Jacelon algebra and Robert's classification theorem. Let W2 be the Razak-Jacelon algebra studied in [16], which has trivial K-groups and a unique tracial state and no unbounded traces. The Razak-Jacelon algebra W2 is constructed as an inductive limit C∗-algebra of Razak's building block in [36], that is, f ∈ C([0, 1]) ⊗ Mm(C) f (0) = diag( c, .., c, 0n), f (1) = diag( A(n, m) =  k z}{ c ∈ Mn(C) k+1 c, .., c), z}{   FINITE GROUP ACTIONS ON CERTAIN STABLY PROJECTIONLESS C∗-ALGEBRAS 7 where n and m are natural numbers with nm and k := m n − 1. Let O2 denote the Cuntz algebra generated by 2 isometries S1 and S2. For every λ1, λ2 ∈ R there exists by universality a one-parameter automorphism group α of O2 given by αt(Sj) = eitλj Sj. Kishimoto and Kumjian showed that if λ1 and λ2 are all nonzero of the same sign and λ1 and λ2 generate R as a closed subgroup, then O2 ⋊α R is a simple stably projectionless C∗-algebra with unique (up to scalar multiple) densely defined lower semicontinuous trace in [26] and [27]. Moreover Robert [37] showed that W2 ⊗ K is isomorphic to O2 ⋊α R for some λ1 and λ2. (See also [8].) We say that A is a 1-dimensional non-commutative CW (NCCW) complex if A is a pullback C∗-algebra of the form A yπ1 C([0, 1]) ⊗ F π2−−−−→ E yρ δ0⊕δ1−−−−→ F ⊕ F where E and F are finite-dimensional C∗-algebras and δi is the evaluation map at i. Razak's building block A(n, m) is a 1-dimensional NCCW complex. Robert clas- sified inductive limit C∗-algebras of 1-dimensional NCCW complexes with trivial K1-groups (and C∗-algebras that are stably isomorphic to such inductive limits) in [37]. In particular, if A is a simple stably projectionless C∗-algebra, then the clas- sification invariant of A is the 3-tuple (K0(A), (T (A), T1(A)), rA) where rA is the pairing between K0(A) and T (A) (see [37, Proposition 6.2.3 and Corollary 6.2.4]). For a homomorphism φ from A to B, we denote by K0(φ) and T (φ) the induced ho- momorphism from K0(A) to K0(B) by φ and the induced map from T (B) to T (A) by φ, respectively. The following theorem is an immediate consequence of Robert's classification theorem (see [37, Theorem 1.0.1, Proposition 3.1.7, Proposition 6.2.3 and Corollary 6.2.4]). Theorem 2.11. (Robert) Let A be a simple stably projectionless C∗-algebra that is stably isomorphic to one inductive limit C∗-algebra of 1-dimensional NCCW complexes with trivial K1- groups. (i) If α and β are automorphisms of A, then α is approximately unitarily equivalent to β if and only if K0(α) = K0(β) and T (α) = T (β). (ii) Let ϕ be an automorphism of a countable abelian group K0(A) and γ an automorphism of a topological cone T (A) with γ(T1(A)) = T1(A). If ϕ and γ are compatible with the pairing rA, then there exists an automorphism α of A such that K0(α) = ϕ and T (α) = γ. Robert's classification theorem shows that if A is a simple approximately finite dimensional (AF) algebra with a unique tracial state and no unbounded traces, then A ⊗ W2 is isomorphic to W2. In general, Robert conjectured that if A and B are separable nuclear C∗-algebras, then T (A) is isomorphic to T (B) as non-cancellative topological cones if and only if A ⊗ W2 ⊗ K is isomorphic to B ⊗ W2 ⊗ K (see [41]). Note that every simple inductive limit C∗-algebra of 1-dimensional NCCW com- plexes has strict comparison and stable rank one. Moreover these C∗-algebras are Z-stable by [45, Corollary 9.2]. 3. The Rohlin property In this section we introduce the Rohlin property for finite group actions on σ- unital C∗-algebras and study some basic properties of finite group actions with the Rohlin property. 8 NORIO NAWATA Definition 3.1. An action α of a finite group G on a σ-unital C∗-algebra A is said to have the Rohlin property if there exists a partition of unity {eg}g∈G ⊂ F (A) consisting of projections satisfying αg(eh) = egh for any g, h ∈ G. A family of projections {eg}g∈G is called Rohlin projections of α. If A is unital, then the definition above coincides with the definition in [12]. Example 3.2. Let G be a finite group, and let µG be the action of G on MG∞ in [12, Example 3.2], that is, µG g = ∞On=1 Ad(λg) for any g ∈ G where λg is the left regular representation of G and we identify B(ℓ2(G)) with MG(C). Define an action νG of G on MG∞ ⊗ W2 ∼= W2 by νG := µG ⊗ id. Then νG has the Rohlin property. Indeed, let {hn}n∈N be an approximate unit for W2. Since µG has the Rohlin property (see [12]), there exist Rohlin projections {(pg,n)n}g∈G of µG in (MG∞)∞. Put eg := [(pg,n ⊗ hn)n] ∈ F (MG∞ ⊗ W2) for any g ∈ G. Then we see that {eg}g∈G are Rohlin projections of νG. The example above shows that for any finite group G, there exists an action of G on W2 with the Rohlin property. Moreover we see that W2 ⋊νG G is isomorphic to W2 because MG∞ ⋊ µG G is a simple unital AF algebra with a unique tracial state. If G is a finite abelian group, then we have the following characterization of the Rohlin property. Proposition 3.3. Let α be an action of a finite abelian group G on a σ-unital C∗-algebra A. Then α has the Rohlin property if and only if there exists a unitary representation u of G on F (A) such that for any g ∈ G and γ ∈ G. αg(u(γ)) = γ(g)u(γ) Proof. Assume that α has the Rohlin property, and choose Rohlin projections {eg}g∈G of α. Define a map u from G to F (A) by u(γ) := Xg∈G γ(g)eg for any γ ∈ G. Then u is a unitary representation such that αg(u(γ)) = γ(g)u(γ) for any g ∈ G and γ ∈ G. Conversely, assume that there exists a unitary representation u of G on F (A) such that αg(u(γ)) = γ(g)u(γ) for any g ∈ G and γ ∈ G. For any g ∈ G, put eg := 1 G Xγ∈ G γ(g)u(γ). Then {eg}g∈G are Rohlin projections of α. (cid:3) The following lemma is an analogous result of [12, Lemma 3.3] in our case. FINITE GROUP ACTIONS ON CERTAIN STABLY PROJECTIONLESS C∗-ALGEBRAS 9 Lemma 3.4. Let A be a σ-unital C∗-algebra with almost stable rank one, and let α and β be actions of a finite group G on A with the Rohlin property. Assume that αg and βg are approximately unitarily equivalent for any g ∈ G. Then for every finite set F ⊂ A and positive number ǫ, there exists a unitary element v in A such that the following hold for any x ∈ F : kβg(x) − Ad(v∗) ◦ αg ◦ Ad(v)(x)k < ǫ, g ∈ G, k[v, x]k < ǫ + sup g∈G kβg(x) − αg(x)k. Proof. Put F1 := [g∈G βg(F ), and there exist unitary elements {vg}g∈G in M (A) such that kvgβg(y)v∗ g − αg(y)k < ǫ 2 for any y ∈ F1. We choose Rohlin projections {eg}g∈G of α. Then there exist positive contractions {fg}g∈G in A∞ such that [fg] = eg for any g ∈ G. Note that we have fgfha = 0 if g 6= h, f 2 g a = fga and αg(fh)a = fgha for any a ∈ A. Set w := Xg∈G vgfg ∈ A∞. For any a ∈ A, we have w∗wa = Xg∈GXh∈G = Xg∈GXh∈G fgv∗ g vhafh fgv∗ g vhfha = Xg∈GXh∈G g vhafgfh = Xg∈G v∗ afg = a. Since A ⊂ GL( A), there exists a bounded sequence (zn)n consisting of invertible elements in A such that w = (zn)n in ( A)∞. Put un := zn(z∗ 2 and u := (un)n ∈ ( A)∞. Then un is a unitary element in A and we have nzn)− 1 k(zn − un)ak = kun((z∗ nzn) 1 2 a − a)k = k(z∗ nzn) 1 2 a − ak → 0 for any a ∈ A. Therefore we have wa = ua in A∞. Note that we may not have aw = au. For x ∈ F , we have kβg(x) − Ad(u∗) ◦ αg ◦ Ad(u)(x)k = kuβg(x)u∗ − αg(uxu∗)k = kwβg(x)w∗ − αg(wxw∗)k = kXh∈G = kXh∈G = kXh∈G = kXh∈G vhβg(x)fhv∗ fhvhβg(x)v∗ vhxfhv∗ h)k fhvhxv∗ h)k h − αg(Xh∈G h − αg(Xh∈G gh −Xh∈G fghvghβg(x)v∗ fghαg(vhxv∗ h)k fgh(vghβg(x)v∗ gh − αg(vhxv∗ h))k. 10 NORIO NAWATA Since fhfka = 0 for any a ∈ A if h 6= k, for x ∈ F , we have kβg(x) − Ad(u∗) ◦ αg ◦ Ad(u)(x)k {kvghβg(x)v∗ gh − αg(vhxv∗ h)k} ≤ sup h∈G ≤ sup h∈G ǫ 2 < ǫ. ≤ + sup h∈G {kvghβgh(βh−1(x))v∗ gh − αgh(βh−1 (x))k + kαgh(βh−1(x)) − αg(vhxv∗ h)k} {kαh(βh−1(x)) − vhβh(βh−1 (x))v∗ hk} For x ∈ F , we have k[u, x]k = kuxu∗ − xk = kwxw∗ − xk = kXg∈G fg(vgxv∗ g − x)k ≤ sup g∈G {kvgβg(βg−1 (x))v∗ g − αg(βg−1 (x))k + kαg(βg−1 (x)) − xk} < ǫ + sup g∈G kβg(x) − αg(x)k. Set v := un for sufficiently large n. Then we obtain the conclusion. (cid:3) We can show the following theorem by Lemma 3.4 and the Bratteli-Elliott-Evans- Kishimoto intertwining argument [10]. Indeed, the same proof as [12, Theorem 3.5] works by using Lemma 3.4 instead of [12, Lemma 3.3]. Theorem 3.5. Let A be a separable C∗-algebra with almost stable rank one, and let α and β be actions of a finite group G on A with the Rohlin property. Assume that αg and βg are approximately unitarily equivalent for any g ∈ G. Then there exists an approximately inner automorphism θ of A such that θ ◦ αg ◦ θ−1 = βg, g ∈ G. Remark 3.6. Izumi showed such a statement for unital separable C∗-algebras A without assuming that A has stable rank one. We do not know that we can show the theorem above without assuming that A has almost stable rank one. Note that every simple separable Z-stable stably projectionless C∗-algebra has almost stable rank one (see Section 2.4). Corollary 3.7. Let α be an action of a finite group G on W2 with the Rohlin property. Then α is conjugate to νG in Example 3.2. Proof. By Razak's classification theorem [36], every automorphism of W2 is ap- proximately inner. Therefore Theorem 3.5 implies the conclusion. (cid:3) We shall show that there exists an action α of Z2 on W2, which does not have the Rohlin property. This action is locally representable (see Definition 4.7) and will be classified in Section 5. Example 3.8. Let β be an action of Z2 on M2∞ defined by β = ∞On=1 Ad(12n−1 ⊕ (−1)) on ∞On=1 M2n (C). Define a Z2-action α on M2∞ ⊗ W2 ∼= W2 by α := β ⊗ id. Then it is easily seen that β is an outer action on M2∞ but there exists a unitary such that β = Ad(U ) and τ (U ) = 0 where τ is element U in a II1 factor πτ (M2∞ ) ′′ FINITE GROUP ACTIONS ON CERTAIN STABLY PROJECTIONLESS C∗-ALGEBRAS 11 the unique tracial state on M2∞. Hence we have α = Ad(U ⊗ 1), and W2 ⋊α Z2 has exactly two extremal tracial states by Proposition 2.5. (See also [35, Example 2.9].) Therefore α is not cocycle conjugate to ν Z2 because W2 ⋊α Z2 is not isomorphic to W2. Corollary 3.7 implies that α does not have the Rohlin property. Note that the invariant ǫ(α) defined in Definition 2.7 is equal to 0. Let α be an action of a finite group G on a C∗-algebra A. Then the K-groups of A have G-module structures induced by α. We denote by (K0(A), α) and (K1(A), α) such G-modules. Izumi showed that there exist strong K-theoretical constraints for the Rohlin property in the unital case [12, Theorem 3.13] and [13, Theorem 3.3]. We do not use projections and unitary elements in A for the definition of the Rohlin property. Nevertheless, we can show that there exist the same K-theoretical constraints by the following proposition. Proposition 3.9. Let A be a separable simple C∗-algebra, and let α be an action of a finite group G on A with the Rohlin property. Then there exists a unital separable C∗-algebra B, and an action β of G on B with the Rohlin property such that (Ki(A), α) is isomorphic to (Ki(B), β) as G-modules for i = 1, 2. Proof. Let O∞ be the Cuntz algebra generated by an infinite sequence of isometries. Then A⊗O∞ is purely infinite (see, for example, [38, Theorem 4.1.10]). Since α⊗id is outer, (A⊗O∞)⋊α⊗idG is simple and purely infinite by [27, Lemma 10]. Therefore we see that (A ⊗ O∞)α⊗id is purely infinite, and hence (A ⊗ O∞)α⊗id has a nonzero projection p. Put B := p(A ⊗ O∞)p and β := (α ⊗ id)p(A⊗O∞)p. Choose Rohlin projections {eg = [(fn,g)n]}g∈G of α in F (A) where (fn,g)n ∈ A∞. Then {((fn,g ⊗ 1)p)n}g∈G are Rohlin projections of β, and hence β has the Rohlin property. Since the inclusion map i from p(A⊗O∞)p into A⊗O∞ induces isomorphisms of K-groups by Brown's theorem [4] and O∞ is KK-equivalent to C, we see that (Ki(B), β) is isomorphic to (Ki(A), α) as G-modules for i = 1, 2. (cid:3) We refer the reader to [13] for details of K-theoretical constraints. The following corollary is an immediate consequence of Proposition 3.9 and [13, Theorem 3.6]. Corollary 3.10. Let A be a simple separable C∗-algebra such that either K0(A) or K1(A) is isomorphic to Z. Then there exist no non-trivial finite group actions on A with the Rohlin property. For given K-theoretical invariants, Izumi constructed model actions with the Rohlin property on Kirchberg algebras [13, Theorem 5.3]. As an application of this construction, we obtain the following corollary by a similar argument as in the proof of Proposition 3.9 and [13, Corollary 5.5]. Corollary 3.11. Let α be an action of a finite group G with the Rohlin property on a simple separable nuclear C∗-algebra A in the UCT class. Then A ⋊α G is in the UCT class. 4. Approximately representable actions In this section we introduce the approximate representability for finite group actions on σ-unital C∗-algebras and study some basic properties of approximately representable actions. Moreover we give some examples of Rohlin actions on W2 by using a classification result of some approximately representable actions. Definition 4.1. An action α of a finite abelian group G on a σ-unital C∗-algebra A is said to be approximately representable if there exist elements {w(g)}g∈G in (Aα)∞ such that a map u from G to F (Aα) given by u(g) = [w(g)] is a unitary representation of G and αg(a) = w(g)aw(g)∗ 12 NORIO NAWATA in A∞ for any a ∈ A and g ∈ G. If A is unital, then the definition above coincides with the definition in [12]. Remark 4.2. For a general finite group G, we can define the approximate repre- sentability as in [12, Remark 3.7]. Put F (Aα, A) := A∞ ∩ (Aα)′/Ann(Aα, A∞) (see Section 2.3). An action α of a finite group G on a σ-unital C∗-algebra A is said to be approximately representable if there exist elements {w(g)}g∈G in A∞ ∩ (Aα)′ such that a map from G to F (Aα, A) given by u(g) = [w(g)] is a unitary representation of G and αg(a) = w(g)aw(g)∗, αg(u(h)) = u(ghg−1), in A∞, in F (Aα, A), a ∈ A, g ∈ G g, h ∈ G. We shall show that the approximate representability is the "dual" notion of the Rohlin property. Note that if α is an action of a finite group G on A, then we have (Aα)∞ = (A∞)α. In a similar way, we see the following lemma. Lemma 4.3. Let α be an action of a finite group G on A, and let B be a subalgebra of Aα. If w is an element in A∞ ∩ B′ such that wa = αg(w)a for any a ∈ A and g ∈ G, then there exists an element y ∈ (Aα)∞ ∩ B′ such that wa = ya in A∞ for any a ∈ A. Proof. Let (wn)n be a representative of w. Define yn := 1 and put y = (yn)n. Then y has the desired property. GPg∈G αg(xn) ∈ Aα, (cid:3) See [12, Lemma 3.8] for the unital case of the following proposition. Proposition 4.4. Let α be an action of a finite abelian group G on a σ-unital C∗-algebra A. Then (i) α has the Rohlin property if and only if α is approximately representable; (ii) α is approximately representable if and only α has the Rohlin property. Proof. (i) Assume that α has the Rohlin property. Then there exists a unitary representation u of G on F (A) such that αg(u(γ)) = γ(g)u(γ) for any g ∈ G and γ ∈ G by Proposition 3.3. Choose elements {w(γ)}γ∈ G in A∞ such that [w(γ)] = u(γ), and let λg be the implementing unitary of αg in M (A ⋊α G). Then we have aλgw(γ)λ∗ g = aγ(g)w(γ) for any a ∈ A, g ∈ G and γ ∈ G. This equation implies w(γ)∗aλgw(γ) = γ(g)aλg, and hence αγ(x) = w(γ)∗xw(γ) for any x ∈ A ⋊α G and γ ∈ G. Since we have (A ⋊α G) α = A, α is approximately representable. The converse follows from a similar computation. (ii) Assume that α is approximately representable, and take elements {w(g)}g∈G in (Aα)∞ as in Definition 4.1. Then we have g = w(g)aw(g)∗ λgaλ∗ a ∈ A, g ∈ G and λgw(h)λ∗ g = w(h) g, h ∈ G where λg is the implementing unitary of αg in M (A ⋊α G). Put z(g) := w(g)∗λg ∈ (A ⋊α G)∞ for any g ∈ G. By the equations above, we have aλhz(g) = aλhw(g)∗λg = aw(g)∗λhλg = aw(g)∗λgλh = w(g)∗λgaλh = z(g)aλh FINITE GROUP ACTIONS ON CERTAIN STABLY PROJECTIONLESS C∗-ALGEBRAS 13 for any a ∈ A and g, h ∈ G. Therefore {zg}g∈G are elements in (A ⋊α G)∞. Define a map v from G to F (A ⋊α G) by v(g) = [z(g)]. Then it can be easily checked that v is a unitary representation and αγ(v(g)) = γ(g)v(g) for any γ ∈ G and g ∈ G. Proposition 3.3 implies that α has the Rohlin property. Conversely, assume that α has the Rohlin property. Then there exists a unitary representation u of G on F (A ⋊α G) such that αγ(u(g)) = γ(g)u(g) for any γ ∈ G and g ∈ G by Proposition 3.3. Choose elements {z(g)}g∈G in (A ⋊α G)∞ such that [z(g)] = u(g). Then we have αγ(λgz(g)∗)x = γ(g)λgγ(g)z(g)∗x = λgz(g)∗x for any x ∈ A ⋊α G, g ∈ G and γ ∈ G where λg is the implementing unitary of αg in M (A ⋊α G). Since we have λgz(g)∗b = bλgz(g)∗ for any b ∈ Aα and (A ⋊α G) α = A, Lemma 4.3 implies that there exist elements {y(g)}g∈G in A∞ ∩ (Aα)′ such that y(g)x = λgz(g)∗x for any x ∈ A ⋊α G and g ∈ G. For any a ∈ A and g, h ∈ G, we have αh(y(g))a = λhλgz(g)∗λ∗ ha = λhλgλ∗ haz(g)∗ = λgz(g)∗a = y(g)a. Hence there exist elements {w(g)}g∈G in (Aα)∞ such that w(g)a = λgz(g)∗a for any a ∈ A and g ∈ G by Lemma 4.3. It can be easily checked that a map v from G to F (Aα) given by v(g) = [w(g)] is a unitary representation of G and αg(a) = w(g)aw(g)∗ for any a ∈ A and g ∈ G. Consequently α is approximately representable. (cid:3) By Theorem 3.5, Proposition 4.4 and the Takesaki-Takai duality, we can show the following classification result of some approximately representable actions. See [12, Corollary 3.9] for the unital case. Corollary 4.5. Let α and β be actions of a finite abelian group G on a separable C∗-algebra A, and let eα and eβ be the projections in M (A ⋊α G) and M (A ⋊β G), respectively, defined in Section 2.2. Assume that α and β are approximately representable and A ⋊α G and A ⋊β G have almost stable rank one. Then α and β are conjugate if and only if there exists an isomorphism θ from A ⋊α G onto A ⋊β G such that θ ◦ αγ ◦ θ−1 is approximately unitarily equivalent to βγ for any γ ∈ G and ¯θ(eα) is Murray-von Neumann equivalent to eβ (in M (A ⋊β G)). Proof. The only if part is obvious. We will show the if part. By Theorem 3.5 and Proposition 4.4, there exists an approximately inner automorphism θ1 of A ⋊β G 1 = βγ for any γ ∈ G. Replacing θ with θ1 ◦ θ, such that θ1 ◦ θ ◦ αγ ◦ θ−1 ◦ θ−1 we may assume that θ is an isomorphism from A ⋊α G onto A ⋊β G such that θ ◦ αγ = βγ ◦θ for any γ and ¯θ(eα) is Murray-von Neumann equivalent to eβ because θ1 is approximately inner. Then there exists an isomorphism Ψ from A ⋊α G ⋊ α G βg ◦ Ψ for any g ∈ G and ¯Ψ(eα) is Murray-von onto A ⋊β G ⋊ β ¯β. Hence the Takesaki-Takai duality Neumann equivalent to eβ in M (A ⋊β G ⋊ β theorem implies that there exists an automorphism Θ of A ⊗ K(ℓ2(G)) such that Θ intertwines α ⊗ Ad(ρ) and β ⊗ Ad(ρ). Let e ∈ K(ℓ2(G)) be the projection onto the constant functions. Then we have (A, α) ∼= ((1 ⊗ e)(A ⊗ K(ℓ2(G)))(1 ⊗ e), α ⊗ Ad(ρ)(1⊗e)(A⊗K(ℓ2(G)))(1⊗e)) G such that Ψ ◦ αg = G) ∼= ( ¯Θ(1 ⊗ e)(A ⊗ K(ℓ2(G))) ¯Θ(1 ⊗ e), β ⊗ Ad(ρ) ¯Θ(1⊗e)(A⊗K(ℓ2(G))) ¯Θ(1⊗e)). ¯β, we see Since ¯Ψ(eα) is Murray-von Neumann equivalent to eβ in M (A ⋊β G ⋊ β that ¯Θ(1⊗e) is Murray-von Neumann equivalent to 1⊗e in M (A⊗K(ℓ2(G)))β⊗Ad(ρ) G) 14 NORIO NAWATA (see Section 2.2). Therefore we have (A, α) ∼= ( ¯Θ(1 ⊗ e)(A ⊗ K(ℓ2(G))) ¯Θ(1 ⊗ e), β ⊗ Ad(ρ) ¯Θ(1⊗e)(A⊗K(ℓ2(G))) ¯Θ(1⊗e)) ∼= ((1 ⊗ e)(A ⊗ K(ℓ2(G)))(1 ⊗ e), β ⊗ Ad(ρ)(1⊗e)(A⊗K(ℓ2(G)))(1⊗e)) ∼= (A, β). (cid:3) Remark 4.6. Let s be a strictly positive element in Aα. Define a positive element sα in A ⋊α G by sα := sλg 1 G Xg∈G where λg is the implementing unitary of αg in M (A ⋊α G). It can be checked that sα(A ⋊α G) = eα(A⋊α G) and the Cuntz class [sα] does not depend on the choice of s. Since A ⋊α G and A ⋊β G have almost stable rank one, we see that the condition that ¯θ(eα) is Murray-von Neumann equivalent to eβ is equivalent to the condition that θ(sα) is Cuntz equivalent to sβ by [7, Theorem 3]. We shall consider examples of approximately representable actions. Definition 4.7. Let C be a class of C∗-algebras. An action α of a finite group G on a C∗-algebra A is said to be locally representable if there exists an α-invariant increasing sequence of C∗-subalgebras {An}n∈N whose union is dense in A, and a unitary representation u(n) of G in M (An) such that αg(a) = Ad(u(n)(g))(a) for any a ∈ An and g ∈ G. Moreover if An is contained in C for any n ∈ N, we say that α is locally representable in C. Note that we do not assume that An ⊂ An+1 is a nondegenerate inclusion in the definition above. Hence M (An) may not be contained in M (An+1). We denote by CF the class of all finite dimensional C∗-algebras. We shall show that every locally representable action on a separable C∗-algebra is approximately representable. Lemma 4.8. Let α be an action of a finite group G on a separable C∗-algebra A. Assume that there exists an α-invariant increasing sequence of C∗-subalgebras {An}n∈N whose union is dense in A. Then there exists an approximate unit {hn}n∈N for A such that hn ∈ (An)α for any n ∈ N. Proof. Let {Fn}n∈N be an increasing sequence of finite subsets in A such that Fn ⊂ An and ∪∞ n=1Fn is dense in A. Since (An)α ⊂ An is a nondegenerate inclusion, there exists a positive contraction hn in (An)α such that khnx − xk < 1 n for any x ∈ Fn. Then {hn}n∈N is an approximate unit for A. (cid:3) Proposition 4.9. If α is a locally representable action of a finite group G on a separable C∗-algebra A, then α is approximately representable. Proof. Let {An}n∈N be an increasing sequence of C∗-subalgebras of A and {u(n)}n∈N unitary representations as in Definition 4.7. There exists an approximate unit {hn}n∈N for A such that hn ∈ (An)α for any n ∈ N by Lemma 4.8. For any g ∈ G, put w(g) := (u(n)(g)hn)n ∈ A∞. Then we have αg(a) = w(g)aw(g)∗ in A∞ for any a ∈ A and g ∈ G, and hence w(g) ∈ A∞ ∩ (Aα)′. Since we have u(n)(g)hn = hnu(n)(g) for any n ∈ N and g ∈ G, it can be easily checked that a FINITE GROUP ACTIONS ON CERTAIN STABLY PROJECTIONLESS C∗-ALGEBRAS 15 map v from G to F (Aα, A) given by v(g) = [w(g)] is a unitary representation of G and αg(v(h)) = v(ghg−1) for any g, h ∈ G. Therefore α is approximately representable. (cid:3) Izumi showed that if β is an outer action of a finite group G on a separable simple unital nuclear C∗-algebra A, then the action β ⊗ id of G on A ⊗ O2 ∼= O2 has the Rohlin property [12, Corollary 4.3]. This result can be regarded as an equivariant version of Kirchberg-Phillips's theorem [20, Theorem 3.8]. We cannot show such a statement for W2. Indeed, Example 3.8 shows that there exists an outer action β on M2∞ such that the action β ⊗ id on M2∞ ⊗ W2 ∼= W2 does not have the Rohlin property. Note that this action is not strongly outer and (M2∞ ⊗ W2) ⋊β⊗id Z2 has two extremal tracial states. Hence it is natural to consider the following question. Question 4.10. Let A be a separable simple nuclear C∗-algebra with a unique tracial state and no unbounded traces, and let β be a strongly outer action of a finite group G on A. Does the action β ⊗ id of G on A ⊗ W2 have the Rohlin property? We shall show the following (non-trivial) partial result of the question above as an application of Corollary 4.5. Corollary 4.11. Let A be a simple AF algebra with a unique tracial state and no unbounded traces, and let β be a strongly outer action of a finite abelian group G on A. Assume that β is approximately representable and A ⋊β G is an AF algebra. Then (A ⊗ W2, β ⊗ id) ∼= (W2, νG) where νG is the action in Example 3.2, and hence β ⊗ id has the Rohlin property. Proof. Proposition 2.1 and [22, Theorem 3.1] imply A ⋊β G is a simple C∗-algebra with a unique tracial state. It is easy to see that A ⋊β G has no unbounded traces. Since (A ⊗ W2) ⋊β⊗id G is isomorphic to (A ⋊β G) ⊗ W2, (A ⊗ W2) ⋊β⊗id G is iso- morphic to W2. On the other hand, W2 ⋊ νG G is isomorphic to W2 and Proposition 4.9 implies that νG is approximately representable. Let θ be an isomorphism from W2 ⋊ νG G onto (A ⊗ W2) ⋊β⊗id G. We denote by ωβ⊗id and ων the unique tracial state on (A ⊗ W2) ⋊β⊗id G and the unique tracial state on W2 ⋊νG G, respectively. Then we have ¯ωβ⊗id(eα⊗id) = ¯ων(eνG ) = 1 G . Hence we see that θ(eνG ) is Murray-von Neumann equivalent to eα⊗id by Proposi- tion 2.9. Since every automorphism of W2 is approximately inner, we obtain the conclusion by Corollary 4.5. (cid:3) Remark 4.12. (i) If an action β on a C∗-algebra A is locally representable in CF, then A ⋊β G is an AF algebra. (ii) There exists a strongly outer action β of a finite abelian group on M2∞ such that β is locally representable in CF and β does not have the Rohlin property. Indeed, define an action β of Z2 on M2∞ by β = ∞On=1 Ad(12n−1+1 ⊕ −12n−1−1) on ∞On=1 M2n (C). Then β is such an action (see [35, Example 3.7] for the proof). But the Corollary above shows that the action β ⊗ id on M2∞ ⊗ W2 ∼= W2 has the Rohlin property. 16 NORIO NAWATA 5. Symmetries of W2 Let CR be the class of C∗-algebras A in Robert's classification theorem [37, Theorem 1.0.1], that is, A is either a 1-dimensional NCCW complex with trivial K1-group, an inductive limit C∗-algebra of such C∗-algebras or a C∗-algebra stably isomorphic to one such inductive limit. It is obvious that the outer actions on W2 in Example 3.2 and Example 3.8 are locally representable in CR. In this section we shall classify the locally representable outer Z2-actions (W2, α) in CR up to conjugacy and cocycle conjugacy. We denote by τ the unique tracial state on W2 in this section. Lemma 5.1. Let A be a simple stably projectionless C∗-algebra with no unbounded traces, and let α be an outer action of a finite group G on A. Assume that α is locally representable in CR. Then A⋊αG is a simple stably projectionless C∗-algebra with no unbounded traces and is contained in CR. Proof. It is easy to see that A ⋊α G has no unbounded traces and Aα is stably pro- jectionless. Since α is outer, A ⋊α G is simple by [22, Theorem 3.1]. Hence we see that A ⋊α G is a simple stably projectionless C∗-algebra with no unbounded traces. Let {An}n∈N be an increasing sequence of C∗-subalgebras of A and {u(n)}n∈N uni- tary representations as in Definition 4.7. Then we have A ⋊α G = ∞[n=1 An ⋊ Ad(u(n)) G. Since An ⋊ sional C∗-algebra, we see that A ⋊α G is contained in CR. Ad(u(n)) G is isomorphic to the tensor product of An and a finite dimen- (cid:3) Theorem 5.2. Let α and β be outer actions of Z2 on W2. Assume that α and β are locally representable in CR. Then α is conjugate to β if and only if W2 ⋊α Z2 is isomorphic to W2 ⋊β Z2 and ǫ(α) = ǫ(β) where ǫ(α) is the invariant defined in Definition 2.7. Proof. It is easy to see that if α is conjugate to β, then W2 ⋊α Z2 is isomorphic to W2 ⋊β Z2 and ǫ(α) = ǫ(β). We will show the converse. Let θ be an isomorphism from W2 ⋊α Z2 onto W2 ⋊β Z2. First, we shall show K0(α) = −id. Put A := W2 ⋊α Z2, and let i1 and i2 denote the inclusion from A into A ⋊ α Z2 and the inclusion from A ⋊ α Z2 into Z2 ∼= A⊗M2(C), respectively. Since A⋊ αZ2 is isomorphic to W2⊗M2(C) A⋊ αZ2⋊ α and K0(W2) = 0, we have K0(i2 ◦ i1) = 0. The Takesaki-Takai duality implies i2 ◦ i1(a) =(cid:18) a 0 0 α(a) (cid:19) for any a ∈ A (see Section 2.2). Hence we have K0(α) = −id. Note that the same argument shows K0( β) = −id. (i) The case where W2 ⋊α Z2 and W2 ⋊β Z2 have a unique tracial state. We denote by ωα and ωβ the unique tracial state on W2 ⋊α Z2 and the unique tracial state on W2 ⋊β Z2, respectively. We see that β is approximately unitarily equivalent to θ◦ α◦θ−1 by Lemma 5.1 and Robert's classification theorem (Theorem 2.11). Moreover we have ¯ωα(eα) = ¯ωβ(eβ) = 1 2 , and hence θ(eα) is Murray-von Neumann equivalent to eβ by Proposition 2.9. Therefore Corollary 4.5 shows that α is conjugate to β. (ii) The case where W2 ⋊α Z2 and W2 ⋊β Z2 have two extremal tracial states. Let ωα,0 and ωα,1 be extremal tracial states on W2 ⋊α Z2 and ωβ,0 and ωβ,1 extremal FINITE GROUP ACTIONS ON CERTAIN STABLY PROJECTIONLESS C∗-ALGEBRAS 17 tracial states on W2 ⋊β Z2. Since W2 ⋊α Z2 ⋊ α Z2 and W2 ⋊β Z2 ⋊ β Z2 have a unique tracial state, we have ωα,0 ◦ α = ωα,1 and ωβ,0 ◦ β = ωβ,1. Hence we see that β is approximately unitarily equivalent to θ ◦ α ◦ θ−1 by Lemma 5.1 and Robert's classification theorem (Theorem 2.11). Choose a unitary element Uα in πτ (W2) such that α = Ad(Uα) and U 2 α = 1 (see Proposition 2.5 and Remark 2.6). By Remark 2.6, we have either ′′ or (¯ωα,0(eα), ¯ωα,1(eα)) = ( 1 2 (1 + τ (Uα)), 1 2 (1 − τ (Uα))) (¯ωα,0(eα), ¯ωα,1(eα)) = ( 1 2 (1 − τ (Uα)), 1 2 (1 + τ (Uα))). In the same way, we can compute ¯ωβ,0(eβ) and ¯ωβ,1(eβ). Since ǫ(α) = ǫ(β), we have either ¯ωα,0(eα) = ¯ωβ,0(eβ) and ¯ωα,1(eα) = ¯ωβ,1(eβ) or ¯ωα,0(eα) = ¯ωβ,1(eβ) and ¯ωα,1(eα) = ¯ωβ,0(eβ). Hence replacing θ with θ ◦ α if necessary, we have ¯ω(θ(eα)) = ¯ω(eβ) for any ω ∈ T1(W2 ⋊β Z2). Therefore α is conjugate to β by Proposition 2.9 and Corollary 4.5. (cid:3) Remark 5.3. (i) With notation as above, W2 ⋊α Z2 is isomorphic to W2 ⋊β Z2 if and only if T (W2 ⋊α Z2) ∼= T (W2 ⋊β Z2) as topological cones and K0(W2 ⋊α Z2) ∼= K0(W2 ⋊β Z2) as abelian groups, and that these isomorphisms are compatible with the pairing between T and K0 by Lemma 5.1 and Robert's classification theorem. (ii) If W2 ⋊α Z2 and W2 ⋊β Z2 have a unique tracial state, then α is conjugate to β if and only if W2 ⋊α Z2 is isomorphic to W2 ⋊β Z2. We shall consider the classification up to cocycle conjugacy. Proposition 5.4. Let A be a simple exact σ-unital stably projectionless C∗-algebra with a unique tracial state and no unbounded traces, and let α and β be actions of a finite abelian group G on A. Assume that A has strict comparison and almost stable rank one. Then α is cocycle conjugate to β if and only if α is conjugate to β. Proof. In general, it is easy to see that if α is cocycle conjugate to β, then α is β, conjugate to β. We will show the if part. Since we see that α is conjugate to there exists an automorphism θ of A ⊗ K(ℓ2(G)) such that θ intertwines α ⊗ Ad(ρ) and β ⊗ Ad(ρ) by the Takesaki-Takai duality theorem. Note that A ⊗ K(ℓ2(G)) is a simple exact σ-unital stably projectionless C∗-algebra with a unique tracial state ω and no unbounded traces, which has strict comparison and almost stable rank one. Let e ∈ K(ℓ2(G)) be the projection onto the constant functions. Then we have (A, α) ∼= (¯θ(1 ⊗ e)(A ⊗ K(ℓ2(G)))¯θ(1 ⊗ e), β ⊗ Ad(ρ) ¯θ(1⊗e)(A⊗K(ℓ2(G))) ¯θ(1⊗e)) by the proof of Corollary 4.5. Since ω is the unique tracial state on A ⊗ K(ℓ2(G)), we have ¯ω(¯θ(1 ⊗ e)) = ¯ω(1 ⊗ e). By Proposition 2.9, there exists a partial isometry v in M (A ⊗ K(ℓ2(G))) such that v∗v = ¯θ(1 ⊗ e) and vv∗ = 1 ⊗ e. For any g ∈ G, define u(g) := vβg ⊗ Adρg(v∗). Then we have (1 ⊗ e)u(g)(1 ⊗ e) = u(g) for any g ∈ G, and we may regard u as a β-cocycle in M (A). It can be easily checked that (¯θ(1 ⊗ e)(A ⊗ K(ℓ2(G)))¯θ(1 ⊗ e), β ⊗ Ad(ρ) ¯θ(1⊗e)(A⊗K(ℓ2(G))) ¯θ(1⊗e)) ∼= (A, βu), 18 NORIO NAWATA and hence we obtain the conclusion. (cid:3) Theorem 5.5. Let α and β be outer actions of Z2 on W2. Assume that α and β are locally representable in CR. Then α is cocycle conjugate to β if and only if W2 ⋊α Z2 is isomorphic to W2 ⋊β Z2. Proof. The only if part is obvious. We will show the if part. Let θ be an isomor- phism from W2 ⋊α Z2 onto W2 ⋊β Z2. The proof of Theorem 5.2 implies that β is approximately unitarily equivalent to θ ◦ α ◦ θ−1. Hence α is conjugate to β by Theorem 3.5 and Proposition 4.4. Proposition 5.4 shows that α is cocycle conjugate to β. (cid:3) The following example is based on Blackadar's construction of symmetries on M2∞ in [2] and Robert's classification theorem. Example 5.6. [9, Theorem 5.2.1 and Theorem 5.2.2] show that there exists a simple stably projectionless C∗-algebra A such that A is expressible as an inductive limit C∗-algebra of 1-dimensional NCCW complexes with trivial K1-groups and (K0(A), (T (A), T1(A)), rA) = (Z, (R+, ∅), 0). By [33, Proposition 5.2], there exists a hereditary subalgebra B of A such that B has a unique tracial state ω and no unbounded traces. Since B is stably isomorphic to A by Brown's theorem [4], B is contained in CR and (K0(B), (T (B), T1(B)), rB ) = (Z, (R+, {ω}), 0). By Robert's classification theorem (Theorem 2.11), there exists an automorphism β of B such that K0(β) = −id. Define a homomorphism Φi from M2i(C) ⊗ B to M2i+1(C) ⊗ B by Φi(x) =(cid:18) x 0 (idM2i (C) ⊗ β)(x) (cid:19) , 0 (M2i(C) ⊗ B, Φi). Then we have K0(Φi) = 0, and hence K0(D) = and put D := lim → 0. It can be easily checked that D is a simple stably projectionless C∗-algebra with a unique tracial state and no unbounded traces. Therefore D is isomorphic to W2 by Robert's classification theorem. We shall construct a locally representable Z2-action (W2, α) in CR. It is easy to see that Φi is unitarily equivalent to a homomorphism Ψi such that   x11 0 x21 0 x21 x22 (cid:19)) = Ψi((cid:18) x11 x12 for any(cid:18) x11 x12 lim → αi of M2i(C) ⊗ B by x21 x22 (cid:19) ∈ M2i(C)⊗B. Since lim → 0 (idM2i−1 (C) ⊗ β)(x22) 0 (idM2i−1 (C) ⊗ β)(x12) x12 0 x22 0 0 (idM2i−1 (C) ⊗ β)(x21) 0 (idM2i−1 (C) ⊗ β)(x11)   (M2i(C)⊗B, Ψi) is isomorphic to D, (M2i (C) ⊗ B, Ψi) is isomorphic to W2. For any i ∈ N, define an automorphism x21 x22 (cid:19)) := Ad((cid:18) 1 αi((cid:18) x11 x12 x21 x22 (cid:19) ∈ M2i(C) ⊗ B. Then we have Ψi ◦ αi = αi+1 ◦ Ψi, and 0 −1 (cid:19))((cid:18) x11 x12 x21 x22 (cid:19)) 0 for any (cid:18) x11 x12 hence there exists an automorphism α of W2 such that α(a) = αi(a) for any a ∈ M2i(C) ⊗ B and i ∈ N. It is obvious that α is a locally representable Z2-action in CR. We shall consider the fixed point algebra W α is isomorphic to an inductive limit C∗-algebra lim → 2 . It can be easily checked that W α 2 (M2i−1(C)⊗B⊕M2i−1(C)⊗B, Θi) FINITE GROUP ACTIONS ON CERTAIN STABLY PROJECTIONLESS C∗-ALGEBRAS 19 where Θi((x1, x2)) = ((cid:18) x1 0 (idM2i−1 (C) ⊗ β)(x2) (cid:19) ,(cid:18) x2 0 0 (idM2i−1 (C) ⊗ β)(x1) (cid:19)) 0 2 ) = Z[ 1 for any (x1, x2) ∈ M2i−1 (C)⊗B⊕M2i−1(C)⊗B. Therefore we have K0(W α 2 ]. It is easy to see that α is outer, and hence W2 ⋊α Z2 is stably isomorphic to W α 2 . Consequently we obtain a locally representable outer Z2-action (W2, α) in CR such that K0(W2 ⋊α Z2) = Z[ ] and K1(W2 ⋊α Z2) = 0. 1 2 Note that the dual action α on W2 ⋊α Z2 has the Rohlin property by Proposition 4.4. Remark 5.7. Blackadar showed that if A is a C∗-algebra with K0(A) = K1(A) = 0 and β is a Z2-action on A, then K0(A ⋊β Z2) and K1(A ⋊β Z2) are uniquely 2-divisible (see [12, Lemma 4.4]). (Note that an abelian group Γ is uniquely 2- divisible if and only if Γ ⊗Z Z[ 1 2 ] is isomorphic to Γ.) For every pair of countable abelian uniquely 2-divisible groups Γ0 and Γ1, Izumi constructed an approximately representable outer action β of Z2 on O2 such that K0(O2⋊β Z2) = Γ0 and K1(O2⋊β Z2) = Γ1 [12, Theorem 4.8 (3)]. If the Robert conjecture is true, then we can construct an approximately representable outer action β of Z2 on W2 such that K0(W2 ⋊β Z2) = Γ0 and K1(W2 ⋊β Z2) = Γ1 by using α in Example 5.6 and the same construction of Izumi. Indeed, there exists a simple separable nuclear stably projectionless C∗-algebra B with a unique tracial state and no unbounded traces such that K0(B) = Γ0 and K1(B) = Γ1 (see, for example, [25] and [33, Proposition 5.2]). If the Robert conjecture is true, then B ⊗ W2 is isomorphic to W2. Define an action β on B ⊗W2 ∼= W2 by β := id⊗α. Then β is an approximately representable outer action such that K0(W2 ⋊β Z2) = Γ0 and K1(W2 ⋊β Z2) = Γ1 by the Kunneth formula. Acknowledgments The author would like to thank the people in University of Copenhagen, where this work was done, for their hospitality. He is also grateful to Hiroki Matui for many helpful discussions and valuable suggestions. References [1] E. B´edos, On the uniqueness of the trace on some simple C ∗-algebras, J. Operator Theory 30 (1993), no. 1, 149 -- 160. [2] B. Blackadar, Symmetries of the CAR algebra, Ann. of Math. (2) 131 (1990), no. 3, 589 -- 623. [3] B. Blackadar, Operator Algebras : Theory of C-Algebras and von Neumann Algebras, En- cyclopaedia of Mathematical Sciences, 122, Springer, 2006. [4] L. G. Brown, Stable isomorphism of hereditary subalgebras of C ∗-algebras, Pacific J. Math. 71 (1977), 335 -- 348. [5] A. Connes, Outer conjugacy classes of automorphisms of factors, Ann. Sci. Ecole Norm. Sup. (4) 8 (1975), no. 3, 383 -- 419. [6] A. Connes, Periodic automorphisms of the hyperfinite factor of type II1, Acta Sci. Math. (Szeged) 39 (1977), no. 1-2, 39 -- 66. [7] K. Coward, G. A. Elliott and C. Ivanescu, The Cuntz semigroup as an invariant for C∗- algebras, J. Reine Angew. Math. 623 (2008), 161 -- 193. [8] A. Dean, A continuous field of projectionless C∗-algebras, Canad. J. Math. 53 (2001), no. 1, 51 -- 72. [9] G. A. Elliott, An invariant for simple C ∗-algebras, Canadian Mathematical Society. 1945 -- 1995, Vol. 3, 61 -- 90, Canadian Math. Soc., Ottawa, ON, 1996. [10] D. E. Evans and A. Kishimoto, Trace scaling automorphisms of certain stable AF algebras, Hokkaido Math. J. 26 (1997), no. 1, 211 -- 224. [11] R. H. Herman and V. F. R. Jones, Models of finite group actions, Math. Scand. 52 (1983), no. 2, 312 -- 320. 20 NORIO NAWATA [12] M. Izumi, Finite group actions on C∗-algebras with the Rohlin property, I, Duke Math. J. 122 (2004), no. 2, 233 -- 280. [13] M. Izumi, Finite group actions on C∗-algebras with the Rohlin property, II, Adv. Math. 184 (2004), no. 1, 119 -- 160. [14] M. Izumi, Group actions on operator algebras, in: Proceedings of the International Congress of Mathematicians, vol. III, Hyderabad, India, 2010, Hindustan Book Agency, 2010, pp. 1528 -- 1548. [15] M. Izumi and H. Matui, Z2-actions on Kirchberg algebras, Adv. Math. 224 (2010), no. 2, 355 -- 400. [16] B. Jacelon, A simple, monotracial, stably projectionless C-algebra, J. Lond. Math. Soc. (2) 87 (2013), no. 2, 365 -- 383. [17] X. Jiang and H. Su, On a simple unital projectionless C∗-algebra, Amer. J. Math. 121 (1999), no. 2, 359 -- 413. [18] T. Katsura and H. Matui, Classification of uniformly outer actions of Z2 on UHF algebras, Adv. Math. 218 (2008), no. 3, 940 -- 968. [19] E. Kirchberg, Central sequences in C∗-algebras and strongly purely infinite algebras, Operator Algebras: The Abel Symposium 2004, 175 -- 231, Abel Symp., 1, Springer, Berlin, 2006. [20] E. Krichberg and N. C. Phillips, Embedding of exact C∗-algebras in the Cuntz algebra O2, J. Reine Angew. Math. 525 (2000), 17 -- 53. [21] A. Kishimoto, On the fixed point algebra of a UHF algebra under a periodic automorphism of product type, Publ. Res. Inst. Math. Sci. 13 (1977/78), no. 3, 777 -- 791. [22] A. Kishimoto, Outer automorphisms and reduced crossed products of simple C ∗-algebras, Comm. Math. Phys. 81 (1981), no. 3, 429 -- 435. [23] A. Kishimoto, The Rohlin property for automorphisms of UHF algebras, J. Reine Angew. Math. 465 (1995), 183 -- 196. [24] A. Kishimoto, Unbounded derivations in AT algebras, J. Funct. Anal. 160 (1998), no. 1, 270 -- 311. [25] A. Kishimoto, Pairs of simple dimension groups, Internat. J. Math. 10 (1999), no. 6, 739 -- 761. [26] A. Kishimoto and A. Kumjian, Simple stably projectionless C∗-algebras arising as crossed products, Canad. J. Math. 48 (1996), no. 5, 980 -- 996. [27] A. Kishimoto and A. Kumjian, Crossed products of Cuntz algebras by quasi-free automor- phisms, in Operator algebras and their applications (Waterloo, ON, 1994/1995), 173 -- 192, Fields Inst. Commun., 13, Amer. Math. Soc., Providence, RI, 1997. [28] H. Matui, Z-actions on AH algebras and Z2-actions on AF algebras, Comm. Math. Phys. 297 (2010), no. 2, 529 -- 551. [29] H. Matui, ZN -actions on UHF algebras of infinite type, J. Reine Angew. Math. 657 (2011), 225 -- 244. [30] H. Matui and Y. Sato, Z-stability of crossed products by strongly outer actions, Comm. Math. Phys. 314 (2012), 193 -- 228. [31] H. Matui and Y. Sato, Z-stability of crossed products by strongly outer actions II , preprint, arXiv:1205.1590v1 [math.OA]. [32] H. Nakamura, Aperiodic automorphisms of nuclear purely infinite simple C ∗-algebras, Er- godic Theory Dynam. Systems 20 (2000), no. 6, 1749 -- 1765. [33] N. Nawata, Picard groups of certain stably projectionless C∗-algebras, to appear in J. London Math. Soc. doi: 10.1112/jlms/jdt013. [34] G. K. Pedersen, C ∗-Algebras and Their Automorphism Groups, Academic Press, London- New York-San Francisco, 1979. [35] N. C. Phillips, Finite cyclic group actions with the tracial Rokhlin property, preprint, arXiv:math/0609785 [math.OA] [36] S. Razak, On the classification of simple stably projectionless C∗-algebras, Canad. J. Math. 54 (2002), no. 1, 138 -- 224. [37] L. Robert, Classification of inductive limits of 1-dimensional NCCW complexes, Adv. Math. 231 (2012), no. 5, 2802 -- 2836. [38] M. Rørdam, Classification of nuclear, simple C∗-algebras, in:Classification of nuclear C ∗- algebras. Entropy in operator algebras, 1 -- 145, Encyclopaedia of Mathematical Sciences, 126, Springer, 2002. [39] M. Rørdam, The stable and the real rank of Z-absorbing C∗-algebras, Internat. J. Math. 15 (2004), no. 10, 1065 -- 1084. [40] J. Rosenberg, Appendix to O. Bratteli's paper on "Crossed products of UHF algebras", Duke Math. J. 46 (1979), no. 1, 25 -- 26. [41] L. Santiago, Reduction of the dimension of nuclear C∗-algebras, preprint, arXiv:1211.7159v1 [math.OA]. FINITE GROUP ACTIONS ON CERTAIN STABLY PROJECTIONLESS C∗-ALGEBRAS 21 [42] S. Sakai, C ∗-algebras and W ∗-algebras, Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 60. Springer-Verlag, New York-Heidelberg, 1971. [43] Y. Sato, The Rohlin property for automorphisms of the Jiang-Su algebra, J. Funct. Anal. 259 (2010), no. 2, 453 -- 476. [44] H. Takai, On a duality for crossed products of C ∗-algebras, J. Funct.l Anal. 19 (1975), 25 -- 39. [45] A. Tikuisis, Nuclear dimension, Z-stability, and algebraic simplicity for stably projectionless C ∗-algebras, preprint, arXiv:1210.2237v2 [math.OA] to appear in Mathematische Annalen. Department of Mathematics and Informatics, Graduate school of Science, Chiba University,1-33 Yayoi-cho, Inage, Chiba, 263-8522, Japan E-mail address: [email protected]
1101.4221	1	1101	2011-01-21T20:06:23	Helson and subdiagonal operator algebras	[ "math.OA" ]	This article shows how the work of Henry Helson, especially the two papers of Helson and Lowdenslager, came to influence the development of the theory of non self adjoint operator algebras acting on Hilbert space.	math.OA	math	HELSON AND SUBDIAGONAL OPERATOR ALGEBRAS WILLIAM ARVESON 1. Commutative origins Henry Helson is known for his work in harmonic analysis, function theory, invariant subspaces and related areas of commutative functional analysis. I don't know the extent to which Henry realized, however, that some of his early work inspired significant developments in noncommutative directions - in the area of non-self adjoint operator algebras. Some of the most definitive results were obtained quite recently. I think he would have been pleased by that - while vigorously disclaiming any credit. But surely credit is due; and in this note I will discuss how his ideas contributed to the noncommutative world of operator algebras. It was my good fortune to be a graduate student at UCLA in the early 1960s, when the place was buzzing with exciting new ideas that had grown out of the merger of classical function theory and the more abstract theory of commutative Banach algebras as developed by Gelfand, Naimark, Raikov, Silov and others. At the same time, the emerging theory of von Neumann algebras and C ∗-algebras was undergoing rapid and exciting development of its own. One of the directions of that noncommutative development - though it went unrecognized for many years - was the role of ergodic theory in the structure of von Neumann algebras that was pioneered by Henry Dye [Dye59], [Dye63]. That Henry would become my thesis advisor. I won't say more about the remarkable development of noncommutative ergodic theory that is evolving even today since it is peripheral to what I want to say here. I do want to describe the development of a class of non-self-adjoint operator algebras that relates to analytic function theory, prediction theory and invariant subspaces: Subdiagonal operator algebras. It is rare to run across a reference to Norbert Wiener's book on prediction theory [Wie57] in the mathematical literature. That may be partly because the book is directed toward an engineering audience, and partly because it was buried as a classified document during the war years. Like all of Wiener's books, it is remarkable and fascinating, but not an easy read for students. It was inspirational for me, and was the source from which I had learned the rudiments of prediction theory that I brought with me to UCLA as a graduate student. Wiener was my first mathematical hero. Date: 25 October, 2010. 1 2 WILLIAM ARVESON Dirichlet algebras are a broad class of function algebras that originated in efforts to understand the disk algebra A ⊆ C(T) of continuous complex- valued functions on the unit circle whose negative Fourier coefficients vanish. Several paths through harmonic analysis or complex function theory or pre- diction theory lead naturally to this function algebra. I remind the reader that a Dirichlet algebra is a unital subalgebra A ⊆ C(X) (X being a com- pact Hausdorff space) with the property that A + A∗ = {f + ¯g : f, g ∈ A} is sup-norm-dense in C(X); equivalently, the real parts of the functions in A are dense in the space of real valued continuous functions. One cannot overestimate the influence of the two papers of Helson and Lowdenslager ([HL58], [HL61]) in abstract function theory and especially Dirichlet alge- bras. Their main results are beautifully summarized in Chapter 4 of Ken Hoffman's book [Hof62]. Along with a given Dirichlet algebra A ⊆ C(X), one is frequently pre- sented with a distinguished complex homomorphism φ : A → C and because A + A∗ is dense in C(X), one finds that there is a unique probability measure µ on X (of course I really mean unique regular Borel probability measure) that represents φ in the sense that (1.1) φ(f ) = ZX f dµ, f ∈ A. Here we are more concerned with the closely related notion of weak∗- Dirichlet algebra A ⊆ L∞(X, µ), in which uniform density of A + A∗ in C(X) is weakened to the requirement that A + A∗ be dense in L∞(X, µ) relative to the weak∗-topology of L∞. Of course we continue to require that the linear functional (1.1) should be multiplicative on A. 2. going noncommutative von Neumann algebras and C ∗-algebras of operators on a Hilbert space H are self-adjoint -- closed under the ∗-operation of B(H). But most operator algebras do not have that symmetry; and for non-self-adjoint algebras, there was little theory and few general principles in the early 1960s beyond the Kadison-Singer paper [KS60] on triangular operator algebras (Ringrose's work on nest algebras was not to appear until several years later). While trolling the waters for a thesis topic, I was struck by the fact that so much of prediction theory and analytic function theory had been captured by Helson and Lowdenslager, while at the same time I could see diverse examples of operator algebras that seemed to satisfy noncommutative vari- ations of the axioms for weak∗-Dirichlet algebras. There had to be a way to put it all together in an appropriate noncommutative context that would retain the essence of prediction theory and contain important examples of operator algebras. I worked on that idea for a year or two and produced a Ph.D. thesis in 1964 -- which evolved into a more definitive paper [Arv67]. HELSON AND SUBDIAGONAL OPERATOR ALGEBRAS 3 At the time I wanted to call these algebras triangular; but Kadison and Singer had already taken the term for their algebras [KS60]. Instead, these later algebras became known as subdiagonal operator algebras. Here are the axioms for a (concretely acting) subdiagonal algebra of oper- ators in B(H). It is a pair (A, φ) consisting of a subalgebra A of B(H) that contains the identity operator, is closed in the weak∗-topology of B(H), all of which satisfy SD1: A + A∗ is weak∗-dense in the von Neumann algebra M it generates. SD2: φ is a conditional expectation, mapping M onto the von Neumann subalgebra A ∩ A∗. SD3: φ(AB) = φ(A)φ(B), for all A, B ∈ A. What [SD2] means is that φ should be an idempotent linear map from M onto A ∩ A∗, that carries positive operators to positive operators, is continuous with respect to the weak∗-topology, and is faithful in the sense that for every positive operator X ∈ M, φ(X) = 0 =⇒ X = 0. We also point out that these axioms differ slightly from the original axioms of [Arv67], but are equivalent when the algebras are weak∗-closed. Examples of subdiagonal algebras: (1) The pair (A, φ), A being the algebra of all lower triangular n × n matrices, A ∩ A∗ is the algebra of diagonal matrices, and φ : Mn → A ∩ A∗ is the map that replaces a matrix with its diagonal part. (2) Let G be a countable discrete group which can be totally ordered by a relation ≤ satisfying a ≤ b =⇒ xa ≤ xb for all x ∈ G. There are many such groups, including finitely generated free groups (commutative or noncommutative). Fix such an order ≤ on G and let x 7→ ℓx be the natural (left regular) unitary representation of G on its intrinsic Hilbert space ℓ2(G), let M be the weak∗-closed linear span of all operators of the form ℓx, x ∈ G, and let A be the weak∗- closed linear span of operators of the form ℓx, x ≥ e, e denoting the identity element of G. Finally, let φ be the state of M defined by φ(X) = hXξ, ξi, X ∈ M, ξ = χe. If we view φ as a conditional expectation from M to the algebra of scalar multiples of the identity operator by way of X 7→ φ(X)1, then we obtain a subdiagonal algebra of operators (A, φ). (3) There are natural examples of subdiagonal algebras in II1 factors M that are based on ergodic measure preserving transformations that will be familiar to operator algebraists (see [Arv67]). In order to formulate the most important connections with function theory and prediction theory, one requires an additional property called finiteness in [Arv67]: there should be a distinguished tracial state τ on the von Neumann algebra M generated by A that preserves φ in the sense that τ ◦ φ = τ . Perhaps we should indicate the choice of τ by writing (A, φ, τ ) rather than (A, φ), but we shall economize on notation by not doing so. 4 WILLIAM ARVESON Recall that the simplest form of Jensen's inequality makes the following log f is integrable assertion about functions f 6= 0 in the disk algebra: around the unit circle, and the geometric mean of f satisfies (2.1) 1 2π ZT f (eiθ) dθ ≤ exp 1 2π ZT log f (eiθ) dθ. In order to formulate this property for subdiagonal operator algebras we require the determinant function of Fuglede and Kadison [FK52] - defined as follows for invertible operators X in M: ∆(X) = exp τ (log X), X denoting the positive square root of X ∗X. There is a natural way to extend the definition of ∆ to arbitrary (noninvertible) operators in M. For example, when M is the algebra of n × n complex matrices and τ is the tracial state, ∆(X) turns out to be the positive nth root of det X. Corresponding to (2.1), we will say that a finite subdiagonal algebra (A, φ) with tracial state τ satisfies Jensen's inequality if (2.2) ∆(φ(A)) ≤ ∆(A), A ∈ A, and we say that (A, φ) satisfies Jensen's formula if (2.3) ∆(φ(A)) = ∆(A), A ∈ A ∩ A−1. It is not hard to show that (2.2) =⇒ (2.3). Finally, the connection with prediction theory is made by reformulating a classical theorem of Szego, one version of which can be stated as follows: For every positive function w ∈ L1(T, dθ) one has inf f ZT 1 + f 2w dθ = expZT log w dθ, f ranging over trigonometric polynomials of the form a1eiθ + · · · + aneinθ. In the noncommutative setting, there is a natural way to extend the definition of determinant to weak∗-continuous positive linear functionals ρ on M, and the proper replacement for Szego's theorem turns out to be the following somewhat peculiar statement: For every weak∗-continuous state ρ on M, (2.4) inf ρ(D + A2) = ∆(ρ), the infimum taken over D ∈ A∩A∗ and A ∈ A with φ(A) = 0 and ∆(D) ≥ 1. In the 1960s, there were several important examples for which I could prove properties (2.2), (2.3) and (2.4); but I was unable to establish them in general. The paper [Arv67] contains the results of that effort. Among other things, it was shown that every subdiagonal algebra is contained in a unique maximal one, and that maximal subdiagonal algebras admit factorization: Every invertible positive operator in M has the form X = A∗A for some A ∈ A ∩ A−1. Factorization was then used to show the equivalence of these three properties for arbitrary maximal subdiagonal algebras. HELSON AND SUBDIAGONAL OPERATOR ALGEBRAS 5 3. Resurrection and Resurgence I don't have to say precisely what maximality means because, in an im- portant development twenty years later, Ruy Exel [Exe88] showed that the concept is unnecessary by proving the following theorem: Every (necessarily weak∗-closed) subdiagonal algebra is maximal. Thus, factorization holds in general and the three properties (2.2), (2.3), (2.4) are always equivalent. Encouraging as Exel's result was, the theory remained unfinished because no proof existed that Jensen's inequality, for example, was true in general. Twenty more years were to pass before the mystery was lifted. In penetrating work of Louis Labuschagne and David Blecher [Lab05], [BL08], [BL07a], [BL07b] it was shown that, not only are the three desired properties true in general, but virtually all of the classical theory of weak∗-Dirichlet function algebras generalizes appropriately to subdiagonal operator algebras. I hope I have persuaded the reader that there is an evolutionary path from the original ideas of Helson and Lowdenslager, through 40 years of sporadic progress, to a finished and elegant theory of noncommutative oper- ator algebras that embodies a remarkable blend of complex function theory, prediction theory, and invariant subspaces. References [Arv67] W. Arveson. Analyticity in operator algebras. Amer. J. Math., 89(3):578 -- 642, July 1967. [BL07a] D. Blecher and L. Labuschagne. Noncommutative function theory and unique extensions. Studia Math., 178(2):177 -- 195, 2007. math.OA/0603437. [BL07b] D. Blecher and L. Labuschagne. von Neumann algebraic H p theory. In Function spaces, volume 435 of Contemp. Math., pages 89 -- 114. Amer. Math. Soc., 2007. math.OA/0611879. [BL08] D. Blecher and L. Labuschagne. A Beurling theorem for noncommutative Lp. J. Oper. Th., 59:29 -- 51, 2008. math.OA/0510358. [Dye59] H. A. Dye. On groups of measure preserving transformations, I. Amer. J. Math., 81(1):119 -- 159, 1959. [Dye63] H. A. Dye. On groups of measure preserving transformations, II. Amer. J. Math., 85:551 -- 576, 1963. [Exe88] R. Exel. Maximal subdiagonal algebras. Amer. J. Math., 110:775 -- 782, 1988. [FK52] B. Fuglede and R. V. Kadison. Determinant theory in finite factors. Ann. Math., 55:520 -- 530, 1952. [HL58] H. Helson and D. Lowdenslager. Prediction theory and Fourier series in several variables. Acta Math., 99:165 -- 202, 1958. [HL61] H. Helson and D. Lowdenslager. Prediction theory and Fourier series in several variables II. Acta Math., 106:175 -- 213, 1961. [Hof62] K. Hoffman. Banach Spaces of Analytic Functions. Prentice-Hall, Englewood Cliffs, 1962. [KS60] R. V. Kadison and I. M. Singer. Triangular operator algebras. Amer. J. Math., 82:227 -- 259, 1960. [Lab05] L. Labuschagne. A noncommutative Szego theorem. Proc. Amer. Math. Soc., 133:3643 -- 3646, 2005. [Wie57] N. Wiener. Extrapolation, Interpolation, and Smoothing of Stationary Time Se- ries. Technology press of M.I.T. John Wiley and Sons, New York, 1957.
1804.05701	2	1804	2018-05-11T15:00:33	The Jordan lattice completion and a note on injective envelopes and von Neumann algebras	[ "math.OA", "math.FA" ]	The article associates two fundamental lattice constructions with each regular unital real ordered Banach space (function system). These are used to establish certain results in the theory of operator algebras, specifically relating the injective envelope of a separable C-algebra with its enveloping von Neumann algebra in a given faithful separable representation. The last section investigates on lattices of projections arising in injective C-algebras and von Neumann algebras and certain nonlinear maps sending projections to projections which are essentially determined by their values on positive projections.	math.OA	math	THE JORDAN LATTICE COMPLETION AND A NOTE ON INJECTIVE ENVELOPES AND VON NEUMANN ALGEBRAS U. HAAG In memory of my parents Kaethe and Rudolf Abstract. The article exhibits certain relations between the injective enve- lope I(A) of a C ∗-algebra A and the von Neumann algebra generated by a representation λ of A provided it is injective. In the first section two canoni- cal lattice constructions associated with any function system X (= underlying real ordered Banach space of an operator system) are defined and their gen- eral properties and interrelations are investigated. The first corresponds to the (positive) injective envelope of X which is denoted L1(X) and is a monotone complete regular unital Banach lattice (= injective abelian C ∗-algebra). It is the unique minimal complete Banach lattice containing X . The second is a somewhat maximal reasonable enveloping lattice construction which is called the Jordan lattice associated with X denoted L(X) . Both constructions are used to prove the main Theorem of the second section relating the injective envelope of a separable C ∗-algebra with its enveloping von Neumann algebra in a given faithful ∗-representation. The last (third) section deals with lattices of projections of specific (noncommutative) C ∗-algebras like von Neumann al- gebras and injective C ∗-algebras. The notion of P-map is defined and certain properties of these maps are exhibited. 0. Introduction. The following text is fairly selfcontained assuming only some general knowledge in the fields of operator theory, C∗-algebras and (real ordered) Banach spaces, so it should be accessible also to the interested reader who is not an outright expert in these subjects. Specific results which are used without proof may be found in gen- eral textbooks available as cited in the references. The author believes that many of the results and methods presented here are completely new opening perspectives for many different future applications beyond those given in the paper. This ac- counts in particular to the use of enveloping lattice constructions and consideration of general monotonous maps (as opposed to positive linear maps) which provide a powerful tool for many problems in a lattice environment. In a more limited sense this might also be the case for the (only partially monotonous) P-maps described in the Appendix although any computations in a noncommutative projection lattice are much more delicate than in the setting of a (linear) function lattice. The reader should be warned that the logical dependence of the presented results is not always in chronological order, but that the proofs of certain assertions might depend on some results which are proved afterwards so that the reader is requested to skip Date: September 19, 2018 Contact:[email protected]. 1 2 U. HAAG to the corresponding passages when needed. This is done in favour of grouping together statements which are closely related in content rather than obeying an accurate choronlogy of logical dependence. We now proceed to give a brief account of the contents of this paper. In the first section the two fundamental lattice constructions are presented. This is done in the context of real (dually) ordered Banach spaces (see below for the precise definition). It is shown that such a space V admits an order isomorphic embedding into a C∗-algebra (of continuous functions on a compact space) so that if V is unital it is order isomorphic with a function system (in the sense of [7]). In the latter case the prototype for such an embedding is given by the lattice construction L1(V) which turns out as a monotone complete abelian C∗-algebra and is equal to the (positive) injective envelope of V , i.e. for any embedding of unital real ordered Banach spaces W ⊆ Z there exists a positive linear map Z −→ L1(V) extending a given positive linear map W ֒→ L1(V) . Moreover L1(V) posseses the following rigidity property: any monotonous extension γ : L1(V) −→ L1(V) of the identity map of V is the identity map. Using this construction it is shown that any unital complete real ordered Banach lattice is order isomorphic with a monotone complete (hence injective) abelian C∗-algebra. If L1(V) is the minimal (complete) Banach lattice containing V as a subspace, then the second lattice construction L(V) gives the somewhat maximal Banach lattice generated by V . Contrary to the prior case it is not generally monotone complete and in order to obtain a monotone complete lattice one defines the Jordan lattice completion as It is then shown in Proposition 1 that there exists a unique surjective monotonous map Lq(V) = L1(cid:0)L(V)(cid:1) . π : Lq(V) ։ L1(V) extending the identity map of V . It turns out that π is a ∗-homomorphism of the underlying abelian C∗-algebras. In case that A = V is a (noncommutative) C∗-algebra there is a so called Jordan squaring operation defined on Lq(A) which is closely related to the C∗-square in A and another partially inverse Jordan square- root operation which extends the squareroot operation on A . These are investigated in detail and the Jordan squaring operation is then related to the (commutative) C∗-square in Lq(A) and certain of its quotients which results make up a major part of Theorem 1. In addition one obtains for any embedding of function systems X ⊆ Y a canonical normal embedding of Jordan lattices Λ : Lq(X) ֒→ Lq(Y) . The fact that Λ is normal (and ∗-homomorphic) is quite surprising and is proved in Theorem 1. In the course of its proof two canonical monotonous retractions r , r : Lq(Y) ։ Lq(X) are employed which themselves are seen to posess very specific properties and these maps remain essential for the rest of the paper (for example in the proofs of Proposi- tion 3 and Theorem 2). A somewhat isolated result in the first section is Proposition 2 which states that any completely positive complete quotient map q : X ։ Mn(C) of an operator system onto a matrix algebra which is an operator order epimorphism admits approximate completely positive splittings which are completely bounded by n . This result is however not used in the course of the paper and is included for its own interest. THE JORDAN LATTICE COMPLETION 3 The second section is much shorter than the first and contains in Theorem 2 what might be called the main result of this article as far as operator algebras are con- cerned. It exhibits a close relation between the injective envelope I(A) of a sep- arable C∗-algebra A and its enveloping von Neumann algebra R in some given faithful ∗-representation on a separable Hilbert space provided it is injective. In addition the Up-Down-property for I(A) is proved analogous to the Up-Down- property of R . In the Corollary of Theorem 2 (which is not really a Corollary but a Theorem in its own right) the section also contains another fundamental result which may be seen as a starting point of the whole paper as far as injective abelian C∗-algebras are concerned. It states that such an algebra is injective not only for positive linear maps, but also for ∗-homomorphisms (of abelian C∗-algebras) and linear lattice maps. This result is used throughout the paper. Finally the last section is supplemented as an Appendix because its content is not logically related to the first two sections, being related only in spirit. It can be considered as an autochtone essay whose perception is independent of the prior article and is of interest for its own sake. The section investigates on conditions needed for a C∗-algebra in order that its subset of positive projections is a lattice (resp. complete lattice). The notion of AΣ-algebra is defined together with the notion of P-algebra, and the notion of P-maps of P-algebras. These maps are essentially determined by their values on projections and they send projections to projections. Theorem P gives an account of the available results concerning these concepts. In particular it is shown that von Neumann algebras as well as injective C∗-algebras are AΣ-algebras whence their subset of positive projections forms a complete lattice. Some other interesting results are added which we do not spell out here in detail. 1. The Jordan lattice completion. In this paper we are primarily interested in C∗-algebras. Some of the following constructions however can be done in the more general setting of ordered real Banach spaces. By this we mean a real Banach space V with a distinguished closed convex cone of positive elements V+ such that V+ ∩ −V+ = {0} and V = V+ − V+ . The ordered Banach space is unital if the supremum of the unit ball, denoted 1 , exists in V and is of of norm one, i.e. 1 − x ≥ 0 whenever kxk ≤ 1 . The usual definition of a unital ordered Banach space includes a regularity condition, i.e. 0 ≤ x ≤ y implies kxk ≤ kyk or equivalently that every element dominated by 1 is of norm less or equal to one, cf. [3]. We will refer to such a space as a regular (unital) ordered Banach space. A (linear) positive injection φ : V ֒→ W of ordered real Banach spaces is termed an order isomorphic embedding if φ(x) ≥ 0 implies x ≥ 0 and φ is bounded from below. The real ordered Banach space V is called dually ordered if its dual is again an ordered real Banach space with positive cone V∗ + given by the functionals attaining only nonnegative values on V+ and if V+ = V∗∗ + ∩ V . This is equivalent to the property that V admits an order isomorphic embedding into a C∗-algebra. If such an embedding exists one sees by restriction of the positive functionals of A to the image of V that any given element of V∗ , which is bounded also on the image of V in A because the inverse map of the embedding is bounded, can be extended to a continuous functional on Asa from the Hahn-Banach theorem, which then decomposes into 4 U. HAAG the difference of two positive functionals. Since A+ = A∩ A∗∗ + also V+ = V∩ V∗∗ + follows. On the other hand if the two conditions are satisfied then the first of the two lattice constructions below gives an order isomorphic embedding of V into a regular complete Banach lattice L1(V) which, in turn may be embedded into a regular unital ordered Banach space itself, and thus admits an order isomorphic embedding into the commutative C∗-algebra of continuous functions on its state space (endowed with the relative w∗-topology, cf. [3], chap. 2). Let V be a dually ordered real Banach space. To each subset C ⊆ V+ of positive elements one attributes an element C in a real vector space L1(V) called the infimum of C . There are several natural equivalence relations available in order to build a linear space, we consider the following: C∼1C′ if and only if for each element a ∈ V such that a ≤ c for every c ∈ C this implies a ≤ c′ for all c′ ∈ C′ and vice versa. For each set C let Cc = {a ∈ V a ≤ C} denote the subset of elements all of which are smaller or equal than each element in C . Then putting (Cc)c = {c ∈ V+ c ≥ Cc} one obtains a maximal representative for the equivalence class of C . Given two elements C = inf {cλ}λ and D = inf {dµ}µ as above their sum is defined to be C + D = inf {cλ + dµ cλ ∈ C , dµ ∈ D}λ,µ and one checks that this definition is compatible with the equivalence relation hence well defined. One also has 2 C = {2cλ cλ ∈ C} ∼1 {cλ + cµ cλ , cµ ∈ C} = C + C since if a ∈ V is any element such that a ≤ 2cλ for all λ then also a ≤ cλ + cµ for all λ , µ . Therefore the scalar multiplication of elements C with positive rational numbers is well defined. Approximating a real positive number by positive rational numbers from below one sees that scalar multiplication extends to the (positive) reals. The convex cone of such elements generates a real linear space L1(V) by taking arbitrary differences A = C − D , subject to the equivalence relation generated by ∼1 and (C + E) − (D + E) ≃1 C − D with E ⊆ V+ a subset of positive elements, and an order on this space is obtained by A ≥ B ⇐⇒ A − B ≃1 C ≥ 0 . One finds that for positive elements the equivalence relation ≃1 which by definition is a stabilized version of ∼1 , i.e. C ≃1 D iff there exists a positive element E such that C + E∼1D + E , in fact reduces to ∼1 so this equivalence relation has cancellation. Namely, suppose C + E∼1D + E . Let e0 ∈ E be a fixed element and denoting Ec = {b ∈ V b ≤ E} the set of elements all of which are smaller or equal than every element in E consider the positive subset e0 − Ec = {e0 − b b ∈ Ec} . Then since taking sums is compatible with ∼1 one gets C + E + (e0 − Ec)∼1D + E + (e0 − Ec) . On the other hand E + (e0 − Ec)∼1{e0} since a ≤ e + e0 − b for all e ∈ E , b ∈ Ec implies b + (a − e0) ∈ Ec and by induction b + n(a − e0) ∈ Ec for all b ∈ Ec , n ∈ N . This then implies that a− e0 ≤ 0 since otherwise there would exist a positive functional φ ∈ V+ with φ(a − e0) > 0 so that supb∈Ec {φ(b)} = +∞ which is impossible since Ec is bounded above. Therefore a ≤ e0 follows and the reverse implication is trivial. Thus C +{e0}∼1D +{e0} and since {e0} consists of a single element one concludes that C∼1D . One then sees by a similar argument that a general element A = C − D is equal to the infimum of the set {c − a c ∈ C , a ∈ Dc} . Conversely any set of selfadjoint elements which is bounded below uniquely determines an element in L1(V) by taking its infimum, since each subset bounded THE JORDAN LATTICE COMPLETION 5 from below is given as the difference of a positive subset and a single positive element of V (viewed as a positive subset). In particular C ≥ D ⇐⇒ Dc ⊆ Cc . For an arbitrary subset {Cλ}λ of positive elements its infimum inf λ Cλ which is the largest element smaller or equal than each Cλ is well defined and is given by the element corresponding to the union ∪λ Cλ . By shifting with a suitable positive element (resp. positive scalar if V is unital) one may equally define an infimum for every set of general (selfadjoint) elements {Aλ}λ which is bounded below and correspondingly, by the symmetry A 7→ −A a supremum for every set of elements bounded above. For example taking each Cλ = {cλ} to consist of a single positive element one gets inf λ {Cλ} = inf C = sup Cc with C = {cλ}λ = ∪Cλ . In particular one has the lattice operations A ∧ B , A ∨ B denoting the (unique !) maximal element smaller or equal to A and B , resp. the unique minimal element larger or equal to both A and B . If V is not unital and A = C − D , B = E − F these operations may be defined as A ∧ B = inf(cid:8)C + F , E + D(cid:9) − (cid:0)D + F(cid:1) , A ∨ B = sup(cid:8)C + F , E + D(cid:9) − (cid:0)D + F(cid:1) . Taking B = 0 the unique minimal positive decomposition of A is given by A = A+ − A− = (A ∨ 0) + (A ∧ 0) . Indeed, suppose given two different positive decompositions A = C − D = C′ − D′ we claim that A = (C ∧C′) − (D ∧D′) . To see this it is sufficient by the symmetry A 7→ −A to prove C − D = A ≤ (C ∧ C′) − (D ∧ D′) ⇐⇒ C + (D ∧ D′) ≤ (C ∧ C′) + D using the relation C + D′∼1C′ + D . Let a ∈ Asa be an element smaller or equal than each element in C + (D∪D′) and given an arbitrary element b ∈ (C ∪C′) +D . Then either b ∈ C +D in which case a ≤ b follows trivially, or b ∈ C′ +D ∼ C +D′ which again implies a ≤ b . Using induction the minimal positive decomposition of A is given by A = A+ − A− = inf λ {Cλ A = Cλ − Dλ} − inf λ {Dλ A = Cλ − Dλ} . Then A+ ≥ A ∨ 0 with A = (A ∨ 0) − ((A ∨ 0) − A) a positive decomposition of A . From uniqueness one must have A+ = A ∨ 0 and hence A− = −(A ∧ 0) . For C ≥ 0 a natural norm (satisfying the triangle inequality) is given by kCk = inf {kdk d ∈ D , D∼1C} = inf {kck c ∈ (Cc)c} . This norm is extended to general elements A = C − D by defining kAk = max(cid:8)kA+k , kA−k(cid:9) ≥ 0 . The triangle inequality is readily checked since (A + B)± ≤ A± + B± . Define L1(V) to be the resulting normed linear space which clearly is a complete vector lattice. Being monotone complete it is a Banach space, hence a regular unital 6 U. HAAG Banach lattice. Namely if (cid:8)An(cid:9) ⊆ L1(V) is a Cauchy sequence then the sequence converges in norm towards the element A = lim inf n An = sup n (cid:8) inf m≥n(cid:8)Am(cid:9)(cid:9) . Thus L1(V) is a regular complete Banach lattice containing a canonical subspace if V is a (unital) C∗-algebra (or order isomorphic with V . In many cases, e.g. for that matter a regular unital ordered Banach space = function system), the embedding V ֒→ L1(V) is isometric with respect to the original norm of V (see the analogous argument below for L(A) ). In any case if V is a dually ordered Banach space the embedding V ֒→ L1(V) is order isomorphic. To see this consider a general selfadjoint element x ∈ V and let x = inf {x} = sup{x} be its image in L1(V) . Then its minimal positive decomposition is given by x = x+ − x− with x± = inf { c 0 ≤ c , ±x ≤ c} so that x− = 0 if and only if x ≥ 0 . Since the dual V∗ is positively generated,there exists α ≥ 1 such that V is α-normal which implies kxk = max{kx+k , kx−k} ≥ 2 + α−1 kxk ≥ α−1 α−1 3 kxk showing that the embedding V ֒→ L1(V) is bounded below. Clearly, L1(V) is If V is unital L1(V) is a unital regular, i.e. 0 ≤ x ≤ y implies kxk ≤ kyk . complete regular Banach lattice, and as such isometrically order isomorphic with an (injective) commutative C∗-algebra by the Corollary below. In case of a general dually ordered Banach space one may adjoin a unit element by considering the linear space eL1 = L1(V) + R 1 , where 1 = sup{x ∈ L1(V)kxk ≤ 1} = sup{x ∈ L1(V) x ≥ 0 , kxk ≤ 1} and ordering given by the positive cone eL1(V)+ = (cid:8)p = α a + β (1 − b) (cid:12)(cid:12) α , β ≥ 0 , 0 ≤ a , b ≤ 1(cid:9) . A norm is imposed on positive elements by the formula kpk1 = sup(cid:8)kc+k (cid:12)(cid:12) c ≤ p , c ∈ L1(V)(cid:9) which is well defined due to L1(V) being a regular lattice and satisfies the triangle inequality. Check that p ≥ c implies p ≥ c+ for c ∈ L1(V) . This follows since c+∧c− = 0 implies (b+c+)∧c− = (b∧c−)+c+ and c = d+e implies c+ ≤ d++e+ . For p ∈ L1(V) one recovers the original norm. For general selfadjoint elements define kxk1 = inf(cid:8)max{kpk1 , kqk1} (cid:12)(cid:12) x = p − q , p, q ≥ 0(cid:9) and check that the axioms for a norm are satisfied. In particular if x = p − q ∈ L1(V) with p = a + λ(1 − b) , q = c + µ(1 − d) , a , b , c , d ∈ L1,+ one has λ = µ and the triangle inequality gives kxk1 ≥ kxk since kpk1 ≥ kx+k , kqk1 ≥ kx−k =⇒ kxk1 ≥ kx+k + kx−k = kxk so the embedding L1(V) ֒→eL1(V) is isometric and order isomorphic proving that V admits an order isomorphic embedding into a regular unital ordered Banach space and a forteriori into a (commutative) C∗-algebra. The details of this ar- gument are left to the reader. In general the construction is not functorial with respect to arbitrary positive linear maps, meaning that in general there is no nat- ural positive linear extension L1(φ) : L1(V) → L1(W) of a given positive linear map φ : V → W . If however π : V ։ W is a complete order epimorphism so that the image set of the lower complement of any complemented positive subset is equal to the complement of the image set of C in W there C ⊆ V+ , C =(cid:0)Cc(cid:1)c THE JORDAN LATTICE COMPLETION 7 exists a unique monotonous extension L1(φ) : L1(cid:0)V(cid:1) ։ L1(cid:0)W(cid:1) of φ which must be linear if φ is linear (see below). One may wonder if there are some simpler notions ensuring functoriality. One notes the following: if π : V ։ W is a sim- ple order epimorphism, i.e. if π(V+) = W+ and in addition the kernel of π is positively generated, then for any two elements x , y ∈ V+ any element b ∈ W with b ≤ π(x) , b ≤ π(y) lifts to an element a ∈ V with a ≤ x , a ≤ y . Since π is an order epimorphism there exist lifts a′ , a′′ of b with a′ ≤ x , a′′ ≤ y . Then a′ − a′′ = c − d with c , d ≥ 0 , c , d ∈ ker π , so that putting a = a′ − c = a′′ − d gives the result. This argument however only applies to finite subsets of V+ and fails for arbitrary complemented positive subsets. A necessary and sufficient con- dition that φ admits a unique monotonous extension is the following: given any complemented positive subset C ⊆ V+ then inf φ(cid:0)C(cid:1) must be larger or equal to sup φ(cid:0)Cc(cid:1) which is encoded in the relation φ(cid:0)Cc(cid:1) ⊆ φ(cid:0)C(cid:1)c . Then uniqueness of any extension can be encoded by the condition φ(cid:0)Cc(cid:1)c alently sup φ(cid:0)C(cid:1)c ≤ inf φ(cid:0)Cc(cid:1)c or equiv- since the reverse implication is trivial. Indeed if this relation holds for all complemented subsets then any monotonous extension is uniquely determined. Conversely suppose that the condition fails for some positive complemented subset C . For any positive element C ∈ L1(V) write C = inf C = inf(cid:8)c ∈ V(cid:12)(cid:12) c ≥ C(cid:9) = sup(cid:8)b ∈ V(cid:12)(cid:12) b ≤ C(cid:9) = sup Cc . Putting φ(cid:0)C(cid:1) = inf(cid:8)φ(C)(cid:9) and φ(cid:0)C(cid:1) = sup(cid:8)φ(Cc)(cid:9) gives two different monoto- nous extensions of φ to L1(V) . Since the functoriality of L1 for linear (positive) maps is so restricted one is forced or maybe seduced to consider a wider notion of functoriality. Definition. (a) A positive map m : V+ → W+ of two ordered Banach spaces is called monotonous iff = (cid:0)φ(cid:0)C(cid:1)c(cid:1)c It is called homogenous iff x ≤ y ⇒ m(x) ≤ m(y) . m(α x) = α m(x) , ∀ α ∈ R , x ∈ V . (b) A positive map c : V → W of two ordered Banach spaces is called convex if it is homogenous, monotonous and if the following condition holds ∀ x , y ≥ 0 . c(x + y) ≤ c(x) + c(y) , Similar c is called concave if it is monotonous, homogenous and if c(x + y) ≥ c(x) + c(y) , ∀ x , y ≥ 0 . (c) A monotonous map of unital dually ordered Banach spaces φ : V ։ W is called a complete order epimorphism iff for any complemented subset C ⊆ V , i.e. C = (Cc)c the set φ(cid:0)Cc(cid:1) is normdense in φ(cid:0)C(cid:1)c . An amiable aspect of a convex map c defined on a Banach lattice L and assuming c(x) = c(x+) − c(x−) with respect to the minimal positive decomposition of x and further assuming that the kernel ker c = {x ∈ L c(x) = 0} is positively generated, so that x ∈ ker c implies x± ∈ ker c , is that the kernel must be a linear subspace, and c is linear for addition by elements from ker c since for every positive z ∈ L+ , d ∈ ker c , d ≥ 0 one has c(z) ≤ c(z + d) ≤ c(z) + c(d) = c(z) . 8 U. HAAG We will later see some interesting examples of convex and concave maps. A real ordered Banach space V whose positive cone V+ admits an order unit e ∈ V+ such that the relation x ≤ e holds for any x ∈ B1(V) in the unit ball is called a preunital ordered Banach space. Corollary. A preunital dually ordered real Banach space V is injective in the category of preunital ordered Banach spaces and positive linear maps if and only if it is a complete Banach lattice. Also, any regular unital complete Banach lattice is isometrically order isomorphic to an injective commutative C∗-algebra and any preunital complete Banach lattice is order isomorphic to a regular unital complete Banach lattice, hence to a commutative C∗-algebra. Proof. Suppose that V is a complete Banach lattice and given a (preunital) in- clusion of preunital ordered real Banach spaces X ⊆ Y together with a positive linear map φ : X → V . Pick an arbitrary element y0 ∈ Y+\X+ and define φ0 : X + Ry0 → V by linear extension of φ0X = φ and φ0(y0) = sup{φ(x) x ∈ X , x ≤ y0} . Since the inclusion X ֒→ Y is preunital the supremum is well defined. Suppose that y = x + αy0 ≥ 0 , α ∈ R , x ∈ X is a positive element in X + R y0 . If α ≤ 0 then On the other hand if α > 0 then φ0(y) = inf {φ(z) z ∈ X , z ≥ y ≥ 0} ≥ 0 . φ0(y) = sup{φ(z) z ∈ X , z ≤ y0} ≥ φ(0) = 0 . One proceeds by (transfinite) induction to extend φ to a positive map φ : Y → V . This proves that any complete Banach lattice is injective for positive linear maps of preunital ordered Banach spaces. For the converse assume that the preunital dually ordered Banach space V is injective for such maps. Then V admits an order isomorphic preunital embedding into some preunital complete Banach lattice L from the argument above and the identity map of V extends to a positive map ι : L → V . Since any subset C ⊆ V which is bounded below has an infimum x ∈ L the element ι(x) is an infimum for C in V by positivity of ι . Therefore any subset C ⊂ V which is bounded below has an infimum and V is a complete lattice. To prove the second statement one notes that if X is a regular unital complete Banach lattice, it is in particular a regular unital Banach space, so it admits a unital isometric and order isomorphic embedding into a unital commutative (injective) C∗-algebra A . By positive injectivity of X there exists a positive linear projection Φ : A → A with Φ(A) = X . Then the unital operator space X (with respect to the operator space structure determined by the embedding into A ) admits a structure as a commutative C∗-algebra (compare with [6], Theorem unique (!) 6.1.3). Now suppose that X is any preunital complete Banach lattice. As such it is certainly a dually ordered Banach space, and applying the functor L1 gives an order isomorphism with a regular preunital complete Banach lattice, i.e. 0 ≤ x ≤ y implies kxk ≤ kyk , and moreover the norm of a general element is given in terms of the minimal positive decomposition x = x+ − x− as kxk = max {kx+k , kx−k} . THE JORDAN LATTICE COMPLETION 9 Let 1 denote the supremum over the unit ball of the regular complete Banach lattice L1(X) and k1k ≥ 1 its norm. Define a new norm on X by kxk = inf(cid:8)α ≥ 0 (cid:12)(cid:12) x ≤ α1(cid:9) , x ≥ 0 and kxk = max{kx+k , kx−k} in the general case and check that the axioms of a norm are satisfied, defining a structure of a regular unital complete Banach lattice such that kxk ≤ kxk ≤ k1k kxk (cid:3) If V is unital we may call L1(V) the positive injective envelope of V since it is the smallest positively injective Banach space containing V (by an order isomor- phic embedding if not by isometric embedding) and has a corresponding rigidity property: every monotonous positive map m : L1(V) → L1(V) which restricts to the identity map of V is equal to the identity map. Remark. There are several generalizations of the L1-construction, one is given by the following scheme: a regular unital real ordered Banach cone is a closed unital affine subset C ⊆ V of a regular unital real ordered Banach space V , i.e. for all c , d ∈ C , α ∈ R+ and γ ∈ R one has c + d ∈ C , α c ∈ C , γ 1 ∈ C . The order relation on the set C is induced by the order on V . Then also the subset −C ⊆ V is a regular unital real ordered Banach cone, and one defines an equivalence relation for subsets C ⊆ C bounded below (resp. positive) by C ∼1 C′ ⇐⇒ Cc :=(cid:8)b ∈ −C(cid:12)(cid:12) b ≤ c , ∀c ∈ C(cid:9) =(cid:8)b ∈ −C(cid:12)(cid:12) b ≤ c′ , ∀c′ ∈ C′(cid:9) = C′ c . Dividing by the equivalence relation ∼1 yields a set L1(C) . As above one checks that addition of two elements as well as multiplication with positive scalars is well defined and that ∼1 has cancellation, i.e. C + E ∼1 D + E =⇒ C ∼1 D . Then also subtraction of two elements is defined by (cid:2)C(cid:3) −(cid:2)D(cid:3) = (cid:2)C(cid:3) + (cid:2)−Dc(cid:3) and one obtains the structure of a real ordered normed space by imposing the cor- responding norms and ordering as above which in addition is a monotone complete lattice. For C = V one obtains the original definition. Another generalization is to arbitrary partially ordered sets (cid:0)S,≥(cid:1) by which we understand ordered sets such that for each pair of elements x , x ∈ S there exist elements w , z ∈ S with w ≤ x , y ≤ z . Given a partially ordered set S let Cc(S) denote the set of complemented subsets C ⊆ S which are bounded below so there exists an element b ∈ S with b ≤ c for all c ∈ C and C is the upper complement of its lower complement Cc = {b ∈ S b ≤ c , ∀c ∈ C} . The subset Cc(S) is naturally ordered by inclusion, i.e. C ≥ D ⇐⇒ C ⊆ D . Also let Cc(S) denote the set of complemented subsets which are bounded above ordered by inclusion, so that there is a natural order isomorphism Cc(S) ≃ Cc(S) with C corresponding to its com- plement Cc . Note that an arbitrary intersection of complemented subsets bounded below is a complemented subset bounded below, and an arbitrary intersection of complemented subsets bounded above is a complemented subset bounded above if one agrees to letting the void set be both upper and lower complement of the whole 10 U. HAAG set S . Then the partially ordered set L1(S) = Cc(S) ≃ Cc(S) is a monotone complete lattice with respect to the operations C ∧ D = (cid:0)Cc ∩ Dc(cid:1)c , C ∨ D = C ∩ D , i.e. any subset of elements which is bounded below has an infimum and any subset bounded above has a supremum, and there is a natural order isomorphic inclusion S ⊆ L1(S) by s 7→ {s}c ≡ {s}c . Then S is called comparable iff there exist monotonous functions r+ : S −→ R , r− : S −→ R such that any subset C ⊆ S with r−(C) := inf {r−(c) c ∈ C} > −∞ is bounded below, and any subset B ⊆ S with r+(B) := sup{r+(b) b ∈ B} < ∞ is bounded above. Then if S is comparable the monotone complete lattice L1(S) is injective in the category of comparable partially ordered sets and r+-bounded (resp. r−- bounded) monotonous maps. As an example of a comparable partially ordered set consider any unital subset S ⊆ V in a regular unital real ordered Banach space meaning that α 1 ∈ S for any α ∈ R . Define r+(cid:0)x(cid:1) = inf(cid:8)α ∈ R(cid:12)(cid:12) α 1 ≥ x(cid:9) , r−(cid:0)x(cid:1) = sup(cid:8)β ∈ R(cid:12)(cid:12) β 1 ≤ x(cid:9) . We now consider a second equivalence relation by which subsets of positive ele- ments generate a linear lattice. In contrast with the previous construction this one is functorial for arbitrary positive maps. The construction can be done for an arbi- trary ordered Banach space as above, but since we want to impose some additional structure we assume from the start that the base space is a (unital) C∗-algebra A . We only note that if V is an arbitrary ordered real Banach space the positive linear map sending V onto its canonical image in the positively regular Banach lattice L(V) is order isomorphic if and only if V is dually ordered by an argu- ment as above. Thus the construction assigns to an arbitrary ordered real Banach space a canonically defined dually ordered Banach space together with a surjective positive linear map relating the two which is an order isomorphism only if V is dually ordered itself. Assume now that V = Asa is the real subspace of selfadjoint elements of a unital C∗-algebra A . Much of the alternative construction parallels the preceding one. To each subset C ⊆ A+ of positive elements one attributes an element C in a real vector space L(A) called the infimum of C . There is a simple equivalence relation C ∼ C′ if and only if each element c ∈ C is larger or equal to some element in the closed convex hull of C′ and vice versa. For each set C let Cc denote the set of positive elements each of which is larger than some element in the closed convex hull of C . Then Cc is the maximal representative in the simple equivalence class C of the set C . The equivalence class of C is then denoted C and sometimes abusively identified with the maximal representative Cc . A preorder on the set of equivalence classes is obtained by the corresponding onesided relation, i.e. C ≥ D iff Cc ⊆ Dc . Then one may define the upper com- plement of the virtual positive element P = C − D ≥ 0 to consist of all elements P c = {a ∈ A+ (a+D)c ⊆ Cc} which are larger or equal than P . This coincides for elements defined by single positive sets with the maximal representative set whence the notation but note that the positive element defined by the upper complement may be strictly larger then the difference C − D in L(A) as defined below. Given two elements C = inf {cλ}λ and D = inf {dµ}µ as above their sum is defined to THE JORDAN LATTICE COMPLETION 11 be C + D = inf {cλ + dµ cλ ∈ C , dµ ∈ D}λ,µ and one checks that this definition is compatible with the equivalence relation hence well defined. One also has 2 C = {2cλ cλ ∈ C} ∼ {cλ + cµ cλ , cµ ∈ C} = C + C since each cλ+cµ is a convex combination of the elements 2cλ , 2cµ . Therefore as in the preceding construction the scalar multiplication of elements C with positive real numbers is well defined. The convex cone of such elements generates a real linear space L(A) (the same notation will be used for the norm completion with respect to the norm defined below) by taking arbitrary differences A = C − D (in general the notation C is reserved for elements defined by a specific representative positive set C , it may happen in certain instances that C is also used to denote general selfadjoint elements in L(A) which meaning should be clear from the context), subject to the equivalence relation with E ⊆ A+ an arbitrary subset of positive elements, and an order on this space is obtained by (C + E) − (D + E) ≃ C − D A ≥ B ⇐⇒ A − B ≃ C − D ≥ 0 with Dc ⊆ Cc . For elements defined by positive sets the equivalence relation ≃ coincides with ∼ so this equivalence relation has cancellation. To see this let D , E be given such that C + E ∼ D + E . One can assume without loss of generality that C , D and E are maximal for ∼ so they are equal to their closed convex hull and the same is true for the sets C + E , D + E . We must show that D ⊆ C . Let d ∈ D , e ∈ E and ǫ > 0 be given. There exist elements c ∈ C , e′ ∈ E satisfying c + e′ ≤ d + e + ǫ1 . Then again there exist c′ ∈ C , e′′ ∈ E with c′ + e′′ ≤ d + e′ + ǫ1 and inductively sequences (c(k))k ⊆ C , (e(k))k ⊆ E with For given N ∈ N one gets on adding these relations c(k) + e(k+1) ≤ d + e(k) + ǫ1 . which implies (c(k) + e(k+1)) ≤ d + NXk=1 e(N +1) ≤ d + 1 N 1 N e(k) + ǫ1 e + ǫ1 . 1 N c(k) + 1 N 1 N NXk=1 NXk=1 Define the positive set E ′ to be the union of E and all elements of the form { 1 k e(k+1)} with respect to all elements d , e and ǫ > 0 as above. One concludes on letting N → ∞ that D + ǫ1 ⊆ C + E ′ ⊆ C and by symmetry one has C + ǫ1 ⊆ D . Since ǫ > 0 was arbitrary one concludes C ∼ D . For an arbitrary set {Cλ}λ of basic positive elements its infimum inf λ Cλ which is the largest element smaller or equal than each Cλ is well defined and is given by the element corresponding to the union ∪λ Cλ . By shifting with a fixed positive element C one may then define an infimum for subsets of general (selfadjoint) elements {Aλ}λ which are bounded below and are brought to the form above by the shift Aλ 7→ Aλ +C . This applies in particular to finite subsets {A1 = C1 − D1 , ··· , An = Cn − Dn} . By shifting each k=1 Dk , then taking the infimum of the corresponding basic positive Ak with Pn denoting the (unique !) maximal element smaller or equal to A and B , resp. the unique minimal element larger or equal to both A and B . Taking B = 0 the unique minimal positive decomposition of A is given by A = A+ − A− = (A ∨ 0) + (A ∧ 0) . If A = P − Q is any positive decomposition of A then P ≥ A ∨ 0 with A = (A∨0) − ((A∨0)−A) a positive decomposition of A which therefore is necessarily minimal. From uniqueness one gets A+ = A∨ 0 and hence A− = −(A∧ 0) . There also exists a minimal positive decomposition by basic positive elements, i.e. infima of positive sets. Indeed, suppose given two different positive decompositions by basic positive elements we claim that A = (C ∧C′) − (D ∧D′) . To see this it is sufficient by the symmetry A 7→ −A to prove A = C − D = C′ − D′ C − D = A ≤ (C ∧ C′) − (D ∧ D′) ⇐⇒ C + (D ∧ D′) ≤ (C ∧ C′) + D 12 U. HAAG elements and shifting back the result by −Pn the set {Ak} . In particular one has the lattice operations k=1 Dk one obtains the infimum of A ∧ B , A ∨ B ǫ > 0 a convex combination b = Pk λkbk ∈ C + D′ ⊆ C + (D ∪ D′) such that using the relation C + D′ ∼ C′ + D . Let a = c + d ∈ (C ∪ C′) + D be given. If c ∈ C then a ∈ C + (D ∪ D′) follows trivially. If c ∈ C′ there exists for each a ≥ b + ǫ1 whence the result. It follows that any selfadjoint element has a unique minimal decomposition as a difference of basic positive elements. In particular any positive element P = C − D with C ≥ D has a unique minimal representation by such elements. A (general) basic element of L(A) is an element which is the infimum of a given subset of selfadjoint elements of A which is bounded below. Correspondingly, an element will be called antibasic iff it is the supremum of a given subset of selfadjoint elements bounded above. A positive antibasic element will sometimes be denoted P to indicate that it is the supremum of its lower complement Pc = {a ∈ Asa a ≤ P} . Then any element A = C − D = (cid:0)R − D(cid:1) − (cid:0)R − C(cid:1) = P − Q can be decomposed as a difference of antibasic positive elements as well where R ≥ 0 is a positive scalar exceeding both kCk and kDk . For given P ≥ 0 consider the set of all positive elements a ≥ 0 such that inf {a} ≥ P . This set P c is a maximal set for ∼ called the upper complement of P which constitutes a basic positive element P c ≥ P abusively identified with the corresponding maximal subset. By definition it is the smallest such element having this property and agrees with P if and only if P is basic itself. By symmetry the supremum of the lower complement Pc = {a ∈ Asa a ≤ P} constitutes an element Pc ≤ P of L(A) which is the largest antibasic element with this property. One may also consider the double lower complement Pcc = (Pc)c , which is basic, and the antibasic double upper complement respectively given by P cc = (P c)c . Then the triple upper and lower complement are defined recursively by P ccc = (P cc)c = (P c)cc and Pccc = (Pcc)c = (Pc)cc respectively. One has the relations Pc ≤ P ≤ P c , Pc ≤ Pccc = P cc ≤ P ccc = Pcc ≤ P c THE JORDAN LATTICE COMPLETION 13 but P need not to be related by order to neither of the elements P cc , Pcc . If P is basic then taking lower complements stabilizes already at the first step, i.e. Pccc = Pc and similarly taking upper complements stabilizes at the first step for antibasic elements. For C ≥ 0 a natural norm (satisfying the triangle inequality) is given by kCk = inf {kdk d ∈ D , D ∼ C} = inf {kck c ∈ Cc} . This norm is extended to general positive elements by kPk = kP ck and to general selfadjoint elements A = C − D by defining kAk = max(cid:8)kA+k , kA−k(cid:9) ≥ 0 . The triangle inequality is readily checked since (A + B)± ≤ A± + B± . If a = inf {a} , a ∈ Asa is the image of the selfadjoint element a = a+ − a− , a+a− = 0 in L(A) then a+ = a+ − a+ ∧ a− where a+ is the infimum of the set of all positive elements majorizing a , and a− is the infimum of all positive elements majorizing −a , both of which are obviously maximal for the equivalence relation ∼ . One easily checks that b ≥ a , b ≥ 0 implies kbk ≥ ka+k and b ≥ −a , b ≥ 0 implies kbk ≥ ka−k (for example by extending a functional ρ of C∗(a, 1) satisfying ρ(a) = ρ(a+) = ka+k to a functional of A by the Hahn-Banach Theorem, then ρ(b) ≥ ka+k for each b ∈ a+ ). Thus the norm of a is the same as the usual C∗- norm of a . Let us just remark that if instead considering a general ordered (say α-normal) real Banach space the construction still yields a positive homomorphism V → L(V) which is bounded from below on positive elements, but not necessarily bounded from below or even injective on general elements. Thus if V is not dually ordered there must exist for each ǫ > 0 an element x ∈ V such that every positive element which is larger than x is also larger than −x up to addition of a positive element of norm less or equal than ǫ since the minimal positive decomposition of the image x in L(V) is given by x = x+ − x− , x± = x ± − x + ∧ x − ± = inf {c ∈ V+ c ≥ ±x} . We want to extend both the squaring operation where x as well as taking square roots of positive elements to L(A) . It turns out that defining a squaring operation which is literally an extension of the square for single elements and has the properties to be expected is a rather strenuous adventure and appears to be unnatural. Instead we will give a stratified extension of the squaring operation which compensates for not being a proper extension by some other favourable aspects, e.g. being monotone. For an element C represented by a positive set define a basic squaring operation by the formula (1) C2 = inf(cid:8)c2 c ∈ Cc(cid:9) pC = inf(cid:8)√c c ∈ Cc(cid:9) . to be the element represented by the set of squares of the maximal representative of C and similarly (2) If c = inf {c} , c ≥ 0 is the image of a positive element of A then c2 ≤ inf {c2} whereas √c = inf {√c} by operator monotonicity of the square root. Of course for commutative A both operations are proper extensions since in this case also the square of operators is monotone. Note that in general these operations need not be 14 U. HAAG strictly inverse to each other, i.e. (√C)2 may be strictly smaller than C . Yet one does have the onesided inversion formula qC2 = C . (3) This follows since the set of squares of the maximal representative of C is again maximal for any convex combination of elements in this set is larger or equal to some element in the set. Indeed, for a given convex combination of two positive elements λa + (1 − λ)b , 0 ≤ λ ≤ 1 one has the inequality (λa + (1 − λ)b)2 ≤ λa2 + (1 − λ)b2 , (4) and taking square roots of this set will give back the original element C by operator monotonicity of the square root. To extend both definitions to general positive elements one is guided by consideration of certain convexity (concavity) properties in case of single operators. Any reasonable definition of the square root for general positive elements should be monotone, in particular should satisfy √P ≤ √P c . The inverse of inequality (4) for convex combinations of squareroots for two positive elements a , b ≥ 0 reads (5) Both inequalites (4) , (5) generalize to L(A) on replacing a , b by basic positive elements C , D , (4′) λ √a + (1 − λ) √b ≤ pλ a + (1 − λ) b . (λC + (1 − λ)D)2 ≤ λC2 + (1 − λ)D2 . λpC + (1 − λ)pD ≤ pλC + (1 − λ)D . (5′) Indeed, if p is an element of the upper complement on the right side of (5′) it can be represented in the form pλ c + (1 − λ) d with c ∈ Cc , d ∈ Dc up to considering larger elements or convex combinations. (4′) follows much in the same way. Assuming a + d = c the formula (5) applied to a and d replacing b renders (6) 1 P = inf √1 − λ√d(cid:1) λ , C , D (cid:26) 1 √a ≤ √λ(cid:0)pC − for every 0 < λ ≤ 1 . For P ≥ 0 one defines (7) √ √λ(cid:0)√c − √1 − λpD(cid:1) (cid:12)(cid:12) 0 < λ ≤ 1 , P ≤ C − D(cid:27) . From this definition it is immediate that the square root is monotone, i.e. P ≤ Q implies √P ≤ √Q . There is only one problem that the infimum may not be well defined in L(A) which contrary to L1(A) is not a complete lattice. The solution of this problem is to consider the following iterated lattice construction Lq(A) = L1(L(A)) . The canonical isometric embedding L(A) ֒→ Lq(A) is supplemented by a canonical surjective contraction π : Lq(A) ։ L1(A) (see below) induced by the canonical surjection L(A) ։ L1(A) . Then the squareroot of P is well defined as an element of Lq(A) and we may extend the definition to arbitrary positive elements of Lq(A) represented by a subset P = inf {Pλ}λ∈Λ by (8) pP = sup(cid:8)√Q (cid:12)(cid:12) 0 ≤ Q ≤ Pλ , ∀λ ∈ Λ(cid:9) . From monotonicity of the square root and the fact that L(A) is a lattice already (so that the lower complement of a positive subset can be replaced by the positive part of the lower complement in the notion of equivalence) this is well defined irrespective THE JORDAN LATTICE COMPLETION 15 of the representative set and monotone. Check that the definition (7) coincides with the previous definition for basic positive elements. It is then sufficient to verify whenever P ≤ C − D or Rewriting the left side as 1 pP ≤ √λ(cid:0)pC − pP + r 1 − λ λ pD ≤ √P √1 − λpD(cid:1) √λpC . √D√λ√1 − λ + (1 − λ) λ λ 1 (9) and applying (5′) with C replaced by λ−1 P and D replaced by (1− λ)−1 D gives the result. In complete analogy with the case of the square root one defines the square in general by the formula P 2 = sup λ , C , D (cid:26) λ(cid:0)C2 − From property (4′) one checks that this coincides with the previous definition for basic positive elements. Moreover the formula makes sense for arbitrary selfadjoint elements. For A ∈ L(A) define (10) 1 − λ D2(cid:1) (cid:12)(cid:12) 0 ≤ λ < 1 , P ≥ C − D(cid:27) . 1 − λ D2(cid:1) (cid:12)(cid:12) 0 ≤ λ < 1 , A ≥ C − D(cid:27) . It is immediate that sqr(A) is monotone and positive, and that sqr(A) = 0 for A ≤ 0 . Now put (11) λ , C , D (cid:26) λ(cid:0)C2 − sqr(A) = sup 1 1 A2 = sqr(A) + sqr(−A) to extend the definition of the square to arbitrary selfadjoint elements. The defini- tion of sqr (and a forteriori the squaring operation for selfadjoint elements) extends to Lq(A) in a similar fashion representing a general element A as the supremum of its lower complement in L(A) and putting sqr A = sup{sqr AA ≤ A} . The lat- tice Lq(A) endowed with the two operations of taking squares and squareroots (of positive elements) will be called the Jordan lattice associated with A . For each el- ement in Lq(A) there are defined upper (basic) and lower (antibasic) complements defined analogously as above such that one has the relations Ac ≤ Acc ≤ Acc ≤ Ac , Ac ≤ A ≤ Ac . An example for a convex positive map is given by the projections uc , (lcc) : Lq(A)+ → Lq(A)+ onto the subset of positive (complementary) basic elements defined by P 7→ Pc , P 7→ Pc , (cid:0)P 7→ Pcc(cid:1) . (cid:0)P 7→ Pcc(cid:1) . On the other hand examples of concave projections onto the subset of positive (complementary) antibasic elements are given by lc , (ucc) : Lq(A)+ → Lq(A)+ Homogeneity and monotonicity are obvious and convexity/concavity of uc/lc are easily checked. To prove convexity of the double lower complement one notes that it factors over the (linear) quotient map to L1(A) on applying a canonical convex 16 U. HAAG lift cv : L1(A) → L(A) as in the proof of Proposition 1 below. From the squaring operation one derives a Jordan type binary operation by the formula (12) h A , Bi = 1 2(cid:2)(A + B)2 − A2 − B2(cid:3) . (C + D + E)2 − (C + D)2 − (C + E)2 − (D + E)2 + C2 + D2 + E 2 ≏ 0 In order that this operation should be distributive with respect to addition in either variable one must have (∗) for every triple (C , D , E) of basic positive elements. The expression on the left side of (∗) is a quadratic cloud associated with the given triple and a main goal in the following is to find suitable conditions and relations making the quadratic cloud disappear. An important special case when this happens is if A is commutative. One then checks that the product (12) gives a distributive, commutative and asso- ciative Banach algebra product on L(A) which extends the product of A and can be shown to satisfy the C∗-condition khA , Aik = kAk2 . The details are left to the reader. Consider the complexification Lq(A)c = Lq(A)⊗RC . Although we have not yet imposed a norm on the complexification we may consider the original real subspace Lq(A) as consisting of selfadjoint elements with respect to the antilinear involution X = A + i B 7→ A − i B = X ∗ for A , B ∈ L(A) . It is then convenient to define kXk = sup(cid:8)kωX + ωX ∗k (cid:12)(cid:12) ω ∈ C , ω = 1(cid:9) . (C + iD) · (C − iD) = inf(cid:8)cµ,νc∗ As a preamble to the construction of an associative product (on a certain quotient space of Lq(A)c ) based on the Jordan bracket consider the following extension of the basic squaring operation on the subset of elements of the form X = C + iD with C , D ≥ 0 . One defines (13) which operation is well defined by uniqueness of the maximal representatives Cc , Dc and coincides with the basic squaring operation in case of a positive element. Note that the operation (13) is monotone in the sense that (C + iD)· (C − iD) ≥ (C′ + iD′)· (C′ − iD′) whenever C ≥ C′ and D ≥ D′ . From (13) one may derive another type of binary operation (of two positive elements say) which corresponds to the Lie bracket of two operators. The Lie type operation is given by µ,ν cµ,ν = cµ + i dν , cµ ∈ Cc , dν ∈ Dc(cid:9) i (14) ⌊ C , D ⌋ = In order that this operation should be distributive with respect to addition in the second variable one must have (∗∗) 2(cid:2)(cid:0)C + iD(cid:1)·(cid:0)C − iD(cid:1) − C · C − D · D(cid:3) . (cid:0) C + i (D + E)(cid:1) ·(cid:0) C − i(D + E)(cid:1) − ( C + iD ) · ( C − iD ) − ( C + iE ) · ( C − iE ) − ( D + E )2 + C2 + D2 + E 2 and similarly for the first variable (which case can be reduced to the one considered above by multiplying with the complex number i). The expression on the left side of (∗∗) is a quadratic cloud of second type associated with the given triple of positive elements. Assuming that all quadratic clouds of first and second type disappear under suitable restrictions and relations one may define a general (bilinear) product by linear extension of ≏ 0 (15) C · D = 1 2 hC , D i + 1 2 ⌊C , D ⌋ . THE JORDAN LATTICE COMPLETION 17 The relations needed to get associativity will be referred to as a quadratic cloud of third type associated with a given triple of elements and determining their precise form is left to the reader. Banach space V . Define L(cid:0)C(cid:1) ⊆ L(cid:0)V(cid:1) to be the subspace which is linearly gen- of L(cid:0)C(cid:1) can be written as a difference of two such elements or equivalently as a Remark. As with the L1-functor the L-construction admits various generaliza- tions. One is to unital affine subcones C ⊆ V in a regular unital real ordered erated by (positive) basic elements {C} with C ⊆ C so that an arbitrary element difference of two positive antibasic elements which are suprema of some given sub- sets in −C which are bounded above. The Proposition below then also applies to this more general situation. Proposition 1. For any unital ordered Banach space V the canonical surjection L(V) ։ L1(V) obtained by assigning with each positive basic element in L(V) its corresponding equivalence class in L1(V) is a complete order epimorphism hence extends naturally to a positive linear surjection which is a surjective ∗-homomorphism of the underlying (commutative) C∗-algebras. Moreover π is the unique monotonous extension of the identity map on V . π : Lq(V) ։ L1(V) Proof. To prove the statement one writes Lq(V) = L1(L(V)) and L1(V) ≃ L1(L1(V)) as the functor L1 is the identity when applied to a complete lattice. One notes that there are two canonical monotonous and homogenous extensions of any positive linear map φ : V → W to cv(φ) : L1(V) −→ L1(W) , cc(φ) : L1(V) −→ L1(W) where cv(φ) is convex and cc(φ) concave. These are obtained by composing the natural linear positive composition L(φ) L(V) −→ L(W) ։ L1(W) on the left with the two lifts induced for positive elements x ∈ L1(V)+ by the assignment x = [C] x = [C] 7→ Ccc = cv(x) , 7→ Cc = cc(x) respectively where C is any basic element representing x and extending to a mo- notonous R-homogenous map by cv(x) = cv(x+) − cv(x−) where x = x+ − x− denotes the minimal positive decomposition of x in L1(V) . Note that for positive elements one always has cc([C]) ≤ cv([C]) . Our strategy will be to show that the kernel of cv(π) is positively generated, so that from the general properties of convex maps it drops to a convex map on the quotient modulo ker cv(π) which then being an order isomorphism and the identity restricted to V must be the identity map. Then also cv(π) must be linear. Since the kernel of π : L(V) ։ L1(V) is positively generated, the order ideal it generates in Lq(V) is contained in the kernel of cv(π) . Namely if 0 ≤ c ≤ d with d ∈ ker π then c ∈ ker cv(π) follows from monotonicity. 18 U. HAAG and if 0 ≤ c1, c2 ∈ ker cv(π) one gets cv(π)(c1 + c2) ≤ cv(π)(c1) + cv(π)(c2) = 0 so that 0 = cv(π)(−(c1 + c2)) ≤ cv(π)(c1 − c2) ≤ cv(π)(c1 + c2) = 0 follows by monotonicity. Thus the order ideal as above is completely contained in the kernel of cv(π) so by the observation on positively generated kernels of convex maps the map cv(π) drops to a convex map π on the quotient lattice which canonically contains a copy of L1(V) , but in fact this copy makes up the whole space since the image of a general element A ∈ Lq(V) is less or equal than the image of Ac and larger or equal than the image of Ac from monotonicity. But these two latter elements are contained in L(V) and identified modulo the kernel of π which has been divided out, so that the image of A must be contained in the subspace L1(V) and the complementary subspace is empty. Therefore from rigidity the induced convex map on the quotient is the identity map. But then the original convex map cv(π) must also be linear and since its kernel is positively generated it is a ∗-ideal of the underlying C∗-algebra as in the Corollary above. The quotient Lq(V)/ ker cv(π) is unital isometric to L1(V) so the natural identification must be a ∗-isomorphism proving the first assertion. The argument also proves that pi restricted to L(V) is a complete order epimorphism since for a general element A ∈ Lq(cid:0)V(cid:1) uniquely determining a complemented subset C ⊆ L(cid:0)V(cid:1) one gets π(cid:0)Ac(cid:1) = π(cid:0)Ac(cid:1) with Ac ∈ C and Ac ∈ Cc . To see uniqueness of π assume given any monotonous extension Cλ,r − Dλ,r ≥ Cr − Dr whenever Cλ−Dλ ≥ C−D showing that r is a retraction for elements in the image of L(V) . Also for a positive basic element P one gets P r ≥ Cr − Dr whenever P ≥ C−D showing for one thing that r is well defined since r(P) ≤ r(P c) = (P c)r and that one recovers the original definition in case of a positive basic element. In fact it is easily seen that for any (even nonpositive) basic element one has r(A) = ρ : Lq(V) −−−→ L1(V) extending the identity map on V . If A ∈ Lq(V) is any element then necessarily π(A) = inf(cid:8)[c](cid:12)(cid:12) c ∈ V , c ≥ A(cid:9) ≥ ρ(A) ≥ sup(cid:8)[a](cid:12)(cid:12) a ∈ V , a ≤ A(cid:9) = π(A) so that ρ(A) = π(A) follows (cid:3) Remark. (i) For any order isomorphic unital embedding of unital ordered Banach spaces λ : V ֒→ W there exists a canonical convex, monotonous and positively homogenous retraction r : Lq(W) ։ Lq(V) for the functorial map L(V) ֒→ L(W) induced by restriction of a positive basic element C ∈ L(W)+ to the subset of elements Cr = {c ∈ V+ λ(c) ∈ C} , i.e. r(C) = Cr . Clearly if Cλ = L(λ)(C) is the image of a positive basic element in L(V) then r(Cλ) = C . Also it is easy to see that (C + D)r ≤ Cr + Dr and Cr ≤ Dr whenever C ≤ D , so that r is convex and monotonous on the convex subcone of positive basic elements. To extend the definition to more general elements define r(A) = sup(cid:8)Cr − Dr (cid:12)(cid:12) C − D ≤ A(cid:9) ∈ Lq(V) . From definition r is monotonous on L(W) (if well defined). If Cλ is an image of a basic element in L(V) and D is an arbitrary basic element then (Cλ + D)r = Cλ,r + Dr whence one has the relation THE JORDAN LATTICE COMPLETION 19 Ar where Ar = A ∩ V . Then the condition of positivity of basic elements in the definition of the extension of r is seen to be superfluous and can be replaced by considering differences of arbitrary basic elements. Positive homogeneity of r is more than obvious, then we may define r(A) = sup(cid:8)r(A) (cid:12)(cid:12) A ≤ A(cid:9) in order to extend the definition of r to Lq(W) . Combining this with the convex lift cv : L1(W) → L(W) and the linear surjection π : Lq(V) ։ L1(V) of Lemma 1 one obtains a convex monotonous retraction r1 : L1(W) cv−→ L(W) r −→ Lq(V) π−→ L1(V) for the functorial maps cv(λ) , cc(λ) . Note also that r is generally magnifying, (λ ◦ r)(A) ≥ A . Namely any element can be represented as a supremum of i.e. basic elements A = supµ Aµ . Then obviously (λ ◦ r)(Aµ) ≥ Aµ for each index µ whence This in turn shows that r is increasing normal, because (λ ◦ r)(A) = (λ ◦ r)(sup Aµ) ≥ sup Ar,λ µ ≥ A . r(sup Aµ) ≥ sup r(Aµ) = r(sup (λ ◦ r)(Aµ)) ≥ r(sup Aµ) hence equality in each instance. (ii) Since any element A ∈ L(W) can also be written as a difference of (positive or arbitrary) antibasic elements A = P − Q we may consider another functorial monotonous retraction r : Lq(W) ։ Lq(V) induced by restriction of antibasic elements, i.e. By symmetry with the above argument the general assignment r r(P) = P = sup{a ∈ V λ(a) ≤ P} . r r(A) = inf(cid:8)P r − Q P − Q ≥ A(cid:9) defines a monotonous retraction r : Lq(W) ։ Lq(V) for λ which extends the above definition for antibasic positive elements. Moreover r is generally reducing, i.e., (λ ◦ r)(A) ≤ A and decreasing normal, i.e. r(inf Aµ) = inf r(Aµ) . One obtains the general relation r(A + B) = inf(cid:8)P r (cid:12)(cid:12) P ≥ A + B(cid:9) ≤ inf(cid:8)(P − A)r + Ar (cid:12)(cid:12) A ≤ A , P ≥ A + B(cid:9) r (cid:12)(cid:12) Q ≥ B(cid:9) = r(A) + r(B) . ≤ sup(cid:8)Ar (cid:12)(cid:12) A ≤ A(cid:9) + inf(cid:8)Q In the following A denotes a C∗-algebra as usual, but note that the same concepts can be applied to a regular unital ordered Banach space V (isometrically order isomorphic with an operator system X ⊆ B(H) ). Let L(A) ⊆ L(A) denote the subcone of basic elements which may be identified with their maximal representative set Ac ⊆ Asa , and is closed under addition and multiplication by positive scalars. Also let C(S(A)) be the subcone of upper semicontinuous functions on the state space of A and bC(S(A)) its linear envelope. Define a positive linear map s : L(A) → bC(S(A)) by linear extension of the affine map (16) where ρ ∈ S(A) and check that it is well defined and strictly monotone in the sense that A (cid:8) B implies s(A) (cid:8) s(B) . This implies that s is an order isomorphic injection. Indeed, suppose that s(A) ≥ 0 for some A = C−D . Then s(C) ≥ s(D) . s(Ac)(ρ) = inf {ρ(a) a ∈ Ac} s : L(A) −→ C(S(A)) , 20 U. HAAG Choosing an arbitrary c ∈ Cc and any state σ one gets sσ(Dc ∪ {c}) = sσ(D) which implies c ∈ Dc by strict monotonicity of s . Therefore C ≥ D and A ≥ 0 . To see that s is strictly monotone it is sufficient to prove that whenever C ⊆ A+ is a maximal closed convex subset of positive elements (i.e. x ∈ C implies y ∈ C for y ≥ x ) and d ≥ 0 is a single positive element not contained in C then there exists a positive functional φ ∈ (A∗)+ separating C and {d} . To prove this choose a small convex open neighbourhood C0 ⊃ C such that d is not contained in the closure of C0 and a point c0 ∈ C . Let C = C0 − c0 = {c − c0 c ∈ C0} . Then C is a convex open neighbourhood of 0 with associated Minkowski functional m(x) = inf {s > 0 s−1x ∈ C} . One checks that m(x) = 0 whenever x ≥ 0 since A+ ⊆ C . This implies that any functional dominated by m must be negative. Put z = d − c0 /∈ C so that m(z) > 1 . Define ψ0(sz) = s for s ∈ R and extend ψ0 to a (necessarily negative) functional ψ on A dominated by m from the Hahn-Banach theorem. Then ψ separates C and {d} and the same holds for the positive functional φ = −ψ proving the statement. For a given state ρ ∈ S(A) let sρ : L(A) ։ C be the induced state on L(A) , i.e. sρ(A) = s(A)(ρ) . Then the kernel of sρ is positively generated, since if C , D are given with sρ(C) = sρ(D) = r then also sρ(C∧D) = r . Also consider the restriction of s to the pure states where P(A) denotes the set of pure states of A , and in case that M is a von Neumann algebra consider the restriction to the normal states r : L(A) −→ bC(P(A)) sν : L(M ) −→ bC(Sν (M )) , rν : L(B(H)) −→ bC(P ν(B(H))) . resp. if M = B(H) the restriction to the vector states (= pure normal states) The maps r , sν and rν fail to be injective in general. We will sometimes also consider the subset Sν f of (normal) states which are finite convex combinations of vector states. f . For a fixed normal state ρ ∈ Sν For the following constructions we consider concrete operator subsystems X ⊆ B(H) , i.e. a specific unital completely isometric representation is taken into ac- count and we therefore consider the subset of states on X which are restrictions of states in Sν f let C0,ρ ⊆ L(X)+ be the subcone generated by basic positive elements which are restrictions E r = r(E) of basic pos- itive elements E ∈ L(B(H)) such that sν ρ(E) = sν(E)(ρ) = 0 and C0,ρ ⊆ L+(X) the positive subcone generated by images of basic positive elements E as above under the antibasic restriction map r as in part (ii) of the Remark after Propo- sition 1. Then C0,ρ (resp. C0,ρ ) generates an order ideal J0,ρ (resp. J0,ρ ) in L(X) (by restriction of the corresponding order ideal in Lq(X) ) whose positive part consists of all positive elements Q ∈ L(X) which are dominated by some ele- ment E ∈ C0,ρ (resp. P ∈ C0,ρ ). Since always r ≤ r one obtains J0,ρ ⊆ J0,ρ . Let J0 = r(cid:0)Tρ∈P ν J0,ρ(cid:1) ⊆ L(X) be the order ideal which is generated by the restriction of the intersection of all kernels ker rν ρ \ρ∈P ν THE JORDAN LATTICE COMPLETION 21 ρ , resp. Jρ ⊆Tn i=1 ker rν ρi , ρ =Pn which is the same as the order ideal generated by Tρ C0,ρ from the fact that Tρ C0,ρ generates the order ideal Tρ ker rν ρ and that r is generally magnifying in the sense that (λ◦ r)(A) ≥ A where λ : Lq(X) ֒→ Lq(B(H)) denotes the functorial (uniquely determined, normal) embedding by the given representation (to be proved in Theorem 1 below). In case that X = B(H) both restriction maps are the identity and we simply write J0,ρ and J0 respectively. If a ∈ Xsa is any selfadjoint element put a = inf {a} , and a± = ac ± where a = a+ − a− denotes the minimal positive decomposition of a in L(A) . Note that each element a+ ∧ a− is the (basic) restriction of the corresponding element in L(B(H)) viewing a as an element in B(H) . Then the same holds for the squares of such elements (in case that X = A is a sub-C∗-algebra) which are all contained in J0 by Lemma 2 below. Put Jρ = J0,ρ + J0 (resp. Jρ = J0,ρ + J0 ). For any collection of normal states F ⊆ Sν f let J0,F and J0,F be the order ideal generated by r(cid:0)Tρ∈F J0,ρ(cid:1) and r(cid:0)Tρ∈F J0,ρ(cid:1) respectively, and JF = J0,F + J0 , JF = J0,F + J0 . Put π0,F : L(X) ։ L(X)/J0,F , resp. πF : L(X) ։ LF(X) = L(X)/JF and similarly π0,F , πF denote the induced surjections all of which are (restrictions of) homomorphisms due to the fact that the kernel is positively generated hence an ideal in Lq(X) . It will follow from Theorem 1 that for a pure normal state of B(H) one has Jρ = J0,ρ = ker rν ρ , and J0,ρ ⊆ ker sν f , so the definition "makes sense" (at least for X = B(H) ) in that these order ideals are not unreasonably large. Since the antibasic restriction map is in general reducing, i.e. (λ ◦ r)(A) ≤ A with λ as above one easily finds that J0,ρ ⊆ ker sν ρ . However ρ . We let Sν the basic restriction of J0,ρ to Lq(X) may not lie in the kernel of sν 0 denote the collection of vector states which is the same as P ν(B(H)) . Restricted to a general operator subsystem however these may no longer be pure states so that in this setting the notation Sν Suppose that M ⊆ B(H) is a von Neumann algebra. Consider the subspace of L(M ) generated by positive basic elements which are infima of w∗-closed subsets C ⊆ M+ . It is not immediately clear that the sum of two maximal w∗-closed positive subsets is again w∗-closed. Therefore we consider some related notions: A subset R ⊆ M ⊆ B(H) is said to be boundedly w∗-closed iff each bounded component R ∩ Br(M ) is w∗-closed. Any w∗-closed set is boundedly w∗-closed. This notion defines a topology on M by taking as closed sets the boundedly w∗-closed sets which will be called the w∗ 0-closed set is necessarily monotone complete and sequentially w∗-closed, since any w∗-convergent sequence must be bounded, the same does not seem to hold for arbitrary nets though. From monotone completeness one induces that any w∗ 0-continuous functional must be normal, so the dual space for M endowed with the w∗ 0-topology coincides with the predual of M . Taking w∗ 0-closures gives a natural projection from positive (closed) convex subsets to boundedly w∗-closed positive convex subsets which can also be realized by infinitely iterating the procedure of taking the union of the w∗-closures of its bounded subsets, i.e. defining i=1 λiρi for a state ρ ∈ Sν 0 seems more appropriate. 0-topology. A positive w∗ C1 = [n∈N w∗ C ∩ Bn(M ) and Cν =Sω Cω as the union over the corresponding transfinite sequence of itera- tions, the successor of each Cω being defined in the manner above and for indices ω 22 U. HAAG without precessor applying the formula to the union of all preceding indices κ < ω . Each subset Cω is contained in the w∗-closure of C so the process must eventually come to an end giving the w∗ 0-closure of C (of course it is much simpler to use the w∗ 0-closure operation in one step, but the construction above is more appealing to imagination which points will lie in the closure). If C is boundedly w∗-closed al- ready then Cν = C . The subspace of differences of basic elements corresponding to boundedly w∗-closed positive convex sets forms a sublattice Lν (M ) ⊆ L(M ) since the sum of two w∗-compact sets is again w∗-compact as well as the closed convex hull of the union of finitely many such sets. Then the assignment C 7→ Cν extends to a well defined linear and positive projection pν : L(M ) ։ Lν (M ) ⊆ L(M ) onto the sublattice Lν (M ) . Also the subspace of differences of boundedly gener- ated positive basic elements which are infima of bounded w∗-closed positive subsets constitute a sublattice Lν 0 (M ) ⊆ Lν (M ) , and it will be shown that this sublattice has an order isomorphic representation by sν f , i.e. the restriction of s to the subset f . Note first that for a bounded set the notions w∗-closed (= σ-weakly closed) and Sν weak operator closed agree, so the same is true for positive boundedly generated maximal (= closed under norm and addition of positive elements) convex subsets C which are w∗-closed, resp. weak operator closed if and only if there exists a w∗- closed (= weak operator closed) bounded subset B ⊂ C having the property that for each c ∈ C there is a b ∈ B with b ≤ c . For suppose that C is not weak opera- tor closed. Then there exists a net (cλ) ⊆ C converging weakly to c /∈ C . For each λ choose an element bλ ≤ cλ , bλ ∈ B . Then the bounded net (bλ)λ admits a con- vergent subnet (bλµ ) → b converging weakly to an element b ∈ B . Since bλ ≤ cλ for each index λ this implies ρ(b) ≤ ρ(c) for each vector state ρ whence b ≤ c and c ∈ C follows. Considering the quotient lattices of L(M ) modulo the kernels of the map sν and sν f both of which are order ideals one may also define the lattices (M ) = L(M )/ ker sν corresponding to the weak Lw(M ) = L(M )/ ker sν operator topology and the w∗-topology respectively. It is obvious that the image of a positive basic element C in Lw(M ) is the same as the image of the element corre- sponding to the weak operator closure of Cc , and that the map to Lw∗ (M ) factors over taking w∗-closures of the underlying sets. Although these are not sublattices of L(M ) the sublattice Lν 0 (M ) ⊆ L(M ) also embeds naturally, isometrically and order isomorphic into Lw(M ) and Lw∗ (M ) by the obvious identifications. To see this it is sufficient to prove that for any weak operator closed maximal convex sub- set C ⊆ B(H)+ and d ≥ 0 , d /∈ C there exists a state in Sν f separating C and {d} . Since C is weak operator closed there exists a weak operator open convex neighbourhood of {d} not intersecting C . The Hahn-Banach separation theorem ([13], Theorem 2.4.7) gives the existence of a normal selfadjoint functional ρ in the linear span of functionals x 7→ hx ξ , ηi , ξ , η ∈ H separating C and {d} . Since the set C is closed under addition of positive elements this functional must be definite, for otherwise if ρ = ρ+ − ρ− denotes the Jordan decomposition of ρ there exists for each ǫ > 0 a positive element k ∈ M+ satisfying ρ+(k) ≤ ǫ , ρ−(1 − k) ≤ ǫ so that ρ(C) = R is the real line which means that the functional cannot separate C and {d} . Therefore ρ may be assumed to be a normal state in Sν f proving (order isomorphic) injectivity of sν 0 (M ) (compare with the argument above). A similar separation argument by arbitrary normal states works for max- imal w∗-closed positive sets. In case of Lν 0 (M ) we may also substitute the dense subset Sν f of normal states which are finite convex combinations of vector f restricted to Lν f and Lw∗ Λ ⊆ Sν THE JORDAN LATTICE COMPLETION 23 sis {ξλ}λ∈Λ of H , where PF =Pn states whose support projection PF is subordinate to some given orthonormal ba- i=1 pλi , F = {λ1,··· , λn} ⊆ Λ and pλ is the minimal projection corresponding to the basis vector ξλ ∈ H , since this subset is f ⊆ Sν normdense in Sν with respect to the norm in M ∗ hence w∗-dense in Sν viewed as a subset of Lν 0(M )∗ . Lemma 1. For an arbitrary collection F ⊆ Sν f and any C∗-subalgebra A ⊆ B(H) the basic squaring operation (1) as well as the operation (13) for basic positive elements drops to the quotient L0,F (A) = L(A) / J0,F , and to the quotient L (A) = L(A) / JF . In particular if C , D ≥ 0 with C − D ∈ JF , then F πF (C2) = πF (D2) , πF (C2) = πF (D2) . i=1 , 0 ≤ λi ≤ 1 ,P∞ i=1 λi = 1 such that ρ(x) =P∞ Proof. To prove the first statement of the Lemma we begin by considering the special case A = B(H) . Let ρ be any normal state. Then there exists a sequence of pairwise orthogonal vector states {ρi}∞ i=1 and a decreasing sequence of positive numbers {λi}∞ i=1 λiρi(x) . To each ρi corresponds a minimal projection pi ∈ B(H) such that ρi(x) = tr(pi x) . Assume first that ρ is a pure (vector) state with corresponding minimal projection p . Let E ≥ 0 be a basic positive element such that rν ρ (E) = 0 . Then for any given ǫ > 0 there exists e ∈ E with ρ(e) ≤ ǫ . We claim that this implies rν ρ (e2) ≤ 4ǫ2 where as usual e = inf {e} . If ρ(e) = 0 then ρ(e2) = 0 and we are done. Otherwise assume 0 < ρ(e) ≤ ǫ . One has pep = ρ(e)p so that we may find e′ ≥ e with e′p = pe′ = (ρ(e) + ǫ)p by adding an element ǫp + R (1 − p) to e with ǫ > 0 as above and R ≫ 0 a sufficiently large scalar, and simultanously erasing the matrix coefficients pe(1 − p) and (1 − p)ep . Then rν ρ (e2) ≤ ρ((e′)2) = ρ (E 2) ≤ 4ǫ2 and since ǫ can be chosen arbitrarily (ρ(e) + ǫ)2 ≤ 4ǫ2 . Then also rν small this settles the case of a vector state. In fact, basically the same argument gives rν ρ (C)2 , and the reverse implication is an easy consequence of the ρ (C)2 for any positive basic element Schwarz inequality for ρ . Therefore rν C ∈ L(B(H)) . The method of proof immediately generalizes to the case of a finite convex combinations of pairwise orthogonal vector states provided that the matrix coefficients piepj = 0 for i 6= j are all zero. In the general case there exists for any given ǫ > 0 an element e ∈ E with ρi(e) ≤ ǫ for each i = 1 ,··· , n . Upon i=1 pi one can substitute e by an element e′ ≥ e with pie′pj = 0 whenever i 6= j and ρi(e′) ≤ 2i ǫ so that if ǫ is chosen sufficiently small these values become arbitrarily small. Then one ρ(E 2) = 0 . It is rather surprising may apply the argument above to show that sν that the corresponding result for an arbitrary normal state seems to be false (or extremely hard to prove), comparing with the fact that any normal state is in the norm closure of Sν f (B(H)) . By monotonicity of the square it follows a forteriori that Q2 ∈ J0,ρ for 0 ≤ Q ≤ E . Let A ⊆ B(H) be a C∗-algebra. If E ∈ J0,ρ(A) is the restriction of an element E ′ ∈ J0,ρ(B(H)) then E 2 is the restriction of (E ′)2 since the squareroot will give back the original elements. Therefore also E 2 ∈ J0,ρ . Then π0,ρ((C + E)2) ≥ π0,ρ(C2) by monotonicity of sqr and positivity of π0,ρ . On adding a positive element of the form P cP with P =Pn ρ (C2) ≤ rν ρ (C2) = rν 24 U. HAAG the other hand from (2′) one gets π0,ρ((C + E)2) ≤ 1 λ π0,ρ(C2) + 1 1 − λ π0,ρ(E 2) = 1 λ π0,ρ(C2) for each 0 < λ < 1 . This proves π0,ρ((C + E)2) = π0,ρ(C2) . Similarly π0,ρ((C − E)2) ≤ π0,ρ(C2) whenever C ≥ E , while on the other hand π0,ρ((C − E)2) ≥ λ(cid:0)π0,ρ(C2) − 1 1 − λ π0,ρ(E 2)(cid:1) = λ π0,ρ(C2) for 0 ≤ λ < 1 proving the reverse inequality. If X = C + i D = Y + (P + i Q) with P c , Qc ∈ Jν 0,ρ then putting X = Y +(P c +i Qc) it is easy to see from monotonicity of the basic operation (13) that π0,ρ(X·X ∗) ≥ π0,ρ(Y·Y ∗) . On the other hand one gets for each 0 < λ < 1 π0,ρ(X·X ∗) ≤ π0,ρ(X·X ∗ ) ≤ 1 λ π0,ρ(Y·Y ∗) + 1 1 − λ π0,ρ((P c + i Qc)·(P c − i Qc)) = 1 λ π0,ρ(Y·Y ∗) ρ(C2) = sν by operator convexity of x 7→ xx∗ proving the reverse inequality. Note however that the identity sν ρ(C)2 is no longer valid if ρ is not a pure state even in the case A = B(H) . In order to prove the second statement we have to presuppose certain results to be proved in Theorem 1 below, namely that the inclusion λ : L(A) ֒→ L(B(H)) extends canonically to a normal multiplicative inclusion λ : Lq(A) ֒→ Lq(B(H)) which then drops to an injective ∗-homomorphism Also the (homogeneized) basic restriction map r as well as the antibasic restriction map drop to monotonous retractions λ : Lq(A) / JF −→ Lq(B(H)) /(cid:0)J0 + λ(JF )(cid:1) . r , r : Lq(B(H)) /(cid:0)J0 + λ(JF )(cid:1) ։ Lq(A) / JF r(P) ≤ r(P + Q) ≤ r(P) + r(Q) ≡ r(P) for λ since r is convex on positive elements while in case that 0 ≤ Q ∈ J0 + λ(JF ) . Then one easily sees that r is generally magnifying whereas r is generally reducing. Then modulo J0 one has the identity (cid:2)C2(cid:3) = (cid:2)C · C(cid:3) = (cid:2)C(cid:3) ·(cid:2)C(cid:3) for any positive basic element C as Here we have used the relation λ(C)2 ≡ λ(C ·C) modulo J0 which is a special case of a general result to be proved below in Theorem 1 but in principle follows from ρ (C)2 derived above. Then if C = D + Q with Q ∈ JF we the result rν can use the C∗-square to see that [C2] = r(cid:2)λ(C2)(cid:3) = r(cid:2)λ(C)2(cid:3) = r(cid:2)λ(C · C)(cid:3) = (cid:2)C(cid:3) ·(cid:2)C(cid:3) . ρ (C2) = rν πF(cid:0)C2(cid:1) = πF(cid:0)C(cid:1) · πF(cid:0)C(cid:1) = πF(cid:0)D(cid:1) · πF(cid:0)D(cid:1) = πF(cid:0)D2(cid:1) . The argument in the case of (13) is quite similar and we leave it to the reader (cid:3) THE JORDAN LATTICE COMPLETION 25 Let a = a+ − a− be a selfadjoint element of A with a± ≥ 0 , a+a− = 0 and C ⊆ A an abelian C∗-subalgebra containing a . Define (18) sqr(a) = sup 1 λ,C,c,d(cid:26)λ(cid:0)c2 − 1 − λ d2(cid:1) (cid:12)(cid:12)(cid:12) c , d ∈ C+ , c − d ≤ a(cid:27) . If a ≥ 0 then simply sqr(a) = a2 by monotonicity of the square in a commutative C∗-algebra and (4) , but note that (18) is in general not the same as sqr(a) , a = inf {a} as defined above. Let a± = (±a) ∨ 0 and a± = inf { b ≥ 0 b ≥ ±a} = ac ± . Then a± = a± − (a+ ∧ a−) so that a = a+ − a− is the minimal basic positive decomposition of a . One has the following result Lemma 2. With notation as above sqr(a) = sqr(a+) = a2 + , sqr(a) ≤ a2 + . If A = B(H) then +)c , sqr(a)cc = (a2 π(sqr(a)) = π(a2 Moreover (a+ ∧ a−)c = 0 for every a ∈ Asa , resp. rν normal state ρ ∈ P ν(B(H)) . Proof. + is obvious since any element which commutes with a also commutes with a+ so we only need to prove the converse on fixing C = C∗(a) . Let R ≥ 0 be a positive scalar. One has +) = π(a2 ρ (a+ ∧ a−) = 0 for every pure sqr(a) ≤ sqr(a+) = a2 +) . (Ra+ + a−)2(cid:1)(cid:9) −)(cid:1)(cid:9) ≥ a2 sqr(a) ≥ sup λ,R(cid:8)λ(cid:0)(R + 1)2a2 + − 1 1 − λ and hence putting λR = (R + 1)−1 sqr(a) ≥ sup R (cid:8)λR(cid:0)(R + 1)2a2 1 (R2a2 1 + − R + as R → ∞ . This proves the first result. 1 − λR + + a2 + − a2 − which converges uniformly to a2 By monotonicity of sqr one has sqr(a) ≤ sqr(a+) = a2 which implies sqr(a)cc ≤ (a2 +)c so we only need to prove the reverse inequality, i.e. given an arbitrary state ρ ∈ S(A) we must prove the inequality sρ(sqr(a)cc) ≥ sρ((a2 +)c) provided that A = B(H) . We first do the case where A = Mn(C) is a matrix algebra. In this case it is sufficient to consider only vector states since both elements are in the image of the canonical concave lift + L1(cid:0)Mn(C)(cid:1) cc−→ L(cid:0)Mn(C)(cid:1) to the quotient map which factors over rν as shown in the proof of Theorem 1. Given a ∈ Mn(C)sa one can assume that aij = 0 for i 6= j and there exist two projections p+ , p− commuting with a such that p+ + p− = 1 and a+ = p+a , a− = p−a . Then ρ decomposes as ρ = ρ+ + ρ− + ρ+/− where ρ+(x) = ρ(p+ x p+) , ρ−(x) = ρ(p− x p−) , One may further decompose p+ as a sum of minimal projections p+ =Pm ρ+/−(x) = ρ(p+ x p− + p− x p+) . k=1 ek which correspond to eigenvectors of a+ and p− as a sum of minimal projections em+1 , ··· , en corresponding to eigenvectors of a− . Assume now that there exists an index 1 ≤ k ≤ m and an index m + 1 ≤ l ≤ n such that trs = tsr = 0 for all 26 U. HAAG 1 ≤ r , s ≤ n unless , r , s ∈ {k , l} . This means that ρ factors canonically over the completely positive projection Q onto the corresponding copy of M2(C) with Q sqr(a) Q = sqr(Q a Q) , Q a+ Q = (Q a Q)+ and we may as well assume n = 2 . Then ρ(x) = tr (t x) with Without loss of generality τ ∈ R+ . Suppose first that t+ a+ < t− a− . Then the element t1 + t2 = 1 , τ τ = t+ t− . −τ t−(cid:19) , t = (cid:18) t+ −τ c = λ (cid:18)t− τ τ t+(cid:19) is contained in a+ if λ > 0 is chosen large enough, and ρ(c2) = 0 so that sρ(sqr(a)) ≥ sρ(a2 +) follows trivially. Approximating a in norm by elements with slightly smaller positive part one finds that the same result holds in case that t+ a+ = t−a− . On the other hand if t+ a+ > t− a− the element d = λ(cid:18)t− τ τ t+(cid:19) is contained in a− if λ > 0 is chosen large enough with ρ(x d) = ρ(d x) = 0 for every x ∈ A . In particular (19) with c ∈ a+ any element minimizing the value of ρ(c2) . As we will see this implies the same result. Assume for the moment that we are given a functional ρ such that for the minimal value ρ(c2) , c ∈ a+ there exists d ∈ a− with ρ(c d + d c) = 0 . We will see that this implies sρ(sqr(a)) ≥ sρ(a2 +) . For each 0 ≤ λ < 1 and each R ≥ 0 one has the estimate ρ(c d + d c) = 0 by definition of sqr . Then, given c ∈ a+ , d ∈ a− as above sqr(a) ≥ λ(cid:18)(cid:0)(R + 1) a+(cid:1)2 sρ(sqr(a)) ≥ λ(cid:0)(R + 1)2ρ(c2) − − 1 1 − λ(cid:0)R a+ + a−(cid:1)2(cid:19) (R2ρ(c2) + ρ(d2))(cid:1) 1 − λ 1 since ρ(c d + d c) = 0 . At λR = (R + 1)−1 the expression to the right attains the value ρ(c2) + R−1 ρ(d2) which converges to the desired estimate sρ(sqr(a)) ≥ ρ(c2) = sρ(a2 +) as R → ∞ . In particular one gets sρ(sqr(a)c) ≥ sρ((a2 +)c) . The whole argument given above extends to the slightly more general setting if one only assumes aij = 0 whenever 1 ≤ i ≤ m , m + 1 ≤ j ≤ n or m + 1 ≤ i ≤ n , 1 ≤ j ≤ m . If ρ is a given state of M2-type as above so that ρ = ρ ◦ ϕ with ϕ : Mn(C) ։ M2(C) the canonical retraction corresponding to the specified pair of indices (k, l) with 1 ≤ k ≤ m , m + 1 ≤ l ≤ n inducing a surjection of a+ onto (aϕ)+ where aϕ = inf {ϕ(a)} , and if ϕ(c) minimizes the value of ρ(ϕ(c)2) one needs to check that it is possible to extend the 2 × 2-matrix ϕ(c)2 (or some positive element which takes the same value at ρ ) to an n × n-matrix c2 satisfying c ∈ a+ . Since we may add arbitrarily large elements on the diagonal entries different from (k, k) and (l, l) and scalar multiples of the complementary onedimensional projection with respect to t (assuming that ρ is a pure state) this THE JORDAN LATTICE COMPLETION 27 is readily achieved. This means that the whole argument is invariant under unitary transformations of the form Ad (U ⊕ V ) with U a unitary matrix in Mm(C) , and V a unitary in the complementary copy of Mn−m(C) . If ρ is any pure state on Mn(C) it corresponds to a minimal projection t ∈ Mn(C) , t2 = t = t∗ via ρ(x) = tr(t x) . Put t1 = p+tp+ , t12 = p+tp− , t2 = p−tp− . Applying a unitary transformation of type Ad (U ⊕ V ) it is possible to diagonalize t1 and t2 both of which are rank one operators so that only two nonzero diagonal entries remain, i.e. ρ ◦ Ad (U ⊕ V ) is of the type considered above and sρ(sqr(a)) ≥ sρ(a2 +) follows. Since for each fixed A the map s(A) (resp. s(B) ) viewed as a map from S(A) to R extends to a R+-homogenous map on the cone of positive functionals which is an infimum of affine maps if A is basic, resp. a supremum of affine maps if B is antibasic one concludes that if given positive functionals ρ , ρ′ , ρ′′ such that ρ = ρ′ + ρ′′ and sρ′ (A) ≥ sρ′ (B) , sρ′′ (A) ≥ sρ′′ (B) then sρ(A) ≥ sρ(B) follows. Therefore sρ(sqr(a)c) ≥ sρ((a2 +)c) for every state ρ (being a convex combination of irreducible states). Also, since Q = sqr(a)c − (a2 +)c is a basic element with s(Q) ≥ 0 one concludes that Q ≥ 0 whence sqr(a)c ≥ (a2 +)c . This again implies sqr(a)cc ≥ (a2 +)c . (cid:3) As we have seen the reverse inequality is trivial. This accounts for the case A = Mn(C) . The argument given above can be generalized to infinite dimensions, i.e. the case A = B(H) where H is a Hilbert space of arbitrary dimension, the most important ingredient being the existence of a matrix decomposition p+ + p− = 1 corresponding to the positive and negative part of a plus pure normal states corresponding to minimal projections via the trace functional. The rest of the argument easily adapts to cover infinite dimensions since the normal states suffice to determine positivity of basic elements in L(B(H)) . This then also proves the ρ (a+ ∧ a−) = 0 for any pure normal state ρ ∈ P ν(B(H)) and any assertion rν selfadjoint element a ∈ B(H) +) is in fact valid for arbitrary C∗-algebras Remark. The formula π(sqr(a)) = π(a2 as follows from Theorem 1. The Lemma also implies (a2 +)c , as sqr(a)cc ≤ +)cc ≤ (a2 (a2 +)c . Moreover the proof of the Lemma shows that for A = B(H) and given pure normal state ρ one either has rν ρ (a−) = 0 so In particular a+ ∧ a− ∈ ker π for arbitrary C∗-algebras that a+ ∧ a− ∈ J0,ρ . A by the following scheme. Although there is in general no natural linear map L1(A) → L1(B(H)) associated with a given ∗-representation λ : A ֒→ B(H) there is some connection, namely a selfadjoint element A ∈ L(A) whose image in L(B(H)) is in the kernel of π (with respect to L1(B(H)) ) is in the kernel of π for A . To see this it is sufficient consider positive elements P which are identified with their lower complements in L1(A) . If P maps to 0 in L1(B(H)) then any element in the lower complement with respect to B(H) is negative or zero, which implies that the same holds for the lower complement with respect to A whence π(P) = 0 . In particular a+ ∧ a− ∈ ker π holds for arbitrary C∗-algebras A since if Λ : L(A) ֒→ L(B(H)) denotes the functorial linear map associated with some faithful ∗-representation λ : A ֒→ B(H) then Λ is (the restriction of) a ∗-homomorphism by Theorem 1 ± where aλ = inf {λ(a)} . In case of injective hence a lattice map and Λ(a±) = aλ A one also has a positive linear retraction Υ : L(B(H)) ։ L(A) functorially ρ (a+) = 0 or else rν +)cc = (a2 28 U. HAAG extending a given completely positive linear retraction υ : B(H) ։ A for λ and there is a commutative diagram (20) L(A) Λy L(B(H)) π π −−−→ L1(A) cv(υ)x cc(υ) −−−→ L1(B(H)) . To see this consider the composition of the upper horizontal map of the diagram with either map cc(λ) and cv(λ) respectively and check that the image of a positive element P by the former composition is smaller or equal than its image under the linear composition L(A) ֒→ L(B(H)) ։ L1(B(H)) whereas the image of the same element by the latter (convex) composition is larger or equal than π ◦ Λ(P) . On the other hand any of the compositions cc(λ) ◦ cc(υ) = cc(λ) ◦ cv(υ) = cv(λ) ◦ cc(υ) = cv(λ) ◦ cv(υ) = id equals the identity map from rigidity. Therefore by the sandwich principle both compositions cv(υ) ◦ π ◦ Λ = cc(υ) ◦ π ◦ Λ must equal π . Lemma 3. Let H be a Hilbert space of arbitrary dimension and B(H) the algebra of bounded operators on H . Then for any pure normal state ρ ∈ P ν(B(H)) the induced functional rν ρ : L(B(H)) → R ρ (C2) = rν ρ (C)2 for every basic positive element C ≥ 0 and every element satifies rν A with rν (A) = 0 is contained in J0,ρ . This applies in particular to the quadratic clouds of first, second and third type with respect to any given triple (C , D , E) of basic positive elements which are in the kernel of π0 = π0,P ν (B(H)) . Let A = B(H) and ρ be a given pure normal state. The first statement Proof. is a consequence of the proof of Lemma 1. For E ∈ L(B(H)) a (positive) basic element put E ρ = E − rν ρ . It is sufficient to show that E ρ ∈ J0,ρ . Since E ρ is basic its minimal basic positive decomposition is given by ρ(cid:0)E(cid:1) 1 ∈ ker rν ρ − E − E ρ = E + ρ = (E ρ,+)c − (E ρ,−)c where E ρ = E ρ,+ −E ρ,− denotes the minimal positive decomposition. Given ǫ > 0 choose e ∈ E c with ρ(e) ≤ rν ρ (eρ) ≤ ǫ , where eρ = inf {e − rν ρ ((eρ)+) = 0 or else rν ρ ((eρ)−) = 0 which implies ρ (E)1} . By Lemma 2 one always has either rν ρ (E) + ǫ . Then 0 = rν ρ (E ρ) ≤ rν Then rν ρ (E − 0 ≤ rν ρ (E + ρ ) ≤ ǫ by linearity of rν E ≡ rν ρ ((eρ)+) ≤ ǫ . ρ ) ≤ rν ρ and therefore since ǫ was arbitrary ρ (E) 1 mod J0,ρ . The argument above shows that i.e. the kernel of rν A = C − D ∈ ker rν ρ is equal to J0,ρ ρ =⇒ A = Cρ − Dρ ∈ J0,ρ , (cid:3) The Lemma shows in particular that the induced map rν : π00(L(B(H))) → bC(Pν(B(H))) THE JORDAN LATTICE COMPLETION 29 is injective where π00 = π0,S ν 0 in the terminology introduced before Lemma 1. We now consider some specific notions defined for injective operator systems. Al- though injective operator systems carry a uniquely determined structure as a C∗- algebra it is sometimes convenient to "forget" the multiplicative structure. In the following we will occasionally consider unital completely positive linear represen- tations of (mostly injective) operator systems λ : X → B(H) . Such a represen- tation will be called transitive iff the image contains an irreducible subalgebra of B(H) , in particular this implies that given any finitedimensional orthogonal pro- jection P ∈ B(H) one has P λ(X+) P = B(P H)+ and P λ(X1) P = B(P H)1 where X+ denotes the positive cone and X1 denotes the unit ball of X . By Kadison's transitivity theorem (c.f. [12], Theorem 2.7.5) these properties hold for any irreducible ∗-representation of a C∗-algebra A . The representation λ will be called relatively transitive iff the image contains a subalgebra which is strongly dense in the strong closure λ(X)′′ . An injective representation of an injective operator system I is supposed to mean a unital completely positive linear map λ : I → B(H) such that λ factors as a product of a complete quotient map π : I ։ J with injective image J = λ(I) and a completely isometric unital embed- ding ι : J ⊆ B(H) . The representation is called multiplicative iff π is a surjective ∗-homomorphism. Note however that a multiplicative injective representation need not be a ∗-homomorphism since ι is only assumed to be completely isometric. The representation is called split injective iff the quotient map π admits a com- pletely positive unital linear cross section s : J → I . Two injective representations λ : I → B(H1) and µ : I → B(H2) are (spatially) equivalent iff there exists a uni- tary U ∈ B(H1 , H2) such that µ(x) = U λ(x) U ∗ . The injective representation λ is weakly contained in the representation µ iff there exists a Hilbert space K and a projection P ∈ B(H2⊗K) commuting with µ⊗ 1K such that λ is equivalent to P (µ ⊗ 1K) . The injective representation λ is stable if there exists an injective representation µ and a separable infinite dimensional Hilbert space K such that λ is equivalent to µ ⊗ 1K . We also introduce the following notion: two states ρ , ρ′ ∈ S(A) of a C∗-algebra A are (algebraically) equivalent iff there exists an element u ∈ A , kuk = 1 such that and one checks that the relation is reflexive and transitive. For example if ρu(x) = ρ(cid:0)u x u∗(cid:1) is equivalent to ρ with u as above then ρ(x) = ρu(cid:0)u∗ x u(cid:1) is equivalent to ρu . Also if ρu is equivalent to ρ for u ∈ A , kuk = 1 and (ρu)v is equivalent to ρu for v ∈ A , kvk = 1 then (ρu)v = ρuv is equivalent to ρ . More generally the states ρ , ρ′ are stably equivalent iff there exists a natural number n ∈ N such that the induced state ρp(X) = ρ(p X p) of Mn(A) is equivalent to ρ′ p(X) = ρ′(p X p) where p ∈ Mn(C) denotes the minimal projection corresponding to the left upper corner. One again checks that this relation is reflexive and transitive. It turns out that two equivalent states ρ , ρ′ of a C∗-algebra A give rise to equivalent cyclic let ξ ∈ Hρ be the canonical representations. This is seen in the following way: cyclic vector of the representation λρ associated with ρ and ξ′ ∈ Hρ′ the canonical The definition entails the identity ρ′(x) = ρ(cid:0) u x u∗(cid:1) . ρ(cid:0)u u∗(cid:1) = 1 30 U. HAAG cyclic vector of λρ′ . Then the assignment U : Hρ → Hρ′ , U (a ξ) = (a u) ξ′ preserves scalar products and by symmetry has a twosided inverse, thus defines a unitary intertwining the representations λρ and λρ′ . In case of two equivalent pure states also the converse statement is true and the element u giving the equivalence can be chosen unitary as follows from [12], Theorem 2.7.5. Any completely posi- tive unital representation can be subdivided into "cyclic" portions by considering restrictions to Hilbert subspaces K ⊆ H containing a cyclic vector ξ ∈ K with K equal to the closure of the subspace generated by all vectors {x ξ x ∈ C∗(λ(I))} . Note however that in general a subrepresentation of an injective representation need not be injective if not completely isometric. Similarly the direct sum of two injec- tive representations need not be injective itself. If ρ , σ ∈ S(X) are two states of an operator system X then ρ will be called separated from σ iff given any convex decomposition σ = λ ρ + (1 − λ) σ′ with σ′ ∈ S(X) this implies λ = 0 . The pair of states (ρ , σ) will be called separated iff ρ is separated from σ and vice versa σ is separated from ρ . An easy example of separated states is given by a pair of different pure states ρ , σ ∈ P(X) . To get a better feeling for the notion of separated states consider the case of a matrix algebra X = Mn(C) and check that in this case ρ is separated from σ if and only if the contraction of ρ to the orthogonal complement of the support of σ is nontrivial. More generally in case of a finite dimensional operator system X one has that ρ is separated from σ if and only if the restriction of ρ to the positive kernel of σ (= intersection of the kernel of σ with the positive cone X+ ) is nontrivial. If Σ ⊆ S(X) is a subset of states of X then Σ will be called positively separating for X iff given x ∈ X the condition ρ(x) ≥ 0 for all ρ ∈ Σ implies x ≥ 0 . The pair (X , Σ) will be called a separating pair iff Σ consists of mutually separated states and is positively separating for X . An injective representation λ : X → B(H) will be called separating iff there exists a subset Σ ⊆ Sν 0 of vector states such that (R, Σ) is a separated pair if R = λ(X)′′ denotes the strong closure of X , and supertransitive iff it is the direct sum of separable transitive representations (in the following the term supertransi- tive representation will usually be understood to include the notion split injective representation when speaking of injective operator systems whereas a (relatively) transitive representation of an injective C∗-algebra is not necessarily required to be injective itself unless stated explicitely). In case a transitive decomposition exists it is necessarily unique up to equivalence. In case of a supertransitive representation we assume a fixed decomposition into relatively transitive separable split injective factors and in case of a separating injective representation given a fixed subset of vector states Σ as above as being part of the representation data. Example. Suppose given an injective C∗-algebra I . Let A ⊆ I be a unital sepa- rable C∗-subalgebra and λρ an irreducible ∗-representation of A associated with some pure state ρ ∈ P(A) . Then there exists a completely isometric embedding ιρ : I(Aρ) → B(Hρ) extending λρ where I(Aρ) denotes the injective envelope of Aρ = λρ(A) . By injectivity of I and rigidity of I(A) there exists a completely positive projection ΦA : I → I with range completely isometric to I(A) which factors as a product of a completely positive retraction ρA : I ։ I(A) and a completely isometric unital embedding ιA : I(A) ֒→ I extending the embedding A ⊆ I . Then there is a completely positive map I ։ I(A) → ιρ (I(Aρ)) factoring THE JORDAN LATTICE COMPLETION 31 over ρA which extends λρ (and for simplicity is denoted by the same letter). In case that A ։ Aρ admits a completely positive cross section (i.e. if the asso- ciated extension is semisplit) one also obtains a completely positive cross section I(Aρ) → I(A) ֒→ I whose existence is due to injectivity of I(A) plus rigidity of I(Aρ) showing that the representation λρ is injective. In any case it is transitive from the fact that it extends an irreducible ∗-representation of A . This shows that any injective C∗-algebra admits a faithful supertransitive representation by sum- ming up over all such transitive building blocks with respect to arbitrary separable subalgebras of I . Clearly this representation is completely isometric restricted to an arbitrary separable subalgebra A ⊆ I hence completely isometric for I which is the inductive limit of its separable subalgebras. An injective C∗-algebra will be called separably determined iff it identifies with the injective envelope of some separable C∗-subalgebra A and separably representable iff it admits a faithful in- jective representation on separable Hilbert space. In the first case I = I(A) admits a faithful separable and separating supertransitive representation. With notation as above let ρI denote the (possibly mixed) state of I = I(A) represented by the vector state corresponding to ρ with respect to the representation λρ . Since λρ(A) ⊆ B(Hρ) is an irreducible subalgebra there exist for any two different unit vectors ξ , η ∈ Hρ a unitary u ∈ A with u ξ = η where u = λρ(u) (compare [12], Theorem 2.7.5). Then we may regard u as a unitary of I(A) and since A is represented homomorphically one gets λρ(u x u∗) = u λρ(x) u∗ which can be seen from the Stinespring dilation of λρ . Therefore any two vector states are unitarily equivalent (and separated being extensions of different pure states of A ) for I(A) . Then one may consider another inequivalent pure state ρ′ of A and repeat the whole construction with respect to ρ′ . If a state of I in the new set should coincide with some state already constructed one must have Aρ = Aρ′ and ρ ∼ ρ′ contra- dicting the assumption that ρ and ρ′ are inequivalent. Otherwise one obtains a subset of states having trivial intersection with the previous ones. Suppose there exists a vector state ρξ in the first transitive building block and a vector state ρξ′ in the second block such that ρξ and ρξ′ are not separated for I . Then their re- strictions to A are also not separated giving a contradiction since these restrictions represent different (even inequivalent) pure states of A . Proceeding in this manner one arrives at a separating supertransitive representation λA which is completely isometric restricted to A (since the atomic representation of A is faithful) hence completely isometric for I(A) . In particular the representation is injective. By separability of A it contains a faithful separable supertransitive subrepresentation. Now if I is the injective envelope of the separable subalgebra A then it is also the injective envelope of any larger separable subalgebra A ⊆ B ⊆ I so we may do the whole construction above with respect to B . For any pure state ρ of A there exists a pure state ρ′ of B restricting to ρ . Then Aρ ⊆ Bρ′ . On the other hand there may exist pure states of B which restrict to mixed states of A . For each ρ as above choose a pure state ρ′ lying above ρ and consider the correspond- ing subrepresentation λA B of λB which still is faithful hence injective because the GNS-representation space Hρ′ for ρ′ contains the representation space Hρ for ρ . Moreover choose λρ′ so that its contraction to Hρ equals λρ . Let Vρ : Hρ ֒→ Hρ′ denote the corresponding isometry. Then one sees that λA is a quotient of λA B by ρ . In this way one obtains a projective system {λAµ}µ of separating supertransitive injective representations of I with µ ≤ ν iff Aν ⊆ Aµ . the contraction Qρ V ∗ 32 U. HAAG Lemma 4. Let X be an operator system. If given ρ , σ ∈ S(X) such that ρ is separated from σ then for each ǫ > 0 there exists a positive element c ∈ X+ with ρ(c) = 1 and σ(c) ≤ ǫ . Proof. Consider the set of twodimensional operator subsystems {Xκ} , Xκ ⊆ X generated by a single positive element cκ of norm one and the order unit 1 . Then for each ǫ > 0 there exists an index κ = κ(ǫ) such that for the restrictions ρκ , σκ ∈ S(Xκ) the maximal number 0 ≤ λκ ≤ 1 with σκ = λκ ρκ + (1 − λκ) σ′ κ satisfies λκ < ǫ . For suppose this is not the case so there exists λ0 > 0 with λκ ≥ λ0 for all κ . Since any positive element is contained in some twodimensional operator subsystem this implies σ ≥ λ0 ρ whence there exists a convex decomposition σ = λ0 ρ + (1 − λ0) σ′ giving a contradiction. Given ǫ > 0 choose κ with λκ ≤ ǫ . By maximality of λκ and finite dimensionality of Xκ one gets that ρκ is separated from σ′ κ must be nontrivial and there exists a positive element c ≥ 0 in this kernel with ρ(c) = 1 and σ(c) = λκ ρ(c) ≤ ǫ (cid:3) The following Proposition is a sort of spin-off result which is not used in the sequel but is included for its own interest. One would like to improve the statement to an exact result but for most purposes in analysis 'approximately' is sufficient. If V , W are matrix ordered spaces admitting an order unit 1V and 1W respectively (are order isomorphic to operator systems, see [5] for the definition) then given ǫ > 0 a linear map φ : V → W is completely ǫ-positive with respect to the order units iff for all positive elements x ∈ Mn(V )+ , x ≤ 1V,n one has φ(x) ≥ −ǫ1W,n . A map φ : V → W of (ordered) Banach spaces is (positively) ǫ-contractive if kφ(x)k ≤ 1 + ǫ for kxk ≤ 1 (and x ≥ 0 respectively). Let X be an operator system. If Proposition 2. κ . Then the positive kernel of σ′ ψ : X ։ Mn(C) is a unital linear completely positive complete quotient map which is an operator order epimorphism, i.e. ψk(cid:0)Mk(X)+(cid:1) = Mk(cid:0)Mn(C)(cid:1)+ for each k ≥ 1 then for any given ǫ > 0 the quotient map ψ admits a completely positive linear map φǫ : Mn(C) ֒→ X with kφǫkcb ≤ n such that φǫ is ǫ-close to a cross section with respect to the completely bounded norm, i.e. k(ψ ◦ φǫ) − idMn(C)kcb ≤ ǫ . Proof. Let {Fλ}λ∈Λ denote the collection of finitedimensional operator subsys- tems of X ordered by inclusion with corresponding inclusion map ιλ : Fλ ֒→ X . Then given ǫ > 0 there exists an index λǫ such that the restriction of ψ to Fλ is approximately an operator order epimorphism, i.e. ψk(cid:0)Mk(Fλ)+(cid:1) is ǫ-normdense in Mk(cid:0)Mn(C)(cid:1)+ for each k = 1,··· , n and the completely bounded norm of the inverse of the induced completely contractive map ψλ : Gλ = Fλ (cid:14) (cid:0)ker ψ ∩ Fλ(cid:1) ∼−→ Mn(C) is bounded by 1 + ǫ for all λ ≥ λǫ , i.e. the restriction ψλ of ψ to Fλ is nearly a complete quotient map. Let Gλ be matrix ordered by the image matrix order cones of Fλ under the quotient map, i.e. the positive cone of Mk(Gλ) identifies with some ǫ-normdense subcone of Mk(Mn(C))+ by the linear completely contractive ∼−→ Mn(C) . The complete quotient map Fλ ։ Gλ induces a isomorphism Gλ completely isometric complete order isomorphic injection of preduals Gλ∗ ֒→ Fλ∗ . ∼−→ Mn(C)∗ As we will see the completely ǫ-positive identification (ψ λ )∗ : Gλ∗ −1 THE JORDAN LATTICE COMPLETION 33 (in the sense that the negative part of the image of a positive contractive element ∼−→ has norm less or equal to ǫ ) has a completely positive perturbation φλ∗ : Gλ∗ −1 Mn(C)∗ which is ǫ-close to (ψ λ )∗ with respect to the completely bounded norm. Since the (pre)dual of Mn(C) being selfdual with respect to the matrix order structure is injective, see Theorem 5.1 of [5], the completely positive map φλ∗ admits a completely positive extension φλ∗ : Fλ∗ ։ Mn(C)∗ , which can be assumed completely bounded by n as will turn out. One has the following general scheme: given a (finite dimensional) operator system F its dual F ∗ is a dually ordered Banach space as defined above (resp. matrix ordered space in the sense of [5]) and for each selfadjoint bounded functional φ ∈ F ∗ there ex- ists a decomposition as a difference of positive functionals φ = φ+ − φ− with kφk = kφ+k + kφ−k . The case of a C∗-algebra is given by the Jordan decomposi- tion. In the general case a unital completely positive representation ι : F ֒→ B(H) dualizes to a positive complete quotient map ι∗ : B(H)∗ ։ F ∗ so we may lift a self- adjoint element φ ∈ F ∗ to a selfadjoint element φ ∈ B(H)∗ with kφk = kφk and consider the image φ = φ+ − φ− of the Jordan decomposition of φ to get the result. Then there exists a canonical unitization V + of any real ordered Banach space V having the property that given any element x ∈ V there is a positive decomposition x = x+ − x− , x± ∈ V+ such that kxk ≥ max{kx+k , kx−k} . Namely consider the space V + R 1 with order structure determined by the closed convex cone generated by V+ and all positive scalar multiples of elements {1 − x x ∈ V+ , kxk ≤ 1} . Define the norm of an element y = x + λ 1 by kyk = kxk + λ and check that this makes V + a unital ordered Banach space in the sense given at the be- ginning of the section, i.e. 0 ≤ y , kyk ≤ 1 implies 1 − y ≥ 0 . The space V + admits a canonical decomposition as a direct 1-sum V + ≃ V ⊕1 C 1 since there exists a natural contractive retraction p : V + ։ V which sends x + λ1 to x , and another natural contractive retraction V + ։ C 1 sending x + λ 1 to λ 1 . The second retraction is clearly positive while the first is not. However in case of Mn(C) which is order isomorphic with its dual by the map which sends a positive functional ρ to t ∈ Mn(C)+ if ρ(x) = tr(tx) one obtains a posi- tive retraction p+ : Mn(C)+ ։ Mn(C)∗ by sending the unit element 1 to the ∗ preimage of 1n ∈ Mn(C) under this identification. Since the order isomorphic ∼−→ Mn(C) is contractive one finds that the resulting identification ι : Mn(C)∗ retraction is positive. Define a matrix order on Mn(C)+ ∗ by taking the closure of the affine sum of all cones of the form α(cid:0)Mr(Mn(C)∗)+(cid:1)+α∗ ⊆ Mk(Mn(C)+ ∗ ) with α ∈ Mk r(C) a scalar matrix and (cid:0)Mr(Mn(C)∗)+(cid:1)+ the positive cone of the uni- tization of Mk(Mn(C)∗) as defined above as positive cone for Mk(Mn(C)+ ∗ ) . One checks that this defines a matrix order restricting to the previously defined order on each subspace Mk(Mn(C)∗)+ from the relation kPi αi xα∗ i k ≤ kPi αi α∗ i k kxk and α 1r α∗ = α α∗ 1k for α ∈ Mk r(C) a scalar matrix and x ∈ Mr(Mn(C)∗) . In the same manner one extends the given matrix orders on Fλ∗ and Gλ∗ to matrix orders on F + λ∗ respectively and checks that the canonical unital extensions of the maps considered above remain completely positive or completely ǫ-positive as the case may be. Letting r(cid:0)Mn(C)+ tor systems associated with the unital matrix ordered spaces Mn(C)+ λ∗ and G+ λ∗ respectively, consider the composition of contractive resp. completely order λ∗(cid:1) denote the opera- λ∗ and G+ ∗(cid:1) , r(cid:0)F + λ∗(cid:1) and r(cid:0)G+ ∗ , F + 34 U. HAAG isomorphic maps and ρ : r(Mn(C)+ ∗ ) r−1 −→ Mn(C)+ ∗ ρ+ : r(Mn(C)+ ∗ ) r−1 −→ Mn(C)+ ∗ p −→ Mn(C)∗ p+−→ Mn(C)∗ ι−→ Mn(C) ι−→ Mn(C) , where the first map is an expansive unital complete order isomorphism, the last map is a completely contractive order isomorphism and the middle map is contractive in the first instance and completely positive unital in the second instance. Note that in each case the maps Mn(C)∗ r −−−→ r(cid:0)Mn(C)+ ∗(cid:1) , Fλ∗ r −−−→ r(cid:0)F + λ∗(cid:1) , Gλ∗ r λ∗(cid:1) −−−→ r(G+ are positively isometric. This follows since for a positive element x ≥ 0 in Fλ∗ say, its norm in r(cid:0)F + which is equal to kxk by construction of F + λ∗ . Also we may write p+ = p + π0 with π0 : Mn(C)+ ∗ → Mn(C)∗ the completely positive map with kernel Mn(C)∗ and sending the unit element to the (preimage of the) unit matrix 1n . The unital completely positive map ρ+ corresponds to a unital positive map (state) λ∗(cid:1) is given by the minimal value α ≥ 0 such that x ≤ α1 defined by linear extension of sρ+ : Mn(Mn(C)+ ∗ ) −→ C sρ+(cid:0)Ekl ⊗ x(cid:1) = 1 n (ρ+(x))kl where Ekl ∈ Mn(C) denotes the matrix unit with (i, j)-th matrix coefficient equal to δik δjl (Kronecker δ-function, see [11], chap. 6). Similarly ρ corresponds to a linear functional sρ : Mn(Mn(C)+ ∗ → C such that sρ+ = sρ + sπ0 using the corresponding formulas. Via the completely ǫ-positive unital identifications G+ linear ǫ-positive) maps ∗ ) = Mn(C) ⊗ Mn(C)+ ∗ one obtains corresponding linear (resp. λ∗ ρ : Mn(G+ + corresponding to ρ+◦ φ λ∗ and transforming correspondingly under the adjoint action of the unitary group of Mn(C) . One now defines a positive suplinear positively homogenous map λ∗) −→ C ∼−→ Mn(C)+ ρ+ , sλ sλ λ∗ and ρ◦ φ + σλ : Mn(F + λ∗) −→ C λ∗) by λ∗ is ǫ-close to sλ σλ(c) = sup(cid:8)sλ and easily checks that the restriction of σλ to G+ ρ . Define +(x) = σλ(x) + sπ0(x) where the latter is the obvious (linear positive) extension σλ to Mn(F + λ∗) of the map as above. Then there exists from the Hahn-Banach Theo- ρ (x) ≥ σλ(x) for each x ∈ Mn(F + λ∗) ρ+ is using the natural unital order isomorphic injection ιλ : Mn(G+ λ∗) ֒→ Mn(F + the formula ρ (a) (cid:12)(cid:12) ιλ(a) ≤ c , a ∈ Mn(G+ λ∗)(cid:9) ρ satisfying esλ eφ+ λ∗ : F + +(x) . This necessitates that esλ ∼→ Mn(C)+ unital and positive. Then the associated linear map λ∗ −→ Mn(C) rem a linear extension esλ ρ+ (x) = esλ and similarly esλ which extends ρ+ via the identification G+ (compare with Theorem 6.1 of [11]). But then the restriction of eφλ∗ associated is completely positive ρ of sλ ρ (x) + sπ0(x) ≥ σλ λ∗ ∗ THE JORDAN LATTICE COMPLETION 35 ρ to Fλ∗ which extends ρ is also completely positive since π0 is trivial for Fλ∗ . We still need to show that the linear map with esλ is completely bounded by n which will be the case if eφ+ λ∗ is positively contractive implying that φλ∗ restricted to Fλ∗ is positively bounded by n whence also the dual map eφλ∗ −−−→ Mn(C) −−−→ Mn(C)∗ φλ∗ : Fλ∗ ∼ is positively bounded by n and being completely positive is completely bounded by n . However the map φλ : Mn(C) −−−→ Fλ r(cid:0)F + λ∗(cid:1) r−1 −−−→ F + λ∗ eφ+ λ −−−→ Mn(C) is completely contractive since unital and completely positive with r−1 positively isometric restricted to Fλ∗ which proves the assertion. It is then obvious that φλ is ǫ-close to a cross section for the quotient map as kψλ ◦ φλ − idMn(C)kcb = kψλ ◦ φλ − idMn(C)kcb = kψλ ◦(cid:0)φλ − ψ proving the Proposition −1 λ (cid:1)kcb ≤ ǫ (cid:3) Proposition 3. Let X ⊆ B(H) be a relatively transitive injective operator sub- system with strong closure R ⊆ B(H) acting on the separable Hilbert space H and let F ⊆ Ψ := P ν(B(H)) is a subset of vector states such that L(R) ∩ J0,F is contained in the kernel of L(R) ։ L1(R) the order ideal JR,X 0,F ⊆ Lq(X) gener- 0,F = J0,F ∩ L(R) , i.e. elements of L(R) in the kernel 0,F and all elements of the form ated by r(cid:0)JR (cid:8)r(P) − r(P)(cid:9) with P ∈ Lq(R) a positive element are contained in the kernel of the natural surjection π : Lq(X) ։ L1(X) as in Proposition 1. Both the basic restriction map and the antibasic restriction map drop to convex (resp. concave) maps 0,F(cid:1) where JR of sF , and the order ideal JR,X 0,F generated by JR,X Of these r0,F is affine on the subcone L(R)(cid:14) JR 0,F generated by images of basic elements. Also both the basic and antibasic restriction maps drop to a common linear map r0,F : L(R)(cid:14) JR r0,F : L(R)(cid:14) JR r0,F : L(R)(cid:14) JR 0,F −→ L1(cid:16)L(X)(cid:14) JR,X 0,F(cid:17) , 0,F(cid:17) . 0,F −→ L1(cid:16)L(X)(cid:14) JR,X 0,F −→ L1(cid:16)L(X)(cid:14) JR,X 0,F(cid:17) which is the restriction of a corresponding ∗-homomorphism on the L1-completion. There exists a unique monotonous extension πR X : Lq(cid:0)R(cid:1) −−−→ L1(cid:0)X(cid:1) of the natural identification of the subspaces X ⊆ R ⊆ Lq(R) and X ⊆ L1(X) respectively. If X ⊆ B(H) is supertransitive and letting ΨR ⊆ Ψ denote the subset of vector states corresponding to minimal projections in R the uniquely determined ∗-homomorphism πΨR : L1(cid:0)L(cid:0)R(cid:1) (cid:14) ker sΨR(cid:1) ≃ l∞(cid:0)ΨR(cid:1) −−−→ L1(cid:0)R(cid:1) 36 U. HAAG admits a ∗-homomorphic cross section of the form extending the identity map on R . sR : L1(cid:0)R(cid:1) −−−→ L1(cid:0)L(cid:0)R(cid:1) (cid:14) ker sΨR(cid:1) Proof. Assume given the relatively transitive injective operator subsystem contain- ing a strongly dense sub-C∗-algebra A ⊆ X ⊆ R ⊆ B(H) acting on the separable Hilbert space H . From the Up-Down-Theorem (Theorem 2.4.3 of [12]) any selfad- joint element x ∈ Rsa is the infimum of a sequence (yn)n ց x with each yn the supremum of a sequence (znm)m ր yn with znm ∈ Asa ⊆ Xsa . Let X ⊆ J ⊆ R denote the operator subsystem generated by all elements which are suprema of monotone increasing nets in Xsa . Clearly if C ⊆ L(R) ⊆ L(B(H)) is a positive basic element in J0,F then rJ R (C) ⊆ L(J ) ⊆ L(B(H)) is also contained in J0,F since each element in C can be approximated from above by elements in J , i.e. rJ R (L(R) ∩ J0,F ) = L(J ) ∩ J0,F . Let us show that this is also true for arbitrary R(cid:0)P(cid:1) ∈ Lq(J ) ∩ J0,F . Since (positive) elements P = C − D ∈ L(R) ∩ J0,F , i.e. rJ R(cid:0)D(cid:1) R(cid:0)C(cid:1) − rJ R(cid:0)C(cid:1) we only need and modulo ker sF the image of C agrees with the image of rJ R(cid:0)D(cid:1) modulo ker sF . For to show that the image of D agrees with the image of rJ every vector state ρ ∈ F , given ǫ > 0 and any d ∈ D one may choose an element dρ,ǫ ∈ J with ρ(d)− ρ(dρ,ǫ) < ǫ . Then the basic element Dρ,ǫ = inf {dρ,ǫ} ∈ L(J ) R(cid:0)D(cid:1) and sρ(cid:0)Dρ,ǫ(cid:1) ≥ sρ(cid:0)D(cid:1) − ǫ . Since ǫ > 0 was arbitrary this satisfies Dρ,ǫ ≤ rJ R(cid:0)D(cid:1)(cid:1) and the reverse relation is trivial. R(cid:0)D(cid:1)(cid:1) ≥ sΨ(cid:0)D(cid:1) = sΨ(cid:0)rJ proves that sF(cid:0)rJ We must show that the basic restriction map from J to X sends J0,F ∩ Lq(J ) into ker π . The (abstract) operator system X is monotone complete by injectivity. If J + ⊆ J sa denotes the subcone of elements which can be realized as suprema of some monotone increasing net of elements in Xsa , i.e. J sa = J + − J + and given any monotonous retraction r : Lq(J ) ։ Lq(X) for the (unique by Theorem 1 below) extension of the natural embedding L(X) ⊆ L(J ) consider the image of any element y ∈ J + in L1(X) under the composition R(cid:0)P(cid:1) ≤ rJ rJ r π Lq(J ) −−−→ Lq(X) −−−→ L1(X) . If y = supλ zλ let z ∈ X be the supremum of the same monotone increasing net in X , i.e. y ≤ z , so that (π ◦ r)(inf {y}) ≤ (π ◦ r)(inf {z}) = π(inf {z}) . On the other hand (π ◦ r)(inf {y}) ≥ (π ◦ r)(inf {zλ}) = inf {zλ} for each λ . Then since is unique irrespective of the chosen retraction r . Next consider a general element y = y+ − y− ∈ J sa , y± ∈ J +. Put z = z+ − z− with z± = r(y±) which does not depend on r . Then inf {z} = supλ(cid:8)zλ(cid:9) in L1(X) one gets that the value (π ◦ r)(inf {y}) = inf {z} π(z) = (π ◦ r)(y+ − z−) ≤ (π ◦ r)(y) ≤ (π ◦ r)(z+ − y−) = π(z) A , B ∈ Lq(J ) with (π ◦ r)(A) = (π ◦ r)(A) and (π ◦ r)(B) = (π ◦ r)(B) . Then by concavity of r , convexity of r and linearity of π one gets so that π◦r is uniquely determined on J . The subset of elements (cid:8)A(cid:12)(cid:12) (π◦r)(cid:0)A(cid:1) = (π ◦ r)(cid:0)A(cid:1)(cid:9) ⊆ Lq(J ) is a subalgebra. To see this suppose given two elements (π◦ r)(cid:0)A + B(cid:1) ≥ (π◦ r)(cid:0)A(cid:1) + (π◦ r)(cid:0)B(cid:1) = (π◦ r)(cid:0)A(cid:1) + (π◦ r)(cid:0)B(cid:1) ≥ (π◦ r)(cid:0)A + B(cid:1) THE JORDAN LATTICE COMPLETION 37 hence equality since generally r ≤ r . Therefore the subset of such elements is a linear space being closed under addition and change of sign. Also (π◦ r)(cid:0)A∧ B(cid:1) = (π◦ r)(cid:0)A)∧ (π◦ r)(cid:0)B(cid:1) = (π◦ r)(cid:0)A(cid:1)∧ (π◦ r)(cid:0)B(cid:1) ≥ (π◦ r)(cid:0)A∧ B(cid:1) which again implies equality by r ≤ r . Here we have used the fact that generally r commutes with the wedge-operation and π is a lattice map (a ∗-homomorphism). Similarly one shows that (π ◦ r)(cid:0)A ∨ B(cid:1) = (π ◦ r)(cid:0)A ∨ B(cid:1) so that the subset in question is a sublattice. If P ≥ 0 is a positive element in the subset then (π ◦ r)(cid:0)P· P(cid:1) ≥ (π ◦ r)(cid:0)P(cid:1)· (π ◦ r)(cid:0)P(cid:1) = (π ◦ r)(cid:0)P)· (π ◦ r)(cid:0)P(cid:1) ≥ (π ◦ r)(cid:0)P· P(cid:1) . Since π is a homomorphism one only needs to check for this, but in fact one has the identities since by monotonicity of the square root also r(cid:0)P · P(cid:1) ≥ r(cid:0)P(cid:1) · r(cid:0)P(cid:1) , r(cid:0)P · P(cid:1) = r(cid:0)P(cid:1) · r(cid:0)P(cid:1) , 2(cid:1) ≥ r(cid:0)P(cid:1) 1 r(cid:0)P 2 , 1 r(cid:0)P · P(cid:1) ≤ r(cid:0)P(cid:1) · r(cid:0)P(cid:1) r(cid:0)P · P(cid:1) = r(cid:0)P(cid:1) · r(cid:0)P(cid:1) r(cid:0)P 2(cid:1) ≤ r(cid:0)P(cid:1) 1 2 , 1 r and r are P-maps in the sense of the Appendix below. Since the sub- i.e. set in question is a sublattice it is also closed under taking C∗-squares for arbi- trary selfadjoint elements A hence it is a subalgebra of Lq(J ) which in particular contains the subalgebra C∗(J ) ⊆ Lq(J ) generated by J . Since X is injective there exists a completely positive linear retraction υ : J ։ X which as seen above is uniquely determined and induces a functorial linear quotient lattice map Υ = L(υ) : L(J ) ։ L(X) extending (possibly nonuniquely) to a surjective ∗- homomorphism Υ : Lq(J ) ։ Lq(X) denoted by the same letter for simplicity (compare with Theorem 1). Clearly Υ sends the C∗-subalgebra C∗(J ) ⊆ Lq(J ) onto the C∗-subalgebra C∗(X) ⊆ Lq(X) generated by X . Since π ◦ r agrees with π ◦ r for elements in C∗(J ) the homomorphism Υ sends the intersection of the kernel of π with C∗(J ) into ker π ∩ C∗(X) so that Υ drops to a surjective ∗-homomorphism on the corresponding quotient C∗-algebras which by the Corollary of Theorem 2 has a ∗-homomorphic extension υ : π(cid:0)C∗(J )(cid:1) ։ π(cid:0)C∗(X)(cid:1) Υ : L1(J ) ։ L1(X) . We will see that Υ is in fact the unique monotonous extension of υ : J ։ X . To ker sΨ in the following way: for elements y ∈ J − = −J + put this end construct an order isomorphic monotonous embedding ρ : J ֒→ L(X) (cid:14) mod ker sΨ ρ(y) ≡ inf(cid:8)x ∈ X(cid:12)(cid:12) x ≥ y(cid:9) ≡ r({y}) for y ∈ J + put ρ(y) ≡ {υ(y)}− inf(cid:8)x ∈ X(cid:12)(cid:12) x ≥ υ(y)−y(cid:9) ≡ sup(cid:8)x ∈ X(cid:12)(cid:12) x ≤ y(cid:9) w,z∈(J −)+(cid:8)ρ(z)(cid:12)(cid:12) y = w−z(cid:9) w,z∈(J −)+(cid:8)ρ(w)(cid:12)(cid:12) y = w−z(cid:9)− and for general elements y ∈ J put ρ(y) ≡ inf inf mod ker sΨ mod ker sΨ 38 U. HAAG (cid:8)x ∈ X x ≥ y±(cid:9) ⊆ J equals y± . Then the statement follows considering the and check that this definition yields a monotonous map on J . Since an arbitrary element y ∈ J has a positive decomposition of the form y = y+ − y− with y± ≥ 0 , y± ∈ J − one gets a well defined map. To see that ρ is a cross section for the quotient map to L1(J ) note that L(X)∩ ker sΨ ⊆ ker π and that the image of ρ(y) under the function representation sΨ agrees with the image of y ∈ J ⊆ B(H) since for each single decomposition y = y+ − y− the infimum in J of all elements elements y± as functions on the state space S(J ) or by restriction on the space Ψ whence for two different decompositions y = y1,+ − y1,− = y2,+ − y2,− one also has y = (y1,+ ∧ y2,+) − (y1,− ∧ y2,−) in C(Ψ) so the same relation holds in L1(J ) . It is easy to see that this relation passes to the limit for arbitrary positive decompositions as above since the composition L(X) ⊆ L(J ) ։ L1(J ) is a positive linear lattice map (and commutes with taking arbitrary infima of basic elements). Therefore ρ must be an order isomorphic (linear) cross section. By monotonicity of ρ one gets r(y) = sup(cid:8)x ∈ X(cid:12)(cid:12) x ≤ y(cid:9) ≡ sup(cid:8)ρ(x)(cid:12)(cid:12) x ≤ y , x ∈ X(cid:9) ≤ ρ(y) ≤ inf(cid:8)ρ(w)(cid:12)(cid:12) w ≥ y , w ∈ X(cid:9) ≡ inf(cid:8)w ∈ X(cid:12)(cid:12) w ≥ y(cid:9) = r(y) mod ker sΨ . Then ρ extends to a monotonous map larger than r and smaller than r with respect to some (∗-homomorphic) extension of the quotient map modulo ker sΨ to r : Lq(J ) −−−→ L1(cid:0)L(X) (cid:14) ker sΨ(cid:1) Lq(X) −−−→ L1(cid:0)L(X) (cid:14) ker sΨ(cid:1) (compare the proof of Theorem 1), e.g. putting r(A) ≡ (cid:0)sup(cid:8)ρ(y)(cid:12)(cid:12) y ∈ J , y ≤ A(cid:9) ∨(cid:2)r(A)(cid:3)(cid:1) ∧(cid:2)r(A)(cid:3) . Similarly the (inverse) order isomorphic identification ρ(J ) ≃ J extends to a monotonous map q : L1(cid:0)L(X) (cid:14) ker sΨ(cid:1) ։ L1(J ) and one has q◦ r = π by uniqueness of π showing on one hand that q is surjective, and secondly uniquely determined (on the image of Lq(J ) ) , since two different choices for q would lead to two different choices for π which is impossible (compare with Proposition 1). Restricted to the subspace X ⊆ Lq(X) the composition of q with any monotonous extension Υ : L1(J ) ։ L1(X) of υ is equal to π so that again by uniqueness of π one finds that Υ is uniquely determined (on the image of Lq(X) ) and is a ∗-homomorphism since as we have seen there exists a ∗-homomorphic extension of υ . Then q must be equal to the composition of the inclusion Lq(X) ⊆ Lq(J ) with π modulo ker sΨ (on the image of Lq(X) ). It is now obvious that the diagram Lq(J ) π y L1(J ) Υ −−−→ r −−−→ L1(cid:0)L(X)(cid:14) ker sΨ(cid:1) π y L1(X) is commutative (at least on the subalgebra of elements for which π ◦ r = π ◦ r from (20) since r is generally magnifying and r is generally reducing). Let L(J +) denote the sublattice of L(J ) generated by basic elements which are infima of THE JORDAN LATTICE COMPLETION 39 some given subset {cµ cµ ∈ J +} which is bounded below so that L(J +) ⊆ L(J ) ⊆ L(R) . Similarly if J − = −J + define L(J −) ⊆ L(J ) to be the sublattice generated by basic elements which are infima of some subset {dµ dµ ∈ J −} . There is a commutative diagram r , r L(J −) −−−→Lq(X) π y π y Υ L1(J ) −−−→L1(X) . Namely the subalgebra of elements for which π ◦ r coincides with π ◦ r contains the functorial image of Λq(X) from (20) as well as all elements contained in some identfication of J in both spaces since composing j with either π◦r or π◦r yields a monotonous extension of υ which by uniqueness must be equal to Υ . Then if A ∈ L(J −) is a basic element one gets monotonous lift j : q(cid:0)L(X) / ker sΨ(cid:1) ⊆ L1(J ) → Lq(J ) extending the canonical (π ◦ r)(cid:0)A(cid:1) = (π ◦ r)(cid:0)(cid:0)A − Ac(cid:1) + (cid:0)Ac(cid:1)(cid:1) ≥ (π ◦ r)(cid:0)A − Ac(cid:1) + (π ◦ r)(cid:0)Ac(cid:1) = (π ◦ r)(cid:0)Ac(cid:1) = (π ◦ r)(cid:0)Ac(cid:1) = (Υ ◦ π)(cid:0)A(cid:1) and be an antibasic element. Then (π ◦ r)(cid:0)A(cid:1) ≤ (π ◦ r)(cid:0)A − Ac(cid:1) + (π ◦ r)(cid:0)Ac(cid:1) = (Υ ◦ π)(cid:0)A(cid:1) whence equality follows in both instances. Here we have used concavity of r and that Ac = (Ac)cc ≤ A is in the image of some monotonous lift j as above for a basic element since π factors over the quotient map modulo ker sΨ and the image of nous extension r of ρ by the assignment L(J −) modulo ker sΨ agrees with L(X) / ker sΨ . Now let B = sup(cid:8)yλ(cid:9) ∈ L(J ) On the other hand if {yλ} ⊆ J + one has (cid:2)r(cid:0)B(cid:1)(cid:3) ≥ r(cid:0)B(cid:1) on specifying a monoto- since r(cid:0){y}(cid:1) = ρ(y) if y ∈ J + . Therefore if B ∈ L(J −) one has (π◦r)(cid:0)A+B(cid:1) ≥ (π◦r)(cid:0)A(cid:1) + (π◦r)(cid:0)B(cid:1) = (Υ◦π)(cid:0)A(cid:1) + (Υ◦π)(cid:0)B(cid:1) = (Υ◦π)(cid:0)A+B(cid:1) (Υ ◦ π)(cid:0)B(cid:1) = (π ◦ r)(cid:0)B(cid:1) ≥ (π ◦ r)(cid:0)B(cid:1) . r(cid:0)A(cid:1) := sup(cid:8)ρ(y)(cid:12)(cid:12) y ∈ J , y ≤ A(cid:9) (π ◦ r)(cid:0)B(cid:1) = (Υ ◦ π)(cid:0)B(cid:1) . For a general element A + B ∈ L(J −) one gets by concavity of r so that π ◦ r is linear on L(J −) and since (π ◦ r)(A) = −(π ◦ r)(−A) the same is true for π ◦ r both of which agree with Υ ◦ π restricted to L(J −) . Note that J ⊆ L(J −) since for y ∈ J one has the identity inf {y} = inf {y+} − inf {y−} for any decomposition y = y+ − y− with y± ∈ J − . From uniqueness of π any map Lq(J −) ։ L1(J ) extending the canonical identification of J is unique by choosing some monotonous extension Lq(J −) ֒→ Lq(J ) of L(J −) ⊆ L(J ) . If two different extensions of this identification would exist for Lq(J −) then also for Lq(J ) on choosing certain extensions of these maps via the chosen embedding above. Also the inclusion L(X) ⊆ L(J −) extends to an order isomorphic linear inclusion Lq(X) ⊆ Lq(J −) and in fact this extension is unique and normal since its composition with any inclusion Lq(J −) ⊆ Lq(J ) extending L(J −) ⊆ L(J ) is necessarily unique and normal from Theorem 1. Therefore we may consider the 40 U. HAAG C∗-subalgebra eL(J −) ⊆ Lq(J −) generated by L(J −) + Lq(X) and extend all the uniqueness results and commuting diagrams as above to this larger space. With respect to some chosen extension i : Lq(cid:0)J −(cid:1) ֒→ Lq(cid:0)J(cid:1) of the natural inclusion eL(cid:0)J −(cid:1) ⊆ Lq(cid:0)J(cid:1) define the minimal and maximal retrac- tions r− , r− : Lq(cid:0)J(cid:1) −−−→ Lq(cid:0)J −(cid:1) , r−(cid:0)A(cid:1) = sup(cid:8)B(cid:12)(cid:12) i(cid:0)B(cid:1) ≤ A(cid:9) , r−(cid:0)A) = inf(cid:8)D(cid:12)(cid:12) i(cid:0)D(cid:1) ≥ A(cid:9) . Consider the sublattice L(cid:0)J +(cid:1) ⊆ L(cid:0)J(cid:1) which is linearly generated by basic el- ements C , C ⊆ J + . To each such element corresponds an element of Lq(cid:0)J −(cid:1) by taking the supremum over the set C viewed as elements of L(cid:0)J −(cid:1) which is denoted C− for simplicity, similarly the corresponding image in Lq(cid:0)J −(cid:1) of an an- tibasic element B with B ⊆ J − is denoted B− . Since i(cid:0)B−(cid:1) ≥ B and i(cid:0)C−(cid:1) ≤ C one gets any such retraction is linear for the given subspace and consequently the image Then by convexity of the maximal and concavity of the minimal retraction one gets for a general element A = C + B r−(cid:0)B(cid:1) = r−(cid:0)B(cid:1) = B− . r−(cid:0)C(cid:1) = r−(cid:0)C(cid:1) = C− , r−(cid:0)A(cid:1) ≤ r−(cid:0)C(cid:1) + r−(cid:0)B(cid:1) = r−(cid:0)C(cid:1) + r−(cid:0)B(cid:1) ≤ r−(cid:0)A(cid:1) ⊆ Lq(cid:0)J −(cid:1) is a quotient space of L(cid:0)J +(cid:1) and any retraction showing that any retraction for i is uniquely determined on L(cid:0)J +(cid:1) . Therefore space L(cid:0)J +(cid:1)− + Lq(cid:0)X(cid:1) . Now J − is a unital affine subcone factors over the image space L(cid:0)J +(cid:1)− J− ⊆ L(cid:0)J −(cid:1) is given by the element υ(c) ∈ X ⊆ L(cid:0)J −(cid:1) one easily finds that L1(cid:0)J −(cid:1) ≃ L1(cid:0)X(cid:1) . From (the Remark before) Proposition 1 there is a unique of J so the space L1(J −) is well defined and is an injective commutative C∗- algebra (see the Remark above). Since the lower complement of an element c ∈ rX : L(cid:0)J +(cid:1) + Lq(cid:0)X(cid:1) −−−→ Lq(cid:0)X(cid:1) surjective ∗-homomorphism π− : L(cid:0)J −(cid:1) −−−→ L1(cid:0)J −(cid:1) ≃ L1(cid:0)X(cid:1) extending the map υ : J − ։ X which is a complete order epimorphism. Then there is a commutative diagram L(cid:0)J −(cid:1) r , r π− y L1(cid:0)X(cid:1) ∼ −−−→Lq(cid:0)X(cid:1) π y −−−→L1(cid:0)X(cid:1) since π− obviously factors over Υ whence π− = π ◦ Υ . Being a complete order epimorphism this map has a unique extension to Lq(cid:0)J −(cid:1) . Then from the argument THE JORDAN LATTICE COMPLETION 41 above there also is a commutative diagram L(cid:0)J +(cid:1) r , r π y L1(cid:0)J(cid:1) Υ −−−→Lq(cid:0)X(cid:1) π y −−−→L1(cid:0)X(cid:1) . For any D ∈ L(J ) there exists a larger element C ∈ L(J +) , C ≥ D with C − D ∈ ker sΨ since each element d ∈ D can be pointwise approximated from above on Ψ by elements in J + . If C ∈ ker π one concludes that also r(cid:0)C(cid:1) is in the kernel of π . For a general positive element P = C − D ∈ L(J ) ∩ ker π first replace C by the corresponding larger element C+ ∈ L(J +) . Then for any vector state ρ ∈ Ψ , given ǫ > 0 there exists Dρ,ǫ ∈ L(J −) with Dρ,ǫ ≤ D and sρ(cid:0)Dρ,ǫ(cid:1) ≥ sρ(cid:0)D(cid:1) − ǫ proving that P +−P =(cid:0)C+ −C(cid:1)−(cid:0)r−(cid:0)D(cid:1)−D(cid:1) ∈ ker sF = \ρ∈F ker sρ =⇒ r(cid:0)P +(cid:1) ∈ ker π P + = C+ − r−(cid:0)D(cid:1) ≥ C − D = P since r− must be reducing on basic elements. Therefore r(cid:0)P(cid:1) ∈ ker π proving the first assertion by the commutative diagram with Lq(cid:0)J −(cid:1) r , r π y L1(cid:0)J(cid:1) −−−→Lq(cid:0)X(cid:1) π y −−−→L1(cid:0)X(cid:1) . 0,F(cid:17) 0,F −−−→ L1(cid:16)L(X)(cid:14) JR,X 0,F −−−→ Lq(X)(cid:14) JR,X Υ r0,F : L(R)(cid:14) JR By convexity the (homogeneized) basic restriction map r drops to a convex map where the map to the right is any convenient (∗-homomorphic) extension of the natural quotient map. The restriction of basic elements from L(R) to L(X) fac- tors over the restriction to L(J +) the latter being linear (affine) on the subcone L(R) ⊆ L(R) of (positive) basic elements and implementing an affine order isomor- phism modulo JR 0,F thereby agreeing on basic elements with the induced antibasic restriction map rJ R from L(R) to Lq(J ) modulo JR 0,F which is well defined be- cause rJ R (A) ≤ rJ R (A + E) ≤ rJ R (A) + rJ R (E) (compare the Remark after Proposition 1) and both sides of the inequality are congruent modulo J0,F for E ∈ J0,F . Now r0,F restricted to L(R)(cid:14) J0,F factors over the basic restriction map r+ 0 : L(J +)(cid:14) J0,F −−−→ L(X)(cid:14) JR,X 0,F induced by basic restriction from L(J +) to L(X) which is seen to coincide with the positive linear map induced by Υ : L(J +) ։ L(X) where Υ = L(υ) . This 0,F ⊆ ker π . proves that the map r0,F is affine on the subcone L(R)(cid:14) J0,F with JR,X 42 U. HAAG 0,F is a positive Let us show that JR,X element then as shown above 0,F ∈ ker π . If [P] = [C] − [D] ∈ L(R) (cid:14) JR so that r0,F(cid:0)[C](cid:1) ≡ r0,F(cid:0)[C](cid:1) mod ker π , r0,F(cid:0)[D](cid:1) ≡ r0,F(cid:0)[D](cid:1) mod ker π r0,F(cid:0)[P](cid:1) ≤ r0,F(cid:0)[C](cid:1)−r0,F(cid:0)[D](cid:1) ≡ r0,F(cid:0)[C](cid:1)−r0,F(cid:0)[D](cid:1) ≤ r0,F(cid:0)[P](cid:1) mod ker π. Therefore JR,X 0,F ⊆ ker π and modulo this order ideal both r0,F and r0,F agree on images of basic elements where they are affine. Then suppose given two general positive elements [P] = [C] − [D] , One gets [Q] = [E] − [F ] ∈ L(R)(cid:14) JR 0,F . r0,F(cid:0)[P] + [Q](cid:1) = r0,F(cid:0)[C + E] − [D + F ](cid:1) ≥ r0,F(cid:0)[C + E](cid:1) − r0,F(cid:0)[D + F ](cid:1) ≡ r0,F(cid:0)[C](cid:1) + r0,F(cid:0)[E](cid:1) − r0,F(cid:0)[D](cid:1) − r0,F(cid:0)[F ](cid:1) mod JR,X 0,F and the reverse inequality follows from convexity. So r0,F and r0,F drop to a common map which is affine on positive elements hence linear by homogeneity. Then the common positive linear map ≥ r0,F(cid:0)[P](cid:1) + r0,F(cid:0)[Q](cid:1) (π ◦ r) = (π ◦ r) : L(cid:0)R(cid:1) −−−→ L1(cid:0)X(cid:1) is a complete order epimorphism being equal to the composition of the complete order epimorphisms π : L(cid:0)R(cid:1) −−−→ L1(cid:0)R(cid:1) , L1(cid:0)R(cid:1) −−−→ L1(cid:0)X(cid:1) where the second map is any monotonous extension of the identity map of X (hence necessarily surjective) whence it has a unique monotonous extension πR X : Lq(cid:0)R(cid:1) −−−→ L1(cid:0)X(cid:1) . Since r and r are P-maps in the sense of the Appendix below (they send pro- jections to projections) one easily finds that πR X is a ∗-homomorphism whence If there also exists a ∗-homomorphic extension of r0,F to L1(cid:0)L(cid:0)R(cid:1) / JR 0,F(cid:1) . X is supertransitive then R = Qj B(Hj) is a product of type I factors and if ΨR = P ν(R) ⊆ P ν(B(H)) = Ψ denotes the subset of vector states corresponding to minimal projections in R then the order ideal J0,ΨR ∩ L(R) coincides with the kernel of the function representation sΨR : L(R) → l∞(ΨR) from Theorem 1, so that also J0,ΨR ∩ L(J ) coincides with the kernel of sΨR : L(J ) → l∞(ΨR) . Again from Theorem 1 one has an isomorphism L1(cid:0)L(cid:0)R(cid:1) (cid:14) ker sΨR(cid:1) ≃ l∞(cid:0)ΨR(cid:1) ≃ L1(cid:0)L(cid:0)J(cid:1) (cid:14) ker sΨR(cid:1) . Consider the isomorphic images of L(cid:0)J(cid:1) and L(R) modulo ker sΨR . Any bound- edly generated element of L(R) (the subspace of differences of boundedly generated basic elements, see above) defines a uniformly continuous function in C(ΨR) with respect to the induced metric on ΨR as subspace of minimal projections in R since generally the image of a basic element in l∞(ΨR) is an upper semicontinuous function, but in order to get a discontinuity or even a nonuniformly continuous function one needs elements of arbitrary large norm which are not dominated by any element in a given bounded subset, i.e. if A ∈ L(R) is boundedly generated, THE JORDAN LATTICE COMPLETION 43 then there exists for every ǫ > 0 a δ > 0 such that ksρ(A)− sρ′(A)k < ǫ whenever kρ− ρ′k < δ . On the other hand each continuous function in C(ΨR) is equal to the image of a complemented basic (and antibasic) element in L(R) . To see this choose for the given continuous function f ∈ C(ΨR) , given ǫ > 0 and for any given point ρ ∈ ΨR an element in xρ,f ∈ R with ρ(xρ,f ) = f (ρ) and ρ′(xρ,f ) ≥ f (ρ′) − ǫ which is easily achieved on letting ρ correspond to the eigenvector for the minimal eigenvalue of xρ,f and choosing xρ,f sufficiently large on the orthogonal comple- ment of ρ . Then the image of the basic element Af,ǫ = inf {xρ,f} in l∞(ΨR) satisfies Similarly one can pointwise approximate the function f from below by an antibasic element Bf,ǫ satisfying sΨR(cid:0)Af,ǫ(cid:1) ≤ f ≤ sΨR(cid:0)Af,ǫ(cid:1) + ǫ 1 . sΨR(cid:0)Bf,ǫ(cid:1) − ǫ 1 ≤ f ≤ sΨR(cid:0)Bf,ǫ(cid:1) by the corresponding procedure which also entails Bf,ǫ − ǫ 1 ≤ Af,ǫ + ǫ 1 c cc f,ǫ ≤ B so that f can be uniformly approximated by images of complemented basic (an- tibasic) elements which moreover approximately agree with the images of their since any element b ≤ Bf,ǫ satisfies the relation b ≤ a + 2 ǫ 1 for every element a ≥ Af,ǫ by construction. Therefore one also has the relations Bf,ǫ ≤ B f,ǫ ≤ Af,ǫ + 2 ǫ 1 , Bf,ǫ − 2 ǫ 1 ≤ (cid:0)Af,ǫ(cid:1)c ≤ (cid:0)Af,ǫ(cid:1)cc ≤ Af,ǫ =⇒ sΨR(cid:0)(cid:0)Af,ǫ(cid:1)c(cid:1), sΨR(cid:0)(cid:0)Af,ǫ(cid:1)cc(cid:1) ≤ f ≤ sΨR(cid:0)(cid:0)Af,ǫ(cid:1)c(cid:1)+ǫ 1, sΨR(cid:0)(cid:0)Af,ǫ(cid:1)cc(cid:1)+ǫ 1 complements. Thus if f = sΨR(cid:0)A(cid:1) for some A ∈ L(R) then the images of π(A) the algebra of (bounded) continuous functions C(cid:0)ΨR(cid:1) is contained in sΨR(cid:0)L(R)(cid:1) (resp. sΨR(cid:0)L(J )(cid:1) ) and can also be identified with a subalgebra of L1(R) (resp. sJ : L1(cid:0)J(cid:1) −−−→ L(cid:0)J(cid:1) (cid:14) ker sΨR sR : L1(cid:0)R(cid:1) −−−→ L(cid:0)R(cid:1) (cid:14) ker sΨR , in l∞(ΨR) using both the convex and the concave lift cv, cc : L1(R) → L(R) composed with sΨR must agree with each other and with f = sΨR(A) . Then L1(J ) ). From the Corollary of Theorem 2 there exist ∗-homomorphic extensions of this identification which are cross sections for π (cid:3) Remark and Definitions. (i) Let λu : A → B(Hu) denote the universal represen- tation of the abstract C∗-algebra A with associated (canonical) order isomorphic embedding Λu : Lq(A) ֒→ Lq(B(Hu)) (see Theorem 1). Writing Lν 0(B(Hu)) = L1(cid:0)L(B(Hu)) (cid:14) ker rν(cid:1) there is a (semicanonical) positive linear surjection πν 0 : Lq(B(Hu)) ։ Lν 0(B(Hu)) extending the natural quotient map modulo the kernel of rν . Consider the composite map Γu : Lq(A) Λu −−−→ Lq(B(Hu)) −−−→ Lν 0(B(Hu) πν 0 which is injective and order isomorphic (any state of A is represented by a vec- tor state of B(Hu) ). Then again Γu admits a positive linear retraction Ψ : 0(B(Hu)) ։ Lq(A) and the (commutative) C∗-product in Lq(A) may be defined Lν 0(B(Hu)) by the formula P· P = Ψ(cid:0)Γu(P)· Γu(P)(cid:1) . recurring to the product in Lν For a basic positive element D ∈ L(B(Hu)) Theorem 1 below gives πν 0 (D) = 0 (D2) . Therefore if C ∈ L(A) is a basic positive element then C · C ≤ C2 since πν 0 (D)·πν 44 U. HAAG C · C = Ψ((Cγ)2) where Cγ = Γu(C) which is represented by the basic positive element inf {λu(c) c ∈ C} . Then the element (Cγ)2 is smaller or equal to the element represented by the positive set {λu(c)2 = λu(c2) c ∈ C} which by the property of Ψ being a retraction for Γu is mapped back to C2 . (ii) For a C∗-algebra A let IA ⊆ L(A) denote the order ideal which is the restric- tion of the order ideal in Lq(A) generated by all positive elements {C2 − C · C} with C ∈ L(A) a positive basic element. For simplicity IB(H) will usually be abbreviated to I . Define LI(A) to be the L1-completion of the quotient lattice L(A)(cid:14) IA and πI : Lq(A) −→ LI(A) = L1(cid:0)L(A)(cid:14) IA(cid:1) any chosen positive linear extension of the natural quotient map. Then the basic squaring operation (1) drops to LI(A) where it coincides with the C∗-square. At this point it is not clear whether this quotient lattice is nontrivial or what it looks like for a general C∗-algebra but it will be shown in Theorem 1 that the quotient map to LI(A) factors the quotient map to L1(A) so the definition makes good sense. (iii) Let X ⊆ B(H) be an operator subsystem and given any collection of vector 0 the lattice L1(cid:16)L(X)/J0,F(cid:17) embeds into L1(cid:16)B(H))/J0,F(cid:17) which states F ⊆ Sν is isomorphic to l∞(F ) as will be shown in the proof of Theorem 1. Let πX L(X) ։ L(X)(cid:14) J0,F denote the quotient map. For each ρ ∈ F consider the basic element E ρ,X ∈ L(X) which is the infimum of the set {c ∈ X c ≥ 0 , ρ(c) ≥ 1} . Define a monotonous increasing normal map 0,F : by the assignment jX : l∞(F ) −→ Lq(X) α 7→ jX(α) = sup ρ∈F α(ρ)E ρ,X if α ∈ l∞(F ) is a realvalued function. Restricted to images of positive basic elements jX is a lift for π0,F . To see this note that for a positive basic element C ∈ L(X)+ one has supρ(cid:8)rν ρ (C)E ρ,X(cid:9) ≤ C by construction of the basic elements {E ρ,X} so the result follows from monotonicity. (iv) One often encounters the following situation: the quotient map L(V) ։ L(V) / J modulo an order ideal J ⊆ L(V) extends (mostly nonuniquely) to a ∗-homomorphism of L1-completions Lq(cid:0)V(cid:1) −−−→ L1(cid:0)L(cid:0)V(cid:1) (cid:14) J(cid:1) which however need not be surjective unless the original quotient map admits a monotonous splitting. If this is not the case we say that the extended map is an approximate quotient map. Theorem 1. For any order isomorphic unital embedding of operator systems V ⊆ W ⊆ B(H) the induced functorial positive linear map L(V) ֒→ L(W) extends uniquely to an injective normal ∗-homomorphism Lq(V) −→ Lq(W) . For any state σ ∈ S(V) the induced functional sσ : L(V) −→ R THE JORDAN LATTICE COMPLETION 45 extends (nonuniquely) to a multiplicative functional on Lq(V) . Assume that X ⊆ B(H) is an operator subsystem. Let F ⊆ Sν vector states. Put LF J0,F ⊆ L(B(H)) the order ideal given by the intersection of all kernels rν ρ , ρ ∈ F . In case that F = Sν 0(X) etc.. For any inclusion of operator subsystems X ⊆ Y ⊆ B(H) there exist ∗-homomorphic extensions of the canonical embeddings/quotients of L(X) , L(Y) , L(B(H)) making a commutative diagram 0 (X) = L1(cid:0)L(X) / J0,F ,X(cid:1) where J0,F ,X = J0,F ∩ L(X) with 0 we simply write Lν 0 be any collection of πF Lq(X) 0,Xy LF 0 (X) I −−−→ Lq(Y) πF 0,Yy I F 0 −−−→ LF 0 (Y) 0 (B(H)) ≃ l∞(F ) and in case that X is separating for F (i.e. 0 (B(H)) . The surjection π : Lq(X) ։ L1(X) factors over πν 0,X . with I the functorial normal ∗-homomorphism induced by the inclusion X ⊆ Y . One has LF if the images of F in S(X) are mutually separated) the map I F is an isomorphism 0 (X) ≃ LF LF Moreover if A ⊆ B(H) is an irreducible C∗-subalgebra the image of the basic squaring operations (1) , (13) in Lν 0(A) agrees with the C∗-square and the C∗- product X · X ∗ for any element of the form X = C + i D respectively. More generally for any (abstract) separable C∗-algebra A the L1-completion Lr 0(A) = L1(cid:0)r(cid:0)L(A)(cid:1)(cid:1) of the quotient of L(A) modulo the kernel of 0 is a commutative W ∗-algebra isomorphic to l∞(P(A)) such that the basic squaring operations (1) , (13) drop to this algebra where they coincide with the correspond- ing C∗-squares and -products. The canonical surjection L(A) ։ r(L(A)) extends (nonuniquely) to a ∗-homomorphism r : L(A) −→ l∞(cid:0)P(A)(cid:1) πr 0 : Lq(A) −−−→ Lr 0(A) ≃ l∞(P(A)) factoring π and factoring over πI . Suppose that I is an injective C∗-algebra. For any (equivalence class of) multi- plicative injective representation µ : I −→ B(H) there exists a canonical order ideal Jµ 0 ⊆ L(I) such that the basic squaring operations (1) , (13) drop to the quotient modulo the ideal Jµ 0 which is a C∗-algebra where they coincide with the image of the corresponding C∗-squares and -products. The quotient map may be extended (nonuniquely) to a ∗-homomorphism πµ 0 : Lq(I) −−−→ Lµ In particular let JI = Tµ Iµ representations of I up to equivalence and put 0(cid:1) . 0 (I) = L1(cid:0)L(I) / Jµ 0 where {µ} ranges over all multiplicative injective L0(I) = L1(cid:0)L(I) / JI(cid:1) . The canonical quotient maps modulo JI may be extended (nonuniquely) to a ∗- homomorphism factoring π and factoring over πI such that for any multiplicative injective repre- sentation µ of I ( a suitable choice of) πµ 0 factors over π0 . π0 : Lq(I) −−−→ L0(I) 46 U. HAAG For any C∗-algebra A the Jordan squaring operations (1) , (9) , (10) , (11) , (13) drop to L1(A) where they agree with the corresponding C∗-squares and -products, i.e. for any A ∈ Lq(A) one has π(A)2 = π(A2) . Moreover if P1/2 denotes the C∗- squareroot of the positive element P ≥ 0 and √P is the Jordan lattice squareroot of P as above then π(P2) = π(P)2 , (21) In particular, every positive linear map φ : C → Lq(A)c = Lq(A)⊗C with domain a (complex) C∗-algebra C satisfies the Schwarz inequality π(pP) ≤ π(P1/2) . mod ker π φ(x x∗) ≥ φ(x0)2 + φ(x1)2 (22) with respect to the squaring operation in Lq(A) as defined above and the C∗- product in C modulo the kernel of π where x0 = (x + x∗)/2 , x1 = i(x∗ − x)/2 . Also for any selfadjoint element a ∈ A and a = {a} = a+ − a− the minimal basic positive decomposition of a in L(A) the image of a+ ∧ a− in L1(A) is trivial (so π(a) = π(a+) − π(a−) is the minimal positive decomposition of a in L1(A) ). Proof. Let V ⊆ W ⊆ B(H) be an order isomorphic unital embedding of operator systems and ιq : Lq(V) ֒→ Lq(W) any monotonous extension of the functorial linear map ι : L(V) ֒→ L(W) . Let A = supλ {Aλ} ∈ Lq(V) be the supremum of certain basic elements and note that since any element of Lq(V) is the supremum of elements in L(V) and every element is the supremum of basic elements, the same is true for an arbitrary element A ∈ Lq(V) . If Aι λ} ∈ Lq(W) . Then A is the infimum of all antibasic elements λ = ι(Aλ) consider the supremum A = supλ {Aι A = C − D = sup(cid:8)C − sup{d} (cid:12)(cid:12) d ∈ D(cid:9) A = inf(cid:8)Bµ (cid:12)(cid:12) Bµ ∈ L(W) , A ≤ Bµ(cid:9) larger or equal to A . From monotonicity one gets A ≤ ιq(A) so that any monot- onous retraction rq : Lq(W) ։ Lq(V) for ι must satisfy rq(A) = A . Consider the functorial retraction r induced by restriction of antibasic elements as in part (ii) of the Remark after Proposition 1. Then (ι ◦ r)(Bµ) ≤ Bµ for each index µ which by monotonicity implies ιq(A) = (ιq ◦ r)(A) ≤ A hence equality. Thus the extension ιq is necessarily unique and (increasing) normal since the image of the supremum of an arbitrary set {Aλ} is seen to agree with the image of the supremum of a corresponding subset of basic elements, each of which is dominated by some Aλ and this image in turn agrees with the supremum of the images. Similarly one finds that ιq is decreasing normal, hence normal. Now let a state σ ∈ S(V) be given. It is easy to see that the kernel of the map sσ : L(V) −→ R is positively generated. Indeed, if sσ(C) = sσ(D) then this common value is equal to sσ(C ∧D) so that the positive elements C − C ∧D and D − C ∧D are contained in the kernel of sσ . Thus sσ is a lattice map which from the Corollary of Theorem 2 can be extended to a corresponding lattice map sσ : Lq(V) ։ R the kernel of which is again positively generated hence must be a ∗-ideal of the underlying commutative C∗-algebra (since the kernel of a positive linear map of C∗-algebras THE JORDAN LATTICE COMPLETION 47 is a ∗-ideal whenever positively generated). Then sσ is uniquely determined on the sub-C∗-algebra generated by L(V) . From this one easily sees that the normal linear injection ιq : Lq(V) ֒→ Lq(W) as above which certainly is a lattice map is in fact a ∗-homomorphism since every state of W determines a state of V so that the composition extends to a multiplicative embedding together with a compatible multiplicative (lattice map) extension L(cid:0)V(cid:1) ֒→ L(cid:0)W(cid:1) s −−−→ l∞(cid:0)S(W)(cid:1) Lq(cid:0)V(cid:1) −−−→ l∞(cid:0)S(W)(cid:1) Lq(cid:0)W(cid:1) −−−→ l∞(cid:0)S(W)(cid:1) from the Corollary of Theorem 2. Now let X ⊆ B(H) be an operator subsystem. From Lemmas 1 and 3 one gets that Lν 00(B(H)) = L(B(H))/J0 admits the structure of an associative algebra with corresponding Jordan product induced by the squaring operation of basic positive elements, i.e. π00(C) · π00(D) + π00(D) · π00(C) = π00((C + D)2 − C2 − D2) , 00(B(H)) → l∞(P ν(B(H))) is an algebra homomor- and that the induced map l : Lν phism. Since the range is a commutative C∗-algebra and l is injective one finds that 00(B(H)) is commutative. If rν (C) ≥ rν (D) then π00(D) = π00(C ∧ D) which Lν shows that l is order isomorphic, so as far as the order and algebra structure are concerned Lν 00(B(H)) may be identified with a dense subalgebra of a selfadjoint real commutative C∗-algebra, i.e. realvalued continuous functions on a compact space X . Since the norm is monotone for positive elements and l is contractive as well as order isomorphic one deduces the inequalities −kAk · kBk ≤ A · B ≤ kAk · kBk from the corresponding relations in C(X) where A , B ∈ Lν selfadjoint elements. This implies kA·Bk ≤ kAk·kBk , i.e. Lν algebra. The inequality 00(B(H)) are arbitrary 00(B(H)) is a Banach −kP + Qk ≤ P − Q ≤ kP + Qk valid for any two positive elements P , Q ≥ 0 implies kP − Qk ≤ kP + Qk so that from Theorem 4.2.5 of [1] Lν 00(B(H)) is isometrically isomorphic to a commutative C∗-algebra. Then also l being injective must be an isometry, so Lν 00(B(H)) ≃ CR(X) ⊆ l∞(Pν (B(H))) . From injectivity of l∞(Pν(B(H))) there exists a monotonous (of course even linear positive) extension l : Lν 0(cid:0)B(H)(cid:1) = 00(cid:0)B(H)(cid:1)(cid:1) −→ l∞(Pν (B(H))) of l . Every pure normal state ρ ∈ P ν(B(H)) L1(cid:0)Lν notes the following: if φ =Pn and ρ′(x) = 1 . Then suppose given a linear combination φ = Pn defines a pure state rρ of Lν 00(B(H)) . Considering the closure of the linear span 00(B(H))∗ denote this Banach space by V . One first of the functionals {rρ} in Lν i=1 λi rρi ≥ 0 is a positive functional (on Lν 00(B(H)) ) which is a finite linear combination of the {rρ} then each coefficient λi ∈ R for i = 1 , ··· , n must be positive. For given any two different pure normal states ρ , ρ′ ∈ P ν(B(H)) and ǫ > 0 there exists x ≥ 0 in B(H) such that ρ(x) ≤ ǫ i=1 λi rρi ≥ 0 with λ1 < 0 say. Without loss of generality we may assume that λj > 0 for j = 2 , ··· , n . For given ǫ < −λ1 choose elements {xj}j=2,··· ,n such that 48 U. HAAG ρj(xj) ≤ ǫ , ρ1(xj) = 1 and consider the positive element P of Lν 00(B(H)) defined by the infimum of the set {xj j = 2 , ··· , n} . Then clearly φ (P) ≤ λ1 + ǫ < 0 contradicting the fact that φ ≥ 0 . In particular the vectors {rρ} are linear in- dependent and the onedimensional projection ωρ : V → V defined by continuous extension of the map given by the formula ωρ(Xκ kXκ λκ rκ) = λρ rρ λκ rκk = Xκ λκ , on the linear span of the {rρ} is positive for every ρ ∈ P ν(B(H)) . The argument also shows that 0(B(H)) −→ V ∗ is necessarily surjective. If x ∈ V ∗ i.e. V ≃ l1(P ν(B(H))) so that V ∗ ≃ l∞(P ν (B(H))) . As we will see the extension l : Lν + is any positive element then x is the supremum of the elements {x(rν κ) = λρ . Now each ηρ is in the image of Lν 00(B(H)) and, if X ⊆ B(H) is transitive also in the image of Lν 00(X) = L(X) / J0,X . This follows from the fact that for every pure normal state κ 6= ρ there exists an element xκ ∈ X+ with κ(xκ) = 0 , ρ(xκ) = 1 . Putting E ρ,X = inf {c ∈ X+ ρ(c) ≥ 1} the image of the basic positive element E ρ,X is clearly equal to ηρ (transitive case). Now it is easy to see that the positive ρ )ηρ} where ηρ(Pκ λκ rν 0(cid:1) of functions vanishing injective envelope of the commutative C∗-algebra c0(cid:0)Sν at infinity is equal to l∞(cid:0)Sν 0(cid:1) whence the latter is also the injective envelope of 00(cid:0)B(H)(cid:1) ⊇ c0(cid:0)Sν 0(cid:1) in the transitive case). Since Lν any positive linear extension of rν to C∗(L(B(H))) ⊆ Lq(B(H)) must be a C∗- homomorphism from the Schwarz inequality and the fact that C · C ≤ C2 for a positive basic element there always exists a multiplicative positive linear extension to Lq(B(H)) by the Corollary of Theorem 2. We denote any chosen (nonunique) such extension by 00(cid:0)X(cid:1) ⊇ c0(cid:0)Sν 0(cid:1) (resp. Lν πν 0 : Lq(B(H)) ։ l∞(P ν(B(H))) = Lν 0(B(H)). 0(X) ֒→ Lν 00(X) ֒→ Lν 0,X) ⊆ ker πν 0(B(H)) ։ Lν 00(B(H)) where Lν C∗(cid:0)L(X)(cid:1) ⊆ Lq(cid:0)X(cid:1) . Choose any ∗-homomorphic extension Lν and note that it restricts canonically to a ∗-homomorphism Lq(X) → Lν 0(B(H)) via the unique normal multiplicative embedding Lq(X) ֒→ Lq(B(H)) yielding a ∗- homomorphic injection Lν 00(X) denotes the image of 0(B(H)) 0(X) which 0,X to be the 0 ◦ I . One may if necessary change the original homomorphism 0 using the methods as above starting from of this map. Then it admits an extremal retraction pX : Lν is a ∗-homomorphism from the Corollary of Theorem 2. Define πν composition pX ◦ πν πν 0 in order that I(ker πν the canonical ∗-homomorphism 0,X(cid:1)(cid:1) −−−→ L00(cid:0)B(H)(cid:1) C∗(cid:0)L(cid:0)B(H)(cid:1) + I(cid:0)ker πν sending I(cid:0)ker πν 0,X(cid:1) + J0 to zero. Then assume this compatibility condition. The image of Lq(cid:0)X(cid:1) under the composition πν 0,X(cid:0)Lq(X)(cid:1) ⊆ 0(cid:0)X(cid:1) so again from the Corollary of Theorem 2 one may find a ∗-homomorphic Lν extension I0 : Lν 0(X) −→ Lν 0(B(H)) making a commutative diagram with the other maps as in the Theorem. The same line of arguments extends to the case (X) ⊆ Lν,F of Lν,F (B(H)) for an arbitrary subset F ⊆ P ν(B(H)) . In particular (B(H)) ≃ l∞(F ) and the same holds for Lν,F Lν,F (X) in case that X is separating 0 ◦ I is ∗-isomorphic to πν 0 0 0 0 THE JORDAN LATTICE COMPLETION 49 0 by an argument as before because Lν,F for the subset F ⊆ Sν minimal projection of l∞(cid:0)F(cid:1) in this case. The statement that π factors over πν 0,X is obvious from Proposition 1. This proves the existence of a commutative diagram as above in the case Y = B(H) . However the argument in case of an arbitrary inclusion X ⊆ Y is much the same. Now let A ⊆ B(H) be an irreducible C∗-subalgebra. Then the natural inclusion ιq : Lq(A) ֒→ Lq(B(H)) drops to an isomorphism (X) contains every 0 Lν 0(A) = L1(cid:0)L(A) / J0,A(cid:1) ∼−→ Lν 0(B(H)) ≃ l∞(P ν(B(H))) and as noted in the proof of Lemma 1 the basic squaring operations (1) , (13) agree with the corresponding C∗-squares and -products for the image of any positive basic elements in L(B(H)) implying in particular that for any positive basic element C ∈ L(A) the image of ιq(C)2 agrees with the image of C ·C modulo J0 . Then we only need to show the congruence of ιq(C2) and ιq(C)2 , and similarly the congruence of ιq(X·X ∗) and ιq(X )·ιq(X )∗ modulo J0 where X = C + i D ∈ L(A)c . In the first case write If we can show the estimate 0 ≤ ιq(C2) − ιq(C)2 = sup = sup a (cid:8)ιq(C2) − inf {a2} (cid:12)(cid:12) a ∈ ιq(C)c(cid:9) c (cid:8) inf {c2 − a2} (cid:12)(cid:12) c ∈ C(cid:9) (cid:12)(cid:12)(cid:12) a ∈ ιq(C)co . a n inf c (cid:8) inf {c2 − a2} (cid:12)(cid:12) c ∈ C(cid:9) ≤ ǫ 1 inf inf modulo J0 for each given a ∈ ιq(C)c and ǫ > 0 we get the result since ǫ may be chosen arbitrary small and the image function of ιq(C2)− ιq(C)2 in l∞(P ν(B(H))) is the pointwise supremum of the functions corresponding to basic elements as above fixing a . Then it is sufficient to check that inf c n inf {P (c2 − a2) P} (cid:12)(cid:12) c ∈ C for any finitedimensional orthogonal projection P ∈ B(H) . From the Schwarz inequality one gets P a2 P ≥ (P aP )2 = a2 P and it is sufficient to check νo ≤ inf c (cid:8) inf {P c2P − a2 c (cid:8) inf {P (c2 − a2) P} (cid:12)(cid:12) c ∈ C(cid:9) ≤ ǫ P P} (cid:12)(cid:12) c ∈ C(cid:9) ≤ ǫ P . Since a ∈ ιq(C)c there exists c ∈ C with c ≤ a which implies cP ≤ aP and invoking Kadison's transitivity theorem revisited, Theorem 2.7.5 of [12], there exists a positive element d ∈ A+ with cP + dP = aP . Then for given ǫ > 0 there exists a finitedimensional projection Q ≥ P with kP (c + d) (1 − Q)k ≤ ǫ and we may choose a positive element hǫ ∈ B(H)+ with kP hǫ Pk ≤ ǫ 3 kak and P (c + d + hǫ) (Q − P ) = (Q − P ) (c + d + hǫ) P = 0 . Once more invoking Theorem 2.7.5 of [12] there exists a positive element dǫ ∈ A+ of norm khǫk with Q dǫ Q = Q hǫ Q and Q dǫ (1 − Q) = 0 = (1 − Q) dǫ Q . Putting cǫ = c + d + dǫ ∈ C one finds 3 ǫ + ǫ2 so that altogether that kP c2 k(P cǫ P )2 − a2 Pk ≤ ǫ if ǫ > 0 is chosen small enough. The argument in case of X ·X ∗ is very similar and we leave it to the reader. For an irreducible representation λρ obtained from a pure state ρ ∈ P(A) the above result gives an isomorphism Lν Lr 0(A) is given by a diagonal subalgebra of the direct product of such quotients where the product ranges over the different folia of pure states of A . Since each 0(cid:0)λρ(A)(cid:1) ≃ l∞(cid:0)Pλρ (A)(cid:1) where Pλ(A) = (cid:8)ρ ∈ P(A) ρ = ρξ , ξ ∈ Hλ(cid:9) . Then ǫ P − (P cǫ P )2k ≤ ǫ2 and k(P cǫ P )2 − a2 Pk ≤ 2 50 U. HAAG factor is a C∗-algebra such that the Jordan square coincides with the C∗-square the same is true for the direct product map. By an argument as above there exists a multiplicative extension πr 0 : Lq(A) −−−→ Lr 0(A) = L1(cid:0)r(L(A))(cid:1) extending the natural map r : L(A) → l∞(P(A)) . To see that Lr 0(A) ≃ l∞(P(A)) it is enough to note that the image of L(A) in l∞(P(A)) contains each function δρ , δρ(ρ) = 1 , δρ(σ) = 0 for σ 6= ρ which is the case since any two states in P(A) are separated. By an argument as used in the proof above the L1-completion of the image of L(A) must agree with the whole algebra l∞(P(A)) . That πr 0 factors π is again obvious from Proposition 1. This implies that for an arbitrary C∗-algebra A the image of the Jordan square of a basic positive element in L1(A) coincides with the C∗-square since the induced map Lr 0(A) ։ L1(A) is necessarily a ∗- homomorphism. From monotonicity of the Jordan square one then gets π(P2) ≤ π((Pc)2) = π(P) · π(P) while on the other hand P2 ≥ P · P implying the reverse inequality. Similarly one gets for a general selfadjoint element A ∈ Lq(A) the estimate sqr(A) ≥ sup λ,A (cid:26)λ(cid:18)(A + r 1)2 − λ,A (cid:26)λ(cid:18) (A + r 1) · (A + r 1) − ≥ sup 1 1 − λ 1 r2 1(cid:19) (cid:12)(cid:12)(cid:12) A ≤ A , A + r 1 ≥ 0 . 0 ≤ λ < 1(cid:27) r2 1(cid:19)(cid:12)(cid:12)(cid:12) A ≤ A, A + r 1 ≥ 0, 0 ≤ λ < 1(cid:27) 1 − λ = sup A≤A(cid:8)A+ · A+(cid:9) = A+ · A+ where the first identity in the third line can be checked pointwise on the spectrum of Lq(A) since for each homomorphism into R the supremum is taken for r → ∞ and the second identity follows by normality of the commutative C∗-square. On the other hand π(sqr(A)) ≤ π(A2 +) = π(A+) · π(A+) so that one gets an equality. Therefore also π(A2) = π(sqr(A)) + π(sqr(−A)) = π(A+)·π(A+) + π(A−)·π(A−) = π(A)·π(A). The second inequality of (21) follows since π(pP)2 = π(pP 2 ) ≤ π(pPc 2 ) ≤ π(P) by taking squareroots. The statement that the image of a+∧a− is trivial in L1(A) follows from the Remark after Lemma 2. Now suppose that I is an injective C∗-algebra. Assume first that µ : I → B(H) is a faithful injective representation and let υ : B(H) ։ I be a completely positive retraction (left inverse) for µ . One gets induced maps M : Lq(I) −→ Lq(B(H)) , Υ : L(B(H)) −→ L(I) , the former which is a normal ∗-homomorphism. Υ sends the order ideal Jr C∗(cid:0)L(cid:0)B(H)(cid:1)(cid:1) ⊆ Lq(cid:0)B(H)(cid:1) which is the kernel of the (multiplicative) function representation 0 ⊆ C∗(cid:0)L(I)(cid:1) for the pure states of B(H) onto a corresponding order ideal Jµ,υ with quotient Lµ,υ(I) = C∗(cid:0)L(I)(cid:1) / Jµ,υ (I) = L1(cid:0)Lµ,υ(I)(cid:1) . If C , D ∈ r : C∗(cid:0)L(cid:0)B(H)(cid:1)(cid:1) −−−→ l∞(cid:0)P(cid:0)B(H)(cid:1)(cid:1) . Put Lµ,υ 0 ⊆ 0 0 THE JORDAN LATTICE COMPLETION 51 L(I) are positive basic elements which agree modulo Jµ,υ images in L(R) . Then 0 let Cµ , Dµ denote their and (cid:0)Dµ(cid:1)2 ≡ Cµ · Cµ = M(cid:0)C · C(cid:1) , ≡ Dµ · Dµ = M(cid:0)D · D(cid:1) (cid:0)Cµ(cid:1)2 The relation Υ(cid:0)(cid:0)Cµ(cid:1)2(cid:1) ≥ C2 follows from the Schwarz inequality applied to υ and the reverse inequality follows from (cid:0)Cµ(cid:1)2 Υ(cid:0)(cid:0)Cµ(cid:1)2(cid:1) = C2 , Υ(cid:0)(cid:0)Dµ(cid:1)2(cid:1) = D2 . ≤ M(cid:0)C2(cid:1) . Then mod Jµ,υ so that since the quotient map modulo the order ideal Jµ,υ homomorphism the result C2 ≡ C · C , D2 ≡ D · D is necessarily a ∗- mod Jr 0 0 0 C2 ≡ D2 mod Jµ,υ 0 follows, i.e. the basic squaring operation is well defined on the quotient and agrees with the C∗-square of the corresponding elements. The argument in case of the operation (13) is much the same. Then the quadratic clouds of all types plus the Lie bracket associated with the Jordan squaring operation must be contained in Jµ,υ . In particular the image of L(I) is a C∗-subalgebra of Lµ,υ for any given triple of positive basic elements and J0 ∩ L(I) ⊆ Jµ,υ (I) . Define 0 0 0 Jµ 0 = \υ Jµ,υ 0 and check that Lµ properties by diagonal embedding into the direct product Lµ For a general multiplicative injective representation µ = ι ◦ π let Jµ preimage of the corresponding ideal Jι L(I) ։ L(J) . Then 0 (I) resp. L0(I) as defined in the Theorem has the required (I) . 0 denote the 0 ⊆ L(J) under the canonical surjection 0 (I) ֒→Qµ,υ Lµ,υ 0 Also from the argument above the canonical surjection L(I) ։ Lι(J) Lµ 0 (I) = Lι 0(J) . extends by the method used above (nonuniquely) to a ∗-homomorphism Lq(I) −−−→ Lµ 0 (I) 0 (I) . From the definition of Lµ since Lµ(I) is a C∗-subalgebra of Lµ 0 (I) it is immediately clear that if µ is multiplicative then the basic squaring operation (1) drops to the quotient Lµ(I) where it coincides with the corresponding C∗-operation since the quotient map Π : L(I) ։ L(J) sends a basic square C2 to Π(C)2 . Then check that if M denotes the set of equivalence classes of multiplicative injective representations of I the approximate quotient L0(I) of Lq(I) as defined in the Theorem has the required properties by diagonal embedding into the direct product L0(I) ֒→ Yµ∈M Lµ 0 (I) . In particular the map Lq(I) −→ Lµ 0 (I) factors over L0(I) . That π factors over π0 follows from Proposition 3. One first does the case where I is separably deter- mined, i.e. is the injective envelope I = I(A) of a separable C∗-algebra A . Then there exists a faithful supertransitive injective representation µ : I ֒→ B(H) on a 52 U. HAAG separable Hilbert space H with strong closure R ≃ ΠP B(P H) ⊆ B(H) a product of type I factors. Let υ : B(H) ։ I be a completely positive retraction factoring as υ : B(H) ΠP Ad P −−−→ R υR −−−→ I . Let Jν 0,R ⊇ Jν 0 denote the order ideal of L(B(H)) which is the kernel of the function representation for all vector states corresponding to minimal projections in R . Then 0(cid:1) ⊆ Υ(cid:0)Jν 0,R(cid:1) = ΥR(cid:0)Jν 0,R ∩ L(R)(cid:1) ⊆ ker π from Proposition 3, whence π factors over π0 in this case. In the general case choose for each separable sub-C∗-algebra A ⊆ I a completely positive embedding iA : I(A) ֒→ I and a completely positive retraction rA : I ։ I(A) extending the identity map of A . The completely positive embedding 0 ⊇ Jr 0 = Υ(cid:0)Jr Jµ,υ admits a completely positive retraction r : ΠA I(A) ։ I , extending to a monoto- nous retraction I ΠA rA I(A) −−−→ YA L1(cid:0)I(A)(cid:1) −−−→ L1(cid:0)I(cid:1) . Π1 : YA Choose a faithful supertransitive representation µ : I ֒→ B(H) ≃ B(cid:0)⊕AHA(cid:1) which is a direct product of separable supertransitive representations µA : I ։ I(A) ֒→ B(HA) factoring over rA for each separable subalgebra A ⊆ I with strong closure RA ⊆ B(HA) and a completely positive retraction of the form −−−→ I . υ : B(H) RA = R Then each single map factors as so that υR ΠP Ad P −−−→ YA πA : Lq(cid:0)I(A)(cid:1) ։ L1(cid:0)I(A)(cid:1) −−−→ Lq(cid:0)I(cid:1) 0 (cid:1) ⊆ L(cid:0)rA(cid:1)(cid:0)JµA,υ L(cid:0)I(A)(cid:1) ΠA πA (cid:1) = JµA,rA◦υ L1(cid:0)I(A)(cid:1) −−−→ YA π 0 −−−→ L1(cid:0)I(cid:1) −−−→ L1(cid:0)I(A)(cid:1) ⊆ ker πA is in the kernel of the composite map 0 Lq(cid:0)I(A)(cid:1) Lq(iA) L(cid:0)rA(cid:1)(cid:0)Jµ,υ −−−→ YA L(cid:0)I(cid:1) ΠA L(rA) 0 from Proposition 3. Thus Jµ,υ which is equal to π by uniqueness Π1 −−−→ L1(cid:0)I(cid:1) (cid:3) Definition. A positive element P ∈ L(A) is a basic square if it is the square of a positive basic element. More generally P ∈ Lq(A)+ is a quasibasic square if it is the supremum of such elements, i.e. P is quasibasic iff P = sup(cid:8) C2 (cid:12)(cid:12) C2 ≤ P(cid:9) . P = sup(cid:8)C (cid:12)(cid:12) C ≤ P(cid:9) . THE JORDAN LATTICE COMPLETION 53 One now defines an order contractive, monotonous and homogenous projection P (i.e. P (P) ≤ P for P ≥ 0 , P (P) ≤ P (Q) whenever P ≤ Q and P (α P) = α P (P) for α ∈ R+ ) of the positive elements onto the subset of quasibasic squares by the formula Also define a stabilized version of P by 2 P (P) = sup(cid:8)pC Q(P) = sup(cid:8)pC + r 1 2 (cid:12)(cid:12) C ≤ P(cid:9) . − r 1 (cid:12)(cid:12) C ≤ P , r ∈ R+(cid:9) and a concave projection onto the subset of quasibasic positive elements by Q0(P) = sup(cid:8)C (cid:12)(cid:12) C ≤ P(cid:9) . It is immediate from the definition that Q , Q0 are order contractive. Lemma 5 below shows that P is concave modulo ker π0 . One may extend P (and the same with Q , Q0 ) to a monotonous R-homogenous map on arbitrary selfadjoint ele- ments using the minimal positive decomposition putting P (A) = P (A+) − P (A−) . Since P is order contractive on positive elements one gets P (A+) ∧ P (A−) = 0 so the extension is again a projection. For r ≥ 0 a positve real number and Qr,f (C) = f (r)C f : R+ → R+ a positive function put Qr,f (C) = f (r)(cid:0)√C + r 1 − r 1(cid:1) and 2 Lemma 5. If A is injective the composition P0 : Lq(A) is concave with P0 = P0 ◦ P . Also P −→ Lq(A) π0−→ L0(A) (π0 ◦ Qr,f )(C) r→∞ −−−→ (π0 ◦ Qr,f )(C) for any boundedly generated positive basic element C and any positive function f with f (r)/r → 0 as r → ∞ . For an arbitrary positive basic element C one has the asymptotic identity π(cid:0)Q(C)(cid:1) = sup r→∞(cid:8)π(cid:0)pC + r 1 2 − r 1(cid:1)(cid:9) = π(C) . in particular π ◦ Q ◦ lcc = π ◦ Q ◦ uc = π and π ◦ Q = π ◦ Q0 . Proof. Monotonicity and homogeneity of P and Q are obvious. Assume given two basic positive elements C , D ≥ 0 . Relation (5′) and monotonicity of the (commutative) C∗-square gives pC + D ·pC + D ≥ (√λpC + √1 − λpD) · (√λpC + √1 − λpD) for every positive real number 0 ≤ λ ≤ 1 . Then the relation follows from the identity valid for any two positive real numbers α , β ≥ 0 pC ·pC + pD ·pD ≤ pC + D ·pC + D (√λ α + √1 − λ β)2 α2 + β2 = sup 0≤λ≤1 by evaluation at any point of the spectrum of the commutative C∗-algebra Lq(A) . If A is injective the C∗-square and the lattice operation (9) , (10) , (11) agree modulo ker π0 so that 2 π0(cid:0)pC + pD 2(cid:1) ≤ π0(cid:0)pC + D 2(cid:1) 54 U. HAAG follows and extends from definition of P to P0(P) + P0(Q) ≤ P0(P + Q) for arbitrary positive elements. Concavity implies in particular that the supremum in the definition of π0 ◦ Q resp. Q is attained for r → ∞ , so that Q(P + r 1) = Q(P) + r 1 for any positive scalar r ≥ 0 follows. Next we prove the asymptotic We first do the generic case A = B(H) and assume that C is boundedly generated, i.e. the infimum over a bounded positive subset of B(H) . Let ρ ∈ P ν(B(H)) be any vector state. Then identity π0(cid:0)Qr,f (C)− Qr,f (C)(cid:1) → 0 for r → ∞ and positive function f as above. 2 rν ) = rν ρ (pC + r 1 since the C∗-square coincides with the Jordan lattice square modulo the kernel of rν ρ . Given r ≥ 0 choose cr ∈ C such that ρ(√cr + r 1) ≤ rν ρ (pC + r 1)2 ρ (pC + r 1) + Then approximate the restriction of ρ to the commutative C∗-algebra generated by cr and 1 by a suitable convex combination of multiplicative states ρ = 1 2 r 3 . 2 1 r3 , ρ(cr)−ρ(cr) ≤ 1 r2 , ρ(c2 r)−ρ(c2 r) ≤ 1 r2 . i=1 λi = 1 with One obtains the estimate Pn i=1 λi si ,Pn ρ(√cr + r 1)−ρ(√cr + r 1) ≤ ρ(cid:0)pC + r 1(cid:1)2 nXi=1 since √cr + r 1 ≤ √r 1 + √cr/2 √r . Now λipsi(cr + r 1) ≥ ρ(√cr + r 1) = rν ≥ ρ(√cr + r 1)2 − nXi=1 + √r − ρ(c2 r) 8√r3 = ρ(cr) 2√r −3 ) 1 r − o(√r λi si(cr) 2√r + √r − si(c2 r) 8√r3! i − y2 with x2 Then since xi = si(cr)/2√r + √r ≥ yi =psi(cr) + r ≥ √r are real positive numbers i )/(xi + yi) is of order √r−3 . ρ(√cr + r 1)2 ≥ ρ(cr) + r − − o(r−2) = ρ(cr) + r − o(r−1) r) − ρ(cr)2 remains bounded independent of r by bounded i = si(cr)2/4r so that xi − yi = (x2 r) − ρ(cr)2 i − y2 because the value ρ(c2 generation of C , so that altogether ρ(c2 4r f (r) rν ρ(cid:0)pC + r 1 2(cid:1) ≥ f (r)(cid:0)rν r (cid:19) ρ(cid:0)C(cid:1) + r(cid:1) − o(cid:18) f (r) the reverse inequality being trivial and as a consequence Qr,f (C) converges to Qr,f (C) for every f with f (r)/r → 0 as r → ∞ . For f (r) = r or if C is not boundedly generated this statement is no longer true. Counterexamples are given below. To prove the last assertion for B(H) note that the assertion lim r→∞ π0(cid:0)Qr,f(cid:0)A(cid:1) − Qr,f(cid:0)A(cid:1)(cid:1) = 0 THE JORDAN LATTICE COMPLETION 55 also makes sense for arbitrary (nonpositive) boundedly generated basic elements since there always exists r0 ≥ 0 such that A + r 1 ≥ 0 for r ≥ r0 so the limit is well defined. Let any positive basic element C ≥ 0 be given and a ∈ B(H) , a ≤ C . Then it is sufficient to show that given ǫ > 0 there exists rǫ ≥ 0 such that the element a is in the lower complement of √C + r 1 − r 1 + ǫ 1 for r ≥ rǫ which implies 2 π(cid:0)C(cid:1) ≤ sup r→∞(cid:8)π(cid:0)pC + r 1 2 − r 1(cid:1)(cid:9) ≤ π(cid:0)Q(C)(cid:1) the reverse inequality being trivial. But certainly Qr,f (a) ≤ Qr,f (C) in case that the left side is defined. On the other hand one has the uniform con- vergence √a + r 1 2 − r 1 r→∞ −−−→ a 2 since this is true modulo ker π0 and the maximal sets corresponding to a and to √a + r 1 − r 1 are both weakly closed so that the restriction to the image of rν is an order isomorphism on this operator subsystem (compare with the argument in the proof of Lemma 2). For this note that the underlying set of √a + r 1 2 is the inverse of the strongly closed set underlying √a + r 1 by taking positive squareroots, hence strongly and a forteriori weakly closed by convexity. This proves the Lemma in the case A = B(H) . For the general case of injective A consider the universal representation λu : A → B(Hu) of A and the induced maps Λu : L(A) −→ L(B(Hu)) , Υ : Lq(B(Hu)) −→ Lq(A) as defined above. The argument above shows that for boundedly generated C one has uniform convergence lim r≥0(cid:8)f (r)(cid:0)pΛu(C + r 1) 2 in Lq(B(Hu)) modulo ker π0 . On the other hand lim 2 r≥0(cid:8)f (r)(cid:0)pΛu(C + r 1) showing that − Λu(C+r 1)(cid:1)(cid:9) ≤ lim π0(cid:0)Qr,f (C)(cid:1) r→∞ −−−→ π0(cid:0)Qr,f (C)(cid:1) . − Λu(C + r 1(cid:1)(cid:1)(cid:9) = 0 r≥0(cid:8)f (r) Λu(cid:0)pC + r 1 2 − (C+r 1)(cid:1)(cid:9) ≤ 0 In the same way one gets the last statement for f ≡ 1 and arbitrary positive basic element C modulo ker π Put P = π ◦ P : Lq(A)+ → L1(A)+ and Q = π ◦ Q = π ◦ Q0 . (cid:3) 2. Injective envelopes and enveloping von Neumann algebras. If X ⊆ B(H) is an operator system its injective envelope is denoted I(X) . A a completely isometric unital embedding ι : I(X) ֒→ B(H) one has (σ ◦ ι)(x) = x . positive linear map σ : B(H) → L1(cid:0)I(X)(cid:1) will be called a retraction (for ι ) if given The retraction σ is called proper if σ(cid:0)B(H)(cid:1) = I(X) ⊆ L1(cid:0)I(X)(cid:1) . 56 U. HAAG (i) Let λ : A → B(H) be a faithful unital ∗-representation of Theorem 2. the unital C∗-algebra A with strong closure given by the injective von Neumann algebra A′′ acting on the separable Hilbert space H . Let ι : I(A) ֒→ A′′ be any completely isometric embedding extending the identity map of A (which exists by injectivity of A′′ ). Then there exists a unique (completely) positive retraction ρ : A′′ ։ I(A) for ι and the kernel of ρ contains a canonical ∗-ideal J ⊳ A′′ which is trivial only if A′′ is (completely isometric to) the injective envelope of A . As a consequence if A is a separable C∗-algebra every selfadjoint element x ∈ I(A)sa is the monotone decreasing limit of a sequence (yn) ց x such that each yn ∈ I(A)sa is the monotone increasing limit (anm) ր yn of elements anm ∈ Asa (Up-Down-Theorem for I(A) ). (ii) If X is an operator system contained in an abelian C∗-algebra then I(X) = L1(X) and each selfadjoint element x ∈ I(X)sa is the least upper bound of the subset {aλ ∈ Xsa aλ ≤ x} , and the greatest lower bound of the subset {aµ ∈ Xsa x ≤ aµ} . In particular any monotone complete abelian C∗-algebra is injective (this of course is well known, cf. Theorem 4.3.6 of [1]). Proof. We begin with part (ii) concerning an operator subsystem of an abelian C∗- algebra. The assumption implies that also I(X) is abelian since the C∗-product of I(X) can be defined recurring to the C∗-product in some injective abelian C∗- algebra A ⊇ X on extending the inclusion of X to a unital isometric inclusion I(X) ⊆ A from injectivity, by the formula x · y = Φ(cid:0)x y(cid:1) , x , y ∈ I(cid:0)X(cid:1) where Φ : A → A is a positive projection with range equal to I(X) (compare with the proof of Theorem 6.1.3 of [6]). Let x ∈ I(X)sa be given and {aλ aλ ∈ X , aλ ≤ x} be the subset of elements in Xsa which are smaller or equal than x . Let x be the least upper bound of this set in I(X) which exists by monotone completeness of the injective envelope (Theorem 6.1.3 of [6]). Then x ≤ x . Consider the subspaces Ax = X + C x ⊆ I(X) and Ax = X + C x ⊆ I(X) and define a map ν : Ax → Ax extending the identity map of X in the obvious way by sending x to x . We claim that ν is positive (and hence completely contractive since unital with I(X) abelian). To see this let a positive element in Ax be given which can be written as y = a + γ x ≥ 0 with γ ∈ R and a ∈ Xsa . Suppose that γ < 0 . Then since x ≤ x one has ν(y) ≥ y ≥ 0 . We may therefore assume γ > 0 . Then ν(y) is equal to the least upper bound of the set {γaλ + a aλ ∈ X , aλ ≤ x} which equals the least upper bound of the set {bλ ∈ X bλ ≤ a+γx} , hence ν(y) ≥ 0 as desired. Extending ν to a completely positive map of I(X) into I(X) and using rigidity gives that x = x . The case of x being equal to the greatest lower bound of elements in Xsa which are larger follows by symmetry. This proves the special Up/Down-property of I(X) . Now it is easy to see that the monotone complete abelian C∗-algebras L1(X) and I(X) are naturally (completely) order isomorphic. If A is a monotone complete abelian C∗-algebra then by the foregoing argument each element x ∈ I(A)sa is the least upper bound of all elements {a ∈ A a ≤ x} . But this set also has a least upper bound x in A with x ≥ x whereas the set {b ∈ Asa b ≥ x} has a greatest lower bound x in A , so that x ≤ x ≤ x ≤ x and equality follows in each instance, i.e. A = I(A) so A must be injective proving (ii). THE JORDAN LATTICE COMPLETION 57 Let λ : A → B(H) be a faithful representation of A as in part (i) of the Theorem with strong closure given by the injective von Neumann algebra R = λ(A)′′ . To save notation put I = I(A) . Any completely positive extension ι : I ֒→ R ⊆ B(H) of λ then is a separable relatively transitive injective representation of I so that from Proposition 3 any monotonous map Lq(R) ։ L1(I) extending the canonical isometric identification of I in both spaces is necessarily unique. In particular any (completely) positive retraction ρ : R ։ I must be unique since any two different such maps can be extended to (different) positive retractions on the level of Lq(R) by injectivity also showing that any retraction in the broad sense of the Definition above is necessarily proper (and unique) restricted to R since we know that a proper retraction exists. Let Jν 0,R ⊆ L(R) denote the order ideal which is the kernel of the function representation rν corresponding to the vector states of B(H) restricted to L(R) . From Proposition 3 the order ideal JR,ι(I) ⊆ L(I) generated by basic restriction of Jν 0,R is contained in the kernel of L(R) ։ L1(R) . Let 0,R is contained in ker π since Jν 0 0 , rν rν 0 : Lν 0(cid:0)R(cid:1) −−−→ L1(cid:16)L(cid:0)I(cid:1) (cid:14) JR,I 0 (cid:17) = Lν 0(cid:0)I(cid:1) be the maps induced by basic restriction and antibasic restriction respectively as in Proposition 3 so that the canonical map Lq(R) ։ L1(I) factors as Lq(cid:0)R(cid:1) −−−→ Lν 0(cid:0)R(cid:1) rν 0 , rν 0 −−−→ Lν 0(cid:0)I(cid:1) −−−→ L1(cid:0)I(cid:1) . If x ∈ J + is the supremum of the monotone increasing net (yλ)λ ր x with yλ ∈ ι(I) and x ∈ ι(I) is the supremum of the same net in ι(I) then the positive element zx = x− x ∈ J − is in the kernel of ρ and in fact in the kernel of rν 0 since its image in Lν 0(R) coincides with the image of the basic element which is the basic restriction of the basic element Cx = inf E x = inf λ (cid:8)x − yλ(cid:9) ∈ I(cid:0)L(cid:0)I(cid:1)(cid:1) λ (cid:8)x − yλ(cid:9) ∈ Jν 0,R . 0(R) is given by the image of the elements The completely positive linear map ι is necessarily an A-module map, i.e. ι(axb) = a ι(x) b for a , b ∈ A , x ∈ I , since ι is multiplicative restricted to A . This fact can be checked from considering some ∗-homomorphic Stinespring dilation of ι and assuming a , b to be unitary elements since an arbitrary element is a linear combination of unitaries. Therefore zaxa = azxa so that ρ(a zx a) = 0 for every a ∈ Asa . Let J + A ⊆ J + denote the subset of elements which are suprema of increasing nets of elements in Asa . Then if b = supν aν ∈ J + A the image of b zx b in Lν µ≥ν(cid:8)aν(x− x)aν(cid:9) ≤ inf µ≥ν(cid:8)aν(x− x)aν(cid:9) = lim sup of Lq(cid:0)R(cid:1) since the net {aνzxaν}ν is strongly convergent to b zx b . That these 0(cid:0){zaν xaν}(cid:1)(cid:9) images are the same is seen from the relations 0(cid:0)lim inf zaν x aν = sup zaν x aν lim inf sup {zaνxaν}(cid:1) ≥ sup 0(cid:0)lim sup ≥ πν µ≥ν(cid:8)πν {zaν xaν}(cid:1) ≥ πν 0(cid:0){zaν xaν}(cid:1)(cid:9) = inf 0(cid:0)lim inf µ≥ν(cid:8)πν {zaν xaν}(cid:1) sup πν inf inf ν ν ν ν ν ν ν ν ν 58 U. HAAG which follow from monotonicity and the fact that πν elements and increasing normal on antibasic elements. Then 0 is decreasing normal on basic = π inf ν ≤ inf ν = inf ν ν zaν x aν(cid:1) (π ◦ r)(cid:0)lim sup λk zaµk xaµk(cid:17)(cid:12)(cid:12)(cid:12) µk ≥ ν , sup(r(cid:16) nXk=1 sup((π ◦ r)(cid:16) nXk=1 sup( nXk=1 nXk=1 λk zaµk xaµk(cid:17)(cid:12)(cid:12)(cid:12) µk ≥ ν , λk (π ◦ r)(cid:16)zaµk xaµk(cid:17)(cid:12)(cid:12)(cid:12) µk ≥ ν , nXk=1 λk = 1 , λk ≥ 0)! λk = 1) nXk=1 λk = 1) = 0 . ν k inf lim inf bk zx bk = sup The first equality holds since r is generally decreasing normal and the definition of antibasic restriction applied to antibasic elements, the second inequality then follows from monotonicity of π plus the fact that π is increasing normal restricted to antibasic elements, then the third equality follows since the composition π ◦ r = π ◦ r is known to be linear. Therefore ρ(b zx b) = 0 follows for any b ∈ J + A . If c ∈ Rsa is a general element it can be represented as the limit of a monotonous decreasing sequence (or net) (bk)k ց c , bk ∈ J + A . Then again the image of c zx c in Lν l≥k(cid:8)bk(x− x)bk(cid:9) ≤ inf l≥k(cid:8)bk(x− x)bk(cid:9) = lim sup 0(R) is represented by the images of the elements since the sequence (bkzxbk)k is strongly convergent to czxc . Now an argument as before applies to show that ρ(c zx c) = 0 . Therefore the ∗-ideal generated by all elements of the form {zx} ⊆ J − is contained in the kernel of ρ . If this ideal is trivial, then zx = 0 for any x ∈ J + implying J = ι(I) and a forteriori R = ι(I) which proves the first assertion of part (i). To prove the Up-Down property of I(A) in case of a separable C∗-algebra A note that there exists a faithful separable supertransitive representation ι : I(A) ֒→ B(H) extending a separable faithful ∗-representation λ : A −→ B(H) which is a direct sum of irreducible representations. Then if R = λ(A)′′ ⊆ B(H) denotes the strong closure of A we may apply Proposition 3 to compute the unique completely positive retraction ρ : R ։ I(A) for a given element x = inf n supm {anm} where (anm) ր bn is a monotonous increasing sequence with strong limit bn ∈ JA ⊆ R and (bn) ց x is a monotonous decreasing sequence with limit x in R . One has bk zx bk sup k k in Lν 0(R) from monotonicity. Therefore n sup sup πν 0 (x) = inf n m (cid:8)πν m (cid:8)anm(cid:9)(cid:19) = π inf sup( rXk=1 λr (π ◦ r)(cid:0)anmk(cid:1)(cid:12)(cid:12)(cid:12) m (cid:8)anm(cid:9)(cid:19) 0(cid:18)inf 0 (anm)(cid:9) ≥ πν sup(r(cid:16) rXk=1 λk = 1)! λk anmk(cid:17)(cid:12)(cid:12)(cid:12) rXk=1 λk = 1) = inf n nbno ≤ ρ(x) . rXk=1 sup n (π ◦ r)(cid:18)inf n ≤ inf n THE JORDAN LATTICE COMPLETION 59 On the other hand since inf n bn ≥ x in R one also gets the reverse inequality from monotonicity of ρ hence equality proving that ρ(x) = inf n sup m (cid:8)ρ(anm)(cid:9) is equal to the limit of the monotone decreasing sequence (bn) ց ρ(x) in I(A) with each bn the limit of the monotone increasing sequence (anm) ր bn . Since each element of I(A)sa is of the form ρ(x) of some selfadjoint element x ∈ Rsa the Up-Down-property for I(A) holds for any selfadjoint element, and the Down-Up property follows by symmetry. In particular if the identity x = inf n sup m (anm) = sup n inf m (dnm) holds in R with respect to monotone increasing nets (anm) ր bn and monotone decreasing nets (dnm) ց cn with anm , dnm ∈ A the corresponding identity also holds in I(A) (cid:3) The following is not really a corollary of Theorem 2 but rather a preamble to the whole text which we have chosen to present here because of its close relation with the results of Theorem 2. After working out the proof the author found that most of its results have been known for a quite a while, c.f. [2], Lemma 8. The only result that may be new is the injectivity of monotone complete abelian C∗-algebras with respect to linear lattice maps, at least the author has found no reference for this. Also the assumption that the order injection ι of the Corollary is multiplicative restricted to A is unnecessary by the Remark following Lemma 8 in [2] that given a positive linear unital order injection ι : A ֒→ B of commutative C∗-algebras any positive linear retraction B ։ A for ι is necessarily a ∗-homomorphism restricted to C∗(ι(A)) ⊆ B . Corollary. Let A be a commutative C∗-algebra with injective envelope I = I(A) . Given a unital order isomorphic embedding ι : I ֒→ B into a commutative C∗-algebra which is multiplicative restricted to A there exists a positive retraction ρ : B ։ I which is a ∗-homomorphism, in particular any commutative injective C∗-algebra is a C∗-quotient of l∞(Z) for some set Z . Any commutative injective C∗-algebra I is injective in the three categories of positive linear maps of function systems, ∗-homomorphisms of abelian C∗-algebras and lattice maps of linear func- tion lattices (which are order isomorphic to a linear sublattice of a commutative C∗-algebra). Proof. Choose a dense ∗-linear subspace Y ⊆ B together with a well ordered basis {cλ}λ∈Λ consisting of selfadjoint elements of norm one say such that for fixed index λ the element cλ is linear independent from the closure of the linear span {cκ κ < λ} and of norm one in the corresponding quotient space, also assuming that the subset {cλ} ∩ ι(A) generates a dense subspace J ⊆ ι(A) and exhausts the leading halfopen interval of all indices 1 ≤ λ < λ0 . To save notation one puts λ0 = 0 disregarding all indices λ < λ0 since for these the map ρω0 is canonically defined. Let Yλ denote the closure of the ∗-linear subspace generated by the set {cκ κ ≤ λ} and Y<λ the closure of the linear span of {cκ κ < λ} . Inductively define ρλ : Yλ → I by linear extension of ρλ(cλ) = inf(cid:8)ρ<λ(y)(cid:12)(cid:12) y ∈ Y<λ , y ≥ cλ(cid:9) , 60 U. HAAG assuming by induction that ρ<λ is well defined and positive. One checks positivity of ρλ . Let x = b + γ cλ ≥ 0 with b ∈ Y<λ be given. First assume that γ > 0 . Then On the other hand if γ < 0 then ρλ(x) = inf(cid:8)ρ<λ(y)(cid:12)(cid:12) y ∈ Y<λ , y ≥ x(cid:9) ≥ 0 . ρλ(x) = sup(cid:8)ρ<λ(y)(cid:12)(cid:12) y ∈ Y<λ , y ≤ x(cid:9) ≥ 0 since 0 is contained in Y<λ and x ≥ 0 . Proceeding by induction and extending by continuity where necessary this results in a positive ∗-linear map ρ : B → I . Then one needs to check that ρ is also multiplicative, i.e. a ∗-homomorphism, which is equivalent to validity of the Schwarz identity ρ(x2) = ρ(x)2 whence for selfadjoint elements x . Since the Schwarz inequality ρ(x2) ≥ ρ(x)2 holds by virtue of positivity one only needs to check the inverse Schwarz inequality ρ(x2) ≤ ρ(x)2 . This can be done inductively on each fixed subspace Yλ assuming the Schwarz identity for all elements in Y<λ and continuous extension arguments. Let b ∈ Y<λ + R+ cλ ≥ 0 be given. Then ρ(b2) ≤ inf(cid:8)ρ(a2)(cid:12)(cid:12) a ∈ Y<λ , a ≥ b(cid:9) = inf(cid:8)ρ(a)2(cid:12)(cid:12) a ∈ Y<λ , a ≥ b(cid:9) = ρ(b)2 . The first equality in the second line follows by normality of the squaring operation in (monotone complete) commutative C∗-algebras. Any selfadjoint element of Yλ can be written as a difference of elements as above. If b , c ≥ 0 are two such elements then = (cid:0)inf(cid:8)ρ(a)(cid:12)(cid:12) a ∈ Y<λ , a ≥ b(cid:9)(cid:1)2 2 ρ(cid:0)bc(cid:1) = ρ(cid:0)(b + c)2(cid:1) − ρ(cid:0)b2(cid:1) − ρ(cid:0)c2(cid:1) = 2 ρ(cid:0)b(cid:1) ρ(cid:0)c(cid:1) ρ(cid:0)(b − c)2(cid:1) = ρ(cid:0)b2(cid:1) − 2 ρ(cid:0)bc(cid:1) + ρ(cid:0)c2(cid:1) = ρ(cid:0)b − c(cid:1)2 proving the Schwarz identity for arbitrary selfadjoint elements in Yλ The induction starts by noting that the Schwarz identity trivially holds for elements of A . There- fore the construction yields a well defined ∗-homomorphism which necessarily is a retraction since it extends the identity map of A . Since any injective C∗-algebra is its own injective envelope and admits a faithful ∗-representation in some l∞(Z) this also proves that any injective C∗-algebra can be represented as a C∗-quotient of some l∞(Z) . Then basically the same argument shows that given an inclusion A ⊆ B of commutative C∗-algebras and a ∗-homomorphism ρ : A −→ I into a com- mutative injective C∗-algebra there exists a positive linear extension ρ : B −→ I of ρ which satisfies the Schwarz identity ρ(x2) = ρ(x)2 for each selfadjoint element x ∈ Bsa , i.e. ρ is a ∗-homomorphism. In particular every commutative injective C∗-algebra A is a complete function lattice, i.e. the lattice operations are normal. Namely using a unital ∗-homomorphic embedding j : A ֒→ l∞(Z) there exists an extremal retraction r : l∞(Z) ։ A which is a ∗-homomorphism, in particular a lattice map so that sup µ (cid:0)xµ ∧ y(cid:1) = r(cid:18)sup µ (cid:0)j(xµ) ∧ j(y)(cid:1)(cid:19) = r(cid:18)(cid:0)sup µ j(xµ)(cid:1) ∧ j(y)(cid:19) = (cid:0)sup µ Now the above argument can be used to show that I is injective also for lattice maps. Suppose given a sublattice inclusion of linear function lattices A ⊆ B (both xµ(cid:1) ∧ y . THE JORDAN LATTICE COMPLETION 61 sublattices of the selfadjoint part of some commutative C∗-algebra). One proceeds as above choosing a well ordered real basis {cλ}λ∈Λ of B such that the leading halfopen interval {cλ λ < λ0} is a basis for A and extends the given lattice map ρ : A −→ I to a positive linear map ρ : B −→ I in the manner above. Then one needs to show that this extremal extension is a lattice map. One proceeds by induction. Assume that for given fixed index λ ∈ Λ the identity ρ(a ∨ y) = ρ(a) ∨ ρ(y) holds for all elements a ∈ Y<λ , y ∈ B . Let x = b + γ cλ be given with γ ∈ R . One has ρ(x ∨ y) = ρ(cid:0)(cid:0)b ∨ (y − γcλ)(cid:1) + γ cλ(cid:1) = ρ(cid:0)b ∨ (y − γ cλ)(cid:1) + γ ρ(cid:0)cλ(cid:1) = ρ(cid:0)b) ∨ ρ(cid:0)y − γ cλ(cid:1) + γ ρ(cid:0)cλ(cid:1) = ρ(cid:0)x(cid:1) ∨ ρ(cid:0)y(cid:1) by linearity of ρ . Thus by induction the relation will hold if it holds for all x ∈ A , y ∈ B . Assume by induction that for some fixed index µ ∈ Λ the relation holds whenever y ∈ Y<µ . Put y = c + γ cµ with γ > 0 . By monotonicity one has ρ(x ∨ y) ≥ ρ(x) ∨ ρ(y) so we only need to prove the reverse inequality. Then ρ(x ∨ y) ≤ inf(cid:8)ρ(x ∨ d)(cid:12)(cid:12) d ≥ y , d ∈ Y<µ(cid:9) = inf(cid:8)ρ(x) ∨ ρ(d)(cid:12)(cid:12) d ≥ y , d ∈ Y<µ(cid:9) = ρ(x) ∨ inf(cid:8)ρ(d)(cid:12)(cid:12) d ≥ y , d ∈ Y<µ(cid:9) = ρ(x) ∨ ρ(y) . Now suppose that γ < 0 . One has ρ(cid:0)(x − y) ∨ 0(cid:1) = ρ(cid:0)x − y(cid:1) ∨ 0 from the argument above so that by linearity of ρ one gets ρ(cid:0)x ∨ y(cid:1) = ρ(cid:0)(x − y) ∨ 0(cid:1) + ρ(y) = (cid:0)(cid:0)ρ(x) − ρ(y)(cid:1) ∨ 0(cid:1) + ρ(y) = ρ(x) ∨ ρ(y) . Therefore this relation holds by induction for all x , y ∈ B . The relation follows by the symmetry x 7→ −x , y 7→ −y , exchanging ∧ for ∨ (cid:3) ρ(x ∧ y) = ρ(x) ∧ ρ(y) 3. Appendix: Projection lattices and P-algebras. A C∗-algebra A will be called an AΣ-algebra iff A is monotone complete and in addition the monotone closure of each unital abelian sub-C∗-algebra is contained in an abelian subalgebra (i.e. each maximal abelian subalgebra is monotone complete). A is a complete AΣ-algebra iff Mn(A) is an AΣ-algebra for every n ∈ N . Every von Neumann algebra is a complete AΣ-algebra, and as we will see also every injective C∗-algebra. A unital C∗-algebra A is a P-algebra if any selfadjoint element x ∈ Asa can be approximated in norm from below and from above by finite linear combinations of positive projections in some abelian subalgebra containing x , and a P-lattice algebra if it is a P-algebra and its subset of positive projections is a lattice for the induced order. A P +-algebra is a P-algebra A such that any element x ∈ A admits a polar decomposition x = v x = x∗ v with v ∈ A a partial isometry. An AΣ+-algebra is an AΣ-algebra which is a P +-algebra. A (possibly discontinuous) positively homogenous map σ : A → B of unital C∗-algebras satisfying σ(x+iy) = σ(x)+iσ(y) for any two selfadjoint elements x , y ∈ Asa which is monotonous (and hence continuous) on each abelian subalgebra C ⊆ A will be called a P-map iff it 62 U. HAAG maps positive projections to positive projections and if σ(α1 + x) = ασ(1) + σ(x) for each element x ∈ A and α ∈ C . One is primarily interested in the values of a P-map on positive elements. Given a P-map as above defined for positive elements there are also alternative ways to extend the domain of definition in a unique manner to arbitrary elements and to decide which extension is more suited may depend on the particular problem one is considering. One is by defining σ(x) = σ(x+)−σ(x−) whenever x = x+ − x− is selfadjoint with x = x+ + x− , x± ≥ 0 and σ(x + iy) = σ(x)+iσ(y) for any two selfadjoint elements x , y ∈ Asa which yields a monotonous (and hence continuous) map on each abelian subalgebra C ⊆ A which in addition is real homogenous. Such an extension will be called a homogeneized P-map. We will mostly consider P-maps which map orthogonal projections p⊥q to orthogonal projections. However to have the opportunity of considering more general P-maps we call such maps orthogonal P-maps or P o-maps. A P-map which maps each pair of complementary projections to a pair of complementary projections will be called complemented or a P c-map. Any P c-map is orthogonal, but the converse need not be true. The dual notion is a coorthogonal P-map or P co-map which sends each pair of coorthogonal projections p⊤q , i.e. [p , q] = 0 , p∨q = 1 to a pair of coorthogonal projections. A complemented P-map is both orthogonal and coorthogonal. A P-map which maps each pair of commuting projections to a pair of commuting projections will be called a P a-map. A P a-map such that σ(p ∧ q) = σ(p) ∧ σ(q) for each pair of commuting projections [p , q] = 0 will be called a P-A-wedge- map or P a∧-map, and a P-A-lattice-map if in addition it sends complementary projections to complementary projections. Correspondingly one may define P a∨- maps, and P ∧-maps resp. P ∨-maps of P-lattice algebras which instead of only considering the operations ∧,∨ for two commuting projections (which are defined in any C∗-algebra) respect these operations for general pairs of projections. We shall introduce another notion which on first sight does not seem as natural as the previous ones and is a bit more complicated to formulate: a mixed P-lattice map or P x-map is a P c-map σ : A → B such that there exists a decomposition P(A) = P+ ∪ P− of the subset of projections of A and a (possibly nontransitive) order relation ≺ defined on pairs of projections in P− called construction time order relation, subject to the following properties: e ∈ P− ⇐⇒ ec ∈ P+ , e ≤ f , f ∈ P− =⇒ e ∈ P− , e , f ∈ P− =⇒ either e ≺ f =⇒ either e ∧ f ≺ f e ≺ f or else or else f ≺ e , ec ∧ f ≺ f and such that the restriction s of σ to P(A) satisfies the following: if e , f ∈ P− with e ≺ f one has s(e ∨ f c) = s(e) ∨ s(f c) ⇐⇒ s(ec ∧ f ) = s(ec) ∧ s(f ) . Check that in case of commuting projections [e , f ] = 0 this implies the stronger conditions s(e ∨ f c) = s(e ∧ f ) + s(f c) = s(e) ∨ s(f c) s(ec ∨ f c) = s(ec ∧ f ) + s(f c) = s(ec) ∨ s(f c) , ⇐⇒ and correspondingly s(ec ∧ f ) = s(e)c ∧ s(f ) since in this case (e ∧ f ) ∨ f c = e ∨ f c , (ec ∧ f ) ∨ f c = ec ∨ f c respectively and s is complemented so that s(e ∧ f ) ∨ s(f c) = s(e ∧ f ) + s(f c) etc.. If given a decomposition as above together with a construction time order relation defined only on pairs of commuting projections THE JORDAN LATTICE COMPLETION 63 a P c-map σ (resp. s ) will be called a P ax-map, iff the above conditions are satisfied for pairs of commuting projections in P− , i.e. disregarding the condition s(e ∨ f c) = s(e) ∨ s(f c) for general (noncommuting) e , f ∈ P− with e ≺ f . Note however that a P ax-map also satisfies the following conditions for any pair of commuting projections e , f ∈ P− , [e , f ] = 0 , namely s(e ∧ f ) = s(e) ∧ s(f ) , s(ec ∨ f c) = s(ec) ∨ s(f c) since (cid:0)s(e) ∧ s(f )(cid:1)c = s(ec) ∨ s(f c) ≥ s(ec ∧ f ) + s(f c) = s(ec ∨ f c) = s(e ∧ f )c in case that e ≺ f , the reverse inequality follows trivially from monotonicity so that taking complements gives s(e ∧ f ) = s(e) ∧ s(f ) . A similar argument works in case that f ≺ e . In particular any P x-map is a P ax- map. A P xx-map is a P c-map σ : A → B such that there exists a decomposition P(A) = P+ ∪ P− as above with the following properties e , f , e ∨ f ∈ P− =⇒ s(e) ∨ s(f ) = s(e ∨ f ) . It will be shown in the proof of Theorem P below that any P ax-map and any P xx- [e , f ] = 0 implies [s(e) , s(f )] = 0 . The map automatically has decoration a , i.e. following compiles these notions into a general concept. A ∗-decorated P-map or P ∗- map of P-lattice algebras is a P-map with one or more of the additional properties ∗ ∈ {o, co, c, a, a∧, a∨, ax, x, xx,∧,∨} . A P a∧-map is also called of concave type since it satisfies σ(x + y) ≥ σ(x) + σ(y) for any two commuting positive elements x , y ≥ 0 , [x , y] = 0 (see below). Similarly a P a∨-map will be called of convex type. A decorated P-map respecting all three lattice operations (and hence has all other decorations as well) will be called a P-lattice map, i.e. a P-lattice map satisfies σ(p ∨ q) = σ(p) ∨ σ(q) , σ(p ∧ q) = σ(p) ∧ σ(q) , σ(1 − p) = 1 − σ(p) for any pair of positive projections p , q ∈ A . One would like to be dealing only with P-lattice maps since these have the strongest properties but then these maps occur very sparsely in nature, not even ∗-homomorphisms of P-lattice algebras are guaranteed to be P-lattice maps. On the other hand P-A-lattice maps still have extremely strong implications (the restriction to any abelian subalgebra is a ∗-homomorphism), and include the notion of ∗-homomorphisms. Note how- ever that even a P-lattice map need not be overall monotonous or continuous. Simple counterexamples can be constructed by discontinuous bijective P c-maps σ : M2(C) → M2(C) on realizing that any such map is a P-lattice map. For the following concept we introduce some notation: put V = (cid:18)0 1 1 0(cid:19) , W = (cid:18)1 0 0 −1(cid:19) , p = 1 + V 2 , q = 1 + W 2 , and pc = 1 − p , qc = 1 − q which are symmetries and projections in M2(C) . Let A , B be complete P-algebras and Eq ⊆ P(M2(A)) denote the subset of pro- jections which are smaller or equal to q , i.e. Eq can be identified with P(A) via the embedding A ֒→ M2(A) corresponding to the left upper corner. Let F ∈ M2(A) , F = F ∗ = F −1 be a symmetry of the form F = e ⊗ 1 + (1 − e) ⊗ V for some e ∈ P(A) and put Aq,F = F(cid:0)A q + A qc(cid:1) F ⊆ M2(A) . A P2-map is a 64 U. HAAG P-map σ : A → B such that there exists a P-map σ2 : M2(A) → M2(B) with σ(x) p = σ2(xp) for x ∈ A such that σ2 is of concave type restricted to each subalgebra Aq,F and satisfies p σ2(cid:0)X(cid:1) p = σ2(cid:0)p X p(cid:1) for any X ∈ Aq,F . An (abstract) lattice Λ is said to be of finite type if each totally ordered subset in a finitely generated sublattice is finite, and a function lattice iff it satisfies the identities (x ∧ y) ∨ z = (x ∨ z) ∧ (y ∨ z) and (x ∨ y) ∧ z = (x ∧ z) ∨ (y ∧ z) for any three elements x , y , z ∈ Λ . A function lattice is clearly of finite type, another example is the lattice of projections of any finitedimensional C∗-algebra, which in fact is of finite depth, i.e. the size of each totally ordered subset is uniformly bounded by some constant N . A complete lattice will be called a complete function lattice if it is a function lattice and in addition for each monotone increasing net (xλ) ր x converging up to an element x and any y one has the identities x∧ y = supλ (xλ ∧ y) , resp. for any monotone decreasing net (wµ) ց w and any z one has w ∨ z = inf µ (wµ ∨ z) . Also recall from [10] that a modular lattice is a lattice satisfying (e ∨ f ) ∧ g = e ∨ (f ∧ g) whenever e ≤ g . A complemented lattice is a lattice Λ containing a unique maximal element 1 and a unique minimal element 0 which is equipped with an involutary map x 7→ xc = 1 − x , (xc)c = x such that x ≤ y ⇐⇒ yc ≤ xc , (x ∧ y)c = (xc ∨ yc)c . Then x and y are orthogonal iff x ≤ yc . For orthogonal elements x⊥y we also write the lattice operations in additive/ subtractive form x ∨ y = x + y and yc ∧ xc = yc − x = xc − y respectively. Then one assumes the following restricted commutation property: (x + y)∧ (x + z) = x + (y ∧ z) . A complemented lattice Λ will be called signed iff given a disjoint decomposition Λ = Λ− ∪ Λ+ (a signature) such that e ∈ Λ− ⇐⇒ ec ∈ Λ+ and e ≤ f , f ∈ Λ− implies e ∈ Λ− . A signature will be called polar iff the following condition holds: given a pair of orthogonal elements e ⊥ f with e + f ∈ Λ+ then either e ∈ Λ+ or else f ∈ Λ+ . Corre- spondingly a lattice Λ is called polar iff for each pair of nonorthogonal elements c , d ∈ Λ , c (cid:2) dc there exists a polar decomposition Λ = Λ−∪ Λ+ with c , d ∈ Λ+ . Any (unital complemented) sublattice Λ0 ⊆ Λ of a polar lattice is again polar (by restricting any suitable polar signature from Λ to Λ0 ). In fact the notion only re- quires knowledge of the A-lattice structure of Λ (complements and orthogonal sums of elements). Any projection lattice of a commutative C∗-algebra is polar since a polar decomposition corresponds to the restriction of a multiplicative functional in this case and a commutative C∗-algebra has a separating family of such function- is polar in this case. Apart from these examples no matrix algebra of higher (or infinite) dimension admits a polar signature. If such a signature would exist on als. Another case of a polar projection lattice is P(cid:0)M2(C)(cid:1) . In fact any signature P(cid:0)M3(R)(cid:1) say, one finds that the corresponding region of the real projective plane consisting of minimal projections of positive signature must be simply connected and convex. Moreover for each point of the boundary the maximal length of the intersection of some geodesic containing the boundary point with this region must be the same and equal to π/2 .The only solutions for this setting are a disc of diam- eter π/2 , the equilateral geodesic triangle of side length π/2 and more generally THE JORDAN LATTICE COMPLETION 65 any uneven regular geodesic polygon of diameter π/2 . In each case there remain orthogonal systems lying completely outside of the region in question. Therefore a polar signature cannot exist for P(cid:0)Mk(C)(cid:1) for any k ≥ 3 . Thus considering full projection lattices of C∗-algebras the notion of polar signature is confined more or less to commutative algebras. The situation might be quite different however if one considers (countable) sublattices of projections (for a countable sublattice the connectedness property makes no sense). A complemented monotonous map of signed lattices r : Λ → Λ′ will be called signed iff r(Λ±) ⊆ Λ′ ± . Specifying a sig- nature on a lattice Λ is equivalent to specifying a complemented monotonous map τ : Λ ։ {0, 1} into the trivial lattice by letting Λ− = τ −1(0) , Λ+ = τ −1(1) . Cor- respondingly, specifying a polar signature on Λ is equivalent to giving an A-lattice map τ : Λ ։ {0, 1} into the trivial lattice. An A-lattice is a partially ordered set Γ containing a unique maximal element 1 and a unique minimal element 0 which is equipped with an order reversing involution e 7→ ec = 1 − e , (ec)c = e and an addition (e, f ) 7→ e + f of orthogonal elements e ⊥ f ⇐⇒ e ≤ f c such that e + ec = 1 and e + f ∈ Γ is the smallest element larger than both e and f , then there is also defined a subtraction operation (e, f ) 7→ f − e for ordered pairs e ≤ f with (f − e) + e = f . The subset of projections of any C∗-algebra is a natural example of an A-lattice. A faithful A-lattice representation is an or- der isomorphic map ι : Γ ֒→ Λ of an A-lattice Γ into a complemented lattice Λ respecting complements and orthogonal sums such that ι(c)⊥ ι(d) ⇒ c⊥ d . An A-sublattice Γ ⊆ Λ of a complemented lattice will be called proper iff e ∧ f ∈ Γ for any pair of commuting elements e , f ∈ Γ , [e , f ] = 0 . A P-lattice algebra A will be called of finite type iff its subset of projections P(A) is a (complemented) lattice of finite type and a modular P-lattice algebra iff its subset of projections is a modular lattice. The bulk of results given below also works if one considers AW ∗-algebras in the sense of [10] and their C∗-quotients instead of AΣ-algebras. In particular the result that the subset of projections in an AW ∗-algebra forms a complete lattice is well known, however the proof given here is a bit different from the one given in [10] and has as consequence that for any normal inclusion of AΣ-algebras one gets a corresponding normal inclusion of projection lattices, which might be known in case of inclusions of von Neumann algebras but does not seem to have been made explicit in any paper known to the author. It also seems that the AW ∗-algebras encountered in nature are mostly monotone complete or sequentially monotone complete, so that AΣ-algebras constitute a natural and interesting category. Some of the results of the Theorem below are sharpened by the subsequent Corollary. If A is an AΣ-algebra, the subset of positive projections P(A) Theorem P. forms a complete lattice which linearly generates A (up to norm closure) and for any surjective ∗-homomorphism q : A ։ B the quotient B is a P-algebra. In case that q is normal B is an AΣ-algebra and q admits a natural normal (nonunital) ∗-homomorphic cross section, there exists a central projection pq ∈ Z(A) with ker q = (1 − pq) A and B ≃ pq A . If q is sequentially normal whence B is sequentially monotone complete, then B is a P-lattice algebra and q is a P- lattice map. If B is any C∗-quotient of the AΣ-algebra A such that B is a P- lattice algebra with quotient map q a P-lattice map (for example if A is abelian), and B0 ⊆ B is a unital separable subalgebra there exists a P c-map σ : B → A which is a cross section for the quotient map restricted to some normdense 66 U. HAAG subalgebra D0 ⊆ B0 . If B is abelian σ can be chosen a P-A-lattice map (= In particular if B = q(A) is any C∗- P-lattice map), i.e. a ∗-homomorphism. quotient of an AΣ-algebra A such that q is a P-lattice map, then any normal element in B lifts to a normal element in A , and any unitary in B lifts to a unitary in A . Alternatively, if A is of finite type σ can be chosen to be a cross section and a P ∧,o-map restricted to D0 and a P o-map on B , or (by duality) a P ∨,co-map cross section restricted to D0 , and a P co-map on B . In case that q is sequentially normal σ can be chosen a (nonunital) P-lattice-map restricted to D0 and a P c-map on B (with respect to the unit s(1) ∈ P(A) ). A P-lattice map (or even a P-A-lattice map) is a ∗-homomorphism restricted to any abelian subalgebra. A P2-map is monotonous (and a forteriori continuous). Any P-map σ : A → B of P-algebras is uniquely determined on its (monotonous) restriction s : P(A) → P(B) to the subset of positive projections and satisfies the identity σ(x2) = σ(x)2 for any x ≥ 0 . A homogeneized P o-map satisfies the Schwarz inequality σ(x2) ≥ σ(x)2 for any selfadjoint x ∈ Asa . A homogeneized P a-map of concave type satisfies the Schwarz inequality σ(x x∗) ≥ σ(x) σ(x∗) for any normal element x , and a homogeneized P a-map of convex type satisfies the reverse Schwarz inequality σ(xx∗) ≤ σ(x) σ(x∗) for normal x . A P c-map is (automatically) real homogenous. A P a,c-map defined on positive elements of a P +-algebra A can be extended to a map of arbitrary elements which sends unitaries to unitaries. Any monotone complete complemented lattice Λ has the following extension property with respect to P c-maps (monotonous maps commuting with taking complements): given an inclusion of A-lattices Γ ⊆ Γ′ and a P c-map s : Γ → Λ there exists a P c-extension s′ : Γ′ → Λ with s′Γ = s . Any injective C∗-algebra and any von Neumann algebra is a complete AΣ+-algebra. In particular any normal quotient of an injective C∗-algebra is injective. Proof. Assume that A is an AΣ-algebra. We first show that the subset of its positive projections forms a complete lattice. Given two projections p, q ∈ A consider the two monotone decreasing sequences of alternating products yn = (q p) (q p) ··· (q p) (xn)n ց x, } Clearly, √x ≤ inf n √x2n = inf n xn = x by monotonicity of the squareroot xn = (p q) (p q) ··· (p q) whence x (and similarly y ) is a projection. Then one checks the relations {z p, (yn)n ց y, n−times } q. n−times {z x y x = inf n (x yn x) = inf n (x xn+1 x) = x, y x y = inf n (y xn y) = inf n (y yn+1 y) = y since both Ad x and Ad y are normal from which x = y follows. (That Ad x is normal for positive x in a monotone complete C∗-algebra is easily seen if x is invertible, however in the general case one can approximate x in norm by invertible positive elements of the form xǫ = x + ǫ1 whence the result). Then x ≤ p , x ≤ q and we claim that x is the largest positive element with this property whence x = p ∧ q . Given 0 ≤ z ≤ p, q one has p z = z = q z then also xn z = z = z xn for every n so that z x z = z2 and z (1 − x) = 0 = (1 − x) z follows. Thus the projections form a lattice putting p ∨ q = 1 − ((1 − p) ∧ (1 − q)) . Then suppose given a monotone decreasing net of projections (pλ)λ ց p . One needs to show that p = p2 ⇐⇒ √p = p . By monotonicity of the squareroot one has √p ≤ inf λ √pλ = p THE JORDAN LATTICE COMPLETION 67 and the reverse relation is trivial proving p = p2 . Then the projections form a complete lattice. From the second property of an AΣ-algebra the monotone completion of each unital sub-C∗-algebra generated by a single positive element x ≥ 0 is an abelian subalgebra hence injective so that since any commutative injective C∗-algebra is a C∗-quotient of some l∞(Z) (see the Corollary of Theorem 2) any positive element can be approximated in norm from above and from below by finite positive linear combinations of projections. On any abelian subalgebra the ∧- and ∨-operations defined for projections generalize to the usual lattice operations for arbitrary selfadjoint elements in a commutative setting. Given any positive element x ≥ 0 in A there is a minimal projection px with pxx = xpx = x called the support projection. Indeed, if p and q are two projections with this property one readily checks that also (p ∧ q)x = x , hence the projections {pλ,x} with this property form a monotone decreasing net in P(A) the subset of all positive projections in A and the infimum px satisfies x (1 − px) x = sup λ x (1 − pλ,x) x = 0 y = (1 − pβ y ) pα x . Then pα x x ≥ α pα by normality rq = sup(cid:8)px (cid:12)(cid:12) x ∈ ker q , x ≥ 0(cid:9) lies in ker q and is the maximal whence px x = x follows. Analogously one defines for each α ∈ R+ and x ≥ 0 the spectral projection px,α ≤ px to be the maximal projection commuting with x and smaller than px such that px,α x ≤ α px,α resp. pα x to be the minimal projection commuting with x such that pα x ≥ 1 − px,α . For α ≤ β define pα,β y . Let q : A ։ B be a surjective normal ∗-homomorphism. Then the subset of contractive positive elements contained in ker q has a supremum rq since for each x ∈ ker q , 0 ≤ x ≤ 1 also px ∈ ker q then projection with this property. Therefore the kernel of q is contained in rq A rq while on the other hand the kernel of q contains rq A + A rq showing that rq is central. Putting pq = (1 − rq) one gets ker q = (1 − pq) A and B ≃ pq A giving a natural normal ∗-homomorphic lift to the quotient map. Now let q : A ։ B be any surjective ∗-homomorphism. We first show that B is a P-algebra. From the fact that each projection in B is contained in an abelian subalgebra which is the quotient of some monotone complete abelian subalgebra in A one finds that any positive projection e ∈ B lifts to a positive projection es ∈ A which fact can be checked for arbitrary C∗-quotients of injective (monotone complete) abelian C∗-algebras on extending a clopen subset of the (totally disconnected) spectrum of the quotient to a clopen subset of the (extremely disconnected) spectrum of the injective extension. Therefore the positive projections P(B) generate B and any positive element x can be approximated in norm from above and below by finite positive linear combinations of pairwise commuting projections commuting with x , i.e. B is a P-algebra. Then we need to show that q is a lattice map in case that q is sequentially normal, i.e. for any pair of positive projections e , f ∈ A the element q(e ∧ f )(=: q(e) ∧ q(f )) is the largest positive element in B dominated by q(e) and q(f ) . Since B is sequentially monotone complete the argument given above yields that B is a P-lattice algebra and it is easy to see that q is a lattice map. Now suppose that q : A ։ B is any surjective ∗-homomorphism which is a P-lattice map with B a P-lattice algebra (e.g. for commutative algebras any ∗- homomorphism is a lattice map) and B0 ⊆ B a separable subalgebra. As a warm up for the more complicated task of constructing P-maps with fancy decorations we begin by constructing a P c-map so that certain general features of the construction 68 U. HAAG are more apparent and the reader can compare and recourse to this much simpler construction on examining the others. The first task is to construct a monotonous complemented map s : P(B) −→ P(A) which is a cross section restricted to some dense countably generated subspace D0 ⊆ B0 . Using a transfinite enumeration of the set of positive projections P(B) = {eγ γ ∈ Γ} with Γ a well ordered set such that the complementary projection of eγ is either the direct successor or the direct precessor of eγ starting with the pair (0, 1) one constructs a monotonous map of the positive projections in B into the set P(A) of positive projections in A by transfinite induction. Put s(0) = 0 , s(1) = 1 . Assume by induction that for some given ordinal γ0 with direct successor γ0 + 1 corresponding to the complementary projection 1− eγ0 one into the set of projections has constructed a coherent lift s of the subset (cid:8)eγ(cid:9)γ<γ0 of A meaning that for all γ , κ < γ0 the relations are satisfied whenever eγ < eκ . Put s(1 − eγ) = 1 − s(eγ) , s(eγ) ≤ s(eκ) E γ0 = [γ<γ0(cid:8)s(eγ)(cid:12)(cid:12) eγ < eγ0(cid:9) , eγ0 = _e∈E γ0(cid:8)e(cid:9) , γ0(cid:1) ∧ (eγ0) , s(eγ0) = (cid:0)(cid:0)eγ0(cid:1) ∨ es E γ0 = [γ<γ0(cid:8)s(eκ)(cid:12)(cid:12) eγ0 < eκ(cid:9) , eγ0 = ^f ∈E γ0(cid:8)f(cid:9) s(1 − eγ0) = 1 − s(eγ0) . and choosing an arbitrary projection es γ0 with q(es γ0) = eγ0 define One proceeds inductively to construct a monotonous map s : P(B) → P(A) . If e⊥f are orthogonal projections then e ≤ 1− f whence s(e) ≤ s(1− f ) = 1− s(f ) , therefore s(e)⊥s(f ) are orthogonal. Consider the normdense subset Pf (B) ⊆ Bsa of the real subspace of selfadjoint elements consisting of finite inear combinations of pairwise commuting projections (any element x ∈ Bsa can be approximated from above and from below by finite linear combinations of projections in the monotone closure of C∗(x) which is abelian). If x ∈ Pf (B) , x ≥ 0 there is a unique decomposition nXk=1 x = αk pk , 0 < p1 < ··· < pn ≤ 1 , αk ∈ R+ , 1 ≤ k ≤ n , whence the {pk} are pairwise commuting projections (contained in some abelian subalgebra). This follows since the corresponding orthogonal decomposition gives the spectral projections and eigenvalues of x which if they exist are unique in any C∗-algebra. Using this decomposition extend s to a lift σ : Pf (B) → Pf (A) by the formula One needs to check that σ is monotonous on the intersection of any abelian sub- algebra C ⊆ B with Pf (B)+ . Let σ(x) = αks(pk) . nXk=1 x = αk pk ≤ βl ql = y nXk=1 mXl=1 THE JORDAN LATTICE COMPLETION 69 denote the canonical decompositions of x ≤ y assuming that [pk , ql] = 0 for all k , l . By subtracting a suitable scalar multiple of 1 from both sides one may assume that the support projection pn of x is strictly smaller than 1 . Then min{αn, βm} pn ≤ min{αn, βm} qm . Subtracting min{αn, βm} s(pn) from the images of both sides the remaining positive term involving the projection s(qm) − s(pn) on the right side may be dropped from consideration since it is orthogonal to the support projection on the left and doesn't affect the validity of the inequality relation (this argument only works in the abelian case). Then one proceeds by induction to prove that σ(x) ≤ σ(y) . From monotonicity on the in- tersection of each abelian subalgebra with Pf (B)+ (which entails continuity) the definition of σ extends to all of Bsa putting σ(x+ − x−) = σ(x+) − σ(x−) and is monotonous and continuous on the positive cone of each abelian subalgebra C ⊆ B , in particular continuous restricted to abelian subalgebras. It may however fail to be monotonous or even continuous in general. Real homogeneity is more than obvious from the definition. If B0 ⊆ B is a separable subalgebra we may choose a dense countably generated subspace D0 ⊆ B0 ⊆ B and a corresponding countable subset of projections P0 ⊆ P(B) such that any element in D0 can be approximated in norm from below and from above by linear combinations of pairwise commuting projections in P0 . Using a transfinite enumeration of P(B) beginning with a (se- quential) enumeration of P0 one checks that the construction above gives a cross section restricted to the subspace of B spanned by P0 hence on D0 since for these only finite infima and suprema are involved and q respects these being a lattice map. This proves that a P c-map with the required properties can be constructed. Suppose that B is abelian. To construct a P-A-lattice map σ : B → A one begins with an enumeration {eγ} of the set P(B) as above. For simplicity we write γ∧ κ resp. γ∨κ to denote the index corresponding to the projection eγ ∧eκ and eγ ∨eκ etc., then assume by induction that for some given finite ordinal γ0 one already has constructed a coherent lift of the uncomplemented sublattice Λγ0 , i.e. Λγ0 is closed under the lattice operations ∧,∨ but not necessarily under c , generated by into the set of projections of A meaning that the set of all projections {eγ}γ<γ0 for all γ , κ ∈ Λγ0 the relations s(eγ ∧ eκ) = s(eγ) ∧ s(eκ) , are satisfied and in general (cid:2)s(eγ) , s(eκ)(cid:3) = 0 s(eγ ∨ eκ) = s(eγ) ∨ s(eκ) , eγ ≤ eκ ⇒ s(eγ) ≤ s(eκ) . is finite for any finite ordinal γ0 . These conditions imply that for any indices γ , κ ∈ Λγ0 . In case of infinite γ0 one uses the fact that any mono- tone complete abelian subalgebra containing all {s(eγ) eγ ∈ Λγ0} is a complete function lattice by the Corollary of Theorem 2 to get this result. We first treat the case of finite γ0 since in this context the sublattice Λγ0+1 generated by Λγ0 and eγ0 is finite and posesses minimal elements not contained in Λγ0 . We may choose one such element denoted eδ0 ≤ eγ0 . Then inductively choose a sequence of Note that Λγ0 putting eγ0 µ = _ω(cid:8)s(eω)(cid:12)(cid:12) ω ∈ Λγ0 , eω ≤ eµ(cid:9) , for arbitrary µ one has µ ∧ s(eγ) ≤ eγ0 eγ0 µ∧γ , eγ0 µ = ^ρ(cid:8)s(eρ)(cid:12)(cid:12) ρ ∈ Λγ0 , eρ ≥ eµ(cid:9) eγ0 ∨ s(eκ) ≥ eγ0 µ∨κ 70 U. HAAG k=1 ⊆ Λγ0+1 such that if Λk elements {eδk}n γ0 denotes the sublattice generated by Λγ0 and {eδ0 , ··· , eδk} the element eδk+1 is a minimal element in the complement Λγ0+1\Λk γ0 = Λγ0+1 . Assume again by induction that one has already constructed a coherent (A-lattice) lift on the sublattice Λk−1 for given k ≤ n . One proceeds as above to obtain a projection z ∈ A with γ0 and Λn γ0 eγ0 δk ≤ z ≤ eγ0 δk , γ0 δk δk eγ0 δk and such that q(z) = eδk . Then we still have to incorporate the correct lattice and commutation relations. Any projection of the form eδk∧γ strictly smaller than eδk is already contained in Λk−1 δk ∧ s(eγ) ≤ z ∧ s(eγ) and replacing z by . Then s(eδk∧γ) = eγ0 , q(z′) = eδk , [z′ , s(eγ)] = 0 and z′ ∧ s(eγ) = commutes with s(eγ) one the new element satisfies z′ ≥ eγ0 s(eδk ∧ eγ) . For the first inequality note that since eγ0 has z′ = (cid:2)z ∧(cid:0)1 − s(eγ)(cid:1)(cid:3) + s(eδk∧γ) ≤ z δk ∧(cid:0)1 − s(eγ(cid:1)(cid:1) + s(eδk∧γ) ≤ z′ . = (cid:0)eγ0 One proceeds in this manner by induction with respect to the induced enumeration to cover all indices γ ∈ Λk−1 leading to a (finite) monotone decreasing sequence of elements (z(n)) . Then put z = inf n z(n) ≥ eγ0 which is a lift of eδk such that z ∧ s(eγ) = s(eδk∧γ) and (cid:2)z , s(eγ)(cid:3) = 0 for each γ ∈ Λk−1 . Then suppose that for some given element eκ ∈ Λk−1 eδk∨κ ∈ Λk−1 z′ = (cid:2)z + (cid:0)s(eδk∨κ) − (cid:0)z ∨ s(eκ)(cid:1)(cid:1)(cid:3) ∧ eγ0 the new element satisfies the same relations as z plus the relation z′ ∨ s(eκ) = s(eδk∨κ) . Proceeding as before by induction with respect to the induced enumer- ation of indices {κ eκ ∈ Λk−1 γ0 } one arrives at an element z such that , eδk∨κ ∈ Λk−1 . Replacing z by δk δk γ0 γ0 γ0 γ0 γ0 z ∧ s(eγ) = s(eδk∧γ) , for all γ, κ, µ with eδk∨κ ∈ E k−1 z ∨ s(eκ) replace z by γ0 z ∨ s(eκ) = s(eδk∨κ) , If eµ , eκ ∈ E k−1 . γ0 (cid:2)z , s(eµ)(cid:3) = 0 , eµ ≤ eδk∨κ but s(eµ) (cid:2) (cid:2)z + (cid:0)z ∨ s(eκ) ∨ s(eµ) − z ∨ s(eκ)(cid:1)(cid:3) ∧ eγ0 δk etc. and repeating this procedure for all pairs µ, κ as above define s(eδk ) to be the supremum (maximum) of the corresponding monotone increasing (finite) sequence of elements. Check that s(eδk ) satisfies the same relations as before plus the relations eµ ≤ eδk ∨ eκ ⇒ s(eµ) ≤ s(eδk ) ∨ s(eκ) Note that eµ ≤ eδk ∨ eκ implies s(eµ) ≤ eγ0 commutative setting. Defining δk ∨ s(eκ) since we are working in a then gives a coherent A-lattice lift for Λk γ0 . We may assume that the union of all uncomplemented sublattices γ0 completing the induction step for finite s(eδk∨κ) := s(eδk ) ∨ s(eκ) is a complemented sublattice. Then the A-lattice condition implies that the map s is complemented restricted to this union. In case of infinite γ0 the complement [γ<∞ Λγ THE JORDAN LATTICE COMPLETION 71 Λγ0+1\Λγ0 may contain no minimal elements. Assume given a coherent A-lattice- map s on the sublattice Λγ0 containing all elements {eγ γ < γ0} together with their finite products (in a commutative setting the wedge of two projections is equal to their product) and vees. Since we don't have to worry about the cross section property we may simply define s(eγ0 ∧ eγ) = eγ0 ∧ s(eγ) , s(eγ0 ∨ eκ) = eγ0 ∨ s(eκ) where eγ0 = sup(cid:8)s(eγ)(cid:12)(cid:12) eγ ∈ Λγ0 , eγ < eγ0(cid:9) and check that it satisfies all the required relations, This is because eγ0 commutes with all images of elements in Λγ0 so that we are working in a complete function lattice whence (cid:0)supλ xλ(cid:1)∧y = supλ (xλ∧y) for any monotone increasing net (xλ) and an arbitrary element y . We now come to the case of arbitrary (noncommu- tative) quotient B (of finite type). The proof uses in parts the special case above but the problem is much more complex in that we have to consider relations of noncommuting projections. One observes the following fact: for any pair of pro- jections e , f ∈ P(B) there exists a unique minimal projection ef larger than e and commuting with f and a unique maximal projection ef smaller than e and commuting with f . To see this put ef = e ∧ f + e ∧(cid:0)1 − f(cid:1) ≤ e ≤ (cid:2)(cid:0)1 − e(cid:1) ∧ f + (cid:0)1 − e(cid:1) ∧(cid:0)1 − f(cid:1)(cid:3)c One has [ef , f ] = 0 = [ef , f ] and ef = e = ef if and only if [e , f ] = 0 . From monotonicity of the assignment e 7→ ef one finds that ef is the minimal projec- tion commuting with f and larger than e . Correspondingly ef is the maximal projection smaller than e commuting with f . Similarly given e and a multi- plet F = (f1 ,··· , fn) of projections there exists a unique minimal projection eF ≥ e and a unique maximal projection eF ≤ e commuting with each fk for k = 1,··· , n . For example if n = 2 we may consider the monotone increasing sequence = ef . e ≤ ef1 ≤ (ef1 )f2 ≤ ((ef1 )f2 )f1 ≤ ··· and check that each projection in this sequence is smaller than any projection larger than e and commuting with both f1 and f2 . Since all projections are contained in the sublattice generated by {e, f1, f2} the sequence must become stationary after finitely many steps. Then the corresponding element commutes with both f1 and f2 and obviously is the minimal projection with this property that dominates e . By induction this argument extends to any finite set F = (f1 ,··· , fn) and the case of eF follows by symmetry since (eF )c = (ec)F . Now let q : A ։ B be as above and assume that A is of finite type which implies that B too is of finite type. We want to construct a P o-map σ : B → A which is a cross section for q and a P ∧-map restricted to a chosen countable sublattice P0 ⊆ P(B) as above. Choose a enumeration of the countable sublattice P0 whose linear span is dense in some the sublattice generated by Eγ0 . We first treat the case of finite γ0 . Assume by induction that one has already constructed a coherent lift s = sγ0 : Λγ0 → P(A) which is a cross section for q meaning that the following must be satisfied: transfinite enumeration (cid:8)eγ(cid:9) of the elements of P(B) beginning with a sequential separable subalgebra containing B0 . Put Eγ0 =(cid:8)eγ(cid:12)(cid:12) γ < γ0(cid:9) and let Λγ0 denote s(e ∧ f ) = s(e) ∧ s(f ) , e ⊥ f =⇒ s(e)⊥ s(f ) 72 U. HAAG also implying e ≤ f =⇒ s(e) ≤ s(f ) . We also assume by induction that if d ∈ Λγ0 is a central element then s(d)∨ s(e) = s(d ∨ e) for any e ∈ Λγ0 and s(dc) = s(d)c . Of course one presets the values s(0) = 0 , s(1) = 1 . Note that if s is a P-lattice map and a cross section then s(cid:0)Λγ0(cid:1) ⊆ P(A) is a sublattice so that for any eγ ∈ Λγ0 the element s(eγ) is minimal (unique) in the lattice s(cid:0)Λγ0(cid:1) with the property that it is a preimage of eγ , in other words the lattice generated by the images of Λγ0 contains no elements which lie in the kernel of q . Let Zγ0+1 ⊆ Λγ0+1 denote the center of Λγ0+1 consisting of those projections commuting with any other projection. Since Λγ0+1 is finite so is its subset of minimal central projections {c1 ,··· , cn} which generate the abelian lattice Zγ0+1 . We may then extend the given lift s to a coherent lift on Λγ0 ∪Zγ0+1 which is an A-lattice-map restricted to Zγ0+1 in the manner above and on noting that each element in Λγ0+1 has a canonical central decomposition as eγ = ck eγ nXk=1 s nXk=1 ck eγ! = nXk=1 s(ck) s(eγ) . extend this lift to a coherent P ∧,a-lift on the sublattice generated by Λγ0 and Zγ0+1 by defining γ0 γ0 enumeration of the minimal central elements of Zγ0+1 Λγ0 , i.e. dl ≤ ckl such that if Λk then eδk is minimal in the complement of Λk−1 the minimal central element of Λk−1 Thus s is an A-lattice-map restricted to Zγ0+1 Zγ0 . Let (cid:8)d1 ,··· , dm(cid:9) be an minimal element ckl ∈ Zγ0+1 . Then choose elements {eδk} ⊆ Λγ0+1\(cid:0)Zγ0+1 Λγ0(cid:1) γ0 is the sublattice of Λγ0+1 generated by Zγ0+1 Λγ0∪(cid:8)eδ1 ,··· , eδk(cid:9) γ0 (cid:8)s(eγ)(cid:12)(cid:12) eγ < eδk(cid:9) , γ0 (cid:8)s(eκ)(cid:12)(cid:12) eκ > eδk(cid:9) . . By abuse of notation let dk be For each k we want to extend s to a coherent lift of the sublattice dk Λk exceeding eδk . Define eδk = sup eδk = inf for some eκ∈Λk−1 eγ ∈Λk−1 γ0 γ0 beginning unchanged. Choose any eδk ≤ zδk ≤ eδk preimage zδk of eδk with with a lift of eδk leaving the images of (cid:0)1 − dk(cid:1) Λk−1 then considering the sublattice eΛk being of finite type a minimal element s(dk) which is a preimage of dk . Obviously . Then being minimal s(dk) must be central, since s(dk) = s(dk) η = s(dk) − s(dk) is a central element in the kernel of q . Define γ0 (cid:1) and zδk it contains γ0 (cid:1)∪{zβk } s(cid:0)Λk−1 s(eδk ) = (cid:0)1 − η(cid:1) zδk , s(eγ) = (cid:0)1 − η(cid:1) s(eγ) . One checks that each element in the sublattice (cid:0)1 − η(cid:1)eΛk for eγ ∈ Λk−1 is minimal with the property of being a preimage for its image under q . Therefore the γ0 and the lattice operations are in one-to-one correspondence γ0 generated by s(cid:0)dk Λk−1 γ0 γ0 elements of (cid:0)1− η(cid:1)eΛk THE JORDAN LATTICE COMPLETION 73 with the elements and lattice operations of dk Λk a P-lattice map. This in turn implies that the complementary map γ0 so that s extends uniquely to eγ 7→ η s(eγ) =: ηγ for eγ ∈ Λk−1 γ0 is a P ∧,o-map which extends to a P-map on dk Λk γ0 putting ηλ = η eλ , eλ = sup eγ ∈Λk−1 γ0 (cid:8)s(eγ)(cid:12)(cid:12) eγ < eλ(cid:9) . Monotonicity and orthogonality of this map are obvious. Then we need to check that it is a wedge-map. Since we are considering finite lattices the suprema are attained so that eλ = s(eγλ) where eγλ < eλ is the unique maximal element in Λk−1 γ0 which is smaller than eλ . Then the equality eγλ ∧ eγµ = eγλ∧µ is straight- forward and s being a wedge map on Λk−1 by induction assumption the result γ0 follows. Define the extension of s to dk Λk γ0 by s(eλ) = s(eλ) + ηλ which again is a P ∧,o-map. By induction the argument carries on to give a P ∧,o- map extension of s to all of Λγ0+1 which again satisfies the extra assumption that the images of central elements remain central completing the induction step. This process leads to a P ∧,o-map defined on P0 . It is then easy to extend this to a P o-map on P(B) by inductively defining s(eγ0) = sup(cid:8)s(eγ)(cid:12)(cid:12) eγ < eγ0 , γ < γ0(cid:9) for infinite γ0 . Monotonicity and orthogonality of this extension are obvious. By duality one obtains a P co-map s which is a P ∨,co-map restricted to P0 putting es(eγ) = 1 − s(ec γ) . sections (cid:8)sγ0 Next assume that q is sequentially normal. Applying the above construction con- sider the sequence of (nonunital) cross sections sγ0 : Λγ0 → P(A) corresponding to the minimal P-lattice-map part of s restricted to Λγ0 for finite index γ0 as above. Then this leads for fixed Λγ0 to a monotone decreasing sequence of cross γ (cid:9) which are the restrictions of sγ to Λγ0 . Since each element of P0 has a finite index the limit of these maps yields a well defined cross section s : P0 → P(A) from the fact that q is sequentially normal. Let us show that it is a P-lattice map with respect to the unit element s(1) . Since all images of a pair of complementary projections {e, 1 − e} under the family of P c-maps {sγ} (with respect to the unit element sγ(1) ) commute with each other it is easy to see that the limit map is again complemented. Then one only needs to show that it respects wedges. Being represented as a limit of a family of monotone decreasing P ∧-maps this is again obvious. Fixing the unit s(1) ∈ P(A) , i.e. replacing P(A) in the manner described above. We now turn to the important subject of constructing cross sections (or general P-maps) having decoration a . We first show that any P-maps of type ax or xx have decoration a . In case of a P ax-map s suppose that [p , q] = 0 are commuting projections. Without loss of generality we may assume that p , q ∈ P+ . We need to show that by P(cid:0)s(1) A s(1)(cid:1) this map may be extended to a P c-map on the whole of P(B) s(p) = s(p)s(q) = (cid:0)s(p) ∨ s(qc)(cid:1) ∧(cid:0)s(p) ∨ s(q)(cid:1) = (cid:2)(cid:0)s(p) ∨ s(qc)(cid:1) − s(qc)(cid:3) + (cid:2)(cid:0)s(p) ∨ s(q)(cid:1) − s(q)(cid:3) . 74 U. HAAG Considering only the second summand of the expression in the second line one has (cid:2)s(p) ∨ s(q) − s(q)(cid:3) = (cid:0)s(p ∨ q) − s(q)(cid:1) = s(cid:0)(p ∨ q) − q(cid:1) = s(cid:0)p ∧ qc(cid:1) with (p − p ∧ q)⊥ (q − p ∧ q) =⇒ (cid:0)s(p) − s(p ∧ q)(cid:1)⊥(cid:0)s(q) − s(p ∧ q)(cid:1) implying (cid:2)s(p) , s(q)(cid:3) = 0 . where the first equality follows from p, q ∈ P+ , the second and third from [p , q] = 0 ⇒ (p∨q)−q = p∧qc and p∧qc ≺ qc ⇒ s(p∧qc)+s(q) = s(p∨q) , or in case that pc ∧ qc ≺ qc one gets s(p ∨ q) − s(q) = s(p ∨ q) s(qc) = s(cid:0)(p ∨ q) qc(cid:1) = s(cid:0)p ∧ qc(cid:1) . Therefore s(p)∧s(qc) = s(p)s(q)∧s(qc) = s(qc) s(p)s(q) s(qc) ≥ s(qc) s(p) s(qc) ≥ s(p)∧s(qc) implying that s(qc) s(p) s(qc) is a projection which can be the case only if s(p) commutes with s(q) . In case of a P xx-map assume without loss of generality that p , q ∈ P− so that s(p) = s(p − p ∧ q) + s(p ∧ q) , s(q) = s(q − p ∧ q) + s(p ∧ q) If p ⊥ q are orthogonal projections, then s(p)⊥ s(q) since s is complemented. On the other hand either one of the projections, say p , is contained in Λ− . If also q ∈ Λ− we have p ∨ q ∈ Λ− in case of a polar signature, therefore s(p) + s(q) = s(p) ∨ s(q) = s(p ∨ q) = s(p + q) . If q ∈ Λ+ one gets s(p + q) − s(p) = s(p + q) ∧ s(pc) = s((p + q) ∧ pc) = s(q) so that always s(p) + s(q) = s(p + q) for polar signatures. But then s must be an A-lattice map. To give an easy example notice that any signed P c-map into C is a P ax- map for the maximal (trivial) construction time order relation e , f ∈ P(A)− ⇒ e ≺ f , f ≺ e . Since any P c-map τ : A ։ C is signed for the induced signature P(A)− = τ −1(0) , P(A)+ = τ −1(1) any P c-map into C can be regarded as a P ax-map, even a P x-map for some signature. j(1Λ0 ) 6= 1Λ ) Consider the following problem: given a (possibly nonunital, i.e. inclusion of monotone complete complemented lattices j : Λ0 ⊆ Λ one may ask whether it is possible to construct a monotonous retraction r : Λ ։ Λ0 , i.e. r◦ j = id which is a P a-map, possibly with some additional features. The methods used above yield natural minimal and maximal retractions r , r : Λ ։ Λ0 which are ∧- and ∨-maps respectively on putting r(e) = r(ec)c r(e) = sup(cid:8)x ∈ Λ0(cid:12)(cid:12) j(x) ≤ e(cid:9) , but these will not be P a-maps in general. Simple examples like the inclusion of C ⊕ C ⊆ M2(C) exhibit that they cannot be monotonous or even continuous in In any case the maps are completely natural so they may prove to be general. relevant in a certain setting, e.g. an inclusion of von Neumann algebras N ⊆ M . We will give some simple counterexamples for the existence of A-lattice retractions. To begin with we consider tensor type inclusions j : Mk(C) ֒→ Mk(C) ⊗ Ml(C) with j(p) = p ⊗ 1l . Assume that l = 2 . Putting n = 2 k and fixing some identification Mn(C) ≃ M2(C) ⊗ Mk(C) corresponding to pairwise commuting unital inclusions j : M2(C) ֒→ Mn(C) , j′ : Mk(C) ֒→ Mn(C) consider the subset of projections of Mn(C) having an orthogonal decomposition by minimal projections of the form {p q} with p the image of a minimal projection in M2(C) and q If p q ⊥ p′ q′ are two orthogonal the image of a minimal projection in Mk(C) . else q ⊥ q′ . Therefore the elements which are orthogonal sums of such minimal orthogonal sum of minimal projections in Γ so is P c , so that P ≥ Q and P , Q ∈ Γ also implies P−Q ∈ Γ . We first consider the problem of constructing an A-lattice projections with p , p′ ∈ j(cid:0)M2(C)(cid:1) and q , q′ ∈ j′(cid:0)Mk(C)(cid:1) then either p ⊥ p′ or projections constitute an A-sublattice Γ ⊆ P(cid:0)Mn(C)(cid:1) . Check that if P is an THE JORDAN LATTICE COMPLETION 75 extends to an A-lattice retraction retraction. Such a map necessarily restricts to an A-lattice map τ : j(cid:0)M2(C)(cid:1) ։ C corresponding to some (polar) signature on P(cid:0)M2(C)(cid:1) . This map canonically on the A-sublattice Γ ⊆ P(cid:0)Mn(C)(cid:1) generated by minimal projections of the form add up to 1 whenever 1n =Pn τ an A-lattice-extension of rΓ to all of P(Mn(C)) . Let U ⊆ j(cid:0)M2(C)(cid:1) be the {p q} as above such that rΓ(p q) = τ (p) q . Check that the projections {τ (pk) qk} k=1 pk qk . So the task is to find for a given map rΓ : Γ ։ P(cid:0)Mk(C)(cid:1) unitary group of the image of M2(C) . Put pU = U j(cid:18)(cid:18)1 0 0 0(cid:19)(cid:19) U ∗ = U p1 U ∗ for a unitary U ∈ U so that every minimal projection in j(cid:0)M2(C)(cid:1) is of the form p = pU . Each minimal projection in Mn(C) is determined by giving a collection of positive real numbers {λ1 , ··· , λn} summing up to 1 representing the diagonal and a selfadjoint n× n-matrix (ωij)ij of complex numbers of modulus 1 satisfying ωji = ωij and ωij ωjk = ωik with respect to some chosen basis subordinate to the inclusions j , j′ . Then the projection is of the form p q iff there exists a collection of positive real numbers {µ1 , ··· , µk} summing up to 1 , a k× k-matrix (ρmn)mn of complex numbers of modulus 1 satisfying ρnm = ρ∗ mn , ρmq ρqn = ρmn and 0 ≤ λ ≤ 1 , ω ∈ C , ω = 1 such that λi = λ µi for 1 ≤ i ≤ k , λi = (1 − λ) µi for k + 1 ≤ i ≤ n and ωij = ρij for 1 ≤ i , j ≤ k , ωij = ω ρ(i−l)j for k + 1 ≤ i ≤ n , 1 ≤ j ≤ k and ωij = ρ(i−l)(j−l) for k + 1 ≤ i, j ≤ n . Consider the subset of minimal projections which are of the form r = (cid:0)p1 + pc 1 V(cid:1) p q(cid:0)p1 + pc 1 V(cid:1) with p q ∈ Γ and V = V ∗ = V −1 , V ∈ j′(cid:0)Mk(C)(cid:1) a symmetry. We claim that it exhausts all minimal elements of P(cid:0)Mn(C)(cid:1) . First check that for any collection of positive real numbers {λ1 , ··· , λn} summing up to 1 there exists p q and V such that r has diagonal (λ1 , ··· , λn) , moreover the entries of the selfadjoint matrix (ωij)ij of complex numbers of modulus 1 corresponding to r can be chosen arbitrary for indices 1 ≤ i , j ≤ k and k + 1 ≤ i , j ≤ n since V is an arbitrary 1 is congruent to an arbitrary minimal projection in the range of pc 1 up to a scalar multiple 1 − λ . Then the projection r is fully determined by giving one more matrix coefficient, say ω1n . This will determine the coefficients ωkn = ω1k ω1n for 1 ≤ k ≤ l , and these determine all other coefficients according to ωkl = ωkn ωln . Any such coefficient is achieved by changing the phase ω of p if necessary by the appropriate value. Of course by symmetry there is a representation of r as symmetry of j′(cid:0)Mk(C)(cid:1) so that pc 1 p (V qV ) pc for every unitary U ∈ M2(C) (again U can be chosen a symmetry modulo the commutant of p ). Then one finds that r is dominated by each twodimensional r = (cid:0)pU + pc projection from the set (cid:8)pU qU + pc U V U(cid:1) pU qU(cid:0)pU + pc U (V U qU V U )(cid:9) ⊆ Γ so that the image of r must be smaller than each value τ (pU ) qU + τ (pc U ) V U qU V U which is either equal to qU or else to V U qU V U . As soon as two different values of minimal projections appear in this set which certainly is the case for r /∈ Γ any monotonous extension of rΓ must send r to 0 . But then there exist orthogonal bases containing only U V U(cid:1) 76 U. HAAG exist. Since any P c-map defined on an A-sublattice has a P c-extension one finds that there always exists a P c-retraction r : Mn(C) ։ Mk(C) for j which is an A-lattice map on Γ , however no such extension can be expected to have decoration elements not in Γ . Therefore an A-lattice extension of rΓ to P(cid:0)Mn(C)(cid:1) cannot a if Γ is not a proper A-sublattice of P(cid:0)Mn(C)(cid:1) which seems not to be the case. of the form x = Pn Suppose given a P2-map σ : A → B . We need to show that it is monotonous. Given a proper A-sublattice E ⊆ P(A) let Pf (E) ⊆ Pf (A) denote the subset of elements k=1 αk pk with 0 < p1 < ··· < pn and {p1 , ··· , pn} ⊆ E . Then a P-map σ : A → B is said to be of concave type on E if for any pair of commuting positive elements x , y in the positive affine span of E one has σ(x + y) ≥ σ(x) + σ(y) . Check by induction on the number of different eigenvalues of y ∈ Pf (E) that a P-map σ is of concave type on E iff the inequality σ(cid:0)x + β q(cid:1) ≥ σ(x) + β σ(q) , β ∈ R+ is a P a∧-map restricted to hEi is of concave type. In fact let x =Pn holds for any element x ∈ Pf (E) and q ∈ E with [q , x] = 0 . Any P-map which k=1 αk pk for 0 < p1 < ··· < pn denote its canonical decomposition. Without loss of generality pn < 1 and x ≥ 0 since always σ(α1 + x) = α1 + σ(x) . Then αk pk! σ(x + β q) = min(αn , β) σ(pn ∨ q) ≥ min(αn, β) σ(pn ∨ q) + min(αn, β) σ(pn ∧ q) +σ min(αn, β) pn ∧ q +(cid:0)β − min(αn, β)(cid:1) q +(cid:0)αn − min(αn, β)(cid:1) pn + n−1Xk=1 αk pk! αk pk! + σ (cid:0)β − min(αn, β)(cid:1) q + (cid:0)αn − min(αn, β)(cid:1) pn + ≥ min(αn, β)(cid:0)σ(pn) + σ(q)(cid:1) + σ (cid:0)β − min(αn, β)(cid:1) q + (cid:0)αn − min(αn, β)(cid:1) pn + n−1Xk=1 n−1Xk=1 where the first inequality follows by induction assumption since pn ∧ q < q, pn can be assumed without loss of generality. Then the result follows after finitely many steps by iteration of the same argument. But then again by induction one derives the general inequality σ(x + y) ≥ σ(x) + σ(y) for commuting elements [ x , y ] = 0 in Pf (E) . Let σ2 : M2(A) → M2(B) be the P-map corresponding to the P2-map σ subject to the conditions prescribed in the definition above. Then σ2 is of concave type on each subalgebra Aq,F . To prove monotonicity of σ in this situation assume given two elements 0 ≤ x ≤ y ≤ 1 in Pf (A) with x =Pn l01 βl ql , 0 < q1 < ··· < qm . Then the support projection of x is smaller than the support projection of y , i.e. pn ≤ qm . Let F = pn 12 + (1 − pn) V ∈ M2(A) so that F 2 = 1 and 0(cid:19) F(cid:19) p k=1 αk pk , 0 < p1 < ··· < pn and y =Pm σ(y) p = 2 p σ2(cid:18)F (cid:18)y σ(x) p = 2 p σ2(cid:18)F (cid:18)x 0 0(cid:19) F(cid:19) p , 0 0 0 0 0 THE JORDAN LATTICE COMPLETION 77 and 0 0 0 0 0 0 0 0 0 0 one gets Considering 0(cid:19)(cid:19) . By induction we may assume that σ(x′) ≤ σ(y′) with x′ = x − min(αn, βm) pn and y′ = y − min(αn, βm) pn has already been proved. Putting γ = min(αn, βm) let 0(cid:19) F(cid:19) = σ2(cid:18)(cid:18)x 0 σ2(cid:18)F (cid:18)x 0 eqm = (cid:18)1 qm − pn(cid:19) , epn = (cid:18)pn ex = (cid:18)x′ ey = (cid:18)y 0(cid:19) + γepn , σ(y) p = 2 p σ2(ey) p − γ(cid:0)σ(qm− pn) p + p(cid:1). σ(x) p = 2 p σ2(ex) p − γ σ(qm− pn) p, 0(cid:19)(cid:19) + γ σ2(cid:0)epn(cid:1)(cid:19) p ≤ 2 p σ2(Fey F ) p − γ p . 2 p σ2(ex) p = 2 p(cid:18)σ2(cid:18)(cid:18)x′ Since (cid:2)epn , Fey F(cid:3) = 0 one gets σ2(cid:0)Fey F(cid:1) − γ σ2(cid:0)epn(cid:1) ≥ σ2(cid:0)Fey F − γepn(cid:1) = σ2(cid:18)F (cid:18)y′ 0(cid:19) F + (cid:18)γ 1 0 0(cid:19)(cid:19) qm − pn(cid:19) . 0(cid:19) + γeqm Thus σ(x) ≤ σ(y) is equivalent to 0 2 p(cid:0)σ2(cid:0)Fey F(cid:1) − γ σ2(epn(cid:1) p ≥ σ(cid:0)y′) p + γ p ≥ σ(x′) p + γ p proving σ(x) ≤ σ(y) as claimed. Then let I be an injective C∗-algebra and λ : I → B(H) a faithful unital ∗- representation. Choose any completely positive retraction r : B(H) → I for λ . Then r is an I-module map in the sense that r(xλ(y)) = r(x) y and r(λ(y)x) = yr(x) (compare Lemma 6.1.2 of [6]). Let C ⊆ I be an abelian subalgebra and (xµ) ր x a monotone increasing net in C with limit x ∈ I . Put z = supµ λ(xµ) ∈ B(H) . If y ∈ C is arbitrary one gets since B(H) is an AΣ-algebra. By (transifinite) induction replacing C by C∗(C, y) one concludes that I is an AΣ-algebra. Similarly if x ∈ I is arbitrary then x y = r(z) y = r(cid:0)z λ(y)(cid:1) = r(cid:0)λ(y) z(cid:1) = y r(z) = y x x = r(cid:0)λ(x)(cid:1) = r(cid:0)v λ(x)(cid:1) = r(v)x = x r(v) for some partial isometry v ∈ B(H) with v∗ v ≤ λ(px) and px the support pro- jection of x∗ x . Then the support projection of r(v∗) r(v) is smaller or equal to px from the Schwarz inequality. Then 0 0 and x∗ x = x r(v∗) r(v)x = r(v∗) r(v) (x∗ x) implying r(v∗) r(v) = px and similarly r(v) r(v∗) = px∗ so that r(v) is a partial isometry giving the polar decomposition of x . Thus I is an AΣ+-algebra as claimed. Then, suppose given a P-map of P-algebras σ : A → B . Then σ is determined by its restriction s : P(A) → P(B) to the subset of positive projections. For this k=1 αk pk ∈ Pf (A) with 0 < p1 < ··· < pn ≤ 1 a finite string of increasing projections and positive coefficients αk > 0 , 1 ≤ k ≤ n . By suppose given x =Pn monotonicity σ(x) must be larger than any of the elements (cid:8)(cid:0)Pn j=k αj(cid:1)s(pk)(cid:9) , 78 U. HAAG while on the other hand it must be smaller or equal than each of the elements has the form (xij )ij , 1 ≤ i, j ≤ n with respect to the matrix decomposition of A j=1 αj(cid:1)(cid:0)1 − s(pk)(cid:1)(cid:9) . These conditions imply that σ(x) j=k αj(cid:1)s(pk) +(cid:0)Pn (cid:8)(cid:0)Pn corresponding to the finite set of orthogonal projections (cid:8)s(pk) − s(pk−1) k = 1 ,··· , n(cid:9) putting p0 = 0 such that xij = 0 whenever i 6= j , and xii = j=k αi(cid:1)s(pk) − s(pk−1) for k = 1 ,··· , n . Therefore on the real subspace (cid:0)Pn of selfadjoint elements it is uniquely determined by the values {s(pk)} and the condition σ(α1 + x) = ασ(1) + σ(x) (or else σ(x) = σ(x+) − σ(x−) ). Since the spectral projections of x2 in the canonical decomposition as above coincide with those of x for a positive element x ∈ Pf (A) one checks the relation σ(x2) = σ(x)2 for such elements and the case of general positive elements follows by continuity. If σ is a homogeneized P o-map one has for an element in x ∈ Pf (A) that the spectral projections of x corresponding to positive eigenvalues are orthogonal to the spectral projections corresponding to negative eigenvalues. Therefore σ(x2) = σ(x2 + + x2 −) ≥ σ(x2 +) + σ(x2 −) = σ(x+)2 + σ(x−)2 = σ(x)2 proving the Schwarz inequality for selfadjoint elements in P(A)f . The general case again follows by continuity. If σ is a homogeneized P a-map of concave type it is in particular a P o-map hence satisfies the Schwarz inequality for selfadjoint elements. If z = x + iy is a normal element then σ(zz∗) = σ(x2 + y2) ≥ σ(x2) + σ(y2) ≥ σ(x)2 + σ(y)2 = σ(z) σ(z∗) whence σ satisfies the Schwarz inequality for normal elements. Suppose that σ is a homogeneized P a-map of convex type. If x = x+ − x− is a selfadjoint element then σ(x2) = σ(x2 + + x2 −) ≤ σ(x2 +) + σ(x2 −) = σ(x+)2 + σ(x−)2 = σ(x)2 . Then let z = x + iy be normal with x , y selfadjoint. One computes σ(z z∗) = σ(x2 + y2) ≤ σ(x)2 + σ(y)2 = σ(z) σ(z∗) so that σ satisfies the reverse Schwarz inequality for normal elements. Suppose that σ is a P a,c-map and 0 ≤ x ≤ 1 is a positive contractive element in P(A)f k=1 αk pk as above. Then y = 1 − x is k=1 βk (1 − pk) + βn+1 1 where βk ≥ 0 for all k . From the complementary condition one checks that σ(y) = 1− σ(x) . Therefore if u = x + iy is unitary with x , y selfadjoint one has with canonical decomposition x = Pn of the form y = Pn 1 = σ(cid:0)x2 + (1− x2)(cid:1) = σ(x2) + σ(1− x2) = σ(cid:0)x + ip1 − x2(cid:1) σ(cid:0)x− ip1 − x2(cid:1) = σ(cid:0)x − ip1 − x2(cid:1) σ(cid:0)x + ip1 − x2(cid:1) proving that σ(cid:0)x + i y(cid:1) is unitary. If A has polar decomposition any selfadjoint element can be written in the form x = V x for some not necessarily unique selfadjoint unitary V defining a positive projection pV = (1 + V )/2 ⇐⇒ V = 2 pV − 1 . Defining an extension of σ to selfadjoint elements by σ(V x) = (cid:0)2 σ(pV ) − 1(cid:1) σ(x) and to arbitrary elements by σ(x + iy) = σ(x) + i σ(y) gives that also σ(cid:0)u) = σ(V x) + i σ(W y) = σ(V ) σ(x) + i σ(W ) σ(y) THE JORDAN LATTICE COMPLETION is unitary since (cid:2)σ(V ) , σ(W )(cid:3) = (cid:2)σ(x) , σ(W )(cid:3) = (cid:2)σ(V ) , σ(y)(cid:3) = (cid:2)σ(x , σ(y)(cid:3) = 0 . This proves the Proposition 79 (cid:3) Remark. As a corollary of these results one proves the special exact double cone property of a ∗-homomorphic quotient map q of an P-algebra A (even when q is not an overall P-lattice map) which may be useful in certain instances. Since q is a ∗-homomorphism it has the following general approximate double cone property: given elements a < b in A and elements q(a) ≤ x ≤ q(b) in B there exists a ≤ y ≤ b + ǫ1 with q(y) = x . This result is fairly obvious in the case where a = 0 , b = 1 in which case one has an exact double cone property, i.e. ǫ can be chosen to be zero. This latter result is not as obvious as it may seem, it may be false for general surjective ∗-homomorphisms. In our case however we are dealing with P-algebras and the task is to construct a unital monotonous lift of some abelian P-subalgebra containing 1 and x into a corresponding abelian subalgebra of A . By nuclearity there always exists a positive linear lift, but not necessarily a unital one. However the method above which applies in any case to abelian P-algebras generated by a single selfadjoint element regardless of whether q is generally a lattice map (a ∗-homomorphism of abelian algebras is always a lattice map), shows that there exists such a monotonous P-map of some abelian P-algebra containing x into an abelian extension in A which is a cross section restricted to the separable subalgebra C∗(1, x) (even a ∗-homomorphic cross section). This proves that the double cone problem for a = 0 , b = 1 has an exact solution in the setting of P- algebras. Then also the double cone problem for a = 0 and b = p a projection has an exact solution which follows by passing to the hereditary subalgebra p A p which again is checked to be a P-algebra. The general case may be transformed to the case a = 0 , b = 1 by applying a linear transformation sending a to 0 and replacing x and q(b) , b by 1 respectively for given ǫ > 0 proving (only) the approximate double cone property for arbitrary a ≤ b which however can be shown without the assumption that A is a P-algebra. If A is an AΣ-algebra q also posesses a corresponding (approximate) triple cone property: for given ǫ > 0 , elements a , b ≥ 0 in A and an element 0 ≤ x ≤ q(a) , q(b) there exists an element 0 ≤ y ≤ a + ǫ 1 , b + ǫ1 with q(y) = x . To see this replace b , q(b) by 1 and a by a′ = √b + ǫ 1 , also replacing q(a) and −1 xpq(b) + ǫ 1 x by pq(b) + ǫ 1 respectively. Then solving the approximate double cone problem for 0 ≤ a′ , 0 ≤ x′ ≤ q(a′) gives an element 0 ≤ y′ ≤ a′ + δ 1 with q(y′) = x′ such that the spectral projection p1 y′ and x′ = pq(b) + ǫ 1 q(a)pq(b) + ǫ 1 by pq(b) + ǫ 1 xpq(b) + ǫ1 is contained in the kernel of q and the element −1 a√b + ǫ 1 −1 −1 −1 −1 −1 −1 y′′ = p1 y′ + (1 − p1 y′) y′ satisfies 0 ≤ y′′ ≤ a′+δ 1 , 1 with q(y′′) = x′ giving an element 0 ≤ y ≤ a+ǫ 1 , b+ ǫ 1 with q(y) = x . By induction one finds that there is an approximate multiple cone property for each n ∈ N as long as all cones except possibly one are pointing in the same direction, i.e for any ǫ > 0 and given positive elements 0 ≤ a1 ,··· , an and 0 ≤ x ≤ q(a1) ,··· , q(an) there exists 0 ≤ y ≤ a1 + ǫ 1 ,··· , an + ǫ 1 with q(y) = x . Again there are special cases in which (some of) the relations can be chosen exactly. 80 U. HAAG A projection filter is a subset F ⊆ P(A) of the set of positive projections in a P-lattice algebra A satisfying e ≤ f , e ∈ F =⇒ f ∈ F , ec /∈ F , e , f ∈ F =⇒ e ∧ f ∈ F . An ultrafilter is a projection filter F ⊆ P(A) such that for any p ∈ P(A) either p ∈ F or else pc ∈ F . One easily finds that an ultrafilter defines a polar signature on P(A) , which corresponds to a P-lattice map ωF : A ։ C from the second condition of a projection filter which is supposed to hold whether or not e commutes with f . A projection filter F defines a lattice ideal iff in addition the following condition holds f ∈ F , p ∈ P(A) =⇒ fp = f ∧ p + f ∧ pc ∈ F . Define an equivalence relation on P(A) by concatenation of simple equivalences f ∈ F =⇒ p ∼F p ∨ f c , p ∼F p ∧ f . The equivalence relation is compatible with taking complements since (p ∧ f )c = pc ∨ f c ∼F f c . (p ∨ f c)c = pc ∧ f ∼F pc , It is also compatible with the wedge-operation since f ∈ F implies (p∨f c−p)c ∈ F for arbitrary p ∈ P(A) whence (p ∨ f c) ∧ q ∼F (cid:2)(cid:0)(p ∨ f c − p) + p(cid:1) ∧ q(cid:3) ∧(cid:0)p ∨ f c − p(cid:1)c (p ∧ f ) ∧ q = (p ∧ q) ∧ f ∼F p ∧ q . = p ∧ q , Then for any two equivalent elements p∼F p′ there is a finite chain {f1 , ··· , fn} ⊆ F with so that if p ≤ q ≤ p′ one gets p′ ≤ p ∨ f c∼F p with f = f1 ∧ ··· ∧ fn ∈ F whence p′ = (cid:0)(cid:0)···(cid:0)p ∨ f c 1(cid:1) ∧ f2(cid:1)···(cid:1) ∨ f c q ∼F q ∨ f c = p ∨ f c ∼F p . n Thus the quotient set modulo the equivalence relation is partially ordered by the order relation induced from P(A) and is a complemented lattice denoted P(A)F = P(A) (cid:14) ∼F . A projection filter generating a lattice ideal will be called an ideal filter. An ultrafilter is automatically an ideal filter since either p ∈ F or else pc ∈ F whence f ∈ F =⇒ fp ∈ F . Any intersection of ideal filters is an ideal filter. A unital (abelian) C∗-algebra D is separably injective iff given an inclusion of oper- ator systems (abelian C∗-algebras, linear function lattices) X ⊆ Y such that X is separable together with a completely positive linear map (∗-homomorphism, linear lattice map) s : X → D there exists a completely positive linear (∗-homomorphic, linear lattice map) extension s : Y → D of s . Corollary. If A is an AW ∗-algebra and q : A ։ B a surjective ∗-homomorphism which is a P-lattice map then every separable abelian sub-C∗-algebra Dλ ⊆ B is contained in an injective (abelian) sub-C∗-algebra Iλ ≃ I(Dλ) ⊆ B which admits a ∗-homomorphic cross section sλ : Iλ → A to the quotient map. If q is an arbitrary surjective ∗-homomorphism then any selfadjoint element x ∈ Bsa is contained in an injective abelian sub-C∗-algebra of B (which admits a ∗-homomorphic lift to A ). Any quotient of an injective abelian C∗-algebra is separably injective. THE JORDAN LATTICE COMPLETION 81 Proof. Given a surjective ∗-homomorphism q : A ։ B with A an AW ∗-algebra assume that B is a P-lattice algebra and q is a P-lattice map. Let Dλ ⊆ B be a separable abelian sub-C∗-algebra. From the proof of Theorem P there exists a ∗-homomorphic cross section sλ : Dλ → C ⊆ A for the quotient map where C is some maximal abelian, hence injective sub-C∗-algebra of A . Then from the Corol- lary of Theorem 2 there exists a ∗-homomorphic extension of sλ to the injective envelope I(Dλ) and from rigidity one gets that the extension is injective. There- fore its composition with the quotient map q results in an injective ∗-homomorphic extension ιλ : I(Dλ) ֒→ B of Dλ ⊆ B . Even if q is not an overall P-lattice map its restriction to any maximal abelian (= injective) sub-C∗-algebra C ⊆ A is, so that any separable sub-C∗-algebra Dλ ⊆ D = q(C) is contained in an injective subalgebra isomorphic to I(Dλ) ⊆ D . Therefore D is separably injective We want to close this survey with some abstract nonsense. Consider the following (cid:3) any countable disjoint partition Λ =Sn∈N Λωn of Λ one has χ(Λωn) ∈ G for some If Λ is a set of cardinality larger or equal to the cardinality of the Hypothesis. continuum there exist free ultrafilters G ⊆ P(Λ) with the following property: for (uniquely determined) n ∈ N where χ(Λωn) denotes the characteristic function of the subset Λωn ⊆ Λ . Assuming the hypothesis let a (nonseparable) AW ∗-algebra A be given containing a separable ∗-ideal J ⊳ A with nonseparable quotient B = A/J so that the cardi- nality of the subset of separable abelian subalgebras {Dλ Dλ ⊆ D}λ∈Λ contained in a maximal abelian subalgebra D ⊆ B which are ordered by inclusion has the cardinality of the continuum at least (unless D is separable which case is however trivial) and that there exists a free ultrafilter G ⊆ P(Λ) satisfying the hypothesis above and ∃λ ∈ Λ , q(κ) = 1 , ∀κ ≥ λ =⇒ q ∈ G . Under these assumptions the Corollary above can be improved to the following result. Conjecture. Any quotient algebra B of an AW ∗-algebra A modulo a separable lattice ideal is an AW ∗-algebra and each maximal abelian subalgebra D ⊆ B admits a ∗-homomorphic cross section s : D → A to the quotient map. Proof. Assume given a surjective ∗-homomorphism q : A ։ B having a separable kernel J ⊳ A with A an AW ∗-algebra such that that B is a P-lattice algebra and q is a P-lattice map. Let D ⊆ B be a maximal abelian subalgebra. We will show that there exists a ∗-homomorphic lift s : D → A for the quotient map which entails that D is injective whence B is an AW ∗-algebra. From the proof of Theorem P there exists for any separable subalgebra Dλ ⊆ D a ∗-homomorphic cross section sλ : Dλ → A for q . Let {Dλ}λ∈Λ denote the set of separable subalgebras of D ordered by inclusion and C = q−1(D) ⊆ A the preimage of D . One easily checks that C is a P-lattice subalgebra of A so that the restriction of q to C is a P-lattice map. Then q determines an ideal filter F ⊆ P(C) ⊆ P(A) the subset of projections contained in J as above and generates J linearly up to norm closure since J is a (nonunital) P-lattice algebra. For any index λ ∈ Λ where e ∈ F ⇐⇒ q(e) = 1 . The complement F c = (cid:8)f ∈ P(C) (cid:12)(cid:12) f c ∈ F(cid:9) is 82 U. HAAG choose a ∗-homomorphism sλ : D → A such that sλ is a cross section restricted to Dλ as in the proof of Theorem P. Then consider the direct product homomorphism ∆ : D Πλ sλ −−−→ Yλ∈Λ A . i.e. Choose a free ultrafilter G ⊆ P(Λ) satisfying the hypothesis above and ∃λ ∈ Λ , q(κ) = 1 , ∀κ ≥ λ =⇒ q ∈ G . f ∈ P(cid:0)Yλ f ∈ P(cid:16)Yλ A(cid:1) ,(cid:8)λ ∈ Λ(cid:12)(cid:12) f (λ) = 1(cid:9) ∈ G =⇒ f ∈ GA . there exists an index λ0 ∈ Λ such that constant and contained in D , x(λ) = x(λ0) ∈ D for all λ ≥ λ0 and check that it is a P-lattice algebra. Also Then G defines an ideal filter GA ⊆ P(cid:0)Πλ A(cid:1) by Consider the subalgebra D∆ ⊆ Qλ B of elements {x} which are eventually (cid:0)∆(D)(cid:14) Πλ J(cid:1) ⊆ D∆ by construction. Define the ideal filter GF ⊆ P(cid:0)Πλ C(cid:1) by and note that the lattice ideal it generates contains P(cid:0)ΠλJ(cid:1) whence the quotient of P(cid:0)q−1(cid:0)D∆(cid:1)(cid:1) modulo the lattice ideal generated by GF is equal to P(B) by the C(cid:17) ,(cid:8)λ ∈ Λ(cid:12)(cid:12) q(f (λ))) = 1(cid:9) =⇒ f ∈ GF if {pλ}λ is a distribution of projections in q−1(cid:0)D∆(cid:1) there exists in each coset of assumption that each element of D∆ is eventually constant. By construction GA ⊆ GΣ . If C (and hence J ) is commutative any coset of projections in C which agree modulo J is countable since J is separable. In the general case there may exist uncountably many projections which agree modulo J , however J being separable projections in C which agree modulo J a normdense subsequence of projections {pn} so that the filter GA will determine a unique projection {pλ} → p ∈ C from each such distribution even if the quotient modulo the lattice ideal corresponding to GA should not be equal to P(C) (one has to take care with continuity arguments in a noncommutative projection lattice since the lattice operations themselves are not continuous). This leads to a well defined map P(cid:0)q−1(D∆)(cid:1) −−−→ P(C) which we want to be an A-lattice map. This fact is easily checked since orthogonal projections cannot be close in norm. Also this map obviously lies above the lattice map induced by GF . Therefore its restriction to ∆(P(D)) defines a cross section for q . Being an A-lattice map its composition with ∆ yields a ∗-homomorphic cross section s : D −−−→ A for q . Thus D is monotone complete hence injective. To see that B is an AW ∗- algebra assume given an increasing net of projections (eλ)λ ր e in D . Assume there exists a projection f ∈ B with f ≥ eλ for each λ but f (cid:3) e . Replacing f by f ∧ e we can assume that f < e . But then [f , eλ] = 0 = [f , e] so that {eλ , e , f} are contained in some maximal abelian subalgebra D′ ⊆ B such that the supremum of the net {eλ} in D′ is strictly smaller than e . Since B is a P- lattice algebra we may choosing some well order on the elements of P(D) consider sequences of the form f ≥ fpω1 ≥(cid:0)fpω1(cid:1)pω2 ≥ ··· ≥ eλ ∀ λ THE JORDAN LATTICE COMPLETION 83 where {pωk}k∈N is any sequence of projections in D . Then all of the projections {eλ , e , fpωk} commute with each other so that they are contained in some maxi- mal monotone complete abelian subalgebra of B and if the monotone decreasing sequence (pωk ) should not be stationary after finitely many steps one can again replace f by the smaller projection inf k{pωk} computed in some monotone com- plete abelian subalgebra containing all these elements plus the elements {eλ , e} . Proceeding in this manner one obtains a transfinite monotone decreasing sequence of projections e > f > f1 > ··· > fω ··· ≥ eλ ∀ λ which must eventually become stationary. Then the limit projection f0 = inf fω satisfies e > f0 ≥ eλ and [f0 , p] = 0 for every p ∈ D whence f0 ∈ D giving a contradiction. Thus e is the least upper bound of the projections {eλ} in P(B) whence P(D) ⊆ P(B) is relatively monotone complete and thus B is an AW ∗- algebra (cid:3) Counterexample. Consider the Toeplitz extension T = C∗(v, v∗) of the circle algebra C(T) = C∗(u, u∗) generated by a single isometry v∗v = 1 . If the quotient map q : B(H) ։ Q(H) = B(H)/K(H) onto the Calkin algebra were a P-lattice map then the induced surjection q0 : T ։ C(T) would admit a ∗-homomorphic cross section σ : C(T) → T , so the unitary u would lift to a unitary in T and any Fredholm operator would have index 0 which one knows from index theory not to be the case. References [1] Albiac F., Kalton N., Topics in Banach space theory, Graduate texts in Mathematics 233 2nd edition, Springer New York-Heidelberg-Berlin (2016) [2] Anderson, J., Extreme points in sets of positive linear maps, J. Funct. Anal. 31 (1979), 195 -- 217 [3] Asimov, Ellis, Convexity theory and its applications to functional analysis, Academic Press London/New York (1980) [4] Choi M.-D., Effros E.G., The completely positive lifting problem for C ∗-algebras Ann. Math. 104 (1976), 585 -- 609 [5] Choi M.-D., Effros E.G., Injectivity and operator spaces, J. Funct. Anal. 24 (1977), 156 -- 209 [6] E. Effros, Z.-J. Ruan, Operator Spaces, London Mathematical Sciences Monographs New Series 23, Oxford University Press, 2000. [7] R. V. Kadison, A representation theory for commutative topological algebras, Mem. Amer. Math. Soc. 7 (1951) [8] R. V. Kadison, A generalized Schwarz inequality and algebraic invariants for operator algebras, Ann. Math. 56 (1952), 494 -- 503 [9] R.V. Kadison, Irreducible operator algebras, Proc. Nat. Acad. Sci. USA 43 (1957), 273 -- 276 [10] I. Kaplansky, Projections in Banach algebras, Ann. Math. 53 (1951), 235 -- 249 [11] V. Paulsen, Completely Bounded Maps and Operator Algebras, Cambridge Studies in Advanced Mathematics 78, Cambridge University Press, 2002. 84 U. HAAG [12] G. K. Pedersen, C ∗-algebras and their automorphism groups, London Mathematical Society, Monograph 14 (1979). [13] G. K. Pedersen, Analysis Now Graduate texts in Mathematics 118, Springer Berlin- Heidelberg-New York (1989)
1912.01903	1	1912	2019-12-04T11:32:11	Commutativity in Jordan Operator Algebras	[ "math.OA" ]	While Jordan algebras are commutative, their non-associativity makes it so that the Jordan product operators do not necessarily commute. When the product operators of two elements commute, the elements are said to operator commute. In some Jordan algebras operator commutation can be badly behaved, for instance having elements $a$ and $b$ operator commute, while $a^2$ and $b$ do not operator commute. In this paper we study JB-algebras, real Jordan algebras which are also Banach spaces in a compatible manner, of which C*-algebras are examples. We show that elements $a$ and $b$ in a JB-algebra operator commute if and only if they span an associative sub-algebra of mutually operator commuting elements, and hence operator commutativity in JB-algebras is as well-behaved as it can be. Letting $Q_a$ denote the quadratic operator of $a$, we also show that positive $a$ and $b$ operator commute if and only if $Q_a b^2 = Q_b a^2$. We use this result to conclude that the unit interval of a JB-algebra is a sequential effect algebra as defined by Gudder and Greechie.	math.OA	math	Commutativity in Jordan Operator Algebras John van de Wetering Radboud University Nijmegen, Netherlands [email protected] December 5, 2019 Abstract While Jordan algebras are commutative, their non-associativity makes it so that the Jordan product operators do not necessarily commute. When the product oper- ators of two elements commute, the elements are said to operator commute. In some Jordan algebras operator commutation can be badly behaved, for instance having elements a and b operator commute, while a2 and b do not operator commute. In this paper we study JB-algebras, real Jordan algebras which are also Banach spaces in a compatible manner, of which C∗-algebras are examples. We show that elements a and b in a JB-algebra operator commute if and only if they span an associative sub-algebra of mutually operator commuting elements, and hence operator com- mutativity in JB-algebras is as well-behaved as it can be. Letting Qa denote the quadratic operator of a, we also show that positive a and b operator commute if and only if Qab2 = Qba2. We use this result to conclude that the unit interval of a JB-algebra is a sequential effect algebra as defined by Gudder and Greechie. 1 Introduction A Jordan algebra (E,∗, 1) is a commutative bilinear unital algebra satisfying the Jordan equation (a∗ b)∗ a2 = a∗(b∗ a2). Any associative algebra A (over a field of characteristic different than 2) becomes a Jordan algebra with the special Jordan product a ∗ b := 1 2 (ab+ba). Jordan algebras were originally studied in the context of quantum mechanics as a generalization of the space of observables of a quantum system [9], but due to their close connection to symmetric cones, they have since then seen use in a variety of fields [4, 2, 3]. For a Jordan algebra E we define the Jordan product operator Ta : E → E as Ta(b) := a ∗ b. Two elements a, b ∈ E are said to operator commute when TaTb = TbTa. Using the commutativity of the product, this is easily seen to be equivalent to a ∗ (c ∗ b) = (a ∗ c) ∗ b for all c ∈ E. Hence, all elements of a Jordan algebra operator commute if and only if the algebra is associative. It might then seem reasonable to think that the Jordan algebra generated by a and b (and the unit) is associative if and only if a and b operator commute. This is however not true (see for instance Remark 2.5.2 of [7]). As such a correspondence between associative subalgebras and operator commutation is useful, there are a variety of results known about restrictions that do satisfy this equivalence. For instance if b is idempotent, b2 := b ∗ b = b, then a and b operator commute if and only if they generate an associative algebra [7, Lemma 2.5.5.]. 1 The original class of Jordan algebras to be studied were the Euclidean Jordan algebras. These are finite-dimensional Jordan algebras over the real numbers that are formally real: i = 0 then ai = 0 for all i. Equivalently, these are finite- dimensional algebras that are also real Hilbert spaces with the Jordan product being self-adjoint. For Euclidean Jordan algebras it is known that elements operator commute if and only if they generate an associative subalgebra1. if Pn i a2 Euclidean Jordan algebras were later generalized to an infinite-dimensional setting in the form of JB-algebras. These are real Jordan algebras that are also Banach spaces with the Jordan product interacting suitably with the norm. Examples of JB-algebras include any norm-closed Jordan algebra A ⊆ Asa where Asa denotes the set of self- adjoint elements of a (unital) C∗-algebra. Such a Jordan subalgebra of a C∗-algebra is commonly referred to as a JC-algebra. For JC-algebras it is also known that elements operator commute if and only they span an associative subalgebra2. In this paper we will settle the question for JB-algebras. Letting Qa = 2T 2 a − Ta2 denote the quadratic map of a (that for associative algebras with the special Jordan product reduces to Qa(b) = aba), we will prove the following. Theorem. Let A be a JB-algebra and a, b ∈ A arbitrary. Then the following are equivalent. a) a and b operator commute. b) a and b generate an associative JB-algebra. c) a and b generate an associative JB-algebra of mutually operator commuting ele- ments. d) a and a2 operator commute with b and b2. If one of a and b is positive then these statements are furthermore equivalent to Qab2 = Qba2. We prove this by resorting to the structure theory of JBW-algebras (JBW-algebras are to JB-algebras as von Neumann algebras are to C∗-algebras). Along the way we also give new, more algebraic, proofs of the statements in the above theorem for Euclidean Jordan algebras and JC-algebras. As an application of our results, we show that the binary operation a & b := Q√ab restricted to the unit interval of a JB-algebra satisfies the axioms of a sequential effect algebra [5]. 2 Preliminaries Definition 2.1. A Jordan algebra (E,∗, 1) over a field F is a vector space over F equipped with a unital commutative (not necessarily associative) bilinear operation ∗ : E × E → E that satisfies the Jordan identity: We will refer to this operation as the Jordan product. (a ∗ b) ∗ (a ∗ a) = a ∗ (b ∗ (a ∗ a)) 1For instance, in [4, Lemma X.2.2] it is shown that elements operator commute if and only if they allow a simultaneous diagonalization. From that it easily follows that they generate an associative subalgebra. 2As far as the author is aware, this isn't explicitly stated anywhere, but it follows easily from [6, Lemma 5.1]. 2 Though the definition of a Jordan algebra works for any field, as the theory of Jordan algebras is slightly different for fields of characteristic 2, we will assume that the field of the Jordan algebra has any characteristic other than 2. Example 2.2. Let (A,·, 1) be an associative unital algebra over some field (not of characteristic 2). Then the operation ∗ : A × A → A defined by a ∗ b := 1 2 (a · b + b · a) makes (A,∗, 1) a Jordan algebra. We will refer to this operation as the special Jordan product of the algebra. Any Jordan algebra that is isomorphic to a subset of an associative algebra equipped with this Jordan product is called special. To proceed we need some basic algebraic properties of Jordan algebras, which are most conveniently expressed with some additional notation. Definition 2.3. Let E be a Jordan algebra. 1. We write a0 := 1, a1 := a, a2 := a ∗ a, a3 := a ∗ a2, a4 := a ∗ a3, . . . . Note that since ∗ is not associative it's not a priori clear whether equations like a4 = a2 ∗ a2 hold. 2. Given a ∈ E we write Ta : E → E for the linear operator Ta(b) := a ∗ b. We call these operators product operators. 3. Given two linear operators S, T : E → E we write [S, T ] := ST − T S for the commutator of S and T . Note that because the Jordan product is bilinear, we have Ta+λb = Ta + λTb. The following identities can be found in any textbook on Jordan algebras, but as they show some fundamental results regarding operator commutativity we include them here. These are known as the linearized Jordan equations: Lemma 2.4. Given a Jordan algebra E, and a, b, c ∈ E, we have a) [Ta, Ta2 ] = 0 b) [Tb, Ta2 ] = 2[Ta∗b, Ta] c) [Ta, Tb∗c] + [Tb, Tc∗a] + [Tc, Ta∗b] = 0; Proof. a) The first equation, [Ta, Ta2 ] = 0, is just a reformulation of the Jordan identity: TaTa2 b ≡ a ∗ (b ∗ a2) = (a ∗ b) ∗ a2 ≡ Ta2 Tab. b) Take the equality [Td, Td2 ] = 0 and let d = a ± b: expanding the terms using linearity we are left with [Ta±b, T(a±b)2 ] = 0. After [Ta, Ta2] ± [Tb, Tb2 ] ± ([Tb, Ta2 ] + 2[Ta, Tab]) + ([Ta, Tb2 ] + 2[Tb, Tab]) = 0. Subtracting the equation for d = a+b from the equation for d = a−b and dividing the result by 2 (here we use that the field is not of characteristic 2) we have the desired equation. 3 c) Take the previous equation and replace a by a± c. Using the same trick as before we arrive at the desired equation. Proposition 2.5. Given a Jordan algebra E and a, b, c ∈ E we have Ta∗(b∗c) = TaTb∗c + TbTc∗a + TcTa∗b − TbTaTc − TcTaTb (1) Proof. Apply the operators of (c)) to an element d and bring all the negative terms to the right to get a((bc)d) + b((ac)d) + c((ab)d) = (bc)(ad) + (ac)(bd) + (ab)(cd). Observe that the right-hand side is invariant under an interchange of a and d so that the left-hand side must be as well. This leads to the equality a((bc)d) + b((ac)d) + c((ab)d) = d((bc)a) + b((dc)a) + c((db)a) = ((bc)a)d + b(a(cd)) + c(a(bd)) where we have used the commutativity of the product to move d to the end in the last equality. Translating this back into multiplication operators, using that this equality holds for all d, and bringing some terms to the other side then gives the desired equation. 2.1 Operator commutativity In this section we will collect and prove some results regarding operator commutativity in general Jordan algebras. Definition 2.6. Let E be a Jordan algebra. We say a, b ∈ E operator commute when their Jordan product maps Ta, Tb : E → E commute, or equivalently when a ∗ (c ∗ b) = (a ∗ c) ∗ b for all c ∈ A. We write a b to denote that a and b operator commute. Definition 2.7. Let E be a Jordan algebra, and let S ⊆ E be some subset. We write S′ for the commutator of S, defined as S′ := {a ∈ E ; ∀s ∈ S : a s}. Proposition 2.8. Let E be a Jordan algebra and let a ∈ E. For any n ∈ N, Tan can be written as a polynomial in Ta2 and Ta. Proof. We prove by induction. It is obviously true for n = 1, 2. Suppose it is true for all k ≤ n. Then by Eq. (1) Tan+1 = Ta∗(a∗an−1 ) = TaTan + TaTan + Tan−1 Ta2 − T 2 a Tan−1 − Tan−1 T 2 a . Expanding each of the Tan and Tan−1 as polynomials of Ta and Ta2 finishes the proof. Corollary 2.9. For any a ∈ E and n, m ∈ N, an and am operator commute and an ∗ am = an+m. Corollary 2.10. For any a ∈ E, the Jordan algebra J(a) generated by a (consisting of polynomials in a) is associative. Corollary 2.11. For any a, b ∈ E, b ∈ J(a)′ iff b a, a2. 4 Let us remark that for general Jordan algebras, it could be that a and b operator commute, while a2 does not operator commute with b. It is also possible that while a and b generate an associative subalgebra, they still do not operator commute. See for instance [7, Remark 2.5.2] for explicit examples. With some more restrictions on a and b these two properties are however more related. We call an element p ∈ E idempotent when p2 = p ∗ p = p. Proposition 2.12. Let E be a Jordan algebra with a, p ∈ E where p is idempotent. Then a and p operator commute if and only if a and p generate an associative subalgebra. Proof. See any textbook on Jordan algebras, e.g. [7, Lemma 2.5.5.]. A related result to this is the following: Proposition 2.13. Let E be a Jordan algebra, and let p ∈ E be idempotent. Then {p}′ is a subalgebra of E. Proof. See e.g. [8, Lemma 1]. Proposition 2.14. Let E be a Jordan algebra with a, b ∈ E and suppose a b, b2 and b a, a2. Then a and b span an associative subalgebra of mutually operator commuting elements. Proof. By repeatedly applying Eq. (1) any Tp where p is a polynomial in a and b can be reduced to a polynomial in Ta, Ta2 , Tb, Tb2 and Ta∗b, it hence remains to show that a2 b2, and that a ∗ b operator commutes with a, a2, b and b2. By Lemma 2.4.b) we already get a ∗ b a, b. With the same equation, but now taking b := b2, we see that b2 a2 ⇐⇒ a ∗ b2 a. Applying Eq. (1) to Ta∗b2 we see that it reduces to a polynomial in Tb, Ta, Ta∗b and Tb2 and since Ta commutes with them all, it commutes with Ta∗b2, and hence a2 b2. Taking 2.4.b) with a := a2, b := a, c := b we get [Ta2 , Ta∗b] = −[Ta, Tb∗a2 ] − [Tb, Ta3 ]. As b a, a2 we also have b a3, and hence this last term dissapears. Expanding Tb∗a2 we see that [Ta, Tb∗a2 ] = 0 and hence indeed [Ta2 , Ta∗b] = 0. Showing that b2 a ∗ b follows entirely analogously. Remark 2.15. As shown by the example in Ref. [7, Remark 2.5.2], the conditions a b, b2 are necessary for the previous proposition to hold. It is unclear however whether the further assumption that also b a2 is necessary. In Ref. [8], to the authors knowledge the first paper dedicated to studying operator commutativity in Jordan algebras, it was shown that if E is a Jordan algebra over a field of characteristic zero, and B ⊆ E is a finite-dimensional semi-simple subalgebra, then B′ is a subalgebra itself. In particular, letting B = J(a) for some a ∈ E, if b a, a2, then b ∈ J(a)′, and hence, since J(a)′ is a subalgebra, also b2 a. We will see that when E is a JB-algebra, that a similar property holds. 2.2 The quadratic product The Jordan product is generally not very well-behaved. In many circumstances it turns out to be better to work with the quadratic product. Definition 2.16. Let E be a Jordan algebra and let a ∈ E be arbitrary. We define the quadratic product of a as Qa := 2T 2 a − 2Ta2 . 5 Note that Qa1 = a2 and Q1 = id. For an associative algebra A equipped with the standard Jordan product we easily calculate Qab = aba, hence the name quadratic product. The quadratic product satisfies the following identity, sometimes referred as the fundamental equality of quadratic Jordan algebras. Proposition 2.17. Let E be a Jordan algebra, and a, b ∈ E arbitrary. Then: QQab = QaQbQa Note that for special Jordan algebras this simply reduces to the evidently true equa- tion (aba)c(aba) = a(b(aca)b)a. Nevertheless, in the general case it is surprisingly hard to prove. In most textbooks it is proven only as a consequence of MacDonalds theo- rem (which states that any polynomial equality in two variables that holds for special Jordan algebras, holds for all Jordan algebras) [7, 1, 11]. Although for the special case of finite-dimensional formally real Jordan algebras it can also be proven using analytic means [4]. In Ref. [14] a semi-automated algebraic proof is given. We will use the fundamental identity without further reference troughout this paper. Note that as an easy consequence we have Qa2 = QQa1 = QaQ1Qa = Q2 a. 2.3 JB-algebras Definition 2.18. Let (A,∗, 1,k·k) be a real Banach space (complete normed vector space) that is also a Jordan algebra. The space A is called a JB-algebra (Jordan- Banach) if the Jordan product ∗ satisfies for all a, b ∈ A: a) ka ∗ bk ≤ kakkbk. b) (cid:13)(cid:13)a2(cid:13)(cid:13) = kak2. c) (cid:13)(cid:13)a2(cid:13)(cid:13) ≤ (cid:13)(cid:13)a2 + b2(cid:13)(cid:13). Remark 2.19. In [7], JB-algebras are allowed to not have a unit. We will only deal with unital JB-algebras. Proposition 2.20 ([7, Proposition 3.1.6]). Let A be a JB-algebra. Then A is an order unit space complete in the order-unit norm, and for all a ∈ A: −1 ≤ a ≤ 1 =⇒ 0 ≤ a2 ≤ 1 Conversely, any complete order unit space with a Jordan product satisfying the above equation is a JB-algebra. As a JB-algebra is an order unit space, it comes with a partial order ≤. The positive elements a ≥ 0 in a JB-algebra precisely correspond with the squares: a ≥ 0 ⇐⇒ ∃b : a = b2. Note that the Jordan product maps Ta are not positive. I.e. if a, b ≥ 0, then it is not necessarily the case that a ∗ b ≥ 0. The quadratic map is better behaved: Proposition 2.21 ([7, Proposition 3.3.6]). Let A be a JB-algebra, and let a, b ∈ A be arbitrary. If b ≥ 0 then Qab ≥ 0. Example 2.22. Let A be a unital C∗-algebra. Then the set of self-adjoint elements Asa forms a JB-algebra with the Jordan product a ∗ b := 1 2 (ab + ba). 6 Definition 2.23. Let A be a JB-algebra. We say A is a JC-algebra when there exists a C∗-algebra A so that A is isometrically isomorphic to a norm-closed subset of Asa. Note that the isometry φ : A → Asa mapping a JC-algebra into its C∗-algebra is necessarily a Jordan homomorphism [17]. When studying JC-algebras as a Jordan algebra we can then hence without loss of generality assume it be a Jordan subalgebra of Asa. Definition 2.24. We call a Jordan algebra E Euclidean if it is also a finite-dimensional Hilbert space in such a way that the Jordan product operators are self-adjoint. Proposition 2.25. A Euclidean Jordan algebra (EJA) is a JB-algebra. Conversely, any finite-dimensional JB-algebra is a Euclidean Jordan algebra. Proof. EJAs are precisely formally real finite-dimensional Jordan algebras [4, Proposi- tion VIII.4.2], and as shown in Ref. [7, Corollary 3.1.7], any finite-dimensional formally real Jordan algebra is a JB-algebra. Conversely, by [7, Corollary 3.3.8], any JB-algebra is formally real, and hence any finite-dimensional JB-algebra is an EJA. Let a ∈ A be an element of a JB-algebra. The Jordan algebra spanned by a is associative, and as the Jordan product is continuous in the norm we can take the closure of the algebra, denoted as C(a), and this algebra is still associative and furthermore consists of mutually operator commuting elements. An associative JB-algebra is easily seen to be equivalent to the self-adjoint part of a commutative C∗-algebra, and hence by the Gel'fand representation C(a) is isomorphic to the set of continuous functions from some compact Hausdorff space to the real numbers. As a result we get a functional calculus for a, which in particular allows us to define a square root √a that operator commutes with a. See [7, Section 3.2] for more details. 2.4 JBW-algebras Let A be a JB-algebra. We call a subset S ⊆ A directed when for any two elements s1, s2 we can find a third element s ∈ S such that s1 ≤ s and s2 ≤ s. The subset is bounded when there is some a ∈ A such that for all s ∈ S, s ≤ a. We call A bounded directed-complete when any non-empty bounded directed set has a supremum. We denote the supremum of a bounded directed set S by W S. Definition 2.26. Let f : A → B be a positive map between JB-algebras. We call f normal when f (W S) = W f (S) for any bounded directed set S that has a supremum in A. If B = R then we call such a map a normal state. We say the normal states are separating when f (a) = f (b) for all normal states f implies that a = b. Definition 2.27. A JB-algebra A is a JBW-algebra when it is bounded directed- complete and has a separating set of normal states. Example 2.28. Let A be a von Neumann algebra, i.e. a C∗-algebra that is bounded directed-complete and has a separating set of normal states [10]. Then its set of self- adjoint elements Asa is a JBW-algebra with the standard Jordan product. Definition 2.29. A JBW-algebra A will be called a JW-algebra when it is isomorphic to an ultraweakly closed subset of the self-adjoint elements of a von Neumann algebra. 7 Just as every C∗-algebra A embeds into its double dual A∗∗ which is a von Neumann algebra, so does every JB-algebra A embed into its double dual A∗∗ which is a JBW- algebra [7, Theorem 4.4.3]. Definition 2.30. Let A be a JBW-algebra, denote by V the vector space spanned by its normal states. The weak topology of A is the σ(A, V ) topology, i.e. it is the weakest topology so that every map in V is continuous. Concretely, a net aα converges weakly to a if ω(aα) converges to a for every normal state ω. We collect below a few well-known results regarding the weak and norm topology that we will use throughout the paper without further reference. Proposition 2.31. Let A be a JBW-algebra and let a ∈ A be an arbitrary element. a) Norm convergence implies weak convergence [7, Remark 4.1.3]. b) Let S ⊆ A be a bounded directed subset. Then the net (as)s∈S converges weakly to V S [7, Remark 4.1.3]. c) The operators Ta and Qa are weakly continuous [7, Corollary 4.1.6]. Definition 2.32. Let a ∈ A be an arbitrary element of a JBW-algebra A. Let W (a) denote the weak closure of C(a), i.e. the JBW-algebra generated by a. Since C(a) consists of mutually operator commuting elements and the Jordan prod- uct is weakly continuous, W (a) also consists of mutually operator commuting elements and in particular, it is associative (see Ref. [7, Remark 4.1.10] for the details). Proposition 2.33 ([7, Proposition 4.2.3]). Let A be a JBW-algebra. The linear span of the idempotents of A lies norm-dense in A. In the next result we write JB(W)-algebra to note that the result holds for both JB-algebras and JBW-algebras. Proposition 2.34. Let S ⊆ A be a Jordan subalgebra of the JB(W)-algebra A. Then S′ is a JB(W)-subalgebra. Proof. Let an ∈ S converge to some a (not necessarily in S) in the norm. Suppose b ∈ S′. Then for any c ∈ A, TbTac = b ∗ (c ∗ a) = limn b ∗ (c ∗ an) = limn TbTan c = limn Tan Tbc = limn(b ∗ c) ∗ an = (b ∗ c) ∗ a = TaTbc. So any b ∈ S′ also commutes with everything in the norm closure of S. Without loss of generality we may then assume that S is a JB-subalgebra. Suppose first that A is a JBW-algebra. Then similarly we may assume that S is weakly closed and hence is a JBW-algebra. But then b ∈ S′ iff b p for every idempo- tent p ∈ S′, and hence S′ = Tp2=p∈S {p}′. But as each {p}′ is a Jordan algebra by Proposition 2.13, S′ will also be a Jordan algebra. It is easy to show that any norm or weak convergent net bn ∈ S′ also operator commutes with all in S′, and hence S′ is a JBW-subalgebra. If A is only a JB-algebra, then we can view A as embedded into A∗∗. As A lies weakly dense in A∗∗, the commutator S′ restricted to A agrees whether taken as commutators in A or A∗∗. Hence, if b, b′ ∈ S′, then b ∗ b′ ∈ S′ as S′ is a subalgebra in A∗∗. So S′ is a Jordan subalgebra. It is easily seen to be norm-closed and hence is a JB-algebra. 8 Using this proposition we can derive a stronger version of Proposition 2.14 for JBW- algebras. Proposition 2.35. Let A be a JBW-algebra, and a, b ∈ A arbitrary. If b a and b a2, then b2 a and there is an associative subalgebra B consisting of mutually operator commuting elements such that W (a), W (b) ⊆ B. Proof. If b a and b a2, then b ∈ J(a)′. By weak continuity of the Jordan product, b ∈ W (a)′. By the previous proposition W (a)′ is a subalgebra, so also b2 ∈ W (a)′. Hence in particular b2 a. Then by Proposition 2.14, a and b generate an associative Jordan algebra S of mutually operator commuting elements. Let B be the norm and weak closure of S, then B has the desired properties. 2.5 Structure of JBW-algebras The JC-algebras are special Jordan algebras since they come from the associative prod- uct of the underlying C∗-algebra. The counterpart to that is an exceptional algebra that is not equal to some part of an associative algebra. Definition 2.36. Let A be a JB-algebra. We call A purely exceptional when any Jordan homomorphism φ : A → Asa onto a C∗-algebra A is necessarily zero. Theorem 2.37 ([7, Theorem 7.2.7]). Let A be a JBW-algebra. Then there is a unique decomposition A = Aex⊕ Asp where Asp is a JW-algebra and Aex is a purely exceptional JBW-algebra. Example 2.38 ([12]). A compact Hausdorff space X is called hyperstonean when it is extremally disconnected and C(X) is separated by normal states (i.e. when C(X) is a JBW-algebra). Let E = M3(O)sa denote the exceptional Albert algebra. Denote by C(X, E) the set of continuous functions f : X → E. Then C(X, E) is a purely exceptional JBW-algebra with the Jordan product given pointwise by (f ∗ g)(x) = f (x) ∗ g(x). The above example is actually the only type of purely exceptional JBW-algebra, as the following result by Shultz shows. Theorem 2.39 ([12]). Let A be a purely exceptional JBW-algebra. Then there exists a hyperstonean compact Hausdorff space X, i.e. that is extremally disconnected and where C(X) is separated by normal states, such that A ∼= C(X, M3(O)). Since any JBW-algebra splits up into a direct sum of a JW-algebra and an algebra of the form C(X, M3(O)sa), many questions regarding JBW-algebras can be settled by studying von Neumann algebras and Euclidean Jordan algebras (of which M3(O)sa is an example). Although direct proofs seem preferable, for some results it is not clear how to construct a direct proof, such as for the results of this paper. 3 Main results We are now ready to start proving our main results. We first study operator commuta- tion in JC-algebras. Then we study it in Euclidean Jordan algebras. Then we combine these results with the structure theory of JBW-algebras of the previous section to derive some consequences for general JBW-algebras. Finally, we generalize this to JB-algebras. 9 Recall that an element a in a C∗-algebra is called normal when aa∗ = a∗a. This should not be confused with maps between JBW-algebras preserving suprema that are also called normal. We will not use the latter definition of 'normal' in this section. The following theorem regarding normal elements will be our primary calculational tool in this section. Theorem 3.1 (Fuglede-Putnam-Rosenblum). Let m, n, a ∈ A be elements of a C∗- algebra, with m and n normal and ma = an. Then m∗a = an∗. Proposition 3.2. Let A ⊆ A be a JC-algebra acting on the C*-algebra A. Elements of A operator commute if and only if they commute as elements of A: TaTb = TbTa ⇐⇒ ab = ba. Proof. In [6, Lemma 5.1] they prove this result using representation theory. We present here a more algebraic proof. Obviously, when ab = ba, we have TaTb = TbTa. So let us prove the converse direction. For a, b, c ∈ A we easily calculate: (TaTb − TbTa)c = 1 4 ((ab − ba)c − c(ab − ba)) (2) Hence, when a and b operator commute we have (ab − ba)c = c(ab − ba) for all c ∈ A. In particular, take c = a and c = b to get: 2aba = ba2 + a2b 2bab = ab2 + b2a Now multiply the first equation by b on the right and the second with a on the left: 2abab = ba2b + a2b2 2abab = a2b2 + ab2a As the left-hand sides now agree, we can combine the equations to get ba2b = ab2a. This equation shows that ab is normal: (ab)(ab)∗ = ab2a = ba2b = (ab)∗(ab), and hence by the Fuglede-Putnam-Rosenblum theorem, since (ab)a = a(ba) we also have ba2 = (ab)∗a = a(ba)∗ = a2b and so b and a2 commute. Recall that we had the equation 2aba = ba2 + a2b. Using the operator commutation we get aba = a2b. We claim that this equality in combination with a2b = ba2 implies that ba = ab. As the C∗-algebra A embeds into the von Neumann algebra A∗∗, and these equations continue to hold in this von Neumann algebra it suffices to show this implication for von Neumann algebras. Let r(a) denote the range-projection of a in A∗∗, i.e. the smallest projection such that r(a)a = a. Recall that r(a2) = r(a). By applying the approximate pseudoinverse of a (see Ref. [13, Section 3.4.1] for more details) to the left of aba = a2b we get r(a)ba = r(a)ab = ab. As b commutes with a2, it commutes with r(a2) = r(a), and hence ab = r(a)ba = br(a)a = ba, and we are done. Proposition 3.3. Let A ⊆ A be a JC-algebra acting on the C∗-algebra A. Let a, b ∈ A with at least one of a and b positive. Then the following are equivalent. a) QaQb = QbQa. b) Qab2 = Qba2. c) a and b operator commute. 10 Proof. a) to b) is trivial. For c) to a) we note that a and b operator commute if ab = ba in A, and hence also a and a2 operator commute with b and b2. The result then follows by the definition of Qa and Qb in terms of Ta, Ta2 , Tb and Tb2. It remains to prove b) to c). Suppose Qab2 = Qba2. Written in terms of the associative product of A this becomes ab2a = ba2b. Hence, (ab) is normal. Without loss of generality, assume that a is positive. Since (ab)a = a(ba), by the Fuglede-Putnam-Rosenblum theorem: ba2 = (ab)∗a = a(ba)∗ = a2b, so that b and a2 commute. By positivity of a, √a2 = a. Since √a2 lies in the bicommutant of a2 we then see that b also commutes with a in A and hence they operator commute in A. Remark 3.4. Of course c) implies a) and a) implies b) regardless of positivity of a and b, but for the other implications, the requirement that at least one of a and b is positive is necessary. Take for instance any non-commuting a and b satisfying a2 = b2 = 1 such as a = +ih+ − −ih− and b = 0i h0 − 1i h1. Then b) holds, but a) and c) do not. Keeping b the same, but letting a = 0i h1 + 1i h0 we get a) but not c). For our next results we will need the following powerful theorem which is shown in Theorem 7.2.5 of Ref. [7]. Theorem 3.5 (Shirshov-Cohn for JB-algebras). A JB-algebra generated by two ele- ments (and possibly the unit) is a JC-algebra. Note that the proof given in Ref. [7] applies equally well for JBW-algebras: Theorem 3.6 (Shirshov-Cohn for JBW-algebras). A JBW-algebra generated by two elements (and possibly the unit) is a JW-algebra. Proposition 3.7. Let A be a JB(W)-algebra, and suppose a, b ∈ A either operator commute or at least one of them is positive and they satisfy Qab2 = Qba2. Then the JB(W)-algebra spanned by a and b (and possibly the unit) is associative. Proof. Let B denote the JB(W)-algebra spanned by a and b. By the Shirshov-Cohn theorem, B is a JC-algebra (respectively JW-algebra). Let B denote the C∗-algebra B acts on. If a and b operator commute in A, then they of course also operator commute in B and hence by Proposition 3.2 ab = ba in B. Similarly, but by Proposition 3.3, if a and b are positive and satisfy Qab2 = Qba2, they operator commute (inside of B), and hence also ab = ba in B. In both cases we then also have a2b = ba2 so that a2 operator commutes with b when restricted to B. By Proposition 2.35, there is an associative JB(W)-subalgebra B′ of B containing both a and b. But as B is already the smallest JB(W)-algebra generated by a and b we necessarily have B′ = B. Corollary 3.8. Let a, b ∈ A in a JB-algebra with at least one of a and b being positive. If Qab2 = Qba2, then Qab2 = a2 ∗ b2. Proof. By the previous proposition, a and b span an associative algebra, and the state- ment is true for associative algebras. Proposition 3.9. Let E be a Euclidean Jordan algebra with a, b ∈ E where at least one of a and b is positive. Then the following are equivalent. a) QaQb = QbQa. 11 b) Qab2 = Qba2. c) b and b2 operator commute with a and a2. Proof. a) to b) and c) to a) are trivial, hence it suffices to prove b) to c). Nevertheless, it will be useful to first prove b) to a). So assume that Qab2 = Qba2. Since E is a Euclidean Jordan algebra, it is a Hilbert space, and hence the space of bounded operators on E, B(E), is a C∗-algebra. Recall that Qa for any a ∈ E is a self-adjoint operator and hence, if a is positive, Qa = Q2√a is a positive operator in the Hilbert space sense, i.e. positive in B(E). By the fundamental equality we have QaQ2 aQb so that QaQb is normal as an element of B(E). Hence, analogously to the proof of Proposition 3.3, we can use the Fuglede-Putnam-Rosenblum theorem to conclude that QaQb = QbQa which proves b) to a). b Qa = QQab2 = QQba2 = QbQ2 Now for b) to c), since Qab2 = Qba2, by Proposition 3.7, a, b and 1 span an associa- tive JBW-algebra B which in this case is an EJA. Then a + 1 also lies in B, and hence Qa+1b2 = (a + 1)2 ∗ b2 = Qb(a + 1)2. Using b) to a) we then see that Qa+1Qb = QbQa+1. As also QaQb = QbQa and Ta = Qa+1 − Qa − id we then necessarily have TaQb = QbTa. In a similar way we also get TaQb+1 = Qb+1Ta and hence TaTb = TbTa. As a2 is also part of the same associative JBW-algebra, we can repeat the argument with a2 instead of a to see that Ta2 Tb = TbTa2 . Similarly, we can take b2 instead of b. Lemma 3.10. Let A be a purely exceptional JBW-algebra with a, b ∈ A where at least one of a and b is positive. Then the following are equivalent. a) QaQb = QbQa. b) Qab2 = Qba2. c) b and b2 operator commute with a and a2. Proof. a) to b) and c) to a) are trivial, so only b) to c) remains. Hence, suppose that Qab2 = Qba2. Since A is a purely exceptional, A ∼= C(X, E) where X is a hyperstonean compact Hausdorff space, and E = M3(O)sa is the exceptional Albert algebra and in particular, E is an EJA. Let f, g : X → E denote the functions corresponding to a and b. Then for every x ∈ X, Qf (x)g(x)2 = (Qf g2)(x) = (Qgf 2)(x) = Qg(x)f (x)2. As these are elements of an EJA, we use the previous proposition to see that Tg(x) and Tg(x)2 operator commute with Tf (x) and Tf (x)2 for all x ∈ X. We wish to conclude from this that Tg and Tg2 operator commute with Tf and Tf 2. So let h : X → E be any other function, then we should have for every x ∈ X, (Tf Tgh)(x) = (TgTf h)(x) (and similarly for f 2 and g2), but as (Tf Tgh)(x) = Tf (x)Tg(x)h(x) this of course directly follows. Proposition 3.11. Let A be a JBW-algebra with a, b ∈ A where at least one of a and b is positive. Then the following are equivalent. a) QaQb = QbQa. b) Qab2 = Qba2. c) b and b2 operator commute with a and a2. 12 Proof. Write A = A1 ⊕ A2 where A1 is a JW-algebra, and A2 is a purely exceptional JBW-algebra. If Qab2 = Qba2, then this equation also holds with a and b restricted to A1 or A2. By Proposition 3.3 and Lemma 3.10 the desired result then follows. Theorem 3.12. Let A be a JBW-algebra and a, b ∈ A arbitrary. Then the following are equivalent. a) a and b operator commute. b) a and b generate an associative JBW-algebra. c) a and b generate an associative JBW-algebra of mutually operator commuting elements. d) a and a2 operator commute with b and b2. If one of a and b is positive then these statements are furthermore equivalent to Qab2 = Qba2. Proof. a) to b) follows by Proposition 3.7. c) to d) follows because a2 and b2 are part of the associative algebra of mutually operator commuting elements. d) to a) is of course trivial. It remains to show b) to c). So suppose a and b generate an associative JBW-algebra B. Let c, d ∈ B be positive. By associativity Qcd2 = c2 ∗ d2 = Qdc2, and hence by Proposition 3.11 c and d operator commute. But the positive elements of course span B, so we are done. Now suppose one of a and b is positive. If they span an associative algebra, then of course Qab2 = a2 ∗ b2 = Qba2. For the converse direction we again refer to Proposi- tion 3.11. Theorem 3.13. Let A be a JB-algebra and a, b ∈ A arbitrary. Then the following are equivalent. a) a and b operator commute. b) a and b generate an associative JB-algebra. c) a and b generate an associative JB-algebra of mutually operator commuting ele- ments. d) a and a2 operator commute with b and b2. If one of a and b is positive then these statements are furthermore equivalent to Qab2 = Qba2. Proof. Again, a) to b), c) to d) and d) to a) are trivial. For proving b) to c) suppose a and b generate an associative JB-algebra B. Note that the JB-algebra A embeds into the JBW-algebra A∗∗, and hence B extends to an associative JBW-algebra in A∗∗. The result then follows in the same way as in the previous theorem. To show that Qab2 = Qba2 for one of a and b positive implies that a and b operator commute we again note that the equality continues to hold in A∗∗, and hence by the previous theorem, a and b operator commute in A∗∗, and thus in A. 13 4 The sequential product Let us use our results regarding operator commutation to prove that the unit interval in a JB-algebra is a sequential effect algebra as defined by Gudder and Greechie. Let A be a JB-algebra. Write [0, 1]A for its set of effects, i.e. the elements a ∈ A with 0 ≤ a ≤ 1. The set of effects in a JB-algebra forms a (convex) effect algebra. For an effect a ∈ [0, 1]A we write a⊥ := 1 − a. Lemma 4.1 ([1, Lemma 1.26]). Let a and b be positive elements in a JB-algebra. Then Qab = 0 iff Qba = 0, and in that case a ∗ b = 0. Theorem 4.2. Let A be a JB-algebra. Define the operation & : [0, 1]A×[0, 1]A → [0, 1]A by a & b := Q√ab. Then & satisfies all the axioms of Ref. [5]: a) a & (b + c) = a & b + a & c. b) 1 & a = a. c) If a & b = 0, then also b & a = 0. d) If a & b = b & a, then a & b⊥ = b⊥ & a, and a & (b & c) = (a & b) & c. e) If a & b = b & a and a & c = c & a, then a & (b + c) = (b + c) & a and a & (b & c) = (b & c) & a. Proof. Points a) and b) are trivial. For c), if a & b = 0 = Q√ab, then Qb√a = 0 by Lemma 4.1. Hence also Qba2 ≤ Qba ≤ Qb√a = 0. Applying Lemma 4.1 again gives Qab = 0, so that also Qab2 = 0. But then Qab2 = 0 = Qba2, and hence by Theorem 3.13, a and b span a JBW-algebra of mutually commuting elements. This algebra necessarily contains √b and hence b & a = Q√b√a2 = a ∗ b = 0 by Corollary 3.8 and Lemma 4.1. Note that if a & b = b & a, then by definition Q√a√b = Q√b√a2, so that by Theo- rem 3.13, √a and √b span an associative algebra of mutually commuting elements, and hence Q√aQ√b = Q√bQ√a, and a & b = Q√a√b = a ∗ b. Furthermore, as this algebra also contains a1/4 = p√a, Qa1/4 commutes with Q√b, and hence a1/4 and √b also generate an associative JBW-algebra. For point d) suppose that a & b = b & a. Then √a and √b span an associative JBW- algebra containing 1, and this algebra hence also contains b⊥ = 1 − b and √b⊥. As a result a & b⊥ = Q√ab⊥ = a ∗ b⊥ = Q√b⊥ a = b⊥ & a. Furthermore, as the JBW-algebra spanned by a1/4 and √b is associative we easily verify that Qa1/4√b = qQ√ab and then = √a2 ∗ √b 2 2 2 calculate a & (b & c) = Q√aQ√bc = Qa1/4 Q√bQa1/4 c = QQ a1/4√bc = Q√Q√abc = (a & b) & c. Finally, for point e) suppose that a & b = b & a and a & c = c & a. Then a operator commutes with b and c and hence with b+c. By Theorem 3.13, a and b+c then generate an associative algebra, and hence (b + c) & a = Q√b+ca = (b + c) ∗ a = Q√a(b + c) = a & (b + c) as desired. And at last, as [Q√a, Q√b] = 0 and [Q√a, Q√c] = 0, we calculate QQ√bcQ√a = Q√bQ2√cQ√bQ√a = Q√aQ√bQ2√cQ√b = QaQQ√bc. Then a and Q√bc generate an associative algebra so that we can finally verify that indeed a & (b & c) = (b & c) & a. 14 In Ref. [16] a further assumption of norm-continuity on the sequential product was assumed. This is also satisfied by the sequential product on JB-algebras: Proposition 4.3. Let A be a JB-algebra with & as defined above. Then the map a 7→ a & b is continuous in the norm. Proof. The map a 7→ a & b is given by the composition of a 7→ √a and c 7→ Qcb. The first is an application of the functional calculus and hence is continuous, while the second follows from the continuity of the Jordan product. In Ref. [5] a notion of σ-sequential effect algebra was also introduced. In such an effect algebra, suprema of increasing sequences exist and are compatible in a suitable way with the sequential product. We will show now that the unit interval of a JBW- algebra satisfies an even stronger condition related to suprema of arbitrary directed sets. Proposition 4.4. Let A be a JBW-algebra with & defined as above. Let a ∈ [0, 1]A and S ⊆ [0, 1]A be a directed subset. Then: • a & W S = W a & S (i.e. & is normal in the second argument). • If for all s ∈ S, a & s = s & a, then a & W S = (W S) & a. Proof. The first point follows immediately by normality of the quadratic product. For the second point, if a & s = s & a for all s ∈ S, then a operator commutes with all these s. By the weak continuity of the Jordan product, a then also operator commutes with W S, and the desired result follows. Acknowledgements: The author would like to thank Bas and Bram Westerbaan for many discussions regarding Jordan algebras. The author is supported in part by AFOSR grant FA2386-18-1-4028.. References [1] Erik M Alfsen & Frederic W Shultz (2012): Geometry of state spaces of operator algebras. Springer Science & Business Media, doi:10.1007/978-1-4612-0019-2. [2] Cho-Ho Chu (2011): Jordan structures in geometry and analysis. 190, Cambridge University Press. [3] Cho-Ho Chu (2017): Infinite dimensional Jordan algebras and symmetric cones. Journal of Algebra 491, pp. 357 -- 371, doi:10.1016/j.jalgebra.2017.08.017. [4] Jacques Faraut & Adam Kor´anyi (1994): Analysis on symmetric cones. Clarendon Press Oxford. [5] Stan Gudder & Richard Greechie (2002): Sequential products on ef- pp. 87 -- 111, 49(1), fect algebras. doi:10.1016/S0034-4877(02)80007-6. Reports on Mathematical Physics [6] Harald Hanche-Olsen (1983): On the structure and tensor products of JC-algebras. Can. J. Math 35(6), pp. 1059 -- 1074, doi:10.4153/CJM-1983-059-8. 15 [7] Harald Hanche-Olsen & Erling Størmer (1984): Jordan operator algebras. 21, Pit- man Advanced Pub. Program. [8] Nathan Jacobson (1989): Operator commutativity in Jordan algebras. In: Nathan Jacobson Collected Mathematical Papers, Springer, pp. 169 -- 172. [9] Pascual Jordan (1933): Uber Verallgemeinerungsmoglichkeiten des Formalismus der Quantenmechanik. Weidmann. [10] Richard V Kadison (1956): Operator algebras with a faithful weakly-closed repre- sentation. Annals of mathematics, pp. 175 -- 181, doi:10.2307/1969954. [11] Kevin McCrimmon (2006): A taste of Jordan algebras. Springer Science & Business Media. [12] Frederic W Shultz (1979): On normed Jordan algebras which are Banach dual spaces. Journal of Functional Analysis 31(3), pp. 360 -- 376. [13] Abraham A. Westerbaan (2019): Ph.D. gebras. https://bram.westerbaan.name/thesis.pdf. thesis, Radboud Universiteit Nijmegen. The Category of Von Neumann Al- Available at [14] John van de Wetering (2018): An algebraic semi-automated proof of the fundamen- tal identity of Jordan algebras. arXiv preprint arXiv:1807.00164. [15] John van de Wetering (2018): Three characterisations of the sequential product. J. Math. Phys. Vol. 59-8, doi:10.1063/1.5031089. [16] John van de Wetering (2019): Sequential product spaces are Jordan algebras. Jour- nal of Mathematical Physics 60(6), p. 062201, doi:10.1063/1.5093504. [17] JD Wright & MA Youngson (1978): On isometries of Jordan algebras. Journal of the London Mathematical Society 2(2), pp. 339 -- 344. 16
1705.04495	1	1705	2017-05-12T10:11:59	Convex subshifts, separated Bratteli diagrams, and ideal structure of tame separated graph algebras	[ "math.OA", "math.RA" ]	We introduce a new class of partial actions of free groups on totally disconnected compact Hausdorff spaces, which we call convex subshifts. These serve as an abstract framework for the partial actions associated with finite separated graphs in much the same way as classical subshifts generalize the edge shift of a finite graph. We define the notion of a finite type convex subshift and show that any such subshift is Kakutani equivalent to the partial action associated with a finite bipartite separated graph. We then study the ideal structure of both the full and the reduced tame graph C*-algebras, $\mathcal{O}(E,C)$ and $\mathcal{O}^r(E,C)$, of a separated graph $(E,C)$, and of the abelianized Leavitt path algebra $L_K^{\text{ab}}(E,C)$ as well. These algebras are the (reduced) crossed products with respect to the above-mentioned partial actions, and we prove that there is a lattice isomorphism between the lattice of induced ideals and the lattice of hereditary $D^{\infty}$-saturated subsets of a certain infinite separated graph $(F_{\infty},D^{\infty})$ built from $(E,C)$, called the separated Bratteli diagram of $(E,C)$. We finally use these tools to study simplicity and primeness of the tame separated graph algebras.	math.OA	math	CONVEX SUBSHIFTS, SEPARATED BRATTELI DIAGRAMS, AND IDEAL STRUCTURE OF TAME SEPARATED GRAPH ALGEBRAS PERE ARA AND MATIAS LOLK Abstract. We introduce a new class of partial actions of free groups on totally disconnected compact Hausdorff spaces, which we call convex subshifts. These serve as an abstract frame- work for the partial actions associated with finite separated graphs in much the same way as classical subshifts generalize the edge shift of a finite graph. We define the notion of a finite type convex subshift and show that any such subshift is Kakutani equivalent to the partial action associated with a finite bipartite separated graph. We then study the ideal structure of both the full and the reduced tame graph C-algebras, O(E, C) and Or(E, C), of a separated graph (E, C), and of the abelianized Leavitt path algebra Lab K (E, C) as well. These algebras are the (reduced) crossed products with respect to the above-mentioned partial actions, and we prove that there is a lattice isomorphism between the lattice of induced ideals and the lattice of hereditary D∞-saturated subsets of a certain infinite separated graph (F∞, D∞) built from (E, C), called the separated Bratteli diagram of (E, C). We finally use these tools to study simplicity and primeness of the tame separated graph algebras. Contents Introduction 1. 2. Preliminary definitions 3. Convex subshifts 4. The lattice of induced ideals 5. The ideals associated to hereditary C-saturated subsets of (E, C) 6. The ideals associated to hereditary D∞-saturated subsets of (F∞, D∞) 7. A complete description of the ideals of finite type 8. V-simplicity 9. Primeness References 1 5 10 24 27 30 38 47 52 58 1. Introduction The study of C-algebras associated to partial actions of groups on topological spaces is an important current line of investigation, see for instance [25] and the references therein. Date: September 24, 2018. 2000 Mathematics Subject Classification. Primary 16D70, 46L35; Secondary 06F05, 37B10, 46L80. Key words and phrases. Separated graph, tame graph algebra, ideal structure, simplicity, primeness. The first named author was partially supported by DGI-MINECO (Spain) through the grant MTM2014- 53644-P. The second named author was supported by the Danish National Research Foundation through the Centre for Symmetry and Deformation (DNRF92). 1 2 PERE ARA AND MATIAS LOLK This of course includes the more traditional setting of globally defined group actions, and is in turn generalized by the often useful setting of C-algebras associated to ´etale topological groupoids [45, 41, 27, 2, 35, 48]. In [26, Section 5], Exel describes any Cuntz-Krieger C-algebra OA as a crossed product of a commutative C-algebra by a partial action of a non-abelian free group. This may be interpreted as follows. Given a {0, 1}-matrix A ∈ Mp({0, 1}), one may consider the one- sided shift of finite type XA consisting of the infinite sequences (xn) ∈ {1, . . . , p}N such that A(xn, xn+1) = 1 for all n. Then, a partial action of the free group Fp on the space XA can ∼= C(XA) ⋊ Fp. (This action is spelled out at the beginning be naturally defined so that OA of Section 3.) This description can be generalized to the more general context of graph C-algebras, see e.g. [25, Theorem 37.8], [31]. One may also consider two-sided subshifts X ⊆ {1, . . . , p}Z (see e.g. [37]), where the shift map is a true homeomorphism, and thus the associated C-algebra is simply the crossed product C(X ) ⋊ Z, where the action is induced by the shift. The C-algebra associated to a given two-sided shift is quite different from the C-algebra associated to the corresponding one-sided shift. For instance, in the case of the full shift on {1, . . . , p}, we get the Cuntz algebra Op from the one-sided shift, and we get the group C-algebra C ∗(Zp ≀ Z) of the lamplighter group Zp ≀ Z from the two-sided shift. In this paper, we propose a unified approach to one-sided and two-sided subshifts, under the general notion of a convex subshift. Given a finite alphabet A, define C(A) to be the space of all the (right-)convex subsets of the free group F(A) on A which contain 1. Then the full convex shift on A is defined using the natural action of F(A) on C(A) (see Definition 3.1 for the precise definition). A convex subshift is just the restriction of the full convex shift to a closed invariant subspace, and we say that a convex subshift is of finite type if it can be obtained by forbidding finitely many balls (or patterns). We show that this notion is equivalent to the study of the dynamical systems associated to separated graphs, a concept recently coined by the first-named author and Ruy Exel [7]. Recall that a separated graph is a pair (E, C) consisting of a directed graph E and a set C = Fv∈E0 Cv, where each Cv is a partition of the set of edges whose terminal vertex is v. By [7, Theorem 9.1], the study of the dynamical systems associated to separated graphs can be reduced to the case of bipartite separated graphs, which are those separated graphs (E, C) such that E0 = E0,0 ⊔ E0,1 is the disjoint union of two layers E0,0 and E0,1, and s(E1) = E0,1, r(E1) = E0,0, where s and r are the source and range maps respectively. Given a finite bipartite separated graph (E, C), there exists a partial action θ(E,C) of the free group F = F(E1) on a zero-dimensional metrizable compact space Ω(E, C) such that the tame C-algebra O(E, C) is isomorphic to the partial crossed product C(Ω(E, C)) ⋊ F ([7]) -- this in turn allows for the definition of a reduced tame graph C ∗-algebra Or(E, C) as the corresponding reduced crossed product. A similar result holds for the abelianized Leavitt path algebra Lab K (E, C) over a field with involution K, allowing us to study both types of algebras at the same time. IDEAL STRUCTURE 3 One of our main results is Theorem 3.25, where we prove that any convex subshift of finite type is Kakutani equivalent to the dynamical system coming from a finite bipartite separated graph. This provides a far-reaching generalization of the well-known result that every shift of finite type is conjugate to an edge shift of a finite graph [37, Chapter 2]. The K-theory of the C-algebras associated to separated graphs has been computed in [8]. For a finite bipartite separated graph (E, C), the K0-group of both the reduced and the full tame graph C-algebras of (E, C) can be computed in terms of an infinite separated graph (F∞, D∞), which is the union of a sequence of finite bipartite separated graphs (En, C n). The infinite separated graph (F∞, D∞) has a structure which resembles very much the one of a usual Bratteli diagram. This leads us to define the new concept of a separated Bratteli diagram (Definition 2.8), and to refer to the graph (F∞, D∞) as the separated Bratteli diagram associated to (E, C). Given a crossed product O = A ⋊ G, we say that an ideal J ⊳ O is induced if J = (J ∩ A) ⋊ G, and we denote by Ind(O) the lattice of induced ideals. We show that the structure of induced ideals of the tame graph algebras of a finite separated graph (E, C) is completely determined by its associated separated Bratteli diagram (F∞, D∞). Concretely we obtain lattice isomorphisms Ind(Lab K (E, C)) ∼= Ind(O(E, C)) ∼= Ind(Or(E, C)) ∼= H(F∞, D∞), where H(F∞, D∞) is the lattice of hereditary D∞-saturated subsets of F 0 This generalizes the well-known result for ordinary graph C-algebras [16, 17]. ∞ (see Theorem 4.7). In sharp contrast with the situation for ordinary graphs, the quotient algebra O(E, C)/I(H) of an ideal generated by a hereditary D∞-saturated subset H can be described in terms of the tame algebra of a separated graph only when H is of finite type, meaning that H can be generated by a hereditary C n-saturated subset of vertices of the separated graph (En, C n) for some n ≥ 0. Otherwise, the quotient O(E, C)/I(H) can be described by means of a crossed product of a free group on a compact invariant subset of Ω(E, C), but it will in general not be a tame graph C-algebra of a separated graph. Even so, its K-theory can be computed by using a corresponding separated Bratteli diagram (F∞/H, D∞/H) (Theorem 6.4). This justifies the introduction of general separated Bratteli diagrams, and indicates that interesting examples (some of them of a pathological behaviour) can occur if we allow the consideration of hereditary D∞-saturated of infinite type; see the comment after Example 6.9. We present in some detail an example which is connected with the full two-sided shift. We believe this example is very useful to understand the various aspects of the theory that have been described above. It is also the example that allows to study the group algebra of the lamplighter group from the perspective of the theory of separated graphs. Using this, and generalizing [12], the first-named author and Joan Claramunt have obtained a concrete approximation of this group algebra by a sequence of finite-dimensional algebras [6]. We also study general (i.e. not necessarily induced) ideals of Or(E, C). We say that an ideal J is of finite type if the corresponding induced ideal I(H) = (J ∩C(Ω(E, C))) ⋊r F arises from a hereditary D∞-saturated set H of finite type. By studying a weakening of topological freeness that we dub relative strong topological freeness, we are able to compute the lattice 4 PERE ARA AND MATIAS LOLK of finite type ideals. We also observe that our techniques, when applied to non-separated graphs, yield a complete characterization of the ideal structure. We close the paper by using our tools to perform a study of simplicity and primeness in tame graph algebras of separated graphs. In this paper, we consider three types of tame graph ∗-algebras associated to separated graphs, namely the full tame graph C-algebra O(E, C), the reduced tame graph C-algebra Or(E, C), and the abelianized Leavitt path algebra Lab K (E, C) over a field with involution K. All three types of algebras have been introduced in [7]. The first two types generalize graph C-algebras [44], in the sense that, denoting by T the trivial partition on each r−1(v), we have O(E, T ) = Or(E, T ) = C ∗(E), where C ∗(E) is the graph C-algebra of E. The third type generalizes Leavitt path algebras [4, 14], in the sense that Lab K (E, T ) = LK(E), where LK(E) is the Leavitt path algebra of E. Contents. We now describe the contents of the paper in more detail. In section 2, we recall relevant definitions and constructions from the existing theory of algebras associated with separated graphs, in particular the main results of [7]. We also introduce the notion of a separated Bratteli diagram. In section 3, we introduce a class of partial actions that we call convex subshifts. These serve as an abstract framework for the partial actions associated with finite separated graphs in much the same way as classical subshifts generalize the edge shift of a finite graph. We define the notion of a finite type convex subshift and show that any such subshift is Kakutani equivalent to the partial action associated with a finite bipartite separated graph (Theorem 3.25). Using the language of convex subshifts, we are also able to explain the precise relationship between the partial action of a finite bipartite separated graph (E, C) and its successors (En, C n) for n ≥ 1 (Theorem 3.22). In section 4, we begin our investigation of the lattice of induced ideals of the algebras Lab(E, C), O(E, C) and Or(E, C). The main result (Theorem 4.7) identifies this lattice with the lattice L(M(F∞, D∞)) of order ideals of the monoid M(F∞, D∞), as well as the lattice of hereditary and D∞-saturated subsets of (F∞, D∞), providing both an algebraic and a graph theoretic perspective. In the following section, we study the specific ideals associated with hereditary and C-saturated subsets of (E, C), and we show that the corresponding quotients arise as separated graph algebras from the quotient graph (Theorem 5.5). In section 6, we combine the work of the previous sections to describe the quotient by an arbitrary induced ideal as a limit of separated graph algebras (Proposition 6.2), and we provide a dynamical description of this approximation. We also pay significant attention to a concrete example that illustrates the general theory (Examples 6.7, 6.9, and 6.10). In particular, we show that the crossed product of a two-sided finite type subshift can be realised as a full corner of the tame graph C- algebra associated with a finite bipartite separated graph (Proposition 6.8). We proceed to study general ideals of finite type in section 7, and we provide a complete characterisation in terms of graph theoretic data (Theorem 7.12). This relies on an abstract study of relatively strongly topologically free partial actions that we initiate. We then study V-simplicity of the separated graph algebras in section 8, showing that (up to Morita equivalence) the tame graph C-algebras degenerate to either classical graph C-algebras or group C-algebras of IDEAL STRUCTURE 5 free groups if M(F∞, D∞) ∼= V(Lab(E, C)) is order simple (Theorem 8.1). In section 9, we finally establish a criterion for Ω(E, C) to be a Cantor space (Proposition 9.6), and in this setting we characterize primeness of the algebras Lab K (E, C) and Or(E, C) (Theorem 9.10). 2. Preliminary definitions In this preliminary section, we will recall all the relevant definitions and constructions from the theory of algebras associated with separated graphs. The reader should note that we use the same conventions as in [7] and [8], but opposite to those of [11] and [10]. One important consequence is that all paths should be read from the right. Definition 2.1. ([11]) A separated graph is a pair (E, C) where E = (E0, E1, r, s) is a directed graph, C =Fv∈E0 Cv, and Cv is a partition of r−1(v) (into non-empty subsets) for every vertex v. In case v is a source, i.e. r−1(v) = ∅, we take Cv to be the empty family of subsets of r−1(v). Given an edge e, we shall use the notation Xe for the element of C containing e. If all the sets in C are finite, we say that (E, C) is a finitely separated graph. This necessarily holds if E is column-finite (that is, if r−1(v) is a finite set for every v ∈ E0.) The set C is a trivial separation of E in case Cv = {r−1(v)} for each v ∈ E0 \ Source(E). In that case, (E, C) is called a trivially separated graph or a non-separated graph. Finally, (E, C) is called bipartite if the vertex set admits a partition E0 = E0,0 ⊔ E0,1 with s(E1) = E0,1 and r(E1) = E0,0. As the first of many different algebras assocated with separated graphs, we now define the Leavitt path algebra. Definition 2.2. Let (K, ∗) be a field with involution. The Leavitt path algebra of the sepa- rated graph (E, C) with coefficients in the field K is the ∗-algebra LK(E, C) with generators {v, e v ∈ E0, e ∈ E1}, subject to the following relations: (V) vv′ = δv,v′v and v = v∗ for all v, v′ ∈ E0 , (E) r(e)e = es(e) = e for all e ∈ E1 , (SCK1) e∗e′ = δe,e′s(e) for all e, e′ ∈ X, X ∈ C, and (SCK2) v =Pe∈X ee∗ for every finite set X ∈ Cv, v ∈ E0. The Leavitt path algebra LK(E) is just LK(E, C) where Cv = {r−1(v)} if r−1(v) 6= ∅ and Cv = ∅ if r−1(v) = ∅. An arbitrary field can be considered as a field with involution by taking the identity as the involution. However, our "default" involution over the complex numbers C will be the complex conjugation, and we will write L(E, C) := LC(E, C). We now recall the definition of the graph C-algebra C ∗(E, C), introduced in [10]. Definition 2.3. The graph C-algebra of a separated graph (E, C) is the C-algebra C ∗(E, C) with generators {v, e v ∈ E0, e ∈ E1}, subject to the relations (V), (E), (SCK1), (SCK2). In other words, C ∗(E, C) is the enveloping C-algebra of L(E, C). In case (E, C) is trivially separated, C ∗(E, C) is just the classical graph C-algebra C ∗(E). There is a unique -homomorphism L(E, C) → C ∗(E, C) sending the generators of L(E, C) to their canonical images in C ∗(E, C). This map is injective by [10, Theorem 3.8(1)]. 6 PERE ARA AND MATIAS LOLK Since both LK(E, C) and C ∗(E, C) are universal objects with respect to the same sets of generators and relations, they can be studied in much the same way. A remarkable difference is that the non-stable K-theory V(LK(E, C)) has been computed for any separated graph (E, C), while the structure of V(C ∗(E, C)) is still unknown. However, it is conjectured in [10] that the natural map L(E, C) → C ∗(E, C) induces an isomorphism V(L(E, C)) → V(C ∗(E, C)). See [5, Section 6] for a short discussion on this problem. We now describe V(LK(E, C)) as the graph monoid of (E, C). Definition 2.4. Given a finitely separated graph (E, C), the graph monoid M(E, C) is the abelian monoid with generators av for v ∈ E0 and relations av = Pe∈X as(e) for all v ∈ E0 and X ∈ Cv. Note that there is a natural map M(E) → V(LK(E, C)) given by av 7→ [v], and this is in fact an isomorphism by [11, Theorem 4.3]. One issue with the algebras LK(E, C) and C ∗(E, C) is that the set of partial isometries represented by the edges E1 is not tame, that is, products of edges and their adjoints are not in general partial isometries. This motivated Ruy Exel and the first named author to introduce certain quotients Lab K (E, C) and O(E, C) of LK(E, C) and C ∗(E, C), respectively, that we shall now describe. Definition 2.5. For a finitely separated graph (E, C), let S denote the multiplicative sub- semigroup of LK(E, C) generated by E1 ∪ (E1)∗, and define an ideal (respectively a closed ideal) J = hαα∗α − α α ∈ Si = h[αα∗, ββ∗] : α, β ∈ Si of LK(E, C) (respectively of C ∗(E, C)). Then we set Lab K (E, C) := LK(E, C)/J and O(E, C) := C ∗(E, C)/J. Observe that modding out J precisely forces E1 to be tame in these quotients. We now recall the main construction of [7] which will play an important role in what is to come. Definition 2.6. Let (E, C) denote a finite bipartite separated graph, and write for all u ∈ E0,0. Then (E1, C 1) is the finite bipartite separated graph defined by Cu = {X u 1 , . . . , X u ku} • E0,0 1 := E0,1 and E0,1 1 := {v(x1, . . . , xku) u ∈ E0,0, xj ∈ X u j }, • E1 := {αxi(x1, . . . , bxi, . . . , xku) u ∈ E0,0, i = 1, . . . , ku, xj ∈ X u • r1(αxi(x1, . . . ,bxi, . . . , xku)) := s(xi) and s1(αxi(x1, . . . , bxi, . . . , xku)) := v(x1, . . . , xku), v := {X(x) x ∈ s−1(v)}, where • C 1 j }, X(xi) := {αxi(x1, . . . , bxi, . . . , xku) xj ∈ X u j for j 6= i}. IDEAL STRUCTURE 7 A sequence of finite bipartite separated graphs {(En, C n)}n≥0 with (E0, C 0) := (E, C) is then defined inductively by letting (En+1, C n+1) denote the 1-graph of (En, C n). Finally, (Fn, Dn) i=0(En, C n), and (F∞, D∞) is the infinite layer graph denotes the union Sn (F∞, D∞) := ∞[n=0 (Fn, Dn) = ∞[n=0 (En, C n). Observe that (Fn, Dn) is a finite separated graph, while (F∞, D∞) is a finitely separated graph. By [7, Theorem 5.1 and Theorem 5.7], there are canonical surjective ∗-homomorphisms ∗-homomorphisms LK(En, C n) → LK(En+1, C n+1) and C ∗(En, C n) → C ∗(En+1, C n+1) such that K (E, C) ∼= lim−→ Lab n LK(En, C n) and O(E, C) ∼= lim−→ n C ∗(En, C n). On the level of monoids, the induced monoid homomorphism M(En, C n) ∼= V(LK(En, C n)) → V(LK(En+1, C n+1) ∼= M(En+1, C n+1) is a unitary embedding, which refines the defining relations of M(En, C n). Consequently, the quotient map LK(E, C) → Lab K (E, C) induces a (universal) refinement M(E, C) ∼= V(LK(E, C)) → V(Lab K (E, C)) ∼= M(F∞, D∞). One of the main advantages of dealing with the quotients Lab K (E, C) and O(E, C) is that, by [7, Corollary 6.12], they admit descriptions as crossed products K (E, C) ∼= CK(E, C) ⋊ F and O(E, C) ∼= C(Ω(E, C)) ⋊ F Lab of a topological partial action θ(E,C) : F y Ω(E, C), referred to as the canonical partial (E, C)-action. Here, F is the free group generated by E1, Ω(E, C) is a certain compact, zero-dimensional, metrisable space, and CK(Ω(E, C)) denotes the ∗-algebra of continuous functions Ω(E, C) → K when K is given the discrete topology, that is, the ∗-algebra of locally constant functions Ω(E, C) → K. Every vertex v ∈ F 0 ∞ corresponds to a compact open subset Ω(E, C)v ⊂ Ω(E, C) such that Ω(E, C) =Fv∈E0 The crossed product description enables the definition of yet another tame graph C ∗- Ω(E, C)v for all n. n algebra. Definition 2.7. Let (E, C) denote a finite bipartite separated graph. Then Or(E, C) is the reduced crossed product C(Ω(E, C)) ⋊r F of the canonical partial (E, C)-action. The separated graph (F∞, D∞) resembles very much the structure of a Bratteli diagram [18], but incorporating separations (and up to a change of the convention of the direction of the arrows). We formalize this with the following definition. 8 PERE ARA AND MATIAS LOLK Definition 2.8. A separated (or colored) Bratteli diagram is an infinite separated graph (F, D). The vertex set F 0 is the union of finite, non-empty, pairwise disjoint sets F 0,j, j ≥ 0. Similarly, the edge set F 1 is the union of a sequence of finite, non-empty, pairwise disjoint sets F 1,j, j ≥ 0. The range and source maps satisfy r(F 1,j) = F 0,j and s(F 1,j) = F 0,j+1 for all j ≥ 0. A Bratteli diagram is just a separated Bratteli diagram with the trivial separation. The graph (F∞, D∞) associated to a finite bipartite separated graph (E, C) is an example of a separated Bratteli diagram, and we will refer to it as the separated Bratteli diagram of the separated graph (E, C). We will find later other examples of separated Bratteli diagrams, related to quotients of tame graph C-algebras, see Theorem 6.4. The graph monoid of a Bratteli diagram (F, D) is the graph monoid M(F, D) of the finitely separated graph (F, D), see Definition 2.4. The Grothendieck group G(F, D) of the monoid M(F, D) is an analogue of the dimension group associated to a Bratteli diagram, but it is not a dimension group in general. It is quite sensible to equip G(F, D) with the structure of a pre-ordered abelian group, taking G(F, D)+ := ϕ(M(F, D)), where ϕ : M(F, D) → G(F, D) is the canonical map. The separated Bratteli diagram of a finite bipartite separated graph computes the K0 of the tame graph algebras of the graph, as follows. Theorem 2.9. Let (E, C) be a finite bipartite separated graph, and let (F∞, D∞) be its separated Bratteli diagram. (1) There is a natural group isomorphism K0(O(E, C)) ∼= K0(Or(E, C)) ∼= G(F∞, D∞). (2) There is a natural isomorphism V(Lab K (E, C)) ∼= M(F∞, D∞), and so an isomorphism of pre-ordered abelian groups K0(Lab for any field with involution K. K (E, C)) ∼= G(F∞, D∞) Proof. (2) follows from [7, Theorem 5.1], and (1) follows from [8, Theorem 4.4(c), Corollary 6.9]. (cid:3) In the following, we shall recall a very handy description of the canonical partial (E, C)- action introduced in [7], although we will make a slight change to the original definition, coherent with the conventions of [38] and [39]. We remark that while we only consider finite bipartite graphs, the dynamical picture can be extended to arbitrary finitely separated graphs [38, Definition 2.6 and Theorem 2.10]. For a comprehensive study of the general theory of partial actions and their crossed products, we refer the reader to [25]. Definition 2.10. Suppose that (E, C) is a separated graph, and let bE denote the double of E, i.e. the graph with bE0 := E0, bE1 := E1 ⊔ (E1)−1, r(e) := r(e) = s(e−1) and r(e−1) := s(e) =: s(e). A path in the double of E (with the convention that a path is read from the right to the left) is called an admissible path if IDEAL STRUCTURE 9 • e 6= f for every subpath ef −1, • Xe 6= Xf for every subpath e−1f . We define the range and source of an admissible path to simply be the range and source in the double, and we view the vertices as the set of trivial admissible paths. A closed path in (E, C) is a non-trivial admissible path α with r(α) = s(α), and α is called a cycle if the concatenated word αα is an admissible path as well. Either way, we shall say that α is based at r(α) = s(α). A cycle is called simple if the only vertex repetition occurs at the end. Definition 2.11. Suppose that (E, C) is finite bipartite separated graph, and let F denote the free group on E1. Given ξ ⊂ F and α ∈ ξ, the local configuration ξα of ξ at α is defined as Then Ω(E, C) is the set of ξ ⊂ F satisfying the following: ξα := {σ ∈ E1 ⊔ (E1)−1 σ ∈ ξ · α−1}. (a) 1 ∈ ξ. (b) ξ is right-convex : In view of (a), this exactly means that if eεn 1 ∈ ξ as well for any 1 ≤ m < n. and εi ∈ {±1}, then eεm m · · · eε1 (c) For every α ∈ ξ, one of the following holds: n · · · eε1 1 ∈ ξ for ei ∈ E1 (c1) ξα = s−1(v) for some v ∈ E0,1. (c2) ξα = {e−1 X X ∈ Cv} for some v ∈ E0,0 and eX ∈ X. Observe that for ξ ∈ Ω(E, C), every α ∈ ξ is an admissible path. Ω(E, C) is made into a topological space by regarding it as a subspace of {0, 1}F. Thus it becomes a compact, zero-dimensional, metrisable space, and a topological partial action θ = θ(E,C) : F y Ω(E, C) with compact open domains is then defined by setting • Ω(E, C)α := {ξ ∈ Ω(E, C) α−1 ∈ ξ}, • θα(ξ) := ξ · α−1 for ξ ∈ Ω(E, C)α−1. This partial action is equivalent to the one defined in [7] under the map ξ 7→ ξ−1. We choose to invert the elements as we want a common terminology and notation for both the algebraic and the topological setting. The compact, open sets corresponding to the vertices of E are given by Ω(E, C)v =(cid:26) Ω(E, C)e−1 for some e ∈ s−1(v) Fe∈X Ω(E, C)e for some X ∈ Cv if v ∈ E0,1 if v ∈ E0,0 , and we note that this is independent of e and X by (c1) and (c2), respectively. The reader may think of Ω(E, C)v as the set of configurations "starting" in v, and we shall regard 1 ∈ ξ as the trivial path v when ξ ∈ Ω(E, C)v. See Remark 3.23 for a description of Ω(E, C)v when v ∈ E0 n for n ≥ 1. 10 PERE ARA AND MATIAS LOLK 3. Convex subshifts Given a finite alphabet A and any set F of finite words in A, one can consider the space XF ⊂ AN of infinite one-sided sequences that do not contain any words from F . In classical symbolic dynamics, XF is then usually turned into a dynamical system by equipping it with the one-sided shift σ, but for the purpose of constructing interesting algebras, one would typically add additional dynamical structure. Specifically, a sequence may not only be shifted to the left, but new letters may also be introduced in the beginning, provided that we stay inside XF , of course. Formally, this dynamical system can be regarded as a partial action with the free group F(A) acting on XF : If a ∈ A and x ∈ XF , then a can act on x by a.x = ax whenever ax ∈ XF , while the inverse a−1 can act on points ax ∈ XF by a−1.ax = x = σ(ax). It is a standard fact that any subshift of finite type (meaning that F is finite) arises as the edge shift of some finite graph E with no sinks, and the graph C ∗-algebra may then be recovered as the partial crossed product C ∗(E) ∼= C(XF ) ⋊ F(A) (see for instance [25, Theorem 36.20] or [19, Theorem 3.1]). If E is a finite directed graph, one can also consider the boundary path space ∂E, where -- in addition to the set of right infinite paths -- one also includes the set of finite paths ending in a sink. Then there is natural partial action of F(E1) on ∂E defined as above, and we still have C ∗(E) ∼= C(∂E) ⋊ F(E1). In fact, a similar description exists for arbitrary graphs. In this section, we shall introduce a class of partial actions that one might consider as a generalisation of both the above partial actions on one-sided sequence spaces (including those of infinite type) and the boundary path spaces of finite graphs. We will see later (Example 6.7, Proposition 6.8) that also two-sided shifts can be recasted in this language. The basic idea is to give up the linear nature of a sequence and allow trees instead. The dynamics still arise from shifting, but there are usually many possible shifting directions with no one being canonical. We will refer to these dynamical systems as convex subshifts for reasons that should become apparent soon. The motivation for introducing such systems is two-fold: On one hand, it provides a frame- work for the dynamical systems associated with finite separated graphs in which one has far more flexibility. On the other hand, the dynamical systems of finite bipartite separated graphs encompass a vast class of convex subshifts, in fact all finite type convex subshifts up to Kakutani-equivalence (see Theorem 3.25). As such, convex subshifts are to finite bipartite separated graphs as one-sided subshifts are to finite graphs. Finally, we shall see how the construction of (En, C n) from (E, C) corresponds to a natural construction in the realm of convex subshifts. The contents of this section are closely related to classical subshifts of free groups on the alphabet {0, 1}. Indeed, any convex subshift is the restriction of a free group subshift to a partial action, but the convexity requirement that we impose is not of finite type; hence a convex subshift would typically be regarded as an infinite type subshift. Moreover, it is most beneficial for us to define everything from scratch so that the formal framework of convex subshifts is similar to that of partial actions associated with separated graphs. IDEAL STRUCTURE 11 Throughout this section, A will be a finite alphabet and F = F(A) will be the free group on A. Though some definitions still make sense for infinite alphabets, we choose to deal only with finite ones for the sake of simplicity. Definition 3.1. Denote by C = C(A) the set of right-convex subsets ξ ⊂ F for which 1 ∈ ξ. As in Definition 2.11, right-convexity in this setting simply means that if aεn 1 ∈ ξ for ai ∈ A and εi ∈ {±1}, then aεm 1 ∈ ξ as well for all 1 ≤ m < n. We topologize C as a subspace of P(F) ∼= {0, 1}F, making it into a compact, zero-dimensional, and metrizable space (with many isolated points). The full convex shift on A is then the partial action F y C given by m · · · aε1 n · · · aε1 • Cα := {ξ ∈ C α−1 ∈ ξ} for all α ∈ F. • α.ξ := ξ · α−1 for ξ ∈ Cα−1. Viewing each ξ ∈ C as a tree, rooted in 1 and labelled relative to the root, the action of any α ∈ ξ on ξ is thus simply given by moving the root to α and relabelling accordingly. Remark 3.2. We could also define C to be the set of left-convex subsets with Cα = {ξ ∈ C α ∈ ξ} and α.ξ = α · ξ, but we choose the above convention to match the one used for separated graphs. It is also very convenient that with our convention, α can act on ξ if and only if α ∈ ξ (as opposed to α−1 ∈ ξ). By [29, Section 4] and [28, Proposition 4.5], the full convex shift is conjugate to the universal action for semi-saturated partial representations of F. Definition 3.3. A convex subshift is the restriction of the full convex shift F y C to any closed invariant subspace Ω ⊂ C. Definition 3.4. An n-ball is an element B ∈ C such that α ≤ n for every α ∈ B, together with the radius r(B) = n; a ball is then simply an n-ball for some n. Note that B as a set does not determine the radius r(B), so one has to specify this. Now if ξ ∈ C, we can always consider the n-ball ξn := {α ∈ ξ : α ≤ n}, and if B is a given n-ball, we shall write ξ 6≡ B if (α.ξ)n 6= B for all α ∈ ξ. Finally, if Ω is any convex subshift, we will write Bn(Ω) := {ξn ξ ∈ Ω} and B(Ω) := Gn≥0 Bn(Ω) for the set of allowed n-balls and allowed balls, respectively. With all the relevant terminology in place, we can give an example of a convex subshift. Definition 3.5. Let F denote any set of balls; we can then define a convex subshift by ΩF := {ξ ∈ C ξ 6≡ B for all B ∈ F }, and we shall refer to this as the convex subshift obtained from forbidding F . In fact, this is not an example, but rather the example. 12 PERE ARA AND MATIAS LOLK Proposition 3.6. If F y Ω is a convex subshift, then Ω = ΩF for some set of balls F . Proof. Define F to be all the balls that do not occur in Ω, i.e. let F := B(C) \ B(Ω). Clearly Ω ⊂ ΩF , so let us consider the reverse inclusion. Given any ξ ∈ ΩF and n ≥ 1, it is enough to check that ξn = ηn for some η ∈ Ω since Ω is closed. But since ξn /∈ F , we must have ξn ∈ Bn(Ω), so ξn = ηn for some η ∈ Ω as desired. (cid:3) Definition 3.7. A convex subshift F y Ω is of finite type if Ω = ΩF for some finite set of forbidden balls F . Observe that for such a convex subshift, we can safely assume that all balls of F have the same radius R; we recognize this by saying that Ω is R-step. Observe that in this situation, Ω is generated by BR(Ω) in the following sense: An element ξ ∈ C belongs to Ω if and only if (α.ξ)R ∈ BR(Ω) for all α ∈ ξ. Before venturing on, we need to discuss how one might compare partial actions of different groups. Definition 3.8. Consider partial actions θ : G y Ω and θ′ : H y Ω′ of discrete groups on topological spaces. Then θ and θ′ are called dynamically equivalent and we shall write θ ≈ θ′, if their transformation groupoids Gθ and Gθ′ (see for instance [39, Example 2.3]) are isomorphic as topological groupoids. We now spell out exactly what this means: There is a homeomorphism ϕ : Ω → Ω′ and continuous maps a : [g∈G {g} × Ωg−1 → H and b : [h∈H {h} × Ω′ h−1 → G such that a(g,x)−1 and ϕ(g.x) = a(g, x).ϕ(x), (1) ϕ(x) ∈ Ω′ (2) ϕ−1(y) ∈ Ωb(h,y)−1 and ϕ−1(h.y) = b(h, y).ϕ−1(y), (3) b(a(g, x), ϕ(x)) = g and a(b(h, y), ϕ−1(y)) = h, (4) a(g′g, x) = a(g′, g.x)a(g, x) if g.x ∈ Ωg′−1, (5) b(h′h, y) = b(h′, h.y)b(h, y) if h.y ∈ Ωh′−1 for all g ∈ G, h ∈ H, x ∈ Ωg−1, y ∈ Ω′ h−1. Remark 3.9. Given a locally compact Hausdorff ´etale groupoid G, one can associate to it both a universal and a reduced groupoid C ∗-algebra (see [45]), denoted C ∗(G) and C ∗ r (G), respectively. If G = Gθ for a partial action θ : G y Ω of a discrete group on a locally compact Hausdorff space, then there are identifications C ∗(Gθ) ∼= C0(Ω)⋊θG and C ∗ r (Gθ) ∼= C0(Ω)⋊θ,rG by [2, Theorem 3.3] and [35, Proposition 2.2]. For an ample groupoid G and any field K with involution, there is also a purely algebraic analogue KG, known as the Steinberg algebra of G [47]. If a partial action θ as above acts on a totally disconnected space, then Gθ is ample and ∼= CK(Ω) ⋊θ G, where CK(Ω) denotes the algebra of compactly supported continuous KGθ functions Ω → K, when K is endowed with the discrete topology. Hence if two partial actions ∼= Gθ′, then they have isomorphic crossed θ and θ′ as above are dynamically equivalent, i.e. Gθ IDEAL STRUCTURE 13 products by base-preserving isomorphisms. In this light, groupoids provide a very flexible framework for identifying the crossed products of partial actions. Definition 3.10. Consider partial actions θ : G y Ω and θ′ : H y Ω′ of discrete groups on topological spaces along with a group homomorphism Ψ : G → H. A continuous map ϕ : Ω → Ω′ is then called Ψ-equivariant if (1) ϕ(Ωg) ⊂ Ω′ (2) ϕ(g.x) = Ψ(g).ϕ(x) for all x ∈ Ωg−1. Ψ(g) for all g ∈ G, The pair (ϕ, Ψ) is called a conjugacy if Ψ is an isomorphism and ϕ admits a Ψ−1-equivariant inverse. However, conjugacy is often too rigid a notion and we therefore introduce another type of equivalence in between conjugacy and dynamical equivalence: The pair (ϕ, Ψ) is called a direct dynamical equivalence if (a) ϕ is a homeomorphism, (b) Ωg ∩ Ωg′ = ∅ for all g 6= g′ with Ψ(g) = Ψ(g′), (c) Ω′ h =SΨ(g)=h ϕ(Ωg) for all h ∈ H, and in this case we will write θ ≈−→ θ′. Now let us see that our choice of name and notation is justified. Proposition 3.11. If θ ≈−→ θ′, then θ ≈ θ′. Proof. Apply the notation from above. We then set a(g, x) := Ψ(g) for all g ∈ G, x ∈ Ωg−1, and given h ∈ H, y ∈ Ω′ h−1, we define b(h, y) to be the unique element of G satisfying Then and Ψ(b(h, y)) = h and y ∈ ϕ(Ωb(h,y)−1). ϕ−1(h.y) = ϕ−1(Ψ(b(h, y)).y) = b(h, y).ϕ−1(y) so (1)-(4) of Definition 3.8 surely hold. Finally, if h′ ∈ H and h.y ∈ Ω′ h′−1, then b(a(g, x), ϕ(x)) = b(Ψ(g), ϕ(x)) = g, Ψ(b(h′, h.y)b(h, y)) = h′h and hence y ∈ h−1.(cid:0)ϕ(Ωb(h′,h.y)) ∩ Ω′ = ϕ(cid:0)b(h, y)−1.(ϕ(Ωb(h′,h.y)) ∩ Ω′ h(cid:1) = Ψ(b(h, y))−1.(cid:0)ϕ(Ωb(h′,h.y)) ∩ Ω′ Ψ(b(h,y)))(cid:1) = ϕ(cid:0)Ωb(h,y)−1b(h′,h.y)−1 ∩ Ωb(h,y)−1(cid:1), Ψ(b(h,y))(cid:1) as desired. (cid:3) b(h′h, y) = b(h′, h.y)b(h, y) Remark 3.12. If (ϕ, Φ) is a direct dynamical equivalence from θ : G y Ω to θ′ : H y Ω′, then the induced isomorphisms CK(Ω) ⋊ G → CK(Ω′) ⋊ H and C0(Ω) ⋊(r) G → C0(Ω′) ⋊(r) H are simply given by f δg 7→ f ◦ ϕ−1Ω′ δΦ(g) with inverses f δh 7→PΦ(g)=h f ◦ ϕΩgδg. Φ(g) 14 PERE ARA AND MATIAS LOLK In fact, any dynamical equivalence is a result of two direct dynamical equivalences: Proposition 3.13. If θ ≈ θ′ for partial actions of G and H, respectively, then for some partial action γ of G × H. θ ≈←− γ ≈−→ θ′ Proof. By otherwise replacing θ′ with the conjugate partial action ϕ ◦ θ′ ◦ ϕ−1, we may assume that θ and θ′ act on the same space Ω and ϕ = idΩ. Writing ag := a(g, −) and bh := b(h, −), we then first define domains by Ω(g,h) := a−1 g−1(h−1) = b−1 h−1(g−1) for (g, h) ∈ G × H. To see that the above equality holds, assume that x ∈ a−1 x ∈ Ωg with a(g−1, x) = h−1. Then x ∈ Ωh and g−1(h−1), i.e. b(h−1, x) = b(a(g−1, x), x) = g−1, hence x ∈ b−1 (g, h) by h−1(g−1), so a−1 g−1(h−1) = b−1 h−1(g−1) from symmetry. Then define the action of (g, h).x := g.x = a(g, x).x = h.x for all x ∈ Ω(g,h)−1. It is straightforward to verify that this does indeed define a partial action γ : G × H y Ω. Then simply observe that the pairs (idΩ, πG) and (idΩ, πH), where πG and πH denote the projections onto G and H, respectively, are direct dynamical equivalences. (cid:3) Returning to the world of convex shifts, we recall that any n-step subshift X can be recoded into being 1-step using higher block shifts; one simply replaces the original alphabet with the n-blocks Bn(X) via the map x 7→ [x1 . . . xn][x2 . . . xn+1][x3 . . . xn+2] . . . . In the following, we shall make a similar construction in the world of convex shifts -- although with blocks replaced by balls. Construction 3.14. Let θ : F y Ω denote any convex subshift, let n ≥ 1 and consider the finite alphabet A[n : Ω] :=n(cid:2)(a.ξ)n a←− ξn(cid:3) ξ ∈ Ω, a ∈ A such that a ∈ ξo, where each (cid:2)(a.ξ)n a←− ξn(cid:3) is just a formal symbol. As every symbol is typically represented by many different configurations ξ, we will simply use the notation [B a←− B′] where B, B′ ∈ Bn(Ω) in the future. We then consider the corresponding free group F[n : Ω] := F(A[n : Ω]) and introduce the notation [B′ a−1 ←−− B] := [B a←− B′]−1. IDEAL STRUCTURE 15 Observe that if Ω ⊂ Λ is an inclusion of convex subshifts over A, then we obtain corresponding inclusions A[n : Ω] ⊂ A[n : Λ] and F[n : Ω] ⊂ F[n : Λ]. When Ω = C is the full convex shift, we will simply write A[n] := A[n : C] and F[n] := F[n : C], and define a group homomorphism Ψn : F[n] → F given by Ψn([B a←− B′]) := a. By a slight abuse of notation, we will also refer to Ψn when we really mean the restriction of Ψn to the subgroup F[n : Ω]. Our aim is to define a replacement for the above block-encoding, more specifically a map φn : C(A) → C(A[n]). Given ξ ∈ Ω, we first set φn(ξ, 1) := 1 ∈ F[n] and proceed to define φn(ξ, α) ∈ F[n] for 1 6= α ∈ ξ. Writing α = sm · · · s1 and Bk = ((sk · · · s1).ξ)n for k ≤ m so that B0 = ξn and Bm = (α.ξ)n, we then set φn(ξ, α) := [Bm sm←− Bm−1] · [Bm−1 sm−1←−−− Bm−2] · · · [B1 s1←− B0], allowing us to define φn by φn(ξ) := {φn(ξ, α) α ∈ ξ}. It is clear from the construction that φn(ξ) is a right-convex subset of F[n] containing 1, so that φn(ξ) ∈ C(A[n]). Observe that if ξ ∈ Ω, then φn(ξ, α) ∈ F[n : Ω] for all α ∈ ξ, so that φn restricts to a map Ω → C(A[n : Ω]). The n-ball subshift θ[n] of θ is then simply the restricted action of F[n : Ω] on the image Ω[n] := φn(Ω). We observe that φn has a Ψn-equivariant inverse ψn : Ω[n] → Ω given by ψn(η) := {Ψn(β) β ∈ η}, and as both φn and ψn are obviously continuous, they are in fact homeomorphisms of Ω and Ω[n]. In particular, it follows that Ω[n] is compact and invariant under the action of F[n : Ω], hence a convex subshift. Using the same notation as above, we also define a map by Ψn : {β ∈ η η ∈ C[n]} = {φn(ξ, α) ξ ∈ C, α ∈ ξ} → C Ψn(cid:0)[Bm sm←− Bm−1] · · · [B1 s1←− B0](cid:1) := m[k=0 (sk · · · s1)−1.Bk and note that Ψn(φn(ξ, α)) ⊂ ξ for all α ∈ ξ. Consequently, ψn(η) = [β∈η Ψn(β) for any η ∈ C[n]. In the following we shall see that passing to higher ball shift does indeed allow one to recode any finite type convex shift into a 1-step convex shift. First though, we have to deal with the higher ball shifts of the full convex shift. 16 PERE ARA AND MATIAS LOLK Lemma 3.15. The n-ball subshift F[n] y C[n] of the full convex shift is 1-step. Proof. It should be clear from the above construction that the 1-balls in C[n] are exactly the sets of the form {[B(s) s←− B]}s∈B ∪ {1} for some B, B(s) ∈ Bn(C) such that [B(s) s←− B] ∈ A[n] ∪ (A[n])−1 for all s ∈ B, and we claim that these sets in fact generate C[n] as a 1-step convex subshift of F[n] y C(A[n]). To this end, we let η ∈ C(A[n]) and assume that (β.η)1 is of this form for any β ∈ η. Now set ξ := {Ψn(β) β ∈ η} and observe that ξ ∈ C: we ultimately wish to show that η = φn(ξ) ∈ C[n]. Letting B ∈ Bn(C) denote the n-ball as above corresponding to η1, we will first show that ξn = B. Observe that if [B′ m sm←− Bm−1][B′ m−1 sm−1←−−− Bm−2] · · · [B′ k and sk 6= s−1 s2←− B1][B′ s1←− B] ∈ η, 2 k+1 for all 1 ≤ k ≤ m − 1. In particular, 1 then our assumption implies that Bk = B′ Ψn(β) = β for all β ∈ η, so that ξn = {Ψn(β) β ∈ ηn}. Now if β = [Bm sm←− Bm−1][Bm−1 sm−1←−−− Bm−2] · · · [B2 s2←− B1][B1 s1←− B] ∈ ηn, then of course Ψn(β) = sm · · · s1 ∈ B, and if, conversely, sm · · · s1 ∈ B for m ≤ n, then an inductive application of our standing assumption implies the existence of some β as above, hence ξn = B as desired. Finally, applying our observation to β.η for some arbitrary s1←− B] ∈ η, we see that (Ψn(β).ξ)n = Bm, and so η = φn(ξ) from the way that φn is defined. sm←− Bm−1][Bm−1 sm−1←−−− Bm−2] · · · [B2 s2←− B1][B1 β = [Bm The next lemma shows that if n < R, then we can recover the R-ball ξR from the (R − n)- (cid:3) ball φn(ξ)R−n. Lemma 3.16. If ξ ∈ C and n < R, then ξR = [β∈φn(ξ)R−n Ψn(β). Proof. We apply the notation of the above construction with m ≤ R − n. For the inclusion ⊃, it is enough to check that (sk · · · s1)−1.Bk ⊂ ξR for all k ≤ m, and this holds simply because Bk ⊂ (sk · · · s1).ξR. For the reverse inclusion, take γ ∈ ξR. If γ ≤ R − n, then γ ∈ Ψn(φn(ξ)R−n) ⊂Sβ∈φn(ξ)R−n Ψn(β), so we may assume that γ > R − n and write γ = µα with α = R − n. But then µ ∈ (α.ξ)n, so γ = µα ∈ α−1.(α.ξ)n ⊂ [β∈φn(ξ)R−n Ψn(β). IDEAL STRUCTURE Definition 3.17. Let n < R and B ∈ BR(C); we then define B[n] := φn(B)R−n = {φn(B, α) α ∈ B, α ≤ R − n} ∈ BR−n(C[n]). Lemma 3.18. If n < R, B ∈ BR(C) and ξ ∈ C, then ξR = B if and only if φn(ξ)R−n = B[n]. Proof. Assuming ξR = B, we immediately see that φn(ξ)R−n = {φn(ξ, α) α ∈ ξ}R−n = {φn(ξ, α) α ∈ ξ, α ≤ R − n} = {φn(B, α) α ∈ B, α ≤ R − n} = B[n]. On the other hand, if φn(ξ)R−n = B[n], then ξR = [β∈φn(ξ)R−n Ψn(β) = [β∈φn(B)R−n Ψn(β) = BR = B by Lemma 3.16. 17 (cid:3) (cid:3) Corollary 3.19. The allowed R-balls of the n-ball shift of a convex subshift θ : F y Ω is given by BR(Ω[n]) = {B[n] B ∈ BR+n(Ω)}. Moreover, if Ω is R-step and n < R, then Ω[n] is (R − n)-step. In particular, Ω[R−1] is 1-step. Proof. The first claim follows immediately from Lemma 3.18. For the second one, we must also refer to Lemma 3.15. (cid:3) We can now prove that every finite type convex subshift can be recoded into a 1-step convex subshift. Proposition 3.20. If θ : F y Ω is a convex subshift and n ≥ 1, then θ[n] ≈−→ θ. In particular, any finite type convex subshift is directly dynamically equivalent to a 1-step convex subshift. Proof. We have already established that the map ψn is a Ψn-equivariant homeomorphism Ω[n] → Ω. Now assume that β = [Bm sm←− Bm−1] · [Bm−1 sm−1←−−− Bm−2] · · · [B1 s1←− B0] and s1←− B′ 0] are distinct elements of F[n : Ω] satisfying Ψn(β) = Ψn(β′). Then Bk 6= B′ k for some k, and m−2] · · · [B′ 1 m−1] · [B′ β′ = [B′ m sm←− B′ sm−1←−−− B′ m−1 (cid:0)(sk · · · s1).ψn(η)(cid:1)n = Bk 6= B′ k =(cid:0)(sk · · · s1).ψn(η′)(cid:1)n for all η ∈ Ω[n] ξ ∈ Ωα, we can consider φn(ξ) ∈ Ω[n]. Then β := φn(ξ, α) ∈ φn(ξ) satisfies Ψn(β) = α and β′ = ∅ as desired. Finally, given any α ∈ F and β′ , hence Ω[n] β , η′ ∈ Ω[n] β ∩ Ω[n] ξ = ψn(φn(ξ)) ∈ ψn(Ω[n] β ), 18 PERE ARA AND MATIAS LOLK hence Ωα =SΨn(γ)=α ψn(Ω[n] follows directly from Proposition 3.11 and Corollary 3.19. γ ). We conclude that θ[n] ≈−→ θ. The second part of the claim now (cid:3) One issue we have not yet dealt with is the identification of the n-ball shift and the n-fold 1-ball shift of a given convex subshift. Proposition 3.21. Consider any convex subshift θ : F y Ω. Then the n-ball convex subshift θ[n] is conjugate to the n-fold 1-ball convex subshift θ[1]···[1]. Proof. We will show that θ[n+1] ∼= θ[1][n] for all n ≥ 1 from which the claim follows inductively. We first define a map Φ : A[n+1 : Ω] → A[1 : Ω][n : Ω[1]] by Φ([B2 s←− B1]) :=(cid:2)B[1] 2 s←−B1 B1 1←−−−−− B[1] 2 1 (cid:3) =(cid:2)φ1(B2)n φ1(B1,s) ←−−−−− φ1(B1)n(cid:3), which is easily seen to be a bijection by Lemma 3.19, so we obtain an induced isomorphism Φ : F[n+1 : Ω] → F[1 : Ω][n : Ω[1]]. The pair (ϕ, Φ), where ϕ : C(A[n+1 : Ω]) → C(A[1 : Ω][n : Ω[1]]) is given by ϕ(η) := {Φ(β) β ∈ η}, then defines a conjugacy. In order to see that it restricts to a conjugacy Ω[n+1] → Ω[1][n], it suffices to check that ϕ(B[n+1]) = B[1][n] for all R ≥ 1 and B ∈ BR+n+1(Ω) due to Corollary 3.19. Recalling that B[n+1] = φn+1(B)R = {φn+1(B, α) α ∈ B, α ≤ R} and B[1][n] = φn(φ1(B)R+n)R = {φn(φ1(B), φ1(B, α)) α ∈ B, α ≤ R}, in fact we only have to verify that Φ(φn+1(B, α)) = φn(φ1(B), φ1(B, α)) for all α ∈ B with α ≤ R. Writing α = sm · · · s1 and tk :=(cid:2)(sk · · · s1.B)1 sk←− (sk−1 · · · s1.B)1(cid:3) = φ1(sk−1 · · · s1.B, sk) so that φ1(sk · · · s1.B) = tk · · · t1.φ1(B) and φ1(B, α) = tm · · · t1, we then see that Φ(φn+1(B, α)) = Φ(cid:16)(cid:2)(sm · · · s1.B)n+1 sm←− (sm−1 · · · s1.B)n+1(cid:3) · · ·(cid:2)(s1.B)n+1 s1←− Bn+1(cid:3)(cid:17) =(cid:2)φ1(sm · · · s1.B)n φ1(sm−1···s1.B,sm) ←−−−−− φ1(B)n(cid:3) =(cid:2)(tm · · · t1.φ1(B))n tm←− (tm−1 · · · t1.φ1(B))n(cid:3) · · ·(cid:2)(t1.φ1(B))n t1←− φ1(B)n(cid:3) ←−−−−−−−−−−− φ1(sm−1 · · · s1.B)n(cid:3) · · ·(cid:2)φ1(s1.B)n φ1(B,s1) = φn(φ1(B), tm · · · t1) = φn(φ1(B), φ1(B, α)) as desired. (cid:3) IDEAL STRUCTURE 19 The following theorem explains the precise relationship between the partial actions θ(E,C) and θ(En,C n), adding a dynamical dimension to [7, Theorem 8.3]. Note that while [7, Theorem 5.7] implies Lab(En, C n) ∼= Lab(E, C) and O(En, C n) ∼= O(E, C) for any finite bipartite graph and n ≥ 1, the corresponding result for Or is not immediate. Theorem 3.22. If (E, C) is a finite bipartite graph and n ≥ 1, then θ(En,C n) is conjugate to the n-ball convex subshift (θ(E,C))[n]. In particular, there are base-preserving isomorphisms Lab(En, C n) ∼= Lab(E, C), O(En, C n) ∼= O(E, C) and Or(En, C n) ∼= Or(E, C) induced from the direct dynamical equivalence θ(En,C n) ∼= (θ(E,C))[n] ≈−→ θ(E,C). The isomorphisms on Lab Moreover, there is a canonical bijective correspondence K and O coincides with the ones induced by Φn as defined in [7]. E0 n ∋ v 7→ B(v) ∈ Bn(Ω(E, C)), and the homeomorphism Ω(En, C n) → Ω(E, C)[n] → Ω(E, C) restricts to a homeomorphism Ω(En, C n)v → {ξ ∈ Ω(E, C) ξn = B(v)} for all v ∈ E0 n. Proof. Set A := E1 and Ω := Ω(E, C) for notational simplicity; by Proposition 3.21, it suffices to verify our claims for n = 1. Observe first that B1(Ω) = B0 B0 1(Ω) :=(cid:8){1}⊔{x1, . . . , xku}−1 u ∈ E0,0, xj ∈ X u j(cid:9) and B1 using the standard notation. The alphabet A[1 : Ω] is therefore given by 1(Ω) := {{1}⊔s−1(v) v ∈ E0,1}, 1(Ω), where 1(Ω) ⊔ B1 A[1 : Ω] =(cid:8)(cid:2){x1, . . . , xku}−1 xi←− s−1(s(xi))(cid:3) u ∈ E0,0, xj ∈ X u j , i = 1, . . . , ku(cid:9) and we can define an isomorphism Φ : F[1 : Ω] → F(E1 1) by Since both convex shifts are 1-step, we simply have to check that Φ induces a bijection (cid:2){x1, . . . , xku}−1 xi←− s−1(s(xi))(cid:3) 7→ αxi(x1, . . . , bxi, . . . , xku)−1. We first build some notation: Given any u ∈ E0,0 and (x1, . . . , xku) ∈Qku B1(Ω[1]) → B1(Ω(E1, C 1)), B 7→ Φ(B). j=1 X u j , we define 1(x1, . . . , xku) :=n(cid:2){1} ⊔ s−1(s(xi)) x−1 i←−− {1} ⊔ {x1, . . . , xku}−1(cid:3) : i = 1, . . . , kuo. B0 Also, for all xi ∈ X u i , we set and define Z(xi) :=n{x1, . . . , bxi, . . . , xku}−1 xj ∈ X u j , j 6= io B1 1(v, {z(e)}e∈s−1(v)) :=n(cid:2){1} ⊔ z(e) e←− {1} ⊔ s−1(v)(cid:3) : e ∈ s−1(v)o 20 PERE ARA AND MATIAS LOLK for any v ∈ E0,1 and z(e) ∈ Z(e). It is easily checked that B1(Ω[1]) = B0 where 1(Ω[1]) ⊔ B1 1(Ω[1]), and Now observe that B0 B1 X u 1(x1, . . . , xku) u ∈ E0,0, (x1, . . . , xku) ∈ jo kuYj=1 1(v, {z(e)}e∈s−1(v)) v ∈ E0,1, z(e) ∈ Z(e)o. 1(Ω[1]) :=nB0 1(Ω[1]) :=nB1 1(x1, . . . , xku)) = {1} ⊔(cid:8)αxi(x1, . . . , bxi, . . . , xku) i = 1, . . . , ku(cid:9) 1 (v(x1, . . . , xku)) = {1} ⊔ s−1 Φ(B0 hence Φ maps B0 that 1(Ω[1]) onto the 1-balls of type (c1) (cf. Definition 2.11). Likewise, we see Φ(B1 1(v, {z(e)}e∈s−1(v))) = {1} ⊔(cid:8)αe(z(e))−1 e ∈ s−1(v)(cid:9), 1(Ω[1]) onto the 1-balls of type (c2). We conclude that the pair (ϕ, Φ) with so Φ maps B1 ϕ : Ω[1] → Ω(E1, C 1) given by ϕ(ξ) := {Φ(α) α ∈ ξ} is a conjugacy of F[1 : Ω] y Ω[1] and F(E1 1) y Ω(E1, C 1). Consequently, there is a direct dynamical equivalence θ(En,C n) ∼= (θ(E,C))[n] ≈−→ θ(E,C), which induces base-preserving isomor- phisms by Remark 3.9. It follows from Remark 3.12 that the isomorphisms on Lab and O are exactly the ones of [7]. We now turn to the second part of the claim and set B(v) :=(cid:26) {1} ⊔ s−1(v) if v ∈ E0,0 if v = v(x1, . . . , xku) ∈ E0,1 1 1 ; this clearly defines a bijective correspondence between E0 {1} ⊔ {x1, . . . , xku}−1 1 = E0,1 for every v ∈ E0 1 and B1(Ω(E, C)). Note that the homeomorphism Ω(E1, C 1) → Ω(E, C) is induced by the group homomorphism F(E1 , and it is easily checked that it maps Ω(E1, C 1)v onto {ξ ∈ Ω(E, C) ξ1 = B(v)}. (cid:3) 1) → F(E1) given by αxi(x1, . . . , bxi, . . . , xku) 7→ x−1 i Remark 3.23. Theorem 3.22 shows that for v ∈ E0 be described as Ω(E, C)v = {ξ ∈ Ω(E, C) ξn = B(v)}. n with n ≥ 1, the subspace Ω(E, C)v may Any partial action on a topological space may be viewed as the restriction of a global action [1, Theorem 2.5]. The globalisation is not Hausdorff in general [1, Proposition 2.10], but whenever it is, one may consider the relationship between the C ∗-algebras of the partial action and its globalisation. However, it is also of natural interest to study restrictions of partial actions, in particular in cases where there is no Hausdorff globalisation, and they play a natural role in our main theorem about convex subshifts. IDEAL STRUCTURE 21 Definition 3.24 (See also [21, Definition 3.1], [36, Definition 2.17]). If θ : G y Ω is a partial action on a topological space and U ⊂ Ω is an open subset, then we denote by θU the restricted partial action G y U with domains and U is called G-full if Ug := θg(U ∩ Ωg−1) ∩ U, X = {g.x g ∈ G, x ∈ U ∩ Xg−1}. Finally, two partial actions θ : G y Ω and γ : H y Ω′ are called Kakutani-equivalent if there exist clopen subspaces K ⊂ Ω and K ′ ⊂ Ω′, resp. G- and H-full, such that θK ≈ γK ′. If θ : G y Ω and θ′ : G y Ω′ are Kakutani equivalent partial actions on totally discon- nected, locally compact spaces, then the groupoids Gθ and Gθ′ are Kakutani equivalent in the sense of [21, Definition 3.1] and hence groupoid equivalent by [21, Theorem 3.2]. It follows that there are Morita-equivalences CK(Ω) ⋊θ G ∼M CK(Ω) ⋊θ′ H and C0(Ω)⋊θ,(r) ∼M C0(Ω′) ⋊θ′,(r) H, see for instance [40, Theorem 2.8], [48, Theorem 13] and [20, Theorm 5.1]. We can now state and prove the second main theorem about convex subshifts. Theorem 3.25. If θ : F y Ω is a convex subshift of finite type, then there is a finite bipartite separated graph (E, C) such that θ(E,C) and θ are Kakutani equivalent. Proof. By Proposition 3.20, we may assume that Ω is 1-step. Now let A[1 : Ω] denote disjoint copies of the alphabet A[1 : Ω] with subscripts + and −, and define a finite bipartite separated graph (E, C) by and A[1 : Ω] − + + , ⊔ A[1 : Ω] • E0,0 := B1(Ω) and E0,1 = A[1 : Ω], • E1 := A[1 : Ω] • r([B a←− B′]+) := B and r([B a←− B′]−) := B′, • s([B a←− B′]+) := s([B a←− B′]−) = [B a←− B′], • CB := {XB(s) 1 6= s ∈ B} for all B ∈ E0,0, where for a ∈ A − XB(a−1) := {[B a←− B′]+ B′ ∈ E0,0 such that B a←− B′}, XB(a) := {[B′ a←− B]− B′ ∈ E0,0 such that B′ a←− B}. Then consider the group homomorphism Φ : F(E1) → F[1 : Ω] given by Φ([B a←− B′]+) := [B a←− B′] and Φ([B a←− B′]−) := 1 as well as the compact open F(E1)-full subspace K := Gu∈E0,0 Ω(E, C)u. Equipping K with the restricted partial action θ(E,C)K : F(E1) y K, we claim that ϕ : K → Ω[1] given by ϕ(ξ) := {Φ(α) α ∈ ξ} 22 PERE ARA AND MATIAS LOLK defines a Φ-equivariant homeomorphism, making the pair (ϕ, Φ) into a direct dynamical equivalence. First, let us check that ϕ even maps into Ω[1], so take any ξ ∈ K. Observe that any length two admissible path α ∈ ξ is of the form [B a←− B′]+[B a←− B′]−1 − or [B a←− B′]−[B a←− B′]−1 + , and these are mapped to [B a←− B′] and [B a←− B′]−1, respectively. It follows that any α ∈ ξ of length four is mapped to a length two word, hence any α ∈ ξ of length 2n is mapped to a word of length n. Note also that if α ∈ ξ has odd length, then α = [B a←− B′]−1 + β or α = [B a←− B′]−1 − β for some [B a←− B′] ∈ A[1 : Ω], and in the latter case, Φ(α) = Φ(β). In the former case, there is a unique extension of length α + 1 inside ξ, namely [B a←− B′]−[B a←− B′]−1 + β ∈ ξ, and Φ(α) = Φ([B a←− B′]−α), so in conclusion ϕ(ξ) = {Φ(α) : α ∈ ξ, α is even}. In particular, ϕ is continuous and ϕ(ξ)1 = Φ(ξ2), so we only need to check that Φ(ξ2) ∈ B1(Ω[1]) for any ξ ∈ K. Assuming that ξ ∈ Ω(E, C)B, for every 1 6= s ∈ B there is B(s) ∈ B1(Ω) such that Φ(ξ2) = {1} ⊔ {[B a←− B(a−1)]−1 a ∈ A ∩ B−1} ⊔ {[B(a) a←− B] a ∈ A ∩ B} = {1} ⊔ {[B(s) s←− B] 1 6= s ∈ B}, and this is exactly an element of B1(Ω[1]). At this point we have verified that ϕ is a well-defined continuous Φ-equivariant map, and we now turn to the construction of an inverse. Define a group homomorphism Σ : F[1 : Ω] → F(E1) and a continuous Σ-equivariant map σ : Ω[1] → K by Σ([B a←− B′]) := [B a←− B′]+[B a←− B′]−1 − and σ(η) := conv{Σ(β) β ∈ η}, where conv(H) for a set H ⊂ F(E1) denotes the convex closure. Observing that Φ is a one- sided inverse of Σ, it follows that Σ is injective and hence that σ is a continuous Σ-equivariant map into C(E1); but, we still have to verify that σ maps into K. Since σ(η)2 = conv(Σ(η1)) for all η ∈ Ω[1], it suffices to check that conv(Σ(η1)) ∈ B2(Ω(E, C)). By construction, η1 is of the form η1 = {1} ⊔ {[B(s) s←− B] 1 6= s ∈ B} = {1} ⊔ {[B a←− B(a−1)]−1 a ∈ A ∩ B−1} ⊔ {[B(a) a←− B] a ∈ A ∩ B} for some B, B(s) ∈ B1(Ω), so Σ(η1) = {1} ⊔ {[B a←− B(a−1)]−[B a←− B(a−1)]−1 + a ∈ A ∩ B−1} ⊔ {[B(a) a←− B]+[B(a) a←− B]−1 − a ∈ A ∩ B}. Taking the convex closure of this, we clearly obtain a 2-ball of Ω(E, C), hence σ(η) ∈ Ω(E, C)B ⊂ K IDEAL STRUCTURE 23 as desired. We now claim that ϕ and σ are in fact mutual inverses. Noting that Σ(Φ(α)) = α for α ∈ ξ of even length, we indeed have σ(ϕ(ξ)) = conv{Σ(Φ(α)) α ∈ ξ, α is even} = conv{α α ∈ ξ, α is even} = ξ and ϕ(σ(η)) = ϕ(conv{Σ(β) β ∈ η}) = {Φ(Σ(β)) β ∈ η} = η. Letting F := Im(Σ) ≤ F(E1) so that Σ = Φ−1 F , we conclude that the partial actions F y K and F[1 : Ω] y Ω[1] are conjugate. Finally observing that, by the above observations, Kα = ∅ for all α ∈ F(E1) \ F , we conclude that (ϕ, Φ) is indeed a direct dynamical equivalence ≈−→ θ[1], from which we obtain the desired direct dynamical equivalence as the com- θ(E,C)K position θ(E,C)K (cid:3) ≈−→ θ[1] ≈−→ θ. In view of the above theorem, the study of convex subshifts of finite type boils down to the study of dynamical systems associated with finite bipartite graphs, at least up to Katutani equivalence. In the following sections, we shall see how one can extract information about the open/closed invariant subspaces from the graph, illustrating the usefulness of having a graph representation. We end this section with an application of Theorem 3.25 to a pair of concrete examples. Example 3.26. Given a finite alphabet A, there is of course a finite bipartite separated graph (E, C) corresponding to the full convex shift on A as in Theorem 3.25. However, one can check that E0 = 4(A4 + A2) and E1 = 8A4, so even when A = 2 this is a fairly sizable graph. We shall therefore refrain from drawing it here. Example 3.27. Consider the alphabet A = {a, b} and the 1-step subshift F2 y Ω with B1(Ω) = {u, v}, where u = {1, a±1, b±1} and v = {1, a±1, b} as illustrated just below. We will u = a b b a v = a a b then describe the separated graph (E, C) of Theorem 3.25. We have E0,1 = A[1 : Ω] = {[u a←− u], [u b←− u], [v a←− u], [u a←− v], [u b←− v], [v a←− v]} and E1 = A[1 : Ω] + ⊔ A[1 : Ω] − with r−1(u) = {[u a←− u]±, [u b←− u]±, [v a←− u]−, [u a←− v]+, [u b←− v]+}, r−1(v) = {[v a←− u]+, [u a←− v]−, [u b←− v]−, [v a←− v]±}, 24 PERE ARA AND MATIAS LOLK and the source map is simply the projection E1 = A[1 : Ω] separation is given by + ⊔ A[1 : Ω] − → A[1 : Ω] = E0,1. The Cu = {Xu(a), Xu(a−1), Xu(b), Xu(b−1)} and Cv = {Xv(a), Xv(a−1), Xv(b)}, where Xu(a) = {[u a←− u]−, [v a←− u]−}, Xu(b) = {[u b←− u]−}, Xv(a) = {[u a←− v]−, [v a←− v]−}, Xv(b) = {[u b←− v]−} . Xu(a−1) = {[u a←− u]+, [u a←− v]+}, Xu(b−1) = {[u b←− u]+, [u b←− v]+}, Xv(a−1) = {[v a←− u]+, [v a←− v]+}, We can therefore picture (E, C) as follows: u v [u a←− u] [u b←− u] [v a←− u] [u a←− v] [u b←− v] [v a←− v] 4. The lattice of induced ideals In this section, we describe the lattice of induced ideals of the algebras Lab(E, C), O(E, C) and Or(E, C) for (E, C) finite and bipartite in terms of graph-theoretic data, specifically certain sets of vertices in the infinite layer graph (F∞, D∞) (see Definition 2.6). We first settle the meaning of the various types of ideals that we shall encounter. When dealing with C-algebras, we will consider only closed ideals, so that the word ideal will mean closed ideal in this case. For a general ring R, an ideal I ⊳ R is called a trace ideal if it is generated by the entries of some set of idempotent matrices over R, and we denote the lattice of trace ideals by Tr(R). The lattice of idempotent-generated ideals Idem(R) then sits as a sublattice of Tr(R). Given a crossed product (algebraic or C ∗-algebraic, reduced or universal) O = A ⋊(r) G, we say that an ideal J ⊳ O is induced if J = (J ∩ A) ⋊(r) G, and we denote by Ind(O) the lattice of induced ideals. Finally, if M is an abelian monoid, then a submonoid I ⊂ M is called an order-ideal if x + y ∈ I implies x, y ∈ I. The lattice of order ideals of M will be denoted by L(M). The basic tools in our analysis are the following results. Theorem 4.1. [11, Proposition 10.10] For any ring R, there is a lattice isomorphism L(V(R))) ∼= Tr(R). Moreover, if V(R) is generated by the classes [e] of idempotents of R, then Tr(R) = Idem(R). IDEAL STRUCTURE 25 Theorem 4.2 ([25]). Let G y A be a partial action of a discrete group G on a C ∗-algebra. Then the map J 7→ J ∩ A defines a bijective correspondence between Ind(A ⋊(r) G) and the lattice of invariant ideals of A. Moreover, if I is an invariant ideal of A, then and if G is exact, then (A ⋊r G)/(I ⋊r G) ∼= (A/I) ⋊r G as well. (A ⋊ G)/(I ⋊ G) ∼= (A/I) ⋊ G, Proof. If J is an ideal of A ⋊(r) G, then J ∩ A is a G-invariant ideal of A by [25, Proposition 23.11], and if I is an invariant ideal of A, then I ⋊(r) G is an ideal of A ⋊(r) G by [25, Proposi- tion 21.12 and 21.15]. In particular, we have the above mentioned bijective correspondence. Moreover, (A ⋊ G)/(I ⋊ G) ∼= (A/I) ⋊ G by [25, Proposition 21.15], and in case G is exact, then (A ⋊r G)/(I ⋊r G) ∼= (A/I) ⋊r G as well by [25, Theorem 21.18]. (cid:3) Remark 4.3. We note that it is straightforward to prove a result completely analogous to Theorem 4.2 for partial actions on ∗-algebras. We now recall some definitions from [11], adapted to our choice of conventions. These notions generalize the classical notions of hereditary and saturated subsets of vertices, cf. [44, Chapter 4], to the separated setting. Definition 4.4. [11, Definition 6.3] Let (E, C) be a finitely separated graph. A subset H of E0 is said to be hereditary if for any e ∈ E1, we have r(e) ∈ H implies s(e) ∈ H, and H is said to be C-saturated if for any v ∈ E0 and X ∈ Cv, s(X) ⊂ H implies v ∈ H. We denote by H(E, C) the lattice of hereditary C-saturated subsets of E0. Theorem 4.5. Let (E, C) be a finite bipartite separated graph. Then there are lattice iso- morphisms Idem(Lab K (E, C)) ∼= L(M(F∞, D∞)) ∼= H(F∞, D∞). If H ∈ H(F∞, D∞), the ideal I(H) of Lab the ideal generated by all the projections πn,∞(v), where v ∈ H ∩ E0 Lab K (E, C) is the natural quotient map. Proof. By [7, Corollary 5.9], we have an isomorphism V(Lab(E, C)) ∼= M(F∞, D∞). In partic- ular, V(Lab(E, C)) is generated by the classes of the idempotents in Lab(E, C) corresponding to the vertices in F∞. By Theorem 4.1, we obtain K (E, C) associated to H through this isomorphism is n, and πn,∞ : L(En, C n) → L(M(F∞, D∞)) ∼= L(V(Lab(E, C))) ∼= Tr(Lab(E, C)) = Idem(Lab(E, C)). On the other hand, by [11, Corollary 6.10], we have L(M(F∞, D∞)) ∼= H(F∞, D∞), so that we finally obtain a lattice isomorphism Idem(Lab(E, C)) ∼= H(F∞, D∞). (cid:3) Let Ω be a zero-dimensional metrizable locally compact Hausdorff space, and let K = K(Ω) be the subalgebra of P(Ω) consisting of all the compact open subsets of Ω. Let θ : G y Ω be a partial action of a discrete group G by continuous transformations on Ω such that Ωg ∈ K for all g ∈ G. Observe that K is then automatically G-invariant. The (relative) type semigroup S(Ω, G, K) has been defined in [7, Definition 7.1], see also [34] and [46]. The semigroup 26 PERE ARA AND MATIAS LOLK S(Ω, G, K) is indeed a conical refinement monoid, and we obtain the following description of its lattice L(S(Ω, G, K)) of order ideals. Lemma 4.6. Let θ : G y Ω be a partial action of a discrete group G by continuous trans- formations on Ω such that Ωg ∈ K for all g ∈ G. Then there are mutually inverse, order- preserving maps ϕ : L(S(Ω, G, K)) → OG(Ω), ϕ(I) =[{K ∈ K [K] ∈ I}, ψ : OG(Ω) → L(S(Ω, G, K)), ψ(U) = h[K] K ∈ K, K ⊆ Ui, where OG(Ω) is the lattice of G-invariant open subsets of Ω, and, for T ⊆ S(Ω, G, K), hT i stands for the order ideal of S(Ω, G, K) generated by T . Proof. Write S := S(Ω, G, K) and take I ∈ L(S). Clearly U := ϕ(I) is an open subset of Ω. If x ∈ Ωg−1 ∩ U for some g ∈ G, then there is K ∈ K with [K] ∈ I such that x ∈ Xg−1 ∩ K. But now we have θg(x) ∈ θg(Ωg−1 ∩ K) with [θg(Ωg−1 ∩ K)] = [Ωg−1 ∩ K] ≤ [K] ∈ I, Let U be an invariant open subset of Ω. Then ψ(U) ∈ L(S) by definition of ψ. and so [θg(Ωg−1 ∩ K)] ∈ I because I is an order ideal of S. It follows that U is G-invariant. It is clear that ϕ and ψ are order-preserving maps. We have to show that (ϕ ◦ ψ)(U) = U and (ψ ◦ ϕ)(I) = I for U ∈ OG(X) and I ∈ L(S). For U ∈ OG(Ω), let K ∈ K be such that [K] ∈ ψ(U). Then there are K1, . . . , Kr ∈ K such that Kj ⊆ U for j = 1, . . . , r and [K] ≤ [K1] + [K2] + · · · + [Kr]. Using the refinement property of S and the definition of the type semigroup, one obtains a decomposition K = ⊔n i ∈ K for each i, and g1, . . . , gn ∈ G such that, for each i, K ′ i such that K ′ i) ⊆ Kj for some j = 1, . . . , r. It follows that i ⊆ Ωg−1 i=1K ′ and θgi(K ′ K ′ i i ⊆ θg−1 i (Kj ∩ Ωgi) ⊆ θg−1 i (U ∩ Ωgi) ⊆ U, where the last containment follows from the fact that U is G-invariant. We deduce that K ⊆ U, and so ϕ(ψ(U)) ⊆ U. The other containment U ⊆ ϕ(ψ(U)) follows from the fact that Ω has a basis of compact open subsets. Finally, let I ∈ L(S). It is clear that I ⊆ ψ(ϕ(I)). To show the reverse inclusion, it is enough to check that, if K ∈ K and K ⊆ ϕ(I), then [K] ∈ I. By compactness of K, there are K1, . . . , Kr ∈ K such that [Ki] ∈ I and K ⊆ ∪r i=1Ki, and thus [K] ≤ [K1] + · · · + [Kr] ∈ I. Since I is an order ideal of S, we see that [K] ∈ I, as desired. (cid:3) We can now obtain a description of the lattice of induced ideals of tame graph algebras. Theorem 4.7. Let (E, C) be a finite bipartite separated graph. Then there is a lattice iso- morphism Ind(Lab(E, C)) ∼= Ind(O(r)(E, C)) ∼= L(M(F∞, D∞)) ∼= H(F∞, D∞). IDEAL STRUCTURE 27 Moreover for H ∈ H(F∞, D∞), we have Lab(E, C)/I(H) ∼= CK(Z) ⋊θ∗ where Z := Ω(E, C) \ U with U :=Sv∈H Ω(E, C)v. Z F and O(r)(E, C)/I(H) ∼= C(Z) ⋊(r),θ∗ F, Z Proof. It follow from [7, Theorem 7.4] that there is a natural isomorphism S := S(Ω(E, C), F, K) ∼= M(F∞, D∞). Combining this with Theorem 4.2 (or Remark 4.3), [11, Corollary 6.10] and Lemma 4.6, we obtain Ind(Lab(E, C)) ∼= Ind(O(r)(E, C)) ∼= OF(Ω(E, C)) ∼= L(S) ∼= L(M(F∞, D∞)) ∼= H(F∞, D∞). The last part follows from Theorem 4.2 and the definitions of the lattice isomorphisms. (cid:3) Remark 4.8. We believe it is likely that Theorem 4.5 generalizes to the setting of tame graph C-algebras, at least for the reduced ones. This would mean that we have a lattice isomorphism Proj(Or(E, C)) ∼= L(M(F∞, D∞)) ∼= H(F∞, D∞), where Proj(Or(E, C)) denotes the lattice of ideals of Or(E, C) which are generated by their projections. By Theorem 4.7, this is equivalent to saying that every ideal generated by projections is induced. In Section 7, we will prove this for ideals I ⊳ Or(E, C) of finite type. 5. The ideals associated to hereditary C-saturated subsets of (E, C) In this section, we will analyze the induced ideals of Lab(E, C), O(E, C) and Or(E, C) arising from hereditary C-saturated subsets of E0, as opposed to the general study of ideals corresponding to hereditary D∞-saturated subsets of (F∞, D∞). By Theorem 3.22, (E, C) and (En, C n) give rise to the same algebras for all n ≥ 0, so we can apply the corresponding results to any hereditary C n-saturated subset of the separated graph (En, C n). First, we shall give a concrete description of the hereditary and D∞-saturated closure of a subset H ∈ H(E, C) inside (F∞, D∞). Lemma 5.1. Let (E, C) be a finite bipartite separated graph and take H ∈ H(E, C). Let H1 := {s1(X(x)) x ∈ E1 and s(x) ∈ H} ∪ (H ∩ E0,0 1 ). Then H1 ∈ H(E1,C1) and H ∪ H1 ∈ H(F1,D1) is the hereditary closure of H inside (F1, D1). Proof. H1 is clearly hereditary; if r1(e) ∈ H ∩E0,0 1 = H ∩E0,1, then e ∈ X(x) for some x ∈ E1 with s(x) = r1(e), hence s1(e) ∈ s1(X(x)) ⊆ H1. Next, we show C 1-saturation. Suppose that s(X(x)) ⊆ H1 for some x ∈ E1, and write w := s(x). Also, set x = xi ∈ X u i with u := r(x) and Cu = {X u ku}. If ku = 1, then necessarily w ∈ H by the definition of H1, so suppose that ku > 1. If for some j 6= i, we have s(xj) ∈ H for all xj ∈ X u j , then u ∈ H by C-saturation, and so s(x) = s(xi) ∈ H because H is hereditary. Thus, we may assume that, for all j 6= i, there exists xj ∈ X u j such that s(xj) /∈ H. Now, consider the vertex 1 , . . . , X u i , . . . , X u v := v(x1, . . . , xi−1, xi, xi+1, . . . , xku) ∈ E0,1 1 . 28 PERE ARA AND MATIAS LOLK Then v ∈ H1 because v = s(αxi(x1, . . . , xi−1, xi+1, . . . , xku)) ∈ s(X(xi)) = s(X(x)) ⊆ H1. But by the definition of H1, there must be k ∈ {1, . . . , ku} such that s(xk) ∈ H. Hence k = i and w = s(xi) ∈ H, as desired. It is now clear that H ∪ H1 ∈ H(F1, D1), and H ∪ H1 is obviously nothing but the hereditary closure of H inside (F1, D1). (cid:3) Notation 5.2. Given H ∈ H(E, C), we define a sequence Hn ∈ H(En, C n) in an inductive way, so that Hn := {s(X(x)) x ∈ E1 n−1 and s(x) ∈ Hn−1} ∪ (Hn−1 ∩ E0,0 n ). Then, by Lemma 5.1, H∞ := S∞ (F∞, D∞). n=0 Hn ∈ H(F∞, D∞) is the hereditary closure of H inside We have thus showed the following lemma: Lemma 5.3. Let (E, C) be a finite bipartite separated graph. Then there is an injective order-preserving map H(E, C) → H(F∞, D∞) sending H ∈ H(E, C) to H∞ ∈ H(F∞, D∞). Moreover, the ideal I(H) of Or(E, C) generated by H is precisely the ideal I(H∞), and similar statements hold for O(E, C) and for Lab K (E, C). We also mention the following easy description of the open, invariant subspace associated with H ∈ H(E, C) in terms of configuration spaces. Recall that if ξ ∈ Ω(E, C)v, then 1 ∈ ξ is regarded as the trivial path v, so that r(1) = v in this situation. Lemma 5.4. Let (E, C) be a finite bipartite graph and let H ∈ H(E, C). Then [v∈H∞ Ω(E, C)v = {ξ ∈ Ω(E, C) r(α) ∈ H for some α ∈ ξ}. Proof. Since I(H) = I(H∞), we have U = θF(cid:16) [v∈H Ω(E, C)v(cid:17) = {ξ ∈ Ω(E, C) r(α) ∈ H for some α ∈ ξ}. Given a finitely separated graph (E, C) and a hereditary C-saturated subset H of E0, we denote by E/H the subgraph of E with (E/H)0 := E0 \ H and (E/H)1 := {e ∈ E1 s(e) /∈ H}. Similarly, for any subset X ⊆ E1, define For v ∈ (E/H)0, we set X/H := X ∩ s−1(E0 \ H). (cid:3) which is a partition of r−1 X ∈ Cv with v ∈ E0 \ H, because H is C-saturated. E/H(v), and C/H :=Fv∈E0\H (C/H)v. Observe that X/H 6= ∅ for all (C/H)v := {X/H X ∈ Cv}, IDEAL STRUCTURE 29 Theorem 5.5. Let (E, C) be a finite bipartite separated graph and let H ∈ H(E, C). Then there is a natural ∗-algebra isomorphism K (E, C)/I(H) ∼= Lab Lab K (E/H, C/H). Likewise, there are natural C-algebra isomorphisms O(E, C)/I(H) ∼= O(E/H, C/H) and Or(E, C)/I(H) ∼= Or(E/H, C/H). Proof. As observed in the proof of [11, Corollary 3.12], it is easy to show using universal properties that the map sending v + I(H) 7→ v for v ∈ (E/H)0 and e + I(H) 7→ e for e ∈ (E/H)1 extends to a ∗-isomorphism C ∗(E, C)/I(H) −→ C ∗(E/H, C/H). Likewise, we obtain a ∗-isomorphism LK(E, C)/I(H) ∼= LK(E/H, C/H) for any field with involution K. It is straightforward to check that (cid:0)E1/H1, C 1/H1) =(cid:0)(E/H)1, (C/H)1(cid:1). Indeed we have that (E1/H1)0,0 = (E/H)0,1 = ((E/H)1)0,0 and that (E1/H1)0,1 is the set of all the vertices v(x1, x2, . . . , xk) such that xi ∈ Xi/H for all i, where Cv = {X1, X2, . . . , Xk} for some v ∈ E0,0 \ H = (E/H)0,0. This shows that (E1/H1)0 = ((E/H)1)0, and similarly (E1/H1)1 = ((E/H)1)1 and C 1/H1 = (C/H)1. We thus obtain the following commutative diagram L(E, C)/I(H) ∼= L(E/H, C/H) L(E1, C 1)/I(H1) ∼= L(E1/H1, C 1/H1) ∼= L((E/H)1, (C/H)1) , where all the maps are the canonical ones. Applying this observation inductively gives iden- tifications L(En, C n)/I(Hn) ∼= L((E/H)n, (C/H)n) commuting with the connecting homo- morphisms, and hence we obtain an isomorphism Lab(E, C)/I(H) = Lab(E, C)/I(H∞) ∼= Lab(E/H, C/H) of the limits. The same proof applies to O. We now give a proof for Or, which uses the dynamical interpretation of these algebras. Let F′ denote the free group on (E/H)1, which we can regard as a subgroup of F. Let U := Sv∈H∞ Ω(E, C)v be the open invariant subset of Ω(E, C) associated to H∞, and set Z := Ω(E, C) \ U. By Theorem 4.7, we have Or(E, C)/I(H) = Or(E, C)/I(H∞) ∼= C(Z) ⋊r,θ∗ Z F. Let us denote by θ′ the restriction of θ to Z. If x ∈ E1 and s(x) ∈ H, then the domain and codomain of θ′ x is empty, so we have that Or(E, C)/I(H) ∼= C(Z) ⋊r,(θ′)∗ F′, 30 PERE ARA AND MATIAS LOLK and we only have to show that the action θ′ of F′ on Z is conjugate to θ(E/H,C/H). Observe that by Lemma 5.4, Z = {ξ ∈ Ω(E, C) r(α) /∈ H for all α ∈ ξ}. We now claim that in fact Z = Ω(E/H, C/H), so let ξ ∈ Ω(E/H, C/H). Since every α ∈ ξ satisfies r(α) ∈ (E/H)0 = E0 \ H, we simply need to verify that the local configurations ξα are either of type (c1) or (c2) with respect to (E, C). Assume first that ξ is of type (c1) with respect to (E/H, C/H), i.e. that ξα = s−1 E/H(r(α)). Now since H is hereditary, we must have ξα = s−1 E/H(r(α)) = s−1(r(α)), so ξα is indeed of type (c1) with respect to (E, C). Next, assume that ξα is of type (c2) with respect to (E/H, C/H), i.e. that ξα = {e−1 X/H X/H ∈ (C/H)r(α)} for some eX/H ∈ X/H. From H being C-saturated, we have (C/H)r(α) = {X/H X ∈ Cr(α)}, so setting eX := eX/H for all X ∈ Cr(α), ξα = {e−1 X/H X/H ∈ (C/H)r(α)} = {e−1 X X ∈ Cr(α)} is of type (c2) with respect to (E, C). The converse inclusion is completely straightforward and does not use (explicitly) the assumptions about H being hereditary and C-saturated. Finally, since the dynamics is completely determined by the configurations, and the configuration spaces agree, we conclude that the two partial action are in fact conjugate. Observe that, by Theorem 4.7 and Remark 4.8(1), the same proof applies to O and Lab (cid:3) respectively, so we obtain a second proof for those. 6. The ideals associated to hereditary D∞-saturated subsets of (F∞, D∞) Recall from Section 4 that every induced ideal of a tame graph algebra corresponds to a set H ∈ H(F∞, D∞). In this section, we shall describe the induced ideal, or rather its quotient, in terms of the intersections H ∩ E0 n. We shall also consider a number of examples. The proof of the following lemma is straightforward. Lemma 6.1. Let H ∈ H(F∞, D∞), and, for each n ≥ 0, set H (n) := H ∩ E0 a hereditary C n-saturated subset of E0 n, and H = ∪∞ n=0H (n). n. Then H (n) is We now describe the hereditary and D∞-saturated closure of H (n) inside (F∞, D∞): Fol- ∞, that is, ∞ the hereditary closure of H (n) in F 0 lowing Notation 5.2, we will denote by H (n) H (n) ∞ = ∪m≥nH (n) m , where H (n) m = {s(X(x)) x ∈ E1 m−1 and s(x) ∈ H (n) m−1} ∪ (H (n) m−1 ∩ E0,1 m−1). Observe that H (n) 5.1 that it is (D≥n)-saturated, that is, if v ∈ F 0,m with m ≥ n and X ∈ C m ∞ is just the hereditary closure of H (n) in F 0 ∞, but we proved in Lemma v are such that s(X) ⊆ H (n) ∞ , then v ∈ H (n) ∞ . Now define IDEAL STRUCTURE 31 H n := (H ∩ F 0 n−1) ∪ H (n) ∞ = ( n−1[i=0 H (i)) ∪ H (n) ∞ and observe that H n is exactly the hereditary and D∞-saturated closure of H (n) inside (F∞, D∞). Note also that F∞ \ H has the structure of a separated Bratteli diagram, which we denote by (F∞/H, D∞/H). Here (F∞/H)0,n = F 0,n \ H, D∞/H = ∞Gn=0 C n/H (n). Proposition 6.2. Let (E, C) denote a finite bipartite separated graph, let H ∈ H(F∞, D∞) and apply the above notation. Also, let Un denote the open invariant subspace of Ω(E, C) corresponding to H n, and set Zn := Ω(E, C) \ Un as well as Z :=T∞ Or(E, C)/I(H) ∼= C(Z) ⋊r F ∼= lim−→ C(Zn) ⋊r F ∼= lim−→ n n n=0 Zn. Then Or(En/H (n), C n/H (n)), where the connecting homomorphisms are simply induced from restriction of functions. The same statement holds with O or Lab in place of Or. Proof. Since H n is the hereditary and D∞-saturated closure of H (n), we have I(H n) = I(H (n)) as ideals of Or(E, C). It follows that Or(En/H (n), C n/H (n)) ∼= Or(En, C n)/I(H (n)) ∼= Or(E, C)/I(H n) ∼= C(Zn) ⋊r F from Theorem 3.22 and Theorem 5.5. Now let U denote the open invariant set corresponding n=0 Un and Z = Ω(E, C) \ U. Thus, n=0 H n by construction, so U =S∞ to H. We have H =S∞ we obtain the identification Or(E, C)/I(H) ∼= C(Z) ⋊r F, C(Zn) ⋊r F. The same proof applies to O and Lab. and clearly C(Z) ⋊r F ∼= lim−→n Remark 6.3. Recall from Theorem 3.22 that every v ∈ E0 n for n ≥ 1 corresponds to an n-ball B(v) ∈ Bn(Ω(E, C)). The open set Un of Proposition 6.2 may therefore be described as the set of configurations ξ ∈ Ω(E, C) satisfying the following: There is some v ∈ H (n) and α ∈ ξ for which θα(ξ)n = B(v). In other words, we can describe the set Zn as (cid:3) Zn = {ξ ∈ Ω(E, C) ξ 6≡ B(v) for all v ∈ H (n)}. Thus, the descending filtration (Zn)n exactly removes the configurations with forbidden n- balls at the n'th step. In particular, we see that the restricted action θZ : F y Z is of finite type in the sense of Definition 3.7 if and only if H = H n for some n ≥ 0. As another consequence, we see that every convex shift θ : F y Ω can be represented, up to Kakutani equivalence, by a separated Bratteli diagram: If (E, C) is the graph representing the full convex shift (see Example 3.26), then θ is Kakutani equivalent to the restriction of θ(E,C) to some closed invariant subspace Z. It follows that θ may be described as above 32 PERE ARA AND MATIAS LOLK by the separated Bratteli diagram (F∞/H, D∞/H), where H ∈ H(F∞, D∞) corresponds to U := Ω(E, C) \ Z. We now describe K0 of these quotient algebras in terms of their associated separated Bratteli diagrams. Theorem 6.4. Let (E, C) be a finite bipartite separated graph, let H be a proper heredi- tary D∞-saturated subset of F∞, and let (F∞/H, D∞/H) be its associated separated Bratteli diagram. (1) There is a natural group isomorphism K0(O(E, C)/I(H)) ∼= K0(Or(E, C)/I(H)) ∼= G(F∞/H, D∞/H). (2) There is a natural isomorphism V(Lab K (E, C)/I(H)) ∼= M(F∞/H, D∞/H), and so an isomorphism of pre-ordered abelian groups K0(Lab K (E, C)/I(H)) ∼= G(F∞/H, D∞/H) for any field with involution K. Proof. We adopt the above notation. It is straightforward to check that, for n ≥ 1 we have M(F≥n/H (n) ∞ , D≥n/H (n) ∞ ) ∼= M(F∞/H n, D∞/H n) and G(F≥n/H (n) ∞ , D≥n/H (n) ∞ ) ∼= G(F∞/H n, D∞/H n). Therefore, it follows from Theorem 2.9 and an easy calculation that there are commutative diagrams K0(O(En/H (n), C n/H (n))) ∼=−−−→ G(F∞/H n, D∞/H n) K0(O(En+1/H (n+1), C n+1/H (n+1))) ∼=−−−→ G(F∞/H n+1, D∞/H n+1) for all n ≥ 1. Using Proposition 6.2, we obtain K0(O(E, C)/I(H)) ∼= lim−→ ∼= lim−→ = G(F∞/H, D∞/H). n n G(F∞/H n, D∞/H n) K0(O(En/H (n), C n/H (n))) This gives (1) for O. The same arguments give, using Theorem 2.9, the result for Or and for Lab. (cid:3) We record some useful properties of the monoid M(F∞/H, D∞/H). y y IDEAL STRUCTURE 33 Proposition 6.5. Let (E, C) be a finite bipartite separated graph, let H be a proper heredi- tary D∞-saturated subset of F∞, and let (F∞/H, D∞/H) be its associated separated Bratteli diagram. Then we have a natural monoid isomorphism M(F∞/H, D∞/H) ∼= M(F∞, D∞)/M(H), where M(H) is the order-ideal of M(F∞, D∞) generated by H. In particular, M(F∞/H, D∞/H) is a refinement monoid. Proof. The isomorphism follows from [10, Construction 6.8]. Since M(F∞, D∞) is a refine- ment monoid, the quotient monoid M(F∞, D∞)/M(H) is also a refinement monoid, by [13, Lemma 4.3]. (cid:3) We now consider a number of examples. Example 6.6. For integers 1 ≤ m ≤ n, define the separated graph (E(m, n), C(m, n)), where (1) E(m, n)0 := {v, w} (with v 6= w). (2) E(m, n)1 := {α1, . . . , αn, β1, . . . , βm} (with n + m distinct edges). (3) s(αi) = s(βj) = v and r(αi) = r(βj) = w for all i, j. (4) C(m, n) = C(m, n)v := {X, Y }, where X := {α1, . . . , αn} and Y := {β1, . . . , βm}. See Figure 1 just below for a picture in the case m = 2, n = 3. We refer the reader to [9] and [7, Example 9.3] for more information on this example. v u Figure 1. The separated graph (E(2, 3), C(2, 3)) The C-algebras Om,n and Or m,n studied in [9] are precisely the C-algebras O(E(m, n), C(m, n)) and Or(E(m, n), C(m, n)), respectively, in the notation of the present paper. It was shown in [9] that these two C-algebras are not isomorphic, and we will now show that the C-algebra Or m,n is not simple if m ≥ 2. Let (E, C) := (E(m, n), C(m, n)). It is clear that H(E, C) only contains the trivial sets ∅, E0. Adopting the notation of [9], we write Ωu := Ω(E, C) = X u ⊔ Y u, where X u = ⊔n i=1H u homeomorphisms hu i , vu defining the universal (m, n)-dynamical system. Indeed, we have hu all i, j. j , and Y u is homeomorphic to each of the sets H u i and vh i = ⊔m j=1V u i : Y u → H u i , V u j by j are the universal maps i = θαi and vu j = θβj for j : Y u → V u j . The maps hu 34 PERE ARA AND MATIAS LOLK Let us now describe the separated graph (E1, C 1). We have E0,0 1 = {w} and E0,1 1 = {wij : 1 ≤ i ≤ n, 1 ≤ j ≤ m}. Now there are n + m elements in C 1 w, namely Xi := X(αi) and Yj := X(βj) for i = 1, . . . , n and j = 1, . . . , m. Note that Xi = m and Yj = n. Moreover, s(Xi) = {wij : j = 1, . . . , m} and s(Yj) = {wij : i = 1, . . . , n}. If m = n ≥ 2, then set H := {wij : i 6= j} . Then H is a maximal hereditary and C 1- 1. Moreover, (E1/H, C 1/H) consists of n cycles based at the vertex w, saturated subset of E0 so that Or n,n/I(H) ∼= Or(E1/H, C 1/H) ∼= Mn+1(C ∗ red(Fn)), which is a simple C-algebra. However On,n/I(H) ∼= C ∗(Fn) is not simple. (Incidentally, note that this gives another proof that Or n,n 6= On,n for n ≥ 2.) If 2 ≤ m < n, then set H := {wij : i 6= j and 1 ≤ i ≤ n, 1 ≤ j ≤ m − 1} ∪ {wim : 1 ≤ i ≤ m − 1}. Then H is again a maximal hereditary C 1-saturated subset of E0 1. However, in this case the quotient C-algebra Or m,n/I(H) is not even V-simple, as we will show in Section 8. It is therefore clear that the universal (E1/H, C 1/H)-system is not equivalent to the (m, n)-system (X, Y ) considered in the proof of [9, Proposition 3.9]. Indeed, observe that v−1 i ◦ hi = idY for i = 1, . . . , m − 1 in that example, and this is not necessarily true in a (E1/H, C 1/H)-system. The (m, n)-dynamical system just mentioned shows that the algebra Mm+1(On−m+1) ∼= Mn+1(On−m+1) is a simple quotient of Om,n, where Ok denotes the usual Cuntz algebra. Indeed, we can define a surjective ∗-homomorphism Om,n → Mm+1(On−m+1) by αi 7→(cid:26) 1 ⊗ ei+1,1 si−m+1 ⊗ em+1,1 if i = 1, . . . , m − 1 if i = m, . . . , n , βj 7→ 1 ⊗ ej+1,1 for j = 1, . . . , m, w 7→ 1 ⊗ e1,1 and v 7→Pm+1 algebra Mm+1(On−m+1) appears as a simple quotient of the reduced tame C-algebra Or j=2 1 ⊗ ej,j. However, it is not clear to the authors whether the same m,n. We now present an example relating our theory with classical symbolic dynamics; hopefully, it can also serve as an exemplification of the general theory of the previous sections. We only consider the case where the alphabet is {0, 1}, but similar statements can be made for an arbitrary finite alphabet, considering a corresponding variation of the separated graph considered below. Example 6.7. Let (E, C) be the separated graph described in Figure 2, with Cv = {X, Y } and X = {α0, α1} and Y = {β0, β1}. IDEAL STRUCTURE 35 v β0 β1 α0 α1 0 1 Figure 2. The separated graph underlying the lamplighter group This example has been considered in [7, Example 9.7], where it is observed that vO(F, D)v ∼= C(X ) ⋊σ Z, vLab K (F, D)v ∼= CK(X ) ⋊σ Z, 0 ⊔ Y ′ where X = {0, 1}Z and σ is the usual shift homeomorphism on X . Note that O(E, C) = Or(E, C) in this case. Indeed it can be easily seen that (E, C)-dynamical systems are in one- to-one correspondence with usual dynamical systems (Y, ϕ), where ϕ is a homeomorphism of the compact Hausdorff space Y, with the additional information of a partition Y = Y0 ⊔ Y1 of Y into clopen subsets Y0, Y1. To obtain the corresponding (E, C)-system, take Ω := Y ′ ⊔ Y, where Y ′ = Y ′ i correspond to the vertices labeled by i, the maps αi correspond to the identification of elements of Y ′ i with elements of Yi, and the maps βi are induced by the homeomorphism ϕ. It is easily checked that v(C(Ω) ⋊ F)v ∼= C(Y) ⋊ϕ Z in this situation. The unique equivariant continuous map Ω → Ω(E, C) predicted by [7, Corollary 6.11], restricted to Y, is the fundamental map in symbolic dynamics (see e.g. [37, §6.5]) sending each element x in Y to the sequence (an)n∈Z recording to which of the sets Y0 or Y1 belongs ϕn(x), that is, an = i ⇐⇒ ϕn(x) ∈ Yi, for n ∈ Z. 1 is a disjoint copy of Y, and where Ωv := Y, Y ′ We here give a dynamical interpretation in terms of the associated canonical sequence {(En, C n)} of bipartite separated graphs. Let X = {0, 1}Z as above. For a finite word a1a2 · · · an ∈ {0, 1}n, we will write [a1a2 · · · ai−1aiai+1 · · · an] := {x ∈ X : xj−i = aj for j = 1, 2, . . . , n}. The space Ω(E, C) is of the form Ω(E, C) = X ′ ⊔ X , where X ′ ∼= X = {0, 1}Z. By a slight abuse of language, we will identify X ′ with X notationally, so that we will obtain two partitions of the space {0, 1}Z into clopen subsets for each separated graph (En, C n), and E0,1 corresponding to the sets of vertices E0,0 respectively, and maps between these n n clopen subsets corresponding to the edges E1 n. The first layer of the sequence {(En, C n)}n≥0 corresponds to a trivial decomposition X 0 = X and to the decomposition X 1 1 = [1]. The maps corresponding to the edges are the maps αi : X 1 i → X defined by αi = idX 1 . We describe now the clopen sets corresponding to the separated graph (En, C n) for any n ≥ 1. Let w = a1a2 · · · an ∈ {0, 1}n be a word of length n. If n = 2m i → X and βi : X 1 and βi = σX 1 X 1 0 = [0], i i 36 PERE ARA AND MATIAS LOLK is even, define If n = 2m + 1 is odd, define X n w := [a1 · · · amam+1am+2 · · · a2m]. X n n has exactly 2n elements, and thus F 0 w := [a1 · · · amam+1am+2 · · · a2m+1]. n+1 = E0,1 n has exactly 2n+1 elements. The set F 0 n = E0,0 The vertices in E0,0 n correspond to a decomposition X = Gw∈{0,1}n X n w . Let n = 2m and take w ∈ {0, 1}n. Then C n 0w , βn+1 1w }. The edges αn+1 wi w0 , αn+1 w1 } correspond to the maps, denoted in the same way, w , where each is simply the identity on the respective domain. Similarly, w given by the restriction of σ to the 2 }, where X w 1 = {αn+1 iw : X n+1 iw → X n iw correspond to the maps βn+1 w = {X w 1 , X w 2 = {βn+1 : X n+1 and X w αn+1 wi the edges βn+1 respective domains. wi → X n Now let n = 2m + 1 and take w ∈ {0, 1}n. In this case, we have C n 1 = {αn+1 w1 }. The edges αn+1 1w } and X w 0w , αn+1 2 }, where iw again correspond to the maps, w , acting as the identity on the respective domains, w given by the restriction of wi → X n : X n+1 w = {X w 1 , X w wi X w denoted in the same way, αn+1 while the edges βn+1 σ−1 to the respective domains. wi 2 = {βn+1 iw : X n+1 w0 , βn+1 iw → X n correspond to the maps βn+1 Now it is quite easy to observe that Ω(E, C)v have Ω(E, C)v sending awb to w, where w ∈ {0, 1}2j−2 and a, b ∈ {0, 1}. ∞ , r2j), and in this case, the maps r2j : F 0,2j ∼= lim←−(F 0,2j ∼= X canonically. Indeed, by [8, p. 3008], we are the maps ∞ → F 0,2j−2 ∞ We can now relate the ideals of O(E, C) with some sets of words, and the quotients of O(E, C) with the subshifts of the shift (X , σ). Recall from [37, Chapter 1] that a subshift of X = {0, 1}Z is a subspace XF which can be described as the set of all elements x in X not containing any block from a fixed family F of finite words in the alphabet {0, 1}. (A block of x is a finite subsequence of consecutive terms in x.) The family F is called the family of forbidden words of the subshift. By [37, Theorem 6.1.21], the subshifts of X are exactly the invariant closed subsets of X . A subshift Z is said to be of finite type if there exists a finite set F such that Z = XF . Proposition 6.8. Let (E, C) be the separated graph described in Example 6.7, and adopt the notation used there. (1) For each subset F of words, there is a unique hereditary D∞-saturated subset HF of the separated graph (F∞, D∞) such that F and HF generate the same subshift, that is XF = XHF . (2) XF is a shift of finite type if and only if there is some n < ∞ such that HF is generated n. In this case, θ(En/H,C n/H) and σ are Kakutani equivalent, so in by H := HF ∩ E0 particular there are Morita-equivalences CK(XF ) ⋊σ Z ∼ Lab K (En/H, C n/H) and C(XF ) ×σ Z ∼ O(En/H, C n/H). IDEAL STRUCTURE 37 Proof. (1): By Example 6.7, the set of vertices of the graph F∞ can be identified with the set of all finite words in the alphabet {0, 1}. Given a set F of words, the space XF is a closed invariant subset of X . One can easily show that X \ XF = ∞[n=0 [w∈Wn[j∈Z σj(X n w ), w are the subsets of X defined in Example 6.7. Observe that the set W :=S∞ where for each n ≥ 0, the set Wn is the set of words of length n containing a block coming from F , and X n n=0 Wn is precisely the hereditary closure of F in the graph F∞. The set HF is the D∞-saturation of W , and it generates the same open invariant subset as F and as W . We therefore have XF = XHF . The uniqueness of HF comes from Theorem 4.7. (2): The first statement follows immediately from (1). Using the above identification of θ(En,C n), we see that Fv∈E0,0∩H Ω(En, C n)v = XF , so that the restriction of θ(En/H,C n/H) to the full, clopen subspace Gv∈E0,0 n \H Ω(En/H, C n/H)v is directly dynamically equivalent to σ via the group homomorphism F((En/H)1) → Z given by • αn+1 7→ 0 and βn+1 iw 7→ 1 if n is even, wi • αn+1 iw 7→ 0 and βn+1 7→ −1 if n is odd. wi In particular, θ(En/H,C n/H) and σ are Kakutani equivalent, so we obtain Morita equivalences as above. This concludes the proof. (cid:3) Example 6.9. The cycles in the different graphs (En, C n) are determined by the complements of some hereditary C n-saturated subsets of E0 n. They correspond to periodic orbits in the shift (X , σ). For instance, consider the word w = 0110, which is primitive, that is, it is not a square. The successive rotations of w are the words Consider the separated graph (E3, C 3), and the C 3-saturation H of the set w1 = w, w2 = 1100, w3 = 1001, w4 = 0011. It is easily seen that E0,1 E0,1 3 \ {w1, w2, w3, w4}. 3 \ H = {w1, w2, w3, w4}, and E0,0 v3 = 001, v1 = 110, v2 = 100, 3 \ H = {v1, v2, v3, v4}, where v4 = 011. The separated graph (E3/H, C 3/H) is essentially given by a cycle of length 4. We have O(E, C)/I(H∞) ∼= O(E3/H, C 3/H) ∼= M8(C(T)). Note that if we start with a word which is not primitive, for instance w = 0101, then there is a non-trivial C 2-saturation H2 induced by H in (E2, C 2), and there is a non-trivial C 1-saturation H1 induced by H2 in (E1, C 1), which is such that E0,1 1 \ H1 = {10, 01} (that is, H1 = {00, 11}), and everything is reduced to the 2-cycle generated by 01. 38 PERE ARA AND MATIAS LOLK The periodic orbits give trivial examples of minimal subshifts. More interesting examples are provided by the minimal Cantor subshifts, whose underlying space is a Cantor set. By [49], every strong orbit equivalence class of minimal Cantor systems contains a minimal Cantor subshift. By combining this with [30, Theorem 2.1], we see that, for every ∗-isomorphism class of C ∗-algebra crossed products C(Ω) ⋊ Z of minimal actions on the Cantor set Ω, there is a representative coming from a minimal subshift. These C ∗-algebras are classified by their ordered Grothendieck groups with distinguished order-units, which may be any simple dimension group with order-unit (see [30, Theorems 1.12 and 1.14]). See also [23] and [33] for further information about the conjugacy classes of minimal subshifts. We remark that these examples imply that, in spite of the results in [38], the theory of separated graph C- algebras leads to non-trivial examples (i.e. not coming from ordinary directed graphs) of simple nuclear C-algebras with stable rank one and real rank zero, since it is well-known that the C-algebras associated to minimal Cantor systems enjoy these properties (see [42] and [24, p. 184]). By the results of the second-named author ([38]) this is not possible for the tame C-algebra of a separated graph, but we see now that, factoring out a suitable maximal ideal generated by projections of the algebra O(E, C) appearing in Example 6.8, we may obtain such examples. The following example appears for instance in [22]. Example 6.10 (The even shift). Consider the subshift Y of (X , σ) defined by taking as a set of forbidden words F = {012n+10 n ≥ 0}. That is, in a word of Y there is always In this case, we have H (2) = {010}, an even number of consecutive 1's between two 0's. H (4) = {01110} ∪ H ′ 4 is the family of words of length 4 or 5 containing as subwords the word 010. In general 4, where H ′ H (2i) = {012i−10} ∪ H ′ 2i, where H ′ 012j−10 with j < i. This gives rise to a subshift which is not finite. 2i is the set of words of length 2i or 2i + 1 containing some subword of the form 7. A complete description of the ideals of finite type In this section, we completely determine the structure of the finite type ideals of Or(E, C): Definition 7.1. Let (E, C) denote a finite bipartite separated graph and let (F∞, D∞) be the separated Bratteli diagram of (E, C). Given an arbitrary ideal J ⊳ Or(E, C), there is some HJ ∈ H(F∞, D∞) for which I(HJ) = (J ∩ C(Ω(E, C))) ⋊r F. We will say that any H ∈ H(F∞, D∞) is of finite type if H = H n for some n ≥ 0, and an ideal J is of finite type if HJ is so. Finally, the lattices of finite type vertex sets and ideals will be denoted by Hfin(F∞, D∞) and Ifin(Or(E, C)), respectively. Given any partial action θ : G y Ω and a point x ∈ Ω, the stabiliser of x is the subgroup Stab(x) := {g ∈ G x ∈ Ωg−1 and θg(x) = x}. Recall that θ is called topologically free if, given any open subspace U and any 1 6= g ∈ G, there exists x ∈ U with g /∈ Stab(x). If θ is topologically free, then C0(Ω) ⋊r G has the IDEAL STRUCTURE 39 intersection property by [25, Theorem 29.5], i.e. any non-zero ideal J ⊳ C0(Ω) ⋊r G has non-trivial intersection C0(U) = J ∩ C0(Ω) 6= {0}. It follows that J contains the non-zero induced ideal I := C0(U) ⋊r G, but it does not tell us anything about the quotient J/I. The problem arises when the restriction of θ to Z := Ω \ U is not topologically free, and a partial action is said to be essentially free if all such restrictions are topologically free. However, while topological freeness of θ(E,C) is quite frequent (we recall the characterisation given in [7] just below), essential freeness is extremely rare as shown in [38]. In this section, we will introduce a weakening of topological freeness that still allows one to obtain information about the ideals, and show that it is always enjoyed by θ(E,C). In particular, the restriction θ(E,C)Z to any closed invariant subspace Z of finite type will also have this property. From these observations, we can completely characterize the structure of finite type ideals of Or(E, C). It is also worth noting that our methods, when applied to non-separated graphs, yield a complete description of the ideals of the graph C ∗-algebra. We first recall Condition (L) of [7], using a slightly different terminology: Definition 7.2. Consider any finite bipartite separated graph (E, C). A vertex v ∈ E0 is said to admit a choice if there exists an admissible path β = eα with s(β) = v, and an element Xe 6= X ∈ Cr(e) with X ≥ 2. The graph is said to satisfy Condition (L) if for every simple cycle σ, the base vertex s(σ) admits a choice. Theorem 7.3 ([7, Theorem 10.5]). Let (E, C) denote a finite, bipartite separated graph. Then θ(E,C) is topologically free if and only if (E, C) satisfies Condition (L). We now introduce the appropriate weakening of topological freeness. Definition 7.4. Let θ : G y Ω be any partial action of a discrete group on a locally compact Hausdorff space. For any x ∈ Ω, we shall say that θ is • topologically free in x if, given any 1 6= g ∈ Stab(x) and an open neighbourhood U of x, there exists y ∈ U with g /∈ Stab(y). • strongly topologically free in x if, given any 1 6= g1, . . . , gn ∈ Stab(x) and any open neighbourhood U of x, there exists y ∈ U with g1, . . . , gn /∈ Stab(y). We will denote by ΩTF the set of points x ∈ Ω in which θ is topologically free. If θ is strongly topologically free in every x ∈ ΩTF, it is said to be relatively strongly topologically free. Observe that if x is an interior point of ΩTF, then θ is automatically strongly topologically free in x. In particular, topologically free partial actions are strongly topologically free in all points. Remark 7.5. We will now expand a bit on the situation for θ(E,C), and we first borrow a bit of graph theory from [38, Section 3]. If v ∈ E0 does not admit a choice, every closed path α based at v has a unique word decomposition α = γ−1βγ for a (possibly trivial) admissible path γ and a cycle β, and we will say that α is a simple closed path if γ does not repeat a vertex, and β is a simple cycle. The set Fv := {closed paths based at v} ∪ {1} 40 PERE ARA AND MATIAS LOLK defines a subgroup Fv ≤ F, and every closed path based at v is a reduced product of simple closed paths. It follows that such v admits a unique simple closed path (up to inversion) ∼= Z. Moreover, Ω(E, C)v = {ξ} is a one-point set and Stab(ξ) = Fv, so if and only if Fv ξ /∈ Ω(E, C)TF if and only if v admits a closed path. It follows from the proof of [7, Theorem 10.5] that every η ∈ Ω(E, C) \ Ω(E, C)TF is of the form η = θγ(ξ) for some γ ∈ ξ with ξ as above -- in particular, the complement Ω(E, C) \ Ω(E, C)TF is discrete. Before we progress any further, let us consider a somewhat trivial, but not uninteresting, example. Example 7.6. If θ : G y Ω is any partial action, and x ∈ ΩTF satisfies Stab(x) ∼= Z, then θ is automatically strongly topologically free in x. Indeed, given 1 6= g1, . . . , gn ∈ Stab(x), we may write gi = gki with g a generator of Stab(x), and we can safely assume that ki > 0 for all i. For any open neighbourhood U of x, we consider h := gk1···kn ∈ G and pick y ∈ U with h /∈ Stab(y) using topological freeness in x. Then we obviously have gi /∈ Stab(y) as well, so θ is indeed strongly topologically free in x. The main result of [39] is a characterisation of nuclearity and exactness of O(r)(E, C) in terms of a graph theoretic Condition (N ). Another equivalent condition is that every stabiliser of θ(E,C) is either trivial or isomorphic to Z, so from the above example we obtain the following proposition: Proposition 7.7. If (E, C) satisfies Condition (N ), then the restriction of θ(E,C) to any closed invariant subspace is relatively strongly topologically free. For general separated graphs, we can only handle restrictions to closed invariant subspaces of finite type. Proposition 7.8. If (E, C) is any finite bipartite separated graph, then θ(E,C) is relatively strongly topologically free. Proof. Whenever we write βα for admissible paths α and β in the following, we shall mean the concatenated product, that is, we do not allow for cancellation. Now consider any ξ ∈ Ω(E, C)TF and g1, . . . , gn ∈ Stab(ξ), assuming, without loss of generality, that no pair gi, gj of distinct elements generate a rank one subgroup. For every i, we may write gi = µ−1 i σiµi for cycles σi. Since the action is topologically free in ξ, there exist admissible paths βi and edges xi with Xxi ≥ 2, such that x−1 i must be admissible, and without loss of generality, we can assume the former. Now given any open neighbourhood U of ξ, we have i βiµi ∈ ξ. Either x−1 i βiσi or x−1 i βiσ−1 ξ ∈ Ω(E, C)B := {η ∈ Ω(E, C) B ⊂ η} ⊂ U for a sufficiently big ball B := ξN . We may of course assume that N > x−1 Picking li ≥ 1 such that σli i µi ≥ N and xi 6= yi ∈ Xxi, we then consider the set i βiµi for all i. ω := B ∪ {y−1 i βiσli i µi i = 1, . . . , n}. IDEAL STRUCTURE 41 It should be clear that there exists η ∈ Ω(E, C) with ω ⊂ η, so in particular η ∈ U. Finally, observe that gi /∈ Stab(η) by construction since x−1 i σ−li i βiµi ∈ B ⊂ η and x−1 i βiµi /∈ η · µ−1 (η), i µi = θg li i so θ is indeed strongly topologically free in ξ. (cid:3) Now let us instead consider a non-example. Example 7.9. For n ∈ Z, we define f 1 n, f 2 n ∈ {0, 1}Z2 by n(a, b) =(cid:26) 0 if b > n 1 if b ≤ n f 1 and f 2 n(a, b) =(cid:26) 0 if a > n 1 if a ≤ n . n, f 2 n, 0, 1 n ∈ Z}. Every f i Then the Z2-shift on {0, 1} restricts to an action θ : Z2 y Ω on the compact Hausdorff space n → 0 for n → −∞ and f i Ω := {f 1 n → 1 for n) = 0 ⊕ Z for all n, so ΩTF = {0, 1}. n → ∞. By construction, Stab(f 1 However, θ is not strongly topologically free in 0 or 1 since any f i n is either fixed by a or b. In conclusion, relative strong topological freeness is not automatic. n is isolated with f i n) = Z ⊕ 0 and Stab(f 2 Remark 7.10. We have no examples of partial actions of free groups that are not relatively strongly topologically free, but we suspect that such examples exist. However, it is notable that whenever a partial action θ : F y Ω of a free group is topologically free in x, and we consider only two elements g1, g2 ∈ Stab(x), then given any open neighbourhood U of x we can find y ∈ U with g1, g2 /∈ Stab(y). Indeed, we may assume that g1 and g2 do not generate a free subgroup of rank one, so that the commutator [g1, g2] = g−1 2 g1g2 is non-trivial. Let Vi be an open neighbourhood of x so that θgi(Vi) ⊂ U and consider V := U ∩ V1 ∩ V2. Applying topological freeness with respect to [g1, g2], we obtain y ∈ V with [g1, g2] /∈ Stab(y). Obviously, we cannot have g1, g2 ∈ Stab(y), and if g1, g2 /∈ Stab(y), then we are done. We may therefore assume that exactly one of g1 and g2 belongs to Stab(y). If g1 ∈ Stab(y), we consider y 6= y′ := θg2(y) ∈ U instead. Observe that if g1 ∈ Stab(y′), then 1 g−1 θ[g1,g2](y) = θg−1 2 g1(y′) = θg−1 1 g−1 1 g−1 2 (y′) = θg−1 1 (y) = y, which contradicts our choice of y. We conclude that g1, g2 /∈ Stab(y′) as desired. In case g2 ∈ Stab(y), we apply the exact same argument with g1 and g2 reversed: this is possible since [g2, g1] = [g1, g2]−1 /∈ Stab(y). It is evident from the definition that Ω \ ΩTF is an open invariant subspace, so there is a corresponding "obstruction ideal": Definition 7.11. Given a partial action θ : G y Ω of a discrete group on a locally compact Hausdorff space, we define an ideal J o = J o θ := C0(Ω \ ΩTF) ⋊r G of C0(Ω) ⋊r G. We now put relative strong topological freeness to work. The proof of the following theorem is modelled over that of [25, Theorem 29.5], but the statement is somewhat more general. 42 PERE ARA AND MATIAS LOLK Theorem 7.12. Let θ : G y Ω denote a partial action of a discrete group G on a locally compact Hausdorff space. Suppose that G is exact and that θ is relatively strongly topologically free. If J ∩ C0(Ω) = {0} for an ideal J ⊳ C0(Ω) ⋊r G, then J ⊂ J o. Proof. Assuming that J 6⊂ J o, we take any a ∈ J \ J o and consider w := a∗a ∈ J \ J o. Now consider the commutative diagram C0(Ω) ⋊r G E C0(Ω) p∗ p C0(ΩTF) ⋊r G F C0(ΩTF) , where E and F are the canonical conditional expectations. By exactness of G, we have ker(p∗) = J o, hence p∗(w) 6= 0. Faithfulness of F then implies p(E(w)) = F (p∗(w)) 6= 0, so f := E(w) attains a non-zero value on ΩTF. Let x0 ∈ ΩTF be such that f (x0) = supx∈ΩTF f (x) and take any 0 < ε < f (x0) . By Urysohn's Lemma (applied to the one point compactification of Ω), there exists u ∈ C0(Ω) with 0 ≤ u ≤ 1, u(x0) = 1 and u(x) = 0 whenever f (x) ≥ f (x0)+ε/4. Consider z := uw ∈ J\J o and note that E(z) = uE(w) = uf , hence f (x0) ≤ kE(z)k ≤ f (x0) + ε/4. We claim that there exists a function h ∈ C0(Ω) such that 2 (1) 0 ≤ h ≤ 1, (2) khE(z)hk > kE(z)k − ε, (3) khzh − hE(z)hk < ε. To see this, we first pick b ∈ C0(Ω)⋊alg G with kz −bk < ε/4 and write b = b1 +Pg∈T bgδg for a finite set T ⊂ G\{1}. If T = ∅, then the claim is easily verified, so we may assume that T 6= ∅. Since θ is strongly topologically free in x0, we can find x1 ∈ Ω with b1(x1) − b1(x0) < ε/4 and T ∩ Stab(x1) = ∅. We may apply [25, Lemma 29.4] for every g ∈ T to obtain hg ∈ C0(Ω) with 0 ≤ hg ≤ 1, hg(x1) = 1 and khg(bgδg)hgk < ε from the calculation 2T . Setting h :=Qg∈T hg, (2) then follows khE(z)hk > khE(b)hk − ε/4 = khb1hk − ε/4 ≥ h(x1)b1(x1)h(x1) − ε/4 = b1(x1) − ε/4 > b1(x0) − ε/2 > E(z)(x0) − 3ε/4 = f (x0) − 3ε/4 ≥ kE(z)k − ε. In order to check (3), we first observe that khbh − hb1hk = kXg∈T h(bgδg)hk ≤Xg∈T khg(bgδg)hgk < ε/2, IDEAL STRUCTURE 43 hence khzh − hE(z)hk ≤ khzh − hbhk + khbh − hb1hk + khb1h − hE(z)hk < ε. Having verified the claim, we let π denote the quotient map C0(Ω) ⋊r G → (C0(Ω) ⋊r G)/J. Since z ∈ J and J ∩ C0(Ω) = 0, we have kE(z)k < khE(z)hk + ε = kπ(hE(z)h − hzh)k + ε ≤ khE(z)h − hzhk + ε < 2ε. But at the same time, kE(z)k ≥ f (x0) > 2ε, a contradiction. (cid:3) We need to specialize the above theorem a bit before we can apply it to our setting. The following is an ever useful observation. Lemma 7.13. Let θ : G y Ω denote a partial action of a discrete group on a locally compact Hausdorff space. If x ∈ Ω is isolated, then where 1x denotes the indicator function in x. 1x(C0(Ω) ⋊(r) G)1x ∼= C ∗ (r)(Stab(x)), Proof. By [9, Proposition 6.1 and Corollary 6.3], we have embeddings (r)(Stab(x)) ∼= C({x}) ⋊(r) Stab(x) ֒→ C0(Ω) ⋊(r) Stab(x) ֒→ C0(Ω) ⋊(r) G, C ∗ and the composition clearly maps onto the corner 1x(C0(Ω) ⋊(r) G)1x. (cid:3) Definition 7.14. Let θ : G y Ω denote a partial action on a locally compact Hausdorff space, and suppose that U ⊂ OG(Ω) is a collection of open invariant subsets U ⊂ Ω for which the restriction θZU to ZU := Ω \ U satisfies the following two conditions: (1) θZU is relatively strongly topologically free, (2) the space WU := ZU \ Z TF U is discrete. Observe that if U ⊂ V for U, V ∈ U, then WU \ V ⊂ WV . For any U, we may therefore choose a set of representatives ΛU for the orbit space WU /G such that ΛU \ V ⊂ ΛV whenever U ⊂ V . We then introduce a set U is a proper ideal of C ∗ r (Stab(x))o U ⊂ I x V for all x ∈ ΛU \ V. and equip it with the partial order U )x∈ΛU(cid:1) U ∈ U, I x IU (θ) :=n(cid:0)U, (I x (cid:0)U, (I x U )x∈ΛU(cid:1) ≤(cid:0)V, (I x V )x∈ΛV(cid:1) ⇔ U ⊂ V and I x For notational simplicity, we will usually write I • U )x∈ΛU , and we finally write IU (C0(Ω)⋊r G) for the collection of ideals J ⊳ C0(Ω) ⋊r G satisfying J ∩ C0(Ω) = C0(U) for some U ∈ U. U = (I x Corollary 7.15. Let θ : G y Ω denote a partial action of an exact group on a locally com- pact Hausdorff space with the setup from Definition 7.14. Then there is a canonical order isomorphism with the following properties: IU (θ) → IU (C0(Ω) ⋊r G), (U, I • U ) 7→ J(U, I • U ), 44 PERE ARA AND MATIAS LOLK (1) J(U, I • (2) The quotient J(U, I • U ) ∩ C0(Ω) = C0(U). Proof. Let us write J o U := J o is a canonical identification θZU U )/(C0(U) ⋊r G) is canonically Morita equivalent to Lx∈ΛU . By Definition 7.14(2), every orbit in WU is clopen, so there I x U . J o U ∼= Mx∈ΛU C0(G.x) ⋊r G. U denote the ideal generated by I x r (Stab(x)) sits as a full corner of C0(G.x) ⋊r G for all x ∈ ΛU , and we U inside C0(G.x) ⋊r G. We may then define an ideal U , and letting πU denote the quotient map C0(Ω) ⋊r G → C0(ZU ) ⋊r G, U ) := π−1 U ) 7→ U ( IU ). We proceed to verify the properties of the map (U, I • From Lemma 7.13, C ∗ let I x IU :=Lx∈ΛU I x U ⊳ J o we finally set J(U, I • J(U, I • U ), assuming that U ⊂ V . Observe that the quotient map πU,V : C0(ZU ) ⋊r G → C0(ZV ) ⋊r G V otherwise. It follows that (U, I • r (G.x) ⋊r G to {0} if x ∈ V , and that it maps C0(G.x) ⋊r G ⊂ J o maps C ∗ U identically to C0(G.x) ⋊r G ⊂ J o U ) ⊂ J(V, I • V ). Both (1) and (2) should be obvious from the above, and injectivity is immediate from these. To see that our map is surjective, take any ideal J ⊳ C0(Ω)⋊r G with J ∩C0(Ω) = C0(U) for U ∈ U, and consider JU := J/(C0(U)⋊r G) ⊳ C0(ZU )⋊r G. Since θZU is assumed relatively strongly topologically free, we see that JU ⊂ J o U from Theorem 7.12. In particular, we may write JU U must be proper, for otherwise we would have x ∈ U. Setting I U (cid:3) U ⊳ C0(G.x) ⋊r G. Observe that I x V ) if and only if J(U, I • ∼= Lx∈ΛU I x U for ideals I x U )1x, we finally have J(U, I • x := 1x( I x U ) ≤ (V, I • U ) = J. We are almost ready to put our results to use at this point, but we still need to introduce some bookkeeping. Construction 7.16. Consider any finite bipartite separated graph (E, C), and let (F∞, D∞) be its Bratteli diagram. We denote by C(E, C) the set of vertices v ∈ E0 which admit no choices and a simple cycle α such that every closed path based at v is a power of α, modulo the relation u ∼ v ⇔ u and v belong to the same cycle. It follows from Remark 7.5 that C(E, C) is in canonical bijective correspondence with the orbit space W/F for W := {ξ ∈ Ω(E, C) \ Ω(E, C)TF Stab(ξ) ∼= Z}. Observe that there is a natural map C(E, C) → C(E1, C 1) given by [v] 7→ [v] whenever v ∈ E0,1, and that it is in fact a bijection. This can either be seen by direct arguments, by Theorem 3.22, or most easily by [38, Lemma 5.2]. Consequently, if H ∈ Hfin(F∞, D∞) satisfies H = H n, we may identify C(En/H (n), C n/H (n)) with C(Em/H (m), C m/H (m)) whenever m ≥ n; formally, we do this by setting C(H) := lim−→ m≥n C(Em/H (m), C m/H (m)). IDEAL STRUCTURE 45 Observe that whenever H1 ⊂ H2 for H1, H2 ∈ Hfin(F∞, D∞), we have an inclusion {c ∈ C(H1) c 6⊂ H2} ⊂ C(H2). Indeed, whenever m ≥ n for sufficiently large n, we have representations C(H1) = C(Em/H (m) 1 , C m/H (m) 1 ) and C(H1) = C(Em/H (m) 2 , C m/H (m) 2 ), and H (m) 1 ⊂ H (m) 2 . Consequently, there in an inclusion {c ∈ C(Em/H (m) 1 , C m/H (m) 1 ) c 6⊂ H (m) 2 } ⊂ C(Em/H (m) 2 , C m/H (m) 2 ), and this does not depend on m. Letting Op(T) denote the collection of proper open subsets of T, we finally define a set Ifin(E, C) =(cid:8)(H, T ) H ∈ Hfin(F∞, D∞), T ∈ Op(T)C(H)(cid:9) and equip it with the partial ordering (H1, T1) ≤ (H2, T2) ⇔ H1 ⊂ H2 and T1(c) ⊂ T2(c) for all c ∈ C(H1) with c 6⊂ H2. Theorem 7.17. For any finite bipartite separated graph (E, C), there is a canonical lattice isomorphism Ifin(E, C) → Ifin(Or(E, C)) , (H, T ) 7→ I(H, T ), with the following properties: (1) HI(H,T ) = H. (2) The quotient I(H, T )/I(H) is Morita equivalent to Lc∈C(H) C0(T (c)). In particular, a finite type ideal of Or(E, C) is generated by its projections if and only if it is induced. Proof. Let U := {Ω(E, C)H H ∈ Hfin(F∞, D∞)} and observe that Corollary 7.15 may be applied thanks to Remark 7.5, Proposition 7.8 and Theorem 7.12. Since Fn is C ∗-simple for r (Z) ∼= C(T) all n ≥ 2 [43], we see that I x correspond to proper open subsets T ⊂ T. Finally, if H = H n and U = Ω(E, C)H, then the orbits of points ξ ∈ WU with stabiliser Z correspond canonically to the elements of C(H). The above statement is therefore exactly the conclusion that can be drawn from Corollary 7.15. (cid:3) U = 0 whenever Stab(x) 6∼= Z, and proper ideals of C ∗ Remark 7.18. We have no hope of achieving a similar result for arbitrary ideals of Or(E, C) except in special cases. Indeed, we suspect that for an infinite type subspace Z, the restriction θ(E,C)Z : F y Z need not be relatively strongly topologically free, and one can easily find examples where the space Z \ Z TF is not discrete. We finally apply our results to classical graph C ∗-algebras to provide a new proof for the description of the ideal lattice first obtained by Hong and Szyma´nski in [32]. We first recall a bit of terminology and a few results. 46 PERE ARA AND MATIAS LOLK Definition 7.19 ([16]). Let E denote any directed graph and denote the set of hereditary and saturated subsets by H(E). For any H ∈ H(E), there is a set of breaking vertices for H given by H fin ∞ := {v ∈ E0 \ H : r−1(v) = ∞ and 0 < r−1(v) ∩ s−1(E0 \ H) < ∞}, and pairs (H, B) with B ⊂ H fin quotient graph E/(H, B) as ∞ are called admissible. For any such pair, one defines a (E/(H, B))0 := (E0 \ H) ∪ {β(v) v ∈ H fin (E/(H, B))1 := s−1(E0 \ H) ∪ {β(e) e ∈ E1, s(e) ∈ H fin ∞ \ B}, ∞ \ B} with r, s extended by r(β(e)) := r(e) and s(β(e)) := β(s(e)). We finally order the admissible pairs (H, B) by (H1, B1) ≤ (H2, B2) ⇔ H1 ⊂ H2 and B1 \ B2 ⊂ H2. Theorem 7.20. Let E denote any directed graph. There exists a partial action θE : F y ∂E of F := F(E1) on a totally disconnected, locally compact Hausdorff space ∂E with the following properties: ( 1) C ∗(E) ∼= C0(∂E) ⋊ F(E1) canonically. (2) There is a canonical lattice isomorphism {(H, B) H ∈ H(E) and B ⊂ H fin ∞ } → OF(∂E), (H, B) 7→ U(H, B), and θE∂E\U (H,B) ≈ θE/(H,B) for every admissible pair (H, B). In particular, C ∗(E)/I(H, B) ∼= C ∗(E/(H, B)), where I(H, B) is the ideal induced from U(H, B). Proof. (1) is proven in [25, Chapter 37], and (2) follows from the description of gauge-invariant ideals in [16] (one can also prove (2) directly with fairly little work). (cid:3) As for the separated graphs, we have to introduce some bookkeeping: Definition 7.21. For any directed graph E, we let C(E) denote the set of vertices v ∈ E0 which admit a cycle without any entries, modulo the relation u ∼ v ⇔ u and v belong to the same cycle. Observe that if (H, B) is an admissible pair, then C(E/(H, B)) = C(E/H) since the additional vertices in E/H are all sources. Given any H ∈ H(E), we set C(H) := C(E/H) and observe that if H1 ⊂ H2, then we have an inclusion We may in turn define a lattice {c ∈ C(H1) c 6⊂ H2} ⊂ C(H2). I(E) := {(H, B, T ) H ∈ H, B ⊂ H fin ∞ , T ∈ Op(T)C(H)} IDEAL STRUCTURE 47 ordered by (H1, B1, T1) ≤ (H2, B2, T2) ⇔(H1, B1) ≤ (H2, B2) and T1(c) ⊂ T2(c) for all c ∈ C(H1) with c 6⊂ H2. Finally, we denote the ideal lattice of C ∗(E) by I(C ∗(E)), and given any J ∈ I(C ∗(E)), we write (HJ , BJ ) for the admissible pair satisfying J ∩ C0(∂E) = C0(U(HJ , BJ )). Theorem 7.22. For any directed graph E, there is a canonical lattice isomorphism I(E) → I(C ∗(E)), (H, B, T ) 7→ I(H, B, T ), with the following properties: (1) HI(H,B,T ) = H and BI(H,B,T ) = B. (2) The quotient I(H, B, T )/I(H, B) is Morita equivalent to Lc∈C(H,B) C0(T (c)). Proof. Let U := OF(∂E) and observe that Corollary 7.15 applies due to Theorem 7.20 and Example 7.6, since any stabiliser is either trivial or isomorphic to Z. For any U = U(H, B) ∈ U, the orbits of points x ∈ WU with Stab(x) ∼= Z correspond to the elements of C(H), so Corollary 7.15 reduces to the above statement. (cid:3) Remark 7.23. Observe that Theorem 7.12 also provides a new proof of Szyma´nski's general Cuntz-Krieger Uniqueness Theorem [50] when applied to the boundary path space action θE. Remark 7.24. We finally remark that Theorem 7.12 has an analogue for algebraic crossed products CK(Ω) ⋊ G, where CK(Ω) denotes the algebra of compactly supported locally con- stant function Ω → K when K is given the discrete topology. The assumptions will have to be slightly different: on one hand, there is no need for exactness of the group, but on the other hand, one needs the space Ω to be totally disconnected to have sufficiently many continuous functions. The proof should be a bit simpler, although one will have to avoid C ∗-techniques. The description of the lattice of ideals of the Leavitt path algebra LK(E) of an arbitrary graph E, obtained in [3, Theorem 2.8.10], can also be obtained using this approach. 8. V-simplicity We continue our investigation of the ideal structure with the study of V-simplicity, that is, we want to compute the algebras of graphs (E, C) for which the monoid M(F∞, D∞) is order simple. By Theorem 4.7, this is equivalent to saying that I ∩C(Ω(E, C)) = 0 for every proper ideal I of Or(E, C), or H(F∞, D∞) = {∅, F 0 ∞}. We will simply say that (E, C) is simple in this case, and any of the algebras O(E, C), Or(E, C) and Lab(E, C) will be called V-simple. The second named author proves similar results in [38, Section 4] by studying minimality of the partial action θ(E,C); our study here, on the other hand, is purely graph-theoretical. We state the main result of this section right away: Theorem 8.1. Let (E, C) be a simple finite bipartite separated graph. Then LK(E, C) = Lab K (E, C) and C ∗(E, C) = O(E, C), and one of the following holds: 48 PERE ARA AND MATIAS LOLK (1) LK(E, C) is isomorphic to a simple Leavitt path algebra, and C ∗(E, C) is isomorphic to a simple graph C ∗-algebra of a non-separated graph. (2) LK(E, C) is Morita-equivalent to K[Fn] and C ∗(E, C) is Morita-equivalent to C ∗(Fn), where Fn is a free group of rank n with 1 ≤ n < ∞. r (Fn). In the latter case, Or(E, C) is Morita equivalent to C ∗ From Theorem 8.1 and the fact that the reduced group C-algebra C ∗ r (Fn) of a free group Fn of rank n > 1 is simple [43], we obtain: Corollary 8.2. Let (E, C) be a finite bipartite separated graph. If Or(E, C) is V-simple, then either Or(E, C) is isomorphic to a simple C ∗-algebra, or it is Morita equivalent to C(T). We develop the proof in various steps. We begin with a simple observation. Lemma 8.3. Let (E, C) be a finite bipartite separated graph. Then (E, C) is simple if and only if H(En, C n) = {∅, E0 n} for all n ≥ 0. Proof. Let n ≥ 0 be given. preserving map H(En, C n) → H(F∞, D∞), so H(En, C n) is trivial if H(F∞, D∞) is trivial. It follows from Lemma 5.3 that there is an injective order- Conversely assume that H(En, C n) is trivial for all n ≥ 0. H = ∪n=0H (n), where H (n) := H ∩ E0 H = F 0 ∞. Otherwise H (n) = ∅ for all n, so H = ∅. n ∈ H(En, C n) for all n. If H (n) = E0 If H ∈ H(F∞, D∞), then n for some n, then (cid:3) A useful property of the separated graphs (En, C n) associated to a finite bipartite separated v and n ≥ 1, then s(x) 6= s(y) whenever x, y are graph (E, C) is the following: different elements of X. This follows immediately from the definition of these graphs. if X ∈ C n We now obtain some necessary conditions for V-simplicity. Lemma 8.4. Let (E, C) be a simple finite bipartite separated graph. Then for all v ∈ F 0 ∞, there is at most one X ∈ D∞ v such that X > 1. Proof. It suffices to show the result for all v ∈ E0,0. Suppose there exist distinct X, Y in Cv such that X > 1 and Y > 1. Take a vertex w := v(x1, y1, z1, . . . , zt) ∈ E0,1 1 , where x1 ∈ X, y1 ∈ Y and zi ∈ Zi, where Cv = {X, Y, Z1, . . . , Zt}. The singleton {w} is then hereditary and C 1-saturated in E1, because the elements X of C 1 having edges that start at w are of the form X = X(x), where x ∈ {x1, y1, z1, . . . , zt}, and so they all have more than one element, and moreover the sources of the vertices of X are all different. So H(E1, C 1) is non-trivial, contradicting Lemma 8.3. (cid:3) At this point we can already show the coincidence between the universal graph algebras and their tame quotients. Corollary 8.5. If (E, C) is a simple finite bipartite separated graph, then the natural maps LK(E, C) → Lab K (E, C) and C ∗(E, C) → O(E, C) are isomorphisms. IDEAL STRUCTURE 49 Proof. Let n ≥ 0. By Lemma 8.4, we have [ee∗, f f ∗] = 0 for e, f ∈ E1 n. Therefore it follows from [7, Theorem 5.1(a)] that the maps LK(En, C n) → LK(En+1, C n+1) and C ∗(En, C n) → C ∗(En+1, C n+1) are isomorphisms. Since this holds for each n ≥ 0, we get LK(E, C) ∼= K (E, C) and C ∗(E, C) ∼= O(E, C). Lab (cid:3) Lemma 8.6. Let (E, C) be a simple finite bipartite separated graph. Then, for all w ∈ E0,1 n with n ≥ 1, there exists X ∈ C n such that X = 1 and s(X) = {w}. Proof. The only point where we use that n ≥ 1 is the property that all the source vertices of edges coming from the same set X ∈ C are distinct. Thus, it suffices to prove the result for an arbitrary finite bipartite separated graph (E, C) such that, for every X ∈ C, we have s(x) 6= s(y) whenever x, y are distinct elements of X. Using this condition, we get that if w ∈ E0,1 and X > 1 for all X ∈ C such that w ∈ s(X), then {w} is a non-trivial hereditary and C-saturated subset of (E, C), which contradicts our hypothesis. (cid:3) Definition 8.7. Let (E, C) be a simple finite bipartite separated graph. A vertex v ∈ E0,0 is said to be of type A in case there is a (unique) X ∈ Cv such that X > 1, and v is said to be of type B in case X = 1 for all X ∈ Cv. Note that, by Lemma 8.4, every vertex v ∈ E0,0 is either of type A or of type B. If v ∈ E0,0 is of type A, we denote by X v the unique element in Cv having more than one element. Lemma 8.8. Let (E, C) be a simple finite bipartite separated graph. Then for each w ∈ E0,1 there exists at most one X ∈ C such that X = 1, s(X) = {w}, and X ∈ Cv for a vertex v ∈ E0,0 of type A. Proof. Suppose that X, Y are distinct, X ∈ Cv, Y ∈ Cv′, for v, v′ vertices of type A, that X = Y = 1, and that s(X) = s(Y ) = {w}. Let X = {x} and Y = {y}. Then X(x) and X(y) are two distinct elements of C 1 w, and X(x) > 1, X(y) > 1 because there are X ′ ∈ Cv and Y ′ ∈ Cv′ with X ′ > 1 and Y ′ > 1. This contradicts Lemma 8.4. (cid:3) We say that a type B vertex v is of type B1 if, given any w ∈ E0,1, there is at most one X ∈ Cv such that X = 1 and s(X) = {w}. Say that v is type B2 in case v is not type B1. The following definitions apply to a general finite bipartite separated graph (E, C). Recall that, for e ∈ E1, we denote by Xe the unique element of C such that e ∈ Xe. Definition 8.9. Let (E, C) be a finite bipartite separated graph. An admissible path γ is said to a 1-path in case all the edges e ∈ E1 appearing in the path (with exponent ±1) satisfy that Xe = 1. Length zero paths are also considered 1-paths. Two vertices v, w of E are said to be 1-connected, denoted by v ∼ w, if there is a 1-path from v to w. A 1-cycle is a 1-path which is also a cycle. Remark 8.10. Observe that the relation ∼ on E0 is an equivalence relation. In fact one can easily show that if γ1, γ2 are 1-paths with r(γ1) = s(γ2), then the reduced product γ2 · γ1 (i.e. the path obtained after cancellation of terms ee−1 or e−1e in the concatenation of γ2 and γ1) is also a 1-path, which gives the transitivity of the relation ∼. It is obvious that ∼ is symmetric and reflexive. 50 PERE ARA AND MATIAS LOLK Lemma 8.11. Let (E, C) be a simple finite bipartite separated graph. The following hold: (a) Two distinct vertices of type A are not 1-connected. (b) The vertices of type A and the vertices of type B2 are not 1-connected. (c) Let v be a vertex of type A and let w a vertex of type B such that there is a 1-cycle based at w. Then v is not 1-connected to w. Proof. (a) Let v, v′ be two distinct vertices of type A. We show that v is not 1-connected to v′ by induction on the length of a minimal 1-path between v and v′. Suppose that there is a path e2e−1 from v to v′ such that Xei = 1 for i = 1, 2. Let w be the vertex in E0,1 such 1 that {w} = s(Xe1) = s(Xe2). Then X(e1) and X(e2) are two different elements of C 1 w having more than one element each, contradicting Lemma 8.4. Assume that there are no 1-paths of length ≤ 2(n − 1) between two vertices of type A, and let γ be a path of length 2n between two vertices v, v′ of type A. Write γ = e2ne−1 2n−1 · · · e2e−1 1 , where each Xei has only one element (namely ei). Then the vertices w, w′ of E1 such that {w} = s(Xe1) and {w′} = s(Xe2n) are of type A in E1, because v and v′ are of type A (in E), and moreover there is a 1-path in E1 from w to w′ of length 2(n − 1), leading to a contradiction. This shows that there is no 1-path of length ≤ 2n between distinct vertices of type A. (b) Let v be a vertex of type A and let v′ be a vertex of type B2. We show that v is not 1-connected to v′ by induction on the length of a minimal 1-path between v and v′. Suppose that there is a path e2e−1 from v to v′ such that Xei = 1 for i = 1, 2. Then in the 1 separated graph (E1, C 1), we have a vertex w with {w} = s(Xe1) = s(Xe2) and X, Y ∈ C 1 w, where X = X(e1) and Y = X(e2). If there is Z ∈ Cv′ such that Z 6= Xe2, Z = 1 and s(Z) = s(Xe2), then setting Z = {z}, we get X(z) 6= Y and s(X(z)) = s(Y ). Now let Y = {y′} and X(z) = {z′}. Note that, since v is of type A, we have X = X(e1) > 1. In conclusion, we have that C 1 w has at least three sets X, Y, X(z), with X > 1, Y = X(z) = 1 and s(Y ) = s(X(z)). Thus, in (E2, C 2) we have a vertex {w′} = s(Y ) = s(X(z)) with Cw′ containing two sets X(y′) and X(z′) with more than one element. This contradicts Lemma 8.4. Assume now that there is no Z ∈ Cv′ such that Z 6= Xe2, Z = 1 and s(Z) = s(Xe2). Since v′ is of type B2 by hypothesis, there are two distinct sets Z, T ∈ Cv′ such that Z = T = 1 and s(Z) = s(T ). Now the vertex w is type A in (E1, C 1), and the vertex w′ = s(Z) = s(T ) is type B2 in (E1, C 1). Indeed, if w′ is of type A, then there is an edge f from w′ to a vertex v′′ ∈ E0,0 of type A such that Xf = 1, and then we could apply the argument in the above paragraph to the 1-path f −1z, where Z = {z}, to arrive at a contradiction. Moreover there is a 1-path e′ 2}. Set T = {t}. Then X(z) = X(t) = 1 and s(X(z)) = s(X(t)) in (E1, C 1), so that we are in the same situation as before, but replacing (E, C) with (E1, C 1). Consequently, we arrive at a contradiction with the fact that H(E3, C 3) only contains the trivial subsets. 1)−1 from w to w′ such that X(e2) = {e′ 2(e′ 1} and X(z) = {e′ This shows the case where the length of the path is 2. If we have a path of length 2n, then it gives rise to a path of length 2n − 2 between a vertex of type A and a vertex of type B2 in the graph (E2, C 2), leading to a contradiction. (c) The proof is similar to the proof of (b), we leave the details to the reader. IDEAL STRUCTURE 51 (cid:3) In the following lemma, we describe the V-simple algebras corresponding to graphs with only vertices of type A. Lemma 8.12. Let (E, C) be a simple finite bipartite separated graph, and suppose that the source vertices of edges coming from the same set X ∈ C are all distinct. If all the vertices in E0,0 are of type A, then LK(E, C) = Lab K (E, C) is isomorphic to a Leavitt path algebra LK(E), and C ∗(E, C) = O(E, C) is isomorphic to a graph C-algebra C ∗(E) of a non-separated graph E with H(E) = {∅, E }. 0 − := E1 \ {yw w ∈ E0,1} and E1 Proof. By Lemma 8.6 and Lemma 8.8, for each w ∈ E0,1 there exists a unique Y ∈ C such that Y = 1 and s(Y ) = {w}. We denote by yw the unique edge in such a group Y . It is clear that E1 + := {yw w ∈ E0,1} defines a non-separated orientation in the sense of [38, Definition 3.8], so LK(E, C) ∼= LK(E) and C ∗(E, C) ∼= C ∗(E) by [38, Proposition 3.11], where E is the graph obtained from (E, C) by inverting all the edges of E1 +. Moreover, E contains no hereditary and saturated subsets since (E, C) contains no hereditary and C-saturated subsets. (cid:3) Lemma 8.13. Let (E, C) be a simple finite bipartite separated graph. Assume that v ∈ E0 is of type B, and that v is not 1-connected to a vertex of type A. Then v is a full projection and there are identifications vLab(E, C)v ∼= K[Fv], vO(E, C)v ∼= C ∗(Fv) and vOr(E, C)v ∼= C ∗ r (Fv), where Fv is the group of closed paths based at v. Proof. From (E, C) being simple, we see that v is full. We claim that Ω(E, C)v is in fact a one-point space. If it were not, then there would exist some admissible path eγ with s(γ) = v along with X ∈ Cr(e) satisfying X > 1 and X 6= Xe. We can of course assume that eγ is minimal with these properties. Assuming that γ passes through a type A vertex, we can write γ = γ2γ1, where γ1 is the maximal initial subpath for which r(γ1) is of type A. But then eγ2 is a 1-path between vertices of type A, contradicting Lemma 8.11. We deduce that eγ is itself a 1-path, so that v is 1-connected to a type A vertex, in conflict with our assumption. It follows that the partial action θ(E,C) restricts to the trivial global action Fv y Ω(E, C)v, hence vLab K (E, C)v ∼= CK(Ω(E, C)v) ⋊ Fv ∼= K ⋊ Fv ∼= K[Fv] and by Lemma 7.13. vO(r)(E, C)v ∼= C(Ω(E, C)v) ⋊(r) Fv ∼= C ⋊(r) Fv ∼= C ∗ (r)(Fv) (cid:3) We are now ready to prove Theorem 8.1. Proof of Theorem 8.1. Passing, if necessary, to (E1, C 1), we can assume that for each X ∈ C we have that s(x) 6= s(x′) whenever x, x′ are distinct elements of X. If E0,0 consists entirely of vertices of type A, then we can reach (1) by Lemma 8.12. If E0,0 contains both vertices 52 PERE ARA AND MATIAS LOLK of type A and of type B, and a vertex of type A is 1-connected to a vertex of type B, then its 1-connected component cannot contain any 1-cycle, by Lemma 8.11. In that case, the vertices of type B in it will vanish in the graph (En, C n) for some n. Indeed, let v be a vertex of type A and let γ be a non-trivial 1-path starting at v. By Lemma 8.11(a), all vertices in E0,0 visited by γ are of type B. If the 1-connected component of v does not contain 1-cycles, then all the vertices visited by γ must be distinct. So we may assume that γ is of maximal length. Suppose for instance that γ = e2re−1 2r−1 · · · e2e−1 1 2r)−1e′ 2r−1 · · · (e′ 2)−1 of length 2r − 1 in (E1, C 1), where each e′ is of even length. Then the vertex s(e1) is of type A in (E1, C 1), and there is a 1-path γ′ = (e′ i belongs to X(ei), for i = 2, . . . , 2r. Moreover all the 1-paths of (E1, C 1) starting at s(e1) are of this form, and we conclude that the 1-connected component of the vertex s(e1) only contains 1-paths of length ≤ 2r − 1. The case where γ is of odd length is treated in the same way. Now let n be the maximum of the lengths of all the maximal 1-paths starting at vertices of type A in (E, C). The above argument shows that the graph (En, C n) has the property that no vertex of type A is 1-connected to a vertex of type B. So we can assume, in addition, that no vertex of type A is 1-connected to a vertex of type (cid:3) B. But then Lemma 8.13 applies. 9. Primeness Recall that a ring is called prime if the product of any two non-zero ideals is non-zero. In K (E, C) and Or(E, C) in purely this final section, we characterize primeness of the algebras Lab graph theoretical terms when the configuration space Ω(E, C) is a Cantor space. In order to check this hypothesis, we also develop a test for the existence of isolated points. We first introduce a partial version of a well known concept. Definition 9.1. A partial action θ : G y Ω is called topologically transitive if for any two non-empty open subsets U, U ′ ⊂ G, there exists g ∈ G such that θg(U ∩ Ωg−1) ∩ U ′ 6= ∅. The main technical tool for our analysis is the fact that for a topologically free partial action θ : G y Ω on a totally disconnected compact Hausdorff space, the algebras CK(Ω) ⋊ G and C(Ω) ⋊r G are prime if and only if θ is topologically transitive. This should be clear from the fact that both these crossed products enjoy the intersection property (see [15, Lemma 4.2] and [25, Theorem 29.5]). Throughout this section, we will write id(α) and td(α) for the initial (i.e right most) and left most) edge, respectively, of a non-trivial admissible α when viewed as a terminal (i.e. path in the double bE. Definition 9.2. Let (E, C) denote a finite bipartite separated graph. If B ∈ Bn(Ω(E, C)), then we define a clopen subspace by Ω(E, C)B := {ξ ∈ Ω(E, C) ξn = B}, IDEAL STRUCTURE 53 and we will say that B is an n-ball at v, where v ∈ E0 is such that Ω(E, C)B ⊂ Ω(E, C)v. By the boundary of B, we shall mean the set ∂B := {td(α) α ∈ B is maximal}. Definition 9.3. Let (E, C) denote a finite bipartite graph, and consider a path closed subset A ⊂ E1 ∪ (E1)−1, that is, a subset satisfying for any admissible path α. From (E, C) being bipartite, this may also be phrased as follows: id(α) ∈ A ⇒ td(α) ∈ A • If ef −1 is admissible and f −1 ∈ A, then e ∈ A. • If e−1f is admissible and f ∈ A, then e−1 ∈ A. We will say that such A is ∂-closed if the following holds: (1) Assume s−1(v) ≥ 2 and let e ∈ s−1(v). If s−1(v) \ {e} ⊂ A, then e−1 ∈ A as well. (2) Assume Cv ≥ 2 and let X ∈ Cv. If Y −1 ∩ A 6= ∅ for any Y ∈ Cv \ {X}, then X ⊂ A as well. It is clear that an intersection of ∂-closed sets is again ∂-closed, so every A ⊂ E1 ∪ (E1)−1 is contained in a minimal ∂-closed subset A. We then associate a set of vertices to A by V (A) := {v ∈ E0,0 X −1 ∩ A 6= ∅ for all X ∈ Cv} ∪ {v ∈ E0,1 s−1(v) ⊂ A}. Lemma 9.4. Let (E, C) denote a finite bipartite separated graph, and consider a path closed subset A ⊂ E1 ∪ (E1)−1. Then v ∈ V (A) if and only if there exists some n ≥ 1 and an n-ball B ∈ Bn(Ω(E, C)) at v such that ∂B ⊂ A. Proof. First suppose that v /∈ V (A). We will show that ∂B 6⊂ A for any n ≥ 1 and any n-ball B at v, and we shall argue by induction on n. If n = 1, then this is clear by definition of V (A). Assuming that it holds for some n ≥ 1, consider any (n + 1)-ball B and write Bn := {α ∈ B : α ≤ n}. From our inductive assumption, there exists a maximal admissible path α ∈ Bn such that td(α) /∈ A. If α is also maximal in B, then surely ∂B 6⊂ A, so we may assume that it is not. We then divide into the two cases r(α) ∈ E0,0 and r(α) ∈ E0,1. In the former, we can write α = eβ, and there exists some Xe 6= Y ∈ Cr(α) with Y −1 ∩ A = ∅. By definition of Ω(E, C), we then have y−1α ∈ B for some y ∈ Y . In the other case, we may instead write α = e−1β. Since e−1 /∈ A, we see that f /∈ A for some e 6= f ∈ s−1(r(α)), and we consider the path f α ∈ B. Either way, we have found an element of the boundary ∂B which is not contained in A, so in particular not in A. For the converse implication, we first introduce a bit of handy notation, specifically we m=0 Am. First set A0 := A, let m ≥ 0 and assume that Ak has been defined for all k ≤ m. Then for any e ∈ E1, we declare that e−1 ∈ Am+1 if define a partition A =F∞ mGk=0 e−1 /∈ Ak, s−1(s(e)) ≥ 2 and s−1(s(e)) \ {e} ⊂ mGk=0 Ak. 54 PERE ARA AND MATIAS LOLK Similarly, e ∈ Am+1 if e /∈ mGk=0 Now set Ak, Cr(e) ≥ 2 and Y −1 ∩ mGk=0 Ak 6= ∅ for any Y ∈ Cr(e) \ {Xe}. T := {e ∈ E1 : Cr(e) = 1} ∪ {e−1 ∈ (E1)−1 : s−1(s(e)) = 1} and observe that A ∩ T = A ∩ T . Whenever σ ∈ A, we write mσ for the number satisfying σ ∈ Amσ , and if X −1 ∩ A 6= ∅ for some X ∈ C, we set mX := min{mx−1 x−1 ∈ X −1 ∩ A}. Given v ∈ V (A), we then define nv :=(cid:26) max{me e ∈ s−1(v)} + 1 if v ∈ E0,1 max{mX X ∈ Cv} + 1 if v ∈ E0,0 , there is an n-ball Bn at v satisfying ∂Bn ⊂ Fnv−n claiming that ∂B ⊂ A for an nv-ball B at v. Specifically, we will show that whenever n ≤ nv, k=0 Ak, and we proceed by induction over n. For n = 1, this is clear. Assuming that the claim has been verified for some 1 ≤ n < nv, and letting i be such that v ∈ E0,i, we consider the cases of i + n being odd and even, separately. If it is odd, then there is a unique (n + 1)-ball Bn+1 containing Bn given by with boundary Bn+1 = Bn ∪(cid:8)eα : α ∈ Bn, α = n and e ∈ s−1(r(α)) \ {td(α)−1}(cid:9) ∂Bn+1 = (∂Bn ∩ T ) ∪ [f ∈∂Bn\T s−1(s(f )) \ {f }, k=0 so it follows immediately from the above definition and A being path closed that ∂Bn+1 ⊂ Ak. If i + n is even, then ∂Bn \ T ⊂ E1, and for any f ∈ ∂Bn \ T , Y ∈ Cr(f ) \ {Xf }, Ak. Now define an (n + 1)-ball Fnv−n−1 we can choose an edge ef,Y ∈ Y such that e−1 Bn+1 by f,Y ∈ Fnv−n−1 k=0 Bn+1 := Bn ∪(cid:8)e−1 ∂Bn+1 = (∂Bn ∩ T ) ∪(cid:8)e−1 td(α),Y α : α ∈ Bn, α = n and Y ∈ Cr(α) \ {Xtd(α)}(cid:9) f,Y f ∈ ∂Bn \ T and Y ∈ Cr(f ) \ {Xf }(cid:9), Ak as desired. Considering the particular case n = nv and B := Bnv, (cid:3) and observe that hence ∂Bn+1 ⊂Fnv−n−1 we finally see that ∂B ⊂ A. k=0 Definition 9.5. Let (E, C) denote a bipartite separated graph. A choice path is an admissible path α = eβ such that there exists Xe 6= X ∈ Cr(α) with X ≥ 2. Given any edge f ∈ E1, we will say that f is a dead end if there is no choice path starting with f , and likewise f −1 is a dead end if no choice path starts with f −1. Proposition 9.6. Let (E, C) denote a finite bipartite separated graph, and define ADE := {dead ends of E1 ∪ (E1)−1}. IDEAL STRUCTURE 55 Then Ω(E, C)v contains an isolated point if and only if v ∈ V (ADE). Consequently, Ω(E, C) is a Cantor space if and only if V (ADE) = ∅. Proof. Observe first that A := ADE is path closed. It follows from Lemma 9.4 that v ∈ V (A) if and only if there exists a ball B at v such that ∂B ⊂ A. If this is the case, then Ω(E, C)B is a one-point space, so Ω(E, C)v does indeed contain an isolated point. If v /∈ V (A), then given any ball B at v, there exists σ ∈ ∂B which is not a dead end. Consequently, Ω(E, C)B contains at least two configurations, so Ω(E, C)v does not contain any isolated points. (cid:3) Remark 9.7. It is worth mentioning that if Ω(E, C) is a Cantor space, then θ(E,C) is auto- matically topologically free. Indeed, by [7, Theorem 10.5], θ(E,C) is topologically free if and only if for every vertex v ∈ E0,1 on a cycle, there exists some e ∈ s−1(v) which is not a dead end. And if no such e existed, then Ω(E, C)v would be a one-point space. Definition 9.8. Let (E, C) denote a finite bipartite separated graph. We will say that two sets A, A′ ⊂ E1 ∪ (E1)−1 can be linked if there exist σ ∈ A, σ′ ∈ A′ and an admissible path α such that the concatenation σ−1ασ′ is admissible. If this is not the case, then the pair A, A′ is unlinkable. Morover, it is maximal unlinkable if for any larger pair A ⊂ D, A′ ⊂ D′ we have D and D′ are unlinkable ⇒ A = D, A′ = D′. Finally, (E, C) is said to have the Linking Property if ∂B and ∂B′ can be linked for any two balls B, B′ ∈ B(Ω(E, C)). Lemma 9.9. If (E, C) has the Linking Property, then θ(E,C) is topologically transitive, and if Ω(E, C) is a Cantor space, then the reverse implication holds as well. Proof. Assume first that (E, C) has the Linking Property. Then given any two open sets U, U ′ ⊂ Ω(E, C), there are balls B, B′ such that Ω(E, C)B ⊂ U and Ω(E, C)B′ ⊂ U ′. Since B and B′ can be linked, there exist σ ∈ ∂B, σ′ ∈ ∂B′ and an admissible path α for which σ−1ασ′ is admissible. Let γ ∈ B and γ′ ∈ B′ be such that td(γ) = σ and td(γ′) = σ′, and set β := γ−1αγ′. The situation is depicted below in Figure 3. 1 σ′ σ−1 β α Figure 3. B′ (to the left) and B (to the right) linked by the admissible path α. 56 PERE ARA AND MATIAS LOLK It follows that ∅ 6= θβ(Ω(E, C)B ∩ Ω(E, C)β−1) ∩ Ω(E, C)B′ ⊂ θβ(U ∩ Ω(E, C)β−1) ∩ U ′, so θ(E,C) is indeed topologically transitive. Conversely, assume now that θ(E,C) is topologically transitive and Ω(E, C) is a Cantor space. Consider any two balls B, B′ and assume without loss of generality that both have radius n. If β ∈ F with β < 2n is such that θβ(cid:0)Ω(E, C)B ∩ Ω(E, C)β−1(cid:1) ∩ Ω(E, C)B′ 6= ∅, then take any two distinct points ξ, ξ′ in this open set, using the assumption about isolated points. By the Hausdorff property, these can be separated by open neighbourhoods U1, U ′ 1 of ξ and ξ′, respectively, contained in the above intersection. We have θβ−1(U1) ⊆ Ω(E, C)B, U ′ 1 ⊆ Ω(E, C)B′, and By applying this procedure sufficiently many times, we see that there are non-empty open subsets V ⊆ Ω(E, C)B and V ′ ⊆ Ω(E, C)B′, such that 1 = ∅. θβ(cid:0)θβ−1(U1) ∩ Ω(E, C)β−1(cid:1) ∩ U ′ θβ(cid:0)V ∩ Ω(E, C)β−1(cid:1) ∩ V ′ = ∅ for all β ∈ F with length β < 2n. However, by topological transitivity, there is some β ∈ F for which this intersection is non-empty, so β ≥ 2n. It follows that B and B′ can be linked. (cid:3) We are now in a position to characterize primeness. Theorem 9.10. Assume that Ω(E, C) is a Cantor space. Then either algebra Lab K (E, C) or Or(E, C) is prime if and only if V (A) = ∅ or V (A′) = ∅ for all maximal unlinkable pairs A, A′ ⊂ E1 ∪ (E1)−1. Proof. Observe that, by maximality, A and A′ as above are path closed. By Remark 9.7, θ(E,C) is topologically free, so Lab K (E, C) and Or(E, C) are both prime if and only if θ(E,C) is topologically transitive, which is again equivalent to the Linking Property by Lemma 9.9. So we really have to check that the Linking Property is equivalent to the above condition. But this is clear from Lemma 9.4. (cid:3) Remark 9.11. It is of course also natural to ask if interesting prime algebras can be con- structed from a finite bipartite separated graph (E, C), where the configuration space Ω(E, C) does contain isolated points. However, this is not the case: For sufficiently big n, there must exist a vertex v ∈ E0,1 such that Ω(En, C n)v is a one-point space, i.e. every e ∈ s−1(v) is n a dead end. By topological transitivity, the orbit of this single point is dense, so in par- ticular v can be connected to any other vertex by an admissible path. But then (En, C n) must satisfy Condition (C) of [38, Definition 3.5]. Note that v cannot admit exactly one simple closed path (up to inversion), for then it would generate an ideal Morita equivalent to K[Z] or C(T), depending on the situation, by 7.13. If v does not admit a closed path, then both algebras degenerate to graph algebras of a non-separated graph by [38, Theorem 5.7]. IDEAL STRUCTURE 57 Finally, if v admits at least two simple closed paths (up to inversion), then it generates an ideal Morita equivalent with the group algebra K[Fv] in the algebraic setting and C ∗ r (Fv) in the C ∗-algebraic (here Fv denotes the group of all closed paths based at v), and the quotient is a classical graph algebra. Example 9.12. We now apply our work to a few examples: (1) If (E, C) = (E(m, n), C(m, n)) as in Example 6.6, then ADE = ∅ and every pair of m,n are edges can be linked, so Ω(E, C) is a Cantor space and the algebras Lab prime. m,n and Or (2) Consider the graph (E, C) as pictured just below: x1 y3 y1 x2 y2 Note that ADE = {y−1 follows that Ω(E, C) is a Cantor space. However, the set A := {x−1 not linked to itself and V (A) = {s(y3)}, so the algebras are not prime. 3 } has closure ADE = {x1, x2, y−1 1 , y−1 2 , y−1 3 }, so V (ADE) = ∅. It 2 , y1, y2, y3} is 1 , x−1 (3) Now consider the following variation of the above graph: x1 x2 y1 x3 y2 y3 Once again we have V (ADE) = ∅, so Ω(E, C) is indeed a Cantor space. Observe that there is a unique pair of maximal unlinkable subsets, namely A = A′ := {x1, x2, x3, y−1 1 , y−1 2 , y−1 3 }, and V (A) = ∅. It follows that Lab K (E, C) and Or(E, C) are prime in this case. Remark 9.13. We finally remark that topological transitivity of θ(E,C) can be phrased very simply in terms of the separated Bratteli diagram (F∞, D∞). Since the vertices of F∞ corre- spond to the balls of Ω(E, C), and there is a direct dynamical equivalence θ(En,C n) ≈−→ θ(E,C) for any n, the partial action θ(E,C) is topologically transitive if and only if for all n, any 58 PERE ARA AND MATIAS LOLK n can be connected by an admissible path in (En, C n), or, alternatively, two vertices u, v ∈ E0 that any two vertices in F∞ lay over a common vertex w ∈ F 0 ∞. Consequently, one can give another proof of Theorem 9.10 by checking that the graph theoretical condition passes from (E, C) to (E1, C 1). References [1] F. Abadie, Enveloping actions and Takai duality for partial actions, J. Funct. Anal. 197 (2003), 14 -- 67. [2] F. Abadie, On partial actions and groupoids, Proc. Amer. Math. Soc. 132 (2004), 1037-1047. [3] G. Abrams, P. Ara, M. Siles Molina, Leavitt Path Algebras, to appear in Lecture Notes in Mathematics (Springer). [4] G. Abrams, G. Aranda Pino, The Leavitt path algebra of a graph, J. Algebra 293 (2005), 319 -- 334. [5] P. Ara, Purely infinite simple reduced C-algebras of one-relator separated graphs, J. Math. Anal. Appl. 393 (2012), 493 -- 508. [6] P. Ara, J. Claramunt, Approximating the group algebra of the lamplighter by finite-dimensional algebras. In preparation. [7] P. Ara, R. Exel, Dynamical systems associated to separated graphs, graph algebras, and paradoxical decompositions, Adv. Math. 252 (2014), 748 -- 804. [8] P. Ara, R. Exel, K-theory for the tame C-algebra of a separated graph. Journal of Functional Analysis 269 (2015), 2995-3041. [9] P. Ara, R. Exel, T. Katsura, Dynamical systems of type (m, n) and their C-algebras, Ergodic Theory Dynam. Systems 33 (2013), 1291 -- 1325. [10] P. Ara, K. R. Goodearl, C-algebras of separated graphs, J. Funct. Anal. 261 (2011), 2540 -- 2568. [11] P. Ara, K. R. Goodearl, Leavitt path algebras of separated graphs. J. reine angew. Math. 669 (2012), 165 -- 224. [12] P. Ara, K. R. Goodearl, The realization problem for some wild monoids and the Atiyah problem, Trans. Amer. Math. Soc. http:dx.doi.org/10.1090/tran/6889. Article published electronically on December 7, 2016. arXiv:1410.8838v2 [math.RA]. [13] P. Ara, K. R. Goodearl, K.C. O'Meara, E. Pardo, Separative cancellation for projective modules over exchange rings, Israel J. Math. 105 (1998), 105 -- 137. [14] P. Ara, M. A. Moreno, E. Pardo, Nonstable K-theory for graph algebras, Algebr. Represent. Theory 10 (2007), 157 -- 178. [15] J. Brown, L. O. Clark, C. Farthing, A. Sims, Simplicity of algebras associated to ´etale groupoids, Semi- group Forum 88 (2014), 433 -- 452. [16] T. Bates, J. Hong, I. Raeburn and W. Szyma´nski, The ideal structure of the C-algebras of infinite graphs, Illinois J. Math. 46 (2002), 1159 -- 1176. [17] T. Bates, D. Pask, I. Raeburn and W. Szyma´nski, The C-algebras of row-finite graphs, New York J. Math. 6 (2000), 307 -- 324 (electronic). [18] O. Bratteli, Inductive limits of finite dimensional C-algebras, Trans. Amer. Math. Soc. 171 (1972), 195 -- 234. [19] Toke M. Carlsen and Nadia S. Larsen, Partial actions and KMS states on relative graph C-algebras, J. Funct. Anal. 271 (2016) 2090-2132. [20] L. O. Clark and A. Sims, Equivalent groupoids have Morita equivalent Steinberg algebras, J. Pure Appl. Algebra 219 (2015), 2062 -- 2075. [21] Toke M. Carlsen, Efren Ruiz, and Aidan Sims, Equivalence and stable isomorphism of groupoids, and diagonal-preserving stable isomorphisms of graph C-algebras and Leavitt path algebras, Proc. Amer. Math. Soc. 145 (2017), 1581 -- 1592. [22] M. Dokuchaev, R. Exel, Partial actions and subshifts, preprint 2015, arXiv:1511.00939v1 [math.OA]. IDEAL STRUCTURE 59 [23] T. Downarowicz, A. Maass, Finite-rank Bratteli-Vershik diagrams are expansive, Ergodic Theory Dynam. Systems 28 (2008), 739 -- 747. [24] G. A. Elliott, On the classification of C-algebras of real rank zero, J. Reine Angew. Math. 443 (1993), 179 -- 219. [25] R. Exel, Partial Dynamical Systems, Fell Bundles and Applications. www.mtm.ufsc.br/~exel/papers/pdynsysfellbun.pdf. [26] R. Exel, Amenability for Fell bundles, J. Reine Angew. Math. 492 (1997), 41 -- 73. [27] Ruy Exel, Inverse semigroups and combinatorial C-algebras, Bull. Braz. Math. Soc. (N.S.) 39 (2008), 191 -- 313. [28] R. Exel and M. Laca, Cuntz -- Krieger algebras for infinite matrices, J. reine angew. Math. 512 (1999), 119 -- 172. [29] R. Exel, M. Laca, J. Quigg, Partial dynamical systems and C-algebras generated by partial isometries, J. Operator Theory 47 (2002), 169 -- 186. [30] T. Giordano, I. F. Putnam, G. F. Skau, Topological orbit equivalence and C-crossed products, J. Reine Angew. Math. 469 (1995), 51 -- 111. [31] D. Gon¸calves, D. Royer, Leavitt path algebras as partial skew group rings, Comm. Algebra 42 (2014), 3578 -- 3592. [32] J.H. Hong and W. Szyma´nski, The primitive ideal space of the C -algebras of infinite graphs, J. Math. Soc. Japan 56 (2004), 4564. [33] S-M. Høynes, F inite-rank Bratteli-Vershik diagrams are expansive - -a new proof, arXiv:1411.3371v1[math.DS]. [34] D. Kerr, P. W. Nowak, Residually finite actions and crossed products, Ergodic Theory and Dynamical Systems 32 (2012), 1585 -- 1614. [35] X. Li, Partial transformation groupoids attached to graphs and semigroups, arXiv:1603.09165 [math.OA]. [36] X. Li, Dynamic characterizations of quasi-isometry, and applications to cohomology, arXiv:1604.07375v3 [math.GR]. [37] D. Lind, B. Marcus, An introduction to symbolic dynamics and coding. Cambridge University Press, Cambridge, 1995. [38] M. Lolk, Exchange rings and real rank zero C ∗-algebras associated with finitely separated graphs, preprint 2017. [39] M. Lolk, On nuclearity and exactness of the tame C ∗-algebras associated with finitely separated graphs, preprint 2017. [40] P. S. Muhly, J. N. Renault and D. P. Williams, Equivalence and isomorphism for groupoid C ∗-algebras, J. Operator Theory 17 (1987), 3 -- 22. [41] A. L. T. Paterson, Graph inverse semigroups, groupoids and their C-algebras, J. Operator Theory 48 (2002), 645 -- 662. [42] I. F. Putnam, On the topological stable rank of certain transformation group C-algebras, Ergodic Theory Dynam. Systems 10 (1990), 197 -- 207. [43] Robert T. Powers, Simplicity of the C ∗-algebra associated with the free group on two generators, Duke Math. J. 42 (1975), no. 1, 151 -- 156. [44] I. Raeburn, Graph algebras. CBMS Regional Conference Series in Mathematics, 103. Published for the Conference Board of the Mathematical Sciences, Washington, DC; by the American Mathematical Society, Providence, RI, 2005. [45] J. Renault, A groupoid approach to C-algebras, Lecture Notes in Mathematics, 793. Springer, Berlin, 1980. [46] M. Rørdam, A. Sierakowski, Purely infinite C-algebras arising from crossed products, Ergodic Theory and Dynamical Systems 32 (2012), 273 -- 293. [47] B. Steinberg, A groupoid approach to discrete inverse semigroup algebras, Adv. Math. 223 (2010), 689727. 60 PERE ARA AND MATIAS LOLK [48] A. Sims and D. P. Williams, Renault's equivalence theorem for reduced groupoid C ∗-algebras, J. Operator Theory 68 (2012), 223 -- 239. [49] F. Sugisaki, On the subshift within a strong orbit equivalence class for minimal homeomorphisms, Ergodic Theory Dynam. Systems 27 (2007), 971 -- 990. [50] W. Szyma´nski, General Cuntz-Krieger uniqueness theorem, Internat. J. Math. 13 (2002) 549555. Departament de Matem`atiques, Universitat Aut`onoma de Barcelona, 08193 Bellaterra (Barcelona), Spain. E-mail address: [email protected] Department of Mathematical Sciences, University of Copenhagen, 2100 Copenhagen, Den- mark. E-mail address: [email protected]
1002.2335	1	1002	2010-02-11T19:15:26	Morita equivalence of nest algebras	[ "math.OA", "math.FA" ]	Let N_1 (resp.N_2) be a nest A (resp. B) be the corresponding nest algebra, A_0 (resp. B_0) be the subalgebra of compact operators. We prove that the nests N_1, N_2 are isomorphic if and only if the algebras A, B are weakly-* Morita equivalent if and only if the algebras A_0, B_0 are strongly Morita equivalent. We characterize the nest isomorphisms which implement stable isomorphism between the corresponding nest algebras.	math.OA	math	MORITA EQUIVALENCE OF NEST ALGEBRAS G. K. ELEFTHERAKIS Abstract. Let N1 (resp. N2) be a nest, A (resp. B) be the corresponding nest algebra, A0 (resp. B0) be the subalgebra of compact operators. We prove that the nests N1,N2 are isomorphic if and only if the algebras A, B are weakly−∗ Morita equivalent if and only if the algebras A0, B0 are strongly Morita equivalent. We characterize the nest isomorphisms which implement stable isomorphism between the corresponding nest algebras. 1. Introduction Rieffel, introduced the idea of Morita equivalence in Operator Theory de- veloping the theory of Morita equivalence for C ∗ and W ∗ algebras [17]. After the advent of the theory of operator spaces and operator algebras a parallel Morita theory for non-selfadjoint algebras was developed by Blecher, Muhly and Paulsen [6], [2]. We call this equivalence strong Morita equivalence. Recently, two approaches have been suggested for the Morita equivalence of dual operator algebras. The one was introduced [11] by the author of this article and it is equivalent to the notion of stable isomorphism of dual operator algebras [13]. We call this equivalence ∆−equivalence. The other was introduced by Blecher and Kashyap [3], [14] and it is strictly weaker than ∆−equivalence. This equivalence is called weak−∗ Morita equivalence. It is interesting that if A and B are strongly Morita equivalent approximately unital operator algebras then the second dual operator algebras A∗∗, B∗∗ are weakly−∗ Morita equivalent [3]. New results on weak−∗ Morita equivalence and ∆−equivalence can be found in [4]. In this paper we prove that strong and weak−∗ Morita equivalence is a lattice property for nest algebras. Particularly we prove that if A and B are nest algebras and A0, B0 are the subalgebras of compact operators then A0 and B0 are strongly Morita equivalent if and only if A and B are weakly−∗ Morita equivalent if and only if the nests Lat(A), Lat(B) are isomorphic. The main tool of the proof is that if θ : Lat(A) → Lat(B) is a nest isomorphism we can construct a dual operator A − B bimodule Y and a dual operator B − A bimodule X such that the identity operator of A is the limit in strong operator topology of a net of finite rank contractions (fλ) where every fλ is the norm limit of a sequence (yλ i xλ i )i∈N, where yλ i is a contractive row operator with finite entries from Y and xλ i is a contractive column operator with finite 1 2 G. K. ELEFTHERAKIS entries from X. Similarly we can decompose the identity of the algebra B. This can be considered as a generalization of the Erdos density Theorem for nest algebras [7]. In section 3 we prove that two nest algebras are weakly−∗ Morita equivalent if and only if they are spatially Morita equivalent (definition 3.1). Also we prove that every spatially Morita equivalent dual operator algebra with a nest algebra is weakly−∗ Morita equivalent with this nest algebra. It is interesting that this does not happen for the more general class of operator algebras, the CSL algebras. In section 4 we present a measure-theoretic result which describes when two separably acting nest algebras are stably isomorphic. As it was pointed out in [3] the [12, example 3.7] is an example of weak−∗ Morita equivalent algebras which are not stably isomorphic. Using the results of this paper we give a new proof of the fact that weak−∗ Morita equivalence is strictly weaker than ∆−equivalence. In section 5 we present a counterexample which states that the second duals of two unital strongly Morita equivalent algebras are not necessarily stably isomorphic. In what follows we describe the notions we use in this paper. Since we use extensively the basics of Operator Space Theory, we refer the reader to the monographs [5], [9], [15] and [16] for further details. A (normal) repre- sentation of a (dual) operator algebra A is a (w∗−continuous) completely contractive homomorphism α : A → B(H) on a Hilbert space H. In the case A is unital, we assume that α is unital. Let H, K be Hilbert spaces and A ⊂ B(H) be an algebra. A subspace X ⊂ B(K, H) is called a left module over A if AX ⊂ X. Similarly we can define the right modules over A. A left and right module over A is called a bimodule over A. An abstract left (right) operator module over an abstract operator algebra A is an operator space Y such that there exist a completely contractive bilinear map A× Y → Y (Y × A → Y ). A left and right operator module over A is called an operator bimodule over A. If A is a dual operator algebra and Y is a dual operator space we say that Y is a left (right) dual operator module if the above completely contractive bilinear map is separately w∗−continuous. A left and right dual operator module over A is called a dual operator bimodule over A. Two operator bimodules Y and Z over an operator algebra A are called isomorphic as operator bimodules if there exists a completely isometric and onto A−module map π : Y → Z. We denote Y ∼= Z as operator bimodules. In the case A is a dual operator algebra and Y, Z are dual operator bimodules we denote Y ∼= Z as dual operator bimodules if the above completely isometric and onto A−module map π is w∗−(bi)continuous. MORITA EQUIVALENCE OF NEST ALGEBRAS 3 If Y is a right operator module over an operator algebra A and X is a left operator module over A we denote by Y ⊗h A X the balanced Haagerup tensor product of Y and X which linearizes the completely bounded A−balanced bilinear maps [5, 3.4]. If Y (resp. X) is a left (resp. right) operator module over an operator algebra B then Y ⊗h A X is also a left (resp. right) operator module over B, [6, Lemma 2.4]. If Y is a dual right operator module over a dual operator algebra A and X is a left dual operator module over A we denote by Y ⊗σh A X the balanced normal Haagerup tensor product of Y and X which linearizes the separately w∗−continuous completely bounded A−balanced bilinear maps [13]. In the case Y (resp. X) is a left (resp. right) dual operator module over a dual operator algebra B then Y ⊗σh A X is also a left (resp. right) dual operator module over B, [13]. We give now the two definitions of Morita equivalence using in this paper: Definition 1.1. [6] The operator algebras A, B are called strongly Morita equivalent if there exist an A− B operator module X and a B − A operator module Y such that A ∼= X ⊗h A X as A and B operator bimodules respectively. Definition 1.2. [3] The dual operator algebras A, B are called weakly−∗ Morita equivalent if there exist an A − B dual operator module X and a B − A dual operator module Y such that A ∼= X ⊗σh A X as A and B dual operator bimodules respectively. B Y and B ∼= Y ⊗σh B Y and B ∼= Y ⊗h 1, s1 1, s2 If s1 = (s1 ∞ (X) (resp. C f in n1 6= 0 and s2 = (s2 If X is a subspace of B(H, K), where H and K are Hilbert spaces, we denote ∞ (X)) the space of operators (x1, x2, ...) : H ∞ → K by Rf in (resp. (x1, x2, ...)T : H → K∞) such that xi ∈ X for all i and there exists n0 ∈ N such that xn = 0 for all n ≥ n0. 2, ..., s1 n1, 0, 0, ...), s1 2, ..., s2 n2 6= 0 are operators in Rf in ∞ (X) we denote by (s1, s2) the operator (s1 1, s1 2, ..., s1 which also belongs to Rf in ∞ (X). In the same way if s1, s2, ..., sn ∈ Rf in ∞ (X) we define the operator (s1, s2, ..., sn) ∈ Rf in ∞ (X). Similarly if t1, t2, ..., tn ∈ ∞ (X) we define the operator (t1, t2, ..., tn)T ∈ C f in C f in A nest N is a totally ordered set of projections of a Hilbert space H containing the zero and identity operators which is closed under arbitrary intersections and closed spans. The corresponding nest algebra is n2, 0, 0, ...), s2 ∞ (X). n1, s2 1, s2 2, ..., s2 n2, 0, 0, ...) Alg(N ) = {x ∈ B(H) : N ⊥xN = 0 ∀ N ∈ N}. If N ∈ N we denote by N− the projection onto the closed span of the union ∪ M <N (M(H)). If N− < N we call the projection N⊖N− an atom. If the nest M ∈N 4 G. K. ELEFTHERAKIS has not atoms is called a continuous nest. If the atoms span the identity operator the nest is called a totally atomic nest. An order preserving 1-1 and onto map between two nests is called a nest isomorphism. If N1 and N2 are nests acting on the Hilbert spaces H1, H2 respectively and θ : N1 → N2 is a nest isomorphism we denote by Op(θ) the space of operators x ∈ B(H1, H2) satisfying θ(N)⊥xN = 0 for all N ∈ N1. Observe that Op(θ) is an Alg(N2) − Alg(N1) bimodule. Finally, if X is a normed space we denote by Ball(X) the unit ball of X and by X ∗ its dual space. If H1, H2 are Hilbert spaces and ξ ∈ H2, η ∈ H1 are vectors we denote by ξ ⊗ η∗ the rank 1 operator sending every ω ∈ H1 to < ω, η > ξ ∈ H2, where < ·,· > is the inner product of H1. Also we symbolize the strong operator topology by SOT. 2. Morita equivalence for nest algebras In this section we fix nests N1,N2 acting on the Hilbert spaces H1, H2 re- spectively, and a nest isomorphism θ : N1 → N2. We denote A = Alg(N1), B = Alg(N2), X = Op(θ), Y = Op(θ−1). If Z is a space of operators we denote its subspace of compact operators by Z0. Observe that AY B ⊂ Y, BXA ⊂ X, Y X ⊂ A, XY ⊂ B, A0Y0B0 ⊂ Y0, B0X0A0 ⊂ X0, Y0X0 ⊂ A0, X0Y0 ⊂ B0. The main result of this section is Theorem 2.9. In particular we are going to prove that A0 ∼= Y0 ⊗h B0 X0, B0 ∼= X0 ⊗h A0 Y0, A ∼= Y ⊗σh B X, B ∼= X ⊗σh A Y. Suppose that p = ∨{N ⊖ N− : N ∈ N1}. The following lemmas are used in Theorem 2.5, where we are going to prove a variant of the Erdos density The- orem for nest algebras: There exists a net of finite rank contractions (fλ) ⊂ A converging in SOT topology to the identity operator of H1, where every fλ is the norm limit of a sequence (yλ i ∈ Ball(C f in i ∈ Ball(Rf in i )i∈N where yλ ∞ (Y0)), xλ ∞ (X0)) for all i, λ. i xλ Lemma 2.1. There exists a net (lλ) of finite rank contractions converg- ing in SOT topology to the projection p such that lλ = sλtλ where sλ ∈ Ball(Rf in ∞ (X0)) for all λ. ∞ (Y0)), tλ ∈ Ball(C f in Proof Suppose that p = ∨k∈J pk where pk = Nk⊖(Nk)− for Nk ∈ N1, k ∈ J. Choose a net of finite rank contractions (fi)i∈I converging in SOT topology to the identity operator of H1. If F = {F : F finite subset of J} the family (gF,i)(F,i) indexed by F × I where gF,i = Pk∈F pkfipk is a net. Observe that every gF,i is a finite rank contraction belonging to A. We can easily check that SOT − lim(F,i) gF,i = ∨kpk = p. MORITA EQUIVALENCE OF NEST ALGEBRAS 5 Let f = pkfipk for some k ∈ J, i ∈ I with polar decomposition f = uf. Suppose that n Xj=1 for λj ≥ 0 and ξj orthogonal vectors of pk(H1). Choose a unit vector η in (θ(Nk) ⊖ θ(Nk)−)(H2). Now we have λjξj ⊗ ξ∗ f = j n λjξj ⊗ η∗ · η ⊗ ξ∗ j = yy∗ f = Xj=1 ∞ (Y0)), y∗ ∈ Ball(C f in Ball(Rf in where y = (√λ1ξ1 ⊗ η∗, ...,√λnξn ⊗ η∗). Observe that f = uyy∗ and uy ∈ Suppose now that F = {j1, ..., jn} ⊂ J, i ∈ I and gF,i = Pn By the above arguments pjkfipjk = sktk where sk ∈ Ball(Rf in Ball(C f in k=1 pjkfipjk. ∞ (Y0)), tk ∈ ∞ (X0)). So gF,i = st where ∞ (X0)). s = (s1, ..., sn) ∈ Rf in ∞ (Y0), t = (t1, ..., tn)T ∈ C f in ∞ (X0). Also, since the projections (pk)k∈J are pairwise orthogonal and ksks∗ for all k we have that kk ≤ 1 n n ksk2 = k Xk=1 sks∗ kk = k pjksks∗ kpjkk ≤ 1. Xk=1 Similarly we can prove ktk ≤ 1 and this completes the proof. Lemma 2.2. Suppose that p⊥ 6= 0 ξ, η ∈ Ball(H1) and N ∈ N1 such that ξ = p⊥N(ξ), η = p⊥N ⊥(η). There exist rank 1 operators (sn)n∈N ⊂ Ball(Y ), (tn)n∈N ⊂ Ball(X), such that the operator ξ ⊗ η∗ is the norm limit of the sequence (sntn)n∈N. (cid:3) Proof We define the continuous order preserving map φ : p⊥N1 → [0,kξk2] : p⊥M → kp⊥M(ξ)k2. The nest p⊥N1 is continuous, so φ is onto [0,kξk2]. Choose a strictly increasing sequence (λn) such that λn → kξk2. Choose Nn ∈ N1 such that φ(p⊥Nn) = λn. It follows that Nn < Nn+1 < N for all n ∈ N and p⊥Nn(ξ) → ξ. Similarly we can find a sequence (Mn)n∈N such that N < Mn+1 < Mn for all n ∈ N and p⊥(I − Mn)(η) → η. For every n ∈ N we choose ωn ∈ H2 such that kθ(Mn) ⊖ θ(Nn)(ωn)k = 1. The operator sn = p⊥Nn(ξ) ⊗ (θ(Mn) ⊖ θ(Nn)(ωn))∗ satisfies sn = Nnsnθ(Nn)⊥ and so sn ∈ Ball(Y0). Similarly the operator tn = (θ(Mn) ⊖ θ(Nn)(ωn)) ⊗ p⊥(I − Mn)(η)∗ 6 G. K. ELEFTHERAKIS satisfies tn = θ(Mn)tnM ⊥ n and so tn ∈ Ball(X0). Now we have sntn = p⊥Nn(ξ) ⊗ p⊥(I − Mn)(η)∗ which clearly converges in norm to the operator ξ ⊗ η∗. Lemma 2.3. Suppose that p⊥ 6= 0, ξ ∈ H1 such that kp⊥(ξ)k = 1 and q is the projection onto the space p⊥N ′′ 1 ξ. There exists a sequence of finite rank contractions (rn)n∈N ⊂ A converging in SOT topology to the projection q such i where sn i tn that rn = k · k − limi∈N sn ∞ (X0)) for all i, n ∈ N. i ∈ Ball(C f in i ∈ Ball(Rf in ∞ (Y0)), tn (cid:3) Proof We define the continuous order preserving map φ : N1p⊥ → [0, 1], φ(Np⊥) = kNp⊥(ξ)k2. Since the nest N1p⊥ is continuous φ is onto [0, 1]. Choose Nk,np⊥ the least element in N1p⊥ such that φ(Nk,np⊥) = k 2n , k = 0, 1, ..., 2n. We denote Ek,n = (Nk,n ⊖ Nk−1,n)p⊥, ξk,n = 2 n 2 Ek,n(ξ), rn = fk,n 2n Xk=2 k,n. where fk,n = ξk−1,n ⊗ ξ∗ converges in SOT topology to the operator q. As in [7, Lemma 3.9] we can prove that krnk ≤ 1 and the sequence (rn)n∈N By the above lemma there exist sequences of rank 1 operators (sk,n )i∈N ⊂ → fk,n, i → ∞ for all k, n. , t3,n )i∈N ⊂ Ball(X0), such that sk,n i = (s2,n sn Ball(Y0),(tk,n We denote i = (t2,n tn , ..., s2n,n , ..., t2n,n , s3,n tk,n i )T ), k·k i i i i i i i i tn and we have rn = k · k − limi sn i . Also 2n i i We may assume that sk,n i = Ek−1,nsk,n i ksn i k2 = k Xk=2 sk,n i (sk,n i )∗k. so ksn i k2 = k Ek−1,nsk,n i (sk,n i )∗Ek−1,nk. 2n Xk=2 i k ≤ 1 and the projections (Ek−1,n)k are pairwise orthogonal we i k ≤ 1. Similarly we can prove ktn Since ksk,n have ksn Lemma 2.4. Suppose that p⊥ 6= 0. There exists a net (gλ) of finite rank contractions in A converging in SOT topology to p⊥ such that gλ = k · k − limi∈N sλ ∞ (X0)) for all i ∈ N. i ∈ Ball(C f in i ∈ Ball(Rf in for all λ where sλ ∞ (Y0)), tλ i k ≤ 1. i tλ i (cid:3) MORITA EQUIVALENCE OF NEST ALGEBRAS 7 Proof Using Zorn's Lemma we find a family of vectors ξk : k ∈ L such that the projections qk onto p⊥N ′′ 1 ξk, k ∈ L are pairwise orthogonal and they span p⊥. We assume that kp⊥(ξk)k = 1 for all k ∈ L. From Lemma 2.3 there exist finite rank contractions (rk n)n∈N such that qk = SOT − limn∈N rk n and n = k · k − limi∈N sn,k rk for sequences tn,k i i (sn,k i )i∈N ⊂ Ball(Rf in ∞ (Y0)), (tn,k i )i∈N ⊂ Ball(C f in ∞ (X0)). We define F = {F : F finite subset of L}. If F ∈ F and n ∈ N we define the finite rank contraction gn,F = Pk∈F rk n. The family (gn,F )n,F indexed by N × F is a net. Fix ξ ∈ H1. Observe that for all n ∈ N and so krk n(ξ) − qk(ξ)k2 = kqk(rk Xk∈L n(ξ) − qk(ξ)k2 ≤ 2Xk∈L krk n − IH1)qk(ξ)k2 ≤ 2kqk(ξ)k2 kqk(ξ)k2 < ∞. If n ∈ N and F ∈ F we have (2.1) qk(ξ)k2 kgn,F (ξ) − p⊥(ξ)k2 = kgn,F (ξ) −Xk∈L n(ξ) − qk(ξ)k2 + kp⊥(ξ)k2 −Xk∈F =Xk∈F krk ≤Xk∈L n(ξ) − qk(ξ)k2 + kp⊥(ξ)k2 −Xk∈F krk n(ξ)−qk(ξ)k2 = 0 by the Theorem of dominated convergence Since limn∈N krk kqk(ξ)k2 kqk(ξ)k2 we have lim n∈NXk∈L krk n(ξ) − qk(ξ)k2 = 0. It follows now from (2.1) that lim(n,F ) kgn,F (ξ)− p⊥(ξ)k2 = 0. We proved that SOT − lim(n,F ) gn,F = p⊥. If F = {k1, ..., kr} ⊂ L then gn,F =Pr rkm n = k · k − lim tn,F So gn,F = k · k − limi∈N sn,F i where , ..., sn,kr i = (sn,k1 sn,F tn,kj i = tn,kj n where tn,km i m=1 rkm sn,km i tn,F i = (tn,k1 , ..., tn,kr )T . i∈N ), . i i i i = qkj sn,kj Since sn,kj i orthogonal we conclude that sn,F all (n, F ). This completes the proof. , i qkj and the projections (qkj ) are pairwise ∞ (X0)) for i ∈ Ball(C f in ∞ (Y0)), tn,F (cid:3) i ∈ Ball(Rf in i i 8 G. K. ELEFTHERAKIS i uλ Theorem 2.5. There exists a net of finite rank contractions (fλ)λ∈Λ con- verging in SOT topology to the identity operator IH1 such that fλ = k · k − limi∈N vλ ∞ (X0)) for all λ ∈ Λ. i )i∈N ⊂ Ball(Rf in i )i∈N ⊂ Ball(C f in ∞ (Y0)), (uλ i where (vλ Proof If p⊥ = 0 the conclusion comes from Lemma 2.1. So we may assume that p⊥ 6= 0. From Lemmas 2.1, 2.4 there exists a net (lλ)λ∈Λ of finite rank contractions converging in SOT topology to the projection p such that lλ = ∞ (Y0)), tλ ∈ Ball(C f in sλtλ where sλ ∈ Ball(Rf in ∞ (X0)) for all λ ∈ Λ, and a net (gλ)λ∈Λ of finite rank contractions converging in SOT topology to p⊥ such i xλ that gλ = k · k − limi∈N yλ i for all λ ∈ Λ where yλ i ∈ Ball(Rf in ∞ (Y0)), xλ i ∈ Ball(C f in ∞ (X0)) We denote fλ = lλ + gλ, vλ for all i ∈ N. Observe that IH1 = SOT − limλ∈Λ fλ and fλ = k · k − limi∈N vλ have i = (sλ, yλ i = (tλ, xλ i ), uλ i )T for all λ ∈ Λ, i ∈ N. i . Now we i uλ kvλ =kpsλs∗ i k2 = ksλs∗ λp + p⊥yλ i k ≤ 1 for all λ ∈ Λ, i ∈ N. λ + yλ i (yλ i )∗k i (yλ i )∗p⊥k ≤ 1. (cid:3) Similarly kuλ Theorem 2.6. The algebras A0, B0 are strongly Morita equivalent. Particu- larly A0 ∼= Y0 ⊗h Proof We define the bilinear map Y0×X0 → A0 : (y, x) → yx. This map is completely contractive and B0−balanced, so induces a completely contractive A0−module map π : Y0 ⊗h B0 X0 → A0 : y ⊗B0 x → yx. We shall prove that π is completely isometric. It suffices to prove that if B0 X0, B0 ∼= X0 ⊗h A0 Y0 as operator modules. mi,j then zi,j = yi,j k xi,j k !i,j k , i, j = 1, ..., n yi,j k ⊗B0 xi,j We recall the contractions fλ, (vλ Xk=1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) k(zi,j)i,jk ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j Xk=1 s )s∈N, λ ∈ Λ from Theorem 2.5. If x is a compact operator then x = k·k−limλ xfλ ([7, Proposition 1.18]).It follows that zi,j = k·k− limλPmi,j k ⊗B0 (xi,j k fλ). If ǫ > 0 there exists λ ∈ Λ k(zi,j)i,jk − ǫ <(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j k fλ)!i,j Xk=1 k ⊗B0 (xi,j yi,j s )s∈N, (uλ such that k=1 yi,j (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) − ǫ 2 . . MORITA EQUIVALENCE OF NEST ALGEBRAS 9 s )!i,j . (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Since xi,j s there exists s ∈ N such that ǫ 2 k ⊗B0 (xi,j yi,j k vλ s uλ − s!i,j s ) ⊗B0 uλ (yi,j k vλ s uλ k vλ k vλ k xi,j Since xi,j s ∈ Rf in ∞ (B0) we have k ⊗B0 (xi,j yi,j <(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j Xk=1 k fλ = k · k − lims∈N xi,j (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j k fλ)!i,j Xk=1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) k(zi,j)i,jk − ǫ <(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j Xk=1 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j k !i,j(cid:0)vλ Xk=1 s ⊗B0 uλ (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j k !i,j Xk=1 (cid:13)(cid:13)vλ s(cid:13)(cid:13) s ⊗B0 uλ s(cid:13)(cid:13) ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j mi,j k !i,j Xk=1 Xk=1 s(cid:13)(cid:13)(cid:13)(cid:13)uλ (cid:13)(cid:13)vλ Since ǫ was arbitrary we have k(zi,j)i,jk ≤ (cid:13)(cid:13)(cid:13)(cid:0)Pmi,j s ⊕ ... ⊕ vλ yi,j k xi,j yi,j k xi,j yi,j k xi,j (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) s(cid:1)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) k !i,j k ⊗B0 xi,j (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) s ⊗B0 uλ yi,j k xi,j . We proved that π is completely isometric. It remains to prove that π is onto A0. It suffices to prove that the space Imπ is dense in A0. k=1 yi,j k (cid:1)i,j(cid:13)(cid:13)(cid:13) Let a ∈ Ball(A0) and ǫ > 0. Since a = k · k − limλ fλa there exists λ ∈ Λ s there exists s ∈ N such 2. Since fλ = k · k − lims vλ such that ka − fλak < ǫ that s uλ s uλ kfλ − vλ ǫ sk < 2 s ak < ǫ. But vλ s uλ s a = π(vλ . It follows that ka − vλ s ⊗B0 (uλ completes the proof. Similarly we can prove that B0 ∼= X0 ⊗h s uλ A0 Y0. s a)) and this (cid:3) We define the bilinear map Y × X → A : (y, x) → yx. This map is com- pletely contractive B−balanced and separately w∗−continuous, so induces a completely contractive w∗−continuous map ρ : Y ⊗σh B X → A : y ⊗B x → yx which is also an A−module map. We shall prove that the restriction of ρ on the space Y ⊗h B X is completely isometric and we shall use this fact in Theorem 2.9 to prove that A ∼= Y ⊗σh Lemma 2.7. The restriction of ρ on the space Y ⊗h BX is completely isometric. B X. 10 G. K. ELEFTHERAKIS Proof It suffices to prove that if zi,j = mi,j Xk=1 yi,j k ⊗B xi,j k , i, j = 1, ..., n then (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) s )(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) . (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) It follows that (yi,j k xi,j k vλ k xi,j yi,j yi,j k xi,j (vλ s ⊗B uλ s ⊕ ... ⊕ vλ s ⊗B uλ k vλ s uλ s )!i,j yi,j k xi,j k !i,j yi,j k ⊗B (xi,j k ⊗B (xi,j yi,j k !i,j k(zi,j)i,jk ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j Xk=1 We recall the contractions fλ, (vλ s )s∈N, (uλ s )s∈N, λ ∈ Λ from Theorem 2.5. Fix λ ∈ Λ. If ǫ > 0 there exists s ∈ N such that − ǫ <(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j mi,j k fλ)!i,j Xk=1 Xk=1 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j mi,j s!i,j Xk=1 Xk=1 s ) ⊗B uλ ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j k !i,j Xk=1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j Xk=1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j Xk=1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j Xk=1 for all λ ∈ Λ. Since k !i,j k ⊗B xi,j yi,j ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j Xk=1 λ mi,j Xk=1 ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) mi,j Xk=1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) k fλ)!i,j The second dual operator space A∗∗ 0 of the operator algebra A0 is also an operator algebra with product describing in [5, section 2.5]. The product on A∗∗ 0 extends the product on A0. With this we mean that if ι : A0 → A∗∗ is the canonical embedding then ι(ab) = ι(a)ι(b) for all a, b ∈ A0. Lemma 2.8. The operator algebra A, (resp. B) is isomorphic as dual oper- ator algebra with A∗∗ 0 (resp. B∗∗ 0 k !i,j k ⊗B xi,j yi,j k fλ)!i,j yi,j k ⊗B (xi,j yi,j k ⊗B (xi,j = w∗ − lim yi,j k xi,j k !i,j yi,j k xi,j k !i,j (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ). . (cid:3) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 0 we have MORITA EQUIVALENCE OF NEST ALGEBRAS 11 Proof We denote by C1 the space of trace class operators in H1 and Ω = {c ∈ C1 : (N−)⊥cN = 0 ∀N ∈ N1}. By [7, section 16] the maps µ : C1/Ω → A∗ 0 : µ(c)(a) = tr(ca), σ : A → (C1/Ω)∗ : σ(a)(c) = tr(ac) are surjective isometries. We define the isometry φ = (µ∗)−1 ◦ σ : A → A∗∗ 0 . This map satisfies φ(a) = ι(a) for all a ∈ A0. Since i(ab) = ι(a)ι(b) for all a, b ∈ A0 and φ is w∗−continuous φ is a homomorphism onto A∗∗ 0 . (When we say w∗−continuous we mean that φ is B(H1)∗ − A∗ 0 continuous.) ments, there exists a w∗−continuous isometry If n ∈ N the algebra Mn(A) is also a nest algebra, so by the above argu- ∼ φ: Mn(A) → (Mn(A)0)∗∗ = Mn(A0)∗∗ such that ∼ ι ((ai,j)) ∼ φ ((ai,j)) = ∼ ι: Mn(A0) → Mn(A0)∗∗ is the canonical embed- for all (ai,j) ∈ Mn(A0), where ding. By [5, 1.4.11] there exists a w∗−continuous isometry τ : Mn(A0)∗∗ → Mn(A∗∗ ι ((ai,j)) = (ι(ai,j)) for all (ai,j) ∈ Mn(A0). So we have a w∗−continuous isometry τ◦ τ◦ φ: Mn(A) → Mn(A∗∗ ∼ φ ((ai,j)) = (ι(ai,j)) 0 ) such that τ ( 0 ) satisfying ∼ ∼ for all (ai,j) ∈ Mn(A0). But the map is a w∗−continuous map satisfying φn : Mn(A) → Mn(A∗∗ 0 ) : (bi,j) = (φ(bi,j)) φn((ai,j)) = (φ(ai,j)) = (ι(ai,j)) for all (ai,j) ∈ Mn(A0). So φn is equal to τ◦ Mn(A) we have φn = τ◦ φ is a completely isometric map and this completes the proof. = φ . So φn is isometry for all n ∈ N. We proved that (cid:3) ∼ ∼ φ in Mn(A0). Since Mn(A0) w∗ We are now ready to present the main theorem of this paper: Theorem 2.9. A. The following are equivalent: (i) The nests N1,N2 are isomorphic. (ii) The algebras A0, B0 are strongly Morita equivalent. (iii) The algebras A, B are weakly−∗ Morita equivalent. B. If θ : N1 → N2 is a nest isomorphism, X = Op(θ), Y = Op(θ−1) then: (i) A0 ∼= Y0 ⊗h (ii) A ∼= Y ⊗σh A0 Y0, as operator modules, A Y, as dual operator modules. B0 X0, B0 ∼= X0 ⊗h B X, B ∼= X ⊗σh 12 G. K. ELEFTHERAKIS Proof A.(i)⇒ (ii) This is Theorem 2.6. (ii)⇒(iii) If A0 and B0 are strongly Morita equivalent then the operator algebras A∗∗ 0 0 are weakly−∗ Morita equivalent, [3, section 3] . So by Lemma 2.8 A and B∗∗ and B are weakly−∗ Morita equivalent. (iii)⇒(iv) Let (A, B, V, U) be a weak−∗ Morita context [3]. It follows that there exist completely contractive separately w∗−continuous bilinear maps (·,·) : V × U → A which is A−module and B−balanced map and [·,·] : U × V → B which is B−module and A−balanced map satisfying (y, x)y′ = y[x, y′], x′(y, x) = [x′, y]x ∀x, x′ ∈ U, y, y′ ∈ V, and A = spanw({(y, x) : x ∈ U, y ∈ V }), B = spanw({[x, y] : x ∈ U, y ∈ V }). If N ∈ N1 we define θ(N) the projection onto the space generated by vectors of the form [xN, y](ω), x ∈ U, y ∈ V, ω ∈ H2. Since b[xN, y] = [bxN, y] for all b ∈ B we have θ(N)⊥Bθ(N) = 0 so θ(N) ∈ N2. Also if N1 ≤ N2 then θ(N1) ≤ θ(N2) and so θ is an order preserving map from N1 into N2. Similarly if M ∈ N1 we define σ(M) the projection onto the space generated by vectors of the form (yM, x)(ξ), x ∈ U, y ∈ V, ξ ∈ H1. The map σ : N2 → N1 is an order preserving map. If x, x′ ∈ U, y ∈ V and N ∈ N1 then θ(N)⊥[xN, y] = 0 ⇒ [θ(N)⊥xN, y] = 0 ⇒ [θ(N)⊥xN, y]x′ = 0 ⇒θ(N)⊥xN(y, x′) = 0. Since the operators (y, x′) span the algebra A we have θ(N)⊥xN = 0 ⇒ xN = θ(N)xN ∀x ∈ U, N ∈ N1. yM = σ(M)yM ∀ y ∈ V, M ∈ N2. If x, x′ ∈ U, y, y′ ∈ V, N ∈ N1 we have [xN ⊥, y][x′N, y′] = [[xN ⊥, y]x′N, y′] =[xN ⊥(y, x′)N, y′] = 0 because (y, x′) ∈ Alg(N1) It follows that [xN ⊥, y]θ(N) = 0 ⇒ [x, N ⊥yθ(N)] = 0 for all x ∈ U, y ∈ V, (2.2) Similarly (2.3) and so (2.4) N ⊥yθ(N) = 0 ⇒ yθ(N) = Nyθ(N), ∀ y ∈ V, N ∈ N1 Similarly we can prove MORITA EQUIVALENCE OF NEST ALGEBRAS 13 (2.5) xσ(M) = Mxσ(M) ∀ x ∈ U, M ∈ N2 If N ∈ N1 and x ∈ U, y ∈ V then (y, x)N = (y, xN) = (y, θ(N)xN) because of (2.2). The last operator is equal to (yθ(N), xN) = σ(θ(N))(y, x)N because of (2.3). It follows that N ≤ σ(θ(N)). of (2.4). The last operator is equal to Similarly (y, x)∗N ⊥ = (N ⊥(y, x))∗ = (N ⊥y, x)∗ = (N ⊥yθ(N)⊥, x)∗ because (N ⊥y, θ(N)⊥x)∗ = (N ⊥y, θ(N)⊥xσ(θ(N))⊥)∗ because of (2.5). The last operator is equal to σ(θ(N))⊥(N ⊥y, θ(N)⊥x)∗. k=1(yi σ(θ(N))⊥ and so N = σ(θ(N)). Since IH1 = w∗ − limiPni k) ⊂ U we have N ⊥ ≤ Similarly we can prove that M = θ(σ(M)) for all M ∈ N2. This completes k) ⊂ V, (xi the proof of the fact that θ is a nest isomorphism. k)∗ for (yi k, xi Claim B-(i) follows from Theorem 2.6. B. Let θ : N1 → N2 be a nest isomorphism and X = Op(θ), Y = Op(θ−1). Let ρ : Y ⊗σh B X → A be the map which was defined above Lemma 2.7. Let B X)) for a fixed n ∈ N. By [3, Corollary 2.8] there exists B X)) converging in w∗ topology to z. It follows → ρ(z) in Mn(A). If f, g are finite rank operators in A we denote z ∈ Ball(Mn(Y ⊗σh a net (zi) ⊂ Ball(Mn(Y ⊗h that ρ(zi) w∗ f n = f ⊕ f ⊕ ... ⊕ f and similarly for gn. We have that f nρ(zi)gn k·k → f nρ(z)gn ⇒ ρ(f nzign) k·k → ρ(f nzgn). From Lemma 2.7 it follows that f nzgn ∈ Mn(Y ⊗h B X) and kρ(f nzgn)k = kf nzgnk for all finite rank operators f, g in A. We recall the finite rank con- tractions (fλ)λ∈Λ from Theorem 2.5. For λ, µ ∈ Λ we have λ zf n kf n Since zf n µk = kρ(f n λ ρ(z)f n λ zf n µk ≤ kρ(z)k for all µ ∈ Λ. Now µ )µ∈Λ we obtain kzk ≤ kρ(z)k. We proved that the B X → A is a complete isometry. From Theorem 2.6 and its µ = w∗ − limλ f n taking the w∗−limit of (zf n map ρ : Y ⊗σh proof we have that A0 = span(Y0X0). Since A = A0 λ zf n µ )k = kf n µ we have kzf n µk ≤ kρ(z)k. we have w∗ A = spanw(Y X) = spanw({ρ(y ⊗B x) : y ∈ Y, x ∈ X}). By the Krein-Smulian Theorem the space Imρ is w∗−closed and so ρ is onto A. Similarly we can prove that B ∼= X ⊗σh A Y, as dual operator modules. (cid:3) 14 G. K. ELEFTHERAKIS 3. Spatial Morita equivalence and nest algebras In this section we shall investigate the relation between weak−∗ and spatial Morita equivalence for nest algebras. We give the definition of spatial Morita equivalence: Definition 3.1. (I. G. Todorov) Let C, D be w∗−closed algebras acting on the Hilbert spaces K1, K2 respectively. We say that C and D are spatially Morita equivalent if there exists a D − C bimodule V ⊂ B(K1, K2) and a C−D bimodule U ⊂ B(K2, K1) such that C = spanw(UV ), D = spanw(V U). We also need the following notions. If L is a set of projections acting on the Hilbert space H the set Alg(L) = {x ∈ B(H) : p⊥xp = 0, ∀ p ∈ L} is an algebra. An algebra A is called reflexive if there exists a set of projections L such that A = Alg(L). In the special case where L is a complete lattice of commuting projections containing the zero and identity operators the algebra Alg(L) is called a CSL algebra and the lattice L is called a CSL lattice. Obviously, nest algebras are CSL algebras. If A is an algebra acting on the Hilbert space H the lattice {p ∈ pr(B(H)) : p⊥xp = 0, ∀ x ∈ A} is called the lattice of A and we denote it by Lat(A). If L is a CSL lattice then Lat(Alg(L)) = L, [1], [8]. Two spatially Morita equivalent algebras are not always weakly−∗ Morita equivalent even in the case one of them is a CSL algebra: Example 3.1. Let C be a nest algebra. We denote the algebras A = C ⊕ C and B = {(cid:18) a b − a 0 b (cid:19) : a, b ∈ C}. Observe that A is a CSL algebra whose lattice is Lat(A) = {p ⊕ q : p, q ∈ Lat(C)}. Since the center of C is trivial [7, Corollary 19.5] the center of A is Z(A) = C ⊕ C and the center of B is Z(B) = {(cid:18) λ µ − λ 0 µ (cid:19) : λ, µ ∈ C}. We also denote the spaces X = {(cid:18) a −a b (cid:19) : a, b ∈ C}, Y = {(cid:18) a b 0 b (cid:19) : a, b ∈ C}. 0 MORITA EQUIVALENCE OF NEST ALGEBRAS 15 We can check that X is an A − B bimodule, Y is a B − A bimodule and XY = A, Y X = B. So the algebras A, B are spatially Morita equivalent. If A and B were weakly−∗ Morita equivalent by [3, Theorem 3.7] they would have isomorphic centers through a completely isometric homomorphism. This is a contradiction because Z(A) is a von Neumann algebra and Z(B) is a non- selfadjoint algebra. Despite the above example, in [10] we proved that two CSL algebras are spatially Morita equivalent if and only if their lattices are isomorphic, so by Theorem 2.9 we conclude the following theorem: Theorem 3.2. Two nest algebras are spatially Morita equivalent if and only if they are weakly−∗ Morita equivalent. Also despite the example 3.1 we have the following theorem: Theorem 3.3. Let A be a nest algebra, B be a unital dual operator algebra and β be a completely isometric normal representation of B such that A and β(B) are spatially Morita equivalent. It follows that A and B are weakly−∗ Morita equivalent. (cid:3) Proof By [10, Theorem 4.1, remark 4.2] β(B) is a nest algebra whose nest is isomorphic with the nest of A. The conclusion comes from Theorems 2.9 and 3.2. Theorem 3.4. (Blecher-Kashyap) If A, B are weakly−∗ Morita equivalent unital dual operator algebras, for every completely isometric normal repre- sentation α of A there exists a completely isometric normal representation β of B such that the algebras α(A), β(B) are spatially Morita equivalent. Proof Suppose that (A, B, X, Y ) is a weakly−∗ Morita context [3]. We use now arguments from the beginning of the 4th section of [3]. If α is a completely isometric normal representation of A on the Hilbert space H the tensor product K = Y ⊗σh A H with its norm is a Hilbert space on which B is represented through the w∗−continuous complete isometry β given by β(b)(y ⊗ h) = (by) ⊗ h ∀ b ∈ B, y ∈ Y, h ∈ H. Also Blecher and Kashyap prove that the maps φ : Y → B(H, K), ψ : X → B(K, H) given by φ(y)(h) = y ⊗ h and ψ(x)(y ⊗ h) = α((x, y))(h) are w∗−continuous complete isometries. See in [3] for the properties of the bilin- ear map (·,·) : X × Y → A. We can easily check that ψ(X) is an α(A)− β(B) bimodule, φ(Y ) is a β(B) − α(A) bimodule and α(A) = spanw(ψ(X)φ(Y )), β(B) = spanw(φ(Y )ψ(X)). (cid:3) Corollary 3.5. If A is a nest algebra and B is a unital dual operator algebra which are weakly−∗ Morita equivalent then there exists a completely isometric normal representation β of B such that β(B) is a nest algebra. 16 G. K. ELEFTHERAKIS Proof By the above Theorem there exists a completely isometric normal representation β of B such that the algebras A and β(B) are spatially Morita equivalent. From [10, remark 4.1] the algebra β(B) is reflexive and from [10, Theorem 4.2] the lattice of β(B) is isomorphic with the nest of A. So β(B) is a nest algebra. (cid:3) Corollary 3.6. If A is a CSL algebra which is not a nest algebra then A is not weakly−∗ Morita equivalent with anyone nest algebra. Proof By the above corollary if A was weakly−∗ Morita equivalent with a nest algebra then it would have a normal completely isometric representation α such that α(A) is a nest algebra. This is a contradiction because as we can easily check α(Lat(A)) = Lat(α(A)). (cid:3) 4. A stable isomorphism theorem for nest algebras In this section we are going to present a new theorem which characterizes the stable isomorphism of separably acting nest algebras. Definition 4.1. Two dual operator algebras C, D are called stably iso- morphic if there exists a Hilbert space H and a completely isometric, w∗- − ⊗ bicontinuous isomorphism from the algebra C ⊗ B(H) onto the algebra D − B(H), where − ⊗ is the normal spatial tensor product. We give two relevant definitions: Definition 4.2. [10] Let C, D be w∗ closed algebras acting on Hilbert spaces H1 and H2 respectively. If there exists a TRO M ⊂ B(H1, H2), i.e. a sub- space satisfying MM∗M ⊂ M, such that C = spanw(M∗DM) and D = spanw(MCM∗) we write C M∼ D. We say that the algebras C, D are TRO equivalent if there exists a TRO M such that C M∼ D. Definition 4.3. [11] Let C, D be abstract dual operator algebras. These al- gebras are called ∆−equivalent if they have completely isometric normal representations φ, ψ such that the algebras φ(C), ψ(D) are TRO-equivalent. In [13] we proved the following theorem: Theorem 4.1. Two unital dual operator algebras are stably isomorphic if and only if they are ∆−equivalent. ∆−equivalence implies weak−∗ Morita equivalence [3, section 3]. The con- verse does not hold. The counterexample is [12, example 3.7]. We shall give a new proof of this fact in Theorem 4.7. [12, Theorem 3.2] implies the following corollary: MORITA EQUIVALENCE OF NEST ALGEBRAS 17 Corollary 4.2. Two nest algebras are ∆−equivalent if and only if they are TRO-equivalent. In what follows if X is a subset of B(H) where H is a Hilbert space we denote by X ′ the commutant of X and by X ′′ the algebra (X ′)′. In [10] we proved the following criterion of TRO-equivalence for reflexive algebras: Theorem 4.3. Two reflexive algebras C, D are TRO-equivalent if and only if there exists a ∗−isomorphism δ : (C∩C ∗)′ → (D∩D∗)′ such that δ(Lat(C)) = Lat(D). Comparing Theorems 4.1, 4.3 and Corollary 4.2 we take the following: 1 → N ′′ Corollary 4.4. The nest algebras Alg(N1), Alg(N2) are stably isomorphic if and only if there exists a ∗−isomorphism δ : N ′′ 2 such that δ(N1) = N2. In the rest of this section we fix two nests N1,N2 acting on the separable Hilbert spaces H1, H2 respectively and we denote A = Alg(N1), B = Alg(N2). We use now extensively notions from [7, section 7]. If ξ (resp. ω) is a unit separating vector for the algebra N ′′ 2 ) we define the order isomor- phism φξ (resp. ψω) from N1 (resp. N2) onto a closed subset of the interval [0, 1] given by φξ(N) = kN(ξ)k2 (resp. ψω(M) = kM(ω)k2). Suppose that [0, 1]\ φξ(N1) = ∪n(ln, rn) and [0, 1]\ ψω(N2) = ∪n(tn, sn). If m is the Lebesgue measure we define the measures µξ, νω given by 1 (resp. N ′′ µξ(S) = m(S ∩ φξ(N1)) + Xrn∈S νω(S) = m(S ∩ ψω(N2)) + Xsn∈S (rn − ln) (sn − tn), for every Borel subset S of [0, 1]. We denote M1 (resp. M2) the nest {Ms : 0 ≤ s ≤ 1} ⊂ B(L2([0, 1], µξ)) (resp. {Ns : 0 ≤ s ≤ 1} ⊂ B(L2([0, 1], νω))) where Ms (resp. Ns) is the projection onto the space L2([0, s], µξ) (resp. L2([0, s], νω)). The algebra N ′′ is ∗−isomorphic with the algebra L∞([0, 1], µξ) (resp. L∞([0, 1], νω)) acting on the Hilbert space L2([0, 1], µξ) (resp. L2([0, 1], νω)) through an isomorphism mapping the nest N1 (resp. N2) onto M1 (resp. M2). We denote by AbsHom([0, 1]) the set of order homeomorphisms α : [0, 1] → [0, 1] which satisfy the property m(S) = 0 ⇒ m(α(S)) = 0. The theorem be- low describes when two separably acting nest algebras are stably isomorphic. 1 Theorem 4.5. The algebras A, B are stably isomorphic if and only if there exist separating unit vectors ξ for N ′′ 2 and α ∈ AbsHom([0, 1]) such that α(φξ(N1)) = ψω(N2). 1 , ω for N ′′ 18 G. K. ELEFTHERAKIS Proof Suppose that the algebras A, B are stably isomorphic. From Corol- 2 such that δ(N1) = N2. 2 . Taking compositions we lary 4.4 there exists a ∗−isomorphism δ : N ′′ Fix separating unit vectors ξ for N ′′ obtain a ∗−isomorphism 1 → N ′′ 1 , and ω for N ′′ ∼ δ: L∞([0, 1], µξ) → L∞([0, 1], νω) ∼ δ (M1) = M2. Every isomorphism between maximal abelian self- such that adjoint algebras is implementing by a unitary. So the nests M1,M2 are unitarily equivalent. By [7, Theorem 7.23] there exists α ∈ AbsHom([0, 1]) such that α(φξ(N1)) = ψω(N2). Conversely if there exist such ξ, ω and α, by the same theorem there ex- ists a unitary u ∈ B(L2([0, 1], µξ), L2([0, 1], νω)) such that u∗M2u = M1. It follows that L∞([0, 1], µξ) = u∗L∞([0, 1], νω)u. Taking compositions we take a ∗−isomorphism δ : N ′′ 2 such that δ(N1) = N2. Again from Corollary 4.4 we conclude that the algebras A and B are stably isomorphic. Remark 4.6. If there exist separating unit vectors ξ for N ′′ 2 and α ∈ AbsHom([0, 1]) such that α(φξ(N1)) = ψω(N2) then for all separating unit vectors ξ1 for N ′′ 2 there exists α1 ∈ AbsHom([0, 1]) such that α1(φξ1(N1)) = ψω1(N2). This is a consequence of [7, Proposition 7.22]. 1 and ω1 for N ′′ 1 , ω for N ′′ 1 → N ′′ (cid:3) We give a new proof of the following result: Theorem 4.7. Weak−∗ Morita equivalence is strictly weaker than ∆−equivalence. Proof Let C be the Cantor set, γ be an order homeomorphism of [0, 1] such that m(γ(C)) > 0. Suppose that [0, 1]\C = ∪n(ln, rn) and [0, 1]\γ(C) = ∪n(tn, sn). We denote by µ the measure µ(S) = Xrn∈S (rn − ln) and by ν the measure ν(S) = m(S ∩ γ(C)) + Xsn∈S (sn − tn). We denote M1 (resp. M2) the nest {Ms : 0 ≤ s ≤ 1} ⊂ B(L2([0, 1], µ)) (resp. {Ns : 0 ≤ s ≤ 1} ⊂ B(L2([0, 1], ν))) where Ms (resp. Ns) is the projection onto the space L2([0, s], µ) (resp. L2([0, s], ν)). The map θ : M1 → M2 : Ms → Nγ(s) is a nest isomorphism so by Theorem 2.9 the algebras A = Alg(M1), B = Alg(M2) are weakly−∗ Morita equiv- alent. If the algebras A, B were ∆−equivalent by Theorem 4.5 there would exist unit vectors ξ for M′′ 2 and α ∈ AbsHom([0, 1]) such that α(φξ(M1)) = ψω(M2). From [7, Proposition 7.22] we have that m(φξ(M1)) = 1, ω for M′′ MORITA EQUIVALENCE OF NEST ALGEBRAS 19 m(C) = 0 and since m(γ(C)) > 0 we have that m(ψω(M2)) > 0. This is a contradiction. (cid:3) 5. A counterexample in Morita equivalence In this section we shall use the notions of TRO equivalence, of ∆−equivalence, of stable isomorphism and we shall consider nest and CSL algebras. See the appropriate definitions in sections 1, 3 and 4. If C and D are unital opera- tor algebras which are strongly Morita equivalent then for every ǫ > 0 there exists a completely bounded isomorphism from C ⊗min K onto D⊗min K with kρkcb < 1 + ǫ and kρ−1kcb < 1 + ǫ, where K is the C ∗−algebra of compact operators on a separable infinite dimensional Hilbert space H and ⊗min is the spatial tensor product [6, Corollary 7.10]. It follows that for every ǫ > 0 there exists a completely bounded w∗−continuous isomorphism σ from C ∗∗ ⊗ B(H) onto D∗∗ ⊗ B(H) with kσkcb < 1 + ǫ and kσ−1kcb < 1 + ǫ, where ⊗ is the normal spatial tensor product. One can wonder now, if the operator algebras C ∗∗ and D∗∗ are stably isomorphic. − − − In this section we give a negative answer to this question. We present a counterexample of unital strongly Morita equivalent algebras C and D whose second duals are not stably isomorphic. Also for the algebras C ∗∗ and D∗∗ there exist normal completely isometric representations φ and ψ respectively such that for every ǫ > 0 there exists an invertible bounded operator Tǫ ǫ ψ(D∗∗)Tǫ and satisfying kTǫk < 1 + ǫ, φ(C) = T −1 ǫ k < 1 + ǫ, φ(C ∗∗) = T −1 kT −1 ǫ ψ(D)Tǫ. Two nests N ,M acting on the separable Hilbert spaces H, K respectively are called similar if there exists an order isomorphism θ : N → M which preserves dimension of intervals. We say that an invertible operator S ∈ B(H, K) implements θ if θ(N) is the projection onto SN(H) for all N ∈ N . In what follows if C is an operator algebra, ∆(C) is its diagonal C ∩ C ∗. We fix similar nests N ,M as above with corresponding nest algebras A = Alg(N ) and B = Alg(M) such that ∆(A) is a totally atomic maximal abelian selfadjoint algebra (masa in sequel) and ∆(B) is a masa with a nontrivial continuous part, [7, example 13.15]. Suppose that θ : N → M is an order isomorphism implementing similarity for N ,M. We denote by A0 (resp. B0) the algebra of compact operators belonging to A (resp. B) and by A1 (resp. B1) the operator algebra A0 + CIH (resp. B0 + CIK ). We denote by X the space Op(θ) and by Y the space Op(θ−1). Theorem 5.1. [7, Theorem 13.20](Davidson) For every ǫ > 0 there exists an invertible bounded operator Sǫ which implements θ such that kSǫk < 1 + ǫ,kS−1 ǫ k < 1 + ǫ. (Observe that Sǫ ∈ X and S−1 ǫ ∈ Y for all ǫ > 0.) 20 G. K. ELEFTHERAKIS Suppose that j : A1 → A∗∗ w∗ 1 the space j(A0) . is the canonical embedding. We denote by JA 1 = JA + CI Lemma 5.2. (i) A∗∗ (ii) JA ∩ CI = 0. Proof (i) Since λ ≤ ka + λIHk for all compact operators a the functional ρ : A1 → C : a + λIH → λ belongs to A∗ 1 by the Goldstine Theorem there exists a net (ai + λiIH) ⊂ A0 + CI converging in w∗−topology to x. Since (λi) converges to ρ(x) we have that (ai) converges to a ∈ JA and so x = a + ρ(x) ∈ JA + CI. 1. If x ∈ A∗∗ (ii) Since ρA0 = 0 if λI ∈ JA then λ = 0. So JA ∩ CI = 0. Suppose that ι : A0 → A∗∗ is the canonical embedding. In lemma 2.8 we have proved that there exists a w∗−continuous completely isometric onto homomorphism φ : A → A∗∗ 0 extending ι. 0 extends to a w∗−continuous completely con- The map φA1 : A1 → A∗∗ 1 → A∗∗ φ (j(a)) = φ(a) for all a ∈ A1. Also satisfying tractive map 1 extends to a w∗−continuous the completely contractive map jA0 : A0 → A∗∗ ∧ κ (ι(a)) = j(a) for all completely contractive map a ∈ A0. So the map ∧ κ: A∗∗ 0 → A∗∗ 0 → A∗∗ 0 satisfies ∧ φ: A∗∗ such that ∧ κ: A∗∗ (cid:3) ∧ ∧ 0 1 0 ∧ κ (ι(a)) = ∧ φ (j(a)) = φ(a) = ι(a) ∧ φ ◦ φ ◦ ∧ ∧ κ is a complete isometry. We denote by θ the w∗−continuous completely isometric homomorphism 0 . Therefore ∧ κ= idA∗∗ for all a ∈ A0. It follows that κ ◦φ : A → A∗∗ 1 . Observe that φ ◦ ∧ θ(A) = ∧ κ (φ(A)) = ∧ κ (ι(A0) w∗ ) = j(A0) w∗ = JA. Suppose that p is the projection θ(idA). Lemma 5.2 implies that p⊥ 6= 0 and 1 = JA ⊕ Cp⊥. A∗∗ Lemma 5.3. The algebra A∗∗ 1 isomorphic with the algebra A ⊕ C acting on the Hilbert space H ⊕ C. is completely isometric and w∗− continuously Proof We define the map θ and the projection p as in the above discussion. We define the completely isometric normal representation π : A∗∗ 1 = JA ⊕ Cp⊥ → B(H ⊕ C) : a ⊕ λp⊥ → θ−1(a) ⊕ λ which is onto A ⊕ C. (cid:3) MORITA EQUIVALENCE OF NEST ALGEBRAS 21 For every ǫ > 0 we denote by Tǫ the bounded invertible operator Sǫ⊕ idC ∈ B(H ⊕ C, K ⊕ C). Also we denote the spaces U = X ⊕ C ⊂ B(H ⊕ C, K ⊕ C) and V = Y ⊕ C ⊂ B(K ⊕ C, H ⊕ C). Observe that U is a B ⊕ C − A ⊕ C bimodule and V is an A ⊕ C − B ⊕ C bimodule. is the canonical embedding we have π(j(a)) = a ⊕ 0 for all a ∈ A0 and π(j(idA1)) = idH⊕C. So 1 ) = A ⊕ C. If j : A1 → A∗∗ By the above lemma π(A∗∗ 1 π(j(A1)) = span{a ⊕ 0, idH⊕C, a ∈ A0}. Similarly if j2 : B1 → B∗∗ 1 completely isometric onto homomorphism ρ : B∗∗ is the canonical embedding there exists a normal 1 → B ⊕ C such that ρ(j2(B1)) = span{b ⊕ 0, idK⊕C, b ∈ B0}. Since S−1 ǫ B0Sǫ = A0 and S−1 ǫ BSǫ = A we have that ǫ ρ(j2(B1))Tǫ = π(j(A1)), T −1 T −1 ǫ ρ(B∗∗ 1 )Tǫ = π(A∗∗ 1 ) for all ǫ > 0. In the following lemmas 5.4, 5.5 we identify the algebra A∗∗ 1 with A ⊕ C, 1 with B ⊕ C, the algebra A1 with π(j(A1)) and the algebra the algebra B∗∗ B1 with ρ(j2(B1)). Lemma 5.4. The algebras A∗∗ 1 and B∗∗ 1 are weakly−∗ Morita equivalent. Proof Let U, V and Tǫ, ǫ > 0 be as in the above discussion. The com- : (v, u) → vu is separately 1 −module map. So induces the w∗- pletely contractive bilinear map V × U → A∗∗ w∗-continuous, B∗∗ continuous completely contractive and A∗∗ 1 1 −balanced and A∗∗ U → A∗∗ τ : V ⊗σh B∗∗ 1 1 −module map 1 u → vu. 1 : v ⊗B∗∗ We shall prove that τ is isometric: If (vi) ⊂ V, (ui) ⊂ U and ǫ > 0 we have: ǫ Tǫvi) ⊗B∗∗ 1 n n n (T −1 vi ⊗B∗∗ the last norm is equal with (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=1 Since Tǫvi ∈ UV ⊂ B∗∗ (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=1 T −1 ǫ ⊗B∗∗ ≤ kT −1 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 1 ui(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=1 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 1 (Tǫviui)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ǫ ⊗B∗∗ ≤ (1 + ǫ)2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) viui(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ǫ kkTǫk(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=1 Xi=1 We let ǫ → 0 and we have that viui(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 1 ui(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=1 Xi=1 vi ⊗B∗∗ (T −1 n n n n n . 1 ui(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) viui)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=1 viui(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) . . 1 Tǫ)( 22 G. K. ELEFTHERAKIS Similarly we can prove that τ is completely isometric. Since A = spanw(Y X) we have that A∗∗ 1 = spanw(V U) and so by the Krein-Smulian Theorem τ is onto A∗∗ 1 . The proof of the fact B∗∗ V is similar. A∗∗ (cid:3) 1 ∼= U ⊗σh 1 Lemma 5.5. The algebras A1 and B1 are strongly Morita equivalent. Proof It suffices to prove that they have equivalent categories of left op- erator modules [2]. If C is an operator algebra we denote by Cmod the category of left operator modules over C. We assume that every Z ∈ Cmod is essential, i.e. the linear span of CZ is dense in Z. If Z1, Z2 ∈ Cmod the space of morphisms HomC(Z1, Z2) is the space of completely bounded maps F : Z1 → Z2 which are C−module maps. dual operator module over A∗∗ 1 Z ∗∗ F (Z) the subspace of U ⊗σh We fix an operator T = Tǫ0 for ǫ0 > 0. If Z ∈ A1mod then Z ∗∗ is a left in a canonical way [5, 3.8.9]. We denote by A∗∗ 1 Since U ⊗σh A∗∗ 1 F (Z) = span(T a ⊗A∗∗ 1 z : a ∈ A1, z ∈ Z). Z ∗∗ is a left operator module over B∗∗ 1 and b(T a ⊗A∗∗ 1 z) = (bT a) ⊗A∗∗ 1 z = T (T −1bT a) ⊗A∗∗ 1 z with T −1bT ∈ A1 for all b ∈ B1, F (Z) is a left operator B1−module. If W ∈ B1mod we denote by G(W ) the subspace of V ⊗σh 1 w : a ∈ A1, w ∈ W ). G(W ) = span(aT −1 ⊗B∗∗ W ∗∗ B∗∗ 1 Since V ⊗σh 1 B∗∗ Now W ∗∗ is a left operator module over A∗∗ 1 , clearly G(W ) ∈ A1mod. G(F (Z)) = span(a2T −1 ⊗B∗∗ 1 T a1 ⊗A∗∗ 1 z : a1, a2 ∈ A1, z ∈ Z) is a left operator module over A1 and subspace of the space V ⊗σh The w∗−Morita equivalence A∗∗ Theorem 3.5]) a complete isometry 1 ∼= U ⊗σh 1 ∼= V ⊗σh U, B∗∗ B∗∗ A∗∗ 1 1 U ⊗σh Z ∗∗. B∗∗ V induces ([3, A∗∗ 1 1 1 1 A∗∗ B∗∗ U ⊗σh V ⊗σh Z ∗∗ → Z ∗∗ : v ⊗B∗∗ which restricts to a completely isometric map RZ : G(F (Z)) → Z : a2T −1 ⊗B∗∗ 1 T a1 ⊗A∗∗ for all a1, a2 ∈ A1, z ∈ Z. This map is clearly onto Z. 1 u ⊗A∗∗ Every morphism F ∈ HomA1(Z1, Z2) can be extended to a morphism 2 ), the space of w∗−continuous completely ∧ F belonging to Homσ bounded A∗∗ 1 −module maps. (Use for example [5, 1.4.8]). 1 , Z ∗∗ (Z ∗∗ A∗∗ 1 1 z → vuz 1 z → a2a1z ∧ ∧ F ( 1 z) = T a ⊗A∗∗ F )(T a ⊗A∗∗ F ( ∧ ∧ 1 F (z) F ) maps F (Z1) into F (Z2). So we can for all a ∈ A1, z ∈ Z1 the operator define ∧ ∧ MORITA EQUIVALENCE OF NEST ALGEBRAS 23 The weak−∗ Morita equivalence A∗∗ 1 ∼= V ⊗σh 1 B∗∗ U, B∗∗ 1 ∼= U ⊗σh erates ([3, Theorem 3.5]) a normal completely contractive functor the left dual operator modules of A∗∗ 1 such that 1 and B∗∗ 1 A∗∗ ∧ V, gen- F between ∧ F ( Since ∧ F ) : U ⊗σh A∗∗ 1 Z ∗∗ 1 → U ⊗σh 1 A∗∗ Z ∗∗ 2 : u ⊗A∗∗ 1 z → u⊗A∗∗ 1 ∧ F (z). F ( F (F ) = F )F (Z1) : F (Z1) → F (Z2). We can easily check that F (F ) ∈ HomB1(F (Z1),F (Z2)). In this way we define functors F : A1mod → B1mod and G : B1mod → A1mod. Using the above complete isometries {RZ : Z ∈ A1mod} we can prove that the functor GF is equivalent to the identity functor 1 A1 mod and the functor FG is equivalent to the identity functor 1 B1 mod. Theorem 5.6. Strong Morita equivalence of unital operator algebras doesn't imply ∆−equivalence of the second dual operator algebras. (cid:3) Proof We recall the unital operator algebras A1, B1 which are strongly Morita equivalent by the above lemma. We shall prove that the algebras A∗∗ 1 are not ∆−equivalent. Suppose that they are ∆−equivalent. We define the completely isometric normal representation (see Lemma 5.3) 1 , B∗∗ The algebra π(A∗∗ π : A∗∗ 1 → B(H ⊕ C) : a ⊕ λp⊥ → θ−1(a) ⊕ λ. 1 ) = A ⊕ C is a CSL algebra with lattice {N ⊕ 0, N ⊕ C : N ∈ N}. 1 ), σ(B∗∗ 1 ) are TRO equivalent. Since π(A∗∗ 1 = JB ⊕ Cq⊥ where q is the identity of the algebra JB Suppose that B∗∗ and JB is isomorphic with the algebra B. By [12, Theorem 2.7] there exists a completely isometric normal representation σ of B∗∗ 1 on a Hilbert space K1 ⊕ K2 of the form σ(b ⊕ λq⊥) = σ1(b) ⊕ λIK2 for all b ∈ JB, λ ∈ C such that the algebras π(A∗∗ 1 ) is a CSL 1 ) is also a CSL algebra, [10, Remark 5.5]. So the algebra σ(B∗∗ algebra, σ(B∗∗ 1 ) contains a masa. It follows that dimK2 = 1. So we may assume that σ(B∗∗ 1 ) is a CSL algebra acting on K1 ⊕ C. 1 ))) is also a masa. The algebras ∆(π(A∗∗ 1 )) are TRO equivalent [10, Proposition 2.5]. But TRO equivalence between masas is a unitary equivalence (use for example [10, Theorem 3.2]). This is a contradiction because ∆(π(A∗∗ 1 )) = ∆(A) ⊕ C is a totally atomic masa and the masa Since ∆(A) (resp. ∆(B)) is a masa, then ∆(π(A∗∗ 1 )) (resp. ∆(σ(B∗∗ 1 )), ∆(σ(B∗∗ 24 G. K. ELEFTHERAKIS ∆(σ(B∗∗ 1 , B∗∗ A∗∗ 1 )) ∼= ∆(B) ⊕ C has a nontrivial continuous part. So the algebras 1 are not ∆−equivalent. (cid:3) Acknowledgement: I wish to thank Prof. D. Blecher who pointed out the problem in section 5 to me. References [1] W. B. Arveson, Operator algebras and invariant subspaces, Ann. of Math. 100 (1974) 433-532. [2] D. P. Blecher, A Morita theorem for algebras of operators on Hilbert space, J. Pure Appl. Algebra 156 (2001), 156-169. [3] D. P. Blecher and U. Kashyap, Morita equivalence of dual operator algebras, J. Pure Appl. Algebra, 212 (2008), 2401-2412. [4] D. P. Blecher and J. E. Kraus, On a generalization of W ∗−modules, Arxiv: 0910.5404 [5] D. P. Blecher and C. Le Merdy, Operator algebras and their modules -- an operator space approach, Oxford University Press, 2004 [6] D. P. Blecher, P. S. Muhly and V. I. Paulsen, Categories of Operator Modules (Morita Equivalence and Projective Modules), Mem. Amer. Math. Soc. 143 (2000), no.681. [7] K. R. Davidson, Nest algebras, Longman Scientific & Technical, Harlow, 1988 [8] K. R. Davidson, Commutative subspace lattices, Indiana Univ. Math. J. 27 (3) (1978) 479-490. [9] E. Effros and Z.-J. Ruan, Operator Spaces, Clarendon Press, Oxford, 2000 [10] G. K. Eleftherakis, TRO equivalent algebras, Houston J. of Mathematics, to appear. [11] G. K. Eleftherakis, A Morita type equivalence for dual operator algebras, J. Pure Appl. Algebra 212 (2008) no 5, 1060-1071 [12] G. K. Eleftherakis, Morita type equivalences and reflexive algebras, J. Operator Theory, to appear [13] G. K. Eleftherakis and V. I. Paulsen, Stably isomorphic dual operator algebras, Math. Ann. 341 (2008), no 1, 99-112 [14] U. Kashyap, A Morita theorem for dual operator algebras, J. Funct. Analysis, 256 (2009) 3545-3567. [15] V. I. Paulsen, Completely Bounded Maps and Operator Algebras, Cambridge University Press, 2002 [16] G. Pisier, Introduction to Operator Space Theory, Cambridge University Press, 2003 [17] M. Rieffel, Morita equivalence for C-algebras and W-algebras, J. Pure Appl. Algebra 5 (1974), 51-96 Department of Mathematics, University of Athens, Panepistimioupolis 157 84, Athens, Greece. E-mail address: [email protected]
1212.0715	2	1212	2013-07-15T09:02:49	Purely infinite crossed products by endomorphisms	[ "math.OA" ]	We study the crossed product $C^$-algebra associated to injective endomorphisms, which turns out to be equivalent to study the crossed product by the dilated autormorphism. We prove that the dilation of the Bernoulli $p$-shift endomorphism is topologically free. As a consequence, we have a way to twist any endomorphism of a $\D$-absorbing $C^$-algebra into one whose dilated automorphism is essentially free and have the same $K$-theory map than the original one. This allows us to construct purely infinite crossed products $C^*$-algebras with diverse ideal structures.	math.OA	math	PURELY INFINITE CROSSED PRODUCTS BY ENDOMORPHISMS EDUARD ORTEGA AND ENRIQUE PARDO To Ingvar Ortega Redalen Abstract. We study the crossed product C ∗-algebra associated to injective endomorphisms, which turns out to be equivalent to study the crossed product by the dilated autormorphism. We prove that the dilation of the Bernoulli p-shift endomorphism is topologically free. As a consequence, we have a way to twist any endomorphism of a D-absorbing C ∗-algebra into one whose dilated automorphism is essentially free and have the same K-theory map than the original one. This allows us to construct purely infinite crossed products C ∗-algebras with diverse ideal structures. Introduction The study of group actions on C ∗-algebras has been largely developed by several authors during the last years. To this end, a key strategy is to characterize properties of the crossed product C ∗-algebra by looking at the dynamical properties of the action. This is very intuitive in the commutative case, but becomes more subtle when the C ∗-algebra is non-commutative. An intensive work on that area was done by Olesen and Pedersen in [14], where they explored a certain subset of the dual group Γ of G, called the Connes' Spectrum, which measured the obstruction of the automorphism to be inner. In this way, they were able to characterize when certain crossed products C ∗-algebra were simple. An associated problem is to examine conditions on the group action so that the ideals in the crossed product are separated by the base algebra; this allows us to have control on the ideal structure of the crossed product. Renault [19] stated implicitly that essential freeness of a group G acting on a C ∗-algebra A might be enough to guarantee that A separates the ideals of the reduced crossed product A ⋊r G. Sierakowski [21] studied this problem and presented another way of ensuring this separation property on A ⋊r G. For, he defined a generalized version of the Rokhlin property, which he called the residual Rokhlin* property. Hence, he showed that A separates the ideals in A ⋊r G provided that the action of G on A is exact and satisfy the residual Rokhlin* property. In [6], Cuntz defined the fundamental Cuntz algebras On. He also represented these al- gebras as crossed products of a UHF-algebra by an endomorphism, and he used this rep- resentation to prove the simplicity of these algebras. He saw this construction as a full 2010 Mathematics Subject Classification. Primary 46L35; Secondary 06A12, 06F05, 46L80. Key words and phrases. Crossed product, dilation, shift endomorphism, Rokhlin property, purely infinite. This research was supported by the NordForsk Research Network "Operator Algebras and Dynamics" (grant #11580). The second author was partially supported by PAI III grants FQM-298 and P07-FQM- 7156 of the Junta de Andaluc´ıa. Both authors were partially supported by the DGI-MICINN and European Regional Development Fund, jointly, through Project MTM2011-28992-C02-02 and by 2009 SGR 1389 grant of the Comissionat per Universitats i Recerca de la Generalitat de Catalunya. 1 2 EDUARD ORTEGA AND ENRIQUE PARDO corner of an ordinary crossed product. Later, Paschke [15] gave an elegant generalization of Cuntz's results, and described the crossed product of a unital C ∗-algebra by an endomor- phism α : A → A, written A ×α N, as the C ∗-algebra generated by A and an isometry V , such that V aV ∗ = α(a). Endomorphisms of C ∗-algebras appeared elsewhere, and this led Stacey to give a modern description of their crossed products in terms of covariant representations and universal properties [22]. As noticed by Cuntz, when the endomorphism is injective, it is possible to transform it into automorphism and the isometries into unitaries via a direct limit construction. So, the crossed product by an endomorphism can be seen as a full corner of a crossed product by an automorphism. This allows to import results from the well-developed theory of crossed products by groups. That fact can be extended to actions by cancellative semigroups; for a more general exposition look at [13] and the references therein. Along his work, Cuntz defined purely infinite simple C ∗-algebras, and he proved that the Cuntz algebras On are among those C ∗-algebras. Purely infinite simple Cuntz-Krieger algebras (a generalization of Cuntz algebras introduced in [7]) where classified by using K- Theoretical invariants [18]. However, the most spectacular result about purely C ∗-algebras came from works of E. Kirchberg and N.C. Phillips [11, 17], who proved that separable, unital purely infinite simple C ∗-algebras in the bootstrap class are classified (up to isomorphism) by their K-theory. There were some definitions that generalize the notion of purely infinite C ∗-algebras to the non-simple case. The most accepted, and useful, was due to Kirchberg and Rørdam [12]. In [11] Kirchberg gave also some nice classification results for non-simple purely infinite C ∗-algebras through bivariant KK-theory. Our main goals in this paper are to give conditions on an endomorphism α ∈ End(A) to guarantee that: (1) A separates ideals of A ×α N; (2) A ×α N is purely infinite. Our fundamental technique is seeing A ×α N as a full corner of a crossed product of ¯A ×¯α Z, where ¯A is the dilation of A by α. After reducing the situation of a crossed product by an automorphism, we will use results of Sierakowski [21], and of Pasnicu and Rørdam [16]. We will also give a concrete example of endomorphism satisfying the residual Rokhlin* property: it is the so called shift endomorphism, that was constructed by Cuntz [6] and later studied by Dykema and Rørdam [8]. The contents of this paper can be summarized as follows: In Section 1, we introduce basic notation, and we construct the dilation of an injective endomorphism. We then review results on actions by a single automorphism, i.e. actions of Z, and we show equivalent conditions for this action being topological and essentially free. We finally recall the residual Rokhlin* property defined by Sierakowski. In Section 2, we prove that the dilation of the shift endomorphism is topologically free. In particular, given a strongly self-absorbing C ∗- algebra with a non-trivial projection D, the shift endomorphism is topologically free. Even more, this allows us to twist any endomorphism on a D-absorbing C ∗-algebra into one that is essentially free and that induces the same K-theory map. In Section 3, we give conditions for a crossed product by a single endomorphism being (non-simple) purely infinite. Finally, in Section 4, we construct various interesting examples of simple and non-simple purely infinite C ∗-algebras. In particular, in an easy way, we construct purely infinite C ∗-algebras with torsion in their K1 groups. Throughout the article we will denote by N the subsemigroup of Z given by {1, 2, . . .} and Z+ the monoid N ∪ {0}. PURELY INFINITE CROSSED PRODUCTS BY ENDOMORPHISMS 3 1. N and Z actions on C ∗-algebras Every injective endomorhism α of a C ∗-algebra A can be canonically dilated to an auto- morphism ¯α into a bigger C ∗-algebra ¯A, and therefore one has an intuitive way to extend the spectral theory of automorphisms to endomorphisms. We will see later that this is the correct one since the associated crossed products are strongly Morita equivalent. Given an endomorphism α : A −→ A, we define the inductive system {Ai, γi}i∈N given by Ai := A and γi = α for every i ∈ N. Let ¯A := lim−→{A, α} the dilation of A by α. For any i ∈ N, αi,∞ : A −→ ¯A denotes the canonical map. The diagram A α / A α / A α / · · · α A α / α α / A α / / A α / / · · · / ¯A ¯α / ¯A gives rise to an automorphism ¯α : ¯A −→ ¯A, that is called the dilation of α. Recall that given an automorphism α ∈ Aut (A) we can define the spectrum of α, denoted by Spec (α), as the Gelfand spectrum of α viewed as the element of the Banach algebra B(A), bounded linear operators of A, and the Connes' spectrum of α is defined as T(α) := \B∈Hα(A) Spec (αB) where Hα(A) is the set of all the hereditary and α-invariant sub-C ∗-algebras B, i.e., α(B) = B. Given a C ∗-algebra A we define by I(A) the set of the closed two-sided ideals of A, and given an endomorphism α ∈ End (A) we denote by I α(A) the set of all α-invariant ideals I of A, i.e., α−1(I) := {x ∈ A : α(x) ∈ I} = I. The following equivalent statements follows from Olesen and Pedersen [14, Theorem 10.4] and Sierakowski [21]. These authors studied actions of more general groups G, but in the situation that G = Z all simplifies in the following way: Given an automorphism α ∈ Aut (A) the following statements are equivalent: (1) T(α) = T, (2) αn is properly outer [9] for every n ∈ N, i.e., kαn I − Ad Uk = 2 for every α-invariant ideal I of A and U a unitary in M(I), (3) the induced action Z y bA on the space of equivalence classes of irreducible represen- tations of A is topologically free, i.e., {[π] ∈ bA : [π ◦ αn] = [π] then n = 0} is dense in bA, (4) A has the intersection property, i.e., given any non-zero ideal J of A ×α Z we have that J ∩ A 6= 0, (5) α satisfies the Rokhlin* property [21] (defined below). Moreover, if A is α-simple, i.e., α(I) = I implies that I = 0, A, then (1) − (5) are equivalent to (6) αn is multiplier outer for every n ∈ Z \ 0, i.e., αn 6= Ad U for every unitary U in the multiplier algebra M(A). The Rokhlin* property was defined by Sierakowski [21] for more general groups G. It is weaker that G y bA being topologically free [21, Theorem 2.11], and both are equivalent / / / / / 4 EDUARD ORTEGA AND ENRIQUE PARDO when A is commutative. However, if G is discrete (e.g. G = Z) then α having the Rokhlin* property implies that α is properly outer [21, Theorem 2.19]. Observe that given any ideal J of A ×α Z we have that the ideal J ∩ A is a α-invariant ideal of A, and therefore the map Φ : I(A ×α Z) −→ I α(A) defined by J 7−→ J ∩ A is a surjective map. In general this map is not injective, but a necessary condition for this map being injective can be given. First observe that given any α-ideal I of A we have the following short exact sequence, 0 −→ I ×α Z −→ A ×α Z −→ (A/I) ×α Z −→ 0 , and therefore the following statements are equivalent: (1) T(αA/I) = T for every α-invariant ideal I of A, (2) the induced action Z y bA is essentially free, i.e., the induced action Z y X in any α-invariant closed subset X of bA is topologically free, (3) A separates ideals of A ×α Z, i.e., Φ is injective and hence bijective, (4) every ideal J of A ×α Z is of the form I ×α Z for some α-invariant ideal I, (5) α satisfies the residual Rokhlin* property. We recall the definition of the Rokhlin* property given by Sierakowski: Given a C ∗-algebra A, set A∞ := l∞(A)/c0(A), where l∞(A) is the C ∗-algebra of all bounded functions from N into A, and c0(A) is the ideal of l∞(A) consisting of all sequences (an) such that kank −→ 0. There is a natural inclusion A ⊆ A∞. Moreover, given an automorphism α ∈ Aut (A), there is a natural extension to an automorphism α ∈ Aut ((A∞)∗∗). An automorphism α ∈ Aut (A) of a separable C ∗-algebra A has the Rokhlin* property provided that there exists a projection p = (pn) ∈ (A∞)∗∗ ∩ A′, that we will call Rokhlin projection, such that: (1) Given any k ∈ Z \ {0}, we have that αk(p)p = 0, (2) Given any a ∈ A \ {0} there exists k ∈ Z such that aαk(p) 6= 0. If given any α-invariant ideal I of A the induced automorphism αA/I ∈ Aut (A/I) has the Rokhlin* property, then we say that has the residual Rokhlin* property. Definition 1.1. Let A be separable C ∗-algebra. Given an endomorphism α ∈ End (A) we say that α satisfies the (residual) Rokhlin* property if its dilation ¯α does. 2. The Rokhlin Property and the p-shift endomorphism Given an endomorphism α ∈ End (A), we will fix conditions on a sequence (an) ∈ A∞ to construct a Rokhlin projection for the dilated endomorphism ¯α ∈ Aut ( ¯A), and hence to guarantee that α has the Rokhlin* property. Lemma 2.1. Let A be a separable C ∗-algebra and let α : A −→ A be an injective endomor- phism. Given a sequence x = (xn) ∈ A∞, define the sequence y = (αn,∞(xn)) ∈ ¯A∞, where ¯A is the dilation of A by α. Then: (1) If for every k ∈ N we have that kxnαk(xn)k −→ 0 when n → ∞, then y ¯αk(y) = 0 for every k ∈ Z. (2) If for every l ∈ N and a ∈ A we have that k[αn(a), xl+n]k −→ 0 when n → ∞, then [b, y] = 0 for every b ∈ ¯A. PURELY INFINITE CROSSED PRODUCTS BY ENDOMORPHISMS 5 (3) If for every l ∈ N and a ∈ A we have that kαn(a)xl+nk −→ kak when n → ∞, then kbyk = kbk for every b ∈ ¯A. Proof. (1) Suppose there exists a sequence x = (xn) ∈ A∞ satisfying kxnαk(xn)k −→ 0 for every k ∈ N when n → ∞. Now observe that, given k, n ∈ N, we have that ¯αk(yn) = ¯αk(αn,∞(xn)) = αn,∞(αk(xn)) So, given k ∈ N it follows that ¯α−k(yn) = ¯α−k(αn,∞(xn)) = αn+k,∞(xn) . and k ¯αk(yn)ynk = kαn,∞(αk(xn)xn)k = kαk(xn)xnk −→ 0 when n → ∞, and k ¯α−k(yn)ynk = kαn+k,∞(xnαk(xn))k = kxnαk(xn)k −→ 0 when n → ∞. Thus, y and ¯αk(y) are pairwise orthogonal elements of ¯A∞ for every k ∈ Z\{0}, as we wanted. Now, given any b ∈ ¯A and ε > 0, there exist a ∈ A and l ∈ N such that kb−αl,∞(a)k < ε/3. (2) By hypothesis, there exists Na ∈ N such that kαn(a)xl+n − xl+nαn(a)k < ε/3 for every n ≥ Na. So, we have that kαl,∞(a)yn − ynαl,∞(a)k = kαn,∞(αn−l(a)xn) − αn,∞(xnαn−l(a))k = kαn−l(a)xn − xnαn−l(a)k < ε/3 for every n ≥ l + Na. Therefore, we have that kbyn − ynbk ≤ kbyn − αl,∞(a)ynk + kαl,∞(a)yn − ynαl,∞(a)k + kynαl,∞(a) − ynbk < ε/3 + ε/3 + ε/3 = ε for every n ≥ l + Na. Thus, we have that yb = by for every b ∈ ¯A. (3) By hypothesis there exists Na ∈ N, such that kαn(a)xl+nk ≥ kak−ε/3 for every n ≥ Na. So we have that kαl,∞(a)ynk = kαn,∞(αn−l(a)xn)k = kαn−l(a)xnk ≥ kak − ε/3 , for every n ≥ l + Na. Therefore, we have that kαl,∞(a)ynk ≤ kαl,∞(a)yn − bynk + kbynk < ε/3 + kbynk, for every n ≥ l + Na, and thus kαl,∞(a)k − ε < kαl,∞(a)k − ε/3 − ε/3 ≤ kαl,∞(a)ynk − ε/3 < kbynk for every n ≥ l + Na. But since kbk − ε ≤ kak = kαl,∞(a)k, it follows that kbk − 2ε ≤ kbynk for every n ≥ l + Na. As ε is arbitrary, we have that kbk = kbyk, as desired. (cid:3) Observe that, given a projection x = (xn) ∈ A∞ satisfying the hypothesis (1) − (3) of the above Lemma, the sequence p = (αn,∞(xn)) ∈ ¯A∞ is a Rokhlin projection for the automor- phism ¯α. Observe that then α satisfies even a stronger property than the Rokhlin* property, because kpbk = kbk for every b ∈ ¯A. 6 EDUARD ORTEGA AND ENRIQUE PARDO Given a unital simple C ∗-algebra A with a non-trivial projection p, Dykema and Rørdam defined in [8] the so called Bernoulli p-shift endomorphism ∆p : A⊗∞ −→ A⊗∞ by x 7−→ p⊗x for every x ∈ A⊗∞. Choosing ⊗ = ⊗max or ⊗min we have that ∆p is injective. They proved that any power of the dilation automorphism ¯∆p is multiplier outer. Hence, since A⊗∞ is simple, ¯∆p must be properly outer. We will prove a more general result. Proposition 2.2. Let A be a nuclear unital C ∗-algebra with a non-trivial projection p, and let ∆p : A⊗∞ −→ A⊗∞ be the p-shift endomorphism. Then, ∆p satisfies the Rokhlin* property. Proof. To simplify notation set B := A⊗∞ and α := ∆p. Given n ∈ N, we define the projection qn := 1 ⊗ · · ·(2n) · · · ⊗ 1 ⊗ (1 − p) ⊗ p ⊗ · · ·(2n) · · · ⊗ p ⊗ 1A⊗∞ ∈ B . Observe that q = (qn) ∈ B∞ is a projection, and then by Lemma 2.1 it is enough to check the following: (1) kqnαk(qn)k −→ 0 when n → ∞, for every k ∈ N, (2) k[αn(b), ql+n]k −→ 0 when n → ∞, for every l ∈ N and b ∈ B, (3) kαn(b)ql+nk −→ kbk when n → ∞, for every l ∈ N and b ∈ B, to prove that q′ = (αn,∞(qn)) ∈ ¯B∞ is a Rokhlin projection for ¯α. First, we have that qn and αk(qn) are orthogonal projections for every k ∈ N, so (1) holds. Now given any b ∈ B and ε > 0, there exists j ∈ N and a ∈ A⊗j such that kb − a ⊗ 1A⊗∞k < ε/2. Therefore, given any l, n ∈ N we have that αn(a ⊗ 1A⊗∞) = p ⊗ · · ·(n) · · · ⊗ p ⊗ a ⊗ 1A⊗∞ , and ql+n = 1 ⊗ · · ·(2(l+n)) · · · ⊗ 1 ⊗ (1 − p) ⊗ p ⊗ · · ·(2(l+n)) · · · ⊗ p ⊗ 1A⊗∞ ∈ B . Thus, whenever n ≥ j − 2l, it follows that ql+nαn(a ⊗ 1A⊗∞) = αn(a ⊗ 1A⊗∞)ql+n = αn(a ⊗ 1A⊗∞)ql+n = p ⊗ · · ·(n) · · · ⊗ p ⊗ a ⊗ zl,n (∗) where zl,n is a projection of A⊗∞. Therefore, we have that kαn(b)ql+n − ql+nαn(b)k ≤ kαn(b)ql+n − αn(a ⊗ 1A⊗∞)ql+nk + kαn(a ⊗ 1A⊗∞)ql+n − ql+nαn(a ⊗ 1A⊗∞)k + kql+nαn(a ⊗ 1A⊗∞) − ql+nαn(b)k < ε/2 + 0 + ε/2 = ε for every n ≥ j − 2l. Then k[αn(b), ql+n]k −→ 0 when n → ∞, so (2) holds. Finally, observe that using the canonical isomorphism C ⊗∞ ∼= C ⊗n ⊗ C ⊗j ⊗ C ⊗∞, and applying (∗) and the nuclearity of A it follows that for every n ≥ j − 2l. So, we have that kαn(a ⊗ 1A⊗∞)ql+nk = kak kαn(b)k − ε < kαn(a ⊗ 1A⊗∞)k − ε/2 = kαn(a ⊗ 1A⊗∞)ql+nk − ε/2 < kαn(b)ql+nk for every n ≥ j − 2l. Then, q′ = (qn) also satisfies (3). Thus, by Lemma 2.1 p = (αn,∞(qn)) ∈ ¯B∞ is a Rokhlin projection for ∆p. (cid:3) PURELY INFINITE CROSSED PRODUCTS BY ENDOMORPHISMS 7 ∆p defined on the UHF-algebra Mn∞ := N∞ Example 2.3. Given any n ∈ N we set p := e1,1 ∈ Mn(C). Then, the p-shift endomorphism i=1 Mn(C) is the Cuntz's endomorphism. By Proposition 2.2, the automorphism ¯∆p : Mn∞ ⊗ K −→ Mn∞ ⊗ K satisfies the Rokhlin* property. Example 2.4. Let A = C ⊕ C and let p := (1, 0). Then we have that A⊗∞ ∼= C(X), where X = {0, 1}N is the Cantor set. In this case, the p-shift endomorphism ∆p : C(X) −→ C(X) is defined as ∆p(f ) = f ◦ δ for every f ∈ C(X), where δ(x) = (0, x) for every x ∈ X. By Proposition 2.2, ¯∆p is properly outer. Let us consider the ideal I = {f ∈ C(X) : f (0) = 0}, that is a ∆p-invariant ideal. Then C(X)/I ∼= C and (∆p)C(X)/I = (∆p)C(X)/I = Id, so it is not properly outer. Now, we will study the p-shift endomorphism on an interesting class of C ∗-algebras. Let D be a unital and separable strongly self-absorbing C ∗-algebra [23]. Then, D ∼= D⊗n ∼= D⊗∞ for every n ∈ N. Also, D has an approximately inner flip, i.e. the homomorphism σ : D ⊗ D −→ D ⊗ D defined by σ(a ⊗ b) = b ⊗ a for every a, b ∈ D is approximately unitary equivalent to the identity map. Recall also that every strongly self-absorbing C ∗-algebra is nuclear. We will first compute which map induces ∆p at the level of K-theory. Recall that a strongly self-absorbing C ∗-algebra D satisfying the Universal Coefficients Theorem (UCT) has trivial K1 group [23, Proposition 5.1] and torsion-free K0 group. Therefore, the Kunneth formulas say that we have an isomorphism K0(D⊗∞) = K0(D ⊗D⊗∞) ∼= K0(D)⊗K0(D⊗∞) induced by the map [a]⊗[b] 7−→ [a⊗b] for projections a ∈ D, b ∈ D⊗∞. Thus, if we define [p]· x := [p]⊗x for every x ∈ K0(D⊗∞), the following Lemma comes straightforward. Lemma 2.5. Let D be a strongly self-absorbing C ∗-algebra satisfying the UCT with a non- trivial projection p. Then K0(∆p) = [p] · IdK0(D⊗∞). We will prove some results about ideal structure for suitable tensor products of C ∗-algebras. We thank Nate Brown for the proof of the following result. Lemma 2.6. Let A be an exact simple C ∗-algebra, and let B be any C ∗-algebra. If I ✁A⊗B, then there exists J ✁ B such that I = A ⊗ J. Proof. By [4, Corollary 9.4.6], I = span{A ⊙ IB : IB ✁ B, A ⊙ IB ⊆ I}. Consider which is clearly an ideal of B. Now, pick J := span{IB ✁ B : A ⊙ IB ⊆ I}, X := A ⊙ (span{IB : IB ✁ B, A ⊙ IB ⊆ I}) , which is a dense linear subspace of A⊙J. We will show that X is also a dense linear subspace of I. For, notice that span{A ⊙ IB : IB ✁ B, A ⊙ IB ⊆ I} = A ⊙ (span{IB : IB ✁ B, A ⊙ IB ⊆ I}) . Since closures are unique, we conclude that I = A ⊗ J, as desired. (cid:3) As a consequence, we have the following C ∗-algebra version of Azumaya-Nakayama's The- orem. 8 EDUARD ORTEGA AND ENRIQUE PARDO Proposition 2.7. If A is an exact simple C ∗-algebra and B is any C ∗-algebra, then the map I 7→ A ⊗ I defines a bijection between ideals of B and ideals of A ⊗ B Proof. Clearly, it is a well-defined map. By Lemma 2.6, it is a surjective map. In order to prove injectivity of the map, notice that any ideal can be seen as a linear subspace of A ⊗ B via the identification I = C ⊗ I = C ⊙ I. By [5, Proposition 4.7.3], if I ✁ B, then (A ⊙ I) ∩ B = (A ⊙ I) ∩ (C ⊙ I) = C ⊙ I = I. Thus, taking closures we have (A ⊗ I) ∩ B = (A ⊗ I) ∩ (C ⊗ I) = C ⊗ I = I, whence the above defined map is injective. So we are done. (cid:3) We say that a C ∗-algebra A absorbs D if A ∼= A ⊗ D. Theorem 2.8. Let D be a strongly self-absorbing C ∗-algebra with a non-trivial projection p, and let A be a separable unital C ∗-algebra that absorbs D. Then, given any injective endomorphism α ∈ End (A), the endomorphism α ⊗ ∆p ∈ End (A ⊗ D⊗∞) satisfies the residual Rokhlin* property. Moreover, if D satisfies the UCT then K∗(α ⊗ ∆p) = [p] · K∗(α). Proof. Let α ∈ End (A) be an injective endomorphism and let ∆p : D⊗∞ −→ D⊗∞ be the p-shift endomorphism. By Proposition 2.7, all the ideals of A ⊗ D⊗∞ are of the form I ⊗ D⊗∞ for ideals I of A. Hence, an ideal I ⊗ D⊗∞ is α ⊗ ∆p-invariant ideal if and only if I is an α-invariant ideal of A. Given any α-invariant ideal I of A, let us consider the quotient B := (A⊗D⊗∞)/(I ⊗D⊗∞). Observe that we can identify B with (A/I) ⊗ D⊗∞, as a ⊗ x 7−→ a ⊗ x for every a ∈ A and x ∈ D⊗∞. Moreover, with this identification, we have that (α ⊗ ∆p)B = αA/I ⊗ ∆p, and to simplify notation set β := αA/I ⊗ ∆p. Let us consider the projection q = (qn) ∈ (D⊗∞)∞ defined in the proof of Proposition 2.2. If 1A is the unit of A, then we define the non-zero projection q′ = (βn,∞(1A⊗qn)) ∈ ¯B∞, where ¯B is the dilation of B with respect to β. We claim that q′ is a Rokhlin projection for ¯β. For this it is enough to check conditions (1) − (3) of Lemma 2.1. Since βk(1A ⊗ qn) = α(1A) ⊗ ∆k p(qn) for every k, n ∈ N, it follows that βk(1A ⊗ qn) · (1A ⊗ qn) = α(1A) ⊗ (∆k p(qn) · qn) = 0, since ∆k p(qn) · qn = 0 for every k ∈ N. Thus, (1) is checked. Now, given a ∈ (A/I) ⊗ D⊗∞ and ε > 0 there exist r, s ∈ N such that b :=Pr i=1 ai ⊗ (xi ⊗ 1D⊗∞) for some ai ∈ A and xi ∈ D⊗s and such that ka − bk < ε/2. Then, as we see in the proof of Proposition 2.2, given l ∈ N for every n ≥ s − 2l we have that ∆n p (xi ⊗ 1D⊗∞)ql+n = ql+n∆n p (xi ⊗ 1D⊗∞) = p ⊗ · · ·(n) · · · ⊗ p ⊗ xi ⊗ zl,n for every i = 1, . . . , r and for some projection zl,n ∈ D⊗∞. Therefore, a standard argument shows that kβn(a)(1A ⊗ ql+n) − (1A ⊗ ql+n)βn(a)k < ε for every n ≥ s − 2l, and then k[βn(a), ql+n]k −→ 0 when n → ∞. Therefore it follows (2). PURELY INFINITE CROSSED PRODUCTS BY ENDOMORPHISMS 9 Finally, we have that kβn( rXi=1 ai ⊗ (xi ⊗ 1D⊗∞))(1A ⊗ ql+n)k = k = k rXi=1 rXi=1 αn(ai) ⊗ (∆n p (xi ⊗ 1D⊗∞)ql+n)k αn(ai) ⊗ (p ⊗ · · ·(n) · · · ⊗ p ⊗ xi ⊗ zl,n)k , and using the canonical isomorphism A ⊗ D⊗∞ ∼= A ⊗ D⊗s ⊗ D⊗∞ and the nuclearity of D it follows that k rXi=1 αn(ai) ⊗ (p ⊗ · · ·(n) · · · ⊗ p ⊗ xi ⊗ zl,n)k = k = k rXi=1 rXi=1 αn(ai) ⊗ xikkp ⊗ · · ·(n) · · · ⊗ p ⊗ zl,nk αn(ai) ⊗ xik = k rXi=1 ai ⊗ xik , for every n ≥ s − 2l. Another standard argument shows us that kak − ε = kβn(a)k − ε < kβn(a)ql+nk , for every n ≥ s − 2l. Thus, kβn(a)ql+nk −→ kak when n → ∞, and then condition (3) is verified. Therefore q′ is a Rokhlin projection for ¯β, and hence β satisfies the Rokhlin* property. Since B denotes (A ⊗ D⊗∞)/(I ⊗ D⊗∞) for every arbitrary α-invariant ideal I of A, we have that α ⊗ ∆p satisfies the residual Rokhlin* property. Finally if D satisfies the UCT then K1(D) = 0 and K0(D) is torsion-free, and hence by the Kunneth formulas we have that K∗(A ⊗ D⊗∞) ∼= K∗(A) ⊗ K∗(D⊗∞) and K∗(α ⊗ ∆p) = K∗(α) ⊗ K∗(∆p), so K∗(α ⊗ ∆p) = [p] · K∗(α) by Lemma 2.5. (cid:3) Notice that if A absorbs D, then by [23, Theorem 2.3] any isomorphism ϕ : A 7−→ A ⊗ D is approximately unitary equivalent to IdA ⊗ 1D. Lemma 2.9. If D is a strongly self-absorbing C ∗-algebra, A is a C ∗-algebra that absorbs D, and ϕ : A → A ⊗ D is an isomorphism, then for any I ✁ A we have that ϕ(I) ⊆ I ⊗ D. Moreover, if A has finitely many ideals, then ϕ(I) = I ⊗ D. Proof. By [23, Theorem 2.3], ϕ is approximately unitary equivalent to IdA ⊗ 1D. Now, as (IdA ⊗ 1D)(I) ⊆ I ⊗ D, given any unitary u ∈ M(A ⊗ D) we have u(IdA ⊗ 1D)(I)u∗ ⊆ I ⊗ D. Hence, ϕ(I) ⊆ I ⊗ D. For the last statement, notice that by the previous statement and Proposition 2.7, for each I ✁ A there exists a unique JI ✁ A such that ϕ(I) = JI ⊗ D ⊆ I ⊗ D, and thus JI ⊆ I again by Proposition 2.7. Hence, there is a monotone decreasing bijection (−)I from the set of ideals of A to itself. Notice that 0 and A are fixed by (−)I . If (−)I is not the identity map, as the set of ideals is finite, there exists K ✁ A different from 0 and A maximal for the property KI 6= K. By the above argument there exists a unique L ✁ A such that LI = K, and notice that K ⊆ L. Assuming K 6= L will contradict the maximality of K. So, K = L, but then K = LI = KI ( K, which is impossible. So we are done. (cid:3) 10 EDUARD ORTEGA AND ENRIQUE PARDO Lemma 2.10. Let B, C be C ∗-algebras, let β : B −→ B be an injective endomorphism, and let δ : C −→ B be an isomorphism. If β satisfies the residual Rokhlin* property, then so does δ−1 ◦ β ◦ δ. Proof. First define α := δ−1 ◦ β ◦ δ, so it is easy to check that the dilated automorphism ¯α = δ−1 ◦ β ◦ δ of ¯C is the composition ¯δ−1 ◦ ¯β ◦ ¯δ, where ¯δ is the induced isomorphism between ¯C and ¯B. Observe also that I α(C) = {δ−1(I) : I ∈ I β(B)}, and then given any I ∈ I β(B) we have that αA/δ−1(I) = δ−1 C/δ−1(I) ◦ βB/I ◦ δC/δ−1(I) Finally, let p ∈ ( ¯B∞)∗∗ ∩ ¯B′ be a Rokhlin projection of β, then since δ induces an isomor- (cid:3) phism δ : ( ¯C ∞)∗∗ −→ ( ¯B∞)∗∗ we have that δ−1(p) is a Rokhlin projection for α. Recall that if D is a strongly self-absorbing C ∗-algebra in the UCT class and A is a C ∗- algebra that absorbs D, then the isomorphism ϕ : A −→ A ⊗ D induces the isomorphism of K-theory K∗(ϕ) : K∗(A) −→ K∗(A) ⊗ K0(D) given by x 7−→ x ⊗ [1D] for every x ∈ K∗(A). We define [q] · x := K∗(ϕ)−1(x ⊗ [q]) for every projection q ∈ D. Corollary 2.11. Let D be a strongly self-absorbing C ∗-algebra with a non-trivial projection p, and let A be a separable unital C ∗-algebra with finitely many ideals such that A absorbs D. Then, given any injective endomorphism α ∈ End (A) there exists an endomorphism β ∈ End (A) such that I α(A) = I β(A) and satisfies the residual Rokhlin* property. Moreover, if D satisfies the UCT then K∗(β) = [p] · K∗(α). Proof. We define β := ϕ−1 ◦ (α ⊗ ∆p) ◦ ϕ where ϕ is any isomorphism A ∼= A ⊗ D⊗∞. First observe that that by Lemma 2.9 we have that I β(A) = {ϕ−1(I ⊗D⊗∞) : I ∈ I α(A)} = I α(A). Now by Theorem 2.8 and Lemma 2.10 we have that β has the residual Rokhlin* property. If D satisfies the UCT we have that K∗(β) = K∗(ϕ−1) ◦ K∗(α ⊗ ∆p) ◦ K∗(ϕ). Therefore, K∗(β)(x) = K∗(ϕ−1) ◦ K∗(α ⊗ ∆p) ◦ K∗(ϕ)(x) = K∗(ϕ−1) ◦ K∗(α ⊗ ∆p)(x ⊗ [1D⊗∞]) = K∗(ϕ−1)(K∗(α)(x) ⊗ [p ⊗ 1D⊗∞]) = [p ⊗ 1D⊗∞] · K∗(α)(x) for every x ∈ K∗(A). Finally, since the map [q] 7−→ [q ⊗ 1D⊗∞] for [q] ∈ K0(D) induces an isomorphism between K0(D) and K0(D⊗∞) we have that [p⊗1D⊗∞]·K∗(α)(x) = [p]·K∗(α)(x) for every x ∈ K∗(A). (cid:3) 3. Purely infinite crossed products Given a unital C ∗-algebra A and an injective endomorphism α : A −→ A Stacey [22], following the ideas of Cuntz [6] and Paschke [15], defined the crossed product A ×α N as the universal C ∗-algebra generated by A and an isometry S∞ such that S∞aS∗ ∞ = α(a) for every a ∈ A. It was also shown that A ×α N ∼= α1,∞(1A)( ¯A ×¯α Z)α1,∞(1A) and that α1,∞(1A) is a full projection. Therefore A ×α N and ¯A ×¯α Z are strongly Morita equivalent. So, to study properties like purely infiniteness, simplicity or the ideal structure of A ×α N, it is enough to look at the crossed product ¯A ×¯α Z. From now on, we will identify A ×α N with the corresponding isomorphic corner sub-C ∗- algebra of ¯A ×¯α Z, while A is identified with α1,∞(A) and the generating isometry S∞ of A ×α N with the compression α1,∞(1)U∞α1,∞(1) of the generating unitary U∞ of ¯A ×¯α Z. PURELY INFINITE CROSSED PRODUCTS BY ENDOMORPHISMS 11 There is a bijection λ : I α(A) −→ I ¯α( ¯A) given by I 7−→ P∞ i=1 αi,∞(I), with inverse λ−1(J) = {a ∈ A : α1,∞(a) ∈ J} for every J ∈ I ¯α( ¯A), that makes commutative the diagram I(A ×α N) µ I( ¯A ×¯α Z) Φ Φ I α(A) λ / I ¯α( ¯A) where µ is the bijection induced by the Morita equivalence. Hence, Φ : I( ¯A ×¯α Z) −→ I ¯α( ¯A) is a bijection if and only if so is Φ : I(A ×α N) −→ I α(A). Thus, there is no ambiguity in saying that A separates ideals of A ×α N exactly when ¯A separates ideals of ¯A ×¯α Z. Now, given I ∈ I α(A) we have that α restricts to an endomorphism of I, and therefore we can i=1 αi,∞(I) with ¯I, the dilation of I by αI. Since ¯I is an ¯α-invariant ideal of ¯A, we can identify ¯I × ¯αI Z with an ideal of ¯A ×¯α Z. Moreover, ¯A/ ¯I is isomorphic to A/I, the dilation of A/I by αA/I, where αA/I is the natural endomorphism induced by the quotient. Thus, given I ∈ I α(A) we have the following short exact sequence: identify P∞ 0 −→ ¯I × ¯αI Z −→ ¯A ×¯α Z −→ A/I ×αA/I Z −→ 0 . Given I ∈ I α(A), the ideal hIi of A ×α N generated by I is not necessarily isomorphic to I ×αI N, as it is defined for the non-unital case, but it is isomorphic to α1,∞(1A)( ¯I × ¯αI Z)α1,∞(1A). Indeed, to simplify notation let us denote B := ¯A ×¯α Z and p := α1,∞(1A) ∈ B, so that by the above identification we have that pBp := A ×α N. From the observation that ∞α1,n(y)U∞ = α1,n+1(y) for every n ∈ N and y ∈ I, it follows that BIB = B ¯IB, and hence U ∗ hIi = pBpIpBp = pBIBp = pB ¯IBp . But B ¯IB is ¯I × ¯αI Z, so the claim is proved. Therefore we have that hIi is strongly Morita equivalent to ¯I ×¯α Z, so K∗(hIi) ∼= K∗( ¯I × ¯αI Z), and we can use the Pimsner-Voiculescu six-terms exact sequence, together with the continuity of the K-theory, to compute the K- groups of the ideals of the form hIi for I ∈ I α(A). Moreover, given I, J ∈ I α(A) with I ⊆ J, by the previous argument we have that hJi/hIi = α1,∞(1)(J/I ×αJ/I Z)α1,∞(1) . Remark 3.1. We would like to remark that our definition of invariant ideal slightly differs from the one given by Adji in [1] for two reasons. First, because we only are interested in actions by injective endomorphisms. And second, because we are not interested for a characterization of the gauge invariant ideals as another crossed product. Our goal in this section is to give conditions for the crossed product A ×α N being purely infinite. Notice that, if A is a separable purely infinite C ∗-algebra of real rank zero and α ∈ End (A) is such that it satisfies the residual Rokhlin* property, then the dilation ¯A is a separable purely infinite C ∗-algebra of real rank zero and the action of Z induced by ¯α on d( ¯A) is essentially free. Hence, ¯A ×¯α Z is a purely infinite C ∗-algebra [20, Theorem 3.3], and thus so is A ×α N. Another condition we will consider is related to endomorphisms on C ∗-algebras of real rank zero that acts in a special way at the level of their monoid of projections. Given a C ∗-algebra / / / 12 EDUARD ORTEGA AND ENRIQUE PARDO A, let V (A) be the monoid of Murray-von Neumann equivalence classes of projections of M∞(A). We briefly recall its construction below. We say that two projections p and q in a C ∗-algebra A are Murray-von Neumann equivalent if there is a partial isometry v such that p = vv∗ and v∗v = q. One can extend this relation to the set P(M∞(A)) of projections in M∞(A), and thereby construct V (A) = P(M∞(A))/∼ , where [p] ∈ V (A) stands for the equivalence class that contains the projection p in M∞(A). This set becomes an abelian monoid when endowed with the operation [p]+[q] = [p⊕q], where p⊕q refers to the matrix(cid:0) p 0 0 q(cid:1). Moreover, if I is a closed two sided ideal of A, then V (I) is an ideal of V (A). If A has real rank zero, then all the ideals of V (A) are of the form V (I) for some ideal I of A, and moreover V (A)/V (I) ∼= V (A/I). Now, given any C ∗-algebra endomorphism α : A −→ A, we can extend it to a endomorphism α : M∞(A) −→ M∞(A) by a⊗k 7→ α(a)⊗k for every a ∈ A and k ∈ M∞(C). Hence, it induces a map α∗ : V (A) −→ V (A) by the rule [p] 7−→ [α(p)] for every projection p ∈ M∞(A). Definition 3.2. Let A be a C ∗-algebra and let α ∈ End (A). Then we say that α contracts projections of A if given x ∈ V (A) there exists n ∈ N such that α∗n(x) < x. Example 3.3. Given a C ∗-algebra A and a strongly self-absorbing C ∗-algebra D with a non- n=1 An where An := A⊗D⊗n⊗1D⊗∞. Therefore given any q ∈ A ⊗ D⊗∞ there exists q′ ∈ A ⊗ D⊗n for some n ∈ N such that q ∼ q′ ⊗ 1D⊗∞. Since D is strongly self-absorbing, the flip σ : D ⊗ D −→ D ⊗ D is approximately unitary i where ai ∈ A and qi ∈ D⊗n, then trivial projection p, we have that A⊗D⊗∞ =S∞ equivalent to IdD⊗D. Hence, if we write q′ = Pi ai ⊗ q′ k(1A ⊗ u ⊗ 1D⊗∞)(Xi i ⊗ 1D⊗∞)(1A ⊗ u∗ ⊗ 1D⊗∞) −Xi there exists a unitary u ∈ D⊗n+1 such that ai ⊗ q′ i ⊗ p ⊗ 1D⊗∞k < 1/2 . ai ⊗ p ⊗ q′ Thus, since p is a non-trivial projection of D we have that (IdA ⊗ ∆p)∗[q] = (IdA ⊗ ∆p)∗[q′ ⊗ 1D⊗∞] = [(IdA ⊗ ∆p)(q′ ⊗ 1D⊗∞)] = [Xi < [Xi ai ⊗ p ⊗ q′ i ⊗ 1D⊗∞] = [Xi ai ⊗ q′ i ⊗ p ⊗ 1D⊗∞] ai ⊗ q′ i ⊗ 1D⊗∞] = [q′ ⊗ 1D⊗∞] = [q] . Therefore, IdA ⊗ ∆p contracts projections of A ⊗ D⊗∞. Moreover, given any ideal I of A we have that αA⊗D⊗∞/I⊗D⊗∞ contracts projections of A ⊗ D⊗∞/I ⊗ D⊗∞. Theorem 3.4. Let A be a unital separable C ∗-algebra of real rank zero, let α ∈ End (A) be an injective endomorphism such that: (1) It satisfies the residual Rokhlin* property. (2) Given any ideal I ∈ I α(A) we have that αA/I contracts projections of A/I. Then A ×α N is a purely infinite C ∗-algebra. Proof. It is enough to check that ¯A×¯α Z is purely infinite, since it is strongly Morita equivalent to A ×α N. PURELY INFINITE CROSSED PRODUCTS BY ENDOMORPHISMS 13 First observe that since A is a separable C ∗-algebra with real rank zero so is ¯A, and since ¯α has the residual Rokhlin* property then the action of Z on d( ¯A) is essentially free. Now we claim that given any I ∈ I α(A) we have that ¯α ¯A/ ¯I contracts projections of ¯A/ ¯I. Indeed, first observe that αA/I = ¯α ¯A/ ¯I. Then, it is enough to check that ¯α contracts projec- tions of ¯A. Given any projection p ∈ ¯A there exists a projection q ∈ A and n ∈ N such that α1,n(q) ∼ p, and hence ¯α(α1,n(q)) = α1,n(α(q)). But since α contracts projections of A we have that α∗([q]) < [q], and hence ¯α∗[p] = ¯α∗[α1,n(q)] = [α1,n(α(q))] < [α1,n(q)] = [p] , as desired. Therefore, by [16, Proposition 2.11] it is enough to check that every non-zero hereditary sub-C ∗-algebra in any quotient of ¯A ×¯α Z contains an infinite projection. Since ¯A separates ideals in ¯A×¯αZ, we have that every ideal of ¯A×¯αZ is of the form ¯I ×¯αZ, so ( ¯A×¯αZ)/( ¯I ×¯αZ) ∼= A/I ×αA/I Z. Because of all the assumptions pass to quotients, we can replace A/I by ¯A and αA/I by ¯α. Let x be any positive element of ¯A ×¯α Z. By [20, Lemma 3.2] there exists a ∈ ¯A+ such that a . x. As ¯A has real rank zero, there exists a projection p ∈ ¯A such that p . a, so p . x. Then, there exists n ∈ N such that α∗n([p]) < [p] in V ( ¯A). Hence, there exists t ∈ ¯A such that t¯αn(p)t∗ + z = p for some idempotent 0 6= z ∈ ¯A. If we set r := tαn(p) and we define s := rU np, where U is the unitary in M( ¯A ×¯α Z) that implements ¯α, then s∗s = (pU ∗nr∗)(rU np) = pU ∗nr∗rU np = pU ∗n ¯αn(p)U np = p and ss∗ = (rU np)(pU ∗nr∗) = rU npU ∗nr = r ¯αn(p)r∗ = t¯αn(p)t∗. Hence, ss∗ + z = t¯αn(p)t∗ + z = p, whence p is an infinite projection (since z 6= 0). Therefore, since p . x, by [12, Proposition 2.6] there exist δ > 0 and v ∈ ¯A ×¯α Z such that v∗(x − δ)+v = p. With w = (x − δ)1/2 + v it follows that w∗w = p, whence ww∗ = (x − δ)1/2 + ∈ x( ¯A ×¯α Z)x is a projection equivalent to p. Hence, ww∗ is infinite because so is p. Thus, by [16, Proposition 2.11], we have that ¯A ×¯α Z is a purely infinite C ∗-algebra and hence A ×α N is too. (cid:3) + vv∗(x − δ)1/2 4. Examples In this section we will use the previous results to construct interesting examples of purely infinite crossed products C ∗-algebras with different ideal structures. In particular we are going to build actions on strongly purely infinite C ∗-algebras A, i.e. A ∼= A ⊗ O∞. Since O∞ is a strongly self-absorbing C ∗-algebra with non-trivial projections, when A has finitely many ideals we can apply Corollary 2.11 to perturb any injective endomorphism β ∈ End (A) to another endomorphism α satisfying the residual Rokhlin* property and with K∗(α) = K∗(β). Therefore A ×α N will be a purely infinite C ∗-algebra with a lattice isomorphism Φ : I(A ×α N) −→ I β(A). Example 4.1. Let On be the Cuntz algebra for n = 2, . . . , ∞; by Kirchberg's work we ∼= On. Given m ∈ N with m < n, by Corollary 2.11 there exists an have that On ⊗ O∞ injective endomorphism αm : On −→ On that satisfies the Rokhlin* property and K∗(αm) = m · IdK∗(On). Therefore, since On is a simple purely infinite C ∗-algebra in the UCT class, so 14 EDUARD ORTEGA AND ENRIQUE PARDO is On ×αm N. Thus, by Kirchberg-Philllips classification results, it is enough to compute its K-theory to determine in which isomorphism class lies. First recall that K∗(On) ∼=(cid:26) (Z, 0) if n = ∞ (Z/(n − 1)Z, 0) otherwise . Observe that the endomorphism αm defines a group endomorphism K0(αm) : K0(On) −→ K0(On) given by y 7−→ my for every y ∈ K0(On). By continuity of K-theory we have that K∗(On) ∼= lim−→ K0(αm) Ki(On) , so K0(On) =(cid:26) Z[1/m] Z/kZ lim−→ K0(αm) if n = ∞ if n 6= ∞ where k = [gcd(n − 1, m)(n − 1)] , where given a, b ∈ N we define [ab] := max {c ∈ N : cb and gcd(a, c) = 1}. Then, K0(αm) : K0(On) −→ K0(On) is given by x 7−→ mx for every x ∈ K0(Om), while K1(Om) = 0. Therefore, we can assume that m ≤ k and hence Ker (Id − K0(αm)) = if n = ∞ and m = 1 Z if n = ∞ and m 6= 1 0 Z/kZ if n 6= ∞ and m = 1 Z/lZ otherwise, where l = k/gcd(k, m − 1) , and Coker (Id − K0(αm)) = if n = ∞ and m = 1 Z Z/(m − 1)Z if n = ∞ and m 6= 1 if n 6= ∞ and m = 1 Z/kZ otherwise, where l = k/gcd(k, m − 1) Z/lZ .     So, using the Pimsner-Voiculescu six-term exact sequence, we have K0(On) Id−K0(αm) / K0(On) / K0(On ×αm N) . K1(On ×αm N) 0 0 0 Thus, K0(On ×αm N) ∼= Coker (Id−K0(αm)) and K1(On ×αm N) ∼= Ker (Id−K0(αm)). Hence, by Kirchberg-Phillips Classification Theorems it follows that On ×αm N ∼=  B Om Ol+1 ⊗ Ol+1 otherwise if n = ∞ and m = 1 if n = ∞ and m 6= 1 , where B is the unique Kirchberg algebra with K∗(B) ∼= (Z, Z). / / O O o o o o PURELY INFINITE CROSSED PRODUCTS BY ENDOMORPHISMS 15 Example 4.2. Let E be the following graph (3) •v2 (8) •v1 }③③③③③③③③ !❉❉❉❉❉❉❉❉ (6) •v4 . (4) •v3 !❉❉❉❉❉❉❉❉ }③③③③③③③③ Then, the graph C ∗-algebra C ∗(E) is purely infinite with real rank zero [10], and C ∗(E) ∼= C ∗(E) ⊗ O∞. So, taking β = IdC∗(E), by Theorem 2.8 there exists an injective endomor- phism α : C ∗(E) −→ C ∗(E) that satisfies the residual Rokhlin* property, and I α(C ∗(E)) = I(C ∗(E)). Moreover, K∗(α) = IdK∗(C∗(E)). E has the following hereditary and saturated subsets {∅, {v4}, {v4, v2}, {v4, v3}, {v2, v3, v4}, {v1, v2, v3, v4}} . so by [2] the Hasse diagram of its primitive ideal space X3 is @✁✁✁✁✁✁✁ ^❂❂❂❂❂❂❂ 2 1 4 ^❂❂❂❂❂❂❂ @✁✁✁✁✁✁✁ 3 Given two ideals I, J ∈ I(C ∗(E)) with I ⊆ J, using [3] we can easily compute K∗(J/I). For example: K∗(I{4}) ∼= (Z/5Z, 0) , K∗(I{3,4}) ∼= (Z/5Z ⊕ Z/3Z, 0) , K∗(I{2,4}) ∼= (Z/5Z ⊕ Z/2Z, 0) , K∗(I{2,3,4}) ∼= (Z/5Z ⊕ Z/3Z ⊕ Z/2Z, 0) , K∗(I{1,2,3,4}) ∼= (Z/7Z ⊕ Z/5Z ⊕ Z/3Z ⊕ Z/2Z, 0) , where IX := IZ/IY and Z, Y ⊆ E0 are hereditary and saturated subsets with Y ⊆ Z and X = Z \ Y . Then C ∗(E)×α N is purely infinite with primitive ideal space X3. Since C ∗(E) separates the ideals of C ∗(E) ×α N, we have that all the subquotients of C ∗(E) ×α N are of the form hJi/hIi for I, J ∈ I(C ∗(E)) with I ⊆ J, and hJi/hIi is strongly Morita equivalent to ( ¯J/ ¯I) ×¯α Z. Therefore we can use the Pimsner-Voiculescu six-term exact sequence for K-Theory to deduce that K∗(hJi/hIi) ∼= (K0(J/I), K0(J/I)). Finally observe that, given any ideals I, J ∈ I(C ∗(E)) with I ( J, does not exist any non- zero map K0(hJi/hIi) −→ K1(hJi/hIi), thus C ∗(E) ×α N is K0-liftable. Hence, by results of Pasnicu and Rørdam [16, Theorem 4.2], it must have real rank zero. } ! ! } @ ^ @ ^ 16 EDUARD ORTEGA AND ENRIQUE PARDO Acknowledgments Parts of this work was done during visits of the first author to the Departamento de Matem´aticas de la Universidad de C´adiz (Spain) and of the second author to the Institutt for Matematiske Fag, Norges Teknisk-Naturvitenskapelige Universitet (Trondheim, Norway). The authors thank both host centers for their warm hospitality. References [1] S. Adji, Semigroup Crossed Products and the Structure of Toeplitz Algebras. J. Operator Theory 44 (2000), 139 -- 150. [2] T. Bates, D. Pask, I. Raeburn and W. Szyma´nski. The C ∗-algebras of row-finite graphs. New York J. Math. 6 (2000), 307 -- 324. [3] T. Bates, J.H. Hong, I. Raeburn and W. Szyma´nski. The ideal structure of the C ∗-algebras of infinite graphs. Illinois J. Math. 46 (2002), no. 4, 1159 -- 1176. [4] N.P. Brown and N. Ozawa. "C ∗-algebras and finite-dimensional approximations". Graduate Studies in Mathematics, 88. American Mathematical Society, Providence, RI, 2008. [5] P.M. Cohn. "Algebra". Vol. 2. Second edition. John Wiley & Sons, Ltd., Chichester, 1989. [6] J. Cuntz. Simple C -algebras generated by isometries. Comm. Math. Phys. 57 (1977), 173 -- 185. [7] J. Cuntz and W. Krieger. A class of C-algebras and topological Markov chains. Inventiones Math. 56 (1980), 251 -- 268. [8] K. Dykema and M. Rørdam. Purely infinite simple C-algebras arising from free products. Can. J. Math. 50, No. 2 (1998), 323 -- 341. [9] G. A. Elliott. Some simple C ∗-algebras constructed as crossed products with discrete outer automor- phism groups. Publ. Res. Inst. Math. Sci. 16 (1980), no. 1, 299 -- 311. [10] J.H. Hong and W. Szyma´nski. Purely Infinite Cuntz-Krieger Algebras of Directed Graphs. Bull. London Math. Soc. 35 (2003) 689 -- 696. [11] E. Kirchberg. Das nicht-kommutative Michael-Auswahlprinzip und die Klassifikation nicht-einfacher Algebren. C ∗-Algebras (Munster, 1999), Springer, Berlin, 2000, pp. 92 -- 141 (German). [12] E. Kirchberg and M. Rørdam. Non-simple purely infinite C ∗-algebras. Amer. J. Math. 122 (2000), no. 3, 637 -- 666. [13] M. Laca. From Endomorphisms to Automorphisms and Back: Dilations and Full Corners. J. London Math. Soc. (2) 61 (2000), no. 3, 893 -- 904. [14] D. Olesen and G.K. Pedersen. Applications of the Connes Spectrum to C ∗-dynamical systems, III. J. Funct. Anal. 45 (1982), no. 3, 357 -- 390. [15] W. L. Paschke. The crossed product by an endomorphism. Proc. Amer. Math. Soc. 80 (1980), 113 -- 118. [16] C. Pasnicu and M. Rørdam. Purely infinite C ∗-algebras of real rank zero. J. Reine Angew. Math. 613 (2007), 51 -- 73. [17] N. C. Phillips. A Classification Theorem for Nuclear Purely infinite Simple C ∗-Algebras. Documenta Math. (2000), no. 5, 49 -- 114. [18] M. Rørdam. Classification of certain infinite simple C ∗-algebras. J. Funct. Anal. 131, (1995), 415 -- 458. [19] J. Renault. "A groupoid approach to C ∗-algebras". Lecture Notes in Mathematics, vol.793, Springer, Berlin, 1980. [20] M. Rørdam and A. Sierakowski. Purely Infinite C ∗-Algebras Arising from Crossed Products. Ergod. Theory & Dynam. Sys. 32 (2012), 273 -- 293. [21] A. Sierakowski. The ideal structure of reduced crossed products. Munster J. Math. 3 (2010), 237 -- 261. [22] P.J. Stacey. Crossed products of C ∗-algebras by ∗-endomorphisms. J. Austral. Math. Soc. (Seres A) 54 (1993), no. 2, 204 -- 212. [23] A. Toms and W. Winter. Strongly Self-Absorbing C ∗-algebras. Trans. Amer. Math. Soc. 359 (2007), no.8, 3999 -- 4029. PURELY INFINITE CROSSED PRODUCTS BY ENDOMORPHISMS 17 Department of Mathematical Sciences, NTNU, NO-7491 Trondheim, Norway E-mail address: [email protected] Departamento de Matem´aticas, Facultad de Ciencias, Universidad de C´adiz, Campus de Puerto Real, 11510 Puerto Real (C´adiz), Spain. E-mail address: [email protected] URL: https://sites.google.com/a/gm.uca.es/enrique-pardo-s-home-page/ URL: http://www.uca.es/dpto/C101/pags-personales/enrique.pardo
1212.3933	2	1212	2013-01-19T15:23:38	Circle maps and C*-algebras	[ "math.OA", "math.DS" ]	We consider a construction of C-algebras from continuous piecewise monotone maps on the circle which generalizes the crossed product construction for homeomorphisms and more generally the construction of Renault, Deaconu and Anantharaman-Delaroche for local homeomorphisms. Assuming that the map is surjective and not locally injective we give necessary and sufficient conditions for the simplicity of the C-algebra and show that it is then a Kirchberg algebra. We provide tools for the calculation of the K-theory groups and turn them into an algorithmic method for Markov maps.	math.OA	math	CIRCLE MAPS AND C ∗-ALGEBRAS THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN Abstract. We consider a construction of C ∗-algebras from continuous piecewise monotone maps on the circle which generalizes the crossed product construction for homeomorphisms and more generally the construction of Renault, Deaconu and Anantharaman-Delaroche for local homeomorphisms. Assuming that the map is surjective and not locally injective we give necessary and sufficient conditions for the simplicity of the C ∗-algebra and show that it is then a Kirchberg algebra. We provide tools for the calculation of the K-theory groups and turn them into an algorithmic method for Markov maps. 1. Introduction There are by now a wealth of ways to associate a C ∗-algebra to specific classes of dynamical systems, both reversible and irreversible. There are several different approaches to the construction, but the most successful is arguably the one which uses a groupoid as an intermediate step. Via the groupoid the study of how the C ∗-algebra depends on the dynamical system and which features it captures can be broken into smaller steps, relating the dynamical system to the groupoid and the groupoid to the C ∗-algebra. Since the C ∗-algebra arises as the convolution algebra of the groupoid there are by now a large reservoir of results which can be exploited for this purpose. As concrete examples of very successful studies which have followed this general recipe we mention only the work of Putnam and Spielberg, [PS], and the work of Deaconu and Shultz, [DS]. Another important aspect of the groupoid approach is the interpretation which the construction is given in the non-commutative geometry of Connes where the algebra is seen as a substitute for the poorly behaved quotient of the unit space by the equivalence relation coming from the action of the groupoid. In this picture the classical crossed product algebra coming from the action of a group by home- omorphisms is the non-commutative space representing the space of orbits under the action, and the groupoid which serves as the stepping stone from the dynamical system to the C ∗-algebra is the transformation groupoid. As a purely algebraic object the transformation groupoid of a homeomorphism can easily be described such that the invertibility of the dynamics is insignificant. It is therefore tempting and natural to try to base a generalization of the crossed prod- uct by a homeomorphism on the transformation groupoid. The problem is to equip this groupoid with a sufficiently nice topology which will allow the construction of the convolution algebra. It was shown by Renault, Deaconu and Anantharaman- Delaroche, in increasing generality, [Re], [De], [An], that for a local homeomorphism there is a canonical topology on the transformation groupoid which turns it into an ´etale locally compact Hausdorff groupoid, which is the ideal setting for the forma- tion of the convolution algebra. In [Th2] the second author introduced a way to Version: September 5, 2018. 1 2 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN describe the transformation groupoid of a homeomorphism, and more generally a local homeomorphism, in such a way that it leads to an ´etale groupoid and hence a C ∗-algebra for certain dynamical systems that are not local homeomorphisms. In [Th2] and [Th3] this generalization of the crossed product for homeomorphisms was used and investigated for holomorphic maps of Riemann surfaces. Such maps are open, but fail to be injective in any open set containing a critical point. The purpose with the present paper is to show how the method from [Th2] can be applied and what can be said about the resulting algebras, in a case where the maps are neither locally injective nor open. The class of maps we consider is the class of continuous piecewise monotone maps of the circle, and there is at least two natural ways in which the viewpoint from [Th2] can be applied. We make a thorough study of one of them. Specifically, from a continuous piecewise monotone map φ : T → T we construct an ´etale second countable locally compact Hausdorff groupoid Γ+ φ which is the usual transformation groupoid when φ is an orientation preserving homeomorphism and is equal to the groupoid of Renault, Deaconu and Anantharaman-Delaroche, [An], [De], [Re], when the map is an orientation preserving local homeomorphism. The C ∗-algebra we study is then the reduced C ∗-algebra C ∗ φ , cf. [Re]. φ(cid:1) of the groupoid Γ+ r(cid:0)Γ+ φ(cid:1) will be more messy than the one we obtain for surjective maps, and the r(cid:0)Γ+ φ(cid:1) is equal to the usual C ∗-algebra of a local r(cid:0)Γ+ We describe now our results. We assume that φ is surjective and not locally injective. Without surjectivity the necessary and sufficient condition for simplicity of C ∗ lack of generality seems insignificant at this point. Assuming that φ is not locally injective means that we exclude the surjective local homeomorphisms. This is done with good conscience because C ∗ homeomorphism when φ is a local homeomorphism of positive degree, and equal to that of φ2 when the degree is negative. In addition, the simple C ∗-algebras that arise from the transformation groupoid of a surjective local homeomorphism of the circle are known, cf. [AT], and there is therefore nothing lost by ignoring them here. We then establish the following facts. 1) When φ is transitive, C ∗ 2) C ∗ 3) C ∗ C ∗-subalgebra contains an infinite projection). (Proposition 4.5.) and has no non-critical fixed point x such that φ−1(x)\{x} only contains critical points. (Theorem 5.21.) φ(cid:1) is purely infinite (i.e. every non-zero hereditary r(cid:0)Γ+ r(cid:0)Γ+ φ(cid:1) is simple if and only if φ is exact (or, equivalently, totally transitive) φ(cid:1) is unital, separable, nuclear and satisfies the universal coefficient r(cid:0)Γ+ φ(cid:1) is classified by its K-theory, thanks to the Kirchberg-Phillips r(cid:0)Γ+ It follows from 1)-3) that when φ is exact without an exceptional fixed point the C ∗-algebra C ∗ classification result, [Ph]. To determine the algebra it suffices therefore to calculate its K-theory; a task which is far from trivial. We obtain here a six-terms exact sequence which can be be applied for the purpose (Theorem 6.5.), and we turn it into an effective algorithm for the calculation when φ is Markov in the sense that it takes critical points to critical points. (Section 7.) theorem (UCT) of Rosenberg and Schochet. (Corollary 6.7.) As a couple of key references for our work we mention the work of Shultz [S] and the work of Katsura [Ka]. The work of Shultz is used to show that a continuous piecewise monotone and transitive map on the circle is conjugate to a piecewise linear CIRCLE MAPS AND C ∗-ALGEBRAS 3 map with slopes that are constant in absolute value; an important step towards the proof of the pure infiniteness of C ∗ transitivity and exactness are equivalent in our setting; a fact which is important on the way to establish nuclearity and the UCT . For this purpose we use also the work of Katsura in much the same way it was used in [Th3], namely to show that C ∗ can be realized as a Cuntz-Pimsner algebra when it is simple. This is an important point. While the groupoid picture is the best approach for the study of many of the connections between properties of C ∗ it is the realization of the algebra as a Cuntz-Pimsner algebra which provides the decisive tools for the K-theory calculations. φ(cid:1). His work is also used to show that total r(cid:0)Γ+ φ(cid:1) r(cid:0)Γ+ φ(cid:1) and the dynamical properties of φ, r(cid:0)Γ+ The construction of Γ+ φ depends on the choice of a pseudo-group on T; see Section 2. This freedom is present already for homeomorphisms and local homeomorphisms, and this is why the C ∗-algebra C ∗ ing from a homeomorphism when it is orientation preserving. While the choice of pseudo-group seems canonical for homeomorphisms and local homeomorphisms, this is much less so in our case where at least two choices seem equally natural; either one can work with the pseudo-group of all locally defined local homeomorphisms on T, or one can restrict to those that preserve the orientation of the circle. In the construction of Γ+ φ we choose the latter and we postpone the study of the algebra which results when the largest pseudo-group is chosen. φ(cid:1) only generalizes the crossed product aris- r(cid:0)Γ+ 2. Transformation groupoid C ∗-algebras Let G be an ´etale second countable locally compact Hausdorff groupoid with unit space G(0). Let r : G → G(0) and s : G → G(0) be the range and source maps, respectively. For x ∈ G(0) put Gx = r−1(x), Gx = s−1(x) and Isx = s−1(x) ∩ r−1(x). Note that Isx is a group, the isotropy group at x. The space Cc(G) of continuous compactly supported functions is a ∗-algebra when the product is defined by (f1f2)(g) = Xh∈Gr(g) f1(h)f2(h−1g) and the involution by f ∗(g) = f (g−1). Let x ∈ G(0). There is a representation πx of Cc(G) on the Hilbert space l2(Gx) of square-summable functions on Gx given by The reduced groupoid C ∗-algebra C ∗ to the norm r (G), [Re], is the completion of Cc(G) with respect kf k = sup x∈G(0) kπx(f )k . In [Th2] the second named author introduced an amended transformation groupoid for a self-map on a locally compact space which in certain cases allows us to topolo- gize the transformation groupoid, or a groupoid closely related to the transformation groupoid, in such a way that the result is a well behaved ´etale groupoid. It is this construction we shall consider in this paper for piecewise monotone maps of the circle. We begin therefore by reviewing the construction. Let X be a locally compact Hausdorff space and ψ : X → X a map. Let P be a pseudo-group on X. More specifically, P is a collection of homeomorphisms η : U → V between open subsets of X such that πx(f )ψ(g) = Xh∈Gr(g) f (h)ψ(h−1g). (2.1) 4 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN i) for every open subset U of X the identity map id : U → U is in P, ii) when η : U → V is in P then so is η−1 : V → U, and iii) when η : U → V and η1 : U1 → V1 are elements in P then so is η1 ◦ η : U ∩ η−1(V ∩ U1) → η1(V ∩ U1). For each k ∈ Z we denote by Tk(ψ) the elements η : U → V of P with the property that there are natural numbers n, m such that k = n − m and ψn(z) = ψm(η(z)) ∀z ∈ U. (2.2) The elements of T =Sk∈Z Tk(ψ) will be called local transfers for ψ. We denote by [η]x the germ at a point x ∈ X of an element η ∈ Tk(ψ). Set Gψ = {(x, k, η, y) ∈ X × Z × P × X : η ∈ Tk(ψ), η(x) = y} . We define an equivalence relation ∼ in Gψ such that (x, k, η, y) ∼ (x′, k′, η′, y′) when i) x = x′, y = y′, k = k′ and ii) [η]x = [η′]x. Let [x, k, η, y] denote the equivalence class represented by (x, k, η, y) ∈ Gψ. The quotient space Gψ(P) = Gψ/∼ is a groupoid such that two elements [x, k, η, y] and [x′, k′, η′, y′] are composable when y = x′ and their product is [x, k, η, y] [y, k′, η′, y′] = [x, k + k′, η′ ◦ η, y′] . The inversion in Gψ(P) is defined such that [x, k, η, y]−1 = [y, −k, η−1, x]. The unit space of Gψ can be identified with X via the map x 7→ [x, 0, id, x], where id is the identity map on X. When η ∈ Tk(ψ) we set U(η) = {[z, k, η, η(z)] : z ∈ U} (2.3) where U is the domain of η. It is straightforward to verify that by varying k, η and U the sets (2.3) constitute a base for a topology on Gψ(P). In general this topology is not Hausdorff and to amend this we now make the following additional assumption. Assumption 2.1. When x ∈ X and η(x) = ξ(x) for some η, ξ ∈ Tk(ψ), then the implication x is not isolated in {y ∈ X : η(y) = ξ(y)} ⇒ [η]x = [ξ]x holds. Then Gψ(P) is Hausdorff: Let [x, k, η, y] and [x′, k′, η′, y′] be different elements of Gψ(P). There are then open neighbourhood's W of x and W ′ of x′ such that U(ηW ) = {[z, k, η, η(z)] : z ∈ W } and U ′(η′W ′) = {[z, k′, η′, η′(z)] : z ∈ W ′} are disjoint. This is trivial when (x, k, y) 6= (x′, k′, y′) while it is a straightforward consequence of Assumption 2.1 when (x, k, y) = (x′, k′, y′). Since the range and source maps are homeomorphisms from U(η) onto U and η(U), respectively, it follows that Gψ(P) is a locally compact Hausdorff space because X is. It is also straightforward to show that the groupoid operations are continuous so that we can conclude the following. Theorem 2.2. Let Assumption 2.1 be satisfied. Then Gψ(P) is an ´etale locally compact Hausdorff groupoid. Compared to the notation used in [Th2] we have here emphasized the pseudo- group since there is more than one natural choice when ψ is a piecewise monotone map on the circle. CIRCLE MAPS AND C ∗-ALGEBRAS 5 3. The oriented transformation groupoid of a piecewise monotone map on the circle Let T be the unit circle in the complex plane. We consider T as an oriented space with the canonical counter-clockwise orientation. Consider a continuous map φ : T → T. There is then a unique continuous map f : [0, 1] → R such that f (0) ∈ [0, 1[ and φ(cid:0)e2πit(cid:1) = e2πif (t) for all t ∈ [0, 1]. We will refer to f as the lift of φ. Note that f (1) − f (0) is an integer, the degree of φ. We say that φ is piecewise monotone when there are points 0 = c0 < c1 < · · · < cN = 1 such that f is either strictly increasing or strictly decreasing on the intervals ]ci−1, ci[, i = 1, 2, . . . , N. When φ : T → T is piecewise monotone and t ∈ T we define the φ-valency val(φ, t) of t to be the element of V = {(+, +), (−, −), (+, −), (−, +)} such that val(φ, t) = (+, +) when φ is strictly increasing in all sufficiently small neighborhoods of t; val(φ, t) = (−, −) when φ is strictly decreasing in all sufficiently small open neighborhoods of t; val(φ, t) = (+, −) when φ is strictly increasing in all sufficiently small intervals to the left of t and strictly decreasing in all sufficiently small intervals to the right of t; and finally val(φ, t) = (−, +) when φ is strictly decreasing in all sufficiently small intervals to the left of t and strictly increasing in all sufficiently small intervals to the right of t. The set V is a monoid with the following composition table. x • y y = (+, +) y = (+, −) y = (−, +) y = (−, −) x = (+, +) x = (+, −) x = (−, +) x = (−, −) (+, +) (+, −) (−, +) (−, −) (+, −) (+, −) (−, +) (−, +) (−, +) (+, −) (−, +) (+, −) (−, −) (+, −) (−, +) (+, +) Table 1. The composition table for • This monoid structure will be important here because of the following observation: Lemma 3.1. Let φ, ϕ : T → T be piecewise monotone, and let x ∈ T. Then val(φ ◦ ϕ, x) = val(φ, ϕ(x)) • val(ϕ, x). Let P + be the pseudo-group of all locally defined homeomorphisms of T that are orientation preserving. The elements of P + consist of open subsets U, V in T and a homeomorphism η : U → V such that s, t ∈ U, s < t ⇒ η(s) < η(t). Lemma 3.2. Let φ, ϕ : T → T be continuous and piecewise monotone maps and x, y ∈ T points such that φ(x) = ϕ(y). It follows that there is a η ∈ P + such that a) η(x) = y, and b) φ(t) = ϕ(η(t)) for all t in a neighborhood of x if and only if val(φ, x) = val(ϕ, y). When this is the case, the germ [η]x of η at x is unique. Proof. Straightforward. (cid:3) 6 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN When x, y ∈ T and k ∈ Z we write x k∼ y when k = n − m for some n, m ∈ N such that φn(x) = φm(y) and val (φn, x) = val (φm, y). Set Then Γ+ φ is a groupoid where the composable pairs are Γ+ φ =n(x, k, y) ∈ T × Z × T : x k∼ yo . 2 : y = x′o (2) =n((x, k, y), (x′, k′, y′)) ∈ Γ+ φ Γ+ φ and the product is (x, k, y)(y, k′, y′) = (x, k + k′, y′). The inversion is given by (x, k, y)−1 = (y, −k, x). This groupoid is identical with the groupoid denoted by Gφ(P +) in Section 2. To see this we denote, as in [Th2], by Tk(φ) the set of elements η ∈ P + with the property that for some n, m ∈ N such that n − m = k, the equality φn(z) = φm(η(z)) holds for all z in the domain of η. It follows then from the last statement in Lemma 3.2 that the implication η, η′ ∈ Tk(φ), η(x) = η′(x) ⇒ [η]x = [η′]x holds. We deduce therefore that the map Gφ(cid:0)P +(cid:1) ∋ [x, k, η, y] 7→ (x, k, y) ∈ Γ+ φ is a bijection. Hausdorff groupoid in the topology for which a base is given by sets of the form It follows from Theorem 2.2 that Γ+ φ is an ´etale locally compact Ω (η, U) = {(z, k, η(z)) : z ∈ U} , (3.1) φ : and Note that Γφ(k, n) is closed in T × Z × T. where η ∈ Tk(φ) and U is an open subset of η's domain. However, there is an alternative description of this topology which we now present. Among others it has the virtue that it is obviously second countable. When k ∈ Z, n ∈ N and n + k ≥ 1, n ≥ 1, set Γ+ l = k, φk+n(x) = φn(y), val(cid:0)φk+n, x(cid:1) = val (φn, y)(cid:9) φ (k, n) =(cid:8)(x, l, y) ∈ Γ+ φ (k, n) is the intersection of a closed and an open subset of T×Z×T. Γφ(k, n) =(cid:8)(x, l, y) ∈ T × Z × T : l = k, φk+n(x) = φn(y)(cid:9) . Proof. Write T as the union of non-degenerate closed intervals T =Si I + T as the union of non-degenerate closed intervals T = Sj J + j ∪Sj J − i such that φn+k is increasing on each I + for all i, and i such that none of the intervals overlap in more than one point. Similarly, write such φn is for all j, and such that none increasing on each J + for all j and decreasing on J − j j of the intervals overlap in more than one point. Then for all i and decreasing on I − i i ∪Si I − Lemma 3.3. Γ+ j Γ+ φ (k, n) = (A ∩ Γφ(k, n)) \B, where A =[i,j (cid:0)I + j (cid:1) ∪[i,j (cid:0)I − i × {k} × J + i × {k} × J − j (cid:1) CIRCLE MAPS AND C ∗-ALGEBRAS 7 and B is the finite set consisting of elements (x, k, y) ∈ Γφ(n, k) such that val(cid:0)φn+k, x(cid:1) ∈ {(+, −), (−, +)} or val (φn, y) ∈ {(+, −), (−, +)} while val(cid:0)φn+k, x(cid:1) 6= val (φn, y). (cid:3) It follows from Lemma 3.3 that Γ+ φ (k, n) is a locally compact Hausdorff space in the relative topology inherited from T × Z × T. Lemma 3.4. A subset W of Γ+ φ (k, n) is open in the relative topology inherited from T × Z × T if and only for all (x, k, y) ∈ W there is an element η ∈ Tk(φ) and an open subset U of the domain of η such that (x, k, y) ∈ Ω(η, U) ⊆ W . φ (k, n) is open in the relative topology of Γ+ Proof. Assume first that W ⊆ Γ+ φ (k, n) is open in the relative topology inherited from T × Z × T and consider a point (x, k, y) ∈ W . It follows from Lemma 3.2 that there is an element η ∈ Tk(φ) such that η(x) = y and φk+n(z) = φn(η(z)) for all z in a neighborhood U of x. By continuity of η we can shrink U to arrange that (x, k, y) ∈ Ω(η, U) ⊆ W . This establishes one implication. To prove the other, let η ∈ Tk(φ) and let U be an open subset of the domain of η. We must show that Ω(η, U) ∩ Γ+ φ (k, n) inherited from T × Z × T. Let (x, k, y) ∈ Ω(η, U) ∩ Γ+ It follows from Lemma 3.2 that φn+k(z) = φn (η(z)) for all z sufficiently close to x. If (z, z′) is sufficiently close to (x, y) in T × T and (z, k, z′) ∈ Γ+ and φn+k(z) = φn(z′) imply that z ≤ x ⇔ z′ ≤ y. Combined with the fact that φn+k(z) = φn (η(z)) = φn(z′) we conclude from this that z′ = η(z) when (z, z′) is sufficiently close to (x, y) and (z, k, z′) ∈ Γ+ φ (k, n). That is, there is an open neighborhood V of (x, k, y) in T× Z× T such that V ∩Γ+ φ (k, n). (cid:3) φ (k, n), the conditions val(cid:0)φn+k, z(cid:1) = val (φn, z′) φ (k, n) ⊆ Ω(η, U)∩Γ+ φ (k, n). In combination with Lemma 3.2 it follows from Lemma 3.4 that Γ+ φ (k, n) is an open subset of Γ+ φ (k, n + 1). Therefore is locally compact and Hausdorff in the inductive limit topology, and the disjoint union Γ+ φ (k, n) Γ+ φ (k) = [n≥−k+1 φ = Gk∈Z Γ+ Γ+ φ (k) is a locally compact Hausdorff in the topology where each Γ+ φ (k) is closed and open and has the topology just defined. By Lemma 3.4 this topology is identical with the one we obtain from Theorem 2.2. Note that the topology is second countable since the topology of T × Z × T is. This proves the following Lemma 3.5. Γ+ φ is a second countable locally compact Hausdorff ´etale groupoid. We assume now and throughout the paper that φ : T → T is continuous and piecewise monotone. The objective is to investigate the structure of C ∗ particular, when it is simple. Let us first consider the case where φ is a local homeomorphism so that the groupoid Γφ of Renault, Deaconu and Anantharaman-Delaroche is defined, cf. [Re], [De] and [An]. φ(cid:1); in r(cid:0)Γ+ 8 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN Lemma 3.6. Assume that φ is a local homeomorphism. If the degree of φ is positive, there is an isomorphism Γ+ If the degree of φ is negative there is an isomorphism Γ+ φ ≃ Γφ of topological groupoids. φ ≃ Γφ2 of topological groupoids. Proof. This follows straightforwardly from the observation that val (φ, x) = (+, +) for all x ∈ T when the degree is positive and val (φ, x) = (−, −) for all x ∈ T when the degree is negative. (cid:3) r(cid:0)Γ+ φ(cid:1) = C ∗ r (Γφ) when the degree of φ is positive, and C ∗ It follows from Lemma 3.6 that for a local homeomorphism φ we have the equality C ∗ r (Γφ2) when it is negative. The simple C ∗-algebras of the form C ∗ r (Γψ) for a surjective, local homeomorphism ψ of the circle were all described in [AT] and we will therefore here assume that φ is not locally injective. Thus, for the remaining part of the paper φ will be continuous, piecewise monotone, surjective and not locally injective. r(cid:0)Γ+ φ(cid:1) = C ∗ 4. Transitivity implies piecewise linearity and pure infiniteness Let s > 0. A continuous function g : [0, 1] → R is uniformly piecewise linear with slope s when there are points 0 = a0 < a1 < a2 < · · · < aN = 1 such that g is linear with slope ±s on each interval [ai−1, ai] , i = 1, 2, . . . , N. We say that φ is uniformly piecewise linear with slope s when its lift f : [0, 1] → R is. Recall that a continuous map h : X → X on a compact metric space X is transitive when for each pair U, V of non-empty open sets in X there is an n ∈ N such that hn(U) ∩ V 6= ∅. When h is surjective, as in the case we consider, transitivity is equivalent to the existence of a point with dense forward orbit. Theorem 4.1. Assume that φ is transitive. It follows that there is an orientation- preserving homeomorphism h : T → T such that h ◦ φ ◦ h−1 is uniformly piecewise linear with slope s > 1. Proof. We will show how the theorem follows from the work of Shultz in [S] on discontinuous piecewise monotone maps of the interval. After conjugation by a rotation of the circle we can assume that φ(1) 6= 1 and 1 /∈ φ(C1). (Indeed, since φ is piecewise monotone and transitive there are λ's in T arbitrary close to 1 such that φ(λ) 6= λ. Choose one of them such that λ /∈ φ(C1). Then φ1(t) = λ−1φ(λt) is conjugate to φ, does not fix 1 and all its critical values are different from 1.) Let µ : T → [0, 1[ be the inverse map of [0, 1[∋ t 7→ e2πit. Then τ (t) = µ ◦ φ(cid:0)e2πit(cid:1) is piecewise monotone in the sense of Shultz [S]. Since φ is surjective and 1 /∈ φ (C1 ∪ {1}), it follows that τ is discontinuous at a point in ]0, 1[ and τ ([0, 1]) = [0, 1[. We claim that τ is transitive in the sense of Definition 2.6 in [S]; that is, we claim that for every open non-empty subset U ⊆ [0, 1] there is an n ∈ N such that the left and right hand limits of τ at x. By construction this union is either {τ (x)} Herebτ is the possibly multivalued map on [0, 1] which associates to each x ∈ [0, 1] or {1, 0}. In the latter case 0 = τ (x). It follows therefore thatbτ (A)\{1} = τ (A) for every subset A ⊆ [0, 1]. Thus (4.1) n[i=0bτ k(U) = [0, 1] bτ k(U) ⊇ τ k(U) CIRCLE MAPS AND C ∗-ALGEBRAS 9 i=0 τ k(U) = [0, 1[ for some n ∈ N. As observed above τ is discontinuous at a point in ]0, 1[. It follows therefore for all k. The strong transitivity of φ implies that Sn−1 that 1 ∈bτ ([0, 1[) and hence that (4.1) holds since n[i=0bτ i(U) ⊇bτ n−1[i=0bτ i(U)! ⊇bτ n−1[i=0 τ i(U)! =bτ ([0, 1[) = [0, 1]. It follows now from Propositions 4.3 and 3.6 in [S] that there is a homeomorphism h : [0, 1] → [0, 1] such that f = h ◦ τ ◦ h−1 is uniformly piecewise linear. From the proof of Proposition 3.6 in [S] we see that h is increasing. Since φ is not locally injective there are non-empty open intervals I, I ′ ⊆ T\{1} such that I ∩ I ′ = ∅ and φ(I) = φ(I ′). Then J = µ(I) and J ′ = µ(I ′) are non-empty open intervals in [0, 1[ such that J ∩ J ′ = ∅ and τ (J) = τ (J ′), i.e. τ is not essentially injective in the sense of Definition 4.1 of [S]. Hence the slope s of the linear pieces of f is > 1 by Proposition 4.3 of [S]. Since h(0) = 0, h(1) = 1 and τ (0) = τ (1) we find that f (0) = f (1) and we can therefore define ϕ : T → T such that ϕ (e2πit) = e2πif (t), t ∈ [0, 1]. Then ϕ = g◦φ◦g−1 where g = µ−1 ◦ h ◦ µ. Then g(1) = 1 = limλ→1 g(λ). Hence g is continuous and an orientation preserving homeomorphism on T. It follows that ϕ is a continuous map on T and conjugate to φ. By construction ϕ is uniformly piecewise linear with slope s > 1. (cid:3) We say that a p-periodic point x ∈ T is repelling when there is an open interval I in T and a r > 1 such that x ∈ I and φp(y) − x ≥ r y − x for all y ∈ I. Lemma 4.2. Assume that φ is transitive and uniformly piecewise linear with slope s > 1. Then the periodic points of φ are dense in T and they are all repelling. Proof. Since φ is transitive there is a point in T with dense forward orbit, cf. The- orem 5.9 in [W]. It follows therefore from Corollary 2 in [AK] that φ has periodic points, and then by Corollary 3.4 in [CM] that the periodic points are dense. For each n ∈ N the map φn is uniformly piecewise linear with slope sn > 1. Therefore all periodic points of φ are repelling. (cid:3) Corollary 4.3. Assume that φ is transitive. It follows that Γ+ in the sense of [An]. φ is locally contractive Proof. By Theorem 4.1 we may assume that φ is piecewise linear with slope s > 1. Let U be an open non-empty subset of T. By Lemma 4.2 there is in U a point which is periodic and repelling. Since there are only finitely many critical points there are also only finitely many periodic orbits which contain a critical point. Hence U contains a periodic point x, say of period n, which is repelling and whose orbit does not contain a critical point. Then val (φ2n, x) = (+, +) and there is therefore an open neighborhood W ⊆ U of x and a κ > 1 such that val (φ2n, y) = (+, +) and φ2n(y) − x ≥ κ y − x for all y ∈ W . The proof is then completed exactly as the proof of Proposition 4.1 in [Th4]. (cid:3) Lemma 4.4. Assume that φ is transitive. It follows that Γ+ the sense of [An], i.e. the points in T with trivial isotropy group in Γ+ T. φ is essentially free in φ are dense in 10 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN Proof. A point in T has non-trivial isotropy group only when it is pre-periodic. It suffices therefore to show that the set of pre-periodic points has empty interior in T; a fact which follows easily from the assumed transitivity of φ. (cid:3) Proposition 4.5. Assume that φ is transitive. It follows that C ∗ infinite in the sense that every non-zero hereditary C ∗-subalgebra contains an infinite projection. φ(cid:1) is purely r(cid:0)Γ+ Proof. This follows from Lemma 4.4 and Corollary 4.3, thanks to Proposition 2.4 in [An]. (cid:3) There is one more fact about piecewise monotone circle maps which can be de- duced from the work of Shultz in [S], and which we shall use below. Recall that a continuous map h : X → X on a compact Hausdorff space X is totally transitive when hn is transitive for all n ∈ N, and exact when for all open non-empty subsets U ⊆ X there is an N ∈ N such that hN (U) = X. Lemma 4.6. φ is exact if and only if φ is totally transitive. Proof. It is obvious that exactness implies total transitivity. To prove the converse, return to the notation introduced in the proof of Theorem 4.1 and assume that φ is totally transitive. As observed in that proof, τ is then transitive and not essentially injective in the sense of [S]. It follows therefore from Corollary 4.7 in [S] that there is an N ∈ N and closed sets Ki, i = 1, 2, . . . , N, with mutually disjoint non-empty interiors Int Ki in [0, 1] such that τ N maps Int Ki onto Int Ki and is exact on Ki for each i. The image of Int Ki in T is open and invariant under φN and must therefore be all of T since φ is totally transitive. This implies that N = 1, which means that τ is exact. It follows that φ is exact as well. (cid:3) For x ∈ T, let RO+(x) be the Γ+ φ -orbit of x, i.e. 5. Simplicity Γ+ Γ+ RO+(x) = {y ∈ T : φn(x) = φm(y), val (φn, x) = val (φm, y) for some n, m ∈ N} . A subset A ⊆ T will be called restricted orbit invariant or RO+-invariant when x ∈ A ⇒ RO+(x) ⊆ A. Let Y ⊆ T be a closed RO+-invariant subset. Then the reduction φ and an ´etale groupoid in the topology inherited from is a closed subgroupoid of Γ+ Γ+ φ . The same is true for the reduction φ Y =(cid:8)(x, k, y) ∈ Γ+ φ : x, y ∈ Y(cid:9) φ T\Y =(cid:8)(x, k, y) ∈ Γ+ φ : x, y ∈ T\Y(cid:9) . φ(cid:1) → Cc(cid:0)Γ+ The restriction map Cc(cid:0)Γ+ φ Y(cid:1) extends to a ∗-homomorphism πY : r(cid:0)Γ+ φ(cid:1) → C ∗ φ Y(cid:1) and the inclusion Cc(cid:0)Γ+ r(cid:0)Γ+ φ(cid:1) extends to an em- φ T\Y(cid:1) ⊆ Cc(cid:0)Γ+ r(cid:0)Γ+ φ T\Y(cid:1) ⊆ C ∗ r(cid:0)Γ+ r(cid:0)Γ+ φ(cid:1) which realizes C ∗ φ T\Y(cid:1) as an ideal in C ∗ r(cid:0)Γ+ φ(cid:1). φ(cid:1) πY C ∗ r(cid:0)Γ+ r(cid:0)Γ+ r(cid:0)Γ+ φ Y(cid:1) φ T\Y(cid:1) C ∗ bedding C ∗ It is straightforward to adopt the proof of Lemma 3.2 in [Th3] to obtain the following. Lemma 5.1. Let Y be a closed RO+-invariant subset of T. It follows that 0 C ∗ C ∗ is exact. 0 CIRCLE MAPS AND C ∗-ALGEBRAS 11 In particular, C ∗ invariant subsets of T. We aim now to show that this is the only obstruction. φ(cid:1) is not simple when there are non-trivial closed RO+- r(cid:0)Γ+ For the statement of the next lemma recall that the full orbit of a point x ∈ T is the set {y ∈ T : φn(y) = φm(x) for some n, m ∈ N}. For each j ∈ N we let Cj denote the critical points of φj, i.e. Cj =(cid:8)t ∈ T : val(cid:0)φj, t(cid:1) ∈ {(+, −), (−, +)}(cid:9) . n=0 φ−n(C1) are then the pre-critical points. The elements ofS∞ Lemma 5.2. Assume that there is a point x ∈ T whose full orbit is dense in T. It follows that there is a point in T which is neither pre-periodic nor pre-critical. Proof. Let Pern be the set of points in T of minimal period n. Assume for a contra- diction that T = [n,k∈N φ−k (Pern ∪C1) . φj (I−)! = ∅. By the Baire category theorem this implies that there are k, n ∈ N such that φ−k (Pern ∪C1) contains a non-degenerate interval. Since φk is piecewise mono- tone and C1 finite this implies that Pern contains a non-degenerate interval. Then i=0 φi (I+) and Pern also contains two non-empty open intervals I+, I− such that Sn Sn i=0 φi (I−) are disjoint. It follows that By assumption there is a point x with dense full orbit. Since both I+ and I− contain an element from this orbit it follows that here is a k ∈ N such that φj (I+)! ∩ ∞[j=0 ∞[j=0 φk(x) ∈ ∞[j=0 φj (I+)! ∩ ∞[j=0 φ(cid:1) is simple if and only of RO+(x) is dense in r(cid:0)Γ+ φ(cid:1) implies that RO+(x) is dense for all x by Lemma 5.1. r(cid:0)Γ+ φ(cid:1) will follow if we can show that not all points of T have r(cid:0)Γ+ Proof. Simplicity of C ∗ For the converse assume that RO+(x) is dense for all x. By Corollary 2.18 in [Th1] the simplicity of C ∗ non-trivial isotropy in Γ+ φ . Since a point with non-trivial isotropy is pre-periodic it suffices to show that not all points of T are pre-periodic under φ. This follows from Lemma 5.2. (cid:3) Lemma 5.3. The C ∗-algebra C ∗ T for all x ∈ T. φj (I−)! . (5.1) (cid:3) This contradicts (5.1). Lemma 5.4. Assume that C ∗ φ(cid:1) is simple. Then φ is transitive. r(cid:0)Γ+ Proof. Let E ⊆ T be closed, with non-empty interior and φ-invariant in the sense that φ(E) ⊆ E. By Theorem 5.9 [W] it suffices to show that E = T. For each n, m ≥ 1 set Un,m = {x ∈ T : φn(x) = φm(y), val (φn, x) = val (φm, y) for some y ∈ Int E} where Int E is the interior of E. Note that Un,m is open and non-empty and that Sn,m Un,m is RO+-invariant. It follows therefore from Lemma 5.3 thatSn,m Un,m = 12 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN φ−n(E) we find then that T. By compactness there is an N ∈ N such that T = SN φ−n(E)! ⊆ E. T = φN (T) ⊆ φN N[n=1 n,m=1 Un,m. Since Un,m ⊆ (cid:3) The converse of Lemma 5.4 is not true in general; transitivity of φ does not imply that C ∗ will be given in Theorem 5.21. φ(cid:1) r(cid:0)Γ+ φ(cid:1) is simple. A necessary and sufficient condition for simplicity of C ∗ r(cid:0)Γ+ φ(cid:1). The elements in φ(C1) are r(cid:0)Γ+ n=1 φn(C1) are the post-critical points. Note 5.1. Finite RO+-orbits and quotients of C ∗ the critical values and the elements ofS∞ that a critical point is pre-critical, but not necessarily post-critical. Lemma 5.5. Assume that φ is transitive. Let A ⊆ T be a non-empty RO+-invariant subset which is not dense in T. It follows that A is finite and consists of points that are post-critical and not pre-critical. Proof. By assumption there is an open non-empty interval J ⊆ T such that By Corollary 4.2 of [Y] φ is not only transitive, but also strongly transitive. There is therefore an N ∈ N such that A ∩ J = ∅. (5.2) φi(J) = T. (5.3) N[i=0 def If x ∈ A and val (φj, x) ∈ {(+, −), (−, +)} for some j ≥ 1 we can choose y ∈ J such that φk(y) = x for some k ∈ {1, 2, . . . , N}. It follows from the composition table for Since φ is not locally injective there is a z ∈ T such that val (φ, z) ∈ {(+, −), (−, +)}. φk+1 (J−) T. Set Mi = I ∩ Ci. Let a ∈ T be a non-critical element, = I. Since φ is strongly transitive there is a K ∈ N such thatSK contradicting (5.2). It follows that val (φj, x) ∈ {(+, +), (−, −)} for all j ∈ N when x ∈ A; i.e A consists of points that are not pre-critical. • that val(cid:0)φk+j, y(cid:1) = val(φj, x) • val(φk, y) = val (φj, x). Hence y ∈ RO+(x) ⊆ A, Choose z0 ∈ J and k ∈ {1, 2, . . . , N} such that φk(z0) = z and note that val(cid:0)φk+1, z0(cid:1) ∈ {(+, −), (−, +)}. There are therefore subintervals J+, J− of J such that val(cid:0)φk+1, y(cid:1) = (+, +) when y ∈ J+, val(cid:0)φk+1, y(cid:1) = (−, −) when y ∈ J−, and φk+1 (J+) = {(+, +), (−, −)}. Assume that a /∈SK that φk+1(y) = y′ and val(cid:0)φi+k+2, y(cid:1) = val (φi+1, y′) • val(cid:0)φk+1, y(cid:1) = val (φ, a). It The last two paragraphs show that A ⊆SK becauseSK i=1 φi (I) = i.e. val(φ, a) ∈ i=1 φi(Mi). We claim that RO+(a) ∩ J 6= ∅. To see this note that there is an i ∈ {1, 2, . . . , K} and a y′ ∈ I\Mi such that φi(y′) = a. Then val (φi, y′) ∈ {(+, +), (−, −)} and there is also an element y ∈ J+ ∪ J− such Call a point x ∈ T exposed when RO+(x) is finite. By Proposition 5.3 and Lemma 5.5 it is the possible presence of exposed points which is the only obstruction for simplicity of C ∗ i=1 φi(Mi). This completes the proof (cid:3) i=1 φi(Mi) is finite and consists of post-critical points. follows that y ∈ RO+(a) ∩ J, proving the claim. φ(cid:1). r(cid:0)Γ+ CIRCLE MAPS AND C ∗-ALGEBRAS 13 Corollary 5.6. Assume that φ is transitive. Then C ∗ there are no exposed points. φ(cid:1) is simple if and only if r(cid:0)Γ+ 5.1.1. deg φ ≥ 2. Lemma 5.7. Assume that φ is transitive and that deg φ ≥ 2. RO+(x) is dense in T for all x ∈ T. It follows that Proof. Let n ∈ N. By looking at the graph of a lift f : [0, 1] → R of φ2n one sees that for any x ∈ T, the set An =(cid:8)y ∈ T : φ2n(y) = x, val(cid:0)φ2n, y(cid:1) = (+, +)(cid:9) contains at least deg φ2n elements. Since An ⊆ RO+(x) we conclude that RO+(x) is infinite for all x ∈ T. It follows then from Lemma 5.5 that RO+(x) is dense for all x. (cid:3) Proposition 5.8. Assume that φ is transitive and that deg φ ≥ 2. It follows that C ∗ φ(cid:1) is simple. r(cid:0)Γ+ 5.1.2. deg φ = 1. Before we specialize to the case where the degree is 1 or −1 we need a couple of more general facts. Lemma 5.9. Assume that φ is transitive, but not totally transitive. It follows that there is a p > 1 and closed intervals Ii, i = 0, 1, 2, . . . , p − 1, such that 1) φ(Ii) = Ii+1 (addition mod p), 2) Ii ∩ Int Ij = ∅, i 6= j, i=0 Ii = T, 3) Sp−1 4) φpIi is totally transitive for each i. Proof. This is a special case of Corollary 2.7 in [AdRR]. (cid:3) Note that the number p and the collection {I0, I1, . . . , Ip−1} of intervals in Lemma 5.9 are unique. We will refer to p as the global period of φ, and say that it is 1 when φ is totally transitive. In the following we denote the set of endpoints of the intervals Ii from Lemma 5.9 by E. Lemma 5.10. Assume that φ is transitive but not totally transitive. Then φ−1 (E) \C1 = E. (5.4) Proof. Assume for a contradiction that e ∈ E, but φ(e) /∈ E. There are then intervals Ii, Ii′, Ij as in Lemma 5.9 such that i 6= i′, e ∈ Ii ∩ Ii′ and φ(e) ∈ Int Ij. By continuity of φ and condition 1) from Lemma 5.9 it follows that Ii+1 ∩ Int Ij 6= ∅ and Ii′+1 ∩ Int Ij 6= ∅. Since i + 1 6= i′ + 1 this violates condition 2). Thus φ(E) ⊆ E. (5.5) If e ∈ E is a critical point the images Ii+1 = φ (Ii) and Ii′+1 = φ (Ii′) of the two intervals Ii, Ii′ containing e will both have non-trivial intersection with the same interval Ij containing φ(e); contradicting 2) again. Hence (5.6) Consider then an element x ∈ φ−1(E) and assume that x /∈ C1. Let Ii and Ii′ be the two intervals among the intervals from Lemma 5.9 which contain φ(x). If x /∈ E there is a third interval Ij which contains x in its interior. Since x is not critical E ∩ C1 = ∅. 14 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN it follows that φ(Ij) = Ij+1 has non-trivial intersection with both Int Ii and Int Ii′, contradicting 2) once more. Hence This completes the proof since (5.4) is equivalent to (5.5), (5.6) and (5.7). φ−1(E)\C1 ⊆ E. (5.7) (cid:3) Lemma 5.11. Assume that φ is transitive but not totally transitive. When deg φ = 1 the set E is a p-periodic orbit where p is the global period of φ, and val (φ, x) = (+, +) for all x ∈ E. When deg φ = −1 the global period of φ is 2 and E consists of two distinct fixed points of valency (−, −). Proof. Let Ii, i = 0, 1, · · · , p − 1, be the intervals from Lemma 5.9, and let e− i , be the left endpoint of Ii, defined using the orientation of T. When deg φ = 1 we see by looking at the graph of a lift of φ that val(cid:0)φ, e− (addition mod p). It follows that E = RO+(cid:0)e− orbit of e− 0 . When deg φ = −1 observe first φ has a fixed point x. This fixed point lies in one of the intervals Ii. Since x also lies in Ii+1 and Ii+2 it follows that two of the intervals Ii, Ii+1 and Ii+2 must be the same, i.e. p = 2. By looking at the graph of a lift of φ we see that E consists of two fixed points of valency (−, −). (cid:3) i(cid:1) = (+, +) and φ(cid:0)e− i(cid:1) = e− 0(cid:1), and that this is also the (forward) i+1 Lemma 5.12. If deg φ = 1 there is for all x ∈ T an element y ∈ φ−1(x) such that val (φ, y) = (+, +). If deg φ = −1 there is for all x ∈ T an element y ∈ φ−1(x) such that val (φ, y) = (−, −). Proof. Look at the graph of a lift of φ. (cid:3) Lemma 5.13. Assume that deg φ ∈ {1, −1}. Then RO+(x) is infinite for all x ∈ T that are not periodic under φ. Proof. Let x ∈ T. It follows from Lemma 5.12 that there are sequences {ni} in N and {xi} in T such that φn1(x1) = x, φni(xi) = xi−1, i ≥ 2, and val (φni, xi) = (+, +) for all i. Then xi ∈ RO+(x) for all i. The set {xi : i ∈ N} is infinite when x is not periodic. (cid:3) Lemma 5.14. Assume that deg φ ∈ {−1, 1} and φ is transitive but not totally transitive. Then E is the set of exposed points for φ. Proof. Let e ∈ E and y ∈ RO+(e). There are natural numbers i, j ∈ N such that φi(e) = φj(y) and val (φi, e) = val (φj, y). It follows from (5.4) φj(y) = φi(x) ∈ E and that val (φi, e) ∈ (±, ±) since e ∈ E. This implies first that val(cid:0)φ, φk(y)(cid:1) ∈ (±, ±) for all k ≤ j − 1 and then that φj−1(y) ∈ φ−1(E)\C1 = E. But then φj−2(y) ∈ φ−1(E)\C1 = E and so on. After j steps we conclude that y ∈ E. This shows that E is RO+-invariant. It remains to show that E contains all exposed points. Assume therefore that y0 is an exposed point. It follows from Lemma 5.13 that all exposed points are periodic. Since they are also post-critical by Lemma 5.5 and there are only finitely many critical points it follows that there are only finitely many exposed points. Let m ∈ N be an even number divisible by the global period p and by all the periods of exposed points. Then φm(y0) = y0. Furthermore, if z ∈ φ−m(y0) and val (φm, z) = (+, +) we see that z ∈ RO+(y0) and hence z is exposed. By definition of m this implies that φm(z) = z, i.e. z = y0. To see that there can not be any CIRCLE MAPS AND C ∗-ALGEBRAS 15 z ∈ φ−m(y0) with val (φm, z) = (−, −) observe by looking at the graph of the lift of φm, that since deg φm = 1 the existence of such a z would imply the existence of a z′ ∈ φ−m(y0)\{y0} with val (φm, z′) = (+, +) which is impossible as we have just seen. Now assume for a contradiction that y0 /∈ E. Then y0 lies in the interior of one of the intervals from Lemma 5.9, say Ii. We can then write Ii as the union Ii = J1 ∪ J2 of two closed non-degenerate intervals such that J1 ∩ J2 = {y0}. As we have just seen an element of Ii ∩(φ−m(y0)\{y0}) must be critical for φm and it follows therefore that φm(J1) = J1. This contradicts the total transitivity of φpIi. (cid:3) Lemma 5.15. Assume that φ is totally transitive and that deg φ ∈ {−1, 1}. It follows that there is at most a single exposed point, and it must be a fixed point e such that φ−1(e)\C1 = {e}. Proof. Let m be the same number as in the proof of Lemma 5.14. In that proof it was shown that φ−m(y)\{y} ⊆ Cm (5.8) for every exposed point y. It follows that if there are two exposed points, say e1 and e2, we could write T = J1 ∪ J2 where J1 and J2 are non-degenerate closed intervals such that J1 ∩ J2 = {e1, e2} and φ2m(Ji) = Ji, i = 1, 2. This contradicts the assumed total transitivity of φ. Therefore there is at most at single exposed point e, and it is fixed by φm. When deg φ = 1 it follows from Lemma 5.12 that there is an element z ∈ φ−1(e) such that val (φ, z) = (+, +). Then z is exposed (since z ∈ RO+(e)) and the uniqueness of e implies that z = e, proving that e is a fixed point for φ. To reach the same conclusion when deg φ = −1 it suffices to consider the case where val (φ, e) = (−, −). By Lemma 5.12 there are elements z1, z ∈ T such that φ(z1) = z, φ(z) = e and val (φ, z1) = val (φ, z) = (−, −). Then z1 ∈ RO+(e) and hence z1 = e because e is the only exposed point. It follows that z = φ(e), i.e. φ2(e) = e. Note that val (φ, φ(e)) = (−, −). We claim that φ−1 ({e, φ(e)}) \C1 = {e, φ(e)}. (5.9) To show this let x ∈ φ−1(e)\C1. If val(φ, x) = (+, +) we find that x = e since e is the only exposed point. If val (φ, x) = (−, −) an application of Lemma 5.12 shows that x = φ(e). Consider then an element y ∈ φ−1(φ(e))\C1. If val(φ, y) = (−, −) it follows that y ∈ RO+(e) and hence y = e by uniqueness of e. If instead val(φ, y) = (+, +) an application of Lemma 5.12 shows that that y = φ(e). Having established (5.9) note that it implies that φ(e) is exposed, whence equal to e. To show that φ−1(e)\C1 = {e} we may assume that deg φ = 1, since the other case follows from (5.9). Furthermore, it suffices to show that φ−1(e)\C1 ⊆ {e} since exposed points are not critical by Lemma 5.5. Consider therefore an element x ∈ φ−1(e)\C1. If val (φ, x) = (+, +) it follows that x ∈ RO+(e) and hence x is exposed. Since e is the only exposed points this shows that x = e. Assume then that val(φ, x) = (−, −). If x 6= e a look at the graph for a lift of φ shows that there is then also a point y ∈ φ−1(e)\{e} with val(φ, y) = (+, +) which we have just seen is not possible. Hence x = e (cid:3) To formulate the next proposition we call a point e ∈ T an exceptional fixed point when φ−1(e)\C1 = {e}. 16 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN Proposition 5.16. Assume that φ is transitive and that deg φ ∈ {−1, 1}. Then C ∗ 1) φ is not totally transitive, or 2) φ is totally transitive and there is an exceptional fixed point. φ(cid:1) is simple unless either r(cid:0)Γ+ In case 2) there is an extension 0 B C(T) 0 (5.10) where B is simple and purely infinite. When φ is not totally transitive and deg φ = 1 there is an extension 0 B C ∗ C(T) ⊗ Mp(C) 0, (5.11) where p is the global period of φ and B is simple and purely infinite. When φ is not totally transitive and deg φ = −1 there is an extension 0 B C ∗ C(T) ⊕ C(T) 0, (5.12) C ∗ φ(cid:1) r(cid:0)Γ+ φ(cid:1) r(cid:0)Γ+ φ(cid:1) r(cid:0)Γ+ where B is simple and purely infinite. Proof. Assume that none of the two cases 1) or 2) occur. It follows from Lemma 5.15 that there are no exposed points, and from Corollary 5.6 that C ∗ In case 2) it follows from Lemma 5.15 that there is exactly one exposed point, e, φ(cid:1) is simple. r(cid:0)Γ+ which is a fixed point. From Lemma 5.1 we get then the extension C ∗ φ(cid:1) r(cid:0)Γ+ C ∗ r(cid:0)Γ+ φ {e}(cid:1) 0 B 0 (5.13) r(cid:0)Γ+ φ T\{e}(cid:1). It is easy to see, cf. the proof of Lemma 4.11 in [Th3], r(cid:0)Γ+ φ {e}(cid:1) ≃ C(T). Furthermore, B is purely infinite because B is an ideal in φ(cid:1) which is purely infinite by Proposition 4.5. To conclude that B is simple r(cid:0)Γ+ r(cid:0)Γ+ φ T\{e}(cid:1) are pre-periodic. It follows from Theorem 4.1 where B = C ∗ that C ∗ C ∗ we argue as in the proof of Proposition 4.10 in [Th3]: The elements of T\{e} with non-trivial isotropy in C ∗ that the pre-periodic points are countable, whence T\{e} must contain a point with trivial isotropy. By Corollary 2.18 of [Th1] it suffices therefore to show that T\{e} does not contain any non-trivial (relatively) closed RO+-invariant subsets. Let therefore L be such a set. Then L ∪ {e} is closed and RO+-invariant in T and hence either equal to T or contained in {e} by Lemma 5.5 and Lemma 5.15. It follows that L = ∅ or L = T\{e}. This completes the proof in case 2). In case 1) we argue as above, except that we use Lemma 5.14 to replace Lemma (cid:3) 5.15, and Lemma 5.11 to determine C ∗ To show by example that all the cases mentioned in Proposition 5.16 can occur r(cid:0)Γ+ φ E(cid:1). consider the graph 1 1 CIRCLE MAPS AND C ∗-ALGEBRAS 17 The graph describes the lift of an exact, and hence totally transitive map φ of the circle of degree 1 with an exceptional fixed point. The corresponding C ∗-algebra C ∗ consider the graph φ(cid:1) is an extension as in (5.10). To show that also the extensions (5.11) occur r(cid:0)Γ+ 1 1 This is the graph of the lift of a transitive, but not totally transitive circle map φ of degree 1 for which C ∗ way the following graph describes a transitive circle map of degree −1 which is not totally transitive and for which C ∗ φ(cid:1) is an extension as in (5.11) (with p = 4). In the same r(cid:0)Γ+ φ(cid:1) is an extension as in (5.12). r(cid:0)Γ+ 1 1 5.1.3. deg φ = 0. A point z ∈ T will be called an exceptional critical value when φ−1(z) ⊆ C1. Lemma 5.17. Assume that deg φ = 0 and that φ is surjective. There is at most one exceptional critical value, and for all other elements x ∈ T there are points y± ∈ φ−1(x) such that val (φ, y±) = (±, ±). Proof. Look at the graph of a lift of φ. (cid:3) If y ∈ T is an Lemma 5.18. Assume that φ is transitive and that deg φ = 0. exposed point there is an exceptional critical value e ∈ T such that φ2(e) = φ(e) 6= e, RO+(y) = {e, φ(e)} and φ−1 (φ(e)) \C1 = {e, φ(e)}. Proof. The main part of the proof will be to show that there is an exceptional critical value e such that one of the following holds: i) φ2(e) = φ(e) 6= e, val (φ, e) = (−, −) and RO+(y) = {φ(e)}, ii) φ2(e) = φ(e) 6= e, val (φ, φ(e)) = (+, +) and RO+(y) = {e}, iii) φ2(e) = φ(e) 6= e, RO+(y) = {e, φ(e)}. Assume first that RO+(y) does not contain an exceptional critical value. Let z ∈ RO+(y). By using Lemma 5.17 we can construct yk, k = 0, 1, 2, 3, . . . such that y0 = z, φ(yk) = yk−1 and val (φ, yk) = (+, +), k ≥ 1. Then yk ∈ RO+(y) for all k so there are k 6= k′ such that yk = yk′. It follows that z is periodic and that val (φ, u) = (+, +) for all u in the orbit Orb(z) of z. Hence Orb(z) ⊆ RO+(y). Since this conclusion holds for all z ∈ RO+(y) and since the forward orbits of elements from RO+(y) must intersect we conclude that RO+(y) = Orb(y) and 18 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN val(cid:0)φ, φk(y)(cid:1) = (+, +) for all k ∈ N. Let z ∈ RO+(y). Using Lemma 5.17 again we find u1, v1 ∈ φ−1(z) such that val (φ, u1) = (+, +) and val (φ, v1) = (−, −). Then u1 ∈ RO+(y) and u1 is therefore an element of the orbit of y. Since v1 6= u1 (or since val (φ, v1) = (−, −)), it follows that v1 is not in the orbit of y. If v1 is not an exceptional critical value we can find v2 ∈ φ−1(v1) such that val (φ, v2) = (−, −). It follows that v2 ∈ RO+(y) and v2 must therefore be an element of Orb(y). This contradicts that v1 is not, and we conclude that v1 must be an exceptional critical value e, which by Lemma 5.17 is unique. This shows that z = φ(e) and we conclude therefore that case i) occurs. We consider then the case where RO+(y) contains an exceptional critical value e. By looking at the graph of a lift of φ we see that a non-critical exceptional critical value e can not be fixed since the degree is 0. Thus φ(e) 6= e since exposed points are not critical. To see that φ(e) is a fixed point assume that it is not. Consider first the case where φ(e) is periodic, say of period p > 1. Since φ(e) 6= e it follows from Lemma 5.17 that there is a point b1 ∈ φ−1(φ(e)) such that val (φ, b1) 6= val (φ, e). Then b1 /∈ {e, φ(e)} and we use Lemma 5.17 again to find b2 ∈ φ−1(b1) such that val (φ, b2) 6= val (φ, φ(e)). It follows that b2 /∈ {e, φ(e), b1}. By requiring in each step that val (φ, bi) 6= val (φ, φ(e)) we obtain through repeated application of Lemma 5.17 elements bi, i = 1, 2, . . . , p + 1, such that φ (bk+1) = bk and bk+1 /∈ {e, φ(e), b1, b2, . . . , bk} for all k = 1, 2, . . . , p. Then, for j > p + 1 we require in each step instead that val (φj, bj) = val(φ, e). It is then still automatic that bk+1 /∈ {e, b1, b2, . . . , bk} for all k, while the fact that bj 6= φ(e) follows for j ≥ p + 1 because j is larger than the period of φ(e). Since bj ∈ RO+(e) = RO+(y) when j > p + 1, we have contradicted the assume finiteness of RO+(y). To get the same contradiction when φ(e) is not assumed to be periodic we proceed in the same way, except that the steps between b1 and bp+1 can be bypassed. In any case we conclude that φ2(e) = φ(e). We next argue, in a similar way, that φ−1 (φ(e)) \C1 ⊆ {e, φ(e)}. Indeed, if b1 ∈ φ−1 (φ(e)) \ ({e, φ(e)} ∪ C1) we use Lemma 5.17 to get a sequence bi such that φ (bi+1) = bi, i ≥ 1, and val (φ, bi) = (−, −), i ≥ 2. Then i 6= i′ ⇒ bi 6= bi′, and bi ∈ RO+(e) for infinitely many i; again contradicting the infiniteness of RO+(e). Since e is not pre-critical by Lemma 5.5 we have shown that φ−1 (φ(e)) \C1 = {e, φ(e)}. If val (φ, e) = (−, −) and val (φ, φ(e)) = (+, +) we find now easily that RO+(y) = RO+(e) = {e}, which is case ii), and in all other cases that RO+(y) = RO+(e) = {e, φ(e)}, which is case iii). Finally we argue that the cases i) and ii) are impossible. Indeed, in both cases we must have that val(φ, e) = (−, −) and val(φ, φ(e)) = (+, +) since otherwise e ∈ RO+(φ(e)). But then the two closed intervals J1 and J2 defined such that J1 ∩ J2 = {e, φ(e)} and J1 ∪ J2 = T are both φ-invariant, which contradicts the transitivity of φ. It follows that only case iii) can occur. (cid:3) Lemma 5.19. Assume that φ is transitive and deg φ = 0. Then there are exposed points if and only if φ is not totally transitive. Proof. If φ is not totally transitive there are exposed points by (the proof of) Lemma 5.14. Conversely, if there are exposed points it follows from Lemma 5.18 that there is an exceptional critical value e such that e 6= φ(e) = φ2(e) and {e, φ(e)} is the set of exposed points. Furthermore, φ−1(φ(e))\C1 = {e, φ(e)}. The points e and φ(e) define closed intervals J1 and J2 such that T = J1 ∪ J2, J1 ∩ J2 = {e, φ(e)} and CIRCLE MAPS AND C ∗-ALGEBRAS 19 φ(Ji) = Ji, i = 1, 2, or φ(J1) = J2 and φ(J2) = J1. The first case is ruled out by transitivity, and the second implies that φ is not totally transitive. (cid:3) Proposition 5.20. Assume that φ is transitive and that deg φ = 0. Then C ∗ r (Γφ) is simple if and only if φ is totally transitive. When φ is not totally transitive there is an extension 0 B C(T) ⊗ M2(C) 0 (5.14) where B is simple and purely infinite. C ∗ φ(cid:1) r(cid:0)Γ+ Proof. With Lemma 5.19 and Lemma 5.18 at hand all the necessary arguments can be found in the proof of Proposition 5.16. (cid:3) The following graph describes a transitive circle map of degree 0 which is not totally transitive and for which C ∗ φ(cid:1) is an extension as in (5.14). r(cid:0)Γ+ 1 1 2 0 1 2 5.1.4. Simplicity. We can now finally give a necessary and sufficient condition for simplicity of C ∗ φ(cid:1). r(cid:0)Γ+ φ(cid:1) is simple. r(cid:0)Γ+ Theorem 5.21. The following conditions are equivalent. i) C ∗ ii) φ is totally transitive and there is no exceptional fixed point. iii) φ is exact and there is no exceptional fixed point. Proof. i) ⇒ ii) : Transitivity of φ follows from Lemma 5.4. And then φ must be totally transitive since otherwise the set E considered in Lemma 5.10 will be non- empty, finite and RO+-invariant, as shown in the first paragraph of the proof of Lemma 5.14, and this contradicts simplicity by Lemma 5.3. The absence of an exceptional fixed point follows also from Lemma 5.3 since an exceptional fixed point is its own RO+-orbit. The implication ii) ⇒ i) follows from Propositions 5.20, 5.16 and 5.8, and the implication iii) ⇒ ii) is trivial. The implication ii) ⇒ iii) follows from Lemma 4.6. (cid:3) 6. Nuclearity, UCT and a six-terms exact sequence 6.1. The Cuntz-Pimsner picture of C ∗ φ = Γ+ φ(cid:1). To simplify notation, set r(cid:0)Γ+ φ(cid:1) is the fixed point algebra of the gauge action β = βc on C ∗ φ(cid:1) r(cid:0)Γ+ r(cid:0)R+ Then C ∗ arising from the homomorphism c : Γ+ [Re]. For n ∈ N, set φ → Z defined such that c(x, k, y) = k, cf. φ (0). R+ R+ φ (n) = Γ+ φ (0, n) 20 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN which is an open sub-groupoid of R+ φ . Then R+ φ (n) and Lemma 6.1. Assume that C ∗ are elements z, z′ ∈ T\C1 such that (x, 1, z), (z′, 1, x) ∈ Γ+ φ . C ∗ C ∗ φ(cid:1) =[n r(cid:0)R+ φ(cid:1) is simple. Let x ∈ T\C1. It follows that there r(cid:0)Γ+ (6.1) φ =Sn R+ r(cid:0)R+ φ (n)(cid:1). Proof. If φk(x) ∈ C1 for some k ≥ 1, set z = φ(x) and let z′ be any element of φ−1(x). Then (x, 1, z), (z′, 1, x) ∈ Γ+ φ . Assume therefore now that φn(x) /∈ C1 for all n ≥ 1. Since φ is not locally injective there are open non-empty intervals I± and I such that val(φ, z) = (+, +) for all z ∈ I+, val(φ, z) = (−, −) for all z ∈ I− and φ(I−) = φ(I+) = I. It follows from Theorem 5.21 that φ is exact and there is therefore an N ∈ N such that φN −1(I) = T. impossible since C ∗ k=0 φ−k(O)) 6= ∅ for all j ≥ n0. Since z ∈ O there are k, l ∈ N such that φl(z) = φkpN +d(x). Note that We consider first the case where x is pre-periodic to a finite orbit O of period k=0 φ−k(O) j=0 φ−j(O) which is a finite set. This is φ(cid:1) is simple, cf. Corollary 5.6. Since O is finite it follows that r(cid:0)Γ+ φ−k(O))! = ∅. p. Let d ∈ {−1, 1}. If there is an M ∈ N such that φ−j(O) ⊆ Cj ∪Sj−1 for all j > M, it follows that RO+(x) ⊆SM there is a z ∈ O and an n0 ∈ N such that φ−j(z)\(Cj ∪Sj−1 j 6= j′ ⇒ φ−j(z)\(Cj ∪ j ′−1[k=0 φN (CN ) ∩(cid:16)φ−k′pN +N +l(z)\Ck′pN −N −l(cid:17) = ∅. Since φN (CN ) is a finite set there is therefore a k′ ≥ k such that k′pN − N − l ≥ n0 and φ−k(O))! ∩ φ−j ′ Choose an element a ∈ φ−k′pN +N +l(z)\Ck′pN −N −l. Then a = φN (y±) where y± ∈ I+ ∪ I− and val(φN , y+) = (+, +), val(φN , y−) = (−, −). It follows that φk′pN (y±) = φkpN +d(x) = φk′pN +d(x) and that either (x, d, y+) ∈ Γ+ φ . When d = 1 this gives us z such that (x, 1, z) ∈ Γ+ φ and when d = −1 it gives us z′ such that (x, −1, z′) ∈ Γ+ φ . φ or (x, d, y−) ∈ Γ+ j−1[k=0 (z)\(Cj ′ ∪ Consider then the case where x is not pre-periodic. If there are infinitely many n ∈ N, n > N, such that the only elements of φ−N (φn(x)) \(cid:8)φn−N (x)(cid:9) are critical points for φN , it follows that φn(x) ∈ φN (CN ) for infinitely many n. Since φN (CN ) is a finite set this contradicts that x is not pre-periodic. It follows that when x is not pre-periodic there is an n0 ∈ N such that for all n ≥ n0 there is an element yn ∈ φ−N (φn(x)) which is not in the forward orbit {φj(x) : j ∈ N} of x and also not critical for φN . Note that when j, j′ ≥ n0. Since φN (CN ) is a finite set there is therefore an m > 2 such that mN + d > n0 and φ−kN (yj ′)! = ∅ j 6= j′ ⇒ [k∈N [k∈N φ−kN (yj)! ∩ [k∈N φ−kN (ymN +d)! ∩ φN (CN ) = ∅. CIRCLE MAPS AND C ∗-ALGEBRAS 21 Let a ∈ φ−(m−2)N (ymN +d). Choose y± ∈ I± such that φN (y±) = a. Then φmN (y±) = φmN +d(x) and for one of the elements v in {y+, y−} it holds that val(cid:0)φmN , v(cid:1) = val(cid:0)φmN +d, x(cid:1). Then (x, 1, v) ∈ Γ+ φ(cid:1) is simple. Let x ∈ C1 and let U be an open r(cid:0)Γ+ Lemma 6.2. Assume that C ∗ non-empty subset of T. There are elements µ1, µ2, · · · , µN ∈ Γ+ µ1µ2µ3 · · · µN is defined, s(γ) = x and r(γ) ∈ U\C1. φ when d = 1 and (x, −1, v) ∈ Γ+ φ when d = −1. (cid:3) φ (1) such that γ = Proof. Since φ is exact the backward orbitS∞ N ∈ N and a z ∈ φ−N (x) ∩ U\C1. Set µi = (φi−1(z), 1, φi(z)). j=1 φ−j(x) is dense in N. so there is an (cid:3) 1 contains an approximate unit for C ∗ 1 spans a dense subspace in the fixed point algebra C ∗ Lemma 6.3. Assume that C ∗ the gauge action, i.e. the set φ(cid:1) is simple. Then the first spectral subspace for r(cid:0)Γ+ φ(cid:1) : βλ(a) = λa ∀λ ∈ T(cid:9) , E1 =(cid:8)a ∈ C ∗ r(cid:0)Γ+ φ(cid:1) as a C ∗-algebra. r(cid:0)Γ+ φ(cid:1) restricts to an action on C ∗ r(cid:0)Γ+ r(cid:0)Γ+ φ T\C1(cid:1). Let r(cid:0)Γ+ φ T\C1(cid:1) be the first spectral subspace of the restricted action. We r(cid:0)Γ+ φ T\C1(cid:1)β. To r(cid:0)Γ+ φ T\C1(cid:1)β φ T\C1(cid:1). r(cid:0)Γ+ Proof. The gauge action β on C ∗ V1 = E1 ∩ C ∗ claim that V1V ∗ show this observe that since the closed span of V1V ∗ suffices to show that the span of V1V ∗ Hence it suffices to show that Cc (T\C1) ⊆ Span V1V ∗ 1 . Let f ∈ Cc (T\C1). Let x ∈ supp f . It follows from Lemma 6.1 that there is a bisection U ⊆ Γ+ φ T\C1 such that x ∈ r(U). There is therefore a function g ∈ Cc(U) such that gg∗ ∈ Cc(T\C1) and gg∗(x) = 1. In this way we get a finite collection gi, i = 1, 2, . . . , N, i (y) > 0 for i=1 gig∗ i = f . in Cc(cid:0)Γ+ all y ∈ supp f . There is then a function h ∈ Cc (T\C1) such that f hPN r(cid:0)Γ+ φ T\C1(cid:1)β φ T\C1(cid:1) and it follows that r(cid:0)Γ+ φ T\C1(cid:1) is contained in the C ∗-subalgebra r(cid:0)Γ+ φ T\C1(cid:1) is generated by V1, whence C ∗ r(cid:0)Γ+ φ(cid:1) generated by E1. r(cid:0)Γ+ φ (1)(cid:1) such that g∗ ∗-algebra generated by Cc(cid:0)Γ+ φ(cid:1). Write f ∈ Cc(cid:0)Γ+ xgxf +Xx∈C1 f = f −Xx∈C1 where h = Px∈C1 Note that it follows from Lemma 6.2 that there are elements gx, x ∈ C1, in the xgx ∈ C(T), xgx(x) = 1 and gx is supported in r−1(T\C1) for all x. Consider then an element g∗ g∗ xgxf and h are both supported in r−1 (T\C1). To conclude that f is contained in the C ∗-algebra generated by E1 we may therefore assume that f is supported in r−1 (T\C1). Under this assumption we write Thus the restriction of the gauge action is full on C ∗ C ∗ of C ∗ i ∈ Cc (T\C1) for all i andPN gxf . Note that f −Px∈C1 Since f hgi and gi are elements of V1 for all i this shows that f ∈ Span V1V ∗ 1 . φ T\C1(cid:1) such that gig∗ g∗ xgxf = f −Xx∈C1 g∗ xgxf +Xx∈C1 A similar argument shows that also V ∗ 1 V1 spans a dense subspace of C ∗ x′gx = 0 when x 6= x′, g∗ 1 is an ideal in C ∗ it φ (1) ∩ Γ+ i=1 gig∗ g∗ g∗ xh, . generates C ∗ φ (1) ∩ Γ+ f = f −Xx∈C1 f g∗ xgx +Xx∈C1 f g∗ xgx, (6.2) 22 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN and note that f −Px∈C1 therefore from the first part of the proof that both are elements of the C ∗-algebra generated by E1. Since f g∗ x are elements of Cc(cid:0)Γ+ φ T\C1(cid:1). It follows f g∗ (cid:3) 1 E1 has dense span xgx andPx∈C1 Xx∈C1 f g∗ it follows from (6.2) that f is in the C ∗-algebra generated by E1. Lemma 6.4. Assume that C ∗ in C ∗ x! gx′ xgx = Xx′∈C1 Xx∈C1 r(cid:0)Γ+ φ(cid:1) is simple. It follows that E∗ f g∗ φ(cid:1). r(cid:0)R+ i=1 g∗ i=1 g∗ Let E−1 = E∗ φ (1) = ∅. Let ω be the state on C ∗ It should be observed that the gauge action is generally not full. Even when 1 E1 is an ideal in C ∗ φ(cid:1) it suffices to show that r(cid:0)R+ Proof. Since the closed span of E∗ E∗ 1E1 contains 1 ∈ C(T). Let f ∈ C (T). Let x ∈ T. Provided x /∈ C1 it follows from Lemma 6.1 that there is an element γ ∈ Γ+ φ (1) such that s(γ) = x. When x ∈ C1 set γ = (z, 1, x), where z ∈ φ−1(x). Then s(γ) = x. Hence, regardless of which x we consider there is a bisection U ⊆ Γ+ φ (1) such that x ∈ s(U). It follows that there is a function g ∈ Cc(U) such that g∗g ∈ C(T) and g∗g(x) = 1. In this way we get a i gi ∈ C (T) for all i and i gi = 1. Since (cid:3) gih∗ and gi are elements of E1 for all i, this completes the proof. annihilates the ideal in C ∗ action is responsible for some intriguing features of the KMS states. correspondence and we can consider the associated Cuntz-Pimsner C ∗-algebra OE−1, cf. [Ka]. We can now show that this is another version of C ∗ algebra is simple. C ∗ 1 is a proper ideal. Assume, for example, that φ has two critical points whose forward orbits do not intersect, and that one of them, say c, has infinite forward orbit. Then r−1(c) ∩ Γ+ finite collection gi, i = 1, 2, . . . , N, in Cc(cid:0)Γ+ φ (1)(cid:1) such that g∗ PN i gi > 0. There is then a function h ∈ C (T) such that hPN φ(cid:1) is simple it can easily happen that the ideal in C ∗ φ(cid:1) generated by E1E∗ r(cid:0)R+ r(cid:0)Γ+ φ(cid:1) such that ω(f ) = f (c, 0, c) r(cid:0)R+ φ(cid:1). Then ω(f g∗) = 0 for all f, g ∈ Cc(cid:0)Γ+ φ (1)(cid:1) and it follows that ω when f ∈ Cc(cid:0)R+ φ(cid:1) generated by E1E∗ r(cid:0)R+ φ(cid:1)- r(cid:0)R+ φ(cid:1) when the latter r(cid:0)Γ+ φ(cid:1) ≃ OE−1 and r(cid:0)Γ+ φ(cid:1) is simple. It follows that C ∗ r(cid:0)Γ+ φ(cid:1)(cid:1) φ(cid:1)(cid:1) r(cid:0)Γ+ K0(cid:0)C ∗ φ(cid:1)(cid:1) id −[E−1]0 K0(cid:0)C ∗ r(cid:0)R+ K0(cid:0)C ∗ r(cid:0)R+ φ(cid:1)(cid:1) φ(cid:1)(cid:1) φ(cid:1)(cid:1) , K1(cid:0)C ∗ r(cid:0)Γ+ K1(cid:0)C ∗ r(cid:0)R+ K1(cid:0)C ∗ r(cid:0)R+ φ(cid:1) is the inclusion map and r(cid:0)Γ+ φ(cid:1) → C ∗ r(cid:0)R+ φ(cid:1)(cid:1) φ(cid:1) , C ∗ r(cid:0)R+ r(cid:0)R+ [E−1] ∈ KK(cid:0)C ∗ φ(cid:1) define a representa- φ(cid:1) and E−1 ⊆ C ∗ r(cid:0)Γ+ φ(cid:1) ⊆ C ∗ r(cid:0)Γ+ r(cid:0)R+ φ(cid:1) as defined by Katsura in Definition 2.1 r(cid:0)Γ+ is the KK-theory element represented by E−1. Proof. The inclusions C ∗ tion of the C ∗-correspondence E−1 in C ∗ 1 be the first negative spectral subspace. Then E−1 is a C ∗ Theorem 6.5. Assume that C ∗ there is an exact sequence 1. This asymmetry in the gauge where ι : C ∗ ι1 id −[E−1]1 ι0 (6.3) CIRCLE MAPS AND C ∗-ALGEBRAS 23 (6.4) of [Ka]. Observe that it follows from Lemma 6.4 that the representation is injective in the sense of Katsura, and that φ(cid:1)(cid:1) = K(E−1), r(cid:0)R+ in the notation from [Ka]. It follows now from Proposition 3.3 in [Ka] that our representation of E−1 is covariant in the sense of Definition 3.4 of [Ka]. Then the isomorphism OE−1 ≃ C ∗ [Ka]. Thanks to (6.4) we get now the stated six-terms exact sequence from Theorem 8.6 in [Ka]. ψt(cid:0)C ∗ φ(cid:1) follows from Lemma 6.3 above and Theorem 6.4 in r(cid:0)Γ+ r(cid:0)R+ φ (k)(cid:1) and some consequences. Fix a natural num- 6.2. The structure of C ∗ ber k and let D be a finite subset of T such that Ck ⊆ φ−k(D). Let c1 < c2 < · · · < cN be a numbering of the elements in D and set c0 = cN . Let Ii = ]ci−1, ci[ , i = :]0, 1[→ ]ci−1, ci[ such that 1, 2, . . . , N. For each i we fix a homeomorphism ψIi limt→0 ψIi(t) = ci−1 and limt→1 ψIi(t) = ci. Let Ik be the set of connected compo- nents of T\φ−k(D). Since Ck ⊆ φ−k(D), the map x 7→ val(cid:0)φk, x(cid:1) is constant on each I ∈ Ik and we set val(cid:0)φk, I(cid:1) = val(cid:0)φk, x(cid:1) , x ∈ I. Set k =(cid:8)(I, J) ∈ Ik × Ik : φk(I) = φk(J), val(cid:0)φk, I(cid:1) = val(cid:0)φk, J(cid:1)(cid:9) . Let Bk denote the finite-dimensional C ∗-algebra generated by the matrix units eI,J, where (I, J) ∈ I (2) k . Similarly, we let Ak be the finite-dimensional C ∗-algebra gener- ated by the matrix units ex,y where x, y ∈ φ−k(D), φk(x) = φk(y) and val(cid:0)φk, x(cid:1) = val(cid:0)φk, y(cid:1). When x ∈ φ−k(D) and I ∈ Ik, write I > x when x ∈ I and y > x for all y ∈ I, and I < x when x ∈ I and y < x for all y ∈ I. Define a ∗-homomorphism Ik : Ak → Bk such that I (2) (cid:3) where we sum over the set of pairs (I, J) ∈ I (2) and val(cid:0)φk, I(cid:1) = (+, +), or I < x, J < y and val(cid:0)φk, I(cid:1) = (−, −). Similarly, we define a ∗-homomorphism Uk : Ak → Bk such that k with the properties that x < I, y < J Ik (ex,y) =XI,J eI,J Uk (ex,y) =XI,J eI,J where we sum over the set of pairs (I, J) ∈ I (2) and val(cid:0)φk, I(cid:1) = (+, +), or x < I, y < J and val(cid:0)φk, I(cid:1) = (−, −). Let (I, J) ∈ I (2) such that φk(I) = φk(J) = Ii. Let λI : Ii → I be the inverse of φk : I → Ii, and similarly λJ : Ii → J the inverse of φk : J → Ii. Then k k with the properties that I < x, J < y for all t ∈]0, 1[. Notice that the limits (λI ◦ ψIi(t), 0, λJ ◦ ψIi(t)) ∈ R+ φ (k) λI(ci) = lim x→ci λI(x) and λI (ci−1) = lim x→ci−1 λI(x) both exist. Let f ∈ Cc(cid:0)R+ φ (k)(cid:1). Then the function ]0, 1[∋ t 7→ f (λI ◦ ψIi(t), 0, λJ ◦ ψIi(t)) 24 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN has a unique continuous extension fI,J : [0, 1] → C. This is because lim t→1 f (λI ◦ ψIi(t), 0, λJ ◦ ψIi(t)) = 0 φ (k) and lim t→0 φ (k), and φ (k). fI,J eI,J. f (x, 0, y)ex,y, lim t→1 lim t→0 For x ∈ T, let πx be the ∗-representation used to define the norm on C ∗ f (λI ◦ ψIi(t), 0, λJ ◦ ψIi(t)) = f(cid:0)λI(ci), 0, λJ(ci)(cid:1) φ (k). Similarly, f (λI ◦ ψIi(t), 0, λJ ◦ ψIi(t)) = 0 f (λI ◦ ψIi(t), 0, λJ ◦ ψIi(t)) = f(cid:0)λI(ci−1), 0, λJ(ci−1)(cid:1) φ (k)(cid:1) → C ([0, 1], Bk) such that φ (k)(cid:1) → Ak such that when(cid:0)λI(ci), 0, λJ(ci)(cid:1) /∈ R+ when(cid:0)λI(ci), 0, λJ(ci)(cid:1) ∈ R+ when(cid:0)λI(ci−1), 0, λJ(ci−1)(cid:1) /∈ R+ when(cid:0)λI(ci−1), 0, λJ(ci−1)(cid:1) ∈ R+ We can then define a ∗-homomorphism b : Cc(cid:0)R+ b(f ) =XI,J We can also define a ∗-homomorphism a : Cc(cid:0)R+ a(f ) = X(x,y)∈Ak φ (k)(cid:9). By construction Ik (a(f )) = b(f )(0) where Ak =(cid:8)(x, y) ∈ T2 : (x, 0, y) ∈ R+ r(cid:0)R+ φ (k)(cid:1), fI,J(t)eI,J(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) kπx(f )k =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) X(I,J)∈B : φk(I) = φk(J) = Iio. When x ∈ φ−k(D), we find that f (z, 0, y)ez,y(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) kπx(f )k =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) X(z,y)∈A where A =(cid:8)(z, y) ∈ T2 : (z, 0, y) ∈ R+ φ (k), φk(z) = φk(x), val(φk, z) = val(φk, x)(cid:9). r(cid:0)R+ φ (k)(cid:1) → {(a, b) ∈ Ak ⊕ C ([0, 1], Bk) : Ik(a) = b(0), Uk(a) = b(1)} . the properties that Ik(a) = b(0), Uk(a) = b(1). Then a =P(x,y)∈A λx,yex,y, where φ (k)(cid:1) such and λx,y ∈ C. Since A is a finite set there is a function g ∈ Cc(cid:0)R+ A =(cid:8)(x, y) ∈ φ−k(D) × φ−k(D) : (x, 0, y) ∈ R+ φ (k)(cid:9) By combining (6.5) and (6.6) we find that f → (a(f ), b(f )) is isometric and extends to an injective ∗-homomorphism cf. (2.1). When x /∈ φ−k(D) there is an i and a t such that φk(x) = ψIi(t). Then that g(x, 0, y) = λx,y for all (x, y) ∈ A. Then (a, b) = µk(g) + (0, b′), where b′ ∈ where B =n(I, J) ∈ I (2) k Proof. It remains to show that µk is surjective. Let (a, b) ∈ Ak ⊕ C ([0, 1], Bk) have µk : C ∗ Lemma 6.6. µk is an isomorphism. and Uk(a(f )) = b(f )(1). , , (6.5) (6.6) CIRCLE MAPS AND C ∗-ALGEBRAS 25 C0(]0, 1[) ⊗ Bk. Write b′ =P(I,J)∈I 2 k in the last sum there is an interval Ii such that φk(I) = φk(J) = Ii. Then b′ I,J eI,J , where b′ I,J ∈ C0(]0, 1[). For each (I, J) U = {(λI ◦ ψIi(t), 0, λJ ◦ ψIi(t)) : t ∈]0, 1[} is an open subset of R+ φ (k) and we can define a function hI,J on U such that hI,J (λI ◦ ψIi(t), 0, λJ ◦ ψIi(t)) = b′ I,J (t). In combination with (6.1) it follows that Then hI,J ∈ C0(U) and we can choose a sequence {hn} ⊆ Cc(U) such that limn→∞ hn = hI,J, uniformly on R+ φ it follows that limn→∞ µk(hn) = I,J eI,J is in the range of µk for (cid:3) φ (k). Since U is a bi-section in R+ all (I, J). It follows that (a, b) is in the range of µk. (cid:0)0, b′ I,J eI,J(cid:1), cf. Lemma 2.4 in [Th1], proving that b′ φ(cid:1) is an ASH-algebra with no dimension growth, cf. [T]. r(cid:0)R+ It follows from Lemma 6.6 that C ∗ algebra in the sense of N.C. Phillips. C ∗ Corollary 6.7. The C ∗-algebra C ∗ ficient theorem (UCT) of Rosenberg and Schochet, [RS]. r(cid:0)R+ φ (k)(cid:1) is a recursive sub-homogeneous C ∗- φ(cid:1) is nuclear and satisfies the universal coef- r(cid:0)Γ+ φ(cid:1) is the inductive limit of a sequence of sub-homogeneous C ∗- r(cid:0)R+ φ(cid:1) is simple. To obtain the same conclusion in r(cid:0)Γ+ φ(cid:1) implies that R+ r(cid:0)R+ φ(cid:1) is the fixed- r(cid:0)R+ φ(cid:1) is nuclear, and the work of Tu in [Tu] r(cid:0)Γ+ Proof. Since C ∗ algebras, it is nuclear and satisfies the UCT. The corollary follows therefore from Theorem 6.5 and [Ka] when C ∗ general we must use a different and even more indirect path. First observe that the nuclearity of C ∗ φ is (topologically) amenable by Corollary 6.2.14 (ii) and Theorem 3.3.7 in [A-DR], and then from Proposition 6.1.8 in [A-DR] that C ∗ point algebra of the gauge action it follows then from a recent result of Spielberg, Proposition 9.3 in [Sp], that Γ+ φ is also (topologically) amenable. Then Corollary 6.2.14 (i) from [A-DR] shows that C ∗ (more precisely, Lemme 3.5 and Proposition 10.7 in [Tu]) implies that it satisfies the UCT. (cid:3) φ(cid:1) equals the full groupoid C ∗-algebra of R+ r(cid:0)R+ φ . Since C ∗ We do not consider the full groupoid C ∗-algebra here, [Re], but observe in passing φ ; a fact which φ are canonically isomorphic by that the previous proof established the topological amenability of Γ+ implies that the full and reduced C ∗-algebras of Γ+ Proposition 6.1.8 in [A-DR]. 7. Markov maps We say that φ is Markov when φ(C1) ⊆ C1. Lemma 7.1. Assume that φ is Markov. Then C ∗ transitive. Proof. The Markov condition implies that all post-critical points are critical and it follows therefore from Lemma 5.5 that a transitive Markov map has no exposed points and therefore also that C ∗ φ(cid:1) is simple if and only if φ is r(cid:0)Γ+ φ(cid:1) is simple when φ is transitive and Markov. (cid:3) r(cid:0)Γ+ Lemma 7.2. Assume that φ is Markov and transitive. There is a natural number k ∈ N with the property that when j ≥ k and x ∈ φ (T\C1), there are elements y± ∈ T such that φj (y±) = x, val (φj, y+) = (+, +) and val (φj, y−) = (−, −). 26 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN Proof. Since φ is not locally injective there are non-empty open intervals I± such that val (φ, x) = (+, +) for all x ∈ I+, val (φ, y) = (−, −) for all y ∈ I− and I = φ(I+) = φ(I−) is an open non-empty interval. Since φ is exact (by Lemma 7.1 and Theorem 5.21) there is an N ∈ N such that φN (I) = T. Set k = N + 2 and let j ≥ k. Consider an element x ∈ φ (T\C1). Then x = φ(u) for some u ∈ T\C1. There is an element z ∈ I such that φj−2(z) = u and elements z± ∈ I± such that z± ∈ I± and φ(z±) = z. Then φj(z±) = x and the Markov condition implies that z± are not critical for φj−1 since u /∈ C1. Note that val (φj, z±) = val (φj−1, z) • val (φ, z±). If val (φj−1, z) = (+, +), set y+ = z+ and y− = z−, and if val (φj−1, z) = (−, −), set y+ = z− and y− = z+. (cid:3) In the following we say that a Markov map φ is of order k when the conclusion of Lemma 7.2 holds; i.e. when a) for all j ≥ k and all x ∈ φ(T\C1) there are elements y± ∈ T such that φj (y±) = x, val (φj, y+) = (+, +) and val (φj, y−) = (−, −). Besides our standing assumptions (that φ : T → T is continuous, piecewise mono- tone and not locally injective), we now also assume that φ is a Markov map of order k. Lemma 7.3. Assume that (x, l, y) ∈ Γ+ k + 1. It follows that there is an element z ∈ T such that (x, 1, z) ∈ Γ+ (z, l − 1, y) ∈ Γ+ φ (l, n), where l ∈ Z, n ∈ N and l + n ≥ φ (1, k) and φ (l − 1, n). Proof. If φk(x) ∈ C1, set z = φ(x). It follows from Lemma 3.1 and the composition table for • that val(cid:0)φk+1, x(cid:1) = val(cid:0)φ, φk(x)(cid:1) • val(cid:0)φk, x(cid:1) = val(cid:0)φ, φk(x)(cid:1) • val(cid:0)φk−1, φ(x)(cid:1) = val(cid:0)φk, φ(x)(cid:1) , showing that (x, 1, z) ∈ Γ+ φ (1, k). Now note that it follows from Lemma 3.1 and the composition table for • first that φk(x) is critical for φn+l−k and then, as above, that val (φn, y) = val(cid:0)φn+l, x(cid:1) = val(cid:0)φn+l−k, φk(x)(cid:1) • val(cid:0)φk, x(cid:1) = val(cid:0)φn+l−k, φk(x)(cid:1) • val(cid:0)φk−1, z(cid:1) = val(cid:0)φn+l−1, z(cid:1) . φ (l − 1, n). This shows that (z, l − 1, y) ∈ Γ+ (z, l − 1, y) ∈ Γ+ φ (l − 1, n) since If instead φk(x) /∈ C1 it follows from condition a) that there is a z ∈ T such φ (1, k) and that φk+1(x) = φk(z) and val(cid:0)φk, z(cid:1) = val(cid:0)φk+1, x(cid:1). Then (x, 1, z) ∈ Γ+ val(cid:0)φl−1+n, z(cid:1) = val(cid:0)φl+n−k−1, φk(z)(cid:1) • val(cid:0)φk, z(cid:1) = val(cid:0)φl+n−k−1, φk+1(x)(cid:1) • val(cid:0)φk+1, x(cid:1) = val(cid:0)φl+n, x(cid:1) = val (φn, y) when l + n ≥ k + 2, while val(cid:0)φl−1+n, z(cid:1) = val(cid:0)φk, z(cid:1) = val(cid:0)φk+1, x(cid:1) = val(cid:0)φl+n, x(cid:1) = val (φn, y) when l + n = k + 1. (cid:3) CIRCLE MAPS AND C ∗-ALGEBRAS 27 Lemma 7.4. Assume that n ≥ k + 1. Let (x, 0, y) ∈ Γ+ z1, z2 ∈ T such that (z1, 0, z2) ∈ Γ+ φ (0, n − 1), (x, 1, z1), (y, 1, z2) ∈ Γ+ φ (0, n). There are elements φ (1, k) and in Γ+ φ . (x, 0, y) = (x, 1, z1)(z1, 0, z2)(z2, −1, y) Proof. We obtain from Lemma 7.3 an element z1 ∈ T such that (x, 1, z1) ∈ Γ+ φ (1, k), (z1, −1, y) ∈ Γ+ φ (−1, n) and (x, 0, y) = (x, 1, z1)(z1, −1, y). Then (y, 1, z1) ∈ Γ+ φ (1, n− 1) and a second application of Lemma 7.3 gives a z2 ∈ T such that (y, 1, z2) ∈ Γ+ φ (1, k), (z2, 0, z1) ∈ Γ+ (cid:3) φ (0, n − 1) and (y, 1, z1) = (y, 1, z2)(z2, 0, z1). Let E be the closure of Cc(cid:0)Γ+ r(cid:0)R+ φ (k)(cid:1) , C ∗ E∗E ⊆ C ∗ φ(cid:1). Then r(cid:0)Γ+ φ (1, k)(cid:1) in C ∗ r(cid:0)R+ φ (k + 1)(cid:1) E ⊆ E, EC ∗ r(cid:0)R+ φ (k)(cid:1) ⊆ E. r(cid:0)R+ φ (k)(cid:1) φ(cid:1). r(cid:0)Γ+ We can therefore, in the natural way, consider E as a C ∗-correspondence on C ∗ in the sense of Katsura, [Ka]. Let OE denote the corresponding C ∗-algebra, cf. Def- inition 3.5 in [Ka]. We aim now to show that OE is isomorphic to C ∗ Lemma 7.5. 2) C ∗ 3) C ∗ 1) Span EE∗ = C ∗ r(cid:0)R+ φ (n)(cid:1) ⊆ EC ∗ r(cid:0)R+ φ (n − 1)(cid:1) E∗ when n ≥ k + 1. φ(cid:1) is generated by E. r(cid:0)Γ+ r(cid:0)R+ φ (k + 1)(cid:1). r(cid:0)R+ φ (k + 1)(cid:1). Let h ∈ Cc(cid:0)R+ Proof. 1) It is clear that EE∗ ⊆ C ∗ show that h is contained in Span EE∗ we may assume that h is supported in a bi-section U. Set K = r (supp h) ⊆ T. Let x ∈ K. There is then a unique y ∈ T It follows from Lemma 7.4 that there is a z ∈ T such such that (x, 0, y) ∈ U. that (x, 1, z), (y, 1, z) ∈ Γ+ φ (1, k). Choose open bi-sections W1 ⊆ Γ+ φ (1, k), W2 ⊆ Γ+ φ (−1, k + 1) containing (x, 1, z) and (z, −1, y), respectively, such that W1W2 ⊆ U. We can then define functions f ∈ Cc(W1), g ∈ Cc(W2) such that f (γ) = h(γ′) in a neighborhood of (x, 1, z), where γ′ ∈ U is determined by the condition that r(γ) = r(γ′), and g(γ) = 1 in a neighborhood of (z, −1, y). There is then an open neighborhood Vx of x such that f g(γ) = h(γ) for all γ ∈ r−1(Vx). By compactness of φ (k + 1)(cid:1). To i ∈ E this shows that h is in the span of EE∗. φ (l)(cid:1) ⊆ Span ECc(cid:0)Γ+ 2) follows from Lemma 7.4 in a similar way. 3) By arguments similar to those used for 1) it follows from Lemma 7.3 that K we get in this way a finite collection of functions ψi ∈ C(T), fi, gi ∈ Cc(cid:0)Γ+ φ (1, k)(cid:1) such that h =Pi ψifigi. Since ψifi, g∗ Cc(cid:0)Γ+ = Cc(cid:0)Γ+ φ (−l)(cid:1) φ(cid:1) is φ(cid:1) it suffices then to show that C ∗ r(cid:0)R+ and sinceSl∈Z Cc(cid:0)Γ+ r(cid:0)Γ+ φ (l)(cid:1) is dense in C ∗ r(cid:0)R+ φ (k)(cid:1)- r(cid:0)R+ φ (k + 1)(cid:1) → L(E) such that π(a)b = ab. Then π is φ (l − 1)(cid:1) for all l ∈ Z. Since Cc(cid:0)Γ+ contained in the C ∗-algebra generated by E. This follows from 1), 2) and (6.1). (cid:3) Let L(E) denote the C ∗-algebra of adjointable operators on the Hilbert C ∗ module E and K(E) the ideal in L(E) consisting of the 'compact' operators. Lemma 7.6. Define π : C ∗ φ (l)(cid:1)∗ injective and π(cid:0)C ∗ Proof. This follows from 1) in Lemma 7.5. r(cid:0)R+ φ (k + 1)(cid:1)(cid:1) = K(E). φ(cid:1) ≃ OE, r(cid:0)Γ+ C ∗ Theorem 7.7. Assume that φ is Markov of order k. Then (cid:3) (7.1) 28 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN and there is a six-terms exact sequence ι0 ι1 (7.2) φ(cid:1)(cid:1) K0(cid:0)C ∗ r(cid:0)Γ+ r(cid:0)R+ K1(cid:0)C ∗ φ (k)(cid:1)(cid:1) id −[E]0 K0(cid:0)C ∗ r(cid:0)R+ φ (k)(cid:1)(cid:1) K0(cid:0)C ∗ r(cid:0)R+ φ (k)(cid:1)(cid:1) φ(cid:1)(cid:1) r(cid:0)R+ r(cid:0)Γ+ K1(cid:0)C ∗ K1(cid:0)C ∗ φ (k)(cid:1)(cid:1) φ(cid:1) is the inclusion map and [E] is the KK-theory r(cid:0)R+ φ (k)(cid:1) → C ∗ r(cid:0)Γ+ r(cid:0)R+ φ (k)(cid:1) → K(E). r(cid:0)Γ+ φ(cid:1) and ι : C ∗ φ (k)(cid:1) → C ∗ r(cid:0)R+ r(cid:0)Γ+ φ(cid:1) be the embeddings. It Proof. Let t : E → C ∗ follows then from Lemma 7.6 above and Proposition 3.3 in [Ka] that (ι, t) is covariant in the sense of Definition 3.4 in [Ka]. The isomorphism (7.1) follows then from 3) in Lemma 7.5 and Theorem 6.4 in [Ka]. Then the 6-terms exact sequence (7.2) follows from Lemma 7.6 above and Theorem 8.6 in [Ka]. (cid:3) where ι : C ∗ element defined from π : C ∗ id −[E]1 7.1. The linking groupoid. Lemma 7.8. Let (x, 0, y) ∈ R+ φ (k). There is an element z ∈ T such that (z, 1, x), (z, 1, y) ∈ Γ+ φ (1, k) and (x, 0, y) = (x, −1, z)(z, 1, y) in Γ+ φ . If φk−1(x) /∈ C1 it follows from the Markov condition and the surjectivity of φ that φk−1(x) ∈ φ (T\C1). Furthermore, the Markov condition also ensures that Proof. If φk−1(x) ∈ C1, choose any z ∈ φ−k(cid:0)φk−1(x)(cid:1) and note that val(cid:0)φk+1, z(cid:1) = val(cid:0)φ, φk−1(x)(cid:1) = val(cid:0)φk, x(cid:1). It follows that (z, 1, x), (z, 1, y) ∈ Γ+ val(cid:0)φk−1, x(cid:1) ∈ (±, ±). Since φ is Markov of order k there is therefore an element z ∈ T such that val(cid:0)φk, z(cid:1) = val(cid:0)φk−1, x(cid:1) and φk(z) = φk−1(x). It follows that (cid:3) (z, 1, x), (z, 1, y) ∈ Γ+ φ (1, k). φ (1, k). It follows from Lemma 7.3 and Lemma 7.8 that Γ+ equivalence in the sense of [SW] and hence from Theorem 13 in [SW] that C ∗ is Morita equivalent to C ∗ responding linking algebra. To this end we introduce the the linking groupoid L as follows, cf. Lemma 3 of [SW]. As a topological space L is the disjoint union r(cid:0)R+ φ (k)(cid:1) r(cid:0)R+ φ (k + 1)(cid:1). We need a detailed description of the cor- φ (1, k) is a R+ φ (k + 1) − R+ φ (k)- L = R+ φ (k + 1) ⊔ R+ φ (k) ⊔ Γ+ φ (1, k) ⊔ Γ+ φ (−1, k + 1). φ (k + 1) ⊔ R+ φ (k + 1) ⊔ Γ+ φ (1, k) to the first copy of T, and R+ To give L a groupoid structure, define r, s : L → T ⊔ T ⊆ R+ φ (k) ⊆ L to be the maps coming from the range and source maps on Γ+ φ , but such that r takes R+ φ (−1, k + 1) to the second while s takes R+ φ (−1, k + 1) to the first copy of T, and R+ φ (1, k) to the second. We then set L(2) = {(γ, γ′) ∈ L × L : s(γ) = r(γ′)} and define the product γγ′ for (γ, γ′) ∈ L(2) to be the same as the product in Γ+ φ . It is then straightforward to verify that L is a second countable ´etale locally compact Hausdorff groupoid. φ (k + 1) ⊔ Γ+ φ (k) ⊔ Γ+ φ (k) ⊔ Γ+ identified with C ∗ The C ∗-algebra of the reduction of L to the first copy of T is in a natural way r(cid:0)R+ φ (k + 1)(cid:1) and in the same way the C ∗-algebra of the reduction CIRCLE MAPS AND C ∗-ALGEBRAS 29 r (L) onto r (L). To see how a and b are related to [E], let f ∈ Cc(L) and of L to the second copy of T is identified with C ∗ give us embeddings a : C ∗ full corners of C ∗ r(cid:0)R+ φ (k)(cid:1). These identifications r(cid:0)R+ φ (k)(cid:1) → C ∗ φ (k + 1)(cid:1), f12 ∈ Cc(cid:0)Γ+ write f = f11 + f12 + f21 + f22 where f11 ∈ Cc(cid:0)R+ φ (1, k)(cid:1), f21 ∈ Cc(cid:0)Γ+ φ (k)(cid:1). We can then define Ψ(f ) ∈ L(cid:0)E ⊕ R+ r(cid:0)R+ φ (k + 1)(cid:1) → C ∗ φ (−1, k + 1)(cid:1) and f22 ∈ Cc(cid:0)R+ φ (k)(cid:1) such that Ψ(f )(e, g) = (f11e + f12g, f21e + f22g) . r (L) and b : C ∗ C ∗ canonical embeddings. Then It follows from Theorem 13 in [SW] and Corollary 3.21 in [RW] that Ψ extends to a ∗-isomorphism r (L) → K(cid:0)E ⊕ C ∗ r(cid:0)R+ φ (k)(cid:1)(cid:1) . r(cid:0)R+ Let K(E) → K(cid:0)E ⊕ C ∗ r(cid:0)R+ φ (k)(cid:1)(cid:1) and C ∗ r(cid:0)R+ φ (k)(cid:1) → K(cid:0)E ⊕ C ∗ φ (k)(cid:1)(cid:1) be the r(cid:0)R+ φ (k)(cid:1) r(cid:0)R+ φ (k + 1)(cid:1) r(cid:0)R+ φ (k)(cid:1) commutes, when π is the isomorphism from Lemma 7.6. By relating this diagram to the description of [E]∗ given by Definition 8.3 and Appendix A in [Ka] we find that r(cid:0)R+ φ (k)(cid:1)(cid:1) K(cid:0)E ⊕ C ∗ K(E) r (L) C ∗ C ∗ Ψ C ∗ C ∗ a b π To study [E]∗ further we need information on the structure of C ∗ following observation. [E]∗ = b−1 ∗ ◦ a∗ ◦ ρ∗. (7.3) r (L) and hence the Lemma 7.9. φj (Cj) = φj+1 (Cj+1) for all j ≥ 1. Proof. The inclusion φj (Cj) ⊆ φj+1 (Cj+1) is a general fact and follows from the observation that φ−1 (Cj) ⊆ Cj+1. For the converse, note that Cj+1 = C1 ∪ φ−1(C1) ∪ · · · ∪ φ−j (C1) and by the Markov property φ (Cj+1) ⊆ φ(C1) ∪ φ (φ−1(C1)) ∪ · · · ∪ φ (φ−j (C1)) ⊆ C1 ∪ C1 ∪ φ−1(C1) ∪ · · · ∪ φ−j+1(C1) = Cj, which implies the desired inclusion. (cid:3) Corollary 7.10. φj(Cj) = φ(C1) when j ≥ 1. Set D = φ(C1). Then Cj ⊆ φ−j(D) by Corollary 7.10. Note that φ−k−1 (D) ⊔ φ−k(D) is an L-invariant subset of T ⊔ T. We get therefore an extension 0 C ∗ C ∗ C ∗ r (L) r(cid:0)L(T⊔T)\(φ−k−1(D)⊔φ−k(D))(cid:1) 0 (7.4) By using the same procedure as in Section 6.2 we can give the following alternative description of this extension. As in Section 6.2 let Ik+1 be the set of intervals of connected components in T\φ−k−1(D), considered as subsets of the first copy of T in T ⊔ T, and Ik the set of intervals of connected components of T\φ−k(D), considered as subsets of the second copy of T in T ⊔ T. Let BL be the finite-dimensional C ∗- algebra generated by the matrix units eI,J, I, J ∈ Ik+1 ∪Ik, subject to the conditions that r(cid:0)Lφ−k−1(D)⊔φ−k(D)(cid:1) 30 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN a) φk+1(I) = φk+1(J) and val(cid:0)φk+1, I(cid:1) = val(cid:0)φk+1, J(cid:1) when I, J ∈ Ik+1, b) φk(I) = φk(J) and val(cid:0)φk, I(cid:1) = val(cid:0)φk, J(cid:1) when I, J ∈ Ik, c) φk+1(I) = φk(J) and val(cid:0)φk+1, I(cid:1) = val(cid:0)φk, J(cid:1) when I ∈ Ik+1 and J ∈ Ik, d) φk(I) = φk+1(J) and val(cid:0)φk, I(cid:1) = val(cid:0)φk+1, J(cid:1) when I ∈ Ik and J ∈ Ik+1. Then C ∗ r(cid:0)L(T⊔T)\(φ−k−1(D)⊔φ−k(D))(cid:1) ≃ SBL, Similarly, we can describe the C ∗-algebra C ∗ where we have used the notation SD for the suspension of a C ∗-algebra D, i.e. SD = C0(R) ⊗ D ≃ C0(]0, 1[) ⊗ D. dimensional C ∗-algebra AL generated by the matrix units ex,y, x, y ∈ φ−k−1(D) ⊔ φ−k(D) ⊆ T ⊔ T, subject to the conditions that r(cid:0)Lφ−k−1(D)⊔φ−k(D)(cid:1) as the finite- e) φk+1(x) = φk+1(y) and val(cid:0)φk+1, x(cid:1) = val(cid:0)φk+1, y(cid:1) when x, y ∈ φ−k−1(D), f) φk(x) = φk(y) and val(cid:0)φk, x(cid:1) = val(cid:0)φk, y(cid:1) when x, y ∈ φ−k(D), g) φk+1(x) = φk(y) and val(cid:0)φk+1, x(cid:1) = val(cid:0)φk, y(cid:1) when x ∈ φ−k−1(D) and h) φk(x) = φk+1(y) and val(cid:0)φk, x(cid:1) = val(cid:0)φk+1, y(cid:1) when x ∈ φ−k(D) and y ∈ y ∈ φ−k(D), φ−k−1(D). Then and the extension (7.4) takes the form 0 C ∗ r(cid:0)Lφ−k−1(D)⊔φ−k(D)(cid:1) ≃ AL, r(cid:0)R+ φ (k)(cid:1) and C ∗ r (L) SBL C ∗ AL 0 This extension is compatible with the description of C ∗ coming from Section 6.2 in the sense that we get a commutative diagram of ∗- homomorphisms r(cid:0)R+ φ (k + 1)(cid:1) (7.5) (7.6) Ak+1 AL 0 0 Ak 0. 0 0 0 SBk+1 SBL SBk C ∗ r(cid:0)R+ φ (k + 1)(cid:1) a C ∗ r (L) b C ∗ r(cid:0)R+ φ (k)(cid:1) From the six-terms exact sequences of the upper and lower extensions in (7.6) we get for j = k and j = k + 1 the identifications φ (j)(cid:1)(cid:1) = ker(cid:0)(Ij)0 − (Uj)0(cid:1) r(cid:0)R+ K0(cid:0)C ∗ φ (j)(cid:1)(cid:1) = coker(cid:0)(Ij)0 − (Uj)0(cid:1) , K1(cid:0)C ∗ r(cid:0)R+ (7.8) where Ij, Uj : Aj → Bj are the ∗-homomorphisms introduced in Section 6.2. Then b−1 0 ◦ a0 is realized as a homomorphism (7.7) and b−1 0 ◦ a0 : ker ((Ik+1)0 − (Uk+1)0) → ker ((Ik)0 − (Uk)0) and b−1 1 ◦ a1 as a homomorphism b−1 1 ◦ a1 : coker ((Ik+1)0 − (Uk+1)0) → coker ((Ik)0 − (Uk)0) . CIRCLE MAPS AND C ∗-ALGEBRAS 31 Note that the full matrix summands in each of the C ∗-algebras Bk, BL and Bk+1 are in one-to-one correspondence with I(±) = I × {(+, +), (−, −)}, where I are the intervals of connected components in T\D. In Bk, for example, the element (I, (+, +)) ∈ I(±) labels the C ∗-subalgebra of Bk generated by the matrix units of Bk generated by the matrix units eI ′,J ′ where I ′, J ′ are intervals in Ik such that eI ′,J ′ where I ′, J ′ are intervals in Ik such that φk(I ′) = φk(J ′) = I and val(cid:0)φk, I ′(cid:1) = val(cid:0)φk, J ′(cid:1) = (+, +), while the element (I, (−, −)) ∈ I(±) labels the C ∗-subalgebra φk(I ′) = φk(J ′) = I and val(cid:0)φk, I ′(cid:1) = val(cid:0)φk, J ′(cid:1) = (−, −). Similarly, the full an element x ∈ φ−k(d) such that val(cid:0)φk, x(cid:1) = v. By using these labels we get matrix summands in Ak, AL and Ak+1 are in one-to-one correspondence with the subset D(±) of D × V consisting of the pairs (d, v) ∈ D × V for which there exists isomorphisms K0 (Ak) ≃ K0 (Ak+1) ≃ K0 (AL) ≃ ZD(±) and K0 (Bk) ≃ K0 (Bk+1) ≃ K0 (BL) ≃ ZI(±) ≃ ZI ⊕ ZI −1◦a0 and b1 becomes under these identifications. −1◦a1 with the property that by making them identifications the maps b0 become identities. Therefore, in order to obtain the desired description of [E]∗ 7.2. A description of ρ : C ∗ Section 6.2, set we must determine what the map ρ∗ : K∗(cid:0)C ∗ r(cid:0)R+ φ (k)(cid:1)(cid:1) → K∗(cid:0)C ∗ r(cid:0)R+ φ (k + 1)(cid:1)(cid:1) r(cid:0)R+ r(cid:0)R+ φ (k)(cid:1) → C ∗ φ (k + 1)(cid:1). Using the notation from r(cid:0)R+ φ (k)(cid:1) → r(cid:0)R+ φ (k + 1)(cid:1) → Dk+1. There is therefore a unique ∗-homomorphism Dj = {(a, f ) ∈ Aj, f ∈ C ([0, 1], Bj) : f (0) = Ij(a), f (1) = Uj(a)} when j ≥ 1. It follows from Lemma 6.6 that there are isomorphisms µk : C ∗ Dk and µk+1 : C ∗ Φ : Dk → Dk+1 such that ρ µk C ∗ r(cid:0)R+ φ (k)(cid:1) r(cid:0)R+ φ (k + 1)(cid:1) µk+1 C ∗ Dk Φ Dk+1 (7.9) commutes. The ∗-homomorphism Φ is given by the formula Φ(a, f ) = (χ(a) + µ(f ), ϕ(f )) (7.10) where χ : Ak → Ak+1, µ : C ([0, 1], Bk) → Ak+1, ϕ : C ([0, 1], Bk) → C ([0, 1], Bk+1) are ∗-homomorphisms such that χ and µ have orthogonal ranges. We describe them one by one. The easiest to define is χ; it is simply given by the formula χ(ex,y) = ex,y To define µ and ϕ we use the homeomorphisms ψJ :]0, 1[→ J ∈ I that were introduced in Section 6.2. Consider f ∈ C[0, 1] and a pair (I, J) ∈ I (2) k . Then when φk(x) = φk(y) ∈ D and val(cid:0)φk, x(cid:1) = val(cid:0)φk, y(cid:1). f(cid:0)ψφk(I) µ(f ⊗ eI,J ) = X(x,y)∈NI,J −1(cid:0)φk(x)(cid:1)(cid:1) ex,y 32 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN where where NI,J =(cid:8)(x, y) ∈ φ−k−1(D)2 : x ∈ I, y ∈ J, φk(x) = φk(y) /∈ D(cid:9) . Similarly, ϕ is given by fI1 ⊗ eI1,J1 (7.11) ϕ (f ⊗ eI,J ) = X(I1,J1)∈MI,J MI,J =n(I1, J1) ∈ I (2) k+1 : I1 ⊆ I, J1 ⊆ Jo and fI1 ∈ C[0, 1] is the continuous extension of the function ]0, 1[∋ t 7→ f ◦ ψ−1 φk(I) ◦(cid:16)φφk(I)∩φ−1(φk+1(I1))(cid:17)−1 ◦ ψφk+1(I1)(t). (We remark that this description of ρ is almost identical with the one given in Lemma 3.5 in [Th5] in a different setting.) Define pj : Dj → Aj such that pj(a, f ) = a, and note that pk+1 ◦ Φ is homotopic to (χ + µ ◦ Ik) ◦ pk so that Dk Φ pk Ak χ+µ◦Ik Dk+1 pk+1 Ak+1 (7.12) commutes up to homotopy. It follows that ρ0 = χ0 + µ0 ◦ (Ik)0 on K0(Dk) = ker ((Ik)0 − (Uk)0). 7.3. An algorithm for the calculation of K∗(cid:0)C ∗ φ(cid:1)(cid:1). For each (d, v) ∈ D(±) r(cid:0)Γ+ let [d, v] be the corresponding element in the standard basis for ZD(±). For each d ∈ D we let I + d . We can then define a homomorphism A : ZD(±) → ZD(±) such that d be the interval I + d ∈ I such that d < I + when v = (+, −), A[d, v] = [φ(d), val (φ, d)] A[d, v] = [φ(d), val (φ, d)] + Xz∈I + d ∩φ−1(D) ([φ(z), val (φ, z) • (+, +)] + [φ(z), val (φ, z) • (−, −)]) when v = (−, +), when v = (+, +) and finally A[d, v] = [φ(d), val (φ, d)] + Xz∈I + A[d, v] = [φ(d), val (φ, d)] + Xz∈I + d ∩φ−1(D) d ∩φ−1(D) when v = (−, −). [φ(z), val (φ, z) • (+, +)] [φ(z), val (φ, z) • (−, −)] CIRCLE MAPS AND C ∗-ALGEBRAS 33 Under our identification of K0 (Ak) and K0(Ak+1) with ZD(±), the homomorphism ρ0 = χ0 + µ0 ◦ (Ik)0 is given by A. In the following we let A denote the restriction of A to ker ((Ik)0 − (Uk)0) ⊆ ZD(±). Since (ij)1 : K1 (SBj) → K1 (Dj) is surjective and K1 (SBk) is free there is auto- matically a homomorphism B : K1 (SBk) → K1 (SBk+1) such that the diagram K1 (SBk) (ik)1 K1(Dk) B Φ K1 (SBk+1) (ik+1)1 K1 (Dk+1) (7.13) commutes. In fact, there are generally many choices since (ik+1)1 has a kernel. We describe now a way to choose B such that it is easy to determine from φ. To this end we need some notation. For each I ∈ I, v ∈ {(+, +), (−, −)}, let [I, v] be the corresponding standard basis element in ZI(±). For v ∈ {(+, +), (−, −)}, set (−1)v =(1 when v = (+, +) −1 when v = (−, −). Let J ∈ I and choose J1 ∈ I such that J1 ⊆ φ(J). We choose J1 such that J1 = φ(JX) where JX ⊆ J is an open subinterval and JX ∩ C1 = ∅. Let I ∈ Ik be a unique open subinterval I1 ⊆ I such that φk(I1) = JX. Note that φk+1 is injective such that φk(I) = J and val(cid:0)φk, I(cid:1) = v ∈ (±, ±). Since φk is injective on I there is on I1 and φk+1(I1) = J1. In particular, I1 ∈ Ik+1. Let u ∈ Cc(I1) ⊆ Cc(cid:0)R+ φ (k)(cid:1) be a positively oriented path in T − 1 of degree 1. Then (u ◦ λI ◦ ψJ ) ⊗ eI,I represents (−1)v[J, v] ∈ ZI(±) ≃ K1(SBk). By using (7.10) and (7.11) we find that µk+1 ◦ ρ(u) ∈ SBk+1 ⊆ Dk+1, and in SBk+1 is the element (cid:18)u ◦ λI ◦(cid:16)φφk(I)∩φ−1(φk+1(I1))(cid:17)−1 ◦ ψJ1(cid:19) ⊗ eI1,I1 = (u ◦ λI1 ◦ ψJ1) ⊗ eI1,I1 which corresponds to (−1)v(−1)w [J1, w • v] in ZI(±) where w = val (φ, JX). There- fore a recipe for a construction of B reads as follows: For each J ∈ I choose an open interval J ′ ⊆ J such that J ′ ∩ C1 = ∅ and φ(J ′) ∈ I. Then B[J, v] = (−1)val(φ,J ′) [φ(J ′), val (φ, J ′) • v] . The construction of B generally depends on many choices which lead to different homomorphisms. But the commutativity of (7.13) guarantees that they all leave im ((Ik)0 − (Uk)0) ⊆ ZI(±) globally invariant and induce the same endomorphism B of coker ((Ik)0 − (Uk)0). In view of the 6 terms exact sequence (7.2) it now follows that the endomorphisms A : ker ((Ik)0 − (Uk)0) → ker ((Ik)0 − (Uk)0) , B : coker ((Ik)0 − (Uk)0) → coker ((Ik)0 − (Uk)0) φ(cid:1)(cid:1) in the sense that there are extensions r(cid:0)Γ+ φ(cid:1)(cid:1) and K1(cid:0)C ∗ r(cid:0)Γ+ determine K0(cid:0)C ∗ coker(cid:16)1 − A(cid:17) K0(cid:0)C ∗ φ(cid:1)(cid:1) r(cid:0)Γ+ ker(cid:16)1 − B(cid:17) 0 0 (7.14) 34 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN and 0 0. (7.15) m(d, v)[d, v] Note that the last extension is always split and hence To identify the C ∗-algebra from its K-theory groups it is important to know coker(cid:16)1 − B(cid:17) ker(cid:16)1 − A(cid:17) φ(cid:1)(cid:1) K1(cid:0)C ∗ r(cid:0)Γ+ φ(cid:1)(cid:1) ≃ coker(cid:16)1 − B(cid:17) ⊕ ker(cid:16)1 − A(cid:17) . r(cid:0)Γ+ K1(cid:0)C ∗ φ(cid:1). Note therefore r(cid:0)Γ+ φ(cid:1)(cid:1) represents the unit 1 of C ∗ r(cid:0)Γ+ which element of K0(cid:0)C ∗ that [1] ∈ K0(cid:0)C ∗ r(cid:0)Γ+ φ(cid:1)(cid:1) is the image of [1] ∈ K0(cid:0)C ∗ φ (k)(cid:1)(cid:1) under the map ι0 in r(cid:0)R+ [1] = X(d,v)∈D(±) (7.2). Under the identification K0 (Ak) = ZD(±) we have that unital ∗-homomorphisms this element is always in ker ((Ik)0 − (Uk)0) and gives there- Example 7.11. In this example we show with a fair amount of details how to complete the K-theory calculations for a family of Markov maps by using the recipe described above. Let m, k ∈ N and let gm,k : [0, 1] → R be the continuous piecewise in K0 (Ak), where m(d, v) = #(cid:8)x ∈ φ−k(d) : val(cid:0)φk, x(cid:1) = v(cid:9). Since Ik and Uk are fore rise to an element of coker(cid:16)1 − A(cid:17) which under the embedding coker(cid:16)1 − A(cid:17) ⊆ φ(cid:1)(cid:1). φ(cid:1)(cid:1) from (7.14) gives us the element representing [1] ∈ K0(cid:0)C ∗ r(cid:0)Γ+ r(cid:0)Γ+ K0(cid:0)C ∗ 2(cid:3) and linear map with the properties that gm,k(0) = 0, gm,k has slope 2m on (cid:2)0, 1 2, 1(cid:3). Then gm,k is the lift of a piecewise monotone map φm,k on slope −2k on (cid:2) 1 φm,k(cid:17) is simple, unital, separable, purely infinite and satisfies the UCT. By r(cid:16)Γ+ the circle which is exact and not locally injective. To simplify the calculations we assume that m ≥ 2, k ≥ 2 or that m = k. Then φm,k is Markov of order 1 and C ∗ the Kirchberg-Phillips classification result the algebra is therefore determined by its K-theory groups and the position of the unit in the K0-group. Note that 1 is the only critical value and that D(±) = 1 × V = {(1, (−, +)), (1, (+, −)), (1, (+, +)), (1, (−, −))}. There is only one interval in I, namely I = T\{1}, and I(±) = {(I, (+, +)), (I, (−, −)}. When we take the elements of D(±) and I(±) in that order we find that (I1)0 =(cid:18)1 0 1 0 1 0 0 1(cid:19) , (U1)0 =(cid:18)0 1 1 0 0 1 0 1(cid:19) , It follows that ker ((I1)0 − (U1)0) =(cid:8)(z, z, u, v) ∈ Z4 : z, u, v ∈ Z(cid:9) ≃ Z3, and The matrix A is coker ((I1)0 − (U1)0) ≃ Z. A = 1 2 1 0 1 1 1 1 m + k − 2 0 m − 1 k − 1 m + k − 2 0 k − 1 m − 1  . CIRCLE MAPS AND C ∗-ALGEBRAS 35 and its restriction A to ker ((I1)0 − (U1)0) = Z3 is 2 1 1 1 m + k − 2 m − 1 k − 1 A = m + k − 2 k − 1 m − 1 . A − 1 = m + k − 2 k − 1 m − 2 .  k  . 1 0 0 m − 1 m + k − 2 m − 2 k − 1 m − 1 1 k 1 1 1 After a couple of row-operations we get from A − 1 the matrix Hence and Hence Hence ker(cid:16) A − 1(cid:17) = 0 when k 6= m − 1 while ker(cid:16) A − 1(cid:17) ≃ Z when k = m − 1. To determine coker(cid:16) A − 1(cid:17) note that this cokernel is the same as the cokernel of (cid:18) k m − 1 m − 1 k (cid:19) . Let g be the greatest common divisor of k and m − 1 when m ≥ 1 and set g = k when m = 1. There are then x, y ∈ Z such that x k g + y m−1 g = 1. Then (cid:18) x − m−1 g y k g(cid:19)(cid:18) k m − 1 g + y k k2−(m−1)2 g g2 ! k (cid:1) After a final column operation this shows that the Smith normal form of(cid:0) k m−1 m−1 is (cid:18) x − m−1 g m − 1 y k g(cid:19) ∈ GL2(Z) 0 g(cid:19) 1 x m−1 k (cid:19) =(cid:18)g 0 ! . 0 0 g 0 k2−(m−1)2 g coker(cid:16) A − 1(cid:17) ≃ Zg ⊕ Z k2−(m−1)2 g . From the recipe for the matrix B it is easily seen that we can choose it to be the iden- tity matrix in this case. It follows that the endomorphism B of coker ((I1)0 − (U1)0) is the identity and hence ker(cid:16) B − 1(cid:17) ≃ coker(cid:16) B − 1(cid:17) ≃ coker ((I1)0 − (U1)0) ≃ Z. All in all we get the conclusion that when k = m − 1 and r(cid:16)Γ+ φm,k(cid:17)(cid:17) ≃ Z2 ⊕ Zm−1, K1(cid:16)C ∗ K0(cid:16)C ∗ r(cid:16)Γ+ K0(cid:16)C ∗ φm,k(cid:17)(cid:17) ≃ Z ⊕ Zg ⊕ Z k2−(m−1)2 r(cid:16)Γ+ φm,k(cid:17)(cid:17) ≃ Z2 r(cid:16)Γ+ , K1(cid:16)C ∗ φm,k(cid:17)(cid:17) ≃ Z g (cid:17) ∈ Zg ⊕ Z k2−(m−1)2 . Zk2−(m−1)2 and in the second by(cid:16)−x, m−1 g g when k 6= m − 1. In the first case the class of the unit is represented by m − 1 ∈ 36 THOMAS LUNDSGAARD SCHMIDT AND KLAUS THOMSEN of k and m are interchanged and we find that Let φ−m,−k(z) = φm,k(z). Then φ−m,−k is also exact and Markov, and the calcu- r(cid:16)Γ+ lation of K∗(cid:16)C ∗ φ−m,−k(cid:17)(cid:17) can proceed exactly as above. In comparison the roles r(cid:16)Γ+ K0(cid:16)C ∗ φ−m,−k(cid:17)(cid:17) ≃ Z2 ⊕ Zk−1, K1(cid:16)C ∗ r(cid:16)Γ+ φ−m,−k(cid:17)(cid:17) ≃ Z ⊕ Zg′ ⊕ Z m2 −(k−1)2 r(cid:16)Γ+ φ−m,−k(cid:17)(cid:17) ≃ Z2 r(cid:16)Γ+ , K1(cid:16)C ∗ φ−m,−k(cid:17)(cid:17) ≃ Z when m = k − 1 and K0(cid:16)C ∗ when m 6= k − 1, where g′ is now the greatest common divisor of m and k − 1, or g′ = m when k = 1. g′ [AdRR] [An] [A-DR] [AK] [AT] [CM] [De] [DS] [Ka] [Ph] [PS] [RW] [Re] [RS] [S] [SW] [Sp] [Th1] [Th2] [Th3] References Ll. Alsed´a, M.A. del R`ıo and J.A. Rodr`ıguez, A splitting theorem for transitive maps, J. Math. Anal. Appl. 232 (1999), 359-375. C. Anantharaman-Delaroche, Purely infinite C ∗-algebras arising from dynamical sys- tems, Bull. Soc. Math. France 125 (1997), 199 -- 225. C. Anantharaman-Delaroche and J. Renault, Amenable groupoids, Monographies de L'Enseignement Mathmatique, vol. 36, L'Enseignement Mathmatique, Geneva, 2000. J. Auslander and Y. Katznelson, Continuous maps of the circle without periodic points, Israel Jour. Math. 32 (1979), 375-381. K.K.S. Andersen and K. Thomsen, The C ∗-algebra of an affine map on the 3-torus, Documenta Mathematica 17 (2012), 531-558. E.M. Coven and I. Mulvey, Transitivity and the centre for maps of the circle, Ergod. Th. & Dynam. Sys. 6 (1986), 1-8. V. Deaconu, Groupoids associated with endomorphisms, Trans. Amer. Math. Soc. 347 (1995), 1779-1786. V. Deaconu and F. Shultz, C ∗-algebras associated with interval maps, Trans. Amer. Math. Soce. 359 (2007), 1889-1924. T. Katsura, On C ∗-algebras associated with C ∗-correspondences, J. Func. Analysis 217 (2004), 366-401. N. C. Phillips, A classification theorem for nuclear purely infinite simple C ∗-algebras, Doc. Math. 5 (2000), 49-114. I. Putnam and J. Spielberg, The structure of C ∗-algebras associated with hyperbolic dynamical systems, J. Func. Anal. 163 (1999), 279 -- 299. I. Raeburn and D. P. Williams, Morita Equivalence and Continuous-Trace C ∗-algebras, American Mathematical Society, 1998. J. Renault, A Groupoid Approach to C ∗-algebras, LNM 793, Springer Verlag, Berlin, Heidelberg, New York, 1980. J. Rosenberg and C. Schochet, The Kunneth theorem and the universal coefficient theorem for Kasparov's generalized K-functor, Duke J. Math. 55 (1987), 337-347. F. Shultz, Dimension groups for interval maps II: The transitive case, Ergod. Th. & Dynam. Sys. 27 (2007), 1287-1321. A. Sims and D. Williams, Renault's equivalence theorem for reduced groupoid C ∗- algebras, arXiv:1002.3093v2 J. Spielberg, Groupoids and C ∗-algebras for categories of paths, arXiv:1111.6924v3. K. Thomsen, Semi-´etale groupoids and applications, Annales de l'Institute Fourier 60 (2010), 759-800. K. Thomsen, KM S states and conformal measures, Comm. Math. Phys. 316 (2012), 615-640. DOI: 10.1007/s00220-012-1591-z K. Thomsen, The groupoid C ∗-algebra of a rational map, J. Non-commutative Geom- etry, to appear. CIRCLE MAPS AND C ∗-ALGEBRAS 37 [Th4] [Th5] [T] [Tu] [Y] [W] K. Thomsen, The C ∗-algebra of the exponential function, Proc. Amer. Math. Soc., to appear. K. Thomsen, The homoclinic and heteroclinic C ∗-algebras of a generalized one- dimensional solenoid, Ergod. Th. & Dynam. Sys. 30 (2010), 263-308. A. Toms, K-theoretic rigidity and slow dimension growth, Invent. Math., to appear, arXiv:0910.2061v2. J.-L. Tu, La conjecture de Baum-Connes pour les feuilletages moyennables, K-theory 17 (1999), 215-264. K. Yokoi, Strong transitivity and graph maps, Bull. Polish Acad. Math. 53 (2005), 377-388. P. Walters, An introduction to ergodic theory, Springer Verlag, New York, 1979. E-mail address: [email protected] Institut for Matematik, Aarhus University, Ny Munkegade, 8000 Aarhus C, Den- mark
1606.07465	1	1606	2016-06-20T16:13:05	The cocycle identity holds under stopping	[ "math.OA", "math.PR" ]	In recent work of the authors, it was shown how to use any finite quantum stop time to stop the CCR flow and its strongly continuous isometric cocycles (Q. J. Math. 65:1145-1164, 2014). The stopped cocycle was shown to satisfy a stopped form of the cocycle identity, valid for deterministic increments of the time used for stopping. Here, a generalisation of this identity is obtained, where both cocycle parameters are replaced with finite quantum stop times.	math.OA	math	The cocycle identity holds under stopping Alexander C. R. Belton Kalyan B. Sinha Department of Mathematics and Statistics Lancaster University, United Kingdom Jawaharlal Nehru Centre for Advanced Scientific Research, Bangalore, India [email protected] [email protected] 20th June 2016 Abstract In recent work of the authors, it was shown how to use any finite quantum stop time to stop the CCR flow and its strongly continuous isometric cocycles (Q. J. Math. 65:1145 -- 1164, 2014). The stopped cocycle was shown to satisfy a stopped form of the cocycle identity, valid for deterministic increments of the time used for stopping. Here, a generalisation of this identity is obtained, where both cocycle parameters are replaced with finite quantum stop times. Key words: quantum stopping time; quantum stop time; quantum Markov time; operator cocycle; Markov cocycle; Markovian cocycle; quantum stochastic cocycle; CCR flow. MSC 2010: 46L53 (primary); 46L55, 60G40 (secondary). 1 Introduction The history of stopping times in non-commutative probability begins in 1979, with Hudson's work on stopping canonical Wiener processes [4]. Since then, many authors have contributed to the subject, and it has developed in various directions and settings: abstract von Neumann algebras, to produce first exit times in C ∗ algebras and to stop quantum stochastic integrals, for example. A good introduction for the latter is provided by [5]; see [2] for further references. In this note, we extend a previous result [2, Theorem 7.2], which itself built upon work of Parthasarathy and Sinha [6] and Applebaum [1]. Let V be a strongly continuous isometric cocycle of the CCR flow σ, so that where bV is the identity-adapted projection of the p-adapted process V . The importance of this identity in classical and quantum probability is well known; it has an intimate connection with stochastic integral representation and Feynman -- Kac formulae [7, 3]. Vs+t = bVs σs(Vt) for all s, t ∈ R+, If S is a finite quantum stop time then Theorem 7.2 of [2] gives the stopped cocycle identity 6 1 0 2 n u J 0 2 ] . A O h t a m [ 1 v 5 6 4 7 0 . 6 0 6 1 : v i X r a VS+t = bVS σS(Vt) for all t ∈ R+. (1) It is shown below that the following generalisation of (1) holds: if T is another finite quantum stop time and the CCR flow σ has countable rank then where S ⋆ T is the convolution of S and T . If V acts on an initial space h, it follows that setting VS ⋆ T = bVS σS(VT ), αS(a) = VS(a ⊗ I)V ∗ S for all a ∈ B(h) 1 gives a generalised Evans -- Hudson flow αS which satisfies a non-deterministic version of the mapping-cocycle relation, The notation of [2] is followed throughout. In particular, the algebraic tensor product is denoted by ⊗, with ⊗ the Hilbert-space and ⊗ the ultraweak product. αS⋆T =bαS ◦ σS ◦ αT . 2 Stopped maps with a non-trivial initial space In Sections 6 and 7 of [2], certain maps ES, ΓS and σS are extended to the case of a non-trivial initial space, so that the ambient Fock space F is replaced by h ⊗ F, where h is a complex Hilbert space. In order to familiarise the reader with key ideas and notation from [2], and as the construction of these extensions are not quite immediate, the details are provided in this section, together with some further observations. Notation 2.1. Let F = Γ+(cid:0)L2(R+; k)(cid:1) be Boson Fock space over the complex Hilbert space of square-integrable functions on the half line R+ := [0, ∞), with values in the complex Hilbert space k. Recall the tensor-product decomposition F = Ft) ⊗ F[t, valid for all t ∈ (0, ∞), where Ft) := Γ+(cid:0)L2([0, t); k)(cid:1) and Ft) := Γ+(cid:0)L2([t, ∞); k)(cid:1), given by extending the identification of exponential vectors such that ε(f ) = ε(f [0,t))⊗ε(f [t,∞)) for all f ∈ L2(R+; k). Let I, It) and I[t denote the identity operators on F, Ft) and F[t, respectively, and let E denote the linear span of the set of exponential vectors in F. Definition 2.2. Let S be a finite quantum stop time, so that S : B[0, ∞] → B(F) is a map from the Borel subsets of the extended half line to the set of orthogonal projections on F, such that (i) the map A 7→ hx, S(A)yi is a complex measure for all x, y ∈ F, (ii) the total mass S(cid:0)[0, ∞](cid:1) = I, with S(cid:0){∞}(cid:1) = 0, and (iii) identity adaptedness holds, so that S({0}) = 0 and S(cid:0)[0, t](cid:1) ∈ B(Ft))⊗I[t for all t ∈ (0, ∞). Notation 2.3. For all t ∈ R+, let Et := Γ+(1[0,t)) ∈ B(F) be the second quantisation of the operator obtained by letting this indicator function act by multiplication, so that Et is the orthogonal projection onto Ft) ⊗ ε(0[t,∞)), and let E∞ := I. Proposition 2.4. Let π = {0 = π0 < · · · < πn+1 = ∞} be a finite partition of [0, ∞] and let h be a complex Hilbert space. If eES,π := Ih ⊗ ES,π, where ES,π := S(cid:0)(πj−1, πj](cid:1)Eπj , n+1Xj=1 then ES,π → ES and eES,π → eES := Ih ⊗ ES in the strong operator topology as π is refined, where ES and eES are orthogonal projections. Proof. The proof of [2, Theorem 3.7] gives that ES,π → ES strongly on E, and thus eES,π → eES strongly on h ⊗ E; the result follows by the density of this last space in h ⊗ F. 2 Notation 2.5. For all s ∈ R+, let Γs := Γ+(θs) ∈ B(F) be the second quantisation of the isometric right shift, such that (θsf )(t) = 1[s,∞)(t)f (t − s) for all t ∈ R+, and let Γ∞ := E0. Proposition 2.6. Let π and h be as in Proposition 2.4. If eΓS,π := Ih ⊗ ΓS,π, where ΓS,π := n+1Xj=1 S(cid:0)(πj−1, πj](cid:1)Γπj , then ΓS,π → ΓS and eΓS,π → eΓS := Ih ⊗ ΓS in the strong operator topology as π is refined, where ΓS and eΓS are isometries. Proof. The claims about ΓS,π and Γ follow from the proof of [2, Theorem 3.8], which also gives that k(eΓS,π −eΓS)u ⊗ xk = kuk k(ΓS,π − ΓS)xk → 0 as π is refined, for all u ∈ h and x ∈ F. As ΓS,π and ΓS are isometries, the same is true foreΓS,π andeΓS. ThuseΓS,π →eΓS strongly on h ⊗ F, and so on h ⊗ F, since keΓS,πk = 1 for all π. Notation 2.7. For all t ∈ R+, let the ultraweakly continuous unital ∗-homomorphism σt : B(F) → B(F); X 7→ It) ⊗ ΓtXΓ∗ t , where Γt is regarded here as an isometric isomorphism from F to F[t with inverse Γ∗ t , and let eσt := idB(h) ⊗ σt. Recall that (σt : t ∈ R+) is the CCR flow semigroup with rank dim k. Notation 2.8. Let FS) := ES(F) and F[S := ΓS(F) be the pre-S and post-S spaces, with identity operators IS) and I[S, respectively. Theorem 2.9. Let h and π be as in Proposition 2.4. If where eσ(Z)S,π := (idB(h) ⊗ σS,π)(Z) σS,π : B(F) → B(F); X 7→ n+1Xj=1 for all Z ∈ B(h ⊗ F), σπj (X) S(cid:0)(πj−1, πj](cid:1), then σS,π → σS and eσS,π → eσS := idB(h) ⊗ σS pointwise in the strong operator topology as π is refined, where σS and eσS are ultraweakly continuous unital ∗-homomorphisms. Furthermore, there exist isometric isomorphisms jS : FS) ⊗ F[S → F and eS : FS) ⊗ h ⊗ F[S → h ⊗ F such that and σS(X) = jS(IS) ⊗ ΓSXΓ∗ S)j∗ S for all X ∈ B(F) eσS(Z) =eS(IS) ⊗eΓSZeΓ∗ S)e∗ S for all Z ∈ B(h ⊗ F). (2) (3) Proof. The convergence of σS,π to σS, and the fact that the latter is a unital ∗-homomorphism, follows from [2, Theorem 5.2]. The representation (2) is [2, Proposition 5.3], and this shows that the map X 7→ σS(X) is continuous when B(F) is equipped with the ultraweak topology, as ampliation gives a normal representation of any von Neumann algebra. In particular, the 3 isometric isomorphism map eσS is an ultraweakly continuous unital ∗-homomorphism such that (3) holds, where the because (3) holds if Z is a simple tensor, and both sides are ultraweakly continuous functions of Z. eS : FS) ⊗ h ⊗ F[S → h ⊗ F; x ⊗ u ⊗ y 7→ u ⊗ jS (x ⊗ y), if the finite partition π′ is a refinement of π, then, for any u ∈ h and any f ∈ L2(R+; k) with compact support, It remains to prove that eσS,π converges to eσS. Working as in the proof of [2, Theorem 5.2], k(eσS,π′ −eσS,π)(Z)uε(f )k 6 kS(cid:0)[0, πn](cid:1)ε(f )k sup{k(eσr(Z) − Z)uε(f (· + s)) : r ∈ [0, δπ], s ∈ [0, τ ]} + kS(cid:0)(πn, ∞)(cid:1)uε(f )k (kZk + 1), Using the same argument as in the proof of [2, Theorem 5.2], and noting that r 7→ eσr(Z) is strongly continuous, it now follows thateσS,π(Z)uε(f ) is convergent, as π is refined, for any u ∈ h and any f ∈ L2(R+; k) with compact support; let the limit be denoted by λS(Z)uε(f ) and extend by linearity. Since where δπ := max{πj − πj−1 : j = 1, . . . , n} and f has support contained in [0, τ ] ⊆ [0, ∞). kλS(Z)zk = lim π keσS,π(Z)zk 6 kZk kzk for all z ∈ h ⊗ Ec, where Ec is the linear span of those exponential vectors corresponding to functions with compact support, there exists a bounded linear operator λS(Z) on h ⊗ F which extends the linear in the strong operator topology, everywhere on h ⊗ F. map z 7→ λS(Z)z. Furthermore, the usual approximation argument gives thateσS,π(Z) → λS(Z) To conclude, we use (3) and argue as the proof of [2, Proposition 5.3]. Using the notation of that proof and the identity at the top of [2, p.1158], that hIS,π∩[0,t](f, Γε(g)), σS,π(X)IS,π∩[0,t](f ′, Γε(g′))i = hES,π∩[0,t]ε(f ), ES,π∩[0,t]ε(f ′)i hε(g), Xε(g′)i, together with the ultraweak continuity ofeσS,π, it follows that hu ⊗ IS,π∩[0,t](f, Γε(g)),eσS,π(Z)u′ ⊗ IS,π∩[0,t](f ′, Γε(g′))i = hES,π∩[0,t]ε(f ), ES,π∩[0,t]ε(f ′)i huε(g), Zu′ε(g′)i for all u, u′ ∈ h, f , f ′, g, g′ ∈ L2(R+; k), t ∈ (0, ∞) and Z ∈ B(h ⊗ F). As π is refined, the right-hand side converges to by [2, Theorem 3.7], whereas the left-hand side converges to hES,tε(f ) ⊗ u ⊗ ΓSε(g), (IS) ⊗eΓSZeΓ∗ S)ES,tε(f ′) ⊗ u′ ⊗ ΓSε(g′)i, hu ⊗ jS(ES,tε(f ) ⊗ ΓSε(g)), λS (Z)u′ ⊗ jS(ES,tε(f ′) ⊗ ΓSε(g′))i by [2, Lemma 3.4 and Theorem 3.10]. The result follows. = heS(ES,tε(f ) ⊗ u ⊗ ΓSε(g)), λS (Z)eS(ES,tε(f ′) ⊗ u′ ⊗ ΓSε(g′))i, Remark 2.10. The representations (2) and (3) also give that X 7→ σS(X) and Z 7→ eσS(Z) are continuous on bounded subsets of B(F) and B(h ⊗ F), respectively, when these spaces are equipped with the strong operator topology, since this is true of the ampliation map T 7→ I ⊗ T . 4 3 The cocycle identity with two stop times Definition 3.1 ([2, Definition 4.1]). The convolution S ⋆ T of two finite quantum stop times S and T is S ⋆ T : B(R+) → B(F); A 7→ (S ⊗ T )(cid:0)f −1(A)(cid:1), f : R+ × R+ → R+; (x, y) 7→ x + y where and S ⊗ T : B(R+ × R+) → B(F); A × B 7→ jS(S(A)FS) ⊗ ΓST (B)Γ∗ S)j∗ S . Lemma 3.2. Let S and T be finite quantum stop times. Then (S ⊗ T )(A × B) = S(A)σS(cid:0)T (B)(cid:1) for all A, B ∈ B[0, ∞]. (4) Proof. If t ∈ [0, ∞] then, by Theorems 3.7 and 3.10, together with Lemma 3.4, of [2], jS(S(cid:0)[0, t](cid:1)ESε(f ) ⊗ ΓSx) = jS (ES,tε(f ) ⊗ ΓSx) =Z[0,t] S(ds)ε(f [0,s)) ⊗ Γsx = S(cid:0)[0, t](cid:1)Z[0,∞] = S(cid:0)[0, t](cid:1)jS(ES ε(f ) ⊗ ΓSx) S(ds)ε(f [0,s)) ⊗ Γsx for all f ∈ L2(R+; k) and x ∈ F. Hence jS(S(A)FS) ⊗ I[S)j∗ S = S(A) for all A ∈ B[0, ∞]. It follows that (S ⊗ T )(A × B) := jS(S(A)FS) ⊗ ΓST (B)Γ∗ S)j∗ S = jS(S(A)FS) ⊗ I[S)j∗ S jS(IS) ⊗ ΓST (B)Γ∗ S)j∗ for all A, B ∈ B[0, ∞], where the final identity is a consequence of (2). S = S(A)σS(cid:0)T (B)(cid:1) Remark 3.3. (i) If the quantum stop times S and T are extended by ampliation to act on h⊗F then the identity (4) becomes for all A, B ∈ B[0, ∞]. This extension will be made when appropriate without further comment. (ii) If 0 6 p < q < ∞ and 0 6 r < s < ∞ then Theorem 2.9 implies that (S ⊗ T )(A × B) = S(A)eσS(cid:0)T (B)(cid:1) (S ⊗ T )(cid:0)(p, q] × (r, s](cid:1) = S(cid:0)(p, q](cid:1)eσS(T(cid:0)(r, s](cid:1)) = st.lim mXj=1 π S(cid:0)(πj−1, πj](cid:1)eσπj (T(cid:0)(r, s](cid:1)) ∈ Ih ⊗ B(Fq+s)) ⊗ I[q+s, where π = {p = π0 < · · · < πm = q} is a typical finite partition of the interval [p, q]. 5 Lemma 3.4. Suppose S and T are finite quantum stop times, with T discrete, so that there exists a finite set {t1 < · · · < tm} ⊆ (0, ∞) such that T(cid:0){t1, . . . , tm}(cid:1) = I. Then (S ⋆ T )(C) = for all C ∈ B(R+), mXj=1 S(cid:0)(C − tj)+(cid:1)σS(T(cid:0){tj}(cid:1)) where (C − t)+ := {s ∈ R+ : s + t ∈ C} for all t ∈ R+. Proof. Note first that, by Lemma 3.2, (S ⊗ T )(R+ × {tj}) = mXj=1 mXj=1 S(R+)σS(T(cid:0){tj}(cid:1)) = σS(T(cid:0){t1, . . . , tm}(cid:1)) = I, so = (S ⋆ T )(C) = (S ⊗ T )(cid:0){(x, y) ∈ R2 mXj=1 (S ⊗ T )(cid:0){(x, tj) ∈ R2 mXj=1 S(cid:0)(C − tj)+(cid:1)σS(T(cid:0){tj}(cid:1)). + : x + y ∈ C}(cid:1) + : x + tj ∈ C}(cid:1) = Remark 3.5. If the discrete stopping time T is supported at one point, so that T(cid:0){t}(cid:1) = I for some t ∈ (0, ∞), then S ⋆ T = S + t, where (S + t)(A) := S(cid:0)(A − t)+(cid:1) for all A ∈ B[0, ∞]. Definition 3.6. Let p ∈ B(k) be an orthogonal projection and, for all t ∈ R+, let P[t ∈ B(F[t) be the orthogonal projection such that P[tε(f ) = ε(pf ) for all f ∈ L2([t, ∞); k), where p acts pointwise. A family of bounded operators V = (Vt)t∈R+ ⊆ B(h ⊗ F) is p-adapted if Vt = Vt) ⊗ P[t for every t ∈ R+, where Vt) ∈ B(h ⊗ Ft)). If p = 0 or p = Ik then p-adaptedness is known as vacuum adaptedness or identity adaptedness, respectively. Given a p-adapted family of bounded operators V , the identity-adapted projection bV is the family of operators bV , where bVt := Vt) ⊗ I[t for all t ∈ R+. A p-adapted family of bounded operators V is an isometric cocycle if bVt is an isometry for all t ∈ R+ and for all s, t ∈ R+. A p-adapted isometric cocycle V is strongly continuous if t 7→ Vtz is continuous for all z ∈ h⊗F. Vs+t = bVseσs(Vt) Theorem 3.7. [2, Theorem 6.5, Corollary 6.6 and Theorem 7.2] If S is a finite quantum stop time, V is a strongly continuous isometric p-adapted cocycle and VS,π := n+1Xk=1 Vπk S(cid:0)(πk−1, πk](cid:1) 6 for any finite partition π = {0 = π0 < · · · < πn+1 = t} of [0, t], then VS,π is a contraction and there exists a contraction VS,t ∈ B(h ⊗ F) such that VS,π → VS,t in the strong operator topology as π is refined, for all t ∈ (0, ∞). Furthermore, there exists a contraction VS ∈ B(h ⊗ F) such that VS,t → VS in the strong operator topology as t → ∞, and for all t ∈ R+. VS+t = bVSeσS(Vt) bVS,π∩[0,t]eσS,π(Z) → bVS,teσS(Z) Proof. The only thing which not immediate is the assertion, at end of [2, Proof of Theorem 7.2], that as the partition π is refined, for all Z ∈ B(h ⊗ F) and t > 0. (In fact, a very slightly weaker claim is made.) It follows from [2, Theorem 6.5] that bVS,π∩[0,t] → bVS,t in the strong operator topology, and kbVS,π∩[0,t]k 6 1 for all π, by [2, Lemma 6.4], so the claim holds as long as eσS,π(Z) → eσS(Z) in the strong operator topology. However, this is part of Lemma 3.8. If S and T are finite quantum stop times, with T discrete, then Theorem 2.9. for any strongly continuous isometric p-adapted cocycle V . VS ⋆ T = bVSeσS(VT ) Proof. If T is as in the statement of Lemma 3.4 and t > tm then VS ⋆ T,t = st.lim π = st.lim π n+1Xk=1 Vπk (S ⋆ T )(cid:0)(πk−1, πk](cid:1) n+1Xk=1 mXj=1 Vπk S(cid:0)(πk−1 − tj, πk − tj]+(cid:1)eσS(T(cid:0){tj}(cid:1)), k =( πk − tj if πk > tj, otherwise, πj where π = {0 = π0 < · · · < πn+1 = t} and (x, y]+ = {s ∈ R+ : x < s 6 y}. For j = 1, . . . , m and k = 0, . . . , n + 1, let 0 so that πj is a partition of [0, t − tj]. Then mXj=1 n+1Xk=1 Vπk S(cid:0)(πk−1 − tj, πk − tj]+(cid:1)eσS(T(cid:0){tj}(cid:1)) = = → k+tj Vπj k−1, πj S(cid:0)(πj S(cid:0)(πj n+1Xk=1 mXj=1 k](cid:1)eσS(T(cid:0){tj}(cid:1)) k](cid:1) n+1Xl=1 n+1Xk=1bVπj mXj=1 S(cid:0)(πj mXj=1bVS,t−tjeσS(Vtj )eσS(cid:0)T ({tj})(cid:1) k−1, πj l−1, πj k 7 l ](cid:1)eσπj l (Vtj )eσS(T(cid:0){tj}(cid:1)) in the strong operator topology as π is refined; for the final identity, note that Vs+tS(cid:0)(r, s](cid:1) = bVseσs(Vt)S(cid:0)(r, s](cid:1) = bVs S(cid:0)(r, s](cid:1)eσs(Vt) Hence VS ⋆ T = st.lim t→∞ VS ⋆ T,t = st.lim t→∞ whenever 0 6 r < s < t < ∞. mXj=1bVS,t−tjeσS(Vtj T(cid:0){tj}(cid:1)) = bVSeσS(VT ). Definition 3.9 (Cf. [6, p.322]). A sequence of finite quantum stop times (Sn)n>1 is said to converge to a quantum stop time S, written Sn ⇒ S, if Sn(cid:0)[0, t](cid:1) → S(cid:0)[0, t](cid:1) in the strong operator topology for all but a countable set of points t ∈ R+. Lemma 3.10. Let V be a strongly continuous isometric p-adapted cocycle. If (Sn)n>1 is a sequence of finite quantum stop times such that Sn ⇒ S for some finite quantum stop time S then VSn → VS in the strong operator topology. Proof. The usual approximation argument shows it suffices to prove that k(VSn −VS)uε(f )k → 0 as n → ∞, where u ∈ h and f ∈ L2(R+; k) are arbitrary. From the proof of [2, Corollary 6.6], if S is any finite quantum stop time and s, t ∈ R+ are such that s 6 t then Letting t → ∞ and recalling that S(cid:0){∞}(cid:1) = 0, it follows that k(VS,t − VS,s)uε(f )k 6 kS(cid:0)(s, t](cid:1)uε(f )k. k(VS − VS,s)uε(f )k 6 kS(cid:0)(s, ∞)(cid:1)uε(f )k. Furthermore, from the proof of [2, Theorem 6.5], k(VS,π′ − VS,π)uε(f )k 6 sup{k(Vr − Vπj )uε(f )k : r ∈ [πj, πj+1], j = 0, . . . , m} kS(cid:0)[0, s](cid:1)ε(f )k, where π′ is any refinement of the partition π = {0 = π0 < · · · < πm+1 = s}; refining π′ shows that the same inequality holds with VS,π′ replaced by VS,s. Hence k(VS − VS,π)uε(f )k 6 sup{k(Vr − Vπj )uε(f )k : r ∈ [πj, πj+1], j = 0, . . . , m} kε(f )k + kS(cid:0)(s, ∞)(cid:1)uε(f )k. Now fix ε > 0, choose s ∈ R+ such that Sn(cid:0)[0, s](cid:1) → S(cid:0)[0, s](cid:1) in the strong operator topology and kS(cid:0)(s, ∞)(cid:1)uε(f )k < ε, and note that kSn(cid:0)(s, ∞)(cid:1)uε(f )k < ε for all sufficiently large n. Therefore k(VS − VSn)uε(f )k 6 k(VS − VS,π)uε(f )k + k(VSn − VSn,π)uε(f )k + k(VS,π − VSn,π)uε(f )k < 2 sup{k(Vr − Vπj )uε(f )k : r ∈ [πj, πj+1], j = 0, . . . , m} kε(f )k + 2ε + k(VS,π − VSn,π)uε(f )k < 4ε + k(VS,π − VSn,π)uε(f )k as long as π is chosen to be sufficiently fine, so that sup{k(Vr − Vπj )uε(f )k : r ∈ [πj, πj+1], j = 0, . . . , m} kε(f )k < ε. 8 Finally, if π is chosen so Sn(cid:0)[0, πj](cid:1) → S(cid:0)[0, πj ](cid:1) in the strong operator topology as n → ∞, for j = 0, . . . , m + 1, then, since k(VS,π − VSn,π)uε(f )k 6 m+1Xj=1 kVπj(cid:0)S(πj−1, πj]) − Sn(πj−1, πj](cid:1)uε(f )k → 0 as n → ∞, and ε is arbitrary, the result follows. Lemma 3.11. Let T be a finite quantum stop time, and suppose that the multiplicity space k is separable. There exists a sequence of discrete quantum stop times (Tn)n>1 such that Tn ⇒ T . Furthermore, S ⋆ Tn ⇒ S ⋆ T for any finite quantum stop time S. Proof. As is well known, a spectral measure is strongly right continuous with left limits: if x ∈ F then whereas lim s→t+ S(cid:0)[0, t](cid:1)x − S(cid:0)[0, s](cid:1)x = − lim S(cid:0)[0, t](cid:1)x − S(cid:0)[0, s](cid:1)x = lim S(cid:0)(t, s](cid:1)x = 0, S(cid:0)(s, t](cid:1)x = S(cid:0){t}(cid:1)x. s→t+ s→t− lim s→t− Now suppose {xn : n > 1} is dense in F and let DS := ∪n>1DS(xn). An ε/3 argument shows In particular, the set of discontinuities DS(x) := {t ∈ R+ : S(cid:0){t}(cid:1)x 6= 0} is countable. that t 7→ S(cid:0)[0, t](cid:1)x is continuous on R+ \ D for all x ∈ F, so DS is the set of discontinuities of S on R+. For all n > 1, let the finite partition πn = {0 = πn and max{πn n < ∞} be such that πn k−1 : k = 1, . . . , n} → 0 as n → ∞. Define a discrete quantum stop time 1 < · · · < πn 0 < πn k − πn n → ∞ Tn : B[0, ∞] → B(F); A 7→ n−1Xk=1 1A(πn k−1, πn k )T(cid:0)(πn k ](cid:1) + 1A(πn n)T(cid:0)(πn n, ∞](cid:1), and note that Tn(cid:0)[0, t](cid:1) = 0 ... T(cid:0)[0, πn 1 ](cid:1) T(cid:0)[0, πn n−1](cid:1) I if 0 6 t < πn 1 , if πn 1 6 t < πn 2 , ... if πn if πn n−1 6 t < πn n, n 6 t 6 ∞.  Thus if x ∈ F and t ∈ R+ then πn large, so k > t > πn k−1 for some k ∈ {1, . . . , n} once n is sufficiently T(cid:0)[0, t](cid:1)x − Tn(cid:0)[0, t](cid:1)x = T(cid:0)(πn k−1, t](cid:1)x → T(cid:0){t}(cid:1)x as n → ∞. This last term equals 0 if t ∈ R+ \ DT , and thus Tn ⇒ T . For the final claim, let S be a finite quantum stop time. If t ∈ R+ and f ∈ L2(R+; k) then, by [2, Corollary 3.5], In(t) := k(cid:0)(S ⋆ Tn)(cid:0)[0, t](cid:1) − (S ⋆ T )(cid:0)[0, t](cid:1)(cid:1)ε(f )k2 =Z[0,t] k(T − Tn)(cid:0)[0, t − s](cid:1)Γ∗ sε(f )k2 exp(cid:16)−Z ∞ s kf (u)k2 du(cid:17)kS(ds)ε(f )k2. 9 To prove that S ⋆ Tn ⇒ S ⋆ T , it suffices to show that In(t) → 0 as n → ∞ for all but countably many t ∈ R+, by the usual approximation argument. dominated convergence theorem gives that Now, as n → ∞, so (T − Tn)(cid:0)[0, t − s](cid:1) → T(cid:0){t − s}(cid:1), by the previous working. Thus the kf (u)k2 du(cid:17)kS(cid:0){t − r}(cid:1)ε(f )k2. In(t) → I(t) := Xr∈D∩[0,t] t−rε(f )k2 exp(cid:16)−Z ∞ kT(cid:0){r}(cid:1)Γ∗ t−r Thus I(t) = 0 whenever t 6∈ DS + DT := {s + r : s ∈ DS, r ∈ DT } and the result follows. Remark 3.12. If the multiplicity space k is not separable, the statement of [2, Corollary 3.5] requires strong measurability, not just Borel measurability, of F and G. As t 7→ Γt and t 7→ Γ∗ t are strongly continuous and t 7→ S(cid:0)[0, t](cid:1) is strongly right continuous on R+, all the subsequent proofs in [2] remain valid. Remark 3.13. It is straightforward to construct on a non-separable Hilbert space a spectral measure which has an uncountable set of discontinuities. Thus the separability hypothesis in Lemma 3.11 may not be dropped. Theorem 3.14. Let V be a strongly continuous isometric p-adapted cocycle, and suppose that the multiplicity space k is separable. If S and T are finite quantum stop times then Proof. By Lemma 3.11, there exists a sequence of discrete quantum stop times (Tn)n>1 such that Tn ⇒ T and S ⋆ Tn ⇒ S ⋆ T . Hence VS ⋆ Tn → VS ⋆ T in the strong operator topology, by VS ⋆ T = bVSeσS(VT ). Lemma 3.10. Furthermore, VS ⋆ Tn = VSeσS(VTn) for all n > 1, by Lemma 3.8, so the result follows from another application of Lemma 3.10 together with Remark 2.10, that σS is strong operator continuous on bounded sets. The next two theorems show that stopping an isometric cocycle can be used to produce a form of inner non-unital Evans -- Hudson flow. Theorem 3.15. Let V be a strongly continuous isometric identity-adapted cocycle. The map αS : B(h) → B(h ⊗ F); a 7→ VS(a ⊗ I)V ∗ S is a ∗-homomorphism for any finite quantum stop time S. Furthermore, if the multiplicity space k is separable, the identity holds for any finite quantum stop times S and T , where αS⋆T =bαS ◦eσS ◦ αT (5) Proof. Note that V ∗ bαS : B(h ⊗ F) → B(h ⊗ F); X 7→ VSXV ∗ S . S VS = Ih ⊗ I, by [2, Proposition 6.8]. Thus if a, b ∈ B(h) then αS(a)αS(b) = VS(a ⊗ I)V ∗ S VS(b ⊗ I)V ∗ S = VS(ab ⊗ I)VS = αS(ab), so αS is multiplicative. Linearity and ∗-preservation are immediate. 10 For the second claim, note that bV = V . Hence, by Theorem 3.14, if a ∈ B(h) then αS⋆T (a) = VS⋆T (a ⊗ I)V ∗ T ) V ∗ S S⋆T = VSeσS(VT )(a ⊗ I)eσS(V ∗ = VSeσS(VT (a ⊗ I)V ∗ = (bαS ◦eσS ◦ αT )(a); T ) V ∗ S the penultimate equality holds because σS is unital, soeσS(a ⊗ I) = a ⊗ I. Theorem 3.16. Let V be a strongly continuous isometric vacuum-adapted cocycle. The map βS : B(h) → B(h ⊗ F); a 7→ VS(a ⊗ ES)V ∗ S is a ∗-homomorphism for any finite quantum stop time S. Furthermore, if the multiplicity space k is separable, the identity holds for any finite quantum stop times S and T , where βS⋆T = bβS ◦eσS ◦ βT (6) bβS : B(h ⊗ F) → B(h ⊗ F); X 7→ bVSXbV ∗ S . Proof. Note that V ∗ S VS = eES, by [2, Proposition 6.7]. Thus if a, b ∈ B(h) then S = VS(ab ⊗ ES)VS = βS(ab), S VS(b ⊗ ES)V ∗ βS(a)βS (b) = VS(a ⊗ ES)V ∗ so βS is multiplicative. As above, linearity and ∗-preservation are immediate. For the second claim, let a ∈ B(h) and note that, by Theorem 3.14, βS⋆T (a) = VS⋆T (a ⊗ ES⋆T )V ∗ S S⋆T = bVSeσS(VT )(a ⊗ ES⋆T )eσS(V ∗ T )bV ∗ = bVSeσS(VT (a ⊗ ET )V ∗ T )bV ∗ = (bβS ◦eσS ◦ βT )(a); S for the penultimate equality, note that ES⋆T = σS(ET ), by [2, Theorem 5.4], which implies Remark 3.17. In the context of Theorems 3.15 and 3.16, note that αS(Ih) = VSV ∗ S = βS(Ih). S,π if V is vacuum adapted, where π is any finite partition of [0, t] and π′ is its one-point extension to a partition of [0, ∞]. immediately thateσS(a ⊗ ET ) = a ⊗ ES⋆T . The former identity is immediate, and the latter holds because VS,πeES,π′VS,π = VS,πV ∗ Remark 3.18. If the finite quantum stop time S is deterministic, so that S(cid:0){s}(cid:1) = I for some s ∈ (0, ∞), then VS = Vs and eσS = eσs. αs+t =bαs ◦eσs ◦ αt generalisation of the deterministic mapping-cocycle relation [3] It follows that (5) and (6) are the stop-time for all s, t > 0. 11 Acknowledgements This work was begun during a conference held at the Kerala School of Mathematics, Kozhikode, India; the warm hospitality and stimulating atmosphere provided by the organisers is gratefully acknowledged. It continued during visits of the first author to the Indian Statistical Institute, Kolkata, and the Jawaharlal Nehru Centre for Advanced Scientific Research, Bangalore, with travel supported by the ANCM project of the Indian Statistical Institute. The second author gratefully acknowledges support by a grant from the SERB-Distinguished Fellowship of the Department of Science and Technology, Government of India. References [1] D. Applebaum, Stopping unitary processes in Fock space, Publ. RIMS Kyoto Univ. 24 (1988), no. 5, 697 -- 705. [2] A.C.R. Belton & K.B. Sinha, Stopping the CCR flow and its isometric cocycles, Q. J. Math. 65 (2014), no. 4, 1145 -- 1164. [3] W.S. Bradshaw, Stochastic cocycles as a characterisation of quantum flows, Bull. Sci. Math. 116 (1992), no. 1, 1 -- 34. [4] R.L. Hudson, The strong Markov property for canonical Wiener processes, J. Funct. Anal. 34 (1979), no. 2, 266 -- 281. [5] R.L. Hudson, Stop times in Fock space quantum probability, Stochastics 79 (2007), no. 3 -- 4, 383 -- 391. [6] K.R. Parthasarathy & K.B. Sinha, Stop times in Fock space stochastic calculus, Probab. Theory Related Fields 75 (1987), no. 3, 317 -- 349. [7] M.A. Pinsky, Stochastic integral representation of multiplicative operator functionals of a Wiener process, Trans. Amer. Math. Soc. 167 (1972), 89 -- 104. 12
1101.4210	3	1101	2012-08-27T11:18:51	Endomorphisms of graph algebras	[ "math.OA", "math.DS", "math.FA", "math.GR" ]	We initiate a systematic investigation of endomorphisms of graph C-algebras C(E), extending several known results on endomorphisms of the Cuntz algebras O_n. Most but not all of this study is focused on endomorphisms which permute the vertex projections and globally preserve the diagonal MASA D_E of C(E). Our results pertain both automorphisms and proper endomorphisms. Firstly, the Weyl group and the restricted Weyl group of a graph C-algebra are introduced and investigated. In particular, criteria of outerness for automorphisms in the restricted Weyl group are found. We also show that the restriction to the diagonal MASA of an automorphism which globally preserves both the diagonal and the core AF-subalgebra eventually commutes with the corresponding one-sided shift. Secondly, we exhibit several properties of proper endomorphisms, investigate invertibility of localized endomorphisms both on C*(E) and in restriction to D_E, and develop a combinatorial approach to analysis of permutative endomorphisms.	math.OA	math	Endomorphisms of Graph Algebras Roberto Conti∗, Jeong Hee Hong† and Wojciech Szyma´nski*‡ 19 May, 2012 (revised 18 August, 2012) Abstract We initiate a systematic investigation of endomorphisms of graph C ∗-algebras C ∗(E), extending several known results on endomorphisms of the Cuntz algebras On. Most but not all of this study is focused on endomorphisms which permute the vertex projections and globally preserve the diagonal MASA DE of C ∗(E). Our results pertain both automorphisms and proper endomorphisms. Firstly, the Weyl group and the restricted Weyl group of a graph C ∗-algebra are introduced and investigated. In particular, criteria of outerness for automorphisms in the restricted Weyl group are found. We also show that the restriction to the diagonal MASA of an automorphism which globally preserves both DE and the core AF- subalgebra eventually commutes with the corresponding one-sided shift. Secondly, we exhibit several properties of proper endomorphisms, investigate invertibility of localized endomorphisms both on C ∗(E) and in restriction to DE, and develop a combinatorial approach to analysis of permutative endomorphisms. MSC 2010: 46L40, 46L55, 46L05 Keywords: Cuntz-Krieger algebra, graph algebra, endomorphism, automorphism, per- mutative endomorphism, AF-subalgebra, MASA, subshift ∗This research was supported through the programme "Research in Pairs" by the Mathematisches Forschungsinstitut Oberwolfach in 2011. †This work was supported by National Research Foundation of Korea Grant funded by the Korean Government (KRF -- 2008 -- 313-C00039). ‡This work was supported by: the FNU Rammebevilling 'Operator algebras and applications' (2009 -- 2011), the Marie Curie Research Training Network MRTN-CT-2006-031962 EU-NCG, the NordForsk Research Network 'Operator algebra and dynamics' (grant #11580), the FNU Forskningsprojekt 'Struc- ture and Symmetry' (2010 -- 2012), and the EPSRC Grant EP/I002316/1. 1 1 Introduction The main aim of this article is to carry out systematic investigations of a certain natural class of endomorphisms and, in particular, automorphisms of graph C ∗-algebras C ∗(E). Namely, we focus on those endomorphisms which globally preserve the diagonal MASA DE. This leads to the concept of the Weyl group of a graph algebra, arising from the normalizer of a maximal abelian subgroup of the automorphism group of C ∗(E), con- sisting of those automorphisms which fix the diagonal DE point-wise. We investigate in depth an important subgroup of the Weyl group corresponding to those automorphisms which globally preserve the core AF-subalgebra FE as well -- the restricted Weyl group of C ∗(E). We develop powerful novel techniques of both analytic and combinatorial nature for the study of automorphisms these groups comprise. By analogy with the theory of semi-simple Lie groups, Cuntz introduced in [18] the Weyl group of the simple, purely infinite C ∗-algebras On. Quite similarly with the classical theory, it arises as the quotient of the normalizer of a maximal abelian subgroup of the automorphism group of the algebra. So defined Weyl group is discrete albeit infinite, and the abelian subgroup in question is an inductive limit of higher dimensional tori. Cuntz posed a problem of determining the structure of the important subgroup of the Weyl group, corresponding to those automorphisms in the Weyl group which globally preserve the canonical UHF-subalgebra of On, [18]. After 30 years, this question has been finally answered in [9]. In the present paper, we take this programme one step further, expanding it from Cuntz algebras On to a much wider class of graph C ∗-algebras. The theory of graph C ∗-algebras began in earnest in the late nineties, [33, 32], and since then it has developed into a fully fledged and very active area of research within op- erator algebras. For a good general introduction to graph C ∗-algebras we refer the reader to [39]. In the case of finite graphs, the corresponding C ∗-algebras essentially coincide with Cuntz-Krieger algebras, that were introduced much earlier, [19], in connection with topological Markov chains. The importance of graph algebras (or Cuntz-Krieger alge- bras) stems to a large extent from numerous applications they have found. Not trying to be exhaustive in any way, we only mention: their role in classification of purely infinite, simple C ∗-algebras, [40, 43], and related to that applications to the problem of semipro- jectivity of Kirchberg algebras, [3, 44, 42]; their connection with objects of interest in noncommutative geometry and quantum group theory, [24, 7, 36]; their strong interplay with theory of symbolic dynamical systems, going back to the original paper of Cuntz and Krieger, [19, 4]. It is also worth mentioning that graph C ∗-algebras (similarly to Cuntz algebras) in many ways behave as purely combinatorial objects, and are thus strongly connected to their purely algebraic counterparts, the Leavitt path algebras, [1]. Therefore we believe that many of the results of the present paper are applicable to the latter algebras as well. The analysis of endomorphisms of graph algebras, developed in the present article, owes a great deal to the close relationship between such algebras and the Cuntz algebras On. The C ∗-algebras On were first defined and investigated by Cuntz in his seminal 2 paper [17], and they bear his name ever since. The Cuntz algebras have been extensively used in many a diverse contexts, including classification of C ∗-algebras, quantum field theory, self-similar sets, wavelet theory, coding theory, spectral flow, subfactors and index theory, among others. Systematic investigations of endomorphisms of On, 2 ≤ n < ∞, were initiated by Cuntz in [18]. A fundamental bijective correspondence λu ↔ u between unital ∗- endomorphisms and unitaries in On was established therein. Using this correspondence, Cuntz proved a number of interesting results, in particular with regard to those endo- morphisms which globally preserve either the core UHF-subalgebra Fn or the diagonal MASA Dn. Likewise, investigations of automorphisms of On began almost immediately after the birth of the algebras in question, [18]. Endomorphisms of the Cuntz algebras have played a role in certain aspects of index theory, both from the C ∗-algebraic and von Neumann algebraic point of view, e.g. see [27], [29] and [13]. One of the most interesting applications of endomorphisms of On, found by Bratteli and Jørgensen in [5, 6], is in the area of wavelets. In particular, permutative endomorphisms have been used in this context. These were further investigated by Kawamura, [31]. The present article builds directly on the progress made recently in the study of localized endomorphisms of On by Conti, Szyma´nski and their collaborators. In partic- ular, much better understanding of those endomorphisms of On which globally preserve the core UHF-subalgebra or the canonical diagonal MASA has been obtained in [14] and [23], respectively. In [45, 15, 12, 8, 10], a novel combinatorial approach to the study of permutative endomorphisms of On has been introduced, and subsequently significant progress in the investigations of such endomorphisms and automorphisms has been ob- tained. In particular, a striking relationship between permutative automorphisms of On and automorphisms of the full two-sided n-shift has been found in [9]. In the present article, we extend this analysis of endomorphisms of the Cuntz alge- bras to the much larger class of graph C ∗-algebras. Most of our results (but not all) are concerned with algebras corresponding to finite graphs without sinks, which may be identified with Cuntz-Krieger algebras of finite 0 -- 1 matrices, [19]. From the very be- ginning, the theory of such algebras has been closely related to dynamical systems. In particular, endomorphisms of Cuntz-Krieger algebras have been studied in the context of index theory, [28]. Quasi-free automorphisms of Cuntz-Krieger algebras (and even more generally, Cuntz-Pimsner algebras) have been studied in [30], [46] and [21]. An interesting connection between automorphisms of Cuntz-Krieger algebras and Markov shifts has been investigated by Matsumoto, [34, 35]. The present paper is organized as follows. In Section 2, we set up notation and present some preliminaries. In particular, we introduce a class of endomorphisms {λu : u ∈ UE} of a graph C ∗-algebra C ∗(E) corresponding to unitaries commuting with the vertex projections (whose set we denote UE). In Section 3, an analogue of the Weyl group for graph C ∗-algebras is introduced and investigated. Namely, let DE be the canonical abelian subalgebra of C ∗(E). Then, under a mild hypothesis, the group AutDE (C ∗(E)) of those automorphisms of C ∗(E) which fix DE point-wise is a maximal abelian subgroup of Aut(C ∗(E)) (Propositions 3 3.2 and 3.3). The Weyl group WE of C ∗(E) is defined as the quotient of the normalizer of AutDE (C ∗(E)) by itself. Then we exhibit several structural properties of the Weyl group, analogous to those discovered by Cuntz in the case of On, [18]. In particular, the Weyl group of C ∗(E) is countable and discrete (Proposition 3.5). In Section 4, we investigate an important subgroup of the Weyl group corresponding to those automorphisms which globally preserve both the diagonal DE and the core AF-subalgebra FE, the restricted Weyl group RWE of C ∗(E). We prove certain facts about the normalizer of FE (Theorem 4.2) and give a characterization of automorphisms globally preserving FE (Lemma 4.5). We obtain a very convenient criterion of outerness for a large class of such automorphisms (Corollary 4.7). Then we take a closer look at the action on the diagonal DE induced by the restriction of those automorphisms of the graph algebra which preserve both the diagonal and the core AF-subalgebra. By earlier results of Cuntz-Krieger and Matsumoto, [19, 34, 35], it is known that all shift commuting automorphisms of the diagonal DE extend to automorphisms of C ∗(E). In the case of the Cuntz algebra, it was shown in [9] that all automorphisms of the diagonal which (along with their inverses) eventually commute with the shift can be extended to automorphisms of On. In the present paper, we show that the automorphisms of DE arising from restrictions of AF-subalgebra FE preserving automorphisms of C ∗(E) eventually commute with the shift (Theorem 4.13). This provides a solid starting point for a future complete determination of the restricted Weyl group RWE of a graph C ∗- algebra C ∗(E). In Section 5, we study localized endomorphisms of a graph algebra C ∗(E), that is endomorphisms λu corresponding to unitaries u from the algebraic part of the core AF-subalgebra which commute with the vertex projections. We obtain an algorithmic criterion of invertibility of such localized endomorphisms (Theorem 5.1), as well as a criterion of invertibility of the restriction of a localized endomorphism to the diagonal MASA (Theorem 5.3). These theorems extend the analogous results for Cuntz algebras obtained earlier in [15]. Localized endomorphisms constitute the best understood and most studied class of endomorphisms of the Cuntz algebras, and it is natural to begin systematic investigations of endomorphisms of graph C ∗-algebras from them. In Section 6, a special class of localized endomorphisms corresponding to permu- tation unitaries is investigated. Automorphisms of C ∗(E) of this type give rise to a large and interesting subgroup of the restricted Weyl group RWE. For a permutative endomorphism λu of a graph algebra C ∗(E), combinatorial criteria are given for λu to be an automorphism of C ∗(E) or its restriction to be an automorphism of the diagonal DE (Lemma 6.1, Lemma 6.3 and Theorem 6.4). In this way, we obtain a far-reaching generalization of the techniques developed by Conti and Szyma´nski in [15] for dealing with permutative endomorphisms of the Cuntz algebras. A few examples are worked out in detail, illustrating applications of the combinatorial machinery developed in the present paper. Acknowledgement. We are grateful to the referee for very careful reading of the manuscript and a number of comments which helped to improve the presentation. 4 2 Notation and preliminaries 2.1 Directed graphs and their C ∗-algebras Let E = (E0, E1, r, s) be a directed graph, where E0 and E1 are (countable) sets of ver- tices and edges, respectively, and r, s : E1 → E0 are range and source maps, respectively. The C ∗-algebra C ∗(E) corresponding to a graph E is by definition the universal C ∗- algebra generated by mutually orthogonal projections Pv, v ∈ E0, and partial isometries Se, e ∈ E1, subject to the following relations (GA1) S∗ e Se = Pr(e) and S∗ e Sf = 0 if e 6= f ∈ E1, (GA2) SeS∗ e ≤ Ps(e) for e ∈ E1, e if v ∈ E0 emits finitely many and at least one edge. It follows from the above definition that a graph C ∗-algebra C ∗(E) is unital if and only (GA3) Pv =Ps(e)=v SeS∗ if the underlying graph E has finitely many vertices, in which case 1 =Pv∈E 0 Pv. A path µ of length µ = k ≥ 1 is a sequence µ = (µ1, . . . , µk) of k edges µj such that r(µj) = s(µj+1) for j = 1, . . . , k − 1. We also view the vertices as paths of length 0. The set of all paths of length k is denoted Ek. The range and source maps naturally extend from edges E1 to paths Ek. A sink is a vertex v which emits no edges, i.e. s−1(v) = ∅. A source is a vertex w which receives no edges, i.e. r−1(v) = ∅. By a loop we mean a path µ of length µ ≥ 1 such that s(µ) = r(µ). We say that a loop µ = (µ1, . . . , µk) has an exit if there is a j such that s(µj) emits at least two distinct edges. As usual, for a path µ = (µ1, . . . , µk) of length k we denote by Sµ = Sµ1 · · · Sµk the corresponding partial isometry in C ∗(E). It is known that each Sµ is non-zero, with the domain projection Pr(µ). Then C ∗(E) is the closed span of {SµS∗ ν : µ, ν ∈ E∗}, where E∗ denotes the collection of all finite paths (including paths of length zero). Here we agree that Sv = Pv for v ∈ E0 viewed as path of length 0. Also note that SµS∗ ν is non-zero if and only if r(µ) = r(ν). In that case, SµS∗ ν is a partial isometry with domain and range projections equal to SνS∗ µ, respectively. ν and SµS∗ The range projections Pµ = SµS∗ µ of all partial isometries Sµ mutually commute, and the abelian C ∗-subalgebra of C ∗(E) generated by all of them is called the diagonal subalgebra and denoted DE. If E does not contain sinks and all loops have exits then DE is a MASA (maximal abelian subalgebra) in C ∗(E) by [25, Theorem 5.2]. We set D0 E = span{Pv : v ∈ E0} and, more generally, Dk E = span{Pµ : µ ∈ Ek} for k ≥ 0. There exists a strongly continuous action γ of the circle group U(1) on C ∗(E), called the gauge action, such that γt(Se) = tSe and γt(Pv) = Pv for all e ∈ E1, v ∈ E0 and t ∈ U(1) ⊆ C. The fixed-point algebra C ∗(E)γ for the gauge action is an AF- algebra, denoted FE and called the core AF-subalgebra of C ∗(E). FE is the closed span of {SµS∗ E the linear span of {SµS∗ µ = ν = k}. For an integer m ∈ Z we denote by C ∗(E)(m) the spectral subspace of the gauge action corresponding to m. That is, C ∗(E)(m) := {x ∈ C ∗(E) : γz(x) = zmx, ∀z ∈ U(1)}. In particular, C ∗(E)(0) = C ∗(E)γ. ν : µ, ν ∈ E∗, µ = ν}. For k ∈ N = {0, 1, 2, . . .} we denote by F k ν : µ, ν ∈ E∗, 5 2.2 Endomorphisms determined by unitaries We denote by UE the collection of all those unitaries in the multiplier algebra M(C ∗(E)) which commute with all vertex projections Pv, v ∈ E0. That is UE := U((D0 E)′ ∩ M(C ∗(E))). (1) These unitaries will play a crucial role throughout this paper. Let u ∈ UE. Then uSe, e ∈ E1, are partial isometries in C ∗(E) which together with projections Pv, v ∈ E0, satisfy (GA1) -- (GA3). Thus, by universality in the definition of C ∗(E), there exists a ∗-homomorphism λu : C ∗(E) → C ∗(E) such that1 λu(Se) = uSe and λu(Pv) = Pv, for e ∈ E1, v ∈ E0. (2) Clearly, λu(1) = 1 whenever C ∗(E) is unital. In general, λu may be neither injective nor surjective. However, the following proposition is an immediate consequence of the gauge-invariant uniqueness theorem [2, Theorem 2.1]. Proposition 2.1. If u ∈ UE belongs to the minimal unitization of the core AF-subalgebra FE, then endomorphism λu is automatically injective. Note that {λu : u ∈ UE} is a semigroup with the following multiplication law: λu ◦ λw = λu∗w, u ∗ w = λu(w)u. (3) We say λu is invertible if λu is an automorphism of C ∗(E). For K ⊆ UE we denote λ(K)−1 := {λu ∈ Aut(C ∗(E)) : u ∈ K}. It turns out that λu is invertible if and only if it is injective and u∗ is in the range of its unique strictly continuous extension to M(C ∗(E)), still denoted λu. (Such an extension exists since λu fixes each vertex projection and thus it fixes an approximate unit comprised of finite sums of vertex Indeed, if λu is injective and there exists a w ∈ M(C ∗(E)) such that projections.) λu(w) = u∗ then w must belong to UE and λuλw = λu∗w = id. Thus λu is surjective and hence an automorphism. In this case, we have λ−1 u = λw. In the present paper, we mainly deal with finite graphs without sinks. If E is such a graph then the mapping u 7→ λu establishes a bijective correspondence between UE and the semigroup of those unital ∗-homomorphisms from C ∗(E) into itself which fix all the vertex projections Pv, v ∈ E0. if ρ is such a homomorphism then E ∩ UE then λu is e belongs to UE and ρ = λu (cf. Indeed, [46]). If u ∈ F 1 u := Pe∈E 1 ρ(Se)S∗ automatically invertible with inverse λu∗ and the map F 1 E ∩ UE ∋ u 7→ λu ∈ Aut(C ∗(E)) (4) is a group homomorphism with range inside the subgroup of quasi-free automorphisms of C ∗(E), see [21, 46]. 1The reader should be aware that in some papers (e.g. in [18], [45] and [15]) a different convention: λu(Se) = u∗Se is used. 6 If λu is an endomorphism of C ∗(E) corresponding to a unitary u in the linear span ν : µ, ν ∈ E∗, µ = ν} and the identity (i.e. in the minimal unitization of the of {SµS∗ algebraic part of the core AF-subalgebra), then we call λu localized, cf. [13]. Let E be a finite graph without sinks, and let ϕ(x) = Xe∈E 1 SexS∗ e (5) be the usual shift on C ∗(E), [19]. It is a unital, completely positive map. One can easily verify that the shift is an injective ∗-homomorphism when restricted to the relative commutant of {Pv : v ∈ E0}. We will denote this relative commutant by BE, i.e. BE := (D0 E)′ ∩ C ∗(E). (6) In particular, UE = U(BE). We have ϕ(BE) ⊆ BE and thus ϕ(UE) ⊆ UE. It is also clear that ϕ(FE) ⊆ FE and ϕ(DE) ⊆ DE. For k ≥ 1 we denote uk := uϕ(u) · · · ϕk−1(u), (7) and agree that u∗ we have k stands for (uk)∗. For each u ∈ UE and for any two paths µ, ν ∈ E∗ λu(SµS∗ ν ) = uµSµS∗ ν u∗ ν. (8) The above equality is established with help of the identity Seu = ϕ(u)Se, which holds for all e ∈ E1 because the unitary u commutes with all vertex projections Pv, v ∈ E0, by hypothesis. Indeed, ϕ(u)Se =(cid:16)Xf ∈E 1 Sf uS∗ f(cid:17)Se = SeuS∗ e Se = SeuPr(e) = SePr(e)u = Seu. More generally, if x ∈ C ∗(E) commutes with all vertex projections then for each finite path α. Furthermore, we have Sαx = ϕα(x)Sα Ad(u) = λuϕ(u∗) for u ∈ UE, (9) (10) where Ad(u)(x) = uxu∗, x ∈ C ∗(E). 3 The Weyl group For algebras A ⊆ B (with a common identity element) we denote by NB(A) = {u ∈ U(B) : uAu∗ = A} the normalizer of A in B, and by A′∩B = {b ∈ B : (∀a ∈ A) ab = ba} the relative commutant of A in B. Proposition 3.1. Let E be a finite graph without sinks, and let u ∈ UE. Then the following hold: 7 (1) If uDEu∗ ⊆ DE then λu(DE) ⊆ DE. (2) If λu(DE) = DE then uDEu∗ ⊆ DE. Proof. Part (1) is established in the same way as the analogous statement about On in [18]. For part (2), first note that DE is a C ∗-algebra generated by {ϕk(SeS∗ 0, 1, 2, ...}, where ϕ0 = id. In particular, DE is generated by D1 UE and λu(DE) = DE then uD1 λu(ϕ(DE)) ⊆ DE. Hence uDEu∗ ⊆ DE. Eu∗ = λu(D1 e ) : e ∈ E1, k = E and ϕ(DE). Now, if u ∈ E) ⊆ DE and uϕ(DE)u∗ = uϕ(λu(DE))u∗ = Further note that if E is a finite graph without sinks in which every loop has an exit then DE is a MASA in C ∗(E) and in part (2) of Proposition 3.1 we can further conclude that uDEu∗ = DE, i.e. that u ∈ NC ∗(E)(DE). In that case, by virtue of [25, Theorem 10.1], every u ∈ NC ∗(E)(DE) can be uniquely written as u = dw, where d ∈ U(DE) and w ∈ SE. Here SE denotes the group of all unitaries in U(C ∗(E)) of the form P SαS∗ β (finite sum). Therefore, one can show, as in [15, Section 2] that the normalizer of the diagonal in C ∗(E) is a semi-direct product NC ∗(E)(DE) = U(DE) ⋊ SE. (11) For C ∗-algebras A ⊆ B (with a common identity), we denote by Aut(B, A) the collection of all those automorphisms α of B such that α(A) = A, and by AutA(B) those automorphisms of B which fix A point-wise. Similarly, EndA(B) denotes the collection of all unital ∗-homomorphisms of B which fix A point-wise. Proposition 3.2. Let E be a finite graph without sinks in which every loop has an exit. Then the mapping u 7→ λu establishes a group isomorphism EndDE (C ∗(E)) = AutDE (C ∗(E)) ∼= U(DE). Proof. It follows immediately from formula (3) that the mapping u 7→ λu is a group homomorphism from U(DE) into Aut(C ∗(E)). If µ is a finite path then λu(Pµ) = Ad(uµ)(Pµ) by (8). Hence, if u ∈ DE, then λu fixes DE point-wise. Consequently, the mapping u 7→ λu is a well-defined group homomorphism into AutDE (C ∗(E)). The map is one-to-one, as already noted in Section 2. To see that the map is onto EndDE (C ∗(E)), recall (again from Section 2) that every endomorphism fixing the vertex projections is of the form λw for some w ∈ UE. Since λw fixes DE point-wise, proceeding by induction on µ one shows that w commutes with all projections Pµ. Thus w ∈ U(DE), since DE is a MASA in C ∗(E) by [25, Theorem 5.2]. We note that under the hypothesis of Proposition 3.2 the fixed-point algebra for the action of AutDE (C ∗(E)) on C ∗(E) equals DE. Indeed, each element in this fixed-point algebra is also fixed by all Ad(u), u ∈ U(DE), and thus belongs to DE. For the proof of the following Proposition 3.3, we note the following simple fact. If E is a finite graph without sinks in which every loop has an exit then DE does not contain 8 minimal projections. Indeed, if p is a non-zero projection in DE then there exists a path α such that Pα ≤ p. Extend α to a path ending at a loop, and then further to a path β which ends at a vertex emitting at least two edges, say e and f . Then for the path µ = βe we have Pµ (cid:12) Pα ≤ p. Proposition 3.3. Let E be a finite graph without sinks in which every loop has an exit. Then the normalizer of AutDE (C ∗(E)) in Aut(C ∗(E)) coincides with Aut(C ∗(E), DE). If, in addition, the center of C ∗(E) is trivial, then AutDE (C ∗(E)) is a maximal abelian subgroup of Aut(C ∗(E)). Proof. Let α ∈ Aut(C ∗(E)) be in the normalizer of AutDE (C ∗(E)). Thus for each w ∈ U(DE) there is a u ∈ U(DE) such that αλu = λwα. Then α(d) = αλu(d) = λwα(d) for all d ∈ DE. Whence α(DE) ⊆ DE. Replacing α with α−1 we get the reverse inclusion, and thus α(DE) = DE. This proves the first part of the proposition. Now let α ∈ Aut(C ∗(E)) commute with all elements of AutDE (C ∗(E)). Then, in particular, Ad(u)α(x) = α Ad(u)(x) for all u ∈ U(DE) and x ∈ C ∗(E). Thus u∗α(u) belongs to the center of C ∗(E) and hence it is a scalar. Therefore for each projection p ∈ DE we have α(2p − 1) = ±(2p − 1), and hence α(p) equals either p or 1 − p. The case α(p) = 1 −p is impossible, since taking a projection 0 6= q (cid:12) p we would get α(q) ≤ 1 −p and thus α(q) 6∈ {q, 1 − q}. Hence α fixes all projections, and thus all elements of DE. The claim now follows from Proposition 3.2. The quotient of Aut(C ∗(E), DE) by AutDE (C ∗(E)) will be called the Weyl group of C ∗(E) (cf. [18]), and denoted WE. That is, WE := Aut(C ∗(E), DE)/AutDE (C ∗(E)). (12) Remark 3.4. Since triviality of the center of C ∗(E) plays a role in our considerations, it is worthwile to mention the following. If E is a finite graph without sinks in which every loop has an exit then the following conditions are equivalent. (i) The center of C ∗(E) is trivial. (ii) C ∗(E) is indecomposable. That is, there is no decomposition C ∗(E) ∼= A ⊕ B into direct sum of two non-zero C ∗-algebras. (iii) There are no two non-empty, disjoint, hereditary and saturated subsets X and Y of E0 such that: for each vertex v ∈ E0 \ (X ∪ Y ) there is no loop through v, and there exist paths from v to both X and Y . Indeed, it is shown in [22] that conditions (ii) and (iii) above are equivalent. Obviously, (i) implies (ii). To see that (ii) implies (i), we first note that in the present case DE is a MASA, [25], and thus it contains the center Z(C ∗(E)) of C ∗(E). Thus, the (closed, two-sided) ideal of C ∗(E) generated by any element of Z(C ∗(E)) is gauge-invariant. But for a finite graph E, C ∗(E) contains only finitely many gauge-invariant ideals, [26] (see also [2] for the complete description of gauge-invariant ideals for arbitrary graphs). Hence Z(C ∗(E)) is finite dimensional. Thus if Z(C ∗(E)) is non-trivial then it contains a non-trivial projection and consequently C ∗(E) is decomposable. 9 Just as in the case of the Cuntz algebras, the Weyl group of a graph algebra turns out to be countable. Proposition 3.5. Let E be a finite graph. Then the Weyl group WE is countable. Proof. For each coset in the quotient Aut(C ∗(E), DE)/AutDE (C ∗(E)) choose a represen- tative α and define a mapping from Aut(C ∗(E), DE)/AutDE (C ∗(E)) to ⊕E 1+E 0C ∗(E) by α 7→ ⊕eα(Se)⊕v α(Pv). This mapping is one-to-one and the target space is separable. Thus it suffices to show that its image is a discrete subset of ⊕E 1+E 0C ∗(E). Let α ∈ Aut(C ∗(E), DE) be such that α(x) − x < 1/2 for all x ∈ {Se : e ∈ E1} ∪ {Pv : v ∈ E0}. We claim that αDE = id. To this end, we show by induction on µ that α(Pµ) = Pµ for each path µ. Indeed, if v ∈ E0 then α(Pv) − Pv < 1/2 and thus α(Pv) = Pv since α(Pv) ∈ DE and DE is commutative. This establishes the base for induction. Now suppose that α(Pµ) = Pµ for all paths µ of length k. Let (e, µ) be a path of length k + 1. Then, by the inductive hypothesis, we have α(P(e,µ)) − P(e,µ) = α(Se)Pµα(S∗ e ) − SePµS∗ e ≤ α(Se)Pµα(S∗ e ) − α(Se)PµS∗ e + α(Se)PµS∗ e − SePµS∗ e < 1 2 + 1 2 = 1. Thus α(P(e,µ)) = P(e,µ), since α(P(e,µ)) and P(e,µ) commute. This yields the inductive step. Now suppose that α, β ∈ Aut(C ∗(E), DE) are such that α(x) − β(x) < 1/2 for all x ∈ {Se : e ∈ E1} ∪ {Pv : v ∈ E0}. Then also β−1α(x) − x < 1/2 for all such x, and hence β−1α ∈ AutDE (C ∗(E)) by the preceding paragraph. This completes the proof. For the remainder of this section, we assume that E is a finite graph without sinks in which every loop has an exit. Clearly, each automorphism α of the graph E gives rise to an automorphism of the algebra C ∗(E), still denoted α, such that α(Se) = Sα(e), e ∈ E1, and α(Pv) = Pα(v), v ∈ E0. With a slight abuse of notation, we are identifying automorphisms of graph E with the corresponding automorphisms of C ∗-algebra C ∗(E), and denote this group ΓE. We denote by GE the subgroup of Aut(C ∗(E)) generated by automorphisms of the graph E and λ(SE ∩ UE)−1. That is, GE := hλ(SE ∩ UE)−1 ∪ ΓEi ⊆ Aut(C ∗(E)). (13) If u ∈ SE ∩ UE then λu(DE) ⊆ DE. Consequently, if u ∈ SE ∩ UE and λu is in- vertible then λu belongs to Aut(C ∗(E), DE), since DE is a MASA in C ∗(E). Since each graph automorphism gives rise to an element of Aut(C ∗(E), DE) as well, we have GE ⊆ Aut(C ∗(E), DE). We would like to stress that the class of automorphisms com- prising GE can be viewed as 'purely combinatorial' transformations, which facilitates their algorithmic analysis. Now, let u ∈ SE ∩ UE be such that λu is invertible. Then, as noted in Section 2 u = λw for some w ∈ UE and w ∈ NC ∗(E)(DE) by Proposition 3.1. Let w = dz above, λ−1 10 with d ∈ U(DE) and z ∈ SE. Then we have z ∈ SE ∩UE. Since λu∗dz = λuλdz = id = λ1, we have λu(d)λu(z)u = u ∗ dz = 1. Thus d = 1, by (11), and λ−1 u = λz. This shows that λ(SE∩UE)−1 is a group. If α is an automorphism of E and u ∈ SE∩UE then αλuα−1 = λw for some w ∈ UE, since αλuα−1 fixes all the vertex projections. A short calculation shows that this w belongs to SE. It follows that λ(SE ∩ UE)−1 is a normal subgroup of GE. Let Γ0 E be the normal subgroup of ΓE consisting of those automorphisms which fix all vertex projections. It is easy to verify that Γ0 E ∩ SE. We have E ∩ UE), where we denote P 1 E = λ(P 1 E = F 1 GE = λ(SE ∩ UE)−1ΓE and λ(SE ∩ UE)−1 ∩ ΓE = Γ0 E. (14) Proposition 3.6. Let E be a finite graph without sinks in which every loop has an exit. Then there is a natural embedding of GE into the Weyl group WE of C ∗(E). Proof. Since GE ⊆ Aut(C ∗(E), DE), it suffices to show that GE ∩AutDE (C ∗(E)) = {id}. Indeed, if β ∈ GE then β = λuα for some u ∈ SE ∩ UE and α ∈ Aut(E). If, in addition, β ∈ AutDE (C ∗(E)) then α ∈ Γ0 E ∩ UE. Therefore β = λuλw = λu∗w and u ∗ w ∈ SE. By Proposition 3.2 we also have β = λd for some d ∈ U(DE). Consequently u ∗ w = d = 1, since SE ∩ U(DE) = {1}, and thus β = id. E and thus α = λw for some w ∈ P 1 One of the more difficult issues arising in dealing with automorphisms of graph C ∗-algebras is deciding if the automorphism at hand is outer or inner. The following theorem provides a criterion useful for certain automorphisms belonging to the subgroup GE of the Weyl group WE, see Corollary 4.7 below. Theorem 3.7. Let E be a finite graph without sinks in which every loop has an exit. Then GE ∩ Inn(C ∗(E)) ⊆ {Ad(w) : w ∈ SE}. Proof. By (14), each element of GE is of the form λuα, with u ∈ SE ∩ UE and α ∈ ΓE. If such an automorphism is inner and equals Ad(y) for some y ∈ U(C ∗(E)) then y ∈ NC ∗(E)(DE), and thus we have λuα = Ad(d) Ad(w) for some d ∈ U(DE) and w ∈ SE, e ∈ SE. Since SE ∩ U(DE) = {1}, we have dϕ(d∗) = 1. Therefore d = ϕ(d) and this implies (via a straightforward calculation) that d belongs to the center of C ∗(E). Hence Ad(u) = id and thus λuα = Ad(w), as required. by (11). Hence ρ := λuα Ad(w∗) = λdϕ(d∗). But then dϕ(d∗) = Pe∈E 1 ρ(Se)S∗ 4 The restricted Weyl group In order to define and study the restricted Weyl group of a graph C ∗-algebra, we need some preparation on endomorphisms which globally preserve its core AF-subalgebra. Recall that γ : U(1) → Aut(C ∗(E)) is the gauge action, such that γz(Se) = zSe for all e ∈ E1 and z ∈ U(1). Then FE is the fixed-point algebra for action γ. The same argument as in Proposition 3.1 yields the following. Proposition 4.1. Let E be a finite graph without sinks, and let u ∈ UE. Then the following hold: 11 (1) If uFEu∗ ⊆ FE then λu(FE) ⊆ FE. (2) If λu(FE) = FE then uFEu∗ ⊆ FE. It turns out that in many instances the normalizer of the core AF-subalgebra FE of C ∗(E) is trivial. The following theorem is essentially due to Mikael Rørdam, [41]. Theorem 4.2. Let E be a directed graph with finitely many vertices and no sources. Sup- pose further that the relative commutant of FE in C ∗(E) is trivial. Then u ∈ U(C ∗(E)) and uFEu∗ ⊆ FE imply that u ∈ U(FE). Proof. If x ∈ FE then uxu∗ ∈ FE and thus γt(u)xγt(u)∗ = γt(uxu∗) = uxu∗ for each t ∈ U(1). Consequently, u∗γt(u) belongs to F ′ E ∩ C ∗(E) = C1. Thus, for each t ∈ U(1) there exists a z(t) ∈ C such that γt(u) = z(t)u. Clearly, t 7→ z(t) is a continuous character of the circle group U(1). Hence there is an integer m such that z(t) = tm. If m = 0 then u is invariant under the gauge action and we are done. Otherwise, by passing to u∗ if necessary, we may assume that m > 0. For each vertex v ∈ E0 choose one edge ev with range v and let T = Pv∈E 0 Sev. We have γt(T ) = tT for all t ∈ U(1) and T is an isometry, since E has no sources. Furthermore, T T ∗ 6= 1, for otherwise each vertex would emit exactly one edge. As E0 is finite and there are no sources, this would entail existence of a loop disjoint from the rest of the graph, and consequently F ′ E ∩ C ∗(E) would contain the non-trivial center of C ∗(E), contrary to the assumptions. Now T mu∗ is fixed by the gauge action and thus it is an isometry in FE. Since FE is an AF-algebra, T mu∗ must be unitary. But then T m and thus T itself would be unitary, a contradiction. This completes the proof. Remark 4.3. Several results of this paper depend on triviality of the center Z(C ∗(E)) of the graph algebra C ∗(E), the center Z(FE) of the core AF-subalgebra FE, or the relative commutant F ′ E ∩C ∗(E), respectively. The most interesting case to us is when E is finite without sinks and where all loops have exits. Since in this case DE is maximal abelian in C ∗(E), [25], it follows immediately that Z(C ∗(E)) ⊆ Z(FE) = F ′ E ∩ C ∗(E). The first inclusion above may be strict, even if C ∗(E) is simple (see [37] for examples and discussion). However, when E is strongly connected (i.e. for every pair of vertices v, w there is a path from v to w) and has period 1 (i.e. the greatest common divisor of the lengths of all loops is 1) then FE is simple and thus its center is trivial, [37, Theorem 6.11]. We begin our analysis of automorphisms of C ∗(E) which globally preserve FE by identifying those which fix FE point-wise. Proposition 4.4. Let E be a finite graph without sinks in which every loop has an exit. Suppose also that the center of FE is trivial. Then EndFE (C ∗(E)) = AutFE (C ∗(E)) = {γz : z ∈ U(1)}. 12 Proof. Let α ∈ EndFE (C ∗(E)). Then αDE = id and thus α = λu for some u ∈ U(DE), as noted in Section 2. Then an easy induction on k shows that u commutes with all F k E, and thus u ∈ F ′ E ∩ DE is the center of FE, since DE is a MASA in FE. Hence u is a scalar and α = λu is a gauge automorphism. E ∩ DE. But F ′ In what follows, we will be mainly working with a finite graph E without sinks and sources. For each vertex v ∈ E0 we select exactly one edge ev with r(ev) = v, and define T = Xv∈E 0 Sev, (15) as in the proof of Theorem 4.2. Then T is an isometry such that γz(T ) = zT for all z ∈ U(1). Lemma 4.5. Let E be a finite graph without sinks and sources in which every loop has an exit. Also, we assume that the center of FE is trivial. For an automorphism α ∈ Aut(C ∗(E)) the following conditions are equivalent. (1) αγz = γzα for each z ∈ U(1); (2) For each e ∈ E1 there is a w ∈ FE such that α(Se) = wT , where T is an isometry as in (15); (3) α(FE) = FE. Proof. (1)⇒(3): If α commutes with each γz then α(FE) ⊆ FE, since FE is the fixed- point algebra for the gauge action. Also, αγz = γzα implies α−1γz = γzα−1, and thus α−1(FE) ⊆ FE as well. (3)⇒(1): Suppose α(FE) = FE. Then for each x ∈ FE and z ∈ U(1) we have z (x) = x. Then by Proposition 4.4 there exists an ω(z) ∈ U(1) such that z (x) = γω(z). The mapping z 7→ ω(z) is a continuous character of U(1). αγzα−1γ−1 αγzα−1γ−1 Indeed, γω(zy) = αγzyα−1γ−1 zy = (αγzα−1γ−1 = γω(z)γzγω(y)γyγ−1 zy = γω(z)γω(y). z )γz(αγyα−1γ−1 y )γyγ−1 zy Hence there exists an m ∈ Z such that ω(z) = zm and, consequently, we have αγz = γzm+1α for all z ∈ U(1). Therefore γz(α−1(Se)) = zm+1α−1(Se) for all e ∈ E1. Since α−1 is an automorphism, this implies that C ∗(E) is generated by the spectral subspace C ∗(E)(m+1) of the gauge action. Thus m + 1 = ±1. However, the case m + 1 = −1 is e , we have e , where xe = α−1(T )Se are partial isometries in FE such that impossible. Indeed, let T be an isometry as in (15). Since 1 = Pe∈E 1 SeS∗ α−1(T ) = Pe∈E 1 xeS∗ e Se = Pr(e). Thus x∗ exe = S∗ Xe∈E 1 x∗ exe = Xe∈E 1 Pr(e) > 1, 13 due to our assumptions on the graph E. On the other hand, xex∗ α−1(T )SeS∗ Xe∈E 1 e = Xe∈E 1 This yields a contradictionPe∈E 1 x∗ (1)⇒(2): We have α(Se) = α(Se)T ∗T = Pv∈E 0 α(Se)S∗ exe >Pe∈E 1 xex∗ the gauge action then α(Se)S∗ stably finite. (2)⇒(1): This is clear. ev belongs to FE, and the claim follows. e α−1(T ∗) ≤ α−1(T T ∗) ≤ 1. e, since FE being an AF-algebra is evSev. If α commutes with Now, we turn to the definition of the restricted Weyl group of C ∗(E). If E has no sinks and all loops have exits, then each α ∈ AutDE (C ∗(E)) automatically belongs to Aut(C ∗(E), FE) by Proposition 3.2, above. Thus we may consider the quotient of Aut(C ∗(E), FE, DE) := Aut(C ∗(E), DE) ∩ Aut(C ∗(E), FE) by AutDE (C ∗(E)), which will be called the restricted Weyl group of C ∗(E) and denoted RWE. That is, RWE := Aut(C ∗(E), FE, DE)/AutDE (C ∗(E)). (16) We denote by P k E the collection of all unitaries in U(C ∗(E)) of the form P SαS∗ E is a finite subgroup of U(C ∗(E)), and we have with α = β = k. Clearly, each P k P k E. As shown in [38], every u ∈ NFE (DE) can be uniquely written as u = dw, where d ∈ U(DE) and w ∈ PE. That is, the normalizer of the diagonal in FE is a semi-direct product for each k. We set PE :=S∞ E ⊆ P k+1 k=0 P k E β NFE (DE) = U(DE) ⋊ PE. (17) We denote by GRE the subgroup of Aut(C ∗(E)) generated by automorphisms of the graph E and λ(PE ∩ UE)−1. That is, GRE := hλ(PE ∩ UE)−1 ∪ ΓEi ⊆ Aut(C ∗(E)). (18) By Proposition 3.6, there is a natural embedding of GE into the Weyl group WE of C ∗(E). Its restriction yields an embedding of GRE into the restricted Weyl group RWE of C ∗(E). Furthermore, λ(PE ∩ UE)−1 is a normal subgroup of GRE. Hence we have GRE = λ(PE ∩ UE)−1ΓE and λ(PE ∩ UE)−1 ∩ ΓE = Γ0 E. (19) Similarly to Theorem 3.7, in the restricted case we have the following. Proposition 4.6. Let E be a finite graph without sinks and sources in which every loop has an exit, and such that the relative commutant of FE in C ∗(E) is trivial. Then GRE ∩ Inn(C ∗(E)) ⊆ {Ad(w) : w ∈ PE}. 14 Proof. By Theorem 3.7, every element of GRE ∩ Inn(C ∗(E)) is of the form Ad(w) with w ∈ SE. Since this Ad(w) globally preserves FE, by hypothesis, Theorem 4.2 implies that w ∈ PE. Since for each w ∈ PE the corresponding inner automorphism Ad(w) of C ∗(E) has finite order, Proposition 4.6 immediately yields the following. Corollary 4.7. Let E be a finite graph without sinks and sources in which every loop has an exit, and such that the relative commutant of FE in C ∗(E) is trivial. Then every element of infinite order in GRE has infinite order in Out(C ∗(E)) as well. In general, it is a non-trivial matter to verify outerness of an automorphism of a graph algebra. Corollary 4.7 solves this problem for a significant class of automorphisms. In the case of Cuntz algebras, an analogous result was proved in [45, Theorem 6], and provided a convenient outerness criterion for permutative automorphisms -- probably, the most studied class of automorphisms of On. Now, we turn to analysis of the action of the restricted Weyl group on the diagonal MASA. Let E be a finite graph without sinks in which every loop has an exit. If α is an automorphism of DE then, following [9], we say that α has property (P) if there exists a non-negative integer m such that the endomorphism αϕm commutes with the shift ϕ. In that case, we also say that α satisfies (P) with m. We define AE := {α ∈ Aut(DE) : both α and α−1 have property (P)}. (20) Clearly, AE is a subgroup of Aut(DE). In the case of the Cuntz algebras, it was shown in [9] that the restricted Weyl group is naturally isomorphic to AE. Now, we investigate this problem for graph algebras. Lemma 4.8. Let E be a finite graph without sinks and sources in which every loop has an exit. Also, we assume that the center of FE is trivial. If α ∈ Aut(C ∗(E), DE) ∩ Aut(C ∗(E), FE) then for each e ∈ E1 there exists a partial unitary d ∈ DE and a partial isometry w ∈ FE equal to a finite sum of words of the form SµS∗ ν with µ = ν, and such that α(Se) = dwT, (21) where T is an isometry as in (15). Proof. For each e ∈ E1, Se is a partial isometry normalizing DE (i.e., SeDES∗ e ⊆ DE and S∗ e DESe ⊆ DE). Since α(DE) = DE, it follows that α(Se) normalizes DE as well. By [25, Theorem 10.1], α(Se) equals dv for some partial unitary d ∈ DE and v a partial isometry which can be written as the sum of a finite collection of words SµS∗ ν . Since α commutes with the gauge action by Lemma 4.5, we have γz(v) = zv for each z ∈ U(1). Thus for each of the words in the decomposition of v we have µ = ν + 1. Now let T be as in (15). Then α(Se) = dv = dwT is the desired decomposition, with w = vT ∗. 15 Proposition 4.9. Let E be a finite graph without sinks and sources in which every loop has an exit. Also, we assume that the center of FE is trivial. Let α ∈ Aut(C ∗(E), DE) ∩ Aut(C ∗(E), FE) and let T be an isometry as in (15). For each e ∈ E1 let de and we be as in Lemma 4.8 so that α(Se) = deweT . Then for each path µ of length r we have α(Sµ) = DµWµT r, where Dµ ∈ DE and Wµ, a sum of words in FE, are such that Dµ = (dµ1wµ1T )(dµ2wµ2T ) · · · (dµr−1wµr−1T )dµr(wµr−1T )∗ · · · (wµ1T )∗, Wµ = (wµ1T )(wµ2T ) · · · (wµr−1T )wµrT ∗(r−1). Furthermore, Dµ is a partial unitary and Wµ is a partial isometry. Thus we also have α(Pµ) = DµD∗ µWµT rT ∗rW ∗ µ . Proof. This is established by a somewhat tedious but not complicated inductive argu- ment. We illustrate it with the case r = 2 only. Since T ∗T = 1 and wµ1 is a partial isometry, we have α(Sµ1Sµ2) = (dµ1wµ1T )(dµ2wµ2T ) = dµ1wµ1(w∗ µ1wµ1)(T dµ2T ∗)T wµ2T. But both w∗ normalizing DE). Thus the above expression equals µ1wµ1 and T dµ2T ∗ belong to DE (since wµ1 and T are partial isometries [(dµ1wµ1T )dµ2(wµ1T )∗][(wµ1T )wµ2T ∗]T 2, as required. Note that Dµ = (dµ1wµ1T )dµ2(wµ1T )∗ = dµ1(wµ1T dµ2T ∗w∗ µ1) is a partial unitary in DE, and Wµ = (wµ1T )wµ2T ∗ is a sum of words in FE and a partial isometry (as a product of partial isometries with mutually commuting domain and range projections). Corollary 4.10. Keeping the notation and hypothesis from Proposition 4.9, let M = min{k : (∀e ∈ E1) de ∈ Dk E, we ∈ F k E}. Then Dµ ∈ DM +µ−1 E and Wµ ∈ F M +µ−1 E . Remark 4.11. In fact, the conclusion of Proposition 4.9 remains valid with no hypothe- sis on the graph E and for any endomorphism α of C ∗(E), if we know that α(Se) = deweT for each e ∈ E1. Remark 4.12. Let T =Pv∈E 0 Sev and T ′ =Pv∈E 0 Sfv be two isometries as in (15). Set eU =Pv∈E 0 SfvS∗ ev. Then eU is a partial isometry in the finite dimensional C ∗-algebra E. We may extend it to a unitary U ∈ P 1 F 1 α ∈ Aut(C ∗(E), FE) ∩ Aut(C ∗(E), DE) and α(Se) = deweT = d′ then d′ unique such we which satisfies α(Se) = deweT and we = α(SeS∗ E, and then we have T ′ = UT . Thus if eT ′, as in Lemma 4.8, eU = we. In particular, if we fix T , then for each e ∈ E1 there is a e = de and w′ ew′ e )weT T ∗. 16 Let α be an endomorphism of C ∗(E), where E is a finite graph without sinks. Define a unital, completely positive map Φα : C ∗(E) → C ∗(E) by Φα(x) = Xe∈E 1 α(Se)xα(S∗ e ). Then the following braiding relation holds If α = λu for some u ∈ UE, then Φα = Ad(u) ◦ ϕ. αϕ = Φαα. (22) (23) Now, we are in the position to show how elements of the restricted Weyl group act on the diagonal, by automorphisms of DE (or, equivalently, homeomorphisms of its spectrum) which eventually commute with the shift. Theorem 4.13. Let E be a finite graph without sinks and sources in which every loop has an exit. Also, we assume that the center of FE is trivial. If α ∈ Aut(C ∗(E), DE) ∩ Aut(C ∗(E), FE) then the restriction of α to DE belongs to AE. This yields a group homomorphism Res : Aut(C ∗(E), DE) ∩ Aut(C ∗(E), FE) −→ AE and a group embedding RWE ֒→ AE. Proof. It suffices to show that there exists an m such that α−1ϕm+1 = ϕα−1ϕm on DE. But αϕα−1ϕm = Φαϕm. Thus, we must show that Φαϕm = ϕm+1 on DE for a sufficiently large m. To this end, for each e ∈ E1 write α(Se) = deweT as in Lemma 4.8. Let m be so large that all we belong to F m+1 . Then, for x ∈ DE, using relation (9) we have T ϕm(x) = ϕm+1(x)T and each we commutes with ϕm+1(x). Consequently, we have E Φαϕm(x) = Xe∈E 1 as required. deweT ϕm(x)T ∗w∗ ed∗ e = ϕm+1(x)Xe∈E 1 deweT T ∗w∗ ed∗ e = ϕm+1(x), Finally, the kernel of the Res homomorphism coincides with AutDE (C ∗(E)). Thus, the restriction gives rise to a homomorphic embedding of the restricted Weyl group RWE into AE. In the case of On, it was shown in [9] that the restriction mapping from Theorem 4.13 is surjective. In this way, the restricted Weyl group of the Cuntz algebra has been identified with the group of homeomorphisms which (along with their inverses) eventu- ally commute with the full one-sided n-shift. The more general case of graph algebras is more complicated. We would like to pose it as an open problem to determine the ex- act class of automorphisms of DE (or, equivalently, homeomorphisms of the underlying space) which arise as restrictions of automorphisms of the graph algebra which preserve both the diagonal and the core AF-subalgebra. 17 5 The localized automorphisms Throughout this section, we assume that E is a finite graph without sinks. Recall that an endomorphism λu of C ∗(E) is called localized if the corresponding unitary u belongs to a finite dimensional algebra F k E ∩ UE for some k. Our main aim in this section is to produce an invertibility criterion for localized endomorphisms, analogous to [15, Theorem 3.2]. Let u be unitary in F k E ∩ UE, for some fixed k ≥ 1. Using (7) and (8) we see that E and r ≥ 1. Following [45], for each pair e, f ∈ E1 we λu(x) = Ad(ur)(x) for all x ∈ F r define a linear map au e,f : F k−1 E → F k−1 E by e,f (x) = S∗ au e u∗xuSf , x ∈ F k−1 E . (24) E /D0 Denote Vk := F k−1 maps from Vk to itself. Since au Now we define Au as the subring of L(Vk) generated by {au e,f E, the quotient vector space, and let L(Vk) be the space of linear e,f : Vk → Vk. E, there is an induced map au : e, f ∈ E1}. E) ⊆ D0 e,f (D0 We denote by H the linear span of the generators Se's. Let u be as above. Following [13, p. 386], we define inductively Ξ0 = F k−1 E , Ξr = λu(H)∗Ξr−1λu(H) , r ≥ 1 . (25) E If α, β are paths of length r, then we denote by Tα,β the linear map from F k−1 and thus it eventually stabilizes. If Ξp = Ξp+1 then we have Ξu :=T∞ Then {Ξr}r is a non-increasing sequence of finite dimensional, self-adjoint subspaces of F k−1 r=0 Ξr = Ξp. to itself E r)(x)Sβ for all x ∈ F k−1 defined by Tα,β = au E . Note that Tα,βTµ,ν = 0 if either αµ or βν does not form a path. It easily follows from our definitions that the space Ξr is linearly spanned by elements of the form Tα,β(x), for α, β ∈ Er, x ∈ F k−1 E . α1,β1. We have Tα,β(x) = S∗ αr,βr · · · au α Ad(u∗ Theorem 5.1. Let E be a finite graph without sinks, and let u ∈ UE be a unitary in F k E for some k ≥ 1. Then the following conditions are equivalent: (1) λu is invertible with localized inverse; (2) the sequence of unitaries {Ad(u∗ m)(u∗)}m≥1 eventually stabilizes; (3) the ring Au is nilpotent; (4) Ξu ⊆ D0 E. E ) ⊆ F l Proof. (1) ⇒ (3): u (F k−1 λ−1 l, and consider an element Tα,β = au x = λ−1 u (b). Then x ∈ F l If the inverse of λu is localized then there exists an l such that E. Let α = (e1, e2, · · · , el) and β = (f1, f2, · · · , fl) be paths of length and let u. Let b ∈ F k−1 E and we have b = λu(x) = Ad(ul)(x). Therefore e1,f1 of Al e2,f2au · · · au el,fl E Tα,β(b) = au el,fl · · · au e2,f2au e1,f1(b) = S∗ αAd(u∗ l )(b)Sβ = S∗ αxSβ. 18 Since x can be written asPγ=ρ=l cγ,ρ(x)SγS∗ Tα,β(b) = S∗ cγ,ρ(x)SγS∗ α(cid:16) Xγ=ρ=l ρ for some cγ,ρ(x) ∈ C, we have ρ(cid:17)Sβ =(cα,β(x)Pr(α), 0, if r(α) = r(β) if r(α) 6= r(β) because S∗ and hence we see that Al αSγ = Pr(α) if α = γ, and S∗ u = 0. αSγ = 0 otherwise. This implies that Tα,β(b) ∈ D0 E, (3) ⇒ (4): Let Al u = 0 for some positive integer l. Then Tα,β(b) ∈ D0 and all α, β such that α = β = l. But this immediately yields Ξl ⊆ D0 E for all E and, b ∈ F k−1 consequently, Ξu ⊆ D0 E. E (4) ⇒ (2): Let Ξu ⊆ D0 E, and let l be a positive integer such that Ξl = Ξu. Let b ∈ F k−1 E and thus it commutes with ϕm(u) for all m, since u commutes with the vertex projections. Consequently, for each r ≥ 1 we have and let α, β ∈ El. Then Tα,β(b) belongs to D0 E Ad(u∗ l+r)(b) = Ad(ϕl−1+r(u∗) · · · ϕl(u∗))(cid:16) Xα,β∈El SαTα,β(b)S∗ β(cid:17) Sα Ad(ϕr−1(u∗) · · · u∗)(Tα,β(b))S∗ β = Xα,β∈El = Xα,β∈El SαTα,β(b)S∗ β . the sequence Ad(u∗ m)(b) stabilizes from m = l + 1. Write f , for some be,f ∈ F k−1 E . Then for each m we have Thus for each b ∈ F k−1 u∗ =Pe,f ∈E 1 Sebe,f S∗ E Ad(u∗ m+1)(u∗) = Xe,f ∈E 1 = Xe,f ∈E 1 Ad(ϕ(ϕm−1(u∗) · · · ϕ(u∗)u∗))(Sebe,f S∗ f ) Se Ad(ϕm−1(u∗) · · · ϕ(u∗)u∗)(be,f )S∗ f and, consequently, the sequence Ad(u∗ m)(u∗) stabilizes from m = l + 2. (2) ⇒ (1): Suppose that the sequence Ad(u∗ m)(u∗) eventually stabilizes. Hence Ad(u∗ r)(u∗) = w for all sufficiently large r. It follows that λu(w) = Ad(ur)(w) = u∗ for suitable r depending on w. Thus λu(w)u = u∗u = 1 and, consequently, λu is invertible with inverse λw. This completes the proof. We also include a different and much more direct proof of implication (1) ⇒ (2), that sheds additional light on the equivalent conditions of the theorem and is interesting in its own right. Let λu be invertible, and suppose that there exists an l ∈ N and a unitary v ∈ UE E, u∗ = λu(v) = E such that λuλv = id. Then we have λu(v)u = 1. Since v ∈ F l in F l 19 Ad(ul)(v), and hence Ad(u∗ l )(u∗) = v. Now for r ≥ 1 we have Ad(u∗ l+r)(u∗) = Ad(ϕl+r−1(u∗) · · · ϕ(u∗)u∗)(u∗) = ϕl+r−1(u∗) · · · ϕl(u∗)Ad(u∗ = ϕl+r−1(u∗) · · · ϕl(u∗)vϕl(u) · · · ϕl+r−1(u) = v l )(u∗)ϕl(u) · · · ϕl+r−1(u) since v commutes with ϕm(u) for every m ≥ l. Thus we can conclude that Ad(u∗ stabilizes at v from m = l. m)(u∗) Remark 5.2. Let u ∈ F 1 V1 = D0 Theorem 5.1 trivially implies that λu is an automorphism of C ∗(E). E ∩ UE, so that λu is quasi-free. Since F 0 E = {0} and consequently each au E, we have e,f is a zero map. Therefore Au = {0} and E = D0 E/D0 If u ∈ UE normalizes DE then λu(DE) ⊆ DE. It may well happen that such a restriction is an automorphism of DE even though λu is not invertible. For unitaries in the algebraic part of FE this can be checked in a way similar to [15, Theorem 3.4]. Indeed, let u ∈ UE∩NF k E). Then it follows from (9) that u ∈ NFE (DE). Furthermore, the subspace Dk−1 e,f , e, f ∈ E1. E e,f to Dk−1 We denote by bu e,f the map induced on E /D0 e,f : e, f ∈ E1}. V D u be the subring of L(V D k Also, we consider a nested sequence of subspaces ΞD e,f the restriction of au E. Let AD is invariant under the action of all maps au k ) generated by {bu r of Dk−1 E , defined inductively as E , and by bu (Dk of F k−1 := Dk−1 E E 0 = Dk−1 ΞD E , ΞD r = λu(H)∗ΞD r−1λu(H), r ≥ 1. Each ΞD r is finite dimensional and self-adjoint. We set ΞD Theorem 5.3. Let E be a finite graph without sinks and let u ∈ UE ∩ NF k some k ≥ 1. Then the following conditions are equivalent: E u :=Tr ΞD r . (26) (Dk E), for (1) λu restricts to an automorphism of DE; (2) the ring AD u is nilpotent; (3) ΞD u ⊆ D0 E. Proof. (1) ⇒ (3): Since the algebraic part S∞ phism of S∞ E of DE coincides with the linear span of all projections in DE, every automorphism of DE restricts to an automor- E. Let e2,f2bu E. Thus, there exists an l such that (λuDE )−1(Dk−1 E ) ⊆ Dl · · · bu α = (e1, e2, · · · , el) and β = (f1, f2, · · · , fl) be in El, and let Rα,β := bu el,fl be in (AD 5.1 yields that Rα,β(d) ∈ D0 of implication (3)⇒(4) in Theorem 5.1, this implies that ΞD e1,f1 u )l. Then the same argument as in the proof of implication (1)⇒(3) in Theorem u )l = {0}. But as in the proof E for all d ∈ Dk−1 E . Thus (AD t=0 Dt t=0 Dt u ⊆ D0 E. (3) ⇒ (2): Let ΞD u ⊆ D0 E and let l be a positive integer such that ΞD l = ΞD and let α, β ∈ El. Then Rα,β(d) belongs to D0 E, and this entails (AD u . Let u )l = {0}, d ∈ Dk−1 i.e. the ring AD E u is nilpotent. 20 t=0 Dt E. u is nilpotent. We show by induction on r ≥ k that all Dr (2) ⇒ (1): Suppose that AD Firstly, let r = k and d ∈ Dk E are in the range of λu restricted toS∞ eventually stabilizes at some f ∈S∞ t=0 Dt E. Similarly to the argument in the implication (4)⇒(2) of the proof of Theorem 5.1 one shows that the sequence Ad(ϕm(u∗) · · · ϕ(u∗)u∗)(d) E. It then follows that d = λu(f ). For the inductive step, suppose that r ≥ k and Dr E and ϕr(D1 generated by Dr all y ∈ D1 the inductive hypothesis. Thus the sequence E), it suffices to show that ϕr(y) belongs to λu(S∞ E is in λu(S∞ E. However, ϕr(y) commutes with u and ϕr−1(y) ∈ Dr t=0 Dt E ⊆ λu(S∞ E). Since Dr+1 is E E) for E) by t=0 Dt t=0 Dt Ad(ϕm(u∗) · · · ϕ(u∗)u∗)(ϕr(y)) = ϕ(Ad(ϕm−1(u∗) · · · ϕ(u∗)u∗)(ϕr−1(y))) eventually stabilizes at λ−1 u (ϕr(y)) ∈S∞ t=0 Dt E. It should be noted that, in the setting of Theorem 5.3, it may well happen that λuDE is an automorphism of DE while λu is a proper endomorphism of C ∗(E). In that case there may not exist any unitary w ∈ C ∗(E) such that (λuDE )−1 = λwDE , and thus (λuDE )−1 is not localized in our sense (as defined in the first paragraph of this section). See [15], [12], [16, Theorem 3.8] and [11, Proposition 3.2] for examples and further discussion of this interesting point. 6 The permutative automorphisms Throughout this section, we assume that E is a finite graph without sinks. Our main goal in this section is to give a combinatorial criterion for invertibility of λu, u ∈ PE, analogous to [9, Corollary 4.12]. Both the statement of the criterion (see Theorem 6.4 below) and its proof are quite similar to those given in the case of the Cuntz algebras in [15], suitably generalized to the present case of graph C ∗-algebras. The key idea is to break the process into two steps, Condition (b) and Condition (d), of which the former detects those endomorphisms which restrict to automorphisms of the diagonal DE. It is useful to look at collections of paths of a fixed length beginning or ending at the same vertex. Hence we introduce the following notation. For v, w ∈ E0 and k ∈ N, let v,w := {α ∈ Ek : r(α) = Ek v,w, disjoint unions. If ∗,v := {α ∈ Ek : s(α) = v} and Ek v,∗ := {α ∈ Ek : r(α) = v}, Ek v, s(α) = w}. Then Ek =Sv∈E 0 Ek u ∈ P k E, k > 0, then there exist permutations σv ∈ Perm(Ek v,∗) such that v,∗ =Sv,w∈E 0 Ek ∗,v =Sv∈E 0 Ek u = Xv∈E 0 Xα∈Ek v,∗ Sσv (α)S∗ α. (27) A unitary u ∈ P k exist permutations σv,w ∈ Perm(Ek v,w) such that E, k > 0, commutes with all the vertex projections if and only if there u = Xv,w∈E 0 Xα∈Ek v,w Sσv,w(α)S∗ α. 21 (28) If the unitary u ∈ P k E ∩ UE is understood, as in equation (28), then we will denote by σ = ∪v,w∈E 0σv,w the corresponding permutation of Ek. In that case, we will also write λu = λσ. Now let u ∈ P k E ∩ UE, e, f ∈ E1, and consider the linear map au e,f , as defined in so ordered that the initial (24). With respect to the basis {SµS∗ vectors span Dk−1 E , the matrix of au ν : µ, ν ∈ Ek−1} of F k−1 E e,f has the block form au e,f =(cid:18) bu e,f 0 cu e,f du e,f (cid:19) , similarly to [15, Section 4]. The first block corresponds to the subspace Dk−1 Thus, the map au e,f ∈ L(Vk) has a matrix E au e,f =(cid:18) bu e,f 0 ∗ du e,f (cid:19) , (29) of F k−1 E . (30) to its quotient Vk does not affect the matrix for du with the first block corresponding to the subspace V D k of Vk. Note that the passage from the space F k−1 e,f and thus there is no tilde over it in formula (30). It is an immediate corollary to Theorem 5.1 that an endomorphism λu of C ∗(E) is invertible if and only if the following two conditions are satisfied. E Condition (b): Condition (d): the ring generated by {bu the ring generated by {du e,f : e, f ∈ E1} is nilpotent. e,f : e, f ∈ E1} is nilpotent. The remainder of this section is devoted to the description of a convenient combinatorial interpretation of these two crucial conditions, similar to the one appearing in [15] and used in the analysis of permutative endomorphisms of the Cuntz algebras. By virtue of Theorem 5.3, Condition (b) alone is equivalent to the restriction of λu to the diagonal DE being an automorphism. 6.1 Condition (b) We fix u ∈ P k then bu Since bu α ∈ Ek−1 E ∩ UE and denote by σ the corresponding permutation, as above. If e 6= f e,e, e ∈ E1. e has exactly one 1 in the row corresponding to each e may be identified with a mapping e,f = 0. Thus, it suffices to consider the ring generated by maps bu e (1) = Pr(e), the matrix of bu ∗,r(e), and 0's elsewhere. Consequently, each bu e := bu e : Ek−1 f u ∗,r(e) → Ek−1 ∗,s(e), f u e (α) = β, (31) whenever bu e has 1 in the α -- β entry. If the unitary u is given by a permutation σ then e (α) = β ⇔ ∃g ∈ E1 s.t. σ(e, α) = (β, g). f u (32) e bu The product bu g). Now Condition (b) may be phrased in terms of mappings {f u follows: g corresponds to the composition f u g ◦ f u e (in reversed order of e and e }, as e } rather than {bu 22 There exists an m such that for all e1, . . . , em ∈ E1 if T = f u v ∈ E0 and α ∈ Ek−1 either Ek−1 ∗,v ∩ T −1(α) = ∅ or Ek−1 ∗,v ⊆ T −1(α). e1 ◦ . . . ◦ f u em then for all Taking into account (31) above, we arrive at the following: Condition (b): There exists an integer m ∈ Z such that for all e1, . . . , em ∈ E1 either: (i) f u ∗,r(em) and its range consists of exactly one element. em has the empty domain, or (ii) its domain equals Ek−1 e1 ◦ . . . ◦ f u In the remainder of this section, notation (α, β) indicates either a single path in E∗ or an ordered pair in the cartesian product E∗ × E∗. This will be clear from context. Lemma 6.1. Let E be a finite graph without sinks in which every loop has an exit. Let u ∈ P k E ∩ UE. Then Condition (b) holds for u (and hence λuDE is an automorphism of DE) if and only if there exists a partial order ≤ onSv∈E 0 Ek−1 1. if v ∈ E0 \ r(E1) then each element of Ek−1 ∗,v × Ek−1 ∗,v ∗,v × Ek−1 ∗,v is minimal, each diagonal such that: element (α, α) is minimal, and there are no other minimal elements; 2. if e ∈ E1 and α 6= β ∈ Ek−1 ∗,r(e) then (f u e (α), f u e (β)) ≤ (α, β). e1 ◦ · · · ◦ f u e1 ◦ · · · ◦ f u Proof. At first suppose that Condition (b) holds for u. Define a relation ≤ as follows. For any α ∈ Ek−1 set (α, α) ≤ (α, α). If γ 6= δ then (α, β) ≤ (γ, δ) if and only if there exists a sequence e1, . . . , ed ∈ E1, possibly empty, such that α = (f u ed)(γ) and β = (f u ed)(δ). Reflexivity and transitivity of ≤ are obvious. To see that ≤ is also antisymmetric, suppose that (α, β) ≤ (γ, δ) and (γ, δ) ≤ (α, β). If (α, β) 6= (γ, δ) then, by definition of ≤, α 6= β, γ 6= δ and there exist edges e1, . . . , ed, g1, . . . , gh such that (α, β) = (f u ed)(γ, δ) and (γ, δ) = (fg1 ◦ · · · ◦ fgh)(α, β). Then (α, β) = (f u ed ◦ fg1 ◦ · · · ◦ fgh has two distinct fixed points, a contradiction with Condition (b). Thus (α, β) = (γ, δ) and ≤ is also antisymmetric. Hence ≤ is a partial order satisfying condition 2 above. By the very definition of ≤, if v ∈ E0 \ r(E1) then each element of Ek−1 is minimal, and likewise each diagonal element (α, α) is minimal. If any other element were minimal for ≤ then there would exist α 6= β and e ∈ E1 such that f u e (β) = β. Thus f u e would have two distinct fixed points, contradicting Condition (b). Thus condition 1 holds true as well. ed ◦ fg1 ◦ · · · ◦ fgh)(α, β). That is, f u e (α) = α and f u ∗,v × Ek−1 ∗,v e1 ◦ · · · ◦ f u e1 ◦ · · · ◦ f u e1 ◦ · · · ◦ f u Conversely, if a partial order ≤ with the required properties exists, then counting shows that each sufficiently long composition product of mappings {f u e } either has the empty domain or its range consists of a single element (and the domain is as required, due to (31)). This completes the proof. Remark 6.2. By the diagram of f u e we mean a directed graph with vertices correspond- ing to the union of the domain and the range of the map f u e and with an edge from e , e ∈ E1, we vertex α to β if and only if f u obtain a directed graph whose edges are labelled by {f u e } or simply by the edges of E, see Example 6.5 below. Our Condition (b) is equivalent to existence of a positive integer m such that all words of length m are synchronizing for this labelled graph2. Also note e (α) = β. Combining the diagrams of all f u 2We are grateful to Rune Johansen for pointing this out. 23 that if the conditions of Lemma 6.1 are satisfied and e ∈ E1 is such that s(e) = r(e), then the diagram of f u e is a rooted tree with the root being the unique fixed point, cf. [15, Section 4.1]. 6.2 Condition (d) Again, we fix a u ∈ P k E ∩ UE and denote by σ the corresponding permutation. It is easy to verify that for e, g ∈ E1 each row of the matrix du e,g either may have 1 in one place and 0's elsewhere or may consist of all zeros. This matrix has 1 in (α, β) row and (γ, δ) column if and only if there exists an h ∈ E1 such that SαS∗ δ uSg. In turn, this takes place if and only if e u∗SγShS∗ β = S∗ hS∗ σ(e, α) = (γ, h) and σ(g, β) = (δ, h). (33) ∗,r(e) × Ek−1 e,g, as follows. The domain D(f u For each e, g ∈ E1 we now define a mapping f u e,g) of f u e,g consists of all (α, β) ∈ Ek−1 e,g is non-zero, and e,g(α, β) = (γ, δ) ∈ Ek−1 × Ek−1 for (γ, δ) satisfying (33). the corresponding value is f u Note that, by the very definition of du e,g, in such a case we must necessarily have α 6= β and γ 6= δ. We denote Ψu := Ek−1 × Ek−1 \ {(α, α) : α ∈ Ek−1}. We also denote by ∆u the subset of Ψu consisting of all those (α, β) for which there exist e, g ∈ E1 such that (α, β) belongs to the domain D(f u ∗,r(g) for which the (α, β) row of du e,g). It is a simple matter to verify that in terms of mappings {f u e,g} Condition (d) may be rephrased as follows (cf. [15, Section 4.3]). Condition (d): There exists an m such that for all (e1, g1), . . . , (em, gm) ∈ Ψu the domain of the map f u e1,g1 ◦ . . . ◦ f u em,gm is empty. The proof of the following lemma is essentially the same as that of [15, Lemma 4.10] and thus it is omitted. Lemma 6.3. Let E be a finite graph without sinks in which every loop has an exit. Let u ∈ P k E ∩ UE. Then Condition (d) holds for u if and only if there exists a partial order ≤ on Ψu such that: 1. the set of minimal elements coincides with Ψu \ ∆u; 2. if e, g ∈ E1 and (α, β) ∈ D(f u e,g) then f u e,g(α, β) ≤ (α, β). Combining Lemma 6.1 with Lemma 6.3 we obtain a combinatorial criterion of in- vertibility of permutative endomorphisms, similar to [15, Corollary 4.12]. Theorem 6.4. Let E be a finite graph without sinks in which every loop has an exit, and let u ∈ P k E ∩ UE. Then the endomorphism λu is invertible if and only if conditions of Lemma 6.1 and Lemma 6.3 hold for u. 24 6.3 Examples We give two examples with small graphs illustrating the combinatorial machinery de- veloped in the preceding section. In Example 6.5, we exhibit a proper permutative endomorphism of C ∗(E) which restricts to an automorphism of the diagonal MASA DE. On the other hand, in Example 6.6 we provide an order 2 permutative automor- phism of a Kirchberg algebra C ∗(E) with K0(C ∗(E)) ∼= Z ∼= K1(C ∗(E)), which is neither quasi-free nor comes from a graph automorphism. Example 6.5. Consider the following graph E (the Fibonacci graph): .................. ........................ ...................................................................................... .................. ........................ .................................................................................................................................... ... .. .. .. .. . .. . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . .. .. .. .. ... .... ... .. .. .. .. . .. . .. .. .. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . • .. . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . .. . .. . . . . . . . . . . . .. . . . .. . .. .. .. ... ... .................................................................................... .. ... ... ................................................................................................................................. e1 e3 ........................ .................. e2 ... ... .. .. . .. . .. . . . . . . . . . . . . . . . . . . . . . • . . . . . . . . . . . . . . . . . . . . . .. . . . .. . .. .. .. .. ... ...... E ∩UE At level k = 2, there are 2 permutations in P 2 ∼= Z2. Denoting edge ej simply by j, the non-trivial transposition is (11, 32). The corresponding map f1 : {1, 3} → {1, 3} is such that f1(1) = 3 and f1(3) = 1. Thus Condition (b) does not hold and, consequently, the corresponding endomorphism is surjective neither on C ∗(E) nor on DE. At level k = 3, there are 24 permutations in P 3 ∼= S3 × Z2 × Z2. These are generated by σ = (132, 321), τ = (111, 132, 321), υ = (113, 323) and ω = (211, 232). Of these 24 permutations, only τ υ and id satisfy Condition (b). However, τ υ does not satisfy Condition (d). Thus, λτ υ is a proper endomorphism of C ∗(E) (i.e., it is not surjective) which restricts to an automorphism of DE. E ∩ UE The maps f1 : {(11), (13), (32)} → {(11), (13), (32)}, f2 : {(11), (13), (32)} → {(21), (23)}, and f3 : {(21), (23)} → {(11), (13), (32)} corresponding to permutation τ υ and involved in verification of Condition (b) are illustrated by the following labelled graph. f3 (11) f1 (13) f1 ................................... .................. .. .. .. ... ... ... .. . ..... ...... .................................... ........... .................. ........................................ f2 ................................................................................................................................................................................................................................................................................................................................................................................................................................. • ................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . f2 f3 • .................. ......................................... .................. ........................................ .. . .. ... .. .. .. • .. . .. • f2 (32) ⋆ .............................. ............................ ... ..... .. ... .. . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . .. .. .... ... ... .. . . .. . . . . . .. . . . . . . . . . .. . . . . . . . . .. . . . . . . . . . .. . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . .. . .. ... f1 ............................................................. (21) (23) 25 Note that the diagram of the map f1, corresponding to an edge whose source and range coincide, gives rise to a rooted tree. This is the left hand side of the diagram above, with the root (the unique fixed point for f1) indicated by a star. Example 6.6. Consider the following graph E: .................. ........................ ................................................................................... .... ... .. .................. ........................ ................................................................................................................................. ................................................................................................................................. .................. ........................ ................................................................................... e6 .................. ........................ ................................................................................. .. .. .. ..... ..................................................................................................................................... ..................................................................................................................................... ... .. .. .. . .. . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . .. . .. .. ... ... . . . .. . .. . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . .. . .. • . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... ... .. .. .. . .. . . . . .. . . . . . . . . . . .. . .. .. . ... .. .... e1 ...... ... .. .. .. .. . .. . . . .. . ... ... .. .. .. . .. . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . • . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . .. . .. .. ... ... . . . . . . . . . . .. . .. .. . ... .. .... ...... ... .. .. .. .. . .. . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . .. . .. .. ... ... .... ... .. . . . . • .. . .. . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . .. . .. ... .. . . . . ... .. .. .. . .. . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . .. . .. .. .. .. ... ........................ .................. e3 ...... ................................................................................ e4 e5 ........................ .................. e2 At level k = 2, there are 8 permutations in P 2 2. Denoting edge ej simply by j, these are generated by transpositions σ = (25, 63), τ = (11, 52) and υ = (36, 44). Of these 8 permutations, only σ and id satisfy Condition (b). Since σ satisfies Condition (d) as well, λσ is an automorphism of C ∗(E). We have E ∩ UE ∼= Z3 λσ(S2) = S6S3S∗ λσ(S6) = S2S5S∗ λσ(Sj) = Sj, 5 + S2S1S∗ 1, 3 + S6S4S∗ 4, j = 1, 3, 4, 5, and it follows immediately that λ2 σ = id. We note that in the present case K0(C ∗(E)) ∼= Z ∼= K1(C ∗(E)) by [20] and [33, Section 4], see also [39]. Thus C ∗(E) is not isomorphic to a Cuntz algebra, and hence this example is not covered in any way by the results of [15]. References [1] G. Abrams and G. Aranda Pino, The Leavitt path algebra of a graph, J. Algebra 293 (2005), 319 -- 334. [2] T. Bates, J. H. Hong, I. Raeburn and W. Szyma´nski, The ideal structure of the C ∗-algebras of infinite graphs, Illinois J. Math. 46 (2002), 1159 -- 1176. [3] B. Blackadar, Shape theory for C ∗-algebras, Math. Scand. 56 (1985), 249 -- 275. [4] M. Boyle, Flow equivalence of shifts of finite type via positive factorizations, Pacific J. Math. 204 (2002), 273 -- 317. [5] O. Bratteli and P. E. T. Jørgensen, Isometries, shifts, Cuntz algebras and mul- tiresolution wavelet analysis of scale N, Integral Equations & Operator Theory 28 (1997), 382 -- 443. [6] O. Bratteli and P. E. T. Jørgensen, Iterated function systems and permutation representations of the Cuntz algebra, Mem. Amer. Math. Soc. 139 (1999). 26 [7] C. Consani and M. Marcolli, Noncommutative geometry, dynamics, and ∞-adic Arakelov geometry, Selecta Math. (N.S.) 10 (2004), 167 -- 251. [8] R. Conti, Automorphisms of the UHF algebra that do not extend to the Cuntz algebra, J. Austral. Math. Soc. 89 (2010), 309 -- 315. [9] R. Conti, J. H. Hong and W. Szyma´nski, The restricted Weyl group of the Cuntz algebra and shift endomorphisms, J. reine angew. Math. 667 (2012), 177 -- 191. [10] R. Conti, J. H. Hong and W. Szyma´nski, Endomorphisms of the Cuntz algebras, Banach Center Publ. 96 (2012), 81 -- 97. [11] R. Conti, J. H. Hong and W. Szyma´nski, The Weyl group of the Cuntz algebra, Oberwolfach Preprint 2011 -- 31, arXiv:1110.4476. [12] R. Conti, J. Kimberley and W. Szyma´nski, More localized automorphisms of the Cuntz algebras, Proc. Edinburgh Math. Soc. 53 (2010), 619 -- 631. [13] R. Conti and C. Pinzari, Remarks on the index of endomorphisms of Cuntz algebras, J. Funct. Anal. 142 (1996), 369 -- 405. [14] R. Conti, M. Rørdam and W. Szyma´nski, Endomorphisms of On which preserve the canonical UHF-subalgebra, J. Funct. Anal. 259 (2010), 602 -- 617. [15] R. Conti and W. Szyma´nski, Labeled trees and localized automorphisms of the Cuntz algebras, Trans. Amer. Math. Soc. 363 (2011), 5847 -- 5870. [16] R. Conti and W. Szyma´nski, Automorphisms of the Cuntz algebras, to appear in the Proceedings of the EU-NCG Meeting in Bucharest (2011), arXiv:1108.0860. [17] J. Cuntz, Simple C ∗-algebras generated by isometries, Commun. Math. Phys. 57 (1977), 173 -- 185. [18] J. Cuntz, Automorphisms of certain simple C ∗-algebras, in Quantum fields-algebras- processes (Bielefield, 1978), pp. 187 -- 196, ed. L. Streit, Springer, Vienna, 1980. [19] J. Cuntz and W. Krieger, A class of C ∗-algebras and topological Markov chains, Invent. Math. 56 (1980), 251 -- 268. [20] J. Cuntz, A class of C ∗-algebras and topological Markov chains. II. Reducible chains and the Ext-functor for C ∗-algebras, Invent. Math. 63 (1981), 25 -- 40. [21] K. J. Dykema and D. Shlyakhtenko, Exactness of Cuntz-Pimsner C ∗-algebras, Proc. Edinburgh Math. Soc. (2) 44 (2001), 425 -- 444. [22] J. H. Hong, Decomposability of graph C ∗-algebras, Proc. Amer. Math. Soc. 133 (2005), 115 -- 126. [23] J. H. Hong, A. Skalski and W. Szyma´nski, On invariant MASAs for endomorphisms of the Cuntz algebras, Indiana Univ. Math. J. 59 (2010), 1873 -- 1892. 27 [24] J. H. Hong and W. Szyma´nski, Quantum spheres and projective spaces as graph algebras, Commun. Math. Phys. 232 (2002), 157 -- 188. [25] A. Hopenwasser, J. R. Peters and S. C. Power, Subalgebras of graph C ∗-algebras, New York J. Math. 11 (2005), 351 -- 386. [26] A. an Huef and I. Raeburn, The ideal structure of Cuntz-Krieger algebras, Ergodic Theory & Dynamical Systems 17 (1997), 611 -- 624. [27] M. Izumi, Subalgebras of infinite C ∗-algebras with finite Watatani indices I. Cuntz algebras, Commun. Math. Phys. 155 (1993), 157 -- 182. [28] M. Izumi, Subalgebras of infinite C ∗-algebras with finite Watatani indices II: Cuntz- Krieger algebras, Duke Math. J. 91 (1998), 409 -- 461. [29] V. F. R. Jones, On a family of almost commuting endomorphisms, J. Funct. Anal. 122 (1994), 84 -- 90. [30] Y. Katayama and H. Takehana, On automorphisms of generalized Cuntz algebras, Internat. J. Math. 9 (1998), 493 -- 512. [31] K. Kawamura, Polynomial endomorphisms of the Cuntz algebras arising from per- mutations. I. General theory, Lett. Math. Phys. 71 (2005), 149 -- 158. [32] A. Kumjian, D. Pask and I. Raeburn, Cuntz-Krieger algebras of directed graphs, Pacific J. Math. 184 (1998), 161 -- 174. [33] A. Kumjian, D. Pask, I. Raeburn and J. Renault, Graphs, groupoids, and Cuntz- Krieger algebras, J. Funct. Anal. 144 (1997), 505 -- 541. [34] K. Matsumoto, On automorphisms of C ∗-algebras associated with subshifts, J. Op- erator Theory 44 (2000), 91 -- 112. [35] K. Matsumoto, Orbit equivalence of topological Markov shifts and Cuntz-Krieger algebras, Pacific J. Math. 246 (2010), 199 -- 225. [36] D. Pask and A. Rennie, The noncommutative geometry of graph C ∗-algebras I: The index theorem, J. Funct. Anal. 233 (2006), 92 -- 134. [37] D. Pask and S.-J. Rho, Some intrinsic properties of simple graph C ∗-algebras, in 'Operator algebras and mathematical physics' (Constant¸a, 2001), 325 -- 340, Theta, Bucharest, 2003. [38] S. C. Power, The classification of triangular subalgebras of AF C ∗-algebras, Bull. London Math. Soc. 22 (1990), 269 -- 272. [39] I. Raeburn, Graph algebras, CBMS Regional Conf. Series in Math. 103. Amer. Math. Soc., Providence, 2005. 28 [40] M. Rørdam, Classification of Cuntz-Krieger algebras, K-Theory 9 (1995), 31 -- 58. [41] M. Rørdam, private communication. [42] J. Spielberg, Semiprojectivity for certain purely infinite C ∗-algebras, Trans. Amer. Math. Soc. 361 (2009), 2805 -- 2830. [43] W. Szyma´nski, The range of K-invariants for C ∗-algebras of infinite graphs, Indi- ana Univ. Math. J. 51 (2002), 239 -- 249. [44] W. Szyma´nski, On semiprojectivity of C ∗-algebras of directed graphs, Proc. Amer. Math. Soc. 130 (2002), 1391 -- 1399. [45] W. Szyma´nski, On localized automorphisms of the Cuntz algebras which preserve in 'New Development of Operator Algebras', R.I.M.S. the diagonal subalgebra, Kokyuroku 1587 (2008), 109 -- 115. [46] J. Zacharias, Quasi-free automorphisms of Cuntz-Krieger-Pimsner algebras, in C ∗- algebras (Munster, 1999), 262 -- 272, Springer, Berlin, 2000. Roberto Conti Dipartimento di Scienze di Base e Applicate per l'Ingegneria Sezione di Matematica Sapienza Universit`a di Roma Via A. Scarpa 16 00161 Roma, Italy E-mail: [email protected] Jeong Hee Hong Department of Data Information Korea Maritime University Busan 606 -- 791, South Korea E-mail: [email protected] Wojciech Szyma´nski Department of Mathematics and Computer Science The University of Southern Denmark Campusvej 55, DK-5230 Odense M, Denmark E-mail: [email protected] 29
1004.3004	2	1004	2010-10-11T10:22:13	Covariant representations of subproduct systems	[ "math.OA" ]	A celebrated theorem of Pimsner states that a covariant representation $T$ of a $C^$-correspondence $E$ extends to a $C^$-representation of the Toeplitz algebra of $E$ if and only if $T$ is isometric. This paper is mainly concerned with finding conditions for a covariant representation of a \emph{subproduct system} to extend to a $C^*$-representation of the Toeplitz algebra. This framework is much more general than the former. We are able to find sufficient conditions, and show that in important special cases, they are also necessary. Further results include the universality of the tensor algebra, dilations of completely contractive covariant representations, Wold decompositions and von Neumann inequalities.	math.OA	math	COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS AMI VISELTER Abstract. A celebrated theorem of Pimsner states that a covariant representation T of a C ∗-correspondence E extends to a C ∗-representation of the Toeplitz algebra of E if and only if T is isometric. This paper is mainly concerned with finding conditions for a covariant representation of a subproduct system to extend to a C ∗- representation of the Toeplitz algebra. This framework is much more general than the former. We are able to find sufficient conditions, and show that in important special cases, they are also necessary. Further results include the universality of the tensor algebra, dilations of completely contractive covariant representations, Wold decompositions and von Neumann inequalities. 0 1 0 2 t c O 1 1 ] . A O h t a m [ 2 v 4 0 0 3 . 4 0 0 1 : v i X r a Contents Introduction 1. Preliminaries and notations 1.1. Basics 1.2. Covariant representations of Hilbert modules 1.3. General subproduct systems 2. The Poisson kernel, representations of the tensor algebra and dilations 3. Pure and relatively isometric covariant representations 4. Fully coisometric covariant representations 5. Conclusions and examples Acknowledgments References 1 4 4 5 6 9 19 24 38 41 41 Introduction This paper treats the interplay between operator algebras and covariant repre- sentations associated with subproduct systems, introduced and studied recently in [40]. This topic is the intersection of two, not completely disjoint, paths of research. The first is the investigation of operator algebras corresponding to product systems. The second concerns with special classes of row contractions, such as commuting or q-commuting or, more generally, "constrained" row contractions. Both paths, briefly 1 COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 2 described hereinafter, greatly generalize much of the Sz.-Nagy -- Foiaş harmonic anal- ysis of Hilbert space contractions ([41]) by considering families of operators, namely covariant representations, satisfying some characteristic conditions. In each of these contexts, certain operator algebras (C∗ and non-selfadjoint) were proved to be uni- versal with respect to the features of the covariant representations. Row contractions have been the subject of extensive study for several decades (e.g. [15, 10, 16, 29, 30, 31, 32, 33, 34, 35]), yielding results on isometric dilations, the Wold decomposition, representations of related operator algebras, the von Neumann inequality and non-commutative functional calculus, to name a few, with an affluence of applications. Motivated by the celebrated paper of Pimsner [28], Muhly and Solel generalized in [20] a large portion of this theory to the framework of covariant representations indexed by Hilbert modules, stimulating a variety of deep follow-ups (see [21, 22, 23, 24, 25]). We sketch out some fundamental results from [28, 20], which are the basis for this paper, referring to §1 for more details. Suppose that E is a C∗-correspondence, that is, a Hilbert C∗-module over a C∗-algebra M equipped with a left M -module structure implemented by a -homomorphism from M to L(E). The (full) Fock space is F (E) := M ⊕ E ⊕ E⊗2 ⊕· · · . Given ζ ∈ E, denote by Sζ the creation (shift) operator in L(F (E)) defined by E⊗n ∋ η 7→ ζ ⊗ η. The C∗-algebra generated by the operators {Sζ : ζ ∈ E} is called the Toeplitz algebra, and is denoted by T (E), and the non-selfadjoint operator algebra generated by these operators is called the tensor algebra, and is denoted by T+(E). The importance of the Toeplitz algebra lies in the fact that it is universal in the following sense. Fix a Hilbert space H. A pair (T, σ) is called a covariant representation of E on H if σ is a C∗-representation of M on H and T (·) is a bimodule map from E to B(H). If (∀ζ, η ∈ E) T (ζ)∗T (η) = σ(hζ, ηi), (0.1) then the covariant representation (T, σ) is said to be isometric, and if T (·) is com- pletely contractive, then (T, σ) is said to be completely contractive. (For example, when M = C and E = Cd, there is a bijection between completely contractive, co- variant representations of E on H and row contractions of length d over H given by T 7→ (T (e1), . . . , T (ed)); now (0.1) is equivalent to T (e1), . . . , T (ed) being isometries with orthogonal ranges.) It is proved in [28, Theorem 3.4] (with minor differences) that there exists a C∗-representation of T (E) on H, mapping Sζ to T (ζ), if and only if (T, σ) is isometric. In other words, the Toeplitz algebra is the universal C∗-algebra generated by an isometric covariant representation of E. Concurrently, a theory was sought to match row contractions subject to restric- tions. For example, row contractions consisting of commuting ([2, 13, 4, 35, 7]) or COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 3 q-commuting ([3, 6, 12]) operators were examined. The general case of constrained row contractions followed ([36, 37]), and proved to have many applications. Subproduct systems provide the means to unify these two theories. Assume that is a family of Hilbert modules over M = X(0) such that X(n + m) X = (X(n))n∈Z+ is an orthogonally complementable sub-bimodule of X(n)⊗ X(m) for all n, m ∈ Z+. This is a subproduct system in the "standard" form. Setting E := X(1) we have X(n) ⊆ E⊗n for all n. Letting pn denote the orthogonal projection of E⊗n onto X(n), the X-shifts are defined over the X-Fock space FX := M ⊕ E ⊕ X(2)⊕ X(3)⊕ . . . ⊆ F (E) by SX n (ζ)η := pn+m(ζ ⊗ η) for ζ ∈ X(n), η ∈ X(m). The Toeplitz algebra T (X) and the tensor algebra T+(X) are now defined to be the C∗- and non-selfadjoint algebras, respectively, generated by the X-shifts in L(FX). Covariant representations of X on H are defined is a suitable manner where Tn : X(n) → B(H) T = (Tn)n∈Z+ for all n. Now, for instance, [28, 20] correspond to product systems (X(n) = E⊗n, so that actually X(n+m) = X(n)⊗X(m) for all n, m); and there is a bijection between row contractions of d commuting operators and completely contractive, covariant representations of the symmetric subproduct system defined by X(n) := (Cd)sn, n ∈ Z+. The main goal of this paper is to generalize the above-mentioned theorem of Pim- sner to this setting. That is, if T is a covariant representation of a subproduct system X on H, when is there a C∗-representation of T (X) on H mapping SX n (ζ) to Tn(ζ)? If this holds, T is said to extend to a C∗-representation. The difficulty starts with the fact that the X-shifts and their adjoints do not satisfy any "isometricity" relation in the spirit of (0.1) (as seen even in the simple example of the symmetric sub- product system). An essential step in the proof of [28, Theorem 3.4] employs (0.1) to reduce every composition of shifts and their adjoints to an operator of the form Sζ1 · · · SζnS∗ηm · · · S∗η1. Such computation is evidently not possible in the subproduct systems case, and new ideas will be utilized to establish our results. The structure of the paper is the following. After presenting some preliminaries in §1, we introduce in §2 the Poisson kernel suitable for our context. It is then used to prove the universality of the tensor algebra T+(X) -- every completely con- tractive, covariant representation extends to a completely contractive representation of T+(X) -- and to prove a dilation theorem for completely contractive, covariant representations. We next address the question of C∗-representability of covariant representations, which is the main part of the work. In view of our dilation theorem, we divide the problem into two: pure covariant representations are handled in §3, while fully coisometric covariant representations are dealt with in §4. In both cases, sufficient conditions are found for C∗-representability. Although we are not able to COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 4 prove them necessary generally, we demonstrate that this is not far from being true by providing examples of important special cases in which equivalence holds. The last section §5 is devoted to deriving some conclusions from the results of the paper and giving more examples. 1. Preliminaries and notations 1.1. Basics. Throughout this paper, H denotes an arbitrary (complex) Hilbert space, and B(H) denotes the Banach space of bounded operators over H. We assume that the reader is familiar with the basic theory of Hilbert C∗-modules, which can be found, e.g., in [18, Ch. 1-4]. As is customary, the action of the C∗- algebra on the Hilbert C∗-module is on the right. The Banach space of all adjointable (bounded) module maps between two Hilbert C∗-modules E, F over the same algebra is denoted by L(E, F ), and L(E) stands for the C∗-algebra L(E, E). A Hilbert W ∗- module is a Hilbert C∗-module over a von Neumann algebra that is self-dual (see [26, §3], [19, §2.5]). In this case, L(E) is a W ∗-algebra. We use the notation (·,·) for the inner product in Hilbert spaces and h·,·i for the rigging in Hilbert modules. A Hilbert C∗-module E over a C∗-algebra M is called a C∗-correspondence in case it is also equipped with a left M -module structure, implemented by a - homomorphism ϕ : M → L(E); that is, a · ζ := ϕ(a)ζ for a ∈ M , ζ ∈ E. In particular, ha · ζ, ηi = hζ, a∗ · ηi. E is called a W ∗-correspondence when E is a Hilbert W ∗-module (in particular, M is a von Neumann algebra) and ϕ is normal. Let M , N be C∗-algebras, and let E, F be Hilbert C∗-modules over M , N re- spectively. Suppose also that σ : M → L(F ) is a -homomorphism. The (interior) tensor product of E and F , E ⊗σ F , is the Hilbert C∗-module over N that contains the algebraic tensor product of E and F , balanced by σ (i.e., (ζa)⊗ η = ζ ⊗ (σ(a)η)), as a dense subspace, with the rigging defined by (∀ζ1, ζ2 ∈ E, η1, η2 ∈ F ) hζ1 ⊗ η1, ζ2 ⊗ η2iE⊗σF := hη1, σ(hζ1, ζ2i)η2iF . Two special cases are worth mentioning. First, if F = H is a Hilbert space, i.e. N = C, then E⊗σ H is also a Hilbert space. Second, if E, F are C∗-correspondences over the same algebra M (and ϕF implements the left action of M on F ), so is E ⊗ F := E ⊗ϕF F . Assume that E1, E2 are Hilbert C∗-modules over M and F1, F2 are Hilbert C∗- modules over N . Suppose also that σj : M → L(Fj), j = 1, 2, are -homomorphisms. If T ∈ L(E1, E2), S ∈ L(F1, F2) and S is an M -module map, that is, Sσ1(a) = σ2(a)S, then there exists a unique operator T ⊗ S ∈ L(E1 ⊗σ1 F1, E2 ⊗σ2 F2) satisfy- ing (T ⊗ S)(ζ ⊗ η) = (T ζ)⊗ (Sη) for ζ ∈ E1, η ∈ F1. Moreover, kT ⊗ Sk ≤ kTk·kSk. COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 5 Tensor products of Hilbert W ∗-modules are discussed more thoroughly in Remark 1.8. For a general (not necessarily selfadjoint) operator algebra A , denote by A 1 its trivial unitization A ⊕ CI if it is not already unital (otherwise, set A 1 := A ). It should be emphasized that the C∗-algebras we consider are not necessarily unital. However, all operator algebra representations, and particularly C∗-algebra represen- tations, are assumed nondegenerate by convention. The majority of the theory developed in this paper is valid in both the C∗-algebra setting and the W ∗ (von Neumann)-algebra setting. To avoid burdensome repetitions in statements, we shall use the letter A where either C or W may be used, yielding, e.g., an A∗-algebra, a Hilbert A∗-module, etc. Unless specifically told otherwise, all definitions, propositions, etc. are valid in both contexts. 1.2. Covariant representations of Hilbert modules. Fix an A∗-correspondence E over an A∗-algebra M . Definition 1.1 ([20, Definition 2.11], [23, Definition 2.15]). A pair (T, σ) is called a covariant representation of E on H if: (1) σ is a C∗-representation of M on H. (2) T is a linear mapping from E to B(H). (3) T is a bimodule map with respect to σ, i.e., T (ζa) = T (ζ)σ(a) and T (aζ) = σ(a)T (ζ) for all ζ ∈ E and a ∈ M . In the W ∗-setting we require, in addition, that σ be normal and that T be continuous with respect to the σ-topology of E and the ultraweak topology on B(H) (see [5, pp. 201-202]). A covariant representation (T, σ) is said to be completely contractive if T is a com- pletely contractive map, when E is viewed as a subspace of the "linking algebra" of M and E (see [20, pp. 398-399]). It is said to be isometric when T (ζ)∗T (η) = σ(hζ, ηi) for all ζ, η ∈ E. An isometric, covariant representation is necessarily completely contractive. Given a completely contractive, covariant representation (T, σ) of E on H, define an operator eT : E ⊗σ H → H by for ζ ∈ E, h ∈ H. It turns out that eT is a well-defined contraction, and that if the left action of M on E is given by the mapping ϕ : M → L(E), then eT (ζ ⊗ h) := T (ζ)h (∀a ∈ M ) eT (ϕ(a) ⊗ IH) = σ(a)eT . (1.1) (1.2) COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 6 Moreover, we have the following useful lemma, for which we assume that σ is a fixed (normal, in the W ∗-setting) representation of M on H. Lemma 1.2 ([20, Lemma 3.5]; [23, Lemma 2.16]; see also the succeeding remark). contractive, covariant representations (T, σ) of E on H and the set of all contractions The mapping (T, σ) 7→ eT defined above is a bijection between the set of all completely eT : E ⊗σ H → H satisfying (1.2). Additionally, (T, σ) is isometric if and only if eT is an isometric operator. Remark 1.3. It is a consequence of the lemma that, in the W ∗-setting, the ultraweak continuity of T is automatic when σ is normal. 1.3. General subproduct systems. In this subsection we let S denote a general Abelian monoid (a semigroup with identity element 0). Definition 1.4 ([40, Definition 1.1]). Let M be an A∗-algebra. A family X = (X(s))s∈S of A∗-correspondences over M is called a subproduct system over M if the following conditions hold: (1) X(0) = M . (2) For all s, t ∈ S there exists a coisometric, adjointable bimodule mapping Us,t : X(s) ⊗ X(t) → X(s + t), so that: a. The maps Us,0 and U0,s are given by the right and left actions of M on X(s), respectively; that is, Us,0(ζ ⊗ a) = ζa and U0,s(a ⊗ ζ) = aζ for all a ∈ M , ζ ∈ X(s). b. The following associativity condition holds for all s, t, r ∈ S: Us+t,r(cid:0)Us,t ⊗ IX(r)(cid:1) = Us,t+r(cid:0)IX(s) ⊗ Ut,r(cid:1) . Remark 1.5. Every Hilbert C∗-module E over M is identified with E ⊗ M (via ζ ⊗ a 7→ ζa). However, when E is a C∗-correspondence, M ⊗ E may only be embedded in E (via a⊗ ζ 7→ aζ). Equality holds if and only if E is essential as a left M -module, that is, the *-homomorphism ϕ implementing the left multiplication is nondegenerate: ϕ(M )E is dense in E. If M is unital (e.g. in the W ∗-setting), this is equivalent to having ϕ(IM ) = IL(E). Definition 1.4 implies that X(t) is essential for all t ∈ S. Example 1.6. A product system is a subproduct system for which all mappings Us,t are unitary. COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 7 Example 1.7 (Subproduct system of Hilbert spaces). Suppose that M = C and X(s) is a Hilbert space for all s ∈ S. This important special case has been recently studied in [8]. We refer the reader to [40] for many other examples of subproduct systems. Remark 1.8. We make a few important comments on tensor products and direct sums of W ∗-correspondences and bounded operators on them, which will be used, sometimes tacitly, throughout. The following facts and related definitions are taken from [26, §3] unless stated otherwise. Fix a von Neumann algebra M . For an arbitrary Hilbert C∗-module X over M , recall that X′, the linear space of bounded module maps from X to M , is a self-dual Hilbert C∗-module over M , that is, X′ is a Hilbert W ∗-module. We refer to X′ as the "self-dual completion" of X. Hilbert W ∗-modules possess a variety of desirable properties. For example, every bounded operator (module map) over a Hilbert W ∗-module is automatically adjointable, and the space of all such operators is a W ∗-algebra. We mention some results that demonstrate the connection between X and X′. Firstly, for an arbitrary Hilbert C∗-module X over M , let the s-topology of X (see when pω is defined [5, p. 202]) be defined by the family of semi-norms (pω)ω∈M + by X ∋ x 7→ ω(hx, xi)1/2. Then X is dense in X′ relative to the s-topology of the latter (see [27, Lemma 2.3] and [5, Proposition 1.4]). Secondly, assume that X, Y are Hilbert C∗-modules over M . Then every bounded module map T : X → Y admits a unique extension to a bounded, adjointable operator T : X′ → Y ′. Suppose that E, F are W ∗-correspondences over M . By the "tensor product" E⊗F we do not mean the usual interior tensor product of [18, Ch. 4] (which is denoted by E⊗C ∗ F for the moment), but rather its self-dual completion. There are hence several essential subtleties that one has to pay attention to. For example, the subspace E ⊗alg F := span{ζ ⊗ η : ζ ∈ E, η ∈ F} is not necessarily norm-dense in E ⊗ F , but rather s-dense there. This "problem" is circumvented when defining bounded operators from E ⊗ F to another Hilbert W ∗-module G by the unique extension property mentioned above. As to comparison of operators, if T1, T2 : E ⊗ F → G E⊗C∗ F , so are bounded module maps that agree on E ⊗alg F , then (T1) that T1 = T2 by virtue of the uniqueness of the extension of the operators Ti from E ⊗C ∗ F to E ⊗ F . If X is a Hilbert C∗-module over M , H is a Hilbert space and σ is a normal representation of M on H, then we may form the Hilbert space X′ ⊗σ H in the usual way. Thus, using the definition of the s-topology and the normality of σ, the subspace X ⊗alg H := span{ζ ⊗ h : ζ ∈ X, h ∈ H} is readily seen to be weakly dense in X′ ⊗σ H. Since a linear subspace is convex, X ⊗alg H is actually norm-dense in ∗ E⊗C∗ F = (T2) COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 8 X′ ⊗σ H. Consequently, X′ ⊗σ H = X ⊗σ H. This observation is relevant, e.g., when E, F are W ∗-correspondences and X = E ⊗C ∗ F . Finally, we comment that if (Ei)I is a family of Hilbert W ∗-modules over M , its Its self-dual completion is called the ultraweak direct sum of (Ei)I, and is also denoted direct sumLI Ei (which is a Hilbert C∗-module over M ) is not necessarily self-dual. byLI Ei where there is no chance of confusion. If Ei is a W ∗-correspondence over M for all i ∈ I thenLI Ei is a W ∗-correspondence over M too. Definition 1.9 ([40, Definition 1.5]). Assume that M is an A∗-algebra and X = (X(s))s∈S is a subproduct system over M . Suppose that T = (Ts)s∈S is a family of linear maps Ts : X(s) → B(H), and denote σ := T0. Then T is called a completely contractive, covariant representation of X on H if: (1) For each s ∈ S, the pair (Ts, σ) is a completely contractive, covariant rep- resentation of the A∗-correspondence X(s) on H in the sense of Definition 1.1. (2) For each s, t ∈ S, ζ ∈ X(s) and η ∈ X(t), Ts+t(Us,t(ζ ⊗ η)) = Ts(ζ)Tt(η). (1.3) An alternative formulation of Condition (2) is obtained by using the bijection equivalent to eTs+t(Us,t ⊗ IH) = eTs(IX(s) ⊗ eTt) operator from X(s) ⊗ (X(t) ⊗σ H) to X(s) ⊗σ H only because (1.2) is satisfied for T = Tt. Operator tensor products of this form will be frequently used throughout. described in Lemma 1.2. For s ∈ S we define the contractions eTs : X(s) ⊗σ H → H by eTs(ζ ⊗ h) := Ts(ζ)h for all ζ ∈ X(s), h ∈ H. The equality in (1.3) is now (see Remark 1.8 in this connection). Observe that IX(s)⊗eTt is a well-defined bounded Lemma 1.10. Under the assumptions of Definition 1.9, the operator eT0 : M ⊗σH → Proof. That eT0 is an isometry is a matter of straightforward calculation. It is onto since σ is nondegenerate by Definition 1.9, (1). H is unitary. (1.4) (cid:3) Let a subproduct system X = (X(s))s∈S be given. We describe next the most natural example of a completely contractive, covariant "representation" of X. The X-Fock space is the A∗-correspondence over M defined to be the (ultraweak, in the W ∗-setting) direct sum of A∗-correspondences FX :=Ms∈S X(s). COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 9 For s ∈ S, let SX s : X(s) → L(FX) be the linear mapping defined by SX s (ζ)η := U X s,t(ζ ⊗ η) for all t ∈ S, η ∈ X(t). The operators SX FX, and the family SX := (SX we write S instead of SX. s , s 6= 0, are called the creation operators of s )s∈S is called the X-shift. When there is no ambiguity The X-Fock space FX being an A∗-correspondence but not necessarily a Hilbert space, the X-shift may not be a covariant representation as is; this is overcome by composing the mappings SX s with a faithful representation of L(FX) on some Hilbert space. The requirements of Definition 1.9 are satisfied, so the X-shift SX is indeed a completely contractive, covariant "representation" of FX. From (1) and (2)a. of Definition 1.4 we infer that SX 0 = ϕ∞, where ϕ∞(a) maps (b, ζ1, ζ2, . . .) ∈ FX to (ab, a · ζ1, a · ζ2, . . .) ∈ FX for all a ∈ M = X(0). Assum- ing the W ∗-setting, we claim that the "representation" map ϕ∞ is normal, that is, continuous with respect to the ultraweak topologies of M and L(FX). Let (ai)I be a bounded net in M that converges ultraweakly to a ∈ M . Then (ϕ∞(ai))I is bounded in L(FX), thus it converges ultraweakly to some A ∈ L(FX) if and only if n=0 X(n) ⊆ FX (see the proof of Proposition 3.10 and the preceding paragraph in [26] and Remark n=0 ω(hζn, ai · ηni), and since, for n ∈ Z+, X(n) is a W ∗-correspondence, the action of left multiplication M → L(X(n)) is normal, and we deduce that ω(hζ, ϕ∞(ai)ηi) → ω(hζ, ϕ∞(a)ηi), hence ϕ∞ is normal. Lemma 1.11. In the W ∗-setting, the image ϕ∞(M ) is a von Neumann algebra, and ϕ∞ is a homeomorphism of M onto ϕ∞(M ), both endowed with their ultraweak topologies. ω(hζ, ϕ∞(ai)ηi) → ω(hζ, Aηi) for all ω ∈ M∗, k ∈ N and ζ, η ∈Lk 1.8). But ω(hζ, ϕ∞(ai)ηi) =Pk Proof. We have just seen that ϕ∞ : M → L(FX) is normal. Since it is visibly also faithful, all we need is to apply Proposition 7.1.15 and Corollary 7.1.16 of [17]. (cid:3) 2. The Poisson kernel, representations of the tensor algebra and dilations Henceforth we focus on subproduct systems relative to the monoid Z+. This section is dedicated to adapting two theorems to the framework of subproduct sys- tems. The first, Theorem 2.15, implies that every completely contractive, covariant representation extends to a representation of the tensor algebra. It is a direct gen- eralization of, e.g., [40, §8]. The second, Theorem 2.22, is a dilation theorem for completely contractive, covariant representations based on [36, Theorem 2.1]. The following is an adaptation of [40, Definition 6.2] to the C∗-setting as well. COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 10 of A∗-correspondences over an A∗-algebra Definition 2.1. A family X = (X(n))n∈Z+ M is a standard subproduct system if X(0) = M and for all n, m ∈ Z+, X(n + m) is an orthogonally complementable sub-bimodule of X(n)⊗ X(m). Setting E := X(1), we have X(n) ⊆ E⊗n (recall that E⊗0 equals M by definition). Let pn ∈ L(E⊗n) pn, stand for the orthogonal projection of E⊗n onto X(n), and define P to beLn∈Z+ the orthogonal projection of F (E) onto FX. The definition implies that the projections (pn)n∈Z+ are bimodule maps, and (∀n, m ∈ Z+) pn+m = pn+m(IE⊗n ⊗ pm) = pn+m(pn ⊗ IE⊗m). Every standard subproduct system is a subproduct system over M -- simply define X(n)⊗X(m). By virtue of [40, Lemma 6.1], we lose nothing by consid- Un,m := (pn+m) ering only standard subproduct systems. Although this lemma is stated there for the W ∗-setting alone, its proof holds verbatim for the C∗-setting as well. In the W ∗- setting, the assumption that X(n+m) is orthogonally complementable is superfluous (see [19, Proposition 2.5.4]). Remark 2.2. The identifications of (X(n)⊗X(m))⊗X(k) with X(n)⊗(X(m)⊗X(k)) and of X(n)⊗M and M ⊗X(n) with X(n) (cf. Remark 1.5) are implicitly employed in the preceding definition. Example 2.3. A standard product system XE satisfies XE(n) = E⊗n for all n ∈ Z+. Here E may be an arbitrary C∗-correspondence that is essential as a left M -module. In the C∗-setting, T 7→ T1 is a bijection between the completely contractive, covariant representations of XE and the completely contractive, covariant representations of E. Note that if X is a standard subproduct system and P is as above, then SX n (ζ)FX for all n ∈ Z+, ζ ∈ X(n). P SXE n (ζ) = Example 2.4. Subproduct systems of finite dimensional Hilbert spaces constitute a very important special case. They were studied thoroughly in [3, 36, 37], mainly in the context of the non-commutative analytic Toeplitz algebra, which is the weak closure of our tensor algebra (see Definition 2.13), called there the non-commutative disc algebra. Nevertheless, in many respects, the results of the present section can be viewed as generalizing parts of [3] and [36, §2]. Particular examples were explored in [40, 42, 43]. Assume that X = (X(n))n∈Z+ is a standard subproduct system, and T is a com- pletely contractive, covariant representation of X. Then (1.3) takes the form Tn+m(pn+m(ζ ⊗ η)) = Tn(ζ)Tm(η), (2.1) COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 11 and (1.4) is now Taking adjoints in (2.2) we hence obtain eTn+m(pn+m ⊗ IH)X(n)⊗X(m)⊗σH = eTn(IX(n) ⊗ eTm). eT ∗n+m = (IX(n) ⊗ eT ∗m)eT ∗n (formally, in the left side of (2.3) we should have composed eT ∗n+m with the embedding of X(n + m)⊗σ H in X(n)⊗ X(m)⊗σ H, but we omit it for the sake of convenience). In particular, (2.2) (2.3) (2.4) (2.5) Iterating (2.3) yields the formula (∀n ∈ Z+) eT ∗n+1 = (IE ⊗ eT ∗n )eT ∗1 = (IX(n) ⊗ eT ∗1 )eT ∗n . eT ∗n =(cid:0)IX(n−1) ⊗ eT ∗1(cid:1)(cid:0)IX(n−2) ⊗ eT ∗1(cid:1)· · ·(cid:0)IE ⊗ eT ∗1(cid:1)eT ∗1 . (∀n ∈ N) is a decreasing sequence of the strong operator topology). Remark 2.5. Equation (2.4) implies that (cid:8)eTneT ∗n(cid:9)n∈Z+ positive contractions, hence s-limn→∞eTneT ∗n exists (where s-lim stands for limit in pletely contractive, covariant representation T of X is called pure if s-limn→∞eTneT ∗n = Definition 2.6. Let X = (X(n))n∈Z+ be a standard subproduct system. A com- 0. Example 2.7. (1) If D is a Hilbert space and π is a C∗-representation of M on D, then the co- , variant representation induced by π is the family S ⊗ ID := (Sn(·) ⊗ ID)n∈Z+ which consists of the induced representation of the operators Sn(ζ) on FX ⊗π D. S ⊗ ID is a pure, completely contractive, covariant representation of X on FX ⊗π D (2) If (cid:13)(cid:13)eT1(cid:13)(cid:13) < 1, then (2.5) yields that (cid:13)(cid:13)eTneT ∗n(cid:13)(cid:13) ≤ (cid:13)(cid:13)eT1(cid:13)(cid:13)2n for all n, hence T is pure. Definition 2.8. Let X = (X(n))n∈Z+ be a completely contractive, covariant representation of X. Define be a standard subproduct system, and let T 2 ∈ B(H), D := Im ∆∗(T ). 1 ∆∗(T ) := (IH − eT1eT ∗1 ) Then ∆∗(T ) is clearly a positive contraction, and it is invertible in case (cid:13)(cid:13)eT1(cid:13)(cid:13) < 1. When the context is understood, we drop the T and simply write ∆∗. Proposition 2.9. Let X = (X(n))n∈Z+ be a standard subproduct system, and let T be a completely contractive, covariant representation of X. Then ∆∗(T ) ∈ σ(M )′. COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 12 Proof. It is enough to prove that eT1eT ∗1 belongs to σ(M )′. Since T1 is a completely contractive, covariant representation of E = X(1), this equality is a result of [20, Lemma 3.6]. (cid:3) By the last proposition, we may consider the bounded operator IX(n) ⊗ ∆∗(T ) ∈ B(X(n) ⊗σ H), and, in a similar fashion, the operator IFX ⊗ ∆∗(T ) ∈ B(FX ⊗σ H). Furthermore, D reduces σ(a) for all a ∈ M . Upon denoting by σ′ the reduced representation, one can form the tensor product Hilbert space X(n) ⊗σ′ D for each n ∈ Z+, as well as FX ⊗σ′ D. For the sake of simplicity, we write σ instead of σ′. The operator-related Poisson kernel has played an important role since its intro- duction (see [44, 4, 35]). The following definition should come as no surprise in light of the corresponding ones in [36, 25]. be a standard subproduct system, and let T Definition 2.10. Let X = (X(n))n∈Z+ be a completely contractive, covariant representation of X. The Poisson kernel of T is the operator K(T ) : H → FX ⊗σ D defined by K(T )h := Mn∈Z+ (IX(n) ⊗ ∆∗(T ))eT ∗n h for all h ∈ H. It is established in the following proposition that K(T ) is well-defined as an element of B(H,FX ⊗σ D). Proposition 2.11. Under the assumptions of Definition 2.10, K(T ) is a contraction. It is an isometry if and only if T is pure. Proof. Given n ∈ Z+ and h ∈ H, we compute (2.6) (2.7) where the last equality is deduced from (2.2) and (2.4). Thus (cid:13)(cid:13)(IX(n) ⊗ ∆∗(T ))eT ∗n h(cid:13)(cid:13)2 ∞Xn=0(cid:13)(cid:13)(IX(n) ⊗ ∆∗(T ))eT ∗n h(cid:13)(cid:13)2 =(cid:0)eTn(IX(n) ⊗ ∆∗(T )2)eT ∗n h, h(cid:1) =(cid:0)eTn(IX(n) ⊗ (IH − eT1eT ∗1 ))eT ∗n h, h(cid:1) =(cid:0)eTneT ∗n h, h(cid:1) −(cid:0)eTn+1eT ∗n+1h, h(cid:1), n→∞(cid:0)eTneT ∗n h, h(cid:1) ≤ (h, h) = (h, h) − lim (eT0eT ∗0 = IH by Lemma 1.10), and we conclude that K(T ) is a well-defined contrac- tion. COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 13 From (2.6) and the boundedness of K(T ) one infers that ifLZ+ (where yn ∈ X(n) ⊗σ D for all n), then yn = y ∈ FX ⊗σ D For h ∈ H we hence have K(T )∗K(T )h = K(T )∗y = ∞Xn=0 eTn(IX(n) ⊗ ∆∗(T ))yn. ∞Xn=0 eTn(IX(n) ⊗ (IH − eT1eT ∗1 ))eT ∗n h ∞Xn=1 = (IH − eT1eT ∗1 )h + n→∞ eTneT ∗n )h. = (IH − s-lim (eTneT ∗n − eTn+1eT ∗n+1)h (2.8) (2.9) Therefore K(T )∗K(T ) = IH if and only if T is pure (Definition 2.6). be a standard subproduct system, and let Proposition 2.12. Let X = (X(n))n∈Z+ T be a completely contractive, covariant representation of X. The following equality holds for all n ∈ Z+ and ζ ∈ X(n): (cid:3) K(T )∗ (Sn(ζ) ⊗ ID) = Tn(ζ)K(T )∗. Proof. We may consider only vectors of the form η ⊗ h when η ∈ X(m) for some m ∈ Z+ and h ∈ D. Indeed we have, owing to (2.2) and (2.8), K(T )∗ (Sn(ζ) ⊗ ID) (η ⊗ h) = K(T )∗(pn+m(ζ ⊗ η) ⊗ h) = eTn+m(pn+m(ζ ⊗ η) ⊗ ∆∗(T )h) = eTn(ζ ⊗ eTm(η ⊗ ∆∗(T )h)) = Tn(ζ)K(T )∗(η ⊗ h). (cid:3) Definition 2.13. Let X = (X(n))n∈Z+ be a standard subproduct system. We define the tensor algebra T+(X) of X to be the (non-selfadjoint, norm closed) subalgebra of L(FX) generated by {Sn(ζ) : n ∈ Z+, ζ ∈ X(n)}. The Toeplitz algebra T (X) of X is the C∗-subalgebra of L(FX) generated by these operators. We also define E(X) to be the operator system span(T+(X)1T+(X)1∗), the closure being taken in the norm operator topology. Remark 2.14. Since the X-shift is a completely contractive, covariant "representa- tion", we deduce from (2.1) that actually T+(X) = span{Sn(ζ) : n ∈ Z+, ζ ∈ X(n)}. The next result implies that T+(X) is the universal operator algebra generated by a completely contractive, covariant representation of X. be a standard subproduct system, and suppose Theorem 2.15. Let X = (X(n))n∈Z+ that T is a completely contractive, covariant representation of X. Then there exists COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 14 a unique unital, completely positive, completely contractive linear map Ψ : E(X) → B(H) that satisfies (∀n ∈ Z+, ζ ∈ X(n)) and Ψ(Sn(ζ)) = Tn(ζ) and Ψ(Sn(ζ)∗) = Tn(ζ)∗ (2.10) (∀a ∈ T+(X), b ∈ E(X)) Ψ(ab) = Ψ(a)Ψ(b). (2.11) Proof. We begin by handling the "strict" case in which (cid:13)(cid:13)eT1(cid:13)(cid:13) < 1. Define a linear map Ψ : L(FX) → B(H) by Ψ(a) := K(T )∗ (a ⊗ ID) K(T ) for all a ∈ L(FX). Ψ is plainly completely contractive and completely positive. Since T is pure, K(T ) is isometric by Proposition 2.11. Therefore Ψ is unital, and the rest of the stated features are a consequence of Proposition 2.12. Indeed, Ψ(Sn(ζ)Sm(η)∗) = K(T )∗ (Sn(ζ)Sm(η)∗ ⊗ ID) K(T ) = K(T )∗ (Sn(ζ) ⊗ ID) (Sm(η)∗ ⊗ ID) K(T ) = Tn(ζ)K(T )∗K(T )Tm(η)∗ = Tn(ζ)Tm(η)∗ (2.12) for all n, m ∈ Z+, ζ ∈ X(n) and η ∈ X(m) (and the proof of (2.10) is similar). This, the norm continuity of Ψ(·) and Remark 2.14 yield that (2.11) is satisfied as stated. We proceed to the general case. Let T be a completely contractive, covariant representation of X. Given 0 < r < 1, define a completely contractive, covariant representation rT of X (on the same Hilbert space H) by rTn := rnTn for all n ∈ Z+. Then(cid:13)(cid:13)reT1(cid:13)(cid:13) ≤ r, so we can associate with rT a function rΨ as above. By (2.12) we now have rΨ(Sn(ζ)Sm(η)∗) = rTn(ζ)rTm(η)∗ = rn+mTn(ζ)Tm(η)∗ −−−→r→1− Tn(ζ)Tm(η)∗ Tn(ζ) and rΨ(Sn(ζ)∗) −−−→r→1− for all n, m ∈ Z+, ζ ∈ X(n), η ∈ X(m), the limit being taken in the norm operator Tn(ζ)∗. Hence, by topology. Similarly, rΨ(Sn(ζ)) −−−→r→1− Remark 2.14, we conclude that limr→1−(rΨ(a)) exists for all a in a dense subspace of E(X). The family {rΨ : 0 < r < 1}, which consists of completely contractive linear maps, is uniformly bounded by 1. Hence, the limit Ψ(a) := limr→1−(rΨ(a)) exists for all a ∈ E(X), forming a linear operator Ψ : E(X) → B(H), which is obviously unital, completely contractive and completely positive, and satisfies (2.10) and (2.11) (from which it follows that Ψ is unique). (cid:3) Restricting Ψ to T+(X) we get a completely contractive representation of the operator algebra T+(X). The next corollary asserts that by this we obtain a bijection. It generalizes the analogous results for product systems, [20, Theorem 3.10] and [24, COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 15 Theorem 2.9]. In the W ∗-setting, a representation of ϕ∞(M ) on a Hilbert space H is called normal if it is continuous with respect to the ultraweak topologies of L(FX) and B(H). Corollary 2.16. Let X be a standard subproduct system. Then there exists a bi- jection between the completely contractive, covariant representations of X and the completely contractive representations of the tensor algebra T+(X) whose restriction to ϕ∞(M ) is normal in the W ∗-setting. This bijection is implemented as follows: a completely contractive, covariant representation T on a Hilbert space H is mapped to the (unique) completely contractive representation ρ of T+(X) on H determined by the equality ρ(Sn(ζ)) = Tn(ζ) (2.13) for all n ∈ Z+ and ζ ∈ X(n). Additionally, each such representation is completely positive, and it extends to a unital, completely positive, completely contractive linear map over E(X) satisfying (2.10) and (2.11). Proof. By Theorem 2.15, there exists a mapping T 7→ ρ of a completely contractive, covariant representation T of X to a completely contractive representation ρ of T+(X) that satisfies (2.13). The restriction of ρ to ϕ∞(M ) satisfies ρ(ϕ∞(a)) = ρ(S0(a)) = T0(a) = σ(a) for all a ∈ M , and therefore it is normal in the W ∗-setting by virtue of the normality of σ and Lemma 1.11. The last part in the statement of the corollary is also clear from the theorem. This map is injective by (2.13). On the other hand, given a completely contractive representation ρ of T+(X) on a Hilbert space H whose restriction to ϕ∞(M ) is normal in the W ∗-setting, define Tn(ζ) := ρ(Sn(ζ)) for all n ∈ Z+ and ζ ∈ X(n). Since the X-shift is a completely contractive, covariant "representation" of X, one can verify that the conditions of Definition 1.9 are fulfilled, thus obtaining a completely contractive, covariant representation T of X on H (see, in particular, Lemma 1.11 and Remark 1.3, and notice that σ := T0 is nondegenerate by Remark 1.5 because ρ is nondegenerate). Consequently, our mapping T 7→ ρ is also a surjection, and the proof is complete. (cid:3) Our next conclusion is an adaptation of von Neumann's inequality to our setting. For this we require the notion of a polynomial. Definition 2.17. Let X = (X(n))n∈Z+ be a standard subproduct system. A poly- nomial over X is a tuple (α, ζ0, ζ1, . . . , ζM ) with α ∈ C, M ∈ Z+ and ζn ∈ X(n) for all 0 ≤ n ≤ M. If T is a completely contractive, covariant representation of X on n=0 Tn(ζn). p(S) ∈ L(FX) is defined similarly. H, define p(T ) ∈ B(H) to be αIH +PM COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 16 Corollary 2.18. Let X be a standard subproduct system, and suppose that T is If p1, . . . , pt and a completely contractive, covariant representation of X on H. q1, . . . , qt are polynomials over X, then (cid:13)(cid:13)(cid:13)(cid:13) tXi=1 pi(T )qi(T )∗(cid:13)(cid:13)(cid:13)(cid:13)B(H) ≤(cid:13)(cid:13)(cid:13)(cid:13) tXi=1 pi(S)qi(S)∗(cid:13)(cid:13)(cid:13)(cid:13)L(FX ) . Proof. The corollary follows from Theorem 2.15 due to Ψ being contractive. (cid:3) We move on to our dilation theorem, beginning with several preliminaries. be a standard subproduct system, and let Definition 2.19. Let X = (X(n))n∈Z+ T, V be completely contractive, covariant representations of X on the Hilbert spaces H,K respectively, where H is a subspace of K. We say that V is a dilation of T if for all n ∈ Z+ and ζ ∈ X(n), Vn(ζ) leaves H⊥ = K⊖H invariant and PHVn(ζ)H = Tn(ζ) (PH denoting the orthogonal projection of K on H); equivalently: Vn(ζ)∗ leaves H invariant and(cid:0)Vn(ζ)∗(cid:1)H = Tn(ζ)∗. This definition is consistent with [40, Definition 5.4] (when X = Y ) and with the standard one for product systems (e.g. [20, Definition 3.1], [23, Definition 2.18]). As is customary, if H,K are Hilbert spaces and W : H → K is isometric, we regard H as a subspace of K. Definition 2.20. Let X = (X(n))n∈Z+ be a standard subproduct system. A com- pletely contractive, covariant representation T of X on H is said to be isometric or fully coisometric if eTn is isometric or coisometric, respectively, for all n ∈ Z+. In both cases, it is enough to check for n = 1 (see (2.2), (2.3) and Lemma 1.10). If X is a product system, then by Lemma 1.2, T is isometric if and only if (T1, σ) is isometric as a covariant representation of E. As an introduction to Theorem 2.22 and the next sections, we summarize several results on product systems. Theorem 2.21 ([28, Theorem 3.4], [20, Theorems 2.12, 3.3], [21, Theorem 2.9], [23, be a standard product system and let V be a Theorem 2.18]). Let X = (X(n))n∈Z+ covariant representation of X on H. Then: (1) In the C∗-setting: V extends to a C∗-representation if and only if it is iso- metric. (2) If V is isometric, it admits a Wold decomposition Vn(ζ) = (Sn(ζ) ⊗ ID) ⊕ Zn(ζ) (up to unitary equivalence) where D is some Hilbert subspace of H and Z is an isometric, fully coisometric, covariant representation of X. (3) Every completely contractive, covariant representation of X possesses a min- imal dilation to an isometric, covariant representation of X. COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 17 Theorem 2.22. Let X = (X(n))n∈Z+ be a standard subproduct system, and suppose that T is a completely contractive, covariant representation of X on H. Then there exists a dilation V of T to a Hilbert space K with the following properties: (1) There exists a Hilbert space U such that K = (FX ⊗σ D) ⊕ U. (2) There exists a completely contractive, fully coisometric, covariant represen- tation Z of X on U such that Vn(ζ) = (Sn(ζ) ⊗ ID) ⊕ Zn(ζ) for all n ∈ Z+, ζ ∈ X(n). Moreover, U = {0} if and only if T is pure. eTn(IX(n) ⊗ Q)eT ∗n = s-lim m→∞eTn(IX(n) ⊗ eTmeT ∗m)eT ∗n = s-lim (note: Remark 2.5 implies that IX(n)⊗Q = s-lim m→∞ is bounded). Define U := Im Q = Im Q Proof. Let Q := s-limn→∞eTneT ∗n (see Remark 2.5). From (2.2) and (2.3) we obtain m→∞eTn+meT ∗n+m = Q IX(n)⊗eTmeT ∗m because(cid:8)eTmeT ∗m(cid:9)m∈Z+ codomain U instead of H. For all n ∈ Z+, eTneT ∗n belongs to σ(M )′ by [20, Lemma 3.6]. Consequently Q ∈ σ(M )′, and thus U reduces σ(a) for all a ∈ M . Write σU for the representation of M on U sending a ∈ M to σ(a)U ∈ B(U). In the W ∗-setting, σU is normal. Given n ∈ Z+, form the Hilbert space X(n) ⊗σU U, which may be viewed as a closed subspace of X(n) ⊗σ H in the natural way. 2 . Let Y : H → U be the operator Q 2 with (2.14) 1 1 1 Fix n ∈ Z+, and define a linear operator Λn : Im Q 2 → X(n) ⊗σU U by (2.15) Λn(Y h) := (IX(n) ⊗ Y )eT ∗n h 1 for all h ∈ H. To see that this is a legitimate definition of a contraction, note 2 ∈ σ(M )′, so that IX(n) ⊗ Y makes sense as a bounded operator from first that Q X(n) ⊗σ H to X(n) ⊗σU U; and for h ∈ H, we obtain from (2.14) that It follows that Λn extends to a bounded operator from U to X(n) ⊗σU U, which we (cid:13)(cid:13)(IX(n) ⊗ Y )eT ∗n h(cid:13)(cid:13)2 =(cid:0)eTn(IX(n) ⊗ Q)eT ∗n h, h(cid:1) =(cid:0)Qh, h(cid:1) = kY hk2 . also denote by Λn. From (2.15) we deduce that ΛnY = (IX(n) ⊗ Y )eT ∗n . Define eZn := Λ∗n. Then eZ∗nY = (IX(n) ⊗ Y )eT ∗n and Y ∗eZn = eTn(IX(n) ⊗ Y ∗), Y ∗eZneZ∗nY = eTn(IX(n) ⊗ Y ∗Y )eT ∗n = eTn(IX(n) ⊗ Q)eT ∗n = Q (see (2.14)), and therefore (2.16) thus (eZneZ∗nY h, Y h) = (Qh, h) = (Y h, Y h), COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 18 we get (for the last equality, note that Y σ(a) = σU (a)Y due to the commutativity of Q whence we conclude that eZneZ∗n = IU . In addition, for ζ ∈ X(n), a ∈ M and h ∈ U, Y ∗eZn((a · ζ) ⊗ h) = eTn((a · ζ) ⊗ Y ∗h) = σ(a)eTn(ζ ⊗ Y ∗h) = σ(a)Y ∗eZn(ζ ⊗ h) = Y ∗σU (a)eZn(ζ ⊗ h) and σ(a); then take adjoints). Since Y ∗ is injective, we infer that eZn((a · ζ) ⊗ h) = σU (a)eZn(ζ ⊗ h). That is, (1.2) holds with eZn in place of eT . By virtue of Lemma A∗-correspondence X(n) on U that is related to eZn in the usual sense of (1.1). Y ∗eZn+m(pn+m ⊗ IU )X(n)⊗X(m)⊗σU U 1.2 there exists a completely contractive, covariant representation (Zn, σU ) of the In a similar fashion we have (from (2.16)), for n, m ∈ Z+, = eTn+m(IX(n+m) ⊗ Y ∗)(pn+m ⊗ IU )X(n)⊗X(m)⊗σU U = eTn+m(pn+m ⊗ IH)X(n)⊗X(m)⊗σH(IX(n)⊗X(m) ⊗ Y ∗) = eTn(IX(n) ⊗ eTm)(IX(n)⊗X(m) ⊗ Y ∗) = eTn(IX(n) ⊗ Y ∗eZm) = Y ∗eZn(IX(n) ⊗ eZm), establishing that eZn+m(pn+m⊗ IU )X(n)⊗X(m)⊗σU U = eZn(IX(n)⊗eZm) by the injectivity of Y ∗. This is precisely (2.2) with Z replacing T . Moreover, Y ∗eZ0(a ⊗ h) = eT0(a ⊗ Let K := (FX ⊗σ D) ⊕ U. Define an operator W : H → K by W := K(T ) Y !. Y ∗h) = σ(a)Y ∗h = Y ∗σU (a)h for all a ∈ M = X(0) and h ∈ U, so Z0 = σU . The requirements of Definition 1.9 are therefore satisfied, making Z = (Zn)n∈Z+ a completely contractive, covariant representation of X on U. Then for h ∈ H we have from (2.9) kW hk2 = kK(T )hk2 + kY hk2 = khk2 − (Qh, h) + (Qh, h) = khk2 , that is, W is an isometry. Define the sequence V = (Vn)n∈Z+ X(n) → B(K) by Vn(ζ) := Sn(ζ) ⊗ ID 0 0 Zn(ζ) ! of linear maps Vn : for n ∈ Z+, ζ ∈ X(n). Since both the induced representation S ⊗ ID and Z are completely contractive, covariant representations of X, the family V is a completely contractive, covariant representation of X on K. COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 19 From (2.16) one may infer by direct calculation that Zn(ζ)∗Y = Y Tn(ζ)∗. Hence we deduce from Proposition 2.12 that Vn(ζ)∗W = (Sn(ζ)∗ ⊗ ID)K(T ) Zn(ζ)∗Y ! = K(T )Tn(ζ)∗ Y Tn(ζ)∗ ! = W Tn(ζ)∗, (2.17) that is, upon identifying H with its image under W we have Tn(ζ)∗ = (Vn(ζ)∗) . H The family V is, in conclusion, a dilation of T to K, and the proof is complete. (cid:3) Corollary 2.23. The first part of [40, Theorem 8.5] follows as a direct consequence of the last theorem, after recalling that kTkcb in their notations equals(cid:13)(cid:13)eT1(cid:13)(cid:13) in ours (see [20, Lemma 3.5]). This alternative proof does not require the Stinespring dilation theorem. 3. Pure and relatively isometric covariant representations We have seen that every completely contractive, covariant representation of a subproduct system may be "extended" to a completely contractive representation of the tensor algebra (Theorem 2.15). Our goal in this section and the next one is to investigate when it is possible to further extend it to a C∗-representation of the Toeplitz algebra in the sense of the following definition. Definition 3.1. Let X = (X(n))n∈Z+ be a standard subproduct system. We shall say that a completely contractive, covariant representation T of X on H extends to a C∗-representation if there exists a C∗-representation π of T (X) on H such that π(Sn(ζ)) = Tn(ζ) for all n ∈ Z+, ζ ∈ X(n). As indicated in the introduction, we split the problem into two. In the present section, we consider chiefly pure covariant representations. A sufficient condition is established in Theorem 3.8, and it is shown that, in some important special cases, this condition is also necessary (Proposition 3.4 and Corollary 3.10). on the subspace X(n) ⊕ X(n + 1) ⊕ X(n + 2) ⊕ · · · of X(n) ⊗ FX (recall that If n ∈ Z+, the definition of eSn implies that eS∗neSn ∈ L(X(n)⊗FX ) is the projection X(n + k) ⊆ X(n) ⊗ X(k) for all k), and that eSneS∗n ∈ L(FX) is the projection on X(n) ⊕ X(n + 1) ⊕ X(n + 2) ⊕ · · · (considered as a subspace of FX). The product system case (Theorem 2.21) provides further motivation for what follows. Notice that an operator R : H1 → H2 is an isometry if and only if it is a partial isometry, and R∗ is surjective. Lemma 3.2. Let X = (X(n))n∈Z+ be a standard subproduct system, and let T be a completely contractive, covariant representation of X on H. If the maps eTn, n ∈ Z+, are all partial isometries, then the following conditions are equivalent: COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 20 (1) For all n ∈ Z+, (2) For all n ∈ Z+ and ζ ∈ X(n), X(n) ⊗σ Im ∆∗ ⊆ ImeT ∗n . ∆∗Tn(ζ)∗Tn(ζ)∆∗ = σ(hζ, ζi)∆∗. (3.1) (3.2) Observe the resemblance and difference between (3.2) and (0.1). Proof. Since eT1eT ∗1 is a projection, so is ∆∗. Fix n ∈ Z+. Condition (3.1) holds if and only if IX(n) ⊗ ∆∗ ≤ eT ∗neTn; equivalently, (cid:13)(cid:13)eT ∗neTn(ζ ⊗ ∆∗h)(cid:13)(cid:13) =(cid:13)(cid:13)ζ ⊗ ∆∗h(cid:13)(cid:13) for all ζ ∈ X(n), h ∈ H. This is exactly Condition (3.2), for(cid:13)(cid:13)eT ∗neTn(ζ ⊗ ∆∗h)(cid:13)(cid:13) =(cid:13)(cid:13)eTn(ζ ⊗ ∆∗h)(cid:13)(cid:13) and eTn(ζ ⊗ ∆∗h) = Tn(ζ)∆∗h (recall that ∆∗ commutes with the image of σ). maps eTn, n ∈ Z+, are all partial isometries, and one (hence both) of the conditions be a standard subproduct system. A com- Definition 3.3. Let X = (X(n))n∈Z+ pletely contractive, covariant representation T of X is relatively isometric if the presented in Lemma 3.2 is fulfilled. (cid:3) To justify this definition, we offer (apart from Theorem 3.7) the following propo- sition, as well as Corollary 3.10 to follow. Unless specifically told otherwise, we use the notation Q for s-limn→∞eTneT ∗n when T is fixed. Proposition 3.4. Let X = (X(n))n∈Z+ be a standard product system, and let T be a pure, completely contractive, covariant representation of X. Then T is isometric (in the C∗-setting this is equivalent to T extending to a C∗-representation) if and only if T is relatively isometric. Proof. Necessity is clear. isometries, without mentioning it explicitly. Since T is pure, we need to demonstrate Sufficiency. We will use repeatedly the fact that the maps eTn, n ∈ Z+, are partial that E ⊗σ Im(I − Q) ⊆ ImeT ∗1 ; equivalently, that E ⊗σ Im(eTneT ∗n − eTn+1eT ∗n+1) ⊆ ImeT ∗1 for all n ∈ Z+. But eTneT ∗n − eTn+1eT ∗n+1 = eTn(IX(n) ⊗ ∆∗)eT ∗n (cf. (2.7)) and ImeTn(IX(n) ⊗ ∆∗)eT ∗n ⊆ ImeTn(IX(n) ⊗ ∆∗), so it is enough to show that (cid:13)(cid:13)(cid:13)eT1(cid:16)ζ ⊗ (eTn(η ⊗ ∆∗h))(cid:17)(cid:13)(cid:13)(cid:13) =(cid:13)(cid:13)(cid:13)ζ ⊗ (eTn(η ⊗ ∆∗h))(cid:13)(cid:13)(cid:13) for all ζ ∈ E, η ∈ E⊗n and h ∈ H. The equality eT1(IE ⊗ eTn) = eTn+1 holds as X is a greater or equal to the right side because IE ⊗ eTn is a contraction. As the reverse product system, and by (3.1), the left side of (3.3) equals kζ ⊗ η ⊗ ∆∗hk, and is thus inequality is obvious, we are done. (3.3) (cid:3) Lemma 3.5. Let X = (X(n))n∈Z+ be a standard subproduct system such that E = X(1) is a Hilbert space (thus so are all the spaces X(n), n ∈ N). Assume that T is COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 21 a completely contractive, covariant representation of X on H. Then for n ∈ Z+ and a fixed orthonormal base (ζi)I of X(n) we have eTn = (Tn(ζi))i∈I as a row vector and eT ∗n = (Tn(ζi)∗)i∈I as a column vector, and consequently (3.4) (3.5) eTneT ∗n =Xi∈I Tn(ζi)Tn(ζi)∗, where in (3.4) and (3.5) the convergence is in the strong operator topology. Proof. If i, j ∈ I and h, k ∈ H, then hζi ⊗ h, ζj ⊗ ki = hh, σ(hζi, ζji)ki = hh, ki 0 i = j else. We conclude that X(n) ⊗σ H =Li∈I ζi ⊗ H and ζi ⊗ H ∼= H for all i ∈ I. The first part of (3.4) is therefore an immediate result of the definition of eTn, and the second part follows from the first. The rest of the lemma is an easy consequence. (cid:3) be a standard subproduct system such that Proposition 3.6. Let X = (X(n))n∈Z+ E = X(1) is a finite dimensional Hilbert space, and suppose that T is a completely contractive, covariant representation of X on H. If T extends to a C∗-representation then it is relatively isometric. Sn(ζi)Sn(ζi)∗) Proof. Let π : T (X) → B(H) be a C∗-representation as in Definition 3.1. We first demonstrate that eTn is a partial isometry for n ∈ Z+. This is equivalent to eTneT ∗n ∈ B(H) being a projection. Indeed, for a fixed orthonormal base (ζi)I of X(n), we have from (3.5) eTneT ∗n =Xi∈I Tn(ζi)Tn(ζi)∗ =Xi∈I π(Sn(ζi)Sn(ζi)∗) = π(Xi∈I establish (3.2) we must verify that Let (ηj)J denote an orthonormal base of E. From Definition 2.8 and the foregoing (the last equality holds since all sums are finite). But Pi∈I Sn(ζi)Sn(ζi)∗ = eSneS∗n, which is a projection, as indicated above. Hence eTneT ∗n is also a projection. we deduce that ∆∗ = IH −Pj∈J T1(ηj)T1(ηj)∗. Fix n ∈ Z+ and ζ ∈ X(n). To ∗ := IFX −eS1eS∗1 = IFX −Pj∈J S1(ηj)S1(ηj)∗ ∈ T (X). Since π is a represen- ∆∗Tn(ζ)∗Tn(ζ)∆∗ = kζk2 ∆∗. Write ∆X tation of T (X), one has ∆∗ = π(∆X ∗ ∆X . But ∆X ∗ ∗ mand C = X(0), so that Sn(ζ)∗Sn(ζ)∆X ∗ ), and it is therefore enough to ascertain that is the projection of FX on the direct sum- = kζk2 ∆X , and the proof is complete. (cid:3) ∗ Sn(ζ)∗Sn(ζ)∆X ∗ = kζk2 ∆X ∗ COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 22 We now present a Wold decomposition for relatively isometric covariant represen- tations. Theorem 3.7. Let X = (X(n))n∈Z+ be a standard subproduct system, and suppose that T is a completely contractive, covariant representation of X on H. Then the following are equivalent: (1) T is relatively isometric. (2) There exist a Hilbert space U, a unitary W : H → K := (FX ⊗σ D) ⊕ U and of X on U such a fully coisometric, covariant representation Z = (Zn)n∈Z+ that W Tn(ζ)W −1 = (Sn(ζ) ⊗ ID) ⊕ Zn(ζ) for all n ∈ Z+, ζ ∈ X(n). Moreover, U may be chosen to be {0} if and only if T is pure. Proof. (1) ⇒ (2). We examine the influence of the assumption that T is relatively isometric on several parts of the proof of Theorem 2.22, commencing by claiming that K(T ) is onto FX ⊗σ D. Our assumptions yield that (eTneT ∗n − eTn+1eT ∗n+1)n∈Z+ is an orthogonal sequence of projections in B(H), whose sum is I − Q. Since eT1eT ∗1 is a projection, we have ∆∗(T ) = IH − eT1eT ∗1 (which is also a projection). Moreover, from (2.7) we learn that ker(cid:0)(IX(n) ⊗ ∆∗(T ))eT ∗n(cid:1) = ker(cid:0)eTneT ∗n − eTn+1eT ∗n+1(cid:1). By (3.1) we obtain X(n)⊗σ D = X(n)⊗σ Im ∆∗(T ) ⊆ ImeT ∗n , and consequently X(n)⊗σ D = Im(cid:0)(IX(n) ⊗ ∆∗(T ))eT ∗n(cid:1). From the foregoing we conclude that X(n)⊗σ D ⊆ Im K(T ) for all n ∈ Z+. But K(T ) is a partial isometry as K(T )∗K(T ) = I − Q by (2.9). Hence its range is closed, thus it equals FX ⊗σ D. The operator Q is a projection, so that U = Im Q and Y : H → U is the coisometry mapping h ∈ H to Qh. Since (ker K(T ))⊥ = Im(I − Q) = ker Y , we have Im W = Im K(T )⊕Im Y = K. In conclusion, W is unitary, so we can read our desired formula for W Tn(ζ)W −1 from (2.17). (2) ⇒ (1). For n ∈ Z+, the map eTn is a partial isometry because it equals W −1(cid:0)(eSn ⊗ ID) ⊕ eZn(cid:1)(IX(n) ⊗ W ), both eSn and eZn are partial isometries and W is unitary. Moreover, as Z is fully coisometric, we obtain ∆∗(T ) = W −1 IFX⊗σD − (eS1eS∗1) ⊗ ID 0 ! W. = W −1 ∆∗(S) ⊗ ID 0 0 0 0 IU − eZ1eZ∗1 ! W COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 23 Hence, using the fact that σ(·) = T0(·) = W −1(cid:0)(ϕ∞(·) ⊗ ID) ⊕ Z0(·)(cid:1)W , we get Zn(ζ)∗ ! · ∆∗(T )Tn(ζ)∗Tn(ζ)∆∗(T ) = W −1 ∆∗(S) ⊗ ID 0 0 ! W 0 ! W 0 ! Sn(ζ)∗ ⊗ ID Zn(ζ) ! ∆∗(S) ⊗ ID 0 · Sn(ζ) ⊗ ID = W −1 (∆∗(S)Sn(ζ)∗Sn(ζ)∆∗(S)) ⊗ ID 0 = W −1 (ϕ∞(hζ, ζi)∆∗(S)) ⊗ ID 0 0 ! W 0 0 0 0 0 0 0 0 = σ(hζ, ζi)∆∗(T ), proving that T is relatively isometric. (cid:3) Remark. In the proof of (2) ⇒ (1) we employed the operator IX(n) ⊗ W . That it is a well-defined, bounded operator is a consequence of the hypotheses. be a standard subproduct system, and suppose Theorem 3.8. Let X = (X(n))n∈Z+ that T is a completely contractive, covariant representation of X. If T is pure and relatively isometric, then it is induced, and it extends to a C∗-representation. Proof. T being pure is equivalent to being able to choose U = {0}. When this is the case, Tn(ζ) = W −1(Sn(ζ) ⊗ ID)W for all n ∈ Z+, ζ ∈ X(n), i.e., T is (unitarily equivalent to) an induced representation. It therefore extends to a C∗-representation in the natural fashion. (cid:3) Corollary 3.9. Let X = (X(n))n∈Z+ be a standard subproduct system such that X(n0) = {0} for some n0 ∈ N. Then every completely contractive, covariant repre- sentation of X that is relatively isometric extends to a C∗-representation. Proof. Every such covariant representation is pure by Definition 2.6 because X(n) = {0} for all n ≥ n0. (cid:3) Combining Theorem 3.8 and Proposition 3.6 we infer the following. be a standard subproduct system such that Corollary 3.10. Let X = (X(n))n∈Z+ E = X(1) is a finite dimensional Hilbert space (see Example 2.4), and suppose that T is a completely contractive, covariant representation of X. If T is pure, then the following conditions are equivalent: (1) T extends to a C∗-representation. (2) T is relatively isometric. COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 24 We wish to emphasize the novelty of this corollary even in the well-studied case of Arveson's symmetric subproduct system SSPd ([4]; see also Example 4.10). Indeed, while it has already been known that a pure, completely contractive, covariant repre- sentation of SSPd extends to a C∗-representation if and only if it is an induced repre- sentation of the SSPd-shifts -- the latter is not a "checkable" condition. On the other hand, our condition of being relatively isometric is very concrete for the subproduct systems considered in the last corollary (and particularly SSPd) owing to Lemma 3.5, or the following observation: if d = dim E and {e1, . . . , ed} is some orthonormal base of E, define Ti := T1(ei), 1 ≤ i ≤ d. For α = (α1, . . . , an) ∈ {1, . . . , d}n, set Tα := Tα1 · · · Tαn. Then [40, Proposition 6.9] implies that eTneT ∗n = Xα∈{1,...,d}n TαT ∗α for all n ∈ N. In conclusion, the conditions in Definition 3.3 can be easily written in terms of the operators T1, . . . , Td. 4. Fully coisometric covariant representations A fully coisometric, covariant representation of a subproduct system X such that E = X(1) is a finite dimensional Hilbert space possesses a dilation to a fully coiso- metric, covariant representation of X that extends to a C∗-representation. This is obtained by a suitable adaptation of [36, Theorem 2.4] and [40, Proposition 7.2]. The proof uses the pair Arveson's extension theorem + Stinespring's dilation theo- rem, as well as properties of the Toeplitz algebra T (X), which depend on E being a finite dimensional Hilbert space. Additionally, as Arveson's extension theorem does not supply a concrete construction for the extension, applying the same techniques to covariant representations of general subproduct systems, one encounters troubles when trying to prove the W ∗-setting continuity condition mentioned in Definition 1.1. For these reasons, using the same lines to prove a generalization of the afore- mentioned dilation theorem seems impossible. This section is devoted to dilations and C∗-extendability of fully coisometric, covariant representations, complementing the developments of the previous section. Suppose that T is a completely contractive, covariant representation of a sub- It is a special case of Theorem 2.15 that kTn(ζ)Tm(ξ)∗k ≤ product system X. kSn(ζ)Sm(ξ)∗k for n, m ∈ Z+, ζ ∈ X(n), ξ ∈ X(m). Assume now that T extends to a C∗-representation. Since this representation maps every "polynomial" of elements of the form Sn(ζ) and their adjoints to the same polynomial with T replacing S, the norm of the latter is less that or equal to the norm of the former. Particularly, kTn(ζ)∗Tm(ξ)k ≤ kSn(ζ)∗Sm(ξ)k. In Theorem 4.1 it is established that every fully COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 25 coisometric, covariant representation admits a dilation that satisfies the last inequal- ity. In Theorem 4.3 we present sufficient conditions for a fully coisometric, covariant representation to extend to a C∗-representation in the C∗-setting. Theorem 4.1. Let X = (X(n))n∈Z+ be a standard subproduct system, and suppose that T is a fully coisometric, covariant representation of X. Then T admits a dilation to a completely contractive, covariant representation V of X, for which the inequality kVn(ζ)∗Vm(ξ)k ≤ kSn(ζ)∗Sm(ξ)k (4.1) holds for all n, m ∈ Z+, ζ ∈ X(n), ξ ∈ X(m). The motivation for using inductive limits in the following proof was [23, Theorem 3.7], where this technique was used for product systems with a positive real parameter instead of the usual Schäffer matrix construction (cf. [23, Theorem 2.18]). Proof. We divide the proof into a few steps. Definition of K. For n, m ∈ Z+, n ≤ m, define an operator un,m : X(n) ⊗σ H → X(m) ⊗σ H by Then un,m is a contractive M -module map. Moreover, if n ≤ m ≤ k, we have un,m := (pm ⊗ IH)(IX(n) ⊗ eT ∗m−n). um,kun,m = (pk ⊗ IH)(IX(m) ⊗ eT ∗k−m)(pm ⊗ IH)(IX(n) ⊗ eT ∗m−n) = (pk ⊗ IH)(pm ⊗ IX(k−m) ⊗ IH)(IX(n) ⊗ IX(m−n) ⊗ eT ∗k−m)(IX(n) ⊗ eT ∗m−n) = (pk ⊗ IH)(IX(n) ⊗ eT ∗k−n) = un,k n≤m (cid:17) is an inductive system of Hilbert spaces that are (cid:16)(X(n) ⊗σ H)n∈Z+, (un,m)n,m∈Z+ (the passage from the second line to the third was done by virtue of (2.3)). Therefore also M -modules. We define K to be its inductive limit, and denote by un : X(n) ⊗σ H → K the canonical contractive M -module maps satisfying umun,m = un for all n ≤ m. Upon identifying X(0) ⊗σ H = M ⊗σ H with H in the usual sense, we see that u0,n = eT ∗n . Since T is fully coisometric, the map u0,n is an isometry for each n ∈ Z+. This implies that u0 : H → K is an isometry. In the course of the proof we shall use repeatedly the following two analytic ("uni- versality") properties of the inductive limit: • The unionSn≥n0 Im un is dense in K for all n0 ∈ Z+. • If n, m ∈ Z+, x ∈ X(n) ⊗σ H and y ∈ X(m) ⊗σ H, then (un,ℓx, um,ℓy)X(ℓ)⊗σH. (unx, umy)K = lim ℓ→∞ COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 26 Definition of the operators Vn(ζ). Given n ∈ Z+ and ζ ∈ X(n), we define an operator Vn(ζ) ∈ B(K) by first setting Vn(ζ)um(η ⊗ h) := un+m(Sn(ζ)η ⊗ h) = un+m(pn+m(ζ ⊗ η) ⊗ h) (4.2) for m ∈ Z+, η ∈ X(m) and h ∈ H. We observe that if ℓ ≥ n + m, then un+m,ℓ(Sn(ζ)η ⊗ h) = (pℓ ⊗ IH)(IX(n+m) ⊗ eT ∗ℓ−(n+m))(pn+m(ζ ⊗ η) ⊗ h) = (pℓ ⊗ IH)(pn+m ⊗ eT ∗ℓ−(n+m))(ζ ⊗ η ⊗ h) = (pℓ ⊗ IH)(IX(n) ⊗ IX(m) ⊗ eT ∗ℓ−(n+m))(ζ ⊗ η ⊗ h) = (pℓ ⊗ IH)(ζ ⊗ um,ℓ−n(η ⊗ h)) = (Sn(ζ) ⊗ IH)um,ℓ−n(η ⊗ h). By the fact that kSn(ζ) ⊗ IHkB(FX⊗σH) ≤ kζk, we have (4.3) kun+m(Sn(ζ)η ⊗ h)kK = lim ℓ→∞kun+m,ℓ(Sn(ζ)η ⊗ h)kX(ℓ)⊗σH ≤ ℓ→∞kum,ℓ−n(η ⊗ h)kX(ℓ−n)⊗σH ≤ kζk · lim = kζkkum(η ⊗ h)kK . Since the union of the ranges of the maps um is dense in K, the mapping Vn(ζ) is well-defined, and it extends to a bounded operator in B(K), also denoted by Vn(ζ). It is important to notice that the (left) module action of M on K is the one implemented by V0 (to see this, simply take n = 0 in (4.2)). V = (Vn)n∈Z+ is a completely contractive, covariant representation of X. The facts that V0 is a representation of M and that the Vn, n ∈ N, are bimodule maps with respect to V0 are inferred directly from (4.2). To establish that the representation V0 is normal in the W ∗-setting, fix m, k ∈ Z+, η ∈ X(m), ξ ∈ X(k) and h, h′ ∈ H. Then (V0(a)um(η ⊗ h), uk(ξ ⊗ h′)) = (um(a · η ⊗ h), uk(ξ ⊗ h′)). Since both σ and the homomorphism that implements the left multiplication on X(m) are normal, the mapping a 7→ a·η⊗h is continuous when M is equipped with the ultraweak topology and X(m) ⊗σ H is equipped with the weak topology. On the other hand, the linear mapping um : X(m) ⊗σ H → K, being bounded, is continuous when both Hilbert spaces are equipped with their weak topologies. Thus a 7→ (um(a · η ⊗ h), uk(ξ ⊗ h′)) is continuous in the ultraweak topology of M . This is enough to guarantee that V0 is normal. If n, m, k ∈ Z+, ζ ∈ X(n), ξ ∈ X(m), η ∈ X(k) and h ∈ H, then Vn(ζ)Vm(ξ)uk(η ⊗ h) = Vn(ζ)um+k(Sm(ξ)η ⊗ h) = un+m+k(Sn(ζ)Sm(ξ)η ⊗ h) = = un+m+k(Sn+m(pn+m(ζ ⊗ ξ))η ⊗ h) = Vn+m(pn+m(ζ ⊗ ξ))uk(η ⊗ h), hence Vn(ζ)Vm(ξ) = Vn+m(pn+m(ζ ⊗ ξ)). COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 27 Fix n ∈ Z+. Define an operator eVn : X(n) ⊗V0 K → K by letting eVn(ζ ⊗ um(η ⊗ h)) := Vn(ζ)um(η ⊗ h) = lim = lim = lim contractive, we obtain where ζ, m, η, h are as usual. If ζ1, . . . , ζp ∈ X(n), m1, . . . , mp ∈ Z+, h1, . . . hp ∈ H ζi ⊗ umi(ηi ⊗ hi)(cid:17)(cid:13)(cid:13)(cid:13)K and ηi ∈ X(mi) for 1 ≤ i ≤ p, then on account of (4.2), (4.3) and the fact that eSn is (cid:13)(cid:13)(cid:13)eVn(cid:16) pXi=1 un+mi,ℓ(Sn(ζi)ηi ⊗ hi)(cid:13)(cid:13)(cid:13)X(ℓ)⊗σH ℓ→∞(cid:13)(cid:13)(cid:13) pXi=1 (Sn(ζi) ⊗ IH)umi,ℓ−n(ηi ⊗ hi)(cid:13)(cid:13)(cid:13)X(ℓ)⊗σH ℓ→∞(cid:13)(cid:13)(cid:13) pXi=1 ℓ→∞(cid:13)(cid:13)(cid:13)(eSn ⊗ IH)(cid:0) pXi=1 ζi ⊗ umi,ℓ−n(ηi ⊗ hi)(cid:1)(cid:13)(cid:13)(cid:13)X(ℓ)⊗σH ζi ⊗ umi,ℓ−n(ηi ⊗ hi)(cid:13)(cid:13)(cid:13)X(n)⊗X(ℓ−n)⊗σH ℓ→∞(cid:13)(cid:13)(cid:13) pXi=1 ζi ⊗ umi(ηi ⊗ hi)(cid:13)(cid:13)(cid:13)X(n)⊗V0K =(cid:13)(cid:13)(cid:13) pXi=1 So eVn is well-defined and contractive on its domain, and it therefore extends to a contraction with domain X(n) ⊗V0 K. From Lemma 1.2, Vn is a completely contrac- tive, covariant representation of X(n) on K. In particular, the maps Vn(·) satisfy the continuity property mentioned in Definition 1.1 in the W ∗-setting (see Remark 1.3). ≤ lim . V is a dilation of T . Let n ∈ Z+, ζ ∈ X(n) be given. Then for h ∈ H one has (u∗0Vn(ζ)u0h, h) = (Vn(ζ)u0h, u0h) = (un(ζ ⊗ h), u0h) K K H = lim ℓ→∞ = lim = lim (un,ℓ(ζ ⊗ h), u0,ℓh)X(ℓ)⊗σH ℓ→∞(cid:0)(pℓ ⊗ IH)(IX(n) ⊗ eT ∗ℓ−n)(ζ ⊗ h),eT ∗ℓ h(cid:1)X(ℓ)⊗σH ℓ→∞(cid:0)eTℓ(pℓ ⊗ IH)(IX(n) ⊗ eT ∗ℓ−n)(ζ ⊗ h), h(cid:1)H . But eTℓ(pℓ ⊗ IH) = eTn(IX(n) ⊗ eTℓ−n) (by (2.2)) and T is fully coisometric, so we conclude that (u∗0Vn(ζ)u0h, h) = (Tn(ζ)h, h). Consequently, u∗0Vn(ζ)u0 = Tn(ζ). If, moreover, m ∈ Z+, η ∈ X(m) and h ∈ H, then by repeated usage of the foregoing, u∗0Vn(ζ)um(η ⊗ h) = u∗0Vn(ζ)Vm(η)u0h = u∗0Vn+m(pn+m(ζ ⊗ η))u0h = Tn+m(pn+m(ζ ⊗ η))h = Tn(ζ)Tm(η)h = u∗0Vn(ζ)u0u∗0Vm(η)u0h = u∗0Vn(ζ)u0u∗0um(η ⊗ h). COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 28 Therefore u∗0Vn(ζ) = u∗0Vn(ζ)u0u∗0, which implies that (u0H)⊥ is invariant under Vn(ζ) because u0u∗0 is a projection. In conclusion, V is a dilation of T . Proof of (4.1). Fix n, m ∈ Z+, ζ ∈ X(n), ξ ∈ X(m). Given p, q, t ∈ Z+, µ1, . . . , µt ∈ X(p), ν1, . . . , νt ∈ X(q) and h1, . . . , ht, h′1, . . . h′t ∈ H, denote x := up(Pt i=1 µi ⊗ hi), y := uq(Pt (Vn(ζ)∗Vm(ξ)x, y) = (Vm(ξ)x, Vn(ζ)y) = j=1 νj ⊗ h′j). Then from (4.2) and (4.3) we obtain = tXi,j=1(cid:0)up+m(Sm(ξ)µi ⊗ hi), uq+n(Sn(ζ)νj ⊗ h′j)(cid:1) tXi,j=1(cid:0)(Sm(ξ) ⊗ IH)up,ℓ−m(µi ⊗ hi), (Sn(ζ) ⊗ IH)uq,ℓ−n(νj ⊗ h′j)(cid:1) ℓ→∞(cid:16)(Sn(ζ)∗Sm(ξ) ⊗ IH)(cid:0) tXi=1 uq,ℓ−n(νj ⊗ h′j)(cid:1)(cid:17). up,ℓ−m(µi ⊗ hi)(cid:1),(cid:0) tXj=1 = lim ℓ→∞ = lim Hence (Vn(ζ)∗Vm(ξ)x, y) ≤ ≤ kSn(ζ)∗Sm(ξ) ⊗ IHk · lim = kSn(ζ)∗Sm(ξ) ⊗ IHk · kxk · kyk ≤ kSn(ζ)∗Sm(ξ)k · kxk · kyk . µi ⊗ hi)(cid:13)(cid:13)(cid:13) · lim ℓ→∞(cid:13)(cid:13)(cid:13)uq,ℓ−n( ℓ→∞(cid:13)(cid:13)(cid:13)up,ℓ−m( tXi=1 tXj=1 νj ⊗ h′j)(cid:13)(cid:13)(cid:13) Since x, y are arbitrary elements of a dense subset of K, inequality (4.1) follows. (cid:3) Remark 4.2. We do not know whether the dilation V constructed in the theorem is automatically fully coisometric. Nevertheless, we do know from (4.2) that eVn has dense range for all n. be a standard subproduct system in the C∗- Theorem 4.3. Let X = (X(n))n∈Z+ setting, and suppose that T is a fully coisometric, covariant representation of X on H, satisfying = kTm(η)hkH (4.4) lim ℓ→∞(cid:13)(cid:13)(pℓ ⊗ IH)(η ⊗ eT ∗ℓ−mh)(cid:13)(cid:13)X(ℓ)⊗σH for all m ∈ N, η ∈ X(m) and h ∈ H. Then T extends to a C∗-representation. Remark 4.4. If m, η, h are as above and ℓ ≥ m, then COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 29 (cid:13)(cid:13)(pℓ+1 ⊗ IH)(η ⊗ eT ∗ℓ+1−mh)(cid:13)(cid:13)2 X(ℓ+1)⊗σH = = (cid:0)(IX(m) ⊗ eTℓ+1−m)(pℓ+1 ⊗ IH)(IX(m) ⊗ eT ∗ℓ+1−m)(η ⊗ h), η ⊗ h(cid:1) = (cid:0)(IX(m) ⊗ eTℓ−m)(IX(ℓ) ⊗ eT1)(pℓ+1 ⊗ IH)(IX(m) ⊗ IX(ℓ−m) ⊗ eT ∗1 ) (IX(m) ⊗ eT ∗ℓ−m)(η ⊗ h), η ⊗ h(cid:1) = (cid:0)(IX(m) ⊗ eTℓ−m)(IX(ℓ) ⊗ eT1)(pℓ+1 ⊗ IH)(IX(ℓ) ⊗ eT ∗1 )(pℓ ⊗ IH) (IX(m) ⊗ eT ∗ℓ−m)(η ⊗ h), η ⊗ h(cid:1) ≤ (cid:0)(IX(m) ⊗ eTℓ−m)(pℓ ⊗ IH)(IX(m) ⊗ eT ∗ℓ−m)(η ⊗ h), η ⊗ h(cid:1) = (cid:13)(cid:13)(pℓ ⊗ IH)(η ⊗ eT ∗ℓ−mh)(cid:13)(cid:13)2 That is, the sequence(cid:8)(cid:13)(cid:13)(pℓ ⊗ IH)(η ⊗ eT ∗ℓ−mh)(cid:13)(cid:13)(cid:9)ℓ≥m is decreasing. Furthermore, (cid:13)(cid:13)Tm(η)h(cid:13)(cid:13)2 =(cid:0)eT ∗meTm(η ⊗ h), η ⊗ h(cid:1) =(cid:0)(IX(m) ⊗ eTℓ−m)eT ∗ℓ eTℓ(pℓ ⊗ IH)(IX(m) ⊗ eT ∗ℓ−m)(η ⊗ h), η ⊗ h(cid:1), X(ℓ)⊗σH . whence (since IX(ℓ)⊗σH − eT ∗ℓ eTℓ is a projection) (cid:13)(cid:13)(pℓ ⊗ IH)(η ⊗ eT ∗ℓ−mh)(cid:13)(cid:13)2 −(cid:13)(cid:13)Tm(η)h(cid:13)(cid:13)2 = =(cid:0)(IX(m) ⊗ eTℓ−m)(IX(ℓ)⊗σH − eT ∗ℓ eTℓ)(pℓ ⊗ IH)(IX(m) ⊗ eT ∗ℓ−m)(η ⊗ h), η ⊗ h(cid:1) =(cid:13)(cid:13)(IX(ℓ)⊗σH − eT ∗ℓ eTℓ)(pℓ ⊗ IH)(IX(m) ⊗ eT ∗ℓ−m)(η ⊗ h)(cid:13)(cid:13)2 =(cid:13)(cid:13)(pℓ ⊗ IH)(η ⊗ eT ∗ℓ−mh) − eT ∗ℓ Tm(η)h(cid:13)(cid:13)2 In conclusion, the limit on the left side of (4.4) always exists, and is greater than or equal to the right side. Moreover, (4.4) is equivalent to ≥ 0. lim ℓ→∞(cid:13)(cid:13)(pℓ ⊗ IH)(η ⊗ eT ∗ℓ−mh) − eT ∗ℓ Tm(η)h(cid:13)(cid:13)X(ℓ)⊗σH Before proving the theorem we turn our attention to condition (4.4), showing that in the case of product systems, it is equivalent to T being isometric. This is the analogue of Proposition 3.4 to the fully coisometric case. The condition is further discussed in Example 4.10. = 0. (4.5) be a standard product system, and suppose Proposition 4.5. Let X = (X(n))n∈Z+ that T is a fully coisometric, covariant representation of X. Then T is isometric (in the C∗-setting this is equivalent to T extending to a C∗-representation) if and only if T satisfies (4.4). Proof. Since pℓ = IE⊗ℓ and the operators IX(m) ⊗eT ∗ℓ−m are isometric for every ℓ ≥ m, (4.4) is equivalent to the equality kη ⊗ hk =(cid:13)(cid:13)eTm(η ⊗ h)(cid:13)(cid:13). This is true for all m, η, h if and only if T is isometric. (cid:3) COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 30 We now return to Theorem 4.3 and its proof. X(n)) and their adjoints. Every such operator can be written asQt S0(I) := IFX ). An S-monomial of this form is said to be of degree Pt Definition 4.6. Let X = (X(n))n∈Z+ be a standard subproduct system. An S- monomial is a composition of finitely many of the operators Sn(ζ) (n ∈ Z+, ζ ∈ i=1 Smi(ξi)∗Sni(ζi) for suitable t ∈ N and ni, mi ∈ Z+, ζi ∈ X(ni), ξi ∈ X(mi) (1 ≤ i ≤ t) (in case M is not unital, it may be implicitly replaced by M 1 where necessary, letting i=1(ni − mi). For k ∈ Z, define Tk(X) to be the closed linear span of all S-monomials of degree k. Evidently, Tk(X) is an operator space contained in T (X). In the special case k = 0, T0(X) is a C∗-subalgebra of T (X). Lemma 4.7. Under the assumptions of Theorem 4.3 there exists, to each k ∈ Z, i=1 Smi(ξi)∗Sni(ζi) of i=1 Tmi(ξi)∗Tni(ζi). Moreover, π0 is a C∗-representation of T0(X) on a contraction πk : Tk(X) → B(H) mapping an S-monomial Qt degree k to Qt H. Proof. Returning to the proof of Theorem 4.1, we claim that (4.4) yields that K = u0H. Indeed, let m ∈ N, η ∈ X(m) and h ∈ H be given, and write g := Tm(η)h ∈ H. Then from (4.5) we deduce that 0 = lim ℓ→∞kum,ℓ(η ⊗ h) − u0,ℓgk = lim ℓ→∞kum,ℓ(η ⊗ h − u0,mg)k . As a result, um(η ⊗ h) = umu0,mg = u0g ∈ u0H, and we conclude that K = u0H. Since u0 is isometric it is, in fact, unitary. In Theorem 4.1 we have established the existence of a dilation V of T , satisfying u∗0Vn(ζ)u0 = Tn(ζ) (4.6) for n ∈ Z+ and ζ ∈ X(n). Consequently, under current circumstances, the operator Vn(ζ) is unitarily equivalent to Tn(ζ). If n ∈ Z+, ζ ∈ X(n) and h ∈ H, then Vn(ζ)u0h = un(ζ ⊗ h), and from (4.3) (or simply the definition of un,ℓ) we have un,ℓ(ζ ⊗ h) = (Sn(ζ) ⊗ IH)u0,ℓ−nh. On the other hand, un(ζ ⊗ h) = u0x for some x ∈ H, hence un(ζ ⊗ h − u0,nx) = 0. This is equivalent to limℓ→∞ kun,ℓ(ζ ⊗ h − u0,nx)k = 0, i.e., Vn(ζ)u0h = u0x and lim ℓ→∞ku0,ℓx − (Sn(ζ) ⊗ IH)u0,ℓ−nhk = 0. (4.7) We now demonstrate a similar relation between Vn(ζ)∗ and Sn(ζ)∗ ⊗ IH. Given n ∈ Z+, ζ ∈ X(n) and h ∈ H, consider y := (Sn(ζ)∗⊗IH)u0,nh = (Sn(ζ)∗⊗IH)eT ∗n h ∈ COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 31 M ⊗ H. If ℓ ≥ 0, then u0,ℓy = (pℓ ⊗ IH)(IM ⊗ eT ∗ℓ )(Sn(ζ)∗ ⊗ IH)eT ∗n h = (pℓ ⊗ IH)(Sn(ζ)∗ ⊗ IX(ℓ)⊗σH)(IX(n) ⊗ eT ∗ℓ )eT ∗n h = (Sn(ζ)∗ ⊗ IH)eT ∗n+ℓh = (Sn(ζ)∗ ⊗ IH)u0,n+ℓh. (#) Equality (#) holds because the vector on which pℓ ⊗ IH acts already belongs to (M ⊗ X(ℓ) ⊗σ H, which we identify with) X(ℓ) ⊗σ H. This implies that (Vn(ζ)∗u0h, u0g)K = (u0h, Vn(ζ)u0g)K = lim ℓ→∞ = lim ℓ→∞ = lim ℓ→∞ (u0,n+ℓh, (Sn(ζ) ⊗ IH)u0,ℓg)X(n+ℓ)⊗σH ((Sn(ζ)∗ ⊗ IH)u0,n+ℓh, u0,ℓg)X(ℓ)⊗σH (u0,ℓy, u0,ℓg)X(ℓ)⊗σH = (u0y, u0g)K. In conclusion, Vn(ζ)∗u0h = u0y and u0,ℓy = (Sn(ζ)∗ ⊗ IH)u0,ℓ+nh for all ℓ. Let us prove that (4.8) (4.9) (cid:13)(cid:13)(cid:13) tYi=1 Tmi(ξi)∗Tni(ζi)(cid:13)(cid:13)(cid:13) ≤(cid:13)(cid:13)(cid:13) tYi=1 Smi(ξi)∗Sni(ζi)(cid:13)(cid:13)(cid:13) for every t ∈ N and ni, mi ∈ Z+, ζi ∈ X(ni), ξi ∈ X(mi) (1 ≤ i ≤ t). Till the end of the proof, the symbol Q will stand for multiplication in reverse order, for the sake of convenience. From (4.6) and u0 being unitary, (4.9) is equivalent to the inequality (cid:13)(cid:13)(cid:13) tYi=1 Vmi(ξi)∗Vni(ζi)u0(cid:13)(cid:13)(cid:13) ≤(cid:13)(cid:13)(cid:13) tYi=1 Smi(ξi)∗Sni(ζi)(cid:13)(cid:13)(cid:13). Fix t and ni, mi, ζi, ξi (1 ≤ i ≤ t) as needed, and let h ∈ H. We prove inductively that for all 0 ≤ p ≤ t there exists yp ∈ H such that Vmi(ξi)∗Vni(ζi)u0h = u0yp and pYi=1 ℓ→∞(cid:13)(cid:13)(cid:13)u0,ℓyp −(cid:16) pYi=1 lim Smi(ξi)∗Sni(ζi) ⊗ IH(cid:17)u0,ℓ−Pp i=1(ni−mi)h(cid:13)(cid:13)(cid:13) = 0. (4.10) For p = 0 choose y0 := h, for which (4.10) surely holds. Assuming the existence of yp has been exhibited, let xp+1 ∈ H be the unique vector such that u0xp+1 = i=1 Vmi(ξi)∗Vni(ζi)u0h = Vnp+1(ζp+1)u0yp. Then (4.7) implies that Vnp+1(ζp+1)Qp lim ℓ→∞(cid:13)(cid:13)u0,ℓxp+1 − (Snp+1(ζp+1) ⊗ IH)u0,ℓ−np+1yp(cid:13)(cid:13) = 0. COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 32 Consequently, from (4.10) and the boundedness of Snp+1(ζp+1) ⊗ IH, lim pYi=1 Smi(ξi)∗Sni(ζi) ⊗ IH(cid:17)u0,ℓ−np+1−Pp ℓ→∞(cid:13)(cid:13)(cid:13)u0,ℓxp+1 −(cid:16)Snp+1(ζp+1) Denote by yp+1 the element of H satisfying u0yp+1 = Qp+1 u0,ℓyp+1 = (Smp+1(ξp+1)∗ ⊗ IH)u0,ℓ+mp+1xp+1. Vmp+1(ξp+1)∗u0xp+1. By (4.8) we deduce that i=1(ni−mi)h(cid:13)(cid:13)(cid:13) = 0. (4.11) i=1 Vmi(ξi)∗Vni(ζi)u0h = The operator Smp+1(ξp+1)∗ ⊗ IH being bounded, we conclude in light of (4.11) that Smi(ξi)∗Sni(ζi) ⊗ IH(cid:17)u0,ℓ−Pp+1 i=1 (ni−mi)h(cid:13)(cid:13)(cid:13) = 0. So by induction, (4.10) holds with p = t. Therefore lim ℓ→∞(cid:13)(cid:13)(cid:13)u0,ℓyp+1 −(cid:16)p+1Yi=1 Vmi(ξi)∗Vni(ζi)u0h(cid:13)(cid:13)(cid:13) = ku0ytk = lim (cid:13)(cid:13)(cid:13) tYi=1 = lim ℓ→∞ku0,ℓytk ℓ→∞(cid:13)(cid:13)(cid:13)(cid:16) tYi=1 Smi(ξi)∗Sni(ζi) ⊗ IH(cid:17)u0,ℓ−Pt Smi(ξi)∗Sni(ζi) ⊗ IH(cid:13)(cid:13)(cid:13) · lim ≤(cid:13)(cid:13)(cid:13) ℓ→∞(cid:13)(cid:13)(cid:13)u0,ℓ−Pt tYi=1 Smi(ξi)∗Sni(ζi)(cid:13)(cid:13)(cid:13) · khk , ≤(cid:13)(cid:13)(cid:13) tYi=1 i=1(ni−mi)h(cid:13)(cid:13)(cid:13) i=1(ni−mi)h(cid:13)(cid:13)(cid:13) (4.12) i i=1 Vm(j) j=1Qt ℓ→∞(cid:13)(cid:13)(cid:13)u0,ℓy −(cid:16) qXj=1 lim proving (4.9). If k ∈ Z, the existence of the mapping πk is shown similarly. Assume i ) (j = 1, . . . , q) is a finite family of S-monomials of degree k, and fix h ∈ H. Using (4.10) q times, we furnish the existence of y ∈ H i )∗Sn(j) i=1 Sm(j) (ζ (j) (ξ(j) i i (ξ(j) i )∗Vn(j) i (ζ (j) i )u0h = u0y and that Qt such thatPq Consequently, just as in (4.12), we have tYi=1 Sm(j) i (ξ(j) i )∗Sn(j) i (ζ (j) i ) ⊗ IH(cid:17)u0,ℓ−kh(cid:13)(cid:13)(cid:13) = 0. COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 33 i Vm(j) i (ζ (j) tYi=1 (ξ(j) i )∗Vn(j) (cid:13)(cid:13)(cid:13) qXj=1 i )u0h(cid:13)(cid:13)(cid:13) ≤ ≤(cid:13)(cid:13)(cid:13) qXj=1 tYi=1 ≤(cid:13)(cid:13)(cid:13) qXj=1 tYi=1 i=1 Smi(ξi)∗Sni(ζi) to Qt degree k to B(H), mapping Qt i )∗Sn(j) i )∗Sn(j) Sm(j) Sm(j) (ξ(j) (ξ(j) i i i i In conclusion, the canonical linear mapping from the span of all S-monomials of i=1 Tmi(ξi)∗Tni(ζi), is a well- defined contraction, which is a multiplicative ∗-mapping when k = 0. It therefore extends to a contraction πk from Tk(X) to B(H). Since T0(X) is a C∗-algebra, π0 is a C∗-representation. (cid:3) ℓ→∞ku0,ℓ−khk (ζ (j) (ζ (j) i ) ⊗ IH(cid:13)(cid:13)(cid:13) · lim i )(cid:13)(cid:13)(cid:13) · khk . Remark 4.8. Condition (4.4) is not only sufficient, but also necessary, to having K = u0H. We omit the details. Proof of Theorem 4.3. Let us begin with some facts on circle actions. Suppose that B is a C∗-algebra with an action α of T on B. Denote the spectral subspaces for α ([14, Definition 2.1]) by (Bk)k∈Z. Then B1 becomes a B0 − B0 Hilbert C∗-bimodule in the sense of [9, Definition 1.8] upon letting ha, biR := a∗b and ha, biL := ab∗ for all a, b ∈ B1. The norm in B1 as a bimodule is the same as its natural one. If now α is semi-saturated, i.e., B is generated as a C∗-algebra by B0 and B1 (see [14, Definition 4.1]), then by [1, Theorem 3.1] and its proof we have B ∼= B0 ⋊B1 Z, the crossed product being the one defined in [1, Definition 2.4]; and moreover, if the crossed product maps are denoted by ιB0, ιB1, then the implementing isomorphism, say φ, makes the following diagram commute: (B0, B1) (ιB0 ,ιB1 ) H H H H H H H H H #H / B (the maps on the diagonal being the inclusions). B0 ⋊B1 Z φ In our framework, we construct the usual gauge action of T on T (X). Given λ ∈ T, λnζn. From the definition of define a unitary Wλ ∈ L(FX) byLn∈Z+ ζn 7→Ln∈Z+ the X-shift it follows that if A is an S-monomial of degree k ∈ Z in T (X), then WλAW ∗λ = λkA. (4.13) # / COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 34 Hence, the formula αλ(A) := WλAW ∗λ defines an automorphism of T (X), and λ 7→ αλ is an action of T on T (X) (in particular, for A ∈ T (X), the function λ 7→ αλ(A) is norm-continuous). Moreover, from (4.13) and [14, Definition 2.4], the kth spectral subspace for α is Tk(X). Hence the gauge action α is semi-saturated, because S0(M ) = ϕ∞(M ) and S1(X(1)) are enough to generate T (X) as a C∗-algebra (there is a subtlety here- see Remark 4.9). Consequently, by [1, Theorem 3.1], T (X) ∼= T0(X) ⋊T1(X) Z. To elaborate, if this isomorphism is implemented by φ and ι0, ι1 are the crossed product maps from T0(X),T1(X), respectively, to T0(X) ⋊T1(X) Z, then the following diagram commutes: (T0(X),T1(X)) N (ι0,ι1) N N N N N N N N N 'N (4.14) T0(X) ⋊ T1(X) Z / T (X) φ From Lemma 4.7 the pair of mappings (π0, π1) is a covariant representation of (T0(X),T1(X)) on H in the sense of [1, Definition 2.1] (properties (i)-(iv) therein are proved by considering first only vectors in the total subsets of T0(X),T1(X) con- sisting of the S-monomials of degree 0, 1 respectively, and then using linearity and continuity; notice that πA(·) is missing from (iii) and (iv)). As a result, the uni- versality property of the crossed product implies that there is a C∗-representation θ : T0(X) ⋊T1(X) Z → B(H) such that the following diagram commutes: (T0(X),T1(X)) N (ι0,ι1) N N (π0,π1) N N N N N N N 'N T0(X) ⋊T1(X) Z / B(H) θ Define π := θ◦φ−1. Since the diagram (4.14) commutes, one infers that π(A) = πk(A) for A ∈ Tk(X) (k = 0, 1). The proof is now complete because, as already mentioned, T0(X) and T1(X) generate T (X) as a C∗-algebra. (cid:3) Remark 4.9. If ζ, η ∈ X(1) then S2(p2(ζ ⊗ η)) = S1(ζ)S1(η), and as the span of the set of "simple tensors" is norm-dense in X(1) ⊗ X(1) in the C∗-setting, the C∗- algebra generated by T0(X) and T1(X) contains S2(X(2)), and similarly Sk(X(k)) for every k ∈ Z+. Nonetheless, in the W ∗-setting, this set of vectors is guaranteed only to be s-dense in the tensor product (see Remark 1.8). Consequently, the C∗- algebra generated by T0(X) and T1(X) is ultraweakly dense in T (X), but there is no reason to expect an equality to hold generally. We consider this limitation quite ' / ' / COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 35 reasonable, as the W ∗-setting is more naturally suitable for ultraweakly continuous representations of ultraweakly closed algebras. By Theorem 4.3, condition (4.4) implies C∗-extendability. While we would like to establish the inverse implication -- at least for subproduct systems with E a finite dimensional Hilbert space (analogously to Corollary 3.10) -- the fully coisometric case is more involved than the pure case. In fact, to our knowledge, no general criterion for C∗-extendability has yet been found even in this fundamental case. However, we will prove that condition (4.4) is indeed equivalent to C∗-extendability in the most well-known subproduct system: the symmetric one. Example 4.10. Let us consider the symmetric subproduct system (see [40, Example 1.3]), defined as follows. For a fixed d ∈ N, take E to be the Hilbert space Cd (of course, M = C). Given n ∈ N, define a projection pn : E⊗n → E⊗n by pn(f1 ⊗ · · · ⊗ fn) := fπ(1) ⊗ · · · ⊗ fπ(n) (4.15) 1 n!Xπ . for all f1, . . . , fn ∈ E, the sum being taken over all permutations π of {1, . . . , n}. Set Esn := pnE⊗n (the n-fold symmetric tensor product of E) and Es0 := C. The symmetric subproduct system is SSPd :=(cid:0)Esn(cid:1)n∈Z+ If {e1, . . . , ed} is a fixed orthonormal basis of E, write Ti := T1(ei), 1 ≤ i ≤ d. There is a bijection between all completely contractive, covariant representations of SSPd on H and all commuting row contractions over H (of length d) implemented by T 7→ (T1, . . . , Td), and we shall identify these two types of objects. A commuting row contraction T is called spherical if Ti is normal for all i = 1, . . . , d and T1T ∗1 +. . .+TdT ∗d = IH. A spherical row contraction is evidently fully coisometric. It was proved in [4, §8] that a fully coisometric, commuting row contraction T extends to a C∗-representation if and only if T is spherical. Theorem. Let T be a fully coisometric, commuting row contraction. Then T is spherical if and only if condition (4.4) holds. Sufficiency is a direct byproduct of Theorem 4.3, but the proof will not require this fact. Proof. Fix a fully coisometric, commuting row contraction T over H. η ∈ Esm and h ∈ H, we will show that If m ∈ N, lim ℓ→∞(cid:13)(cid:13)(pℓ ⊗ IH)(η ⊗ eT ∗ℓ−mh)(cid:13)(cid:13)Esℓ⊗H = kTm(η)∗hk , from which the theorem's conclusion follows. Fixing an orthogonal base {e1, . . . , ed} for E, it suffices to consider only the case η = e⊗m , 1 ≤ p ≤ d, because {k⊗m : k ∈ E} p COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 36 spans Esm, and from (4.5), condition (4.4) being true for all η is equivalent to it being true for η in a spanning subset of Esm. Write [d] := {1, 2, . . . , d}. It follows from [40, Proposition 6.9] (or a direct calcu- lation) that for h ∈ H and ℓ ∈ N we have eT ∗ℓ h = X(i1,...,iℓ)∈[d]ℓ ei1 ⊗ · · · ⊗ eiℓ ⊗(cid:0)T ∗iℓ · · · T ∗i1h(cid:1) . (4.16) We introduce some notations. Given ℓ ∈ N, define an equivalence relation over [d]ℓ by saying that two ℓ-tuples are equivalent if each is a rearrangement of the other. Let Aℓ be a subset of [d]ℓ that contains exactly one element of each equivalence class of [d]ℓ. If (i1, . . . , iℓ) ∈ [d]ℓ and 1 ≤ k ≤ d, write ck = ck(i1, . . . , iℓ) for the number of appearances of k in the series i1, . . . , iℓ. Denote the equivalence class of (i1, . . . , iℓ) by [(i1, . . . , iℓ)]ℓ. The cardinality of [(i1, . . . , iℓ)]ℓ, denoted by a(i1,...,iℓ), equals(cid:0) c1!···cd! (the multinomial coefficient). Rewrite (4.16) as ℓ c1 ··· cd(cid:1) = ℓ! eT ∗ℓ h = X(j1,...,jℓ)∈Aℓ(cid:16) X(i1,...,iℓ)∈[(j1,...,jℓ)]ℓ khk2 =(cid:13)(cid:13)eT ∗ℓ h(cid:13)(cid:13)2 = X(j1,...,jℓ)∈Aℓ ei1 ⊗ · · · ⊗ eiℓ(cid:17) ⊗(cid:0)T ∗jℓ · · · T ∗j1h(cid:1) a(j1,...,jℓ)(cid:13)(cid:13)T ∗jℓ · · · T ∗j1h(cid:13)(cid:13)2 . (this is possible only by virtue of the commutativity of T1, . . . , Td). As a result, (4.17) (4.18) Given (j1, . . . , jℓ−m) ∈ [d]ℓ−m, write (η, j1, . . . , jℓ−m) for the ℓ-tuple (p, . . . , p, j1, . . . , jℓ−m), and observe that pℓ(η⊗ei1⊗· · ·⊗eiℓ−m) is the same for all (i1, . . . , iℓ−m) in [(j1, . . . , jℓ−m)]ℓ−m, and equals 1 ℓ! · ℓ! a(η,j1,...,jℓ−m) · X (i1....,iℓ)∈[(η,j1,...,jℓ−m)]ℓ ei1 ⊗ · · · ⊗ eiℓ ℓ! ( is exactly the number of times each summand is repeated in (4.15)). a(η,j1,...,jℓ−m) Therefore, as [(j1, . . . , jℓ−m)]ℓ−m consists of a(j1,...,jℓ−m) elements, we infer from (4.17) that (pℓ ⊗ IH)(η ⊗ eT ∗ℓ−mh) = X(j1,...,jℓ−m)∈ Aℓ−m a(j1,...,jℓ−m) a(η,j1,...,jℓ−m)(cid:16) X(i1....,iℓ)∈ [(η,j1,...,jℓ−m)]ℓ ei1 ⊗ · · · ⊗ eiℓ(cid:17) ⊗(cid:16)T ∗jℓ−m · · · T ∗j1h(cid:17) . COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 37 Write cp = cp(j1, . . . , jℓ−m). All summands in the last sum are mutually orthogonal, hence we have = (cid:13)(cid:13)(pℓ ⊗ IH)(η ⊗ eT ∗ℓ−mh)(cid:13)(cid:13)2 = X(j1,...,jℓ−m)∈Aℓ−m = Xthe same a2 (j1,...,jℓ−m) a2 (η,j1,...,jℓ−m) (cp + m) · · · (cp + 1) ℓ(ℓ − 1) · · · (ℓ − m + 1) a(η,j1,...,jℓ−m)(cid:13)(cid:13)T ∗jℓ−m · · · T ∗j1h(cid:13)(cid:13)2 a(j1,...,jℓ−m)(cid:13)(cid:13)T ∗jℓ−m · · · T ∗j1h(cid:13)(cid:13)2 (4.19) these tuples alone, the result is dominated by (because ck(η, j1, . . . , jℓ−m) = ck for all k 6= p and cp(η, j1, . . . , jℓ−m) = cp + m). Next, we assert that the sum of all summands in (4.19) corresponding to tuples (j1, . . . , jℓ−m) ∈ Aℓ−m with cp < √ℓ are negligible as ℓ → ∞. For, summing over √ℓ + m ℓ − m + 1!m Xsuch tuples ℓ − m + 1!m X(j1,...,jℓ−m)∈Aℓ−m √ℓ + m a(j1,...,jℓ−m)(cid:13)(cid:13)T ∗jℓ−m · · · T ∗j1h(cid:13)(cid:13)2 ≤ a(j1,...,jℓ−m)(cid:13)(cid:13)T ∗jℓ−m · · · T ∗j1h(cid:13)(cid:13)2 = √ℓ + m ℓ − m + 1!m khk2 (4.20) (j1,...,jℓ−m) (see (4.18)) and the right side converges to 0 as ℓ → ∞. Consequently, letting ℓ grow big enough, we may assume (in particular) that cp ≥ m. For such (ℓ − m)- tuple (j1, . . . , jℓ−m), write (k1, . . . , kℓ−2m) for an (ℓ− 2m)-tuple obtained by removing (any) m repetitions of p from (j1, . . . , jℓ−m). So in conclusion, after removing all (j1, . . . , jℓ−m) ∈ Aℓ−m with cp < √ℓ from the sum in (4.19), we are left with X(j1,...,jℓ−m)∈Aℓ−m (cp + m) · · · (cp + 1) ℓ(ℓ − 1) · · · (ℓ − m + 1) a(j1,...,jℓ−m)(cid:13)(cid:13)T ∗jℓ−m · · · T ∗j1h(cid:13)(cid:13)2 with cp≥ √ℓ = = Xthe same (cp + m) · · · (cp + 1) ℓ(ℓ − 1) · · · (ℓ − m + 1) · a(k1,...,kℓ−2m)(cid:13)(cid:13)T ∗kℓ−2m · · · T ∗k1T m∗p h(cid:13)(cid:13)2. (ℓ − m) · · · (ℓ − 2m + 1) cp · · · (cp − m + 1) · (4.21) Since m is fixed, (ℓ−m)···(ℓ−2m+1) ℓ(ℓ−1)···(ℓ−m+1) → 1 as ℓ → ∞. So for the purpose of the limit of (cid:13)(cid:13)(pℓ ⊗ IH)(η ⊗ eT ∗ℓ−mh)(cid:13)(cid:13)2 we can remove this factor from (4.21). Additionally, since cp ≥ √ℓ, (cp+m)···(cp+1) cp···(cp−m+1) converges to 1 as ℓ → ∞ uniformly for all relevant tuples. COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 38 Removing this factor as well from (4.21) we get X(j1,...,jℓ−m)∈Aℓ−m √ℓ with cp≥ a(k1,...,kℓ−2m)(cid:13)(cid:13)T ∗kℓ−2m · · · T ∗k1T m∗p h(cid:13)(cid:13)2 X(i1,...,iℓ−2m)∈Aℓ−2m with cp(i1,...,iℓ−2m)≥√ℓ−m = a(i1,...,iℓ−2m)(cid:13)(cid:13)T ∗iℓ−2m · · · T ∗i1T m∗p h(cid:13)(cid:13)2 . Had the cp of the (ℓ − 2m)-tuples in the last sum not been restricted, we would have had the sum X(i1,...,iℓ−2m)∈Aℓ−2m a(i1,...,iℓ−2m)(cid:13)(cid:13)T ∗iℓ−2m · · · T ∗i1T m∗p h(cid:13)(cid:13)2 =(cid:13)(cid:13)T m∗p h(cid:13)(cid:13)2 =(cid:13)(cid:13)Tm(η)∗h(cid:13)(cid:13)2 , (by (4.18)), as desired. So all that is left is to see that the difference between the two sums converges to zero. This will be done exactly as before (see (4.20)), denoting now cp = cp(i1, . . . , iℓ−2m): X(i1,...,iℓ−2m)∈Aℓ−2m with cp<√ℓ−m (cp + m) · · · (cp + 1) a(i1,...,iℓ−2m)(cid:13)(cid:13)T ∗iℓ−2m · · · T ∗i1T m∗p h(cid:13)(cid:13)2 = Xthe same (ℓ − m) · · · (ℓ − 2m + 1) ≤ ℓ − 2m + 1!m Xthe same √ℓ ℓ − 2m + 1!m X(t1,...,tℓ−m)∈Aℓ−m ≤ = ℓ − 2m + 1!m khk2 −−−→ℓ→∞ √ℓ √ℓ 0 a(η,i1,...,iℓ−2m)(cid:13)(cid:13)T ∗iℓ−2m · · · T ∗i1T m∗p h(cid:13)(cid:13)2 a(η,i1,...,iℓ−2m)(cid:13)(cid:13)T ∗iℓ−2m · · · T ∗i1T m∗p h(cid:13)(cid:13)2 a(t1,...,tℓ−m)(cid:13)(cid:13)T ∗tℓ−m · · · T ∗t1h(cid:13)(cid:13)2 (see (4.18)). (cid:3) 5. Conclusions and examples In this section we combine some of the previous results, and give more examples. We begin with a general Wold decomposition. Recall that Q = s-limn→∞eTneT ∗n . be a standard subproduct system in the C∗- Theorem 5.1. Let X = (X(n))n∈Z+ setting, and suppose that T is a completely contractive, covariant representation of X on H. Assume that T is relatively isometric and lim ℓ→∞(cid:13)(cid:13)(pℓ ⊗ Q)(η ⊗ eT ∗ℓ−mh)(cid:13)(cid:13)X(ℓ)⊗H = kTm(η)QhkH (5.1) COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 39 for all m ∈ N, η ∈ X(m) and h ∈ H. Then there exist a Hilbert space U, a unitary operator W : H → (FX ⊗σ D) ⊕ U and a fully coisometric, covariant represen- of X on U, which extends to a C∗-representation, such that tation Z = (Zn)n∈Z+ W Tn(ζ)W −1 = (Sn(ζ) ⊗ ID) ⊕ Zn(ζ) for all n ∈ Z+, ζ ∈ X(n). In particular, T extends to a C∗-representation. Proof. From Theorems 2.22 and 3.7 and their proofs, we should ascertain that the covariant representation Z (of X on the Hilbert subspace U of H) mentioned there extends to a C∗-representation. Under present assumptions, Q is the projection of H on U and Y is Q with codomain U. In particular, U = Im Q = Im Y . By (2.16), (pℓ ⊗ IU )(η ⊗eZ∗ℓ−mY h) = (pℓ ⊗ IU )(cid:0)η ⊗ (IX(ℓ−m) ⊗ Y )eT ∗ℓ−mh(cid:1) = (pℓ ⊗ Q)(η ⊗eT ∗ℓ−mh). Since Zm(η)Y h = Tm(η)Qh, we infer from (5.1) that Z satisfies the conditions of Theorem 4.3. This completes the proof. (cid:3) Example 5.2 (Homogeneous ideals). We extend results of [22, §2], showing that most of their hypothesis (H1) is superfluous (also cf. [39], [11, §2], [3, §1] and [40, §7]). Throughout this example we shall be working in the C∗-setting. Let E be a C∗-correspondence that is essential as a left M -module. Write S = (Sn)n∈Z+ for the shift in F (E). Given a two-sided, norm closed ideal I E T+(E), define I (n) := I ∩ {Sn(ζ) : ζ ∈ E⊗n} for n ∈ Z+. We call I homogeneous if I = spanSn∈Z+ I (n) (equivalently: I is invariant under the gauge action on T+(E)). Our standing hy- potheses are: (1) The ideal I E T+(E) is homogeneous. (2) The submodule Y (n) := {ζ ∈ E⊗n : Sn(ζ) ∈ I} is orthogonally complementable in E⊗n for all n and Y (0) = {0}. Given n ∈ Z+, define X(n) to be the orthogonal complement of Y (n) in E⊗n, and let pn denote the orthogonal projection of E⊗n on X(n). If n, m ∈ Z+, then since I is an ideal we obtain Y (n)⊗ E⊗m ⊆ Y (n + m) and E⊗n ⊗ Y (m) ⊆ Y (n + m), that is, pn+m ≤ pn ⊗ IE⊗m and pn+m ≤ IE⊗n ⊗ pm. Therefore, since pn ⊗ IE⊗m and IE⊗n ⊗ pm Im pn+m = X(n + m). In conclusion, X = (X(n))n∈Z+ X(0) = M . Moreover, (1) yields that commute, X(n) ⊗ X(m) = Im(pn ⊗ pm) = (cid:0)Im(pn ⊗ IE⊗m)(cid:1) ∩(cid:0)Im(IE⊗n ⊗ pm)(cid:1) ⊇ IF (E) =(cid:0)span [n∈Z+ is a subproduct system with I (n)(cid:1)F (E) =(cid:0)span [n∈Z+ I (n)E⊗m = Mk∈Z+ = span [n,m∈Z+ I (n)(cid:1)F (E) span [n,m∈Z+ n+m=k I (n)E⊗m = Mk∈Z+ Y (k) COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 40 (for I (n)E⊗m ⊆ Y (n) ⊗ E⊗m ⊆ Y (n + m) = I (n+m)M ). Consequently, FX = F (E) ⊖ IF (E). (5.2) pn be the projection of F (E) on FX. The relation between T+(E) Let P :=Ln∈Z+ and T+(X) is illuminated in the following result. Theorem. T+(X) ∼= T+(E)/I in the sense that the operator algebras are (canoni- cally) completely isometrically isomorphic. Proof. Write SX = (SX n )n∈Z+ for the X-shift. Define κ : T+(E)/I → T+(X) by T +I 7→ P TFX for T ∈ T+(E). The mapping κ is well-defined by (5.2) and κ(Sn(ζ)+ I) = SX n (ζ) for all n ∈ Z+, ζ ∈ X(n). It is contractive since if T ∈ T+(E) and Z ∈ I, then from (5.2), (cid:13)(cid:13)P TFX(cid:13)(cid:13) =(cid:13)(cid:13)P T P(cid:13)(cid:13) =(cid:13)(cid:13)P (T + Z)P(cid:13)(cid:13) ≤(cid:13)(cid:13)T + Z(cid:13)(cid:13), and so(cid:13)(cid:13)P TFX(cid:13)(cid:13) ≤ inf Z∈I(cid:13)(cid:13)T +Z(cid:13)(cid:13) =(cid:13)(cid:13)T +I(cid:13)(cid:13). Using the identification of Mn(T+(E)/ I) with Mn(T+(E))/Mn(I), one proves in a similar fashion that κ is actually com- pletely contractive. Let π be a nondegenerate completely isometric representation of T+(E)/I on some Hilbert space H so that M ∋ a 7→ π(ϕ∞(a) + I) is a (nondegenerate, because E is essential) C∗-representation of M . Define a completely contractive, covariant representation T of X on H by Tn(ζ) := π(Sn(ζ) + I) (n ∈ Z+, ζ ∈ X(n)). Indeed, if n, m ∈ Z+, ζ ∈ X(n) and η ∈ X(m), we obtain Tn(ζ)Tm(η) = π(Sn(ζ)Sm(η) +I) = π(Sn+m(ζ ⊗ η) +I) = π(Sn+m(pn+m(ζ ⊗ η)) +I) (as (IE⊗(n+m) − pn+m)(ζ ⊗ η) ∈ E⊗(n+m) ⊖ X(n + m)). Moreover, since Sn(·) is completely contractive for all n, the same is true for Sn(·) + I, and hence also for Tn(·). Therefore, Theorem 2.15 implies that there is a completely contractive linear mapping ϑ : T+(X) → T+(E)/I satisfying ϑ(SX n (ζ)) = Sn(ζ) + I for all n ∈ Z+, ζ ∈ X(n). Hence ϑ = κ−1. Both κ and ϑ are completely contractive, thus they are completely isometric. This completes the proof. (cid:3) Corollary. d(T,I) = kP T Pk for every T ∈ T+(E). Example 5.3. Fix a von Neumann algebra M ⊆ B(H) and a cp-semigroup (Θn)n∈Z+ on M . It is established in [40, Theorem 2.2] and the preceding discussion that there over the commutant M ′ and a completely exist a subproduct system X = (X(n))n∈Z+ contractive, covariant representation T = (Tn)n∈Z+ of X on H such that (∀n ∈ Z+, a ∈ M ) Θn(a) = eTn(IX(n) ⊗ a)eT ∗n . (5.3) COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 41 This construction is interesting in view of [40, Theorem 2.6]. Let us see when this T is relatively isometric. From (5.3), Tn is a partial isometry if and only if Θn(I) is a projection. Using the notations of [40], we have ∆∗ = I − Θ1(I) and Tn(x) = W ∗Θnx for x ∈ X(n) = LM (H, M ⊗Θn H). Consequently, (3.2) is equivalent to (∀n ∈ Z+, x ∈ X(n)) (I − Θ1(I))x∗WΘnW ∗Θnx(I − Θ1(I)) = x∗x(I − Θ1(I)) (σ is the identity representation of M ′). For instance, this condition is easily satisfied when the operators WΘn are all coisometric, which happens, e.g., when Θ is an e- semigroup (this in turn implies that T is actually isometric, so that X is necessarily a product system by [40, Corollary 2.8]). Example 5.4 (Strictly cyclic subproduct systems). Let M be a C∗-algebra. A C∗- correspondence F over M is called strictly cyclic ([20, p. 419]) if F = P M for some projection P in M(M ), the multiplier algebra of M (the left action of M on F is not restricted). A subproduct system X = (X(n))n∈Z+ over M is strictly cyclic if E = X(1) = M and X(n) is strictly cyclic -- i.e., X(n) = PnM ⊗n -- for all n ≥ 2. When given such X, the projection pn of M ⊗n on X(n) is of a very concrete form: it is multiplication on the left by Pn, resulting in a more simplified condition (4.4). For example, define the left operation of M on E := M to be left multiplication. Then E⊗n may be identified with M via a1 ⊗ · · · ⊗ an 7→ a1 · · · an, and if the projections Pn ∈ M(M ), n ≥ 2, commute with M and satisfy Pn+mM ⊆ PnPmM for each n, m ∈ N (P1 := I), we obtain a strictly cyclic subproduct system by setting X(n) := PnM . Acknowledgments The author would like to express his deep gratitude to Baruch Solel for many in- triguing and fruitful conversations on the content of this paper. He is grateful to the referee for carefully reading the manuscript and making useful suggestions. References [1] B. Abadie, S. Eilers and R. Exel, Morita equivalence for crossed products by Hilbert C ∗- bimodules, Trans. Amer. Math. Soc. 350 (1998), no. 8, 3043 -- 3054. [2] T. Andô, On a pair of commutative contractions, Acta Sci. Math. (Szeged) 24 (1963), 88 -- 90. [3] A. Arias and G. Popescu, Noncommutative interpolation and Poisson transforms, Israel J. Math. 115 (2000), 205 -- 234. [4] W. Arveson, Subalgebras of C ∗-algebras III: Multivariable operator theory, Acta Math. 181 (1998), no. 2, 159 -- 228. [5] M. Baillet, Y. Denizeau and J.-F. Havet, Indice d'une espérance conditionnelle, Compos. Math. 66 (1988), no. 2, 199 -- 236. COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 42 [6] B. V. R. Bhat and T. Bhattacharyya, A model theory for q-commuting contractive tuples, J. Operator Theory 47 (2002), no. 1, 97 -- 116. [7] B. V. R. Bhat, T. Bhattacharyya and S. Dey, Standard noncommuting and commuting dilations of commuting tuples, Trans. Amer. Math. Soc. 356 (2004), no. 4, 1551 -- 1568. [8] B. V. R. Bhat and M. Mukherjee, Inclusion systems and amalgamated products of product systems, preprint, arXiv:0907.0095. [9] L. Brown, J. A. Mingo and N.-T. Shen, Quasi-multipliers and embeddings of Hilbert C ∗- bimodules, Canad. J. Math. 46 (1994), no. 6, 1150 -- 1174. [10] J. W. Bunce, Models for n-tuples of noncommuting operators, J. Funct. Anal. 57 (1984), no. 1, 21 -- 30. [11] K. R. Davidson and D. R. Pitts, Nevanlinna-Pick interpolation for non-commutative analytic Toeplitz algebras, Integral Equations Operator Theory 31 (1998), no. 3, 321 -- 337. [12] S. Dey, Standard dilations of q-commuting tuples, Colloq. Math. 107 (2007), no. 1, 141 -- 165. [13] S. W. Drury, A generalization of von Neumann's inequality to the complex ball, Proc. Amer. Math. Soc. 68 (1978), no. 3, 300 -- 304. [14] R. Exel, Circle actions on C ∗-algebras, partial automorphisms, and a generalized Pimsner- Voiculescu exact sequence, J. Funct. Anal. 122 (1994), no. 2, 361 -- 401. [15] A. E. Frazho, Models for noncommuting operators, J. Funct. Anal. 48 (1982), no. 1, 1 -- 11. [16] A. E. Frazho, Complements to models for noncommuting operators, J. Funct. Anal. 59 (1984), no. 3, 445 -- 461. [17] R. V. Kadison and J. R. Ringrose, Fundamentals of the Theory of Operator Algebras. Volume II: Advanced Theory, Graduate Studies in Mathematics, Vol. 16, Amer. Math. Soc., Providence, 1997. [18] E. C. Lance, Hilbert C ∗-Modules. A Toolkit for Operator Algebraists, London Mathematical Society Lecture Note Series, Vol. 210, Cambridge University Press, Cambridge, 1995. [19] V. M. Manuilov, E. V. Troitsky, Hilbert C ∗-Modules, Translations of Mathematical Mono- graphs, Vol. 226, Amer. Math. Soc., Providence, 2005. [20] P. S. Muhly and B. Solel, Tensor algebras over C ∗-correspondences: representations, dilations, and C ∗-envelopes, J. Funct. Anal. 158 (1998), no. 2, 389 -- 457. [21] P. S. Muhly and B. Solel, Tensor algebras, induced representations, and the Wold decomposition, Canad. J. Math. 51 (1999), no. 4, 850 -- 880. [22] P. S. Muhly and B. Solel, On quotients of tensor algebras and their C ∗-envelopes, Proc. Edinb. Math. Soc. (2) 43 (2000), no. 2, 361 -- 377. [23] P. S. Muhly and B. Solel, Quantum Markov processes (correspondences and dilations), Internat. J. Math. 13 (2002), no. 8, 863 -- 906. [24] P. S. Muhly and B. Solel, Hardy algebras, W ∗-correspondences and interpolation theory, Math. Ann. 330 (2004), no. 2, 353 -- 415. [25] P. S. Muhly and B. Solel, The Poisson kernel for Hardy algebras, Complex Anal. Oper. Theory 3 (2009), no. 1, 221 -- 242. [26] W. L. Paschke, Inner product modules over B ∗-algebras, Trans. Amer. Math. Soc. 182 (1973), 443 -- 468. [27] W. L. Paschke, Inner product modules arising from compact automorphism groups of von Neu- mann algebras, Trans. Amer. Math. Soc. 224 (1976), 87 -- 102. COVARIANT REPRESENTATIONS OF SUBPRODUCT SYSTEMS 43 [28] M. V. Pimsner, A class of C ∗-algebras generalizing both Cuntz-Krieger algebras and crossed products by Z, in: Free Probability Theory, Fields Inst. Commun., Vol. 12, Amer. Math. Soc., Providence, 1997, pp. 189 -- 212. [29] G. Popescu, Isometric dilations for infinite sequences of noncommuting operators, Trans. Amer. Math. Soc. 316 (1989), no. 2, 523 -- 536. [30] G. Popescu, Von Neumann inequality for (B(H)n)1, Math. Scand. 68 (1991), no. 2, 292 -- 304. [31] G. Popescu, Functional calculus for noncommuting operators, Michigan Math. J. 42 (1995), no. 2, 345 -- 356. [32] G. Popescu, Non-commutative disc algebras and their representations, Proc. Amer. Math. Soc. 124 (1996), no. 7, 2137 -- 2148. [33] G. Popescu, Noncommutative joint dilations and free product operator algebras, Pacific J. Math. 186 (1998), no. 1, 111 -- 140. [34] G. Popescu, Universal operator algebras associated to contractive sequences of non-commuting operators, J. London Math. Soc. (2) 58 (1998), no. 2, 469 -- 479. [35] G. Popescu, Poisson transforms on some C ∗-algebras generated by isometries, J. Funct. Anal. 161 (1999), no. 1, 27 -- 61. [36] G. Popescu, Operator theory on noncommutative varieties, Indiana Univ. Math. J. 55 (2006), no. 2, 389 -- 442. [37] G. Popescu, Operator theory on noncommutative varieties, II, Proc. Amer. Math. Soc. 135 (2007), no. 7, 2151 -- 2164. [38] M. A. Rieffel, Induced representations of C ∗-algebras, Adv. Math. 13 (1974), 176 -- 257. [39] D. Sarason, Generalized interpolation in H ∞, Trans. Amer. Math. Soc. 127 (1967), no. 2, 179 -- 203. [40] O. M. Shalit and B. Solel, Subproduct systems, Doc. Math. 14 (2009), 801 -- 868. [41] B. Sz.-Nagy and C. Foiaş, Harmonic Analysis of Operators on Hilbert Space, North-Holland, Amsterdam, 1970. [42] B. Tsirelson, Graded algebras and subproduct systems: dimension two, preprint, arXiv:0905.4418. [43] B. Tsirelson, Subproduct systems of Hilbert spaces: dimension two, preprint, arXiv:0906.4255. [44] F.-H. Vasilescu, An operator-valued Poisson kernel, J. Funct. Anal. 110 (1992), no. 1, 47 -- 72. Department of Mathematics, Technion - Israel Institute of Technology, 32000 Haifa, Israel E-mail address: [email protected]
0905.3509	3	0905	2011-02-24T19:05:41	A treatment of the Cauchy--Schwarz inequality in $C^*$-modules	[ "math.OA", "math.FA" ]	We study the Cauchy--Schwarz and some related inequalities in a semi-inner product module over a $C^$-algebra $\A$. The key idea is to consider a semi-inner product $\A$-module as a semi-inner product $\A$-module with respect to another semi-inner product. In this way, we improve some inequalities such as the Ostrowski inequality and an inequality related to the Gram matrix. The induced semi-inner products are also related to the the notion of covariance and variance. Furthermore, we obtain a sequence of nested inequalities that emerges from the Cauchy--Schwarz inequality. As a consequence, we derive some interesting operator-theoretical corollaries. In particular, we show that the sequence arising from our construction, when applied to a positive element of a $C^$-algebra, converges to its pseudo-inverse.	math.OA	math	A TREATMENT OF THE CAUCHY -- SCHWARZ INEQUALITY IN C ∗-MODULES LJILJANA ARAMBASI ´C 1, DAMIR BAKI ´C 2 AND MOHAMMAD SAL MOSLEHIAN 3 Abstract. We study the Cauchy -- Schwarz and some related inequalities in a semi-inner product module over a C ∗-algebra A . The key idea is to consider a semi-inner product A -module as a semi-inner product A -module with respect to another semi-inner product. In this way, we improve some inequalities such as the Ostrowski inequality and an inequality related to the Gram matrix. The induced semi-inner products are also related to the the notion of covariance and variance. Furthermore, we obtain a sequence of nested inequalities that emerges from the Cauchy -- Schwarz inequality. As a consequence, we derive some interesting operator-theoretical corollaries. In particular, we show that the sequence arising from our construction, when applied to a positive invertible element of a C ∗-algebra, converges to its inverse. 1. Introduction Let A be a C ∗-algebra. A (right) semi-inner product A -module is a linear space X which is a right A -module with a compatible scalar multiplication (λ(xa) = x(λa) = (λx)a for all x ∈ X , a ∈ A , λ ∈ C) endowed with an A -semi- inner product h·, ·i : X × X → A such that for all x, y, z ∈ X , λ ∈ C, a ∈ A , it holds (i) hx, xi ≥ 0; (ii) hx, λy + zi = λhx, yi + hx, zi; (iii) hx, yai = hx, yia; (iv) hx, yi∗ = hy, xi. Obviously, every semi-inner product space is a semi-inner product C-module. We can define a semi-norm on X by kxk = khx, xik 2 , where the latter norm denotes that in the C ∗-algebra A . A pre-Hilbert A -module (or an inner-product module) is a semi-inner product module over A in which k · k defined as above 1 2010 Mathematics Subject Classification. Primary 46L08; Secondary 26D15, 46L05, 46C50, 47A30, 47A63. Key words and phrases. C ∗-algebra, semi-inner product C ∗-module, positive operator, Cauchy -- Schwarz inequality, Gram matrix, covariance -- variance inequality, Ostrowski inequality. 1 2 LJ. ARAMBASI ´C, D. BAKI ´C, M.S. MOSLEHIAN is a norm. A pre-Hilbert A -module X such that (X, k · k) is complete is called a Hilbert C ∗-module. Each C ∗-algebra A can be regarded as a Hilbert A -module via ha, bi = a∗b (a, b ∈ A ). Throughout the paper, fA stands for the minimal unitization of A . By e we denote the unit in fA . If X is an A -module then it can be regarded as an fA -module via xe = x. The basic theory of C ∗-algebras and Hilbert C ∗-modules can be found in [11, 18, 19]. One of the fundamental inequalities in a semi-inner product module (X , h·, ·i) over a C ∗-algebra A is the Cauchy -- Schwarz inequality. It states that hx, yihy, xi ≤ kyk2 hx, xi (x, y ∈ X ) (1.1) which generalizes the classical Cauchy -- Schwarz inequality. Today many gener- alizations of the classical Cauchy -- Schwarz inequality for integrals, isotone func- tionals as well as in the setting of inner product spaces are well-studied; see the book [4]. Moreover, Niculescu [15] and Joit¸a [10] have investigated the reverse of the Cauchy -- Schwarz inequalities in the framework of C ∗-algebras and Hilbert C ∗-modules, see also [14] and references therein. We also refer to another inter- esting paper by Ilisevi´c and Varosanec [9] of this type. Some operator versions of the Cauchy -- Schwarz inequality with simple conditions for the case of equality are presented in [6]. Using the Cauchy -- Schwarz inequality in a new suitably defined semi-inner product on X we improve some known inequalities in semi-inner product mod- ules. The most interesting improvement is the one of the Cauchy -- Schwarz in- equality itself. In this way, we improve some inequalities such as the Ostrowski in- equality (see [1]) and show that the Gram matrix [hxi, xji], where x1, . . . , xn ∈ X , is greater then or equal to some positive element of Mn(A ). The induced semi- inner products are also related to the the notion of covariance and variance. In the last section of the paper we repeat this technique by starting with an induced semi-inner product. This leads to a (possibly finite) sequence of nested inequalities refining the Cauchy -- Schwarz inequality. Namely, for every z ∈ X such that a := hz, zi 6= 0, we construct an increasing sequence of positive elements (pm(a))m ∈ A such that hx, xi ≥ hx, zi pm(a) hz, xi for all x, y ∈ X (see Theorem 3.2); thereby, for m = 0 we get kzk2hx, xi ≥ hx, zi hz, xi, i.e. the Cauchy -- Schwarz inequality. By analyzing the sequence (pm(a))m we obtain some interesting operator-theoretical consequences. In particular, if a is invertible, then (pm(a))m converges in norm to a−1 (Theorems 3.5 and 3.7). Moreover, in CAUCHY -- SCHWARZ INEQUALITY 3 Proposition 3.10 we show that, for a positive operator a on a Hilbert space H , the sequence (apm(a))m converges in norm to the orthogonal projection to Im a if and only if Im a is a closed subspace of H . 2. An induced family of semi-inner products Let (X , h·, ·i) be a semi-inner product module over a C ∗-algebra A . For an arbitrary z ∈ X we define h·, ·iz : X × X → A , hx, yiz := kzk2 hx, yi − hx, zihz, yi. (2.1) It is easy to see that h·, ·iz is a semi-inner product on X . (Note that the case when hz, zi = 0 gives a trivial semi-inner product; however, this does not contradict the definition of the semi-inner product.) In this section we show how one can improve, by using this new class of induced semi-inner products, several results known from the literature that rely on the Cauchy -- Schwarz inequality for the original semi-inner product in an arbitrary module. 2.1. The Gram matrix. We begin with the Gram matrix [hxi, xji] in the matrix C ∗-algebra Mn(A ) of all n × n matrices with entries from A due to it naturally appears in this context. Namely, positivity of the Gram matrix for two elements is strongly related to the Cauchy -- Schwarz inequality. To see this, let us write the Cauchy -- Schwarz inequality in a matrix form. Recall that a matrix" a b c # ∈ b∗ M2(A ) with invertible c ∈ A (resp. a ∈ A ) is positive if and only if a ≥ 0, c ≥ 0 and bc−1b∗ ≤ a (resp. a ≥ 0, c ≥ 0 and b∗a−1b ≤ c); see [3]. Therefore, (1.1) can be written as (2.2) hx, yi " hx, xi hx, yi∗ khy, yike # ≥ 0 , where e ∈ fA is the unit. Since " hx, xi hx, yi∗ khy, yike # ≥" hx, xi hx, yi∗ hx, yi hx, yi hy, yi # ≥ 0, it follows that positivity of the Gram matrix sharpens the Cauchy -- Schwarz in- equality. Remark 2.1. A number of arguments can be simplified if we use positivity of the Gram matrix. For example, it was proved in [9, Theorem 2.1] that for x, y ∈ X 4 LJ. ARAMBASI ´C, D. BAKI ´C, M.S. MOSLEHIAN such that y belongs to the center of A , a stronger version of the Cauchy -- Schwarz inequality holds, namely, hx, yihy, xi ≤ hx, xihy, yi. From positivity of the Gram matrix it follows that for every x, y ∈ X and every ε > 0 we have " hx, xi hx, yi∗ hx, yi hy, yi + εe # ≥ 0, or, equivalently, hx, yi(hy, yi + εe)−1hy, xi ≤ hx, xi. If y belongs to the cen- ter of A , we get, by multiplying by (hy, yi + εe) 2 on both sides, hx, yihy, xi ≤ 1 hx, xi(hy, yi + εe). Since ε > 0 is arbitrary, we have hx, yihy, xi ≤ hx, xihy, yi. Because of this stronger version of the Cauchy -- Schwarz inequality, we can define another family of semi-inner products on X : for every z ∈ X such that z belongs to the center of A , the mapping {·, ·}z : X × X → A , {x, y}z := hz, zihx, yi − hx, zihz, yi is a semi-inner product on X . Let us now consider the Gram matrix [hxi, xji] ∈ Mn(A ) for an arbitrary number of elements x1, . . . , xn in a semi-inner product module (X , h·, ·i). It is known that [hxi, xji] ≥ 0, i.e., the Gram matrix is a positive element of the C ∗-algebra Mn(A ). The interesting fact about this inequality is that it is self- improving, as we show in the following theorem. Theorem 2.2. Let A be a C ∗-algebra and (X , h·, ·i) a semi-inner product A - module. Let n ∈ N and x1, . . . , xn ∈ X . Then for every z ∈ X we have kzk2(cid:2) hxi, xji(cid:3) ≥(cid:2) hxi, zi hz, xji(cid:3). (2.3) Proof. Let us first prove that [hxi, xji] is positive in Mn(A ) (the proof is included for the convenience of the reader, see [11, Lemma 4.2]). Since h·, ·i is a semi-inner product on X it holds that Then * nXi=1 nXi,j=1 xiai, xiai+ ≥ 0, nXi=1 (a1, . . . , an ∈ A ). a∗ i hxi, xji aj ≥ 0, (a1, . . . , an ∈ A ). (2.4) CAUCHY -- SCHWARZ INEQUALITY 5 By [18, Lemma IV.3.2] we know that a matrix [cij] ∈ Mn(A ) is positive if and i cijaj ≥ 0 for all a1, . . . , an ∈ A . Therefore, (2.4) means that the i,j=1 a∗ matrix [hxi, xji] is positive. only ifPn we get(cid:2)hxi, xjiz(cid:3) ≥ 0, which is exactly (2.3). It holds for an arbitrary semi-inner product, so, choosing h·, ·iz instead of h·, ·i, (cid:3) The following corollary is a direct consequence of the preceding theorem. A positive linear mapping Φ : A → B, where B is a C ∗-subalgebra of A , is called a left multiplier if Φ(ab) = Φ(a)b (a ∈ A , b ∈ B). Corollary 2.3. Let (X , h·, ·i) be a semi-inner product A -module, B a C ∗- subalgebra of A and Φ : A → B a positive left multiplier. Then kΦ(hz, zi)k[Φ(hxi, xji)] ≥ [Φ(hxi, zi)Φ(hz, xj)] (2.5) for all x1, · · · , xn, z ∈ X . Proof. Given a left multiplier Φ : A → B, any semi-inner product A -module X becomes a semi-inner product B-module with respect to [x, y]Φ = Φ(hx, yi), (x, y ∈ X ). (2.6) By (2.3), it holds k[z, z]Φk(cid:2)[xi, xj]Φ(cid:3) ≥(cid:2)[xi, z]Φ[z, xj]Φ(cid:3) . (cid:3) Remark 2.4. Let X be a C ∗-algebra regarded as a Hilbert C ∗-module over itself. Since every conditional expectation Φ : A → B is a completely positive left multiplier (cf. [18, IV, §3]), the preceding corollary is an extension of [2, Theorem 1] for conditional expectations. 2.2. A covariance -- variance inequality. Another application of Theorem 2.2 is the covariance -- variance inequality in semi-inner product C ∗-modules. The interested reader is referred to [2, 7, 12] for some generalizations of covariance -- variance inequality. Let us begin with a definition and some known examples. Definition 2.5. Let A be a C ∗-algebra, (X , h·, ·i) be a semi-inner product A - module and x, y, z ∈ X . The covariance covz(x, y) between x and y with respect to z is defined to be the element hx, yiz of A . The element covz(x, x) is said to be the variance of x with respect to z and denoted by varz(x). 6 LJ. ARAMBASI ´C, D. BAKI ´C, M.S. MOSLEHIAN Example 2.6. Given a Hilbert space H , vectors x, y ∈ H and operators S, T ∈ B(H ), covariance and variance of operators was defined in [12] as covx,y(S, T ) = kyk2(SxT x) − (Sxy)(yT x). Observe that covx,y(S, T ) = (SxT x)y. In the case where kxk = 1 and y = x we get the notion of covariance of two operators T and S introduced in [7] as covx(S, T ) = (SxT x) − (Sxx)(xT x) . A notion of covariance and variance of Hilbert space operators was investigated in [7, 17]. In addition, Enomoto [5] showed a close relation of the operator covariance -- variance inequality with the Heisenberg uncertainty principle and pointed out that it is exactly the generalized Schrodinger inequality. For more information on related ideas and concepts we refer the reader to [16, Section 5]. Another remarkable fact is that for a unit vector x ∈ H , the determinant of the positive semidefinite Gram matrix (SxSx) (SxT x) (Sxx) (T xSx) (T xT x) (T xx) (xSx) (xT x) (xx)   is the difference varx(S)varx(T ) − covx(S, T )2 and is nonnegative; see [8]. Example 2.7. Recall that if (Ω, µ) is a probability measure space, then Ef = ance between f and g is defined to be cov(f, g) = E(f g) − Ef Eg and variance of f is cov(f, f ). We can obtain this by considering L2(Ω, µ) as a Hilbert C-module RΩ f dµ is the expectation of the random variable f ∈ L2(Ω, µ). Then the covari- via the usual inner product hf, gi =RΩ f g. Let A be a C ∗-algebra and X be a semi-inner product A -module. The Cauchy -- Schwarz inequality for covz(·, ·) is known as the covariance -- variance in- equality. Therefore, Theorem 2.2 can be also stated in the following form. Theorem 2.8 (Generalized covariance-variance inequality). Let A be a C ∗- algebra and X be a semi-inner product A -module. Let x1, . . . , xn, z ∈ X . Then the matrix [covz(xi, xj)] ∈ Mn(A ) is positive. Assume that A is a C ∗-algebra acting on a Hilbert space, B is one of its C ∗- subalgebras and X is a Hilbert A -module. Let us fix a positive left multiplier mapping Φ and x ∈ X such that kΦ(hx, xi)k = 1. For operators A and B in the CAUCHY -- SCHWARZ INEQUALITY 7 algebra B(X ) of all adjointable operators on X we could define the covariance of A, B and variance of A by cov(A, B) = Φ(hAx, Bxi) − Φ(hAx, xi)Φ(hx, Bxi) and var(A) = cov(A, A), respectively. Observe that, if we regard X as a semi- inner product A -module with respect to [·, ·]Φ defined by (2.6), then we have cov(A, B) = covx(Ax, Bx). Therefore, " var(A) cov(A, B)∗ cov(A, B) var(B) # =" varx(Ax) covx(Ax, Bx) covx(Ax, Bx)∗ varx(Bx) # ≥ 0. 2.3. An Ostrowski type inequality. Here we show that some Ostrowski-type inequalities can be viewed as the Cauchy -- Schwarz inequality with respect to a new semi-inner product. It was proved in [13] that for any three elements in a real inner product space (H, (··)) it holds (cid:12)(cid:12)kzk2(xy) − (xz)(yz)(cid:12)(cid:12)2 ≤(cid:0)kzk2kxk2 − (xz)2(cid:1)(cid:0)kzk2kyk2 − (yz)2(cid:1) . Since here H is a real vector space, this may be written as (xy)z2 ≤ (xx)z(yy)z and this is exactly the Cauchy -- Schwarz inequality for (··)z. Therefore, the Cauchy -- Schwarz inequality for a semi-inner product h·, ·iz on a semi-inner product module X , i.e. (kzk2 hy, xi − hy, zihz, xi)(kzk2 hx, yi − hx, zihz, yi) ≤(cid:13)(cid:13)kzk2 hx, xi − hx, zihz, xi(cid:13)(cid:13) (kzk2 hy, yi − hy, zihz, yi) generalizes the result from [13]. In the special case when hx, zi = 0 we get hz, yi 2 ≤ kzk2 kxk2 (kxk2y2 − hx, yi 2), (2.7) (2.8) which is the Ostrowski inequality in a semi-inner product C ∗-module (see [1]). Since (2.7) is the Cauchy -- Schwarz inequality and (2.8) is its special case, Theo- rem 2.2 improves both of them. Namely, " hx, xiz hx, yi∗ z hx, yiz hy, yiz # ≥ 0, (which is exactly (2.3) for n = 2) improves (2.7), and it improves (2.8) in the case hx, zi = 0. 8 LJ. ARAMBASI ´C, D. BAKI ´C, M.S. MOSLEHIAN 3. A nested sequence of inequalities In this section we show that the Cauchy -- Schwarz inequality can be improved by a sequence of nested inequalities. To do that, let us first fix some notation. Let A be a C ∗-algebra and e ∈ fA . For a positive element a ∈ A , a 6= 0, define f0(a) = a, f1(a) = f0(a)g0(a), . . . g0(a) = kf0(a)ke − f0(a), g1(a) = kf1(a)ke − f1(a), (3.1) fm(a) = fm−1(a)gm−1(a), gm(a) = kfm(a)ke − fm(a), . . . Observe that all fm(a) and gm(a) are polynomials in a. An easy inductive argument shows that all fm(a) and gm(a) are positive elements as well, and for all m ≥ 0 it holds fm+1(a) = fm(f1(a)), gm+1(a) = gm(f1(a)). (3.2) It may happen that fm(a) = 0 for some m ∈ N (see Proposition 3.3 below); then, obviously, fk(a) = 0 for all k ≥ m. On the other hand, if fm(a) 6= 0 for some m, then, by definition, fj(a) 6= 0, ∀j ≤ m. Thus, for each m such that fm(a) 6= 0 we can define kf0(a)k , kf0(a)k + kf0(a)k + p0(a) = e p1(a) = e p2(a) = e . . . pm(a) = e g0(a)2 kf0(a)k·kf1(a)k , kf0(a)k·kf1(a)k + g0(a)2 g0(a)2g1(a)2 kf0(a)k·kf1(a)k·kf2(a)k , (3.3) l=1(cid:16) kf0(a)k +Pm Ql 1 k=0 kfk(a)kQl−1 k=0 gk(a)2(cid:17) . It is convenient here to make the following convention: if m is the last index such that fm(a) 6= 0 then we define pj(a) = pm(a), (j > m). (3.4) Thus, we can treat (pm(a)) as an infinite sequence of positive elements in A even in the case when there is m ≥ 0 such that fm(a) = 0. Remark 3.1. It is obvious that 0 ≤ p0(a) ≤ p1(a) ≤ . . . ≤ pm(a) ≤ . . .. Observe also that pm(a)'s which are defined by (3.3) are all different. Indeed, suppose CAUCHY -- SCHWARZ INEQUALITY 9 pm−1(a) and pm(a) are defined by (3.3) and pm−1(a) = pm(a). Then 1 k=0 kfk(a)k Qm gk(a)2 = 0 m−1Yk=0 which implies g0(a)g1(a) · · · gm−1(a) = 0 and therefore fm(a) = fm−1(a)gm−1(a) = fm−2(a)gm−2(a)gm−1(a) = . . . = f0(a)g0(a) · · · gm−1(a) = 0. This is the contradiction, since pm(a) is defined by (3.3). Theorem 3.2. Let X be a module over a C ∗-algebra A , and let h·, ·i be any A -valued semi-inner product on X . For each z ∈ X such that hz, zi 6= 0 it holds hx, xi ≥ . . . ≥ hx, zi pm(hz, zi) hz, xi ≥ hx, zi pm−1(hz, zi) hz, xi ≥ . . . ≥ hx, zi p0(hz, zi) hz, xi = 1 kzk2 hx, zi hz, xi ≥ 0. Proof. We will prove by induction that hx, xi∗ ≥ hx, zi∗pm(hz, zi∗)hz, xi∗ (3.5) holds true for all m ≥ 0, for each z ∈ X and for every A -valued semi-inner product h·, ·i∗ on X such that hz, zi∗ 6= 0. For m = 0 this is precisely the statement of Theorem 2.2 for n = 1. Suppose that (3.5) holds for some m and for all z and h·, ·i∗ such that hz, zi∗ 6= 0. Choose an arbitrary semi-inner product h·, ·i on X such that hz, zi 6= 0. If fm+1(hz, zi) = 0, there is nothing to prove since then, by our convention, pm+1(hz, zi) = pm(hz, zi). Suppose now that fm+1(hz, zi) 6= 0. Then, by (3.2), fm(f1(hz, zi) = fm(hz, ziz) 6= 0. By the inductive assumption (for the semi-inner product h·, ·iz) it holds hx, xiz ≥ hx, zizpm(hz, ziz)hz, xiz, that is, kzk2 hx, xi ≥ hx, zi hz, xi + hx, zizpm(hz, ziz)hz, xiz. (3.6) 10 LJ. ARAMBASI ´C, D. BAKI ´C, M.S. MOSLEHIAN Observe that hz, ziz = f1(hz, zi), so kzk2e − hz, zi and pm(hz, ziz) commute. Therefore hx, zizpm(hz, ziz)hz, xiz = (kzk2 hx, zi − hx, zi hz, zi)pm(hz, ziz) ·(kzk2 hz, xi − hz, zi hz, xi) = hx, zi (kzk2e − hz, zi)2pm(hz, ziz) hz, xi = hx, zi (g0(hz, zi))2 pm (f1(hz, zi)) hz, xi . Since fm+1(hz, zi) 6= 0 and fm(hz, ziz) 6= 0, the elements pm+1(hz, zi) and pm(hz, ziz) are defined by (3.3). It is easy to verify, using (3.1) and (3.2), that e + g0(hz, zi)2pm(f1(hz, zi)) = kzk2pm+1(hz, zi), (3.7) which, together with (3.6), gives hx, xi ≥ hx, zi pm+1(hz, zi) hz, xi. To complete the proof it only remains to recall that pm(hz, zi) ≥ pm−1(hz, zi) ≥ (cid:3) . . . ≥ p0(hz, zi) ≥ 0, for every z ∈ X . If hz, zi is not a scalar multiple of the unit, then f1(hz, zi) 6= 0 and the pre- ceding theorem strictly refines the inequality from Theorem 2.2. Moreover, if fm(hz, zi) 6= 0 for all m ∈ N, Theorem 3.2 provides an infinite sequence of in- equalities. On the other hand, if fm(hz, zi) = 0 for some m ≥ 0, then, by (3.4), only finitely many inequalities are obtained. The following proposition charac- terizes all such elements z ∈ X . It turns out that the sequence of inequalities obtained in Theorem 3.2 is finite precisely when hz, zi has a finite spectrum. Proposition 3.3. Let a be a positive element of a C ∗-algebra A ⊆ B(H ). Then there exists m ∈ N such that fm(a) = 0 if and only if a has a finite spectrum. Proof. Suppose that there is m ∈ N such that fm(a) = 0. Let λ ∈ σ(a). Then fm(λ) ∈ fm(σ(a)) = σ(fm(a)) = {0}. This shows that σ(a) is contained in a finite set, namely in the set of all zeros of the polynomial fm. To prove the converse, suppose that σ(a) is a finite set. First observe that 0 ∈ σ(fm(a)) for all m ≥ 1. Let m ≥ 1 be such that fm(a) 6= 0 (if such m does not exist, we are done). Since σ(fm+1(a)) = {kfm(a)kλ − λ2 : λ ∈ σ(fm(a))}, and since 0 and kfm(a)k are two different elements of σ(fm(a)) such that kfm(a)k0 − 02 = kfm(a)k · kfm(a)k − kfm(a)k2 = 0, we conclude that card σ(fm+1(a)) < card σ(fm(a)). Since σ(a) is finite, there is m such that σ(fm(a)) = {0}, i.e. fm(a) = 0. (cid:3) CAUCHY -- SCHWARZ INEQUALITY 11 Remark 3.4. Suppose that A = C, i.e. that X is a semi-inner product space. Then for each z ∈ X the spectrum σ(hz, zi) is a singleton, so f1(hz, zi) = 0. Hence, in this situation, the sequence of inequalities from Theorem 3.2 termi- nates already at the first step. In other words, Theorem 3.2 reduces then to Theorem 2.2. Therefore, Theorem 3.2 gives us a new (possibly finite) sequence of inequalities only if the underlying C ∗-algebra is different from the field of complex numbers. Let us first consider the case from the preceding proposition, when the sequence of the inequalities is finite. The following result is interesting in its own. If a ∈ A is such that fM (a) 6= 0 and fM +1(a) = 0 for some M ∈ N, we show that, roughly speaking, pM (a) is the inverse of a. Theorem 3.5. Let a 6= 0 be a positive element in a C ∗-algebra A ⊆ B(H ) with a finite spectrum. Let M be the number with the property fM (a) 6= 0, fM +1(a) = 0. Then apM (a) is the orthogonal projection to the image of a. In particular, if a is an invertible operator, then pM (a) = a−1. Proof. Let us first observe that, since the spectrum of a is finite, Im a is a closed subspace of H . For every λ ∈ R and l ∈ N it holds λ l−1Yk=0 gk(λ) = f0(λ)g0(λ)g1(λ) · · · gl−1(λ) = f1(λ)g1(λ) · · · gl−1(λ) = f2(λ)g2(λ) · · · gl−1(λ) = . . . = fl(λ). Since fM (a) 6= 0 and fM +1(a) = 0 we conclude that p0(a), . . . , pM (a) are defined by (3.3), while, by (3.4), pj(a) = pM (a) for j ≥ M + 1. Therefore, for λ 6= 0 and m = 1, . . . , M it holds pm(λ) = 1 kf0(a)k + 1 λ2 mXl=1 Ql fl(λ)2 k=0 kfk(a)k . (3.8) Let m ∈ {1, . . . , M}. Since fm(a) ≥ 0 (and fm(a) 6= 0), kfm(a)k is the max- imum of the set σ(fm(a)) = fm(σ(a)). Let λm ∈ σ(a) be such that fm(λm) = kfm(a)k. Then gm(λm) = kfm(a)k − fm(λm) = 0 and therefore fj(λm) = 0 for all j ≥ m + 1. Since obviously λm 6= 0, (3.8) gives pj(λm) = pm(λm) for all j ∈ {m, . . . , M}. Therefore, for j ∈ {m, . . . , M} we have pj(λm) = pm(λm) = 1 kf0(a)k + 1 λ2 m m−1Xl=1 fl(λm)2 k=0 kfk(a)k Ql + fm(λm)2 k=0 kfk(a)k! . Qm 12 LJ. ARAMBASI ´C, D. BAKI ´C, M.S. MOSLEHIAN Using fm(λm) = kfm(a)k, for all j ∈ {m, . . . , M} we get pj(λm) = = = 1 kf0(a)k 1 kf0(a)k 1 kf0(a)k + + + 1 λ2 1 λ2 1 λ2 m m−1Xl=1 m m−2Xl=1 m m−2Xl=1 fl(λm)2 k=0 kfk(a)k fl(λm)2 k=0 kfk(a)k fl(λm)2 k=0 kfk(a)k Ql Ql Ql + + + (3.9) fm(λm) fm−1(λm)2 k=0 kfk(a)k k=0 kfk(a)k! Qm−1 k=0 kfk(a)k! Qm−1 Qm−1 k=0 kfk(a)k ! . Qm−1 fm−1(λm)2 + fm(λm) fm(λm) + Observe that for every k ∈ N and every λ ∈ R it holds: fk−1(λ)2 + fk(λ) = fk−1(λ)2 + fk−1(λ)gk−1(λ) = fk−1(λ)(fk−1(λ) + gk−1(λ)) = fk−1(λ)kfk−1(a)k. Therefore, for j ∈ {m, . . . , M}, We now proceed recursively in the same way as (3.10) is obtained from (3.9) to pj(λm) = 1 kf0(a)k = 1 kf0(a)k + + 1 λ2 1 λ2 m m−2Xl=1 m m−2Xl=1 get pj(λm) = = = = = 1 kf0(a)k 1 kf0(a)k 1 kf0(a)k + + + 1 λ2 1 λ2 1 λ2 1 + 1 λ2 kf0(a)k m λm + g0(λm) λmkf0(a)k fm−1(λm)kfm−1(a)k k=0 kfk(a)k ! Qm−1 k=0 kfk(a)k! . Qm−2 fm−1(λm) (3.10) + f2(λm) k=0 kfk(a)k! Q1 kf0(a)kkf1(a)k(cid:19) f2(λm) + + fl(λm)2 k=0 kfk(a)k fl(λm)2 k=0 kfk(a)k Ql Ql m 1Xl=1 Ql m(cid:18) m(cid:18) f1(λm)kf1(a)k kf0(a)kkf1(a)k(cid:19) fl(λm)2 k=0 kfk(a)k kf0(a)kkf1(a)k f1(λm)2 + λmg0(λm) kf0(a)k = kf0(a)k λmkf0(a)k = 1 λm for j ∈ {m, . . . , M}. After all, we have proved: if λm ∈ σ(a) is such that kfm(a)k = fm(λm) for some m ∈ {0, . . . , M} then pj(λm) = 1 λm for all j ∈ {m, . . . , M}. Let us take particular λ ∈ σ(a), λ 6= 0. From fM +1(a) = 0 it follows that fM +1(λ) = 0. Then there exists m ≤ M such that fm(λ) 6= 0 and fm+1(λ) = 0. Then from fm+1(λ) = fm(λ)gm(λ) we get gm(λ) = 0, i.e., fm(λ) = kfm(a)k. CAUCHY -- SCHWARZ INEQUALITY 13 This means that for every λ ∈ σ(a) there is m ≤ M such that pj(λ) = 1 j ∈ {m, . . . , M}. Since σ(a) is finite, there is m ≤ M such that λ for all and therefore Then pm(λ) = pM (λ) = 1 λ , 1 λ , ∀λ ∈ σ(a) \ {0} ∀λ ∈ σ(a) \ {0}. λpM (λ) =( 1, λ ∈ σ(a) \ {0}, 0, λ ∈ σ(a) ∩ {0}. This is precisely what we need to conclude that apM (a) is the orthogonal projec- tion to Im a. In the case when a is invertible, then λpM (λ) = 1 for all λ ∈ σ(a), so apM (a) = e. (cid:3) Let us now consider the sequence (pm(a)) in full generality. Again, we assume that a is a positive operator on some Hilbert space H (i.e. A is represented faithfully on H ). Denote by p the orthogonal projection to Im a. If a has a finite spectrum we have seen in the preceding theorem that apM (a) is equal to p, where M is less than or equal to the number of elements of σ(a); in particular, if a is an invertible operator, then pM (a) = a−1. If σ(a) is an infinite set we know that fm(a) is never equal to 0; thus, (pm(a)) is in this case an increasing sequence of positive elements of B(H ). It would be natural to expect that in this situation the sequence (apm(a)) converges to p in norm. However, this is not true in general, as the following example shows. Example 3.6. Suppose that a is a positive compact operator with an infinite spectrum. Then the sequence (apm(a)) cannot converge to p in norm. Indeed, suppose the opposite. Observe that pm(a) ∈ C ∗(a), for all m ≥ 0, where C ∗(a) denotes the C ∗-algebra generated by a. Since C ∗(a) is closed, the assumption would imply p ∈ C ∗(a). But this is impossible: since σ(a) is an infi- nite set, Im a is an infinite-dimensional subspace and hence p (as a non-compact operator) cannot belong to C ∗(a). In this light, the following theorem is the best possible extension of Theorem 3.5. Theorem 3.7. Let a be a positive element in a C ∗-algebra A ⊆ B(H ) with an infinite spectrum. Then limm→∞ apm(a)a = a. In particular, if a is an invertible operator, limm→∞ pm(a) = a−1. 14 LJ. ARAMBASI ´C, D. BAKI ´C, M.S. MOSLEHIAN Proof. Since σ(a) is not finite, fm(a) 6= 0 for all m ∈ N, so every pm(λ) is defined by (3.3). Take an arbitrary m ∈ N. For every λ ∈ σ(a) we have 1 − λpm(λ) = (cid:18)1 − = g0(λ) kak k=0 gk(λ) k=0 kfk(a)k fl(λ)Ql−1 Ql k=0 gk(λ)2 k=0 kfk(a)k λ − − k=0 gk(λ) f1(λ)g0(λ) k=0 kfk(a)k λQl−1 kak(cid:19) − mXl=1 Ql fl(λ)Ql−1 mXl=1 Ql kakf1(a)k(cid:19) − mXl=2 fl(λ)Ql−1 mXl=2 Ql fm(λ)Qm−1 Qm − − k=0 gk(λ) k=0 kfk(a)k k=0 gk(λ) k=0 kfk(a)k = (cid:18) g0(λ) kak = g0(λ)g1(λ) kakkf1(a)k = . . . k=0 gk(λ) k=0 kfk(a)k k=0 gk(λ) k=0 kfk(a)k = Qm−1 Qm−1 = Qm Qm . k=0 gk(λ) k=0 kfk(a)k λQm Qm = fm+1(λ) k=0 kfk(a)k Qm Then and therefore λ − λ2pm(λ) = ka − apm(a)ak = sup λ∈σ(a) {λ − λ2pm(λ)} = sup λ∈σ(a) { fm+1(λ) k=0 kfk(a)k } ≤ kfm+1(a)k k=0 kfk(a)k . Qm Qm From σ(fk(a)) ⊆ [0, kfk(a)k] and fk+1(λ) = kfk(a)kfk(λ) − fk(λ)2 it follows that σ(fk+1(a)) ⊆ [0, 1 4 kfk(a)k2 for all k. Then ka − apm(a)ak ≤ 4 kfk(a)k2], so kfk+1(a)k ≤ 1 1 4 kfm+1(a)k k=0 kfk(a)k ≤ Qm Qm−1 1 42 kfm−1(a)k2 k=0 kfk(a)k 1 ≤ 1 4m+1 kak. Since m is arbitrary, we conclude that limm→∞ apm(a)a = a. kf1(a)k 1 4m 4m+1 kak ≤ = = = 1 4 Qm kfm(a)k2 k=0 kfk(a)k 1 42 kak2 kak Qm−2 = kfm−1(a)k k=0 kfk(a)k kfm(a)k k=0 kfk(a)k Qm−1 ≤ . . . (cid:3) Remark 3.8. From limm→∞ apm(a)a = a one easily gets limm→∞ apm(a) = p in the strong operator topology (where, as before, p denotes the orthogonal projection to Im a). CAUCHY -- SCHWARZ INEQUALITY 15 By the preceding remark, p is the only possible norm-limit of the sequence (apm(a)). In the following proposition we characterize those positive operators a for which the sequence (apm(a)) converges to p in norm. First we need a lemma. Keeping the notation from the preceding paragraphs, let us also fix the following notational conventions: for a positive operator a ∈ B(H ) on a Hilbert space H denote H1 = Im a and H2 = Ker a. According to the decomposition H = H1 ⊕ H2 we can write a = " a1 0 0 #. For the operators a and a1 we 0 denote by (fm(a)) and (f (1) and (p(1) m (a1)) those defined by (3.3) and (3.4). m (a1)) the sequences defined by (3.1) and by (pm(a)) Lemma 3.9. fm(a) =" f (1) 0 m (a1) 0 0 # and pm(a) =" p(1) 0 m (a1) 0 0 #, ∀m ≥ 0. kf (1) Proof. The first assertion is trivial for m = 0. Let Ij denote the identity op- erator on Hj for j = 1, 2. Observe that kak = ka1k which means kf0(a)k = 0 (a1)k. This implies f1(a) =" a1 0 0 # =" f (1) " a1(kf (1) 0 (a1)kI1 − a1) 0 0 #" kf0(a)kI1 − f0(a1) kf0(a)kI2 # = 0 #. A general inductive argument is 1 (a1) 0 0 0 0 0 0 obtained exactly in the same way. The second assertion now follows from the first one combined with (3.8). (cid:3) Proposition 3.10. Let a ∈ B(H ) be a positive operator and p ∈ B(H ) the orthogonal projection to Im a. Then (apm(a))m converges to p in norm if and only if Im a is a closed subspace of H . Proof. Suppose first that a has a closed range, i.e. Im a = Im a. Then a1 = aIm a : Im a → Im a is a bijection. Since Im a is a Hilbert space, a1 is an invertible operator. By Theorem 3.7, the sequence (a1p(1) m (a1)) converges in norm m (a1) = I1. By the preceding lemma apm(a) =" a1p(1) 0 # 0 # which is the orthogonal projection to Im a = Im a. m (a1) 0 0 converges in norm to" I1 0 0 and limm→∞ a1p(1) Conversely, suppose that (apm(a)) converges in norm. As we already noted, the limit is then necessarily p. By the second assertion of the preceding lemma, I1 is then the norm-limit of the sequence (a1p(1) m (a1)). Since the group of invertible operators is open, it follows that a1p(1) m (a1) is an invertible operator, for m large 16 LJ. ARAMBASI ´C, D. BAKI ´C, M.S. MOSLEHIAN enough. In particular, a1p(1) m (a1) is a surjection and hence a1 is a surjection as well. Thus, Im a1 = Im a. Since, obviously, Im a = Im a1, this shows that a has a closed range. (cid:3) Notice that each positive operator with a finite spectrum has a closed range. Thus, the preceding proposition is in the accordance with Theorem 3.5. At the same time, it provides another explanation of Example 3.6 since a compact positive operator with an infinite spectrum cannot have a closed range. Concluding remarks: (a) To complete our analysis, let us first turn back to the sequence of inequalities from Theorem 3.2. If z ∈ X has the property that σ(hz, zi) is finite then, by Proposition 3.3 and Theorem 3.5, there exists M ∈ N such that fM (hz, zi) = 0, fM +1(hz, zi) 6= 0 and hz, zi pM (hz, zi) is the projection to Im hz, zi . In this case, the sequence of inequalities from Theorem 3.2 is finite and the last term between 1 kzk2 hx, zi hz, xi and hx, xi is hx, zi pM (hz, zi) hz, xi (for all x). The following claim explains the reason: the sequence terminates at that place because hx, zi pM (hz, zi) hz, xi is the maximal element of the set of positive elements under consideration. Claim. Let X be a semi-inner product module over a C ∗-algebra A ⊆ B(H ). For z ∈ X and a = hz, zi ∈ A , let p ∈ B(H ) denote the orthogonal projection to Im a. Suppose that there exists a positive operator h ∈ B(H ) such that for all x ∈ X and every m ≥ 0 it holds hx, xi ≥ hx, zi h hz, xi ≥ hx, zi pm(hz, zi) hz, xi . (3.11) Then aha = a and ah = p. Proof. It follows from (3.11) that hz, zi ≥ hz, zihhz, zi ≥ hz, zipm(hz, zi)hz, zi, ∀m ≥ 0, that is, a ≥ aha ≥ apm(a)a for all m ≥ 0. By Theorem 3.7 (or Remark 3.8), it follows that aha = a. This implies ah = p. (cid:3) Suppose now, as in the discussion preceding the above claim, that there exists M ∈ N such that fM (a) 6= 0 and fM +1(a) = 0. Then a has a finite spectrum, Im a is a closed subspace, and apM (a) = p. So, if h is as in the above claim, then ah = p and therefore apM (a) = ah. By taking adjoints we get ha = pM (a)a and this shows that h and pM (a) coincide on Im a. CAUCHY -- SCHWARZ INEQUALITY 17 If σ(a) is infinite, there is no M as above, but still the sequence (apm(a)a) converges in norm to a. From the proof of the claim it follows that for any h ∈ B(H ) which satisfies left-hand side inequality of (3.11) it holds aha ≤ a. Therefore, we have a kind of best result even in this case, since h which appears in (3.11) is such that aha = limm→∞ apm(a)a = a. (b) Observe that, if b ∈ A is positive and such that kzb 1 2 k ≤ 1, then, by Theorem 2.2, we have hx, xi ≥ kzb 1 2 k2 hx, xi ≥Dx, zb 1 2 , xE = hx, zi b hz, xi 1 2EDzb for every x ∈ X . Thus, the inequalities from Theorem 3.2 can alternatively be derived from the inequalities kzpm(hz, zi) these inequalities directly, we opted for the inductive approach from the proof of Theorem 3.2 since it leads naturally to the sequence (fm(a)) and gives us more insight into the sequence pm(a) which, as we have seen, has many interesting properties. Instead of proving 2 k ≤ 1, m ∈ N. 1 (c) The assertion of Theorem 3.2 can be formulated for Gram matrices as well; the proof requires no essential changes. In this way, one obtains the result that directly improves Theorem 2.2. Acknowledgment. The authors wish to thank the referee for several helpful comments and valuable suggestions. The third author was supported by a grant from Ferdowsi University of Mashhad (No. MP87037MOS). References [1] Lj. Arambasi´c and R. Raji´c, Ostrowski's inequality in pre-Hilbert C ∗-modules, Math. Ineq. Appl. 12 (2009), no. 1, 217 -- 226. [2] R. Bhatia and C. Davis, More operator versions of the Schwarz inequality, Comm. Math. Phys. 215 (2000), no. 2, 239 -- 244. [3] M.-D., Choi, Some assorted inequalities for positive linear maps on C ∗-algebras, J. Oper- ator Theory 4 (1980), 271 -- 285. [4] S.S. Dragomir, A survey on Cauchy -- Bunyakovsky -- Schwarz type discrete inequalities, JI- PAM. J. Inequal. Pure Appl. Math. 4 (2003), no. 3, Article 63, 142 pp. [5] M. Enomoto, Commutative relations and related topics, Surikaisekikenkyusho Kokyuroku, Kyoto University, 1039 (1998), 135 -- 140. [6] J.I. Fujii, Operator-valued inner product and operator inequalities, Banach J. Math. Anal. 2 (2008), no. 2, 59 -- 67. 18 LJ. ARAMBASI ´C, D. BAKI ´C, M.S. MOSLEHIAN [7] M. Fujii, T. Furuta, R. Nakamoto, and S.-E. Takahasi, Operator inequalities and covariance in noncommutative probability, Math. Japon. 46 (1997), no. 2, 317 -- 320. [8] M. Fujii, S. Izumino, R. Nakamoto, and Y. Seo, Operator inequalities related to Cauchy -- Schwarz and Holder -- McCarthy inequalities, Nihonkai Math. J. 8 (1997), no. 2, 117 -- 122. [9] D. Ilisevi´c and S. Varosanec, On the Cauchy -- Schwarz inequality and its reverse in semi- inner product C ∗-modules, Banach J. Math. Anal. 1 (2007), no. 1, 78 -- 84. [10] M. Joit¸a, On the Cauchy -- Schwarz inequality in C ∗-algebras, Math. Rep. (Bucur.) 3(53) (2001), no. 3, 243 -- 246. [11] E.C. Lance, Hilbert C ∗-Modules, London Math. Soc. Lecture Note Series 210, Cambridge Univ. Press, 1995. [12] C.S. Lin, On variance and covariance for bounded linear operators, Acta Math. Sin. (Engl. Ser.) 17 (2001), 657 -- 668. [13] J. Ma, An identity in real inner product space, J. Inequal. Pure and Appl. Math., 8(2) (2007), Art. 48, 4 pp. [14] M.S. Moslehian and L.-E. Persson, Reverse Cauchy-Schwarz inequalities for positive C ∗- valued sesquilinear forms, Math. Inequal. Appl. 4 (2009), no. 4, 701 -- 709. [15] C.P. Niculescu, Converses of the Cauchy -- Schwarz inequality in the C ∗-framework, An. Univ. Craiova Ser. Mat. Inform. 26 (1999), 22 -- 28. [16] K.R. Parthasarathy, An introduction to quantum stochastic calculus, Monographs in Math- ematics, 85. Birkhauser Verlag, Basel, 1992. [17] Y. Seo, Variance in noncommutative probability, Math. Japon. 46 (1997), no. 3, 439 -- 444. [18] M. Takesaki, Theory of Operator Algebras. I, Reprint of the first (1979) edition. Encyclopae- dia of Mathematical Sciences. 124. Operator Algebras and Non-commutative Geometry, 5. Springer-Verlag, Berlin, 2002. [19] N. E. Wegge-Olsen, K-theory and C*-algebras - a friendly approach, Oxford University Press, Oxford, 1993. 1 2 Department of Mathematics, University of Zagreb, Bijenicka cesta 30, 10000 Zagreb, Croatia. E-mail address: [email protected] E-mail address: [email protected] 3 Department of Pure Mathematics, Center of Excellence in Analysis on Al- gebraic Structures (CEAAS), Ferdowsi University of Mashhad, P.O. Box 1159, Mashhad 91775, Iran. E-mail address: [email protected] and [email protected]
1703.02924	3	1703	2018-02-28T17:21:14	A completely bounded non-commutative Choquet boundary for operator spaces	[ "math.OA", "math.FA" ]	We develop a completely bounded counterpart to the non-commutative Choquet boundary of an operator space. We show how the class of completely bounded linear maps is too large to accommodate our purposes. To overcome this obstacle, we isolate the subset of completely bounded linear maps on an operator space admitting a dilation of the same norm which is multiplicative on the generated $C^$-algebra. We view such maps as analogues of the familiar unital completely contractive maps, and we exhibit many of their structural properties. Of particular interest to us are those maps which are extremal with respect to a natural dilation order. We establish the existence of extremals and show that they have a certain unique extension property. In particular, they give rise to $$-homomorphisms which we use to associate to any representation of an operator space an entire scale of $C^$-envelopes. We conjecture that these $C^$-envelopes are all $*$-isomorphic, and verify this in some important cases.	math.OA	math	A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY FOR OPERATOR SPACES RAPHAEL CLOU ATRE AND CHRISTOPHER RAMSEY Abstract. We develop a completely bounded counterpart to the non-commu- tative Choquet boundary of an operator space. We show how the class of completely bounded linear maps is too large to accommodate our purposes. To overcome this obstacle, we isolate the subset of completely bounded linear maps on an operator space admitting a dilation of the same norm which is multiplicative on the generated C ∗-algebra. We view such maps as analogues of the familiar unital completely contractive maps, and we exhibit many of their structural properties. Of particular interest to us are those maps which are extremal with respect to a natural dilation order. We establish the existence of extremals and show that they have a certain unique extension property. In particular, they give rise to ∗-homomorphisms which we use to associate to any operator space an entire scale of C ∗-envelopes. We conjecture that these C ∗-envelopes are all ∗-isomorphic, and verify this in some important cases. 1. Introduction Let X be a compact metrizable space and let A be a uniform algebra on X, that is a closed unital algebra of continuous functions which separates the points of X. The Shilov boundary is the smallest closed subset ΣA ⊂ X with the property that the restriction map is isometric. One way of constructing ΣA is to take the closure of the Choquet boundary of A, which consists of the points ξ ∈ X with the property that there is a unique probability measure on X, namely the point mass at ξ, satisfying f 7→ fΣA , f ∈ A f (ξ) =ZX f dµ, f ∈ A. In other words, the point ξ lies in the Choquet boundary of A if the point evaluation functional f ∈ A extends to a unique state on the C∗-algebra C(X). f 7→ f (ξ), This paper will be primarily concerned with the investigation of similar ques- tions in a non-commutative context. More precisely, we replace uniform algebras by subspaces M of arbitrary C∗-algebras A such that A = C∗(M). Such subspaces are called operator spaces, and throughout the years the surrounding theory has developed into a powerful machine. In particular, it is now well-known that op- erator spaces and operator algebras can be defined in a purely abstract fashion, 2010 Mathematics Subject Classification. Primary 46L07; Secondary 47A20, 46L52. Key words and phrases. Operator space, completely bounded map, non-commutative Choquet boundary, C ∗-envelope. The first author was partially supported by an NSERC Discovery Grant. 1 2 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY independent of any ambient C∗-algebra (these are results of Ruan [38], Choi-Effros [13] and Blecher-Ruan-Sinclair [11] respectively). Despite the sophistication of this theory, it is often desirable to have access to the wealth of structure available for C∗-algebras, and thus the question arises of how to identify some sort of canonical smallest C∗-algebra containing the object of interest. Such a C∗-algebra would be the non-commutative analogue of the Shilov boundary of a uniform algebra. In a seminal paper, Arveson [4] initiated a program to construct the sought after C∗-algebra by developing a non-commutative version of the Choquet boundary. The central objects in his approach are the so-called boundary representations, certain unital completely positive linear maps having a unique extension property, much in the spirit of the defining property for points to lie in the classical Choquet boundary. Although Arveson was not able to fully realize his plan initially, the impact that his approach had is still very much felt to this day. The solution to the problem was eventually found by Hamana who constructed in [25] the C∗-envelope of an operator system using a different argument. Nevertheless, the objects introduced by Arveson were interesting in their own right, and spurred on significant results by Muhly-Solel [30] and Dritschel-McCullough [22]. Drawing from these contributions, Arveson himself later managed to fulfill his initial vision and construct the C∗-envelope of a unital operator space using boundary representations, at least in the separable case [7]. The separability assumption was later removed by Davidson-Kennedy in [18] (see also [27] for related work). There are still interesting unresolved issues regarding boundary representations, such as Arveson's hyperrigidity conjecture [8] which has witnessed recent progress [28],[17]. We mention here that the very foundation of Arveson's boundary representations approach is Stinespring's dilation theorem, which guarantees that unital completely positive maps on C∗-algebras can be dilated to unital ∗-homomorphisms. In partic- ular, the construction of the C∗-envelope of a general (possibly non-unital) operator space presents some difficulties, although there has been a meaningful theory devel- oped in that setting as well. Indeed, Hamana [26] (see also Blecher [9]) associates to any operator space its triple envelope, that is a ternary ring of operators that is the smallest such in the usual universal sense. This object can also be obtained by using a technique close in spirit to Arveson's non-commutative Choquet boundary, as was recently done by Fuller-Hartz-Lupini in [23]. A device known as Paulsen's "off-diagonal" technique and the associated generalization of Stinespring's dilation theorem [33] for completely contractive maps play a central role therein. Heuristically, Arveson's approach yields the existence of what one may call a unital completely positive non-commutative Choquet boundary. The corresponding adaptation of Fuller-Hartz-Lupini yields the existence of a completely contractive non-commutative Choquet boundary. However, the analogy with classical uniform algebra theory in the latter case is less accurate, as the resulting non-commutative Shilov boundary is not an algebra. We note that this apparent imperfection is somehow intrinsic due to the lack of a perfect analogue of the Stinespring dilation theorem. Accordingly, our goal in this paper is two-fold. First, we aim to develop a completely bounded counterpart to the aforementioned Choquet boundaries. Sec- ond, we wish to use this completely bounded non-commutative Choquet boundary to construct a non-commutative Shilov boundary that is still a C∗-algebra. To achieve these objectives, we consider operator spaces along with the extra data of a completely isometric representation on some Hilbert space. Even when we restrict A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 3 our attention to the unital setting, the completely bounded theory faces the usual obstacle related to the absence of a Stinespring dilation. We overcome this difficulty by focusing on a subclass of the completely bounded linear maps. This ultimately allows us to obtain the desired objects, although subtleties arise in our construction that are not present in the works cited above. We now outline the organization of the paper and state our main results. In Section 2 we collect various preliminary notions that we require throughout. In Section 3 we adapt the machinery of [22] and [18] to construct extremal elements in a very general framework. This tool is used in two different contexts later on. In Section 4, we show the following theorem (Theorem 4.7), which illustrates that the class of completely bounded linear maps is not appropriate for our purposes, and that the machinery developed in Section 3 cannot be used to produce a Shilov boundary that is an algebra by means of so-called CBr-extremal elements. Theorem 1.1. Let A ⊂ B(H) be an operator algebra. Let ω : A → B(Hω) be a completely bounded linear map. Assume that ω is CBr(A)-extremal for some r ≥ 1. Then, the following statements are equivalent. (i) The map ω is multiplicative. (ii) The map ω is completely contractive, and there is a ∗-homomorphism σ : (iii) There is a contractive completely positive map Ψ : C∗(A) → B(Hω) that C∗(A) → B(Hω) that agrees with ω on A. agrees with ω on A. This theorem motivates the introduction of a subclass of maps on which we focus in subsequent sections. Roughly speaking, for every r ≥ 1 this subclass P r consists of completely bounded linear maps that have a multiplicative dilation that is well- behaved. To make this precise, we introduce in Section 5 what we call Paulsen's similarity property, and provide a characterization of it (Theorem 5.4) in terms of unital extensions of homomorphisms. We formally introduce the class P r in Section 6, and establish many of its structural properties. We summarize some of them in the following (see Theorem 6.3 and Corollary 6.4). Theorem 1.2. Let M ⊂ B(H) be an operator space and let ϕ : M → B(Hϕ) be an element of P r(M), where r ≥ 1. Then, there is a Hilbert space Kϕ containing Hϕ, a ∗-homomorphism σ : C∗(M) → B(Kϕ) and an invertible operator X ∈ B(Kϕ) with such that kXk = kX−1k ≤ r1/2 ϕ(a) = PHϕ Xσ(a)X−1Hϕ , a ∈ M. In particular, there is a contractive completely positive linear map ψ : C∗(M) → B(Hϕ) with the property that kϕ − ψMkcb ≤ r − 1. Section 7 contains the main technical development of the paper, and is devoted to the study of the extremal elements of the class P r. This is where the benefits of working with this subclass are made manifest, and we obtain the following (see Corollary 7.4, Theorem 7.5 and Theorem 7.6). Theorem 1.3. Let M ⊂ B(H) be an operator space and let AM be the operator algebra that it generates. Then, the following statements hold. 4 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY (1) There is a P r(M)-extremal element ω such that for every n ∈ N and every a ∈ Mn(M) we have kak ≤ kω(n)(a)k. (2) Assume that ω is P r(M)-extremal. Then, ω has a unique P r-extension to AM, and that extension is a completely bounded homomorphism. Fur- thermore, there is an invertible operator X with kXk = kX−1k ≤ r1/2 such that the map defined as ωX (a) = Xω(a)X−1, a ∈ M has a unique contractive completely positive extension to C∗(M). That extension is a ∗-homomorphism. In Section 8, given an operator space M and a completely isometric linear map µ : M → B(Hµ), we define the C∗-envelope that we seek. For any r ≥ 1 we build a ∗-homomorphism εµ,r on C∗(µ(M)) using the extremal elements from the class P r. We then define the C∗r -envelope as C∗e,r(M, µ) = εµ,r(C∗(µ(M))). Although this construction is not independent of the representation µ in general, we show that it is invariant under appropriate isomorphisms, and that it has a universal property (see Theorem 8.4 and Corollary 8.5). Theorem 1.4. Let M,N be operator spaces and let µ : M → B(Hµ), ν : N → B(Hν ) be completely isometric linear maps. Let r ≥ 1 and let τ : µ(M) → ν(N ) be a P- isomorphism. Then, C∗e,r(M, µ) and C∗e,r(N , ν) are unitarily equivalent. Moreover, there is a surjective ∗-homomorphism ρ : C∗(ν(N )) → C∗e,r((M, µ)) such that ρ ◦ τ = εµ,r on µ(M). We elucidate the dependence of the C∗r -envelopes on the parameter r. We con- jecture that C∗e,r(M, µ) = C∗e,1(M, µ) for every r ≥ 1, but are unable to prove it in general. Nevertheless, we provide supporting evidence for the conjecture in some important cases. First, we obtain the following variation on Arveson's boundary theorem [5] (Theorem 8.11). Theorem 1.5. Let M be an operator space and let µ : M → B(Hµ) be a completely isometric linear map such that C∗(µ(M)) contains the ideal K of compact operators on Hµ. Let r ≥ 1 and assume that there is n ∈ N and a ∈ Mn(M) such that kµ(a) + Kk < r−1kak. Then, the C∗-algebras C∗e,r(M, µ) and C∗(µ(M)) are ∗- isomorphic. Finally, we make a connection with the original motivational example of uniform algebras (Theorem 8.12). Theorem 1.6. Let A be a uniform algebra on a compact metrizable space X. Let ΣA ⊂ X denote the Shilov boundary of A. Let α : A → B(Hα) be a completely isometric algebra homomorphism. Then, for each r ≥ 1 the C∗-algebras C∗e,r(A, α) and C(ΣA) are ∗-isomorphic. 2. Preliminaries 2.1. Operator spaces and completely bounded maps. We start by recalling, very briefly, terminology and basic results from operator space theory. A good reference on the subject is [31], which the reader can consult for more details. A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 5 Let H be a Hilbert space and let B(H) denote the C∗-algebra of bounded linear operators on H. A subspace M ⊂ B(H) is called an operator space. If M is self- adjoint and contains the identity, then it is called an operator system. If M is a subalgebra of B(H), we say that it is an operator algebra. As mentioned in the introduction, these three concepts can be defined abstractly, but for our purposes the previous definitions will suffice. Let n ∈ N and let Mn(M) denote the space of n×n matrices with entries from M. If we put H(n) = H⊕ . . .⊕H, then we may view Mn(M) as a subspace of B(H(n)). The norm on Mn(M) is that induced by this identification. If ϕ : M → B(Hϕ) is a linear map, then it induces another linear map ϕ(n) : Mn(M) → B(H(n) ϕ ) defined as ϕ(n)([aij ]i,j) = [ϕ(aij )]i,j , [aij]i,j ∈ Mn(M). If M is an operator algebra and ϕ is a homomorphism, then so is ϕ(n). The com- pletely bounded norm of ϕ is defined as kϕkcb = sup n∈Nkϕ(n)k. The map ϕ is said to be completely bounded if kϕkcb is finite, and completely con- tractive if kϕkcb ≤ 1. If ϕ(n) is isometric for every n ∈ N, then ϕ is said to be If ϕ(n) is positive for every n ∈ N, then ϕ is said to be completely isometric. It is well-known that if M is an operator system and ϕ is completely positive. completely positive, then kϕkcb = kϕ(1)k. Wittstock's extension theorem says that given an operator space M ⊂ B(H) and a completely bounded linear map ϕ : M → B(Hϕ), there exists a linear map Φ : B(H) → B(Hϕ) that agrees with ϕ on M and such that kΦkcb = kϕkcb. If M is an operator system and ϕ is completely positive, then by Arveson's extension theorem Φ can be chosen to be completely positive as well. Finally, we mention a very important observation of Arveson that will be used frequently without mention. If M is a unital operator space, then M + M∗ is an operator system, and any unital completely contractive map on M extends uniquely to a unital completely positive map on M + M∗. 2.2. Unitizations of operator spaces. We typically do not assume that operator spaces contain the identity element of the ambient B(H). Even in the event that they do contain it, we typically do not assume that the maps we consider preserve the identity. Accordingly, there is a certain standard unitization procedure that we will have the occasion to use several times throughout. It associates to an operator space M a unital operator space Υ(M), and to a linear map ϕ on M a unital linear map Υ(ϕ) on Υ(M). The notation established here will be used tacitly throughout the paper. More precisely, let M ⊂ B(H) be an operator space and let A = C∗(M) be the C∗-algebra that it generates. Then, there exist a unital C∗-algebra Υ(A), a unital subspace Υ(M) ⊂ Υ(A) and an injective ∗-homomorphism υ : A → Υ(A) with the property that Υ(A) = υ(A) + Ce and Υ(M) = υ(M) + Ce where e is the unit of Υ(A). To show this, we distinguish two cases. If A is not unital, we put Υ(A) = A + CIH, Υ(M) = M + CIH 6 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY and we simply let υ : A → Υ(A) be the inclusion. If A is unital, we let and we define the injective ∗-homomorphism υ : A → Υ(A) as Υ(A) = A ⊕ C, Υ(M) = M ⊕ C υ(a) = a ⊕ 0, a ∈ A. This establishes the claim. Next, we note that the unitization can also be performed on maps. If ϕ : M → B(Hϕ) is a linear map, then there is a unique unital linear map Υ(ϕ) : Υ(M) → B(Hϕ) that satisfies (Υ(ϕ) ◦ υ)(m) = ϕ(m), m ∈ M. In the following two particular cases, more can be said about the map Υ(ϕ). • Let ϕ be a contractive linear map on M that extends to a contractive com- pletely positive linear map on A. Since υ−1 is contractive and completely positive on υ(A), we may use [12, Proposition 2.2.1] to see that Υ(ϕ) is a unital map on Υ(M) that extends to a completely positive map on Υ(A). • Let M be an operator algebra and let ϕ be a completely contractive ho- momorphism. It is easily verified that Υ(ϕ) is a unital homomorphism, and moreover it is completely contractive by [29, Corollary 3.3] (see also [10, Theorem 2.1.13]). Finally, we note that given a linear map Φ : Υ(M) → B(H), there is a linear map ϕ : M → B(H) such that Υ(ϕ) = Φ if and only if Φ is unital. Moreover, Φ is unital and completely contractive on Υ(M) if and only if Φ = Υ(ϕ) for some linear map ϕ on M that extends to a contractive completely positive linear map on A. 2.3. Ultraproducts. Several of our arguments in the sequel will require the ma- chinery of ultraproducts. The following material is folklore, but we collect it here for the convenience of the reader. Kλ be a Hilbert space containing a fixed Hilbert space H. We denote byQλ∈Λ Kλ Let Λ be a directed set and let U be a cofinal ultrafilter on Λ. For each λ ∈ Λ, let the Banach space of all bounded nets (ξλ)λ∈Λ. The ultraproduct Hilbert space KU of (Kλ)λ∈Λ along U is defined as inner product on KU is given as follows: Given ξ = (ξλ)λ∈Λ ∈Qλ∈Λ Kλ, we denote its canonical image in KU by [ξ]. The for ξ = (ξλ)λ∈Λ ∈ Qλ∈Λ Kλ and η = (ηλ)λ∈Λ ∈Qλ∈Λ Kλ we have U hξλ, ηλiKλ. h[ξ], [η]iKU = lim In particular, we see that the linear map V : H → KU defined as is an isometry. V h = [(h)λ∈Λ], h ∈ H where Kλ! /ZU KU = Yλ∈Λ ZU =((ξλ)λ∈Λ ∈ Yλ∈Λ U kξλk = 0) . Kλ : lim A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 7 For each λ ∈ Λ, let Tλ : Kλ → Kλ be a bounded linear operator. Assume that sup λ∈Λ kTλk < ∞. Then, the ultraproduct operator TU : KU → KU of (Tλ)λ∈Λ along U is the bounded linear operator defined as TU [(ξλ)λ∈Λ] = [(Tλξλ)λ∈Λ], (ξλ)λ∈Λ ∈ Yλ∈Λ Kλ. It is readily verified that kTUk ≤ limU (kTλk)λ∈Λ. completely bounded linear map. Assume that Let M be an operator space. For each λ ∈ Λ, let ϕλ : M → B(Kλ) be a Then, the map defined as is linear with lim U sup λ∈Λkϕλkcb < ∞. lim U (ϕλ)λ∈Λ : M → B(KU ) (ϕλ)λ∈Λ(a) = lim U (ϕλ(a))λ∈Λ, a ∈ M (cid:13)(cid:13)(cid:13)lim (ϕλ)λ∈Λ(cid:13)(cid:13)(cid:13)cb ≤ lim U U (kϕλkcb)λ∈Λ. 3. Dilation orders and maximal maps If M is an operator algebra and each ϕλ is multiplicative, then so is limU (ϕλ)λ∈Λ. Let M be an operator space and let r ≥ 0. For each Hilbert space H we denote by CBr(M,H) the set of linear maps ϕ : M → B(H) with kϕkcb ≤ r. Assume that for every Hilbert space H, we are given a subset Cr(M,H) of CBr(M,H). We will say that Cr(M) is a subclass of CBr(M) to describe such a situation. A partial order ≺ defined on the subclass Cr(M) is said to be a dilation order if whenever ϕ ∈ Cr(M,Hϕ), ψ ∈ Cr(M,Hψ) satisfy ϕ ≺ ψ, then we must have that Hϕ ⊂ Hψ and kϕ(a)ξk ≤ kψ(a)ξk, kϕ(a)∗ξk ≤ kψ(a)∗ξk for every a ∈ M, ξ ∈ Hϕ. We say that the subclass Cr(M) has the limit property with respect to ≺ if given a totally ordered set Λ and a net ϕλ ∈ Cr(M,Hλ), λ ∈ Λ with the property that ϕλ ≺ ϕµ if λ ≤ µ, then we can find an element ψ ∈ Cr(M,∪λHλ) satisfying ϕλ ≺ ψ for every λ ∈ Λ. Moreover, we say that an element ω ∈ Cr(M,Hω) is maximal if whenever δ ∈ Cr(M,Hδ) satisfies ω ≺ δ we must have that kω(a)ξk = kδ(a)ξk, kω(a)∗ξk = kδ(a)∗ξk for every a ∈ M, ξ ∈ Hω. We now describe a general procedure to construct maximal maps [1], which we will use subsequently in two different situations. It is standard in the context of completely contractive maps on operator spaces [23], or that of unital completely positive maps on operator systems [18, 22], but because our context is different we provide the straightforward adaptations of the usual proofs. First, we show that maximality can be achieved "locally". 8 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY Lemma 3.1. Let M be an operator space, let r ≥ 0 and let Cr(M) be a subclass of CBr(M). Assume that Cr(M) has the limit property with respect to the dilation order ≺. Let ϕ ∈ Cr(M,Hϕ). Then, for each a ∈ M and ξ ∈ Hϕ, there is ω ∈ Cr(M,Hω) such that ϕ ≺ ω with the property that if δ ∈ Cr(M,Hδ) and ω ≺ δ then kω(a)ξk = kδ(a)ξk, kω(a)∗ξk = kδ(a)∗ξk. Proof. Let ψ0 = ϕ. Then, there is ψ1 ∈ Cr(M,H1) such that ψ0 ≺ ψ1 and kψ1(a)ξk ≥ sup{kψ(a)ξk : ψ0 ≺ ψ} − 1. Next, choose ψ2 ∈ Cr(M,H2) such that ψ1 ≺ ψ2 and kψ2(a)∗ξk ≥ sup{kψ(a)∗ξk : ψ1 ≺ ψ} − 1/2. Arguing by induction, for each n ≥ 1 we find ψn ∈ Cr(M,Hn) such that ψn−1 ≺ ψn, kψ2n−1(a)ξk ≥ sup{kψ(a)ξk : ψ2n−2 ≺ ψ} − 1/(2n − 1) and kψ2n(a)∗ξk ≥ sup{kψ(a)∗ξk : ψ2n−1 ≺ ψ} − 1/2n. Since Cr(M) has the limit property with respect to ≺, we find ω ∈ Cr(M,Hω) such that ψn ≺ ω for every n ≥ 0. In particular, we have ϕ ≺ ω. Finally, given δ ∈ Cr(M,Hδ) such that ω ≺ δ we have ψn ≺ δ for every n ≥ 1 whence and thus kψ2n−1(a)ξk ≥ kδ(a)ξk − 1/(2n − 1), kψ2n(a)∗ξk ≥ kδ(a)∗ξk − 1/2n kω(a)ξk ≥ kψ2n−1(a)ξk ≥ kδ(a)ξk − 1/(2n − 1), kω(a)∗ξk ≥ kψ2n(a)∗ξk ≥ kδ(a)∗ξk − 1/2n for every n ≥ 1. Hence kω(a)ξk ≥ kδ(a)ξk, kω(a)∗ξk ≥ kδ(a)∗ξk. But the reverse inequalities are always satisfied since ≺ is a dilation order, and we have as desired. kω(a)ξk = kδ(a)ξk, kω(a)∗ξk = kδ(a)∗ξk (cid:3) Using the previous lemma, a standard induction argument yields the existence of maximal elements. Theorem 3.2. Let M be an operator space, let r ≥ 0 and let Cr(M) be a subclass of CBr(M). Assume that Cr(M) has the limit property with respect to the dilation order ≺. Then, for each ϕ ∈ Cr(M,Hϕ) there is a maximal element ω ∈ Cr(M,Hω) such that ϕ ≺ ω. Proof. Let γ0 be an ordinal with the property that there is an enumeration {xα : α < γ0} of M × Hϕ. Using transfinite recursion, we construct a net ϕα ∈ Cr(M,Hα), α ≤ γ0 such that: • ϕ ≺ ϕα ≺ ϕβ if α < β, and A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 9 • for every α we have that if δ ∈ Cr(M,Hδ) satisfies ϕα ≺ δ, then kϕα(a)ξk = kδ(a)ξk, kϕα(a)∗ξk = kδ(a)∗ξk for (a, ξ) ∈ {xβ : β < α}. Put ϕ0 = ϕ. Let α be an ordinal and assume that we have constructed {ϕβ}β<α. We now show how to find ϕα. There are two cases to consider according to whether α is a successor or a limit ordinal. If α is a successor, then we let ϕα ∈ Cr(M,Hα) be the element obtained from applying Lemma 3.1 to ϕα−1 and to the pair (a, ξ) = xα−1. It is readily verified that this has the required properties. Alternatively, if α is a limit ordinal, then since Cr(M) has the limit property we find ϕα ∈ Cr(M,Hα) such that ϕβ ≺ ϕα for every ordinal β < α. We claim that ϕα has the desired property. In other words, we claim that if δ ∈ Cr(M,Hδ) and ϕα ≺ δ then kϕα(a)ξk = kδ(a)ξk, kϕα(a)∗ξk = kδ(a)∗ξk for (a, ξ) ∈ {xβ : β < α}. Fixing β < α we proceed to show that kϕα(a)∗ξk = kδ(a)∗ξk kϕα(a)ξk = kδ(a)ξk, for (a, ξ) = xβ. Since α is a limit ordinal, there is an ordinal β′ such that β < β′ < α. In particular, we see that ϕβ ≺ ϕβ ′ ≺ ϕα and ϕβ ′ ≺ δ so that kϕβ ′(a)ξk = kδ(a)ξk, kϕβ ′(a)∗ξk = kδ(a)∗ξk by the recursive assumption on ϕβ ′ . This forces kϕα(a)ξk = kδ(a)ξk, kϕα(a)∗ξk = kδ(a)ξk and establishes the existence of the collection {ϕα}α≤γ0. Next, we put θ1 = ϕγ0 and X1 = Hγ0. By choice of γ0, we see that kθ1(a)ξk = kδ(a)ξk, kθ1(a)∗ξk = kδ(a)∗ξk for every a ∈ M, ξ ∈ Hϕ and every δ ∈ Cr(M,Hδ) such that θ1 ≺ δ. By repeating the argument of the previous paragraphs and proceeding by (usual) induction, we obtain a sequence of maps such that and θn ∈ Cr(M,Xn), n ≥ 1 ϕ ≺ θn ≺ θn+1 kθn(a)ξk = kδ(a)ξk, kθn(a)∗ξk = kδ(a)∗ξk, a ∈ M, ξ ∈ Xn−1 whenever δ ∈ Cr(M,Hδ) satisfies θn ≺ δ. Let X = ∪nXn. Since Cr(M) has the limit property, we obtain a map ω ∈ Cr(M,X ) such that ϕ ≺ θn ≺ ω, n ≥ 1. Thus, we see that kω(a)ξk = kθn+1(a)ξk = kδ(a)ξk, kω(a)∗ξk = kθn+1(a)∗ξk = kδ(a)∗ξk for every n ∈ N, a ∈ M and ξ ∈ Xn, whenever δ ∈ Cr(M,Hδ) with ω ≺ δ. But we have that ∪nXn is dense in X , so we infer kω(a)ξk = kδ(a)ξk, kω(a)∗ξk = kδ(a)∗ξk 10 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY for every a ∈ M, ξ ∈ X and every δ ∈ Cr(M,Hδ) such that ω ≺ δ. Thus, ω is maximal. (cid:3) 4. Extremals in the class of completely bounded maps Let M be an operator space and let r ≥ 0. In this section, we analyze the natural dilation order on the full class CBr(M). That is, given ϕ ∈ CBr(M,Hϕ) and ψ ∈ CBr(M,Hψ), we write ϕ ≺ ψ if Hϕ ⊂ Hψ and a ∈ M. ϕ(a) = PHϕ ψ(a)Hϕ , It is clear that this defines a dilation order on CBr(M). We say that ω ∈ CBr(M,Hω) is CBr(M)-extremal if whenever δ ∈ CBr(M,Hδ) satisfies ω ≺ δ, then we neces- sarily have that Hω is reducing for δ(M). The technical tool we need to establish the existence of CBr(M)-extremals is the following standard fact. Lemma 4.1. Let M be an operator space. Then, the class CBr(M) has the limit property with respect to the dilation order ≺. Proof. Let Λ be a totally ordered set. For each λ ∈ Λ, let ϕλ ∈ CBr(M,Hλ). Assume that ϕλ ≺ ϕµ whenever µ ≥ λ. Let K = ∪λ∈ΛHλ. For a ∈ M we define a linear operator as follows. Given x ∈ ∪λHλ and y ∈ ∪λHλ, since the spaces Hλ increase with λ we may find an index λ0 ∈ Λ such that both x and y lie in Hλ0 . We then put ψ(a) : ∪λ∈ΛHλ → ∪λ∈ΛHλ hψ(a)x, yi = hϕλ0 (a)x, yi. We claim that ψ(a) is well-defined. Indeed, assume that x and y both lie in Hλ∩Hµ for some λ ≤ µ. Then, using that ϕλ ≺ ϕµ we find which shows that ψ(a) is well-defined. Moreover, it is clear that hϕλ(a)x, yi = hϕµ(a)x, yi kψ(a)k ≤ sup λ kϕλ(a)k ≤ rkak so that we may extend ψ(a) to a bounded linear operator on K with norm at most rkak. We thus obtain a bounded linear map ψ : M → B(K), that is easily seen to satisfy kψkcb ≤ r. By construction, we have that ϕλ(a) = PHλψ(a)Hλ , a ∈ M (cid:3) so that ϕλ ≺ ψ. We then obtain the following useful consequence. Theorem 4.2. Let M be an operator space, let r ≥ 0 and let ϕ ∈ CBr(M,Hϕ). Then, there exists a CBr(M)-extremal ω : M → B(Hω) with ϕ ≺ ω. Proof. The class CBr(M) has the limit property with respect to ≺ in view of Lemma 4.1. By Theorem 3.2, we see that there is a maximal element ω ∈ CBr(M,Hω) such that ϕ ≺ ω. The proof is now completed by noting that such an element must be CBr(M)-extremal. Indeed, let δ ∈ CBr(M,Hδ) such that ω ≺ δ. We calculate for every a ∈ M and ξ ∈ Hω that k(δ(a) − ω(a))ξk2 = kδ(a)ξk2 + kω(a)ξk2 − 2 Rehδ(a)ξ, ω(a)ξi = kδ(a)ξk2 − kω(a)ξk2 = 0 A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 11 so that δ(a)ξ = ω(a)ξ. Likewise, k(δ(a)∗ − ω(a)∗)ξk2 = kδ(a)∗ξk2 + kω∗(a)ξk2 − 2 Rehδ(a)∗ξ, ω(a)∗ξi = kδ(a)∗ξk2 − kω(a)∗ξk2 = 0 so that δ(a)∗ξ = ω(a)∗ξ. Hence, δ(M)Hω ⊂ Hω and δ(M)∗Hω ⊂ Hω, so indeed ω is CBr(M)-extremal. (cid:3) Before we can state and prove the main result of this section, we require some preparation. The following observation is elementary but we will need it repeatedly throughout the paper. As such, we recall the proof here for the reader's convenience. Lemma 4.3. Let M be an operator space and let ϕ : M → B(Hϕ) be a linear map. Assume that there is a Hilbert space K, two isometries and a linear map ψ : M → B(K) such that V1 : Hϕ → K, V2 : Hϕ → K ϕ(a) = V ∗1 ψ(a)V2, a ∈ M. Then, there is a Hilbert space Kϕ containing Hϕ and two unitary operators R1 : Kϕ → K ⊕ K, R2 : Kϕ → K ⊕ K such that When V1 = V2 we may choose R1 = R2. ϕ(a) = PHϕR∗1(ψ(a) ⊕ ψ(a))R2Hϕ, a ∈ M. Proof. For k = 1, 2, define an isometry as Note that V ′k : Hϕ → K ⊕ K V ′kξ = Vkξ ⊕ 0, ξ ∈ Hψ. Define also a unitary operator ϕ(a) = V ′∗1 (ψ(a) ⊕ ψ(a))V ′2 , a ∈ M. Uk : K ⊕ K = VkHϕ ⊕ (VkHϕ)⊥ ⊕ K → Hϕ ⊕ (VkHϕ)⊥ ⊕ K as Uk(Vkξ ⊕ z ⊕ η) = ξ ⊕ z ⊕ η, It is readily verified that ξ ∈ Hϕ, z ∈ (VkHϕ)⊥, η ∈ K. UkV ′k : Hϕ → Hϕ ⊕ (VkHϕ)⊥ ⊕ K is simply the inclusion in the first component, so that (UkV ′k)∗ is the projection PHϕ onto the first component. For every a ∈ M we obtain that ϕ(a) = V ′∗1 (ψ(a) ⊕ ψ(a))V ′2 = (U1V ′1 )∗U1(ψ(a) ⊕ ψ(a))U∗2 (U2V ′2 ) = PHϕU1(ψ(a) ⊕ ψ(a))U∗2 Hϕ. Note that (V1Hϕ)⊥ ⊕ K and (V2Hϕ)⊥ ⊕ K have the same dimension. Choose a unitary operator Z : (V1Hϕ)⊥ ⊕ K → (V2Hϕ)⊥ ⊕ K. 12 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY Then, the operator W = I ⊕ Z : Hϕ ⊕ (V1Hϕ)⊥ ⊕ K → Hϕ ⊕ (V2Hϕ)⊥ ⊕ K is unitary as well, and satisfies PHϕ = PHϕ W . Thus, we have that a ∈ M. ϕ(a) = PHϕW U1(ψ(a) ⊕ ψ(a))U∗2Hϕ , The desired equality now follows upon setting Kϕ = Hϕ ⊕ (V2Hϕ)⊥ ⊕ K, R∗1 = W U1, R2 = U∗2 . Finally, we note that a careful look at the proof reveals that indeed R1 and R2 may be chosen to be equal whenever V1 = V2. (cid:3) Another piece of preparation we need for the main result of this section is the following generalization of Stinespring's dilation theorem. Theorem 4.4. Let M ⊂ B(H) be an operator space and ϕ : M → B(Hϕ) a com- pletely bounded linear map. Then, there is a Hilbert space K, a ∗-homomorphism π : C∗(M) → B(K), another Hilbert space Kϕ containing Hϕ and two unitary operators with the property that R1 : Kϕ → K, R2 : Kϕ → K ϕ(a) = kϕkcbPHϕ R∗1π(a)R2Hϕ, a ∈ M. Proof. The claim is trivial if ϕ = 0, so that upon replacing ϕ with ϕ/kϕkcb, we may assume that ϕ is completely contractive. By [33, Theorems 2.4 and 2.7], we may find a Hilbert space K, two isometries V2 : Hϕ → K and a ∗-homomorphism π : C∗(M) → B(K) such that a ∈ M. ϕ(a) = V ∗1 π(a)V2, V1 : Hϕ → K, The desired conclusion now follows at once from Lemma 4.3. (cid:3) We also require the following elementary non-unital adaptation of the usual Stinespring dilation theorem. Lemma 4.5. Let A be a C∗-algebra and let ϕ : A → B(H) be a contractive completely positive map. Then, there is a Hilbert space K containing H and ∗- homomorphism σ : A → B(K) such that ϕ(a) = PHσ(a)H, a ∈ A. Proof. Consider the unital C∗-algebra Υ(A) and the associated unital completely positive map Υ(ϕ) : Υ(A) → B(H) (see Subsection 2.2). Apply the usual Stine- spring dilation theorem to Υ(ϕ) to find a Hilbert space K containing H and unital ∗-homomorphism π : Υ(A) → B(K) such that Υ(ϕ)(b) = PHπ(b)H, b ∈ Υ(A). Then, the map σ = π ◦ υ : A → B(K) is a ∗-homomorphism such that ϕ(a) = PHσ(a)H, a ∈ A. The last preliminary we need shows that CBr-extremal elements are necessarily non-degenerate. (cid:3) A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 13 Lemma 4.6. Let M be an operator space, let r ≥ 1 and let ω : M → B(Hω) be CBr(M)-extremal. Then, ω(M)Hω = Hω. Proof. Let H′ω = ω(M)Hω. According to the orthogonal decomposition Hω = H′ω ⊕ H′⊥ω , for every a ∈ M we have ω(a) =(cid:20)∗ 0 ∗ 0(cid:21) . If H′⊥ω 6= {0}, then we may choose ϕ : M → B(H′⊥ω ) with kϕkcb = r. Let Φ : M → B(Hω) be defined as Φ(a) =(cid:20)0 0 ϕ(a)(cid:21) , 0 Next, define δ : M → B(Hω ⊕ Hω) as δ(a) =(cid:20)ω(a) Φ(a) 0 (cid:21) , 0 a ∈ M. a ∈ M. A straightforward verification yields that kδkcb = r. Moreover, if we identify Hω with Hω ⊕ {0} ⊂ Hω ⊕ Hω, then we note that ω ≺ δ, yet Hω is not reducing for δ(M) and thus ω is not CBr-extremal. (cid:3) Finally, we come to the main result of this section that elucidates the structure of CBr(M)-extremal elements, at least for operator algebras. Theorem 4.7. Let A ⊂ B(H) be an operator algebra. Let ω : A → B(Hω) be a completely bounded linear map. Assume that ω is CBr(A)-extremal for some r ≥ 1. Then, the following statements are equivalent. (i) The map ω is multiplicative. (ii) The map ω is completely contractive, and there is a ∗-homomorphism σ : (iii) There is a contractive completely positive map Ψ : C∗(A) → B(Hω) that C∗(A) → B(Hω) that agrees with ω on A. agrees with ω on A. Proof. (i) ⇒ (ii) Assume that ω is multiplicative. It is no loss of generality to assume that kωkcb 6= 0. By Theorem 4.4, there is a Hilbert space K, a ∗-homomorphism π : C∗(A) → B(K), another Hilbert space Kω containing Hω and two unitary operators with the property that R1 : Kω → K, R2 : Kω → K ω(a) = kωkcbPHω R∗1π(a)R2Hω , a ∈ A. Using that ω is CBr(A)-extremal we see that R∗1π(A)R2Hω ⊂ Hω. Consequently, we obtain ω(a) = kωkcbR∗1π(a)R2Hω , a ∈ A. 14 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY Now, we calculate for a ∈ A and b ∈ A that cbR∗1π(a)R2R∗1π(b)R2Hω − kωkcbR∗1π(ab)R2Hω ) 0 = R1(ω(a)ω(b) − ω(ab)) = R1(kωk2 = kωkcbπ(a)(kωkcbR2R∗1 − I)π(b)R2Hω = kωkcbπ(a)(kωkcbR2 − R1)R∗1π(b)R2Hω = π(a)(kωkcbR2 − R1)ω(b). By Lemma 4.6, we see that Hω = ω(A)Hω so we find (kωkcbR2 − R1)Hω ⊂ ∩a∈A ker π(a). Hence, we have ω(a) = kωkcbR∗1π(a)R2Hω = R∗1π(a)R1Hω , a ∈ A. In particular, we see that ω is completely contractive. Using again that ω is CBr(A)- extremal we see that Hω is reducing for R∗1π(A)R1 so that R1Hω is invariant for π(C∗(A)). Thus, the map σ : C∗(A) → B(Hω) defined as σ(a) = PHω R∗1π(a)R1Hω , a ∈ C∗(A) is a ∗-homomorphism that agrees with ω on A. This establishes (ii). The implication (ii) ⇒ (iii) is trivial. (iii) ⇒ (i) Assume that there is a contractive completely positive map on Ψ : C∗(A) → B(Hω) that agrees with ω on A. By Lemma 4.5, there is a Hilbert space K containing Hω and a ∗-homomorphism π : C∗(A) → B(K) such that Ψ(a) = PHω π(a)Hω , a ∈ C∗(A). In particular, we have that ω(a) = PHω π(a)Hω , a ∈ A. Since ω is CBr(A)-extremal, we conclude that π(A)Hω ⊂ Hω, and thus ω is mul- tiplicative. (cid:3) Guided by the approach of [4], [22] and [18], one expects the CBr(M)-extremal elements to be the natural analogues of points in the Choquet boundary. The previous theorem shows that the class of completely bounded linear maps is too large to use the associated non-commutative Choquet boundary to produce a non- commutative Shilov boundary that is an algebra. Indeed, if r > 1, then by Theorem 4.2 there are CBr(A)-extremals ω such that kωkcb > 1, and those are not multi- plicative by virtue of Theorem 4.7. If one is willing to settle for weaker algebraic properties of the Shilov boundary, then some meaningful results can be obtained in the completely contractive setting in full generality, as was done in [26], [9] and [23]. This leads to the notion of triple envelope of an operator space. As mentioned in the introduction, we take a different path: we restrict the class of bounded linear maps under consideration in order to obtain extremals that automatically admit a multiplicative extension. and and kθkcb ≤(cid:16)kϕk1/2 cb + (kϕkcb + 1)1/2(cid:17)2 ϕ(a) = PHϕ θ(a)Hϕ , a ∈ M. kXkkX−1k ≤(cid:16)kϕk1/2 cb + (kϕkcb + 1)1/2(cid:17)2 ϕ(a) = V ∗Xπ(a)X−1V, a ∈ M. A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 15 5. Paulsen's similarity property Before we can define the subclass of completely bounded linear maps we are interested in, we make a detour to carefully examine the class of completely bounded homomorphisms. The foundation of our investigation is the following. Theorem 5.1. Let M ⊂ B(H) be an operator space and ϕ : M → B(Hϕ) a completely bounded linear map. Then, there is a Hilbert space Kϕ containing Hϕ and a completely bounded homomorphism θ : C∗(M) → B(Kϕ) with the property that Proof. By [33, Theorems 2.4 and 2.8], there is a Hilbert space K, an isometry V : Hϕ → K, a ∗-homomorphism π : C∗(M) → B(K) and an invertible operator X ∈ B(K) such that By Lemma 4.3, there is a Hilbert space Kϕ containing Hϕ and a unitary operator U : K → Kϕ such that if we let θ : C∗(M) → B(Kϕ) be defined as θ(a) = U Xπ(a)X−1U∗, a ∈ C∗(M) then we find Finally, note that ϕ(a) = PHϕ θ(a)Hϕ , a ∈ M. kθkcb ≤ kXkkX−1k ≤(cid:16)kϕk1/2 cb + (kϕkcb + 1)1/2(cid:17)2 . (cid:3) We also require a celebrated result of Paulsen (see [33, Theorem 3.1] and [32]). Theorem 5.2. Let A be an operator algebra and θ : A → B(H) a completely bounded homomorphism. Then, there is an invertible operator X ∈ B(H) with kXk = kX−1k ≤ kθk1/2 cb + (kθkcb + 1)1/2 such that the map is completely contractive. When θ is unital, it can be arranged that a 7→ Xθ(a)X−1, a ∈ A kXk = kX−1k = kθk1/2 cb . Inspired by Theorem 5.2, we make the following definition. Let A be an operator algebra, let r ≥ 1 and let θ : A → B(H) be a completely bounded homomorphism. Then, we say that θ has Paulsen's similarity property with constant r if there is an invertible operator X ∈ B(H) with kXk = kX−1k ≤ r 16 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY such that the map a 7→ Xθ(a)X−1, a ∈ A is completely contractive. We now single out a refinement of Theorem 5.2. Corollary 5.3. Let A ⊂ B(H) be an operator algebra, let r ≥ 1 and let θ : A → B(Hθ) be a completely bounded homomorphism that has Paulsen's similarity property with constant r. Then, there is a Hilbert space Kθ containing Hθ, a ∗- homomorphism σ : C∗(A) → B(Kθ) and an invertible operator X ∈ B(Hθ) with such that kXk = kX−1k ≤ r Xθ(a)X−1 = PHθ σ(a)Hθ , a ∈ A. Proof. By assumption, there is an invertible operator X ∈ B(Hθ) with kXk = kX−1k ≤ r such that the map θX : A → B(Hθ) defined as θX (a) = Xθ(a)X−1, a ∈ A is a completely contractive homomorphism. We first claim that θX extends to a contractive completely positive map on C∗(A). adjoint projection in θX (A)′. Hence, If A is unital, then P = θX (IH) is a contractive idempotent, and thus a self- and the unital completely contractive map θX (a) = P θX (a)P ⊕ 0, a ∈ A a 7→ P θX (a)P ∈ B(PHθ), a ∈ A extends to a contractive completely positive map on C∗(A), so the same is true of θX . In the non-unital case, we saw in Subsection 2.2 that there is unital completely contractive homomorphism on A+CIH extending θX . We conclude that θX extends to a unital completely positive map on C∗(A + CIH). The claim is established. By Lemma 4.5, there is a Hilbert space Kθ containing Hθ and a ∗-homomorphism σ : C∗(A) → B(Kθ) such that Xθ(a)X−1 = PHθ σ(a)Hθ , a ∈ A as desired. (cid:3) We examine Paulsen's similarity property more closely. First observe that com- pletely contractive homomorphisms trivially have Paulsen's similarity property with constant 1. Moreover, Theorem 5.2 shows that any completely bounded ho- momorphism θ has Paulsen's similarity property with constant r provided that r ≥ kθk1/2 cb + (kθkcb + 1)1/2. In the other direction, if θ has Paulsen's similarity property with constant r, then clearly r ≥ kθk1/2 cb . Any unital completely bounded homomorphism θ has Paulsen's similarity property with the sharp constant kθk1/2 cb , by Theorem 5.2 again. The next result is a characterization of Paulsen's similarity property in terms of the existence of unital completely bounded extensions. It uses the unitization procedure described in Subsection 2.2. Theorem 5.4. Let A be an operator algebra, let r ≥ 1 and let θ : A → B(H) be a homomorphism. Then, the following statements are equivalent. A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 17 (i) The map θ has Paulsen's similarity property with constant r. (ii) The unital homomorphism Υ(θ) : Υ(A) → B(H) satisfies kΥ(θ)kcb ≤ r2. Proof. (i) ⇒ (ii) Assume that θ has Paulsen's similarity property with constant r, so there is an invertible operator X ∈ B(H) with kXk = kX−1k ≤ r such that the map θX : A → B(H) defined as a 7→ Xθ(a)X−1, a ∈ A so that is completely contractive. Thus, the unital homomorphism Υ(θX ) is completely contractive as seen in Subsection 2.2. It is readily verified that b ∈ Υ(A) kΥ(θ)kcb ≤ kXkkX−1kkΥ(θX)kcb ≤ r2. Υ(θX )(b) = X(Υ(θ)(b))X−1, (ii) ⇒ (i) Conversely, assume that the unital homomorphism Υ(θ) : Υ(A) → B(H) satisfies kΥ(θ)kcb ≤ r2. By Theorem 5.2, there is an invertible operator X ∈ B(H) with such that the map kXk = kX−1k = kΥ(θ)k1/2 b 7→ XΥ(θ)(b)X−1, b ∈ Υ(A) cb ≤ r is completely contractive. In particular, we see that the map a 7→ XΥ(θ)(υ(a))X−1 = Xθ(a)X−1, a ∈ A is completely contractive, since υ : A → Υ(A) is completely isometric. We conclude that θ has Paulsen's similarity property with constant r. (cid:3) A useful consequence is the following, that describes how Paulsen's similarity property behaves with respect to direct sums. Corollary 5.5. Let A be an operator algebra and let θ1 : A → B(H1), θ2 : A → B(H2) be completely bounded homomorphisms. Let θ = θ1 ⊕ θ2 : A → B(H1 ⊕H2). Then, the following statements hold. (1) If θ1 and θ2 have Paulsen's similarity property with constants r1 and r2 re- spectively, then θ has Paulsen's similarity property with constant max{r1, r2}. (2) If θ has Paulsen's similarity property with constant r, then θ1 and θ2 both have Paulsen's similarity property with constant r. Proof. (1) Assume that θ1 and θ2 have Paulsen's similarity property with con- stants r1 and r2 respectively. Accordingly, we can find invertible operators X1 ∈ B(H1), X2 ∈ B(H2) with kX1k = kX−1 1 k ≤ r1, kX2k = kX−1 2 k ≤ r2 such that the maps and a 7→ X1θ1(a)X−1 1 , a 7→ X2θ2(a)X−1 2 , a ∈ A a ∈ A are completely contractive. If we put X = X1 ⊕ X2, then kXk = kX−1k ≤ max{r1, r2}, 18 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY and the map a 7→ X(θ1(a) ⊕ θ2(a))X−1, a ∈ A is completely contractive, so that θ1 ⊕ θ2 has Paulsen's similarity property with constant max{r1, r2}. (2) Assume that θ = θ1⊕ θ2 has Paulsen's similarity property with constant r. It suffices to establish the claim for θ1. Using Theorem 5.4, we see that kΥ(θ)kcb ≤ r2. By construction of Υ(θ), we see that H1 and H2 are reducing for Υ(θ)(Υ(A)). We can thus write Υ(θ) = Θ1 ⊕ Θ2 where Θ1 : Υ(A) → B(H1), Θ2 : Υ(A) → B(H2) are unital completely bounded homomorphisms. It is readily verified that Θ1 = Υ(θ1) and it is clear that kΘ1kcb ≤ r2. Another application of Theorem 5.4 shows that θ1 has Paulsen's similarity property with constant r. (cid:3) In light of Theorem 5.4, it is easy to construct homomorphisms θ that do not have Paulsen's similarity property with the smallest possible constant, namely kθk1/2 cb . Example 1. Let which is an idempotent. Consider the homomorphism θ : C → M2(C) uniquely determined by θ(1) = E. Then, θ is a completely bounded homomorphism with E =(cid:20) 2 −2 1 −1 (cid:21) ∈ M2(C) kθkcb = kEk =(cid:13)(cid:13)(cid:13)(cid:13)(cid:20) 5 −5 5 (cid:21)(cid:13)(cid:13)(cid:13)(cid:13) −5 1/2 υ(λ) = λ ⊕ 0, λ ∈ C. = √10. Since C is unital, we have that Υ(C) = C ⊕ C and υ : C → Υ(C) is given by We find and so Υ(θ)(1 ⊕ −1) = Υ(θ)(2 ⊕ 0) − Υ(θ)(1 ⊕ 1) = θ(2) − I =(cid:20) 4 −4 2 −2 (cid:21) −(cid:20) 1 0 0 1 (cid:21) =(cid:20) 3 −4 2 −3 (cid:21) kΥ(θ)kcb ≥ kΥ(θ)(1 ⊕ −1)k =(cid:13)(cid:13)(cid:13)(cid:13)(cid:20) 13 −18 25 (cid:21)(cid:13)(cid:13)(cid:13)(cid:13) > kθkcb. −18 1/2 By Theorem 5.4, we conclude that θ does not have Paulsen's similarity property with constant r = kθk1/2 cb . 6. A subclass of the completely bounded maps The main issue behind the shortcomings of the class CBr(M) exhibited in Sec- tion 4 is the lack of a perfect analogue of Stinespring's dilation theorem. Even though Theorems 4.4 and 5.1 are useful replacements, the fact remains that a com- pletely bounded linear map does not necessarily dilate to a completely bounded homomorphism of the same norm, which conflicts with the machinery developed in Section 3. In this section, we attempt to remedy this problem by restricting our attention to a smaller class of maps. A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 19 We recall at the onset that we always assume that operator spaces are concretely represented on some Hilbert space. This will be a standing assumption throughout this section and the next. In particular, it makes sense to consider the C∗-algebra generated by an operator space, although this depends on the choice of representa- tion. This dependence is not relevant for the purposes of Sections 6 and 7. We will carefully analyze the impact of the choice of representation in Section 8. Let M be an operator space and let r ≥ 1. Given a Hilbert space H, we denote by P r(M,H) the set of linear maps ϕ : M → B(H) for which there is a Hilbert space Kϕ containing H along with a completely bounded homomorphism θ : C∗(M) → B(Kϕ) that has Paulsen's similarity property with constant r1/2 and such that ϕ(a) = PHθ(a)H, a ∈ M. In view of Theorem 5.4, the reader may venture to guess that ϕ ∈ P r(M,H) if and only if Υ(ϕ) ∈ CBr(Υ(M),H). However, simple examples show that such a description unfortunately does not hold (see Example 8 below). The remainder of this section is devoted to exhibiting some of the basic prop- erties of the subclass P r(M). We first note that it is clear that P r(M,H) ⊂ CBr(M,H), and it follows from Theorems 5.1 and 5.2 that there is a positive constant Cr > r depending only on r such that CBr(M,H) ⊂ P Cr (M,H). This seems to indicate that the inclusion P r(M,H) ⊂ CBr(M,H) may be strict. This is indeed the case. Before proceeding with the example illustrating this fact, we note in passing that similar questions were considered in [34]. Example 2. Let H = C, let M = B(H) (so that M = C∗(M)) and let ϕ : M → B(H) be the completely bounded linear map of multiplication by r > 1, so that ϕ(λ) = rλ for λ ∈ M. Let K be a Hilbert space containing H and let θ : C∗(M) → B(K) be a homomorphism such that ϕ(λ) = PHθ(λ)H, λ ∈ M. Then, with respect to the decomposition K = H ⊕ (K ⊖ H) we have θ(1) =(cid:20) r a c (cid:21) . b Since θ is multiplicative, we must have that θ(1) is idempotent, which is easily seen to force b 6= 0 as r2 6= r. Hence kθkcb ≥ kθ(1)k ≥pkbk2 + r2 > r = kϕkcb and we infer that ϕ ∈ CBr(M, C) \ P r(M, C). Although we needed to choose r > 1 in the example above, it can be shown that the inclusion P 1(M,H) ⊂ CB1(M,H) is still strict (see Example 3 below). We note also that the basic idea behind the previous example is to exploit a multiplica- tive "relation" within M that is not preserved by the linear map ϕ. This idea can be extended to identify an obstruction for a map in CBr(M) to lie inside the class P r(M). 20 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY Theorem 6.1. Let M be an operator space and let ϕ : M → B(Hϕ) be a linear map with kϕkcb = r. Assume that there are two elements a0, b0 ∈ M such that ka0k = kb0k = 1, b0a0 = 0, ϕ(a0) = rI and ϕ(b0) 6= 0. Then, ϕ does not lie in P r(M,Hϕ). Proof. Let Kϕ be a Hilbert space containing Hϕ and let θ : C∗(M) → B(Kϕ) be a homomorphism such that ϕ(a) = PHϕ θ(a)Hϕ , a ∈ M. Since b0a0 = 0, we must have θ(b0)θ(a0) = 0 and in particular 0 = PHϕθ(b0)θ(a0)PHϕ = ϕ(b0)ϕ(a0) + PHϕ θ(b0)PH⊥ = rϕ(b0) + PHϕθ(b0)PH⊥ θ(a0)PHϕ . ϕ ϕ θ(a0)PHϕ Hence, we find kPHϕθ(b0)PH⊥ ϕ kkPH⊥ ϕ θ(a0)PHϕk ≥ rkϕ(b0)k > 0. Since ϕ(a0) = rI, we find kθ(a0)k2 ≥ kθ(a0)PHϕk2 = kϕ(a0) + PH⊥ ϕ = r2 + kPH⊥ ϕ θ(a0)PHϕk2 > r2 θ(a0)PHϕk2 and thus kθkcb > r. We conclude that ϕ does not lie in P r(M,Hϕ). (cid:3) Examples satisfying the conditions of the previous theorem are easily constructed. Example 3. Let M be an operator space containing an operator N such that kNk = 1 and N 2 = 0. Choose a linear functional ϕ : M → C with kϕk = 1 and ϕ(N ) = 1. Then, kϕkcb = 1. Applying Theorem 6.1 with a0 = b0 = N shows that ϕ ∈ CB1(M, C) \ P 1(M, C). Obviously, the element a0 from Theorem 6.1 can never be the identity of M. Nevertheless, even if we insist that a map ϕ ∈ CBr(M,H) be unital, it still may not lie in P r(M,H). Example 4. Let M ⊂ M2(C) be the unital operator space consisting of upper triangular Toeplitz matrices, that is For convenience, given x, y ∈ C we use the following notation M =(cid:26)(cid:20)x y 0 x(cid:21) : x, y ∈ C(cid:27) . Tx,y =(cid:20)x y 0 x(cid:21) . Let ϕ : M → C be the unital linear functional defined as x, y ∈ C. If x 6= 0 and y 6= 0, a standard verification shows that ϕ(Tx,y) = x + y, kTx,yk2 > x2 + y2. Thus, ϕ(Tx,y) = x + y ≤ √2px2 + y2 < √2kTx,yk by the Cauchy-Schwarz inequality. Moreover, for every x ∈ C and y ∈ C we have ϕ(T0,y) = y = kT0,yk, ϕ(Tx,0) = x = kTx,0k. A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 21 Hence, we conclude that ϕ(T ) < √2kTk for every T ∈ M with kTk = 1. By compactness, we infer that kϕk < √2, and since ϕ is a functional we find ϕ ∈ CB√2−ε(M, C) for some ε > 0. Let K be a Hilbert space containing H = C as a subspace. Assume that θ : C∗(M) → B(K) is a homomorphism such that ϕ(T ) = PHθ(T )H, T ∈ M. We see that ϕ(T0,1) = 1 so we may write θ(T0,1) =(cid:20) 1 A B C(cid:21) . Since θ is multiplicative and T 2 0,1 = 0, we have θ(T0,1)2 = 0. In particular this forces 1 + AB = 0. Thus, either kAk ≥ 1 or kBk ≥ 1, which implies that kθ(T0,1)k ≥ √2. Since kT0,1k = 1, we conclude that kθkcb ≥ √2. Thus, we see that ϕ ∈ CB√2−ε(M, C) \ P√2−ε(M, C). Before establishing a useful property of maps in the class P r(M), we need the following elementary calculation. Lemma 6.2. Let M be an operator space and let ϕ : M → B(Hϕ) be a linear map. Assume that there is a Hilbert space Kϕ containing Hϕ and a linear map ψ : M → B(Kϕ) such that ϕ(a) = PHϕψ(a)Hϕ , a ∈ M. Let X ∈ B(Hϕ) be invertible with kXk = kX−1k. Then, there is another invertible operator X′ ∈ B(Kϕ) for which the following assertions hold. (a) We have kXk = kX′k = kX−1k = kX′−1k. (b) The space Hϕ is reducing for X′ and X′Hϕ = Hϕ = X′∗Hϕ. (c) We have Xϕ(a)X−1 = PHϕ X′ψ(a)X′−1Hϕ, a ∈ M. Proof. Define the invertible operator X′ : Kϕ → Kϕ as X′ = X ⊕ I according to the decomposition Kϕ = Hϕ ⊕ H⊥ϕ . Properties (a) and (b) are clearly satisfied, so it remains to establish (c). We have and so we find X = X′Hϕ, X−1 = X′−1Hϕ X′PHϕ = PHϕX′, PHϕ X′−1 = X′−1PHϕ Xϕ(a)X−1 = X′ϕ(a)X′−1Hϕ = X′PHϕψ(a)PHϕ X′−1Hϕ = PHϕ X′ψ(a)X′−1Hϕ for every a ∈ M. (cid:3) We now exhibit an important dilation property of maps in the class P r. Theorem 6.3. Let M be an operator space, let r ≥ 1 and let ϕ ∈ P r(M,Hϕ). Then, there is a Hilbert space Kϕ containing Hϕ, a ∗-homomorphism σ : C∗(M) → B(Kϕ) and an invertible operator X ∈ B(Kϕ) with kXk = kX−1k ≤ r1/2 22 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY such that ϕ(a) = PHϕ Xσ(a)X−1Hϕ , a ∈ M. Proof. By assumption, there is a Hilbert space K containing Hϕ and a homomor- phism θ : C∗(M) → B(K) that has Paulsen's similarity property with constant r1/2 and such that ϕ(a) = PHϕ θ(a)Hϕ , a ∈ M. Applying Corollary 5.3 to θ, we obtain a Hilbert space Kϕ containing K, a ∗- homomorphism σ : C∗(M) → B(Kϕ) and an invertible operator X ∈ B(K) with such that kXk = kX−1k ≤ r1/2 Xθ(a)X−1 = PKσ(a)K, a ∈ C∗(M). By Lemma 6.2, we can find an invertible operator X′ ∈ B(Kϕ) such that and We conclude that kX′k = kXk ≤ r1/2, kX′−1k = kX−1k ≤ r1/2 θ(a) = PKX′−1σ(a)X′K, a ∈ C∗(M). ϕ(a) = PHϕX′−1σ(a)X′Hϕ , a ∈ M. (cid:3) Given an operator space M, a positive number r and a Hilbert space H, we de- note by CPr(M,H) the set of linear maps ϕ : M → B(H) that admit a completely positive extension ψ : C∗(M) → B(H) satisfying kψkcb ≤ r. An interesting conse- quence of the previous theorem is an upper bound for the distance of an element of P r(M,H) to the set CP1(M,H). Before stating it, we need a simple preliminary calculation. Let A be a unital C∗-algebra and let X ∈ A be a positive invertible operator with kXk = kX−1k. Then, for each a ∈ A we have ka − XaX−1k ≤ k(I − X)ak + kXa(I − X−1)k ≤ kak(kI − Xk + kXkkI − X−1k) ≤ kak(1 + kXk) max{kI − Xk,kI − X−1k}. Since X is positive and satisfies kXk = kX−1k, we may use the spectral theorem to conclude that max{kI − Xk,kI − X−1k} ≤ max{kXk − 1, 1 − 1/kXk} so that ka − XaX−1k ≤ kak(1 + kXk) max{kXk − 1, 1 − 1/kXk}. We can now establish the announced distance estimate. Corollary 6.4. Let M be an operator space, let r ≥ 1 and let ϕ ∈ P r(M,H). Then, there is a linear map ψ ∈ CP1(M,H) with the property that kϕ − ψkcb ≤ r − 1. A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 23 Proof. By Theorem 6.3, there is a Hilbert space Kϕ containing H, a ∗-homomorphism σ : C∗(M) → B(Kϕ) and an invertible operator X ∈ B(Kϕ) with kXk = kX−1k ≤ r1/2 such that ϕ(a) = PHXσ(a)X−1H, a ∈ M. Using the polar decomposition if necessary, we may assume that X is positive. Let now ψ : M → B(H) be the linear map defined as ψ(a) = PHσ(a)H, a ∈ M. Then, ψ ∈ CP1(M,H). Using the calculation preceding the corollary, we see that kϕ − ψkcb ≤ (1 + r1/2) max{r1/2 − 1, 1 − 1/r1/2} = (1 + r1/2)(r1/2 − 1) = r − 1. We can now show that P 1(M,H) = CP1(M,H) for every Hilbert space H. (cid:3) Corollary 6.5. Let M be an operator space and let ϕ : M → B(H) be a linear map. Then, ϕ ∈ P 1(M,H) if and only if ϕ admits a contractive completely positive extension to C∗(M). Proof. If ϕ ∈ P 1(M,H) then ϕ ∈ CP1(M,H) by Corollary 6.4. Conversely, assume that ϕ admits a contractive completely positive extension to C∗(M). By Lemma 4.5, there is a Hilbert space Kϕ containing H and ∗-homomorphism π : C∗(M) → B(Kϕ) such that ϕ(a) = PHπ(a)H, a ∈ M. Thus, ϕ ∈ P 1(M,H). (cid:3) We make a few remarks regarding the previous result. First, note that if M is unital and ϕ : M → B(H) is a unital completely contractive map, then ϕ admits a unital (hence contractive) completely positive extension to C∗(M), so that ϕ ∈ P 1(M,H) by Corollary 6.5. This shows that Example 4 is somewhat sharp: the norm of the functional ϕ cannot be taken to be 1 therein. Moreover, we mention that Example 2 shows that Corollary 6.5 fails beyond the completely contractive setting. Indeed, that example exhibits a completely positive map ϕ : C → C with kϕkcb = r > 1 such that ϕ does not lie in P r(C, C). The reverse inclusion also fails as the next example illustrates. Example 5. Let H be a Hilbert space and let X ∈ B(H) be a non-unitary invertible operator with kXk = kX−1k = r1/2. In particular, r > 1. Let θ : B(H) → B(H) be defined as a ∈ B(H). θ(a) = XaX−1, It is clear that θ ∈ P r(B(H),H). Moreover, by applying θ to rank-one operators it is easily verified that kθk = r > 1, so that kθk > kθ(I)k. Thus, θ is not completely positive. 24 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY A useful consequence of Corollary 6.5 is the following corollary. It guarantees that certain maps preserve the classes P r(M). Corollary 6.6. Let M ⊂ B(HM) and N ⊂ B(HN ) be operator spaces, let τ ∈ P 1(M,HN ) such that τ (M) ⊂ N and let ϕ ∈ P r(N ,Hϕ). Then, ϕ ◦ τ ∈ P r(M,Hϕ). Proof. By assumption, there is a Hilbert space Kϕ containing Hϕ and a homomor- phism θ : C∗(N ) → B(Kϕ) that has Paulsen's similarity property with constant r1/2 and is such that ϕ(a) = PHϕ θ(a)Hϕ , a ∈ N . By Corollary 5.3, there is an invertible operator X ∈ B(Kϕ) with kXk = kX−1k ≤ r1/2 such that the homomorphism θX : C∗(N ) → B(Kϕ) defined as θX (a) = Xθ(a)X−1, a ∈ C∗(N ) is contractive and completely positive. Next, use Corollary 6.5 to find a contractive completely positive map Ψ : C∗(M) → B(HN ) that agrees with τ on M. By Arveson's extension theorem there is a contractive completely positive map Ξ : B(HN ) → B(Kϕ) that agrees with θX on C∗(N ). Then, Ξ ◦ Ψ is contractive and completely positive, and it agrees with θX◦τ on M. Hence, by Lemma 4.5 we obtain a Hilbert space K containing Kϕ and a ∗-homomorphism π : C∗(M) → B(K) such that By Lemma 6.2, we see that θX ◦ τ (a) = PKϕπ(a)Kϕ , a ∈ M. θ ◦ τ (a) = PKϕπX (a)Kϕ, a ∈ M where πX : C∗(M) → B(K) is a homomorphism that has Paulsen's similarity property with constant r1/2. Finally, we note that ϕ ◦ τ (a) = PHϕθ ◦ τ (a)Hϕ = PHϕ πX (a)Hϕ , so that ϕ ◦ τ ∈ P r(M,Hϕ). a ∈ M (cid:3) We close this section by describing a relationship between the classes P r(M) for different values of the parameter r ≥ 1. The key technical tool is the following, that we require in later sections as well. See Subsection 2.3 for some background on ultraproducts. Lemma 6.7. Let M be an operator space, let Λ be a directed set and let U be a cofinal ultrafilter on Λ. For each λ ∈ Λ, let ϕλ ∈ P rλ(M,Hλ). Assume that (rλ)λ∈Λ is a bounded. Then, the ultraproduct limU (ϕλ)λ∈Λ yields an element of P r(M,HU ), where r = limU (rλ)λ∈Λ and HU is the ultraproduct Hilbert space of (Hλ)λ∈Λ along U . Proof. For each λ ∈ Λ, there is a Hilbert space Kλ containing Hλ along with a completely bounded homomorphism θλ : C∗(M) → B(Kλ) that has Paulsen's similarity property with constant r1/2 and such that λ Thus, for each λ ∈ Λ there is an invertible operator Xλ ∈ B(Kλ) with ϕλ(a) = PHλθλ(a)Hλ, a ∈ M. kXλk = kX−1 λ k ≤ r1/2 λ A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 25 and such that the homomorphism Ξλ : C∗(M) → B(Kλ) defined as Ξλ(a) = Xλθλ(a)X−1 λ , a ∈ C∗(M) is completely contractive. Let X ∈ B(KU ) be defined as limU (Xλ)λ∈Λ. Then, X is invertible with kXk ≤ r1/2 and kX−1k ≤ r1/2. Upon renormalizing, we may assume that kXk = kX−1k ≤ r1/2. Let θ : C∗(M) → B(KU ) be the homomorphism defined as limU (θλ)λ∈Λ. Note that Xθ(a)X−1 = lim U (Xλθλ(a)X−1 λ )λ∈Λ, a ∈ C∗(M) so that θ has Paulsen's similarity property with constant r1/2. Finally, we observe that HU ⊂ KU , and if ξ = (ξλ)λ∈Λ, η = (ηλ)λ∈Λ are two elements ofQλ∈Λ Hλ then we have hθ(a)[ξ], [η]iKU = lim U = lim U = h(lim U (hθλ(a)ξλ, ηλiKλ)λ∈Λ (hϕλ(a)ξλ, ηλiHλ)λ∈Λ (ϕλ)λ∈Λ)(a)[ξ], [η]iHU . We conclude that limU (ϕλ)λ∈Λ lies in P r(M,HU ). (cid:3) We can now describe a certain continuity property of the class P r(M) with respect to r. Theorem 6.8. Let M be an operator space, let r ≥ 1 and let H be a Hilbert space. Then, P r(M,H) = ∩ε>0 P r+ε(M,H). Proof. It follows from the definition that P r(M,H) ⊂ ∩ε>0 P r+ε(M,H). Con- versely, let ϕ ∈ ∩ε>0 P r+ε(M,H). Thus, ϕ ∈ ∩n P r+1/n(M,H). Let U be a cofinal ultrafilter on N. By Lemma 6.7 we see that limU (ϕ)n∈N ∈ P r(M,HU ). Finally, let V : H → HU be the isometry defined as A standard verification yields V ξ = lim U (ξ)n∈N, ξ ∈ H. ϕ(a) = V ∗(cid:16)lim U (ϕ)n∈N(cid:17) (a)V, a ∈ M so an application of Lemma 4.3 shows that indeed ϕ ∈ P r(M,H). (cid:3) 7. Extremals in the class P r(M) We emphasize once more that we always assume that operator spaces are con- cretely represented on some Hilbert space. Let M be an operator space and let r ≥ 1. In this section we restrict the dilation order defined on the class CBr(M) in Section 4 to the subclass P r(M). We recall the notation and terminology here for convenience. Let ϕ ∈ P r(M,Hϕ) and ψ ∈ P r(M,Hψ). We write ϕ ≺ ψ if Hϕ ⊂ Hψ and a ∈ M. ϕ(a) = PHϕ ψ(a)Hϕ , Clearly, this is a dilation order on P r(M). We say that an element ω ∈ P r(M,Hω) is P r(M)-extremal if whenever δ ∈ P r(M,Hδ) satisfies ω ≺ δ, we necessarily have that Hω is reducing for δ(M). It is an easy consequence of Lemma 6.2 that 26 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY P r(M)-extremal elements are preserved by unitary equivalence. This will be used throughout, often without mention. We will also require the following simple ob- servation. Lemma 7.1. Let M be an operator space and let r ≥ 1. Let ω ∈ P r(M,Hω) be a P r(M)-extremal element and let X ⊂ Hω be a reducing subspace for ω(M). If we define ω′ : M → B(X ) as ω′(a) = ω(a)X , a ∈ M then ω′ ∈ P r(M,X ) and it is a P r(M)-extremal element. Proof. It is straightforward to see that ω′ ∈ P r(M,X ). Let δ′ ∈ P r(M,Hδ) such that ω′ ≺ δ′ and define δ : M → Hδ ⊕ (Hω ⊖ X ) as δ(a) = δ′(a) ⊕ ω(a)(Hω⊖X ), a ∈ M. Upon invoking Corollary 5.5, it is readily verified that δ ∈ P r(M,Hδ ⊕ (Hω ⊖X )). Moreover, we note that ω ≺ δ. Hence, Hω is reducing for δ(M), which implies in particular that X is reducing for δ′(M). (cid:3) Our immediate goal is to establish the existence of P r(M)-extremal elements. For that purpose, we first show that P r(M) has the limit property with respect to the dilation order ≺. This is the first instance where our working with this smaller class of linear maps creates difficulties that are not present in the standard setting of [22], [18] and [23]. Indeed, on top of the usual inductive limit procedure of Lemma 4.1, we have to use the ultraproduct machinery from Subsection 2.3 to keep track of the multiplicative dilations. Lemma 7.2. Let M be an operator space and let r ≥ 1. Then, the class P r(M) has the limit property with respect to the dilation order ≺. Proof. Let Λ be a totally ordered set. For each λ ∈ Λ, let ϕλ ∈ P r(M,Hλ). Assume that Hλ ⊂ Hµ and ϕλ(a) = PHλϕµ(a)Hλ , a ∈ M whenever µ ≥ λ. Set H = ∪λHλ. We need to find an element ψ ∈ P r(M,H) such that ϕλ ≺ ψ for every λ ∈ Λ. Let ψ : M → B(H) be the map constructed from the collection (ϕλ)λ∈Λ as in the proof of Lemma 4.1. It is clear that ϕλ ≺ ψ for every λ ∈ Λ. We need only verify that ψ ∈ P r(M,H). Let U be a cofinal ultrafilter on Λ and let HU be the ultraproduct Hilbert space of (Hλ)λ∈Λ along U. For each µ ∈ Λ we define an isometry Vµ : Hµ → HU as follows. If ξ ∈ Hµ then Vµξ = [(ξλ)λ∈Λ], where ξλ =(ξ 0 if λ ≥ µ if λ < µ. Using that U is cofinal, it is easily verified that there exists another isometry V : H → HU such that V ξ = Vµξ if ξ ∈ Hµ. Now, let ϕ : M → B(HU ) be defined as ϕ = limU (ϕλ)λ∈Λ. We show that ψ(a) = V ∗ϕ(a)V, a ∈ M. A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 27 Indeed, assume that ξ, η ∈ Hµ. Then, using again the fact that U is cofinal we find hV ∗ϕ(a)V ξ, ηiH = hϕ(a)V ξ, V ηiHU = lim U (hϕλ(a)ξ, ηiHλ )λ∈Λ = lim U (cid:0)hϕµ(a)ξ, ηiHµ(cid:1)λ∈Λ = hϕµ(a)ξ, ηiHµ = hψ(a)ξ, ηiH. Since H = ∪µHµ we conclude that ψ(a) = V ∗ϕ(a)V, a ∈ M as claimed. By Lemma 6.7 we see that ϕ ∈ P r(M,HU ), so that an application of Lemma 4.3 shows that ψ ∈ P r(M,H). (cid:3) The following is now a straightforward consequence. Theorem 7.3. Let M be an operator space, let r ≥ 1 and let ϕ ∈ P r(M,Hϕ). Then, there exists a P r(M)-extremal element ω ∈ P r(M,Hω) such that ϕ ≺ ω and such that dimHω ≤ (1 + ℵ0 dimM) dimHϕ Proof. The class P r(M) has the limit property with respect to the dilation order ≺ by Lemma 7.2. By Theorem 3.2, we see that there is a maximal element ζ ∈ P r(M,K) such that ϕ ≺ ζ. We claim that ζ is P r(M)-extremal. Indeed, let δ ∈ P r(M,Hδ) such that ζ ≺ δ. We calculate for every a ∈ M and ξ ∈ K that k(δ(a) − ζ(a))ξk2 = kδ(a)ξk2 + kζ(a)ξk2 − 2 Rehδ(a)ξ, ζ(a)ξi = kδ(a)ξk2 − kζ(a)ξk2 = 0 so that δ(a)ξ = ζ(a)ξ. Likewise, k(δ(a)∗ − ζ(a)∗)ξk2 = kδ(a)∗ξk2 + kζ∗(a)ξk2 − 2 Rehδ(a)∗ξ, ζ(a)∗ξi = kδ(a)∗ξk2 − kζ(a)∗ξk2 = 0 so that δ(a)∗ξ = ζ(a)∗ξ. Hence, δ(M)K ⊂ K and δ(M)∗K ⊂ K, so indeed ζ is P r(M)-extremal. Let Hω = Hϕ + C∗(ζ(M))Hϕ. This is the smallest reducing subspace for ζ(M) that contains Hϕ. By choosing a Hamel basis for ζ(M), we can find a dense subset of C∗(ζ(M)) with cardinality at most ℵ0 dim ζ(M). Hence, there is a total subset of Hω with cardinality at most By applying the Gram-Schmidt algorithm, we find dim Hϕ + ℵ0 dim ζ(M) dim Hϕ. dimHω ≤ (1 + ℵ0 dim ζ(M)) dim Hϕ ≤ (1 + ℵ0 dimM) dimHϕ. We define ω : M → B(H) as ω(a) = ζ(a)Hω , a ∈ M. It is clear that ϕ ≺ ω, and it follows from Lemma 7.1 that ω ∈ P r(M,Hω) is P r(M)-extremal. (cid:3) For our purposes, it will be relevant to know if the collection of P r(M)-extremals "completely norms" the operator space M, in the following precise sense. 28 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY Corollary 7.4. Let M be an operator space and let r ≥ 1. Then, there is a P r(M)-extremal element ω ∈ P r(M,Hω) such that for every n ∈ N and every a ∈ Mn(M) we have kak ≤ kω(n)(a)k. Moreover, we have that dim Hω ≤ (dimM)2(1 + ℵ0 dim M). Proof. There is an isometric ∗-homomorphism π : C∗(M) → B(H). Using a Hamel basis, we may find a dense subset S ⊂ C∗(M) with cardS ≤ ℵ0 dim M. For each a ∈ S and m ∈ N, choose a unit vector ξa,m ∈ H such that Let H′ denote the smallest closed subspace containing the sets kπ(a)ξa,mk ≥ kak(1 − 1/m). {ξa,m : a ∈ S, m ∈ N} and Then, H′ is reducing for π(C∗(M)). Let π′ : C∗(M) → B(H′) be the ∗-homomorphism defined as {π(C∗(M))ξa,m : a ∈ S, m ∈ N}. π′(a) = π(a)H′ , a ∈ C∗(M). By construction, we see that π′ is isometric, and hence completely isometric. Now, H′ contains a total subset of cardinality at most By applying the Gram-Schmidt algorithm, we find ℵ0 cardS + ℵ0(cardS)2. dimH′ ≤ ℵ0 cardS + ℵ0(cardS)2 ≤ ℵ0(dim M)2. Let ϕ : M → B(H′) be defined as ϕ(a) = π′(a), a ∈ M. Clearly, we have that ϕ ∈ P r(M,H′). By Theorem 7.3, there is a P r(M)-extremal element ω ∈ P r(M,Hω) such that ϕ ≺ ω and dim Hω ≤ (1 + ℵ0 dim M) dimH′ ≤ (dimM)2(1 + ℵ0 dim M). Since ≺ is a dilation order, we have for every n ∈ N and every a ∈ Mn(M). kω(n)(a)k ≥ kϕ(n)(a)k = kak (cid:3) Let M be an operator space, let AM denote the operator algebra it generates, and let r ≥ 1. An element ϕ ∈ P r(M,Hϕ) is said to have the unique extension property relative to P r(M) if there exits a unique element Φ ∈ P r(AM,Hϕ) that agrees with ϕ on M, and if this unique map Φ is a homomorphism that has Paulsen's similarity property with constant r1/2. It is clear that ϕ ∈ P r(M,Hϕ) has the unique extension property relative to P r(M) if and only if the following two statements hold: (a) there is Φ ∈ P r(AM,Hϕ) that agrees with ϕ on M, and (b) every Ψ ∈ P r(AM,Hϕ) that agrees with ϕ on M is a homomorphism that has Paulsen's similarity property with constant r1/2. We now arrive at an important property of P r(M)-extremals. A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 29 Theorem 7.5. Let M be an operator space, let r ≥ 1 and let ω ∈ P r(M,Hω) be a P r(M)-extremal element. Then, ω has the unique extension property relative to P r(M). Proof. By assumption, there is a Hilbert space K containing Hω along with a homomorphism θ : C∗(M) → B(K) that has Paulsen's similarity property with constant r1/2 and such that ω(a) = PHω θ(a)Hω , a ∈ M. The map Ω : AM → B(Hω) defined as a ∈ AM clearly lies in P r(AM,Hω) and agrees with ω on M. It remains to prove that any such extension is a homomorphism that has Paulsen's similarity property with constant r1/2. Ω(a) = PHω θ(a)Hω , For that purpose, let Ψ ∈ P r(AM,Hω) that agrees with ω on M. Then, there is a Hilbert space Kω containing Hω along with a homomorphism ρ : C∗(M) → B(Kω) that has Paulsen's similarity property with constant r1/2 and such that Ψ(a) = PHω ρ(a)Hω , a ∈ AM. Let δ denote the restriction of ρ to M, so that δ ∈ P r(M,Kω) and ω ≺ δ. Since ω is assumed to be P r(M)-extremal, we conclude that Hω is reducing for δ(M), and hence for ρ(AM). Thus, we have Ψ(a) = ρ(a)Hω , a ∈ AM which shows that Ψ is multiplicative. Invoking Corollary 5.5 we see that Ψ has Paulsen's similarity property with constant r1/2. (cid:3) Although we will not need this fact, we remark that the argument used in the previous proof can be used to show that the unique multiplicative extension of ω to AM must be P r(AM)-extremal. We leave the simple details to the reader. We emphasize here that Theorem 7.5 justifies our considering the subclass P r(M) rather than the full class CBr(M), since in this context every extremal extends to be multiplicative on AM. This stands in contrast with the situation for general completely bounded linear maps, as illustrated by Theorem 4.7. Next, we examine another unique extension property. An element ϕ ∈ CP1(M,Hϕ) is said to have the unique extension property relative to CP1(M) if there ex- ists a unique element Φ ∈ CP1(C∗(M),Hϕ) that agrees with ϕ on M, and if moreover Φ is a ∗-homomorphism. By definition, an element ϕ ∈ CP1(M,Hϕ) has at least one contractive completely positive extension to C∗(M). Hence, ϕ ∈ CP1(M,Hϕ) has the unique extension property relative to CP1(M) if and only if every Ψ ∈ CP1(C∗(M),Hϕ) that agrees with ϕ on M is a ∗-homomorphism. We can now refine Theorem 7.5. Theorem 7.6. Let M be an operator space and let r ≥ 1. Let ω ∈ P r(M,Hω) be P r(M)-extremal. Then, the following statements hold. (1) There is an invertible operator X ∈ B(Hω) with kXk = kX−1k ≤ r1/2 such that the map ωX : M → B(Hω) defined as ωX (a) = Xω(a)X−1, a ∈ M belongs to P 1(M,Hω). 30 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY (2) Let Y ∈ B(Hω) be an invertible operator with kY k = kY −1k ≤ r1/2 and let ωY : M → B(Hω) be defined as ωY (a) = Y ω(a)Y −1, a ∈ M. If the map ωY belongs to P 1(M,Hω), then it has the unique extension property relative to CP1(M). Proof. (1) By Theorem 7.5 there is a homomorphism Ω : AM → B(Hω) that has Paulsen's similarity property with constant r1/2 and that agrees with ω on M. By Corollary 5.3, there is an invertible operator X ∈ B(Hω) such that kXk = kX−1k ≤ r1/2, a Hilbert space Kω containing Hω and a ∗-homomorphism π : C∗(M) → B(Kω) such that ωX (a) = Xω(a)X−1 = PHω π(a)Hω , a ∈ M. In particular, we have that ωX ∈ CP1(M,Hω) and thus ωX ∈ P 1(M,Hω) by Corollary 6.5. (2) We see that ωY ∈ CP1(M,Hω) by Corollary 6.5. Let Ψ : C∗(M) → B(Hω) be a contractive completely positive map that agrees with ωY on M. We need to show that Ψ is a ∗-homomorphism. By Lemma 4.5, there is a Hilbert space K containing Hω and a ∗-homomorphism π : C∗(M) → B(K) such that In particular, we see that Ψ(a) = PHω π(a)Hω , a ∈ C∗(M). ωY (a) = PHω π(a)Hω , a ∈ M. We can now invoke Lemma 6.2 to find another invertible operator Y ′ ∈ B(K) with kY ′k = kY ′−1k ≤ r1/2, such that the space Y ′Hω = Hω = Y ′∗Hω is reducing for Y ′ and such that ω(a) = Y −1ωY (a)Y = PHω πY (a)Hω , a ∈ M where πY : C∗(M) → B(K) is defined as πY (a) = Y ′−1π(a)Y ′, a ∈ C∗(M). If we denote by δ the restriction of πY to M, then δ ∈ P r(M,K) and ω ≺ δ. Since ω is assumed to be P r(M)-extremal, we conclude that Y ′−1π(M)Y ′Hω ⊂ Hω, (Y ′−1π(M)Y ′)∗Hω ⊂ Hω or π(M)Y ′Hω ⊂ Y ′Hω, π(M)∗Y ′∗−1Hω ⊂ Y ′∗−1Hω. Since π is a ∗-homomorphism and Y ′Hω = Hω = Y ′∗−1Hω, this translates to π(M)Hω ⊂ Hω, π(M∗)Hω ⊂ Hω and thus π(C∗(M))Hω ⊂ Hω. But recall that Ψ(a) = PHω π(a)Hω , a ∈ C∗(M) so that Ψ is a ∗-homomorphism. (cid:3) A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 31 Given an operator space M and two linear maps ϕ ∈ CP1(M,Hϕ), ψ ∈ CP1(M,Hψ) we use our standard notation ϕ ≺ ψ to mean that Hϕ ⊂ Hψ and ϕ(a) = PHϕ ψ(a)Hϕ , a ∈ M. A linear map ω ∈ CP1(M,Hω) is said to be CP1(M)-extremal if whenever δ ∈ CP1(M,Hδ) satisfies ω ≺ δ, we have that Hω is reducing for δ(M). Interestingly, CP1(M)-extremality and the unique extension property relative to CP1(M) are equivalent. This is reminiscent of the classical setting of unital completely positive maps [22], and the proof is very similar. We provide it below for completeness. Lemma 7.7. Let M be an operator space and let ω ∈ CP1(M,Hω). Then, ω is CP1(M)-extremal if and only if it has the unique extension property relative to CP1(M). Proof. Throughout the proof we assume that M ⊂ B(HM). Assume first that ω is CP1(M)-extremal. Let Ψ : C∗(M) → B(Hω) be a contractive completely positive map that agrees with ω on M. We must show that Ψ is a ∗-homomorphism. By Lemma 4.5, there is a Hilbert space Kω containing Hω and a ∗-homomorphism π : C∗(M) → B(Kω) such that a ∈ C∗(M). Ψ(a) = PHω π(a)Hω , In particular, we have ω(a) = PHω π(a)Hω , a ∈ M. By assumption, we see that Hω is reducing for π(M), and hence for π(C∗(M)). This immediately implies that Ψ is a ∗-homomorphism. Conversely, assume that ω has the unique extension property relative to CP1(M). Let δ ∈ CP1(M,Hδ) be such that ω ≺ δ. There is a contractive completely positive map ∆ : C∗(M) → B(Hδ) that agrees with δ on M. By Lemma 4.5, there is a Hilbert space Kδ containing Hδ and a ∗-homomorphism π : C∗(M) → B(Kδ) such that ∆(a) = PHδ π(a)Hδ , If we let Ψ : C∗(M) → B(Hω) be defined as Ψ(a) = PHω ∆(a)Hω , a ∈ C∗(M). a ∈ C∗(M) then we see that Ψ is contractive and completely positive, and it agrees with ω on M. By assumption, we see that Ψ is a ∗-homomorphism. Let now a ∈ C∗(M). Applying the Schwarz inequality for completely positive maps [31, Exercise 3.4] to a contractive completely positive extension of ∆ to B(HM), we obtain whence and thus ∆(a)∗∆(a) ≤ ∆(a∗a) PHω ∆(a)∗∆(a)PHω ≤ PHω ∆(a∗a)PHω Ψ(a)∗Ψ(a) + PHω π(a)∗PH⊥ ω π(a)PHω ≤ Ψ(a∗a). But Ψ is a ∗-homomorphism so that PHω π(a)∗PH⊥ ω π(a)PHω ≤ 0 32 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY ω π(a)Hω = 0. Since a ∈ C∗(M) is arbitrary, we conclude that π(C∗(M))Hω ⊂ or PH⊥ Hω and in particular δ(M)Hω ⊂ Hω and δ(M)∗Hω ⊂ Hω, so that Hω is reducing for δ(M). Hence, ω is CP1(M)-extremal. (cid:3) We saw in Theorem 7.5 that P r(M)-extremals have the unique extension prop- erty relative to P r(M). We will see in Example 6 below that the converse is false. Hence, the previous lemma does not extend to the class P r(M) for r > 1. The following is an easy consequence of Corollary 6.5. Corollary 7.8. Let M be an operator space and let r ≥ 1. Let ω ∈ P r(M,Hω) be P r(M)-extremal. Let Y ∈ B(Hω) be an invertible operator such that kY k = kY −1k ≤ r1/2 and let ωY : M → B(Hω) be defined as ωY (a) = Y ω(a)Y −1, a ∈ M. If the map ωY belongs to P 1(M,Hω), then it is P 1(M)-extremal. Proof. Simply combine Corollary 6.5, Theorem 7.6 and Lemma 7.7. (cid:3) Theorem 7.6 along with Lemma 7.7 shows that any P r(M)-extremal is similar to a CP1(M)-extremal, and it is natural to wonder whether the converse holds. Our next task is to show that this is not the case. First, we need the following It shows that if ω is P r(M)-extremal, result which is of independent interest. then kωkcb cannot be too small, and ω does not belong to P s(M) for any s < r. This fact is not obvious from the definition of extremality. Theorem 7.9. Let M be an operator space, let r ≥ 1 and let ω ∈ P r(M,Hω). Assume that ω is a non-zero P r(M)-extremal element. Then, the following state- ments hold. (1) The map ω does not belong to P s(M,Hω) if 1 ≤ s < r. In particular, the unique homomorphism Ω ∈ P r(AM,Hω) that agrees with ω on M satisfies kΥ(Ω)kcb = r. (2) The map ω satisfies kωkcb ≥ r 16 − 5 4 . Proof. (1) Let 1 ≤ s < r and suppose on the contrary that ω ∈ P s(M,Hω). Then, there is a Hilbert space Kω containing Hω along with a homomorphism θ : AM → B(Kω) that has Paulsen's similarity property with constant s1/2 and such that Let ε > 0 and define a linear map ρε : AM → B(K(4) ω(a) = PHω θ(a)Hω , a ∈ M. ω ) as ρε(a) = θ(a) 0 εθ(a) 0 0 0 θ(a) 0 εθ(a) 0 0 0 0 0 0 0  , a ∈ AM. It is readily verified that ρε is a homomorphism with kρεkcb ≤ (1 + 2ε)kθkcb. In addition, if we let be defined as Θ : AM → B(K(4) ω ) Θ(a) = θ(a) ⊕ θ(a) ⊕ 0 ⊕ 0, a ∈ AM A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 33 then lim ε→0kΥ(ρε) − Υ(Θ)kcb = 0. Since θ has Paulsen's similarity property with constant s1/2, we may use Corollary 5.5 along with Theorem 5.4 to conclude that kΥ(Θ)kcb ≤ s < r. Thus, there is ε0 > 0 small enough so that kΥ(ρε0)kcb < r. Another application of Theorem 5.4 yields that ρε0 has Paulsen's similarity property with constant r1/2. Note that Hω can be identified with in which case Hω ⊕ 0 ⊕ 0 ⊕ 0 ⊂ K(4) ω , ω(a) = PHω ρε0 (a)Hω , a ∈ M. But Hω is not reducing for ρε0 (M), since ω is non-zero. This contradicts the fact that ω is P r(M)-extremal. We conclude that ω ∈ P r(M,Hω)\∪s<r P r(M,Hω). Since ω is P r(M)-extremal we know from Theorem 7.5 that there is a unique homomorphism Ω : AM → B(Hω) that agrees with ω on M and that has Paulsen's similarity property with constant r1/2, but not with constant s1/2 for any s < r. Using Theorem 5.4, we conclude that kΥ(Ω)kcb = r. as (2) Let ε > 0 and define a completely bounded linear map δ : M → B(Hω ⊕Hω) r < 4(5 + 4p1 + ε2kωkcb). Letting ε → 0 and rearranging yields the announced inequality. (cid:3) Using the previous theorem, we show that a CP1(M)-extremal element, let alone a map merely similar to one, is not necessarily a P r(M)-extremal if r is allowed to be arbitrarily large. δ(a) =(cid:20) ω(a) εω(a) 0(cid:21) , 0 a ∈ M. Then, we have kδkcb = √1 + ε2kωkcb. By Theorem 5.1, there is a Hilbert space Kω containing Hω ⊕ Hω and a homomorphism θ : C∗(M) → B(Kω) such that δ(a) = PHω⊕Hω θ(a)Hω⊕Hω , a ∈ M. Moreover, we can assume that kθkcb ≤ C where C =(cid:18)(cid:16)p1 + ε2kωkcb(cid:17)1/2 +(cid:16)1 +p1 + ε2kωkcb(cid:17)1/2(cid:19)2 . By Theorem 5.2, we see that θ has Paulsen's similarity property with constant √C + √1 + C. Thus, δ ∈ P (√C+√1+C)2(M,Hω ⊕ Hω). On the other hand, note that Hω can be identified as the first coordinate in Hω⊕Hω, in which case ω(a) = PHω δ(a)Hω , a ∈ M. The subspace Hω is clearly not reducing for δ(M), since ω is non-zero. However, ω is assumed to be P r(M)-extremal, so we must have that (√C + √1 + C)2 > r. Now, C ≤ 4(1 + √1 + ε2kωkcb) so that 34 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY Example 6. Let ω : B(H) → B(H) be the identity map. Trivially, we see that ω has the unique extension property relative to P r(B(H)) for every r ≥ 1, and also relative to CP1(B(H)). In particular, we see that ω is CP1(B(H))-extremal by Lemma 7.7. On the other hand, by virtue of part (2) of Theorem 7.9, we see that ω is not P r(M)-extremal as soon as r 5 16 − 4 > 1 since kωkcb = 1. The previous example is a bit artificial. The following is a more satisfying ex- ample that does not rely on forcing r to be large so that we can apply Theorem 7.9. We exhibit a linear map ω with kωkcb = r and with the property that there is an invertible operator X such that kXk = kX−1k = r1/2 and such that the map a 7→ Xω(a)X−1, a ∈ M is CP1(M)-extremal, yet ω itself is not P r(M)-extremal. The basic idea is to append a direct summand to create room. Example 7. Let H be a Hilbert space and let ε > 0. Let Hδ = H ⊕ H and define a linear map δ : B(H) → B(Hδ) as δ(a) =(cid:20) a εa 0(cid:21) , 0 a ∈ B(H). We see that kδkcb = √1 + ε2. Using Theorem 5.1 and arguing as in the proof of part (2) of Theorem 7.9, we infer that there is r ≥ 1 large enough so that δ ∈ P r(B(H),Hδ). Compressing to the copy of H corresponding to the first coordinate in Hδ, we observe also that a = PHδ(a)H, a ∈ B(H). Next, fix an invertible operator X ∈ B(H) such that kXk = kX−1k = r1/2 ≥ 1. Let ω : B(H) → B(H ⊕ H) be the homomorphism defined as ω(a) = XaX−1 ⊕ a, a ∈ B(H). Then, kωkcb = r and ω ∈ P r(B(H),H⊕H). Consider the linear map ∆ : B(H) → B(H ⊕ Hδ) defined as ∆(a) = XaX−1 ⊕ δ(a), a ∈ B(H). By Corollary 5.5, we have that ∆ ∈ P r(B(H),H⊕Hδ) and clearly ω ≺ ∆. Since the copy of H corresponding to the first coordinate in Hδ is not reducing for δ(B(H)), the space H⊕H ⊂ H⊕Hδ is not reducing for ∆(B(H)). Thus, ω is not P r(B(H))- extremal. On the other hand, let Y ∈ B(H ⊕ H) be the invertible operator defined as Y = X⊕I. Then, kY k = kY −1k ≤ r1/2. Consider the map ωY : B(H) → B(H⊕H) defined as ωY (a) = Y −1ω(a)Y = a ⊕ a, a ∈ B(H). Clearly, ωY has the unique extension property relative to CP1(B(H)). Hence, ωY is CP1(B(H))-extremal by Lemma 7.7. Thus, ω is similar to a CP1(B(H))-extremal element, yet it is not P r(B(H))-extremal itself. A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 35 Finally, we exhibit an example that shows that given a linear map ϕ, in general it is not sufficient that kΥ(ϕ)kcb ≤ r to force ϕ to lie in P r(M). This was mentioned in passing at the beginning of Section 6. Example 8. We return to the map from Example 2. Let H = C, let M = B(H) and ϕ : M → B(H) be defined as ϕ(λ) = 2λ for each λ ∈ M. Assume that ϕ ∈ P 3(M, C). Then, by Theorem 7.3, there is a P 3(M)-extremal element ω ∈ P 3(M,Hω) with ϕ ≺ ω. Since M is a C∗-algebra, we see from Theorem 7.5 that ω(1) is an idempotent. Moreover, we have that 2 = ϕ(1) = PHω(1)H which is easily seen to force kω(1)k > 2. Next, we note that kΥ(ω)kcb = 3 by Theorem 7.9. Since M is unital, we have that Υ(M) = M ⊕ C and 2kω(1)k − 1 ≤ k2ω(1) − IHωk = kΥ(ω)(1 ⊕ −1)k ≤ kΥ(ω)kcb = 3 whence kω(1)k ≤ 2, which is a contradiction. Thus, ϕ /∈ P 3(M, C). On the other hand, a straightforward verification shows that Υ(ϕ)(λ ⊕ µ) = 2λ − µ, λ ∈ M, µ ∈ C. Hence, kΥ(ϕ)kcb = kΥ(ϕ)k ≤ 3. 8. A scale of C∗-envelopes for representations of operator spaces In this section, we define a scale of C∗-envelopes associated to operator spaces. Traditionally, one would expect such envelopes to depend only on the completely isometric isomorphism class of the operator space, and not on the choice of rep- resentation, as is the case in [26],[9],[23]. However, since our present setting is that of possibly non-unital operator spaces, to obtain C∗-envelopes we will need to consider an operator space along with an associated completely isometric represen- tation on some Hilbert space. We emphasize that unlike ours, the usual construc- tions ([26],[9],[23]) do not require this additional piece of data, and although their resulting envelopes are not algebras, they exhibit stronger universality properties. Let M be an operator space and put d(M) = (1 + ℵ0 dimM)(dim M)2. For each cardinal number n ≤ d(M) we fix a Hilbert space Hn of dimension n, and implicitly identify every other such space with Hn. This is a standard procedure to avoid set theoretic difficulties, and this convention will be used tacitly henceforth. Next, let r ≥ 1 and let µ : M → B(Hµ) be a completely isometric linear map. For a cardinal number c we let Er(µ, c) be the subset of elements ω ∈ P r(µ(M), Hc) that are P r(µ(M))-extremal. Given ω ∈ Er(µ, c), we denote by Ir(ω) the collection of invertible operator X ∈ B(Hc) such that kXk = kX−1k ≤ r1/2 and with the property that the map ωX : µ(M) → B(Hc) defined as ωX (a) = Xω(a)X−1, a ∈ µ(M) belongs to P 1(µ(M), Hc). We know from Theorem 7.6 that Ir(ω) is non-empty, and that for each X ∈ Ir(ω) there is a unique ∗-homomorphism πω,X : C∗(µ(M)) → B(Hc) such that πω,X (a) = Xω(a)X−1, a ∈ µ(M). 36 We let RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY Hr,µ = Mc≤d(M) Mω∈Er(µ,c) MX∈Ir(ω) Hc and we define a ∗-homomorphism εµ,r : C∗(µ(M)) → B(Hr,µ) as εµ,r(a) = Mc≤d(M) Mω∈Er(µ,c) MX∈Ir(ω) πω,X (a), a ∈ C∗(µ(M)). We verify that εµ,r(M) can be identified with µ(M) in a meaningful way. Corollary 8.1. Let M be an operator space, let µ : M → B(Hµ) be a completely isometric linear map and let r ≥ 1. Then, r−1kak ≤ kε(n) for every n ∈ N and every a ∈ Mn(µ(M)). Proof. It is clear that εµ,r is completely contractive as it is a ∗-homomorphism. Furthermore, we may use Corollary 7.4 to obtain a cardinal number c such that c ≤ d(M) and an element ω ∈ P r(µ(M), Hc) that is P r(µ(M))-extremal and such that for every n ∈ N and every a ∈ Mn(µ(M)) we have kak ≤ kω(n)(a)k. If X ∈ Ir(ω), then µ,r(a)k ≤ kak kε(n) µ,r(a)k ≥ kX (n)ω(n)(a)X−1(n)k ≥ r−1kak for every n ∈ N and every a ∈ Mn(µ(M)). For r ≥ 1, we define the C∗r -envelope of the pair (M, µ) as C∗e,r(M, µ) = εµ,r(C∗(µ(M))). If we let Jµ,r = ker εµ,r, then we see that C∗(µ(M))/Jµ,r ∼= C∗e,r(M, µ). particular, if C∗(µ(M)) is simple, then the surjective ∗-homomorphism εµ,r : C∗(µ(M)) → C∗e,r(M, µ) is necessarily injective, so that C∗e,r(M, µ) and C∗(µ(M)) are ∗-isomorphic for every r ≥ 1. In addition, if µ(M) is a C∗-algebra to begin with, then in fact Jµ,r is trivial and µ(M) can be identified with the C∗r -envelope of (M, µ), as we show next. Corollary 8.2. Let M be an operator space, let µ : M → B(Hµ) be a completely isometric linear map and let r ≥ 1. If µ(M) is C∗-algebra, then C∗e,r(M, µ) is ∗-isomorphic to C∗(µ(M)). Proof. By Corollary 8.1, we see that the surjective ∗-homomorphism εµ,r : C∗(µ(M)) → C∗e,r(M, µ) is injective. Going back to the more interesting general case where µ(M) is merely an opera- tor space, we wish to establish that C∗e,r(M, µ) does not depend on the isomorphism class of M in an essential way. What kind of map should be considered an isomor- phism in this context is an interesting question. A natural answer would be that an isomorphism should be an invertible map that preserves the class P r. Accord- ingly, we call a bijective linear map τ between two concretely represented operator spaces M and N a P-isomorphism if τ and τ−1 both lie in the class P 1. If τ is a P-isomorphism then for every r ≥ 1 we see by virtue of Corollary 6.6 that (cid:3) In (cid:3) A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 37 ϕ ◦ τ−1 ∈ P r(N ,Hϕ) and ψ ◦ τ ∈ P r(M,Hψ) whenever ϕ ∈ P r(M,Hϕ) and ψ ∈ P r(N ,Hψ). Let us exhibit two concretes instances of P-isomorphisms. First, by Corollary 5.3, we see that a bijective map τ : M → N is a P-isomorphism if it is assumed to and AN denote the operator algebras generated by M and N respectively. Second, if M and N are both unital and τ : M → N is a unital completely isometric linear isomorphism, then τ is necessarily a P-isomorphism by Corollary 6.5. extend to a completely isometric algebra isomorphismbτ : AM → AN , where AM Moreover, we have the following. Lemma 8.3. Let M and N be operator spaces and let µ : M → B(Hµ), ν : N → B(Hν ) be completely isometric linear maps. Let r ≥ 1 and let c be a cardinal number. Let τ : µ(M) → ν(N ) be a P-isomorphism and let ω ∈ Er(ν, c). Then ω ◦ τ ∈ Er(µ, c) and Ir(ω) = Ir(ω ◦ τ ). Proof. Let δ ∈ P r(µ(M),Hδ) such that ω ◦ τ ≺ δ. Then, we have that δ ◦ τ−1 ∈ P r(ν(N ),Hδ) and ω ≺ δ◦τ−1. Hence, the space Hc is reducing for δ◦τ−1(ν(N )) = δ(µ(M)). This shows that ω ◦ τ is P r(µ(M))-extremal. Next, let X ∈ I(ω), so that ωX ∈ P 1(ν(N ), Hc). Since τ is a P-isomorphism, we also have ωX ◦ τ ∈ P 1(µ(M), Hc). Hence, X ∈ Ir(ω◦ τ ). The same argument can be used to establish the reverse inclusion, by symmetry. (cid:3) We now prove that the C∗r -envelope is invariant under P-isomorphisms. Theorem 8.4. Let M and N be operator spaces and let ν : N → B(Hν ) µ : M → B(Hµ), be completely isometric linear maps. Let r ≥ 1 and let τ : µ(M) → ν(N ) be a P-isomorphism. Then, the maps εµ,r ◦ τ−1 and εν,r are unitarily equivalent on ν(N ). In particular, C∗e,r(M, µ) and C∗e,r(N , ν) are unitarily equivalent. Proof. We first note that d(M) = d(N ). By Lemma 8.3, we see that ζ ∈ Er(ν, c) if and only if ζ = ω ◦ τ−1 for some ω ∈ Er(µ, c). Moreover, we have that Ir(ω) = Ir(ω ◦ τ−1) for every ω ∈ Er(µ, c). For b ∈ ν(N ) we see that εµ,r ◦ τ−1(b) = Mc≤d(M) Mω∈Er(µ,c) MX∈Ir(ω) X(ω ◦ τ−1(b))X−1 which, after a unitary permutation of the coordinates of Hr,µ, becomes Mc≤d(M) Mζ∈Er(ν,c) MX∈Ir(ζ) εµ,r ◦ τ−1(b) = U εν,r(b)U∗, Hence, there is a unitary operator U : Hr,ν → Hr,µ such that b ∈ ν(N ). Xζ(b)X−1 = εν,r(b). Consequently, C∗e,r(M, µ) = C∗ (εµ,r(µ(M))) = C∗(cid:0)εµ,r ◦ τ−1(ν(N ))(cid:1) = U C∗ (εν,r(ν(N ))) U∗ = U C∗e,r(N , ν)U∗. (cid:3) 38 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY Now, the reader may wonder exactly how the construction of the C∗r -envelope depends on the completely isometric representation µ. More precisely, if µ1 : M → B(H1), µ2 : M → B(H2) are completely isometric linear maps, it is not clear at the moment what relation exists, if any, between C∗e,r(M, µ1) and C∗e,r(M, µ2). Indeed, Theorem 8.4 only guarantees invariance of the envelope under the assumption that µ2 ◦ µ−1 is a P- isomorphism. The following example shows that in general the C∗r -envelopes can differ. We are grateful to Ken Davidson for suggesting the idea therein. Example 9. Let M = C and let r ≥ 1. Let µ1 : M → C be the identity map and let 1 µ2 : M → M2(C) be the completely isometric linear map defined as µ2(λ) =(cid:20)0 λ 0(cid:21) , 0 λ ∈ C. Now, µ1(M) = C is a C∗-algebra, whence C∗e,r(M, µ1) ∼= C by Corollary 8.2 . On the other hand, it is easily verified that C∗(µ2(M)) = M2(C) and so it is simple. In that case, we saw previously that C∗e,r(M, µ2) ∼= C∗(µ2(M)) = M2(C). Thus, C∗e,r(M, µ2) is not ∗-isomorphic to C∗e,r(M, µ1). Nevertheless, we may use Theorem 8.4 to establish the following universality property, which justifies the terminology "envelope". Corollary 8.5. Let M and N be operator spaces and let ν : N → B(Hν ) µ : M → B(Hµ), be completely isometric linear maps. Let r ≥ 1 and let τ : µ(M) → ν(N ) be a P-isomorphism. Then, there is a surjective ∗-homomorphism ρ : C∗(ν(N )) → C∗e,r(M, µ) such that ρ ◦ τ = εµ,r on µ(M). Proof. By Theorem 8.4, there is a unitary operator U : Hr,ν → Hr,µ such that U C∗e,r(N , ν)U∗ = C∗e,r(M, µ) and Define ρ : C∗(ν(N )) → C∗e,r(M, µ) as εµ,r ◦ τ−1(b) = U εν,r(b)U∗, b ∈ ν(N ). ρ(b) = U εν,r(b)U∗, b ∈ C∗(ν(N )). Then, ρ is a surjective ∗-homomorphism and ρ ◦ τ = εµ,r on µ(M). (cid:3) In light of the preceding developments, it seems relevant here to point out that the embedding εµ,r : µ(M) → εµ,r(µ(M)) is not known to be a P-isomorphism. We will revisit this issue below in Theorem 8.10. Next, we wish to relate the C∗-algebra C∗e,1(M, µ) and the usual (unital) C∗- envelope of µ(M). For that purpose, we introduce some terminology. Let M be a unital operator space and let H be a Hilbert space. We denote by UCB1(M,H) A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 39 the set of unital completely contractive linear maps ϕ : M → B(H). Let K be a Hilbert space containing H and let δ ∈ UCB1(M,K). If δ satisfies ϕ(a) = PHδ(a)H, a ∈ M then we write ϕ ≺ δ. An element ω ∈ UCB1(M,Hω) is said to be UCB1(M)- extremal if whenever δ ∈ UCB1(M,Hδ) satisfies ω ≺ δ then we must have that Hω is reducing for δ(M). We now relate two kinds of extremality. Lemma 8.6. Let M be an operator space and let ω ∈ CP1(M,Hω). Then ω is CP1(M)-extremal if and only if Υ(ω) is UCB1(Υ(M))-extremal. Proof. Assume that Υ(ω) is UCB1(Υ(M))-extremal. Let δ ∈ CP1(M,Hδ) such that ω ≺ δ. Then, Υ(δ) ∈ UCB1(Υ(M),Hδ) and it is easily verified that Υ(ω) ≺ Υ(δ). We conclude that Hω is reducing for Υ(δ)(Υ(M)), and in particular for δ(M). Hence ω is CP1(M)-extremal. Conversely, assume that ω is CP1(M)-extremal. Let ∆ ∈ UCB1(Υ(M),H∆) such that Υ(ω) ≺ ∆. Recall now that the map υ : M → Υ(M) is the restriction of a ∗-homomorphism, so that ∆ ◦ υ ∈ CP1(M,H∆). We note that ω ≺ ∆ ◦ υ. We conclude that Hω is reducing for ∆(υ(M)). Since ∆ is unital, in fact Hω must be reducing for ∆(Υ(M)), so that Υ(ω) is UCB1(M)-extremal. (cid:3) We now recall the construction of the standard unital C∗-envelope for a (con- crete) unital operator space M. If c is a cardinal number, then we let B(M, c) de- note the collection of elements β ∈ UCB1(M, Hc) that are UCB1(M)-extremal. It is well-known that every β ∈ B(M, Hc) extends to a unique unital ∗-homomorphism πβ : C∗(M) → B(Hc). Then, the unital C∗-envelope of M can be defined as where κ is the unital ∗-homomorphism defined as U C∗e (M) = κ(C∗(M)) κ = Mc≤d(M) Mβ∈B(M,c) πβ that is completely isometric on M. The reader should consult [4],[22],[3],[18] for more details. Note that the usual name for U C∗e (M) is simply the C∗-envelope and the standard notation is C∗e (M), but in the context of this paper we want to emphasize the fact that the maps involved in the previous construction are all unital. We can now relate the unital C∗-envelope to the C∗1 -envelope. Theorem 8.7. Let M be an operator space and let µ : M → B(Hµ) be a completely isometric linear map. Then, C∗e,1(M, µ)+CIH1,µ is ∗-isomorphic to U C∗e (Υ(µ(M))). Proof. By Subsection 2.2, Corollary 6.5 and Lemma 8.6, we see that the UCB1(Υ(µ(M)))- extremal elements are precisely those of the form Υ(ω) for some P 1(µ(M))-extremal element ω. Hence, we find that is unitarily equivalent to πβ κ = Mc≤d(Υ(µ(M))) Mβ∈B(Υ(µ(M)),c) Mc≤d(Υ(µ(M))) Mω∈E1(µ,c) πΥ(ω). 40 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY Since ℵ0 dimM = ℵ0 dim Υ(µ(M)) we conclude that d(Υ(µ(M))) = d(M) and thus U C∗e (Υ(µ(M))) = C∗ Mc≤d(Υ(µ(M))) Mβ∈B(Υ(µ(M)),c) is unitarily equivalent to β(b) : b ∈ Υ(µ(M))  Since I1(ω) consists of unitary operators for every ω ∈ E1(µ, c), we see that U C∗e (Υ(µ(M))) is ∗-isomorphic to C∗ Mc≤d(M) Mω∈E1(µ,c) C∗ Mc≤d(M) Mω∈E1(µ,c) MV ∈I1(ω) = CIH1,µ + C∗ Mc≤d(M) Mω∈E1(µ,c) MV ∈I1(ω) = CIH1,µ + C∗ Mc≤d(M) Mω∈E1(µ,c) MV ∈I1(ω) Υ(ω)(b) : b ∈ Υ(µ(M))  . V Υ(ω)(b)V ∗ : b ∈ Υ(µ(M))  V ω(a)V ∗ : a ∈ µ(M)  πω,V (a) : a ∈ µ(M)  = CIH1,µ + C∗e,1(M, µ). (cid:3) The outstanding issue that remains to be addressed is the relationship, for a given µ, between the C∗r -envelope and the C∗s -envelope for s ≥ r. We elucidate it partially with the following. Theorem 8.8. Let M be an operator space, let µ : M → B(Hµ) be a completely isometric linear map and let s ≥ r ≥ 1. Then, εµ,r(a) = Yrγs,r(Y −1 a ∈ µ(M) where Yr ∈ B(Hr,µ) and Ys ∈ B(Hs,µ) are invertible operators with s εµ,s(a)Ys)Y −1 , r kYrk = kY −1 r k ≤ r1/2, and γs,r is a completely contractive map. Proof. For notational convenience, we define ιµ,r : µ(M) → B(Hr,µ) as kYsk = kY −1 s k ≤ s1/2 ιµ,r(a) = Mc≤d(M) Mω∈Er(µ,c) MX∈Ir(ω) ω(a), a ∈ µ(M). Let c ≤ d(M) and let ω ∈ Er(µ, c). Then ω ∈ P s(µ(M), Hc). By virtue of Theorem 7.3, there is a Hilbert space Kω containing Hc with dimKω ≤ (1 + ℵ0 dim M)c ≤ (1 + ℵ0 dimM)d(M) = d(M) along with a linear map ζω ∈ P s(µ(M),Kω) that is P s(µ(M))-extremal and such that ω(a) = PHcζω(a)Hc , a ∈ µ(M). A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 41 There is a cardinal number cω ≤ d(M) and a unitary operator Uω : Kω → Hcω such that ζ′ω ∈ Es(µ, Hcω ), where ζ′ω(a) = Uωζω(a)U∗ω, a ∈ µ(M). We note that ∪c≤d(M){ζ′ω : ω ∈ Er(µ, c)} ⊂ ∪c≤d(M)Es(µ, c). We can then define a completely contractive surjective linear map PHc U∗ωζ′ω(a)UωHc as γs,r Mc≤d(M) Mζ∈Es(µ,c) MX∈Is(ζ) for every a ∈ µ(M). Note then that γs,r ◦ ιµ,s = ιµ,r on µ(M). Put γs,r : ιµ,s(µ(M)) → ιµ,r(µ(M)) ζ(a) = Mc≤d(M) Mω∈Er(µ,c) MX∈Ir(ω) Yr = Mc≤d(M) Mω∈Er(µ,c) MX∈Ir(ω) Ys = Mc≤d(M) Mζ∈Es(µ,c) MX∈Is(ζ) kYsk = kY −1 r k ≤ r1/2, X X s k ≤ s1/2. which satisfy kYrk = kY −1 For a ∈ µ(M) we have Yrιµ,r(a)Y −1 r = εµ,r(a), Ysιµ,s(a)Y −1 s = εµ,s(a) whence εµ,r(a) = Yrιµ,r(a)Y −1 r = Yrγs,r(ιµ,s(a))Y −1 r = Yrγs,r(Y −1 s εµ,s(a)Ys)Y −1 r . (cid:3) The relationship between the C∗1 -envelope and the C∗r -envelope is clearer, as the next result shows. Theorem 8.9. Let M be an operator space, let µ : M → B(Hµ) be a completely isometric linear map and let r ≥ 1. Then, there is a surjective ∗-homomorphism Γr : C∗e,1(M, µ) → C∗e,r(M, µ) such that Γr ◦ εµ,1 = εµ,r. Proof. Let c be a cardinal number satisfying c ≤ d(M), let ω ∈ Er(µ, c) and let X ∈ Ir(ω). Then, we have ωX ∈ E1(µ, c) by Corollary 7.8. Moreover, we note that if ζ ∈ E1(µ, c), then I1(ζ) coincides with the set of unitary operators on Hc, so in particular it contains the identity I. Thus, we see that there is an element ζ(ω, X) ∈ E1(µ, c) such that πζ(ω,X),I = πω,X . Put Π = {ζ(ω, X) : ω ∈ Er(µ, c), X ∈ Ir(ω)} ⊂ E1(µ, c). We define a map γr on C∗e,1(M, µ) to be the corresponding compression, so that γr Mc≤d(M) Mζ∈E1(µ,c) MU∈I1(ζ) πζ,U (a) = Mc≤d(M)Mζ∈Π πζ,I (a) 42 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY for every a ∈ C∗(µ(M)). Clearly, γr is a ∗-homomorphism. By choice of Π, there is a unitary operator V such that V ∗γr Mc≤d(M) Mζ∈E1(µ,c) MU∈I1(ζ) πζ,U (a) V = Mc≤d(M) Mω∈Er(µ,c) MX∈Ir(ω) for every a ∈ C∗(µ(M)). Define Γr : C∗e,1(M, µ) → C∗e,r(M, µ) as πω,X (a) Γr(b) = V ∗γr(b)V, b ∈ C∗e,1(M, µ). Then, we see that Γr is a ∗-homomorphism and that Γr ◦ εµ,1 = εµ,r. In particular Γr is surjective. (cid:3) The question arises whether there is a ∗-isomorphism π : C∗e,r(M, µ) → C∗e,1(M, µ) satisfying π ◦ εµ,r = εµ,1 on µ(M). It is not immediately clear what the answer should be, especially in view of Examples 6 and 7. Nevertheless, we suspect that this ∗-isomorphism does exist, although we cannot prove it in general. We can however establish the following related fact. Theorem 8.10. The following statements hold. (1) Let M be an operator space, let µ : M → B(Hµ) be a completely isometric linear map and let r ≥ 1. Assume that the map εµ,r : µ(M) → εµ,r(µ(M)) is a P-isomorphism. Then, there is a ∗-isomorphism π : C∗e,r(M, µ) → C∗e,1(M, µ) satisfying π ◦ εµ,r = εµ,1 on µ(M). (2) Let A be an operator algebra, let α : A → B(Hα) be a completely isometric homomorphism and let r ≥ 1. Assume that there is a ∗-isomorphism π : C∗e,r(A, α) → C∗e,1(A, α) satisfying π ◦ εα,r = εα,1 on α(A). Then, the map εα,r : α(A) → εα,r(α(A)) is a P-isomorphism. Proof. (1) Recall that C∗e,r(M, µ) = C∗(εµ,r(µ(M))). Since εµ,r : µ(M) → εµ,r(µ(M)) is a P-isomorphism, we may apply Corollary 8.5 to obtain a surjective ∗-homomorphism ρ : C∗e,r(M, µ) → C∗e,1(M, µ) such that ρ ◦ εµ,r = εµ,1 on µ(M). On the other hand, by Theorem 8.9 there is a surjective ∗-homomorphism Γr : C∗e,1(M, µ) → C∗e,r(M, µ) such that Γr ◦ εµ,1 = εµ,r on µ(M). We now see that (Γr ◦ ρ)(εµ,r(a)) = εµ,r(a), a ∈ µ(M). We conclude that Hence, ρ is injective as desired. Γr ◦ ρ(a) = a, a ∈ C∗e,r(M, µ). (2) By Corollary 8.1, we see that εα,1 : α(A) → εα,1(α(A)) is a completely isometric algebra isomorphism, and thus εα,1 is a P-isomorphism. On the other hand, we see that εα,r = π−1 ◦ εα,1 on α(A). Since π is a ∗-isomorphism, we infer that εα,r : α(A) → εα,r(α(A)) is a P-isomorphism as well. (cid:3) Thus, for an operator algebra A and a completely isometric homomorphism α on it, whether the map εα,r : α(A) → εα,r(α(A)) is a P-isomorphism is equivalent to the existence of a ∗-isomorphism π : C∗e,r(A, α) → C∗e,1(A, α) satisfying some additional natural condition. In general, we make the following conjecture. A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 43 Conjecture. Let M be an operator space, let µ : M → B(Hµ) be a com- pletely isometric linear map and let r ≥ 1. Then, the C∗-algebras C∗e,1(M, µ) and C∗e,r(M, µ) are ∗-isomorphic. We close the paper by verifying this conjecture in some cases of interest. We already observed that if C∗(µ(M)) is simple, then C∗e,r(M, µ) and C∗(µ(M)) are ∗-isomorphic for every r ≥ 1. On the other hand, if we assume that C∗(µ(M)) contains the ideal K of compact operators on Hµ, then we can also achieve some partial success. The following can be viewed as a variation of an important fact called Arveson's boundary theorem [5]. Theorem 8.11. Let M be an operator space and let µ : M → B(Hµ) be a com- pletely isometric linear map such that C∗(µ(M)) contains the ideal K of compact operators on Hµ. Let q : C∗(µ(M)) → C∗(µ(M))/K denote the quotient map. Let r ≥ 1 and assume that there is an integer n ∈ N and an element a ∈ Mn(µ(M)) such that kq(n)(a)k < r−1kak. Then, the map εµ,r : C∗(µ(M)) → C∗e,r(M, µ) is a ∗-isomorphism. Proof. Basic representation theory for C∗-algebras [21] stipulates that the ∗-homo- morphism is unitarily equivalent to εµ,r : C∗(µ(M)) → B(Hr,µ) id(c) ⊕ σ ◦ q for some cardinal number c and some ∗-homomorphism σ on C∗(µ(M))/K. By assumption, we see that k(σ ◦ q)(n)(a)k < r−1kak. In view of Corollary 8.1, we conclude that c 6= 0. In particular εµ,r is injective and thus εµ,r : C∗(µ(M)) → C∗e,r(M, µ) (cid:3) is a ∗-isomorphism. Many important classical examples of operator spaces fit into the framework of the previous theorem, such as the higher-dimensional Toeplitz algebra of multivari- ate operator theory. Example 10. Fix a positive integer d ≥ 2 and let Bd ⊂ Cd denote the open unit ball. The Drury-Arveson space H 2 d is the reproducing kernel Hilbert space on Bd with reproducing kernel given by the formula k(z, w) = 1 1 − hz, wiCd , z, w ∈ Bd. A function ϕ : Bd → C is a multiplier for H 2 This is a Hilbert space of holomorphic functions on Bd, and it is a very natural higher-dimensional analogue of the classical Hardy space on the unit disc. It can be identified with the symmetric Fock space over Cd [6], [19]. d if ϕf ∈ H 2 d for every f ∈ H 2 d . Examples of such functions include the holomorphic polynomials in d variables. Now, every multiplier ϕ gives rise to a bounded linear multiplication operator Mϕ ∈ B(H 2 d ), and the identification ϕ 7→ Mϕ allows us to view the multiplier algebra as an operator algebra on H 2 d . The book [2] is an excellent reference on these topics. Let Ad ⊂ B(H 2 d ) denote the norm closure of the polynomial multipliers. This algebra is of tremendous importance in multivariate operator theory [6],[36],[37], [14] and is also the target of intense research in function theory [16],[20],[15]. One 44 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY of the distinguishing features of Ad is that it is not a uniform algebra, in the sense that the norm of a multiplier does not coincide with its supremum norm over Bd. In fact, something more precise is known to hold. The so-called Toeplitz algebra Td = C∗(Ad) contains the ideal K of compact operators on H 2 d , and the quotient Td/K is ∗-isomorphic to the C∗-algebra of continuous functions on the unit sphere [6, Theorem 5.7]. Moreover, since d ≥ 2 the quotient map q : Td → Td/K is not bounded below on Ad (see [6, Theorem 3.3] or [19, Theorem 2.4]). In particular, we see that C∗e,r(Ad, id) is ∗-isomorphic to Td for every r ≥ 1, by Theorem 8.11. Crucial to the preceding example was the fact that quotient map q : Td → Td/K is not bounded below on Ad, which fails when d = 1. The algebra A1 is simply the usual disc algebra consisting of the holomorphic functions on the open unit disc which extend to be continuous on the closed disc. The identification of the C∗r -envelope of A1 requires some classical uniform algebra machinery. In fact, we obtain the following more general result. Theorem 8.12. Let X be a compact metrizable space and let A ⊂ C(X) be a closed unital subalgebra which separates the points of X. Let ΣA ⊂ X denote the Shilov boundary of A. Let α : A → B(Hα) be a completely isometric algebra homomorphism. Then, for each r ≥ 1 the C∗-algebras C∗e,r(A, α) and C(ΣA) are ∗-isomorphic. Proof. Let σ : A → C(ΣA) be the completely isometric algebra homomorphism given by restriction. Since completely isometric algebra isomorphisms are P- isomorphisms, by virtue of Theorem 8.4 we see that it suffices to show that C∗e,r(A, σ) and C(ΣA) are ∗-isomorphic. It follows from the Stone-Weierstrass theorem that C∗(σ(A)) = C(ΣA). Now, we note that εσ,r(1) is a self-adjoint projection, so that there is a unital commutative C∗-algebra A such that C∗e,r(A, σ) = εσ,r(C(ΣA)) = A ⊕ {0}. The maximal ideal space of A can be identified with a compact subset Y ⊂ ΣA. If we denote by g the Gelfand transform of A, then we know that g : A → C(Y ) is a ∗-isomorphism by the Gelfand-Naimark theorem. Moreover, note that g ◦ εσ,r(f ) = fY , f ∈ C(ΣA). Let ξ ∈ ΣA be a peak point for A, so that there is f ∈ A such that f (ξ) = 1 and f (x) < 1 if x ∈ ΣA \ {ξ}. Assume that ξ /∈ Y . Since Y is compact, there is a natural number n large enough so that kf nY kC(Y ) < 1/r. Then, we see that kg ◦ εσ,r(f n)k < 1/r which implies that kεσ,r(f n)k < 1/r. But since kf nk = 1 and f n ∈ A, this contradicts Corollary 8.1. We thus conclude that Y contains all peak points for A. Since Y is closed, this implies that Y = ΣA (see the discussion following [24, Theorem 11.6] or [35, Chapter 8] for instance). Thus A is ∗-isomorphic to C(ΣA). In particular, we see that C∗e,r(A, σ) is ∗-isomorphic to C(ΣA). (cid:3) References [1] Jim Agler, An abstract approach to model theory, Surveys of some recent results in operator theory, Vol. II, 1988, pp. 1 -- 23. MR976842 (90a:47016) [2] Jim Agler and John E. McCarthy, Pick interpolation and Hilbert function spaces, Gradu- ate Studies in Mathematics, vol. 44, American Mathematical Society, Providence, RI, 2002. MR1882259 (2003b:47001) A COMPLETELY BOUNDED NON-COMMUTATIVE CHOQUET BOUNDARY 45 [3] William Arveson, Notes on the unique extension property. unpublished notes available at https://math.berkeley.edu/∼arveson/Dvi/unExt.pdf. [4] [5] [6] [7] [8] , Subalgebras of C ∗-algebras, Acta Math. 123 (1969), 141 -- 224. MR0253059 (40 #6274) , Subalgebras of C ∗-algebras. II, Acta Math. 128 (1972), no. 3-4, 271 -- 308. MR0394232 (52 #15035) , Subalgebras of C ∗-algebras. III. Multivariable operator theory, Acta Math. 181 (1998), no. 2, 159 -- 228. MR1668582 (2000e:47013) , The noncommutative Choquet boundary, J. Amer. Math. Soc. 21 (2008), no. 4, 1065 -- 1084. MR2425180 (2009g:46108) , The noncommutative Choquet boundary II: hyperrigidity, Israel J. Math. 184 (2011), 349 -- 385. MR2823981 [9] David P. Blecher, The Shilov boundary of an operator space and the characterization theo- rems, J. Funct. Anal. 182 (2001), no. 2, 280 -- 343. MR1828796 [10] David P. Blecher and Christian Le Merdy, Operator algebras and their modules -- an operator space approach, London Mathematical Society Monographs. New Series, vol. 30, The Claren- don Press, Oxford University Press, Oxford, 2004. Oxford Science Publications. MR2111973 [11] David P. Blecher, Zhong-Jin Ruan, and Allan M. Sinclair, A characterization of operator algebras, J. Funct. Anal. 89 (1990), no. 1, 188 -- 201. MR1040962 [12] Nathanial P. Brown and Narutaka Ozawa, C ∗-algebras and finite-dimensional approxima- tions, Graduate Studies in Mathematics, vol. 88, American Mathematical Society, Providence, RI, 2008. MR2391387 (2009h:46101) [13] Man Duen Choi and Edward G. Effros, Injectivity and operator spaces, J. Functional Analysis 24 (1977), no. 2, 156 -- 209. MR0430809 [14] Raphael Clouatre and Kenneth R. Davidson, Absolute continuity for commuting row con- tractions, J. Funct. Anal. 271 (2016), no. 3, 620 -- 641. MR3506960 [15] Raphael Clouatre and Kenneth R. Davidson, Duality, convexity and peak interpolation in the Drury-Arveson space, Adv. Math. 295 (2016), 90 -- 149. MR3488033 [16] Serban Costea, Eric T. Sawyer, and Brett D. Wick, The Corona theorem for the Drury- Arveson Hardy space and other holomorphic Besov-Sobolov spaces on the unit ball in Cn, Anal. PDE 4 (2011), no. 4, 499 -- 550. [17] Kenneth Davidson and Matthew. Kennedy, Choquet order and hyperrigidity for function systems, preprint (2016). [18] Kenneth R. Davidson and Matthew Kennedy, The Choquet boundary of an operator system, Duke Math. J. 164 (2015), no. 15, 2989 -- 3004. MR3430455 [19] Kenneth R. Davidson and David R. Pitts, Nevanlinna-Pick interpolation for non- commutative analytic Toeplitz algebras, Integral Equations Operator Theory 31 (1998), no. 3, 321 -- 337. MR1627901 (2000g:47016) [20] Kenneth R. Davidson, Christopher Ramsey, and Orr Moshe Shalit, Operator algebras for analytic varieties, Trans. Amer. Math. Soc. 367 (2015), no. 2, 1121 -- 1150. MR3280039 [21] Jacques Dixmier, C ∗-algebras, North-Holland Publishing Co., Amsterdam-New York-Oxford, 1977. Translated from the French by Francis Jellett, North-Holland Mathematical Library, Vol. 15. MR0458185 [22] Michael A. Dritschel and Scott A. McCullough, Boundary representations for families of representations of operator algebras and spaces, J. Operator Theory 53 (2005), no. 1, 159 -- 167. MR2132691 (2006a:47095) [23] Adam H. Fuller, Michael Hartz, and Martino Lupini, Boundary representations of operator spaces, and compact rectangular matrix convex sets, preprint arXiv:1610.05828 (2016). [24] Theodore W. Gamelin, Uniform algebras, Prentice-Hall, Inc., Englewood Cliffs, N. J., 1969. MR0410387 [25] Masamichi Hamana, Injective envelopes of operator systems, Publ. Res. Inst. Math. Sci. 15 (1979), no. 3, 773 -- 785. MR566081 [26] , Triple envelopes and Silov boundaries of operator spaces, Math. J. Toyama Univ. 22 (1999), 77 -- 93. MR1744498 [27] Craig Kleski, Boundary representations and pure completely positive maps, J. Operator The- ory 71 (2014), no. 1, 45 -- 62. MR3173052 [28] , Korovkin-type properties for completely positive maps, Illinois J. Math. 58 (2014), no. 4, 1107 -- 1116. MR3421602 46 RAPHA EL CLOU ATRE AND CHRISTOPHER RAMSEY [29] Ralf Meyer, Adjoining a unit to an operator algebra, J. Operator Theory 46 (2001), no. 2, 281 -- 288. MR1870408 [30] Paul S. Muhly and Baruch Solel, An algebraic characterization of boundary representations, Nonselfadjoint operator algebras, operator theory, and related topics, 1998, pp. 189 -- 196. MR1639657 [31] Vern Paulsen, Completely bounded maps and operator algebras, Cambridge Studies in Ad- vanced Mathematics, vol. 78, Cambridge University Press, Cambridge, 2002. MR1976867 (2004c:46118) [32] Vern I. Paulsen, Completely bounded homomorphisms of operator algebras, Proc. Amer. Math. Soc. 92 (1984), no. 2, 225 -- 228. MR754708 (85m:47049) [33] , Every completely polynomially bounded operator is similar to a contraction, J. Funct. Anal. 55 (1984), no. 1, 1 -- 17. MR733029 (86c:47021) [34] Vern I. Paulsen and Ching Yun Suen, Commutant representations of completely bounded maps, J. Operator Theory 13 (1985), no. 1, 87 -- 101. MR768304 [35] Robert R. Phelps, Lectures on Choquet's theorem, Second Ed., Lecture Notes in Mathematics, vol. 1757, Springer-Verlag, Berlin, 2001. MR1835574 [36] Gelu Popescu, von Neumann inequality for (B(H)n )1, Math. Scand. 68 (1991), no. 2, 292 -- 304. MR1129595 (92k:47073) [37] , Operator theory on noncommutative varieties, Indiana Univ. Math. J. 55 (2006), no. 2, 389 -- 442. MR2225440 (2007m:47008) [38] Zhong-Jin Ruan, Subspaces of C ∗-algebras, J. Funct. Anal. 76 (1988), no. 1, 217 -- 230. MR923053 Department of Mathematics, University of Manitoba, Winnipeg, Manitoba, Canada R3T 2N2 E-mail address: [email protected] E-mail address: [email protected]
1211.1548	1	1211	2012-11-07T13:45:11	Examples of non exact 1-subexponential $C^*$-algebras	[ "math.OA", "math.FA" ]	This is a complement to our previous paper on the arxiv on quantum expanders and geometry of operator spaces. We show that there is a non-exact $C^$-algebra that is 1-subexponential, and we give several other complements to the results of that paper. Our example can be described very simply using random matrices: Let ${X_j^{(m)}\mid j=1,2,...}$ be an i.i.d. sequence of random $m\times m$-matrices distributed according to the Gaussian Unitary Ensemble (GUE). For each $j$ let $u_j(\omega)$ be the block direct sum defined by $$u_j(\omega)= \oplus_{m\ge 1} X_j^{(m)}(\omega)\in \oplus_{m\ge 1} M_m.$$ Then for almost every $\omega$ the $C^$-algebra generated by ${u_j(\omega) \mid j=1,2,...}$ is 1-subexponential but is not exact. The GUE is a matrix model for the semi-circular distribution. We can also use instead the analogous circular model.	math.OA	math	Examples of non exact 1-subexponential C∗-algebras by Gilles Pisier Texas A&M University College Station, TX 77843, U. S. A. and Universit´e Paris VI IMJ, Equipe d'Analyse Fonctionnelle, Case 186, 75252 Paris Cedex 05, France October 31, 2018 Abstract This is a supplement to our previous paper on the arxiv [13]. We show that there is a non-exact C ∗-algebra that is 1-subexponential, and we give several other complements to the results of that paper. Our example can be described very simply using random matrices: Let {X (m) j = 1, 2,···} be an i.i.d. sequence of random m × m-matrices distributed according to the Gaussian Unitary Ensemble (GUE). For each j let uj(ω) be the block direct sum defined by j uj(ω) = ⊕m≥1X (m) j (ω) ∈ ⊕m≥1Mm. Then for almost every ω the C ∗-algebra generated by {uj(ω) j = 1, 2,···} is 1-subexponential but is not exact. The GUE is a matrix model for the semi-circular distribution. We can also use instead the analogous circular model. Consider the direct sum B = ⊕m≥1Mm. By definition, for any x = ⊕m≥1x(m) ∈ B we have kxk = supm≥1 kx(m)k. We equip Mm with its normalized trace τm. Let uj = ⊕muj(m) be elements of B. Let A be the unital C ∗-algebra generated by u1, u2,··· , un. For simplicity we set u0 = 1. Let C be a unital C ∗-algebra that we assume generated by c1, c2,··· and equipped with a faithful tracial state τ . We again set c0 = 1. We say (following [8]) that {uj(m) 0 ≤ j ≤ n} tends strongly to {cj 0 ≤ j ≤ n} when m → ∞ if it tends weakly (meaning "in moments" relative to τm and τ ) and moreover kP (ui(m))k → kP (ci)k for any (non-commutative) polynomial P . This implies that for any n + 1-tuple of such polynomials P0, P1,··· , Pn, for any k and any aj ∈ Mk we have (0.1) lim m→∞kXn 0 aj ⊗ Pj(ui(m))k = kXn 0 aj ⊗ Pj(ci)k. In particular we have (0.2) lim m→∞kXn 0 aj ⊗ uj(m)k = kXn 0 aj ⊗ cjk. Let I0 ⊂ B denote the ideal of sequences (xm) ∈ B that tend to zero in norm (usually denoted by c0({MNm}). Let Q : B → B/I0 be the quotient map. It is easy to check that for any polynomial P we have kQ(P (uj))k = kP (cj)k. So that, if we set I = I0 ∩ A, we have a natural identification A/I = C. 1 Let Pd denote the linear space of all polynomials of degree ≤ d in the non commutative variables n). We will need to consider the space Mk ⊗ Pd. It will be convenient to (X1,··· , Xn, X ∗ systematically use the following notational convention: 1 ,··· , X ∗ ∀1 ≤ j ≤ n Xn+j = X ∗ j . A typical element of Mk ⊗ Pd can then be viewed as a polynomial P = P aJ ⊗ X J with coefficients in Mk. Here the index J runs over the disjoint union of the sets {1,··· , 2n}i with 1 ≤ i ≤ d. We also add symbolically the value J = 0 to the index set and we set X 0 equal to the unit. We denote by P (u(m)) ∈ Mk ⊗ Mm (resp. P (c) ∈ Mk ⊗ C) the result of substituting {uj(m)} It follows from the strong convergence of {uj 0 ≤ j ≤ n} to (resp. {cj}) in place of {Xj}. {cj 0 ≤ j ≤ n} that for any d and any P ∈ Mk ⊗ Pd we have kP (u(m))k → kP (c)k. In particular this implies (actually this already follows from weak convergence) With a similar convention we will write e.g. P (c) = P aJ ⊗ cJ . kP (c)k ≤ lim inf ∀k ∀d ∀P ∈ Mk ⊗ Pd (0.3) m→∞ kP (u(m))k. Remark 0.1. Let us write P as a sum of monomials P = P aJ ⊗ X J as above. We will assume that the operators {cJ} are linearly independent. From this assumption follows that there is a constant c2(n, d) such that XJ kaJk ≤ c2(n, d)kP (c)k. Indeed, since the span of the cJ 's is finite dimensional, the linear form that takes P to its cJ - coefficient is continuous, and its norm (that depends obviously only on (n, d)) is the same as its c.b. norm. Of course this depends also on the distribution of the family {cj} but we view this as fixed from now on. We will consider the following assumption: (0.4) n X1 τ (cj2) > k n X1 uj ⊗ ¯cjkA⊗min ¯C. Notation. Let α ⊂ N be a subset (usually infinite in the sequel). We denote B(α) = ⊕m∈αMm. uj(α) = ⊕m∈αuj(m) ∈ B(α). We will denote by A(α) ⊂ B(α) the unital C ∗-algebra generated by {uj(α) 1 ≤ j ≤ n}. With this notation A = A(N). We also set Ed(α) = {P (u(α)) P ∈ Pd}. It will be convenient to set also uα j (m) = 0 whenever m 6∈ α. Fix a degree d ≥ 1. Then for any real numbers m ≥ 1 and t ≥ 1 we define Cd(m, t) = sup m′≥m sup k≤t{kP (u(m′))k P ∈ Mk ⊗ Pd, kP (c)k ≤ 1}. 2 Theorem 0.2. Assume that for any d ≥ 1 there are a > 0 and D > 0 such that Cd(aN D, N ) → 1 when N → ∞. Assume moreover that (0.3) holds. Then for any subset α ⊂ N the unital C ∗-algebra A(α) generated by {uj(α) 1 ≤ j ≤ n} is 1-subexponential. Moreover, if we assume (0.4) then it is not exact. Proof. For subexponentiality, we need to show that for any fixed ε > 0 and any finite dimensional subspace E ⊂ A(α) the growth of N 7→ KE(N, 1 + ε) is subexponential. Since the polynomials in {uj(α)} are dense in A(α), by perturbation it suffices to check this for E ⊂ Ed(α). Thus we may as well assume E = Ed(α). Then we may choose N0 large enough so that Cd(aN D, N ) < 1 + ε for all N ≥ N0. We claim that for all N ≥ N0 we have KE(N, (1 + ε)2) ∈ O(N 2D) when N → ∞. To verify this, let P ∈ MN ⊗ Pd. Then, recalling (0.3), we have (0.5) kP (c)k ≤ sup m≥aN D kP (u(m))k ≤ Cd(aN D, N )kP (c)k. Let α′ = α ∩ [1, aN D). Let T : E → B(α′) ⊕ C be the linear mapping defined for all P in Pd by T (P (u(α)) = P (u(α′)) ⊕ P (c). We may assume α infinite (otherwise the subexponentiality is trivial). Then (0.3) shows that kTkcb ≤ 1. Conversely, by (0.5) we have k(T −1)Nk ≤ Cd(aN D, N ) < 1 + ε. Let E be the range of T . This shows that dN (E, E) < 1 + ε. We have E ⊂ ⊕k<aN D Mk ⊕ E′ where E′ is a finite dimensional subspace of C (included in the span of polynomials of degree d). Since C is exact, there is an integer K such that E′ is completely (1 + ε)-isomorphic to a subspace of MK , so that E is completely (1 + ε)-isomorphic to a subspace of ⊕k<aN D Mk ⊕ MK . Therefore we have for any N ≥ N0 KE(N, (1 + ε)2) ≤ 1 + 2 + ··· + [aN D] + K and hence our claim follows, proving the 1-subexponentiality. We now show that A(α) is not exact. Recall the notation B(α) = ⊕m∈αMm. By Kirchberg's results (see e.g. [11, p. 286]), if A(α) is exact then the inclusion map V : A(α) → B(α) satisfies for any C ∗-algebra C the mapping V ⊗ IdC : A(α) ⊗min C → B(α) ⊗max C is the following: Let M U denote the bounded (and is actually contractive). Let U be any free ultrafilter on α. von Neumann algebra ultraproduct of {Mm m ∈ α}, with each Mm equipped with τm. Recall that M U is finite (cf. e.g. [11, p. 211]). We may view C as embedded in M U . Let M be the von Neumann algebra generated by C. Note that M can also be identified (as von Neumann algebra) with the von Neumann algebra generated by C when we view it as embedded in B(L2(τ )). We have a quotient map Q1 : B(α) → M U and a (completely contractive) conditional expec- tation Q2 from M U to M. Let q : A(α) → M be the composition q = Q2Q1V . By the above, q ⊗ IdC : A(α) ⊗min C → M ⊗max C must be bounded (and actually contractive). However, if we take C = ¯C, this implies since cj = q(uj) X1 cj ⊗ ¯cjkM⊗max ¯C ≤ k uj ⊗ ¯cjkA(α)⊗min ¯C ≤ k uj ⊗ ¯cjkA⊗min ¯C. X1 X1 k n n n But now using the fact that left and right multplication acting on L2(τ ) are commuting represen- tations on M, we immediately find X1 cj ⊗ ¯cjkM⊗max ¯C τ (cj2) ≤ k X1 n n 3 and this contradicts (0.4). This contradiction shows that A(α) is not exact. Remark 0.3. Let Y (m) denote a random m × m-matrix with i.i.d. complex Gaussian entries with mean zero and second moment equal to m−1/2, and let (Y (m) ) be a sequence of i.i.d. copies of Y (m). We will use the matrix model formed by these matrices (sometimes called the "Ginibre ensemble"), for which it is known ([15]) that we have weak convergence to a free circular family {cj}. Moreover, by [5] we have also almost surely strong convergence of the random matrices to the free circular system. Actually, the inequalities from [5] that we will crucially use are stated there mostly for the GUE ensemble, i.e. for self adjoint Gaussian matrices with a semi-circular weak limit. These can be defined simply by setting j j = √2ℜ(Y (m) X (m) j ). Note we also have an identity in distribution sj = √2ℜ(cj). We call this the self-adjoint model. However, as explained in [5] , it is easy to pass from one setting to the other by a simple "2×2-matrix trick". Since we prefer to work in the circular setting, we will now indicate this trick. When working in the self-adjoint model, of course we consider only polynomials of degree d in (X1,··· , Xn). Fix k. Then the set of polynomials of degree ≤ d with coefficients in Mk of the form ) is included in the corresponding set of polynomials of degree ≤ d of the form P (Y (m) P (X (m) ). Conversely, any P (Y (m) 2n ). Indeed, the real and imaginary parts of Y (m) . This is clear when the coefficients are arbitrary in Mk. However, the results of [5] are stated for self-adjoint coefficients aJ in Mk. Then the trick consists in replacing the general coefficients aJ by self-adjoint ones defined by ) can be viewed as a polynomial of degree ≤ d in (X (m) are independent copies of X (m) ,··· , X (m) 1 j j j j j aJ = (cid:18)0 aJ J 0(cid:19) ∈ M2k. a∗ Let P = P aJ ⊗X J . One then notes that k P (s)k = kP (s)k and similarly k P (X (m) )k. Thus by simply passing from k to 2k we can deduce the strong convergence for general coefficients, as expressed in (0.1) and (0.2) from the case of self-adjoint coefficients. )k = kP (X (m) j j The following Lemma is well known. Lemma 0.4. Let F be any scalar valued random variable that is in Lp for all p < ∞. Fix a > 0. Assume that sup p≥1 p−akFkp ≤ σ. Then ∀t > 0 P{F > t} ≤ e exp −(eσ)−1/at1/a. Proof. By Tchebyshev's inequality, for any t > 0 we have tpP{F > t} ≤ (σpa)p, and hence P{F > t} ≤ (t−1σpa)p ≤ exp−p log(t/(σpa)). Assuming t/(eσ) ≥ 1, we can choose p = (t/(eσ))1/a and then we find P{F > t} ≤ exp−(eσ)−1/at1/a and, a fortiori, the inequality holds. Now if t/(eσ) < 1, we have exp−(eσ)−1/at1/a > e−1 and hence e exp −(eσ)−1/at1/a > 1 so that the inequality trivially holds. We will use concentration of measure in the following form: Lemma 0.5. There is a constant c1(n, d) > 0 such that for any k and any P ∈ Mk ⊗ Pd with kP (c)k ≤ 1, we have ∀t > 0 P{kP (Y (m))k − EkP (Y (m))k > t} ≤ e exp −(t2/dm1/d/c1(n, d)). 4 Proof. This follows from a very general concentration inequality for Gaussian random vectors, that can be derived in various ways. We choose the following for which we refer to [9]. Consider any sufficiently smooth function (meaning a.e. differentiable) f : Rn → R and let P denote the canonical Gaussian measure on Rn. Assuming f ∈ Lp(P) we have kf − Efkp ≤ (π/2)kDf (x).ykLp(P(dx)P(dy)). Let γ(p) denote the Lp-norm of a standard normal Gaussian variable (in particular γ(p) = kfkp for f (x) = x1). Recall that γ(p) ∈ O(√p) when p → ∞. Thus the last inequality implies that there is a constant β such that kf − Efkp ≤ β√pkkDf (x)k2kLp(P(dx), where kDf (x)k2 denotes the Euclidean norm of the gradient of f at x. Clearly this remains true for any f on Cn (with the gradient computed on R2n). We will apply this to a function f defined on (Cm2 )n. We need to first clarify the notation. We identify Cm2 with Mm. Then we define f on (Cm2 )n by f (w1,··· , wn) = kg(w1,··· , wn)k with g(w1,··· , wn) = P (m−1/2w1,··· , m−1/2wn, m−1/2w∗ 1,··· , m−1/2w∗ n). Note that for this choice of f the derivative Dz in any direction z satisfies Dzf ≤ kDzgk and hence taking the sup over z in the Euclidean unit sphere, we have pointwise kDfk2 ≤ sup z kDzgk. We now invoke Remark 0.1. Using the bound in that remark, we are reduced to the case when P = Y J . Then Dzg is the sum of at most d terms of the form m−1/2azib so that km−1/2azibk ≤ m−1/2kakkzikkbk and hence since kzik ≤ kzik2, km−1/2azibk ≤ m−1/2kakkbk. Recollecting all the terms , this yields a pointwise estimate at the point w ∈ M n m z kDzgk ≤ c3(n, d)m−1/2 sup{km−1/2wjk 1 ≤ j ≤ n}d−1. sup Thus we obtain and a fortiori kf − Efkp ≤ β√pc3(n, d)m−1/2k sup 1≤j≤nkY (m) j kd−1kp, kf − Efkp ≤ β√pc3(n, d)m−1/2k X1≤j≤n kY (m) j kd−1kp ≤ β√pc3(n, d)m−1/2nkkY (m) 1 kd−1kp, Now by general results on integrability of Gaussian vectors (see [7, p. 134]), we know that there is an absolute constant c5 such that kkY (m) 1 kd−1kp = kY (m) 1 Lp(d−1)(Mm) ≤ (c5pp(d − 1)EkY (m) kd−1 1 k)d−1 and since we know that EkY (m) 1 Thus we obtain k → 2 when m → ∞ it follows that kkY (m) 1 kd−1kp ≤ (c10pp(d − 1))d−1. kf − Efkp ≤ c4(n, d)m−1/2pd/2, and the conclusion follows from the preceding Lemma. 5 Remark 0.6. It will be convenient to record here an elementary consequence of Lemma 0.5. Let F = kP (Y (m))k and let tm = EkP (Y (m))k, so that we know ∀t > 0 P{F > t + tm} ≤ ψm(t) with ψm(t) = e exp −(t2/dm1/d/c1(n, d)). We have E(cid:0)(F/2 − tm)1{F/2>tm}(cid:1) = Z ∞ tm P{F/2 > t}dt ≤ Z ∞ tm P{F > t + tm}dt ≤ Z ∞ tm ψm(t)dt and hence (0.6) EF 1{F/2>tm} ≤ 2tmP{F/2 > tm} + 2Z ∞ tm ψm(t)dt. The next result is a consequence of the results of Haagerup and Thorbjørnsen [5] and of them with Schultz [6]. Let us first recall the result from [5] that we crucially need. Theorem 0.7 ([5, 6]). Let χd(k, m) denote the best constant such that for any P ∈ Mk ⊗ Pd we have EkP (Y (m) j )k ≤ χd(k, m)kP (c)k. Then for any 0 < δ < 1/4 lim m→∞ χd([mδ], m) = 1. Proof. By homogeneity we may assume kP (c)k ≤ 1. Then by Remark 0.1 we also have (0.7) XJ kaJk ≤ c2(n, d). j Fix ε > 0 and t > 1 + ε. Consider a function ϕ ∈ C ∞ c (R, R) with values in [0, 1] such that ϕ = 0 on [−1, 1] and ϕ(x) = 1 for all x such that 1 + ε < x < t and ϕ(x) = 0 for x > 2t. Let P (m) = P (X (m) ) and P (∞) = P (sj). By Remark 0.3 we can reduce our estimate to the case of a polynomial in (X (m) ) with self-adjoint coefficients and with (sj) in place of (cj). Thus we now assume kP (s)k ≤ 1. Clearly we still have a bound of the form (0.7). Then by [6] (and by very carefully tracking the dependence of the various constants in [6]) we have for m ≥ c13(n, d) (0.8) j E(cid:8)(τk ⊗ τm)ϕ(P (m))(cid:9) = (τk ⊗ τ )ϕ(P (∞)) + Rm(ϕ) where cε depends only on ε. Note ϕ(P (∞)) = 0. Therefore Rm(ϕ) ≤ k3m−2c9(n, d)cεt3 where (0.9) (0.10) E(cid:8)(τk ⊗ τm)ϕ(P (m))(cid:9) ≤ k3m−2c9(n, d)cεt3. Since kP mk ∈ (1 + ε, t) ⇒ (τk ⊗ τm)ϕ(P m) ≥ 1/(km) by Tchebyshev's inequality we find P{kP (m)k ∈ (1 + ε, t)} ≤ (km)k3m−2c9(n, d)cεt3 = k4m−1c9(n, d)cεt3. Thus we obtain EkP (m)k ≤ 1 + ε + k4m−1c9(n, d)cεt4 + E(kP (m)k1{kP (m)k>t}). 6 We will now invoke (0.6): choosing t = 2tm = 2EkP (m)k we find EkP (m)k ≤ 1 + ε + k4m−1c9(n, d)cεt4 m + 2tmψm(tm) + 2Z ∞ tm ψm(t). Now by (0.7) and by Holder we have tm ≤ c2(n, d) sup J EkX (m)J k ≤ c2(n, d) sup J E(kX (m) 1 kJ) but by a well known result essentially due to Geman [3] (cf. e.g. have c9(d) = sup m E(kX (m) 1 kd) < ∞. [14, Lemma 6.4]), for any d we Therefore we have tm ≤ c′ hence we have proved 2(n, d). We may assume tm > 1 (otherwise there is nothing to prove) and EkP (m)k ≤ 1 + ε + k4m−1c′ 9(n, d)cε + 2c′ 2(n, d)ψm(1) + 2Z ∞ 1 Thus for any ε > 0 we conclude χd(k, m) ≤ 1 + ε + k4m−1c′ 9(n, d)cε + 2c′ 2(n, d)ψm(1) + 2Z ∞ 1 ψm(t)dt. ψm(t)dt. From this estimate it follows clearly that for any 0 < δ < 1/4 lim sup m→∞ χd([mδ], m) ≤ 1 + ε. Lemma 0.8. Fix integers d, k, m. Let χd(k, m) denote the best constant appearing in Theorem 0.7. Then for any ε > 0 there are positive constants c7(n, d, ε) and c8(n, d, ε) such that if k is the largest integer such that m ≥ c7(n, d, ε)k2d the set Ωd,ε(m) = {∀P ∈ Mk ⊗ Pd kP (Y (m)(ω))k ≤ (1 + ε)(χd(k, m) + ε)kP (c)k} satisfies P(Ωd,ε(m)c) ≤ e exp(cid:16)−m1/d/c8(n, d, ε)(cid:17) . Proof. For any P ∈ Mk ⊗ Pd with kP (c)k ≤ 1, we have by Lemma 0.5 for any t > 0 P{kP (Y (m))k > t + χd(k, m)} ≤ e exp −(t2/dm1/d/c1(n, d)). Let N be a δ-net in the unit ball of the space Pd equipped with the norm P 7→ kP (c)k. Since dim(Mk ⊗ Pd) = c6(n, d)k2 for some c6(n, d), it is known that we can find such a net with Let Ω1 = {∀a ∈ N ,kP (Y (m))k > t + χd(k, m)}. Clearly N ≤ (1 + 2/δ)c6(n,d)k2 . P(Ω1) ≤ Ne exp −(t2/dm1/d/c1(n, d)) ≤ e exp(cid:16)2c6(n, d)δ−1k2 − t2/dm1/d/c1(n, d)(cid:17) . 7 Thus if we choose m so that (roughly ) t2/dm1/d/c1(n, d) = 4c6(n, d)δ−1k2 we find an estimate of the form Note that on the complement of Ω1 we have P(Ω1) ≤ e exp(cid:16)−t2/dm1/d/2c1(n, d)(cid:17) . ∀P ∈ N kP (Y (m))k ≤ t + χd(k, m). By a well known result (see e.g. [10, p. 49-50]) we can pass from the set N to the whole unit ball at the cost of a factor close to 1, namely we have on the complement of Ω1 ∀P ∈ Mk ⊗ Pd kP (Y (m))k ≤ (1 − δ)−1(t + χd(k, m))kP (c)k. Thus if we set t = ε and δ ≈ ε/2, we obtain that if m = c7(n, d, ε)k2d we have a set Ω′ 1 = Ωc 1 with P(Ω′ 1) > 1 − e exp(cid:16)−ε2/dm1/d/2c1(n, d)(cid:17) , such that for any ω ∈ Ω′ 1 we have ∀P ∈ Mk ⊗ Pd kP (Y (m)(ω))k ≤ (1 + ε)(χd(k, m) + ε)kP (c)k. Theorem 0.9. For each j let uj(α)(ω) be the block direct sum defined by uj(α)(ω) = ⊕m∈αY (m) j (ω) ∈ ⊕m∈αMm. Let α ⊂ N be any infinite subset. Then for almost every ω the C ∗-algebra A(α)(ω) generated by {uj(α)(ω) j = 1, 2,··· } is 1-subexponential but is not exact. Moreover, these results remain valid in the self-adjoint setting, if we replace uj(α)(ω) by uj(α)(ω) = ⊕m∈αX (m) j (ω) ∈ ⊕m∈αMm. Proof. We will give the proof for α = N. The proof for a general subset is identical. By Lemma 0.8 for any degree d and ε > 0 we have Xm P(Ωd,ε(m)c) < ∞. Therefore the set Vd,ε = lim inf m→∞ Ωd,ε(m) has probability 1. Furthermore (since we may use a sequence of ε's tending to zero) we have P(∩d,εVd,ε) = 1. Now if we choose ω in ∩d≥1,ε>0Vd,ε the operators uj(α)(ω) satisfy the assumptions of Theorem 0.2, and hence A(α)(ω) is 1-subexponential. Note that n X1 τ (cj2) = n. Since kujk = supm kuj(m)k and limm→∞ kuj(m)(ω)k = 2 a.s. we know that supm kuj(m)k < ∞ a.s.. Therefore, by concentration (or by the integrability of the norm of Gaussian random vectors, see [7]) E(kujk2) = E(sup m kuj(m)k2) < ∞, 8 and since the uj's have the same distribution E(kujk2) = E(ku1k2). By Fatou's lemma E lim inf n→∞ n−1 n X1 kujk2 ≤ lim inf n→∞ En−1 n X1 kujk2 = E(ku1k2) < ∞ and hence there is a measurable set Ω0 ⊂ Ω with P(Ω0) = 1 such that ∀ω ∈ Ω0 lim inf n→∞ n−1 n X1 kuj(ω)k2 < ∞. Therefore if we choose ω in the intersection of ∩d,εVd,ε∩ Ω0 (which has probability 1) we find almost surely n k X1 uj(ω) ⊗ ¯cjkA⊗min ¯C ≤ 2 max{kX uju∗ jk1/2,kX u∗ j ujk1/2} ≤ 2( n X1 kuj(ω)k2)1/2 ∈ O(√n) so that (0.4) is satisfied and A(α)(ω) is not exact. Lastly, since {uj(α)(ω) j ∈ α} has the same distribution as {√2ℜuj(α)(ω) j ∈ α} the random C ∗-algebra they generate has "the same distribution" as A(α)(ω), whence the last assertion. Remark 0.10. It seems clear that our results remain valid if we replace (Y (m) ) by an i.i.d. sequence of uniformly distributed m × m unitary matrices, but, at the time of this writing, we have not yet been able to extract the suitable estimates (as in Theorem 0.7) from the proof of Collins and Male [2] that strong convergence holds in this case. However, while this would simplify our example, by eliminating the need for estimates of kujk, it apparently would not significantly change the picture. j Acknowledgment. I am very grateful to Mikael de la Salle for useful remarks. References [1] N.P. Brown and N. Ozawa, C ∗-algebras and finite-dimensional approximations, Graduate Studies in Mathematics, 88, American Mathematical Society, Providence, RI, 2008. [2] B. Collins and C. Male, The strong asymptotic freeness of Haar and deterministic matrices, To appear. [3] S. Geman, A Limit Theorem for the Norm of Random Matrices, Ann. Prob., 8 (1980), 252- 261. [4] U. Haagerup and S. Thorbjørnsen, Random matrices and K-theory for exact C ∗-algebras, Doc. Math. 4 (1999), 341 -- 450 (electronic). [5] U. Haagerup and S. Thorbjørnsen, A new application of random matrices: Ext(C ∗ red(F2)) is not a group. Ann. of Math. 162 (2005), 711 -- 775. [6] U. Haagerup, H. Schultz and S. Thorbjørnsen, A random matrix approach to the lack of projections in C ∗ red(F2). Adv. Math. 204 (2006), no. 1, 1-83. [7] M. Ledoux, The Concentration of Measure Phenomenon, American Mathematical Society, Providence, RI, 2001. 9 [8] C. Male, The norm of polynomials in large random and deterministic matrices with an ap- pendix by Dimitri Shlyakhtenko, arXiv:1004.4155v5 [9] G. Pisier, Probabilistic methods in the geometry of Banach spaces, Probability and Analysis (Varenna, 1985), 167-241, Lecture Notes in Math., 1206, Springer, Berlin, 1986. [10] G. Pisier, The volume of Convex Bodies and Banach Space Geometry . Cambridge University Press.1989. [11] G. Pisier, Introduction to operator space theory, Cambridge University Press, Cambridge, 2003. [12] G. Pisier, Remarks on B(H) ⊗ B(H). Proc. Indian Acad. Sci. 116 (2006), no. 4, 423-428. [13] G. Pisier, Quantum Expanders and Geometry of Operator Spaces, arxiv 2012. [14] H. Schultz, Non-commutative polynomials of independent Gaussian random matrices, The real and symplectic cases, Probab. Theory Rel. Fields 131 (2005) 261309. [15] D. Voiculescu, K. Dykema and A. Nica, Free random variables, American Mathematical Society, Providence, RI, 1992. 10
1101.1200	2	1101	2011-02-11T15:05:27	Quantum Brownian Motion on noncommutative manifolds: construction, deformation and exit times	[ "math.OA", "math-ph", "math.FA", "math-ph" ]	We begin with a review and analytical construction of quantum Gaussian process (and quantum Brownian motions) in the sense of [25],[10] and others, and then formulate and study in details (with a number of interesting examples) a definition of quantum Brownian motions on those noncommutative manifolds (a la Connes) which are quantum homogeneous spaces of their quantum isometry groups in the sense of [11]. We prove that bi-invariant quantum Brownian motion can be 'deformed' in a suitable sense. Moreover, we propose a noncommutative analogue of the well-known asymptotics of the exit time of classical Brownian motion. We explicitly analyze such asymptotics for a specific example on noncommutative two-torus A{\theta}, which seems to behave like a one-dimensional manifold, perhaps reminiscent of the fact that A{\theta} is a noncommutative model of the (locally one-dimensional) 'leaf-space' of the Kronecker foliation.	math.OA	math	Quantum Brownian Motion on noncommutative manifolds: construction, deformation and exit times. Biswarup Das 1 Debashish Goswami 2 Abstract We begin with a review and analytical construction of quantum Gaussian process (and quantum Brownian motions) in the sense of [25],[10] and others, and then formulate and study in details (with a number of interesting examples) a definition of quantum Brownian motions on those noncommutative manifolds (a la Connes) which are quantum homogeneous spaces of their quantum isometry groups in the sense of [11]. We prove that bi-invariant quantum Brownian motion can be 'deformed' in a suitable sense. Moreover, we propose a noncommutative analogue of the well-known asymptotics of the exit time of classical Brownian motion. We explicitly analyze such asymptotics for a specific example on noncommutative two-torus Aθ, which seems to behave like a one-dimensional manifold, perhaps reminiscent of the fact that Aθ is a noncommutative model of the (locally one-dimensional) 'leaf-space' of the Kronecker foliation. 1 Introduction There is a very interesting confluence of Riemannian geometry and probability theory in the domain of (classical) stochatistic geometry. The role of the Brownian motion on a Riemannian manifold cannot be over-estimated in this context; in fact, classical stochastic geometry is almost synon- imous with the analysis of Brownian motion on manifolds. Since the inception of the quantum or noncommutative analogues of Riemannian geometry and the theory of stochastic processes few dacades ago, in the name of noncommutative geometry ( a la Connes) and quantum probability respectively, it has been a natural problem to explore the possibility of interaction and confluence of them. However, there is not really much work in this direction yet. In [26], some case-studies have been made but no general theory was really formulated. The aim of the present paper is to formulate at least some general principle of quantum stochastic geometry using a quantum analogue of Brownian motion on homogeneous spaces. The first problem in this context is a suitable noncommutative generalization of Brownian motion, or somewhat more generaly, quantum diffusion or Gaussian processes on manifolds. In the theory of Hudson-Parthasarathy quantum stochastic analysis, a quantum stochastic flow is thought of as (quantum) diffusion or Gaussian if its quantum stochastic flow equation does not have any 'Poisson' or 'number' coefficients (see [19], [26] and references therein for details). An important question in this context is to characterize the quantum dynamical semigroups which arise as the vacuum expectation semigrooups of quantum Gaussian processes or quantum Brownian motions. In the classical case, such criteria formulated in terms of the 'locality' of the generator are quite 1Indian Statistical Institute 2Indian Statistical Institute 1 well-known. However, there is no such intrinsic characterization in the general noncommutative framework, except a few partial results, e.g. [26, p.156-160], valid only for type I algebras. On the other hand, in the algebraic theory of quantum Levy processes a la Schuermann et al, there are simple and easily verifiable necessary and sufficient conditions for a quantum Levy process on a bialgebra to be of Gaussian type. This means, in some sense, we have a better understand- ing of quantum Gaussian processes on quantum groups. On the other hand, for any Riemannian manifold M , the group of Riemannian isometries ISO(M ) is a Lie group, and Gaussian processes or Brownian motions on the group of isometries induces similar processes on the manifold. For a compact Riemannian manifold the canonical Brownian motion genearted by the (Hodge) Laplacian arises in this way from a bi-invariant Brownian motion on ISO(M ). Moreover, whenever ISO(M ) acts transitively on M , i.e. when M is a homogeneous space for ISO(M ), any covariant Brownian motion does arise from a bi-invariant Brownian motion on ISO(M ). All these facts suggest that an extension of the framework of Schuermann et al to quantum homogeneous spaces is called for, and this is indeed one of the objectives of the present article. We also treat these concepts from an analytical viewpoint, realizing quantum Gaussian processes and quantum Brownian motions as bounded operator valued quantum stochastic flows. We then make use of the quantum isometry groups (recently developed by the second author and his collaborators, see, e.g. [11, 4, 2, 5]) of noncommutative manifolds described by spectral triples and define (and study) quantum Gaus- sian process or quantum Borwnian motion on those noncommutative manifolds which which are 'quantum homogeneous spaces' for their quantum isometry groups. For constructing interesting noncommutative examples, we investigate the problem of 'deform- ing' quantum Gaussian processes in the framwork of Rieffel ([24]), and prove in particular that any bi-invariant quantum Gaussian process can indeed be deformed. This has helped us to explicitly describe all the Gaussian generators for certain interesting noncommutative manifolds. Finally, using our formulation of quantum Brownian motion on noncommutative manifolds, we propose an analogue of the classical results about the asymptotics of exit time of Brownian motion from a ball of small volume (see, for example,[22]). We carry it out explicitly for noncommutative two-torus, and obtain quite remarkable results. The asymptotic behaviour in fact differs sharply from the commutative torus, and resembles the asymptotics of a one-dimensional manifold, which is perhaps in agreement with the fact that the noncommutative two-torus is a model for the 'leaf space' of the Kronecker foliation, and this 'leaf space' is locally (i.e. restricted to a foliation chart) is one dimensional. 2 Preliminaries 2.1 Brownian Motion on Classical Manifolds and Lie-groups Let M be a compact Riemannian manifold of dimension d, equipped with the Riemannian metric h·, ·i . Let Expx : TxM → M denote the Riemannian exponential map, given by Expx(v) = γ(1), where v ∈ TxM and γ : [0, 1] → M is the geodesic such that γ(0) = x, γ′(0) = v. The Laplace-Beltrami operator ∆ on 2 M is defined by: ∆f (x) := d2 dt2 f (Expx(tYi)) t=0 , (1) dXi=1 where f ∈ C 2(M ) and {Yi}d i=1 is a set of complete orthonormal basis of TxM. This definition is independent of the choice of orthonormal basis of TxM. If x1, x2, ...xd be a local chart at x, then writing ∂i for ∂ ∂xi , ∆ can be written as: ∆f (x) = dXi,j=1 gij(x)∂j ∂kf (x) − dXi,j,k=1 gjk(x)Γi jk(x)∂if (x), (2) where (gjk) = (gjk)−1, gjk(x) := h∂j, ∂kix , and Γi Definition 2.1. The Hodge Laplacian on C∞(M ) is the elliptic differential operator defined in terms of local coordinates (x1, x2, ...xn) as: jk are the Christoffel symbols. ∆0f = − where f ∈ C∞(M ) and g ≡ ((gij )). 1 pdet(g) nXi,j=1 ∂ ∂xj (gijpdet(g) ∂ ∂xi f ), It may be noted that the Hodge Laplacian on M and the Laplace-Beltrami operator both has similar second order terms and in case M = Rd, they coincide. It is well known that a standard d-dimensional Brownian Motion on Rd has the Hodge Laplacian : (Σ, F, P ) → M will be called a diffusion process 0 = m and the generator of the process, say L, when restricted to C∞c (M ) as its generator. An M valued Markov process X m t starting at m ∈ M if X m will be a second order elliptic differential operator i.e. Lf (x) = dXi,j=1 aij(x)∂i∂jf (x) + bi(x)∂if (x), dXi=1 where ((aij(·))) is a nonsingular positive definite matrix. We will sometimes use the term Gaussian process for such a Markov process. The diffusion process will be called a Riemannian Brownian motion, if L restricted to C∞c (M ) is the Hodge Laplacian restricted to the C∞(M ). Remark 2.2. It may be noted that the standard text books e.g. [28, 15] refer to a Markov process as a Riemmanian Brownian motion if its generator is a Laplace-Belatrami operator. We differ from this usual convention. However our convention will agree with the usual convention in context of symmetric spaces as will be explained later. The Markov semigroup associated with standard Brownian motion, given by (Ttf )(m) = IE(f (X m dilation" of the heat semigroup. t )) is called the heat-semigroup. The Brownian Motion gives a "stochastic 3 Diffusion processes on classical manifolds are important objects of study as many geometrical invariants can be obtained by analyzing the exit time of the motion from suitably chosen bounded domains. For example, Proposition 2.3. [22] Consider a hypersurface M ⊆ Rd with the Brownian motion process X m t starting at m. Let Tε = inf {t > 0 : kX m t − mk = ε} be the exit time of the motion form an extrinsic ball of radius ε around m. Then we have IEm(Tε) = ε2/2(d − 1) + ε4H 2/8(d + 1) + O(ε5), where H is the mean curvature of M. Proposition 2.4. [14] Let M be an n-dimensional Riemannian manifold with the distance function d(·, ·), and X x t ) (known as the radial part of X x t form a ball of radius ǫ around x, ǫ being fixed. Then t be the Brownian motion starting at x ∈ M. Let ρt := d(x, X x t ). Let Tǫ be the first exit time of X x IE(ρ2 t∧Tǫ ) = nt − 1 6 S(x)t2 + o(t2), where S(x) is the scalar curvature at x. We shall need a slightly modified version of the asymptotics described by Proposition 2.3, using the expression obtained in [12], of the volume of a small extrinsic ball as described below: Let Vm(ǫ) denote the ball of radius ǫ around m ∈ M. Let n be the intrinsic dimension of the manifold. Then we have Vm(ǫ) = where αn := 2Γ( 1 2 )nΓ( n 2 )−1 and K1, K2 are constants depending on the manifold. The intrinsic dimension n of the hypersurface M is obtained from IE(τǫ) as the unique integer αnǫn n (cid:0)1 − K1ǫ2 + K2ǫ4 + O(ǫ6)(cid:1)m , (3) n satisfying limǫ→0 IE(τǫ ) 2 m V ǫ Observe that V (ǫ) ǫ2 → ( αn n ) Proposition 2.3 can be recast as 2 n = ∞ if m is just less than n 6= 0 if m 6= n = 0 if m > n. 4 n and V (ǫ) n ǫ4 → ( αn n ) 2 4 n as ǫ → 0+. So the asymptotic expression of IE(τǫ) = 1 2(d − 1) ( V (ǫ)n αn 2 n + ) H 2 8(d + 1) ( V (ǫ)n αn 4 n + O(V (ǫ) 5 n ). ) In particular, we get the extrinsic dimension d and the mean curvature H by the following formulae: d = 1 2 (1 + lim ǫ→0 1 IE(τǫ) ( nV (ǫ) 2 n ), ) IE(τǫ) − 1 αn 2(d−1) ( nV (ǫ) αn V (ǫ) 4 n 2 n ) . (4) (5) H 2 = 8(d + 1)( αn n ) 4 n lim ǫ→0 4 If there is a Lie group G which has a left (right) action on M, then it is natural to study the t diffusion processes Xt ≡ {X m t , m ∈ M } which are left (right) invariant in the sense that t = X g·m · g = X m·g (X m g · X m ) almost everywhere for all g ∈ G, m ∈ M. In particular, if M = G, t t we shall call X e t (where e is the identity element of G) the cannonical left (right) invariant diffusion process, and we will usually drop the adjective left or right. For such a diffusion process, the generator L = Pi AiXi + 1 2Pi,j BijXiXj, where (Bij)i,j is a non-negetive definite matrix and {X1, ...Xd} is a basis of the Lie-algebra G. The diffusion process defined above is called bi-invariant if it is both left and right invariant. We also note that such processes constitute a special class of the so-called Levy process on groups [15] i.e. a stochastic process which has almost surely cadlag paths, left (right) independent increment and left (right) stationary increments (see [15] for details). Proposition 2.5. ([13]) A necessary and sufficient condition for a Diffusion process Motion on a Lie group G to be bi-invariant is the following: AjC l kj = 0, BijC l kj + BjlC i kj = 0 (1 ≤ i, k, l ≤ d), where C l kj are the Cartan coefficients of G. If M is a symmetric space (i.e. the isometry group G acts transitively on M ), it is interesting to study the diffusion processes on M which are covariant i.e. αg ◦ L = L ◦ αg for all g ∈ G, where L is the generator of the diffusion process and α : G × M → M is the action of G on M. Proposition 2.6. [15] Let G be a Lie group and let K be a compact subgroup. If gt is a right K invariant left Levy process in G with g0 = e, then its one point motion from o = eK in M = G/K is a G invariant Feller process in M. Conversely, if xt is a G invariant Feller process in M with x0 = o, then there is a right K invariant left Levy process gt in G with g0 = e such that its one-point motion in M from o is identical to the process xt in distribution. Suppose that G is compact. The proof of Proposition 2.6, as in [15] then implies that any covariant diffusion process xt on M can be realized as restriction of a corresponding right K invariant diffusion process on G. 2.2 Quantum Stochastic Calculus We refer the reader to [19] and [26] for the basics of Hudson-Parthasarathy (H-P for short) formal- ism and Evans-Hudson (E-H) formalism of quantum stochastic calculus which we briefly review here. Let V, W be vector spaces and V0 ⊆ V, W0 ⊆ W be vector subspaces. We will denote by Lin(V0, W0) the space of all linear maps with domain V0 and range lying inside W0. For T ∈ Lin(V0 ⊗ W0, V ⊗ W), ξ, η ∈ W0, denote by hξ, Tηi the map from V0 to V defined by hu, hξ, Tηi vi = hu ⊗ ξ, T (v ⊗ η)i , for u, v ∈ V0. Furthermore for L ∈ Lin(V0, V ⊗ W), denote by hξ, Li the operator defined by hhξ, Li u, vi = hL(u), v ⊗ ξi and define hL, ξi := hξ, Li∗ , whenever it exists. For a Hilbert space H, Γ(H) will denote the symmetric fock space over H and for f ∈ H, e(f ) will denote the exponential vector on f (see [26]). 5 2.2.1 Hudson-Parthasarathy equation Let h and k0 be Hilbert spaces with subspaces V0 ⊆ h and W0 ⊆ k0. Consider a quadruple of operators (A, R, S, T ), such that D(R), D(S) and D(A) are subspaces of h, D(T ) is a subspace of h ⊗ k0. Suppose A ∈ Lin(D(A), h), R ∈ Lin(D(R), h ⊗ k0), S ∈ Lin(D(S), h ⊗ k0) and T ∈ Lin(D(T ), h ⊗ k0). Furthermore assume that V0 ⊆ \ξ,η∈W0 (D(hξ, Ri) ∩ D(hS, ηi) ∩ D(hξ, Tηi) ∩ D(A)) . Let Γ := Γ(L2(R+, k0)). A family of operators (Vt)t≥0 ∈ Lin(h ⊗ Γ, h ⊗ Γ) is called a solution of an H-P equation with initial Hilbert space h, noise space k0 and initial condition V0 = id, if it satisfies an equation of the form hue(f ), Vtve(g)i = hue(f ), ve(g)i+Z t 0 (cid:10)ue(f ), Vs ◦(cid:0)A + hf (s), Si + hR, g(s)i +(cid:10)f (s), Tg(s)(cid:11)(cid:1) ve(g)(cid:11) ds, for u, v ∈ V0, f, g being step functions taking values in W0. We will denote the above equation symbolically by dVt = Vt ◦ (a†S (dt)+aR(dt) + ΛT (dt) + Adt), V0 = id. (6) 2.2.2 Evans-Hudson equation Let A ⊆ B(h) be a C∗ or von-Neumann algebra such that there exists a dense (in the appropriate topology) ∗-subalgebra A0. Suppose L is a densely defined map on A such that A0 ⊆ D(L). Assume the following: 1. There exists a ∗-representation π : A → A ⊗ B(k0) (normal in case A is a von-Neumann algebra); 2. a π-derivation δ : A0 → A ⊗ k0, such that : δ(x)∗δ(y) = L(x∗y) − x∗L(y) − L(x∗)y, for all x, y ∈ A0. Let σ := π − idB(h⊗k0). Then a family of ∗-homomorphism jt : A → A′′ ⊗ B(Γ), t ≥ 0 is said to satisfy an E-H equation with initial condition j0 = id, if hue(f ), jt(x)ve(g)i = hue(f ), xve(g)i +Z t 0 hue(f ), js ◦ (L(x) + hf (s), δ(x)i + hδ(x∗), g(s)i (7) for all x ∈ A0 u, v ∈ h, f, g being step functions. We will write the above equation symbolically as +(cid:10)f (s), σ(x)g(s)(cid:11))ve(g)ids, djt = jt ◦ (a†δ(dt)+aδ† (dt) + Λσ(dt) + Ldt) j0 = id. 6 (8) We will call any of the two equations described above a quantum stochastic differential equation (QSDE for short). If the solution of an H-P equation is unitary, then the solution will be called the H-P dilation of the vacuum semigroup Tt(x) := he(0), Vt(x ⊗ 1Γ)V ∗t e(0)i and jt is called an E-H dilation of the vacuum semigroup Tt(x) := he(0), jt(x)e(0)i . It is well known that solutions of such QSDE are cocycles (see [26, 19]). Definition 2.7. A semigroup (Tt)t≥0 : A → A is called a quantum dynamical semigroup (QDS for short) if for each t, Tt is a contractive completely positive map on A (normal in case A is a von-Neumann algebra). The semigroup is callled conservative if Tt(1) = 1 for all t ≥ 0. Typical examples of QDS are the Markov semigroups associated with a Markov process as well as the vacuum semigroups described above. Theorem 2.8. Let R : D(R) → h ⊗ k0 be a densely defined closed operator with D(R) ⊆ h, for Hilbert spaces h, k0. Suppose there exists a dense subspace W0 ⊆ k0, such that u ⊗ ξ ∈ D(R∗) for u ∈ D(R), ξ ∈ W0. Let H be a densely defined self adjoint operator on h such that iH − 1 2 R∗R (= G) as well as −iH − 1 2 R∗R (= G∗) generate C0 semigroups in h. Furthermore, suppose that both of D(G) and D(G∗) are contained in D(R). Then the QSDE: dUt = Ut ◦ (a†R(dt) − aR(dt) + (iH − 1 2 R∗R)dt) U0 = id; (9) has a unique solution which is unitary. 2 R∗R R∗ R Proof. Let Z =(cid:18)iH − 1 0(cid:19) . Suppose H = uH, R = vR be the polar decomposition of 0 (cid:19) . 2 R(n)∗R(n), where R(n) := R(1+ Rn )−1, and Z (n) =(cid:18)A(n) R(n)∗ H and R respectively. Put A(n) := iu(1+ Hn )−1H− 1 Then it can be verified that all the conditions of Theorem 7.2.1 in page 174 of [26] hold. Thus the above equation has a contractive solution Ut, t ≥ 0. Now observe that in the notation of Theorem 7.2.3 in page 179 of [26], Lγ η (I) = 0, for all γ, η ∈ C ⊕ W0. We will prove that βλ = {0}. Formally define L(x) = R∗(x ⊗ 1k0)R + xG + G∗x, where G = iH − 1 2 R∗R. Then the conditions (Ai) and (Aii) in page 39 of [26] hold. So by Theorem 3.2.13 in page 46 of [26], there is a minimal R(n) Let D ⊆ h be the subspace such that for x ∈ D, R∗(x ⊗ 1k0 )R + xG + G∗x ∈ B(h), i.e. L(x) ∈ B(h) semigroup say (eTt)t≥0 on B(h), whose form generator ]Lmin on a certain dense subspace is of the form R∗(x ⊗ 1k0)R + xG + G∗x. We prove that eTt is conservative: for x ∈ D. Note that 1 := 1B(h) ∈ D. Let (eT∗,t)t≥0 be the predual semigroup of (eTt)t≥0. It is known (see chapter 3 of [26]) that for σ ∈ B1(h) (B1(h) is the space of trace class operators on h), the linear span of operators ρ of the form ρ = (1 − G)−1σ(1 − G)−1, denoted by B say, belongs to D(]Lmin (ρ) = RρR∗ + Gρ + ρG∗ for ρ ∈ B, and B is a core for ]Lmin (ρ)). Since B is a core being the generator of (eT∗,t)t≥0. Moreover we have ]Lmin . Now for a ∈ D, ρ ∈ B, tr(L(a)ρ) = tr(a]Lmin ∗ ∗ ), ]Lmin ∗ ∗ ∗ 7 for ]Lmin ∗ , we have tr(L(a)ρ) = tr(a]Lmin (ρ)) for all ρ ∈ D(]Lmin ). Observe that for ρ ∈ D(]Lmin ), ∗ (cid:18)t−1Z t 0 eT∗,s(ρ)ds(cid:19)(cid:19) ∗ (10) tr eTt(a) − a t ∗ ∗ ρ! = tr a eT∗,t(ρ) − ρ = tr(cid:18)L(a)(cid:18)t−1Z t t !! = tr(cid:18)a]Lmin 0 eT∗,s(ρ)ds(cid:19)(cid:19) ; which proves that a ∈ D(]Lmin) and by continuity, L(a) = ]Lmin(a), for all a ∈ D. Now L(1) = 0 Theorem 3.2.16 of [26], βλ = {0}. The same set of arguments hold for G = −iH − 1 which implies that ]Lmin(1) = 0, i.e. (eTt)t≥0 is conservative. Thus by condition (v) in page 48 of implies that eβλ = {0} (in the notation of Theorem 7.2.3 of [26]). Moreover eLγ 2 R∗R, which η(I) = 0. Thus all the conditions of Theorem 7.2.3 in page 179 of [26] hold, which proves that the solution is unitary. The uniqueness follows from ([18, Theorem p. ]). ✷ 2.3 Compact Quantum Group We shall refer the reader to [16] and the references therein for the basics of Compact Quantum Group, which we briefly review here. Suppose A, B are ∗-algebras. If both A and B are C∗ algebras, then A ⊗ B, will mean the injective tensor product, otherwise they will denote the algebraic tensor product. Definition 2.9. A compact quantum group (CQG for short) is a unital C∗-algebra Q ⊆ B(k) equipped with a unital ∗-homomorphism (called coproduct) ∆ : Q → Q ⊗ Q such that ∆(Q)(Q ⊗ 1) as well as ∆(Q)(1 ⊗ Q) are dense in Q ⊗ Q. Given a CQG Q, there exists a unique state h on Q called the Haar state (see [16]) satisfying (h ⊗ id)∆(a) = (id ⊗ h)∆(a) = h(a)1Q. Moreover, we have a dense ∗-algebra Q0 ⊆ Q which is a Hopf∗-algebra, equipped with counit ǫ : Q0 → C and antipode κ : Q0 → Q0, satisfying (ǫ ⊗ id)∆ = (id ⊗ ǫ)∆ = id. Throughout the discussion, we will use Sweedler's notation for CQG i.e. ∆(a) = a(1) ⊗ a(2), for all a in Q0. For a map X ∈ B(H1 ⊗ H2), we will use the notation X(12) to denote the operator X ⊗ IH3 and the notation X(13) to denote the operator Σ23X(12)Σ23, where Σ23 ∈ U (H1 ⊗ H2 ⊗ H3) is the flip between H2 and H3. Definition 2.10. A map U : H → H ⊗ Q, where H is a hilbert space is called an unitary (co)representation of the CQG Q on the Hilbert space H, if eU ∈ M(K(H) ⊗ Q) defined by eU (ξ ⊗ b) := U (ξ)(1 ⊗ b), for ξ ∈ H and b ∈ Q, is an unitary operator which further satisfies (idH ⊗ ∆)eU = eU(12)eU(13). If dimension of dimH = n < ∞, we may alternatively represent U by the Q-valued n × n k=1 is an orthonormal basis of H. We will call n the invertible matrix ((hU ei, ej i))i,j, where {ek}n dimension of the representation U. By G.N.S construction using h, let Q ⊆ B(L2(h)). Then ∆ viewed as ∆ : L2(h) → L2(h) ⊗ Q becomes an unitary representaion (say U ) such that ∆(x) = U (x ⊗ 1)U∗. Moreover, Q0 is the linear span of the matrix coefficients of all finite dimensional unitary inequivalent corepresentations (see 8 [4, 16]). Furthermore, L2(h) = ⊕πHπ and Q0(⊆ L2(h)) = ⊕alg vector space of dimension d2 dimensional irreducibles π of dimension dπ by the Peter-Weyl theory for CQG[16]. π Hπ where Hπ is a finite dimensional π obtained from the decomposition of ∆ (viewed as U ) into finite 2.3.1 Action of a compact quantum group on a C∗-algebra We say that a CQG (Q, ∆) (co)-acts on a unital C∗-algebra B, if there is a unital C∗ homomorphism (called an action) α : B → B ⊗ Q, satisfying the following: 1. (α ⊗ id) ◦ α = (id ⊗ ∆) ◦ α, 2. the linear span of α(B)(1 ⊗ Q) is dense in B ⊗ Q. It has been shown in [23] that (2) is equivalent to the existence of a dense ∗-subalgebra B0 ⊆ B such that α(B0) ⊆ B0 ⊗alg Q0. We say that an action α is faithful, if there is no proper Woronowicz C∗-subalgebra (see [4],[16]) Q1 of Q such that α is a C∗ action of Q1 on B. We refer the reader to [4] and the references therein for details of C∗ action. For a CQG (Q, ∆), denote by IrrQ, the set of inequivalent, unitary irrereducible representations of Q and let uγ be a co-representation of Q of dimension dγ, for γ ∈ IrrQ. We will call a vector subspace V ⊆ B a subspace correponding to uγ if • dimV = dγ, k=1 ek ⊗ uγ ki, for some orthonormal basis {ej }dγ j=1 of V. • α(ei) =Pdγ Proposition 2.11. [23] Let α be an action of a CQG (Q, ∆) on a C∗-algebra B. Then there exists vector subspaces {Wγ}γ∈IrrQ of B such that 1. B = ⊕γ∈IrrQ 2. For each γ ∈ IrrQ, there exists a set Iγ and vector subspaces Wγi, i ∈ Iγ, such that Wγ a. Wγ = ⊕i∈Iγ Wγi. b. Wγi corresponds to uγ for each i ∈ Iγ. 3. Each vector subspace V ⊆ B corresponding to uγ is contained in Wγ. 4. The cardinal number of Iγ doesn't depend on the choice of {Wγi}i∈Iγ . It is denoted by cγ and called the multiplicity of uγ in the spectrum of α. Definition 2.12. A CQG (Q′, ∆′) is called a quantum subgroup of another CQG (Q, ∆) if there is a Woronowicz C∗-ideal J of Q such that (Q′, ∆′) ∼= (Q, ∆)/J . Definition 2.13. [23] Suppose a CQG (Q, ∆) acts on a C∗-algebra B. Then B is called 1. A quotient of (Q, ∆) by a quantum subgroup (S, ∆S ) if: a) B is C∗-isomorphic to the algebra C := {x ∈ Q : (π ⊗ id)∆(x) = 1 ⊗ x}, b) the action α is given by α := ∆C, 9 where π is the CQG morphism from Q to S. 2. Embeddable, if there exists a faithful C∗-homomorphism ψ : B → Q such that ∆ ◦ ψ = (ψ ⊗ id) ◦ α. 3. Homogeneous if the multiplicity of the trivial representation of Q in the spectrum of α (see [23]) be 1. Henceforth, we will refer to B as a quantum space. It can be easily shown that a quantum space is homogeneous if and only if the corresponding action is ergodic (i.e. α(x) = x ⊗ I implies x is a scalar multiple of the identity of B. Proposition 2.14. [23] Let α be the action of a CQG (Q, ∆) on a C∗-algebra B. Then a) (B, α) is quotient ⇒ (B, α) is embeddable ⇒ (B, α) is homogeneous. b) In the classical case, (B, α) is quotient ⇐⇒ (B, α) is embeddable ⇐⇒ (B, α) is homogeneous. We refer the reader to [23] for more discussions on these three types of quantum spaces. denotes the multiplication in B. Definition 2.15. For a linear map P : Q0 → B, where B is a ∗-algebra, define eP : Q0 → Q0 ⊗ B by eP := (id ⊗ P) ◦ ∆. For two such maps P1, P2, define P1 ∗ P2 := mB ◦ (P1 ⊗ P2) ◦ ∆, where mB It follows that (idQ0 ⊗ mB) ◦ (fP1 ⊗ idB) ◦fP2 = ^P1 ∗ P2. Observe that (id ⊗ eP)∆ = ∆ ◦ eP. 2.3.2 Rieffel Deformation Let θ = ((θkl)) be a skew symmetric matrix of order n. We denote by C∗(Tn θ ) the universal C∗- algebra generated by n unitaries (U1, U2, ...Un) satisfying UkUl = e2πθklUlUk, for k 6= l. If θkl = θ0 for k < l, where θ0 ∈ R, we will denote the corresponding universal C∗-algebra by C∗(Tn ) and W θ0 will denote the ∗-subalgebra generated by unitaries U1, U2, ...Un. Let A be a unital C∗-algebra on which there is a strongly continuous ∗-automorphic action σ of Tn. Denote by τ the natural action of Tn on C∗(Tn θ ) given on the generators U′i s by τ (z)Ui = ziUi, where z = (z1, z2, ...zn) ∈ Tn. Let τ−1 denote the inverse action s → τ−s. We refer the reader to [24] for the original approach of Rieffel using twisted convolution. Definition 2.16. The fixed point algebra of A ⊗ C∗(Tn C∗(Tn Aθ. θ ), under the action (σ × τ−1), i.e. (A ⊗ , is called the Rieffel deformation of A under the action σ of Tn, and is denoted by θ ))σ×τ −1 There is a natural isomorphism between (Aθ)−θ and A, given by the identification of A with the generated by elements of the form ap ⊗ U p ⊗ (U′)p, n , (U′)p := (U′1)p1(U′2)p2....(U′n)pn, U′1, U′2, ...U′n be- −θ) and ap belongs to the spectral subspace of the action σ corresponding subalgebra of (A ⊗ C∗(Tn where p = (p1, p2, ...pn) ∈ Zn, U p := U p1 ing the generators of C∗(Tn to the character p. θ ))(σ⊗id)×τ −1 θ ) ⊗ C∗(Tn 1 U p2 2 ...U pn 10 Let Q be a CQG with coproduct ∆ and assume that there exists a surjective CQG morphism π : Q → C(Tn) which identifies C(Tn) as a quantum subgroup of Q. For s ∈ Tn, let Ω(s) denote the state defined by Ω(s) := evs ◦ π, where evs denotes evaluation at s. Define an action of T2n on Q by (s, u) → χ(s,u), where χ(s,u) := (Ω(s) ⊗ id) ◦ ∆ ◦ (id ⊗ Ω(−u)) ◦ ∆. It has been shown in [29] θ −θ 0(cid:19) can be given a unique CQG that the Rieffel deformation Qeθ,−eθ of Q with respect toeθ :=(cid:18) 0 structure such that the such that the Hopf∗ algebra of Qθ,−θ is isomorphic as a coalgebra with the cannonical Hopf∗ algebra of Q. 2.3.3 Quantum Isometry group We begin by defining spectral triple (also called spectral data). We shall refer the reader to [7] and [4] for details. Definition 2.17. An odd spectral triple or spectral data is a triple (A∞, H, D) where H is a separable Hilbert space, A∞ is a ∗-subalgebra of B(H),(not necessarily norm closed) and D is a self adjoint (typically unbounded) operator such that for all a in A∞, the operator [D, a] has a bounded extension. Such a spectral triple is also called an odd spectral triple. If in addition, we have γ in B(H) satisfying γ = γ = γ1, Dγ = γD and [a, γ] = 0 for all a in A∞, then we say that the quadruplet (A∞, H, D, γ) is an even spectral triple. The operator D is called the Dirac operator corresponding to the spectral triple. Since in the classical case, the Dirac operator has compact resolvent if the manifold is compact, we say that the spectral triple is of compact type if A∞ is unital and D has compact resolvent. A spectral triple (A∞, H, D) will be called Θ summable if e−tD2 is a trace class operator (t ≥ 0). Next we discuss the notion of Hilbert space of k-forms in non-commutative geometry. Proposition 2.18. Given an algebra B, there is a (unique upto isomorphism)B −B bimodule Ω1(B) and a derivation δ : B → Ω1(B), satisfying the following properties: 1. Ω1(B) is spanned as a vector space by elements of the form aδ(b) with a, b belonging to B; and 2. for any B − B bimodule E and a derivation d : B → E, there is an unique B − B linear map η : Ω1(B) → E such that d = η ◦ δ. D and are called the Hilbert spaces of zero and one forms respectively (see [4]). The bimodule Ω1(B) is called the space of universal 1-forms an B and δ is called the universal derivation. Given a Θ-summable spectral triple, (A∞, H, D), it is possible to define an inner product structure on Ω0(A∞) ≡ A∞ and Ω1(A∞). The corresponding Hilbert spaces are are denoted by D and H1 H0 We now define quantum isometry group. Let (A∞, H, D) be a Θ-summable spectral triple which is admissible in the sense that it satisfies the regularity conditions (i)-(v) as given in [11, pages 9-10]. Let L := −d∗ DdD, which is a densely defined self-adjoint operator on H0 and is called the Laplacian of the spectral triple. We will denote by Q′,L the category whose objects are triplets (S, ∆, α) where 11 (S, ∆) is a CQG acting smoothly and isometrically on the given noncommutative manifold, with α being the corresponding action. Proposition 2.19. [11] For any admissible spectral triple (A∞, H, D), the category Q′,L has a uni- versal object denoted by (QISOL, α0). Moreover, QISOL has a coproduct ∆0 such that (QISOL, ∆0) is a CQG and (QISOL, ∆0, α0) is a universal object in the category Q′,L. The action α0 is faithful. The reader may see [11] and [4] for further details of QISOL. We now give some examples of quantum isometry groups. 1. non-commutative 2-tori: The non-commutative 2-tori C∗(T2 θ) is the universal C∗-algebra generated by a pair of unitaries U, V such that U V = e2πiθV U i.e. Rieffel deformation of C(T2) with respect to(cid:18) 0 −θ 0(cid:19) . The C∗ algebra underlying the quantum isometry group of θ the standard spectral triple on C∗(T2 ⊕4 for even k, k = 1, 2, ...8. Define i=1(C(T2)⊕ C∗(T2 θ) (see [7]) is given by θ)) (see [5]). Let Uk1, Uk2 be the generators of C(T2) for odd k and C∗(T2 θ) M = A1 A2 C∗1 C∗2 B1 B2 D∗1 D∗2 C1 C2 A∗1 A∗2 D1 D2 B∗1 B∗2  , where A1 = U11 + U41, A2 = U62 + U72, B1 = U52 + U61, B2 = U12 + U22, C1 = U21 + U31, C2 = U51 + U82, D1 = U71 + U81, D2 = U32 + U42. Then the coproduct ∆ and the counit ǫ θ), k=1 Mik ⊗ Mkj, ǫ(Mij) = δij. The action of the QISO on C∗(T2 are given by ∆(Mij) =P4 say α, is given by α(U ) = U ⊗ (U11 + U41) + V ⊗ (U52 + U61) + U−1 ⊗ (U21 + U31) + V −1 ⊗ (U71 + U81), α(V ) = U ⊗ (U62 + U72) + V ⊗ (U12 + U22) + U−1 ⊗ (U51 + U82) + V −1 ⊗ (U32 + U42). 2. The θ deformed sphere S2n−1 , for a skew symmet- ric matrix θ is the universal C∗-algebra generated by 2n elements {zµ, zµ}µ=1,2,..2n, satisfying the relations: : The non-commutative manifold S2n−1 θ θ • (zµ)∗ = zµ; • zµzν = e2πiθµν zν zµ, zµzν = e2πiθνµzν zµ; µ=1 zµzµ = 1. • P2n The quantum isometry group of the spectral triples on Sn whose CQG structure is described as follows: It is generated by (aµ θ , as described in [7, 8] is Oθ(n) ν )µ,ν=1,2,...n, satisfying: ν , bµ (a) aµ (b) aµ (c) bµ ν aτ ν bτ ν bτ ρ = λµτ λρνaτ ρ = λµτ λρνbτ ρ = λµτ λρνbτ ρaµ ρaµ ρbµ ν , aµ ν , aµ ν , bµ ρ aµ ρ = λτ µλνρa∗τ ν a∗τ ν , ρ aµ ρ = λτ µλνρb∗τ ν b∗τ ν , ρ bµ ν b∗τ ρ = λτ µλνρb∗τ ν , 12 µ=1(a∗µ α bµ β + bµ αa∗µ β ) = 0, µ=1(a∗µ α aµ (d) Pn ν ) =Pn The coproduct ∆ is given by ∆(aµ λ=1 bµ ∆(bµ The action of the QISO on S2n−1 λ=1 aµ αb∗µ λ ⊗ bλ β + bµ β ) = δαβ1,Pn ν ) =Pn ν +Pn α(zµ) =Xν (zν ⊗ aµ θ ν + zν ⊗ bµ λ=1 aµ λ ⊗ aλ ν +Pn ν ), α(zµ) =Xν (zν ⊗ aµ ν + zν ⊗ b µ ν ). λ=1 bµ λ ⊗ b∗λ ν , ν ; and the counit ǫ is given by ǫ(aµ ν ) = δµν , ǫ(bµ ν ) = 0. λ ⊗ a∗λ , say α is given by 3. The free sphere Sn−1 + : The free sphere denoted by Sn−1 is defined as the universal C∗ algebra generated by elements {xi}n−1 i = 1. Consider the spectral triples as described in Theorem 6.4 in page 13 of [2]. It has been shown (see [2]) that the quantum isometry group associated to this spectral triple is the free orthogonal group C∗(O+(n)) which is described as the universal C∗-algebra generated by n2 elements {xij}n satisfying i=1 satisfying xi = x∗i and Pn−1 i=1 x2 i,j=1 + a. xij = x∗ij for i, j = 1, 2, ...n; b. Pn k=1 xkixkj = δij 1,Pn For more examples, we refer the reader to [4]. k=1 xikxjk = δij1. 2.3.4 Algebraic Theory of Levy processes on involutive bialgebras We refer the reader to [10] and [25] for the basics of the algebraic theory of Levy processes on involutive bialgebras, which we briefly review here. Definition 2.20. Let B be an involutive bialgebra with coproduct ∆. A quantum stochastic process (lst)0≤s≤t on B over some quantum probability space (A, Φ) (i.e. A is a unital ∗-algebra, Φ is a positive functional such that Φ(1) = 1) is called a Levy process, if the following four conditions are satisfied: 1. (increment property) We have lrs ∗ lst = lrt for all 0 ≤ r ≤ s ≤ t, ltt = 1 ◦ ǫ for all t ≥ 0, where lrs ∗ lst := mA ◦ (lrs ⊗ lst) ◦ ∆. 2. (independence of increments) The family (lst)0≤s≤t is independent, i.e. the quantum random variables ls1t1, ls2t2, ....lsntn are independent for all n ∈ IN and all 0 ≤ s1 ≤ t1 ≤ ....tn. 3. (Stationarity of increments) The marginal distribution φst := Φ ◦ lst of jst depends only on the difference t − s. 4. (Weak continuity) The quantum random variables lst converge to lss in distribution for t → s. Define lt := l0t. Due to stationarity of increments, it is meaningful to define the marginal distributions of (lst)0≤s≤t by φt−s = Φ ◦ lst. Lemma 2.21. ([10]). The marginal distributions (φt)t≥0 form a convolution semigroup of states on B i.e. they satisfy 13 1. φ0 = ǫ, φt ∗ φs = φt+s for all s, t ≥ 0, and limt→0 φt(b) = ǫ(b) for all b ∈ B. 2. φt(1) = 1 and φt(b∗b) ≥ 0 for all t ≥ 0 and all b ∈ B. This convolution semigroup characterizes a Levy process on an involutive bialgebra. Definition 2.22. A functional l : B → C is called conditionally completely positive (CCP for short) functional if l(b∗b) ≥ 0 whenever ǫ(b) = 0. The generator of the above convolution semigroup of states is a CCP functional on the bialgebra B. Proposition 2.23. (Schoenberg correspondence)[10] Let B be an involutive bialgebra, (φt)t≥0 a convolution semigroup of linear functionals on B and l be its generator, i.e. l(a) = d dt t=0φt(a).. Then the following are equivalent: 1. (φt)t≥0 is a convolution semigroup of states. 2. l : B → C satisfies l(1) = 0, and it is hermitian and CCP. Next we define Schurmann triple on B. Definition 2.24. Let B be a unital ∗-algebra equipped with a unital hermitian character ǫ. A Schurmann triple on (B, ǫ) is a triple (ρ, η, l) consisting of 1. a unital ∗-representation ρ : B → L(D) of B on some pre-Hilbert space D, 2. a ρ − ǫ − 1-cocycle η : B → D, i.e. a linear map η : B → D such that η(ab) = ρ(a)η(b) + η(a)ǫ(b) for all a, b ∈ B, 3. and a hermitian linear functional l : B → C that satisfies l(ab) = l(a)ǫ(b) + ǫ(a)l(b) + hη(a∗), η(b)i for all b ∈ B. A Schurmann triple is called surjective if the cocycle η is surjective. Upto unitary equivalence, we have a one-to-one correspondence between Levy processes on B, convolution semigroup of states on B and surjective Schurmann triples on B. Choosing an orthonormal basis (ei)i of D, we can write η as η(·) =Pi ηi(·)ei. The η′is will be called the 'coordinate' of the cocycle η. We will denote by VA, the vector space of ǫ-derivations on A0, i.e. for VA consists of all maps η : A0 → C, such that η(ab) = η(a)ǫ(b) + ǫ(a)η(b). Lemma 2.25. Let l be the generator of a Gaussian process on A0. Suppose that (l.η.ǫ) be the surjective Schurmann triple associated to l. Let d := dimVA. Then there can be atmost d coordinates of η. Proof. Let (ηi)i be the coordinates of η. Observe that ηi is an ǫ-derivation for all i. It is enough to i=1 λiηi(a) = 0, for all a ∈ A0. This prove that {ηi}i is a linearly independent set. Suppose thatPk implies thatDη(a),Pk associated noise space. Since {η(a) : a ∈ A0} is total in k0, we havePk i=1 λieiE = 0, for all a ∈ A0, where (ei)i is an orthonormal basis for k0, the that λi = 0 for i = 1, 2, ...k. Hence proved. ✷ i=1 λiei = 0 which implies 14 Proposition 2.26. [10] For a ganerator l of a Levy process, the following are equivalent: 1. lK 3 = 0, K = kerǫ, 2. l(b∗b) = 0 for all b ∈ K 2, 3. l(abc) = l(ab)ǫ(c) − ǫ(abl(c) + l(bc)ǫ(a) − ǫ(bc)l(a) + l(ac)ǫ(b) − ǫ(ac)l(b), 4. ρK = 0, for any surjective Schurmann triple, 5. ρ = ǫ1 for any surjective Schurmann triple i.e. the process is "Gaussian", 6. ηK 2 = 0 for any Schurmann triple, 7. η(ab) = η(a)ǫ(b) + ǫ(a)η(b) for any Schurmann triple. A generator l satisfying any of the above conditions is called a Gaussian generator or the generator of a Gaussian process. Definition 2.27. We will call a Gaussian Levy process the algebraic Quantum Brownian Motion (QBM for short) if span of the maps {ηi}i is the whole of VA, where ηi are the 'coordinates' of the cocycle of the unique (upto unitary equivalnece) surjective Schurmann triple associated to l. It is known [25] that the following weak stochastic equation hlt(x)e(f ), e(g)i = ǫ(x) he(f ), e(g)i dτD{lτ ∗ (l + hg(τ ), ηi + hη, f (τ )i +Dg(τ ), (ρ − ǫ)f (τ )E)}(x)e(f ), e(g)E , (11) +Z t 0 which can be symbolically written as dlt = lt ∗ (dA†t ◦ η + dΛt ◦ (ρ − ǫ) + dAt ◦ η† + ldt) with the initial conditions l0 = ǫ1 has a unique solution (lst)0≤s≤t such that lst is an algebraic levy process on A0. Then using this algebraic quantum stochastic differential equation, it can be proved that jt =elt satisfies an EH type equation as defined in subsection 2.2 with δ =eη, L =el, σ = ]ρ − ǫ. However, it is not clear whether jt(x) ∈ A0 ⊗alg B(Γ(L2(R+, k0))). We shall prove later that at least for Gaussian generators, this will be the case i.e. jt(x) is bounded. 15 3 Quantum Brownian Motion on non-commutative manifolds 3.1 Analytic construction of Quantum Brownian motion Let (Q, ∆) be a CQG, Q0 be the corresponding Hopf-∗ algebra and h be the Haar state on Q. Let Q0 := ⊕Hπ be the decomposition obtained by Peter-Weyl theory as in section 2.3. Theorem 3.1. Let (Tt)t≥0 be a QDS on Q such that it is left covariant in the sense that (id ⊗ Tt) ◦ ∆ = ∆ ◦ Tt. Let L be the generator of (Tt)t≥0. Then there exist a CCP functional l on Q0 such thatel = L. Proof. The generator eL is CCP in the sense that ∂L(x, y) = L(x∗y) − L(x∗)y − x∗L(y) is a CP kernel (see [26]). The left covariance condition implies that for each t ≥ 0, Tt as well as L keep each of the spaces Hπ invariant. Consequently L(Q0) ⊆ Q0, so that it makes sense to define l = ǫ ◦ L. Moreover for x, y ∈ Q0, ǫ ◦ ∂L(x, y) = l ((x − ǫ(x))(y − ǫ(y))) , so that l is CCP in schurmann's sense. Hence our claim is proved. ✷ We shall prove the converse of Theorem 3.1 for the Gaussian generators. For this, we need a few preparatory lemmae. Lemma 3.2. In Sweedler's notation, h(a(1)b)a(2) = h(ab(1))κ(b(2)). Proof. h(ab(1))κ(b(2)) = ((h ⊗ 1) ◦ ∆) (ab(1))κ(b(2)) = (h ⊗ id)(cid:0)∆(ab(1))(id ⊗ κ)(b(2))(cid:1) = (h ⊗ id){∆(a)∆(b(1))(id ⊗ κ(b(2)))} = (h ⊗ id) [∆(a){(id ⊗ mQ)(∆ ⊗ id)(id ⊗ κ)∆(b)}] = (h ⊗ id) [∆(a){(id ⊗ mQ)(id ⊗ id ⊗ κ)(∆ ⊗ id)∆(b)}] = (h ⊗ id) [∆(a){(id ⊗ mQ)(id ⊗ id ⊗ κ)(id ⊗ ∆)∆(b)}] = (h ⊗ id) [∆(a){(id ⊗ mQ ◦ (id ⊗ κ)∆)∆(b)}] = (h ⊗ id) [∆(a){(id ⊗ ǫ)∆(b)}] = (h ⊗ id) [∆(a)(b ⊗ 1)] = h(a(1)b)a(2). ✷ Corollary 3.3. For any functional P : Q0 → C, h(cid:16)eP(a)b(cid:17) = h(cid:16)a( ]P ◦ κ)(b)(cid:17) . Proof. h(eP(a)b) = (h ⊗ id) [(id ⊗ P)∆(a)(b ⊗ 1)] = (id ⊗ P) [(h ⊗ id)(∆(a)(b ⊗ 1))] = (id ⊗ P)(cid:2)h(ab(1))κ(b(2))(cid:3) = h(ab(1))P(κ(b(2))) = h(ab(1)P(κ(b(2)))) = h(a(id ⊗ P)(id ⊗ κ)∆(b)) = h(a( ]P ◦ κ)(b)). ✷ 16 (12) (13) Lemma 3.4. Let η : Q0 → C be an ǫ-derivation. Put δ := (id ⊗ η) ◦ ∆. Then h(δ(a)) = 0 for all a ∈ Q0. Proof. h(δ(a)) = (h ⊗ id)(id ⊗ η) ◦ ∆(a) = (id ⊗ η)(h ⊗ id) ◦ ∆(a) = η(h(a)1Q) = h(a)η(1Q) = 0 for all a ∈ Q0, (14) where we have used the fact that (h ⊗ id) ◦ ∆(a) = (id ⊗ h) ◦ ∆(a) = h(a)1Q. ✷ Let (l, η, ǫ) be the surjective Schurmann triple for l, so that on Q0, we have 0(x)∗θi ǫ-derivation for each i. Define θi 0 := (id ⊗ ηi) ◦ ∆ for each i. Observe that l(a∗b) − ǫ(a∗)l(b) − l(a∗)ǫ(b) = hη(a), η(b)i . We recall that η : Q0 → k0, for some Hilbert space k0 so that η(a) = Pi ηi(a)ei, (ei)i being an orthonormal basis for k0 and ηi : Q0 → C being an 0(x)k ≤ kx∗(1)x(1)k Pi ηi(x(2))ηi(x(2)) < kx(1)k2kη(x(2))k2 < ∞, so that δ :=Pi θi kPi θi Q0 ⊆ L2(h). By Corollary 3.3, h(L(a∗)b) = h(a∗gl ◦ κ(b)) i.e. hL(a), biL2(h) = Da,gl ◦ κ(b)EL2(h) 0 ⊗ ei = (id⊗η)◦∆ is a derivation from Q0 to Q⊗k0. Now L is a densely defined operator with D(L) = . Thus L has an adjoint which is also densely defined. Thus L is L2(h)-closable, and we denote its closure by the same notation L. Note that a linear map S : Q0 → Q0 is left covariant i.e. (id ⊗ S)∆ = ∆ ◦ S if and only if S(Hπ) ⊆ Hπ for all π. In such a case, we will denote by Sπ the map SHπ . Lemma 3.5. Let l : Q0 → C be a CCP functional and (l, η, ǫ) be the surjective Schurmann triple associated with it. Then L =el on Q0 has Christinsen-Evans form i.e. L(x) = R∗(x ⊗ 1k0)R − R∗Rx − x R∗R + i[T, x], 1 2 1 2 for densely defined closable operators R and T, T ∗ = T. Proof. Let R := δ : Q0(cid:0)⊆ L2(h)(cid:1) → L2(h) ⊗ k0, where δ := (id ⊗ η) ◦ ∆. For x ∈ Q0, consider the quadratic forms (cid:10)Φ(x)y, y′(cid:11)L2(h) = h(cid:0)L(y∗x∗y′) − L(y∗x∗)y′ − y∗L(x∗y′) + y∗L(x∗)y′(cid:1) ; (cid:10)L(x)y, y′(cid:11)L2(h) = h(y∗L(x∗)y′) and 1 = 1 2 h(L(y∗x∗)y′ − y∗x∗L(y′) − L(y∗)x∗y′ + y∗L(x∗y′)), 2D[L − gl ◦ κ, x]y, y′EL2(h) 2i (L −gl ◦ κ) on Q0, we get hL(x)y, y′iL2(h) = h(Φ(x) + i[T, x]) y, y′iL2(h) for x ∈ Q0. Note that 2 R∗Rx − x 1 where Φ(x) = R∗(x ⊗ 1k0)R − 1 adding (17) to it, we get zero. So by taking T = 1 T is covariant, hence we have T = ⊕πTπ and since each Hπ is finite dimensional and T ∗π = Tπ by corollary 3.3, we have that T has a self-adjoint extension on L2(h) which is the L2-closure of T in Q0. ✷ 2 R∗R. Observe that by subtracting (16) from (15) and (17) (15) (16) 17 For a set of vectors {h1, h2, ....} in any vector space, we will denote by hhii = 1, 2, ...iC algebraic linear span over C. We are now in a position to prove the converse of Theorem 3.6 for Gaussian generators, which gives a left covariant QDS on Q and Gaussian generator. Theorem 3.6. Given a Gaussian CCP functional l on Q0, there is a unique covariant QDS on Q such that its generator is an extension ofel. Proof. Note that in notation of Lemma 3.5, we have R∗R =el + gl ◦ κ, T = 1 2i (el − gl ◦ κ) and hence 2 R∗R = −gl ◦ κ. Hence (id ⊗ G) ◦ ∆ = ∆ ◦ G. So each Gπ generates a semigroup in G := iT − 1 Hπ say T π 2 (R∗R)π, with T ∗π = Tπ. Take St := ⊕πT π t , which is a C0, contractive semigroup in L2(h). There exists a minimal semigroup (Tt)t≥0 on B(L2(h)), such that its generator, say Lmin, is of the form given in Lemma 3.5 when restricted to a suitable dense domain (see [26]). Now following the arguments used in proving t which is contractive, since the generator is of the form iTπ − 1 Theorem 2.8, we can conclude that Lmin =el on Q0. Thus Lmin(Hπ) ⊆ Hπ. Furthermore, each Hπ being finite dimensional, Tt(x) = etLmin )n(x), which converges in the norm for x ∈ Hπ. Thus in particular we see that Tt(Hπ) ⊆ Hπ for all π and all t ≥ 0 i.e. (id⊗Tt)◦∆ = ∆◦Tt. ✷ (x) =Pn tn n! (Lmin π π Theorem 3.7. The QDS generated by a Gaussian generator l as in Theorem 3.6, always admits E-H dilation which is implemented by unitary cocycles. D(R∗) for all u ∈ V0 and ξ ∈ W0. The proof of Theorem 3.6 implies that G := iT − 1 Proof. We will apply Theorem 2.8 with H = T. Let V0 = Q0 and W0 = heii = 1, 2, 3...iC, where i ⊗ hei. Thus u ⊗ ξ ∈ 2 R∗R generates (ei)i is an orthonormal basis for k0. Observe that by Lemma 3.4, R∗ = −Pi θ0 a C0 contractive semigroup in L2(h). Noting that G∗ is an extension of −el, using arguments as in Theorem 3.6, we can prove that G∗ generates a C0 contractive semigroup in L2(h). Thus all the conditions of Theorem 2.8 hold, and we get unitary cocycles (Ut)t≥0 satisfying an H-P equation. Then jt : B(L2(h)) → B(L2(h)) ⊗ B(Γ) defined by jt(x) := Ut(x ⊗ 1Γ)U∗t , is a ∗-homomorphic EH flow satisfying the stochastic differential equation: djt =jt ◦(cid:16)aδ†(dt) − a†δ(dt) + Ldt(cid:17) j0 = id, (18) on Q0, where δ(x) = (id ⊗ η) ◦ ∆(x) = [R, x] for x ∈ Q0. We need to show that jt(Q0) ⊆ Q′′ ⊗ B(Γ) i.e. he(f ), jt(x)e(g)i ∈ Q′′ for f, g ∈ Γ and x ∈ Q0. Let lt be the algebraic Levy process associated with l, satisfying equation (11) with ρ = ǫ. For x ∈ Q0 and ξ, ξ′ belonging to k0, let T ξ,ξ′ t and φξ,ξ′ t denote the mapsDe(χ[0,t]ξ), jt(·)e(χ[0,t]ξ′)E and (x) which will be shown towards the end of the proof. Let D denote the linear span of elements of the form e(f ) where f is a step function taking values in (ei)i. By the theorems in [27, 21], D is dense in Γ. i biχ[ti−1,ti], where t0 = 0, tk = t, and (cid:10)e(χ[0,t]ξ), lt(·)e(χ[0,t]ξ′)(cid:11) repectively. We claim that for all x ∈ Q0, T ξ,ξ′ Consider the step functions f = Pk i aiχ[ti−1,ti] and g = Pk (x) = gφξ,ξ′ t t 18 ai, bi belong to {ei : i ∈ IN}. Then note that for x ∈ Q0, t2−t1 ◦ ......T ak ,bk t1−t0 ◦ T a2,b2 t−tk−1 ^ ^ ^ φa1,b1 φa2,b2 φak ,bk t2−t1 ◦ ...... t1−t0 ◦ t−tk−1 (cid:10)e(f ), jt(x)e(g)(cid:11) = T a1,b1 = eA(x) =De(f ),elt(x)e(g)E ∈ Q0, t2−t1 ∗ ......φak ,bk t−tk−1 he(f ), jt(x)e(g)i ∈ Q′′ for all f, g ∈ Γ x ∈ Q0. where A(x) = (φa1,b1 t1−t0 ∗ φa2,b2 = (x) (x) (19) )(x). Since D is total in Γ, this implies that the map The proof of the theorem will be complete once we show that for x ∈ Q0, ξ, ν ∈ k0, we have T ξ,ν t (x) = gφξ,ν t (x). This can be achieved as follows: Fix an x ∈ Q0. From the cocycle property, it follows that T ξν t is a convolution semigroup of states on Q0. Since lt and jt satisfy equations (11) and (18) respectively, )t≥0 is L = l + hξ, ηi + η†ν and the it follows that the generator of the convolution semigroup (φξ,ν generator of the semigroup (T ξν t )t is eL. By the fundamental Theorem of coalgebra (see [10]), there is a finite dimensional coalgebra say Cx containing x. It follows that eL(Cx) ⊆ Cx. Note that Cx being finite dimensional, the mapeL : Cx → Cx is bounded with keLk = Mx(say), where Mx depends is a C0-semigroup on Q and φξ,ν on x. Now t t (20) (21) T ξ,ν t 0 T ξ,ν s (x) = x +Z t (eL(x))ds = x + teL(x) +Z t s1=0Z s1 = x + teL(x) + 2!eL2(x) + sn (eLn(x))dsk ≤ ethξ,νi tn(Mx)nkxk sn=0 T ξ,ν s2=0 t3 t2 T ξ,ν t T ξ,ν s2 (eL2(x))ds 3!eL3(x) + ..... +Z t 2!eL2(x) + ...... t2 n! ^(L ∗ L)(x) + t2 2! (x) = x + teL(x) + =eǫ(x) + teL(x) + = gφξ,ν (x), where = φξ,ν t t Now kR sn−1 t3 3! ^(L ∗ L ∗ L)(x) + .... (x) =(cid:18)ǫ + tL + t2 2! (L ∗ L) + t3 3! (L ∗ L ∗ L) + ....(cid:19) (x). s1=0Z s1 s2=0Z s2 s3=0 ....Z sn−1 sn=0 T ξ,ν sn (eLn(x))ds; → 0 as n → ∞. Thus ✷ We will call jt a Quantum Gaussian process on Q. If l generates the algebraic QBM (as defined after Proposition 2.26), then we will call jt the Quantum Brownian motion (QBM for short) on Q. Remark 3.8. If l = l ◦ κ, we will call the above QBM symmetric. This is because under the given condition, (Tt)t≥0 generated by L becomes a symmetric QDS i.e. h(Tt(x)y) = h(xTt(y)). 19 The following result, which is probably well-known, demonstrates the equivalence of the quan- tum and classical definitions of Gaussian processes on compact Lie-groups. Theorem 3.9. Let G be a compact Lie-group. Then a generator of a quantum Gaussian process (QBM) on Q = C(G) is also the generator of a classical Gaussian process (QBM) and vice-versa. associated with the map L is covariant with respect to left action of the group. Moreover, (Tt)t≥0 is a Feller semigroup. Thus by Theorem 2.1 in page 42 of [15], we see that C∞(G) ⊆ D(L). Now on C(G), there is a canonical locally convex topology generated by the seminorms kf kn := Proof. Let l be the given generator and let L :=el, as before. Observe that the semigroup (Tt)t≥0 Pi1,i2,...ik:k≤n k∂i1 ∂i2...∂ik (f )k, where ∂il is the generator of the one-parameter group Lexp(tXil ), such that C∞(G) is complete and Q0 is dense in C∞(G) in this topology (see [26] and references therein). Now as L is closable in the norm topology, it is closable in this locally convex topology and hence (by the closed graph theorem) continuous as a map from (C∞(G), {k · kn}n) → (C(G), k · k∞). From this, and using the fact that L commutes with Lg ∀g, it can be shown along the lines of Lemma 8.1.9 in page 193 of [26] that L(f ) ∈ C∞(G). Moreover, we can extend the identity L(abc) = L(ab)c − abL(c) + aL(bc) − L(a)bc + L(ac)b − acL(b) for all a, b, c ∈ C∞(G) by continuity. Thus L is a local operator. Now by the main theorem in [30], this implies that L is a second order elliptic differential operator, and hence generator of a classical Gaussian process. On the other hand, given a generator L of a classical Gaussian process, (id ⊗ L)∆ = ∆ ◦ L implies that in particular, L(Q0) ⊆ Q0. Moreover, it can be verified that L satisfies the identity L(abc) = L(ab)c − abL(c) + aL(bc) − L(a)bc + L(ac)b − acL(b) for a, b, c ∈ C∞(G) and hence in Q0. Thus L is the generator of a quantum Gaussian process as well. ✷ 3.2 Quantum Brownian motion on quantum spaces Suppose G is a compact Lie-group, with Lie-algebra g, of dimension d. There exists an Ad(G)- invariant inner product in g which induces a bi-invariant Riemannian metric in G. Suppose G acts transitively on a manifold M. Then as manifolds, M ∼= H\G, for some closed subgroup H ⊆ G and as the innerproduct on g is in particular Ad(H)-invariant, it induces a G-invariant Riemannian metric on M. Let g and h be the Lie-algebras of G and H respectively. It is a well-known fact (see [15]) that g = h ⊕ p, where p is a subspace such that Ad(H)p ⊆ p and [p, p] ⊆ h. Let {Xi}d i=1 be a basis of g such that {Xi}m i=m+1 is a basis for h. Let π : G → H/G be the quotient map given by π(g) = Hg, for g ∈ G. It follows that if f ∈ C(H/G), Xi(f ◦ π) ≡ 0 for all i = m + 1, ...d. The Laplace-Beltrami operator on M is thus given by i=1 is a basis for p and {Xi}d 1 2 ∆H/Gf (x) = 1 2 mXi=1 X 2 i (f ◦ π)(g), where f ∈ C∞(M ) and x = Hg, or in other words, if {W (i) i=1 denote the standard Brownian motion in Rd, the standard covariant Brownian motion on M (∼= H/G), starting at m is given by Bm t Xi) and exp denotes the exponential map of the Lie group t G. Now suppose M is a compact Riemannian manifold such that the isometry group of M, say G, acts transitively on M. The above discussion applies to M and it may be noted in particular that in this case, the Laplace-Beltrami operator on M coincides with the Hodge-Laplacian on M restricted := m.Bt where Bt := exp(Pm t }d i=1 W (i) 20 to C∞(M ). It follows from Proposition 2.8 in page 51 of [15] and the discussions preceeding it that a Riemannian Brownian motion on a compact Riemannian manifold M is induced by a bi-invariant Brownian motion on G, the isometry group of M, if G acts transitively on M. Furthermore by Proposition 2.13, it follows that if G acts transitively on M, then the action is ergodic i.e. C(M ) is homogeneous. Motivated by this, we amy define a Quantum Brownian motion on a quantum space as follows: Let (A∞, H, D) be a spectral triple satisfying the conditions stated in subsection 2.3.3. Let (Q, ∆) denote the quantum isometry group as obtained in Proposition 2.19, α being the action. Suppose that (Q, ∆) acts ergodically on A∞, i.e. the quantum space A := A∞k·k∞ is homo- geneous. Let l : Q0 → C be the generator of a bi-invariant quantum gaussian process jt(·) on Q i.e. (l ⊗ id) ◦ ∆ = (id ⊗ l) ◦ ∆ on Q0. Define the process kt := (id ⊗ lt) ◦ ∆ : A0 → A ⊗ B(Γ(L2(R+, k0))) on A0. Since α is an ergodic action, it is known that there exists an α-invariant state τ on A (see []). Moreover, in the notation of Proposition 2.11, τ is faithful on A0 := ⊕γ∈IrrQ(cid:0)⊕i∈Iγ Wγi(cid:1) and as a Hilbert space, L2(τ ) := ⊕γ∈IrrQ(cid:0)⊕i∈Iγ Wγi(cid:1) . Theorem 3.10. There exists a unitary cocycle (Ut)t≥0 ∈ Lin(L2(τ )⊗Γ) satisfying an HP equation, where Γ := Γ(L2(R+, k0)) such that kt(x) = Ut(x⊗idΓ)U∗t for x ∈ A0. Thus kt extends to a bounded map from A to A′′ ⊗ B(Γ). Moreover, kt satisfies an EH equation with coefficients (LA, δ, δ†), where LA := (id ⊗ l) ◦ α, δ := (id ⊗ η) ◦ α, and initial condition j0 = id. Proof. Observe that (α ⊗ id) ◦ α = (id ⊗ ∆) ◦ α. Hence, proceeding as in subsection 3.1, with Hπ replaced by Wγ for γ ∈ IrrQ, and L2(h) replaced by L2(τ ), we get the existence of a unitary cocycle (Ut)t≥0 satisfying an HP equation with coefficient matrix (cid:18)iT − 1 initial condition U0 = I, where T, R are the closed extensions of 1 (id ⊗ η) ◦ α respectively. Now, proceeding as in Theorem 3.7, we get our result. ✷ 2 R∗R R∗ R 0(cid:19) , with the 2i (LA − (id ⊗ (l ◦ κ)) ◦ α) and Definition 3.11. A generator of covariant quantum Gaussian process (QBM) on the non-commutative manifold A is defined as a map of the form lA := (id ⊗ l) ◦ α, where l is the generator of some bi-invariant quantum Gaussian process (QBM) on Q. In such a case, the EH flow kt obtained in Theorem 3.10 will be called covariant quantum-Gaussian process (QBM) with the generator LA. We will usually drop the adjective 'covariant'. Observe that (LA ⊗ idQ)α =(idA ⊗ l ⊗ idQ)(α ⊗ idQ)α = (idA ⊗ l ⊗ idQ)(idA ⊗ ∆)α = (idA ⊗ (l ⊗ idQ)∆)α = (idA ⊗ (idQ ⊗ l)∆)α (since (l ⊗ id)∆ = (id ⊗ l)∆) = (α ⊗ l)α = α ◦ LA. (22) It is not clear whether the condition (22) is equivalent to the bi-invariance of the Gaussian generator l on Q. However, let us show that it is indeed so for the class of quantum spaces which are quotient (hence in particular for the classical ones). 21 We recall (see subsection 2.3.1) that A will be called a quotient of the CQG (Q, ∆) by a quantum subgroup H if A is C∗-algebra isomorphic to the algebra {x ∈ Q : (π ⊗ id)∆(x) = 1 ⊗ x}, where π : Q → H is the CQG morphism. Theorem 3.12. Let l : Q0 → C be the generator of a quantum Gaussian process on Q. Suppose (Q, ∆) acts on a quantum space A such that A is a quotient space. Denote the action by α and define LA := (idA ⊗ l)α. Then the following conditions are equivalent: 1. (l ⊗ id)∆ = (id ⊗ l)∆. 2. (l ⊗ idQ)α = (idA ⊗ l)α. 3. (LA ⊗ idQ)α = α ◦ LA. Proof. It can be shown (see [23]) that α = ∆A in case of quotien spaces, where A has been identified with the algebra {x ∈ Q : (π ⊗ id)∆(x) = 1 ⊗ x}. Thus (1) ⇒ (2) is trivial. Let us prove (2) ⇒ (1). It can be shown (see [23, page 5]) that if A is a quotient space, then the subspaces Wγi for γ ∈ IrrQ, as described in Proposition 2.11 are spanned by {uγ j=1 and cardinality of the set Iγ is nγ. So for a fixed i, j, i = 1, 2, ...nγ ; j = 1, 2, ....dγ , we have ij ) = (idA ⊗ l)α(uγ ij) ij}dγ (l ⊗ idQ)α(uγ i.e. l(uγ ik)uγ kj = dγXk=1 uγ ikl(uγ kj) ; dγXk=1 ii) = l(uγ comparing the coefficients, we get l(uγ jj), ij) = 0 for i 6= j, where 1 ≤ i ≤ nγ and ⊕dγ 1 ≤ j ≤ dγ. As a vector space, Q0 = ⊕γ∈IrrQ i=1 Wγi. From the preceding discussions, it follows that (l⊗id)∆(uγ ij ) which implies that (l⊗id)∆ = (id⊗l)∆ i.e. (2) ⇒ (1). (1) ⇒ (3) was already observed right after defining covariant quantum Gaussian process. The proof of the theorem will be completed if we show (3) ⇒ (2). This can be argued as follows: ij ) = (id⊗l)∆(uγ l(uγ Since A is a quotient, we have α = ∆A. Consider the functional ǫA0 , where A0 := A ∩ Q0. Note ⊗ idQ on both sides of (3), we get (l ⊗ idQ)α = LA := (idA ⊗ l)α. ◦ LA = l. So applying ǫA0 that ǫA0 Thus (3) ⇒ (2). ✷ 3.3 Deformation of Quantum Brownian motion Recall the set-up and notations of section 2.3, where the Rieffel deformation of (Q, ∆), denoted by Qθ,−θ, for some skew symmetric matrix θ, of a CQG was described. As a C∗-algebra, it is the fixed point subalgebra (Q ⊗ C∗(T2n , and has the same coalgebra structure as that of Q. θ ))σ×τ −1 Theorem 3.13. Let l be the generator of a quantum Gaussian process and L :=el. Suppose that L ◦ σz = σz ◦ L, for z ∈ T2n. Then we have the following: (i) (L ⊗ id)((Q0 ⊗alg W)σ×τ −1 ) ⊆ (Q0 ⊗alg W)σ×τ −1 ; (ii) Lθ := (L ⊗ id) (Q0 ⊗algW)σ×τ −1 is a generator of a quantum Gaussian process; 22 (iii) with respect to the natural identification of (Qθ,−θ)−θ,θ with Q, we have (Lθ)−θ = L. Proof. Notice that the counit ǫ and the coproduct ∆ remains the same in the deformed algebra, as the coalgebra Q0 is vector space isomorphic to (Q0 ⊗alg W)σ×τ −1 . By our hypothesis, σz ◦L = L◦σz , which implies (i). Since L is a CCP map, it follows that Lθ is a CCP map. Moreover, since we have the identity l(abc) = l(ab)ǫ(c) − ǫ(ab)l(c) + l(bc)ǫ(a) − ǫ(bc)l(a) + l(ac)ǫ(b) − ǫ(ac)l(b) for a, b, c ∈ Q0, it follows that lθ := ǫ ◦ Lθ also satisfies the same identity on the coalgebra (Q0 ⊗alg W)σ×τ −1 . Thus lθ, or equivalently Lθ, generates a quantum Gaussian process on Qθ,−θ, which proves (ii). (iii) follows from the natural identification of (Qθ,−θ)−θ,θ with Q and an application of the result in (ii). ✷ We have the following obvious corollary: Corollary 3.14. For a bi-invariant quantum Gaussian process, the conclusion of Theorem 3.13 hold. Thus we have a 1 − 1 correspondence given by L ↔ Lθ, between the set of quantum Gaussian processes on Q and Qθ,−θ. In case Q is co-commutative, i.e. Σ ◦ ∆ = ∆, where Σ is the flip operation, it is easily seen that any quantum Gaussian process on Q will be bi-invariant and so the 1 − 1 correspondence L ↔ Lθ holds for arbitrary quantum Gaussian processes in such a case. It is not clear, however, whether we can get 1 − 1 correspondence between bi-invariant QBM on the deformed and undeformed CQGs. Theorem 3.15. If in the setup of Theorem 3.13, we have Q = C(G) for a compact Lie-group G with abelian Lie-algebra g, then the hypothesis of Theorem 3.13 and hence the conclusion hold. Proof. Let G = GeFi∈Λ Gi, where e ∈ G is the identity element and Ge, Gi are the connected components of G, Ge being the identity component. Let the coproduct of the Rieffel-deformed algebra Qθ,−θ be denoted by ∆θ (note that it is the same coproduct as the original one). Observe that since the action σ is strongly continuous, z · Ge ⊆ Ge ∀z ∈ T2n, or equivalently, we have σz (C(Ge)) ⊆ C(Ge). Thus one has the following decomposition: (C(G))θ,−θ := (C(Ge))θ,−θ ⊕ (B)θ,−θ, where B := ⊕i∈ΛC(Gi) and C(Ge)θ,−θ itself is a quantum group satisfying ∆θ (C(Ge)θ,−θ) ⊆ C(Ge)θ,−θ ⊗ C(Ge)θ,−θ. Note that since Ge is an abelian Lie-group, C(Ge)θ,−θ is a co-commutative quantum group. We claim that l is supported on C(Ge)θ,−θ. Observe that χGe (the indicator function of Ge) ∈ C(Ge). Moreover, we have σz (χGe ) = χGe . Thus χGe is identified with χθ := θ ))σ×τ −1 χGe ⊗ 1 ∈ (C(G) ⊗ C∗(T2n is a self-adjoint idempotent in C(G)θ,−θ. It now suffices to show that l((1 − χθ . Let Ge (l, η, ǫ) be a Schurmann triple for l. Now )a) = 0 for all a ∈ (C(G)0 ⊗alg hUii = 1, 2, ...2niC)σ×τ −1 . In particular, χθ Ge Ge l((1 − χθ Ge )a) = l(1 − χθ Ge )ǫ(a) + ǫ(1 − χθ Ge Now as (1 − χθ Ge By conditions 2 and 6 of Proposition 2.26, we have l(1 − χθ Ge ), and clearly ǫ(1 − χθ Ge )2 = (1 − χθ Ge ) = 0, which implies that 1 − χθ Ge ∈ ker(ǫ)2. ) = 0. This implies )l(a) +Dη(1 − χθ Ge ), η(a)E . ) = η(1 − χθ Ge 23 that l((1 − χθ Ge co-commutative quantum group, we have )a) = 0 for all a ∈ (C(G)0 ⊗alg hUii = 1, 2, ...2niC)σ×τ −1 . Now as (C(Ge))θ,−θ is a (l ⊗ id)∆θ = (id ⊗ l)∆θ on C(Ge)θ,−θ. (23) Let z = (u, v) for u, v ∈ Tn. Let us recall that σz = (Ω(u) ⊗ id)∆(id ⊗ Ω(−v))∆, where we have Ω(u) := evu ◦ π, π : C(G) → C(Tn) being the surjective CQG morphism. Let R(x) := σ(0,x) and L(x) := σ(x,0) for x ∈ Tn. By equation (23), we have l(R(u)a) = l(L(u)a) for all a ∈ C(Ge)θ,−θ. Now L(u)(C(Gi)θ,−θ) ⊆ C(Gi)θ,−θ and R(u)(C(Gi)θ,−θ) ⊆ C(Gi)θ,−θ for all i and l(C(Gi)θ,−θ) = 0, which, in combination with equation (23), gives l(R(u)a) = l(L(u)a) for all a ∈ C(G)θ,−θ. From this, it easily follows that L ◦ σz = σz ◦ L for all z ∈ T2n. ✷ Moreover, in subsection 3.4, we shall see that condition of Theorem 3.13 is indeed necessary, i.e. there may not be a 'deformation' of a general quantum Gaussian generator. 3.4 Computation of Quantum Brownian motion In this subsection, we compute the generators of QBM on the QISO of various non-commutative manifolds. We refer the reader to subsection 2.3.3 for a recollection of the description of QISO of the non-commutative manifolds which we will consider here. a. non-commutative 2-tori: Recall from subsection 2.3.3 that C∗(T2 θ) is the universal C∗- algebra generated by a pair of unitaries U, V satisfying the relation U V = e2πiθV U. The QISO of C∗(T2 θ) is a Rieffel deformation of the compact quantum group C(cid:0)T2 ⋊ (Z2 2 ⋊ Z2)(cid:1) (see [4]). Moreover, T2⋊(Z2 2 ⋊Z2) is a Lie-group with abelian Lie-algebra. Hence an application of Theorem 3.15 and Theorem 3.13 leads to the conclusion that the generators of quantum Gaussian processes on the QISO of C∗(T2 θ) are precisely those coming from QISO(C(T2))=ISO(T2)∼= C(T2 ⋊ (Z2 2 ⋊ Z2)) i.e. they are of the form lθ, where l is a 2 ⋊ Z2), i.e. on its identity component T2. generator of classical Gaussian process on T2 ⋊ (Z2 It can be seen by a direct computation that the space of ǫ-derivations on QISO(C∗(T2 θ)) is same as the space of ǫ-derivations on C(T2 ⋊ (Z2 2 ⋊ Z2)). Moreover, all the ǫ-derivations are supported on the identity component namely C(T2), which remains undeformed as a quantum subgroup of QISO(C∗(T2 θ)). Thus it follows that in this case, a QBM on the undeformed CQG remains a QBM on the deformed CQG. Using the action α as described in subsection 2.3.3, we can construct a QBM on C∗(T2 described in section 3.2, and conclude that θ) as Theorem 3.16. Any QBM kt on C∗(T2 θ) is essentially driven by a classical Brownian mo- θ)′′ ⊗ B(Γ(L2(R+, C2))) ∼= B(L2(ω1, ω2)), tion on T2, in the sense that kt : C∗(T2 θ) → C∗(T2 where (ω1, ω2) is the 2-dimensional standard Wiener measure, is given by kt(a)(ω1, ω2) = α (a). (e2πiω1 ,e2πiω2 ) We now give an intrinsic characterization of a qunatum Gaussian (QBM) generator on C∗(T2 Let A0 denote the ∗-subalgebra spanned by the unitaries U, V. θ) : 24 Theorem 3.17. A linear CCP map L : A0 → A0 is a generator of a quantum Gaussian process (QBM) on C∗(T2 θ) if and only if L satisfies: 1. L(abc) = L(ab)c − abL(c) + L(bc)a − bcL(a) + L(ac)b − acL(b), for all a, b, c ∈ A0. 2. (L ⊗ id) ◦ α = α ◦ L, where α is the action of T2 on C∗(T2 θ). Moreover, L will generate a QBM if and only if l(1,1) − l(1,0) − l(0,1) < 2qRe(l(1,0))Re(l(0,1)), where l(U ) = l(1,0)U, l(V ) := l(0,1), l(U V ) := l(1,1) U V. Proof. Suppose that L is the generator of a quantum Gaussian process (QBM) on C∗(T2 θ). No- tice that condition (2.) implies that U, V, U V are the eigenvectors of L. Let the eigenvalues be denoted by l(1,0), l(0,1) , l(1,1) respectively. Then there exists a Gaussian (Brownian) functional l on QISO(C∗(T2 θ))(= Q) with surjective Schurmann triple (l, η, ǫ), such that L = (id ⊗ l)α. Let (ηi)i=1,2 be the coordinates of η. Then since l(abc) = l(ab)ǫ(c) − ǫ(ab)l(c) + l(bc)ǫ(a) − ǫ(bc)l(a) + l(ac)ǫ(b) − ǫ(ac)l(b) for a, b, c ∈ Q0, we have condition 1. of the present theorem. Condition (2.) follows by a direct computation, along with the fact that if l is generates a θ). QBM, then η1, η2 spans the space VC∗(T2 Conversely, suppose that we are given a CCP functional L, satisfying conditions (1.) and (2.). Choose two vectors (c1, c2), (d1, d2) ∈ R2 such that c2 2 = −2Re(l(0,1)), and c1d1 + c2d2 = l(1,1) − l(1,0) − l(0,1). Consider the two ǫ-derivations η1 := c1η(1) + d1η(2) and η2 := c2η(1) + d2η(2). Define a CCP finctional lnew on Q as : lnew(U11) = l(1,0) and lnew(U12) = l(0,1), lnew(Ukj) = 0 for k > 1, j = 1, 2, and extend the definition p=1 η1(a∗)ηp(b). Note that we have lnew(abc) = lnew(ab)ǫ(c)−ǫ(ab)lnew(c)+lnew(bc)ǫ(a)−ǫ(bc)lnew(a)+lnew(ac)ǫ(b)−ǫ(ac)lnew(b) for a, b, c ∈ Q0. It follows that Lnew := (id ⊗ Lnew)α satisfies conditions (1.) and (2.) Thus L = Lnew on A0 and since Lnew generates a quantum Gaussian process (QBM) on C∗(T2 θ), so does L. ✷ to (Q)0 by the rule l(a∗b) = l(a∗)ǫ(b) + l(b)ǫ(a∗) +P2 1 + d2 1 + c2 2 = −2Re(l(1,0)), d2 Remark 3.18. It follows from this that in a similar way, we can also characterize generators of quantum Gaussian processes on quantum spaces on which Tn acts ergodically. b. The θ deformed sphere Sn θ : Theorem 3.19. (i) Suppose that l is the generator of a quantum Gaussian process on Oθ(2n). Then it satisfies the following: There exists 2n complex numbers {z1, z2, .....z2n} with Re(zi) ≤ 0 for all i and A ∈ M2n(C) with Aii = 0 ∀ i and [Aij − zi − zj]ij ≥ 0, such that l(ai i) = zi, l(ai∗i aj j) = Aij i, j = 1, 2, ....2n. (24) Conversely, given 2n complex numbers {z1, z2, .....z2n} and A ∈ M2n(C), such that Re(zi) ≤ 0, Aii = 0 ∀ i and [Aij − zi − zj]ij ≥ 0, there exists a unique map l, such that l generates a quantum Gaussian process and satisfies equation (24). 25 (ii) The generator of a quantum Gaussian process say l generates a QBM if and only if the matrix is invertible. ν) − l(aµ∗µ ) − l(aν (cid:2)l(aµ∗µ aν ν )(cid:3)µ,ν ∈ M2n(C) (25) (iii) l generates a bi-invariant quantum Gaussian process if and only if zβ = z for all β = 1, 2, ..2n, where z ∈ R such that z ≤ 0. Proof. Let us first calculate all possible ǫ-derivations. Let η be an ǫ-derivation on this CQG. Put η(aµ ν , µ.ν = 1, 2, ...n. Using condition (a), we get ν ) = dµ ν ) = cµ ν , η(aµ ν , η(bµ ν , η(bµ cµ ν δτ ρ + cτ ρδµ ν = λµτ λρν (cµ ν δτ ρ + cτ ρδµ ν ); ν ) = bcµ ν ) =cdµ ν = 0 for µ 6= ν. Likewise using conditions (b) and (c), we get ν = 0 for µ 6= ν. Using condition (d) with α = β, we arrive at the following putting τ = ρ, we get cµ ν = dµ relations: ν = bdµ bcµ α = 0 (since η(1) = 0), α = 0, α = 0; α + cα dα α + dα bcα bdα α +bdα β = cβδαβ,bcα this implies that cα β = −cβδαβ for n complex numbers {c1, c2, ...cn}. It may be noted that all the above steps are reversible, and hence this also characterizes ǫ-derivations on Oθ(2n). Note that the space of ǫ-derivations, VOθ(2n) is 2n-dimensional and is spanned by n ǫ-derivations {η(1), η(2), ...η(2n)}, where (η(k)(aα β ))α,β = Ekk, where Eij denote an elementary matrix. Now we prove (i) as follows: Let l be the generator of a quantum Gaussian process. Let the surjective Schurmann triple of l be (l, η, ǫ). Let (ηi)i be the coordinates of η, which are ǫ-derivations. By Lemma 2.25, there can be atmost 2n such coordinates. Let ηi(aα αβ such that αβ = c(i) c(i) β ) = l′αβ. Then using β δαβ. Suppose that l(aα the relations among the generators of Oθ(2n), as given in subsection 2.3.3, we arrive at the following results: αβ and ηi(aα∗β ) = bc(i) β ) = c(i) β ) = lαβ and l(bα β δαβ and bc(i) αβ = −c(i) lαβ = 0 for all α 6= β, c(i) lαα + lαα = −Xi l′αβ = 0 for all α, β. α 2 for all α = 1, 2, ...2n, Moreover, we have l(a∗b) − l(a∗)ǫ(b) − ǫ(a∗)l(b) = hη(a), η(b)i , so that by taking zi l(ai j)]ij we have the result. i), A := [l(xi∗i aj := Conversely, suppose that we are given 2n complex numbers {z1, z2, ...z2n} such that Re(zi) ≤ 0 for all i and A ∈ M2n(C), satisfying the hypothesis. Let B := [Aij − zi − zj]ij. Suppose that 26 P := B 1 i=1 by ηk := P2n η :=P2n 2 . Let us define 2n ǫ-derivations (ηi)2n i=1 ηi ⊗ ei, where {ei}i is the standard basis of C2n. Define a CCP map l on Oalg i=1 Pikη(i), k = 1, 2, ....2n. Let θ (2n) by the prescription l(ai j) = 0 ∀ i, j and extending the map to Oalg θ (2n) by the rule l(a∗b) = l(a∗)ǫ(b) + ǫ(a∗)l(b) + hη(a), η(b)i . Such a map is clearly the generator of a quantum Gaussian process on Oθ(2n) and it satisfies l(ai∗i aj j) = Aij. The uniqueness follows from the fact that a generator of a quantum Gaussian process on Oθ(2n) must satisfy the identity: j ) = 0 for i 6= j, l(bi i) = zi, l(ai l(abc) = l(ab)ǫ(c) − ǫ(ab)l(c) + l(bc)ǫ(a) − ǫ(bc)l(a) + l(ac)ǫ(b) − ǫ(ac)l(b), for all a, b, c ∈ Oalg θ (2n). For proving (ii), let us proceed as follows: Let l be the generator of a QBM and let (l, η, ǫ) be the surjective Schurmann triple associated with l. Suppose that (ηi)i are the coordinates of η. Then by hypothesis, {η1, η2, ...η2n} forms . Then j)]ij = P∗P, which i η(i). Consider the 2n × 2n matrix P such that Pij := c(j) a basis for V. Let ηk =Pi c(k) P∗P is an invertible matrix. Moreover, we have [l(ai∗i aj proves our claim. j) − l(ai∗i ) − l(aj i j) − l(ai∗i ) − l(aj Conversely, suppose that l is the generator of a quantum Gaussian process, such that B := [l(ai∗i aj j)]ij is an invertible matrix. Let (l, η, ǫ) be the surjective Schurmann i η(i), for all k. Let ]ij. Then we have P∗P = B, which implies that the matrix P is invertible, and i=1 forms a basis for VOθ (2n), which proves the claim. triple associated to l. Let (ηi)i be the coordinates of η. Let ηk = Pi c(k) P := [c(j) hence {ηi}2n i (iii) follows by a direct computation using the formula for coproduct, as given in subsection 2.3.3. ✷ We have the following obvious corollary, which follows from (iii) of the theorem above and the definition of quantum Gaussian process on quantum homogeneous space. Corollary 3.20. A map L , which generates a qunatum Gaussian process on S2n−1 θ , S2n−1 θ satisfy: L S2n−1 θ (zµ) = czµ, for some real number c ≤ 0. Remark 3.21. Notice that the space of ǫ-derivations on the undeformed algebra O(2n) has dimension more than 2n, since there are ǫ-derivations, which takes non-zero values on (bµ ν )µν and hence there are quantum Gaussian processes on O(2n) such that their generators take non-zero values on bα β , and so there is no 1-1 correspondence between quantum Gaussian processes on the deformed and undeformed algebra in this case. c. The free orthogonal group O+(2n): We refer the reader to subsection 2.3.3 again, for the definition and formulae for the free orthogonal group. Before stating the main theorem, we introduce some notations for convenience. Let A ∈ Mn(2n−1)(C). We will index the elements of A by the set IN 4 instead of IN 2 as follows: 27 Let A =  A1 A2 . . . A2n−1  , where Ai is a n(2n − 1) × (2n − i + 1) matrix, such that (Ai)kl =a(i,i+k,1,1+l)χ{1,2,...2n−1} (l) + a(i,i+k,2,3+(l−2n))χ{2n,2n+1,...4n−3} (l) + a(i,i+k,3,4+(l−(4n−2)))χ{4n−2,...6n−2} (l) + . . . + a(i,i+k,2n−1,2n) χ{n(2n−1)} (l), for k = 1, 2, ...2n − i + 1, where χB denotes the indicator function of the set B. We now state the main theorem: Theorem 3.22. (i) There exists a 1-1 correspondence between generators of quantum Gaus- sian processes on O+(2n) and matrices L := [Lij] ∈ M2n(C) and A := [Aij] ∈ Mn(2n−1)(C), satisfying a. B ∈ Mn(2n−1)(C), defined by B := [a(i,j,k,l) − Lij − Lkl], i < j, k < l is positive definite, b. Lij + Lji := −Pi−1 k=1 a(k,i,k,j) +Pj−1 k=i+1 a(i,k,k,j) −P2n (ii) l will generate a QBM if and only if the matrix B, defined above, is invertible. k=j+1 a(i,k,j,k), i < j. (iii) There exists no bi-invariant quantum Gaussian process on O+(2n). Proof. Using the relations among the generators, as given in subsection 2.3.3, it is seen that the epsilon-derivations on this algebra are given by η(xij) = Aij; such that Aij = −Aji. Clearly this characterizes the ǫ-derivations on the CQG. Observe that the space of ǫ-derivations, VO+(2n) has dimension n(2n − 1). A basis for the space is given by {η(ij)}i<j, such that η(ij)(xij) = 1, η(ij)(xji) = −1 and η(ij)(xkl) = 0 for k 6= i, j or l 6= i, j. So after a suitable re-indexing, let us denote the basis by {η(p)}n(2n−1) We prove (i): p=1 . Let l be the generator of a quantum Gaussian process on Oθ(2n), with the surjective Schur- mann triple (l, η, ǫ). Let (ηi)i be the coordinates of η. By Lemma 2.25, there can be atmost 28 n(2n−1) coordinates. Let A(i) := ((ηi(xkl)))kl. Now using the relations among the generators, as described in subsection 2.3.3, we see that such that ((l(xij)))i,j = L Lij + Lji = − n(2n−1)Xs=1 2nXk=1 ik A(s) A(s) jk , i < j. 1 2 . Define n(2n − 1) ǫ-derivations by ηp :=Pn(2n−1) k=1 Ppkη(k), Thus by taking a(i,j,k,l) := l(xijxkl) and Lij := l(xij), the conclusion follows. Conversely, suppose that we are given matrices L ∈ M2n(C), A ∈ Mn(2n−1)(C) satisfying the hypothesis in (i). Let P := B p = 1, 2, ...n(2n − 1). Define a CCP map by the prescription l(xij) := Lij, and extending the definition to Oalg ηp(a)ηp(b), where Oalg + (2n) is the ∗-algebra generated by xij, i, j = 1, 2, ...n(2n − 1). Clearly such a functional satisfies l(xijxkl) = a(i,j,k,l), i < j, k < l. The uniqueness follows from the fact that l satisfies l(abc) = l(ab)ǫ(c) − ǫ(ab)l(c) + l(bc)ǫ(a) − ǫ(bc)l(a) + l(ac)ǫ(b) − ǫ(ac)l(b), a, b, c ∈ Oalg + (2n). (ii) follows from the fact that the invertibility of the matrix B implies the invertibility of the matrix P := B + (2n) by the rule l(a∗b) = l(a∗)ǫ(b) + ǫ(a∗)l(b) +Pn(2n−1) , as defined in (i), forms a basis for VO+(2n). 2 , so that {ηi}n(2n−1) p=1 1 i=1 (iii) can be proven as follows: Theorem 3.23. Suppose L is the generator of a bi-invariant QBM on the free orthogonal group. Then L ≡ 0. Proof. Since L is bi-invariant, we have (id ⊗ L)∆(xij) = (L ⊗ id)∆(xij ) and (id ⊗ L)∆(xijxkl) = (L ⊗ id)∆(xijxkl) where i 6= j and k 6= l; comparing the coefficients in (26) and (27), we get (26) (27) L(xij) = 0 for i 6= j; L(xijxkl) = 0 for i 6= j and k 6= l; substituting k = i, l = j, (i 6= j) in the second equation, we get 0 = L(xijxkl) =Xp≥1 η(p)(xij)η(p)(xij) =Xp≥1 η(p)(xij)2, i 6= j; where ηp is an ǫ-derivation for each p. This implies η(p) ≡ 0, since η(p)(xii) = 0. Thus L becomes an ǫ-derivation. But L(xij) = 0 for i 6= j. Thus we have L ≡ 0. ✷ 29 Remark 3.24. Theorem 3.23 implies that there does not exists any quantum Brownian mo- tion on the quantum space S+ 2n−1 (i.e. the free sphere) in the sense described in subsection 3.2. ✷ 4 Exit time of Quantum Brownian motion on non-commutative torus. 4.1 Motivation and formulation We shall first recast the classical results about the assymptotics of exit time of Brownian motion in a form which will be easily generalized to the quantum set-up. Let M be a Riemannian manifold of Dimension d which is also a homogeneous space. Therefore M can be realized as K/G, where G is the isometry group of M and K is a compact subgroup of G. For m ∈ M, let Bm t denote the standard Brownian motion on M starting at m, as described in section 3.2. Let eA denote the universal enveloping von-Neumann algebra of C(M ). Let us define a map jt : eA → eA ⊗ B(L2(IP )) by: jt(f )(x, ω) := f (Bx eA, where IP denote the d-dimensional Wiener measure. r denote a ball of radius r around x ∈ M. Let τBx t (ω)), for f ∈ C(M ) and extending the map to be the exit time of the Brownian motion from the ball Bx r ∀ 0 ≤ s ≤ t}, so that we have r . Then {τBx > t} = {Bx s ∈ Bx Let Bx r χ{τBx function on the set A. In terms of the map jt, we have s ∈Bx r >t} =Vs≤t(cid:16)χ{Bx r r }(cid:17) , whereV denotes infimum and for a set A, χA denotes the indicator >t}(·) =^s≤t )(x, ·) =^s≤t ((evx ⊗ id) ◦ js(χBx js(χBx ))(·). r r χ{τ Bx r r as τBx ([0, t)) = 1 − ∧s≤t(js(χBx Now by the Wiener-Ito isomorphism (see [19]), L2(IP )e=Γ(L2(R+, Cd)). Thus we may view τBx a family of projections in eA ⊗ B(Γ(L2(R+, Cd))) defined by > t)dt =R ∞0 De(0), {(evx ⊗ 1)(cid:16)∧s≤tjs(χBx )(cid:17)}e(0)E dt, since τBx We recall from subsection 2.1, the asymptotic behaviour of IE(τBx is a positive IE(τBx random variable. Note that the points of M are in 1 − 1 correspondence with the pure states and {Pr = χBx One can slightly generalize this as follows: ) =R ∞0 IP (τBx r }r≥0 is a family of projections on eA satisfying vol(Pr) → 0 as r → 0 and evx(Pr) = 1 ∀r. Choose a sequence (xn)n ∈ M and positive numbers ǫn such that xn → x and ǫn → 0. Now (·) has the same distributioin as the random variable ) as r → 0. Now one has )) . r r r r r r r for large n0 the random variable χ χ{Bx ǫn } for each s ≥ 0. Thus, s ∈Bx {B xn s ∈B xn ǫn } IE(τ ) = IE(τBx ǫn B xn ǫn ) =Z ∞ 0 De(0), {(evxn ⊗ id)(cid:16)∧s≤tjs(χ B xn ǫn )(cid:17)}e(0)E dt 30 which implies that the asymptotic behaviour of IE(τ ) and IE(τBx ǫn ) will be the same. B xn ǫn For a non-commutative generalization of the above, we need the notion of quantum stop time. There are several formulations of this concept [1, 20, 3]. The one most suitable for us is the following: Definition 4.1. [3][Barnette] Let (At)t≥0 be an increasing family of von-Neumann algebras (called a filtration). A quantum random time or stop time adapted to the filtration (At)t≥0 is an increasing family of projections (Et)t≥0, E∞ = I such that Et is a projection in At and Es ≤ Et whenever 0 ≤ s ≤ t < +∞. Furthermore, for t ≥ s, Et ↓ Es as t ↓ s. Observe that by our definition, τBr ([0, t)) is adapted to the filtration (At)t≥0, where Suppose that we are given an E-H flow jt : A → A′′ ⊗ B(Γ(L2(R+, k0))), where A is a C∗ or von-Neumann algebra. For a projection P ∈ A, the family {1 − ∧s≤t (js(P ))}t≥0 defines a quantum At := eA ⊗ B(Γt])(cid:0)Γt] := Γ(cid:0)L2([0, t], Cn)(cid:1) (cid:1), for τBr ([0, t]) ∈ At ⊗ 1Γ[t . random time adapted to the filtration (cid:0)A′′ ⊗ B(Γt])(cid:1)t≥0 . Let us assume, furthermore, that A is Definition 4.2. We refer to the quantum random time {1 −Vs≤t js(P )}t≥0 as the 'exit time from the C∗ or von-Neumann closure of the 'smooth algebra' A∞ of a Θ-summable, admissible spectral triple and jt is a QBM on it. the projection P '. Motivated by the Propostion 2.3 and the discussion after it, we would like to formulate a quantum analogue of the exit time asymptotics and study it in concrete examples. Let τ be the non-commutative volume form corresponding to the spectral triple, and assume that we are given a family {Pn}n≥1 of projections in A, and a family {ωn}n≥1 of pure states of A such that • ωn is weak∗ convergent to a pure state ω, • ωn(Pn) = 1 for all n, • vn ≡ τ (Pn) → 0 as n → ∞. Definition 4.3. Let γn :=R ∞0 dtDe(0), (ωn ⊗ id) ◦Vs≤t js(Pn)e(0)E . We say that there is an exit time asymptotic for the family {Pn; ωn} of intrinsic dimension n0 if lim n→∞ γn 2 v m n and = ∞ if m is just less than n0 6= 0 if m 6= n = 0 if m > n γn = c1v 2 n0 n + c2v 4 n0 n + · · ·ckv 2k n0 n + O(v n 2k+1 n0 ) as n → ∞. (28) It is not at all clear whether such an asymptotic exists in general, and even if it exists, whether it is independent of the choice of the family {Pn; ωn}. If it is the case, one may legitimately think 31 of c1, c2 as geometric invariants and imitating the classical formulae (4) and (5), the extrinsic dimension d and the mean curvature H of the non-commutative manifold may be defined to be d := 1 2c1 ( n0 αn0 2 n0 + 1, ) H 2 := 8(d + 1)c2( 4 n0 . ) αn0 n0 (29) (30) 4.2 A case-study: non-commutative Torus. Fix an irrational number θ ∈ [0, 1]. We refer the reader to [[9],page 173] for a natural class of projections in C∗(T2 θ), which we will be using in this section. Let tr be the canonical trace in C∗(T2 θ), given by tr(Pm,n amnU mV n) = a00. This trace will θ) as a concrete C∗-subalgebra of B(eH), where eH denote the so-called universal enveloping θ) in B(eH). For (x, y) ∈ T2, let α(x,y) denote the canonical action θ) given by α(x,y)(Pm,n amnU mV n) = Pm,n xmynamnU mV n. For a projection P, be taken as an analogue of the volume form in C∗(T2 C∗(T2 Hilbert space for C∗(T2 i.e. the weak closure of C∗(T2 of T2 on C∗(T2 let A(t,s)(P ) := As,t(P ). Note that each α(x,y) extends as a normal automorpihsm of W ∗(T2 C∗(T2 θ), there are two conditional expectations denoted by φ1, φ2, which are defined as: θ) be the universal enveloping von-Neumann algebra of it. θ). Throughout the section, we will consider θ), and let W ∗(T2 θ). On φ1(A) :=Z 1 0 α (1,e2πit ) (A)dt, φ2(A) :=Z 1 0 α (e2πit ,1) (A)dt. By universality of W ∗(T2 Let X = {A ∈ W ∗(T2 θ), φ1, φ2 extend on W ∗(T2 θ) A = f−1(U )V −1 + f0(U ) + f1(U )V, f1, f0 ∈ L∞(T), f−1(t) := f1(t + θ)}. θ) as well. Lemma 4.4. The subspace X is closed in the ultraweak topology. 0 (U ) + f (β) −1 (U )V −1 + f (β) (U ), φ1(AβV ) = f (β) Proof. Let Aβ := f (β) Now φ1(Aβ) = f (β) map, which implies that f (β) ultraweakly convergent, to f0(U ), f1(U ), f−1(U ) (say), and clearly f−1(t) = f1(t + θ). ✷ Lemma 4.5. Suppose f1(t)f1(t + θ) = 0 and A ∈ X. Define (U )V be a convergent net in the ultraweak topology. (U ) Since φ1 is a normal −1 (U ) (all of which are elements of L∞(T)) are −1 (U ) and φ1(AβV −1) = f (β) (U ) and f (β) 0 1 1 (U ), f (β) 1 0 As,t := f−1(e2πisU )V −1e−2πit + f0(e2πisU ) + f1(e2πisU )V e2πit. Suppose s, s′ ∈ [0, 1) be such that s − s′ ≤ ǫ denotes the Lebesgue measure of a Borel subset C ⊆ R. Then As,t · As′,t′ ∈ X. 4 where 0 < ǫ < θ, and supp(f1) < ǫ, where C Proof. It suffices to show that the coefficient of V 2 in As,t · As′,t′ is zero. By a direct computation, the coefficient of V 2 is g(l) := f1(s+l)f1(s′ +l−θ)e2πi(t+t′). But (s+l)−(s′+l−θ) = θ+s−s′ > ǫ. Now by hypothesis, we have supp(f1) < ǫ, so that f1(s + l) · f1(s′ + l − θ) = 0 and hence the lemma is proved. ✷ 32 Lemma 4.6. Suppose A = f−1(U )V −1 + f0(U ) + f1(U )V and f1(l)f1(l + θ) = 0, for l ∈ [0, 1). Then A2n ∈ X, for n ∈ IN . Proof. The coefficient of V 2 in A2 is f1(l)f1(l + θ) for l ∈ [0, 1) and this is zero by the hypoethesis. Hence A2 ∈ X. The coefficient of V in A2 is f (2) 1 (l) := f1 (f0 + τθ (f0)) , where τθ is left translation by θ. We have f (2) 1 (l + θ) = 0, so that applying the same argument as before, we conclude that A4 ∈ X. Proceeding like this we get the required result. ✷ 1 (l)f (2) Lemma 4.7. Suppose P = f−1(U )V −1 + f0(U ) + f1(U )V, such that P 2 = P and supp(f1) < ǫ. Then(cid:0)As,t(P )(cid:1)V(cid:16)As′,t′ (P )(cid:17) ∈ X for s − s′ < ǫ 4 . Proof. We start with the following well-known formula due to von-Neumann: where P, Q are projections and PV Q denotes the projection onto R(P ) ∩ R(Q). Thus in particular: P ∧ Q = SOT − lim n→∞ (P · Q)n, As,t(P )^ As′ ,t′ (P ) = SOT − lim n→∞ {As,t(P ) · As′,t′ (P )}n. Now by the hypothesis, s − s′ < ǫ As,t(P ) · As′,t′ (P ) ∈ X. The coefficient of V in As,t(P ) · As′ ,t′ (P ) is 4 and supp(f1) < ǫ. It follows by Lemma 4.5, that 1 (l) := {f1(s + l)f0(s′ + t − θ)e2πit + f0(s + l)f1(s′ + t)e2πit′ f (2) }. One may check that f (2) {As,t (P ) · As′,t′ (P )}2n ∈ X for n ≥ 1. Now by Lemma 4.4, the subspace X is closed in the SOT topology. Thus 1 (l + θ) = 0 for s − s′ < ǫ 4 . Thus by Lemma 4.6, 1 (l)f (2) SOT − lim n→∞ {As,t (P ) · As′ ,t′ (P )}2n ∈ X; i.e. (cid:0)As,t(P )(cid:1)V(cid:16)As′ ,t′ (P )(cid:17) ∈ X. ✷ Lemma 4.8. Let P = f−1(U )V −1 + f0(U ) + f1(U )V and A = f (A) be projections, (f−1, f0, f1) and (f (A) Then A ≤ As,t (P ) and A ≤ As′,t′ (P ) if and only if the following hold: (U )V ) satisfying the conditions given in [[9],page 173]. −1 (U )V −1 + f (A) (U ) + f (A) −1 , f (A) , f (A) 0 1 0 1 • f1(s + l)f (A) 1 (l − θ) = 0; −1 (l + θ) = 0; • f−1(s + l)f (A) • f0(s + l)f (A) 0 (l) + f1(s + l)f (A) −1 (l − θ)e2πit + f−1(s + l)f (A) 1 (l + θ)e−2πit = f (A) 0 (l); • f1(s + l)f (A) 0 (l − θ)e2πit + f0(s + l)f (A) 1 (l) = f (A) 1 (l); • f−1(s + l)f (A) 0 (l + θ)e−2πit + f0(s + l)f (A) −1 (l) = f (A) −1 (l); 33 (l − θ) = 0; −1 (l + θ) = 0; 1 • f1(s′ + l)f (A) • f−1(s′ + l)f (A) • f0(s′ + l)f (A) 0 • f1(s′ + l)f (A) • f−1(s′ + l)f (A) 0 0 (l) + f1(s′ + l)f (A) −1 (l − θ)e2πit′ + f−1(s′ + l)f (A) 1 (l + θ)e−2πit′ = f (A) 0 (l); (l − θ)e2πit′ + f0(s′ + l)f (A) 1 (l) = f (A) 1 (l); (l + θ)e−2πit′ + f0(s′ + l)f (A) −1 (l) = f (A) −1 (l); for l ∈ [0, 1). Proof. It follows by comparing the coefficients of V −1, V and 1 from the equations As,t(P )A = A; As′ ,t′ (P )A = A. ✷ Lemma 4.9. For two projections A and B such that A = f (A) −1 (U )V −1 + f (A) 0 B = f (B) −1 (U )V −1 + f (B) 0 we have A ≤ B if and only if • f (B) 1 (l)f (A) 1 (l − θ) = 0; • f (B) 1 (l + θ)f (A) 1 (l + 2θ) = 0; (U ) + f (A) 1 (U )V, (U ) + f (B) 1 (U )V ; • f (B) 0 (l)f A 0 (l) + f (B) 1 (l)f (A) 0 (l) + f (B) 1 (l + θ)f (A) 1 (l + θ) = f (A) 0 (l); • f (B) 1 (l)f (A) 0 (l − θ) + f (B) 0 (l)f (A) 1 (l) = f (A) 1 (l); • f (B) 1 (l + θ)f (A) 0 (l + θ) + f (B) 0 (l)f (A) 1 (l + θ) = f (A) 1 (l + θ); for l ∈ [0, 1). Proof. It follows by comparing the coefficients of V, V −1, 1 in the equation BA = A. ✷ Lemma 4.10. Let P = f−1(U )V −1 + f0(U ) + f1(U )V such that P is a projection and suppose f0(t) = 0 for some t. Then f1(t) = f1(t + θ) = 0. Proof. The fact that P 2 = P implies that f0(t) − (f0(t))2 = f1(t − θ)2 + f1(t)2 (see [9],page 173), f0(t + θ) − (f0(t + θ))2 = f1(t)2 + f1(t + θ)2. (31) The first expression in (31) implies that f1(t) = 0. Moreover we have f1(t + θ) (1 − f0(t) − f0(t + θ)) = 0 [9, page173]; so that if f0(t + θ) = 0 implies f1(t + θ) = 0; else if f0(t + θ) = 1, the second expression in (31) gives f1(t + θ) = 0. ✷ 34 f0(t) =  ǫ−1t if 0 ≤ t ≤ ǫ 1 if ǫ ≤ t ≤ θ ǫ−1(θ + ǫ − t) if θ ≤ t ≤ θ + ǫ 0 if θ + ǫ ≤ t ≤ 1 f1(t) =(pf0(t) − f0(t)2 if θ ≤ t ≤ θ + ǫ 0 if otherwise. For a set A ⊆ R and real numbers a ∈ R, τa(A) := A + a. Define functions f0 and f1 by: It is known (see [9]) that P := f−1(U )V −1 + f0(U ) + f1(U )V is a projection in C∗(T2 θ). Theorem 4.11. Let P = f−1(U )V −1 + f0(U ) + f1(U )V be a projection with f0, f1 as described above. Consider the projections As,t(P ), As′ ,t′ (P ) such that s − s′ < ǫ 4 . Then (cid:0)As,t (P )(cid:1)^(cid:16)As′,t′ (P )(cid:17) = χS (U ), for the set S = X1 ∩ X2 ∩ X3 ∩ X4, where X1 = τ−s({xf1(x) = 0}), X2 := τ−s′({xf1(x) = 0}), X3 := τ−s({xf0(x) = 1}) and X4 := τ−s′({xf0(x) = 1}). Proof. The hypothesis of the theorem and Lemma 4.7 together implies that (cid:0)As,t(P )(cid:1)^(cid:16)As′ ,t′ (P )(cid:17) ∈ X. (U ) = χS (U ). Thus B ≤ As,t(P ), B ≤ As′ ,t′ (P ). Again if A = f (A) Let B = χS (U ). Then it follows that the conditions of Lemma 4.8 hold with f (A) −1 (U )V −1+f (A) f (A) (U )V 0 be a projection, then it may be easily observed that A ≤ As,t(P ) and A ≤ As′,t′ (P ) together with Lemma 4.10 implies that f1, f0 is zero outside S. An application of Lemma 4.9 implies that f1, f0 must vanish outside S if and only if A ≤ B. Hence the theorem is proved. 1 = 0 and (U )+f (A) 0 1 ✷ It is worthwhile to note that the conclusion of the above theorem holds if we replace U by U k, V by V k, and θ by {kθ} ({·} denoting the fractional part). Let Pn = f (kn) −1 (U kn) + f (kn) 0 (U kn) + f (kn) such that {knθ} → 0. Put ǫn := {knθ}2 (W (1) ). Define jt : W ∗(T2 , W (2) θ) → W ∗(T2 t t 1 (U kn)U kn, be projections as described in [9, page 173] . Consider a standard Brownian motion in R2, given by θ) ⊗ B(Γ(L2(R+, C2))) by jt(·) := α (·). (1) 2πiW t (e (2) 2πiW t ) ,e Theorem 4.12. Almost surely, Vs≤t (js(Pn)(ω)) ∈ W ∗(U ), for all n, i.e. (js(Pn)) ∈ W ∗(U ) ⊗ B(Γ(L2(R+, C2))), ^s≤t for each n. 35 Proof. In the strong operator topology, ^0≤s≤t (js(Pn)) = lim m→∞^i {j it 2m (Pn) ∧ j (i+1)t 2m (Pn)}. (32) Now almost surely a Brownian path restricted to [0, t] is uniformly continuous, so that the for sufficiently large m, and for almost all ω, W (1) can be made small, uniformly for all i it 2m (Pn)} ∈ W ∗(U ) ∩ X by Theorem 4.11. Now Lemma (Pn) ∧ j (i+1)t 2m − W (1) 2m (i+1)t 2m 4.4 implies that W ∗(U ) ∩ X is closed in the WOT-topology. Thus such that i = 0, 1, ..2m. SoVi{j it m→∞^i lim {j it 2m (Pn) ∧ j (i+1)t 2m (Pn)} ∈ W ∗(U ) ∩ X. ✷ Let zn = e 2πi 3{knθ} 4 pure states on C∗(T2 in the beginning, consider . Consider the sequence of states φzn := evzn ◦E1. By [17], this is a sequence of θ) converging in the weak-∗ topology to φ1 := ev1 ◦ E1. Following the discussion e(0), (φzn ⊗ 1) ◦ ^0≤s≤t (js(Pn))e(0)+ . A direct computation shows that this is equal to IP {e2πiW (1) s ∈ B, 0 ≤ s ≤ t} = IP {τ [ −{knθ} 4 , {knθ} 4 ] > t}, where B := {e2πix : x ∈ [−{knθ}4 , {knθ}4 ]}. So we have a family of (τn)n random times defined by τn([t, +∞)) = ^0≤s≤t (js(Pn)); 0De(0), (φzn ⊗ 1) ◦V0≤s≤t(js(Pn))e(0)E dt can be taken as the expectation of the random time τn. Note that here the analogue for balls of decreasing volume is (Pn)n, such that tr(Pn) = {knθ} → 0, tr being the canonical trace in W ∗(T2 θ). Now, by Proposition 2.3, we have so thatR t 0 e(0), (φzn ⊗ 1) ◦ ^0≤s≤t Z t 8 (cid:19) + = 2 sin2(cid:18) {knθ} = IE(τ −{knθ} 2 3 {knθ} 4 ) ] 4 [ , = {knθ}2 25 + {knθ}4 211.3 (js(Pn))e(0)+ dt sin4(cid:18) {knθ} 8 (cid:19) + O(cid:18)sin5(cid:18) {knθ} 8 (cid:19)(cid:19) + O({knθ}5), since the mean curvature of the circle viewed inside R2 is 1. (33) 36 1 2√2 Remark 4.13. In view of equations (4),(29) and (30), we see that the 'intrinsic dimension' n0 = 1, the 'extrinsic diimension' d = 5, and the 'mean curvature' is . As we have already remarked in the introduction, the instrinsic one-dimensionality may be interpreted as a manifestation of the local one-dimensionality of the 'leaf space' of the Kronecker foliation (see [7] for details). It is worth pointing out that the spectral behaviour of the standard Dirac operator or the Laplacian coming from it for this noncommutative manifold is identical with that of the commutative two-torus, and thus it does not recognize the one-dimensionality of the leaf space of Kronecker foliation. Thus, it is a remarkable success of our (quantum) stochastic analysis using exit time to reveal the association of the noncommutative geometry of Aθ with the leaf space of Kronecker foliation, and also to distinguish it from the commutative two-torus. All these give a good justification for developing a general theory of quantum stochastic geometry. References [1] S. Attal and K. B. Sinha. Stopping semimartingales on Fock space. In Quantum probability communications, QP-PQ, X, pages 171 -- 185. World Sci. Publ., River Edge, NJ, 1998. [2] Teodor Banica, Debashish; Goswami. Quantum isometries and noncommutative spheres. Comm. Math. Phys., (1), 2009. [3] Chris Barnett and Ivan F. Wilde. Quantum stopping-times. In Quantum probability & related topics, QP-PQ, VI, pages 127 -- 135. World Sci. Publ., River Edge, NJ, 1991. [4] Jyotishman Bhowmick. Quantum isometry groups. Phd Thesis. http://arxiv.org/abs/0907.0618v1, 2010. [5] Jyotishman Bhowmick and Debashish Goswami. Quantum isometry groups: examples and computations. Comm. Math. Phys., 285(2):421 -- 444, 2009. [6] Berndt A. Brenken. Representations and automorphisms of the irrational rotation algebra. Pacific J. Math., 111(2):257 -- 282, 1984. [7] Alain Connes. Noncommutative geometry. Academic Press Inc., San Diego, CA, 1994. [8] Alain Connes and Michel Dubois-Violette. Noncommutative finite-dimensional manifolds. I. Spherical manifolds and related examples. Comm. Math. Phys., 230(3):539 -- 579, 2002. [9] Kenneth R. Davidson. C∗-algebras by example, volume 6 of Fields Institute Monographs. American Mathematical Society, Providence, RI, 1996. [10] Uwe Franz. The Theory of Quantum Levy Processes. Habilitation thesis EMAU Greifswald. arXiv:math/0407488v1, 2004. [11] Debashish Goswami. Quantum group of isometries in classical and noncommutative geometry. Comm. Math. Phys., 285(1):141 -- 160, 2009. [12] Alfred Gray. The volume of a small geodesic ball of a Riemannian manifold. Michigan Math. J., 20:329 -- 344 (1974), 1973. 37 [13] Kiyosi Ito. Brownian motions in a Lie group. Proc. Japan Acad., 26(8):4 -- 10, 1950. [14] M. Liao and W. A. Zheng. Radial part of Brownian motion on a Riemannian manifold. Ann. Probab., 23(1):173 -- 177, 1995. [15] Ming Liao. L´evy processes in Lie groups, volume 162 of Cambridge Tracts in Mathematics. Cambridge University Press, Cambridge, 2004. [16] Ann Maes and Alfons Van Daele. Notes on compact quantum groups. Nieuw Arch. Wisk. (4), 16(1-2):73 -- 112, 1998. [17] Debashish Goswami and Lingaraj Sahu. Invariants for Normal Completely Positive Maps on the Hyperfinite II1 Factor. http://arxiv.org/abs/math/0601358, Proc. Ind. Acad. Sci. (Math. Sci.). [18] A. Mohari and Kalyan B. Sinha. Quantum stochastic flows with infinite degrees of freedom and countable state Markov processes. Sankhy¯a Ser. A, 52(1):43 -- 57, 1990. [19] K. R. Parthasarathy. An introduction to quantum stochastic calculus, volume 85 of Monographs in Mathematics. Birkhauser Verlag, Basel, 1992. [20] K. R. Parthasarathy and Kalyan B. Sinha. Stop times in Fock space stochastic calculus. In Proceedings of the 1st World Congress of the Bernoulli Society, Vol. 1 (Tashkent, 1986), pages 495 -- 498, Utrecht, 1987. VNU Sci. Press. [21] K. R. Parthasarathy and V. S. Sunder. Exponentials of indicator functions are total in the boson Fock space Γ(L2[0, 1]). In Quantum probability communications, QP-PQ, X, pages 281 -- 284. World Sci. Publ., River Edge, NJ, 1998. [22] Mark A. Pinsky. Mean exit time from a bumpy sphere. Proc. Amer. Math. Soc., 122(3):881 -- 883, 1994. [23] Piotr Podle´s. Symmetries of quantum spaces. Subgroups and quotient spaces of quantum SU(2) and SO(3) groups. Comm. Math. Phys., 170(1):1 -- 20, 1995. [24] Marc A. Rieffel. Deformation quantization for actions of Rd. Mem. Amer. Math. Soc., 106(506):x+93, 1993. [25] Michael Schurmann. White noise on bialgebras, volume 1544 of Lecture Notes in Mathematics. Springer-Verlag, Berlin, 1993. [26] Kalyan B. Sinha and Debashish Goswami. Quantum stochastic processes and noncommutative geometry, volume 169 of Cambridge Tracts in Mathematics. Cambridge University Press, Cambridge, 2007. [27] M Skeide. Indicator functions of intervals are totalizing in the symmetric fock space. in (Accardi, L. Kuo, H.-H. Obata, N., Saito, K., Si Si, Streit, L., eds.) Trends in Contempo- rary Infinite Dimensional Analysis and Quantum Probability, Volume in honour of Takeyuki Hida, Istituto Italiano di Cultura (ISEAS), Kyoto 2000 (Rome, Volterra-Preprint 1999/0395), (0395), 1999. 38 [28] Daniel W. Stroock and S. R. Srinivasa Varadhan. Multidimensional diffusion processes. Classics in Mathematics. Springer-Verlag, Berlin, 2006. Reprint of the 1997 edition. [29] Shuzhou Wang. Deformations of compact quantum groups via Rieffel's quantization. Comm. Math. Phys., 178(3):747 -- 764, 1996. [30] Kosaku Yosida. A characterization of the second order elliptic differential operators. Proc. Japan Acad., 31:406 -- 409, 1955. 39
1208.6162	1	1208	2012-08-30T13:13:33	Z is universal	[ "math.OA" ]	We use order zero maps to express the Jiang-Su algebra Z as a universal C-algebra on countably many generators and relations, and we show that a natural deformation of these relations yields the stably projectionless algebra W studied by Kishimoto, Kumjian and others. Our presentation is entirely explicit and involves only -polynomial and order relations.	math.OA	math	Z IS UNIVERSAL BHISHAN JACELON AND WILHELM WINTER Abstract. We use order zero maps to express the Jiang-Su algebra Z as a universal C∗-algebra on countably many generators and relations, and we show that a natural deformation of these re- lations yields the stably projectionless algebra W studied by Kishimoto, Kumjian and others. Our presentation is entirely explicit and involves only ∗-polynomial and order relations. 1. Introduction In Elliott's programme to classify simple, nuclear C∗-algebras using K-theoretic invariants, the Jiang-Su algebra Z plays a particularly prominent role (see [18]). While there are various ways of characterizing Z (see for example [4] and [14]), its most concise description (due to the second named author, in [20]) is as the unique initial object in the category of strongly self-absorbing C∗-algebras. Here, a separable, unital C∗-algebra D 6= C is strongly self-absorbing if there is an isomorphism ϕ : D → D ⊗ D that is approximately unitarily equivalent to the first factor embedding, cf. [16]. The statement that Z is an initial object in this category is equivalent to saying that every strongly self-absorbing C∗-algebra absorbs Z tensorially (i.e. is 'Z-stable'). Apart from Z, the known strongly self-absorbing algebras are: the Cuntz algebras O2 and O∞, UHF algebras of infinite type, and such UHF algebras tensored with O∞. These all admit presentations as universal C∗-algebras (see Section 5 for a discussion), and Theorem 3.1 of this article provides such a description for Z which, although complicated, is explicit and algebraic in the sense that it involves only ∗-polynomial and order relations. The proof relies on the 'order zero' presentations of prime dimension drop algebras described in [14] (see Section 2), and gives a construction of Z as an inductive limit of such algebras with connecting maps defined in terms of generators and relations. The Jiang-Su algebra may be thought of as a stably finite analogue of O∞, and the C∗-algebra W constructed in [3] (and studied in another form in [5]) has been similarly proposed as a stably finite analogue of O2. The conjecture that W ⊗ W ∼= W, while still open, is known to have interesting consequences. For example, it is shown in [3] that among the C∗-algebras classified in [11], those that are simple and have trivial K-theory would absorb W tensorially. On the other hand, L. Robert proves in [12] that the Cuntz semigroup of a W-stable C∗-algebra is determined by the cone of its lower semicontinuous 2-quasitraces. These results indicate that W may play an important role in the classification of nuclear, stably projectionless C∗-algebras. In this article, we examine the structure of W rather than its role in classification, by showing in Theorem 4.3 how to obtain W as a nonunital deformation of Z. The paper is organized as follows. In Section 2 we establish notation and recall various basic facts about order zero maps and dimension drop algebras. Section 3 contains the presentation of Z as a universal C∗-algebra (Theorems 3.1 and 3.3), and Section 4 contains the corresponding description of W (Theorem 4.3). We conclude with some open questions in Section 5. Date: May 29, 2018. 2000 Mathematics Subject Classification. 46L35, 46L05. Key words and phrases. Jiang-Su algebra, strongly self-absorbing C∗-algebra, stably projectionless C∗-algebra, order zero map, classification. Research supported by the DFG through SFB 878, by the ERC through AdG 267079 and by the EPSRC through EP/G014019/1 and EP/I019227/1. 1 2 BHISHAN JACELON AND WILHELM WINTER Acknowledgements. The first author would like to thank Simon Wassermann, Stuart White, Rob Archbold and Ulrich Krahmer for carefully reading the version of this article that appeared in his doctoral thesis. 2. Preliminaries In this section, we collect some well-known facts about order zero maps and dimension drop algebras that are used throughout the article. (Detailed exposition of order zero maps can be found in [21] and [22].) We denote by eij (or e(n) ij ) the canonical (i, j)-th matrix unit in Mn = Mn(C). Recall that a completely positive (c.p.) map ϕ : B → A has order zero if it preserves orthogonality. Every completely positive and contractive (c.p.c.) order zero map ϕ : B → A (for B unital) is of the form ϕ(·) = πϕ(·)ϕ(1B) = ϕ(1B)πϕ(·) for a ∗-homomorphism πϕ : B → A∗∗ called the supporting ∗-homomorphism of ϕ. We frequently use the notion of positive functional calculus provided by this decomposition: if f ∈ C0(0, 1] is positive with kfk ≤ 1 then the map f (ϕ) : B → A given by f (ϕ)(·) := πϕ(·)f (ϕ(1B)) is a well-defined c.p.c. order zero map. It is easy to see that if p ∈ B is a projection, then f (ϕ)(p) = f (ϕ(p)). On the other hand, if ϕ(1B) is a projection, then ϕ is in fact a ∗-homomorphism. Finally, c.p.c. order zero maps B → A correspond bijectively to ∗-homomorphisms C0((0, 1], B) → A. For B = Mn, one way of interpreting this fact is to say that the cone C0((0, 1], Mn) is the universal C∗- algebra generated by a c.p.c. order zero map on Mn. Equivalently, it is easy to check that C0((0, 1], Mn) is the universal C∗-algebra on generators x1, . . . , xn subject to the relations R(0) n given by kxik ≤ 1, x1 ≥ 0, xix∗ i = x2 1, x∗ j xj ⊥ x∗ i xi for 1 ≤ i 6= j ≤ n i xj). One can therefore view (2.1) (for example by mapping xj to t1/2 ⊗ e1j, so that t ⊗ eij corresponds to x∗ the statement C0((0, 1], Mn) = C∗(ϕ ϕ c.p.c. order zero on Mn) (2.2) as an abbreviation for these relations. Remark 2.1. In the case n = 2, C0((0, 1], M2) is the universal C∗-algebra C∗(x kxk ≤ 1, x2 = 0). Therefore, if A is a C∗-algebra and v ∈ A is a contraction with v2 = 0, then there is a unique c.p.c. order zero map ψ : M2 → A with ψ1/2(e12) = v (so that ψ(e11) = vv∗ and ψ(e22) = v∗v). By a prime dimension drop algebra, we mean a C∗-algebra of the form Zp,q := {f ∈ C([0, 1], Mp ⊗ Mq) f (0) ∈ Mp ⊗ 1q, f (1) ∈ 1p ⊗ Mq}, (2.3) where p and q are coprime natural numbers. The Jiang-Su algebra Z is the unique inductive limit of prime dimension drop algebras which is simple and has a unique tracial state (see [4]). The order zero notation (2.2) essentially appears in [14, Proposition 2.5], where the presentation of prime dimension drop algebras described in [4, Proposition 7.3] is reinterpreted in terms of order zero maps. Specifically, the prime dimension drop algebra Zp,q is the universal unital C∗-algebra C∗(α, β α c.p.c. order zero on Mp, β c.p.c. order zero on Mq, α(1p) + β(1q) = 1, [α(Mp), β(Mq)] = 0), with generators corresponding to the obvious embeddings of C0([0, 1), Mp) and C0((0, 1], Mq) into Zp,q. When q = p + 1, there is another presentation of Zp,p+1 in terms of order zero maps that does not involve a commutation relation. The following is essentially contained in [14, Proposition 5.1], and we note that these relations have already proved highly useful, for example in [19], [21], [15] and [8]. Proposition 2.2. Let Z (n) be the universal unital C∗-algebra C∗(ϕ, ψ Rn), where Rn denotes the set of relations: (i) ϕ and ψ are c.p.c. order zero maps on Mn and M2 respectively; (ii) ψ(e11) = 1 − ϕ(1n); (iii) ψ(e22)ϕ(e11) = ψ(e22). Then Z (n) ∼= Zn,n+1. In Section 3, we use Proposition 2.2 to write Z as a limit of dimension drop algebras in a universal way. We make analogous use of Proposition 4.1, a nonunital version of Proposition 2.2, to present W. Z IS UNIVERSAL 3 3. Generators and relations for the Jiang-Su algebra (Z (q(k)), αk). In this section, we will construct an inductive system (Z (q(k)), αk), where q(k) = p3k for some fixed p ≥ 2 (p = 2 will do) and Z (q(k)) = C∗(ϕk, ψk Rq(k)) ∼= Zq(k),q(k)+1 (as in Proposition 2.2), and we will check that the inductive limit is simple with a unique tracial state. It will then follow from the classification theorem of [4] that Z ∼= lim −→ If this procedure is to provide an explicit presentation of Z as a universal C∗-algebra, we need to be able to describe the connecting maps αk in terms of generators and relations. (This is perhaps the key difference between our construction and the original construction of Z as an inductive limit in [4].) In other words, for every k ∈ N we will find c.p.c. order zero maps ϕk : Mq(k) → Z (q(k+1)) and ψk : M2 → Z (q(k+1)) that satisfy the relations Rq(k) of Proposition 2.2. By universality, we will then have unital connecting maps αk : Z (q(k)) → Z (q(k+1)) with αk ◦ ϕk = ϕk and αk ◦ ψk = ψk. Before giving the connecting maps, it is instructive to note that there are obvious choices for ϕk and ψk. Since q(k + 1) = q(k)3, we can identify Mq(k+1) with Mq(k) ⊗ Mq(k) ⊗ Mq(k) (and e(q(k+1)) with e(q(k)) ). We could then set ϕk = ϕk+1 ◦ (idMq(k) ⊗ 1q(k) ⊗ 1q(k)) and ψk = ψk+1; it is easy to see that these maps satisfy the relations Rq(k), but the corresponding inductive limit certainly would not be simple. The idea is therefore to define ϕk in such a way as to ensure that [0, 1] is chopped up into suitably small pieces under the induced ∗-homomorphism αk; ψ1/2 (e12) will then be some partial-isometry-like element that facilitates the relations Rq(k). 11 ⊗ e(q(k)) 11 ⊗ e(q(k)) 11 11 k One way of doing this is as follows. Define ρk : Mq(k) → Mq(k+1) by ρk = (idMq(k) ⊗ 1q(k)−1 ⊗ 1q(k)) ⊕ Mi=1  q(k) i q(k) (cid:0)idMq(k) ⊗ eq(k),q(k) ⊗ eii(cid:1)  . (3.1) Note that ρk is c.p.c. order zero, with supporting ∗-homomorphism πρk = idMq(k) ⊗ 1q(k) ⊗ 1q(k). We may then define ϕk := ϕk+1 ◦ ρk. For this to work, we need to be able to transport the defect 1 − ϕk+1(ρk(1q(k))) = (1 − ϕk+1(1q(k+1))) + ϕk+1(1q(k+1) − ρk(1q(k))) underneath ϕk+1(ρk(e(q(k)) )), and the basic idea is to do this in two steps. Step 1. Use ψk+1(e12) to transport the corner πψk+1 (e11)(1 − ϕk+1(ρk(1q(k))))πψk+1 (e11) underneath 11 πψk+1 (e22)ϕk+1(e(q(k+1)) 11 )πψk+1 (e22) ≤ ϕk+1(e(q(k+1)) 11 ) ≤ ϕk+1(ρk(e(q(k)) 11 )). Step 2. Use a partial isometry vk+1 ∈ Mq(k+1) to transport (a projection bigger than) 1q(k+1)−ρk(1q(k)) , so that ϕk+1(vk+1) transports underneath (a projection smaller than) ρk(e(q(k)) the rest of 1 − ϕk+1(ρk(1q(k))) underneath ϕk+1(ρk(e(q(k)) ) − e(q(k+1)) )) − ϕk+1(e(q(k+1)) Although this is essentially the right idea, it needs fine-tuning in the guise of functional calculus. We achieve this in Theorem 3.3 by adjusting the relations for Z (q(k)), while for Theorem 3.1, we modify ϕk and ψk using the following piecewise linear functions: ). 11 11 11 11 ✓ ✓ 3 16 ✻ 1 ✓ ✓ ✻ 1 d g ❅ ❅ ❅ ❅ 1 4 1 2 3 4 ✻ 1 ✻ 1 ✲ 1 ✲ 1 f 1 2 3 4 1 4 h 1 4 1 2 3 4 ✲ 1 ✲ 1 These are chosen so that, writing ¯d(t) = d(1 − t), we have g = f − h, hf = h, (1 − f ) ¯d = 1 − f and g ¯d = g. (3.2) 4 BHISHAN JACELON AND WILHELM WINTER For use in Section 4, we also note that if d is the function d(t) = d(t(1 − t)) then we have (f − f 2) d = f − f 2 and g d = g. Finally, to accomplish Step 2, we choose a partial isometry vk+1 ∈ Mq(k+1) such that and vk+1v∗ k+1 = 1q(k) ⊗ eq(k),q(k) ⊗ 1q(k)−1 (3.3) (3.4) v∗ k+1vk+1 = (e11 ⊗ 1q(k)−1 ⊗ 1q(k)) + (e11 ⊗ eq(k),q(k) ⊗ eq(k),q(k)) − (e11 ⊗ e11 ⊗ e11). This is possible since both of these projections have rank q(k)2 − q(k); since they are orthogonal, we moreover have v2 k+1 = 0. This vk+1 then satisfies: (i) v∗ (ii) v∗ (iii) vk+1v∗ k+1vk+1 ⊥ e11 ⊗ e11 ⊗ e11 = e(q(k+1)) k+1vk+1 is dominated by ρk(e(q(k)) 11 11 k+1 acts like a unit on (in fact, vk+1v∗ ) (and therefore by ρk(e(q(k)) k+1 is orthogonal to e(q(k+1)) ); and ) − e(q(k+1)) 11 11 11 too); 1q(k+1) − ρk(1q(k)) = q(k) Mi=1 (cid:18)1 − i q(k)(cid:19) (1q(k) ⊗ eq(k),q(k) ⊗ eii). (3.5) Theorem 3.1. Let the functions d, f, g, h ∈ C0(0, 1], the partial isometries vk+1 ∈ Mq(k+1), and the c.p.c. order zero maps ρk : Mq(k) → Mq(k+1) be as above for each k ∈ N. Define ZU to be the universal unital C∗-algebra generated by c.p.c. order zero maps ϕk on Mq(k) (k ∈ N) and ψk on M2 (k ∈ N) such that for each k, these maps satisfy the relations Rq(k), i.e. and ψk(e11) = 1 − ϕk(1q(k)) ψk(e22)ϕk(e11) = ψk(e22), together with the additional relations Sq(k) given by ϕk = f (ϕk+1) ◦ ρk, (3.6) (3.7) (3.8) d(ψk+1)(e12) (3.9) ψ1/2 k (e12) = (cid:0)1 − f (ϕk+1)(1q(k+1)) + g(ϕk+1)(1q(k+1) − ρk(1q(k)))(cid:1)1/2 + h(ϕk+1)(1q(k+1) − ρk(1q(k)))1/2f (ϕk+1)(vk+1). Then ZU ∼= Z. Proof. For each k, define ϕk : Mq(k) → Z (q(k+1)) = C∗(ϕk+1, ψk+1 Rq(k+1)) and ψk : M2 → Z (q(k+1)) by (3.10) ϕk = f (ϕk+1) ◦ ρk and where and ψ1/2 k (e12) = γk + δk, (3.11) γk := (cid:0)1 − f (ϕk+1)(1q(k+1)) + g(ϕk+1)(1q(k+1) − ρk(1q(k)))(cid:1)1/2 d(ψk+1)(e12) (3.12) δk := h(ϕk+1)(cid:0)1q(k+1) − ρk(1q(k))(cid:1)1/2 (3.13) We need to check that ϕk and ψk satisfy the relations Rq(k). First, it is obvious that ϕk is c.p.c. order zero since ϕk+1 and ρk are, and f is contractive. Next, to show that (3.11) genuinely defines a c.p.c. order zero map ψk, it suffices to check that γk + δk is a contraction that squares to zero (see Remark 2.1). In fact, this would follow automatically from the relations (3.6) and (3.7) for ϕk and ψk (where, for the moment, we interpret ψk(e11) and ψk(e22) as notation for ψ1/2 (e12)∗ and f (ϕk+1)(vk+1). (e12) ψ1/2 k k Z IS UNIVERSAL 5 (e12)∗ ψ1/2 ψ1/2 would imply that k k (e12) respectively). Indeed, 1 − ϕk(1q(k)) is certainly a contraction, and (3.6) and (3.7) ψk(e22) ψk(e11) = ψk(e22)(1 − ϕk(1q(k))) = ψk(e22) − n Xi=1 ψk(e22) ϕk(e11) ϕk(eii) = 0, (3.14) k (e12)(cid:17)2 = 0. Let us now check that ϕk and ψk really do satisfy these relations. and hence that (cid:16) ψ1/2 Claim 1: ψk(e11) = 1 − ϕk(1q(k)). Proof of Claim 1. First note that, using (3.7) and property (i) of the partial isometry vk+1, we have d(ψk+1)(e12)f (ϕk+1)(v∗ k+1) = d1/2(ψk+1)(e12)d1/2(ψk+1)(e22)ϕk+1(e11)f (ϕk+1)(v∗ k+1vk+1v∗ k+1) = 0. Therefore, the cross terms γkδ∗ k and δkγ ∗ k in the expansion of ψk(e11) = ψ1/2 k (e12) ψ1/2 k (e12)∗ vanish. Using the fact that f h = h, and property (iii) of vk+1, we have h(ϕk+1)(1q(k+1) − ρk(1q(k)))f (ϕk+1)(vk+1)f (ϕk+1)(v∗ k+1) = h(ϕk+1)((1q(k+1) − ρk(1q(k))vk+1v∗ k+1) = h(ϕk+1)(1q(k+1) − ρk(1q(k))). Thus, δkδ∗ k = h(ϕk+1)(1q(k+1) − ρk(1q(k))). From (3.6) we have d(ψk+1)(e11) = d(ψk+1(e11)) = d(1 − ϕk+1(1q(k+1))) = ¯d(ϕk+1(1q(k+1))), where ¯d(t) = d(1 − t) as in (3.2), whence we also obtain (1 − f (ϕk+1)(1q(k+1)))d(ψk+1)(e11) = (1 − f )(ϕk+1(1q(k+1))) ¯d(ϕk+1(1q(k+1))) = (1 − f )(ϕk+1(1q(k+1))) = 1 − f (ϕk+1)(1q(k+1)). Similarly, we have g(ϕk+1)(1q(k+1))d(ψk+1)(e11) = g(ϕk+1)(1q(k+1)), hence We therefore have γkγ ∗ follows that g(ϕk+1)(1q(k+1) − ρk(1q(k)))d(ψk+1)(e11) = g(ϕk+1)(1q(k+1) − ρk(1q(k))). k = 1 − f (ϕk+1)(1q(k+1)) + g(ϕk+1)(1q(k+1) − ρk(1q(k))). Since g + h = f , it ψk(e11) = γkγ ∗ k + δkδ∗ k = 1 − f (ϕk+1)(1q(k+1)) + g(ϕk+1)(1q(k+1) − ρk(1q(k))) + h(ϕk+1)(1q(k+1) − ρk(1q(k))) = 1 − f (ϕk+1)(ρk(1q(k))) = 1 − ϕk(1q(k)). Claim 2: ψk(e22) ϕk(e11) = ψk(e22). Proof of Claim 2. Since f h = h and vk+1 is a partial isometry with property (ii), we have h(ϕk+1)(1q(k+1)−ρk(1q(k)))f (ϕk+1)(vk+1)f (ϕk+1)(ρk(e11)) = h(ϕk+1)((1q(k+1) − ρk(1q(k)))vk+1) = h(ϕk+1)(1q(k+1) − ρk(1q(k)))f (ϕk+1)(vk+1). Thus, δk ϕk(e11) = δk. Next, it follows from (3.7), upon approximating d1/2 and f uniformly by polynomials, that d1/2(ψk+1)(e22)f (ϕk+1)(e11) = f (1)d1/2(ψk+1)(e22) = d1/2(ψk+1)(e22). 11 11 ) − e(q(k+1)) ⊥ (ρk(e(q(k)) Since e(q(k+1)) d1/2(ψk+1)(e22)f (ϕk+1)(ρk(e11)) = d1/2(ψk+1)(e22), d(ψk+1)(e12). Therefore, γk ϕk(e11) = γk, and so ψk(e22) ϕk(e11) = (γ ∗ is order zero, we therefore have hence d(ψk+1)(e12)f (ϕk+1)(ρk(e11)) = k)(γk + δk) ϕk(e11) = ψk(e22). We have now shown that ϕk and ψk satisfy the relations Rq(k). This means that, for any k ∈ N, (3.8) and (3.9) do not introduce any new relations on ϕk+1 and ψk+1; thus, the sub-C∗-algebra generated ) and f (ϕk+1) k + δ∗ 11 6 BHISHAN JACELON AND WILHELM WINTER by ϕk+1 and ψk+1 within ZU is isomorphic to the universal C∗-algebra on relations Rq(k+1) (that is, to Z (q(k+1))), and moreover contains the sub-C∗-algebra generated by ϕk and ψk. Therefore, by Proposition 2.2, ZU is isomorphic to an inductive limit of prime dimension drop algebras. The strategy for the remainder of the proof is to pass from the abstract picture of ZU as a uni- (Z (q(k)), αk), where the (unital) versal C∗-algebra, to a concrete description as an inductive limit lim −→ connecting maps αk : Z (q(k)) → Z (q(k+1)) are determined by (3.8) and (3.9) (i.e. αk ◦ ϕk = ϕk and αk ◦ ψk = ψk). We will obtain explicit descriptions of the maps αk, and use these to show that ZU is simple and has a unique tracial state. For each k ∈ N, let us fix an identification of Z (q(k)) with Zq(k),q(k)+1 via the order zero map Mq(k) → Zq(k),q(k)+1 (which, abusing notation, we also call ϕk) defined by: ϕk(a)(t) = uk(t)(a ⊗ 1q(k))uk(t)∗ ⊕ (1 − t)(a ⊗ eq(k)+1,q(k)+1) (3.15) for a ∈ Mq(k) and t ∈ [0, 1]. (Here, uk is a unitary in the algebra C([0, 1], Mq(k) ⊗ Mq(k)), included nonunitally in the top left corner of C([0, 1], Mq(k) ⊗ Mq(k)+1), with uk(0) = 1 and uk(1) implementing the flip in Mq(k) ⊗ Mq(k).) It is easy to write down a suitable ψk, but for the purpose of computing the connecting map Zq(k),q(k)+1 → Zq(k+1),q(k+1)+1 (also called αk), this is not necessary. k for the map evt ◦ αk : Zq(k),q(k)+1 → Mq(k+1) ⊗ Mq(k+1)+1, where evt denotes evaluation at t. Then αt k is a finite-dimensional representation of Zq(k),q(k)+1, so is a direct sum of finitely many irreducible representations πt m(t) of Zq(k),q(k)+1 (corresponding up to unitary equivalence and, at the endpoints, up to multiplicity, to point evaluations). Since C∗(ϕk(1q(k))) ⊂ Zq(k),q(k)+1 separates the points of [0, 1], it is easy to see that the unitary equivalence classes of πt For each t ∈ [0, 1], let us write αt m(t) can be determined by computing αt k(ϕk(1q(k))). To do this, note that 1, . . . , πt 1, . . . , πt f (ϕk+1)(b)(t) = uk+1(t)(b ⊗ 1q(k+1))uk+1(t)∗ ⊕ f (1 − t)(b ⊗ eq(k+1)+1,q(k+1)+1) (3.16) for b ∈ Mq(k+1), and recall the definition (3.1) of ρk. We then have, for a ∈ Mq(k) and t ∈ [0, 1], αt k(ϕk(a)) = f (ϕk+1)(ρk(a))(t) i q(k) q(k) Mi=1 q(k) (cid:0)a ⊗ eq(k),q(k) ⊗ eii ⊗ 1q(k+1)(cid:1)  uk+1(t)∗ q(k) (cid:0)a ⊗ eq(k),q(k) ⊗ eii ⊗ eq(k+1)+1,q(k+1)+1(cid:1)  ϕk(a)(cid:18)1 − = uk+1(t)(a ⊗ 1q(k)−1 ⊗ 1q(k) ⊗ 1q(k+1))uk+1(t)∗ ⊕ uk+1(t)  ⊕ f (1 − t)(a ⊗ 1q(k)−1 ⊗ 1q(k) ⊗ eq(k+1)+1,q(k+1)+1) ⊕ f (1 − t) Mi=1  ∼u  Mi=1 Mm=1  ⊕ ϕk(a)(cid:18)1 − Mi=1  q(k) (cid:19) if (1 − t)  , q(k)(cid:19)  ⊕  Mm=1 q(k)(q(k)−1) q(k)−1 q(k+1) q(k) i i ϕk(a)(1 − f (1 − t))  where ∼u denotes unitary equivalence. Write hi(t) = 1 − if (1−t) (so that, in fact, hq(k) = 1 − f (1 − t) = h(t)). By our earlier reasoning it then follows that, for every t ∈ [0, 1], there is a unitary wk(t) ∈ Mq(k+1) ⊗ Mq(k+1)+1 such that   ⊕  evh(t)  ⊕  k = wk(t)    wk(t)∗. evhi(t)  Mm=1 Mm=1 Mi=1 Mi=1   q(k)(q(k)−1) (3.17) ev i q(k)−1 q(k+1) αt q(k) q(k) q(k) It could be that t 7→ wk(t) is not continuous, but this does not matter. (Moreover, it is not difficult to show that, up to approximate unitary equivalence, continuity may be assumed anyway.) Z IS UNIVERSAL 7 We can also give a description of the connecting map αk,k+n = αk+n−1 ◦ ··· ◦ αk. For each j ∈ N, let Λj be the sequence of continuous functions given by listing each constant function i/q(j) (for 1 ≤ i ≤ q(j)− 1) with multiplicity q(j + 1), then h with multiplicity q(j)(q(j)− 1) and then each hi for 1 ≤ i ≤ q(j). Then αk,k+n is fibrewise unitarily equivalent to the direct sum of all maps of the form evF1◦···◦Fn with Fj ∈ Λk+j−1 for 1 ≤ j ≤ n. Let us write T (A) for the space of tracial states on a C∗-algebra A. Recall that every tracial state on Zq(j),q(j)+1 is of the form R tr ◦ evt(·)dµ(t) for some Borel probability measure µ on [0, 1], where tr is the unique tracial state on Mq(j) ⊗ Mq(j)+1. In particular, every such trace extends to a trace on C([0, 1], Mq(j) ⊗ Mq(j)+1), and is invariant under fibrewise unitary equivalence. Since ZU ∼= lim T (Zq(k),q(k)+1). −→ Thus T (ZU ) is an inverse limit of nonempty compact Hausdorff spaces, so is nonempty. That is, ZU has at least one tracial state. For uniqueness, we need to show that for every k ∈ N, every ǫ > 0, and every b ∈ Zq(k),q(k)+1 we have Zq(k),q(k)+1 with unital connecting maps αk, we have T (ZU ) ∼= lim ←− (3.18) for all sufficiently large n and every τ1, τ2 ∈ T (Zq(k+n),q(k+n)+1). The key observation for this is that for each j, most of the elements in the sequence Λj defined above are constant functions. In fact, the proportion of functions in Λj that are not constant is τ1(αk,k+n(b)) − τ2(αk,k+n(b)) < ǫ q(j)(q(j) − 1) + q(j) q(j + 1)(q(j) − 1) + q(j)(q(j) − 1) + q(j) = q(j)4 − q(j)3 + q(j)2 = q(j)2 − q(j) + 1 . (3.19) q(j)2 1 Since F1 ◦ ··· ◦ Fn is constant if any of the Fi are constant, it follows that for fixed b ∈ Zq(k),q(k)+1, αk,k+n(b) is fibrewise unitarily equivalent to a direct sum of continuous Mq(k) ⊗ Mq(k)+1-valued func- tions, most of which are constant except for a small corner. But any two tracial states agree on the constant pieces, and the small corner has trace at most kbkQk+n−1 q(j)2−q(j)+1 , which of course converges to 0 as n → ∞. Thus (3.18) holds, and so ZU has a unique tracial state. It is well known that, to establish simplicity, it suffices to show the following (see for example [14, Theorem 3.4]): if b is a nonzero element of Zq(k),q(k)+1, then αk,r(b) generates Zq(r),q(r)+1 as a (closed, two-sided) ideal for every sufficiently large r (which is the case if and only if αt k,r(b) is nonzero for every t ∈ [0, 1]). Suppose that b is such an element, so that there is an interval in (0, 1) of width ǫ > 0 on which b is nonzero. For each n ∈ N and t ∈ [0, 1], αt k,k+n+1(b) contains summands unitarily equivalent to b(cid:16)h(n)(cid:16) q(k+n)(cid:17)(cid:17) for 1 ≤ i ≤ q(k + n) − 1, where h(n) := z h ◦ ··· ◦ h. Moreover, h(n) is of the form j=k n 1 i 4n . But this is true for all { } 0 ≤ t ≤ ln/4n ln/4n ≤ t ≤ (1 + ln)/4n (1 + ln)/4n ≤ t ≤ 1 0, 4nt − ln, 1, h(n)(t) =   as n → ∞. Thus ZU is simple. q(k+n) < ǫ 1 p3k+n −→ 0 for some ln, and so it suffices to show that for large n we have large n since q(k+n) = 4n 4n It now follows from the classification theorem [4, Theorem 6.2] that ZU ∼= Z. (cid:3) Remark 3.2. One point that should be emphasized is that, despite the use of functional calculus, the relations of Theorem 3.1 really are algebraic, or at least C∗-algebraic in the sense that they involve only ∗-polynomial and order relations. This can be made explicit by encoding the relations (3.2) satisfied by the functions d, f , g and h into the relations for the building blocks Z (q(k)). More specifically, it is not difficult to derive from Proposition 2.2 that the dimension drop algebra Zn,n+1 is isomorphic to the universal C∗-algebra on generators ϕ, ψ and h with relations: (i) ϕ, ψ and h are c.p.c. order zero maps on Mn, M2 and C respectively (in particular, h is just a positive contraction); (ii) [ψ(e11), ϕ(Mn)] = [h, ϕ(Mn)] = 0; (iii) ψ(e11)h = h; (iv) h(1 − ϕ(1n)) = 1 − ϕ(1n); 8 BHISHAN JACELON AND WILHELM WINTER (v) ψ(e22)ϕ(e11) = ψ(e22). (It is a straightforward exercise in functional calculus to write down inverse isomorphisms between the universal C∗-algebra determined by these relations and Z (n) ∼= Zn,n+1.) The following is then proved in exactly the same way as Theorem 3.1. Theorem 3.3. The Jiang-Su algebra Z is isomorphic to the universal unital C∗-algebra generated by c.p.c. order zero maps ϕk on Mq(k) (k ∈ N) and ψk on M2 (k ∈ N), and positive contractions hk (k ∈ N), together with (for each k ∈ N) the relations: [ψk(e11), ϕk(Mq(k))] = [hk, ϕk(Mq(k))] = 0, ψk(e11)hk = hk, hk(1 − ϕk(1q(k))) = 1 − ϕk(1q(k)), ψk(e22)ϕk(e11) = ψk(e22), 1 √2 (1 + hk)1/2ψ1/2 k ϕk = ϕk+1 ◦ ρk, (e12) = (hk+1 + (1 − hk+1)ϕk+1(vk+1v∗ k+1(vk+1), + (1 − ψk+1(e11))1/2ϕ1/2 k+1))1/2ψ1/2 k+1(e12) where the c.p.c. order zero maps ρk : Mq(k) → Mq(k+1) and the partial isometries vk ∈ Mq(k) are as in (3.1) and (3.4) respectively. (cid:3) The article [10] (or, in a much more general setting, [11]) contains a classification by tracial data of 4. W as a universal C∗-algebra simple inductive limits of building blocks Wn,m := {f ∈ C([0, 1], Mn⊗Mm) f (0) = a⊗1m, f (1) = a⊗1m−1, a ∈ Mn}, n, m ∈ N, m > 1. (4.1) Such building blocks are easily seen to be stably projectionless, and it can moreover be shown that they have trivial K-theory (this is why the classifying invariant is purely tracial). The classification is also complete in the sense that every permissible value of the invariant is attained -- see [17] or [3, Proposition 5.3]. Then, W may be defined as the unique C∗-algebra in this class which has a unique tracial state (and no unbounded traces). An explicit construction of W is given in [3], and in this section we obtain another one by adapting the previous section's universal characterization of Z. To begin with, notice that Wn,n+1 is isomorphic to a subalgebra of the dimension drop algebra Zn,n+1; the following indicates that it in fact may be thought of as its nonuntial analogue (compare with Proposition 2.2). Proposition 4.1. Let W (n) be the universal C∗-algebra C∗(ϕ, ψ Rn), where Rn denotes the set of relations: (i) ϕ and ψ are c.p.c. order zero maps on Mn and M2 respectively; (ii) ψ(e11) = ϕ(1n)(1 − ϕ(1n)); (iii) ψ(e22)ϕ(e11) = ψ(e22). Then W (n) ∼= Wn,n+1. Proof. The proof is almost identical to that of Proposition 2.2, but we include it here for completeness. Define ϕ : Mn → Wn,n+1 by ϕ(a)(t) = (a ⊗ 1n) ⊕ (1 − t)(a ⊗ en+1,n+1) for a ∈ Mn and t ∈ [0, 1]. Then ϕ is clearly a c.p.c. order zero map. Equivalently, if we write xi(t) = (e1i ⊗ 1n) ⊕ (1 − t)1/2(e1i ⊗ en+1,n+1) = ϕ1/2(e1i)(t) for 1 ≤ i ≤ n, then the xi satisfy the order zero relations R(0) n and ϕ(eij) = x∗ i xj. Next, define v(t) = t1/2(1 − t)1/2 ej1 ⊗ en+1,j. n Xj=1 Z IS UNIVERSAL 9 Then vv∗ = ϕ(1n)(1 − ϕ(1n)) and vx1 = v, and (so) kvk ≤ 1 and v2 = 0. In particular, there is a unique c.p.c. order zero map ψ : M2 → Wn,n+1 with ψ1/2(e12) = v, i.e. ej1 ⊗ en+1,j, ψ(e12)(t) = t(1 − t) n Xj=1 so that ψ(e11) = vv∗, ψ(e22) = v∗v and ϕ and ψ satisfy all of the relations Rn. We have Next, we check that v and the xi generate Wn,n+1 as a C∗-algebra. Write A := C∗({v, x1, . . . , xn}). and v∗xi(t) = t1/2(1 − t)(e1i ⊗ e1,n+1) v∗xivxj (t) = t(1 − t)3/2(e1j ⊗ e1i) for 1 ≤ i, j ≤ n. Thus, for t ∈ (0, 1), the elements v∗xi(t) and v∗xivxj (t) give all matrix units {e1k ⊗ e1l}1≤k≤n,1≤l≤n+1, so generate all of Mn ⊗ Mn+1, and so the irreducible representation evt : Wn,n+1 → Mn ⊗ Mn+1 restricts to an irreducible representation of A. For t ∈ {0, 1}, the xi generate all the matrix units of Mn in the endpoint irreducible representation ev∞ : Wn,n+1 → Mn. Thus every irreducible representation of Wn,n+1 restricts to an irreducible representation of A. Also, since x1(s) is not unitarily equivalent to x1(t) for distinct s, t ∈ (0, 1), it follows that inequivalent irreducible representations of Wn,n+1 restrict to inequivalent irreducible representations of A. Therefore, by Stone- Weierstrass (i.e. [2, Proposition 11.1.6]), we do indeed have C∗({v, x1, . . . , xn}) = Wn,n+1. It remains to show that these generators of Wn,n+1 enjoy the appropriate universal property: for every representation { ϕ, ψ} of the given relations, we need to show that there is a ∗-homomorphism Wn,n+1 → C∗( ϕ, ψ) sending ϕ to ϕ and ψ to ψ. By [7, Lemma 3.2.2], it suffices to consider the case where { ϕ, ψ} is an irreducible representation on some Hilbert space H (i.e. has trivial commutant in B(H)). Note that the irreducible representations of Wn,n+1 are (up to unitary equivalence), the evaluation maps evt : Wn,n+1 → Mn(n+1) for t ∈ (0, 1) together with the endpoint representation ev∞ : Wn,n+1 → Mn. We will therefore show that (again, up to unitary equivalence) ϕ = evt ◦ ϕ and ψ = evt ◦ ψ for some t ∈ (0, 1) ∪ {∞}. ψ1/2(e12) ϕ1/2(e1i), so that ψi(e11) = ϕ(1n)(1 − ϕ(1n)) and ψi(e22) ϕ(eii) = ψi(e22). Define For each i ∈ {1, . . . , n}, let ψi : M2 → C∗( ϕ, ψ) be the c.p.c. order zero map defined by ψ1/2 (e12) = i z := ψ(e11) + ψi(e22) ∈ C∗( ϕ, ψ). n Xi=1 Then and [z, ϕ(e1j)] = ψ1(e22) ϕ(e1j) − ϕ(e1j) ψj(e22) = 0, [z, ψ(e12)] = ψ2(e12) + n Xi=1 ψi(e22) ψ1/2(e11) ψ1/2(e12) − n Xi=1 ψ(e12) ϕ(e11) ϕ(eii) ψi(e22) = ψ2(e12) + 0 − ψ2(e12) = 0, so z is central in C∗( ϕ, ψ), and is therefore ζ1 for some scalar ζ. Moreover, z is positive with kzk = k ψ(e11)k = k ϕ(1n)(1 − ϕ(1n))k ≤ 1/4, so 0 ≤ ζ ≤ 1/4. If ζ = 0 then ψ = 0 and ϕ(1n) is therefore a projection. It follows that ϕ is a ∗-homomorphism giving an irreducible representation of Mn on H. Thus (up to unitary equivalence) H = Cn and ϕ = ev∞ ◦ ϕ. Suppose that ζ > 0. Then ζ ψ(e11) = z ψ(e11) = ( ψ(e11))2, so p := ζ −1 ψ(e11) and qi := ζ −1 ψi(e22) are equivalent orthogonal projections with p + q1 + ··· + qn = 1. Since p commutes with ϕ(Mn), the maps p ϕ(·)p and (1 − p) ϕ(·)(1 − p) are c.p.c. order zero. In fact, ζ ϕ(1n)(1 − p) = ϕ(1n)(z − ψ(e11)) = z − ψ(e11) = ζ(1 − p), 10 BHISHAN JACELON AND WILHELM WINTER i.e. (1 − p) ϕ(1n)(1 − p) = 1 − p. Thus, (1 − p) ϕ(·)(1 − p) is a unital c.p.c. order zero map into the corner (1 − p)B(H)(1 − p) ∼= B((1 − p)H), so is a ∗-homomorphism into this corner. Also, p ϕ(1n)p commutes with (the WOT-closure of) the corner pC∗( ϕ, ψ)p = pC∗( ϕ)p (which, by irreducibility, is all of pB(H)p ∼= B(pH)) so p ϕ(1n)p = tp for some t ∈ [0, 1]. So t−1p ϕ(·)p is also a ∗-homomorphism, and is in fact an irreducible representation of Mn on pH. In particular, up to unitary equivalence, pH = Cn and p ϕ(·)p = t · idMn . unitary equivalence) (1 − p)H = Cn2 a 7→ diag(a, . . . , a). Finally, since Moreover, since every qi is equivalent to p, they all have trace n (= tr(p)). Thus (again up to (so H = Cn(n+1)) and (1 − p) ϕ(·)(1 − p) : Mn → Mn2 is just t(1 − t)p = tp(p − tp) = ϕ(1n)p(p − ϕ(1n)p) = p ϕ(1n)(1 − ϕ(1n)) = p ψ(e11) = ζp, we have t(1 − t) = ζ. Therefore, ϕ = p ϕ(·)p + (1 − p) ϕ(·)(1 − p) = ev1−t ◦ ϕ and, since ζ −1/2 ψ1/2(e12) is a partial isometry implementing an equivalence between q1 and p, ψ = ev1−t ◦ ψ (up to conjugation by a unitary). Thus Wn,n+1 has the required universal property. (cid:3) Remark 4.2. It should also be possible to detect ∗-homomorphisms from Wn,n+1 to a stable rank one C∗-algebra A at the level of the Cuntz semigroup W (A) (just as for Zn,n+1 in [14, Proposition 5.1]). The existence of hxi ∈ W (A) and a positive contraction y ∈ A with nhxi = hyi and hy − y2i ≪ hxi (where ≪ denotes the relation of compact containment) is probably necessary and sufficient, but perhaps this is not the most useful characterization. Finally, we present W as a nonunital deformation of Z. Theorem 4.3. Choose positive functions d, f, g, h ∈ C0(0, 1], partial isometries vk+1 ∈ Mq(k+1), and c.p.c. order zero maps ρk : Mq(k) → Mq(k+1) as in Theorem 3.1. Define WU to be the universal C∗- algebra generated by c.p.c. order zero maps ϕk on Mq(k) (k ∈ N) and ψk on M2 (k ∈ N) such that for each k, these maps satisfy the relations Rq(k), i.e. and ψk(e11) = ϕk(1q(k))(1 − ϕk(1q(k))) ψk(e22)ϕk(e11) = ψk(e22), (4.2) (4.3) (4.4) (4.5) together with the additional relations Sq(k) given by ϕk = f (ϕk+1) ◦ ρk, ψ1/2 k (e12) = f (ϕk+1)(ρk(1q(k)))1/2(cid:18)h(ϕk+1)(1q(k+1) − ρk(1q(k)))1/2f (ϕk+1)(vk+1) d(ψk+1)(e12)(cid:19). +(cid:0)1 − f (ϕk+1)(1q(k+1)) + g(ϕk+1)(1q(k+1) − ρk(1q(k)))(cid:1)1/2 Then WU ∼= W. Proof. The proof is essentially the same as that of Theorem 3.1, so we omit most of the details. As before, let us write ϕk = f (ϕk+1) ◦ ρk and ψ1/2 k (e12) = γk + δk, where this time γk := f (ϕk+1)(ρk(1q(k)))1/2λkd(ψk+1)(e12) and with and δk := f (ϕk+1)(ρk(1q(k)))1/2µkf (ϕk+1)(vk+1), λk := (1 − f (ϕk+1)(1q(k+1)) + g(ϕk+1)(1q(k+1) − ρk(1q(k))))1/2 To show that ψk(e11) = ϕk(1q(k))(1 − ϕk(1q(k))), we proceed exactly as in the proof of Claim 1. The only difference is that we now have µk := h(ϕk+1)(1q(k+1) − ρk(1q(k)))1/2. d(ψk+1)(e11) = d(ψk+1(e11)) = d(ϕk+1(1q(k+1))(1 − ϕk+1(1q(k+1)))) = d(ϕk+1(1q(k+1))), Z IS UNIVERSAL 11 where d(t) = d(t(1 − t)) as in (3.3). We also have f (ϕk+1)(ρk(1q(k)))(1 − f (ϕk+1)(1q(k+1))) = πϕk+1(ρk(1q(k+1)))(f − f 2)(ϕk+1(1q(k+1))). Since d(f − f 2) = f − f 2, this therefore gives f (ϕk+1)(ρk(1q(k)))(1 − f (ϕk+1)(1q(k+1)))d(ψk+1)(e11) = f (ϕk+1)(ρk(1q(k)))(1 − f (ϕk+1)(1q(k+1))), and the rest of the argument carries over mutatis mutandis. (Note in particular that λk and µk both commute with f (ϕk+1)(ρk(1q(k)))1/2.) The proof that ψk(e22) ϕk(e11) = ψk(e22) is literally the same as the proof of Claim 2. We now know that WU is isomorphic to an inductive limit lim −→ (Wq(k),q(k+1), βk). Moreover, argu- ing exactly as before, we see that the connecting maps βk are (fibrewise) unitarily equivalent to the connecting maps αk obtained earlier. That is, there are unitaries zk(t) ∈ Mq(k+1) ⊗ Mq(k+1)+1 such that βt k = zk(t)    q(k+1) Mm=1 q(k)−1 Mi=1 ev i q(k)   ⊕  q(k)(q(k)−1) Mm=1 q(k) evh(t)  ⊕ Mi=1  evhi(t)   zk(t)∗  (4.6) for every t ∈ [0, 1]. The same arguments as with ZU show that WU is simple and has a unique tracial state. (One has to perhaps be slightly careful about the existence of a trace, since the space of tracial states of a nonunital C∗-algebra need not be compact. But this is not an issue.) The only minor technicality is that, since the building blocks Wq(k),q(k+1) are nonunital and the connecting maps βk are degenerate, WU may have unbounded traces. However, one can easily show, using (3.19), that this is not the case. It therefore follows from the classification theorem of [10] (or indeed from the more general result proved in [11]) that WU ∼= W. (cid:3) Corollary 4.4. There exists a trace-preserving embedding of W into Z. Such an embedding is canon- ical at the level of the Cuntz semigroup, and is unique up to approximate unitary equivalence. Proof. This follows immediately from Theorem 4.3 and Theorem 3.1. The result can already be deduced from the main theorem of [11], which also gives the uniqueness statement. (cid:3) 5. Outlook It might be interesting to characterize other C∗-algebras as we have done for Z and W. It should 5.1. in particular be possible, for any n ≥ 2, to obtain a universal construction of a simple, monotracial, stably projectionless C∗-algebra Wn with (K0(Wn), K1(Wn)) = (0, Z/(n − 1)Z). Candidate building blocks could be of the form {f ∈ C([0, 1], Mm ⊗ M(n−1)(m+1)) : f (0) = a ⊗ 1(n−1)(m+1), f (1) = a ⊗ 1(n−1)m, a ∈ Mm}, which at least have the right K-theory. Of course, W2 is just W, obtained as in Theorem 4.3. It was proved in [11] that W ⊗K ∼= O2 ⋊ R for certain 'quasi-free' actions of R on the Cuntz algebra O2 (see for example [5] and [1]). More generally, one would expect (i.e. the Elliott conjecture predicts) that Wn ⊗ K ∼= On ⋊ R, and in this sense Wn might be thought of as a stably projectionless analogue of On. (Similar speculation is made in the article [9].) It is unclear what interpretation the corresponding universal unital algebras might have. Note for example that the Jiang-Su algebra is not stably isomorphic to a crossed product of a Kirchberg algebra by R (when simple, such a crossed product is either traceless or stably projectionless -- see [6, Proposition 4]). 5.2. One of our motivations for presenting Z as a universal C∗-algebra was to find a direct proof of its strong self-absorption (i.e. one that does not rely on classification). To put this problem into context, consider the other strongly self-absorbing C∗-algebras. On the one hand, UHF algebras of infinite type can also be described in terms of order zero generators and relations, for example: M2∞ ∼= C∗((ϕk)∞ k=1 ϕk order zero on Mq(k), ϕk(1q(k)) = ϕk(1q(k))2, ϕk = ϕk+1 ◦ idq(k) ⊗ 1q(k) ⊗ 1q(k)) 12 BHISHAN JACELON AND WILHELM WINTER i=1 s∗ (where q(k) is still 23k ), and the proof of strong self-absorption in this case amounts to linear algebra. On the other hand, while O2 and O∞ are presented simply as C∗(s1, s2 s∗ 2) and i sj = δij) respectively, the proofs that O2 ⊗ O2 ∼= O2 and O∞ ⊗ O∞ ∼= O∞ require some C∗((si)∞ difficult analysis (see for example [13]). It is conceivable that our presentation of Z lies somewhere in the middle of this spectrum. That being said, it is at least possible to show from our relations, in connection with [11], that the has stable rank one and strict i si = 1 = s1s∗ 1 + s2s∗ U C∗-algebra Z ⊗∞ comparison, and then use the main theorem of [11] to deduce strong self-absorption.) is strongly self-absorbing. (One can show that Z ⊗∞ Meanwhile, it remains an open problem to prove that W ⊗ W ∼= W. U References [1] Andrew Dean. A continuous field of projectionless C∗-algebras. Canad. J. Math., 53(1):51 -- 72, 2001. [2] Jacques Dixmier. Les C∗-alg`ebres et leurs repr´esentations. Cahiers Scientifiques, Fasc. XXIX. Gauthier-Villars & Cie, ´Editeur-Imprimeur, Paris, 1964. [3] Bhishan Jacelon. A simple, monotracial, stably projectionless C∗-algebra. arXiv preprint math.OA/1006.5397; to appear in J. London Math. Soc., 2010. [4] Xinhui Jiang and Hongbing Su. On a simple unital projectionless C∗-algebra. Amer. J. Math., 121(2):359 -- 413, 1999. [5] Akitaka Kishimoto and Alex Kumjian. Simple stably projectionless C∗-algebras arising as crossed products. Canad. J. Math., 48(5):980 -- 996, 1996. [6] . Crossed products of Cuntz algebras by quasi-free automorphisms. In Operator algebras and their applications (Waterloo, ON, 1994/1995), volume 13 of Fields Inst. Commun., pages 173 -- 192. Amer. Math. Soc., Providence, RI, 1997. [7] Terry A. Loring. Lifting solutions to perturbing problems in C∗-algebras, volume 8 of Fields Institute Monographs. American Mathematical Society, Providence, RI, 1997. [8] Hiroki Matui and Yasuhiko Sato. Strict comparison and Z-absorption of nuclear C∗-algebras. arXiv preprint math.OA/1111.1637; to appear in Acta Math., 2011. [9] Norio Nawata. Picard groups of certain stably projectionless C∗-algebras. arXiv preprint math.OA/1207.1930, 2012. [10] Shaloub Razak. On the classification of simple stably projectionless C∗-algebras. Canad. J. Math., 54(1):138 -- 224, 2002. [11] Leonel Robert. Classification of inductive limits of 1-dimensional NCCW complexes. arXiv preprint math.OA/1007.1964; to appear in Adv. Math., 2010. [12] . The cone of functionals on the Cuntz semigroup. arXiv preprint math.OA/1102.1451; to appear in Math. Scand., 2011. [13] Mikael Rørdam. A short proof of Elliott's theorem: O2 ⊗O2 ∼= O2. C. R. Math. Rep. Acad. Sci. Canada, 16(1):31 -- 36, 1994. [14] Mikael Rørdam and Wilhelm Winter. The Jiang-Su algebra revisited. J. Reine Angew. Math., 642:129 -- 155, 2010. [15] Yasuhiko Sato. The Rohlin property for automorphisms of the Jiang-Su algebra. J. Funct. Anal., 259(2):453 -- 476, 2010. [16] Andrew S. Toms and Wilhelm Winter. Strongly self-absorbing C∗-algebras. Trans. Amer. Math. Soc., 359(8):3999 -- 4029 (electronic), 2007. [17] Kin-Wai Tsang. On the positive tracial cones of simple stably projectionless C∗-algebras. J. Funct. Anal., 227(1):188 -- 199, 2005. [18] Wilhelm Winter. Localizing the Elliott conjecture at strongly self-absorbing C∗-algebras. arXiv preprint math.OA/0708.0283; to appear in J. Reine Angew. Math., 2007. [19] [20] [21] [22] Wilhelm Winter and Joachim Zacharias. Completely positive maps of order zero. Munster J. Math., 2:311 -- 324, 2009. . Decomposition rank and Z-stability. Invent. Math., 179(2):229 -- 301, 2010. . Strongly self-absorbing C∗-algebras are Z-stable. J. Noncommut. Geom., 5(2):253 -- 264, 2011. . Nuclear dimension and Z-stability of pure C∗-algebras. Invent. Math., 187(2):259 -- 342, 2012. Mathematisches Institut, Einsteinstr. 62, 48149 Munster, Germany E-mail address: [email protected] E-mail address: [email protected]
1511.07505	1	1511	2015-11-23T23:30:38	Tensor splitting properties of n-inverse pairs of operators	[ "math.OA" ]	In this paper we study n-inverse pairs of operators on the tensor product of Banach spaces. In particular we show that an n-inverse pair of elementary tensors of operators on the tensor product of two Banach spaces can arise only from l- and m-inverse pairs of operators on the individual spaces. This gives a converse to a result of Duggal and M\"uller, and proves a conjecture of the second named author. Our proof uses techniques from algebraic geometry, which generalize to other relations among operators in a tensor product. We apply this theory to obtain results for n-symmetries in a tensor product as well.	math.OA	math	Tensor splitting properties of n-inverse pairs of operators Stepan Paul and Caixing Gu May 14, 2018 Abstract In this paper we study n-inverse pairs of operators on the tensor product of Banach spaces. In particular we show that an n-inverse pair of elementary tensors of operators on the tensor product of two Banach spaces can arise only from l- and m-inverse pairs of operators on the individual spaces. This gives a converse to a result of Duggal and Muller [13], and proves a conjecture of the second named author [16]. Our proof uses techniques from algebraic geometry, which generalize to other relations among operators in a tensor product. We apply this theory to obtain results for n-symmetries in a tensor product as well. 1 Introduction Let B(X) be the algebra of all bounded linear operators on a Banach space X. For S, T ∈ B(X), we define the functional calculus βn(S, T ) = n (−1)n−k(cid:18)n Xk=0 k(cid:19)SkT k. (1) As in Sid Ahmed [24] and Duggal and Muller [13], we say S is a left n-inverse of T (or T is a right n-inverse of S, or (S, T ) is an n-inverse pair ) if βn(S, T ) = 0. If βn(S, T ) = 0, but βn−1(S, T ) 6= 0, we say S is a strict left n-inverse of T . In fact, these definitions make sense for elements S and T in an arbitrary C-algebra with identity. This definition is of course a generalization of the definition of an ordinary left inverse- that S is a left inverse of T if and only if Loosely speaking, the expression (1) is obtained by substituting S for x and T for y in ST − 1 = 0. the expansion n (xy − 1)n = always keeping powers of S to the left of powers of T . (−1)n−k(cid:18)n k(cid:19)xkyk, Xk=0 1 The concept of n-inverse pairs of operators is motivated by the n-isometries studied early in [2, 3, 4, 23] on Hilbert spaces and more recently in [9, 10, 11, 12, 14, 17, 25] on Hilbert spaces and in [6, 8, 15, 20] on Banach spaces. An operator T on a Hilbert space H is called an n-isometry if βn(T ∗, T ) = 0, that is, if T ∗ is a left n-inverse of T. If X and Y are Banach spaces, we let X⊗Y denote the completion, endowed with a reasonable uniform cross norm, of the algebraic tensor product X ⊗ Y of X and Y. The initial objective of this paper is to prove the following theorem. Theorem 1.1. Suppose S1, T1 ∈ B(X) and S2, T2 ∈ B(Y ). Then the following are equiva- lent: (a) S1 ⊗ S2 is a (strict) left n-inverse of T1 ⊗ T2 in B(X⊗Y ). (b) There exist positive integers l, m with l + m = n + 1 and λ ∈ C∗ so that S1 is a (strict) left l-inverse of λT1 in B(X) and S2 is a (strict) left m-inverse of (1/λ)T2 in B(Y ). That (b) implies (a) was proved by Duggal and Muller in Theorem 2.3 of [13]. A corollary of their result for n-isometric tensor products is proved in Theorem 2.10 of [12], which answers questions about m-isometric elementary operators acting on Hilbert-Schmidt operator ideals studied in [9] and [10]. The other implication was conjectured by the second named author in Conjecture 20 of [16], and verified for small n and under some technical conditions. With some additional work, Theorem 7 of [14] for n-isometric elementary operators can be viewed as a corollary of this result. An elementary operator (acting on Hilbert-Schmidt operator ideals) of length one is equivalent to the tensor product of two operators by [18]. See [14] for more general elementary operators (such as generalized derivations) that are m-isometries. In this paper, we will prove Theorem 1.1, and generalize to a more general set of relations among elements of a C-algebra. Specifically, for any polynomial p(x, y), we consider the relation obtained by substituting S for x and T for y into p(x, y)n, always keeping powers of S to the left of powers of T . Of particular interest are the cases where p(x, y) = xy − 1 as already discussed, and where p(x, y) = x − y. The latter yields Xk=0 γn(S, T ) = n (−1)n−k(cid:18)n k(cid:19)SkT n−k. This relation is studied in [21] and [22] for bounded operators S and T on a Hilbert space. In this context, we say T is in the nth Helton class of S and write T ∈ Heltonn(S) if γn(S, T ) = 0. Furthermore, we say T is an n-symmetry if γn(T ∗, T ) = 0. The n-symmetric operators were introduced and studied in connection with Sturm-Liouville conjugate point theory by Helton [19] and studied in [5]. They are inspriational in the study of m-isometries and more general hereditary roots in [3] and [25]. Interestingly, we prove in Section 4 that a direct analogue of Theorem 1.1 is possible essentially in exactly the two cases xy − 1 and x − y, and no others. 2 We consider several applications of this theory in Sections 5 and 6, and perhaps most interestingly prove the following pair of theorems. If H and K are Hilbert spaces, we denote by H⊗K the Hilbert space tensor product of H and K. Theorem 1.2. Suppose H and K are Hilbert spaces, T1 ∈ B(H) and T2 ∈ B(K), and both T1 and T2 are left invertible. Then the following are equivalent: (a) T1 ⊗ T2 is an n-symmetry in B(H⊗K). (b) There exist positive integers l, m with l + m = n + 1 and λ ∈ C with λ = 1 so that λT1 is an l-symmetry in B(H) and ¯λT2 is an m-symmetry in B(K). Theorem 1.3. Suppose H and K are Hilbert spaces, T1 ∈ B(H) and T2 ∈ B(K). Then the following are equivalent: (a) T1 ⊗ IK + IH ⊗ T2 is an n-symmetry in B(H⊗K). (b) There exist positive integers l, m with l + m = n + 1 and λ ∈ C so that T1 + λIH is an l-symmetry in B(H) and T2 − λIK is an m-symmetry in B(K). The outline of this paper is as follows. In Sections 2 and 3, we lay down the alge- braic foundation for dealing with expressions such as (1) and show that our problem can In Section 4, we take advantage of the be considered in a commutative algebra setting. commutativity to prove our main theoretical results. In particular, we will use Hilbert's Nullstellensatz extensively, and we use the notion of the height of an ideal see that quasi- homogeneous polynomials play a special role. In Section 5, we need to briefly explain how the more general algebra results from previous sections imply Theorem 1.1, and show how the theory applies to n-symmetries and the Helton class of an operator. Finally in Section 6, we study the nilpotent pertubation of a left n-inverse. In doing so, we see that the algebraic results apply in a much stronger way for n-symmetries, leading to the proof of Theorem 1.3. 2 Definitions and Algebraic Foundation Let C denote the field of complex numbers, and let C∗ denote the set of nonzero complex numbers. For us, a C-algebra A is a complex vector space which is also an algebra with an identity. For elements S, T of a C-algebra A, we define βn(S, T ) and n-inverses as in the introduction. Note that if S is a left n-inverse of T, then S is a left m-inverse of T for all m ≥ n. This follows from the recursive formula βn(S, T ) = Sβn−1(S, T )T − βn−1(S, T ). (2) It is also true that T has a left n-inverse for all n if and only if T is left-invertible. This follows from (2) and βn(S, T ) = S n Xk=1 (−1)n−k(cid:18)n k(cid:19)Sk−1T k! + (−1)n. 3 Because of this fact, we avoid the term n-invertible even though it appears in the literature. We also define γn(S, T ) as in the introduction. We begin by giving a general algebraic formalism for the construction of expressions such as βn(S, T ) and γn(S, T ). In particular, we will make precise our loose explanation in the introduction that "powers of S are kept to the left of powers of T ". This is important, for example, because if A is non-commutative, then of course βn(S, T ) may not be equal to (ST − 1)n. To deal with this discrepancy, we define a vector space homomorphism from the free commutative C-algebra on x, y to the free C-algebra on X, Y defined on the monomial basis of C[x, y] by Φ : C[x, y] → ChX, Y i Φ(xiyj) = X iY j. Here, C[x, y] is the commutative C-algebra of formal polynomials in two commuting variables x, y, and ChX, Y i is the C-algebra of formal polynomials in two non-commuting variables X, Y (i.e. formal linear combinations of words in X, Y ). In what follows, set R = C[x, y], F = ChX, Y i. Again, Φ is only a vector space homomorphism and not a C-algebra homomorphism, so the multiplicative structure is not preserved. For example, yx = xy in R, but Φ(yx) = XY 6= Y X = Φ(y)Φ(x). However, Φ is exactly the map we need to construct expressions like βn and γn because Φ ((xy − 1)n) = βn(X, Y ), Φ ((x − y)n)) = γn(X, Y ). Since F is a free object in the category of C-algebras, for any C-algebra A and S, T ∈ A, there is a unique C-algebra homomorphism so that κ : F → A κ(X) = S, κ(Y ) = T. For a given element ω ∈ F , we may write ω(S, T ) for κ(ω). Using this notation, S is a left n-inverse of T if and only if Φ((xy − 1)n)(S, T ) = 0. Although Φ is not a C-algebra homomorphism, it does behave like one in a crucial way shown in the following proposition. Proposition 2.1. The inverse image of any two-sided ideal in ChX, Y i under Φ is an ideal in C[x, y]. 4 Proof. Let I be any two-sided ideal in ChX, Y i. Since Φ is a vector space homomorphism, it will suffice to check that Φ−1(I) is closed under multiplication by a monomial. Suppose f (x, y) =P kijxiyj, is such that f ∈ Φ−1(I), where all but finitely many kij are nonzero. Then for any a, b ≥ 0, we have Φ(xayb · f (x, y)) = Φ(cid:16)X kijxi+ayj+b(cid:17) =X kijX i+aY j+b = X a ·(cid:16)X kijX iY j(cid:17) · Y b. The righthand side is clearly in I, so xayb · f (x, y) is in Φ−1(I). Notice that the proof of Proposition 2.1 would not go through with three or more vari- If Ψ : C[x, y, z] → ChX, Y, Zi is the ables, and indeed the conclusion would not hold. analogous vector space homomorphism for three variables, and I ⊂ ChX, Y, Zi is the two- sided ideal generated by XZ, then xz ∈ Ψ−1(I) but xyz /∈ Ψ−1(I). Lemma 2.2. Let I be an ideal in C[x, y]. The inverse image under Φ of the ideal generated by Φ(I) is equal to I. Proof. This follows immediately from the observation that Φ(I) = Φ(R) ∩ hΦ(I)i. 3 Tensor Products This paper concerns itself with questions about how relations in the tensor product of two possibly non-commutative C-algebras descend to relations in the individual C-algebras and vice versa. In this section, we will build a general framework which will allow us to use Proposition 2.1 to convert such questions to the commutative algebra setting. Our motivation lies in proving Theorem 1.1, but in this section we maintain a much more general perspective that can be applied to other relations such as γn(S, T ), and also to nilpotent perturbations in Section 6. In what follows, suppose A1 and A2 are C-algebras, and Si, Ti ∈ Ai. Also, define i = 1, 2 κi : F → Ai, X 7→ Si Y 7→ Ti Now let be any injective C-algebra homomorphism. Letting S = C[x1, y1, x2, y2], we also remind ourselves of the C-algebra isomorphism δ : R → R ⊗ R µ : R ⊗ R xαyβ ⊗ xγyδ → 7→ xα S 1 yβ 1 xγ 2yδ 2 . We then define ǫ and κ so that the following diagram commutes. 5 Φ R δ F ǫ κ (3) S µ R ⊗ R Φ ⊗ Φ F ⊗ F κ1 ⊗ κ2 A1 ⊗ A2 In particular, ǫ(X) := (Φ ⊗ Φ)δ(x), ǫ(Y ) := (Φ ⊗ Φ)δ(y), κ := (κ1 ⊗ κ2)ǫ Recall the following linear algebra fact about tensor products. Lemma 3.1. If φ1 : V1 → W1 and φ2 : V2 → W2 are vector space homomorphisms, then ker(φ1 ⊗ φ2) = (ker(φ1) ⊗ V2) + (V1 ⊗ ker(φ2)). Proof. This is a standard result, but we include the proof for completeness. Let Vi = V ′ i ⊕ ker(φi) be a splitting of the epimorphism φi : Vi → im(φi) for i = 1, 2. Then V1 ⊗ V2 = (V ′ 1 ⊗ V ′ 2) ⊕ (V ′ 1 ⊗ ker(φ2)) ⊕ (ker(φ1) ⊗ V ′ 2) ⊕ (ker(φ1) ⊗ ker(φ2)). However, the restriction φ1 ⊗ φ2 : V ′ the splitting. Hence 1 ⊗ V ′ 2 → im(φ1) ⊗ im(φ2) is an isomorphism because of ker(φ1 ⊗ φ2) = (V ′ 1 ⊗ ker(φ2)) ⊕ (ker(φ1) ⊗ V ′ = (ker(φ1) ⊗ V2) + (V1 ⊗ ker(φ2)). 2) ⊕ (ker(φ1) ⊗ ker(φ2)) For any ideal I ⊆ R, we write for the corresponding ideals in S. Let I ′ = µ(I ⊗ R), I ′′ = µ(R ⊗ I) I := Φ−1(ker κ1), J := Φ−1(ker κ2), so that I and J are ideals in R by Lemma 2.1. We also will define for any p ∈ R, p := µδ(p). In what follows, we adopt the above notation under the assumptions that A1, A2 are C algebras, Si, Ti ∈ Ai for i = 1, 2, and that δ : R → R ⊗ R is any injective C-algebra homomorphism. The following theorem allows us to prove facts about relations in A1 ⊗ A2 in the commu- tative algebra S. Theorem 3.2. Let p ∈ R. Then p ∈ I ′ + J ′′ if and only if Φ(p) ∈ ker κ. 6 Proof. By Lemma 3.1, we have Hence by the commutativity of (3), Φ(p) ∈ ker(κ) if and only of if ker(κ1Φ ⊗ κ2Φ) = I ⊗ R + R ⊗ J. δ(p) ∈ I ⊗ R + R ⊗ J Applying the isomorphism µ, (4) is equivalent to p ∈ I ′ + J ′′. The following two results will be used in proving all of our major theorems. Lemma 3.3. Suppose that p, q1, q2 ∈ R and p ∈ hq1(x1, y1), q2(x2, y2)i. If Φ(ql 1) ∈ ker κ1, Φ(qm 2 ) ∈ ker κ2, then Φ(pn) ∈ ker κ, where n = l + m − 1. Proof. By assumption, there exist f, g ∈ S such that pn = n Xk=0(cid:18)n k(cid:19)f n−kq1(x1, y1)n−kgkq2(x2, y2)k. (4) (5) Notice that every summand of (5), is in either I ′ or J ′′. Thus Φ(pn) ∈ ker κ. The following theorem is a variation on the above lemma, which operates under more technical conditions. However, it will allow us to obtain sharp "if and only if" statements like Theorems 1.1, 1.2, and 1.3. Theorem 3.4. Suppose that p, q1, q2 ∈ R are such that there exist f, g ∈ R so that 1. p(x1, x2, y1, y2) = f (x2, y2)q1(x1, y1) + g(x1, y1)q2(x2, y2). 2. Φ(qL 1 ) ∈ ker κ1 and Φ(qM 2 ) ∈ ker κ2 for large enough L and M. 3. For any i, j, f iqj 1 ∈ I. Then Φ(pn) ∈ ker κ if and only if there exist l and m so that l + m = n+ 1 and Φ(ql and Φ(qm 2 ∈ J if and only if qj 1 ∈ I if and only if qj 2 ∈ J, and giqj 1) ∈ ker κ1 2 ) ∈ ker κ2. Proof. The "if" part of the statement follows directly from Lemma 3.3. In fact, if we assume that l, m are minimal so that Φ(ql 2 ) ∈ ker κ2, then applying (5) with n = l + m − 1, we have 1) ∈ ker κ1 and Φ(qm pl+m−2 = f (x2, y2)l−1q1(x1, y1)l−1g(x1, y1)m−1q2(x2, y2)m−1 in S/(I ′ + J ′′). Applying µ−1 to both sides of the expression, we have 1 ⊗ f l−1qm−1 /∈ I and f l−1qm−1 δ(pl+m−2) = gm−1ql−1 2 By Condition 3, we see that gm−1ql−1 (3), Φ(pl+m−2) /∈ ker κ. than n + 1, proving the "only if" part of the theorem. /∈ J, so by the commutativity of Thus if we assume Φ(pn) ∈ ker κ, the minimal l, m chosen above must add to no more 2 1 in R/I ⊗ R/J. 7 Remark 3.5. Condition 3 of Lemma 3.4 is a rather technical condition, and we take care to include assumptions throughout the exposition that ensure it will be satisfied. However, we note that Condition 3 is satisfied when f, q2 generate R and g, q1 generate R. This follows from Lemma 3.6 below. Lemma 3.6. Suppose g ∈ R, and I ⊆ R is an ideal so that gm ∈ I, but gm−1 /∈ I. If f ∈ R is such that f, g generate R, then for any k ≥ 0, f kgm−1 /∈ I. Proof. Since f, g generate R, there exist α, β so that αf + βg = 1. By way of contradiction, let k be minimial so that f kgm−1 ∈ I. Then αf kgm−1 + βf k−1gm = (αf + βg)f k−1gm−1 = f k−1gm−1. But since gm ∈ I, both sides of the equation must by in I, violating the minimality of k. The following lemma, which relies on Hilbert's Nullstellensatz, will assist us in showing that Condition 2 of Theorem 3.4 is satisfied in all of its applications in later sections. For an ideal I ⊆ C[x1, . . . , xn], we write V (I) for the variety along which the ideal vanishes. That is, V (I) = {a ∈ Cnf (a) = 0 ∀f ∈ I}. Lemma 3.7. Suppose p ∈ √I ′ + J ′′ in S. Then for any (a, b) ∈ V (I), p(a, b, x, y) ∈ √J in R, and for any (c, d) ∈ V (J), p(x, y, c, d) ∈ √I in R. Proof. Notice that V (I ′ + J ′′) = {(a, b, c, d)(a, b) ∈ V (I), (c, d) ∈ V (J)}. Since p ∈ √I ′ + J ′′, p vanishes at every point of V (I ′ + J ′′). So if (a, b) ∈ V (I), then p(a, b, c, d) = 0 for all (c, d) ∈ V (J), and thus by Hilbert's Nullstellensatz, p(a, b, x, y) ∈ √J. Similarly, if (c, d) ∈ V (J), p(x, y, c, d) ∈ √I. 4 Tensor-splitting properties In this section, we will apply the results of Section 2 to the case where δ : R → R ⊗ R is defined by (6) Applying the definitions of Section 3, we see that ǫ(X) = X ⊗ X and ǫ(Y ) = Y ⊗ Y , and δ(x) = x ⊗ x, δ(y) = y ⊗ y. κ(ω) = ω(S1 ⊗ S2, T1 ⊗ T2). In particular, We introduce the notation κΦ((xy − 1)n) = βn(S1 ⊗ S2, T1 ⊗ T2). pλ,µ(x, y) := p(λx, µy), pλ(x, y) := p(x, λy) 8 for p(x, y) ∈ R, λ, µ ∈ C. Notice that pa,b(x, y) = p(x, y, a, b) = p(a, b, x, y) Now proving Theorem 1.1 entails showing that, for p(x, y) = xy − 1, 1/λ) ∈ ker κ2 λ) ∈ ker κ1 and Φ(pm Φ(pl Φ(pn) ∈ ker κ ⇐⇒ for some l, m > 0 with l + m = n + 1, λ ∈ C∗ . (7) It is natural to ask if (7), or some weaker version, is true for other polynomials p ∈ R. We therefore make the following definitions. Definition 4.1. Suppose p ∈ R. If for every pair of C-algebras A1, A2, and every Si, Ti ∈ Ai (not both zero), i = 1, 2 Φ(pn)(S1 ⊗ S2, T1 ⊗ T2) = 0 =⇒ a,b)(S1, T1) = 0, Φ(pm Φ(pl c,d)(S2, T2) = 0 for some a, b, c, d ∈ C, large enough l, m (8) we say p has a weak tensor-splitting property. If additionally we can replace the phrase "large enough l, m" in (8) with "some positive l, m with l + m = n + 1" whenever Si is left-invertible and Ti is right-invertible, then we say p has a strong tensor-splitting property. Definition 4.2. Suppose p ∈ R. If for every pair of C-algebras A1, A2, and every Si, Ti ∈ Ai, i = 1, 2 Φ(pl)(S1, T1) = 0, Φ(pm)(S2, T2) = 0 =⇒ Φ(pn)(S1 ⊗ S2, T1 ⊗ T2) = 0 for some large enough n (9) we say p has a weak tensor-product property. If additionally we can replace the phrase "large enough n" in (9) with "n = l + m − 1", then we say p has a strong tensor-product property. In what follows, we use Theorem 3.2 to characterize polynomials with the above properties by working in the commutative algebra setting. Note that applying the notation of Section 2 to our situation where δ is defined in (6), p(x1, x2, y1, y2) = p(x1x2, y1y2). Our first main result is the following. Theorem 4.3. Every p(x, y) ∈ C[x, y] satisfies a weak tensor-splitting property. Proof. Suppose Φ(pn) ∈ ker(κ). By Theorem 3.2, this means pn ∈ I ′ + J ′′, where I = Φ−1(ker κ1) and J = Φ−1(ker κ2). Since Si, Ti are assumed not to both be zero, I and J are proper ideals of R, and so V (I) and V (J) are non-empty by the Weak Nullstellensatz. Applying Lemma 3.7, we see that there exist a, b, c, d so that pc,d ∈ √I and pa,b ∈ √J, a,b) ∈ ker(κ2) for large enough and thus (again by Theorem 3.2) Φ(pl l, m. c,d) ∈ ker(κ1) and Φ(pm 9 This theorem may seem surprisingly strong, but as we shall see in Corollary 4.6, it turns out the assumption that p ∈ √I ′ + J ′′ is quite restrictive for most polynomials p. The height of a prime ideal p is the length n of a maximal ascending chain of prime ideals p0 ( p1 ( · · · ( pn = p. The height of any ideal I is the infimum of the heights of prime ideals containing I. If I ⊆ C[x, y], then the height of I is also equal to the complex codimension of V (I) in C2. The only height-0 ideal is all of R, and height-2 ideals in R are ones large enough that their vanishing set is a finite (or empty) set of points. If A is any C-algebra, we say S ∈ A is algebraic if S satisfies a polynomial relation; that is there exist scalars ci so that cnSn + · · · + c1S + c01 = 0. In an operator algebra, the algebraic elements are called algebraic operators. In this sense, algebraic operators behave like finite matrices. Proposition 4.4. Let A be any C-algebra, and suppose κ : F → A is a C-algebra homo- morphism as before. Let S = κ(X), T = κ(Y ). If Φ−1(ker κ) has height 2, then S and T are both algebraic. Proof. Suppose I = Φ−1(ker κ) has height 2 so that V (I) is a finite collection of points pi = (ai, bi), i = 1, . . . , r. Hence √I is the product of the maximal ideals mpi = hx−ai, y−bii, so I contains α(x, y) = (x − a1)k · · · (x − ar)k and β(x, y) = (y − b1)k · · · (y − br)k for some k. But then Φ(α), Φ(β) ∈ ker(κ), so Φ(α)(S) = 0 and Φ(β)(T ) = 0, and thus S and T are algebraic. Because of Proposition 4.4, the only "interesting" (i.e. not both algebraic) pairs S, T in an algebra A are ones for which I has height 1. In what follows, we characterize polynomials p ∈ R for which a height 1 ideal may arise in our context. A polynomial q(x1, . . . , xn) ∈ C[x1, . . . , xn] is called quasi-homogeneous if there exist coprime integers w1, . . . , wn called weights and d called the quasi-degree so that for any λ ∈ C∗, q(λw1x1, . . . , λwnxn) = λdq(x1, . . . , xn). Equivalently, if we write q(x1, . . . , xn) = Xi1,...,in ki1,...,inxi1 1 · · · xin n , then ki1,...,in 6= 0 implies w1i1 + · · · wnin = d. Note that some authors restrict to the case where w1, . . . , wn, d are all positive, but we do not make that part of the definition. Proposition 4.5. Let p(x, y) ∈ R be irreducible. Then the following are equivalent: (a) p is quasi-homogeneous. (b) p(x, y) = A(xαyβ − B) or p(x, y) = A(Bxα − yβ) for some coprime positive integers α and β and constants A, B ∈ C. (c) p ∈ hpi′ + hpi′′. 10 (d) There exist height 1 ideals I, J ⊂ R so that p ∈ √I ′ + J ′′. (e) p has a strong tensor-product property. (f ) p has a weak tensor-product property. Proof. That (a) is equivalent to (b) is a standard result, and we omit the proof. If α = β = 1 and A = B = 1, then the implication (b) implies (c) follows from x1x2y1y2 − 1 = (x1y1 − 1)x2y2 + (x2y2 − 1), x1x2 − y1y2 = (x1 − y1)x2 + (x2 − y2)y1. The general case follows from substituting xα for x and yβ/B for y. (10) Condition (c) clearly implies (d), and (e) clearly implies (f). By Lemma 3.3, (c) implies (e). If p has a weak tensor-product property, then let A1 = A2 = F/Φ(p), with canonical maps κi : F → Ai. Then I = J = hpi are height 1 ideals, and since Φ(p) ∈ ker κi for i = 1, 2, the weak tensor-product property implies p ∈ √I ′ + J ′′. Thus (f) implies (d). Finally, we must prove (d) implies (a). Suppose I and J are height-1 ideals in R and pn ∈ I ′ + J ′′. First, note that if p(x, y) = x or p(x, y) = y, then p is quasi-homogeneous. Otherwise, the vanishing set V (J) contains some (c, d) with c, d both nonzero. Then by Lemma 3.7, we have pc,d ∈ √I. Since I has height 1, √I has height 1, and hpc,di ⊆ √I. But since pc,d is irreducible, hpc,di is prime of height 1, so √I = hpc,di, and V (I) is the codimension 1 set of (a, b) so that pc,d(a, b) = 0. Now suppose (a0, b0) ∈ V (I) with a0, b0 nonzero, and let p(x, y) =Xi,j kijxiyj 0bj (where all but finitely many kij are zero). Let (a, b) ∈ V (I) be such that the constants λ = a/a0 and µ = b/b0 are nonzero and have modulus other than 1. Then since pa,b and pa0,b0 are both irreducible and each generate √J, the polynomials must be constant multiples of one another. Hence, there is a constant M so that if ki,j 6= 0, Mai 0 = aibj; that is, M = λiµj. Therefore, if ki1j1 and ki2j2 are both nonzero, λi1µj1 = λi2µj2, so λi1−i2 = µj2−j1. Let α, β be coprime integers so that α(i1 − i2) = β(j2 − j1). Then by the Chinese Remainder Theorem, there exists η so that ηα = λ, ηβ = µ. Letting d = αi1 + βj1, we have M = ηd, and so for any i, j with ki,j 6= 0, ηαi+βj = ηd; since η 6= 1, d = αi + βj. Therefore pa0,b0 and, as a corollary, p are quasi-homogeneous with weights α and β and quasi-degree d. Corollary 4.6. If p is irreducible and not quasi-homogeneous, and then either S1 and T1 are both algebraic, or S2 and T2 are both algebraic. Φ(pn)(S1 ⊗ S2, T1 ⊗ T2) = 0, (11) 11 The authors view Corolloary 4.6 as an indication that unless p is quasi-homogeneous, then the condition (11) is in some sense too strong to be useful. In particular, if p is not quasi- homogeneous, (11) implies that one of the two pairs of operators Si, Ti was uninteresting in the sense that both operators were algebraic. This also sheds new light on Theorem 4.3; the weak tensor-splitting property of p may simply arise as a bi-product of the implied algebraic- ness of the original operators. For this reason, we consider the tensor-splitting property to really only be meaningful in the case where p is quasi-homogeneous. Theorem 4.7. Any irreducible quasi-homogeneous polynomial in two variables has a strong tensor-splitting property. Specifically, suppose p(x, y) = A(xαyβ − B) or p(x, y) = A(Bxα − yβ) and let η be such that ηβ = B; then if Φ(pn)(S1 ⊗ S2, T1 ⊗ T2) = 0 and Si is left-invertible and Ti is right-invertible for i = 1, 2, then there exists λ ∈ C∗ and l, m ≥ 0 with l + m ≤ n + 1 so that Φ(pl λ)(S1, T1) = 0, Φ(pm η/λ)(S2, T2) = 0. Proof. We will restrict to the cases that p(x, y) = x − y and p(x, y) = xy − 1. To go to the general cases, substitute xα and yβ/B for x and y respectively. Suppose Φ(pn) ∈ ker(κ). Since we can assume that Si, Ti both have one-sided inverses in Ai for i = 1, 2, then I = Φ−1(ker κ1) and J = Φ−1(ker κ2) are both proper ideals, and there exist points (a, b) ∈ V (I), (c, d) ∈ V (J) so that a, b, c, d are all nonzero. We will now prove that p, pλ, p1/λ satisfy the conditions of Theorem 3.4, whose conclusion is exactly what we are trying to prove. The following variation on (10) shows Condition 1, that p ∈ hpλ, p1/λi, is satisfied. x1x2y1y2 − 1 = (x2y2/λ)(λx1y1 − 1) + 1(x2y2/λ − 1), x1x2 − y1y2 = x2(x1 − λy1) + λy1(x2 − y2/λ). We also just showed Condition 2, that pL λ ∈ I and pM 1/λ ∈ J, is satisfied as well. then since xy/λ, xy/λ − 1 generate R, as do 1, λxy − 1, we may apply Remark 3.6. For Condition 3, we treat the two cases of xy− 1 and x− y separately. If p(x, y) = xy− 1, If p(x, y) = x − y, we prove Condition 3 directly. Suppose (λy)i(x − λy)j ∈ I. Then 0 = κ2Φ((λy)i(x − λy)j) = λiκ2(Φ((x − λy)j)Y i) = λiκ2(Φ((x − λy)j))T i 2 Since T2 is right-invertible, this implies that Φ((x − λy)j) ∈ ker κ2, so (x − λy)j ∈ I. By a similar argument, if xi(x − y/λ)j ∈ J, then (x − y/λ)j ∈ J. Remark 4.8. Note that as long as A1, A2 are nonzero, we do not need to include the invertibility of Si, Ti as a separate condition for the case of p(x, y) = xy − 1. Indeed, the only time this condition is invoked in the proof is to show that a, b, c, d are nonzero, but this is implied by the fact that (abxy − 1)L ∈ I and (cdxy − 1)M ∈ J and that I, J are proper ideals. 12 5 Operator Theoretic Results We now briefly explain the operator theoretic consequences of the results of the previous section. Lemma 5.1. For Banach algebras B1, B2, we let ι : B1 ⊗ B2 → B1⊗B2 be the inclusion of the algebraic tensor product into its completion with respect to some reasonable uniform cross norm. Then for any ω ∈ F and α, β ∈ B1⊗B2, we have ω(α, β) = 0 if and only ι(ω(α, β)) = 0. Proof. Follows immediately from the injectivity of ι. The above lemma may be somewhat trivial, but for us it means that results from ear- lier sections regarding the algebraic tensor product of C-algebras apply equally well to the (completed) tensor product of Banach algebras. Proof of Theorem 1.1. Note that for Banach spaces X and Y, B(X) and B(Y ) are nonzero C-algebras with identities. The Banach algebra B(X⊗Y ) is isomorphic to a completion of B(X) ⊗ B(Y ) with respect to a reasonable uniform cross norm. Thus, if we apply Theorem 4.7 with A1 = B(X) and A2 = B(Y ), the algebraic assumption that Φ((xy − 1)n)(S1 ⊗ S2, T1 ⊗ T2) = 0, in B(X) ⊗ B(Y ) is equivalent to the same condition in B(X⊗Y ) by Lemma 5.1, and so the same conclusion holds. We should also remark that the statement of Theorem 1.1 also holds when the paren- thetical strictness condition is added to both (a) and (b). This follows from the original statement of Theorem 1.1 along with the recursive condition in (2). We now move on to some applications of the theory to p(x, y) = x − y. Recall from the introduction that Helton classes were defined for operators on a Hilbert space, but the definition extends easily to arbitrary C-algebras. In particular if S, T are elements of an arbitrary C-algebra A, we say T ∈ Heltonn(S) if Φ((x − y)n)(S, T ) = 0. Theorem 5.2. Assume S1, T1 ∈ B(X) and S2, T2 ∈ B(Y ), and that S1, S2 are left-invertible, and T1, T2 are right-invertible.Then the following are equivalent. (a) The tensor product T1 ⊗ T2 on X⊗Y belongs to Heltonn(S1 ⊗ S2). (b) There exist m and l such that m + l = n + 1 and λ ∈ C∗ so that λT1 ∈ Heltonl(S1) and (1/λ)T2 ∈ Heltonm(S2). 13 Proof. Using the same logic as in the proof of Theorem 1.1, we can apply Theorem 4.7 and Lemma 5.1 with p(x, y) = x − y. In preparation to prove Theorem 1.2, we recall the definition of an n-symmetry from the introduction that if H is a Hilbert space, we say T ∈ B(H) is an n-symmetry if γn(S, T ) = Φ((x − y)n)(T ∗, T ) = 0. In what follows, we write σ(T ) and σap(T ) for the spectrum and approximate point spectrum of T respectively. Lemma 5.3. If T is not nilpotent and λT ∈ Heltonn(T ∗), then λ = 1. Proof. Let r be the spectral radius of T , so that there exists ρ ∈ σap(T ) with ρ = r. Note that by the non-nilpotency condition, ρ 6= 0. Then there exists a sequence of unit vectors hi ∈ H such that k(T − ρIH)hik → 0 as i → 0. hγn(T ∗, λT )hi, hii = = n n Xk=0 Xk=0 (−1)n−k(cid:18)n (−1)n−k(cid:18)n k(cid:19)(cid:10)(T ∗)k(λT )n−khi, hi(cid:11) k(cid:19)(cid:10)(λT )n−khi, T khi(cid:11) Thus, as i → ∞, we have hγn(T ∗, λT )hi, hii → k(cid:19)(cid:10)(λρ)n−khi, ρkhi(cid:11) k(cid:19)(λρ)n−k ¯ρk hhi, hii n n (−1)n−k(cid:18)n Xk=0 (−1)n−k(cid:18)n Xk=0 = (λρ − ¯ρ)n. = However, by assumption, γn(T ∗, λT ) = 0 and ρ 6= 0, so λ = 1. Proof of Theorem 1.2. We can apply Theorem 5.2 directly to show that (b) implies (a) in Theorem 1.2. To prove the converse, assume T1 ⊗ T2 is an n-symmetry, so that Theorem 5.2 tells us Φ((x − y)l)(T ∗ 1 , λT1) = 0, Φ((x − y)m)(T ∗ 2 , (1/λ)T2) = 0. Thus it only remains to be shown that λ = 1, but this is the content of Lemma 5.3. 6 Nilpotent perturbation of a left n-inverse Inspired by the construction of n-isometries using the sum of isometries and nilpotent oper- ators in [1] and [7], the second named author proved in Theorem 2 of [16] that if S is a left m-inverse of T and Q is a nilpotent operator of order l such that QS = SQ, then S + Q is 14 an n-inverse of T , where n = l + m − 1. In this section we show that the theorem can be extended in a certain context, and prove the converse using the algebraic geometry approach developed in previous sections. The same result actually applies in a much stronger way to Helton classes, where we use it to prove Theorem 1.3. We achieve these theorems using the same framework built in Section 3, but the map δ : R → R ⊗ R is different than in Section 4. Proposition 6.1. Let A1, A2 be C-algebras and S, T ∈ A1 and Q ∈ A2 with S, T not nilpotent, Q nonzero. Also let p ∈ C[x, y]. If Φ(pl)(S + λ1, T ) = 0 in A1 and (Q − λ1)m = 0 in A2, then Φ(pn)(S ⊗ 1 + 1 ⊗ Q, T ⊗ 1) = 0 where n = l + m − 1. Proof. We will apply Lemma 3.3 using δ : R → R ⊗ R defined by δ(x) = x ⊗ 1 + 1 ⊗ x and δ(y) = y ⊗ 1. Say κ1 : F → A1 is defined by κ1(X) = S, κ1(Y ) = T , and κ2 : F → A2 is defined by κ2(X) = Q and κ2(Y ) = 0. First notice that if p ∈ R, then µδ(p) = p(x1 + x2, y1). Supposing p(x, y) = P kijxiyj, we can write 1 p(x1 + x2, y1) =X kij(x1 + x2)iyj =X kij(x1 + λ)iyj =(cid:16)X kij(x1 + λ)iyj = p(x1 + λ, y1) + (x2 − λ)g 1 + (x2 − λ)gij 1(cid:17) + (x2 − λ)g for some gij for some g Let q1(x, y) = p(x+λ, y) and q2(x, y) = x−λ, and we can see then that p ∈ hq1(x1, y1), q2(x2, y2)i. Also, notice that Φ(ql Φ(pn) ∈ ker κ. 1) ∈ ker κ1 and Φ(qm 2 ) ∈ ker κ2 by assumption. Thus by Lemma 3.3, We say p(x, y) is linear in x if p(x, y) can be written as α(y)x + β(y) for some α, β. Theorem 6.2. Let A1, A2 be C-algebras and S, T ∈ A1 and Q ∈ A2 with S, T not both zero and Q nonzero. Also let p ∈ C[x, y] be any irreducible polynomial. Then the following are equivalent: (a) Φ(pn)(S ⊗ 1 + 1 ⊗ Q, T ⊗ 1) = 0. (b) There exist positive integers l, m with l+m = n+1 and λ ∈ C so that Φ(pl)(S+λ1, T ) = 0 and (Q − λ1)m = 0. Proof. We will use the same definitions of δ and κi as in the proof of Proposition 6.1. Because of Proposition 6.1, we need only prove that (a) implies (b). Since Q is assumed to be nonzero, ker κ2 is a proper ideal, so V (J) is non-empty, where J = Φ−1(ker κ2). Therefore, let (λ, d) be any point in V (J) and set q1(x, y) := p(x, y, λ, d) = p(x + λ, y). 15 By the linearity assumption we can write p(x, y) = α(y)x + β(y). We claim that there Then by Lemma 3.7, q1 ∈ √I. exists (a, b) ∈ V (I) so that α(b) 6= 0. Indeed, if no such (a, b) exists, then α(y) ∈ √I, but then since q1(x, y) = α(y)(x + λ) + β(y) ∈ √I, we also have β(y) ∈ √I, and since p was assumed to be irreducible, this means I = R, contrary to the assumption that S and T are not both zero. Therefore, pick (a, b) ∈ V (I) so that α(b) 6= 0. Then by Lemma 3.7, q2(x, y) := p(a, b, x, y) = p(x + a, b) = p(a + λ, b) + α(b)(x − λ) = α(b)(x − λ) ∈ √I. And since α(b) 6= 0, this means q2 ∈ √I. We now show that p, q1, q2 satisfy the conditions of Theorem 3.4. The proof of Proposi- tion 6.1 shows that Condition 1 is satisfied, and we just showed that Conditions 2 is satisfied. We use Remark 3.6 to prove Condition 3. In particular, q2 and 1 clearly generate R, and since α and β share no roots, q1 = α(y)(x + λ) + β(y) and α(y) also generate R. Corollary 6.3. Let X, Y be Banach spaces, and assume S, T ∈ B(X) and Q ∈ B(Y ). Then the following are equivalent: (a) The tensor sum S ⊗ IY + IX ⊗ Q on X⊗Y is a strict left n-inverse of T ⊗ IY . (b) There exist positive l and m such that l + m = n − 1 and some constant λ ∈ C so that S + λIX is a strict left l-inverse of T and Q − λIY is a nilpotent operator of order l. Proof. Follows directly from Proposition 6.2 using p(x, y) = xy − 1 and Lemma 5.1. Corollary 6.3 was proven in Theorem 22 of [16] using operator theoretic techniques, but here we see that it follows purely from algebraic considerations. We could make a similar statement about Helton classes, but we will see in Proposition 6.5 that a much stronger result is possible when p(x, y) = x − y. It is natural to ask how Corollary 6.3 can be applied to n-isometries. The following result (stated here in the equivalent tensor product of operators instead of elementary operators) is proved in Theorem 12 of [14]. Theorem 6.4. Suppose H and K are Hilbert spaces. Assume S ⊗ IK + IH ⊗ Q is a strict n-isometry and σ(S ⊗ IK + IH ⊗ Q) 6=(cid:8)±e±iαeiθ for some α, θ ∈ [0, 2π)(cid:9) . (12) Then there exist m and l such that m + 2l = n + 2, and S + λIH (or (Q − λIK)) is a strict m-isometry and Q + λIK (or S − λIH) is a nilpotent operator of order l for some constant λ ∈ C. The spectral condition (12) is necessary by Proposition 14 of [14]. Somewhat surprisingly, Proposition 6.2 can be strengthened significantly when applied to p(x, y) = x − y, and so we treat this case separately. 16 q1(x, y) := p(x + λ, y) = x − y + λ ∈ Then let (a, b) ∈ V (I), so that p(a + c, b + d) = a− b + λ, and thus a− b = −λ. Therefore, by Theorem 4.3, √I. √J. Proposition 6.5. Suppose A1, A2 are C-algebras and Si, Ti ∈ Ai are nonzero for i = 1, 2. Then the following are equivalent: (a) γn(S1 ⊗ 1 + 1 ⊗ S2, T1 ⊗ 1 + 1 ⊗ T2) = 0. (b) There exist l, m ≥ 0 and λ ∈ C so that l + m = n + 1 and γl(S1 + λ1, T1) = 0 and γm(S2 − λ1, T2) = 0. Proof. We define δ : R → R ⊗ R by δ(x) = x ⊗ 1 + 1 ⊗ x and δ(y) = y ⊗ 1 + 1 ⊗ y and let p(x, y) = x − y. Thus p(x1, y1, x2, y2) = x1 + x2 − y1 − y2. Let κi : F → Ai via κi(X) = Si and κi(Y ) = Ti. Since Si, Ti are nonzero, I and J are proper ideals. Thus there exists (c, d) ∈ V (J), and so by Theorem 4.3 if we let λ = c − d, q2(x, y) := p(x − λ, y) = x − y − λ ∈ We have now shown that p, q1, q2 satisfy the conditions of Theorem 3.4. In particular, p(x1, y1, x2, y2) = 1q1(x1, y1) + 1q2(x2, y2), so Condition 1 is satisfied. We just showed that Condition 2 holds, and Condition 3 is satisfied trivially. Remark 6.6. We remark that condition (b) of Proposition 6.5 could have been equivalently stated as γl(S1 + α1, T1 − β1) = 0 and γm(S2 − α1, T2 + β1) = 0 by making the substitution α + β = λ. Surprisingly, for n-symmetries, we have the following nice result without any spectral condition. We first state a lemma which follows from formula (4) in Lemma 7 of [17]. Lemma 6.7. If γn(T ∗ + λI, T − λI) = 0, then λ is pure imaginary so that T − λI is an n-symmetry. Proof. Let α ∈ σap(T ), so that there exists a sequence of unit vectors hi ∈ H so that k(T − αI)hik → 0 as i → 0. hγn(T ∗ + λI, T − λI)hi, hii = = = n n Xk=0 Xk=0 Xk=0 n (−1)n−k(cid:18)n (−1)n−k(cid:18)n (−1)n−k(cid:18)n k(cid:19)(cid:10)(T ∗ + λ)k(T − λI)n−khi, hi(cid:11) k(cid:19)h(T − λI)n−khi, (T + ¯λ)khii k(cid:19)(cid:10)(λT )n−khi, T khi(cid:11) 17 Thus, as i → ∞, we have hγn(T ∗ + λI, T − λI)hi, hii → n n (−1)n−k(cid:18)n Xk=0 (−1)n−k(cid:18)n Xk=0 = (α − ¯α − 2λ)n. = k(cid:19)(cid:10)(α − λ)hi, (α + ¯λ)hi(cid:11) k(cid:19)(α − λ)n−k( ¯α + λ)k hhi, hii imaginary. Hence since γn(T ∗ + λI, T − λI) = 0, we must have λ = (α − ¯α)/2, and so λ is pure Thus, T ∗ + λI = (T − λI)∗, so the original assumption shows that T ∗ − λI is an n- symmetry. Proof of Theorem 1.3. We start by proving that (b) implies (a), so we begin by supposing that γm(T ∗ Then by Proposition 6.5, if n = l + m − 1, we have 1 + ¯λIH, T1 + λIH ) = 0, γl(T ∗ 2 − ¯λIK, T2 − λIK) = 0. γn((T ∗ 1 + ¯λIH ) ⊗ IK + IH ⊗ (T ∗ 2 − ¯λIK),(T1 + λIH) ⊗ IK + IH ⊗ (T2 − λIK)) = γn(T ∗ = 0 1 ⊗ IK + IH ⊗ T ∗ 2 , T1 ⊗ IK + IH ⊗ T2) which proves the claim. To prove that (a) implies (b), we assume γn(T ∗ 1 ⊗ IK + IH ⊗ T ∗ 2 , T1 ⊗ IK + IH ⊗ T2) = 0 Using Remark 6.6 and Proposition 6.5 we can conclude that for some λ and some positive l, m with l + m = n + 1, we have γl(T ∗ 1 − λIH , T1 + λIH) = 0, γm(T ∗ 2 + λIK, T2 − λIH) = 0 Finally, applying Lemma 6.7, we see that λ is pure imaginary so that ¯λ = −λ, and the claim is proved. References [1] J. Agler, Sub-Jordan operators: Bishop's theorem, spectral inclusion, and spectral sets, J. Operator Theory 7 (2)(1982) 373-395. [2] J. Agler, A disconjugacy theorem for Toeplitz operators, American Journal of Mathe- matics, 112 (1990) 1–14. 18 [3] J. Agler, W. Helton, M. Stankus, Classification of hereditary matrices, Linear Algebra Appl. 274 (1998) 125–160. [4] J. Agler, M. Stankus, m-isometric transformations of Hilbert space, I, Integral Equa- tions Operator Theory 21 (4)(1995) 383–429. [5] J. A. Ball and J. W. Helton, Nonnormal dilations, disconjugacy, and constrained spectral factorization, Integral Equations Operator Theory 312 (1980) 216–309. [6] F. Bayart, m-Isometries on Banach spaces, Math. Nachr. 284 (2011), 2141–2147. [7] T. Berm´udez, A. Martin´on and J. Noda, An isometry plus a nilpotent operator is an m-isometry and applications, J. Math. Anal. Appl. 407(2) (2013) 505-512. [8] T. Berm´udez, A. Martin´on, J. Noda, Products of m-isometries, Linear Algebra Appl. 438 (2013) 80–86. [9] F. Botelho, J. Jamison, Isometric properties of elementary operators, Linear Algebra Appl. 432 (2010) 357–365. [10] F. Botelho, J. Jamison and B. Zheng, Strict isometries of any orders, Linear Algebra Appl. 436 (2012) 3303–3314. [11] M. Ch¯o, S. Ota and K. Tanahashi, Invertible weighted shift operators which are m- isometries, Proc. Amer. Math. Soc. 141 (2013) 4241–4247. [12] B. P. Duggal, Tensor product of n-isometries, Linear Algebra Appl. 437 (1) (2012) 307–318. [13] B. P. Duggal and V. Muller, Tensor product of left n-invertible operators, Studia Math. 215 (2013), 113–125. [14] C. Gu, Elementary operators which are m-isometries, Linear Algebra Appl. 451 (2014) 49–64. [15] C. Gu, The (m, q)-isometric weighted shifts on lp-spaces, Integral Equations Operator Theory 82 (2015) 157–187. [16] C. Gu, Structures of left n-invertible operators and their applications, Studia Math. 226 (3) (2015) 189–211. [17] C. Gu and M. Stankus, Some results on higher order isometries and symmetries: prod- ucts and sums with a nilpotent, Linear Algebra Appl. 469 (2015) 500–509. [18] J. Eschmeier, Tensor products and elementary operators, J. Reine Angew. Math. 390 (1988) 47–66. [19] J.W. Helton, Jordan operators in infinite dimensions and Sturm-Liouville conjugate point theory, Trans. Amer. Math. Soc. 170 (1972) 305–331. 19 [20] P. Hoffman, M. Mackey and M. ´O Searcoid, On the second parameter of an (m, p)- isometry, Integral Equations Operator Theory 71 (2011) 389–405. [21] I. Kim, Y. Kim, E. Ko, J. E. Lee, Some connections between an operator and its Helton class, J. Math. Anal. Appl. 340 (2) (2008) 1235–1240. [22] I. Kim, Y. Kim, E. Ko, J. E. Lee, Inherited properties through the Helton class of an operator, Bull. Korean Math. Soc. 48 (1) (2011) 183–195. [23] S. Richter, A representation theorem for cyclic analytic two-isometries, Trans. Amer. Math. Soc. 328 (1991) 325–349. [24] O. A. M. Sid Ahmed, Some properties of m-isometries and m-invertible operators on Banach spaces, Acta Math. Sci. Ser. B Engl. Ed. 32 (2012), 520–530. [25] M. Stankus, m-Isometries, n-symmetries and other linear transformations which are hereditary roots, Integral Equations Operator Theory 75 (2013) 301–321. Stepan Paul Department of Mathematics University of California, Santa Barbara Santa Barbara, CA 93106 e-mail: [email protected] Caixing Gu Department of Mathematics California Polytechnic State University San Luis Obispo, CA 93407 e-mail: [email protected] Mathematics Subject Classification (2010). Primary 47A80, 47A10, Secondary 47B47 Keywords: Banach algebra, algebraic operator, quasi-homogeneous polynomial, n-inverse, n-isometry, tensor product 20
1012.3362	1	1012	2010-12-15T15:36:57	Spectral Invariance of Besov-Bessel Subalgebras	[ "math.OA" ]	Using principles of the theory of smoothness spaces we give systematic constructions of scales of inverse-closed subalgebras of a given Banach algebra with the action of a d-parameter automorphism group. In particular we obtain the inverse-closedness of Besov algebras, Bessel potential algebras and approximation algebras of polynomial order in their defining algebra. By a proper choice of the group action these general results can be applied to algebras of infinite matrices and yield inverse-closed subalgebras of matrices with off-diagonal decay of polynomial order. Besides alternative proofs of known results we obtain new classes of inverse-closed subalgebras of matrices with off-diagonal decay .	math.OA	math	SPECTRAL INVARIANCE OF BESOV-BESSEL SUBALGEBRAS ANDREAS KLOTZ Abstract. Using principles of the theory of smoothness spaces we give sys- tematic constructions of scales of inverse-closed subalgebras of a given Banach algebra with the action of a d-parameter automorphism group. In particular we obtain the inverse-closedness of Besov algebras, Bessel potential algebras and approximation algebras of polynomial order in their defining algebra. By a proper choice of the group action these general results can be applied to alge- bras of infinite matrices and yield inverse-closed subalgebras of matrices with off-diagonal decay of polynomial order. Besides alternative proofs of known results we obtain new classes of inverse-closed subalgebras of matrices with off-diagonal decay. This work is a continuation and extension of results presented in [20]. 1. Introduction We aim at systematic constructions of inverse-closed subalgebras of a given Ba- nach algebra A. Recall that a subalgebra B of A is called inverse-closed in A, if (1) b ∈ B invertible in A implies b−1 ∈ B . Many equivalent notions for this relation are used in the literature, e.g.,one says that B is a spectral subalgebra of A [30], or B is spectrally invariant in A, see [19] for a collection of synonyms. A prototypical result is Wiener's Lemma, which states precisely that the Wiener algebra of trigonometric series with absolutely convergent coefficients is inverse- closed in the algebra of continuous functions on the torus. Another example is the algebra Cm(X) of m times continuously differentiable functions on a closed interval X, which is inverse-closed in C(X) by the iterated quotient rule (note that the algebra property of Cm(X) follows from the iterated product rule). Our original interest was the construction of Banach algebras of matrices with off-diagonal decay that are inverse-closed in B(ℓ2), the bounded operators in ℓ2, see [20]. A key result is Jaffard's theorem. Theorem ([25]). If the entries of the matrix A satisfy A(k, l) ≤ Ck − l−r for some C > 0 and r > 0, and A is invertible in B(ℓ2), then A−1(k, l) ≤ C′k − l−r for some C′ > 0. In other words Jaffard's theorem states that the Banach algebra C∞ r , consisting r = supk,lA(k, l)(1 + k − lr), is inverse- of matrices A with finite norm kAkC∞ closed in B(ℓ2). Generalizations and variants of this theorem have been obtained in, e.g., [5, 7, 21, 22, 25]. Date: May 18, 2017. 2000 Mathematics Subject Classification. 41A65, 42A10, 47B47. Key words and phrases. Banach algebra, matrix algebra, Besov spaces, Bessel potential spaces, inverse closedness, spectral invariance, off-diagonal decay, automorphism group, Jackson-Bernstein theorem. A. K. was supported by National Research Network S106 SISE of the Austrian Science Foun- dation (FWF) and the FWF project P22746N13. 1 2 ANDREAS KLOTZ Inverse-closed subalgebras are often related to the concept of smoothness, an el- ementary example is again the algebra Cm(X). In more generality, the domain of a densely defined, closed and symmetric derivation on the C∗ algebra A was shown to be inverse-closed in A by Bratteli and Robinson [11], an extension of this result to general Banach algebras was given in [26]. Related concepts of smooth- ness that define inverse-closed subalgebras are the differentials norms [10], the Dp algebras [26, 27], or the Leibniz seminorms [36]. Although it might be not immediately obvious the examples of matrices with off-diagonal decay fit into this picture, as off-diagonal decay can be decribed by smoothness conditions. In particular, the formal commutator δ(A) = [X, A] , X = Diag((k)k∈Z), is a derivation on B(ℓ2), and its domain defines an algebra of matrices with off-diagonal decay that is inverse-closed in B(ℓ2) [20, 3.4]. The proof of the spectral invariance of B ⊆ A makes often detailed use of some specific properties of the involved algebras. The standard proof for Wiener's Lemma is a prime example of an application of the Gelfand theory. Proofs of the spectral invariance of Banach algebras of matrices involve the theorem of Bochner-Philips [5, 6], interpolation arguments [21, 40, 41] or commutator estimates [25]. As it turns out, all of these proof methods use some related concepts of smoothness [28]. In a previous publication [20] we obtained systematic constructions of inverse- closed subalgebras of a given Banach algebra with additional smoothness condi- tions. In particular it was shown in [20] that (1) the domain of a (not necessaryly densely defined) closed derivation of a symmetric Banach algebra A is inverse-closed in A, (2) the subalgebra C(A) of continuous elements of a Banach algebra A with d-parameter automorphism group Ψ and the associated Holder-Zygmund spaces Λ∞ r (A) are inverse-closed in A, (3) the approximation spaces of polynomial order of a symmetric Banach al- gebra A are inverse-closed in A, where the approximating subspaces are adapted to the algebra multiplication (see Section 2.4). Applied to subalgebras of B(ℓ2) the theory yields scales of inverse-closed subalgebras of matrices with off-diagonal decay, including the Banach algebras of matrices used in the literature cited above, but also new classes of inverse-closed subalgebras of matrices with off-diagonal decay constructed by approximation with banded matrices. In this article we generalize the approach (2). In its essence this approach is based on the product and quotient rules of real analysis. The identities (2) ∆t(f g) = Ttf ∆tg + ∆tf g , ∆t(1/f ) = − ∆tf Ttf · f , where Ttf (x) = f (x − t) is the translation operator, imply that the smoothness of f (resp. g) is preserved by the product f g, and by the reciprocal 1/f . To give an example, the membership of the continuous function f in the Holder-Zygmund space Λ∞ r (L∞(R)) for 0 < r < 1 is defined by the condition k∆tf k∞ ≤ Ctr for some C > 0, and Equation (2) implies that k∆t(1/f )k∞ ≤ C′tr for a constant C′ > 0. In [20] we have (amongst other things) adapted this approach to more general Banach algebras. If translation is replaced by the action of a d-parameter automor- phism group on the Banach algebra A, the noncommutative form of the product and SPECTRAL INVARIANCE OF BESOV-BESSEL SUBALGEBRAS 3 quotient rule (2) impliy that the noncommutative Holder-Zygmund space Λ∞ is an inverse-closed subalgebra of A [20, 3.21]. r (A) In this article we extend this approach to cover Besov and Bessel potential spaces. In more detail, the organization of the paper is as follows. After introducing notation we describe some classes of matrices that will serve as examples for the theory to be developed, and we define the approximation spaces needed. In Section 3 we introduce smoothness for a Banach algebra A by the action of a d-parameter automorphism group, and review basic properties. Besov spaces are defined in Section 3.2, and it is proved that they form inverse-closed subalgebras of A. A useful property not only for simplifying several proofs but also of conceptual interest is a reiteration theorem for Besov spaces (Theorem 3.7). In [2] a similar theorem was proved by interpolation methods for Besov spaces of operators. An examination of the details of this proof shows that it uses similar ingredients (and is of similar complexity) as ours, which is valid in the more general setting of a Banach algebra with automorphism group. Instead of interpolation theory the proof given here uses estimates for the moduli of smoothness. In Section 3.3 we identify Besov spaces as approximation spaces and obtain a Littlewood-Paley-decomposition of them. For the discussion of Bessel potential spaces Pr(A) in Section 3.4 we introduce the concept of Cw-continuity [2, 3] in order to cover relevant examples of Banach algebras of matrices with off-diagonal decay. A description of Bessel potential spaces by hypersingular integrals is used to prove the algebra properties and the inverse-closedness of Pr(A) in A. Again, we illustrate the abstract concepts by constructing subalgebras of matri- ces with off-diagonal decay. The membership of a matrix A in a Besov or Bessel subalgebra of B(ℓ2) is then equivalent to a form of off-diagonal decay, so this appli- cation of the general theory describes scales of inverse-closed subalgebras of matrices with off-diagonal decay. Long proofs have been moved to the appendices. Acknowledgments. The author wants to thank Karlheinz Grochenig for many helpful discussions. 2. Resources ∗ = Cd \ {0}, and ∗ = Rd \ {0}. The symbol ⌊x⌋ denotes the greatest integer smaller or equal to the 2.1. Notation. The d-dimensional torus is Td = Rd/Zd. Let Cd Rd real number x. xα = xα1 A multi-index α = (α1, . . . , αd) is a d-tuple of nonnegative integers. We set d f (x) is the partial derivative. The j=1 αj, and β ≤ α means that βj ≤ αj for j = 1, . . . , d. d , and Dαf (x) = ∂α1 1 · · · ∂αd 1 · · · xαd degree of xα is α = Pd More generally, xp =(cid:0)Pd k=1x(k)p(cid:1)1/p denotes the p-norm on Cd. Positive constants will be denoted by C, C′,C1,c, etc., The same symbol might denote different constants in each equation. If f and g are positive functions, f ≍ g means that C1f ≤ g ≤ C2f . We sometimes use the notation f . g (f & g) to express that there is a constant C > 0 such that f ≤ Cg (f ≥ Cg). standard dual pairing between ℓp(Zd) and its dual ℓp′ The standard basis of ℓp(Zd) is ek = (δjk)j∈Zd , hx, yi = Pk∈Zd x(k)y(k) is the scalar product as hx, yi =Pk∈Zd x(k)y(k). This should not lead to confusion. A submultiplicative weight on on Zd is a positive function v : Zd → R such that v(0) = 1 and v(x + y) ≤ v(x)v(y) for x, y ∈ Zd. The standard polynomial weights are vr(x) = (1 + x)r for r ≥ 0. The weighted spaces ℓp w(Zd) are defined by the norm kxkℓp w(Zd) = kxwkℓp(Zd). If w = vr we will simply write kxkℓp (Zd). If p = 2, we define the r (Zd). 4 ANDREAS KLOTZ The Schwartz space of rapidly decreasing functions on Rd is denoted by S (Rd). The Fourier transform of f ∈ S (Rd) is F f (ω) = f (ω) =RRd f (x)e−2πiω·x dx. This definition is extended by duality to S ′(Rd), the space of tempered distributions. The same symbols are also used for the Fourier transform on Zd and Td. The continuous embedding of the normed space X into the normed space Y is denoted as X ֒→ Y . The operator norm of a bounded linear mapping A : X → In the special case of operators A : ℓ2(Zd) → ℓ2(Zd) we write Y is kAkX→Y . kAkB(ℓ2(Zd)) = kAkℓ2(Zd)→ℓ2(Zd) or simply kAkB(ℓ2). We will consider Banach spaces with equivalent norms as equal. 2.2. Inverse closed subalgebras of Banach algebras. All Banach algebras are assumed to be unital. To verify that a Banach space A with norm k kA is a Banach algebra it is sufficient to prove that kabkA ≤ CkakAkbkA for some constant C. The expression kak′ A = supkbkA=1kabkA is an equivalent norm on A and satisfies kabk′ A. A ≤ kak′ Akbk′ Definition 2.1 (Inverse-closedness). If A ⊆ B are Banach algebras with common multiplication and identity, we call A inverse-closed in B, if a ∈ A and a−1 ∈ B implies a−1 ∈ A. (3) Inverse-closedness is equivalent to spectral invariance. This means that the spec- trum σA(a) = {λ ∈ C : a − λ not invertible in A} of an element a ∈ A satisfies σA(a) = σB(a), for all a ∈ A. The relation of inverse-closedness is transitive: If A is inverse-closed in B and B is inverse-closed in C, then A is inverse-closed in C. Remark. Spectral invariance is a generalization of Wiener's Lemma, which states precisely that the Wiener algebra F ℓ1(Zd) of absolutely convergent Fourier series is inverse-closed in C(Td). See [19] for a concise overview of the importance of the concept of inverse-closedness. 2.3. Examples of Matrix Algebras. To describe the most common forms of off- diagonal decay, let us fix some notation. An infinite matrix A over Zd is a function A : Zd × Zd → C. The m-th side diagonal of A is the matrix A(m) with entries A(m)(k, l) =(A(k, l), 0, k − l = m, otherwise. A matrix A is banded with bandwidth N , if Let 1 < p ≤ ∞, r > d(1 − 1/p), or p = 1, and r ≥ 0. The space Cp r consists of all A = Xm∞≤N A(m). matrices A with finite norm kAkCp r =(cid:18)Xk∈Zd Xl∈Zd with the standard change for p = ∞. A(l, l − k)p(1 + k)rp(cid:19)1/p The following special cases have obtained particular interest. The Jaffard algebra C∞ r consists of the matrices A for which A(k, l) ≤ C(1 + k − l)−r, so the norm of C∞ r describes polynomial decay off the diagonal. The algebra of convolution-dominated matrices C1 r , r ≥ 0, (sometimes called the Baskakov-Gohberg-Sjostrand algebra) consists of all matrices A, for which there is a h ∈ ℓ1 r(Zd) such that Ax(k) ≤ h ∗ x(k), where x denotes the vector with components (x(k))k∈Zd . SPECTRAL INVARIANCE OF BESOV-BESSEL SUBALGEBRAS 5 If 1 < p ≤ ∞ and r > d(1 − 1/p) or p = 1 and r ≥ 0 the Schur algebra Sp r is defined by the norm kAkS p r = maxn sup k∈Zd(cid:0)Xl∈Zd A(k, l)pvr(k − l)p(cid:1)1/p , sup l∈Zd(cid:0)Xk∈Zd A(k, l)pvr(k − l)p(cid:1)1/po with the standard change for p = ∞. Remarks. (1) The scales Sp S∞ r = C∞ r . (2) It follows immediately from the definitions that Cp r ֒→ Sp r . r and Cp r are identical at the endpoint p = ∞, i.e. We note that the norms above depend only on the absolute values of the matrix entries. Precisely, a matrix norm on A is called solid, if B ∈ A and A(k, l) ≤ B(k, l) for all k, l implies A ∈ A and kAkA ≤ kBkA. In particular, for a solid norm we have k AkA = kAkA, where A is the matrix with entries A(k, l) = A(k, l) for k, l ∈ Zd. The following result summarizes the main properties of the matrix classes Cp r and Sp r . See [5, 21, 25, 40] for proofs. Proposition 2.2. Assume that A is one of the matrix classes Cp r for r > d(1 − 1/p), if p > 1, and r ≥ 0, if p = 1. Then A is a solid Banach ∗-algebra with respect to matrix multiplication and taking adjoints as the involution. Every A is continuously embedded into the algebra B(ℓp(Zd)) of bounded operators on ℓp(Zd) for 1 ≤ p ≤ ∞. With the exception of S1 0 [42] every class A is inverse-closed in B(ℓp(Zd)), 1 ≤ p ≤ ∞. In particular, A is symmetric. r or Sp In the sequel we will construct algebras that are inverse-closed in one of the r , and are therefore inverse-closed in B(ℓ2(Zd)) by Proposi- r , Sp standard algebras Cp tion 2.2. We generalize the definitions above. Definition 2.3. A matrix algebra A (over Zd) is a Banach algebra of matrices that is continuously embedded in B(ℓ2(Zd)). We drop the reference to the index set Zd whenever possible. Lemma 2.4. If A is a matrix algebra, the selection of matrix elements is contin- uous. Proof. A(k, l) = hAek, eli ≤ kAkB(ℓ2) ≤ CkAkA. (cid:3) 2.4. Approximation Spaces and Algebras. Let the index set Λ be either R+ 0 or N0. An approximation scheme on the Banach algebra A is a family (Xσ)σ∈Λ of closed subspaces of A that satisfy X0 = {0}, Xσ ⊆ Xτ for σ ≤ τ , and Xσ · Xτ ⊆ If A possesses an involution, we further assume that 1 ∈ X1 Xσ+τ , σ, τ ∈ Λ. and Xσ = X ∗ σ for all σ ∈ Λ. The σ-th approximation error of a ∈ A by Xσ is Eσ(a) = inf x∈Xσ ka − xkA. We define approximation spaces E p r (A) by the norm (4) kakp E p r = Ek(a)p(k + 1)rp 1 k + 1 , for Λ = N0 , ∞ Xk=0 for 1 ≤ p < ∞ with the obvious change for p = ∞. If Λ = R+ 0 an equivalent norm is kakp σ+1 . Algebra properties of approximation spaces of E p r approximation spaces are discussed in [1, 20]. In particular, in [20] the following result is proved. 0 Eσ(a)p(σ + 1)rp d σ =R ∞ Proposition 2.5. If A is a symmetric Banach algebra and (Xσ)σ∈Λ an approxi- mation scheme, then E p r (A) is inverse-closed in A. 6 ANDREAS KLOTZ If A is a matrix algebra and TN = TN (A) denotes the set of matrices in A with bandwidth smaller than N , TN = {A ∈ A : A = Xk∞<N A(k)} then the sequence (TN )N ≥0 is an approximation scheme for A. The closure of all banded matrices in A is the space of band-dominated matrices in A [34, 35]. In [20] we obtained the following constructive description of E ∞ r (C1 0 ): The ap- proximation space E ∞ r (C1 0 ) consists of all matrices A satisfying k A(0)kB(ℓ2) < ∞, 2rkX2k≤l<2k+1 k A(l)kB(ℓ2) = 2rkX2k≤l<2k+1 sup m∈Zd A(m, m − l) ≤ C for all k ≥ 0. Theorem 3.14 is a more general result of this type. 3. Smoothness in Banach Algebras An important observation in [20] was that the off-diagonal decay of matrices can be described by smoothness properties, using derivations and the action of the automorphism group χt(A) = P A(k)e2πikt. In our treatment we focused on Holder-Zygmund spaces and on spaces of m times differentiable elements. Now we extend our research and cover the more general Besov and Bessel potential spaces. We also establish the isomorphism between Besov spaces and approximation spaces of polynomial order. In all cases we obtain results on the inverse-closedness of the smoothness spaces in their defining algebra. It turns out that the investigations can be carried out with no additional ef- fort for Banach algebras with the bounded action of a d-parameter automorphism group. Here we obtain new methods for the construction of scales of inverse-closed subalgebras. 3.1. Automorphism Groups and Continuity. Let A be a Banach algebra. A (d-parameter) automorphism group acting on A is a set of Banach algebra auto- morphisms Ψ = {ψt}t∈Rd of A that satisfy the group properties ψsψt = ψs+t for all s, t ∈ Rd. If A is a ∗-algebra we assume that Ψ consists of ∗-automorphisms. simplify some proofs, we assume that Ψ is uniformly bounded, that is, In order to MΨ = sup t∈Rd kψtkA→A < ∞ . The abstract theory works for more general group actions [2, 23, 43]. An element a ∈ A is (strongly) continuous, if (5) kψt(a) − akA → 0 for t → 0. The continuous elements of A are denoted by C(A). For t ∈ Rd \ {0} the generator δt is (6) δt(a) = lim h→0 ψht(a) − a h The domain D(δt, A) of δt is the set of all a ∈ A for which this limit exists. The canonical generators of Ψ are (δek )1≤k≤d, and Ψ is the automorphism group generated by (δek )1≤k≤d. If α ∈ Nd ed . In [20, 3.15] precise conditions are given, under which condition two derivations commute. In particular, this is true for all cases that will be encountered in this text. Each generator δt is a closed derivation, that is δt(ab) = aδt(b) + δt(a)b for all a, b ∈ D(δt, A), and the operator δt is a closed operator on its domain. If A is a Banach ∗-algebra, then δt is a ∗-derivation [12]. 0 is a multi-index, then δα = δα1 e1 · · · δαd SPECTRAL INVARIANCE OF BESOV-BESSEL SUBALGEBRAS 7 Proposition 3.1 ([20, 3.4]). If A is symmetric, and δ is a closed ∗-derivation then D(δt, A) is inverse-closed in A. We call the action of Ψ periodic if ψt+ek = ψt for all 1 ≤ k ≤ d and all t ∈ Rd. Definition 3.2. A matrix algebra A is called homogeneous [16, 17], if χt : A 7→ MtAM−t, χt(A)(k, l) = e2πi(k−l)·tA(k, l) for all k, l ∈ Zd and t ∈ Rd define uniformly bounded mappings on A, where Mt, t ∈ Rd, is the modulation operator Mtx(k) = e2πik·tx(k), k ∈ Zd. Clearly χ = {χt}t∈Rd defines an automorphism group on A. The algebra of bounded operators on ℓ2 is a homogeneous matrix algebra, and so are all solid matrix algebras. In the literature on group actions it is often assumed that Ψ is strongly contin- uous on all of A, i.e. A = C(A). This is no longer true for most matrix algebras, and in general C(A) is a closed and inverse-closed subalgebra of A [20, 3.14]. In [20] the spaces C(A) have been identified for the algebras C1 r , C∞ r and Sp r . the opportunity to state the full result for the algebras Cp r , S1 r . We use Proposition 3.3. If A is one of the algebras C∞ C(A) 6= A. r , Sp r , B(ℓ2(Zd)) for r ≥ 0, then C(Cp r ) = Cp r , 1 ≤ p < ∞ , C(C∞ r ) = {A ∈ C∞ r : lim k∞→∞ k A(k)kC∞ r = lim k A(k)kB(ℓ2)(1 + k)r = 0} k∞→∞ C(Sp r ) = {A ∈ Sp r : lim N→∞ lim N→∞ sup k∈Zd Xs∞>N k∈Zd Xs∞>N sup A(k, k − s)p(1 + s)rp = 0 and A(k − s, k)p(1 + s)rp = 0} . The method of proof is as in [20]. 3.2. Besov Spaces. The theory of vector valued Besov spaces is well established [9, 13, 31, 44]. The main results of this section are the algebra properties of vector- valued Besov spaces derived from a given Banach algebra A. Though possibly known, we were not able to find any references, so full proofs of the results are included. Let A be a Banach algebra with automorphism group Ψ. Define the kth differ- t = (ψt − id)k, t ∈ Rd. For a step size h > 0, the modulus of h(a) = supt≤hk∆takA. If k > 1, the kth modulus ence operator as ∆k continuity of a ∈ A is ωh(a) = ω1 of smoothness of a is ωk h(a) = supt≤hk∆k h(a, A) = ωk t akA. Definition 3.4. Let 1 ≤ p ≤ ∞, r > 0, l = ⌊r⌋ + 1. The (vector valued) Besov space Λp r(A) consists of all a ∈ A for which the seminorm aΛp r (A) =( (cid:0)RR+ (h−rωl kakA + suph>0 h−rωl h(a))p dh h (cid:1)1/p , h(a) , 1 ≤ p < ∞ p = ∞ is finite. The parameter r is the smoothness parameter. The Besov norm is kakΛp r (A) = kakA + aΛp r (A). 8 ANDREAS KLOTZ Actually, replacing l by any integer k > ⌊r⌋ in the preceding definition yields an r(A). In addition, we will need the following norm equiva- equivalent norm for Λp lences. (7) kakΛp (t−rk∆k r (A) ≍ kakA +(cid:16)ZRd ≍ kakA +(cid:16) ∞ Xl=0(cid:0)2rlωk td(cid:17)1/p t akA)p dt 2−l(a)(cid:1)p(cid:17)1/p . If l ∈ N0, and l ≤ r, these norms are further equivalent to kakA + Xα=l kδα(a)kΛp r−l(A) . The Besov spaces Λp 1 ≤ p, q ≤ ∞ and 0 < r < s, then Λp See, e.g., [9, 13, 31, 44] for these and other basic properties. r(A) are Banach spaces for all 1 ≤ p ≤ ∞ and r > 0. r(A) ֒→ Λq r(A). If p < q then Λp s(A) ֒→ Λq If r(A). Lemma 3.5. If A is a Banach algebra with (bounded) automorphism group Ψ, then Ψ is a bounded automorphism group on Λp r(A) for every 1 ≤ p ≤ ∞, and r > 0. Proof. Assume that p < ∞. If a ∈ Λp r(A), k > ⌊r⌋ and s ∈ Rd, then for every s > 0 kψsakΛp r (A) = kψsakA +(cid:16)ZRd (t−rk∆k t ψsakA)p dt ≤ MΨkakΛp r (A) , td(cid:17)1/p since δk t ψs = ψsδk t , so ψs is bounded on Λp r(A). The proof for p = ∞ is similar. (cid:3) Proposition 3.6 ([13, 3.1.5, 3.4.3]). If k > ⌊r⌋ then C(Λp C(Λ∞ r(A)) = Λp r (A)) = λ∞ r(A), r (A) = {a ∈ A : 1 ≤ p < ∞ , lim h→0 h−rωk h(a) = 0} . Does the iteration of the construction of Besov spaces yield refined smoothness spaces? Theorem 3.7 (Reiteration theorem). If 1 ≤ p, q ≤ ∞ and r, s > 0 then (8) Λq s(Λp r(A)) = Λq r+s(A) . A proof of is in appendix A. Remarks. A proof of the reiteration formula for the Banach algebra of bounded operators on a Banach space X and the automorphism group ψ obtained by conju- gation with an automorphism group on X has been given in[2], using interpolation theory. We think the reiteration formula is of some conceptual interest. Note that the classical notion of Besov spaces on Rd does not even allow to formulate the result. We use (8) to simplify proofs of approximation results. The main result of this section treats the algebra properties of Besov spaces. Theorem 3.8. Let A be a Banach algebra with automorphism group Ψ. For all parameters 1 ≤ p ≤ ∞ and r > 0, the Besov space Λp r(A) is a Banach subalgebra of A. Moreover, Λp r(A) is inverse-closed in A. Proof. We treat the case r < 1 first. To show that Λp use the identity r(A) is a Banach algebra we (9) ∆t(ab) = ψt(a)∆t(b) + ∆t(a)b . SPECTRAL INVARIANCE OF BESOV-BESSEL SUBALGEBRAS 9 Taking norms we obtain k∆t(ab)kA ≤ kψt(a)kAk∆t(b)kA + k∆t(a)kAkbkA ≤ MΨkakAk∆t(b)kA + k∆t(a)kAkbkA . This implies a similar relation for the Besov-seminorms, namely, abΛp r (A) ≤ MΨkakAbΛp r (A) + kbkAaΛp r (A). So kabkΛp r(A) = kabkA + abΛp r (A) ≤ CkakΛp r (A)kbkΛp r (A) , and the assertion follows. To show that Λp r(A) is inverse-closed in A we assume that a ∈ Λp ible in A. It is sufficient to verify that a−1Λp computation we obtain r(A) is invert- r (A) is finite. By a straightforward (10) ∆t(a−1) = −ψt(a−1) ∆t(a) a−1 . This implies that a−1 has a finite Λp r(A)-norm. In the general case we can use the reiteration theorem (Theorem 3.7) and the transitivity of inverse-closedness, and prove the statement by induction. Assume that the statement is proved for all smoothness parameters smaller than s > 0. As Λp r−s(Λp s(A)) is inverse-closed in Λp s(A) is inverse-closed in A by hypotheses, the theorem is proved. (cid:3) s(A)) for r > s, the preceding argument yields Λp s(A) for s < r < s + 1. As Λp r(A) = Λp r−s(Λp 3.3. Characterization of Besov Spaces as Approximation Spaces. As in the case of function spaces, Λp r(A) can be characterized by approximation properties. This was carried out for Λ∞ r (A) in [20], so our treatment is very brief. We focus on the approximation of a Banach algebra A with automorphism group Ψ by smooth elements. Definition 3.9 (Bernstein inequality). An element a ∈ A is σ-bandlimited for σ > 0, if there is a constant C such that for every multi-index α (11) kδα(a)kA ≤ C(2πσ)α . An element is bandlimited, if it is σ-bandlimited for some σ > 0. If A is a Banach algebra with automorphism group Ψ, then, X0 = {0}, Xσ = {a ∈ A : a is σ-bandlimited}, σ > 0 is an approximation scheme for A [20, Lemma 5.8]. From now on we use this approximation scheme without further notice. In particular, if A is ahomogeneous matrix algebra, we obtain the following characterization of bandlimited elements Proposition 3.10 ([20, 5.7]). A matrix A is banded with bandwidth N in the homogeneous matrix algebra A, if and only if it is N -bandlimited with respect to the group action {χt}. Theorem 3.11 (Jackson Bernstein Theorem). Let A be a Banach algebra with automorphism group Ψ, and assume that r > 0 and 1 ≤ p ≤ ∞. If {Xσ : σ ≥ 0} is the approximation scheme of bandlimited elements, then (12) The proof is in appendix B. Λp r(A) = E p r (A) . 10 ANDREAS KLOTZ Littlewood-Paley Decomposition. The norms of Besov spaces are not easily com- putable. An equivalent explicit norm for these spaces can be obtained by means of a Littlewood-Paley decomposition. First we need some technical preparation: If µ ∈ M(Rd) and a ∈ C(A), the action of µ on a is defined by (13) µ ∗ a =ZRd ψ−t(a)dµ(t). This action is a generalization of the usual convolution and satisfies similar prop- erties: If f ∈ C∞ c (Rd) then (14) kµ ∗ akA ≤ MΨkµkM(Rd) kakA . δα(f ∗ a) = Dαf ∗ a ∈ C(A) for every multi-index α. See [13] for details and proofs. In particular, if the group action is periodic, the action of µ on a is (15) µ ∗ a =ZTd ψ−t(a) dµ(t) = Xk∈Zd F (µ)(k)a(k) , converges in the C1-sense. Now assume that ϕ ∈ S (Rd) with supp ϕ ⊆ {ω ∈ Rd : 2−1 ≤ ω∞ ≤ 2}, ϕ(ω) > 0 where a(k) = RTd ψ−t(a)e2πik·t dt is the k-th Fourier coefficient of a and the sum for 2−1 < ω∞ < 2, and Pk∈Z ϕ(2−kω) = 1 for all ω ∈ Rd \ {0}. Set ϕk(ω) = ϕ(2−kω), k ∈ N0, so ϕk(x) = 2kdϕ0(2kx), and let ϕ−1 = 1 − P∞ { ϕk}k≥−1 is a dyadic partition of unity. k=0 ϕk. Then Proposition 3.12. Let { ϕk}k≥−1 be a dyadic partition of unity, and 1 ≤ p ≤ ∞, r > 0. An element a ∈ A is in Λp r(A), if and only if (16) Paley decomposition P∞ (cid:18) ∞ Xk=−1 2rkpkϕk ∗ akp A(cid:19)1/p < ∞ . The expression (16) defines an equivalent norm on Λp r(A). Moreover the Littlewood- k=0 ϕk ∗ a converges to a in the norm of A. The special case p = ∞ was proved in [20] with a weak type argument. This approach does not work for p < ∞, so we adapt a proof in [9], see Appendix C. Approximation of Polynomial Order in Homogeneous Matrix Spaces. Lemma 3.13. If {ϕk}k≥−1 is a dyadic partition of unity and the action of Ψ on A is periodic, then for a ∈ C(A), (17) ϕk ∗ a = X⌊2k−1⌋≤l∞<2k+1 ϕk(l)a(l) . Proof. Let ϕΠ k (t) =Pl∈Zd ϕk(t + l) denote the periodization of ϕk. Then ϕk(l)e2πil·t ϕΠ k (t) = X⌊2k−1⌋≤l∞<2k+1 by Poisson's summation formula. Equation (17) now follows, combining (15) with ϕk ∗ a =ZRd ψ−t(a)ϕk(t) dt =ZTd ψ−t(a)ϕΠ k (t) dt (cid:3) Equation (17) allows us to obtain a characterization of the approximation spaces for homogeneous matrix algebras by the Littlewood-Paley decomposition of its el- ements. SPECTRAL INVARIANCE OF BESOV-BESSEL SUBALGEBRAS 11 Proposition 3.14. Let A be a homogeneous matrix algebra, r > 0, and Φ = {ϕk}k≥−1 a dyadic partition of unity. Then the norm on the approximation space E p r (A) = Λp r(A) is equivalent to (18) kAkE p If A is solid, then (19) kAkE p p A(cid:19)1/p . r (A) ≍(cid:18) ∞ Xk=0 2kpr(cid:13)(cid:13)(cid:13) X⌊2k−1⌋≤l∞<2k+1 r (A) ≍(cid:18) ∞ 2kpr(cid:13)(cid:13)(cid:13) X⌊2k⌋≤l∞<2k+1 Xk=−1 ϕk(l) A(l)(cid:13)(cid:13)(cid:13) A(cid:19)1/p A(l)(cid:13)(cid:13)(cid:13) p . Remark. If the matrix algebra A is solid, similar results can be obtained for ap- proximation spaces of non-polynomial order [28]. Proof. The results for general homogeneous matrix algebras follow from the Jackson Bernstein Theorem (Theorem 3.11) and the Littlewood-Paley decomposition. We still have to prove the norm equivalence (19). Set Ck = kP2k≤l<2k+1 A(l)kA. The solidity of A implies that, for k ≥ −1, Bk = kϕk ∗ AkA ≤ k X2k−1≤l∞<2k+1 A(l)kA = Ck−1 + Ck On the other hand, since φk−1 + φk + φk+1 ≡ 1 on {ξ : 2k−1 ≤ ξ2 ≤ 2k+1}, we obtain Ck ≤ Bk−1 + Bk + Bk+1. So kAkp k=−1 2kprCk, E p and this is (19). (cid:3) k=0 2kprBk ≍ P∞ r (A) ≍ P∞ We apply the preceding results to the example of Cp r (see Section 2.3 for the definition). We obtain E q s (Cp r ) = E q s+r(Cp 0 ) , kAkE q r (Cp s ) ≍(cid:16) ∞ Xj=0 2jq(r+s)(cid:0) X⌊2−1⌋≤k∞<2j kA[k]kp ℓ∞(Zd)(cid:1)q/p(cid:17)1/q ≍ kAkE q r+s(Cp 0 ) . In particular, E p s (Cp r ) = Cp r+s . If p 6= q these norms define new classes of inverse-closed subalgebras of B(ℓ2) with a form of off-diagonal decay suited to approximation with banded matrices. These results should be compared to the definition of discrete Besov spaces [32]. 3.4. Bessel Potential Spaces. Bessel potentials allow us to define an analogue of polynomial weights in a Banach algebra with an automorphism group. For homogeneous matrix algebras the Bessel potential spaces are weighted algebras. We define the Bessel kernel Gr by its Fourier transform, F Gr(ω) = (1 + 2πω2 2)−r/2 , r > 0. Cw groups. In analogy to the case of real functions we would like to define an element of the Bessel potential space Pr(A) as an element of the form a = Gr ∗ y for some y ∈ A. However, the action Gr ∗ y is defined only for y ∈ C(A). Using a weaker form of continuity for the action of the automorphism group we can extend the convolution "∗" to the whole algebra for all examples of matrix algebras in Section 2.3. 12 ANDREAS KLOTZ Definition 3.15 ([4, 12]). Let A be a Banach algebra with automorphism group Ψ. For a ∈ A, a′ ∈ A′ define Ga′,a(t) = ha′, ψt(a)i. Assume that A′ Ψ = {a′ ∈ A′ : Ga′,a is continuous for all a ∈ A} is a norm fundamental subspace of A′, that is kakA = sup{ha′, ai : a′ ∈ A′ Ψ, ka′kA′ ≤ 1} for all a ∈ A. Assume that A is equipped with the weak topology σ(A, A′ respect to the functionals in A′ (20) Ψ, and Ψ) with the convex hull of every σ(A, A′ Ψ)-compact set has σ(A, A′ Ψ)-compact closure. In this case we call Ψ a Cw group and denote A with the σ(A, A′ Aw, if necessary. Ψ) topology by Condition (20) ensures the existence of the "convolution integral" (13) as a Ψ) is quasi-complete, i.e., bounded Cauchy nets Pettis integral, see below. If σ(A, A′ converge, then condition (20) is automatically satisfied [29]. Example 3.16. (1) If A′ Ψ is the predual of A (in particular, if A is a von Neumann algebra) the quasi-completeness is a consequence of the Banach-Alaoglu theorem. (2) If A is a Banach function space in the sense of [8], and A′ Ψ is a norm fundamental order ideal of the Koethe dual A∼, then it is known that (A, σ(A, A′ Ψ)) is quasi-complete [8, 1.5.2]. (3) If A = C(A) then A′ Ψ = A′. It is well-known that ψt is strongly continuous at a ∈ A, if and only if it is continuous with respect to the σ(A, A′)- topology [13, 24], so in this case Aw = A. In this case the condition (20) is a consequence of the Krein-Smulian theorem. Remarks. If the group action is uniformly bounded the space A′ subspace of A′. Indeed, if a′ k ∈ A′ Ψ and a′ Ψ is a norm-closed lim t→0 Ga′,a(t) − Ga′,a(0) = lim t→0 = lim k k → a′ in norm, then ha′ ha′, ψt(a) − ai = lim t→0 lim k k, ψt(a) − ai = 0 . ha′ lim t→0 k, ψt(a) − ai We do not have general conditions when the action of χ on a homogeneous matrix algebra is a Cw-group. For the specific examples of matrix algebras introduced in Section 2.3 we can prove that χ is a Cw group. Example 3.17. Recall that B(ℓ2) is the dual of the trace class operators B1, B(ℓ2) = (B1)′ and the finite rank operators are dense in B1. Adapting a continuity argument from [16] we verify that (B(ℓ2))′ χ ⊇ B1. Indeed, for x, y ∈ ℓ2(Zd) and the rank one operator (x ⊗ y)z = hz, yix we obtain Gx⊗y,A(t) − Gx⊗y,A(t) = tr((x ⊗ y)χt(A)) − tr((x ⊗ y)A) = hx, (χt(A) − A)yi =hx, MtAM−t(y − Mty)i As limt→0kz − Mtzkℓ2(Zd) = 0 for every z ∈ ℓ2(Zd), it follows that Gx⊗y,A is continuous. So, if A′ is a finite rank operator then GA′,A is continuous. As A′ Ψ is norm closed in A′, and the finite rank operators are dense in B1 we obtain the continuity of GA′,A for all A′ ∈ B1. We have shown that the space A′ Ψ contains B1. This implies that A′ Ψ is norm fundamental, and so χ is a Cw-group on B(ℓ2). SPECTRAL INVARIANCE OF BESOV-BESSEL SUBALGEBRAS 13 Example 3.18. If A = Sp space on Z2d with r we can argue as follows: Let ℓ∞,p mr (Z2d) the mixed norm k(x(k, l))k,l∈Zd kℓ∞,p mr = sup k∈Zd (Xl∈Zd x(k, l)p(1 + k − l)rp)1/p and define (jx)(k, l) = x(l, k). Then we obtain the isometric isomorphism From [15, Lemma 1.12] (and using standard facts about sequence spaces, e.g [29, 30.3] we conclude that Sp r ∼= ℓ∞,p mr (Z2d) ∩ j(cid:0)ℓ∞,p mr (Z2d)(cid:1)(cid:1)∼ ∼= ℓ1,p′ mr (Z2d)(cid:1) . m−r (Z2d) + j(cid:0)ℓ1,p′ mr (Z2d) ∩ j(cid:0)ℓ∞,p (cid:0)ℓ∞,p It is routine to verify that GA′,A is continuous for A′ ∈ ℓ1,p′ and A ∈ Sp r , so Example 3.16 (2) verifies that Sp r is a Cw-group. m−r (Z2d)(cid:1) m−r (Z2d) + j(cid:0)ℓ1,p′ m−r (Z2d)(cid:1) We need the concept of Cw-groups not only to extend the action of a measure defined in (13) to the whole of A, but also to give a weak type description of this action. Proposition 3.19 ([3, 1.2]). If Ψ is a Cw- group for the Banach algebra A, then for each µ ∈ M(Rd) and each a ∈ A there is an element, denoted as µ ∗ a ∈ A, such that ha′, ψ−t(a)i dµ(t) ha′, µ ∗ ai =ZRd µ ∗ a =ZRd for all a′ ∈ A′ Ψ . As usual we write (21) ψ−t(a) dµ(t) . We obtain the norm inequality kµ ∗ ak ≤ MΨkakAkµkM(Rd) . Remarks. Clearly in special cases the existence of the integral (21) can be verified directly. In particular, if A = C(A) the integral exists in the sense of Bochner. The following result is straightforward. Proposition 3.20. If Ψ is a Cw- group for the Banach algebra A, and Ga′,a(t) = ha′, ψt(a)i for a′ ∈ A′ Ψ, a ∈ A, then kakA ≍ sup{kGa′,ak∞ : a′ ∈ A′ Ψ, ka′kA′ ≤ 1} . Moreover, (22) Ga′,µ∗a = µ ∗ Ga′,a . Before defining Bessel potential spaces we list properties of the Bessel kernel that will be needed in the sequel. Lemma 3.21 ([38, V.5]). r (L1(Rd)) , kGrkL1(Rd) = 1, (1) Gr ∈ Λ∞ (2) Gr ∗ Gs = Gr+s for all r, s > 0, (3) Gr ∗ S = {Gr ∗ ϕ : ϕ ∈ S } = S . Definition 3.22. Let A be a Banach space and Ψ a Cw- group acting on A (this includes the case A = C(A)). The Bessel potential space of order r > 0 is Pr(A) = Gr ∗ A = {a ∈ A : a = Gr ∗ y for some y ∈ A} with the norm kGr ∗ ykPr(A) = kykA. 14 ANDREAS KLOTZ We have to verify that the definition of the norm on Pr(A) is consistent, that is, we show that the convolution with Gr is injective on A. We use a weak type argument. Let y ∈ A with Gr ∗ y = 0. This is equivalent to Ga′,Gr∗y(t) = Gr ∗ Ga′,y(t) = 0 for all t ∈ Rd and all a′ ∈ A′ function ϕ ∈ S and obtain Ψ. Now we proceed as in [38, V.3.3]. We choose a test ZRd (Gr ∗ Ga′,y)(t)ϕ(t) dt =ZRd Ga′,y(t)(Gr ∗ ϕ)(t) dt = 0 . By Lemma 3.21 (3) the convolution with Gr is surjective on S , and so it follows that Ga′,y = 0 for all a′ ∈ A′ Ψ, that is, y = 0. An immediate consequence of Definition 3.22 is the embedding Pr(A) ֒→ A. Indeed, if a ∈ Pr(A), then a = Gr ∗ y for a y ∈ A, and (23) kakA ≤ kGrkL1(Rd)kykA = kGrkL1(Rd)kakPr(A). As Gr ∗ Gs = Gr+s for r, s > 0 we obtain a useful reiteration property for the Bessel potential spaces. Proposition 3.23. If A is a Banach algebra and Ψ a Cw-automorphism group on A, then for all r, s > 0 Pr(Ps(A)) = Pr+s(A) . Proof. We have to verify that Ψ is a Cw-automorphism group on Pr(A). For this we show that the dual pairing defined by ha′, Gr ∗ yiA′ Ψ×Pr(A) = ha′, yiA′×A yields a norm-fundamental subspace of Pr(A)′. As ha′, Gr ∗ yiA′ Ψ×Pr (A) ≤ ka′kA′kykA = ka′kA′kGr ∗ ykPr(A) Ψ×Pr(A) is continuous for every a′ ∈ A′ the mapping z 7→ ha′, ziA′ A′ definition t 7→ ha′, ψtziA′ Finally, A′ Ψ ⊃ Ψ. Moreover, a straightforward computation shows that ka′kPr(A)′ = ka′kA′. By Ψ and each z ∈ Pr(A). Ψ×Pr (A) is continuous for each a′ ∈ A′ Ψ is norm fundamental, as we have for z = Gr ∗ y Ψ, so Pr(A)′ sup{ha′, yiA′ Ψ×Pr(A) : a′ ∈ A′ Ψ, ka′kPr(A)′ ≤ 1} = sup{ha′, yiA′×A : a′ ∈ A′ =kykA = kzkPr(A) Ψ, ka′kA′ ≤ 1} 3.4.1. Characterization by Hypersingular Integrals. Lemma 3.24. If a ∈ Pr(A), then kakPr(A) ≍ supka′kA′ ≤1kGa′,akPr(L∞), where the dual pairing in Ga′,a is the one of A′ Ψ × A. Proof. Let a = Gr ∗ y. Then (cid:3) kakPr(A) =kykA ≍ sup kGa′,yk∞ ka′kA′ ≤1 kGr ∗ Ga′,ykPr(L∞) = sup ka′kA′ ≤1 = sup ka′kA′ ≤1 kGa′,Gr ∗ykPr(L∞) . (cid:3) We state a special case of a result by Wheeden [46] (see also [37],[38, V.6.10]). Proof. We only show the second statement. As (cid:13)(cid:13)(cid:13)Zǫ≤t2 ∆t(a) tr 2 dt td 2(cid:13)(cid:13)(cid:13)A 2 ∆t(a) ∆t(a) tr 2 t−r kakA + sup kakA + sup ǫ>0(cid:13)(cid:13)(cid:13)Zt2≥ǫ ǫ>0(cid:13)(cid:13)(cid:13)Zǫ≤t2≤1 ≤(cid:13)(cid:13)(cid:13)Zǫ≤t2≤1 2(cid:13)(cid:13)(cid:13)A ≤(cid:13)(cid:13)(cid:13)Zǫ≤t2≤1 2(cid:13)(cid:13)(cid:13)A ≤ C(kakA +(cid:13)(cid:13)(cid:13)Zǫ≤t2≤1 ∆t(a) tr 2 ∆t(a) tr 2 dt td dt td . . dt td dt td 2(cid:13)(cid:13)(cid:13)A 2(cid:13)(cid:13)(cid:13)A +(cid:13)(cid:13)(cid:13)Zt2≥1 2(cid:13)(cid:13)(cid:13)A + (1 + MΨ)kakAZt2≥1 2(cid:13)(cid:13)(cid:13)A dt td dt td ∆t(a) tr 2 ) , ∆t(a) tr 2 t−r 2 dt td 2 (cid:3) SPECTRAL INVARIANCE OF BESOV-BESSEL SUBALGEBRAS 15 Theorem 3.25. Let 0 < r < 2. A function f is an element of Pr(L∞(Rd)) if and only if f ∈ L∞(Rd) and (24) If (24) holds, (25) t−r 2 ∆t(f ) sup ǫ>0(cid:13)(cid:13)(cid:13)Zt2≥ǫ < ∞. dt td 2(cid:13)(cid:13)(cid:13)L∞(Rd) kf kL∞(Rd) + sup t−r 2 ∆t(f ) ǫ>0(cid:13)(cid:13)(cid:13)Zt2≥ǫ < ∞ dt td 2(cid:13)(cid:13)(cid:13)L∞(Rd) defines an equivalent norm on Pr(L∞(Rd)). Combining Lemma 3.24 with Theorem 3.25 we obtain the first statement of the following theorem. Theorem 3.26. Let A be a Banach algebra and Ψ a Cw-automorphism group acting on it. For 0 < r < 2 the norm kakPr(A) is equivalent to (26) This norm is further equivalent to the proof of the other inequality works in a similar way. Next we compare Bessel potential spaces with Besov spaces. Proposition 3.27. If A is Banach algebra with Cw-automorphism group Ψ, then Λ1 r(A) ֒→ Pr(A) ֒→ Λ∞ r (A) if r > 0. Proof. For the proof of the embedding Pr(A) ֒→ Λ∞ Gr ∗ y, y ∈ A. The seminorm aΛ∞ r (A) can be estimated for k > ⌊r⌋ as r (A) let a ∈ Pr(A) with a = aΛ∞ r (A) = sup t6=0 k∆k t (Gr ∗ y)kA tr = sup k t6=0 ∆k t (Gr) tr ∗ ykA ≤ sup k t6=0 ∆k t (Gr) tr kL1(Rd)kykA = kGrkΛ∞ r (L1)kakPr(A) , and this is the desired embedding. We still have to verify the first inclusion. Assume first that 0 < r < 1. By Theorem 3.26, for an a ∈ Pr(A) kakPr(A) ≍ kakA + sup t−r 2 ∆t(a) ǫ>0(cid:13)(cid:13)(cid:13)Zt2≥ǫ t−r 2 k∆t(a)kA dt td 2 dt td 2(cid:13)(cid:13)(cid:13)A ≤ kakA +ZRd = kakΛ1 r(A) . 16 ANDREAS KLOTZ In the general case we proceed by induction. Assume that the statement is true for all positive values up to s > 0, and s < r < s + 1. Then Λ1 r(A) = Λ1 r−s(Λ1 s(A)) ⊆ Pr−s(Λ1 s(A)) ⊆ Pr−s(Ps(A)) = Pr(A) , where we have used the reiteration theorems for the Bessel and the Besov spaces (Theorem 3.7). (cid:3) Another application of the reiteration theorem and the representation of the norm of Pr(A) by the hypersingular integral (26) shows how Besov spaces and Bessel potential spaces interact. Proposition 3.28. If A is a Banach algebra with Cw-automorphism group Ψ, then for all r, s > 0 and 1 ≤ p ≤ ∞ (27) Pr(Λp s(A)) = Λp s(Pr(A)) = Λp r+s(A) . Proof. Again, we need to know first that Ψ is a Cw-automorphism group on Λp If p < ∞ then C(Λp follows from r(A). If p = ∞ the assertion r(A) by Proposition 3.6. r(A)) = Λp kakΛ∞ r (A) = sup t6=0 sup{ha′, ∆k t (a) tr i : a′ ∈ A′ Ψ, ka′kA′ ≤ 1} . The details are similar to the proof of the analogue statement in Proposition 3.23 and are left to the reader. Using the reiteration theorems for Bessel potential spaces and Besov spaces, it suffices to prove the proposition only for 0 < r, s < 1. We show first that Pr(Λp s(A)), so a = Gr ∗ y with y ∈ Λp s(Pr(A)). Assume that a ∈ Pr(Λp s(A). We obtain the following estimate. s(A)) ֒→ Λp kakp Λp s (Pr(A)) =ZRd =ZRd =ZRd =ZRd k∆t(a)kp tsp Pr(A) dt td k∆t(Gr ∗ y)kp Pr(A) tsp kGr ∗ ∆t(y)kp Pr(A) tsp k∆t(y)kp A tsp dt td dt td dt td =kykp s (A) = kGr ∗ ykp Λp Pr(Λp s (A)) . Now let a = Gr ∗ y ∈ Pr(Λp s(A)). Then kakp s (A)) =kGr ∗ ykp Pr(Λp s (A)) = kykp Λp Pr(Λp s (A) =ZRd k∆t(y)kp A tsp dt td 2 =ZRd =kGr ∗ ykΛp k∆t(Gr ∗ y)kp Pr(A) tsp dt td 2 s (Pr(A)) . s (Pr (A)) = kakΛp Consequently Pr(Λp s(Pr(A)). Finally, Proposition 3.27 implies that s(A)) = Λp s(Λ1 r(A)) ֒→ Λp Λp s(Pr(A)) ֒→ Λp s(Λ∞ r (A)) , and the first and last space in this chain equal Λp for Besov spaces (Theorem 3.7). r+s(A) by the reiteration theorem (cid:3) SPECTRAL INVARIANCE OF BESOV-BESSEL SUBALGEBRAS 17 Algebra Properties. The characterization of Bessel potential spaces by a hypersin- gular integral yields the Banach algebra properties of Pr(A). Theorem 3.29. If A is a Banach algebra with Cw-group Ψ, then the Bessel po- tential space Pr(A) is a Banach subalgebra of A for every r > 0. Moreover, Pr(A) is inverse-closed in A. For functions in Pr(L∞(Rd) this result is in Strichartz [39]. Proof. We treat the case r < 1 first. Let a, b ∈ Pr(A). Using ∆t(ab) = ∆t(a)∆t(b) + a∆t(b) + ∆t(a)b we obtain (28) (cid:13)(cid:13)(cid:13)Zǫ≤t2≤1 tr 2 ∆t(ab) dt td 2(cid:13)(cid:13)(cid:13)A aZǫ≤t2≤1 +(cid:13)(cid:13)(cid:13) ≤(cid:13)(cid:13)(cid:13)Zǫ≤t2≤1 2(cid:13)(cid:13)(cid:13)A ∆t(b) tr 2 dt td ∆t(a)∆t(b) tr 2 dt td +(cid:13)(cid:13)(cid:13)(cid:16)Zǫ≤t2≤1 2(cid:13)(cid:13)(cid:13)A tr 2 ∆t(a) dt td . 2(cid:17)b(cid:13)(cid:13)(cid:13)A The second and third term of the expression on the right hand side of the inequality are dominated by kakAkbkPr(A) + kakPr(A)kbkA . kakPr(A)kbkPr(A) . For the estimation of the first term in (28) we use the embedding Pr(A) ֒→ r (A) (Proposition 3.27), so k∆takA . tr 2kakPr(A), with a similar estimate for Λ∞ b. Therefore ∆t(a)∆t(b) tr 2 dt td . kakPr(A)kbkPr(A)Z0≤t2≤1 tr 2 dt td 2 ≤ CrkakPr(A)kbkPr(A) , and Cr does not depend on ǫ. Combining the estimates we have proved that (cid:13)(cid:13)(cid:13)Zǫ≤t2≤1 2(cid:13)(cid:13)(cid:13)A kabkPr(A) . kakPr(A)kbkPr(A). For the verification of the inverse-closedness of Pr(A) in A we use a similar argument: Expand the identity ((10)) to obtain ∆t(a−1) = −∆t(a−1)∆t(a)a−1 − a−1∆t(a)a−1. So (29) (cid:13)(cid:13)(cid:13)Zǫ≤t2≤1 ∆t(a−1) tr 2 dt td 2(cid:13)(cid:13)(cid:13)A ≤(cid:13)(cid:13)(cid:13)Zǫ≤t2≤1 ∆t(a−1)∆t(a)a−1 tr 2 dt td 2(cid:13)(cid:13)(cid:13)A a−1∆t(a)a−1 +(cid:13)(cid:13)(cid:13)Zǫ≤t2≤1 2kakΛ∞ tr 2 r (A) , and, as Λ∞ r (A) is inverse- . dt td 2(cid:13)(cid:13)(cid:13)A As a ∈ Λ∞ closed in A, k∆t(a−1)kA . tr r (A), we know that k∆t(a)kA . tr 2ka−1kΛ∞ r (A). The first term on the right hand side of (29) can be dominated by Zǫ≤t2≤1 k∆t(a−1)kAk∆takAka−1kA tr 2 dt td 2 The second term can be estimated as . ka−1kΛ∞ r (A)kakΛ∞ r (A)kakA (cid:13)(cid:13)(cid:13)Zǫ≤t2≤1 a−1∆t(a)a−1 tr 2 dt td 2(cid:13)(cid:13)(cid:13)A a−1(cid:16)Zǫ≤t2≤1 =(cid:13)(cid:13)(cid:13) .ka−1k2 AkakPr(A). ∆t(a) tr 2 dt td 2(cid:17)a−1(cid:13)(cid:13)(cid:13)A As Λ∞ r (A) is inverse-closed in A we obtain the inverse-closedness of Pr(A) in A. 18 ANDREAS KLOTZ If r ≥ 1 we can proceed by induction. Assume that we have already proved that Ps(A) is inverse-closed in A, and s < r < s + 1. By what we have just proved Pr(A) = Pr−s(Ps(A)) is inverse-closed in Ps(A). As Ps(A) is inverse-closed in A by hypotheses we are done. (cid:3) 2)r/2 for 3.4.2. Application to Weighted Matrix Algebras. We call v∗ r > 0 the Bessel weight of order r. If A is a Banach space of matrices, we say that a matrix A is in the weighted matrix space Avr , where vr is the standard polynomial weight vr(k) = (1 + k)r, if the matrix with entries A(k, l)vr(k − l) is in A. The norm in Avr is kAkAvr = k(A(k, l)vr(k − l))k,l∈Zd kA. In a similar way we introduce r . Av∗ r (k) = (1 + 2πk2 Proposition 3.30. If A is a homogeneous matrix algebra, and χ is a Cw- group on A, then Av∗ r = Pr(A) . Proof. By definition A is in Pr(A), if there is a A0 ∈ A such that A = Gr ∗ A0. This is equivalent to A(k) = (1 + 2πk2)−r/2 A0(k) , or A0(k) = (1 + 2πk2)r/2 A(k), and therefore kAkPr(A) = kA0kA = kAkAv∗ r , i.e., A ∈ Av∗ r . (cid:3) Proposition 3.31. If A is a homogeneous matrix algebra, χ is a Cw- group on A, and v∗ r = Pr(A) is a matrix algebra. This algebra is inverse-closed in A. r , r > 0, is a Bessel weight, then Av∗ Proof. This is an application of Theorem 3.29. (cid:3) Proposition 3.31 applies in particular to the weighted subalgebras of B(ℓ2). For solid matrix algebras the standard polynomial weights vr can be taken in- stead of v∗ r . Corollary 3.32. If A is a solid matrix algebra, and χ is a Cw- group on A, then Avr is an inverse-closed subalgebra of A. We state the results of Proposition 3.27 and Proposition 3.28 for weighted matrix algebras. Proposition 3.33. If A is a homogeneous matrix algebra, and r, s > 0, then Λ1 Λp r(A) ֒→ Av∗ s ) = E p r(Av∗ r ֒→ Λr r (Av∗ ∞(A), s ) = (Λp Example 3.34. For the Schur algebras Sp r+s(A) = E p r+s(A). r(A))v∗ s = Λp r we obtain 0 ) . s+r(Sp s (Sp E q r ) = E q Appendix A. Proof of the Reiteration theorem We give a proof of Theorem 3.7. We need some properties of the moduli of smoothness. Lemma A.1. If l, k ∈ N, l ≥ k, t ∈ Rd and h > 0, then t(x)kX ≤ (MΨ + 1)kk∆l−k t (1) k∆l (2) (x)kX and ωl h(x) ≤ (MΨ + 1)kωl−k h (x) , k ωk t (x) ≍ sup hj ≤t 1≤j≤k Yj=1 k(cid:0) ∆hj(cid:1)xkX . SPECTRAL INVARIANCE OF BESOV-BESSEL SUBALGEBRAS 19 The proof of (1) is an easy calculation in complete analogy to the corresponding Item 2 is properties of the moduli of smoothness for functions. See, e.g., [18]. proved in [8, 5.4.11]. Proof of the Reiteration theorem. We assume first that x is in Λq mate kxkΛq relations of Besov spaces imply that kxkΛp mate xΛq r (A) + xΛq r (A) ≤ CkxkΛq r (A)). As kxkΛq r (A)) = kxkΛp s+r(A) and esti- r (A)) and the inclusion s (Λp r+s(A), it suffices to esti- s(Λp s(Λp r (A)). s (Λp Assume that ⌊r⌋ < m and ⌊s⌋ < n, m, n ∈ N. Using the norm equivalences in (7) we can write (30) xΛq s(Λp r (A)) ≍(ZR+hh−sωn+m h = kh−sωn+m (x, Λp h (x, Λp r(A))iq dh h )1/q ∗ , r(A))kLq where kf kLq . An estimate of the modulus of smoothness is ∗ =(cid:0)R ∞ ωn+m 0 f (t)q dt (x, Λp t (cid:1)1/q h r(A)) = sup u≤h k∆n+m u (31) ≤ sup u≤h k∆n+m u . sup u≤h k∆n+m u xkΛp r (A) xkA + sup u≤h ∆n+m u xkA + sup u≤h ∆n+m u xΛp r (A) xΛ1 r (A) , where the last inequality uses the embedding Λ1 r(A) ֒→ Λp r(A) for p ≥ 1. Inserting this estimate into (30) we obtain (32) xΛq s (Λp r (A)) . xΛq s (A) + kh−s sup u≤h ∆n+m u xΛ1 ∗ . r (A)kLq With φ(v, u) = k∆n+m v ∆n+m u xkA the Lq ∗-norm in (32) can be rewritten as t−r sup v≤t φ(v, u) dt t kLq ∗ kh−s sup u≤hZR+ u≤hZ h 0 ≤ kh−s sup t−r sup v≤t φ(v, u) dt t kLq ∗ + kh−s sup u≤hZ ∞ h t−r sup v≤t φ(v, u) dt t kLq ∗ =: I + II. We can estimate the first term further using Hardy's inequality. h dt t (cid:21)q dh t (cid:21)q dh h dt t−rφ(v, u) sup v≤t 0 0 0 sup h−sq sup u≤h(cid:20)Z h Iq =Z ∞ ≤Z ∞ h−sq(cid:20)Z h . Z ∞ 0 (cid:18)t−(r+s) . Z ∞ 0 (cid:18)t−(r+s)ω2(n+m) v,u≤h v,u≤t sup (∗∗) (∗) 0 t t−rφ(v, u) t φ(v, u)(cid:19)q dt (x, A)(cid:19)q dt t = xq r+s(A) , Λq (∗) by Hardy's inequality, and (∗∗) using Lemma A.1(2). For the second term we use (1) of Lemma A.1 to get φ(v, u) = k∆n+m v ∆n+m u xkA . k∆n+m u xkA. 20 ANDREAS KLOTZ Then supv≤t φ(u, v) is independent of t, and (33) u≤hZ ∞ IIq .Z ∞ 0 (cid:18)h−s sup =Z ∞ h−(r+s)q sup u≤h h 0 t−rk∆n+m u xkA dt t (cid:19)q dh h k∆n+m u xkq A dh h = xq r+s(A). Λq I and II together give the desired estimate. s(Λp For the converse assume that x ∈ Λq r(A)). Then, (34) xq Λq r+s(A) ≍ZRd(cid:18)t−(r+s)k∆n+m t−rq(cid:18)Zη≥t ≍ZRd t xkA(cid:19)q dt td η−spk∆n+m t xkp A dη ηd(cid:19)q/p dt td 2 , where we have used t−s ≍ (cid:0)Rη≥tη−sp dη for the last equivalence. As t kA for η ≥ t , we can dominate the right hand xkA ≤ supv≤ηk∆m v ∆n 2(cid:1)1/p ηd k∆n+m side of (34) by t t−rq sup u≤t(cid:18)Zη≥t u≤t(cid:18)ZRd t−rq sup η−sp sup v≤η k∆n v ∆m u xkp A η−sp sup v≤η k∆n v ∆m u xkp A td dη ηd(cid:19)q/p dt ηd(cid:19)q/p dt td dη ZRd ≤ZRd ≤ZRd t−rq sup u≤t (∆m u xΛp s (A))q dt td ≤ xq Λq s (A)). r (Λp (cid:3) Appendix B. Jackson Bernstein Theorem Proposition B.1 ([20, 5.12]). Let a ∈ A and σ > 0. (1) There is a σ-bandlimited element aσ ∈ C(A) such that ka − aσkA ≤ Cω1/σ(a) with C independent of σ and a. (2) If δα(a) ∈ C(A), for all multi-indices α with α = k then there exists a σ-bandlimited element aσ ∈ A such that ka − aσkA ≤ Cσ−k Xα=k 1/σ(δαa) . ω2 Corollary B.2. If a ∈ Λp r(A) for r > 0, then a ∈ E p r (A). Proof. We use the integral version of the norm for an approximation space in (4) and assume that 1 ≤ p < ∞. The proof for p = ∞ is simpler and done in [20]. Assume first that 0 < r < 1. Then, by Proposition B.1(1), Z ∞ 1 (cid:0)Eσ(a)σr(cid:1)p dσ σ ≤ CZ 1 0 (cid:0)ωτ (a)τ −r(cid:1)p dτ τ ≤ Cap r (A) , Λp and so the approximation norm is dominated by the Besov norm. Likewise, if r = k + η, 0 < η ≤ 1, and k ∈ N, Proposition B.1(2) yields Z ∞ 1 (cid:0)Eσ(a)σr(cid:1)p dσ σ ≤ C Xα=kZ 1 0 (cid:0)ω2 τ (δα(a))τ −η(cid:1)p dτ τ SPECTRAL INVARIANCE OF BESOV-BESSEL SUBALGEBRAS and again kakE p r (A) is dominated by the Besov norm. 21 (cid:3) Before proving the converse implication in Theorem 3.11, i.e., the Bernstein-type result, we need a mean-value property of automorphism groups. Lemma B.3 ([20, 5.15]). If a is σ-bandlimited, then (35) k∆takA ≤ Cσ t kakA . Proposition B.4. Let a ∈ A, and r > 0, 1 ≤ p ≤ ∞. a ∈ Λp r(A). If a ∈ E p r (A), then If r (A), the representation theorem of approximation theory (see, e.g [33, 3.1]) Proof. We adapt a standard proof [14] and verify the statement for p < ∞. a ∈ E p implies that (36) a = ∞ Xk=0 ak, with ak ∈ X2k and 2krpkakkp A < ∞ , ∞ Xk=0 where (Xσ)σ≥0 is the approximation scheme of bandlimited elements, and kakE p ∞ Xk=0 r (A) ≍(cid:0) 2krpkakkp , A(cid:1)1/p where the infimum is taken over all admissible representations as in (36). An k=0 ak is convergent in A. Note application of Holders inequality shows that P∞ that (36) implies that kakkA ≤ C2−krfor all k ∈ N0. We assume first that 0 < r < 1. We need an estimate for the norm of ∆ta. (37) k∆takA ≤ ≤ M M Xk=0 Xk=0 k∆takkA + ∞ Xk=M+1 k∆takkA k∆takkA + (MΨ + 1) ∞ kakkA , Xk=M+1 where the value of M will be chosen later. Lemma B.3 implies that k∆takkA ≤ C2kt kakkA for all k ∈ N. Substituting back into (37) yields (38) k∆takA ≤ C(cid:16) M Xk=0 2ktkakkA + ∞ Xk=M+1 kakkA(cid:17) . We use this relation for the estimation of the Besov seminorm. aΛp r (A) ≍(cid:18) ∞ Xl=0(cid:0)2lrω2−l(a)(cid:1)p(cid:19)1/p .(cid:18) ∞ 2lrp(cid:16) Xl=0 Xk=0 M 2k2−lkakkA + ∞ Xk=M+1 kakkA(cid:17)p(cid:19)1/p . We split this expression into two parts and assume that M = l in the inner sums. aΛp r (A) .(cid:18) ∞ Xl=0 2l(r−1)p(cid:16) l Xk=0 2kkakkA(cid:17)p(cid:19)1/p +(cid:18) ∞ Xl=0 ∞ 2lrp(cid:16) Xk=l+1 kakkA(cid:17)p(cid:19)1/p . 22 ANDREAS KLOTZ We apply Hardy's inequalities to both terms on the right hand side and obtain aΛp 2l(r−1)p2lpkalkp A(cid:17)1/p +(cid:16) ∞ Xl=0 2lrpkalkp A(cid:17)1/p ∞ r (A) .(cid:16) Xl=0 = 2(cid:16) Xl=0 ∞ 2lrpkalkp . A(cid:17)1/p As the representations a = P∞ r (A) . r (A), using again the representation theorem. Next we consider the case r = kakE p m + η for m ∈ N0 and 0 < η < 1. The Bernstein inequality implies that k=0 ak were arbitrary we conclude that aΛp kδα(ak)kA ≤ C(2π2k)αkakkA k=0 δαak converges in A for all α with for all k ∈ N and α ∈ Nd α ≤ m and its sum must be δα(a) , as each δj is closed on D(δα). We now apply the above estimates δα(a) instead of a and deduce that δα(a) must be in Λp η(A) for α ≤ k. Thus a ∈ Λp 0. Consequently P∞ r(A). If r is an integer, then we have to use second order differences and a corresponding version of the mean value theorem. The argument is almost the same as above (see [45] for details in the scalar case). (cid:3) Combining Propositions B.2 and B.4, we have completed the proof of Theo- rem 3.11. Appendix C. Littlewood-Paley decomposition Proof of Proposition 3.12. We include the derivation of the relevant results to keep the presentation self-contained. We follow [9], but we use approximation arguments where feasible. We use some obvious facts of the dyadic partition of unity (ϕk)k≥−1. By definition, supp ϕk = 2k supp ϕ ⊆ {ω : 2k−1 ≤ ω∞ ≤ 2k+1} for k ≥ 0, and supp ϕ−1 ⊆ {ω : ω∞ ≤ 1}. As the intersection of supp( ϕk) with supp( ϕl) is nonempty only for l ∈ {k − 1, k, k + 1} we obtain that ϕk = ϕk ∗ (ϕk−1 + ϕk + ϕk+1) if k ≥ 0, and ϕ−1 = ϕ−1 ∗ (ϕ−1 + ϕ0). Assume first that (16) holds. Then kϕk ∗ akA ≤ C2−rk, and so P∞ k=−1 ϕk ∗ a is norm convergent in A. A standard weak type argument shows that the limit is actually a. For a ∈ Λp r(A) and m > ⌊r⌋ we use kakΛp t (ϕk ∗ a)kA ≤ Cmkϕk ∗ akA by Lemma A.1 (1), and k∆m As k∆m C′tm2mkkϕk ∗ akA by repeated application of Lemma B.3 we conclude that r (A) ≍ kakA +(cid:0)P∞ k=0(2rkωm . t (ϕk ∗ a)kA ≤ 2−k (a))p(cid:1)1/p (39) k∆m t (ϕk ∗ a)kA ≤ C1 min(1, tm2mk)kϕk ∗ akA . As an immediate consequence we obtain (40) and so (41) min(1, tm2mk)kϕk ∗ akA , ωm t(a) ≤ C1 ∞ Xk=−1 ∞ 2rjωm 2−j (a) ≤ C1 2(j−k)r2kr min(1, 2−(j−k)m)kϕk ∗ akA . Xk=1 The right hand side of this relation can be written as a convolution. If we set u(l) = min(1, 2−lm)2lr for l ∈ Z, and v(l) = 2lrkϕl ∗ akA if l > −1 and 0 else, then u and v are sequences in ℓ1(Z), and the right hand side of (41) is just (u ∗ v)(j). SPECTRAL INVARIANCE OF BESOV-BESSEL SUBALGEBRAS 23 So k(cid:0)2rjωm 2−j (a)(cid:1)j∈Nkℓp(N) ≤ Ckukℓ1(Z)kvkℓp(Z), and this means that ∞ (42) kakΛp r (A) ≤ C(cid:16) so (16) implies that a ∈ Λp r(A). For the other inequality we use kakΛp m < r ≤ m + 1. First we show that , A(cid:17)1/p 2rkpkϕk ∗ akp Xk=−1 r (A) ≍ kakA +Pα=mkδα(a)kΛp r−m(A) with (43) and kϕk ∗ akA ≤ C2−mkkϕk ∗ δα(a)kA , m = α 2−k (δαa). kϕk ∗ δα(a)kA ≤ Cω2 (44) For the proof of these relations choose an even function Φ ∈ S(Rd) such that Φ ≡ 1 on supp ϕ0, and Φ ≡ 0 in a neighbourhood of 0. Set Φk(t) = 2kdΦ(2kt), then kΦkk1 = kΦk1 and Φk ∗ ϕk = ϕk. The function η(α) : ω → (2πiω)−α Φ(ω) is an element of S . Again, if we set η(α) k k1 = kη(α)k1. Then Φk(ω) = Φ(2−kω) = 2−kα(2πiω)α η(α) k (w) , and so, assuming that α = m, we obtain ϕk(ω) = 2−km η(α) k (ω)(2πiω)α ϕk(ω) for all ω ∈ Rd, which implies k (t) = 2kdη(α)(2kt), then kη(α) ∗ δα(ϕk ∗ a) = 2−k mη(α) the last equality by (14). Now (43) follows immediately. ϕk ∗ a = 2−k mη(α) k k ∗ ϕk ∗ δα(a), For the proof of (44) set y = δα(a) and yk = ϕk ∗ y = Φk ∗ ϕk ∗ y = Φk ∗ yk. We obtain ϕk ∗ y = Φk ∗ yk =ZRd Φk(t)ψ−t(yk) dt Φk(t)(cid:8)ψ−t(yk) − 2yk + ψt(yk)(cid:9) dt = 1 2ZRd as RRd Φk = 0 and Φk(−t) = Φk(t). Changing variables we obtain Φ(u)ψ−2−ku∆2 2−ku(yk) dt = 1 ϕk ∗ y = 1 = 1 2ZRd 2ZRd 2ZRd Φ(u)ψ−2−ku(cid:0)ϕk ∗ ∆2 2−ku(y)(cid:1) dt. Φk(t)ψ−t∆2 t (yk) dt , Taking norms we get kϕk ∗ ykA ≤ Mψ Φ(u)kϕkk1ω2 2−ku(y) dt. Φ(u)(1 + u2)ω2 2−k (y) dt 2 ZRd 2 kϕ0k1ZRd ≤ Mψ ≤ Cω2 2−k (y) , where the estimate for ω2 to show. 2−ku(y) follows from Lemma A.1. This is what we wanted The proof of the reverse inclusion now follows by putting (43) and (44) together. 2rkkϕk ∗ akA ≤ C2(r−m)kkϕk ∗ δα(a)kA ≤ C2(r−m)kω2 2−k (δα(a)) , and so (45) ∞ Xk=−1 ∞ 2rpkkϕk ∗ akp A ≤ C(cid:0)kakp ≤ C′(kakp A + Xk=0 2(r−m)pkω2 2−k (δα(a))p(cid:1) r−m(A)) ≤ C"kakp A + δα(a)p r−m(A) . Λp Λp We have shown that a ∈ Λp (42) and (45). r(A) implies (16). The norm equivalence follows from (cid:3) 24 ANDREAS KLOTZ References [1] J. M. Almira and U. Luther. Inverse closedness of approximation algebras. J. Math. Anal. Appl., 314(1):30 -- 44, 2006. [2] W. O. Amrein, A. Boutet de Monvel, and V. Georgescu. C0-groups, commutator methods and spectral theory of N -body Hamiltonians, volume 135 of Progress in Mathematics. Birkhauser Verlag, Basel, 1996. [3] W. Arveson. On groups of automorphisms of operator algebras. J. Functional Analysis, 15:217 -- 243, 1974. [4] W. Arveson. The harmonic analysis of automorphism groups. In Operator algebras and ap- plications, Part I (Kingston, Ont., 1980), volume 38 of Proc. Sympos. Pure Math., pages 199 -- 269. Amer. Math. Soc., Providence, R.I., 1982. [5] A. G. Baskakov. Wiener's theorem and asymptotic estimates for elements of inverse matrices. Funktsional. Anal. i Prilozhen., 24(3):64 -- 65, 1990. [6] A. G. Baskakov. Asymptotic estimates for elements of matrices of inverse operators, and harmonic analysis. Sibirsk. Mat. Zh., 38(1):14 -- 28, i, 1997. [7] A. G. Baskakov. Estimates for the elements of inverse matrices, and the spectral analysis of linear operators. Izv. Ross. Akad. Nauk Ser. Mat., 61(6):3 -- 26, 1997. [8] C. Bennett and R. Sharpley. Interpolation of operators, volume 129 of Pure and Applied Mathematics. Academic Press Inc., Boston, MA, 1988. [9] J. Bergh and J. Lofstrom. Interpolation Spaces. An Introduction. Springer-Verlag, Berlin, 1976. Grundlehren der Mathematischen Wissenschaften, No. 223. [10] B. E. Blackadar and J. Cuntz. The structure of stable algebraically simple C ∗-algebras. Amer. J. Math., 104(4):813 -- 822, 1982. [11] O. Bratteli and D. W. Robinson. Unbounded derivations of C ∗-algebras. Comm. Math. Phys., 42:253 -- 268, 1975. [12] O. Bratteli and D. W. Robinson. Operator algebras and quantum statistical mechanics. 1. Texts and Monographs in Physics. Springer-Verlag, New York, second edition, 1987. C ∗- and W ∗-algebras, symmetry groups, decomposition of states. [13] P. L. Butzer and H. Berens. Semi-groups of operators and approximation. Die Grundlehren der mathematischen Wissenschaften, Band 145. Springer-Verlag New York Inc., New York, 1967. [14] P. L. Butzer and R. J. Nessel. Fourier analysis and approximation. Academic Press, New York, 1971. Volume 1: One-dimensional theory, Pure and Applied Mathematics, Vol. 40. [15] M. Cwikel, P. G. Nilsson, and G. Schechtman. Interpolation of weighted Banach lattices. A characterization of relatively decomposable Banach lattices. Mem. Amer. Math. Soc., 165(787):vi+127, 2003. [16] K. DeLeeuw. An harmonic analysis for operators. I: Formal properties. Ill. J. Math., 19:593 -- 606, 1975. [17] K. DeLeeuw. An harmonic analysis for operators. II: Operators on Hilbert space and analytic operators. Ill. J. Math., 21:164 -- 175, 1977. [18] R. A. DeVore and G. G. Lorentz. Constructive approximation, volume 303 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin, 1993. [19] K. Grochenig. Wiener's Lemma: Theme and Variations. An Introduction to Spectral Invari- ance. Applied and Numerical Harmonic Analysis. Birkhauser, Boston, 2009. Inzell Lectures on Harmonic Analysis. [20] K. Grochenig and A. Klotz. Noncommutative approximation: Inverse-closed subalgebras and off-diagonal decay of matrices. Constructive Approximation, 32:429 -- 446, 2010. [21] K. Grochenig and M. Leinert. Symmetry and inverse-closedness of matrix algebras and func- tional calculus for infinite matrices. Trans. Amer. Math. Soc., 358(6):2695 -- 2711 (electronic), 2006. [22] K. Grochenig and Z. Rzeszotnik. Banach algebras of pseudodifferential operators and their almost diagonalization. Ann. Inst. Fourier (Grenoble), 58(7):2279 -- 2314, 2008. [23] Y. Grushka and S. Torba. Direct theorems in the theory of approximation of vectors in a Ba- nach space with exponential type entire vectors. Methods Funct. Anal. Topology, 13(3):267 -- 278, 2007. [24] E. Hille and R. S. Phillips. Functional analysis and semi-groups. American Mathematical Society Colloquium Publications, vol. 31. American Mathematical Society, Providence, R. I., 1957. rev. ed. [25] S. Jaffard. Propri´et´es des matrices "bien localis´ees" pr`es de leur diagonale et quelques appli- cations. Ann. Inst. H. Poincar´e Anal. Non Lin´eaire, 7(5):461 -- 476, 1990. SPECTRAL INVARIANCE OF BESOV-BESSEL SUBALGEBRAS 25 [26] E. Kissin and V. S. Shul′man. Dense Q-subalgebras of Banach and C ∗-algebras and un- bounded derivations of Banach and C ∗-algebras. Proc. Edinburgh Math. Soc. (2), 36(2):261 -- 276, 1993. [27] E. Kissin and V. S. Shul′man. Differential properties of some dense subalgebras of C ∗- algebras. Proc. Edinburgh Math. Soc. (2), 37(3):399 -- 422, 1994. [28] A. Klotz. Noncommutative Approximation: Smoothness, Approximation and Invertibility in Banach Algebras. PhD thesis, University of Vienna, 2010. [29] G. Kothe. Topological vector spaces. I. Translated from the German by D. J. H. Garling. Die Grundlehren der mathematischen Wissenschaften, Band 159. Springer-Verlag New York Inc., New York, 1969. [30] T. W. Palmer. Banach algebras and the general theory of ∗-algebras. Vol. I, volume 49 of Encyclopedia of Mathematics and its Applications. Cambridge University Press, Cambridge, 1994. Algebras and Banach algebras. [31] J. Peetre. New thoughts on Besov spaces. Mathematics Department, Duke University, Durham, N.C., 1976. Duke University Mathematics Series, No. 1. [32] A. Pietsch. Eigenvalues of integral operators. I. Math. Ann., 247(2):169 -- 178, 1980. [33] A. Pietsch. Approximation spaces. J. Approx. Theory, 32(2):115 -- 134, 1981. [34] V. S. Rabinovich, S. Roch, and B. Silbermann. Fredholm theory and finite section method for band-dominated operators. Integral Equations Operator Theory, 30(4):452 -- 495, 1998. Dedicated to the memory of Mark Grigorievich Krein (1907 -- 1989). [35] V. S. Rabinovich, S. Roch, and B. Silbermann. Limit operators and their applications in operator theory, volume 150 of Operator Theory: Advances and Applications. Birkhauser Verlag, Basel, 2004. [36] M. A. Rieffel. Leibniz seminorms for "matrix algebras converge to the sphere". arXiv:0707.3229v3 [math.OA], 2010. [37] E. M. Stein. The characterization of functions arising as potentials. Bull. Amer. Math. Soc., 67:102 -- 104, 1961. [38] E. M. Stein. Singular integrals and differentiability properties of functions. Princeton Math- ematical Series, No. 30. Princeton University Press, Princeton, N.J., 1970. [39] R. S. Strichartz. Multipliers on fractional Sobolev spaces. J. Math. Mech., 16:1031 -- 1060, 1967. [40] Q. Sun. Wiener's lemma for infinite matrices with polynomial off-diagonal decay. C. R. Math. Acad. Sci. Paris, 340(8):567 -- 570, 2005. [41] Q. Sun. Wiener's lemma for infinite matrices. Trans. Amer. Math. Soc., 359(7):3099 -- 3123 (electronic), 2007. [42] R. Tessera. The inclusion of the schur algebra in B(ℓ2) is not inverse-closed. Monatsh. Math., DOI 10.1007/s00605-010-0216-x, 2010. [43] S. Torba. Inverse theorems in the theory of approximation of vectors in a Banach space with exponential type entire vectors. arXiv:0809.4030v1 [math.FA], 2008. [44] H. Triebel. Theory of function spaces. II, volume 84 of Monographs in Mathematics. Birkhauser Verlag, Basel, 1992. [45] R. M. Trigub and E. S. Bellinsky. Fourier analysis and approximation of functions. Kluwer Academic Publishers, Dordrecht, 2004. [46] R. L. Wheeden. On hypersingular integrals and Lebesgue spaces of differentiable functions. Trans. Amer. Math. Soc., 134:421 -- 435, 1968. Faculty of Mathematics, University of Vienna, Nordbergstrasse 15, A-1090 Vienna, AUSTRIA E-mail address: [email protected]
1812.06483	1	1812	2018-12-16T15:22:15	Operator system structures and extensions of Schur multipliers	[ "math.OA" ]	For a given C-algebra $\mathcal{A}$, we establish the existence of maximal and minimal operator $\mathcal{A}$-system structures on an AOU $\mathcal{A}$-space. In the case $\mathcal{A}$ is a W-algebra, we provide an abstract characterisation of dual operator $\mathcal{A}$-systems, and study the maximal and minimal dual operator $\mathcal{A}$-system structures on a dual AOU $\mathcal{A}$-space. We introduce operator-valued Schur multipliers, and provide a Grothendieck-type characterisation. We study the positive extension problem for a partially defined operator-valued Schur multiplier $\varphi$ and, under some richness conditions, characterise its affirmative solution in terms of the equality between the canonical and the maximal dual operator $\mathcal{A}$-system structures on an operator system naturally associated with the domain of $\varphi$.	math.OA	math	OPERATOR SYSTEM STRUCTURES AND EXTENSIONS OF SCHUR MULTIPLIERS YING-FEN LIN AND IVAN G. TODOROV Abstract. For a given C-algebra A, we establish the existence of max- imal and minimal operator A-system structures on an AOU A-space. In the case A is a W-algebra, we provide an abstract characterisa- tion of dual operator A-systems, and study the maximal and minimal dual operator A-system structures on a dual AOU A-space. We in- troduce operator-valued Schur multipliers, and provide a Grothendieck- type characterisation. We study the positive extension problem for a partially defined operator-valued Schur multiplier ϕ and, under some richness conditions, characterise its affirmative solution in terms of the equality between the canonical and the maximal dual operator A-system structures on an operator system naturally associated with the domain of ϕ. 1. Introduction The problem of completing a partially defined matrix to a fully defined positive matrix has attracted considerable attention in the literature (see e.g. [5] and [8] and the references therein). Given an n by n matrix, only a subset of whose entries are specified, this problem asks whether the remain- ing entries can be determined so as to yield a positive matrix. For block operator matrices, this problem was considered in [14], where the authors showed that it is closely related to questions about automatic complete posi- tivity of certain positive linear maps. More specifically, one associates to the pattern κ of the partially defined matrix (that is, the set of all given entries) the operator system S(κ) of all fully specified matrices supported by κ. The positive completion problem is then linked to the question of whether the operator-valued Schur multiplier with domain S(κ) is completely positive. A continuous infinite dimensional version of the scalar-valued completion problem was considered in [11], where the authors characterised the operator systems possessing the positive completion property in terms of an approx- imation of its positive cone via rank one operators. The original motivation behind the present paper was the study of the operator-valued, infinite di- mensional and continuous, analogue of the positive completion problem. We relate the question to the automatic complete positivity of operator-valued Date: 2 December 2018. 1 2 Y.-F. LIN AND I. G. TODOROV Schur multipliers; in fact, we characterise the extendability of Schur mul- tipliers in terms of an equality between operator system structures on an associated Archimedean order unit (AOU) -vector space. One of the fundamental representation theorems in Operator Space The- ory is Choi-Effros Theorem [13, Theorem 13.1], which characterises operator systems (that is, unital selfadjoint linear subspaces S of the space B(H) of all bounded linear operators on a Hilbert space H) abstractly, in terms of properties of the cones of positive elements in the S-valued matrix space Mn(S). Operator A-systems, that is, the operator systems which admit a bimodule action by a unital C-algebra A, can be characterised similarly in a way that takes into account the extra A-module structure [13, Corollary 15.13]. Dual operator systems -- that is, operator systems that are also dual operator spaces -- were characterised by D. P. Blecher and B. Magajna in [4]. However, no analogous representation of dual operator A-systems, where A is a W-algebra, has been known. The idea of viewing operator spaces as a quantised version of Banach spaces has been very fruitful in Functional Analysis [6]. Operator systems can in a similar vein be thought of as a quantised version of Archimedean order unit (AOU) -vector spaces. The possible quantisations, or operator system structures, on a given AOU space, were first studied in [15], where it was shown that every AOU space possesses two extremal operator system structures. However, no similar development has been achieved for dual AOU spaces or for AOU A-spaces. In this paper, we unify all aforementioned strands of questions. We pro- vide a Choi-Effros type representation theorem for dual operator A-systems. We study the operator A-system structures on a given AOU A-space, as well as the dual operator A-system structures on a given dual AOU A-space. The latter results are new even in the case where A coincides with the complex field. We introduce infinite dimensional measurable operator-valued Schur multipliers, and provide a characterisation that generalises their well-known description by A. Grothendieck [9] in the scalar case (see also [10] and [17]). Finally, we study the positive extension problem for operator-valued Schur multipliers, and characterise the possibility of such an extension by equality of the canonical and the maximal dual operator D-system structures on the domain of the given Schur multiplier. Our context is that of an arbitrary (albeit standard) measure space (X, µ), which includes as a sub-case the discrete case and thus the finite case considered in [14]. In this context, the algebra D is the maximal abelian selfadjoint algebra corresponding to L∞(X, µ). Our results are a far reaching generalisation of the results of V. I. Paulsen, S. Power and R. R. Smith [14]; in particular, they provide a dif- ferent view on the positive completion problem for block operator matrices considered therein. The paper is organised as follows. After collecting some preliminaries in Section 2, we establish, in Section 3, the existence of the minimal and the maximal operator A-system structures on a AOU A-space V , OMINA(V ) OPERATOR SYSTEM STRUCTURES 3 and OMAXA(V ). In case V is a C-algebra, OMINA(V ) was essentially defined in [20], in relation with the problem of automatic complete positivity of A-module maps, whose completely bounded version was first considered by R. R. Smith in [23] (see also the subsequent paper [19]). We show that OMAXA(V ) (resp. OMINA(V )) is characterised by the automatic complete positivity of A-bimodule positive maps from V into any operator A-system (resp. from any operator A-system into V ). In Section 4, we provide a characterisation theorem for dual operator A- systems and, in Section 5, we define dual AOU A-spaces and undertake a development, analogous to the one in Section 3, for dual operator A-system structures. In Section 6, we introduce the operator-valued version of measurable Schur multipliers and provide a Grothendieck-type characterisation, noting the special case of positive Schur multipliers. In Section 7, we study par- tially defined operator-valued Schur multipliers and their extension proper- ties to a fully defined positive Schur multiplier. Associated with the domain κ ⊆ X × X of the Schur multiplier is an operator system S(κ). Our analysis depends on the presence of sufficiently many operators of finite rank in S(κ). We note that, of course, this holds true trivially in the classical matrix case. Under such richness conditions on the domain κ, we show that the positive extension problem for operator-valued Schur multipliers defined on κ has an affirmative solution precisely when the canonical operator system structure of S(κ) coincides with its maximal dual operator D-system structure. We denote by (·, ·) the inner product in a Hilbert space, and we use h·, ·i to designate duality paring. We will assume some basic facts and notions from Operator Space Theory, for which we refer the reader to the monographs [3, 6, 13, 18]. 2. Preliminaries In this section we recall basic results and introduce some new notions that will be needed subsequently. If W is a real vector space, a cone in W is a non-empty subset C ⊆ W with the following properties: (a) λv ∈ C whenever λ ∈ R+ := [0, ∞) and v ∈ C; (b) v + w ∈ C whenever v, w ∈ C. A -vector space is a complex vector space V together with a map ∗ : V → V which is involutive (i.e. (v∗)∗ = v for all v ∈ V ) and conjugate linear (i.e. (λv + µw)∗ = λv∗ + µw∗ for all λ, µ ∈ C and all v, w ∈ V ). If V is a -vector space, then we let Vh = {x ∈ V : x∗ = x} and call the elements of Vh hermitian. Note that Vh is a real vector space. An ordered -vector space [16] is a pair (V, V +) consisting of a -vector space V and a subset V + ⊆ Vh satisfying the following properties: (a) V + is a cone in Vh; (b) V + ∩ −V + = {0}. 4 Y.-F. LIN AND I. G. TODOROV Let (V, V +) be an ordered -vector space. We write v ≥ w or w ≤ v if v, w ∈ Vh and v − w ∈ V +. Note that v ∈ V + if and only if v ≥ 0; for this reason V + is referred to as the cone of positive elements of V . An element e ∈ Vh is called an order unit if for every v ∈ Vh there exists r > 0 such that v ≤ re. The order unit e is called Archimedean if, whenever v ∈ V and re + v ∈ V + for all r > 0, we have that v ∈ V +. In this case, we call the triple (V, V +, e) an Archimedean order unit -vector space (AOU space for short). Note that (C, R+, 1) is an AOU space in a canonical fashion. Let A be a unital C-algebra. Recall that a (complex) vector space V is said to be an A-bimodule if it is equipped with bilinear maps A × V → V , (a, x) → a · x and V × A → V , (x, a) → x · a, such that (a · x) · b = a · (x · b), (ab) · x = a · (b · x), x · (ab) = (x · a) · b and 1 · x = x for all x ∈ V and all a, b ∈ A. If V and W are A-bimodules, a linear map φ : V → W is called an A-bimodule map if φ(a · x · b) = a · φ(x) · b, for all x ∈ V and all a, b ∈ A. Definition 2.1. Let A be a unital C-algebra. An AOU space (V, V +, e) will be called an AOU A-space if V is an A-bimodule and the conditions (1) (2) and (3) are satisfied. (a · x)∗ = x∗ · a∗, x ∈ V, a ∈ A, a · e = e · a, a ∈ A, a∗ · x · a ∈ V +, x ∈ V +, a ∈ A, For a complex vector space V , we let Mm,n(V ) denote the complex vector space of all m by n matrices with entries in V , and often use the natural identification Mm,n(V ) ≡ Mm,n ⊗ V . We write At for the transpose of a matrix A ∈ Mm,n(V ). We set Mn(V ) = Mn,n(V ), Mm,n = Mm,n(C) and Mn = Mn(C); we write In for the identity matrix in Mn. If V is an AOU A-space, we equip Mn(V ) with an involution by letting (xi,j)∗ = (x∗ j,i) and set (4) (ai,j) · (xi,j) =  n Xp=1 ai,p · xp,j i,j n and (xi,j) · (bi,j) = Xp=1  xi,p · bp,j i,j , whenever (xi,j) ∈ Mm,n(V ), (ai,j) ∈ Mk,m(A) and (bi,j) ∈ Mn,l(A), m, n, k, l ∈ N. Let A be a unital C-algebra and (V, V +, e) be an AOU A-space. We write en for the element of Mn(V ) whose diagonal entries coincide with e, while its off-diagonal entries are equal to zero. A family (Pn)n∈N, where Pn ⊆ Mn(V )h is a cone with Pn ∩ (−Pn) = {0}, n ∈ N, will be called a matrix ordering of V . A matrix ordering (Pn)n∈N will be called an operator OPERATOR SYSTEM STRUCTURES 5 A-system structure on V if P1 = V +, (5) A∗ · X · A ∈ Pn, whenever X ∈ Pm and A ∈ Mm,n(A), and en ∈ Mn(V ) is an Archimedean order unit for Pn for every n ∈ N. Condition (5) will be referred to as the A-compatibility of (Pn)n∈N. The triple S = (V, (Pn)n∈N, e) is called an operator A-system (see [13]); we write Mn(S)+ = Pn. Note that if B ⊆ A is a unital C-subalgebra, then every operator A-system is also an operator B-system in a canonical fashion. Op- erator C-systems are called simply operator systems. We note that every operator system has a canonical operator space structure (see [13]). Note that condition (2) is not a part of the standard definition of an operator A- system; it is however automatically satisfied, as easily follows from Theorem 2.2 below. Let H be a Hilbert space and B(H) be the space of all bounded linear op- erators on H. We write B(H)+ for the cone of all positive operators in B(H). We identify Mn(B(H)) with B(H n), where H n denotes the direct sum of n copies of H, and write Mn(B(H))+ = B(H n)+, n ∈ N. It is straightforward to see that B(H) is an operator system when equipped with the adjoint operation as an involution, the matrix ordering (Mn(B(H))+)n∈N, and the identity operator I as an Archimedean matrix order unit. Given AOU spaces (V, V +, e) and (W, W +, f ), a linear map φ : V → W is called unital if φ(e) = f , and positive if φ(V +) ⊆ W +. A linear map s : V → C is called a state on V if s is unital and positive. Let S and T be operator systems with units e and f , respectively. For a linear map φ : S → T , we let φ(n,m) : Mn,m(S) → Mn,m(T ) be the (linear) map given by φ(n,m)((xi,j)i,j) = (φ(xi,j))i,j, and set φ(n) = φ(n,n). The map φ is called n-positive if φ(n) is positive, and it is called completely positive if it is n-positive for all n ∈ N. A bijective completely positive map φ : S → T is called a complete order isomorphism if its inverse φ−1 is completely positive. In this case, we call S and T are completely order isomorphic; if φ is moreover unital, we say that S and T are unitally completely order isomorphic. Further, φ is called a complete isometry if φ(n) is an isometry for each n ∈ N. We note that a unital surjective map φ : S → T is a complete isometry if and only if it is a complete order isomorphism [3, 1.3.3]. We refer the reader to [13] for the general theory of operator systems and operator spaces, and in particular for the definition and basic properties of completely bounded maps. The following characterisation, extending the well-known Choi-Effros representation theorem for operator systems [13, Theorem 13.1], was established in [13, Corollary 15.12]. Theorem 2.2. Let A be a unital C-algebra and S be an operator system. The following are equivalent: (i) S is unitally completely order isomorphic to an operator A-system; 6 Y.-F. LIN AND I. G. TODOROV (ii) there exist a Hilbert space H, a unital complete isometry γ : S → B(H) and a unital -homomorphism π : A → B(H) such that γ(a · x) = π(a)γ(x) for all x ∈ S and all a ∈ A. We note that, if A is a unital C-algebra and S is an operator system that is also an operator A-bimodule satisfying (1), then S is an operator A-system precisely when the family (Mn(S)+)n∈N is A-compatible. 3. The extremal operator A-system structures In this section, we show that any AOU A-space can be equipped with two extremal operator A-system structures, and establish their universal properties. We first consider the minimal operator A-system structure. Note that, in the case where the AOU A-space is a C-algebra containing A, this operator system structure was first defined and studied in [20]. Let A be a unital C-algebra and (V, V +, e) be an AOU A-space. For n ∈ N, let C min n (V ; A) = {X ∈ Mn(V )h : C ∗ · X · C ∈ V +, for all C ∈ Mn,1(A)}. Remark 3.1. Suppose that (V, V +, e) is an AOU A-space and that B is a unital C-subalgebra of A. Then (V, V +, e) is also an AOU B-space in the natural fashion. Clearly, C min In particular, C min n (V ; C); note that the latter set coincides with the cone C min n (V ; A) is contained in C min (V ) introduced in [15, Definition 3.1]. n (V ; A) ⊆ C min n (V ; B). n Theorem 3.2. Let A be a unital C-algebra and (V, V +, e) be an AOU A-space. Then (C min (V ; A))n∈N is an operator A-system structure on V . Moreover, if (Pn)n∈N is an operator A-system structure on V then Pn ⊆ C min n (V ; A) for each n ∈ N. n Proof. Since V + is a cone, C min [15, Theorem 3.2] and Remark 3.1, C min X ∈ C min hence n (V ; A) is a cone, too. As a consequence of (V ; A)) = {0}. If m (V ; A), A ∈ Mm,n(A) and C ∈ Mn,1(A) then AC ∈ Mm,1(A) and n (V ; A) ∩ (−C min n C ∗ · (A∗ · X · A) · C = (AC)∗ · X · (AC) ∈ V +, n (V ; A). Thus, the family (C min n showing that A∗ · X · A ∈ C min A-compatible. (V ; A))n∈N is Suppose that (Pn)n∈N is an operator A-system structure on V . If X ∈ Pn n (V ; A). n (V ; A). It will follow from the proof of Theorem 3.7 below n (V ; A). To see that en is Archimedean, n (V ; A) for every r > 0. Let C ∈ Mn,1(A). then, by A-compatibility, C ∗ · X · C ∈ P1 = V +, and hence X ∈ C min Thus, Pn ⊆ C min that en is an order unit for C min suppose that X + ren ∈ C min Using (2), we have C ∗ · X · C + rC ∗C · e = C ∗ · (X + ren) · C ∈ V +, Let ǫ > 0 and T = (C ∗C + ǫ1)−1/2 ∈ A. We have that for all r > 0. C ∗ · X · C + rC ∗C · e + rǫe ∈ V +, for all r > 0 OPERATOR SYSTEM STRUCTURES 7 and hence, by (2) and (3), T (C ∗ · X · C)T + re ∈ V +, for all r > 0. Since e is Archimedean for V +, we have that T (C ∗ ·X ·C)T ∈ V +. Applying (3) again, we conclude that C ∗ · X · C = T −1(T (C ∗ · X · C)T )T −1 ∈ V +; thus X ∈ C min n (V ; A) and the proof is complete. (cid:3) (V ; A))n∈N the minimal operator A-system structure on V , We call (C min n and let The following theorem describes its universal property. Part (i) below was established in [20] in the case V is a C-algebra containing A. OMIN A(V ) =(cid:0)V, (C min n (V ; A))n∈N, e(cid:1) . Theorem 3.3. Let A be a unital C-algebra and (V, V +, e) be an AOU A-space. (i) Suppose that S is an operator A-system and φ : S → V is a posi- tive A-bimodule map. Then φ is completely positive as a map from S into OMIN A(V ). (ii) If T is an operator A-system with underlying space V and positive cone V +, such that for every operator A-system S, every positive A-bimodule map φ : S → T is completely positive, then there exists a unital A-bimodule map ψ : T → OMINA(V ) that is a complete order isomorphism. Proof. (i) Let S be an operator A-system and φ : S → V be a positive A-bimodule map. Suppose that X = (xi,j) ∈ Mn(S)+ and C = (ai)n i=1 ∈ Mn,1(A). Then C ∗ · X · C ∈ S +; since φ is a positive A-bimodule map, we have C ∗ · φ(n)(X) · C = n Xi,j=1 a∗ i · φ(xi,j) · aj = φ  n Xi,j=1 a∗ i · xi,j · aj  = φ(C ∗ · X · C) ∈ V +. n Thus, φ(n) maps Mn(S)+ into C min (V ; A) and hence φ is completely positive. (ii) Suppose that the operator A-system T satisfies the properties in (ii). Since the identity id : OMINA(V ) → V is a positive A-bimodule map, we have that id : OMINA(V ) → T is completely positive. On the other hand, the identity id : T → V is also positive and A-bimodular. By (i), id : T → OMINA(V ) is completely positive, and we can take ψ = id. (cid:3) We next consider the maximal operator A-system structure. For n ∈ N, set Dmax n (V ; A) =( k Xi=1 A∗ i · xi · Ai : k ∈ N, xi ∈ V +, Ai ∈ M1,n(A)) and let Dmax(V ; A) = (Dmax n (V ; A))n∈N. 8 Y.-F. LIN AND I. G. TODOROV Remark 3.4. Suppose that (V, V +, e) is an AOU A-space and that B is a unital C-subalgebra of A. Clearly, Dmax (V ; A). Given any AOU space (V, V +, e), in [15] the authors defined (V ; B) ⊆ Dmax n n Dmax n (V ) =( k Xi=1 Bi ⊗ xi : k ∈ N, xi ∈ V +, Bi ∈ M + n) . (V ) = Dmax n is the sum of matrices of the form A∗A, where n Since every matrix B ∈ M + A ∈ M1,n, we have that Dmax Lemma 3.5. Let A be a unital C-algebra and (V, V +, e) be an AOU A- space. Let Pn ⊆ Mn(V )h be a cone, n ∈ N, such that the family (Pn)∞ n=1 is A-compatible and P1 = V +. Then Dmax Proof. Let n ∈ N. If A ∈ M1,n(A) then (V ; A) ⊆ Pn, for each n ∈ N. (V ; C1). n n A∗ · V + · A = A∗ · P1 · A ⊆ Pn. Thus Dmax n (V ; A) ⊆ Pn. (cid:3) If x1, . . . , xn ∈ V we let diag(x1, . . . , xn) denote the element of Mn(V ) with x1, . . . , xn on its diagonal (in this order) and zeros elsewhere. Proposition 3.6. Let A be a unital C-algebra and (V, V +, e) be an AOU A-space. The following hold: (i) Dmax n (V ; A) = {A∗ · diag(x1, . . . , xm) · A : A ∈ Mm,n(A), xi ∈ V +, i = 1, . . . , m, m ∈ N}; (ii) Dmax(V ; A) is an A-compatible matrix ordering on V and e is a ma- trix order unit for it. Proof. (i) Let Dn denote the right hand side of the equality in (i). We first observe that Dn is a cone in Mn(V )h. If x1, . . . , xm ∈ V + and A = (ai,k)i,k ∈ Mm,n(A) then the (i, j)-entry of A∗ · diag(x1, . . . , xm) · A is equal thus, Dn ⊆ Mn(V )h. It is clear that Dn is closed under taking multiples with non-negative real numbers. Fix elements A∗ · diag(x1, . . . , xm) · A, and B∗ · diag(y1, . . . , yk) · B of Dn. Letting C = [A B]t, we have A∗ · diag(x1, . . . , xm) · A + B∗ · diag(y1, . . . , yk) · B = C ∗ · diag(x1, . . . , xm, y1, . . . , yk) · C ∈ Dn ; in other words, Dn is a cone. If B ∈ Mn,l(A) then B∗ · (A∗ · diag(x1, . . . , xm) · A) · B = (AB)∗ · diag(x1, . . . , xm) · (AB) ∈ Dl, n=1 is A-compatible. By (3), D1 = V +. Lemma 3.5 now implies and so (Dn)∞ that Dmax (V ; A) ⊆ Dn for n ∈ N. n k=1 a∗ k,i · xk · ak,j and, by (1), to Pm a∗ k,i · xk · ak,j!∗ = m Xk=1 m Xk=1 a∗ k,j · xk · ak,i; OPERATOR SYSTEM STRUCTURES 9 On the other hand, if x1, . . . , xm ∈ V + then, letting Ei ∈ M1,m(A) be the row with 1 at the ith coordinate and zeros elsewhere, we have that m diag(x1, . . . , xm) = E∗ i · xi · Ei ∈ Dmax m (V ; A). Xi=1 Since the family Dmax(V ; A) is A-compatible, A∗ · diag(x1, . . . , xm) · A ∈ Dmax n (V ; A), A ∈ Mm,n(A). Thus, Dn ⊆ Dmax n (V ; A) and (i) is established. (ii) By Remark 3.4 and [15, Proposition 3.10], en is an order unit for (V ; A). (cid:3) (V ; C1). By Remark 3.4 again, en is an order unit for Dmax Dmax n n For n ∈ N, let C max n (V ; A) = {X ∈ Mn(V ) : X + ren ∈ Dmax n (V ; A) for every r > 0}. Theorem 3.7. Let A be a unital C-algebra and (V, V +, e) be an AOU A-space. Then (C max (V ; A))n∈N is an operator A-system structure on V . Moreover, if (Pn)n∈N is an operator A-system structure on V then n C max n (V ; A) ⊆ Pn for each n ∈ N. Proof. Write Cn = C max Cn ⊆ C min Dmax n (V ; A), n ∈ N. By Theorem 3.2 and Lemma 3.5, (V ; A); thus, Cn ∩ (−Cn) = {0}. Since en is an order unit for (V ; A) ⊆ Cn, we have that en is an order unit for Cn. n Suppose that X ∈ Mn(V )h is such that X + ren ∈ Cn for every r > 0. (V ; A) and Dmax n n Let ǫ > 0; then X + ǫen =(cid:16)X + ǫ 2 en(cid:17) + ǫ 2 en ∈ Dmax n (V ; A) and hence X ∈ Cn. Thus, en is an Archimedean matrix order unit for Cn. It remains to show that the family (Cn)n∈N is A-compatible. To this end, let X ∈ Cn for some n ∈ N and A ∈ Mn,m(A). By Proposition 3.6, there exists R > 0 such that Rem − A∗ · en · A ∈ Dmax m (V ; A). Let r > 0. Since X + r A-compatible (Proposition 3.6), we have R en ∈ Dmax n (V ; A) and the family Dmax(V ; A) is A∗ · X · A + rem = (cid:16)A∗ ·(cid:16)X + r R en(cid:17) · A(cid:17) + r(cid:18)em − 1 R A∗ · en · A(cid:19) ∈ Dmax m (V ; A). It follows that A∗ · X · A ∈ Cm. Thus, (Cn)n∈N is an operator A-system structure on V . Suppose that (Pn)n∈N is an operator A-system structure on V and X ∈ Cn for some n ∈ N. By Lemma 3.5, X + ren ∈ Pn for all r > 0 and since en is an Archimedean order unit for Pn, we conclude that X ∈ Pn. Thus, Cn ⊆ Pn, and the proof is complete. (cid:3) 10 Y.-F. LIN AND I. G. TODOROV We call (C max n and let (V ; A))n∈N the maximal operator A-system structure on V OMAX A(V ) = (V, (C max n (V ; A))n∈N, e). Remark. Recall that, given an AOU space (V, V +, e), the maximal op- erator system structure (C max (V ))n∈N on V was defined in [15] by letting C max (V ) defined in Remark 3.4. It follows that the maximal operator system OMAX(V ) defined in [15] coincides with OMAXC(V ). (V ) be the Archimedeanisation of the cone Dmax n n n Theorem 3.8. Let A be a unital C-algebra and (V, V +, e) be an AOU A-space. (i) Suppose that S is an operator A-system and φ : V → S is a positive A-bimodule map. Then φ is completely positive as a map from OMAX A(V ) into S. (ii) Suppose that T is an operator A-system with underlying space V and positive cone V +, such that for every operator A-system S, every positive A- bimodule map φ : T → S is completely positive. Then there exists a unital A- bimodule map ψ : T → OMAXA(V ) that is a complete order isomorphism. n (V ; A) imply that φ(n)(Dmax Proof. (i) Let S is an operator A-system and φ : V → S be a posi- tive A-bimodule map. The modularity property of φ and the definition (V ; A)) ⊆ Mn(S)+. Suppose that of Dmax X ∈ C max (V ; A). Letting z = φ(e), we now have that φ(n)(X) + r(z ⊗ In) ∈ Mn(S)+ for every r > 0. Since Mn(S)+ is closed, this implies that φ(n)(X) ∈ Mn(S)+. Thus, φ is completely positive. n n (ii) is similar to the proof of Theorem 3.3 (ii). (cid:3) Remark. Let A be a C-algebra and AA (resp. SA) be the category, whose objects are AOU A-spaces (resp. operator A-systems) and whose morphisms are unital positive (resp. unital completely positive) maps. It is easy to see that the correspondences V → OMINA(V ) and V → OMAXA(V ) are covariant functors from AA into SA. We finish this section with considering the case where V = Mk and A coincides with its subalgebra Dk of all diagonal matrices. Proposition 3.9. We have that Mk = OMINDk (Mk) = OMAXDk (Mk). Proof. Suppose that X = (Xi,j)i,j belongs to Mn(OMINDk (Mk))+. Let ξ = (λi,1, . . . , λi,k)n i=1 be a vector in Cnk. Let Di = diag(λi,1, . . . , λi,k), and write ξi for the vector (λi,1, . . . , λi,k) in Ck, i = 1, . . . , n. Letting e be the vector in Ck with all entries equal to one, we have n n (Xξ, ξ) = (Xi,jξj , ξi) = (D∗ i Xi,j Dj e, e). Xi,j=1 Xi,j=1 It follows by the assumption that (Xξ, ξ) ≥ 0; thus, X ∈ M + Theorem 3.2, Mk = OMINDk(Mk). nk and, by OPERATOR SYSTEM STRUCTURES 11 Now fix X = (Xi,j)i,j ∈ M + nk. Since X is the sum of rank one operators in M + nk, in order to show that X ∈ Mn(OMAXDk (Mk))+, it suffices to assume that X is itself of rank one. Write X = RR∗, where R ∈ Mnk,1, and suppose that R = (R1, . . . , Rn)t, where Ri ∈ Mk,1, i = 1, . . . , n. We have that X = (RiR∗ i,j=1. Let J ∈ Mk be the matrix with all its entries equal to one, and let Di be the diagonal matrix whose entries coincides with the vector Ri, i = 1, . . . , n. Then X = (DiJ D∗ i,j=1, showing that X ∈ Mn(OMAXDk (Mk))+. By Theorem 3.7, Mk = OMAXDk(Mk). (cid:3) j )n j )n Remark. We note that the minimal and the maximal operator A-system structure are in general distinct. Indeed, this is the case even when V = Mk and A = CI [15]. 4. Dual operator A-systems In this section, we establish a representation theorem for dual operator A-systems. An operator system S is called a dual operator system if it is a dual operator space, that is, if there exists an operator space S∗ such that (S∗)∗ ∼= S completely isometrically [4]. Here, and in the sequel, we denote by X ∗ the operator space dual [3] of an operator space X , and we use the same notation for the dual Banach space of a normed space X ; it will be clear from the context with which category we are working. Let S be an operator system. If H is a Hilbert space and φ : S → B(H) is a unital complete isometry such that φ(S) is weak* closed, then φ(S), and therefore S, is a dual operator space; thus, in this case, S is a dual operator system. The converse statement was established by Blecher and Magajna in [4]. Theorem 4.1 ([4]). If S is a dual operator system then there exists a Hilbert space H, a weak* closed operator system U ⊆ B(H) and a unital surjective complete order isomorhism φ : S → U that is also a a weak* homeomor- phism. Remark 4.2. Suppose that S is a dual operator system and S∗ is an op- erator space such that, up to a complete isometry, S = (S∗)∗. Then Mn(S) is an operator system in a canonical fashion; in fact, if S ⊆ B(H) for some Hilbert space H, then Mn(S) ⊆ B(H n). By [3, 1.6.2], up to a complete n)∗, where ⊗ is the projective operator space isometry, Mn(S) = (S∗ ⊗M ∗ tensor product. It follows that Mn(S) is a dual operator system, and its canonical weak* topology coincides with the topology of entry-wise weak* convergence: for a net ((xα i,j)i,j)α ⊆ Mn(S) and an element (xi,j)i,j ∈ Mn(S), we have (cid:0)(xα i,j(cid:1)i,j )α →w∗ α (xi,j)i,j ⇐⇒ (cid:10)xα i,j, φ(cid:11) → α hxi,j, φi , i, j = 1, . . . , n, φ ∈ S∗. 12 Y.-F. LIN AND I. G. TODOROV Recall that a W-algebra is a C-algebra that is also a dual Banach space; by Sakai's Theorem [21], every W-algebra possesses a faithful - representation on a Hilbert space H, whose image is a von Neumann al- gebra (that is, a weak* closed subalgebra of B(H) containing the identity operator), which is also a weak* homeomorphism. Definition 4.3. Let A be a W-algebra. An operator system S will be called a dual operator A-system if (i) S is an operator A-system, (ii) S is a dual operator system, and (iii) the map from A × S into S, sending the pair (a, x) to a · x, is sepa- rately weak continuous. Note that, if S is a dual operator system then the involution is weak* continuous, and thus (1) implies that if S is in addition a dual operator A-system then the map A × S × A → S, (a, x, b) → a · x · b, is separately weak* continuous. If S and T are dual operator systems, a linear map φ : S → T will be called normal if it is weak* continuous. Suppose that H is a Hilbert space, γ : S → B(H) is a unital complete order isomorphism such that γ(S) is weak* closed and γ : S → γ(S) is a weak* homeomorphism, and π : A → B(H) is a unital normal -homomorphism such that γ(a · x) = π(a)γ(x) for all x ∈ S and all a ∈ A. It is clear that, in this case, S is a dual operator A-system. Theorem 4.7 below establishes the converse of this fact. The result is both a weak version of Theorem 2.2 and an A-module version of Theorem 4.1. We will need two lemmas. Recall that, if A is a W-algebra and n ∈ N then Mn(A) is a W-algebra in a canonical way. Remark 4.4. Let A be a W-algebra and S be a dual operator A-system. It is straightforward to verify that Mn(S) is a dual operator Mn(A)-system, when it is equipped with the action defined in (4). Lemma 4.5. Let A be a W-algebra, S be a dual operator A-system and φ : S → C be a normal state. Then the functional ω : A → C given by ω(a) = φ(a · 1), a ∈ A, is a normal state of A and (6) φ(a · x · b) ≤ ω(aa∗)1/2ω(b∗b)1/2, for all a ∈ M1,m(A), b ∈ Mm,1(A), x ∈ Mm(S) with kxk ≤ 1, and m ∈ N. Proof. Let H, γ and π be as in Theorem 2.2, and let φ′ : γ(S) → C be given by φ′(γ(x)) = φ(x), x ∈ S. If a, b ∈ A then ω(ab) = φ((ab) · 1) = φ′(γ((ab) · 1)) = φ′(π(ab)γ(1)) = φ′(π(ab)) = φ′(π(a)γ(1)π(b)) = φ′(γ(a · 1 · b)) = φ(a · 1 · b). OPERATOR SYSTEM STRUCTURES 13 Thus, ω(a∗a) = φ(a∗ · 1 · a) ≥ 0 for every a ∈ A, and hence ω is positive. Moreover, ω(1) = φ(1) = 1 and hence ω is a state. By the separate weak* continuity of the A-module action on S, the state ω is normal. Suppose that φ′ has the form ∞ If x ∈ Mm(S), kxk ≤ 1, a ∈ φ′(T ) = (T ξi, ξi), T ∈ γ(S), i=1 kξik2 = 1. M1,m(A) and b ∈ Mm,1(A), then Xi=1 where (ξi)i∈N ⊆ H with P∞ φ(a · x · b) = (cid:12)(cid:12)(cid:12) φ′(cid:16)π(1,m)(a)γ(m)(x)π(m,1)(b)(cid:17)(cid:12)(cid:12)(cid:12) Xi=1(cid:16)π(1,m)(a)γ(m)(x)π(m,1)(b)ξi, ξi(cid:17)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Xi=1(cid:12)(cid:12)(cid:12)(cid:16)γ(m)(x)π(m,1)(b)ξi, π(m,1)(a∗)ξi(cid:17)(cid:12)(cid:12)(cid:12) 2!1/2 ∞ ≤ ∞ Xi=1(cid:13)(cid:13)(cid:13) π(m,1)(b)ξi(cid:13)(cid:13)(cid:13) Xi=1(cid:13)(cid:13)(cid:13) ∞ ∞ ≤ 2!1/2 π(m,1)(a∗)ξi(cid:13)(cid:13)(cid:13) = φ′(π(b∗b))1/2φ′(π(aa∗))1/2 = ω(aa∗)1/2ω(b∗b)1/2. (cid:3) We will need the following modification of a result of R. R. Smith [23] on automatic complete boundedness. Its proof is a straightforward modification of the proof of [23, Theorem 2.1] and is hence omitted. Theorem 4.6. Let A be a unital C-algebra, S be an operator A-system and ρ : A → B(H) be a cyclic -representation. Suppose that Φ : S → B(H) is a linear map such that Φ(a · x · b) = ρ(a)Φ(x)ρ(b) for all x ∈ S and all a, b ∈ A. If Φ is contractive then Φ is completely contractive. Theorem 4.7. Let A be a W-algebra and S be a dual operator A-system. Then there exist a Hilbert space H, a unital complete order embedding γ : S → B(H) with the property that γ(S) is weak closed and γ is a weak* homeomorphism, and a unital normal -homomorphism π : A → B(H), such that (7) γ(a · x) = π(a)γ(x), x ∈ S, a ∈ A. Proof. The proof is motivated by the proof of [4, Theorem 1.1] and relies on ideas which go back to the proof of Ruan's Theorem [6, Theorem 2.3.5]. Fix n ∈ N and let B = Mn(A). By Remark 4.4, Mn(S) is a dual operator B- system. Let x ∈ Mn(S) be a selfadjoint element of norm one and ǫ ∈ (0, 1). By the proof of Theorem 1.1 given in [4], there exists a normal state φ on Mn(S) such that (8) φ(x) > 1 − ǫ. 14 Y.-F. LIN AND I. G. TODOROV Let ω : B → C be the normal state given by ω(b) = φ(b · 1), b ∈ B. By Lemma 4.5, (9) φ(a · y · b) ≤ ω(aa∗)1/2ω(b∗b)1/2, for all y ∈ Mnm(S) with kyk ≤ 1, a ∈ M1,m(B) and b ∈ Mm,1(B), m ∈ N. Let ρ : B → B(H) be the GNS representation arising from ω and ξ be its corresponding unit cyclic vector. By [24, Proposition III.3.12], ρ is normal. It follows that there exists a normal unital -representation θ : A → B(K) such that, up to unitary equivalence, H = K ⊗ Cn and ρ = θ(n). Inequality (9) implies φ(a∗ · y · b) ≤ kρ(b)ξkkρ(a)ξkkyk, a, b ∈ B, y ∈ Mn(S). Thus, the sesqui-linear form Ly : (ρ(B)ξ) × (ρ(B)ξ) → C given by Ly(ρ(b)ξ, ρ(a)ξ) = φ(a∗ · y · b), a, b ∈ B, is bounded and has norm not exceeding kyk. It follows that there exists a linear operator Φ(y) : ρ(B)ξ → ρ(B)ξ such that (10) and (11) (Φ(y)ρ(b)ξ, ρ(a)ξ) = φ(a∗ · y · b), a, b ∈ B, kΦ(y)k ≤ kyk. Since ρ(B)ξ in dense in H, the operator Φ(y) can be extended to an operator on H. By (10), the map Φ : Mn(S) → B(H) is linear and hermitian and, by (11), it is contractive. For a, b, c, d ∈ B, by (10), we have (Φ(c∗ · y · d)ρ(b)ξ, ρ(a)ξ) = (ρ(c∗)Φ(y)ρ(d)ρ(b)ξ, ρ(a)ξ). The density of ρ(B)ξ in H now implies that (12) Φ(c∗ · y · d) = ρ(c∗)Φ(y)ρ(d), c, d ∈ B, y ∈ Mn(S). We show that Φ is weak* continuous. Suppose that (yα)α ⊆ Mn(S) is a net of contractions such that yα →α y in the weak* topology, for some y ∈ Mn(S). Fix δ > 0, η, ζ ∈ H, and choose a, b ∈ B such that kρ(b)ξ − ηk < δ and kρ(a)ξ − ζk < δ. OPERATOR SYSTEM STRUCTURES 15 Let α0 be such that φ(a∗ · yα · b) − φ(a∗ · y · b) < δ if α ≥ α0. For α ≥ α0 we have (Φ(yα)η, ζ) − (Φ(y)η, ζ) ≤ (Φ(yα)η, ζ) − (Φ(yα)ρ(b)ξ, ρ(a)ξ) + (Φ(yα)ρ(b)ξ, ρ(a)ξ) − (Φ(y)ρ(b)ξ, ρ(a)ξ) + (Φ(y)ρ(b)ξ, ρ(a)ξ) − (Φ(y)η, ζ) = (Φ(yα)η, ζ) − (Φ(yα)ρ(b)ξ, ρ(a)ξ) + φ(a∗ · yα · b) − φ(a∗ · y · b) + (Φ(y)ρ(b)ξ, ρ(a)ξ) − (Φ(y)η, ζ) ≤ (Φ(yα)η, ζ) − (Φ(yα)ρ(b)ξ, ζ) + (Φ(yα)ρ(b)ξ, ζ) − (Φ(yα)ρ(b)ξ, ρ(a)ξ) + φ(a∗ · yα · b) − φ(a∗ · y · b) + (Φ(y)ρ(b)ξ, ρ(a)ξ) − (Φ(y)η, ρ(a)ξ) + (Φ(y)η, ρ(a)ξ) − (Φ(y)η, ζ) ≤ δ(kζk + kηk + kρ(a)ξk + kρ(b)ξk + 1) ≤ δ(2kζk + 2kηk + 2δ + 1). We thus showed that Φ(yα) →α Φ(y) in the weak operator topology; since the net (Φ(yα))α is bounded, the convergence is in fact in the weak* topology. It follows from Shmulyan's Theorem that the map Φ is weak* continuous. Identity (12) easily implies that there exists a (normal) map Ψ : S → B(K) such that Φ = Ψ(n). Since Φ is hermitian and contractive, so is Ψ. By (12) and Theorem 4.6, the map Φ, and hence Ψ, is completely contractive. Now (12) implies (13) Ψ(a · z · b) = θ(a)Ψ(z)θ(b), z ∈ S, a, b ∈ A. By (10), 1 = φ(1) = (Φ(1)ξ, ξ) ≤ kΦ(1)kkξk2 ≤ 1. Thus Φ(1)ξ = ξ; by (12), Φ(1)ρ(b)ξ = ρ(b)Φ(1)ξ = ρ(b)ξ, b ∈ B, and since ξ is cyclic for ρ, we conclude that Φ(1) = 1. Ψ(1) = 1. It follows that The map Ψ, constructed in the previous paragraph, depends on the el- ement x ∈ Mn(S), and on the chosen ǫ. Note that, by (8) and (10), (resp. θ) as above, over all selfadjoint x ∈ Mn(S) with norm one, all n ∈ N, and all ǫ ∈ (0, 1). The map γ is unital, weak* continuous, hermitian, and has the property that if x ∈ Mn(S) is selfadjoint then kxk = 1 implies (cid:13)(cid:13)Ψ(n)(x)(cid:13)(cid:13) > 1 − ǫ. Let γ (resp. π) be the direct sum of the maps Ψ (cid:13)(cid:13)γ(n)(x)(cid:13)(cid:13) = 1. This easily yields that γ is completely positive and has a completely positive inverse. As in the proof of [4, Theorem 1.1], the image of γ is weak* closed and γ is a weak* homeomorphism onto its range. In addition, π is a normal -representation as a direct sum of such. Condition (7) follows from (13). (cid:3) 16 Y.-F. LIN AND I. G. TODOROV 5. The dual extremal operator A-system structures In this section, we study dual versions of the extremal operator A-system structures considered in Section 3. We start with the definition of a dual AOU space. Note first that, if (V, V +, e) is an AOU space then the expression kvk = sup{f (v) : f a state on V } defines a norm on V , called the order norm [16]; in the sequel we equip V with its order norm. If V is a dual Banach space, the weak continuous functionals on V will be called normal functionals. Definition 5.1. A dual AOU space is an AOU space (V, V +, e), which is also a dual Banach space, and (i) the involution is weak* continuous; (ii) V + is weak* closed, and (iii) for v ∈ V , kvk = sup{f (v) : f a normal state on V }, and the weak* topology of V is determined by normal states of V . Suppose that (V, V +, e) is a dual AOU space, and let V∗ be the predual of V . Note that the algebraic tensor product V∗ ⊗ M ∗ n can be canonically embedded into the dual of Mn(V ). By the weak* topology on Mn(V ) we will mean the topology arising from this duality; thus, (xα i,j) →α (xi,j) if and only if xα i,j →α xi,j for every i, j. Definition 5.2. Let A be a W-algebra. A dual AOU space (V, V +, e) will be called dual AOU A-space if (i) (V, V +, e) is an AOU A-space, and (ii) the left (and hence the right) A-module action is separately weak continuous. Definition 5.3. Let A be a W-algebra and (V, V +, e) be a dual AOU A- space. A matrix ordering (Cn)n∈N on V will be called a dual operator A- system structure on V if (V, (Cn)n∈N, e) is a dual operator A-system whose weak topology coincides with that of V , and C1 = V +. Theorem 5.4. Let A be a W-algebra, (V, V +, e) be a dual AOU A-space and (Cn)n∈N be an operator A-system structure on V . The following are equivalent: (i) (Cn)n∈N is a dual operator A-system structure on V ; (ii) Cn is weak closed for each n ∈ N. Proof. (i)⇒(ii) Let S = (V, (Cn)n∈N, e). By Theorem 4.7, there exist a Hilbert space H and a complete order embedding γ : S → B(H) such that γ(S) is weak* closed and γ is a weak* homeomorphism. Clearly, Mn(γ(S))+ is weak* closed in Mn(B(H)). Note that the weak* topology on Mn(B(H)) = B(H n) is given by entry-wise weak* convergence. On the other hand, since γ is a weak* homeomorphism, we have that if ((xα i,j))α ⊆ Mn(V ) and (xi,j) ∈ Mn(V ) then (xα i,j) →α γ(xi,j) for every i, j. It follows that Cn is weak* closed. i,j) →α (xi,j) weak* if and only if γ(xα OPERATOR SYSTEM STRUCTURES 17 (ii)⇒(i) Let S = (V, (Cn)n∈N, e). For each n, let Pn = {φ : V → Mn : weak* continuous unital completely positive map} . Cn and let J : V → B(H) be the map given by J(x) = Let H = ⊕n∈N ⊕φ∈Pn ⊕n∈N⊕φ∈Pn φ(x). It is clear that J is a weak* continuous completely positive map. In addition, by condition (iii) from Definition 5.1, J is isometric. To show that J is a complete order isomorphism, assume that J (n)(X) ≥ 0 for some X = (xi,j) ∈ Mn(V )h and that, by way of contradiction, X does not belong to Cn. The space Mn(V ), equipped with the topology of weak* convergence, is a locally convex topological vector space. By a geometric form of the Hahn-Banach Theorem, there exists a functional s : Mn(V ) → C, continuous with respect to the topology of entry-wise weak* convergence, such that s(Cn) ⊆ R+ but s(X) < 0. By [13, Theorem 6.1], the map φs : V → Mn, given by φs(x) = (si,j(x))i,j (and where si,j(x) = s(Ei,j ⊗ x)), is completely positive. It is clear that φs is normal. In addition, φ(n) does not map X to a positive matrix. After normalisation, we may assume that φs is contractive. s Let P = φs(e); then P is a positive contraction. Assume that rank(P ) = k and let Q be the projection onto ker(P )⊥. It was shown in the proof of [13, Theorem 13.1] that, if A ∈ Mn,k and B ∈ Mk,n are matrices such that A∗P A = Ik and AB = Q, and ψ is the mapping given by ψ(x) = A∗φs(x)A, then ψ is a (unital completely positive) map such that ψ(n)(X) is not positive. Clearly, ψ is normal, and hence an element of Pk. This contradicts the fact that J (n)(X) ≥ 0. To show that J is a weak* homeomorphism, suppose that J(xα) →α J(x) in the weak* topology, for some net (xα) ⊆ V and some element x ∈ V . Then φ(xα) → φ(x) for all normal positive functionals φ. By condition (iii) of Definition 5.1, xα → x in the weak* topology of V . We finally note that J(V ) is weak* closed in B(H). Suppose that J(xα) → T , where T ∈ B(H) and (xα)α ⊆ V is a net such that the net J(xα)α is bounded. Since J is an isometry, (xα)α is also bounded, and hence has a subnet (xβ)β, weak* convergent to an element of V , say x. Since J is weak* continuous, we conclude that T = limβ J(xβ) = J(x), and hence T ∈ J(V ). By the Krein-Smulyan, J(V ) is weak* closed. By the previous paragraphs, the weak* topology of V coincides with the weak* topology of the operator system S. It now follows that the A-module operations on S are separately weak* continuous; thus, S is a dual operator A-system and the proof is complete. (cid:3) As the next two statements show, if (V, V +, e) is a dual AOU A-space then the minimal operator A-system structure defined in Section 3 is auto- matically a dual minimal operator A-system structure. Theorem 5.5. Let A be a W-algebra and (V, V +, e) be a dual AOU A- space. Then (C min n (V ; A))n∈N is a dual operator A-system structure. 18 Y.-F. LIN AND I. G. TODOROV Proof. Since the A-module actions on V are weak continuous, C min is weak* closed for each n ∈ N. By Theorem 5.4, (C min operator A-system structure. n (V ; A) (V ; A))n∈N is a dual (cid:3) n Theorem 5.6. Let A be a W-algebra and (V, V +, e) be a dual AOU A- space. (i) Suppose that S is a dual operator A-system and φ : S → V is a normal positive A-bimodule map. Then φ is completely positive as a map from S into OMINA(V ). (ii) If T is a dual operator A-system with underlying space V and posi- tive cone V +, such that for every dual operator A-system S, every normal positive A-bimodule map φ : S → T is completely positive, then there exists a unital normal A-bimodule map ψ : T → OMINA(V ) that is a complete order isomorphism and a weak homeomorphism. Proof. (i) is a direct consequence of Theorem 3.3 (i). The proof of (ii) follows by a standards argument, similar to the one given in the proof of Theorem 3.3 (ii). (cid:3) In the remainder of the section, we consider the dual maximal operator A- system structure. For a W-algebra A and a dual AOU A-space (V, V +, e), set W max n (V ; A) = C max n (V ; A) w∗ , n ∈ N. Theorem 5.7. Let A be a W-algebra and (V, V +, e) be a dual AOU A- space. Then (W max (V ; A))n∈N is a dual operator A-system structure on V . Moreover, if (Pn)n∈N is a dual operator A-system structure on V then W max (V ; A) ⊆ Pn for each n ∈ N. n n Proof. By Theorem 3.7, (C max (V ; A))n∈N is an operator system A-structure on V . It follows by the separate weak* continuity of the A-module actions on V and the definition of the Mn(A)-module operations on Mn(V ) (see (4)) that the family (W max (V ; A))n∈N is A-compatible. n n Since the element e is a matrix order unit for (Dmax It follows that (V, (W max dition (ii) of Definition 5.1, V + = W max closed, Theorem 5.4 implies that it is a dual operator A-system. (V ; A))n∈N, e) is an operator A-system; by con- (V ; A). Since its cones are weak* n 1 Suppose that (Pn)n∈N is a dual operator A-system structure on V . Fix (V ; A) ⊆ Pn. By Theorem 5.4, Pn is weak* (cid:3) n ∈ N. By Theorem 3.7, C max n closed. It follows that W max (V ; A) ⊆ Pn. n We denote by OMAXw∗ A (V ) the operator system (V, (W max n (V ; A))n∈N, e). sition 3.6) and Dmax order unit for (W max trix order unit for (W max X + ren ∈ W max topology and W max n n n n (V ; A))n∈N (see Propo- (V ; A) ⊆ W max (V ; A) for each n ∈ N, e is a matrix (V ; A))n∈N. To show that e is an Archimedean ma- (V ; A))n∈N, suppose that X ∈ Mn(V ) is such that (V ; A) for all r > 0. Since X + ren →r→0 X in the weak* n n n (V ; A) is weak* closed, X ∈ W max (V ; A). n OPERATOR SYSTEM STRUCTURES 19 Theorem 5.8. Let A be a W-algebra and (V, V +, e) be a dual AOU A- space. (i) Suppose that S is a dual operator A-system and φ : V → S is a normal positive A-bimodule map. Then φ is completely positive as a map from OMAXw∗ A (V ) into S. (ii) If T is a dual operator A-system with underlying space V and posi- tive cone V +, such that for every dual operator A-system S, every normal positive A-bimodule map φ : T → S is completely positive, then there exists a unital normal A-bimodule map ψ : T → OMAXw∗ A (V ) that is a complete order isomorphism and a weak homeomorphism. Proof. (i) By Theorem 3.8 (i), φ(n)(C max weak* continuous and Mn(S)+ is weak* closed, φ(n)(W max n (V ; A)) ⊆ Mn(S)+. Since φ is (V ; A)) ⊆ Mn(S)+. (cid:3) n (ii) similar to the proof of Theorem 3.3 (ii). Remark. Let A be a W-algebra and Aw∗ A ) be the category, whose objects are dual AOU A-spaces (resp. dual operator A-systems) and whose morphisms are weak continuous unital positive (resp. weak* con- tinuous unital completely positive) maps. It is easy to see that the corre- spondences V → OMINw∗ A (V ) are covariant functors from Aw∗ A (V ) and V → OMAXw∗ A (V ) = OMINA(V ) as per Theorem 5.5. A into Sw∗ A , here OMINw∗ A (resp. Sw∗ 6. Inflated Schur multipliers In this section, we introduce an operator-valued version of classical mea- surable Schur multipliers, and characterise them in a fashion, similar to the well-known descriptions in the scalar-valued case [9, 17]. Let (X, µ) be a standard measure space. We denote by χα the characteris- tic function of a measurable set α ⊆ X. If f and g are measurable functions defined on X, we write f ∼ g when f (x) = g(x) for almost all x ∈ X. Throughout the section, let H = L2(X, µ) and fix a separable Hilbert space K. For a function a ∈ L∞(X, µ), let Ma be the operator on H given by Maf = af , f ∈ H, and set D = {Ma : a ∈ L∞(X, µ)} . We denote by H ⊗ K the Hilbertian tensor product of H and K. Note that H ⊗K is unitarily equivalent to the space L2(X, K) of all weakly measurable functions g : X → K such that kgk2 :=(cid:0)RX kg(x)k2dµ(x)(cid:1)1/2 If U ⊆ B(H) and V ⊆ B(K), we denote by U ¯⊗V the spacial weak* tensor product of U and V. We write M(X, B(K)) for the space of all functions F : X → B(K) such that, for all ξ0 ∈ K, the functions x → F (x)ξ0 and x → F (x)∗ξ0 are weakly measurable. Note that D ¯⊗B(K) can be canonically identified with the space L∞(X, B(K)) of all bounded functions F in M(X, B(K)) [24]. Through this identification, a function F gives rise to the operator MF ∈ B(L2(X, K)), defined by < ∞. (MF ξ)(x) = F (x)(ξ(x)), x ∈ X, ξ ∈ L2(X, K). 20 Y.-F. LIN AND I. G. TODOROV It is easy to see that if k ∈ M(X × X, B(K)) then the function (x, y) → kk(x, y)k is measurable as a function from X × X into [0, +∞]. Let L2(X × X, B(K)) be the space of all functions k ∈ M(X × X, B(K)) for which kkk2 :=(cid:18)ZX×X kk(x, y)k2dµ(x)dµ(y)(cid:19)1/2 < ∞. (Note that the functions from the space L2(X ×X, B(K)) need not be weakly measurable.) If k ∈ L2(X × X, B(K)) and ξ, η ∈ L2(X, K) then, by [24, Lemma 7.5], the function (x, y) → (k(x, y)(ξ(y)), η(x)) is measurable. Stan- dard arguments (see [12, p. 391]) show that the formula (Tkξ, η) =ZX×X (k(x, y)(ξ(y)), η(x)) dµ(y)dµ(x), x, y ∈ X, ξ, η ∈ L2(X, K), defines a bounded operator on L2(X, K) with kTkk ≤ kkk2. If K = C, the operators of the form Tk are precisely the Hilbert-Schmidt operators on H. Remark 6.1. For an element k ∈ L2(X × X, B(K)), we have that Tk = 0 if and only if k(x, y) = 0 for almost all (x, y) ∈ X × X. Proof. Suppose that Tk = 0; then, for ξ, η ∈ K and f, g ∈ L2(X), we have RX×X f (x)g(y)(k(x, y)ξ, η)dµ(y)dµ(x) = 0. Thus, (k(x, y)ξ, η) = 0 almost everywhere. Since K is separable and k(x, y) is bounded for all x, y ∈ X, this implies that k(x, y) = 0 almost everywhere. The converse direction is trivial. (cid:3) We equip the linear space {Tk : k ∈ L2(X × X, B(K))} with the operator space structure arising from its inclusion into B(H ⊗K). Similarly, whenever S is an operator system and S0 ⊆ S is a self-adjoint (not necessarily unital) subspace of S, we equip S0 with the matrix ordering inherited from S, and thus talk about a linear map from S0 into an operator system T being positive or completely positive. For functions ϕ ∈ L∞(X ×X, B(K)) and k ∈ L2(X ×X), let ϕk : X ×X → B(K) be the function given by (ϕk)(x, y) = k(x, y)ϕ(x, y), x, y ∈ X. It is straightforward to check that ϕk ∈ L2(X × X, B(K)). Definition 6.2. A function ϕ ∈ L∞(X × X, B(K)) will be called an (in- flated) Schur multiplier if the map Tk −→ Tϕk, k ∈ L2(X × X), is completely bounded. We will denote by S(X, K) the space of all inflated Schur multipliers with values in B(K). If ϕ ∈ S(X, K) then the map Sϕ : Tk → Tϕk defined on the space S2(H) of all Hilbert-Schmidt operators on H extends to a completely bounded map from K(H) into B(H ⊗ K), which will be denoted in the same way. By taking the second dual of Sϕ, and composing with the OPERATOR SYSTEM STRUCTURES 21 weak* continuous projection from B(H ⊗ K)∗∗ onto B(H ⊗ K), we obtain a completely bounded weak* continuous map from B(H) into B(H ⊗ K) which for simplicity will still be denoted by Sϕ. Theorem 6.3. Let ϕ ∈ L∞(X × X, B(K)). The following are equivalent: (i) ϕ ∈ S(X, K); (ii) there exist functions Ai ∈ L∞(X, B(K)) and Bi ∈ L∞(X, B(K)), i=1 Bi(y)∗Bi(y) converge almost everywhere in the weak* topology, ∞ esssup x∈X (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=1 i ∈ N, such that the seriesP∞ Ai(x)Ai(x)∗(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=1 ϕ(x, y) = (14) and ∞ i=1 Ai(x)Ai(x)∗ andP∞ y∈X (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=1 < ∞, esssup ∞ < ∞, Bi(y)∗Bi(y)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Ai(x)Bi(y), a.e. on X × X, where the sum is understood in the weak* topology. Proof. (ii)⇒(i) Considering Ai, Bi ∈ D ¯⊗B(K), i ∈ N, the assumptions imply that A = (Ai)i∈N (resp. B = (Bi)i∈N) is a bounded row (resp. column) operator. It follows that the map Ψ : B(H) → B(H ⊗ K), given by Ψ(T ) = ∞ Xi=1 Ai(T ⊗ I)Bi, T ∈ B(H), is well-defined and completely bounded. Let k ∈ L2(X × X) ∩ L∞(X × X), ξ, η ∈ K and f, g ∈ L2(X) ∩ L1(X). For almost all (x, y) ∈ X × X, we have k(x, y)f (y)g(x) (ϕ(x, y)ξ, η)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12) ∞ (Bi(y)ξ, Ai(x)∗η) ∞ ≤ kkk∞f (y)g(x) ≤ kkk∞f (y)g(x) Xi=1 Xi=1 ≤ kkk∞f (y)g(x) ∞ Xi=1 ≤ kkk∞f (y)g(x)kAkkBkkξkkηk, kBi(y)ξkkAi(x)∗ηk kBi(y)ξk2!1/2 ∞ Xi=1 kAi(x)∗ηk2!1/2 22 Y.-F. LIN AND I. G. TODOROV while the function (x, y) → f (y)g(x) is integrable with respect to µ × µ. By the Lebesgue Dominated Convergence Theorem, we now have (Ψ(Tk)(f ⊗ ξ), g ⊗ η) Ai(Tk ⊗ I)Bi(f ⊗ ξ), g ⊗ η! k(x, y)f (y)g(x)(Bi(y)ξ, Ai(x)∗η)dµ(x)dµ(y) k(x, y)f (y)g(x) ∞ Xi=1 Ai(x)Bi(y)! ξ, η! dµ(x)dµ(y) k(x, y)f (y)g(x) (ϕ(x, y)ξ, η) dµ(x)dµ(y) ∞ = = ∞ Xi=1 Xi=1ZX×X = ZX×X = ZX×X = ZX×X f (y)g(x) ((ϕk)(x, y)ξ, η) dµ(x)dµ(y) = (Tϕk(f ⊗ ξ), g ⊗ η) . By linearity and the density of L2(X × X) ∩ L∞(X × X) in L2(X × X) and of L2(X) ∩ L1(X) in L2(X), it follows that ϕ ∈ S(X, K) and Ψ = Sϕ. (i)⇒(ii) Let ϕ ∈ S(X, K). For k ∈ L2(X × X), a, b ∈ L∞(X), ξ, η ∈ K and f, g ∈ L2(X), we have (Sϕ(MbTkMa)(f ⊗ ξ), g ⊗ η) = ZX×X a(y)b(x)f (y)g(x) ((ϕk)(x, y)ξ, η) dµ(x)dµ(y) = ((Mb ⊗ I)Sϕ(Tk)(Ma ⊗ I)(f ⊗ ξ), g ⊗ η) . By continuity, Sϕ(BT A) = (B ⊗ I)Sϕ(T )(A ⊗ I), T ∈ K(H), A, B ∈ D. Let Φ1 : K(H)⊗1 → B(H ⊗K) be the map given by Φ1(T ⊗I) = Sϕ(T ); then Φ1 is a completely bounded D⊗1-bimodule map. Using [13, Exercise 8.6 (ii)], we can find a completely bounded weak* continuous D ⊗ 1-bimodule map Φ2 : B(H ⊗ K) → B(H ⊗ K) extending Φ1. By [10], there exist a bounded row operator A = (Ai)∞ i=1 and a bounded column operator B = (Bi)i∈N, where Ai, Bi ∈ D ¯⊗B(K), i ∈ N, such that Φ2(T ) = ∞ Xi=1 AiT Bi, T ∈ B(H ⊗ K). Using the identification D ¯⊗B(K) ≡ L∞(X, B(K)), we consider Ai (resp. Bi) as a function Ai : X → B(K) (resp. Bi : X → B(K)). The boundedness of A and B now imply that there exists a null set N ⊆ X such that the OPERATOR SYSTEM STRUCTURES 23 series ∞ ∞ Xi=1 Ai(x)Ai(x)∗ and Bi(y)∗Bi(y) Xi=1 are weak* convergent whenever x, y 6∈ N . If (x, y) 6∈ N × N then the series i=1 Ai(x)Bi(y) is weak* convergent. As in the first part of the proof, we (cid:3) conclude that ϕ(x, y) coincides with its sum for almost all (x, y). P∞ An inspection of the proof of Theorem 6.3 shows the following description of inflated Schur multipliers. Remark 6.4. The following are equivalent, for a completely bounded map Φ : K(H) → B(H ⊗ K): (i) Φ(BT A) = (B ⊗ I)Φ(T )(A ⊗ I), for all T ∈ K(H) and all A, B ∈ D; (ii) there exists a Schur multiplier ϕ ∈ S(X, K) such that Φ = Sϕ. Definition 6.5. A Schur multiplier ϕ ∈ S(X, K) will be called positive if the map Sϕ : B(H) → B(H ⊗ K) is positive. For the next theorem, note that, if ϕ ∈ L∞(X × X, B(K)) and α ⊆ X is a subset of finite measure then the function ϕχα×α belongs to L2(X×X, B(K)) and hence the operator Tϕχα×α : H → H ⊗ K is well-defined. Theorem 6.6. The following are equivalent, for a Schur multiplier ϕ ∈ S(X, K): (i) ϕ is positive; (ii) the map Sϕ : B(H) → B(H ⊗ K) is completely positive; (iii) for every subset α ⊆ X of finite measure, the operator Tϕχα×α is positive; (iv) there exist functions Ai ∈ L∞(X, B(K)), i ∈ N, such that the series i=1 Ai(x)Ai(x)∗ converges almost everywhere in the weak* topology, P∞ and < ∞, Ai(x)Ai(x)∗(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) esssup ∞ x∈X (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=1 Xi=1 ∞ ϕ(x, y) = Ai(x)Ai(y)∗, a.e. on X × X. Proof. (i)⇒(iii) Let α ⊆ X be a subset of finite measure. Then χα ∈ H; let χα ⊗ χ∗ α be the corresponding (positive) rank one operator. Then Tϕχα×α = Sϕ(χα ⊗ χ∗ α), and the conclusion follows. (iii)⇒(ii) Let n ∈ N, Xi = X for i = 1, . . . , n, Y = X1 ∪ · · · ∪ Xn and ν be the disjoint sum of n copies of the measure µ. Identify Cn ⊗ H with L2(Y, ν), and define ψ : Y ×Y → B(K) by letting ψ(x, y) = ϕ(x, y) if (x, y) ∈ Xi × Xj = X × X. Note that Sψ = idMn ⊗Sϕ and hence ψ ∈ S(Y, K). Let α ⊆ X have finite measure and J ∈ Mn be the matrix all of whose entries 24 Y.-F. LIN AND I. G. TODOROV are equal to 1. Let αi ⊆ Xi be the set that coincides with α, i = 1, . . . , n, and α = ∪n i=1αi; we have that (15) Tψχ α× α ≡ J ⊗ Tϕχα×α . By assumption, Tϕχα×α is positive; thus, by (15), Tψχ α× α is positive. For g ∈ L∞(Y, ν) ∩ L2(Y, ν) and h ∈ L∞(α), we have Since the set (Sψ(g ⊗ g∗)h, h) =(cid:0)Tψχ α× α(gh), gh(cid:1) ≥ 0. (cid:8)h ∈ L2(Y, ν) : ∃ a set of finite measure α ⊆ X with h ∈ L∞(α)(cid:9) is dense in L2(Y, ν), we have that Sψ(g ⊗ g∗) ∈ B(H ⊗ K)+. By weak* continuity, Sψ(T ) ∈ B(H ⊗ K)+ whenever T ∈ B(L2(Y, ν))+. Thus, Sψ is positive, that is, Sϕ is n-positive. (ii)⇒(i) is trivial. (ii)⇒(iv) follows from the proof of Theorem 6.3 by noting that in the case Sϕ is completely positive, one can choose Bi = A∗ (iv)⇒(i) follows from the proof of Theorem 6.3. i , i ∈ N. (cid:3) 7. Positive extensions In this section, we apply our results on maximal operator system A- structures to questions about positive extensions of inflated Schur multipli- ers. We first recall some measure theoretic background from [2] and [7], required in the sequel. A subset E ⊆ X × X is called marginally null if E ⊆ (M × X) ∪ (X × M ), where M ⊆ X is null. We call two subsets E, F ⊆ X × X marginally equivalent (resp. equivalent), and write E ∼= F (resp. E ∼ F ), if their symmetric difference is marginally null (resp. null with respect to product measure). We say that E is marginally contained in F (and write E ⊆ω F ) if the set difference E \ F is marginally null. A measurable subset κ ⊆ X × X is called • a rectangle if κ = α × β where α, β are measurable subsets of X; • ω-open if it is marginally equivalent to a countable union of rectan- gles, and • ω-closed if its complement κc is ω-open. Recall that, by [22], if E is any collection of ω-open sets then there exists a smallest, up to marginal equivalence, ω-open set ∪ωE, called the ω-union of E, such that every set in E is marginally contained in ∪ωE. Given a measurable set κ, one defines its ω-interior to be intω(κ) =[ ω {R : R is a rectangle with R ⊆ω κ} . The ω-closure clω(κ) of κ is defined to be the complement of intω(κc). For a set κ ⊆ X × X, we write κ = {(x, y) ∈ X × X : (y, x) ∈ κ}. The subset κ ⊆ X × X is said to be generated by rectangles if κ ∼= clω(intω(κ)) [7, 11]. OPERATOR SYSTEM STRUCTURES 25 For any ω-closed subset κ ⊆ X × X, let S2(κ) =(cid:8)Tk : k ∈ L2(κ)(cid:9) , S0(κ) = S2(κ) where L2(κ) is the space of functions in L2(X × X) which are supported on κ, up to a set of zero product measure. Note that the spaces S2(κ), S0(κ) and S(κ) are D-bimodules. We equip them with the operator space structures inherited from B(H). k·k and S(κ) = S2(κ) w∗ , Partially defined scalar-valued Schur multipliers were defined in [11]. Here we extend this notion to the operator-valued setting. Definition 7.1. Let κ ⊆ X × X be a subset generated by rectangles. A function ϕ ∈ L∞(κ, B(K)) will be called a partially defined Schur multiplier if the map Sϕ from S2(κ) into B(H ⊗ K), given by k ∈ L2(κ), Sϕ(Tk) = Tϕk, is completely bounded. Remark 7.2. For Schur multipliers ϕ, ψ ∈ L∞(κ, B(K)), we have that Sϕ = Sψ if and only if ϕ ∼ ψ. Proof. Suppose ϕ, ψ ∈ L∞(κ, B(K)) are such that Sϕ = Sψ. Then Tϕk = Tψk for every k ∈ L2(κ). By Remark 6.1, ϕk ∼ ψk. It now easily follows that ϕ ∼ ψ. The converse implication follows by reversing the previous steps. (cid:3) Let κ ⊆ X × X be a subset generated by rectangles. We note that the map Sϕ from Definition 7.1 is D-bimodular. In addition, if ψ ∈ S(X, K) is given as in Definition 6.2, then its restriction ψκ : κ → B(K) is an inflated Schur multiplier. Proposition 7.3. Let K be a separable Hilbert space, κ ⊆ X × X a subset generated by rectangles and ϕ ∈ L∞(κ, B(K)). The following are equivalent: (i) ϕ is a Schur multiplier; (ii) there exists a Schur multiplier ψ : X × X → B(K) such that ψκ ∼ ϕ; (iii) there exists a unique completely bounded map Φ0 : S0(κ) → B(H ⊗K) such that Φ0(Tk) = Tϕk, for each k ∈ L2(κ); (iv) there exists a unique completely bounded weak* continuous map Φ : S(κ) → B(H ⊗ K) such that Φ(Tk) = Tϕk, for each k ∈ L2(κ). Proof. (i)⇒(ii) Since ϕ is a Schur multiplier, the map Φ2 : S2(κ) → B(H ⊗ K), given by Φ2(Tk) = Tϕk, extends to a completely bounded linear map Φ0 : S0(κ) → B(H ⊗ K). By continuity, Φ0(BT A) = (B ⊗ I)Φ0(T )(A ⊗ I), T ∈ S0(κ), A, B ∈ D. Let Φ : S0(κ) ⊗ 1 → B(H ⊗ K) be the map given by Φ(T ⊗ I) = Φ0(T ), T ∈ S0(κ). By [13, Exercise 8.6 (ii)], there exists a completely bounded D ⊗ 1-bimodule map Φ1 : B(H ⊗ K) → B(H ⊗ K), extending Φ. Let Ψ : K(H) ⊗ 1 → B(H ⊗ 26 Y.-F. LIN AND I. G. TODOROV K) be the restriction of Φ1; then ΨS0(κ)⊗1 = Φ. Let Ψ : K(H) → B(H ⊗ K) be given by Ψ(T ) = Ψ(T ⊗ I). Clearly, Ψ(BT A) = (B ⊗ I)Ψ(T )(A ⊗ I), T ∈ K(H), A, B ∈ D. By Remark 6.4, there exists ψ ∈ S(X, K) such that Ψ = Sψ. For every k ∈ L2(κ) we have Sψ(Tk) = Sϕ(Tk). By Remark 7.2, ψκ ∼ ϕ. (ii)⇒(iv) Take Φ = SψS(κ). The uniqueness of Φ follows from the fact that the Hilbert-Schmidt operators with integral kernels in L2(κ) are weak* dense in S(κ). (iv)⇒(iii)⇒(i) are trivial. (cid:3) If ϕ : κ → B(K) is a Schur multiplier then we will denote still by Sϕ the weak* continuous map defined on S(κ) whose existence was established in Proposition 7.3 (iv). We say that a subset κ ⊆ X × X is symmetric if κ ∼= κ. We call κ a positivity domain [11] if κ is symmetric, generated by rectangles and the diagonal ∆ := {(x, x) : x ∈ X} is marginally contained in κ. The following was established in [11]: Proposition 7.4. If κ ⊆ X×X is generated by rectangles, then the following are equivalent: (i) S(κ) is an operator system; (ii) κ is a positivity domain. Let ϕ : κ → B(K) be a Schur multiplier. We say that the Schur multiplier ψ : X × X → B(K) is a positive extension of ϕ if ψ is positive and ψκ ∼ ϕ. Proposition 7.5. Let κ be a positivity domain and ϕ : κ → B(K) be a Schur multiplier. The following are equivalent: (i) ϕ has a positive extension; (ii) the map Sϕ : S(κ) → B(H ⊗ K) is completely positive. Proof. (i)⇒(ii) Suppose that ψ : X × X → B(K) is a positive extension of ϕ. By Theorem 6.6, Sψ is completely positive. On the other hand, SψS(κ) = Sψκ. Since ψκ = ϕ, we conclude that Sϕ is completely positive. (ii)⇒(i) Let Φ0 be the restriction of Sϕ to S0(κ) + CI; clearly, Φ0 is a completely positive map. By Arveson's Extension Theorem, there exists a completely positive map Ψ0 : K(H) + CI → B(H ⊗ K) extending Φ0. The restriction Ψ of Ψ0 to K(H) is then a completely positive extension of SϕS0(κ). Let Ψ∗∗ be the second dual of Ψ, and E : B(H ⊗ K)∗∗ → B(H ⊗ K) be the canonical projection. We have that the map Ψ = E ◦ Ψ∗∗ : B(H) → B(H ⊗ K) is completely positive and weak* continuous extension of Sϕ. Let Ψ : B(H) ⊗ 1 → B(H ⊗ K) (resp. Φ : S(κ) ⊗ 1 → B(H ⊗ K)) be the map given by Ψ(T ⊗ I) = Ψ(T ) (resp. Φ(T ⊗ I) = Sϕ(T )); then Ψ is a completely positive extension of map Φ. Note that Φ is a D ⊗ 1-bimodule map. By [13, Exercise 7.4], Ψ is a D ⊗ 1-bimodule map. By Remark 6.4, OPERATOR SYSTEM STRUCTURES 27 there exists ψ ∈ S(X, K) such that Ψ = Sψ; the function ψ is the desired positive extension of ϕ. (cid:3) If S is an operator system, we write S ++ for the cone of all positive finite rank operators in S. If T is an operator system, we call a linear map Φ : S → T strictly positive if Φ(S) ∈ T + whenever S ∈ S ++. We call Φ strictly completely positive if Φ(n) is strictly positive for all n ∈ N. A Schur multiplier ϕ : κ → B(K) will be called strictly positive (resp. strictly completely positive) if the map Sϕ : S(κ) → B(H ⊗ K) is strictly positive (resp. strictly completely positive). Lemma 7.6. Let κ be a positivity domain. Every positive finite rank opera- tor in Mn(S(κ)) has the form (Tki,j )n i,j=1, where ki,j ∈ L2(κ), i, j = 1, . . . , n. Proof. Recall that S2(κ) = {Tk : k ∈ L2(κ)} and S0(κ) = S2(κ) It . Suppose that T ∈ Mn(S(κ))++ follows that Mn(S0(κ)) = Mn(S2(κ)) and let T = (Ti,j)n i,j=1, where Ti,j ∈ S(κ), i, j = 1, . . . , n. Since T has finite rank, so does Ti,j; in particular, Ti,j is a Hilbert-Schmidt operator and, by [7, Lemma 6.1], Ti,j ∈ S2(κ). (cid:3) k·k . k·k Recall that the Banach space projective tensor product T (X) = L2(X, µ) ⊗L2(X, µ) can be canonically identified with the predual of B(H) (and the dual of Indeed, each element h ∈ T (X) can be written as a series h = K(H)). 2 < ∞, and the pairing i=1 kfik2 i=1 kgik2 P∞ i=1 fi ⊗ gi, where P∞ is then given by hT, hi = (T fi, gi), T ∈ B(H). ∞ 2 < ∞ and P∞ Xi=1 We have [2] that h can be identified with a complex function on X × X, defined up to a marginally null set, and given by h(x, y) = fi(x)gi(y). ∞ Xi=1 h = P∞ The positive cone T (X)+ consists, by definition, of all functions h ∈ T (X) that give rise to positive functionals on B(H), that is, functions h of the form 2 < ∞. It is well-known that a function ϕ ∈ L∞(X × X) is a Schur multiplier if and only if, for every h ∈ T (X), there exists h′ ∈ T (X) such that ϕh ∼ h′ (see [17]). In particular, if the measure µ is finite then S(X, C) can be naturally identified with a subspace of T (X). i=1 fi ⊗ fi, where P∞ i=1 kfik2 Theorem 7.7. Let κ ⊆ X × X be a positivity domain. The following are equivalent: (i) for every separable Hilbert space K, every strictly positive Schur mul- tiplier ϕ : κ → B(K) is strictly completely positive; 28 Y.-F. LIN AND I. G. TODOROV (ii) for every n ∈ N, every positive finite rank operator in Mn(S(κ)) is the i=1 ⊆ D norm limit of sums of operators of the form (DiSD∗ and S ∈ S(κ)++. j )i,j, where (Di)n Proof. (i)⇒(ii) We first assume that the measure µ is finite. Suppose that there exists n ∈ N and a positive finite rank operator T ∈ Mn(S(κ)) that is not equal to the limit, in the norm topology, of the operators of the form (DiSD∗ i=1 ⊆ D and S ∈ S(κ)++. By Lemma 7.6, j )n T = (Tki,j )n i,j=1, for some ki,j ∈ L2(κ), i, j = 1, . . . , n. By a geometric form of Hahn-Banach's Theorem, there exist a norm continuous functional ω : Mn(S0(κ)) → C and γ < 0 such that i,j=1, where (Di)n (16) ω(T ) < γ and ω(cid:0)(DiSD∗ j )n i,j=1(cid:1) ≥ 0, S ∈ S(κ)++, (Di)n Let ωi,j : S0(κ) → C be the norm continuous functionals such that i=1 ⊆ D. ω((Si,j)n i,j=1) = n Xi,j=1 ωi,j(Si,j), Si,j ∈ S0(κ), i, j = 1, . . . , n. After extending ωi,j to K(H), we may assume that ωi,j ∈ T (X) for i, j = 1, . . . , n. Suppose first that ωi,j ∈ S(X, C), i, j = 1, . . . , n. Identify ω with the function (denoted by the same symbol) ω : X ×X → Mn, given by ω(x, y) = (ωi,j(x, y))n i,j=1. Since Sω : S2(H) → B(H) ⊗ Mn is given by Sω(Tk) = (Sωi,j (Tk)), k ∈ L2(X × X), and the maps Sωi,j are completely bounded, we have that the map Sω is completely bounded, that is, ω ∈ S(X, Mn). We claim that S(n) ω is not strictly positive. Note that S(n) ω (T ) =(cid:0)Sωi,j (Tkp,q )(cid:1)i,j,p,q . Writing e for the vector in H n with all its entries equal to the constant function 1, we have that γ > ω(T ) = ωi,j(x, y)ki,j(x, y)d(µ × µ)(x, y) n Xi,j=1Zκ = (cid:16)(cid:0)Sωi,j (Tki,j )(cid:1)i,j e, e(cid:17) . (17) Suppose that S(n) itive, which contradicts (17). ω (T ) is positive. Then its submatrix (Sωi,j (Tki,j ))i,j is pos- We now show that Sω is strictly positive. Let S ∈ S(κ)++. Using Lemma 7.6, write S = Tk for some k ∈ L2(κ). We have that Sω(S) = (Tωi,j k)n i,j=1. For i = 1, . . . , n, let ξi ∈ L∞(X, µ) and note that, since µ is finite, ξi ∈ H. OPERATOR SYSTEM STRUCTURES 29 Let Di = Mξi, i = 1, . . . , n, and set ξ = (ξi)n i=1. We have that n (Sω(S)ξ, ξ) = (Tωi,j kξj, ξi) ωi,j(x, y)k(x, y)ξj (x)ξi(y)d(µ × µ)(x, y) n = Xi,j=1 Xi,j=1Zκ = ω(cid:0)(D∗ i SDj)n i,j=1(cid:1) ≥ 0. Since L∞(X, µ) is dense in H, we have that Sω(S) ∈ Mn(B(H))+. Now relax the assumption that ωi,j ∈ S(X, C). By standard arguments (see e.g. the proof of [1, Lemma 3.13]), there exist measurable sets Xm ⊆ X with Xm ⊆ Xm+1, m ∈ N, such that µ(X \ Xm) →m→∞ 0 and the restriction ω(m) i,j of ωi,j to Xm × Xm belongs to S(Xm, C) for all m ∈ N. Let ω(m) : X × X → Mn be the function given by ω(m)(x, y) = (ω(m) i,j (x, y))i,j if (x, y) ∈ Xm × Xm and ω(m)(x, y) = 0 otherwise, and note that ω(m) defines a functional on Mn(K(H)) in the natural way (which will be denoted by the same symbol). Let Pm be the projection from H onto L2(Xm). We have that ω(m)(R) = ω((Pm ⊗ In)R(Pm ⊗ In)), R ∈ Mn(K(H)). Since (Pm ⊗ In)R(Pm ⊗ In) →m→∞ R in norm, for every R ∈ Mn(K(H)), we have that (16) eventually holds true for ω(m) in the place of ω. By the previous paragraph, ω(m) is a Schur multiplier for which Sω(m) is strictly positive, but not strictly completely positive. Finally, relax the assumption that µ be finite. Let (Xm)m∈N be an in- creasing sequence of sets of finite measure such that ∪∞ m=1Xm = X, and let Qm be the projection from H onto L2(Xm), m ∈ N. Let T ∈ Mn(S(κ))++. Since T is a positive operator of finite rank, (QmT Qm)m∈N is a sequence of positive finite rank operators, converging to T in norm. By the first part of the proof, QmT Qm is a norm limit of operators of the form (DiSD∗ j )i,j, where (Di)n i=1 ⊆ D and S ∈ S(κ)++. The conclusion follows. (ii)⇒(i) Let ϕ : κ → B(K) be a Schur multiplier such that Sϕ : S(κ) → B(H ⊗ K) is strictly positive. It follows from the assumption and fact that Sϕ is a D-bimodule map that S(n) ϕ (T ) is positive whenever T ∈ Mn(S(κ))++. (cid:3) Definition 7.8. Let κ be a positivity domain. We call κ rich if Mn(S(κ))+ = Mn(S(κ))++ w∗ for every n ∈ N. Suppose that X is a countable set equipped with counting measure. In this case, positivity domains can be identified with undirected graphs with vertex set X in the natural way. This identification will be made in the subsequent remark and in Theorem 7.12. Remark 7.9. Let X be a countable set. Then any graph κ ⊆ X × X is rich. 30 Y.-F. LIN AND I. G. TODOROV Proof. For X = N, write Qm for the projection onto the span of {ei}m i=1, m ∈ N, where {ei}i∈N is the standard basis of ℓ2. If T ∈ Mn(S(κ))+ then ((Qm ⊗ In)T (Qm ⊗ In))m∈N is a sequence in Mn(S2(κ))++, converging in the weak* topology to T . (cid:3) By Proposition 7.5, if a Schur multiplier ϕ : κ → B(K) has a positive extension then the map Sϕ : S(κ) → B(H ⊗ K) is necessarily positive. We call ϕ admissible if Sϕ is a positive map. The main result of this section is a characterisation of when an admissible Schur multiplier has a positive extension, in terms of the maximal operator D-system structure defined in Section 5. Note that S(κ) is a dual AOU D-space in the natural fashion. Theorem 7.10. Let κ ⊆ X × X be a rich positivity domain. The following are equivalent: (i) for every separable Hilbert space K, every admissible Schur multiplier ϕ : κ → B(K) has a positive extension; (ii) S(κ) = OMAXw∗ D (S(κ)). Proof. (i)⇒(ii) Let ϕ : κ → B(K) be a strictly positive Schur multiplier. Since S(κ)+ = S(κ)++ and Sϕ is weak* continuous, Sϕ is positive. By the assumption and Proposition 7.5, Sϕ is completely positive. In particular, Sϕ is strictly completely positive. By Theorem 7.7 and the fact that the matricial cones of any operator system are norm closed, we have that w∗ (18) Mn(S(κ))++ ⊆ Mn(OMAX D(S(κ)))+. Since κ is rich, by taking weak* closures on both sides in (18) we obtain that (19) Mn(S(κ))+ ⊆ Mn(OMAX w∗ D (S(κ)))+. Since the converse inclusion in (19) always holds, we conclude that S(κ) = OMAXw∗ D (S(κ)). (ii)⇒(i) follows from Theorem 5.8 and Proposition 7.5. (cid:3) Theorem 7.10 and Remark 7.9 have the following immediate corollary. In the case where X is finite, it is a reformulation, in terms of operator system structures, of [14, Theorem 4.6]. Corollary 7.11. Let X be a countable set, equipped with counting measure and κ ⊆ X × X be a symmetric set containing the diagonal. The following are equivalent: (i) for every Hilbert space K, every admissible Schur multiplier ϕ : κ → B(K) has a positive extension; D (S(κ)). (ii) S(κ) = OMAXw∗ Let X be a countable set. Recall that a graph κ ⊆ X ×X is called chordal if every 4-cycle in κ has an edge connecting two non-consecutive vertices of the cycle (see e.g. [14]). OPERATOR SYSTEM STRUCTURES 31 Theorem 7.12. Let X be a countable set and κ ⊆ X × X be a chordal graph. Then S(κ) = OMAXw∗ D (S(κ)). Proof. Fix n ∈ N and let [n] = {1, . . . , n}. Suppose that κ ⊆ X × X is a chordal graph. Let κ(n) = {((x, i), (y, j)) ∈ (X × [n]) × (X × [n]) : (x, y) ∈ κ} . Then κ(n) is a chordal graph on X ×[n]. By [11, Theorem 2.5], every positive operator in Mn(S(κ)) is a weak* limit of rank one positive operators in Mn(S(κ)). Suppose that K is a Hilbert space and ϕ : κ → B(K) is a Schur multiplier such that Sϕ : S(κ) → B(H ⊗ K) is a positive map. Let R ∈ Mn(S(κ)) be a positive rank one operator. After identifying Mn(S(κ)) with S(κ(n)), we see that there exists a subset α ⊆ X × [n] such that R is supported on α × α. Let β = {x ∈ X : ∃ i ∈ [n] with (x, i) ∈ α}. Since α×α ⊆ κ(n), we have that β ×β ⊆ κ. Setting β = β ×[n], we have that α ⊆ β, and hence R is supported on β × β. The restriction ψ of ϕ to β × β is a positive Schur multiplier. By Theorem 6.6, the map Sψ : S(β × β) → B(H ⊗ K) is completely positive. Thus, S(n) ψ (R) ∈ B(H ⊗ K)+. Since Sϕ is weak* continuous, the previous paragraph implies that Sϕ is completely positive. By Proposition 7.5, ϕ has a positive extension and, by Corollary 7.11, S(κ) = OMAXw∗ (cid:3) ϕ (R) = S(n) D (S(κ)). References [1] M. Anoussis, A. Katavolos and I. G. Todorov, Ideals of A(G) and bimodules over maximal abelian selfadjoint algebras, J. Funct. Anal. 266 (2014), 6473-6500. [2] W. B. Arveson, Operator algebras and invariant subspaces, Ann. Math. (2) 100 (1974), 433-532. [3] D. P. Blecher and C. Le Merdy, Operator algebras and their modules -- an oper- ator space approach, Oxford University Press, 2004. [4] D. P. Blecher and B. Magajna, Dual operator systems, Bull. London Math. Soc. 43 (2011) 311-320. [5] H. Dym and I. Gohberg, Extensions of band matrices with band inverses, Linear Algebra Appl. 36 (1981), 1-24. [6] E. Effros and Zh.-J. Ruan, Operator spaces, Oxford University Press, 2000. [7] J. A. Erdos, A. Katavolos, and V. S. Shulman, Rank one subspaces of bimodules over maximal abelian selfadjoint algebras, J. Funct. Anal. 157 (1998) no. 2, 554-587. [8] R. Grone, C. R. Johnson, E. M. Sa, H. Wolkowicz, Positive definite completions of partial Hermitian matrices , Linear Algebra Appl. 58 (1984), 109-112. [9] A. Grothendieck, R´esum´e de la th´eorie m´etrique des produits tensoriels topologiques, Boll. Soc. Mat. Sao-Paulo 8 (1956), 1-79. [10] U. Haagerup, Decomposition of completely bounded maps on operator algebras, un- published manuscript. [11] R. Levene, Y.-F. Lin and I. G. Todorov, Positive completions of Schur multipli- ers, J. Operator Theory 78 (2017), no. 1, 45-69. [12] A. McKee, I. G. Todorov and L. Turowska, Herz-Schur multipliers of dynamical systems, Adv. Math. 331 (2018), 387-438. 32 Y.-F. LIN AND I. G. TODOROV [13] V. I. Paulsen, Completely bounded maps and operator algebras, Cambridge Univer- sity Press, 2002. [14] V. I. Paulsen, S. Power and R. R. Smith, Schur products and matrix completions, J. Funcr. Anal. 85 (1989), 151-178. [15] V. I. Pauslen, I. G. Todorov and M. Tomforde, Operator system structures on ordered spaces, Proc. London Math. Soc. 102 (2011), no. 1, 25-49. [16] V. I. Paulsen and M. Tomforde, Vector spaces with an order unit, Indiana Univ. Math. J. 58 (2009), no. 3, 1319-1360. [17] V. V. Peller, Hankel operators in the perturbation theory of unitary and selfadjoint operators, Funktsional. Anal. i Prilozhen. 19 (1985), no. 2, 37-51, 96. [18] G. Pisier, Introduction to operator space theory, Cambridge University Press, 2003. [19] F. Pop, A. M. Sinclair and R. R. Smith, Norming C-algebras by C-subalgebras, J. Funct. Anal. 175 (2000), 168-196. [20] F. Pop and R. R. Smith, Cones arising from C-algebras and complete positivity, Math. Proc. Camb. Phil. Soc. 145 (2008), 121-127. [21] S. Sakai, C-algebras and W*-algebras, Springer, 1971. [22] V. S. Shulman, I. G. Todorov and L. Turowska, Closable multipliers, Integral Eq. Operator Th. 69 (2011), no. 1, 29-62. [23] R. R. Smith, Completely bounded module maps and the Haagerup tensor product, J. Funct. Anal. 102 (1991), 156-175. [24] M. Takesaki, Theory of operator algebras I, Springer, 2003. Mathematical Sciences Research Centre, Queen's University Belfast, Belfast BT7 1NN, United Kingdom E-mail address: [email protected] Mathematical Sciences Research Centre, Queen's University Belfast, Belfast BT7 1NN, United Kingdom, and School of Mathematical Sciences, Nankai Uni- versity, 300071 Tianjin, China E-mail address: [email protected]
1712.00823	2	1712	2018-10-19T16:44:38	Approaching the UCT problem via crossed products of the Razak-Jacelon algebra	[ "math.OA" ]	We show that the UCT problem for separable, nuclear $\mathrm C^$-algebras relies only on whether the UCT holds for crossed products of certain finite cyclic group actions on the Razak-Jacelon algebra. This observation is analogous to and in fact recovers a characterization of the UCT problem in terms of finite group actions on the Cuntz algebra $\mathcal O_2$ established in previous work by the authors. Although based on a similar approach, the new conceptual ingredients in the finite context are the recent advances in the classification of stably projectionless $\mathrm C^$-algebras, as well as a known characterization of the UCT problem in terms of certain tracially AF $\mathrm C^*$-algebras due to Dadarlat.	math.OA	math	APPROACHING THE UCT PROBLEM VIA CROSSED PRODUCTS OF THE RAZAK -- JACELON ALGEBRA SELC¸ UK BARLAK AND G ´ABOR SZAB ´O Abstract. We show that the UCT problem for separable, nuclear C∗- algebras relies only on whether the UCT holds for crossed products of certain finite cyclic group actions on the Razak -- Jacelon algebra. This observation is analogous to and in fact recovers a characterization of the UCT problem in terms of finite group actions on the Cuntz algebra O2 established in previous work by the authors. Although based on a similar approach, the new conceptual ingredients in the finite context are the recent advances in the classification of stably projectionless C∗- algebras, as well as a known characterization of the UCT problem in terms of certain tracially AF C∗-algebras due to Dadarlat. Introduction A separable C∗-algebra A is said to satisfy Rosenberg -- Schochet's universal coefficient theorem (UCT) if for every separable C*-algebra A′, the following sequence is exact Ext(K∗(A), K∗−1(A′)) / KK∗(A, A′) / Hom(K∗(A), K∗(A′)), where the right hand map is the natural one and the left hand map is supposed to be the inverse of a map that is always defined; see [34]. A separable C∗-algebra satisfies the UCT if and only if it is KK-equivalent to a commutative C∗-algebra. Furthermore, the class of separable, nuclear C∗-algebras satisfying the UCT can be characterized as the smallest class of separable, nuclear C∗-algebras that contains C and is closed under countable inductive limits, the two out of three property for extensions, and KK- equivalences; see [5]. Whereas Skandalis [37] has shown that there exist non- examples within the class of separable, exact, non-nuclear C∗-algebras, it is still open whether all separable, nuclear C∗-algebras satisfy the UCT. This arguably constitutes the most important open question about separable, nuclear C∗-algebras and is commonly referred to as the UCT problem. Due to the recent dramatic progress in the structure and classification theory of simple, nuclear C∗-algebras achieved by many hands -- see among others [24, 25, 14, 10, 40] -- the UCT problem is receiving an increasing amount of attention. 2010 Mathematics Subject Classification. Primary 46L35, 46L55. The first author was supported by the Villum Fonden project grant 'Local and global structures of groups and their algebras' (2014-2018). The second author was supported by EPSRC grant EP/N00874X/1, the Danish National Research Foundation through the Centre for Symmetry and Deformation (DNRF92), and the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement 746272. 1 / / / / 2 SELC¸ UK BARLAK AND G ´ABOR SZAB ´O Over the years, different characterizations and reductions of the UCT problem have emerged. Kirchberg proved in [19] (and later [20] in published form) that every separable, nuclear C∗-algebra is KK-equivalent to some unital Kirchberg algebra, thus localizing the problem at the class of unital Kirchberg algebras. In fact, he showed that it is sufficient to only consider those Kirchberg algebras with vanishing K-theory, which by the Kirchberg -- Phillips classification theory [19, 28] then asks whether such C∗-algebras must be isomorphic to the Cuntz algebra O2; see [7, 8]. Furthermore, by using either the work of Spielberg [38] or combining results of Katsura [18] and Yeend [41, 42], one sees that UCT Kirchberg algebras have specific groupoid models. These in turn give rise to Cartan subalgebras in the sense of Renault [30]. Conversely, it was recently recently shown in [1] that any separable, nuclear C∗-algebra with a Cartan subalgebra satisfies the UCT. As a consequence, the UCT problem turns out to be equivalent to the question whether every unital Kirchberg algebra has a Cartan subalgebra. Using Kirchberg's aforementioned insight, the authors reduced the UCT problem in [4] to crossed products of O2 by certain actions of finite cyclic groups. In addition to Kirchberg's previously known reduction theorem, the other two crucial ingredients of the proof were the Kirchberg -- Phillips O2- absorption theorem [21] and the existence of certain model actions of Zp on O2 for any prime number p. The feature of any of these models is that the associated crossed product has, in a sense, the smallest possible non-trivial K-theory, namely that given by the (p − 1)-fold direct sum of Mp∞; see also [15, 16]1. The crossed product viewpoint has recently been taken up in [1, 2] to characterize the UCT problem in terms of the existence of invariant Cartan subalgebras for certain finite order automorphisms of O2. We would also like to remark that in a similar fashion as for finite cyclic groups, a variant of the proof of [5, 23.15.12] shows that the UCT problem can be reduced to crossed products of O2 by certain actions of the circle group T as well. Building on Lin's classification of tracially AF C∗-algebras [22, 23], Dadar- lat achieved in [9] a reduction of the UCT problem to stably finite, nuclear C∗-algebras, which is similar in spirit to Kirchberg's reduction theorem. He showed that the UCT problem has an affirmative answer if and only if the universal UHF algebra is the only separable, unital, simple, nuclear, tracially AF C∗-algebra whose ordered K-theory is isomorphic to (Q, 0). As such C∗- algebras are automatically monotracial, this in particular reduces the UCT problem to the class of separable, unital, simple, nuclear C∗-algebras with a unique tracial state. Due to work of Sato -- White -- Winter [36] and the recent classification of KK-contractible C∗-algebras with finite nuclear dimension [11], one may obtain a monotracial2 analog of the Kirchberg -- Phillips absorp- tion theorem with the Razak -- Jacelon algebra W in place of O2; see [29, 17] and Theorem 2.2. 1In these references, one sees that the possible K-groups for crossed products of O2 by ap- proximately representable Zp-actions are precisely the modules over the ring Z[1/p, e2πi/p]. The additive group of this ring is in turn isomorphic to Z[1/p]⊕p−1 ∼= K0(cid:0)M ⊕p−1 2Although we will not need it, we point out that using more advanced results around the Toms -- Winter conjecture like [6] yields more general tracial versions in the same fashion. p∞ (cid:1). APPROACHING THE UCT PROBLEM VIA CROSSED PRODUCTS 3 In this short note, we reduce the UCT problem to crossed products of W by certain actions of finite cyclic groups. Similarly as in [4], we con- struct a model action of Zp on W for each prime number p such that the crossed product is in Robert's classifiable class [31], and which is moreover monotracial and KK-equivalent to the (p − 1)-fold direct sum of the UHF algebra Mp∞. The action itself arises as the dual action of another action constructed similarly as in [4, Example 4.12], which we will carry out in detail for the reader's convenience. Once the model action is taken care of, our main result (Theorem 2.3) is then proved using the aforementioned W- absorption theorem for monotracial C∗-algebras and Dadarlat's reduction theorem from [9]. Near the end of the paper, we will argue why our previ- ous characterization of the UCT problem in terms of crossed products of O2 can be directly recovered from the main result of this note, thus showing that we achieve a reduction that is a priori stronger. It remains open whether the UCT problem can be further characterized in terms of the existence of certain Cartan subalgebras in W, akin to a similar phenomenon studied in [1, 2] for O2. This direction of research may be the subject of subsequent work. Acknowledgement. Substantial parts of this work were carried out during research visits of the first author to the University of Aberdeen in July 2017 and to the University of Copenhagen in November 2017, respectively. He is grateful to these institutions for their hospitality and support. 1. The model actions Throughout the paper, we will assume familiarity with K-theory and Kasparov's bivariant KK-theory; see [5] for an introduction. For standard references treating the Rokhlin property for finite group actions, see [15, 16, 26, 35, 12]. Definition 1.1. Let G be a finite group and A a separable C∗-algebra. An action α : G y A is said to have the Rokhlin property, if there exists an approximately central sequence en ∈ A of positive contractions satisfying k(en − e2 n)ak + (cid:13)(cid:13)(cid:13) (cid:16)1 − X g∈G αg(en)(cid:17)a(cid:13)(cid:13)(cid:13) n→∞ −→ 0 for all a ∈ A. Remark 1.2. Let A be a finite C∗-algebra, which has at least one tracial state but no unbounded traces. Recall that an automorphism α on A is called strongly outer, if it is outer, and for every α-invariant tracial state τ on A, the induced automorphism of α on the von Neumann algebra πτ (A)′′ is outer. We will call an action α : G y A of a discrete group strongly outer, if αg is a strongly outer automorphism for all g 6= 1G. Actions with the Rokhlin property are particular examples of strongly outer actions. Definition 1.3. Let G be a finite abelian group and A a separable C∗- algebra. An action α : G y A is said to be approximately representable, if there exist sequences of contractions xg,n ∈ Aα in the fixed point algebra, for g ∈ G, such that the following properties hold for all g, h ∈ G and a ∈ A: 4 SELC¸ UK BARLAK AND G ´ABOR SZAB ´O • (x1G,n)n is an approximate unit in A; • ka(xg,nxh,n − xgh,n)k + k(xg,nxh,n − xgh,n)ak g,n − xg−1,n)ak n→∞ • ka(x∗ • kαg(a) − xg,nax∗ g,n − xg−1,n)k + k(x∗ n→∞ −→ 0. g,nk n→∞ −→ 0; −→ 0; Remark 1.4. For what follows, we call Robert's class CR the class of sepa- rable C∗-algebras that are stably isomorphic to inductive limits of 1-NCCW complexes with trivial K1-groups on the level of building blocks. We will make use of Robert's paper [31] where a classification theory for C∗-algebras in CR is developed in terms of the functor Cu∼ defined there. If one restrict to the simple C∗-algebras A in CR, then it is shown in [31, Section 6] that Cu∼(A) is naturally isomorphic to K0(A) ⊔ LAff ∼ +(T0(A)). Let us denote by C0 R the simple, stably projectionless C∗-algebras in Robert's class with trivial pairing between the K0-group and the traces. In more recent work [13] by Gong -- Lin, specific models for simple, stably projectionless C∗-algebras with trivial pairing maps are constructed. In particular, one sees that for every pair (G0, T ) of a torsion-free abelian group G0 and Choquet simplex T , there exists a C∗-algebra A ∈ C0 R with continuous scale and (K0(A), T (A)) ∼= (G0, T ); see [13, Section 6].3 Moreover, it follows from Robert's aforementioned results that for every A ∈ C0 R, and every group endomorphism κ on K0(A), there exists a trace- preserving endomorphism β on A with K0(β) = κ, and β is unique up to approximate unitary equivalence. By the usual Elliott intertwining method [32, Corollary 2.3.4], β may be chosen to be an automorphism if κ is an automorphism. We will use this fact in the proof of the proposition below: The reader should now recall the Razak -- Jacelon algebra W from [29, 17]. Proposition 1.5 (cf. [4, Example 4.12]). Let p ≥ 2 be a prime number. Then there exists a strongly outer, approximately representable action γ : Zp y W such that W ⋊γ Zp ∼KK M ⊕p−1 p∞ . Proof. Using the above Remark 1.4, we choose a C∗-algebra B ∈ C0 unique tracial state, no unbounded traces, and K0(B) ∼= Z[1/p]⊕p−1. R with a Since B satisfies the UCT, it follows that B ∼KK M ⊕p−1 of these C∗-algebras are isomorphic; see [34]. p∞ , as the K-groups As p is prime, we note that the K0-group is in fact isomorphic to the additive group of the ring Z[1/p, ξp], where ξp = e2πi/p. Then there ex- ists an automorphism β on B such that, under this identification, one has K0(β)(x) = ξp · x. Similarly as in [4]4, we will now reconstruct B as a new inductive limit that will allow us to replace β by a Zp-action having the same invariant. Consider Bn = Mpn−1 ⊗ B and βn = idMpn−1 ⊗β for n ≥ 1. 3The continuous scale assumption can be avoided if one replaces the Choquet simplex by a topological cone with a suitable scale. However we do not need this level of generality for our applications. 4The reference [26, Section 5] may be more appropriate for p = 2. APPROACHING THE UCT PROBLEM VIA CROSSED PRODUCTS 5 From now on, we will use the index set Zp for the canonical matrix units in Mp, and identify X ⊕p ∼= X ⊕Zp for any X being either a C∗-algebra or a group. We will also denote βj n for all n, all j ∈ Zp and j ∈ {0, . . . , p − 1} with j = j + pZ. n = βj We then define via Φn : B⊕p n → B⊕p (n+1) (cid:2)Φn(xi)i∈Zp(cid:3)j = diag (cid:0)βi n(xi+j)(cid:1)i∈Zp for j ∈ Zp. Since this is just a combination of diagonal embeddings using compositions of βn, we may describe this inductive limit on the level of K-theory by the following commutative diagram K0(B⊕p n ) K0(Φn) / K0(B⊕p (n+1)) ∼= ∼= Z[1/p, ξp]⊕p ϕ / Z[1/p, ξp]⊕p where ϕ is given by multiplication with the p × p-matrix ξp−2 p ξp−3 p ... ... 1 Ξp = (ξk−l )p−1 l,k=0 = ξ2 p ξp . . . · · · · · · ξp 1 1 p   ξp−1 p ... ... ξ2 p ξp . . . · · · · · · ξp−1 p · · · · · · · · · · · · ξp−1 p ξp−2 p ... ... ξp 1   . As the K-groups are uniquely p-divisible, using the simple relation Ξ2 p = p · Ξp for this matrix yields that the inductive limit of the K-groups is isomorphic to the cokernel of ϕ, which is easily seen to be isomorphic to Z[1/p, ξp] ∼= Z[1/p]⊕p−1. Now the limit is clearly separable, stably projectionless and is in Robert's class CR. As the connecting maps Φn are clearly non-degenerate, full, and trace-collapsing, it follows that the inductive limit B := lim −→ {B⊕p n , Φn} is simple and has a unique tracial state, no unbounded traces and (thus automatically) a trivial pairing map. As its K-theory is isomorphic to that of B, there exists an isomorphism B ∼= B. We will henceforth use this identification without much further mention. This allows us to construct an inductive limit action α : Zp y B ∼= B arising on each building block from the action α(n) : Zp y B⊕p n given by From the definition of the connecting maps, the formula α(n) j (xi)i∈Zp = (xi−j)i∈Zp. Φn ◦ α(n) j = α(n+1) j ◦ Φn / / 6 SELC¸ UK BARLAK AND G ´ABOR SZAB ´O is evident for all n and j ∈ Zp. In particular, α is well-defined. 1 ) = ϕ ◦ σ1 = Ξp · (δi,j+1)i,j∈Zp = ξp · Ξp = (ξp · ) ◦ K0(Φn). Moreover, the standard identification K0(B⊕p n ) ∼= Z[1/p, ξp]⊕p turns the Zp-action K0(α) into the canonical shift σ : Zp y Z[1/p, ξp]⊕p on the level of abelian groups. Using the above diagrams, this allows us to see that K0(Φn) ◦ K0(α(n) In other words, K0(α1)(x) = ξp · x for all x ∈ K0(B) ∼= Z[1/p, ξp]. From the construction of α, it is clear that it has the Rokhlin property. In fact on every building block, one has a canonical equivariant embedding from C(Zp) equipped with the shift into the center of M(B⊕p n ), and so α has the Rokhlin property arising as an equivariant inductive limit of Rokhlin actions. Applying [35, Theorem 3], we deduce that B ⋊α Zp is a C∗-algebra in the class C0 R. It clearly has a unique and bounded trace, and is Morita equivalent to the fixed point algebra Bα, whose K-theory is naturally isomorphic to5 {x ∈ K∗(B) K∗(α1)(x) = ξp · x = x} = 0. In summary, we deduce that B⋊αZp ∼= W. We obtain the action γ : Zp y W as the dual γ = α under this identification, which has the desired property by Takai duality [39]. As α has the Rokhlin property, γ will be approximately representable by [26, Proposition 4.4]; in fact locally B-representable by [4, Theorem 3.4]. It is also strongly outer as the crossed product, which is isomorphic to B, has a unique trace. This finishes the proof. (cid:3) Remark 1.6. We note that most of the above proof of Proposition 1.5 is given in detail only in order for this note to be more self-contained. In fact, right after choosing β ∈ Aut(B) near the beginning of the proof, one can deduce from Remark 1.4 that β defines a Zp-action up to approximate unitary equivalence. Due to classification, one has B ∼= B ⊗ Mp∞ as the K-groups of B are uniquely p-divisible. Applying the much more general existence result [4, Theorem 2.3] would then immediately give the Rokhlin action α : Zp y B with α1≈uβ, which implies K0(α1)(x) = ξp · x and allows one to finish the proof in the same manner as above. 2. Main theorem The following insight due to Dadarlat forms the basis of our main result. See [22, 23] for details surrounding tracially AF C∗-algebras. We note that the equivalence (i)⇔(ii) is actually proved in Dadarlat's work, whereas the equivalence (ii)⇔(iii) follows from the simple fact that any C∗-algebra as in (ii) is automatically monotracial. We will only use the equivalence (i)⇔(iii) in the sequel. Theorem 2.1 (see [9, Theorem 1.2]). The following are equivalent: (i) The UCT holds for every separable, nuclear C∗-algebra; (ii) if A is any separable, unital, simple, nuclear, tracially AF C∗-algebra such that one has an order-isomorphism (K0(A), K1(A)) ∼= (Q, 0), then A is isomorphic to the universal UHF algebra; (iii) the UCT holds for every separable, unital, simple, nuclear C∗-algebra with a unique tracial state. 5See [3, Theorem 4.9]. In the simple unital case, this observation is [15, Theorem 3.13]. APPROACHING THE UCT PROBLEM VIA CROSSED PRODUCTS 7 The following arises as an application of the recent advances in the struc- ture theory of simple C∗-algebras. This should be understood as a mono- tracial variant of the Kirchberg -- Phillips O2-absorption theorem [21]. Theorem 2.2. Let A be a separable, unital, simple, and nuclear C∗-algebra. Then A has a unique tracial state if and only if A ⊗ W ∼= W. Proof. As W is Z-stable, one has A ⊗ W ∼= (A ⊗ Z) ⊗ W, and hence we may assume that A is Z-stable. Let us show the "if" part. We may assume that A has tracial states. For if A is traceless, then it is a Kirchberg algebra -- see [33] -- and thus A ⊗ W ∼= O2 ⊗ K by Kirchberg -- Phillips classification [28] as this tensor product is a KK-contractible, non-unital Kirchberg algebra. If A has more than one trace, then this immediately gives rise to distinct tracial states on A ⊗ W, so it cannot be isomorphic to W, which has a unique trace. Thus A is indeed monotracial. For the "only if" part, we assume that A is monotracial. As it is also Z-stable by assumption, it follows from the main result of [36] that A has finite nuclear dimension. In particular, the tensor product A ⊗ W also has finite nuclear dimension. But then it follows from [11, Theorem 7.5] that A ⊗ W is classifiable. As it is KK-contractible, has a unique tracial state and no unbounded traces, it must be isomorphic to W. (cid:3) We now come to the main result of this note, which is a finite analog of [4, Theorem 4.17]. Theorem 2.3. Let q1, q2 ≥ 2 be two distinct prime numbers. Then the following are equivalent: (i) The UCT holds for all separable, nuclear C∗-algebras; (ii) for p ∈ {q1, q2} and for all strongly outer, approximately representable actions α : Zp y W, the crossed product W ⋊α Zp satisfies the UCT. Proof. The implication (i)⇒(ii) is obviously trivial. (ii)⇒(i): Suppose that the UCT fails for some separable, nuclear C∗- algebra. Then by Theorem 2.1, it fails for a separable, unital, simple, nuclear C∗-algebra A with a unique tracial state. There exists a natural short exact sequence C0(cid:0)R, A ⊗ M(q1q2)∞(cid:1) / A ⊗ Zq∞ 1 ,q∞ 2 / A ⊗ (Mq∞ 1 ⊕ Mq∞ 2 ). Since Zq∞ 1 or A ⊗ Mq∞ 2 . particular, we may choose p ∈ {q1, q2} and assume that A ∼= A ⊗ Mp∞. 2 ∼KK C, the UCT must fail for A ⊗ Mq∞ 1 ,q∞ In Let γ : Zp y W be the action from Proposition 1.5. Then the UCT fails for A ⊗ M ⊕p−1 p∞ ∼KK A ⊗ (W ⋊γ Zp). Theorem 2.2 implies A ⊗ W ∼= W. So if use this identification and set α = idA ⊗γ : Zp y A ⊗ W ∼= W, then we get that W ⋊α Zp ∼= A ⊗ (W ⋊γ Zp) ∼KK A ⊗ M ⊕p−1 p∞ ∼= A⊕p−1 does not satisfy the UCT. Clearly α is strongly outer and approximately representable as γ had these properties. This finishes the proof. (cid:3) Let us now explain how Theorem 2.3 is a priori stronger than our previous characterization of the UCT problem [4] in terms of crossed products on O2. / / / / 8 SELC¸ UK BARLAK AND G ´ABOR SZAB ´O If all crossed products of Remark 2.4. Let p ≥ 2 be a prime number. the form O2 ⋊α Zp satisfy the UCT, then so do all crossed products of the form W ⋊α′ Zp.6 The same is true if we restrict the statement to actions α : Zp y O2 and α′ : Zp y W that are assumed to be outer and/or approximately representable. In particular, Theorem 2.3 in combination with [4, Theorem 3.4] formally implies [4, Theorem 4.17 (1)⇔(3)]. Proof. Let α′ : Zp y W be an action. It is well-known that the unital inclusion C ⊂ O∞ is a KK-equivalence; see [8]. Then α′ ⊗ idO∞ : Zp y W ⊗ O∞ is an action on W ⊗ O∞ ∼= O2 ⊗ K such that its crossed product satisfies the UCT precisely when W ⋊α′ Zp satisfies the UCT. Now let β : Zp y O2 ⊗ K be some arbitrary action. As all non-trivial projections in O2⊗K are equivalent, there exists a unitary z ∈ U (M(O2⊗K)) with Ad(z) ◦ β ◦ (1O2 ⊗ idK) = 1O2 ⊗ idK. We denote zj = zβ(z) · · · βj−1(z) for j ∈ Zp, where j = 0, . . . , p − 1 with j = j + pZ. Note that this yields the formula Ad(zj) ◦ βj ◦ (1O2 ⊗ idK) = 1O2 ⊗ idK for all j ∈ Zp. Let us consider the family {wi,j}i,j∈Zp in U (M(O2 ⊗ K)) given by wi,j = ziβi(zj)z∗ i+j. Then we obtain a cocycle action (β′, w) : Zp y O2 ⊗ K via j = Ad(zj) ◦ βj. By definition, β is exterior equivalent to (β′, w). It is also β′ immediate from the construction that β′ fixes elements in 1O2 ⊗K pointwise, and that the family {wi,j} commutes with 1O2 ⊗ K. Hence (β′, w) restricts to a cocycle action (γ, v) : Zp y O2 = O2 ⊗ 1 = M(O2 ⊗ K) ∩ (1O2 ⊗ K)′ such that (β′, w) = (γ ⊗ idK, v ⊗ 1); see [3, Proposition 1.7]. Using the Packer -- Raeburn 2-cocycle vanishing trick [27, Theorem 3.4], there exists some genuine action δ : Zp y O2 ⊗ Mp ∼= O2 that is cocycle conjugate to (γ ⊗ idMp, v ⊗ 1Mp). To summarize, the action β is equivalently Morita equivalent to δ, and on the level of crossed products we obtain (O2 ⊗ K) ⋊β Zp ∼= (O2 ⊗ K) ⋊β′,w Zp ∼= (O2 ⋊γ,v Zp) ⊗ K ∼= (O2 ⋊γ,v Zp) ⊗ Mp ⊗ K ∼= (O2 ⋊δ Zp) ⊗ K. The claim now follows when we combine all steps so far. The second part of the statement follows as the class of all outer and/or approximately repre- sentable actions is invariant under tensoring with idO∞ and under equivari- ant Morita equivalence; see [3, Propositions 3.14 and 4.23] for details. (cid:3) Remark 2.5. It is unclear whether there is any direct formal converse to Remark 2.4. In any case, it is not hard to see that the assignment α′ 7→ α′⊗idO∞ for outer actions α′ : Zp y W does not remember all the dynamical information. In fact, let us briefly sketch how to obtain two non-cocycle conjugate actions the images of which become cocycle conjugate under this assignment. Consider γ to be the action constructed in Proposition 1.5. Similarly as in the proof of Proposition 1.5, we may consider D to be R with continuous scale, K0(D) ∼= Z[1/p, ξp], and two the C∗-algebra in C0 extremal tracial states τ0 and τ1. By Robert's classification, there exists 6Although we do not need it, the argument works in fact for arbitrary finite groups in place of Zp. APPROACHING THE UCT PROBLEM VIA CROSSED PRODUCTS 9 and τi ◦ β = τ1−i. This allows β ∈ Aut(D) satisfying K0(β) = ξp · one to run the rest of the argument in the completely analogous fashion (or to proceed as suggested in Remark 1.6), and obtain an approximately ∼= D. Clearly γ and δ are representable action δ : Zp y W with W ⋊δ Zp not cocycle conjugate as their crossed products are non-isomorphic. In fact, they have KK-equivalent crossed products, but γ is strongly outer while δ is not. However, the actions γ ⊗ idO∞ and δ ⊗ idO∞ can be identified with ap- proximately representable actions on W ⊗ O∞ ∼= O2 ⊗ K whose crossed products are isomorphic. Since the dual actions act the same way on K- theory, namely via multiplication by ξp on the K0-group Z[1/p, ξp], these actions turn out to be cocycle conjugate; see [15, 16]. We end this note with two questions, which are motivated by analogous known problems about the Cuntz algebra O2. Question 2.6. Is every action of a finite group on the Razak -- Jacelon al- gebra W approximately representable? What about actions of countable discrete groups? Question 2.7 (cf. [1, 2]). Let p ≥ 2 be a prime number and let α : Zp y W be a strongly outer, approximately representable action. Does W ⋊α Zp satisfy the UCT precisely when α fixes a Cartan subalgebra of W globally? References [1] S. Barlak, X. Li: Cartan subalgebras and the UCT problem. Adv. Math. 316 (2017), pp. 748 -- 769. [2] S. Barlak, X. Li: Cartan subalgebras and the UCT problem, II (2017). URL https://arxiv.org/abs/1704.04939. [3] S. Barlak, G. Szab´o: Sequentially split ∗-homomorphisms between C∗-algebras. Int. J. Math. 27 (2016), no. 12. 48 pages. [4] S. Barlak, G. Szab´o: Rokhlin actions of finite groups on UHF-absorbing C∗-algebras. Trans. Amer. Math. Soc. 369 (2017), pp. 833 -- 859. [5] B. Blackadar: K-theory for operator algebras. Second edition. Cambridge University Press (1998). [6] J. Bosa, N. Brown, Y. Sato, A. Tikuisis, S. White, W. Winter: Covering dimension of C∗-algebras and 2-coloured classification. Mem. Amer. Math. Soc., to appear (2016). URL http://arxiv.org/abs/1506.03974. [7] J. Cuntz: Simple C∗-algebras generated by isometries. Comm. Math. Phys. 57 (1977), no. 2, pp. 173 -- 185. [8] J. Cuntz: K-theory for certain C∗-algebras. Ann. of Math. 113 (1981), pp. 181 -- 197. [9] M. Dadarlat: Some remarks on the universal coefficient theorem in KK-theory. In Operator algebras and mathematical physics (Constanta), pp. 65 -- 74 (2003). [10] G. A. Elliott, G. Gong, H. Lin, Z. Niu: On the classification of simple C∗-algebras with finite decomposition rank, II (2015). URL http://arxiv.org/abs/1507.03437. [11] G. A. Elliott, G. Gong, H. Lin, Z. Niu: The classification of simple sepa- rable KK-contractible C∗-algebras with finite nuclear dimension (2017). URL https://arxiv.org/abs/1712.09463. [12] E. Gardella, L. Santiago: Equivariant ∗-homomorphisms, Rokhlin contraints and equivariant UHF-absorption. J. Funct. Anal. 270 (2016), no. 7, pp. 2543 -- 2590. [13] G. Gong, H. Lin: On classification of non-unital simple amenable C∗-algebras, II (2017). URL https://arxiv.org/abs/1702.01073. [14] G. Gong, H. Lin, Z. Niu: Classification of simple amenable Z-stable C∗-algebras (2015). URL http://arxiv.org/abs/1501.00135. 10 SELC¸ UK BARLAK AND G ´ABOR SZAB ´O [15] M. Izumi: Finite group actions on C∗-algebras with the Rohlin property I. Duke Math. J. 122 (2004), no. 2, pp. 233 -- 280. [16] M. Izumi: Finite group actions on C∗-algebras with the Rohlin property II. Adv. Math. 184 (2004), no. 1, pp. 119 -- 160. [17] B. Jacelon: A simple, monotracial, stably projectionless C∗-algebra. J. Lond. Math. Soc. 87 (2013), no. 2, pp. 365 -- 383. [18] T. Katsura: A class of C∗-algebras generalizing both graph algebras and homeomor- phism C∗-algebras IV, pure infiniteness. J. Funct. Anal. 254 (2008), pp. 1161 -- 1187. [19] E. Kirchberg: The Classification of Purely Infinite C∗-Algebras Using Kasparov's Theory (2003). Preprint. [20] E. Kirchberg: Central sequences in C∗-algebras and strongly purely infinite algebras. Operator Algebras: The Abel Symposium 1 (2004), pp. 175 -- 231. [21] E. Kirchberg, N. C. Phillips: Embedding of exact C∗-algebras in the Cuntz algebra O2. J. reine angew. Math. 525 (2000), pp. 17 -- 53. [22] H. Lin: Tracially AF C∗-algebras. Trans. Amer. Math. Soc. 353 (2001), no. 2, pp. 693 -- 722. [23] H. Lin: Classification of simple C∗-algebras of tracial topological rank zero. Duke Math. J. 125 (2004), no. 1, pp. 91 -- 118. [24] H. Matui, Y. Sato: Strict comparison and Z-absorption of nuclear C∗-algebras. Acta Math. 209 (2012), no. 1, pp. 179 -- 196. [25] H. Matui, Y. Sato: Decomposition rank of UHF-absorbing C∗-algebras. Duke Math. J. 163 (2014), no. 14, pp. 2687 -- 2708. [26] N. Nawata: Finite group actions on certain stably projectionless C∗-algebras with the Rohlin property. Trans. Amer. Math. Soc. 368 (2016), no. 1, pp. 471 -- 493. [27] J. A. Packer, I. Raeburn: Twisted crossed products of C∗-algebras. Math. Proc. Cambridge Philos. Soc. 106 (1989), no. 2, pp. 293 -- 311. [28] N. C. Phillips: A classification theorem for nuclear purely infinite simple C∗-algebras. Doc. Math. 5 (2000), pp. 49 -- 114. [29] S. Razak: On the classification of simple stably projectionless C∗-algebras. Can. J. Math. 54 (2002), no. 1, pp. 138 -- 224. [30] J. Renault: Cartan subalgebras in C∗-algebras. Irish Math. Soc. Bulletin 61 (2008), pp. 29 -- 63. [31] L. Robert: Classification of inductive limits of 1-dimensional NCCW complexes. Adv. Math. 231 (2012), no. 5, pp. 2802 -- 2836. [32] M. Rørdam: Classification of Nuclear C∗-Algebras. Encyclopaedia of Mathematical Sciences. Springer (2001). [33] M. Rørdam: The stable and the real rank of Z-absorbing C∗-algebras. Int. J. Math. 15 (2004), pp. 1065 -- 1084. [34] J. Rosenberg, C. Schochet: The Kunneth Theorem and the Universal Coefficient Theorem for Kasparov's generalized K-functor. Duke Math. J. 55 (1987), no. 2, pp. 431 -- 474. [35] L. Santiago: Crossed products by actions of finite groups with the Rokhlin property. Int. J. Math. 26 (2015). 31 pages. [36] Y. Sato, S. White, W. Winter: Nuclear dimension and Z-stability. Invent. Math. 202 (2015), pp. 893 -- 921. [37] G. Skandalis: Une notion de nucl´earit´e en K-th´eorie (d'apr`es J. Cuntz). K-Theory 1 (1988), no. 6, pp. 549 -- 573. [38] J. Spielberg: Graph-based models for Kirchberg algebras. J. Operator Th. 57 (2007), pp. 347 -- 374. [39] H. Takai: On a duality for crossed products by C∗-algebras. J. Funct. Anal. 19 (1975), pp. 25 -- 39. [40] A. Tikuisis, S. White, W. Winter: Quasidiagonality of nuclear C∗-algebras. Ann. Math. 185 (2017), pp. 229 -- 284. [41] T. Yeend: Topological higher-rank graphs and the C∗-algebras of topological 1- graphs. Contemp. Math. 414 (2006), pp. 231 -- 244. [42] T. Yeend: Groupoid models for the C∗-algebras of topological higher-rank graphs. J. Operator Theory 57 (2007), pp. 95 -- 120. APPROACHING THE UCT PROBLEM VIA CROSSED PRODUCTS 11 Europa-Universitat Flensburg, Institut fur mathematische, naturwissen- schaftliche und technische Bildung, Abteilung fur Mathematik und ihre Di- daktik, Auf dem Campus 1b, DE-24943 Flensburg, Germany E-mail address: [email protected] Department of Mathematics, KU Leuven, Celestijnenlaan 200b, box 2400, B-3001 Leuven, Belgium E-mail address: [email protected]
1611.01883	2	1611	2017-02-02T04:05:40	The classification of some generalised Bunce-Deddens algebras	[ "math.OA" ]	We use $K$-theory to prove an isomorphism theorem for a large class of generalised Bunce-Deddens algebras constructed by Kribs and Solel from a directed graph $E$ and a sequence $\omega$ of positive integers. In particular, we compute the torsion-free component of the $K_0$-group for a class of generalised Bunce-Deddens algebras to show that supernatural numbers are a complete invariant for this class.	math.OA	math	THE CLASSIFICATION OF SOME GENERALISED BUNCE -- DEDDENS ALGEBRAS JAMES ROUT Abstract. We use K-theory to prove an isomorphism theorem for a large class of generalised Bunce -- Deddens algebras constructed by Kribs and Solel from a directed graph E and a sequence ω of positive integers. In particular, we compute the torsion- free component of the K0-group for a class of generalised Bunce -- Deddens algebras to show that supernatural numbers are a complete invariant for this class. 1. Introduction In [7] Kribs and Solel introduced a family of direct limit C ∗-algebras constructed from directed graphs E and sequences ω = (nk)∞ k=1 of natural numbers such that nknk+1 for all k ∈ N. They called these C ∗-algebras generalised Bunce -- Deddens algebras. The graph E consisting of a single vertex connected by a single loop-edge generates the classical Bunce -- Deddens algebras. Supernatural numbers have been used to classify UHF algebras ([4, Theorem 1.12]) and the classical Bunce -- Deddens algebras ([1, Theorem 3.7] and [2, Theorem 4]). Kribs showed in [6, Theorem 5.1] that the generalised Bunce -- Deddens algebras corresponding to the graph BN consisting of a single vertex with N loops, are classified by their associated supernatural numbers in the sense that C ∗(BN , ω) ∼= C ∗(BN , ω′) if and only if [ω] = [ω′]. The special case N = 1 is Bunce and Deddens' theorem. Kribs and Solel later showed in [7, Theorem 7.5] that the generalised Bunce -- Deddens algebras corresponding to the simple cycle with j edges, are classified by their associated supernatural numbers; again the special case j = 1 is the original result of Bunce and Deddens. Kribs and Solel asked in [7, Remark 7.7] for what class of graphs E a similar classification theorem could be obtained. Here we prove that such a theorem can be obtained for the class of generalised Bunce -- Deddens algebras corresponding to a given strongly connected finite directed graph E such that 1 is an eigenvalue of the vertex matrix, and the only roots of unity that are eigenvalues are the PE-th roots of unity, where PE is the period of the graph E. In [10, Proposition 3.11] it was shown that if [ω] = [ω′] then C ∗(E, ω) ∼= C ∗(E, ω′) for row-finite directed graphs E with no sinks or sources. The main result of this article (Theorem 6.1) shows that if C ∗(E, ω) ∼= C ∗(E, ω′) then [ω] = [ω′] for strongly connected finite directed graphs E such that 1 is an eigenvalue of At E and such that the only roots of unity that are eigenvalues of At E are the PE-th roots of unity. We prove this by studying the torsion-free component of K0(C ∗(E, ω)); we assume that 1 is an eigenvalue of At E to ensure that this is nontrivial. The Perron -- Frobenius theorem (see [3, Theorem 8.2.1]) Date: September 18, 2018. 2010 Mathematics Subject Classification. 46L35 (primary); 46L80 (secondary). Key words and phrases. graph C ∗-algebra; Bunce -- Deddens algebra; K-theory; classification. This research is supported by an Australian Government Research Training Program (RTP) Scholarship. 1 2 JAMES ROUT says that if 1 is an eigenvalue of At E, then the PE-th roots of unity are also eigenvalues of At E. The hypothesis that these are the only roots of unity that are eigenvalues of At E is nontrivial. The nonnegative inverse eigenvalue problem asks which sets of n complex numbers λ1, . . . , λn occur as the eigenvalues of some n × n nonnegative matrix. Deep results of [5] regarding this problem show that it is possible for any collection of roots of unity to appear as eigenvalues of a nonnegative matrix. If 1 is not an eigenvalue of At E, then K0(C ∗(E, ω)) is purely torsion and another argu- ment (perhaps along the lines of [6, Theorem 5.1]) will be needed. We have not addressed that case in this article. (Ail i=0 about cokernels. We show that the matrix Pnk/l−1 We begin in Section 3 with some calculations for the sums of powers of matrices and E)t, where l := limj→∞ gcd(PE, nj) and gcd(PE, nk) = l, is invertible if the only eigenvalues of At E are the PE-th roots of unity (Lemma 3.2). We recall the equivalence relation ∼l on E0 established in [10, Lemma 4.2] to show that coker(1 − Al E) (Corollary 3.6). i=1 coker(1 − At In Section 4 we compute K1(C ∗(E, ω)) for strongly connected finite directed graphs E such that the only roots of unity that are eigenvalues of At E are the PE-th roots of unity. We show that the torsion-free component is isomorphic to l copies of K1(C ∗(E)) (Theorem 4.1). We do this by showing that ker(1 − AE(n))t ∼= ker(1 − An E)t for n ≥ 1 (Lemma 4.2), and by showing that K1(C ∗(E(nk)) → K1(C ∗(E(nk+1)) induces the identity map on ker(1 − Al E) for all k such that gcd(PE, nk) = l. i=1 ker(1 − At E)t ∼=Ll E)t ∼=Ll In Section 5 we compute the torsion-free component of K0(C ∗(E, ω)) for strongly con- nected finite directed graphs E such that 1 is an eigenvalue of At E and such that the only roots of unity that are eigenvalues of At E are the PE-th roots of unity. We show that this group is isomorphic to l copies of the torsion-free component of K0(C ∗(E)) adjoined the supernatural number [ω] associated to ω (Theorem 5.3). We do this by showing that coker(1 − AE(n)t) ∼= coker(1 − An E)t for n ≥ 1 (Lemma 5.5), and by showing that the map K0(C ∗(E(nk)) → K0(C ∗(E(nk+1)) induces the multiplication by nk+1/nk map on coker(1 − Al E) modulo torsion (Proposition 5.15). Finally, in Section 6 we prove that if C ∗(E, ω) ∼= C ∗(E, ω′), then [ω] = [ω′] for strongly connected finite directed graphs E such that the only roots of unity that are eigenvalues of At E are the PE-th roots of unity (Theorem 6.1). We prove this by recovering the supernatural number [ω] associated to ω from the torsion-free component of K0(C ∗(E, ω)) (Theorem 6.3). E)t ∼=Ll i=1 coker(1 − At 2. Background 2.1. Directed graphs and their C ∗-algebras. We use the convention for graph C ∗- algebras appearing in Raeburn's book [9]. So if E = (E0, E1, r, s) is a directed graph, then a path in E is a word µ = e1 . . . en in E1 such that s(ei) = r(ei+1) for all i, and we write r(µ) = r(e1), s(µ) = s(en), and µ = n. As usual, we denote by E∗ the collection of paths of finite length, and En := {µ ∈ E∗ : µ = n}; we also write E<n := {µ ∈ E∗ : µ < n}. We borrow the convention from the higher-rank graph literature in which we write, for example vE∗ for {µ ∈ E∗ : r(µ) = v}, and vE1w for {e ∈ E1 : r(e) = v and s(e) = w}. The vertex matrix of E is then the E0 × E0 integer matrix with AE(v, w) = vE1w. We say that E is finite if E0 is finite, that E is row-finite if vE1 is finite for all v ∈ E0, and that E has no sources if each vE1 is nonempty. A directed graph is strongly connected if for every pair of vertices v, w ∈ E0, there exists µ ∈ E∗\E0 such that THE CLASSIFICATION OF SOME GENERALISED BUNCE -- DEDDENS ALGEBRAS 3 r(µ) = v and s(µ) = w. The vertex matrix AE is irreducible if and only if the graph E is strongly connected. The period PE of a strongly connected directed graph E is given by PE = gcd{µ : µ ∈ E∗, r(µ) = s(µ)} (see for example [8, Section 6] with k = 1). The group PE Z is then equal to the subgroup generated by {µ : µ ∈ vE∗v} for any vertex v of E, and so is equal to {µ − ν : µ, ν ∈ vE∗v} for any v. If E is finite or row-finite and has no sources, then a Cuntz -- Krieger E-family in a C ∗-algebra A is a pair (s, p), where s = {se : e ∈ E1} ⊆ A is a collection of partial isometries and p = {pv : v ∈ E0} ⊆ A is a set of mutually orthogonal projections such that s∗ e for all v ∈ E0. The graph algebra C ∗(E) is the universal C ∗-algebra generated by a Cuntz -- Krieger ese = ps(e) for all e ∈ E1, and pv =Pe∈vE 1 ses∗ E-family [9, Proposition 1.21]. Theorem 7.1 of [9] says that the K-theory of C ∗(E) is given by K1(C ∗(E)) ∼= ker(1 − At E), and K0(C ∗(E)) ∼= coker(1 − At E). 2.2. Multiplicative sequences and supernatural numbers. A multiplicative sequence is a sequence ω = (nk)∞ k=1 of natural numbers with nknk+1 for all k ∈ N. We say that a multiplicative sequence ω = (nk)∞ j=1, and write ωω′, if for each k ∈ N there exists j(k) ∈ N such that nkmj(k). Define an k=1 : nknk+1 for all k} by ω ∼ ω′ if ωω′ and ωω′. The equivalence relation ∼ on {(nk)∞ supernatural number [ω] associated to ω is the collection [ω] := {ω′ : ωω′ and ω′ω}. k=1 divides a multiplicative sequence ω′ = (mj)∞ 2.3. Generalised Bunce -- Deddens algebras. Let E = (E0, E1, r, s) be a row-finite directed graph with no sources, and fix n ≥ 1. Define sets E(n)0 := E<n and E(n)1 := {(e, µ) : e ∈ E1, µ ∈ s(e)E<n}, and maps sn(e, µ) := µ and rn(e, µ) =(eµ r(e) if µ < n − 1 if µ = n − 1. Then E(n) = (E(n)0, E(n)1, rn, sn) is a row-finite directed graph with no sources. For µ ∈ E∗, we write [µ]n for the unique element of E<n such that µ = [µ]nµ′ for some µ′ with µ′ ∈ nN; we think of [µ]n as the residue of µ modulo n. By Theorem 3.4 and Proposition 3.6 of [10] there exist injective homomorphisms jn,mn : C ∗(E(n)) → C ∗(E(mn)) such that jn,mn(sn,(e,µ)) = Xτ ∈s(e)E<mn,[τ ]n=µ smn,(e,τ ), and jn,mn(pn,ν) = Xτ ∈E<mn,[τ ]n=ν pmn,τ , for n, m ∈ N and e ∈ E1, µ ∈ s(e)E<n and ν ∈ E<n. Kribs and Solel define the generalised Bunce -- Deddens algebra associated to a multi- plicative sequence ω = (nk)∞ k=1 by C ∗(E, ω) := lim −→ (C ∗(E(nk)), jnk,nk+1). 4 JAMES ROUT 3. Applications of Perron-Frobenius theory In this section we analyse the invertibility of the E0 × E0 matrix P(nk/l)−1 E)t, where l = gcd(PE, ω) := limj→∞ gcd(PE, nj) and k is such that gcd(PE, nk) = l. We also show that coker(1 − Al E). These results will be very useful when we compute the K1(C ∗(E, ω)) in Section 4 and the torsion-free component of the K0(C ∗(E, ω)) in Section 5. E)t is isomorphic to l copies of coker(1 − At (Ail i=0 Lemma 3.1. For each n ≥ 1, let Rn be the polynomial over C given by Rn(x) =Pn−1 The roots of Rn are the n-th roots of unity excluding 1. Proof. We have (1 − x)Rn(x) = 1 − xn, so the roots of (1 − x)Rn are the n-th roots of unity. The only root of 1 − x is 1, so every nth root of unity other than 1 is itself a root of Rn. Since the degree of Rn is n − 1, these are all the roots of Rn. (cid:3) Lemma 3.2. Let E be a strongly connected finite directed graph, let ω = (nk)∞ k=1 be a multiplicative sequence, and let l = gcd(PE, ω). Then PE/l and nk/l are coprime for all k such that gcd(PE, nk) = l. Hence, if the only roots of unity that are eigenvalues of At E i=0 xi. are the PE-th roots of unity, then 0 6∈ σ(cid:0)Rnk/l(Al E)t(cid:1) for k such that gcd(PE, nk) = l. Proof. Suppose for contradiction that k ≥ K and that PE/l is not coprime to nk/l. Say p 6= 1 satisfies p(PE/l) and p(nk/l). Then plPE and plnk. This implies that pl ≤ l, which is a contradiction. For the second statement, we have σ((Al E)t) ∩ T = {e(2πij/PE )l : j ∈ N ∪ {0}} = {e2πij/(PE /l) : j ∈ N ∪ {0}}, by the spectral mapping theorem. By Lemma 3.1, the roots of Rnk/l are the nk/l-th roots of unity. Since gcd(PE/l, nk/l) = 1, we have that e2πji/(nk/l) 6∈ σ((Al E)t) for any j ∈ N ∪ {0}. So 0 6∈ σ(Rnk/l(Al (cid:3) Lemma 3.3. Let E be a strongly connected finite directed graph. Then At At Eδv + Im(1 − E)t). Eδv = (1 − At E)δv ∈ Im(1 − At E), so At Eδv + Im(1 − (cid:3) E) = δv + Im(1 − At E) for all v ∈ E0. Proof. Fix v ∈ E0. We have that δv − At At E) = δv + Im(1 − At We now show that coker(1 − Al E). E)t ∼=Ll i=1 coker(1 − At E). By [10, Lemma 4.2] there is an equivalence relation ∼l on E0 such that v ∼l w if and only if λ ∈ lZ for all λ ∈ vE∗w. We enumerate the equivalence classes for ∼l. Fix v ∈ E0, and let Λ0 = [v]. Now iteratively fix e ∈ E1 with r(e) ∈ Λi and let Λi+1 = [s(e)], where addition in the subscript is modulo l. Then Λ0, . . . , Λl−1 is an enumeration of the equivalence classes in E0/ ∼l. Lemma 3.4. Let E be a strongly connected finite directed graph. Let ω = (nk)∞ multiplicative sequence, and let l := gcd(PE, ω). There is an isomorphism k=1 be a Θ : coker(1 − Al E)t → ZΛi/(1 − Al E)tZΛi l−1 Mi=0 satisfying Θ(δv + Im(1 − Al E)t) = (0, . . . , 0, δv + (1 − Al E)tZΛj , 0, . . . , 0), where v ∈ Λj for some 0 ≤ j ≤ l − 1 and δv + (1 − Al E)tZΛj appears in the j-th position. THE CLASSIFICATION OF SOME GENERALISED BUNCE -- DEDDENS ALGEBRAS 5 i=0 Proof. Fix 0 ≤ j ≤ l − 1, and v ∈ Λj. Since E0 = Fl−1 θ : ZE 0 → Ll−1 i=0 Λi, there is an isomorphism ZΛi such that θ(δv) = (0, . . . , 0, δv, 0, . . . , 0), where δv is in the j-th position. Our choice of Λ0, . . . , Λl−1 ensures that (Al E)tδv ∈ ZΛj . Hence θ((1−Al Al Therefore θ descends to an isomorphism Θ : coker(1 − Al satisfying the desired formula. E)tδv) = (0, . . . , 0, (1−Al E)tδv = Pw∈E 0 vElwδw ∈ ZΛj and so (1 − E)tδv, 0, . . . , 0) ∈Ll−1 i=0(1−Al ZΛi/(1 − Al E)tZΛi. E)tZΛi (cid:3) i=0 E)t → Ll−1 Lemma 3.5. Let E be a strongly connected finite directed graph. Let ω = (nk)∞ k=1 be a multiplicative sequence, and let l = gcd(PE, ω). For each 0 ≤ j ≤ l − 1, there is an isomorphism Φj : ZΛj /(1 − Al E)tZΛj → ZE 0/ Im(1 − AE)t satisfying Φj(δv + (1 − Al E)tZΛj ) = δv + (1 − At E)ZE 0 , for some v ∈ Λj. E)t ⊆ Im(1 − At Proof. Fix 0 ≤ j ≤ l − 1. The formula (1 − Al E)t, it follows that that Im(1 − Al the map ZΛj → ZE 0 given by δv 7→ δv for v ∈ Λj, descends to a homomorphism Φj : ZΛj /(1 −Al E), for v ∈ Λj. E)tZΛj → ZE 0/ Im(1 −AE)t satisfying Φj(δv + (1 −Al E)t(cid:1) shows E)tZΛj ) = δv + Im(1 −At E)tZΛj ⊆ Im(1 − Al E). Since (1 − Al E)t = (1 − At E)(cid:0)Pl−1 i=0(Ai We must show that Φj is an isomorphism. To see that Φj is surjective, fix 0 ≤ k ≤ l − 1 and v ∈ Λk. Then (Aj−k E )tδv ∈ ZΛj and δv + Im(1 − At E) = (Aj−k E )tδv + Im(1 − At E )tδv + (1 − Al E) = Φj(cid:0)(Aj−k E)tZΛj(cid:1). That is, a ∈ Im(1 − At To see that Φj is injective, fix a =Pv∈Λj Pw∈Λk bwδw for each 0 ≤ k ≤ l − 1. Since a ∈ ZΛj , we have 0 = aΛk = ((1 − At E)b)Λk = Ebk−1, for all 0 ≤ k ≤ l − 1, k 6= j, where subtraction in the subscript is modulo E)k−jbj for each 0 ≤ k ≤ l − 1, k 6= j, where subtraction in the E)b where b = Pw∈E 0 bwδw. Let bk := bΛk = avδv ∈ ZΛj such that Φj(a+(1−Al E). Say a = (1 − At bk − At l. Therefore bk = (At superscript is modulo l. Hence E)tZΛj ) = 0. a = (1 − At E)b = (1 − At E)(b0 + · · · + bl−1) = (1 − At E)(cid:16) l−1 Xk=0 (Ak E)t(cid:17)bj = (1 − Al E)tbj ∈ (1 − Al E)tZΛj . (cid:3) Corollary 3.6. Let E be a strongly connected finite directed graph. Let ω = (nk)∞ k=1 be a multiplicative sequence, and let l = gcd(PE, ω). There is an isomorphism ρ : coker(1 − Al i=1 coker(1 − At E) satisfying ρ(cid:0)δv + Im(1 − Al E)t(cid:1) =(cid:0)0, . . . , 0, δv + Im(1 − At where v ∈ Λj for some 0 ≤ j ≤ l − 1 and δv + Im(1 − At E), 0, . . . , 0(cid:1), E) appears in the j-th position. E)t →Ll Proof. Define ρ := (cid:0)Ll−1 i=0 Φi(cid:1) ◦ Θ. It follows from Lemma 3.5 and Lemma 3.6 that ρ is (cid:3) an isomorphism that satisfies the desired formula. 6 JAMES ROUT 4. Computing K1(C ∗(E, ω)) In this section we compute K1(C ∗(E, ω)) where E is a strongly connected finite graph E such that the only roots of unity that are eigenvalues of At E are the PE-th roots of unity, and ω is a multiplicative sequence. The main result of this section is the following. Theorem 4.1. Let E be a strongly connected finite graph and suppose that the only roots of unity that are eigenvalues of At k=1 be a multiplicative sequence and let l := gcd(PE, ω). Then E are the PE-th roots of unity. Let ω = (nk)∞ l K1(cid:0)C ∗(E, ω)(cid:1) = Mi=1 ker(1 − At E). To prove Theorem 4.1 we need a series of results. We begin by studying ker(1 − At E(n)) for n ≥ 1. Let {δv : v ∈ E0} be the generators of ZE 0 and let {δµ,n : µ ∈ E<n} be the generators of ZE<n. For 0 ≤ k ≤ n − 1 and a =Pµ∈E<n aµδµ,n ∈ ZE<n, we define ak :=Pµ∈Ek aµδµ,n ∈ ZE<n and aZEk := Pµ∈Ek aµδµ,k ∈ ZEk. For b = Pv∈E 0 bvδv ∈ ZE 0, we define ιn(b) := Pv∈E 0 bvδv,n ∈ ZE<n. Lemma 4.2. Let E be a row-finite directed graph with no sources and let n ≥ 1. There E)t satisfying ψn(a) = aE 0 for a ∈ is an isomorphism ψn : ker(1 − At ker(1 − At E(n)) → ker(1 − An E(n)). E(n)) → ZE 0 by ψn(a) = aZE0 for a ∈ (1 − At E(n)). We check E)t. Let a ∈ ker(1 − At E)t(aZE0 ) = ((1 − An E(n)). Then E(n))ta0)ZE0 = 0. E)t, and hence ψn descends to a homomorphism ker(1 − At E(n)) → Proof. Define ψn : ker(1 − At that ψn(ker(1 − At (1 − An E(n)) ⊆ ker(1 − An E)t(ψn(a)) = (1 − An So ψn(a) ∈ ker(1 − An ker(1 − An E)t which we also label ψn. Define ϕn : ker(1 − An We check ϕn(ker(1 − An E(n))i(ιn(b)) for b ∈ ker(1 − An E)t. E)t. Then (1 − At E(n)). Let b ∈ ker(1 − An i=0 (At E)t) ⊆ ker(1 − At E(n))(ϕn(b)) = (1 − At E)t → ZE<n by ϕn(b) =Pn−1 E(n))(cid:16) Xi=0 E(n))t(b)(cid:1) = 0. = ιn(cid:0)(1 − An E(n))t(ιn(b)) = (1 − An (Ai n−1 E(n))t(ιn(b))(cid:17) So ϕn(b) ∈ ker(1 − At ker(1 − At E(n)) which we also label ϕn. E(n)), and hence ϕn descends to a homomorphism ker(1 − An E)t → We check that ϕn is an inverse for ψn. Let a ∈ ker(1 − At E(n)). Fix k < n. We have 0 = (1 − At E(n))(ak) =(ak − At aEn−1 − At E(n)(ak+1) E(n)(a0) if k 6= n − 1 if k = n − 1. So an−1 = At step yields an−i = (Ai i=0 (Ai ϕn(aZE0 ) =Pn−1 E(n)(a0). Then an−2 = At E(n))t(a0). Repeating this E(n))t(a0) for i < n. Since a0 = ιn(aZE0 ), we have ϕn(ψn(a)) = E(n))ta0 = a. E(n)(an−1) = (A2 THE CLASSIFICATION OF SOME GENERALISED BUNCE -- DEDDENS ALGEBRAS 7 Now, we check that ψn is an inverse for ϕn. Let v ∈ E0 and 0 ≤ i < n. Repeated δv,n ∈ span{δµ,n : µ ∈ En−i}. Thus applications of (5.1) shows that (cid:0)Ai E(n)(cid:1)t (4.1) Now, let b ∈ ker(1 − An E(n)(cid:1)tδv,n(cid:1)(cid:12)(cid:12)ZE0 =(δv (cid:0)(cid:0)Ai E)t. By (4.1), we have 0 if i = 0 otherwise. ψn(ϕn(b)) =(cid:16) n−1 Xi=0 (Ai E(n))t(ιn(b))(cid:17)(cid:12)(cid:12)(cid:12)ZE0 = b. (cid:3) Suppose E is a row-finite directed graph with no sources. Define the skew-product graph E ×1 Z as the graph with edge set (E ×1 Z)1 = E1 × Z and vertex set (E ×1 Z)0 = E0 × Z and range and source maps defined by r(e, k) = (r(e), k − 1) and s(e, k) = (s(e), k). For each n ≥ 1, we denote by sn,((e,µ),k) and pn,(µ,k) the generators of C ∗(E(n) ×1 Z). Proposition 6.7 of [9] gives a natural action βE(n) of Z on C ∗(E(n) ×1 Z) such that (βE(n))m(sn,((e,µ),k)) = sn,((e,µ),k+l). By [9, Lemma 7.10] there is an isomorphism φE(n) of C ∗(E ×1 Z) onto the crossed product C ∗(E(n)) ⋊ T such that φE(n) ◦ (βE(n))m = γn m ◦ φE(n), where γn is the dual of the gauge action γn of C ∗(E(n)). Lemma 4.3. Let E be a row-finite directed graph with no sources, and let n, m ∈ N. There is a homomorphism in,mn : C ∗(E(n) ×1 Z) → C ∗(E(mn) ×1 Z) such that in,mn(sn,((e,µ),1)) = Xτ ∈s(e)E<mn,[τ ]n=µ in,mn(pn,(µ,1)) = Xτ ∈E<mn,[τ ]n=µ pmn,(τ,1), smn,((e,τ ),1) and for all n ≥ 1. Proof. Let (iC∗(E(n)), iT) be the universal covariant representation of (C ∗(E(n)), T, γn). Recall the injective homomorphism jn,mn : C ∗(E(n)) → C ∗(E(mn)). We show that jn,mn is T-equivariant. For e ∈ E1 and µ ∈ s(e)E<n and z ∈ T, we have jn,mn(γn z (sn,(e,µ)) = jn,mn(zsn,(e,µ)) = Xτ ∈E<mn,[τ ]n=µ zsmn,(e,τ ) = γmn z (jn,mn(sn,(e,µ))), and similarly for µ ∈ E<n, jn,mn(γn z (pn,µ)) = (γmn z (jn,mn(pn,µ)). By [12, Corollary 2.48] there is a homomorphism jn,mn×1 : C ∗(E(n))⋊T → C ∗(E(mn))⋊ T satisfying (jn,mn × 1)(iC∗(E(n))(a)iT(z)) = iC∗(E(mn))(jn,mn(a))iT(z) for all a ∈ C ∗(E(n)) and z ∈ T. 8 JAMES ROUT Define in,mn := φ−1 E(mn) ◦ (jn,mn × 1) ◦ φE(n). Let ((e, µ), 1) ∈ E(n)1 ×1 Z and let f1(z) = z for z ∈ T. We calculate E(mn) ◦ (jn,mn × 1) ◦ φE(n)(cid:1)(sn,((e,µ),1)) = φ−1 (cid:0)φ−1 = φ−1 E(n)(cid:0)(jn,mn × 1)(iA(s(n,(e,µ)))iT(f1))(cid:1) E(n)(cid:16)iA(cid:16) Xτ ∈s(e)E<mn,[τ ]n=µ smn,(e,τ )(cid:17)iT(f1)(cid:17) = Xτ ∈s(e)E<mn,[τ ]n=µ smn,((e,τ ),1). Similarly, for (µ, 1) ∈ E(n)0 ×1 Z, we have E(mn) ◦ (jn,mn × id) ◦ φE(n)(cid:1)(pn,(µ,1)) = φ−1 (cid:0)φ−1 = φ−1 E(n)(cid:0)(jn,mn × id)(iA(pn,µ)iT(f1))(cid:1) E(n)(cid:16)iA(cid:16) Xτ ∈E<mn,[τ ]n=µ pmn,τ(cid:17)iT(f1)(cid:17) pmn,(τ,1). (cid:3) = Xτ ∈E<mn,[τ ]n=µ Proposition 4.4. Let E be a row-finite directed graph with no sources and let n, m ∈ N. There are isomorphisms K1(C ∗(E(n))) → ker(1 − An E)t and K1(C ∗(E(mn))) → ker(1 − Amn E )t such that the following diagram commutes. K1(C ∗(E(n))) K1(C ∗(E(mn))) K1(jn,mn) ker(1 − An E)t x 7→ x ker(1 − Amn E )t Proof. The naturality of the Pimsner -- Voiculescu diagram gives the following commutative diagram (see [9, Lemma 7.12]). K1(C ∗(E(n))) K1(jn,mn) K1(C ∗(E(mn))) ker(1 − (βE(n))−1 ∗ ) ker(1 − (βE(mn))−1 ∗ ) By [9, Lemma 7.13] there is an injection σn : ZE<n → K0(C ∗(E(n) ×1 Z)) satisfying σn(δµ,n) = [pn,(µ,1)]0. Define φn,mn : ZE<n → ZE<mn by φn,mn(δµ,n) =Pτ ∈E<mn,[τ ]n=µ δτ,mn for µ ∈ E<n. We claim that the following diagram commutes. K0(C ∗(E(n) ×1 Z)) K0(C ∗(E(mn) ×1 Z)) K0(in,mn) (4.2) σn ZE<n φn,mn σmn ZE<mn THE CLASSIFICATION OF SOME GENERALISED BUNCE -- DEDDENS ALGEBRAS 9 To prove this claim, fix µ ∈ E<n. Then (σ−1 mn ◦ K0(in,mn) ◦ σn)(δµ,n) = (σ−1 = σ−1 mn ◦ K0(in,mn))([p(n,µ,1)]0) mn([in,mn(p(n,µ,1)]0) = σ−1 mn(cid:16) Xτ ∈E<mn,[τ ]n=µ [p(mn,τ,1)]0(cid:17) δτ,mn = Xτ ∈E<mn,[τ ]n=µ = φn,mn(δµ,n). It follows from [9, Theorem 7.16] that σn restricts to an isomorphism of ker(1 − At ∗ ). Restricting diagram 4.2 to the subgroups ker(1 − (βE(n))−1 E(n)) ∗ ) ⊆ E(n)) ⊆ ZE<n yields the following commuting diagram. onto ker(1 − (βE(n))−1 K0(C ∗(E(n)) ×1 Z)) and ker(1 − At ker(1 − (βE(n))−1 ∗ ) ker(1 − (βE(mn))−1 ∗ ) K0(in,mn) ker(1 − At E(n)) φn,mn ker(1 − At E(mn)) Now, we claim that the following diagram commutes. ker(1 − At E(n)) φn,mn ker(1 − At E(mn)) ψn ψmn ker(1 − An E)t x 7→ x ker(1 − Amn E )t To prove this claim, fix x ∈ ker(1 − At E(n)). Then ψmn(φn,mn(x)) = ψmn(cid:16) Xµ∈E<n xµ Xτ ∈E<mn,[τ ]n=µ δτ,mn(cid:17) = Xv∈E 0 xvδv = ψn(x). Combining the preceding commutative diagrams gives the desired commutative dia- (cid:3) gram. Proof of Theorem 4.1. By [11, Theorem 6.3.2], we have By Proposition 4.4, we have −→(cid:0)K1(C ∗(E(nk)), K1(jnk,nk+1)(cid:1). K1(cid:0)C ∗(E, ω)(cid:1) ∼= lim K1(C ∗(E(nk)), K1(jnk,nk+1)(cid:1) ∼= lim (Ajl −→(cid:0) ker(1 − Ank E )t, x 7→ x(cid:1). E)t is invertible for k such that gcd(PE, nk) = −→ (cid:0) lim j=0 ker(1 − Ank gcd(PE, ω). So By Lemma 3.2 the matrix Pnk/l−1 E )t = ker(cid:16)(cid:16) Xj=0 E )t, x 7→ x(cid:1) ∼= ker(1 − Al (cid:0) ker(1 − Ank E)t(cid:17)(1 − Al for k such that gcd(PE, nk) = gcd(PE, ω). Hence (Ajl nk/l−1 E)t. E)t(cid:17) = ker(1 − Al E)t, 10 JAMES ROUT Combining the previous three isomorphisms gives an isomorphism E)t ∼=Ll Now, ker(1 − Al Corollary 3.6. So ker(1 − Al K1(C ∗(E, ω)) ∼= ker(1 − Al E)t ∼= Zr, where r = rank coker(1 − Al E)t. E)t = l · rank coker(1 − At i=1 ker(1 − Al E)t, giving the result. E) by (cid:3) 5. Computing the torsion-free component of K0(C ∗(E, ω)) In this section we calculate the torsion-free component of K0(C ∗(E, ω)). We will use this group in Section 6 to recover the supernatural number [ω] associated to ω. In order to state the main theorem of this section, we need the following lemma. Lemma 5.1. Let A be a free abelian group and let ω = (nk)∞ quence. Define an equivalence relation ∼ on A × N, by (a, j) ∼ (a′, j′) if k=1 be a multiplicative se- max{nj, nj′} nj a = max{nj, nj′} nj′ a′, and define Ah 1 ωi := {(a, j) : a ∈ A, j ∈ N} / ∼ . ω(cid:3) is a torsion-free abelian group under the operation Then A(cid:2) 1 if j′ ≥ j if j ≥ j′. [(a + (nj/nj′) · a′, j)] [(a − a, i)] = [(0, i)], so [(−a, i)] is an inverse for [(a, i)]. If k · [(a, i)] = [(0, i)], then [(k · a, i)] = [(0, i)], so k · a = 0 forcing a = 0 since Proof. Closure, associativity, and commutativity follow easily since A is abelian. Let 0 be the identity element of A. Then [(0, i)] + [(a, i)] = [(0 + a, i)] = [(a, i)] so [(0, i)] is [(a, j)] + [(a′, j′)] =([((nj′/nj) · a + a′, j′)] Moreover, rank A(cid:2) 1 ω(cid:3) = rank A. an identity for A(cid:2) 1 ω(cid:3). Fix a ∈ A and let −a be the inverse. Then [(a, i)] + [(−a, i)] = A is free abelian. To see that rank A(cid:2) 1 ω(cid:3) = rank A, let {aα} be a maximal linearly independent subset of A. Suppose [(0, i)] =Pα cα · [(aα, i)] for cα ∈ N with all but finitely many nonzero. Then [(0, i)] = Pα[(cα · aα, i)] = [(Pα cα · aα, i)], so 0 = Pα cα · aα, independent subgroup of A(cid:2) 1 ω(cid:3) . To see that it is maximal, take c ∈ N and b ∈ A. Then Pα cα[(aα, i)] + c[(b, i)] = [(Pα cα · aα + c · b, i)] = [(0, i)], by the maximality of {aα}. (cid:3) ωi ∼= AN Zh 1 Remark 5.2. We have Ah 1 and since {ai} is linearly independent, cα = 0 for all α. Hence {[(aα, i)]} is a linearly ωi via the map [a, j] → aN 1 nj the elements [(a, j)] as formal fractions and write a/nj for [(a, j)]. . We will regard We now state the main theorem of this section about the torsion-free component of K0(C ∗(E, ω)). Recall that the torsion subgroup of an abelian group A consists of the nonzero elements of A which have finite order. Theorem 5.3. Let E be a strongly connected finite directed graph. Let PE denote the period of E, and let l = gcd(PE, ω). Suppose 1 is an eigenvalue of At E and that the only E are the PE-th roots of unity. Let ω = (nk)∞ roots of unity that are eigenvalues of At k=1 THE CLASSIFICATION OF SOME GENERALISED BUNCE -- DEDDENS ALGEBRAS 11 be a multiplicative sequence. Let torE denote the torsion subgroup of K0(C ∗(E)), and tor(E,ω) the torsion subgroup of K0(C ∗(E, ω)). There is an isomorphism Ψ : K0(C ∗(E, ω))/ tor(E,ω) → l Mi=1 (cid:0)K0(C ∗(E))/ torE(cid:1)h 1 ωi satisfying Ψ([1C∗(E,ω)]0 + torE,ω) = ([1C∗(E)]0 + torE, . . . , [1C∗(E)]0 + torE). To prove Theorem 5.3 we need a series of lemmas. We begin by studying K0(C ∗(E(n)) ∼= coker(1 − At E(n)) for n ≥ 1. Lemma 5.4. Let E be a row-finite directed graph with no sources and let n ≥ 1. Then (5.1) Moreover, δµ,n − δs(µ),n ∈ Im(1 − At Proof. Let µ ∈ En \ E0. We calculate At E(n)δµ,n =(δµ2...µµ,n if µ ∈ E<n\E0 if µ ∈ E0. Pλ∈µEn δλ2...λn,n E(n)) for each µ ∈ E<n. At E(n)δµ,n = Xν∈E<n = At E(n)(ν, µ)δν,n = Xν∈E<n X ν∈E<n,e∈E 1r(ν),[eν]n=µ δν,n = δµ2...µµ,n. µE(n)1νδν,n Let µ ∈ E0. Then At E(n)δµ,n = Xν∈E<n At E(n)(ν, µ)δν,n = Xν∈E<n µE(n)1νδν,n = Xλ∈µEn δλ2...λn,n. The final statement clearly holds when µ ∈ E0, so let µ ∈ E<n\E0. Repeated applica- E(n))tδµ,n ∈ (cid:3) tions of the first case of (5.1) give (Aµ Im(1 − At E(n))tδµ,n = δs(µ),n, so δµ,n −δs(µ),n = (1−Aµ E(n)). Lemma 5.5. Let E be a row-finite directed graph with no sources and let n ≥ 1. There is an isomorphism ψn : coker(1 − An E)t) = δv,n + Im(1 − At E(n)) for v ∈ E0. E(n)) satisfying ψn(δv + Im(1 − An E)t → coker(1 − At Proof. Define a map ψn : ZE 0 → coker(1 − At show that ψn(Im(1 − An (5.1) give (An E)t) ⊆ Im(1 − At E(n)) by ψn(δv) = δv,n + Im(1 − At E(n)). We E(n)). Let v ∈ E0. Repeated applications of E)t(w, v)δw = E)tδv(cid:1), so ψn(cid:0)(An ψn(cid:0)(1 − An E(n))tδv,n = Pλ∈vEn δs(λ),n = Pw∈E 0 vEnwδw,n = Pw∈E 0(An E)tδv(cid:1) = (1 − An E(n))tδv,n + Im(1 − An E(n))t = Im(1 − An E)t → coker(1 − At Thus ψn descends to a homomorphism coker(1 − An label by ψn, satisfying ψn(δv + Im(1 − An E)t) = δv,n + Im(1 − At E(n)) for v ∈ E0. E(n))t ⊆ Im(1 − At E(n)). E(n)), which we also 12 JAMES ROUT Define a map ϕn : ZE<n → coker(1 − An show that ϕn(Im(1 − At µ ∈ E<n\E0, then E(n)) = Im(1 − An E)t by ϕn(δµ,n) = δs(µ) + Im(1 − An E(n))δµ,n ∈ Im(1 − At E)t. Take (1 − At E)t. We E(n)). If ϕn((1 − At E(n))δµ,n) = ϕn(δµ,n − δµ2...µµ,n) = δs(µ) − δs(µ) + Im(1 − An E)t = Im(1 − An E)t, by the first case of (5.1). If µ ∈ E0, then applying the second case of (5.1) at the first equality, we have ϕn(cid:0)δv,n − At E(n)δv,n(cid:1) = ϕn(cid:0)δv,n − Xλ∈vEn δλ2...λn,n(cid:1) δs(λ) + Im(1 − An E)t vEnwδw + Im(1 − An E)t = δv − Xλ∈vEn = δv − Xw∈E 0 = δv − Xw∈E 0 = (1 − An = Im(1 − An E)t. (An E)t(w, v)δw + Im(1 − An E)t E)tδv + Im(1 − An E)t Thus ϕn descends to a homomorphism coker(1 − At label ϕn. E(n)) → coker(1 − An E)t, which we also To show that ψn is an isomorphism, we show that ψn and ϕn are mutually inverse. Let µ ∈ E<n. Then ψn(ϕn(δµ,n + Im(1 − At E(n)))) = ϕn(δs(µ) + Im(1 − An E)t) = δs(µ),n + Im(1 − At E(n)) = δµ,n + Im(1 − At E(n)) by Lemma 5.4, so ψn ◦ ϕn is the identity on coker(1 − At E(n)). Now, let v ∈ E0. Then ϕn(ψn(δv + Im(1 − An E)t)) = ϕn(δv,n + Im(1 − At E(n))) = δv + Im(1 − An E)t, so ϕn ◦ ψn is the identity on coker(1 − An E)t. (cid:3) Remark 5.6. For each n ≥ 1, let σn : coker(1−At E(n)) → K0(C ∗(E(n))) be the isomorphism of [9, Theorem 7.16]. Looking into the proof of [9, Theorem 7.1] shows that this isomor- phism is given by σn(δµ,n + Im(1 − At E(n))) = [pµ,n]0 for µ ∈ E<n. So σn ◦ ψn : coker(1 − An for v ∈ E0. By Lemma 5.4, we have [pµ,n]0 − [pv,n]0 = σn(δµ,n − δv,n + Im(1 − At E)t → K0(C ∗(E(n))) is an isomorphism satisfying (σn ◦ ψn)(cid:0)δv + Im(1 − An for any µ ∈ E<nv. So (σn ◦ ψn)(cid:0)δv + Im(1 − An E)t(cid:1) = [pµ,n]0 for any µ ∈ E<nv. Lemma 5.7. Let E be a row-finite directed graph with no sources and let n, m ∈ N. The following diagram commutes. E)t(cid:1) = [pv,n]0 E(n))) = 0 THE CLASSIFICATION OF SOME GENERALISED BUNCE -- DEDDENS ALGEBRAS 13 ZE 0 Pm−1 i=0 (Ain E )t ZE 0 coker(1 − An E)t coker(1 − Amn E )t σn ◦ ψn σmn ◦ ψmn K0(C ∗(E(n))) K0(C ∗(E(mn))) K0(jn,mn) Proof. Let ηn := σn ◦ ψn, and fix v ∈ E0. Then K0(jn,mn)(ηn)(δv + Im(1 − An E)t)) = K0(jn,mn)([pv,n]0) Now, by Remark 5.6, we have = Xµ∈vE<mn,µ∈nN [pµ,mn]0 = m−1 Xi=0 Xµ∈vEin [pµ,mn]0. (ηmn)(cid:16) m−1 Xi=0 (Ain E )tδv + Im(1 − Amn E )t(cid:17) = ηmn(cid:16) m−1 m−1 Xi=0 Xµ∈vEin δs(µ) + Im(1 − Amn E )t(cid:17) = = m−1 Xi=0 Xµ∈vEin Xi=0 Xµ∈vEin (ηmn)(δs(µ) + Im(1 − Amn E )t) [ps(µ),mn]0 = m−1 Xi=0 Xµ∈vEin [pµ,mn]0. (cid:3) Corollary 5.8. Let E be a row-finite directed graph with no sources and let n, m ∈ N. There exists a homomorphism φn,mn : coker(1 − An E )t satisfying φn,mn(δv + Im(1 − An E)t → coker(1 − Amn E )t for v ∈ E0. E)t) =Pµ∈vE<mn,µ∈nN δs(µ),mn + Im(1 − Amn Proof. Let ηn := σn ◦ ψn, and define φn,mn : coker(1 − An φn,mn := η−1 mn ◦ K0(jn,mn) ◦ ηn. Let v ∈ E0. By Remark 5.6, we have φn,mn(δv + Im(1 − An E)t → coker(1 − Amn E )t by E)t) = (η−1 = (η−1 mn ◦ K0(jn,mn) ◦ ηn)(δv + Im(1 − An mn ◦ K0(jn,mn))([ps(µ),n]0) E)t) [pµ,mn]0(cid:17) [ps(µ),mn]0(cid:17) = η−1 = η−1 mn(cid:16) Xµ∈vE<mn,µ∈nN mn(cid:16) Xµ∈vE<mn,µ∈nN = Xµ∈vE<mn,µ∈nN δs(µ),mn + Im(1 − Amn E )t. (cid:3) We now look at direct limits of quotients of abelian groups by their torsion subgroups. We seek to apply the following result to the sequence (coker(1 − Ank E )t, φnk,nk+1)∞ k=1. Lemma 5.9. Let (Gk, φk,k+1) be a directed system of abelian groups. Let tork := tor(Gk) Gk). For each k there exists a homomorphism for each k ∈ N, and tor∞ := tor(lim −→ 14 JAMES ROUT φk,k+1 : Gk/ tork → Gk+1/ tork+1 such that φk,k+1(g+tork) = φk,k+1(g)+tork+1. Moreover, there is an isomorphism q∞ : lim −→ (Gk, φk,k+1)/ tor∞ → lim −→ (Gk/ tork, φk,k+1) such that q∞(φk,∞(g) + tor∞) = φk,∞(g + tork). Proof. Write Qk := Gk/ tork. For each k ∈ N, let qk : Gk → Qk be the quotient map. Let r ∈ tork. Then there exists n ≥ 1 such that nr = 0, and then nφk,k+1(r) = φk,k+1(nr) = 0. So φk,k+1(tork) ⊆ tork+1, and hence qk+1 ◦ φk,k+1 descends to a homomorphism φk,k+1 : Qk → Qk+1 such that φk,k+1(g + tork) = φk,k+1(g) + tork+1 for all g ∈ Gk. So ( φk+1,∞ ◦ qk+1) ◦ φk,k+1 = φk+1,∞ ◦ φk,k+1 ◦ qk = φk,∞ ◦ qk for all k ∈ N. Therefore the universal property of lim −→ q∞ : lim −→ (Gk, φk,k+1) → lim −→ (Qk, φk,k+1) satisfing q∞ ◦ φk,∞ = φk,∞ ◦ qk. (Gk, φk) gives a homomorphism We show that q∞ descends to a homomorphism satisfying the desired formula. Let p ∈ tor∞. Then there exists r ∈ Gk and n ≥ 1 such that 0 = np = nφk,∞(r) = φk,∞(nr). By [11, Proposition 6.2.5(ii)] we have ker φk,∞ =Sm≥0 ker φk,k+m, so there exists m ≥ 0 such that 0 = φk,k+m(nr) = nφk,k+m(r), giving φk,k+m(r) ∈ tork+m. Therefore q∞(p) = q∞(φk,∞(r)) = q∞(φk+m,∞(φk,k+m(r))) = φk+m,∞(qk+m(φk,k+m(r))) = 0. So q∞(tor∞) ⊆ {0}, and hence q∞ descends to a homomorphism q∞ : lim (Gk, φk,k+1)/ tor∞ → −→ lim −→ (Qk, φk,k+1) satisfying q∞(φk,∞(g) + tor∞) = q∞(φk,∞(g)) = φk,∞(qk(g)) = φk,∞(g + tork) for all g ∈ Gk. It remains to show that q∞ is an isomorphism. We do this by finding an inverse. As in the first paragraph, we find that φk,∞(tork) ⊆ tor∞ since φk,∞ is a homomorphism. Therefore φk,∞ descends to a homomorphism ψk,∞ : Qk → lim (Gk, φk,k+1)/ tor∞ satisfying −→ ψk,∞(g + tork) = φk,∞(g) + tor∞ for each g ∈ Gk. We have ψk+1,∞( φk,k+1(g + tork)) = ψk+1,∞(φk,k+1(g) + tork+1) = φk+1,∞(φk,k+1(g)) + tor∞ = φk,∞(g) + tor∞ = ψk,∞(g + tork). So ψk+1,∞ ◦ φk,k+1 = ψk,∞, and hence the universal property of lim −→ momorphism ψ : φk,∞(g) + tor∞ for all g ∈ Gk. (Qk, φk) gives a ho- (Gk, φk,k+1)/ tor∞ satisfying ψ( φk,∞(g + tork)) = (Qk, φk) → lim −→ lim −→ We check that ψ is an inverse for q∞. Let g ∈ Gk. Then q∞(ψ( φk,∞(g + tork))) = q∞(φk,∞(g) + tor∞) = q∞(φk,∞(g)) = φk,∞(qk(g)) = φk,∞(g + tork). We also have ψ(q∞(φk,∞(g) + tor∞)) = ψ(q∞(φk,∞(g)) = ψ( φk,∞(qk(g))) = φk,∞(g) + tor∞ . So q∞ ◦ ψ is the identity on φk,∞(Qk) and ψ ◦ q∞ is the identity on φk,∞(Gk)/ tor∞, and hence by continuity, ψ and q∞ are mutually inverse. (cid:3) THE CLASSIFICATION OF SOME GENERALISED BUNCE -- DEDDENS ALGEBRAS 15 For n ≥ 1, the torsion subgroup of coker(1 − An E)t is {a + Im(1 − An E)t : a ∈ ZE 0 , ma ∈ Im(1 − An E)t for some m ∈ N}. Define Tn := {a ∈ ZE 0 : ma ∈ Im(1 − An n (torn) where qn : ZE 0 → coker(1 − An E)t for some m ∈ N}. E)t is the quotient map. So Tn = q−1 Proposition 5.10. Let E be a strongly connected finite directed graph. Suppose 1 is an eigenvalue of At E are the PE-th roots of unity. Let ω = (nk)∞ k=1 be a multiplicative sequence and let l := gcd(PE, ω). Then Tnk = Tl for all k such that gcd(PE, nk) = l. E and that the only roots of unity that are eigenvalues of At Proof. Fix k ≥ K. Let C :=Pnk/l−1 i=0 (1 − Ank E )t = (1 − Al (Ail E)t. We have nk/l−1 E)t(cid:16) Xi=0 (Ail E)t(cid:17) = (1 − Al E)tC. (5.2) So Im(1 − Ank mx ∈ Im(1 − Ank E )t ⊆ Im(1 − Al E)t. Now take x ∈ Tnk. Then there exists m ∈ N such that E )t ⊆ Im(1 − Al E)t. Hence Tnk ⊆ Tl. For the reverse inclusion, take x ∈ Tl. Then mx = (1 − Al E)ty, for some m ∈ N and y ∈ ZE 0. Equation (5.2) gives (m det C)x = (det C)(1 − Al E)ty = (1 − Al E)tC(det C)C −1y = (1 − Ank E )t(det C)C −1y ∈ Im(1 − Ank E )t. By Lemma 3.2 det C 6= 0, so Tl ⊆ Tnk. Remark 5.11. If we could compute det C, we could compute det(1 − Ank E )t. Then (when 1 is not an eigenvalue), we could calculate K0(C ∗(E, nk)) and try to use Kribs' argument for [6, Theorem 5.1] to prove a classification result for the generalised Bunce -- Deddens algebras constructed from a finite strongly connected graph whose vertex matrix does not have eigenvalue 1. (cid:3) E and that the only roots of unity that are eigenvalues of At Lemma 5.12. Let E be a strongly connected finite directed graph. Suppose 1 is an eigen- value of At E are the PE-th roots of unity. Let ω = (nk)∞ k=1 be a multiplicative sequence, and let l := gcd(PE, ω). For each k such that gcd(PE, nk) = l, there is an isomorphism τ : coker(1 − Ank E )t/ tornk → ZE 0/Tl satisfying (5.3) τ (a + Im(1 − Ank E )t + tornk) = a + Tl . for a ∈ ZE 0 Proof. To see that the formula (5.3) is well-defined, suppose (a + Im(1 − Ank (b + Im(1 − Ank where t ∈ tornk, that is, t = c + Im(1 − Ank Im(1 − Ank E )t ⊆ Im(1 − Al a map τ satisfying (5.3). E )t) + tornk = E )t + t, E )t for some c ∈ Tnk. Then a − b − c ∈ E)t ⊆ Tl. By Proposition 5.10, c ∈ Tl, so a − b ∈ Tl. So there is E )t) + tornk, where a, b ∈ ZE 0. Then a + Im(1 − Ank E )t = b + Im(1 − Ank The map τ is clearly a surjective group homomorphism. To see that it is injective, suppose a + Tl = b + Tl for a, b ∈ ZE 0. We have a = b + c, for some c ∈ Tl, and hence a + Im(1 − Ank E )t + E )t = b + c + Im(1 − Ank E )t. So a + Im(1 − Ank E )t = b + Im(1 − Ank 16 JAMES ROUT E )t + tornk = (cid:3) E and that the only roots of unity that are eigenvalues of At E )t. By Proposition 5.10, c ∈ Tnk . Therefore a + Im(1 − Ank E )t + tornk. c + Im(1 − Ank b + Im(1 − Ank Corollary 5.13. Let E be a strongly connected finite directed graph. Suppose that 1 is an eigenvalue of At E are the PE-th roots of unity. Let ω = (nk)∞ k=1 be a multiplicative sequence, let l := gcd(PE, ω). For each k such that gcd(PE, nk) = l, there is an isomorphism θnk : coker(1 − Ank E )t/ tornk → E)t) + torl coker(1 − Al for a ∈ ZE 0 . Proof. Fix k such that gcd(PE, nk) = l. The previous Lemma gives an isomorphism coker(1 − Ank E )t) + tornk 7→ a + Tl, where a ∈ ZE 0. The result follows since ZE 0/Tl is isomorphic to coker(1 −Al E)t/ torl via a+ Tl 7→ a + Im(1 − Al (cid:3) E)t/ torl given by θnk(cid:0)(a + Im(1 − Ank E )t) + tornk(cid:1) = (a + Im(1 − Al E)t + torl. We take θnk to be the composition of these isomorphisms. E )t/ tornk → ZE 0/Tl satisfying (a + Im(1 − Ank We give another description of the torsion-free abelian group A(cid:2) 1 Lemma 5.14. Let A be a free abelian group and let ω = (nk)∞ k=1 be a multiplicative sequence, and let mk := nk+1/nk for all k ∈ N. Define maps Mk : A → A by Mk(a) = mk · a, and let Mk,∞ be the natural map A → lim (A, Mk). There is an isomorphism −→ φ : lim −→ ω(cid:3) satisfying φ(Mk,∞(a)) = a/nk, for each k ∈ N and a ∈ A. ω(cid:3) of Lemma 5.1. Proof. Fix k ∈ N. Define jk,∞ : A → A(cid:2) 1 a homomorphism by definition of the operation on A(cid:2) 1 ω(cid:3) by jk,∞(a) = a/nk for a ∈ Z. This jk,∞ is ω(cid:3). We calculate jk+1,∞(Mk(a)) = (mk · a)/nk+1 = (nk+1/nk) · (a/nk+1) = a/nk = jk,∞(a). So the universal property of lim (A, Mk) induces a homomorphism φ satisfying the desired formula. It remains to check −→ that φ is an isomorphism. To see that φ is injective, fix a ∈ A such that φ(Mk,∞(a)) = (A, Mk) ∼= A(cid:2) 1 0. Then a/nk = 0, so a = 0. To see that φ is surjective, fix a/nk ∈ Ah 1 φ(Mk,∞(a)) = a/nk. Proposition 5.15. Let E be a strongly connected finite directed graph. Suppose that 1 is an eigenvalue of At E are the PE- th roots of unity. Let ω = (nk)∞ k=1 be a mulitiplicative sequence, and let l := gcd(PE, ω). Fix K such that gcd(PE, nK) = l, and define ω′ := (n′ k = nK+k−1 for k ≥ 2. For each k ≥ 1, the map φn′ such that the following diagram commutes. E and that the only roots of unity that are eigenvalues of At ωi. Then descends to a map φn′ k=1 where n′ 1 = l and n′ k)∞ k,n′ k,n′ (cid:3) k+1 k+1 coker(1 − An′ k E )t/ torn′ θn′ k k φn′ k ,n′ k+1 coker(1 − A k+1 n′ E )t/ torn′ k+1 θn′ k+1 coker(1 − Al E)t/ torl M ′ k coker(1 − Al E)t/ torl Proof. Fix k ≥ 1. Applying the first assertion of Lemma 5.9 we see that φn′ scends to a homomorphism φn′ satisfying φn′ : coker(1 − An′ (g) + torn′ → coker(1 − A E )t/ torn′ (g + torn′ n′ E k+1 k k k+1 k,n′ )t/ torn′ de- , k+1 k,n′ k+1 k+1 k,n′ ) = φn′ E − 1)t. Note that Bk + m′ k,n′ k+1 k+1 . k k Define Bk := Pm′ k−1 E − 1)t = (An′ that (Ain′ Lemma 5.10. Thus Im Bk + Im(1 − Al E − 1)t(Pi−1 (Ain′ i=0 k k j=0(Ajn′ k−1 E )t), so Im Bk ⊆ Im(1 − An′ E)t ⊆ torl. k1 = Pm′ i=0 k (Ain′ k E )t. We have = Tl by k E )t ⊆ Tn′ k THE CLASSIFICATION OF SOME GENERALISED BUNCE -- DEDDENS ALGEBRAS 17 Fix x ∈ ZE 0. By the preceding paragraph, we have θn′ k+1(cid:0) φn′ k,n′ k+1(cid:0)x + Im(1 − An′ k E )t + torn′ m′ k−1 k(cid:1)(cid:1) = θn′ k+1(cid:0)φn′ Xi=0 (Ain′ = k,n′ k+1(cid:0)x + Im(1 − An′ E )t(cid:1) + torn′ k k+1(cid:1) k E )tx + Im(1 − Al E)t + torl k1)(x) + Im(1 − Al E)t + torl = (Bk + m′ = (m′ = M ′ k1)(x) + Im(1 − Al k(x + Im(1 − Al E)t + torl E)t + torl) = (M ′ k ◦ θn′ k )(x + Im(1 − An′ k E )t + torn′ k ). (cid:3) Recall the isomorphism ρ : coker(1 − Al i=1 coker(1 − At E) of Lemma 3.6 satis- fying E)t →Ll ρ(cid:0)δv + Im(1 − Al E)t(cid:1) =(cid:0)0, . . . , δv + Im(1 − At E), . . . , 0(cid:1), where v ∈ Λj for some 0 ≤ j ≤ l − 1, and δv + Im(1 − At E) appears in the j-th position. Lemma 5.16. Let E be a strongly connected finite directed graph. Suppose that 1 is an eigenvalue of At E are the PE-th roots of unity. Let ω = (nk)∞ k=1 be a multiplicative sequence, and let l = gcd(PE, ω). There i=1 K0(C ∗(E)) such that the following diagram E and that the only roots of unity that are eigenvalues of At is an isomorphism ψ : K0(C ∗(E(l))) →Ll commutes. coker(1 − Al E)t σl ◦ ψl K0(C ∗(E(l))) ρ ψ i=1 coker(1 − At E) Ll i=1 σ1 i=1 K0(C ∗(E)) Ll Ll Moreover, ψ(cid:16)Pµ∈E<l[ps(µ),l]0(cid:17) = ([1C∗(E)]0, . . . , [1C∗(E)]0). Proof. We define ψ := (Ll so is ψ. Using Lemma 3.3 at the second equality, we have i=1 σ1)◦ρ◦(σl◦ψl)−1. Since ρ, σl◦ψl, and σ1 are all isomorphisms, We now show that ψ satisfies the second statement. Fix 0 ≤ i ≤ l − 1, and v ∈ Λi. l−1 Xj=0 ρ(cid:16) E)t(cid:17) =(cid:16)(Al−i (Aj E)tδv + Im(1 − Al E )tδv + Im(1 − At E), . . . , (Al−i−1 E )tδv + Im(1 − At E)(cid:17) =(cid:16)δv + Im(1 − At E), . . . , δv + Im(1 − At E)(cid:17). 18 Hence, ψ(cid:16) Xµ∈E<l JAMES ROUT [ps(µ),l]0(cid:17) E)t(cid:17)(cid:17) δs(µ) + Im(1 − Al l l σ1(cid:17) ◦ ρ ◦ (σl ◦ ψl)−1(cid:17)(cid:16) Xµ∈E<l [ps(µ),l]0(cid:17) =(cid:16)(cid:16) Mi=1 =(cid:16)(cid:16) σ1(cid:17) ◦ ρ(cid:17)(cid:16) Xµ∈E<l Mi=1 =(cid:16)(cid:16) σ1(cid:17) ◦ ρ(cid:17)(cid:16) Xv∈E 0 Xj=0 Mi=1 =(cid:16) σ1(cid:17)(cid:16) Xv∈E 0 Mi=1 δv + Im(1 − At (Aj l−1 l l = ([1C∗(E)]0, . . . , [1C∗(E)]0). E)tδv + Im(1 − Al E)t(cid:17) E), . . . , Xv∈E 0 δv + Im(1 − At E)(cid:17) (cid:3) Proof of Theorem 5.3. Fix K such that gcd(PE, nK) = gcd(PE, ω), and let ω′ = (n′ where n′ unital isomorphism C ∗(E, ω) ∼= C ∗(E, ω′) by [10, Proposition 3.11]. Hence k)∞ k=1 k. Since [ω] = [ω′], we have a k = nK+k−1 for k ≥ 2. Let m′ 1 = l and n′ k+1/n′ k = n′ So it suffices to prove the theorem for ω′. (cid:0)K0(C ∗(E, ω)), [1C∗(E,ω)](cid:1) ∼=(cid:0)K0(C ∗(E, ω′)), [1C∗(E,ω′)](cid:1). −→(cid:0)K0(C ∗(E(n′ k))), K0(jn′ )(cid:1)(cid:1). By [11, Theorem 6.3.2] there is k,n′ k+1 Let torω′ := tor(cid:0) lim an isomorphism K0(C ∗(E, ω′)) ∼= lim k))), K0(jn′ k,n′ k+1 )(cid:17) −→(cid:16)K0(C ∗(E(n′ 1,∞)(cid:16) Xµ∈E<n′ 1 [pµ,n′ 1]0(cid:17). [1C∗(E,ω′)] 7→ K0(jn′ This isomorphism descends to an isomorphism satisfying satisfying k))), K0(jn′ k,n′ k+1 )(cid:17)/ torω′ K0(C ∗(E, ω′))/ tor(E,ω′) [1C∗(E,ω′)]0 + tor(E,ω′) 7→ K0(jn′ ∼= lim −→(cid:16)K0(C ∗(E(n′ 1,∞)(cid:16) Xµ∈E<n′ 1 δs(µ) ∈ ZE 0, and let tor∞ := tor(cid:0) lim ◦ ψn′ k k 1 [pµ,n′ 1]0(cid:17) + torω′ . −→(cid:0) coker(1 − An′ E )t, φn′ )−1 discussed in Remark 5.6 induce an isomorphism k k,n′ k+1(cid:1)(cid:1). k))), K0(jn′ k,n′ k+1 (coker(1 − An′ k E )t, φn′ k,n′ k+1 )/ tor∞ [pµ,n′ 1]0(cid:17) + torω′ −→ )(cid:1)/ torω′ ∼= lim 1,∞(cid:16)(σn′ 1,∞(x + Im(1 − An′ 1 ◦ ψn′ 7→ φn′ = φn′ 1)−1(cid:16) Xµ∈E<n′ 1 [pµ,n′ 1]0(cid:17)(cid:17) + tor∞ 1 E )t) + tor∞ . The isomorphisms (σn′ lim Let x := Pµ∈E<n′ −→(cid:0)K0(C ∗(E(n′ 1,∞)(cid:16) Xµ∈E<n′ K0(jn′ 1 satisfying THE CLASSIFICATION OF SOME GENERALISED BUNCE -- DEDDENS ALGEBRAS 19 By Lemma 5.9 there is an isomorphism lim −→ (coker(1 − An′ ∼= lim E )t, φn′ k,n′ −→ 1,∞(x + Im(1 − An′ E )t) + tor∞ 7→ φn′ By Proposition 5.15 there is an isomorphism )/ tor∞ satisfying φn′ k+1 k 1 (coker(1 − An′ 1,∞(x + Im(1 − An′ k E )t/ torn′ k , φn′ k,n′ k+1 ) 1 E )t + torn′ 1). lim −→ satisfying φn′ (coker(1 − An′ 1,∞(x + Im(1 − An′ k E )t/ torn′ k , φn′ k,n′ k+1 1 E )t + torn′ 1) 7→ Mn′ (coker(1 − Al ) ∼= lim −→ 1,∞(x + Im(1 − Al E)t/ torl, Mn′ ) k E)t + torl). By Lemma 5.14 there is an isomorphism (coker(1 − Al lim −→ 1,∞(x + Im(1 − Al E)t/ torl, Mn′ k ) ∼=(cid:0) coker(1 − Al E)t/ torl(cid:1)h 1 ω′i E)t + torl)/n′ 1. E)t + torl) 7→ (x + Im(1 − Al satisfying mn′ The isomorphism ηl := σl ◦ ψl : coker(1 − Al to an isomorphism ηl : coker(1 − Al isomorphism E)t → K0(C ∗(E(l))) of Remark 5.6 descends E)t/ torl → K0(C ∗(E(l)))/ torE(l). This ηl induces an (cid:0) coker(1 − Al (cid:0)x + Im(1 − Al E)t/ torl(cid:1)h 1 E)t + torl(cid:1)/n′ ω′i ∼=(cid:0)K0(C ∗(E(l)))/ torE(l)(cid:1)h 1 ω′i, 1 7→ ηl(x + Im(1 − Al E)t + torl)/n′ 1 =(cid:16) Xµ∈E<l [ps(µ),l]0 + torE(l)(cid:17)/n′ 1. satisfying Ll satisfying The isomorphism of Lemma 5.16 descends to an isomorphism K0(C ∗(E(l)))/ torE(l) → i=1 K0(C ∗(E))/ torE, and this induces an isomorphism l (cid:0)K0(C ∗(E(l)))/ torE(l)(cid:1)h 1 [ps(µ),l]0 + torE(l)(cid:17)/n′ K0(C ∗(E))/ torE(cid:17)h 1 ω′i ∼=(cid:16) ω′i, Mi=1 1 7→ ψ(cid:16) Xµ∈E<l [ps(µ),l]0 + torE(l)(cid:17)/n′ 1 (cid:16) Xµ∈E<l = ([1C∗(E)]0 + torE, . . . , [1C∗(E)]0 + torE)/l, since n′ 1 = l. Composing the isomorphisms of the previous seven paragraphs gives an isomorphism Ψ : K0(C ∗(E, ω′))/ tor(E,ω′) → l Mi=1 (cid:0)K0(C ∗(E))/ torE(cid:1)h 1 ω′i satisfying Ψ([1C∗(E,ω′)]) = ([1C∗(E)]0 + torE, . . . , [1C∗(E)]0 + torE)/l. (cid:3) Remark 5.17. In the proof of Theorem 5.3, we needed to apply Lemma 5.16 to relate the torsion-free component of K0(C ∗(E(l))) back to the torsion-free component of K0(C ∗(E)). This uses Corollary 3.6, which requires Lemma 3.5, where it is crucial that the power of At E)t matches the number of equivalence classes for the equivalence relation ∼l. We also needed to apply Corollary 5.13 to obtain an isomorphism between the E in the term (1 − Al 20 JAMES ROUT torsion-free component of K0(C ∗(E(l))) and the torsion-free component of K0(C ∗(E(nk))) for all k such that gcd(PE, nk) = l. This uses Lemma 5.10 which depends on Lemma 3.2 explaining why we require that the only roots of unity that are eigenvalues of At E are the PE-th roots of unity. 6. Classification of C ∗(E, ω) In this section we use Theorem 5.3 to prove the following isomorphism theorem. Theorem 6.1. Fix a strongly connected finite directed graph E. Let ω = (nk)∞ ω′ = (n′ the only roots of unity that are eigenvalues of At C ∗(E, ω) ∼= C ∗(E, ω′) if and only if [ω] = [ω′]. k=1 and E and that E are the PE-th roots of unity. Then k=1 be multiplicative sequences. Suppose 1 is an eigenvalue of At k)∞ To prove this theorem we need some preliminary results. Lemma 6.2. Let D ⊆ N. Suppose D = m for some 1 ≤ m ≤ ∞, enumerate D in increasing order, (d1, d2, . . . , dm), and define a nondcreasing sequence lcm(D) by lcm(D) := (d1, lcm(d1, d2), lcm(d1, d2, d3), . . . lcm(d1, d2, . . . , dm), lcm(d1, d2, . . . , dm), . . . ). Then lcm(D) is a multiplicative sequence such that dk lcm(D) for all 1 ≤ k ≤ m. More- over, if ω = (nk)∞ k=1 is another multiplicative sequence such that dkω for all 1 ≤ k ≤ m, then [lcm(D)] divides [ω]. Proof. Clearly lcm(D)k lcm(D)k+1 for each k ≥ 1. It is also clear that, for each 1 ≤ k ≤ m, dk lcm(D)l for all l ≥ k, and so dk lcm(D). For the final statement, fix ω such that dkω for each 1 ≤ k ≤ m. For each 1 ≤ k ≤ m, there exist natural numbers l1, . . . , lk such that d1nl1, . . . , dknlk. Let l(k) = max{l1, . . . , lk}. Then dinl(k) for each 1 ≤ i ≤ k, so lcm(d1, . . . , dk)nl(k). (cid:3) If A is a free abelian group, a ∈ A and n ≥ 1, we write na if there exists a′ ∈ A such that na′ = a. Theorem 6.3. Fix a strongly connected finite directed graph E, and a generalised Bunce -- Deddens algebra C ∗(E, ω). Suppose that the only roots of unity that are eigenvalues of At E are the PE-th roots of unity. Set and let D := {n ≥ 1 : n(cid:0)[1C∗(E,ω)]0 + tor(E,ω)(cid:1) ∈ K0(C ∗(E, ω))/ tor(E,ω)} d := lcm{n ≥ 1 : n(cid:0)[1C∗(E)]0 + torE(cid:1) ∈ K0(C ∗(E))/ torE}. Then [ω] = [l · lcm(D)]/d. Proof. There is an isomorphism θ : K0(C ∗(E))/ torE → ZN , where N = rank K0(C ∗(E)). Let (u1, . . . , uN ) := θ([1C∗(E)]0 + torE) ∈ ZN . We claim that gcd(u1, . . . , uN ) = d. Let e1, . . . , eN be the generators of ZN , and let n ≥ i=1 uiθ−1(ei) = θ−1(u1, . . . , uN ) = 1 such that nui for each 1 ≤ i ≤ N. Then n divides PN [1C∗(E)]0 + torE. So nd, and hence gcd(u1, . . . , uN )d. Now, fix n ≥ 1 such that n([1C∗(E)]0+torE). Then there exists a ∈ K0(C ∗(E)) such that na+torE = [1C∗(E)]0+torE. We then have that nθ(a+torE) = (u1, . . . , uN ). So n is a com- mon divisor of u1, . . . , uN , and hence n gcd(u1, . . . , uN ). So gcd(u1, . . . , uN ) is a common multiple of {n ≥ 1 : n(cid:0)[1C∗(E)]0 + torE(cid:1) ∈ K0(C ∗(E))/ torE}, giving d gcd(u1, . . . , uN ), and so gcd(u1, . . . , uN ) = d. THE CLASSIFICATION OF SOME GENERALISED BUNCE -- DEDDENS ALGEBRAS 21 Next we claim that for n ≥ 1, we have n lcm(D) if and only if n ∈ D. If n ∈ D, it is clear that n lcm(D). For the other direction, suppose n lcm(D). Then there is an i ≥ 1 such that n lcm(d1, . . . , di). Since d1, . . . , di ∈ D, we have that lcm(d1, . . . , di) divides [1C∗(E,ω)]0 + tor(E,ω), and so n ∈ D. We now show that [lcm(D)] divides [dω/l]. Fix n ≥ 1. Then n ∈ D ⇐⇒ n(cid:0)[1C∗(E,ω)]0 + tor(E,ω)(cid:1) ∈ K0(C ∗(E, ω))/ tor(E,ω) l Mi=1 (cid:16)K0(C ∗(E))/ torE(cid:17)h 1 ωi ⇐⇒ n(cid:0)[1C∗(E))]0 + torE, . . . , [1C∗(E)]0 + torE(cid:1)/l ∈ ⇐⇒ n(cid:0)[1C∗(E)]0 + torE(cid:1)/l ∈(cid:16)K0(C ∗(E))/ torE(cid:17)h 1 ωi N Zh 1 ωi Mi=1 ⇐⇒ n(u1, . . . , uN )/l ∈ ⇐⇒ n(d/l) ∈ Zh 1 ωi ⇐⇒ n1 ∈ Zh (dω)/li ⇐⇒ n(dω/l). 1 Hence ndω for all n ∈ D, and so [lcm(D)] divides [dω/l] by Lemma 6.2. To see that [dω/l] divides [lcm(D)], fix k ≥ 1. We have that nk1 ∈ Z(cid:2) 1 (dnk/l)(d/l) ∈ Z(cid:2) 1 ω(cid:3), so ω(cid:3). The above string of implications gives us (dnk/l) lcm(D) for each k ≥ 1, so [dω/l] divides [lcm(D)], and the result follows. (cid:3) We now prove Theorem 6.1. Proof of Theorem 6.1. Suppose that [ω] = [ω′]. Then C ∗(E, ω) ∼= C ∗(E, ω′) by [10, Proposition 3.11]. Now suppose that C ∗(E, ω) ∼= C ∗(E, ω′). Let l = gcd(PE, ω) and l′ = gcd(PE, ω). Since C ∗(E, ω) ∼= C ∗(E, ω′), the number of summands in Theorem 5.3 must be equal, so l = l′. Let d be as in Theorem 6.3. Let and let D := {n ≥ 1 : n(cid:0)[1C∗(E,ω)]0 + tor(E,ω)(cid:1) ∈ K0(C ∗(E, ω))/ tor(E,ω)} D′ := {n ≥ 1 : n(cid:0)[1C∗(E,ω′)]0 + tor(E,ω′)(cid:1) ∈ K0(C ∗(E, ω′))/ tor(E,ω′)}. Fix n ≥ 1. Since C ∗(E, ω) ∼= C ∗(E, ω′), we have that n divides [1C∗(E,ω)]0 + tor(E,ω) precisely when n divides [1C∗(E,ω′)]0 + tor(E,ω′), so D = D′. By Theorem 6.3 we have that [ω] = [l · lcm(D)]/d = [l · lcm(D′)]/d = [ω′]. (cid:3) Remark 6.4. Theorem 6.1 says that for a given graph E and [ω] 6= [ω′], we have C ∗(E, ω) 6= C ∗(E, ω′). One might ask whether this can be extended to say that given graphs E and F and given [ω] 6= [ω′], we must have C ∗(E, ω) 6= C ∗(F, ω′). The following example demonstrates that the answer is no. Let C1 be the graph consisting of a single vertex connected by a single loop and let C3 be the graph with three vertices connected by a single cycle. Let ω = (3, 6, 12, 24, . . . ) and let ω′ = (1, 2, 4, 8, 16, . . . ). Note that ω = 3 ω′. 22 JAMES ROUT Since C1(3) = C3, we have that C ∗(C1, ω) ∼= C ∗(C3, ω′). This illustrates why Theorem 6.1 applies only to generalised Bunce -- Deddens algebras constructed from the same graph. 7. Acknowledgments The results in this article are from my PhD thesis. Thanks to my PhD supervisors Aidan Sims and Dave Robertson for their guidance and support during my PhD and during the writing of this article. It has been great learning from such generous and talented mathematicians. Thanks to Gunar Restorff for pointing out an error in Lemma 3.4. Thanks to Mike Boyle for bringing [5] to my attention and for a helpful email conversation about the spectra of nonnegative integer matrices. Thanks to Toke Meier Carlsen for helpful conversations. References [1] J. Bunce and J. Deddens, C ∗-algebras generated by weighted shifts, Indiana Univ. Math. J. 23 (1973), 257 -- 271. [2] J. Bunce and J. Deddens, A family of simple C ∗-algebras related to weighted shift operators, J. Func. Anal. 19 (1975), 13 -- 24. [3] F.R. Gantmacher, Matrix Theory vol. 2, Amer. Math. Soc., Providence, 2000. [4] J. Glimm, On a certain class of operator algebras, Trans. Amer. Math. Soc. 95 (1960), 318 -- 340. [5] K.H. Kim, N.S. Ormes, and F.W. Roush, The spectra of nonnegative integer matrices via formal power series, J. Amer. Math. Soc. 13 (2000), 773 -- 806. [6] D.W. Kribs, Inductive limit algebras from periodic weighted shifts on Fock space, New York J. Math. 8 (2002), 145 -- 159. [7] D.W. Kribs and B. Solel, A class of limit algebras associated with directed graphs, J. Australian Math. Soc. 82 (2007), 345 -- 368. [8] M. Laca, N.S. Larsen, S. Neshveyev, A. Sims and S.B.G. Webster, Von Neumann algebras of strongly connected higher-rank graphs, Math. Ann. 363 (2015), 657 -- 678. [9] I. Raeburn, Graph algebras, Published for the Conference Board of the Mathematical Sciences, Washington, DC, 2005, vi+113. [10] D. Robertson, J. Rout and A. Sims, KMS states on generalised Bunce -- Deddens algebras and their Toeplitz extensions, Bul. Malaysian Math. Sci. Soc., 2015, 1 -- 35. [11] M. Rørdam, F. Larsen, N.J. Laustsen, An introduction to K-theory for C ∗-algebras, London Math. Society Student Texts, vol. 49, Cambridge University Press, 2000. [12] D.P. Williams, Crossed products of C ∗-algebras, American Math. Society, Providence, RI, 2007, xvi+528. School of Mathematics and Applied Statistics, University of Wollongong, Wollon- gong NSW 2522, Australia E-mail address: [email protected]
1505.07755	4	1505	2016-03-07T18:14:02	Liberation theory for noncommutative homogeneous spaces	[ "math.OA", "math.QA" ]	We discuss the liberation question, in the homogeneous space setting. Our first series of results concerns the axiomatization and classification of the families of compact quantum groups $G=(G_N)$ which are "uniform", in a suitable sense. We study then the quotient spaces of type $X=(G_M\times G_N)/(G_L\times G_{M-L}\times G_{N-L})$, and the liberation operation for them, with a number of algebraic and probabilistic results.	math.OA	math	LIBERATION THEORY FOR NONCOMMUTATIVE HOMOGENEOUS SPACES TEODOR BANICA Abstract. We discuss the liberation question, in the homogeneous space setting. Our first series of results concerns the axiomatization and classification of the families of compact quantum groups G = (GN ) which are "uniform", in a suitable sense. We study then the quotient spaces of type X = (GM × GN )/(GL × GM −L × GN −L), and the liberation operation for them, with a number of algebraic and probabilistic results. Introduction The notion of noncommutative space goes back to an old theorem of Gelfand, stating that any commutative C ∗-algebra must be of the form C(X), for a certain compact space X. In view of this result, one can define the category of "noncommutative compact spaces" to be the category of C ∗-algebras, with the arrows reversed. The category of usual compact spaces embeds then covariantly into this category, via X → C(X). Once again by using the Gelfand theorem, each noncommutative space X can be thought of as appearing as "liberation" of its classical version Xclass, which is obtained by dividing the corresponding algebra C(X) by its commutator ideal. We will be interested here in the liberation operation, in the algebraic manifold context. Given a family of noncommutative polynomials Pi ∈ C < z1, . . . , zN >, the associated noncommutative manifold X, and its classical version Xclass, are given by: X = Spec(cid:16)C ∗(cid:16)z1, . . . , zN(cid:12)(cid:12)(cid:12)Pi(z1, . . . , zN ) = 0(cid:17)(cid:17) S Xclass = n(z1, . . . , zN ) ∈ CN(cid:12)(cid:12)(cid:12)Pi(z1, . . . , zN ) = 0o S Here the family of polynomials {Pi} is assumed to be such that the biggest C ∗-norm on the universal ∗-algebra < z1, . . . , zNPi(z1, . . . , zN ) = 0 > is bounded. The liberation operation Xclass → X can be axiomatized in the quantum group con- text, the idea being that the category of pairings, which encodes in an abstract way the commutation relations ab = ba, must be replaced by a new category of partitions. The 2000 Mathematics Subject Classification. 46L65 (46L54). Key words and phrases. Liberation theory, Homogeneous space. 1 2 TEODOR BANICA theory here, based on Woronowicz's fundamental work in [35], [36], on Wang's free quan- tum groups [31], [32], on the Weingarten formula [4], [5], [12], [34], and on the liberation philosophy in free probability theory [9], [21], [27], [30], was developed in [8]. In the homogeneous space case, where the general study goes back to [10], [22], [23], some related theory, concerning spaces of type X = GN /GN −M , was developed in [6], and then in [7]. Such spaces were shown to have a number of interesting features, making them potential candidates for an algebraic manifold extension of [8], provided that the family of compact quantum groups G = (GN ) producing them satisfies: (1) The "easiness" condition in [8], stating that we must have SN ⊂ GN , for any (2) The "uniformity" condition, stating that we must have GN ∩ U + N −M = GN −M , N ∈ N, with these inclusions being of a certain special type. with respect to the standard embedding U + N −M ⊂ U + N . We will review here this work, by using some new ideas, from [1], [2], [3]. On one hand, we will replace the easiness assumption by a condition of type HN ⊂ GN , which will allow us to use a twisting parameter q = ±1. On the other hand, we will study more general quotient spaces, depending on parameters L ≤ M ≤ N, as follows: X = (GM × GN )(cid:14)(GL × GM −L × GN −L) Our main result will be a verification of the Bercovici-Pata liberation criterion, for certain variables associated χ ∈ C(X), in a suitable L, M, N → ∞ limit. There are many questions raised by the present work. Here are some of them: (1) A first question concerns the full classification of the quantum groups satisfying the above conditions. For some recent advances here, see [15], [25], [28], [29]. (2) A second question concerns the validity of the quantum isometry group formula G+(X +) = G(X)+, in relation with the rigidity results in [11], [16]. (3) Yet another question regards the possible applications of the present formalism to free probability invariance questions, in the spirit of [13], [18]. (4) Finally, there are as well several interesting questions in relation with the axiom- atization problem for the noncommutative algebraic manifolds [17]. Finally, regarding the general presentation of the paper: (1) We use Woronowicz's quantum group formalism in [35], [36], with the extra axiom S2 = id. A reference here is the book [20]. Regarding the noncommutative homogeneous spaces, whose axiomatization is known to run into several difficulties [7], [10], [14], [22], [23], [26], we use here a simplified formalism, best adapted to our examples, that we intend to fully clarify in a forthcoming paper. (2) The present work is a technical continuation of [1], [2], [3], [6], [7], with the aim of basically enlarging a list of examples. As already mentioned, the general theory is not available yet. We have therefore opted for an "example-first" presentation. This is actually in tune with the easy quantum group literature, where most of the basic examples were constructed and studied long before [8]. LIBERATION THEORY FOR NONCOMMUTATIVE HOMOGENEOUS SPACES 3 The paper is organized as follows: in 1-2 we discuss the orthogonal and unitary cases, in 3-4 we introduce all the quantum groups that we are interested in, and we study the associated homogeneous spaces, and in 5-6 we discuss probabilistic aspects. I would like to thank the referee for a careful reading of the Acknowledgements. manuscript, and for a number of useful suggestions. This work was partly supported by the "Harmonia" NCN grant 2012/06/M/ST1/00169. 1. Partial isometries In this section and in the next one we discuss the construction of the homogeneous spaces that we are interested in, in the case where the underlying groups or quantum groups are ON , UN , or their twists ¯ON , ¯UN , or their free versions O+ N , U + N . We begin with the classical case. Best is to start as follows: Definition 1.1. Associated to any integers L ≤ M ≤ N are the spaces OL U L M N =nT : E → F isometry(cid:12)(cid:12)(cid:12)E ⊂ RN , F ⊂ RM , dimR E = Lo M N =nT : E → F isometry(cid:12)(cid:12)(cid:12)E ⊂ CN , F ⊂ CM , dimC E = Lo where the notion of isometry is with respect to the usual real/complex scalar products. As a first observation, at L = M = N we obtain the groups ON , UN . More generally, at of RN , CN , studied in [2], which are by definition given by: M = N we obtain the various components of the semigroups eON ,eUN of partial isometries OL N N N[L=0 eON = U L N N N[L=0 eUN = Yet another interesting specialization is L = M = 1. Here the elements of O1 1N are the isometries T : E → R, with E ⊂ RN one-dimensional, and such an isometry is uniquely determined by the element T −1(1) ∈ RN , which must belong to the sphere SN −1 . Thus, we have O1 . Similarly, in the complex case we have U 1 In general, the most convenient is to view the elements of OL 1N = SN −1 M N , U L M N as rectangular 1N = SN −1 R R . C matrices, and to use matrix calculus for their study: Proposition 1.2. We have identifications of compact spaces OL U L M N ≃nU ∈ MM ×N (R)(cid:12)(cid:12)(cid:12)UU t = projection of trace Lo M N ≃nU ∈ MM ×N (C)(cid:12)(cid:12)(cid:12)UU ∗ = projection of trace Lo with each partial isometry being identified with the corresponding rectangular matrix. 4 TEODOR BANICA Proof. We can indeed identify the partial isometries T : E → F with their corresponding extensions U : RN → RM , U : CN → CM , obtained by setting UE⊥ = 0, and then identify these latter linear maps U with the corresponding rectangular matrices. (cid:3) As an illustration, at L = M = N we recover in this way the usual matrix description See [2]. Finally, at L = M = 1 we obtain the usual description of SN −1 of ON , UN . More generally, at M = N we recover the usual matrix description of eON ,eUN . Now back to the general case, observe that the isometries T : E → F , or rather their extensions U : KN → KM , with K = R, C, obtained by setting UE⊥ = 0, can be composed with the isometries of KM , KN , according to the following scheme: , SN −1 . R C KN B∗ / KN B(E) E U T / KM A / KM F A(F ) In other words, the groups OM × ON , UM × UN act respectively on OL the identifications in Proposition 1.2 made, the statement here is: M N , U L M N . With Proposition 1.3. We have action maps as follows, which are transitive, OM × ON y OL UM × UN y U L M N M N : : (A, B)U = AUBt (A, B)U = AUB∗ whose stabilizers are respectively OL × OM −L × ON −L and UL × UM −L × UN −L. Proof. We have indeed action maps as in the statement, which are transitive. Let us compute now the stabilizer G of the point U = (1 0). Since the elements (A, B) ∈ G 0 0 satisfy AU = UB, their components must be of the form A = (x b). Now since 0 A, B are both unitaries, these matrices follow to be block-diagonal, and we obtain: a), B = (x ∗ ∗ 0 G =(cid:26)(A, B)(cid:12)(cid:12)(cid:12)A =(cid:18)x 0 0 a(cid:19) , B =(cid:18)x 0 0 b(cid:19)(cid:27) We conclude that the stabilizer of U = (1 0 0 0) is parametrized by triples (x, a, b) belonging (cid:3) respectively to OL × OM −L × ON −L and UL × UM −L × UN −L, as claimed. M N : Finally, let us work out the quotient space description of OL M N , U L Theorem 1.4. We have isomorphisms of homogeneous spaces as follows, OL M N = (OM × ON )/(OL × OM −L × ON −L) U L M N = (UM × UN )/(UL × UM −L × UN −L) with the quotient maps being given by (A, B) → AUB∗, where U = (1 0 0 0). / / / / / O O / / O O / / O O O O ∆(uij) =Xk uik ⊗ ukj , ε(uij) = δij , S(uij) = u∗ ji LIBERATION THEORY FOR NONCOMMUTATIVE HOMOGENEOUS SPACES 5 Proof. This is just a reformulation of Proposition 1.3 above, by taking into account the fact that the fixed point used in the proof there was U = (1 (cid:3) 0 0 0). Once again, the basic examples here come from the cases L = M = N and L = M = 1, where the quotient spaces at right are respectively ON , UN and ON /ON −1, UN /UN −1. In fact, in the general L = M case we obtain the following spaces, considered in [7]: OM M N = (OM × ON )/(OM × ON −M ) = ON /ON −M U M M N = (UM × UN )/(UM × UN −M ) = UN /UN −M For some further information on these spaces, we refer to [2], [7]. 2. Liberations and twists We discuss now some noncommutative versions of the above constructions. We use the quantum group formalism of Woronowicz [35], [36], with the extra axiom S2 = id. In other words, we consider pairs (A, u) consisting of a C ∗-algebra A, and a unitary matrix u ∈ MN (A), such that the following formulae define morphisms of C ∗-algebras: These morphisms are called comultiplication, counit and antipode. We write A = C(G), and call G a compact matrix quantum group. For full details here, see [20]. We recall from [1], [5] that the compact groups ON , UN can be twisted, by replacing the commutation relations ab = ba, ab∗ = b∗a between the standard coordinates uij(g) = gij with the following commutation/anticommutation relations: ab× =(−b×a for a 6= b on the same row or column of u otherwise b×a Here b× = b, b∗, and the precise statement is that these relations, when applied to a ij)), respectively matrix u = (uij) which is orthogonal (u = ¯u, ut = u−1, where ¯u = (u∗ biunitary (u∗ = u−1, ut = ¯u−1) produce quantum groups ¯ON , ¯UN . See [1], [5]. We can liberate OL M N , U L M N , and then twist them, as follows: Definition 2.1. Associated to any integers L ≤ M ≤ N are the algebras C(OL+ C(U L+ M N ) = C ∗(cid:16)(uij)i=1,...,M,j=1,...,N(cid:12)(cid:12)(cid:12)u = ¯u, uut = projection of trace L(cid:17) M N ) = C ∗(cid:16)(uij)i=1,...,M,j=1,...,N(cid:12)(cid:12)(cid:12)uu∗, ¯uut = projections of trace L(cid:17) M N ), obtained by imposing the twisting relations. M N ), C( ¯U L and their quotients C( ¯OL Observe that the above universal algebras are well-defined, because the trace conditions, which readPij uiju∗ ij =Pij u∗ ijuij = L, show that we have uij ≤ √L. 6 TEODOR BANICA We have inclusions between the various spaces constructed so far, as follows: U L M N / U L+ M N ¯U L M N OL M N OL+ M N ¯OL M N Indeed, the inclusions at right follow from definitions, and those at left come from Proposition 1.2, and from the fact that OL M N , U L M N are stable by conjugation. At the level of basic examples now, we first have the following result: Proposition 2.2. At L = M = 1 we obtain the diagram SN −1 C / SN −1 C,+ ¯SN −1 C SN −1 R SN −1 R,+ ¯SN −1 R consisting of the liberations and twists of the spheres SN −1 R , SN −1 C . Proof. We recall from [1] that the various spheres are constructed as follows, with the symbol × standing for "commutative", "twisted" and "free", respectively: C(SN −1 R,× ) = C ∗ C(SN −1 C,× ) = C ∗ z2 × z1, . . . , zN(cid:12)(cid:12)(cid:12)zi = z∗ i ,Xi × z1, . . . , zN(cid:12)(cid:12)(cid:12)Xi i =Xi 1N , U 1× ziz∗ i = 1! i zi = 1! z∗ Now by comparing with the definition of O1× 1N , this proves our claim. We have as well the following result, once again making the link with [1]: (cid:3) Proposition 2.3. At L = M = N we obtain the diagram ¯UN UN / U + N consisting of the liberations and twists of the groups ON , UN . ON O+ N ¯ON / o o / / O O O O o o O O / o o / / O O O O o o O O / o o / / O O O O o o O O LIBERATION THEORY FOR NONCOMMUTATIVE HOMOGENEOUS SPACES 7 Proof. We recall from [1] that the various quantum groups are constructed as follows, with the symbol × standing once again for "commutative", "twisted" and "free": On the other hand, according to Proposition 1.2 and to Definition 2.1 above, we have the following presentation results: C(O× N ) = C ∗ C(U × N ) = C ∗ C(ON × N N ) = C ∗ C(U N × N N ) = C ∗ ×(cid:16)(uij)i,j=1,...,N(cid:12)(cid:12)(cid:12)u = ¯u, uut = utu = 1(cid:17) ×(cid:16)(uij)i,j=1,...,N(cid:12)(cid:12)(cid:12)uu∗ = u∗u = 1, ¯uut = ut¯u = 1(cid:17) ×(cid:16)(uij)i,j=1,...,N(cid:12)(cid:12)(cid:12)u = ¯u, uut = projection of trace N(cid:17) ×(cid:16)(uij)i,j=1,...,N(cid:12)(cid:12)(cid:12)uu∗, ¯uut = projections of trace N(cid:17) ×(cid:16)(uij)i,j=1,...,N(cid:12)(cid:12)(cid:12)u = ¯u, uut, utu = projections of trace N(cid:17) ×(cid:16)(uij)i,j=1,...,N(cid:12)(cid:12)(cid:12)uu∗, u∗u, ¯uut, ut¯u = projections of trace N(cid:17) We use now the standard fact that if p = aa∗ is a projection then q = a∗a is a projection too. Together with T r(uu∗) = T r(ut¯u) and T r(¯uut) = T r(u∗u), this gives: C(ON × N N ) = C ∗ C(U N × N N ) = C ∗ Now observe that, in tensor product notation, and by using the normalized trace, the conditions at right are all of the form (tr ⊗ id)p = 1, with p = uu∗, u∗u, ¯uut, ut¯u. We therefore obtain (tr ⊗ ϕ)(1 − p) = 0 for any faithful state ϕ, and it follows that the projections p = uu∗, u∗u, ¯uut, ut¯u must be all equal to the identity, as desired. (cid:3) Regarding now the homogeneous space structure of OL× M N , U L× M N , the situation here is more complicated in the twisted and free cases than in the classical case. See [1], [7]. The classical results have, however, some partial extensions. In order to formulate a result, we use the standard coaction formalism for the compact quantum groups, as in [7]. Also, given two noncommutative compact spaces X, Y , we define their product X × Y via the formula C(X × Y ) = C(X) ⊗ C(Y ), with the tensor product being the maximal one. Finally, we use the standard fact that when G, H are compact matrix quantum groups, then so is their product G × H. For more on these topics, see [7], [20]. With these conventions, we have the following result: Proposition 2.4. The spaces U L× M × U × (1) We have an action U × (2) We have a map U × N M × U × N → U L× Similar results hold for the spaces OL× M N have the following properties: y U L× M N , given by uij →Pkl aik ⊗ b∗ M N , given by uij →Pl≤L ail ⊗ b∗ jl. M N , with all the ∗ exponents removed. jl ⊗ ukl. 8 TEODOR BANICA Proof. In the classical case, the transpose of the action map UM × UN y U L quotient map UM × UN → U L 0 0): M N and of the M N are as follows, where J = (1 0 ϕ → ((A, B, U) → ϕ(AUB∗)) ϕ → ((A, B) → ϕ(AJB∗)) But with ϕ = uij we obtain precisely the formulae in the statement. The proof in the orthogonal case is similar. Regarding now the free case, the proof goes as follows: jl ⊗ ukl we have: plbpnb∗ mnust jt ⊗ uklu∗ aik ⊗ b∗ jt ⊗ ukt = Uij aika∗ (1) Assuming uu∗u = u, with Uij =Pkl aik ⊗ b∗ qmaqs ⊗ b∗ jt ⊗ uklu∗ = Xklmt ij =Xij Xklst Xij (V V ∗V )ij =Xpq Xx,y,z≤L (UU ∗U)ij = Xpq Xklmnst aik ⊗ b∗ Also, assuming that we havePij uiju∗ ij = L, we obtain: is ⊗ b∗ jlbjt ⊗ uklu∗ (2) Assuming uu∗u = u, with Vij =Pl≤L ail ⊗ b∗ qyaqz ⊗ b∗ Also, assuming that we havePij uiju∗ ij =Xij Xl,s≤L ij = L, we obtain: Xij mlumt =Xkt st =Xkl jz =Xx≤L jlbjs =Xl≤L jl we have: is ⊗ b∗ pxbpyb∗ UijU ∗ VijV ∗ aixa∗ aika∗ aila∗ 1 ⊗ 1 ⊗ uklu∗ kl = L aix ⊗ b∗ jx = Vij 1 = L By removing all the ∗ exponents, we obtain as well the orthogonal results. In the twisted case the proof is similar. Let us first discuss the orthogonal case. The twisting relations can be written as follows: With this formula in hand, the verification of the extra relations goes as follows: uijupq = (−1)δip+δjq upquij (1) With Uij =Pkl aik ⊗ bjl ⊗ ukl we have, as desired: aikapm ⊗ bjlbqn ⊗ uklumn (−1)δip+δkmapmaik ⊗ (−1)δjq+δlnbqnbjl ⊗ (−1)δkm+δlnumnukl (−1)δip+δjq apmaik ⊗ bqnbjl ⊗ umnukl = (−1)δip+δjq UpqUij UijUpq = Xklmn = Xklmn = Xklmn LIBERATION THEORY FOR NONCOMMUTATIVE HOMOGENEOUS SPACES 9 (2) With Vij =Pl≤L ail ⊗ bjl we have as well, as desired: VijVpq = Xl,m≤L = Xl,m≤L ailapm ⊗ bjlbqm = Xl,m≤L (−1)δip+δjq apmail ⊗ bqmbjl = (−1)δip+δjq VpqVij (−1)δip+δlmapmail ⊗ (−1)δjq+δlmbqmbjl The proof in the unitary case is similar, by adding ∗ exponents where needed. Let us examine now the relation between the above maps. In the classical case, given (cid:3) a quotient space X = G/H, the associated action and quotient maps are given by: (a : G × X → X : p : G → X : (g, g′H) → gg′H g → gH Thus we have a(g, p(g′)) = p(gg′). In our context, a similar result holds: N , U × Theorem 2.5. With G = GM × GN and X = GL M N , where GN = O× N , we have G × G id×p G × X m a G p / X where a, p are the action map and the map constructed in Proposition 2.4. Proof. At the level of the associated algebras of functions, we must prove that the following diagram commutes, where Φ, π are morphisms of algebras induced by a, p: C(X) π C(G) Φ ∆ C(G × X) id⊗π / C(G × G) When going right, and then down, the composition is as follows: (id ⊗ π)Φ(uij) = (id ⊗ π)Xkl aik ⊗ b∗ aik ⊗ b∗ jl ⊗ aks ⊗ b∗ ls On the other hand, when going down, and then right, the composition is as follows, where F23 is the flip between the second and the third components: ∆π(uij) = F23(∆ ⊗ ∆)Xs≤L aik ⊗ aks ⊗ b∗ Thus the above diagram commutes indeed, and this gives the result. ais ⊗ b∗ ls! jl ⊗ b∗ (cid:3) jl ⊗ ukl =Xkl Xs≤L js = F23 Xs≤LXkl / / / / / / 10 TEODOR BANICA In general, going beyond Theorem 2.5 leads to some non-trivial questions. A first issue comes from the fact that the inclusions GL × GM −L × GN −L ⊂ GM × GN are not well- defined, in the free case. There are as well some analytic issues, coming from the fact that the maps in Proposition 2.4 (2) are in general not surjective. See [1], [7]. 3. Uniform quantum groups We discuss in this section a generalization of the above constructions. For this purpose, we first need to axiomatize a suitable class of compact quantum groups, generalizing the classical groups ON , UN , their twists ¯ON , ¯UN , and their free versions O+ N , U + N . Let P (k, l) the set of partitions between an upper row of k points, and a lower row of l points, with each leg colored black or white, and with k, l standing for the corresponding "colored integers". We have the following notion, which appears as a straightforward generalization of the corresponding orthogonal notion from [8]: Definition 3.1. A category of partitions is a collection of sets D = Skl D(k, l), with D(k, l) ⊂ P (k, l), which contains the identity, and is stable under: (1) The horizontal concatenation operation ⊗. (2) The vertical concatenation ◦, after deleting closed strings in the middle. (3) The upside-down turning operation ∗ (with reversing of the colors). Here the vertical concatenation operation assumes of course that the colors match. Regarding the identity, the precise condition is that D(◦,◦) contains the "white" identity ◦ ◦ . By using (3) we see that D(•,•) contains the "black" identity • • , and then by using (1) we see that each D(k, k) contains its corresponding (colored) identity. The basic example of such a category is P itself. Yet another basic example is the category NC of noncrossing partitions. There are of course many other examples. We refer to [8] for the uncolored case, and to [28], [29] for the general colored case. As explained in [8], [19], [29], such categories produce quantum groups. In this paper, however, we will need a modification of this construction, in order to cover as well the twists. We use for this purpose a number of findings from our recent papers [1], [3]. As explained in [1], [3], in order to cover the twists of the quantum groups in [8] we must restrict attention to the categories D ⊂ Peven, where Peven ⊂ P is the category of partitions having all blocks of even size. Such partitions act on tensors, as follows: Definition 3.2. Associated to any π ∈ Peven(k, l) are the linear maps ej1 ⊗ . . . ⊗ ejl ej1 ⊗ . . . ⊗ ejl Tπ(ei1 ⊗ . . . ⊗ eik) = Xj:ker(i ¯Tπ(ei1 ⊗ . . . ⊗ eik) = Xτ ≤π j )≤π ε(τ ) Xj:ker(i j )=τ where {ei} is the standard basis of CN , and ε : Peven → {−1, 1} is the signature map. LIBERATION THEORY FOR NONCOMMUTATIVE HOMOGENEOUS SPACES 11 Here the kernel of a multi-index (i j) = (i1...ik ) is the partition obtained by joining the j1...jl sets of equal indices. Thus, the condition ker(i j) ≤ π simply tells us that the strings of π must join equal indices. As for the signature map ε : Peven → {−1, 1}, this is a canonical extension of the usual signature map ε : S∞ → {−1, 1}, constructed in [1]. Here are a few examples of such linear maps, taken from [1]: T∩(1) = ¯T∩(1) =Xi ei ⊗ ei T/\(ei ⊗ ej) = ej ⊗ ei , , T∪(ei ⊗ ej) = ¯T∪(ei ⊗ ej) = δij ¯T/\(ei ⊗ ej) =(ej ⊗ ei for i = j for i 6= j −ej ⊗ ei In general, Tπ, ¯Tπ can be thought of as coming from a twisting parameter q = ±1. As explained in [1], for π ∈ NCeven we have τ ≤ π =⇒ ε(τ ) = 1, and so Tπ = ¯Tπ. In general, however, the maps Tπ, ¯Tπ are different. We refer to [1] for full details here. We have the following "q-easiness" notion, inspired from [8]: Definition 3.3. A compact quantum group G ⊂ U + N is called quizzy when Hom(u⊗k, u⊗l) = span(cid:16) Tπ(cid:12)(cid:12)π ∈ D(k, l)(cid:17) for any colored integers k, l, for a certain category of partitions D ⊂ Peven. Here the dot stands for a fixed value of q = ±1, with the maps Tπ being used at q = 1, and with the maps ¯Tπ being used at q = −1. Also, the "colored" tensor powers u⊗k, u⊗l are defined by tensoring the corepresentations u◦ = u and u• = ¯u. At the level of basic examples, we have the following result: Proposition 3.4. The following are quizzy quantum groups, UN ON / U + N O+ N ¯UN ¯ON with q ∈ {−1, 1} being by definition 1 at left, −1 at right, and ±1 in the middle. Proof. As explained in [1], [8], the above quantum groups appear indeed from the following categories of partitions, with q ∈ {−1, 1} being as in the statement: P2 NC2 P2 P2 NC2 / P2 / o o / / O O O O o o O O / / o o o o / 12 TEODOR BANICA Here P2 is the set of all pairings, P2 ⊂ P2 is the set of "matching" pairings, whose upper and lower strings connect ◦ − •, and whose through strings connect ◦ − ◦ or • − •, and NC2,NC2 are the corresponding subsets of noncrossing pairings. See [1], [8]. Consider now the group H s N = Zs ≀ SN , with s ∈ {2, 4, . . . ,∞}, which consists of the permutation matrices σ ∈ SN with nonzero entries multiplied by elements of Zs. This group has a free analogue, H s+ N = Zs ≀∗ S+ N , constructed in [4], as follows: (cid:3) C(H s+ N ) = C ∗(cid:16)(uij)i,j=1,...,N(cid:12)(cid:12)(cid:12)uiju∗ ij = u∗ ijuij = pij = magic, us ij = pij(cid:17) Here the "magic" condition states that the entries of p = (pij) are projections, summing ij = pij, disappears by definition The s = 2,∞ specializations of H s up to 1 on each row and column, and the last condition, us at s = ∞. Observe that the classical version of H s+ are the quantum groups in [5], and their complex analogues. We have: Proposition 3.5. The following are quizzy quantum groups, at both q = ±1, N . See [4]. N , denoted respectively HN , H + N and KN , K + N , N is indeed H s N , H s+ KN H s N / K + N H s+ N H + N where H s HN N = Zs ≀ SN , and where H s+ N = Zs ≀∗ S+ N , with s ∈ {2, 4, . . . ,∞}. Proof. As explained in [3], [4], the above quantum groups appear indeed, with parameter q = ±1 as in the statement, from the following categories of partitions: Peven NCeven P s even Peven NC s even NCeven Here P s even ⊂ Peven is the set of partitions having the property that, in each block, the number of white legs equals the number of black legs, modulo s, with all legs counted with coefficient + up, and − down. At right we have the subset NC s even ∩ NC, and the lower and upper objects are the corresponding specializations at s = 2,∞. even = P s We recall that the free complexification (eG,eu) of a compact matrix quantum group (G, u) is obtained by considering the subalgebra C(eG) ⊂ C(T) ∗ C(G) generated by the entries ofeu = zu, where z is the standard generator of C(T). See [24]. (cid:3) / / / O O O O / / O O O O o o o o o o LIBERATION THEORY FOR NONCOMMUTATIVE HOMOGENEOUS SPACES 13 With this convention, we have the following extra example, from [29]: Proposition 3.6. The quantum group K ++ N appears as an intermediate object N = eK + K + N ⊂ K ++ N ⊂ U + N N ) → C(K + with both inclusions being proper, and is quizzy, equal to its own twist. N ) ⊂ C(T) ∗ C(K + Proof. By composing the canonical inclusion C(K ++ obtain a morphism C(K ++ N ) mapping euij → uij, so we have inclusions as in ij = zuiju∗ ijz∗ and qij = eu∗ ijeuij = u∗ N ) with ε ∗ id we ijuij are projections, and are not equal, both these inclusions are proper. the statement. Since the elements pij = euijeu∗ Regarding the easiness claim, this follows from the general theory of the representations of free complexifications [24]. To be more precise, as explained in [29], the associated category NC− even is that of the even noncrossing partitions which, when rotated on one line, have alternating colors in each block. Observe that the inclusions in the statement correspond then to the inclusions at the partition level, which are as follows: NCeven ⊃ NC− even ⊃ NC2 Finally, since NC− The above examples are in fact the only ones that we are interested in, in the classi- even ⊂ NCeven, the quantum group K ++ N equals its own twist. (cid:3) cal/twisted and free cases. In order to axiomatize these objects, we use: Proposition 3.7. For a quizzy quantum group HN ⊂ GN ⊂ U + of partitions NC2 ⊂ D ⊂ Peven, the following are equivalent: N , coming from a category (1) D is stable by removing blocks. (2) GN ∩ U + N −M = GN −M , for any M ≤ N. If these conditions are satisfied, we call both G = (GN ) and D "uniform". Proof. This was proved in [7] in the orthogonal case, and the proof in general is similar. Assume that we have a subgroup K ⊂ U + N −M , with fundamental representation v, and consider the N × N matrix v = diag(v, 1M ). Then, for any π ∈ Peven, we have: Tπ ∈ Hom(v⊗k, v⊗l) ⇐⇒ Tπ′ ∈ Hom(v⊗k′ , v⊗l′ ), ∀π′ ⊂ π With this formula in hand, we deduce that given a subgroup G ⊂ U + representation denoted u, the algebra of functions on K = G ∩ U + N , with fundamental N −M is given by: C(K) = C(U + N −M ).(cid:10)T ∈ Hom(v⊗k, v⊗l), ∀T ∈ Hom(u⊗k, u⊗l)(cid:11) Thus, we have GN ∩ U + N ) is the easy quantum group associated to the category D′ generated by all the subpartitions of the partitions in D. In particular GN ∩ U + N −M = GN −M for any M ≤ N is equivalent to D = D′, as claimed. (cid:3) N −M , where G′ = (G′ N −M = G′ 14 TEODOR BANICA Observe that the quantum groups in Propositions 3.4, 3.5, 3.6 are all uniform. We have in fact the following result, where by "classical/twisted" and "free" we mean \/ ∈ D and D ⊂ NCeven, where D ⊂ Peven is the associated category of partitions: Theorem 3.8. The uniform classical/twisted and free quantum groups are KN H s N HN UN , ¯UN K + N H s+ N H + N ON , ¯ON / K ++ N / U + N O+ N where H s N = Zs ≀ SN , H s+ N = Zs ≀∗ S+ N , with s ∈ {2, 4, . . . ,∞}, and K ++ N . N = eK + Proof. This is a consequence of the recent classification results in [28], [29], the idea being as follows. First, the diagram in the statement being obtained by merging the examples in Propositions 3.4, 3.5, 3.6, all the above quantum groups are quizzy. The uniformity condition is clear as well, for each of the quantum groups under con- sideration. Finally, all these quantum groups are either classical/twisted or free. In order to prove now the converse, in view of the twisting results in [2], it is enough to deal with the q = 1 case. So, consider a uniform category of partitions D ⊂ Peven, as in Proposition 3.7. We must prove that in the classical/free cases, the solutions are: Peven P s even Peven P2 NCeven NC s even NCeven P2 NC− even NC2 NC2 To be more precise, in the classical case, where \/ ∈ D, we must prove that the only solutions are the categories P2,P2, P s even, and that in the free case, where D ⊂ NCeven, we must prove that the only solutions are the categories NC2,NC2,NC− We jointly investigate these two problems. Let B be the set of all possible labelled blocks in D, having no upper legs. Observe that B is stable under the switching of colors operation, ◦ ↔ •. We have two possible situations, as follows: even, NC s even. ) ) / / 0 0 / / / / O O O O / / O O . . / / O O O O 5 5 u u o o o o o o o o o o o o o o o o i i LIBERATION THEORY FOR NONCOMMUTATIVE HOMOGENEOUS SPACES 15 (1) B consists of pairings only. Here the pairings in question can be either all labelled pairings, namely ◦−◦, ◦− •, •−◦, •−•, or just the matching ones, namely ◦−•, •−◦, and we obtain here P2,P2 in the classical case, and NC2,NC2 in the free case. (2) B has at least one block of size ≥ 4. In this case we can let s ∈ {2, 4, . . . ,∞} to be the length of the smallest ◦ . . .◦ block, and we obtain in this way the category P s even in the classical case, and the categories NC− (cid:3) even in the free case. See [28]. The above occurrence of K ++ N is quite an issue, and reminds the "forgotten series" from the orthogonal case [8], found later on, in [33]. Note that this forgotten series has disappeared in the present setting, due to the uniformity axiom. In the unitary case, however, it is quite unclear on how to add an extra axiom, as to avoid K ++ N . even, NC s In what follows we will rather ignore K ++ N , and focus instead on K + Bercovici-Pata computations in [4] to be the "correct" liberation of KN . N , known since the In general now, the full classification of the uniform quizzy quantum groups remains an open problem. For the ongoing program here, we refer to [15], [25], [28], [29]. 4. Homogeneous spaces In this section we associate noncommutative homogeneous spaces, as in sections 1-2 above, to the quantum groups considered in section 3. We will be mostly interested in the quantum groups from Theorem 3.8, so let us first discuss, with full details, the case of the quantum groups H s Definition 4.1. Associated to any partial permutation, σ : I ≃ J with I ⊂ {1, . . . , N} and J ⊂ {1, . . . , M}, is the real/complex partial isometry N appearing there. As in [2], we use: N , H s+ given on the standard basis elements by Tσ(ei) = eσ(i). Tσ : span(cid:16)ei(cid:12)(cid:12)(cid:12)i ∈ I(cid:17) → span(cid:16)ej(cid:12)(cid:12)(cid:12)j ∈ J(cid:17) We denote by SL M N the set of partial permutations σ : I ≃ J as above, with range I ⊂ {1, . . . , N} and target J ⊂ {1, . . . , M}, and with L = I = J. See [2]. In analogy with the decomposition result H s N = Zs ≀ SN , we have: Proposition 4.2. The space of partial permutations signed by elements of Zs, is isomorphic to the quotient space (H s H sL M N =nT (ei) = wieσ(i)(cid:12)(cid:12)(cid:12)σ ∈ SL M × H s N )/(H s M N , wi ∈ Zso L × H s M −L × H s N −L). Proof. This follows by adapting the computations in the proof of Proposition 1.3 above. Indeed, we have an action map as follows, which is transitive: The stabilizer of the point U = (1 0 embedded via (x, a, b) → [(x 0 M × H s H s N : y H sL M N 0 0) follows to be the group H s (A, B)U = AUB∗ a), (x 0 0 0 b)], and this gives the result. L × H s M −L × H s N −L, (cid:3) 16 TEODOR BANICA In the free case now, the idea is similar, by using inspiration from the construction of the quantum group H s+ N = Zs ≀∗ S+ N in [4]. The result here is as follows: Proposition 4.3. The noncommutative space H sL+ M N associated to the algebra C(H sL+ M N ) = C(U L+ ij = u∗ ijuij = pij = projections, us has an action map, and is the target of a quotient map, as in Theorem 2.5 above. Proof. We must show that if the variables uij satisfy the relations in the statement, then these relations are satisfied as well for the following variables: M N ).(cid:10)uiju∗ ij = pij(cid:11) Uij =Xkl aik ⊗ b∗ jl ⊗ ukl , Vij =Xl≤L ail ⊗ b∗ jl Since the standard coordinates aij, bij on the quantum groups H s+ M , H s+ N satisfy the relations xy = xy∗ = 0, for any x 6= y on the same row or column of a, b, we obtain: jlbjm ⊗ uklu∗ aika∗ ik ⊗ b∗ jlbjl ⊗ uklu∗ kl ir ⊗ b∗ Thus, in terms of the projections xij = aija∗ VijV ∗ aila∗ UijU ∗ ij = Xklmn aika∗ im ⊗ b∗ ij = Xl,r≤L ij =Xkl mn =Xkl jlbjr =Xl≤L UijU ∗ xik ⊗ yjl ⊗ pkl , VijV ∗ xil ⊗ yjl ij, yij = bijb∗ ij, we have: aila∗ jlbjl il ⊗ b∗ ij, pij = uiju∗ ij =Xl≤L By repeating the computation, we conclude that these elements are projections. Also, a similar computation shows that U ∗ ijUij, V ∗ ijVij are given by the same formulae. Finally, once again by using the relations of type xy = xy∗ = 0, we have: U s ij =Xkrlr jl1 . . . b∗ aik1 . . . aiks ⊗ b∗ ij =Xlr≤L V s ail1 . . . ails ⊗ b∗ jls ⊗ uk1l1 . . . uksls =Xkl il ⊗ (b∗ as jl1 . . . b∗ jls =Xl≤L jl)s as ik ⊗ (b∗ jl)s ⊗ us kl Thus the conditions of type us ij = pij are satisfied as well, and we are done. (cid:3) Let us discuss now the general case. We use the Kronecker symbols δπ(i) ∈ {−1, 0, 1} from [1], depending on a twisting parameter q = ±1, constructed as follows: δσ(i) =(δker i≤σ ε(ker i)δker i≤σ (untwisted case) (twisted case) With this convention, we have the following result: LIBERATION THEORY FOR NONCOMMUTATIVE HOMOGENEOUS SPACES 17 Proposition 4.4. The various spaces GL standard coordinates of U L+ M N the relations M N constructed so far appear by imposing to the Xi1...is Xj1...js δπ(i)δσ(j)ue1 i1j1 . . . ues isjs = Lπ∨σ with s = (e1, . . . , es) ranging over all the colored integers, and with π, σ ∈ D(0, s). Proof. According to the various constructions in section 1-2 and in the beginning of this section, the relations defining GL M N can be written as follows, with σ ranging over a family of generators, with no upper legs, of the corresponding category of partitions D: δσ(j)ue1 i1j1 . . . ues isjs = δσ(i) Xj1...js We therefore obtain the relations in the statement, as follows: Xi1...is Xj1...js δπ(i)δσ(j)ue1 i1j1 . . . ues i1j1 . . . ues isjs δσ(j)ue1 δπ(i) Xj1...js δπ(i)δσ(i) = Xτ ≤π∨σ Xker i=τ isjs = Xi1...is = Xi1...is = Xτ ≤π∨σ Xker i=τ 1 = Lπ∨σ (±1)2 As for the converse, this follows by using the relations in the statement, by keeping π (cid:3) fixed, and by making σ vary over all the partitions in the category. In the general case now, where G = (GN ) is an arbitary uniform quizzy quantum group, we can construct spaces GL M N by using the above relations, and we have: Theorem 4.5. The spaces GL M N constructed by imposing the relations M N ⊂ U L+ Xi1...is Xj1...js δπ(i)δσ(j)ue1 i1j1 . . . ues isjs = Lπ∨σ with π, σ ranging over all the partitions in the associated category, having no upper legs, are subject to an action map/quotient map diagram, as in Theorem 2.5. Proof. We proceed as in the proof of Proposition 2.4. We must prove that, if the variables uij satisfy the relations in the statement, then so do the following variables: Uij =Xkl aik ⊗ b∗ jl ⊗ ukl , Vij =Xl≤L ail ⊗ b∗ jl 18 TEODOR BANICA Regarding the variables Uij, the computation here goes as follows: δπ(i)δσ(j)U e1 i1j1 . . . U es isjs Xi1...is Xj1...js = Xi1...is Xj1...js Xk1...ksXl1...ls = Xk1...ksXl1...ls δπ(k)δσ(l)ue1 k1l1 . . . ues ksls = Lπ∨σ δπ(i)δσ(j)ae1 i1k1 . . . aes isks ⊗ (bes jsls . . . be1 j1l1)∗ ⊗ ue1 k1l1 . . . ues ksls For the variables Vij the proof is similar, as follows: δπ(i)δσ(j)V e1 i1j1 . . . V es isjs Xi1...is Xj1...js = Xi1...is Xj1...js Xl1,...,ls≤L = Xl1,...,ls≤L δπ(l)δσ(l) = Lπ∨σ δπ(i)δσ(j)ae1 i1l1 . . . aes isls ⊗ (bes jsls . . . be1 j1l1)∗ Thus we have constructed an action map, and a quotient map, as in Proposition 2.4 (cid:3) above, and the commutation of the diagram in Theorem 2.4 is then trivial. The above results generalize some of the constructions in [3]. As explained in [3], there are many interesting questions regarding such spaces, and their quantum isometry groups. In what follows we will advance on some related topics, of probabilistic nature. 5. Integration theory In the remainder of this paper we discuss the integration over GL explicit formulae. Our main result will be the fact that the operations of type GL GL+ M N are indeed "liberations", in the sense of the Bercovici-Pata bijection [9]. The integration over GL M N is best introduced as follows: M N , with a number of M N → Definition 5.1. The integration functional of GL M N is the composition tr : C(GL M N ) → C(GM × GN ) → C Here we use the standard fact, proved by Woronowicz in [35], that any compact quantum of the representation uij →Pl≤L ail ⊗ b∗ group G has a Haar integration functional, RG : C(G) → C, which is by definition the unique positive unital trace subject to the following invariance relations: jl with the Haar functional of GM × GN . (cid:18)ZG ⊗id(cid:19) ∆ϕ =(cid:18)id ⊗ZG(cid:19) ∆ϕ =ZG ϕ LIBERATION THEORY FOR NONCOMMUTATIVE HOMOGENEOUS SPACES 19 Observe that in the case L = M = N we obtain the integration over GN . Also, at L = M = 1 we obtain the integration over the sphere. More generally, at any L = M we obtain the integration over the corresponding row algebra of GM , discussed in [7]. In the general case now, we first have the following result: Proposition 5.2. The integration functional tr has the invariance property (id ⊗ tr)Φ(x) = tr(x)1 with respect to the coaction map given by Φ(uij) =Pkl aik ⊗ b∗ Proof. We restrict the attention to the orthogonal case, the proof in the unitary case being similar. We must check the following formula: jl ⊗ ukl. b∗ l1m1 . . . b∗ lsms j1l1 . . . b∗ b∗ l1m1 . . . b∗ b∗ lsms j1m1 . . . b∗ jsms) (id ⊗ tr)Φ(ui1j1 . . . uisjs) = tr(ui1j1 . . . uisjs) Let us compute the left term. This is given by: j1l1 . . . b∗ j1l1 . . . b∗ jslsZGM jsls ⊗ uk1l1 . . . uksls ai1k1 . . . aisks ⊗ b∗ ak1m1 . . . aksms ⊗Xlr ai1k1 . . . aisks ⊗ b∗ ai1k1 . . . aisksZGM X = (id ⊗ tr)Xkrlr ak1m1 . . . aksmsZGN = Xkrlr Xmr≤L jslsZGN = Xmr≤LXkr X = Xmr≤L(cid:18)id ⊗ZGM(cid:19) ∆(ai1m1 . . . aisms) ⊗(cid:18)id ⊗ZGN(cid:19) ∆(b∗ = Xmr≤LZGM = (cid:18)ZGM ⊗ZGN(cid:19) Xmr≤L ai1m1 . . . aisms ⊗ZGN ai1m1 . . . aisms ⊗ b∗ j1m1 . . . b∗ jsms b∗ j1m1 . . . b∗ jsms By using now the invariance property of the Haar functionals of GM , GN , we obtain: But this gives the formula in the statement, and we are done. (cid:3) We will prove now that tr is in fact the unique positive unital invariant trace on M N ). For this purpose, we will need the Weingarten formula. We recall from section C(GL 4 above that the generalized Kronecker symbols are constructed as follows: δσ(i) =(δker i≤σ ε(ker i)δker i≤σ (untwisted case) (twisted case) With this convention, the integration formula is as follows: 20 TEODOR BANICA Theorem 5.3. We have the Weingarten type formula ZGL M N ui1j1 . . . uisjs = Xπστ ν Lσ∨νδπ(i)δτ (j)WsM (π, σ)WsN (τ, ν) where WsM = G−1 sM , with GsM (π, σ) = M π∨σ. Proof. We make use of the usual quantum group Weingarten formula, for which we refer to [1], [8]. By using this formula for GM , GN , we obtain: ZGL M N b∗ j1l1 . . . b∗ jsls ui1j1 . . . uisjs = Xl1...ls≤LZGM = Xl1...ls≤LXπσ = Xπστ ν Xl1...ls≤L ai1l1 . . . aislsZGN δπ(i)δσ(l)WsM (π, σ)Xτ ν δσ(l)δν(l)! δπ(i)δτ (j)WsM (π, σ)WsN (τ, ν) δτ (j)δν(l)WsN (τ, ν) Let us compute now the coefficient appearing in the last formula. Since the signature map takes ±1 values, for any multi-index l = (l1, . . . , ls) we have: δσ(l)δν(l) = δker l≤σε(ker l) · δker l≤νε(ker l) = δker l≤σ∨ν Thus the coefficient is Lσ∨ν, and we obtain the formula in the statement. (cid:3) We can now derive an abstract characterization of tr, as follows: Proposition 5.4. The integration functional tr constructed above is the unique positive unital C ∗-algebra trace C(GL M N ) → C which is invariant under the action of GM × GN . Proof. We use the method in [7], the point being to show that tr has the following ergod- icity property: (cid:18)ZGM ⊗ZGN ⊗id(cid:19) Φ = tr(.)1 We restrict the attention to the orthogonal case, the proof in the unitary case being similar. We must verify that the following holds: (cid:18)ZGM ⊗ZGN ⊗id(cid:19) Φ(ui1j1 . . . uikjk) = tr(ui1j1 . . . uikjk)1 LIBERATION THEORY FOR NONCOMMUTATIVE HOMOGENEOUS SPACES 21 By using the Weingarten formula, the left term can be written as follows: bj1l1 . . . bjsls · uk1l1 . . . uksls X = Xk1...ksXl1...lsZGM = Xk1...ksXl1...lsXπσ = Xπστ ν ai1k1 . . . aisksZGN δπ(i)δσ(k)WsM (π, σ)Xτ ν δπ(i)δτ (j)WsM (π, σ)WsN (τ, ν) Xk1...ksXl1...ls By using now the formula in Theorem 4.5 above, we obtain: δτ (j)δν(l)WsN (τ, ν) · uk1l1 . . . uksls δσ(k)δν(l)uk1l1 . . . uksls X = Xπστ ν Lσ∨νδπ(i)δτ (j)WsM (π, σ)WsN (τ, ν) Now by comparing with the formula in Theorem 5.3, this proves our claim. Assume now that τ : C(GL M N ) → C satisfies the invariance condition. We have: τ(cid:18)ZGM ⊗ZGN ⊗id(cid:19) Φ(x) = (cid:18)ZGM ⊗ZGN ⊗τ(cid:19) Φ(x) = (cid:18)ZGM ⊗ZGN(cid:19) (id ⊗ τ )Φ(x) = (cid:18)ZGM ⊗ZGN(cid:19) (τ (x)1) = τ (x) On the other hand, according to the formula established above, we have as well: τ(cid:18)ZGM ⊗ZGN ⊗id(cid:19) Φ(x) = τ (tr(x)1) = tr(x) Thus we obtain τ = tr, and this finishes the proof. (cid:3) 6. Probabilistic aspects We discuss now the precise computation of the laws of certain linear combinations of coordinates. A set of coordinates {uij} is called "non-overlapping" if each horizontal index i and each vertical index j appears at most once. With this convention, we have: Proposition 6.1. For a sum χE =P(ij)∈E uij of non-overlapping coordinates we have K π∨τ Lσ∨νWsM (π, σ)WsN (τ, ν) ZGL M N χs E = Xπστ ν where K = E is the cardinality of the indexing set. 22 TEODOR BANICA Proof. In terms of K = E, we can write E = {(α(i), β(i))}, for certain embeddings α : {1, . . . , K} ⊂ {1, . . . , M} and β : {1, . . . , K} ⊂ {1, . . . , N}. In terms of these maps α, β, the moment in the statement is given by: Ms =ZGL M N Xi≤K uα(i)β(i)!s By using the Weingarten formula, we can write this quantity as follows: uα(i1)β(i1) . . . uα(is)β(is) Lσ∨νδπ(α(i1), . . . , α(is))δτ (β(i1), . . . , β(is))WsM (π, σ)WsN (τ, ν) Ms = ZGL M N Xi1...is≤K = Xi1...is≤KXπστ ν = Xπστ ν Xi1...is≤K δπ(i)δτ (i)! Lσ∨νWsM (π, σ)WsN (τ, ν) But, as explained in the proof of Theorem 5.3, the coefficient on the left in the last (cid:3) formula equals K π∨τ . We therefore obtain the formula in the statement. We can further advance in the classical/twisted and free cases, where the Weingarten theory for the corresponding quantum groups is available from [1], [4], [8]. The result here, which justifies our various "liberation" claims, is as follows: Theorem 6.2. In the context of the liberation operations OL M N → H sL+ H sL M N , the laws of the sums of non-overlapping coordinates, M N → OL+ M N , U L M N → U L+ M N , χE = X(ij)∈E uij are in Bercovici-Pata bijection, in the E = κN, L = λN, M = µN, N → ∞ limit. Proof. We use the general theory in [1], [4], [8]. According to Proposition 6.1 above, in terms of K = E, the moments of the variables in the statement are given by: Ms = Xπστ ν K π∨τ Lσ∨νWsM (π, σ)WsN (τ, ν) We use now two standard facts, namely the fact that in the N → ∞ limit the Weingarten matrix WsN is concentrated on the diagonal, and the fact that we have π ∨ σ ≤ π+σ In the regime K = , with equality precisely when π = σ. See [8]. 2 LIBERATION THEORY FOR NONCOMMUTATIVE HOMOGENEOUS SPACES 23 κN, L = λN, M = µN, N → ∞ from the statement, we therefore obtain: K π∨τ Lπ∨τ M −πN −τ K πLπM −πN −π Ms ≃ Xπτ ≃ Xπ = Xπ (cid:18)κλ µ(cid:19)π In order to interpret this formula, we use general theory from [4], [21]: (1) For GN = ON , ¯ON /O+ N , the above variables χE follow to be asymptotically Gauss- ian/semicircular, of parameter κλ µ , and hence in Bercovici-Pata bijection. (2) For GN = UN , ¯UN /U + N the situation is similar, with χE being asymptotically com- plex Gaussian/circular, of parameter κλ µ , and in Bercovici-Pata bijection. (3) Finally, for GN = H s N /H s+ of parameter κλ µ , and once again in Bercovici-Pata bijection. N , the variables χE are asymptotically Bessel/free Bessel (cid:3) The convergence in the above result is of course in moments, and we do not know whether some stronger convergence results can be formulated. Nor do we know whether one can use linear combinations of coordinates which are more general than the sums χE that we consider. These are interesting questions, that we would like to raise here. References [1] T. Banica, Liberations and twists of real and complex spheres, J. Geom. Phys. 96 (2015), 1 -- 25. [2] T. Banica, The algebraic structure of quantum partial isometries, Infin. Dimens. Anal. Quantum Probab. Relat. Top. 19 (2016), 1 -- 36. [3] T. Banica, A duality principle for noncommutative cubes and spheres, J. Noncommut. Geom., to appear. [4] T. Banica, S.T. Belinschi, M. Capitaine and B. Collins, Free Bessel laws, Canad. J. Math. 63 (2011), 3 -- 37. [5] T. Banica, J. Bichon and B. Collins, The hyperoctahedral quantum group, J. Ramanujan Math. Soc. 22 (2007), 345 -- 384. [6] T. Banica and D. Goswami, Quantum isometries and noncommutative spheres, Comm. Math. Phys. 298 (2010), 343 -- 356. [7] T. Banica, A. Skalski and P.M. So ltan, Noncommutative homogeneous spaces: the matrix case, J. Geom. Phys. 62 (2012), 1451 -- 1466. [8] T. Banica and R. Speicher, Liberation of orthogonal Lie groups, Adv. Math. 222 (2009), 1461 -- 1501. [9] H. Bercovici and V. Pata, Stable laws and domains of attraction in free probability theory, Ann. of Math. 149 (1999), 1023 -- 1060. [10] F. Boca, Ergodic actions of compact matrix pseudogroups on C∗-algebras, Ast´erisque 232 (1995), 93 -- 109. [11] A. Chirvasitu, Quantum rigidity of negatively curved manifolds, preprint 2015. 24 TEODOR BANICA [12] B. Collins and P. ´Sniady, Integration with respect to the Haar measure on the unitary, orthogonal and symplectic group, Comm. Math. Phys. 264 (2006), 773 -- 795. [13] S. Curran and R. Speicher, Quantum invariant families of matrices in free probability, J. Funct. Anal. 261 (2011), 897 -- 933. [14] K. De Commer and M. Yamashita, Tannaka-Krein duality for compact quantum homogeneous spaces. I. General theory, Theory Appl. Categ. 28 (2013), 1099 -- 1138. [15] A. Freslon, On the partition approach to Schur-Weyl duality and free quantum groups, preprint 2014. [16] D. Goswami and S. Joardar, Rigidity of action of compact quantum groups on compact, connected manifolds, preprint 2013. [17] I. Klep, V. Vinnikov and J. Volcic, Null- and Positivstellensatze for rationally resolvable ideals, preprint 2015. [18] C. Kostler, R. Speicher, A noncommutative de Finetti theorem: invariance under quantum permu- tations is equivalent to freeness with amalgamation, Comm. Math. Phys. 291 (2009), 473 -- 490. [19] S. Malacarne, Woronowicz's Tannaka-Krein duality and free orthogonal quantum groups, preprint 2016. [20] S. Neshveyev and L. Tuset, Compact quantum groups and their representation categories, SMF (2013). [21] A. Nica and R. Speicher, Lectures on the combinatorics of free probability, Cambridge Univ. Press (2006). [22] P. Podle´s, Quantum spheres, Lett. Math. Phys. 14 (1987), 193 -- 202. [23] P. Podle´s, Symmetries of quantum spaces. Subgroups and quotient spaces of quantum SU(2) and SO(3) groups, Comm. Math. Phys. 170 (1995), 1 -- 20. [24] S. Raum, Isomorphisms and fusion rules of orthogonal free quantum groups and their complexifica- tions, Proc. Amer. Math. Soc. 140 (2012), 3207 -- 3218. [25] S. Raum and M. Weber, The full classification of orthogonal easy quantum groups, Comm. Math. Phys. 341 (2016), 751 -- 779. [26] P.M. So ltan, On actions of compact quantum groups, Illinois J. Math. 55 (2011), 953 -- 962. [27] R. Speicher, Multiplicative functions on the lattice of noncrossing partitions and free convolution, Math. Ann. 298 (1994), 611 -- 628. [28] P. Tarrago and M. Weber, The classification of tensor categories of two-colored noncrossing parti- tions, preprint 2015. [29] P. Tarrago and M. Weber, Unitary easy quantum groups: the free case and the group case, preprint 2015. [30] D.V. Voiculescu, K.J. Dykema and A. Nica, Free random variables, AMS (1992). [31] S. Wang, Free products of compact quantum groups, Comm. Math. Phys. 167 (1995), 671 -- 692. [32] S. Wang, Quantum symmetry groups of finite spaces, Comm. Math. Phys. 195 (1998), 195 -- 211. [33] M. Weber, On the classification of easy quantum groups, Adv. Math. 245 (2013), 500 -- 533. [34] D. Weingarten, Asymptotic behavior of group integrals in the limit of infinite rank, J. Math. Phys. 19 (1978), 999 -- 1001. [35] S.L. Woronowicz, Compact matrix pseudogroups, Comm. Math. Phys. 111 (1987), 613 -- 665. [36] S.L. Woronowicz, Tannaka-Krein duality for compact matrix pseudogroups. Twisted SU(N) groups, Invent. Math. 93 (1988), 35 -- 76. T.B.: Department of Mathematics, Cergy-Pontoise University, 95000 Cergy-Pontoise, France. [email protected]
1612.07257	2	1612	2017-06-16T17:53:26	Obstructions to lifting cocycles on groupoids and the associated $C^*$-algebras	[ "math.OA" ]	Given a short exact sequence of locally compact abelian groups $0 \to A \to B \to C \to 0$ and a continuous $C$-valued $1$-cocycle $\phi$ on a locally compact Hausdorff groupoid $\Gamma$ we construct a twist of $\Gamma$ by $A$ that is trivial if and only if $\phi$ lifts. The cocycle determines a strongly continuous action of $\widehat{C}$ into $\operatorname{Aut} C^(\Gamma)$ and we prove that the $C^$-algebra of the twist is isomorphic to the induced algebra of this action if $\Gamma$ is amenable. We apply our results to a groupoid determined by a locally finite cover of a space $X$ and a cocycle provided by a \v{C}ech 1-cocycle with coefficients in the sheaf of germs of continuous $\mathbb{T}$-valued functions. We prove that the $C^*$-algebra of the resulting twist is continuous trace and we compute its Dixmier-Douady invariant.	math.OA	math	OBSTRUCTIONS TO LIFTING COCYCLES ON GROUPOIDS AND THE ASSOCIATED C∗-ALGEBRAS MARIUS IONESCU AND ALEX KUMJIAN ABSTRACT. Given a short exact sequence of locally compact abelian groups 0 → A → B → C → 0 and a continuous C-valued 1-cocycle φ on a locally compact Hausdorff groupoid Γ we construct a twist of Γ by A that is trivial if and only if φ lifts. The cocycle determines a strongly continuous action of bC into Aut C ∗(Γ) and we prove that the C ∗-algebra of the twist is isomorphic to the induced algebra of this action if Γ is amenable. We apply our results to a groupoid determined by a locally finite cover of a space X and a cocycle provided by a Cech 1-cocycle with coefficients in the sheaf of germs of continu- ous T-valued functions. We prove that the C ∗-algebra of the resulting twist is continuous trace and we compute its Dixmier-Douady invariant. 1. INTRODUCTION It is a well-known fact that given a 1-cocycle φ : Γ → T on an ´etale groupoid Γ, there is an automorphism α of C∗(Γ) such that α(f )(γ) = φ(γ)f (γ) for all f ∈ Cc(Γ) (see [19, Proposition II.5.1]). If φ can be lifted to an R-valued 1- cocycle φ, there is a strongly continuous 1-parameter group of automorphisms α : R → Aut C∗(Γ) such that α = α1. But in general there is a cohomological obstruction to lifting φ. There is a central groupoid extension Σφ of Γ by Z, called a twist, which is trivial precisely when φ can be lifted. Our goal in this paper is to describe the structure of C∗(Σφ) in terms of C∗(Γ) and the automorphism α. Our results will be proven in a somewhat more general form. Given a lo- cally compact Hausdorff groupoid Γ, a short exact sequence of locally compact abelian groups 0 → A i−→ B p −→ C → 0 and a continuous 1-cocycle φ : Γ → C we construct a twist Σφ of Γ by A which is trivial if and only if the cocycle lifts. As above the cocycle φ determines isomorphic to the induced algebra for this action if Γ is amenable (see Theo- a strongly continuous action αφ : bC → Aut C∗(Γ). We prove that C∗(Σφ) is rem 4.3). So in particular, there is an action γ : bB → Aut C∗(Σφ) such that C∗(Σφ) ⋊γ bB is strong Morita equivalent to C∗(Γ) ⋊ In example 3.6 we consider a groupoid Γ determined by a locally finite open cover U := {Ui}i∈I of a space X (see [17]) and take A := Z, B := R and C := T. The 1-cocycle φ arises from a Cech 1-cocycle λ = {λij }i,j∈I with coefficients in αφ bC. The work of the authors was partially supported by their individual grants from the Simons Foundation (#209277 to Marius Ionescu and # to Alex Kumjian). This work was supported by Simons Foundation Collaboration grants #209277 (MI) and #353626 (AK). 1 2 MARIUS IONESCU AND ALEX KUMJIAN T , the sheaf of germs of continuous T-valued functions. If λ satisfies a certain liftability hypothesis, the corresponding twist Σφ is then determined by a Cech 2-cocycle λ⋆ with coefficients in the constant sheaf with fiber Z (we denote the sheaf by Z for the sake of simplicity) that measures the obstruction to lifting λ to a Cech 1-cocycle with coefficients in the sheaf of germs of continuous R- valued functions. Finally we show in example 4.6 that C∗(Σφ) is a continuous trace algebra and compute its Dixmier-Douady invariant using the cohomology class [λ] which is identified with [λ⋆] via the standard isomorphism of Cech cohomology groups H 1(X, T ) ∼= H 2(X, Z). This example was a key motivation for this project and it was suggested by results of [16] (see also [22]). In this note we assume that all topological spaces and groupoids are sec- ond countable locally compact and Hausdorff and all groups are second count- able abelian locally compact and Hausdorff. Hence all spaces are paracompact. Moreover we assume that all groupoids are endowed with a Haar system. The second author would like to thank the first author and his colleagues at the Naval Academy for their hospitality and support during a recent visit. 2. PRELIMINARIES In this section we present first a characterization of the induced algebra from an action of a closed subgroup of a locally compact abelian group on a C∗-algebra. Our characterization is essential in our proof of the main result, Theorem 4.3. For a related characterization of the induced algebra see [6]. Second, we provide conditions that guarantee that the natural map j as defined in equation (1) from the C∗-algebra of a subgroupoid into the multiplier algebra of the C∗-algebra of the groupoid is faithful. Our results generalize well known facts about group C∗-algebras. 2.1. A characterization of the induced algebra. Let H be a closed sub- group of a locally compact abelian group G, let D be a C∗-algebra and let α : H → Aut D be a strongly continuous action. If f : G → D is a continuous function such that f (x − h) = αh(f (x)) for all h ∈ H, x ∈ G then x + H 7→ kf (x)k yields a well-defined continuous function. We define the induced C∗-algebra (see [22, §3.6]) by indG H (D, α) := {f : G → D continuous f (x − h) = αh(f (x)), h ∈ H, x ∈ G and x + H 7→ kf (x)k ∈ C0(H/G)}. There is a canonical translation action β : G → Aut(indG H(D, α)) given by βg(f )(x) := f (x − g). By [16, Lemma 3.1] D ⋊α H is strong Morita equiva- lent to indG H (D, α) ⋊β G. Point evaluation at 0 yields a surjective homomorphism π : indG H (D, α) → D given by π(f ) = f (0). There is a natural translation action τ : G → Aut(C0(G/H)) (where τg(k)(x + H) = k(x − g + H)) and a G-equivariant homomorphism j : C0(G/H) → M (indG H (D, α)) determined by pointwise multiplication. It is easy to check that these maps satisfy the following conditions for all f ∈ indG H (A, α): OBSTRUCTIONS TO LIFTING COCYCLES ON GROUPOIDS AND THE ASSOCIATED C ∗-ALGEBRAS 3 i. For all k ∈ C0(G/H), π(j(k)f ) = k(H)π(f ); ii. for all h ∈ H, π(βh(f )) = αh(π(f )); iii. if π(βg(f )) = 0 for all g ∈ G, then f = 0; iv. and lim x+H→∞ kf (x)k = 0. We show below that these conditions characterize the induced algebra. Note H (D, α) may be identified with the section algebra of a continuous C∗- that indG bundle over G/H with fibers isomorphic to D and factor maps πx+H . Theorem 2.2. Let H be a closed subgroup of a locally compact abelian group G. Let D and E be C∗-algebras, α : H → Aut D and γ : G → Aut E strongly continuous actions, ρ : E → D a surjective homomorphism, and i : C0(G/H) → Z(M (E)) a G-equivariant homomorphism. Suppose that i For all k ∈ C0(G/H) and e ∈ E, ρ(i(k)e) = k(H)ρ(e); ii For all h ∈ H and e ∈ E, ρ(γh(e)) = αh(ρ(e)); iii If ρ(γg(e)) = 0 for all g ∈ G, then e = 0; iv For all e ∈ E, lim kρ(γx(e))k = 0. x+H→∞ Then there is a (unique) G-equivariant isomorphism Ψ : E → indG that: π ◦ Ψ = ρ and j(k)Ψ(e) = Ψ(i(k)e) for all k ∈ C0(G/H) and e ∈ E. H (D, α) such Proof. For e ∈ E, we define a function Ψ(e) : G → D by Ψ(e)(x) := ρ(γ−x(e)). It is straightforward to check that Ψ(e) is continuous and that Ψ(e) ∈ indG H (D, α). Indeed we have Ψ(e)(x − h) = ρ(γh−x(e)) = ρ(γh(γ−x(e))) = αh(ρ(γ−x(e))) = αh(Ψ(e)(x)). It is also routine to check that the map thus defined Ψ : E → indG H (D, α) is an equivariant ∗-homormorphism. Injectivity follows from (iii) above. Let k ∈ C0(G/H) and e ∈ E. Then by (i) above and the G-equivariance of i it follows that for all x ∈ G Ψ(i(k)e)(x) = ρ(γ−x(i(k)e)) = ρ(i(τ−x(k))γ−x(e)) = τ−x(k)(H)ρ(γ−x(e)) = k(x + H)Ψ(e)(x) = (j(k)Ψ(e))(x), and hence j(k)Ψ(e) = Ψ(i(k)e). Thus Ψ is a map of C0(G/H)-algebras. Since ρ(e) = Ψ(e)(0) for all e ∈ E we have π ◦ Ψ = ρ. Next we use Proposition C.24 of [22] (see also [7, Proposition 14.1]), which states that a submodule of continuous sections of an upper semicontinuous C∗- bundle which is fiberwise dense must also be norm dense in the C∗-algebra of sections vanishing at infinity, to prove that Ψ(E) is dense in indG H (D, α). This will prove that Ψ is surjective and therefore an isomorphism. We check the hypotheses a) and b) of the proposition. Let f ∈ Ψ(E) and let k ∈ C0(G/H), then f = Ψ(e) for some e ∈ E and so j(k)f = j(k)Ψ(e) = Ψ(i(k)e) ∈ Ψ(E). This proves condition (a). Now since each factor map is conjugate to π ◦ βx for some x ∈ G, its restriction to Ψ(E) is surjective. Thus (b) holds and the result follows. (cid:3) 4 MARIUS IONESCU AND ALEX KUMJIAN Remark 2.3. Let D1 and D2 be C∗-algebras. A homomorphism φ : D1 → M (D2) is said to be nondegenerate if it extends uniquely to a unital map φ : M (D1) → M (D2) or equivalently if {φ(eλ)}λ converges strictly to the unit of M (D2) for every approximate identity {eλ}λ in D1. Property (iv) above is equiv- alent to the requirement that i : C0(G/H) → M (E) be nondegenerate. Remark 2.4. Note that since C0(G/H) ∼= C∗([G/H)), the map i is determined by a strictly continuous homomorphism u : [G/H → U M (E) where U M (E) is the unitary group of the multiplier algebra M (E). The nondegeneracy of i is equivalent to the requirement that u0 = 1 and condition (i) above is satisfied if and only if ρ(uχe) = ρ(e) for all χ ∈ [G/H and e ∈ E. Remark 2.5. Our characterization of the induced algebra is provided to fa- cilitate the proof of our main result, namely, that the C∗-algebra of a certain groupoid extension regarded as an obstruction to lifting a cocycle is an induced C∗-algebra. One could in principle use Echterhoff's more general characteri- zation (see the main theorem of [6]), but this would have required the identi- fication of the C∗-algebra of the quotient groupoid with a certain quotient of the putative induced algebra. The proof of this identification would use sim- ilar techniques with those in our proof and it would not necessarily lead to a simplification of our arguments. 2.6. Multiplier algebras of groupoid C∗-algebras. Given a C∗-algebra D let M (D) denote its multiplier algebra. We view M (D) as the C∗-algebra L(DD) of adjointable maps on the Hilbert C∗-module DD (see, for example, [18, Section 2.3] and [12]). Recall that DD is a full Hilbert C∗-module via d · e = de and hd, eiD = d∗e. Assume that Σ is a closed subgroupoid of a locally compact Hausdorff groupoid Γ such that Σ0 = Γ0. Assume that Σ is endowed with a Haar system β = {βu}u∈Γ0 and that Γ is endowed with a Haar system λ = {λu}u∈Γ0 . Then there is a ∗-homomorphism j : C∗(Σ) → M (C∗(Γ)) defined for a ∈ Cc(Σ) and f ∈ Cc(Γ) via (1) (j(a)(f ))(γ) =ZΣ a(η)f (η−1γ) d βr(γ)(η) such that j(a)∗ = j(a∗) (see [19, Proposition II.2.4]). It is known that if Σ and Γ are locally compact groups then the map j fails to be faithful in general (see, for example, [4]). However, if Σ is a clopen subgroup of Γ or Γ is an amenable group then the map j is a faithful *-homomorphism of C∗(Σ) into C∗(Γ) (see [21, Proposition 1.2], [4, Theorem 1.3], [3, Corollary 1.5]). We prove next that the above mentioned results hold for groupoid C∗-algebras as well: the map j is faithful if Σ is a clopen subgroupoid of Γ or if Γ is an amenable groupoid. Before proceeding further, we recall the definition of induced representa- tions from closed subgroupoids following [9, Section 2] (see also [19, Section II.2]). Let ΣΓ := Γ ∗ Γ/Σ be the imprimitivity groupoid. Then Cc(Γ) is a pre- Cc(ΣΓ) − Cc(Σ)-imprimitivity bimodule with actions and inner products given by OBSTRUCTIONS TO LIFTING COCYCLES ON GROUPOIDS AND THE ASSOCIATED C ∗-ALGEBRAS 5 F · ϕ(z) =ZΓ ϕ · g(z) =ZΣ hϕ, ψi∗(h) =ZΓ ∗hϕ, ψi(cid:0)[x, y](cid:1) =ZΣ F(cid:0)[z, y](cid:1)ϕ(y) dλs(z)(y) ϕ(zh)g(h−1) dβs(z)(h) ϕ(y)ψ(yh) dλr(h)(y) ϕ(xh)ψ(yh) dβs(x)(h). If L is a representation of C∗(Σ) on B(HL) then the induced representation IndΓ Σ L acts on the completion HInd L of Cc(Γ)⊙ HL with respect to the pre-inner product given on elementary tensors by (2) (ϕ ⊗ h , ψ ⊗ k) = (L(hψ, ϕi∗)h , k). If ϕ ⊗Σ h denotes the class of ϕ ⊗ h in HInd L, then the induced representation is given by (3) for f ∈ Cc(Γ), where IndΓ Σ L(f )(ϕ ⊗Σ h) = f ∗ ϕ ⊗Σ h f ∗ ϕ(γ) =ZΓ f (η)ϕ(η−1γ) dλr(γ)(η). In the following we suppress Σ from ϕ ⊗Σ h to simplify slightly the notation. 2.6.1. The clopen case. Assume first that Σ is a clopen subgroupoid of Γ. Then the restriction of the Haar system λ = {λu}u∈Γ0 to Σ is a Haar system on Σ. We assume in the following that Σ is endowed with this Haar system, that is, β = λΣ. Then the map iΣ : Cc(Σ) → Cc(Γ) defined via iΣ(f )(γ) =(f (γ) 0 if γ ∈ Σ otherwise is a well defined faithful ∗-homomorphism. We prove next that iΣ extends to a faithful ∗-homomorphism of C∗(Σ) into C∗(Γ), generalizing Proposition 1.2 of [21]. Proposition 2.7. With notation as above, iΣ extends to an embedding iΣ : C∗(Σ) → C∗(Γ) and, therefore, C∗(Σ) can be viewed as a subalgebra of C∗(Γ). Proof. We need to show that kiΣ(f )kC ∗(Γ) = kf kC ∗(Σ) for all f ∈ Cc(Σ). Let L be a representation of C∗(Γ). Renault's disintegration theorem (see [20, Propo- sition 4.2] and also [13, Section 7]) implies that L is the integrated form of a unitary representation (µ, Γ0 ∗ H, V ) of Γ. Since Σ is a clopen subgroupoid of Γ, any unitary representation of Γ restricts to a unitary representation of Σ. Therefore kf kC ∗(Σ) ≥ kiΣ(f )kC ∗(Γ) for all f ∈ Cc(H). For the converse inequality, let (L, HL) be a representation of C∗(Σ). Since Σ is a closed subgroupoid of Γ, the representation L can be induced to a rep- resentation IndΓ Σ L of C∗(Γ) as in (3). Let Hres be the closed subspace of HInd L obtained by completing Cc(Σ) ⊙ HL via the inner product in (2). Then U : Hres → HL defined on elementary tensors via U (ϕ ⊗ h) = L(ϕ)h defines a unitary that intertwines L with a subrepresentation of IndΓ Σ L restricted to 6 MARIUS IONESCU AND ALEX KUMJIAN C∗(Σ). It follows that kf kC ∗(Σ) ≤ kiΣ(f )kC ∗(Γ) for all f ∈ Cc(Σ). Therefore kiΣ(f )kC ∗(Γ) = kf kC ∗(Σ) for all f ∈ Cc(Σ) and one can view C∗(Σ) as a subalge- bra of C∗(Γ). (cid:3) 2.7.1. The amenable case. We assume now that (Σ, β) is a closed subgroupoid of (Γ, λ) and Γ is amenable in the sense of [2]. It follows that Σ is amenable as well (see [2, Proposition 5.1.1]). Proposition 2.8. Assume that (Σ, β) is a closed subgroupoid of an amenable groupoid (Γ, λ) such that Σ0 = Γ0. Then the map j defined in (1) is faithful. We recall first the definition of the left regular representation of C∗(Γ). Let µ be a quasi-invariant measure on Γ0 with full support. The left regular rep- resentation of C∗(Γ) is the representation LΓ := IndΓ Γ0 µ induced from µ. By using induction in stages (see [9, Theorem 4]) it follows that if Σ is a closed subgroupoid of Γ such that Σ0 = Γ0 then LΓ is unitarily equivalent to IndΓ Σ LΣ. Proof of Proposition 2.8. The proof is virtually identical with the group case (see, for example, the proof of [4, Thm. 1.3]): by the amenability of Σ, any representation of C∗(Σ) is weakly contained in the left regular representation of C∗(Σ), which is itself contained in the restriction to C∗(Σ) of the left regular representation of C∗(Γ). (cid:3) The authors would like to thank Alcides Buss and Dana P. Williams for useful conversations and suggestions that led to a significant shortening of our original proof of Proposition 2.8. 3. TWISTS AND SHORT EXACT SEQUENCES Let Γ be a locally compact Hausdorff groupoid and let G be a locally compact Hausdorff abelian group. The set of continuous 1-cocyles from Γ to G is defined via ZΓ(G) = Z 1(Γ, G) := { φ : Γ → G φ(γ1γ2) = φ(γ1) + φ(γ2) for all (γ1, γ2) ∈ Γ2 }. Then ZΓ(G) is an abelian group and the map G 7→ ZΓ(G) is a functor. Definition 3.1. Let A be an abelian group and Γ a groupoid. A twist by A over Γ is a central groupoid extension Γ0 × A j −→ Σ π−→ Γ, where Σ0 = Γ0, j is injective, π is surjective, and j(r(σ), a)σ = σj(s(σ), a) for all σ ∈ Σ and a ∈ A. Example 3.2. The semi-direct product Γ × A of Γ and A is called the trivial twist. Recall from [10] that (γ1, a1)(γ2, a2) = (γ1γ2, a1+a2) provided that s(γ1) = r(γ2), and (γ, a)−1 = (γ−1, −a). Then Γ × A is a twist by A via j0(u, a) = (u, a) for (u, a) ∈ Γ0 × A, and π0(γ, a) = γ for (γ, a) ∈ Γ × A. Following Definition 2.5 of [10] we say that two twists by A are properly isomorphic if there is a twist morphism between them which preserves the inclusion of Γ0 × A. The following lemma is a generalization of Proposition 2.2 of [10]. OBSTRUCTIONS TO LIFTING COCYCLES ON GROUPOIDS AND THE ASSOCIATED C ∗-ALGEBRAS 7 Lemma 3.3. A twist Σ by A is properly isomorphic to a trivial twist if and only if there is a groupoid homomorphism τ : Γ → Σ such that πτ = idΓ. Proof. If τ : Γ × A → Σ is a twist isomorphism then we can define τ : Γ → Σ via τ (γ) = τ (γ, 0A), where 0A is the identity of A. It is easy to check that τ is a groupoid homomorphism and that πτ = idΓ. Assume now that there is a groupoid homomorphism τ : Γ → Σ such that πτ = idΓ. Then we can define τ : Γ × A → Σ via τ (γ, a) = τ (γ)j(s(γ), a). We check next that τ is a twist isomorphism. Let (u, a) ∈ Γ0 × A. Then τ (j0(u, a)) = τ (u, a) = τ (u)j(u, a) = τ (u)τ (u)j(u, a) = τ (u)j(u, a)τ (u) = j(u, a). If (γ, a) ∈ Γ × A then π(τ (γ, a)) = π(τ (γ)j(s(γ), a)) = γπ(j(s(γ), a)) = γ = π0(γ, a). (cid:3) Following [10] we write TΓ(A) for the collection of proper isomorphism classes of twists by A and we write [Σ] ∈ TΓ(A). We endow TΓ(A) with the opera- tion [Σ] + [Σ′] := [∇A (Σ ∗Γ Σ′)], where ∇A ∈ HomΓ(A ⊕ A, A) is defined via ∗ ∇A(a, a′) = a + a′ (see [10, Proposition 2.6]). Then TΓ(A) is an abelian group with neutral element [Γ × A]. It can be shown that A 7→ TΓ(A) is a half-exact functor. Suppose that B and C are locally compact abelian groups. If p : B → C is a homomorphism and φ ∈ ZΓ(C), then the obstruction twist determined by φ is defined via (4) Σφ = {(γ, b) ∈ Γ × B : φ(γ) = p(b)}. We establish that Σφ is indeed a twist in the following proposition and show that it has a Haar system in the next section (see equation (5) in Section 4). If Γ is an ´etale groupoid, it has a basis consisting of open bisections, that is, open subsets to which the restrictions of both the range and source maps are injective. Proposition 3.4. Assume that p : B → C is a surjective homomorphism of locally compact abelian groups and let φ ∈ ZΓ(C). Then Σφ is a closed sub- groupoid of Γ × B and it is a twist of Γ by A := ker p. Σφ is properly isometric φ. The to the trivial twist if and only if there is φ ∈ ZΓ(B) such that φ = p∗ projection π1 : Σφ → Γ is a surjective groupoid morphism. If Γ is ´etale and A is discrete then Σφ is ´etale and π1 is a local homeomorphism. Proof. Note that Σφ is a closed subgroupoid of Γ×B because φ and p are contin- uous maps. Since p is surjective it follows that Σφ is a twist by A via j(u, a) = (u, a) and π(γ, b) = γ. We make the obvious identification Σ0 φ = Γ0 × {0B} ≃ Γ0, where 0B is the unit of B. If Σφ is properly isometric to the trivial twist then Lemma 3.3 implies that there is a groupoid homomorphism τ : Γ → Σφ such that π1τ = idΓ. Therefore there is φ ∈ ZΓ(B) such that τ (γ) = (γ, φ(γ)) and, φ for some φ ∈ ZΓ(B), then define hence, φ = p∗ τ : Γ → Σφ via τ (γ) = (γ, φ(γ)). It follows that π1τ = idΓ and Lemma 3.3 implied that Σφ is properly isometric to the trivial twist. φ. Conversely, if φ = p∗ It is easy to check that π1 is a groupoid morphism. Moreover, π1 is surjective since p is surjective. Now suppose that Γ is ´etale and A is discrete. Since π1 : Σφ → Γ is open, in order to prove that π1 is a local homeomorphism it suffices to show that it 8 MARIUS IONESCU AND ALEX KUMJIAN is locally injective. Let (γ, b) ∈ Σφ. Then there are an open bisection U in Γ such that γ ∈ U and a neighborhood V0 of 0B such that V0 ∩ ker p = {0B}. Let V be an open neighborhood of b such that V − V ⊆ V0. Then W := (U × V ) ∩ Σφ is an open neighborhood of (γ, b) in Σφ. Given (γ1, b1), (γ2, b2) ∈ W such that π1(γ1, b1) = π1(γ2, b2), then γ1 = γ2 and p(b1 − b2) = p(b1) − p(b2) = φ(γ1) − φ(γ2) = 0. Hence b1 − b2 ∈ (V − V ) ∩ ker p ⊂ V0 ∩ ker p = {0B} and so b1 = b2. Thus the restriction of π1 to W is injective and therefore a local homeomorphism. It follows easily that Σφ is ´etale. (cid:3) The following result generalizes an initial segment of the exact sequence given in [10, Proposition 3.3]. Proposition 3.5. Given a short exact sequence of locally compact abelian groups 0 → A i−→ B p −→ C → 0, there is an exact sequence 0 → ZΓ(A) i∗−→ ZΓ(B) p∗−→ ZΓ(C) δ−→ TΓ(A), where δ(φ) := [Σφ] for φ ∈ ZΓ(C). Proof. Since ZΓ is left exact we only need to show that Im p∗ = ker δ. We have that φ ∈ ker δ if and only if Σφ is properly isometric to the trivial twist. By Lemma 3.4, this happens if and only if φ = p∗ (cid:3) φ for some φ ∈ ZΓ(B). In the following example we consider the case A = Z, B = R and C = T. In a certain case where the groupoid Γ is equivalent to a space X, we indicate how the construction of the groupoid Σφ is related to the boundary map of Cech co- homology H 1(X, T ) → H 2(X, Z), where H 1(X, T ) is the first Cech cohomology of X with coefficients in T , the sheaf of germs of continuous T-valued func- tions, and H 2(X, Z) is the second Cech cohomology of X with coefficients in the constant sheaf with fiber Z (see [18, §4.1] for background on Cech cohomology). This example is inspired by results in [16]. Example 3.6. Let X be a Hausdorff second countable locally compact space; thus X is also paracompact. Let U := {Ui}i∈I be a locally finite open cover of X. Set Uij := Ui ∩ Uj for i, j ∈ I and similarly Uijk := Ui ∩ Uj ∩ Uk for i, j, k ∈ I. We now construct an ´etale groupoid associated to U (see [17, §1, Remark 3] and [8, Example III.1.0]). Let ΓU := {(x, i, j) : x ∈ Uij} and Γ0 := {(x, i, i) : x ∈ Ui}; U note that Γ0 U may be identified with the disjoint unionFi∈I Ui and that ΓU may be identified with the disjoint unionFi,j∈I Uij. It is routine to check that with the topology obtained from this identification and with the following structure maps ΓU is an ´etale groupoid: s(x, i, j) = (x, j, j) r(x, i, j) = (x, i, i) (x, i, j)−1 = (x, j, i) (x, i, j)(x, j, k) = (x, i, k). OBSTRUCTIONS TO LIFTING COCYCLES ON GROUPOIDS AND THE ASSOCIATED C ∗-ALGEBRAS 9 homeomorpismFi∈I Ui → X (see [11]). Next suppose that we are given a Cech Note that ΓU may also be viewed as the natural groupoid associated to the local 1-cocycle λ = {λij}i,j∈I with coefficients in T . So λij : Uij → T is continuous for all i, j ∈ I and λik(x) = λij (x)λjk(x) for all i, j, k ∈ I and x ∈ Uijk. It is routine to check that the map φ : ΓU → T given by φ((x, i, j)) = λij (x) is a continuous groupoid 1-cocycle. Now suppose that each λij lifts, that is, for each i, j ∈ I there is a continuous function λij : Uij → R such that λij = e ◦ λij where e(t) = e2π√−1t for t ∈ R. We may assume that λii = 0 for all i ∈ I. We observe that for i, j, k ∈ I the formula (λ⋆)ijk(x) := λij (x) + λjk(x) − λik(x) for x ∈ Uijk defines a continuous integer-valued function. A routine computation shows that for all i, j, k, ℓ ∈ I and x ∈ Uijkℓ (λ⋆)jkℓ(x) − (λ⋆)ikℓ(x) + (λ⋆)ijℓ(x) − (λ⋆)ijk(x) = 0, that is, (λ⋆)ijk gives a Cech 2-cocycle with values in Z (i.e. the constant sheaf with fiber Z). It is normalized in the sense that (λ⋆)ijk = 0 if j = i or j = k. As in [17] we may construct a groupoid 2-cocycle φ⋆ by the formula φ⋆((x, i, j), (x, j, k)) = (λ⋆)ijk(x) for all x ∈ Uijk. We obtain a twist Σ by Z over ΓU determined by φ⋆ (see [19, Prop. I.1.14]). Indeed, let Σ := {(n, (x, i, j)) : n ∈ Z, x ∈ Uij}. We identify Γ0 U with Σ0 via the map (x, i, i) 7→ (0, (x, i, i)). The range and source maps factor through those of ΓU . The remaining structure maps are given by (m, (x, i, j))(n, (x, j, k)) := (m + n + (λ⋆)ijk(x), (x, i, k)) (n, (x, i, j))−1 := (−n − (λ⋆)jij (x), (x, j, i)). We claim that Σ ∼= Σφ. Define ξ : Σ → ΓU × R by ξ((n, (x, i, j))) = ((x, i, j), λij (x) + n). We first note that ξ induces an identification of the unit spaces. Given a pair of composable elements (m, (x, i, j)), (n, (x, j, k)) ∈ Σ we have ξ((m, (x, i, j))(n, (x, j, k))) = ξ((m + n + (λ⋆)ijk(x), (x, i, k))) = ((x, i, k), λik(x) + m + n + (λ⋆)ijk(x)) = ((x, i, k), m + n + λij (x) + λjk(x)) = ((x, i, j), λij (x) + m)((x, j, k), λjk(x) + n) = ξ((m, (x, i, j)))ξ((n, (x, j, k))). Hence, ξ is a groupoid homomorphism. Moreover ξ is injective and its image coincides with Σφ. It follows that the isomorphism class of Σ does not depend on the specific choice of the λij. Remark 3.7. With notation as in the above example any (normalized) Cech 2- cocycle with values in the constant sheaf with fiber Z gives rise to a groupoid as above which is isomorphic to a pullback. Let R be the sheaf of germs of contin- uous R-valued functions on X. Since R is a fine sheaf, we have H n(X, R) = 0 for n ≥ 1. Moreover since 0 → Z → R → T → 0 is exact, the connecting map of the long exact sequence of cohomology ∂n : H n(X, T ) → H n+1(X, Z) is 10 MARIUS IONESCU AND ALEX KUMJIAN an isomorphism for n > 0 (see [18, Theorem 4.37]). In Example 3.6 we have [λ⋆] = ∂1([λ]). Remark 3.8. Note that every Cech 2-cocycle with values in an arbitrary sheaf of abelian groups is cohomologous to a normalized cocycle a = {aijk}ijk (that is, aijk = 0 if j = i or j = k). Indeed given a Cech 2-cocycle c = {cijk}ijk a routine calculation shows that ciij (x) = ciii(x) = cjii(x) for all i, j ∈ I and x ∈ Uij. For i, j ∈ I, define a 1-cochain λ by λij = 0 if i 6= j and λii(x) = ciii(x) for x ∈ Uii. Then a := c − d1λ is a normalized 2-cocycle cohomologous to c. (Recall that (d1λ)ijk(x) := λjk(x) − λik(x) + λij (x) for x ∈ Uijk). 4. THE STRUCTURE OF THE C∗-ALGEBRA C∗(Σφ) Let Γ be a locally compact Hausdorff groupoid endowed with Haar system {λu}u∈Γ0 and let 0 → A i−→ B p −→ C → 0, be a short exact sequence of locally compact abelian groups. In the sequel, we identify A with its image i(A) in B. Let φ ∈ ZΓ(C) and let Σφ be the obstruction bB twist as in (4). We prove below that C∗(Σφ) ∼= ind bC C∗(Γ) if Γ is amenable (see Theorem 4.3). We begin by describing the action of bC on C∗(Γ). The following lemma follows immediately from Proposition II.5.1(iii) of [19]. Lemma 4.1. Assume that C is a locally compact abelian group. Given φ ∈ ZΓ(C), the map αφ t (f )(γ) = ht, φ(γ)if (γ), Aut C∗(Γ). for f ∈ Cc(Γ), t ∈ bC, and γ ∈ Γ, defines a strongly continuous action αφ : bC → We define next a Haar system on Σφ. Choose Haar measures µA, µB, and µC on A, B, and, respectively C such that: ZB f (b) dµB(b) =ZCZA f (b + a) dµA(a)dµC (p(b)). One can always make such a choice since A, B, and C are locally compact abelian groups (see, e.g. [5, page 79]). Let π1 : Σφ → Γ be the projection map (see Lemma 3.4). Note that π−1 1 (γ) = {γ} × Aγ where Aγ := {b ∈ B : φ(γ) = p(b)} is a coset in B, since Aγ = bγ + A for some bγ ∈ B with φ(γ) = p(bγ). We write µγ for the measure defined on Aγ via ZAγ f (γ, b) dµγ(b) :=ZA f (γ, bγ + a) dµA(a). The measure µγ is independent of the choice of bγ because µA is a Haar mea- sure on A. We define now a Haar system {νu}u∈Γ0 on Σφ via (5) f (γ, b) dµγ(b)dλu(γ), ZΣu φ f (γ, b) dνu(γ, b) =ZΓuZAγ for all u ∈ Γ0. It is easy to check that (5) defines a Haar system on Σφ using the fact that {λu} is a Haar system on Γ and µA is a Haar measure on A. OBSTRUCTIONS TO LIFTING COCYCLES ON GROUPOIDS AND THE ASSOCIATED C ∗-ALGEBRAS11 The main goal of this section is to prove that C∗(Σφ) is isomorphic to the B C (C∗(Γ), αφ) using Theorem 2.2. We begin by defining a induced algebra ind map ρ : Cc(Σφ) → Cc(Γ) via (6) ρ(f )(γ) =ZAγ f (γ, b) dµγ(b). Lemma 4.2. With notation as above, the map ρ defined in (6) extends to a surjective ∗-homomorphism ρ : C∗(Σφ) → C∗(Γ) which factors through the map ι : C∗(Σφ) → M (C∗(Γ × B)) given in Subsection 2.6. Moreover, if either Γ is amenable or Σφ is a clopen subset of B × Γ, then ι is injective. Proof. We prove first that ρ is a ∗-homomorphism. Let f and g in Cc(Σφ) and let γ ∈ Γ. Then, using the fact that µA is a Haar measure on A, we obtain f (α, b′)g(α−1γ, −b′ + b) dµα(b′)dλr(γ)(α)dµγ (b) g(α−1γ, −b′ + b) dµγ(b)dµα(b′)dλr(γ)(α) ρ(f ∗ g)(γ) =ZAγ (f ∗ g)(γ, b) dµγ(b) f (α, b′)ZAγ =ZAγZΓr(γ)ZAα =ZΓr(γ)ZAα =ZΓr(γ) =(cid:0)ρ(f ) ∗ ρ(g)(cid:1)(γ). ρ(f )(α)ρ(g)(α−1γ) dλr(γ)(α) It is easy to show that ρ(f∗) = ρ(f )∗ for all f ∈ Cc(Σφ). Recall from Subsection 2.6 (see [19, Proposition II.2.4]) that there is a bounded ∗-homomorphism from C∗(Σφ) into M (C∗(Γ×B)). Since C∗(Γ×B) is ∗-isomorphic to C∗(Γ)⊗C∗(B) and C∗(B) is isomorphic to C0( B) it follows that M (C∗(Γ×B)) is ∗-isomorphic to Cb( B, M (C∗(Γ))) where M (C∗(Γ)) is given the strict topology (see [1, Corollary 3.4]). Then ρ is just the composition of the above bounded ∗-homomorphisms with evaluation at 0. Hence ρ extends to a bounded ∗- homomorphism ρ : C∗(Σφ) → M (C∗(Γ)). However, since ρ(Cc(Σφ)) ⊆ Cc(Γ), Cc(Σφ) is dense in C∗(Σφ), and Cc(Γ) is dense in C∗(Γ), it follows that ρ is a bounded ∗-homomorphism from C∗(Σφ) into C∗(Γ). To prove that ρ is surjective let b be a Bruhat approximate cross-section for B over A. Recall (see, for example, Proposition C.1 of [18]) that b : B → [0, ∞) is a continuous function such that supp b∩(K +A) is compact for every compact b(b + a)dµA(a) = 1 for all b ∈ B. Given g ∈ Cc(Γ) define f ∈ Cc(Σφ) via f (γ, b) = g(γ)b(b). It follows immediately that ρ(f ) = g and, hence, ρ is surjective. set K in B, and such thatRA The final assertion follows from Proposition 2.7 if Σφ is a clopen subset of Γ × B. If Γ is amenable the assertion follows from Proposition 2.8 and the fact that Γ × B is amenable (see [2, Prop. 5.1.2]). (cid:3) We are now ready to state and prove the main result of the paper. Theorem 4.3. Let Γ be a locally compact Hausdorff amenable groupoid en- dowed with Haar system {λu}u∈Γ0 and let 0 → A i−→ B p −→ C → 0, 12 MARIUS IONESCU AND ALEX KUMJIAN be a short exact sequence of locally compact abelian groups. Let φ ∈ ZΓ(C) and let Σφ be the obstruction twist as in (4) endowed with the Haar system defined in (5). Then there are a surjective ∗-homomorphism ρ : C∗(Σφ) → C∗(Γ), a strongly continuous action γ : B → Aut(C∗(Σφ)) and a B-equivariant homomorphism i : C0( B/ C) → Z(M (C∗(Σφ))) such that the four conditions of Theorem 2.2 are satisfied. Therefore there is a unique B-equivariant isomorphism Ψ : C∗(Σφ) → ind B C C∗(Γ) defined via for f ∈ Cc(Σφ) and t ∈ B. Ψ(f )(t) = γ−t(f ) Proof. The existence of the map ρ : C∗(Σφ) → C∗(Γ) is proved in Lemma 4.2. The strongly continuous action γ : B → Aut(C∗(Σφ)) is given by γt(f )(σ, b) = ht, bif (σ, b). In order to define the map i : C0( B/ C) → Z(M (C∗(Σφ))) we identify B/ C with A and we also identify C0( A) with C∗(A) via the Fourier transform. Recall that we view the multiplier algebra M (C∗(Σφ)) as the C∗-algebra of the adjointable operators on the Hilbert C∗-module C∗(Σφ)C ∗(Σφ). Then we define i : C∗(A) → Z(M (C∗(Σφ))) via (7) h(a)f ((r(σ), −a)σ) dµA(a) (i(h)f )(σ) =ZA for all σ ∈ Σφ, h ∈ Cc(A) and f ∈ Cc(Σφ). A straightforward but tedious computation shows that i(h) is an adjointable operator on C∗(Σφ)C ∗(Σφ): let f and g be in Cc(A) and let σ = (γ, b) ∈ Σφ. Then hi(h)f , gi(γ, b) = (i(h)f )∗ ∗ g(γ, b) =ZΣr(γ) =ZΓr(γ)ZAγ φ (i(h)f )∗(η, b′)g(η−1γ, −b′ + b) dνr(γ)(η, b′) i(h)f (η−1, −b′)g(η−1γ, −b′ + b) dµγ(b′)dλr(γ)(η) which using (7), interchanging the integrals on A, using the fact that µA is a Haar measure on A, and interchanging back the integrals, equals ZΓr(γ)ZAγ = hf, i(h∗)(g)i(γ, b). f (η−1, −b′)(i(h∗)g)(η−1γ, −b′ + b) dµγ(b′)dλr(γ)(η) Hence i(h) is an adjointable operator on C∗(Σφ)C ∗(Σφ). Moreover, i(h) belongs to Z(M (C∗(Σφ))) since Σφ is a twist. Note that as in Remark 2.4 i is determined by the map u : A → U M (C∗(Σφ)) given as the composition A → M (C∗(A) ⊗ C0(Γ0)) → M (C∗(Σφ)). Then condition (i) of Theorem 2.2 follows from the fact that ρ(uaf ) = ρ(f ) for all a ∈ A and f ∈ Cc(Σφ). The nondegeneracy of i (see Remark 2.3) follows from OBSTRUCTIONS TO LIFTING COCYCLES ON GROUPOIDS AND THE ASSOCIATED C ∗-ALGEBRAS13 the fact that u0 = 1M(C ∗(Σφ)). Hence condition (iv) of Theorem 2.2 holds. Con- dition (ii) follows by a straightforward computation. Since B × Γ is amenable, it follows by Lemma 4.2 that ι : C∗(Σφ) → M (C∗(Γ × B)) ∼= Cb( B, M (C∗(Γ))) is injective. Note that ι is B-equivariant with respect to natural actions. Hence, if ρ(γχ(f )) = 0 for all χ ∈ B, we have f = 0 and so condition (iii) holds. (cid:3) Remark 4.4. As noted in the proof of the above theorem (see Remark 2.4) one can replace conditions (i) and (iv) of Theorem 2.2 by the following two condi- tions on the corresponding map u : A → U M (C∗(Σφ)): condition (i) may be replaced by the requirement that ρ(uaf ) = ρ(f ) for all a ∈ A and f ∈ C∗(Σφ) and condition (iv) may be replaced by the requirement that u0 = 1M(C ∗(Σφ)). Remark 4.5. Theorem 4.3 also holds if one replaces the requirement that Γ be amenable by the requirement that Σφ be a clopen subset of Γ × B (see Lemma 4.2). Example 4.6. With notation as in Example 3.6, we first observe that C∗(ΓU ) is a continuous trace algebra with Prim C∗(ΓU ) = X and trivial Dixmier-Douady invariant δ(C∗(Γ)) = 0, that is, it is strong Morita equivalent to C0(X) (see [11]). Next we let α : Z → Aut C∗(ΓU ) be the action determined by the T- valued groupoid 1-cocycle φ defined by the Cech 1-cocyle λij (see [19, Proposition II.5.1]); so (αnf )(x, i, j) := λij (x)nf (x, i, j) for all f ∈ Cc(ΓU ), (x, i, j) ∈ ΓU and n ∈ Z. It is straightforward to see that α fixes every ideal in C∗(ΓU ). For each i ∈ I the ideal Ji in C∗(ΓU ) corresponding to Ui ⊂ X may be identi- fied with C∗((ΓU )Vi ) where Vi := {(x, j, j) : x ∈ Uij} ⊂ Γ0 U Note that (ΓU )Vi = {(x, j, k) : x ∈ Uijk}; let wi : (ΓU )Vi → R be defined by (hence Vi ∼=Fj∈I Uij). wi(x, j, k) :=(λji(x) 0 if j = k, otherwise. Then wi may be identified with a multiplicative unitary for the ideal C∗((ΓU )Vi ) and α1Ji = Ad wi. Hence, α is locally unitary. A quick calculation shows that λij wi(x, k, k) = wj(x, k, k) for all x ∈ Uijk. Therefore the Phillips-Raeburn obstruction η(α) is the class [λ] ∈ H 1(X, T ), the first Cech cohomology of X with coefficients in T , the sheaf of continuous T-valued functions on X (see [14, §2.10]). Using the usual isomorphism H 1(X, T ) ∼= H 2(X, Z) we identify η(α) with [λ⋆] ∈ H 2(X, Z). Note that the element η(α) may also be identified with the class of Prim C∗(ΓU ) ⋊α Z regarded as a circle bundle under the dual action of T (see [15, p. 224]). By Theorem 4.3 we have C∗(Σφ) ∼= indR Z(C∗(Γ), α). Next we observe that C∗(Σφ) is a continuous trace algebra (see the remark pre- ceding [16, Lemma 3.3]) and that Prim C∗(Σφ) ∼= T × X (see the proof of [16, Proposition 3.4]). Since δ(C∗(Γ)) = 0, it follows by [16, Corollary 3.5] that δ(C∗(Σφ)) = z × η(α) = z × [λ⋆] ∈ H 3(T × X, Z) 14 MARIUS IONESCU AND ALEX KUMJIAN where z is the standard generator of H 1(T, Z). REFERENCES [1] Charles A. Akemann, Gert K. Pedersen, and Jun Tomiyama. "Multipliers of C∗-algebras". In: J. Functional Analysis 13 (1973), pp. 277 -- 301. [2] Claire Anantharaman-Delaroche and Jean Renault. Amenable groupoids. Vol. 36. Monographies de L'Enseignement Math´ematique [Monographs of L'Enseignement Math´ematique]. With a foreword by Georges Skan- dalis and Appendix B by E. Germain. Geneva: L'Enseignement Math´e- matique, 2000, p. 196. ISBN: 2-940264-01-5. [3] M. E. B. Bekka, A. T. Lau, and G. Schlichting. "On invariant subalgebras of the Fourier-Stieltjes algebra of a locally compact group". In: Math. Ann. 294.3 (1992), pp. 513 -- 522. ISSN: 0025-5831. [4] Mohammed E. B. Bekka and Alain Valette. "Lattices in semi-simple Lie groups, and multipliers of group C∗-algebras". In: Ast´erisque 232 (1995). Recent advances in operator algebras (Orl´eans, 1992), pp. 67 -- 79. ISSN: 0303-1179. [5] Anton Deitmar and Siegfried Echterhoff. Principles of harmonic analy- sis. Second. Universitext. Springer, Cham, 2014, pp. xiv+332. ISBN: 978- 3-319-05791-0; 978-3-319-05792-7. [6] Siegfried Echterhoff. "On induced covariant systems". In: Proc. Amer. Math. Soc. 108.3 (1990), pp. 703 -- 706. ISSN: 0002-9939. [7] James M. G. Fell and Robert S. Doran. Representations of ∗-algebras, locally compact groups, and Banach ∗-algebraic bundles. Vol. 1. Vol. 125. Pure and Applied Mathematics. Basic representation theory of groups and algebras. Boston, MA: Academic Press Inc., 1988, pp. xviii +746. ISBN: 0-12-252721-6. [8] A. Haefliger. "Differential cohomology". In: Differential topology (Varenna, 1976). Liguori, Naples, 1979, pp. 19 -- 70. [9] Marius Ionescu and Dana P. Williams. "Irreducible representations of groupoid C∗-algebras". In: Proc. Amer. Math. Soc. 137.4 (2009), pp. 1323 -- 1332. ISSN: 0002-9939. [10] Alex Kumjian. "On equivariant sheaf cohomology and elementary C∗- bundles". In: J. Operator Theory 20.2 (1988), pp. 207 -- 240. ISSN: 0379- 4024. [11] Alexander Kumjian. "Preliminary algebras arising from local homeomor- phisms". In: Math. Scand. 52.2 (1983), pp. 269 -- 278. ISSN: 0025-5521. [12] E. C. Lance. Hilbert C∗-modules. Vol. 210. London Mathematical Soci- ety Lecture Note Series. A toolkit for operator algebraists. Cambridge: Cambridge University Press, 1995, pp. x+130. ISBN: 0-521-47910-X. [13] Paul S. Muhly and Dana P. Williams. Renault's Equivalence Theorem for Groupoid Crossed Products. Vol. 3. NYJM Monographs. Available at http://nyjm.albany.edu:8000/m/2008/3.htm. Albany, NY: State Univer- sity of New York University at Albany, 2008. [14] John Phillips and Iain Raeburn. "Automorphisms of C∗-algebras and sec- ond Cech cohomology". In: Indiana Univ. Math. J. 29.6 (1980), pp. 799 -- 822. ISSN: 0022-2518. REFERENCES 15 [16] [15] John Phillips and Iain Raeburn. "Crossed products by locally unitary au- tomorphism groups and principal bundles". In: J. Operator Theory 11.2 (1984), pp. 215 -- 241. ISSN: 0379-4024. Iain Raeburn and Jonathan Rosenberg. "Crossed products of continuous- trace C∗-algebras by smooth actions". In: Trans. Amer. Math. Soc. 305.1 (1988), pp. 1 -- 45. ISSN: 0002-9947. Iain Raeburn and Joseph L. Taylor. "Continuous trace C∗-algebras with given Dixmier-Douady class". In: J. Austral. Math. Soc. Ser. A 38.3 (1985), pp. 394 -- 407. ISSN: 0263-6115. Iain Raeburn and Dana P. Williams. Morita equivalence and continu- ous-trace C∗-algebras. Vol. 60. Mathematical Surveys and Monographs. Providence, RI: American Mathematical Society, 1998, pp. xiv +327. ISBN: 0-8218-0860-5. [17] [18] [19] Jean Renault. A Groupoid Approach to C∗-Algebras. Vol. 793. Lecture Notes in Mathematics. New York: Springer-Verlag, 1980. [20] Jean Renault. "Repr´esentation des produits crois´es d'alg`ebres de groupoıdes". In: J. Operator Theory 18.1 (1987), pp. 67 -- 97. ISSN: 0379-4024. [21] Marc A. Rieffel. "Induced representations of C∗-algebras". In: Advances in Math. 13 (1974), pp. 176 -- 257. [22] Dana P. Williams. Crossed products of C∗-algebras. Vol. 134. Mathemat- ical Surveys and Monographs. Providence, RI: American Mathematical Society, 2007, pp. xvi+528. ISBN: 978-0-8218-4242-3; 0-8218-4242-0. DEPARTMENT OF MATHEMATICS, UNITED STATES NAVAL ACADEMY, ANNAPOLIS, MD, USA E-mail address: [email protected] DEPARTMENT OF MATHEMATICS, UNIVERSITY OF NEVADA, RENO, NV, USA E-mail address: [email protected]
1302.7303	1	1302	2013-02-28T20:08:46	Unitarization of uniformly bounded subgroups in finite von Neumann algebras	[ "math.OA", "math.MG" ]	This note will present a new proof of the fact that every uniformly bounded group of invertible elements in a finite von Neumann algebra is similar to a unitary group. The proof involves metric geometric arguments in the non-positively curved space of positive invertible operators of the algebra; in 1974 Vasilescu and Zsido proved this result using the Ryll-Nardzewsky fixed point theorem.	math.OA	math	Unitarization of uniformly bounded subgroups in finite von Neumann algebras Martín Miglioli∗ Abstract This note will present a new proof of the fact that every uniformly bounded group of invertible elements in a finite von Neumann algebra is similar to a unitary group. The proof involves metric geometric arguments in the non-positively curved space of positive invertible operators of the algebra; in 1974 Vasilescu and Zsido proved this result using the Ryll-Nardzewsky fixed point theorem. 1 Geometry of the cone of positive operators in a finite algebra The metric geometry of the cone of positive invertible operators in a finite von Neumann algebra was studied in [1, 5]. In this subsection we recall some facts from these papers. Let A be a von Neumann algebra with a finite (normal, faithful) trace τ . Denote by Ah the set of selfadjoint elements of A, by GA the group of invertible elements, by UA the group of unitary operators, and by P the set of positive invertible operators P = eAh = {a ∈ GA : a > 0}; P is an open subset of Ah in the norm topology. Therefore if one regards it as a manifold, its tangent spaces identify with Ah endowed with the uniform norm k · k. We make of P a weak Banach-Finsler manifold by assigning for each a ∈ P the following 2-norm to the tangent space Ta(P) ≃ Ah kxka,2 = ka− 1 2 xa− 1 2 k2, for x ∈ Ah ≃ Ta(P) where One obtains a geodesic distance d2 on P by considering kxk2 = τ (x2) 1 2 for x ∈ Ah. d2(a, b) = inf {Lenght(γ) : γ is a piecewise smooth curve joining a and b}, where smooth means differentiable in the norm induced topology and the lenght of a curve γ : [0, 1] → P is measured using the norm above: Lenght(γ) = Z 1 0 k γ(t)kγ(t),2dt. If A is finite dimensional, i.e. a sum of matrix spaces, this metric is well-known: positively curved Riemannian metric on the set of positive definite matrices [8]. it is the non ∗Supported by ANPCyT, Argentina. 1 If A is of type II1, the trace inner product is not complete, so that P is not a Hilbert-Riemann manifold and (P, d2) is not a complete metric space, see Remark 3.21 in [5]. The following holds • By [1, Th. 3.1 and Th. 3.2] the unique geodesic between a and b for a, b ∈ P is given by and has lenght equal to γa,b(t) = a 1 2 (a− 1 2 ba− 1 2 )ta 1 2 d2(a, b) := Lenght(γa,b) = kln(a− 1 2 ba− 1 2 )k2. • The action of GA on P given by Ig(a) = gag∗ is isometric, i.e. d2(Ig(a), Ig(b)) = d2(a, b), and sends geodesic segments to geodesic segments, i.e. Ig ◦ γa,b = γIg(a),Ig (b) for all a, b ∈ P and g ∈ GA. See the Introduction of [1]. • Let a ∈ P and γ : [0, 1] → P be a geodesic. Then [5, Theorem 4.4] d2(γ0, γ1)2 + 4d2(a, γ 1 2 )2 ≤ 2(d2(a, γ0)2 + d2(a, γ1)2) so the metric space (P, d2) satisfies the semi-parallelogram law (see Definition 2.1 below). • By [1, Cor. 3.4] the distance along two geodesics is convex, i.e. t 7→ d2(γa1,b1 (t), γa2,b2 (t)), [0, 1] → [0, +∞) is convex for a1, b1, a2, b2 ∈ P. This implies d2(γa1,b1 (t), γa2,b2 (t)) ≤ td2(γa1,b1 (0), γa2,b2 (0)) + (1 − t)d2(γa1,b1 (1), γa2,b2 (1)) = = td2(a1, a2) + (1 − t)d2(b1, b2). If t0 ∈ [0, 1] is fixed, the continuity of (a, b) 7→ γa,b(t0), P × P → P in the d2 metric follows from the above inequality. • Let Pc1,c2 := {a ∈ P : c11 ≤ a ≤ c21} for 0 < c1 < c2. In Pc1,c2 the linear metric and the rectifiable distance are equivalent [5, Prop. 3.2], i.e. there are C > 0, C ′ > 0 such that ka − bk2 ≤ Cd2(a, b), d2(a, b) ≤ C ′ka − bk2 a, b ∈ Pc1,c2 Since k · k2 is complete on subsets of A which are closed and bounded in the uniform norm and Pc1,c2 is closed and bounded in the uniform norm (Pc1,c2 , d2) is a complete metric space. Also, for a, b ∈ Pc1,c2 d2(a, b) ≤ C ′ka − bk2 ≤ C ′ka − bk ≤ 2C ′c2 so that Pc1,c2 is bounded in the d2 metric. • Pc1,c2 is geodesically convex: if a, b ∈ Pc1,c2 then γa,b(t) ∈ Pc1,c2 for every t ∈ [0, 1], see [2]. 2 Non-negatively curved metric spaces In this subsection we recall some well-known results from metric geometry. A general reference is [4]. For the convenience of the the reader we include the proof of the Bruhat-Tits fixed point theorem. Definition 2.1. A metric space (X, d) satisfies the semi-parallelogram law if for all x, y ∈ X there is a z ∈ X such that for all w ∈ X the following inequality holds A Bruhat-Tits space is a complete metric space in which the semi-parallelogram law holds. d(x, y)2 + 4d(w, z)2 ≤ 2[d(x, z)2 + d(y, z)2]. 2 Remark 2.2. The point z satisfying this inequality is unique and is called the midpoint between x and y and we denote it by m(x, y). We therefore have a function m : X × X → X called the midpoint map. Lemma 2.3. Serre's Lemma [7, Ch. XI, Lemma 3.1] Let (X, d) be a Bruhat-Tits and S a bounded subset of X. Then there is a unique closed ball Br[y] of minimal radius containing S. Definition 2.4. The center y of the closed ball Br[y] in the previous lemma is called the circumcenter of the bounded set S. Theorem 2.5. Bruhat-Tits fixed point theorem [3] If (X, d) is a Bruhat-Tits space and I : G → Isom(X) is an action of a group G on X by isometries which has a bounded orbit, then the circumcenter of each orbit is a fixed point of the action. Proof. We denote the action by g · x for g ∈ G and x ∈ X. Since the action is isometric and there is a bounded orbit all orbits are bounded. For x ∈ X let Br[y] be the unique closed ball of minimal radius which contains G · x. If g ∈ G then G · x = g · (G · x) ⊆ g · Br[y] = Br[g · y] where the last equality follows since the action is isometric. From the uniqueness of the closed balls of minimal radius containing G · x we conclude that g · y = y. Therefore, g · y = y for every g ∈ G and y is a fixed point of the action. 3 Uniformly bounded subgroups Definition 3.1. A subset A ⊆ P is geodesically convex if γa,b(t) ∈ A for every a, b ∈ A and t ∈ [0, 1]. Definition 3.2. The convex hull of a subset S ⊆ P is the smallest geodesically convex set containing S and we denote it by conv(A). An alternative definition is conv(S) = [n∈N Xn where X1 = S, and Xn+1 = {γa,b(t) : a, b ∈ Xn, t ∈ [0, 1]} inductively for n ≥ 1. Lemma 3.3. If C ⊆ Pc1,c2 is a geodesically convex subset then its closure C in (Pc1,c2 , d2) is geodesically convex. Proof. If a, b ∈ C and t ∈ [0, 1] let (an)n, (bn)n be sequences in C such that an → a, bn → b. γan,bn (t) ∈ C for all n ∈ N and since (a, b) 7→ γa,b(t) is continuous on Pc1,c2 × Pc1,c2, γan,bn (t) → γa,b(t). We conclude that γa,b(t) ∈ C. Theorem 3.4. Let H ⊆ GA be a uniformly bounded subgroup, i.e. suph∈H khk := M < ∞. Then there is an s ∈ PM −1,M such that shs−1 ∈ UA for every h ∈ H. Proof. Consider the action I : H → Isom(P) given by Ih(a) = hah∗ for h ∈ H and a ∈ P. We denote h · a := Ih(a). Since H · 1 = {hh∗ : h ∈ H} ⊆ PM −2,M 2 and PM −2,M 2 is geodesically convex conv(H · 1) ⊆ PM −2 ,M 2. Also, since PM −2 ,M 2 is closed in (P, d2), conv(H · 1) ⊆ PM −2 ,M 2. We adopt the notation of Definition 3.2. X1 = H · 1 is invariant for the action and since the action sends geodesics segments to geodesic segments, if Xn is invariant then Xn+1 is invariant for all n ≥ 1. We conclude that conv(H · 1) = Sn∈N Xn is invariant. Since the action is also isometric conv(H · 1) is an invariant subset and we can restrict the action to this subset. Note that conv(H · 1) is a geodesically convex subset of P, in (P, d2) the semi-parallelogram holds and the midpoint of a, b ∈ P is γa,b( 1 2 ), so this law also holds in (conv(H · 1), d2). Since conv(H · 1) 3 is a closed subset of the complete metric space (PM −2,M 2 , d2), (conv(H · 1), d2) is a complete metric space. We conclude that (conv(H · 1), d2) is a Bruhat-Tits space. Since PM −2,M 2 is bounded in the d2 metric conv(H · 1) is bounded in this metric. Therefore the action has bounded orbits and the Bruhat-Tits fixed point theorem states that the circumcenter a ∈ conv(H · 1) of H · 1 satisfies Ih(a) = hah∗ = a for all h ∈ H. Then 1 = a− 1 2 = a− 1 2 hah∗a− 1 2 = (a− 1 2 ha 1 2 )(a 1 2 h∗a− 1 2 ) 2 ha 1 2 )(a− 1 2 ha 1 2 )∗ for all h ∈ H 2 aa− 1 = (a− 1 1 2 Ha 2 ⊆ UA. so that a− 1 Since a ∈ PM −2,M 2 , then a [6, Prop. 4.2.8]. Taking s = a 1 2 ∈ PM −1,M because the square root is an operator monotone function 1 2 we get the unitarizer stated in the theorem. Remark 3.5. The unitarizability of a uniformly bounded subgroup H of the group of bounded linear operators acting on a Hilbert space was obtained independently in the 50s by Day, Dixmier and Nakamura, see [9] and the references therein, assuming that H is amenable. In that context the unitarizer s was obtained as the square root of the center of mass of {hh∗}h∈H. In the present note the unitarizer is the square root of the circumcenter of that same set; we assume however the existence of a finite trace, and in this setting, Vasilescu and Zsido [10] proved in the 70s the result (without the assumption on amenability) using the Ryll-Nardzewsky fixed point theorem which involves weak topologies. References [1] E. Andruchow and G. Larotonda, Nonpositively Curved Metric in the Positive Cone of a Finite von Neumann Algebra, J. London Math. Soc. (2) 74 (2006), no. 1, 205-218. [2] E. Andruchow, G. Corach and D. Stojanoff, Geometrical Significance of Löwner Heinz in- equality, Proc. Amer. Math. Soc. 128 (2000), no. 4, 1031-1037. [3] F. Bruhat and J. Tits, Groupes réductifs sur un corps local, I. Données radicielles valuées, Inst. Hautes Études Sci. Publ. Math. 41 (1972), 5-252. [4] D. Burago, Y. Burago and S. Ivanov, A Course in Metric Geometry, Amer. Math. Soc., Providence, 2001. [5] C. Conde and G. Larotonda, Spaces of nonpositive curvature arising from a finite algebra. [6] R. V. Kadison and J. R. Ringrose, Fundamentals of the Theory of Operator Algebras. Volume I: Elementary Theory, Amer. Math. Soc., Providence, 1997. [7] S. Lang, Fundamentals of Differential Geometry, Graduate Texts in Mathematics, 191. Springer-Verlag, New York, 1999 [8] G. D. Mostow, Some new decomposition theorems for semi-simple groups, Mem. Amer. Math. Soc. (1955), no. 14, 31-54. [9] M. Nakamura, Z. Takeda, Group representation and Banach limit, Tohoku Math. J. 3 (1951) 132-135. [10] F. H. Vasilescu and L. Zsido, Uniformly bounded groups in finite W ∗-algebras, Acta Sci. Math. (Szeged), 36 (1974), 189-192. 4
1209.0314	1	1209	2012-09-03T11:45:02	Idempotent states on locally compact groups and quantum groups	[ "math.OA", "math.FA", "math.PR" ]	This is a short survey on idempotent states on locally compact groups and locally compact quantum groups. The central topic is the relationship between idempotent states, subgroups and invariant C*-subalgebras. We concentrate on recent results on locally compact quantum groups, but begin with the classical notion of idempotent probability measure. We also consider the `intermediate' case of idempotent states in the Fourier--Stieltjes algebra: this is the dual case of idempotent probability measures and so an instance of idempotent states on a locally compact quantum group.	math.OA	math	IDEMPOTENT STATES ON LOCALLY COMPACT GROUPS AND QUANTUM GROUPS PEKKA SALMI Dedicated to Professor Victor Shulman on the occasion of his 65th birthday Abstract. This is a short survey on idempotent states on locally compact groups and locally compact quantum groups. The central topic is the relationship between idempotent states, subgroups and invariant C-subalgebras. We concentrate on recent results on locally compact quantum groups, but begin with the classical notion of idempotent probability measure. We also consider the 'intermediate' case of idempotent states in the Fourier -- Stieltjes algebra: this is the dual case of idempotent probability measures and so an instance of idempotent states on a locally compact quantum group. This is a short survey on idempotent states on locally compact groups and, more generally, on locally compact quantum groups. Idempotent states arise for example as limits of random walks (as we shall see in section 1) and as limits in ergodic theorems for random walks [8]. Idempotent states are also connected to the construction of the Haar measure of a compact group: taking the Ces`aro limit of convolution powers of a probability measure gives an idempotent probability measure, which is the Haar measure if the original measure is suitably chosen. The same process works in the case of compact quantum groups: this is the construction of Haar state due to Woronowicz [42, 45]. Very recently, idempotent states on locally compact quantum groups have also turned up in connection with Hopf images [3] and Poisson boundaries [16]. As we shall see, idempotent states are inherently related to subgroups. However, there is some evidence against the preceding claim, such as Pal's example of an idempotent state on the Kac -- Paljutkin quantum group that does not arise from the Haar state of a subgroup [24]. In this paper we shall see that in fact also Pal's example is associated with a subgroup (in a different way), and so perhaps there is still hope to connect all idempotent states to subgroups, quotient groups or combinations of these. There are not many new things in this survey article: only a cute new proof to a known result and the already-mentioned insight to the example of Pal. Many of the results, and much more, can be found in the recent papers due to (combinations of) Franz, Skalski, Tomatsu and the author [9, 10, 11, 31, 33]. 1. Random walks and idempotent probability measures Every probability measure µ on a discrete group G determines a random walk: if we start from point s ∈ G, then the probability for taking a step to t ∈ G is where µts−1 := µ({ts−1}). Suppose that we start from the identity e and Xk is the random variable denoting the position after k steps. We can use convolution to describe the random walk: P (s 7→ t) = µts−1 , P (X1 = t) = µt P (X2 = t) = Xs∈G P (X1 = s)P (s 7→ t) = Xs∈G µsµts−1 = (µ ⋆ µ)t ... P (Xk = t) = (µ⋆k)t. 2010 Mathematics Subject Classification. Primary 60B15; Secondary 43A05, 43A35, 46L30, 81R50. 1 2 PEKKA SALMI In general the convolution of measures µ and ν on a locally compact group G is defined by h µ ⋆ ν, f i =ZZ f (st) dµ(s) dν(t) (f ∈ C0(G)); here, and throughout the paper, we consider measures on G as functionals on the C-algebra C0(G) of continuous functions on G vanishing at infinity. In the discrete case the convolution boils down to µ ⋆ ν = Xs,t∈G µsνtδst = Xs∈G (cid:18)Xt∈G µst−1 νt(cid:19)δs where δs denotes the Dirac measure at s. More generally, a probability measure on a locally compact group determines a random walk on that group. From this point of view, we shall see that random walks give rise to idempotent prob- ability measures. An idempotent probabity measure on a locally compact group G is a probability measure µ on G that is an idempotent under the convolution product: µ ⋆ µ = µ. Now suppose that G is a compact group and ν is a probability measure on G. If the sequence of convolution powers of ν converges in the weak* topology, then the limit is an idempotent probability measure, which embodies the limit of the random walk. The convergence of such a sequence of convolution powers is widely studied in probability theory (see for example [12, 13]). Taking a slightly different approach, consider the Ces`aro averages 1 n nXk=1 ν⋆k. The sequence of these averages always converges in the weak* topology. Morevover, the limit µ is an idempotent probability measure. This gives a useful way to generate idempotent probability measures: for example, the Haar measure of a compact group may be constructed this way. (In the case of non-compact groups the situation is more complicated and to make sure that the limit is non-zero, some form of tightness needs to be assumed for the sequence of convolution powers of ν.) 2. Kawada -- Ito theorem Now that we have seen how idempotent probability measures may arise in practice, a natural question is how to characterise these measures. In the case of locally compact abelian groups, we may use the Fourier -- Stieltjes transform to convert an idempotent measure to a characteristic function on the dual group. This trivialises the algebraic side of things, but now the positivity and the normalisation condition become non-trivial. Still, this is a useful approach and, as we shall see in the next section, leads to a simple characterisation. But let us first review the history of the general problem. Already in 1940 Kawada and Ito [17] characterised idempotent probability measures on compact groups as the normalised Haar measures of compact subgroups. It seems that harmonic analysts were unaware of this paper, and Wendel [41] rediscovered the result in 1954 (truth be told, Wendel's main result is an interesting new proof for the existence of Haar measure on a compact group G, using idempotents in the compact semigroup formed by the probability measures on G). Trying to characterise all idempotent measures on a locally compact abelian group, Rudin [29, 30] showed that any idempotent measure is concentrated on a compact subgroup, thereby extending Wendel's result -- or that of Kawada -- Ito -- to locally compact abelian groups. (The full description of idempotent measures on locally compact abelian groups is due to Cohen [4]; the non-abelian case is open.) Independently, Pym [28] and Loynes [23] characterised idempotent probability measures on locally compact groups (not necessarily abelian) as the normalised Haar measures of compact subgroups. However, since the problem took two separate paths, one starting from Kawada -- Ito and another one from Wendel, it is perhaps not that surprising that the problem was solved already in 1954, by Kelley [18]. The three approaches, due to Kelley, Pym and Loynes, are all quite different: Kelley studies operators on C0(G), Pym idempotent measures on semigroups and IDEMPOTENT STATES 3 Loynes operator-valued Fourier transform. There are also other generalisations, for example one due to Parthasarathy [25] to complete separable metric groups, which need not be locally compact. Theorem 1 (Kawada -- Ito). Let µ be an idempotent probability measure on a locally compact group G. Then there is a compact subgroup H of G such that µ is the normalised Haar measure of H (considered as a measure on G). The Kawada -- Ito theorem gives a procedure to construct the Haar measure of a compact group. Start with a probability measure ν whose support generates the compact group G. As mentioned in the preceding section, the Ces`aro averages of convolution powers of ν converge in the weak* topology to an idempotent probability measure. It follows from the Kawada -- Ito theorem, that the limit is the normalised Haar measure of G due to the choice of ν. As another example, consider the circle group T and the Dirac measure δz at some z ∈ T. Then there is an idempotent probability measure µ such that 1 z → µ weak. If z is a rational multiple of π, then µ is the counting measure of the finite subgroup of T generated by z. On the other hand, if z is an irrational multiple of π, then µ is the normalised Lebesgue measure on T. The latter statement amounts to the Weyl equidistribution theorem. nPn k=1 δ∗k 3. Idempotent states in the Fourier -- Stieltjes algebra Let G be a locally compact group. The Fourier -- Stieltjes algebra B(G) is the collection of all coefficient functions ( π(·)ξ ζ ) of strongly continuous unitary representations π of G (here ξ and ζ are elements of the representation space Hπ). The Fourier -- Stieltjes algebra is the dual space of the universal group C-algebra C∗(G) (this determines the norm of B(G)), and B(G) is a Banach algebra under the pointwise multiplication of functions. If G is abelian, B(G) is isomorphic, via Fourier -- Stieltjes transform, to the measure algebra M(bG) of Radon measures on the dual group bG. An idempotent state in B(G) is a state on C∗(G) that is an idempotent: u2 = u. That u is a state means that u is a positive definite function with u(e) = 1, where e denotes the identity of G. In the case of abelian G, the Fourier -- Stieltjes transform takes an idempotent probability measure on bG to an idempotent state in B(G). This explains why the characterisation of idempotent states in B(G) may be viewed as the dual version of the Kawada -- Ito theorem. In fact, many early studies on idempotents in M(G), with G abelian, used the Fourier -- Stieltjes transform to translate the problem to the dual setting. Continuing from the work of Cohen [4], mentioned in the previous section, Host [14] charac- terised all idempotents in B(G) as characteristic functions of sets in the open coset ring of G. However, his characterisation does not immediately lead to the following characterisation of idem- potent states, which is due to Ilie and Spronk [15]. The short proof presented here is new (Ilie and Spronk obtained their result as a corollary of a more general characterisation of contractive idempotents). Theorem 2. Every idempotent state in B(G) is the characteristic function of an open subgroup of G. Proof. First of all, every idempotent in B(G) is a characteristic function of some open (and closed) set H, because B(G) consists of continuous functions. Denote the universal representation of G by . If s ∈ H, then u(s−1) = h u, (s−1) i = h u, (s)∗ i = h u, (s) i = 1, so s−1 is also in H. Moreover, because u is a state, and h u, (s)∗(s) i = u(e) = 1, h u, (s)∗ ih u, (s) i = 1 · 1 = 1. 4 PEKKA SALMI It follows from Choi's theorem on multiplicative domains (see [26, Theorem 3.19]) that u is mul- tiplicative at (s). So for every t ∈ G, u(ts) = u(t)u(s) which implies that H is closed under multiplication. Hence H is an open subgroup and u = 1H. (cid:3) The fact that compact subgroups in the Kawada -- Ito theorem have changed to open ones in the preceding result reflects subgroup duality. Suppose that G is abelian and H is a closed subgroup of G. Then the continuous characters on G that are constant 1 on H form a closed subgroup H ⊥ of bG. Now H is compact if and only if H ⊥ is open, and vice versa. Moreover, the Fourier -- Stieltjes transform maps a measure on G supported by H to a function on bG that is constant on the cosets of H ⊥. 4. Locally compact quantum groups Locally compact quantum groups provide a natural context to discuss the results in the previous sections in a unified manner. We shall walk through the definition due to Kustermans and Vaes [21]. Let G denote a locally compact quantum group. This means that we have a C-algebra C0(G), a non-degenerate ∗-homomorphism ∆ : C0(G) → M(C0(G) ⊗ C0(G)) (where the tensor product is the minimal C-algebraic tensor product and M( · ) denotes the multiplier algebra) such that (id ⊗ ∆)∆ = (∆ ⊗ id)∆ (coassociativity); and span ∆(C0(G))(C0(G) ⊗ 1) = span ∆(C0(G))(1 ⊗ C0(G)) = C0(G) ⊗ C0(G). The map ∆ is called the comultiplication of G. We also need to assume that there exist left and right Haar weights on C0(G), denoted by φ and ψ, respectively. These are so-called KMS-weights, which are densely defined, faithful and lower semicontinuous. The important invariance properties, as with Haar measures, are that whenever ω ∈ C0(G)∗ + and a, b ∈ C0(G)+ are such that φ(a) < ∞ and ψ(b) < ∞. So a locally compact quantum group is given by a C-algebra that has a suitable comultiplication and left and right Haar weights. It is convenient to use the suggestive notation C0(G) for the C-algebra because in the commutative case the C-algebra is C0(G) for some locally compact group G. In this case the comultiplication is given by dualised group multiplication: ∆(f )(s, t) = f (st) (f ∈ C0(G), s, t ∈ G). Note that ∆(f ) ∈ Cb(G×G) = M(C0(G)⊗C0(G)) but ∆(f ) /∈ C0(G)⊗C0(G) unless G is compact or f = 0. Of course the left and right Haar weights are given by integration against the left and right Haar measures, respectively. Whenever G is a locally compact quantum group such that C0(G) is commutative, it is of this form. Next we consider the dual of the commutative case, which is known as the co-commutative case. Let λ be the left regular representation of G. Then the reduced group C-algebra C∗ r(G) is generated by λ(L1(G)) in B(L2(G)). (Hopefully the reader is not confused by the two uses of 'B' as both the Fourier -- Stieltjes algebra and the algebra of bounded operators; the distinction should be clear from the context.) We define a comultiplication on C∗ r(G) by putting ∆(λ(s)) = λ(s) ⊗ λ(s) (s ∈ G). r(G), because the linear span of λ(G) is strictly dense in M(C∗ Note that actually λ(s) is in M(C∗ r(G)) but the above does define a unique comultiplication on C∗ r(G)). In this case the left and right Haar weights coincide and are the so-called Plancherel weight. The construction of this weight uses Tomita -- Takesaki theory [34, section VII.3] (for discrete G, φ(a) = ( aδe δe ) is the usual tracial state). and φ(cid:0)(ω ⊗ id)∆(a)(cid:1) = ω(1)φ(a) ψ(cid:0)(id ⊗ ω)∆(b)(cid:1) = ω(1)ψ(b) IDEMPOTENT STATES 5 Both these examples may be considered as Kac algebras [7]. Every Kac algebra determines a locally compact quantum group, so the latter notion is more general. For an example of a locally compact quantum group that is not a Kac algebra, see section 7, which includes the description of the quantum deformation of SU(2) defined by Woronowicz. For a more thorough introduction to locally compact quantum groups, see for example the survey by Kustermans and Tuset [19, 20] or the book by Timmermann [36]. From now on we shall concentrate on locally compact quantum groups G that are coamenable. That means that there is a state ǫ on C0(G), called the counit of G, such that (id ⊗ ǫ)∆(a) = (ǫ ⊗ id)∆(a) = a for every a ∈ C0(G). In the commutative case coamenability is a vacuous condition (every commu- tative quantum group is coamenable), but a co-commutative quantum group G = bG is coamenable if and only if the locally compact group G is amenable. 5. Classical cases as instances of idempotent states on locally compact quantum groups The notion of idempotent state from the two classical cases -- idempotent probability measures on groups and idempotent states in the Fourier -- Stieltjes algebra -- is easily generalised to the language of locally compact quantum groups. The dual space of the C-algebra C0(G) carries a natural Banach algebra structure: the multiplication is defined by ω ⋆ σ(a) = (ω ⊗ σ)∆(a) (ω, σ ∈ C0(G)∗, a ∈ C0(G)). An idempotent state on a locally compact quantum group G is a state ω on the C-algebra C0(G) that is an idempotent under the product defined above: ω ⋆ ω = ω. A much more difficult task than the definition is to unify the results from the classical cases to general results on locally compact quantum groups. Indeed, it is perhaps not even possible to do so. To even bring forth this discussion we need some further terminology. A locally compact quantum group H is compact if C0(H) is unital, in which case we write C(H) for C0(H). A compact quantum subgroup of a coamenable locally compact quantum group G is a compact quantum group H such that there exists a surjective ∗-homomorphism π : C0(G) → C(H) (π ⊗ π)∆G = ∆Hπ. (The reader is warned that there are other definitions of closed quantum subgroup [38, 39], and it is not clear whether they are all equivalent. In our situation, all the definitions coincide as they do in many other cases; see [5].) A compact quantum subgroup of G always gives rise to an idempotent state on G. Indeed, when a locally compact quantum group H is compact, the left and right Haar weights are actually bounded functionals and coincide. By normalisation, there exists a unique state -- the Haar state -- φH on C(H) that is both left and right invariant. Using the subgroup morphism π, we may pull back φH to obtain an idempotent state ω = φHπ on G. Obviously ω is a state and it is an idempotent due to invariance of φH: ω ⋆ ω =(cid:0)(φHπ) ⊗ (φHπ)(cid:1)∆G =(cid:0)φH ⊗ φH)(π ⊗ π)∆G = (φH ⊗ φH)∆Hπ = φH(1)φHπ = ω. It should be noted that in the case of compact quantum groups, the existence of Haar state follows from the other axioms as shown by Woronowicz [42, 45]. Indeed, this may be done with a similar process of using Ces`aro averages as mentioned after the Kawada -- Ito theorem: the Ces`aro averages of convolution powers of a faithful state converge to the Haar state (however, the resulting Haar state is not necessarily faithful). The assumption that there is a faithful state on the C-algebra, which is true in the separable case, may be dropped, as shown by Van Daele [40]. Now it is possible to at least formulate the statement of the Kawada -- Ito theorem: every idem- potent state on a locally compact quantum group G is a Haar idempotent, that is, of the form ω = φHπ where φH is the Haar state of a compact quantum subgroup H of G and π : C0(G) → C(H) is the associated morphism. The problem is that this statement is false. It is, moreover, easily seen to be false. Let G be an amenable locally compact group with a non-normal open subgroup 6 PEKKA SALMI r(G). H. Then 1H is an idempotent state on C∗ compact quantum subgroup of C∗ of the form C∗ finite case and [31, section 7] for a related discussion on left invariant C-subalgebras in C∗ If 1H were a pullback of the Haar state of a r(G), then the compact quantum subgroup would necessarily be r(G/H). But H not being normal, this is not possible (see [10, Theorem 6.2] for the r(G)). Although this example, obtainable with finite groups, certainly seems to be the most straight- forward counterexample of the Kawada -- Ito theorem for quantum groups, it was not the first one. The first counterexample is due to Pal [24] and it comes from a genuine quantum group: the Kac -- Paljutkin quantum group. We shall next describe this quantum group and Pal's example as well as provide a new insight to this example. 6. Pal's counterexample The underlying C-algebra of the Kac -- Paljutkin quantum group G is C ⊕ C ⊕ C ⊕ C ⊕ M2(C), the basis of which is given by the vectors ek = δ1,k ⊕ δ2,k ⊕ δ3,k ⊕ δ4,k ⊕(cid:18)δ5,k δ7,k δ8,k δ6,k(cid:19) k = 1, 2, . . . , 8. The comultiplication of G is defined by ∆(e1) = e1 ⊗ e1 + e2 ⊗ e2 + e3 ⊗ e3 + e4 ⊗ e4 + 1 2 (e5 ⊗ e5 + e6 ⊗ e6 + e7 ⊗ e7 + e8 ⊗ e8) ∆(e2) = e1 ⊗ e2 + e2 ⊗ e1 + e3 ⊗ e4 + e4 ⊗ e3 + 1 2 (e5 ⊗ e6 + e6 ⊗ e5 + ie7 ⊗ e8 − ie8 ⊗ e7) ∆(e3) = e1 ⊗ e3 + e3 ⊗ e1 + e2 ⊗ e4 + e4 ⊗ e2 + 1 2 (e5 ⊗ e6 + e6 ⊗ e5 − ie7 ⊗ e8 + ie8 ⊗ e7) ∆(e4) = e1 ⊗ e4 + e4 ⊗ e1 + e2 ⊗ e3 + e3 ⊗ e2 + 1 2 (e5 ⊗ e5 + e6 ⊗ e6 − e7 ⊗ e7 − e8 ⊗ e8) ∆(e5) = e1 ⊗ e5 + e5 ⊗ e1 + e2 ⊗ e6 + e6 ⊗ e2 + e3 ⊗ e6 + e6 ⊗ e3 + e4 ⊗ e5 + e5 ⊗ e4 ∆(e6) = e1 ⊗ e6 + e6 ⊗ e1 + e2 ⊗ e5 + e5 ⊗ e2 + e3 ⊗ e5 + e5 ⊗ e3 + e4 ⊗ e6 + e6 ⊗ e4 ∆(e7) = e1 ⊗ e7 + e7 ⊗ e1 − ie2 ⊗ e8 + ie8 ⊗ e2 + ie3 ⊗ e8 − ie8 ⊗ e3 − e4 ⊗ e7 − e7 ⊗ e4 ∆(e8) = e1 ⊗ e8 + e8 ⊗ e1 + ie2 ⊗ e7 − ie7 ⊗ e2 − ie3 ⊗ e7 + ie7 ⊗ e3 − e4 ⊗ e8 − e8 ⊗ e4. Pal's idempotent state is defined by ω(cid:18) 8Xk=1 αkek(cid:19) = 1 4 α1 + 1 4 α4 + 1 2 α6. As we shall see in section 9, we can always associate a certain C-subalgebra to an idempotent state. For Pal's idempotent state ω, the associated C-subalgebra (id ⊗ ω)∆(C(G)) is spanned by the elements a = e1 + e2 + e3 + e4 + e5 + e6 and b = e1 − e2 − e3 + e4 − e5 + e6. Moreover, one can calculate that (1) ∆(a) = a ⊗ a and ∆(b) = b ⊗ b. IDEMPOTENT STATES 7 the chosen viewpoint suits us better). Then C∗(Z2) is spanned by λ(0) and λ(1), where λ denotes the left regular representation of Z2. Define π : C∗(Z2) → C(G) by π(λ(0)) = a and π(λ(1)) = b. Now consider the quantum group bZ2 given by the group C-algebra C∗(Z2) (of course Z2 ∼= bZ2 but By (1), we see that π preserves the quantum group structure of bZ2. There is also a conditional expectation E onto π(C∗(Z2)) defined by E(e1) = E(e4) = (a + b) E(e2) = E(e3) = (a − b) 1 8 1 8 E(e5) = 1 4 (a − b) E(e6) = 1 4 (a + b) E(e7) = E(e8) = 0. of bZ2 is the constant function 1 (considered as an element of B(Z2)). Finally, note The counit ǫbZ2 ◦ π−1 ◦ E. What this shows is that Pal's idempotent state is of the similar form that ω = ǫbZ2 as the idempotent states 1H on group C-algebras: 1H = ǫ bH ◦ π−1 ◦ E where ǫ bH is the counit r(G) is the natural embedding (i.e. zero extension), and E : C∗ r(H)) is the natural conditional expectation (i.e. restriction to H). So although Pal's idempotent is not like the idempotent states in the commutative case (i.e. not a Haar idempotent), it is similar to the idempotent states in the co-commutative case. Thus it is associated with a subgroup but in a different way. of bH (i.e. constant 1 on H), π : C∗ r(G) → π(C∗ r(H) → C∗ These examples of idempotent states that are not Haar idempotents show that a new approach is needed for general locally compact quantum groups. In section 8 we consider another notion, that of left invariant C-subalgebras, that is closely tied with idempotent states as we shall see. There is also the approach of Franz and Skalski, who show in [10] that every idempotent state on a finite quantum group arises from the Haar state of a so-called quantum subhyper group. On a related note, Franz and Skalski [9] also show that idempotent states on a finite quantum group correspond to quantum pre-subgroups in the sense of [1]. In all these approaches one associates idempotent states with structures more general than subgroups, and that is what we shall do with left invariant C-subalgebras in section 9. 7. Positive examples from deformation quantum groups In this section we shall consider idempotent states on some important examples of compact quantum groups, in particular on the quantum deformation of SU(2) introduced by Woronowicz [43, 42]. It turns out that on these deformations of classical groups, SUq(2), Uq(2) and SOq(3), all idempotent states are Haar idempotents. The results in this section are due to Franz, Skalski and Tomatsu [11]. Define C(SUq(2)) as the universal unital C-algebra generated by elements a and c such that (cid:18)a −qc∗ a∗ (cid:19) c is formally a unitary matrix (a 2 × 2 matrix with entries in C(SUq(2))). The comultiplication of SUq(2) is determined by the identity ∆(cid:18)a −qc∗ a∗ (cid:19) =(cid:18)a −qc∗ c a∗ (cid:19) ⊗(cid:18)a −qc∗ a∗ (cid:19) . c c This identity is to be read as follows: on the left-hand side we apply ∆ to each entry and on the right-hand side we take a formal matrix multiplication where we use the tensor product when 'multiplying' entries; then we just equate the entries of the two 2 × 2 matrices. Although the relations given above fully determine the structure of SUq(2), to prove that we actually get a compact quantum group takes some work. Recall however that the existence of the Haar state follows from the general theory of compact quantum groups. Using the representation theory of SUq(2), Franz, Skalski and Tomatsu [11] calculated all the idempotent states on SUq(2) for q ∈ (−1, 0) ∪ (0, 1). It turns out that these are all Haar idem- potents. Namely, the idempotent states on SUq(2) are the Haar state and the Haar idempotents coming from the subgroups T and Zn, 1 ≤ n < ∞. We see that T is a subgroup of SUq(2) by mapping the generator a to the generator z of C(T) and c to 0. Moreover, Zn's are subgroups of 8 PEKKA SALMI T. Already Podle´s [27] showed that these are all the closed quantum subgroups of SUq(2). This result follows also from [11], but of course it takes more work to show that all idempotent states actually arise from these subgroups. Franz, Skalski and Tomatsu also give the complete list of idempotent states for the related deformation quantum groups Uq(2) and SOq(3). Also in these cases all idempotent states are Haar idempotents. 8. Left invariant C-subalgebras In this section we shall consider another notion related to idempotent states besides subgroups. The notion is that of left invariant C-subalgebra (here we could use alternative terminology and call these coideals or homogeneous spaces). Let G be a coamenable locally compact quantum group. For ω ∈ C0(G)∗, define the left and right convolution operators on C0(G) by Lω(a) = (ω ⊗ id)∆(a) Rω(a) = (id ⊗ ω)∆(a) (a ∈ C0(G)). A C-subalgebra X ⊆ C0(G) is said to be left invariant if Lω(X) ⊆ X for all ω ∈ C0(G)∗. A nondegenerate C-subalgebra X of C0(G) is left invariant if and only if ∆ : X → M(C0(G) ⊗ X). (A C-subalgebra is nondegenerate if it contains a bounded approximate identity for the ambient C-algebra.) Consider the commutative case when G is a locally compact group. Then a C-subalgebra X of C0(G) is left invariant if and only if it is left translation invariant; that is, the function Lsf (t) = f (st) is in X whenever f ∈ X and s ∈ G. Lau and Losert [22] have characterised left invariant C-subalgebras of C0(G): a C-subalgebra X ⊆ C0(G) is left invariant if and only if there is a compact subgroup H of G such that X consists of all the functions in C0(G) that are constant on right cosets of H. The latter statement means that X is ∗-isomorphic to C0(G/H). Earlier, de Leeuw and Mirkil [6] gave this characterisation in the case of locally compact abelian groups. Moreover, Takesaki and Tatsuuma [35] produced several related results, characterising closed (left) invariant self-adjoint subalgebras of L∞(G), the Fourier algebra A(G), the group von Neumann algebra VN(G) and the L1 group algebra (here the meaning of 'closed' depends on the context: with A(G) and L1(G) it means norm-closed and with L∞(G) and VN(G) it means weak- closed). The dual version of the Lau -- Losert characterisation for a locally compact amenable group G is given in [31]: a C-subalgebra X ⊆ C∗ r(H) for some open subgroup H. One can also consider strictly closed left invariant C-subalgebras of Cb(G) and M(C∗ r(G)) and obtain in both cases a one-to-one correspondence with closed subgroups of G [32]. r(G) is invariant if and only if X ∼= C∗ Finally, we also have the following result from [31], concerning left invariant C-subalgebras of coamenable locally compact quantum groups. Recall that a conditional expectation on a C- algebra A is a norm 1 projection from A onto a C-subalgebra of A. The following result also employs a symmetry condition that is related to the problem brought out by Pal's counterexample. We postpone the formulation of this symmetry condition until after the theorem. Theorem 3. There is a one-to-one correspondence between compact quantum subgroups of G and symmetric, left invariant C-subalgebras X of C0(G) with a conditional expectation P from C0(G) onto X such that (id ⊗ P )∆ = ∆P . r(G). However, C∗ Let G be an amenable locally compact group and H an open subgroup of G. As noted above, r(H) is an invariant C-subalgebra of C∗ C∗ r(H) is not associated with a compact quantum subgroup unless H is normal, in which case C∗ r(G/H). We shall need an analogue of this normality condition for more general quantum groups. This can be done through the so-called multiplicative unitary of a locally compact quantum group G. There is a canonical way to define a unitary operator W on L2(G) ⊗ L2(G) such that W determines the quantum group G [21, Proposition 3.17]. Here L2(G) denotes the Hilbert space obtained by applying the GNS-construction to the left Haar weight of G. The C-algebra C0(G) is faith- fully represented on L2(G) and it is natural to identify C0(G) with its image in B(L2(G)). The r(H) is associated with C∗ IDEMPOTENT STATES 9 multiplicative unitary W determines the comultiplication via ∆(a) = W ∗(1 ⊗ a)W (a ∈ C0(G)). The notion of multiplicative unitary is central in the theory of locally compact quantum groups; seminal work in this area is due to Baaj and Skandalis [2] and Woronowicz [44]. We say that a C-subalgebra X of C0(G) is symmetric if W (x ⊗ 1)W ∗ ∈ M(X ⊗ B0(L2(G))) whenever x ∈ X (here B0 denotes the compact operators). Tomatsu [37] introduced this type of condition, calling it coaction symmetry (due to the fact that X is symmetric if and only if it is closed under the natural action of the dual quantum group of G). Returning to the co- commutative case, the left invariant C-subalgebra C∗ r(H) associated with an open subgroup H of G is symmetric if and only if H is normal [31]. 9. Idempotent states and left invariant C-subalgebras Although Pal's counterexample showed that we cannot associate all idempotent states to com- pact quantum subgroups, we may still have a chance of associating idempotent states to suitable left invariant C-subalgebras. The results in this section are from [33], many of them generalisa- tions from [9] or [11]. Let G be a coamenable locally compact quantum group. If ω is an idempotent state on G, then Rω(C0(G)) is a left invariant C-subalgebra of C0(G) and Rω is a conditional expectation onto this C-subalgebra. The following result generalises an earlier result due to Franz and Skalski [9] concerning compact quantum groups. Theorem 4. Suppose that G is unimodular (i.e. φ = ψ). There is a one-to-one correspondence between idempotent states ω on G and left invariant C-subalgebras X of C0(G) with a conditional expectation P from C0(G) onto X such that φ ◦ P = φ. The correspondence is given by where ǫ is the counit of G. Xω = Rω(C0(G)), ωX = ǫPX . The preceding result leaves room for improvement: one would like to remove the unimodularity assumption in which case the conditional expectation should preserve both left and right Haar weights. The following result characterises those idempotent states that arise from compact quantum subgroups. It also brings together the symmetry condition from the preceding section. The equivalence between the first and the third condition is proved for compact quantum groups in [11]. Theorem 5. Let ω be an idempotent state on G and let Xω = Rω(C0(G)). The following are equivalent: (1) ω is a Haar idempotent; (2) Xω is symmetric; (3) Nω := { a ∈ C0(G); ω(a∗a) = 0 } is an ideal. The set Nω in the third condition of the the preceding result is always a left ideal, so the condition is automatically satisfied if C0(G) is commutative. Consequently, we get the Kawada -- Ito theorem from section 2 as a corollary. Corollary 6 (Kawada -- Ito). If C0(G) is commutative, then all idempotent states on G are Haar idempotents. Finally, we have the following correspondence result, which does not assume unimodularity, but works only for Haar idempotents. Theorem 7. There is a one-to-one correspondence between Haar idempotents ω on G and sym- metric, left invariant C-subalgebras X of C0(G) with a conditional expectation P from C0(G) onto X such that φ ◦ P = φ and ψ ◦ P = ψ. 10 PEKKA SALMI Note that the preceding theorem improves Theorem 3 in the sense that the condition that (id ⊗ P )∆ = ∆P may be replaced by the more natural condition that P preserves both left and right Haar weights. Acknowledgement. This paper is based on a talk given in the conference 'Operator Theory and its Applications' held in honour of Victor Shulman in Gothenburg 2011; I thank the organisers Ivan Todorov and Lyudmila Turowska for a great conference. I thank Nico Spronk for generous support throughout my postdoctoral stay at University of Waterloo and in particular for enabling my conference visit. I thank Emil Aaltonen Foundation for support during the preparation of this paper. I thank Adam Skalski and the referee for helpful comments improving the paper. References [1] S. Baaj, E. Blanchard and G. Skandalis, Unitaires multiplicatifs en dimension finie et leurs sous-objets, Ann. Inst. Fourier (Grenoble) 49 (1999), 1305 -- 1344. [2] S. Baaj and G. Skandalis, Unitaires multiplicatifs et dualit´e pour les produits crois´es de C ∗-alg`ebres, Ann. Sci. ´Ecole Norm. Sup. (4) 26 (1993), 425 -- 488. [3] T. Banica, U. Franz and A. Skalski, arXiv:1112.5018. Idempotent states and the inner linearity property, preprint, [4] P. J. Cohen, On a conjecture of Littlewood and idempotent measures, Amer. J. Math. 82 (1960), 191 -- 212. [5] M. Daws, P. Kasprzak, A. Skalski and P. M. So ltan, Closed quantum subgroups of locally compact quantum groups, preprint. [6] K. de Leeuw and H. Mirkil, Translation-invariant function algebras on abelian groups, Bull. Soc. Math. France 88 (1960), 345 -- 370. [7] M. Enock and J.-M. Schwartz, Kac algebras and duality of locally compact groups, Springer-Verlag, Berlin, 1992. [8] U. Franz and A. Skalski, On ergodic properties of convolution operators associated with compact quantum groups, Colloq. Math. 113 (2008), 13 -- 23. [9] U. Franz and A. Skalski, A new characterisation of idempotent states on finite and compact quantum groups, C. R. Math. Acad. Sci. Paris 347 (2009), 991 -- 996. [10] U. Franz and A. Skalski, On idempotent states on quantum groups, J. Algebra 322 (2009), 1774 -- 1802. [11] U. Franz, A. Skalski and R. Tomatsu, Idempotent states on compact quantum groups and their classification on U q(2), SU q(2), and SO q(3), to appear in J. Noncomm. Geom., arXiv:0903.2363. [12] U. Grenander, Probabilities on algebraic structures, Almqvist & Wiksell, Stockholm, 1968. [13] G. Hognas and A. Mukherjea, Probability measures on semigroups: convolution products, random walks, and random matrices, second ed., Springer, New York, 2011. [14] B. Host, Le th´eor`eme des idempotents dans B(G), Bull. Soc. Math. France 114 (1986), 215 -- 223. [15] M. Ilie and N. Spronk, Completely bounded homomorphisms of the Fourier algebras, J. Funct. Anal. 225 (2005), 480 -- 499. [16] M. Kalantar, M. Neufang and Z.-J. Ruan, Poisson boundaries over locally compact quantum groups, preprint, arXiv:1111.5828. [17] Y. Kawada and K. Ito, On the probability distribution on a compact group. I, Proc. Phys.-Math. Soc. Japan (3) 22 (1940), 977 -- 998. [18] J. L. Kelley, Averaging operators on C∞(X), Illinois J. Math. 2 (1958), 214 -- 223. [19] J. Kustermans and L. Tuset, A survey of C-algebraic quantum groups, part I, Irish Math. Soc. Bull. (1999), no. 43, 8 -- 63. [20] J. Kustermans and L. Tuset, A survey of C-algebraic quantum groups, part II, Irish Math. Soc. Bull. (2000), no. 44, 6 -- 54. [21] J. Kustermans and S. Vaes, Locally compact quantum groups, Ann. Sci. ´Ecole Norm. Sup. (4) 33 (2000), 837 -- 934. [22] A. T.-M. Lau and V. Losert, Complementation of certain subspaces of L∞(G) of a locally compact group, Pacific J. Math. 141 (1990), 295 -- 310. [23] R. M. Loynes, Fourier transforms and probability theory on a noncommutative locally compact topological group, Ark. Mat. 5 (1963), 37 -- 42. [24] A. Pal, A counterexample on idempotent states on a compact quantum group, Lett. Math. Phys. 37 (1996), 75 -- 77. [25] K. R. Parthasarathy, A note on idempotent measures in topological groups, J. London Math. Soc. 42 (1967), 534 -- 536. [26] V. Paulsen, Completely bounded maps and operator algebras, Cambridge University Press, Cambridge, 2002. [27] P. Podle´s, Symmetries of quantum spaces. Subgroups and quotient spaces of quantum SU(2) and SO(3) groups, Comm. Math. Phys. 170 (1995), 1 -- 20. [28] J. S. Pym, Idempotent measures on semigroups, Pacific J. Math. 12 (1962), 685 -- 698. [29] W. Rudin, Idempotent measures on Abelian groups, Pacific J. Math. 9 (1959), 195 -- 209. IDEMPOTENT STATES 11 [30] W. Rudin, Measure algebras on abelian groups, Bull. Amer. Math. Soc. 65 (1959), 227 -- 247. [31] P. Salmi, Compact subgroups and left invariant C-subalgebras of locally compact quantum groups, J. Funct. Anal. 261 (2011), 1 -- 24. [32] P. Salmi, Subgroups and strictly closed invariant C-subalgebras, preprint, arXiv:1110.5459. [33] P. Salmi and A. Skalski, Idempotent states on locally compact quantum groups, to appear in Q. J. Math., arXiv:1102.2051. [34] M. Takesaki, Theory of operator algebras. II, Springer-Verlag, Berlin, 2003. [35] M. Takesaki and N. Tatsuuma, Duality and subgroups, Ann. of Math. (2) 93 (1971), 344 -- 364. [36] T. Timmermann, An invitation to quantum groups and duality. From Hopf algebras to multiplicative unitaries and beyond, European Mathematical Society (EMS), Zurich, 2008. [37] R. Tomatsu, A characterization of right coideals of quotient type and its application to classification of Poisson boundaries, Comm. Math. Phys. 275 (2007), 271 -- 296. [38] S. Vaes, A new approach to induction and imprimitivity results, J. Funct. Anal. 229 (2005), 317 -- 374. [39] S. Vaes and L. Vainerman, On low-dimensional locally compact quantum groups, Locally compact quantum groups and groupoids (Strasbourg, 2002), IRMA Lect. Math. Theor. Phys., vol. 2, de Gruyter, Berlin, 2003, pp. 127 -- 187. [40] A. Van Daele, The Haar measure on a compact quantum group, Proc. Amer. Math. Soc. 123 (1995), 3125 -- 3128. [41] J. G. Wendel, Haar measure and the semigroup of measures on a compact group, Proc. Amer. Math. Soc. 5 (1954), 923 -- 929. [42] S. L. Woronowicz, Compact matrix pseudogroups, Comm. Math. Phys. 111 (1987), 613 -- 665. [43] S. L. Woronowicz, Twisted SU(2) group. An example of a noncommutative differential calculus, Publ. Res. Inst. Math. Sci. 23 (1987), 117 -- 181. [44] S. L. Woronowicz, From multiplicative unitaries to quantum groups, Internat. J. Math. 7 (1996), 127 -- 149. [45] S. L. Woronowicz, Compact quantum groups, Sym´etries quantiques (Les Houches, 1995), North-Holland, Am- sterdam, 1998, pp. 845 -- 884. Department of Mathematical Sciences, University of Oulu, PL 3000, FI-90014 Oulun yliopisto, Fin- land E-mail address: [email protected]
1804.01733	2	1804	2018-12-11T08:44:22	Ground states of groupoid C*-algebras, phase transitions and arithmetic subalgebras for Hecke algebras	[ "math.OA", "math.NT" ]	We consider the Hecke pair consisting of the group $P^+_K$ of affine transformations of a number field $K$ that preserve the orientation in every real embedding and the subgroup $P^+_O$ consisting of transformations with algebraic integer coefficients. The associated Hecke algebra $C^(P^+_K,P^+_O)$ has a natural time evolution $\sigma$, and we describe the corresponding phase transition for KMS$_\beta$-states and for ground states. From work of Yalkinoglu and Neshveyev it is known that a Bost-Connes type system associated to $K$ has an essentially unique arithmetic subalgebra. When we import this subalgebra through the isomorphism of $C^(P^+_K,P^+_O)$ to a corner in the Bost-Connes system established by Laca, Neshveyev and Trifkovic, we obtain an arithmetic subalgebra of $C^(P^+_K,P^+_O)$ on which ground states exhibit the `fabulous' property with respect to an action of the Galois group $Gal(K^{ab}/H_+(K))$, where $H_+(K)$ is the narrow Hilbert class field. In order to characterize the ground states of the $C^$-dynamical system $(C^(P^+_K,P^+_O),\sigma)$, we obtain first a characterization of the ground states of a groupoid $C^$-algebra, refining earlier work of Renault. This is independent from number theoretic considerations, and may be of interest by itself in other situations.	math.OA	math	GROUND STATES OF GROUPOID C ∗-ALGEBRAS, PHASE TRANSITIONS AND ARITHMETIC SUBALGEBRAS FOR HECKE ALGEBRAS MARCELO LACA, NADIA S. LARSEN, AND SERGEY NESHVEYEV Dedicated to Alain Connes on the occasion of his 70th birthday O r (P + r (P + K , P + K , P + Abstract. We consider the Hecke pair consisting of the group P + K of affine transformations of a number field K that preserve the orientation in every real embedding and the subgroup P + consisting of transformations with algebraic integer coefficients. The associated Hecke algebra C ∗ O ) has a natural time evolution σ, and we describe the corresponding phase transition for KMSβ-states and for ground states. From work of Yalkinoglu and Neshveyev it is known that a Bost-Connes type system associated to K has an essentially unique arithmetic subalgebra. When we import this subalgebra through the isomorphism of C ∗ O ) to a corner in the Bost- Connes system established by Laca, Neshveyev and Trifkovi´c, we obtain an arithmetic subalgebra of C ∗ O ) on which ground states exhibit the 'fabulous' property with respect to an action of the Galois group G(K ab/H+(K)), where H+(K) is the narrow Hilbert class field. In order to characterize the ground states of the C ∗-dynamical system (C ∗ O ), σ), we obtain first a characterization of the ground states of a groupoid C ∗-algebra, refining earlier work of Renault. This is independent from number theoretic considerations, and may be of interest by itself in other situations. K , P + r (P + K , P + r (P + 8 1 0 2 c e D 1 1 ] . A O h t a m [ 2 v 3 3 7 1 0 . 4 0 8 1 : v i X r a Introduction The seminal work of Bost and Connes [1], that showed how a Hecke C ∗-algebra constructed from the inclusion of the ring of integers in the field of rationals relates to class field theory for the field Q, inspired several generalizations to algebraic number fields. One such generalization, for quadratic imaginary fields, by Connes, Marcolli, and Ramachandran [4] eventually led to the construction of Bost-Connes type systems for arbitrary number fields by Ha and Paugam [11], which exhibit the expected phase transition and symmetries [18]. Later, Yalkinoglu [32] proved that these Bost-Connes systems contained arithmetic models, which were subsequently shown to be unique under some natural conditions by Neshveyev, see the Appendix to [32]. In the decades following the Bost-Connes paper there has been a vast amount of work putting forward constructions of C ∗-algebras from number theory and unraveling relationships between operator algebras and number theory, formalized in classification of KMS-states, computation of K-theory, and in results about recovery of number theoretical input from the operator algebra side. Without attempting to be exhaustive, we recall some developments. One successful vein of investigation started with the work by Cuntz [7] on a C ∗-algebra of the ax + b semigroup over N, and continued with [9, 10, 24]. The paper [22] by Laca and Raeburn marked the first C ∗-algebra with a time evolution admitting an interesting phase transition at infinity for ground states, see also [20]. Further results of this type were achieved in [8]. It is natural to ask what information passes between number theory and operator algebras when one takes a number field as a starting point and associates C ∗-algebras to it. In [23], Li showed that the Toeplitz type algebra associated to a number field [8] recovers the field up to arithmetic equivalence given a specified number of roots of unity. Motivated by a study of Bost-Connes type systems, Cornelissen et al. [5] showed that the number field can be recovered from the dual of Date: April 5, 2018; minor changes: December 9, 2018. 1 the abelianized Galois group and the collection of all associated L-series. Recently, Takeishi used K-theoretical techniques inspired by [23] to prove that the C ∗-algebra in the Bost-Connes system associated to a number field K recovers the Dedekind zeta-function of K, [29], thus refining the findings of [18] that showed that the entire Bost-Connes system determines the Dedekind zeta- function. Finally, the full reconstruction of number fields up to isomorphism from the C ∗-algebras of the respective Bost-Connes systems has been recently achieved by Kubota and Takeishi [12], see also [6]. Notwithstanding these spectacular developments, one disadvantage of general Bost-Connes sys- tems is that, as opposed to the original Bost-Connes system for Q, they take class field theory as an input. By contrast, systems based on Hecke algebras rely on simple arithmetic considerations and therefore carry a lot of appeal. In this paper we consider one such system already studied in [21] (see also [14, 15]), which is probably the most successful generalization of the original construction of Bost and Connes to general number fields. Let K be an algebraic number field with ring of integers O and denote by K ∗ + the respective multiplicative groups of totally positive elements, that is, elements that embed as positive numbers in every real embedding of K. We consider the C ∗-dynamical system based on the Hecke C ∗-algebra of the inclusion of affine groups + and O∗ P + O :=(cid:18)1 O0 O∗ +(cid:19) ⊂ (cid:18)1 K 0 K ∗ +(cid:19) =: P + K with respect to the natural dynamics determined by the modular function of the inclusion. The purpose of this paper is to give an explicit description of the phase transition of KMS-states and of ground states of this system and to explore the arithmetic subalgebra and 'fabulous' property of ground states. It has already been observed in [21] that this C ∗-dynamical system exhibits a phase transition for finite inverse temperatures replicating that for Bost-Connes systems. Since the methods used in [21] depend in a crucial way on results for the Bost-Connes systems [18] and induction through C ∗-correspondences [19], it is interesting to examine the phase transition, the symmetries and the arithmetic subalgebra explicitly at the level of Hecke systems. r (P + K , P + The Hecke C ∗-algebra C ∗ O ) has a realization as a groupoid C ∗-algebra, and in order to exploit this in the characterization of equilibrium states we develop tools to characterize ground states and KMSβ-states for dynamics determined by 1-cocycles on certain groupoid C ∗-algebras. These groupoids are often nonprincipal, and in Section 1 we build upon the results of [26] to this effect. We believe our results will be of independent interest in other contexts, so we have presented them in a way that is independent from our original number theoretic motivation. The parametrization of KMS-states from [26] is restated in Theorem 1.1 and a similar characterization for ground states in terms of a boundary set of a 1-cocycle introduced by Renault [27] is obtained in Theorem 1.4. The specific results that are useful for our applications relate to groupoids obtained by reduction of a transformation group to a subset that is not invariant. In Theorem 1.9 we show that ground states are parametrized by all states on the C ∗-algebra of the boundary groupoid and, under some extra assumptions, KMSβ-states correspond to the tracial states of the boundary groupoid, Theorem 1.12. The boundary groupoid often has a noncommutative C ∗-algebra, in which case there is a phase transition for ground states. In Section 2 we review the definition and basic properties of the Hecke C ∗-algebra C ∗ K , P + O ), emphasizing its presentation in terms of distinguished double cosets, Proposition 2.1. This leads to realizations as a semigroup crossed product and as a groupoid C ∗-algebra, Proposition 2.2. The study of the boundary groupoid of the cocycle determined by the norm leads naturally to the consideration of integral ideals of minimal norm in their narrow class (of fractional ideals modulo principal ideals generated by totally positive elements of K). This allows us to realize the boundary r (P + 2 r (P + K , P + r (P + K , P + r (P + K , P + We begin Section 3 by making explicit various symmetry actions on C ∗ O ) for which the ground states that are invariant under the dual coaction of K ∗ C ∗-algebra explicitly as a subhomogeneous C ∗-algebra in Corollary 2.6. We state and prove our explicit characterization of equilibrium states, Theorem 2.7, which is our main result in Section 2. O ). We also O ) to a corner in the Bost-Connes C ∗-algebra AK from [21] exhibit the isomorphism of C ∗ in an explicit form that suits our present purposes. The main result is Theorem 3.3 where we show that if we pull back the corresponding corner of the arithmetic subalgebra for the Bost-Connes system for K obtained in [32] through this isomorphism, then we obtain an arithmetic subalgebra of C ∗ + have the expected 'fabulous' property with respect to the natural action of G(K ab/H+(K)), where H+(K) is the narrow Hilbert class field. O ) have almost as nice properties as the full Bost-Connes systems, including considerations with arithmetic models. An interesting new phenomenon is that the equilibrium states with good arithmetic properties are neither the KMS∞-states, nor the ground states, but a particular class of the latter ones. Acknowledgement This research was carried out through visits of M.L. to the Department of Mathematics at the University of Oslo and of N.S.L. to the Department of Mathematics and Sta- tistics at the University of Victoria. Both are grateful to their respective hosts for their hospitality. To summarize, we see now that the systems defined by the Hecke pairs (P + K , P + 1. Equilibrium states on groupoid algebras Let G be a locally compact second countable (Hausdorff) groupoid. The unit space of G will be denoted by G(0) and the range and source maps by r : G → G(0) and s : G → G(0), respectively. We will assume that G is ´etale, that is, r and s are local homeomorphisms, which among other things implies that G(0) is open in G. For x ∈ G(0) put Gx = r−1(x), Gx = s−1(x), and Gx x = Gx ∩ Gx. Notice that Gx continuous functions on G with compact support is a ∗-algebra under the convolution product x is a group for each x ∈ G(0), called the isotropy subgroup at x. The space Cc(G) of (f1 ∗ f2)(g) = Xh∈Gr(g) f1(h)f2(h−1g) and the involution f ∗(g) = f (g−1). Its C ∗-enveloping algebra is the full groupoid C ∗-algebra C ∗(G). Suppose c : G → R is a continuous 1-cocycle on G; by definition, this means that c is a continuous groupoid homomorphism of G to (R, +). Then there is a unique continuous one-parameter group {σc t : t ∈ R} of automorphisms of C ∗(G) such that t (f )(g) = eitc(g)f (g) for f ∈ Cc(G) and g ∈ G. σc (1.1) Recall that a C ∗-algebra A equipped with a continuous one-parameter group of automorphisms {σt}t∈R is referred to as a C ∗-dynamical system (A, σ). As these are the models for quantum sta- tistical mechanical time evolutions, their equilibrium states, or KMS-states, are of special interest. We refer the reader to [2] for a comprehensive treatment of the theory of KMS-states, and restrict ourselves to stating their definition in order to set the notation. For each value of an inverse temperature parameter β ∈ R, the σ-KMSβ-states of A, alternatively, the KMSβ-states of (A, σ), are the states φ of A such that φ(ab) = φ(bσiβ(a)), for all b ∈ A and all analytic elements a in A, which are those such that the map R ∋ t 7→ σt(a) extends to an entire function on C. There are also related notions of KMS∞-states and ground states that will be addressed shortly. 3 The KMSβ-states of (C ∗(G), σc) for finite β are characterized, in terms of measures on the unit space and traces on the C ∗-algebras of the isotropy subgroups, by the following theorem, which refines an earlier result of Renault [27, Proposition II.5.4]. Theorem 1.1 ([26, Theorem 1.3]). Suppose that c is a continuous R-valued 1-cocycle on a locally compact second countable ´etale groupoid G. Let σc be the corresponding dynamics on C ∗(G). Then, for every β ∈ R, there exists a one-to-one correspondence between the σc-KMSβ-states on C ∗(G) and the pairs (µ,{ϕx}x∈G(0)) consisting of a probability measure µ on G(0) and a µ-measurable field of states ϕx on C ∗(Gx x) such that: (i) µ is quasi-invariant with Radon -- Nikodym cocycle e−βc; (ii) ϕx(ug) = ϕr(h)(uhgh−1) for µ-a.e. x and all g ∈ Gx Namely, the state corresponding to (µ,{ϕx}x) is given by for µ-a.e. x. x and h ∈ Gx; in particular, ϕx is tracial ϕ(f ) =ZG(0) Xg∈Gx x f (g)ϕx(ug)dµ(x) for f ∈ Cc(G). Remark 1.2. KMSβ-states for β 6= 0 are automatically σ-invariant. theorem this implies that ϕx(ug) = 0 for µ-a.e. x ∈ G(0) and all g ∈ Gx x \ c−1(0). In the setting of the above (1.2) For β = 0, σ-invariance is sometimes included in the definition of KMS0-states. In this case (1.2) should be added to conditions (i) and (ii) of the theorem, see [26]. We aim to obtain next an analogous description of the set of ground states. For this purpose it will be most convenient for us to use the characterization of ground states given in [2, Propo- sition 5.3.19 (4)]. It states that given a continuous one-parameter group σ of automorphisms of a C ∗-algebra A, a state ϕ of A is a σ-ground state iff for every smooth compactly supported function F on R with support in (−∞, 0) we have where F is the inverse Fourier transform of F , so ϕ(σ F (a)∗σ F (a)) = 0 for all a ∈ A, and F (x) = 1 2π ZR σ F (a) =ZR F (y)e−ixydy, F (t)σt(a)dt. The main advantage of this formulation is that when A = C ∗(G), with the dynamics σc determined as above by a continuous 1-cocycle c on G, the operators σ F have a very simple form: Following Renault [27] we consider the subset σc F (f )(g) = F (c(g))f (g) for f ∈ Cc(G) and g ∈ G. (1.3) Z = {x ∈ G(0) c ≤ 0 on Gx} = {x ∈ G(0) c ≥ 0 on Gx} ⊂ G(0), which we call the boundary set of the cocycle c. (This is denoted by Min(c) in [27], but we shall eschew this notation here.) The boundary set can very well be empty, but when it is nonempty, we may consider the corresponding reduction of G: GZ = {g ∈ G r(g) ∈ Z, s(g) ∈ Z}, which we call the boundary groupoid of the cocycle c. In general, GZ may fail to be ´etale, so it may not have a good C ∗-algebra associated with it. However, the kernels c−1(0) of many naturally arising cocycles are ´etale groupoids and we have the following result. 4 Lemma 1.3. The set Z is invariant under c−1(0) and the boundary groupoid GZ coincides with the reduction of c−1(0) by Z. As a consequence, if the kernel groupoid c−1(0) is ´etale, then so is the boundary groupoid GZ . Proof. The first part of the lemma is contained in [27, Proposition I.3.16]. For the second part we just need to observe that the property of being ´etale is inherited by reduction to invariant subsets. Indeed, if W ⊂ c−1(0) is open and r defines a homeomorphism of W onto r(W ), then W ∩ r−1(Z) is contained in GZ and r defines a homeomorphism of W ∩ r−1(Z) onto r(W ) ∩ Z. Hence GZ is ´etale provided c−1(0) is ´etale. (cid:3) The following characterization of ground states is a refinement of [27, Proposition II.5.4]. Theorem 1.4. Suppose c is a continuous R-valued 1-cocycle on a locally compact second countable ´etale groupoid G. Let σc be the associated dynamics on C ∗(G) as in (1.1) and let Z ⊂ G(0) be the boundary set of c. If Z 6= ∅ and the boundary groupoid GZ is ´etale, then there is an affine isomorphism of the state space of C ∗(GZ ) onto the σc-ground state space of C ∗(G) that maps the state ψ of C ∗(GZ ) to the unique state ϕψ of C ∗(G) defined by ϕψ(f ) = ψ(fGZ ) for f ∈ Cc(G). If Z = ∅, then there are no σc-ground states on C ∗(G). Proof. Assume ϕ is a σc-ground state. Consider the corresponding GNS-triple (H, π, ξ). By Re- nault's disintegration theorem [27, Theorem II.1.21] the representation π is the integrated form of a representation of G on a measurable field of Hilbert spaces Hx, x ∈ G(0), with respect to a measure class [ν] on G(0). Identifying H withR ⊕ G(0) Hx dν(x), consider the vector field (ξx)x defining ξ. Then, for every f ∈ Cc(G), we have (1.4) (1.5) (π(f )ξ)x = Xg∈Gx D(g)−1/2f (g)gξs(g), where D is the Radon -- Nikodym cocycle defined by the quasi-invariant measure ν. We claim that ξx = 0 for ν-a.e. x ∈ G(0) \ Z. Take a point x0 ∈ G(0) \ Z. There exists g0 ∈ Gx0 such that c(g0) < 0. Choose an open neighborhood W of g0 such that rW : W → r(W ) and sW : W → s(W ) are homeomorphisms and c(W ) ⊂ (c(g0) − δ, c(g0) + δ) for some δ ∈ (0,−c(g0)). Take any smooth compactly supported function F on R such that supp F ⊂ (−∞, 0) and F ≡ 1 on [c(g0) − δ, c(g0) + δ]. Then, for any function f ∈ Cc(G) with support in W , it follows from (1.5) that (π(f )ξ)x =(D(r−1(x))−1/2f (r−1(x))r−1(x)ξs(r−1(x)), 0, if x ∈ r(W ), if x 6∈ r(W ), where r−1 is the inverse of rW . Hence, since σ F (f ) = f by (1.3) and our choice of F , we get kπ(σ F (f ))ξk2 =Zr(W ) D(r−1(x))−1f (r−1(x))2kξs(r−1(x))k2dν(x) =Zs(W ) f (s−1(x))2kξxk2dν(x), where s−1 is the inverse of sW . The assumption that ϕ is a ground state implies that the above integral is zero. Since this is true for any f ∈ Cc(G) supported in W , it follows that ξx = 0 for ν-a.e. x in the neighborhood s(W ) of x0. As x0 ∈ G(0) \ Z was arbitrary, we conclude that ξx = 0 for ν-a.e. x ∈ G(0) \ Z. This proves our claim. As ξ is a unit vector, we see in particular that if there exists a σc-ground state, then Z 6= ∅. Next, consider the representation of GZ on the field (Hx)x∈Z of Hilbert spaces. Integrating it with respect 5 to the measure νZ we get a representation πZ of C ∗(GZ ). The underlying space HZ =R ⊕ Z Hxdν(x) of this representation is a subspace of H. By the claim that we proved we have ξ ∈ HZ. The vector ξ defines a state ψ on C ∗(GZ ). As ϕ(f ) = (π(f )ξ, ξ) = RZ((π(f )ξ)x, ξx)dν(x), it is clear from (1.5) that ϕ(f ) = ψ(fGZ ) for all f ∈ Cc(G). Note that ψ is the only state on C ∗(GZ ) for which this identity can hold, since any function f ∈ Cc(GZ ) extends to a compactly supported continuous function on G. Conversely, assume Z 6= ∅ and take a state ψ on C ∗(GZ ). We want to show that the linear functional ϕψ on Cc(G) defined by ϕψ(f ) = ψ(fGZ ) extends (necessarily uniquely) to a σc-ground state on C ∗(G). For this we induce the GNS-representation of C ∗(GZ ) to a representation of C ∗(G). Induced representations of groupoid algebras have been studied by a number of authors, see [28] for an overview. The construction essentially goes back to [25]. Consider the space Y = s−1(Z) ⊂ G. At the level of groupoids the induction is done through the commuting actions G y Y x GZ arising from the actions of G on itself by left and right translations. These actions define a C ∗(G)-C ∗(GZ )- correspondence H, see [28] and references therein. Namely, the correspondence H is the completion of Cc(Y ) with respect to the C ∗(GZ )-valued inner product The left and right actions of C ∗(G) and C ∗(GZ ) are given by the usual convolution operators: hζ, ηi(g) = Xy∈Gr(g) ζ(y)η(yg). f (h)ζ(h−1y) for f ∈ Cc(G), (f ζ)(y) = Xh∈Gr(y) For every t ∈ R we define an operator Ut on Cc(Y ) by (Utζ)(h) = eitc(h)ζ(h). Since the cocycle c is zero on GZ , these operators preserve the inner product and commute with the right action of Cc(GZ ) ⊂ C ∗(GZ ). Hence they extend to a one-parameter group of unitary operators on the right C ∗-Hilbert C ∗(GZ )-module H. It is also clear from the definition that they implement the dynamics σc: ζ(yg)fZ (g−1) for fZ ∈ Cc(GZ ). (ζfZ)(y) = Xg∈Gs(y) Z Utaξ = σc t (a)Utξ for a ∈ C ∗(G) and ξ ∈ H. Consider the GNS-triple (Hψ, πψ, ξψ) defined by the state ψ on C ∗(GZ ). Then, using the C ∗- correspondence H, we can induce the representation πψ to a representation π of C ∗(G) on H = H ⊗C ∗(GZ ) Hψ. In order to define ϕψ assume first that Z ⊂ G(0) is compact, then its characteristic function 1Z is in Cc(Y ), where we regard Z a subset of Y ; note that Z is open Y , since Z = Y ∩ G(0). In this case ξ = 1Z ⊗ ξψ ∈ H is a unit vector and for each f ∈ Cc(G), we have π(f )ξ = fY ⊗ ξψ. Hence (π(f )ξ, ξ) = ψ(h1Z , fY i) = ψ(fGZ ), so ξ defines the required state ϕψ. It remains to check that ϕψ is a σc-ground state. The dynamics σc is implemented by the unitaries Wt = Ut ⊗ 1 on H. Since Wtξ = ξ, it follows that in order to check that ϕψ is a σc-ground state we have to show that the operators are zero on the vectors π(a)ξ, a ∈ C ∗(G), for all smooth compactly supported functions F with support in (−∞, 0). But this is clear, since for any f ∈ Cc(G) we have, similarly to (1.3), that W F =ZR F (t)Wtdt which is zero as c ≥ 0 on Y . This finishes the proof in the case when Z is compact. 6 W F π(f )ξ = (F ◦ c)Y fY ⊗ ξψ, When Z is only locally compact, for every ρ ∈ Cc(Z), 0 ≤ ρ ≤ 1, consider the vector ξρ = ρ ⊗ ξψ ∈ H of norm ≤ 1. It defines a positive linear functional θρ of norm ≤ 1 on C ∗(G). This functional is σc-ground by the same reasoning as above. We have θρ(f ) = ψ((ρ ◦ r)GZ fGZ (ρ ◦ s)GZ ) for f ∈ Cc(G). It follows that the required state ϕψ is the limit of the net (θρ)ρ in the weak∗ topology. σc-ground since every functional θρ is. It is (cid:3) Remark 1.5. We would like to draw attention at this point to a minor but potentially misleading omission in [26]. The formula analogous to (1.5) used in the proof of [26, Theorem 1.1] should have included the factor D(g)−1/2. This omission has essentially no consequences for the proof of [26, Theorem 1.1] and that result holds as stated; a corrected proof can be found in arXiv:1106.5912v3. Remark 1.6. The proof of the theorem implies that there is a unique completely positive contraction P : C ∗(G) → C ∗(GZ ) such that P (f ) = fGZ for f ∈ Cc(G). Indeed, using the C ∗(G)-C ∗(GZ )- correspondence H defined in the proof, this map is given by P (a) = h1Z , a1Zi, a ∈ C ∗(G), when Z is compact, and it is given by the pointwise norm limit of the maps a 7→ hρ, aρi, where ρ ∈ Cc(Z), 0 ≤ ρ ≤ 1, when Z is only locally compact. Remark 1.7. For graph C ∗-algebras, or more precisely, for their reductions by the projections corresponding to vertices, the above theorem recovers a result of Thomsen [30, Corollary 5.4]. Such a reduction 1vC ∗(E)1v is defined by a groupoid E with the unit space consisting of paths in a graph G starting at a vertex v. Thomsen considers a cocycle defined by a function F on the edges and describes the ground states of 1vC ∗(E)1v in terms of so called F -geodesics. It is not difficult to see that a path starting at v is an F -geodesic if and only if it belongs to the boundary set of the restricted cocycle. (As a side remark, if we considered the entire groupoid defining C ∗(E), then the boundary set of the corresponding cocycle would only be a subset of the F -geodesics.) Recall that, given a C ∗-dynamical system (A, σ), a state on A is called a σ-KMS∞-state if it is the limit in the weak∗ topology of a net of states (ϕi)i such that each ϕi is σ-KMSβi and βi → +∞, cf. [3]. Any KMS∞-state is a ground state by [2, Proposition 5.3.23], but the converse does not necessarily hold. This is clear from the drastic example of O∞ on which the gauge action has a ground state and no KMSβ-states for any β ∈ R∪{∞}, but also, more subtly, from recent examples that have an abundance of KMSβ-states for large β, [22, 8]. The following corollary of Theorem 1.4 makes this phenomenon particularly transparent. Corollary 1.8. In the setting of Theorem 1.4, suppose that Z 6= ∅ and the groupoid c−1(0) is ´etale. Let ϕψ be the σc-ground state on C ∗(G) corresponding to the state ψ of C ∗(GZ ). If ϕψ is a σc-KMS∞-state, then ψ is tracial. In particular, if C ∗(GZ ) is noncommutative, then the set of σc-KMS∞-states on C ∗(G) is strictly smaller than the set of σc-ground states. Proof. Let ψ be a state of C ∗(GZ ) and assume the associated ground state ϕψ is a σc-KMS∞-state. Then its restriction to the subalgebra C ∗(c−1(0)) ⊂ C ∗(G) is tracial. Since GZ is the reduction of c−1(0) by the closed c−1(0)-invariant set Z, the map Cc(c−1(0)) ∋ f 7→ fGZ extends to a homomorphism π : C ∗(c−1(0)) → C ∗(GZ ). As the restriction of ϕψ to C ∗(c−1(0)) equals ψ ◦ π, it follows that ψ is tracial. (cid:3) Next we consider a class of C ∗-dynamical systems arising from the reduction of a transformation groupoid to a subset that is not invariant. These include the systems of number theoretic origin based on Hecke algebras that constitute our main application in the next section. 7 Suppose that Γ is a countable discrete group acting on a second countable, locally compact topological space X. The associated transformation groupoid is the set Γ × X endowed with the product given by (γ′, γx)(γ, x) = (γ′γ, x); its unit space is naturally homeomorphic to X and its C ∗- algebra is the crossed product Γ ⋉ C0(X), which we write with the group on the left for consistency with the groupoid notation. We denote by γ 7→ uγ the canonical embedding of Γ as a group of unitaries into the multiplier algebra of Γ ⋉ C0(X), and we view the canonical embedding of C0(X) simply as inclusion. Then the products uγf for γ ∈ Γ and f ∈ C0(X) span a dense ∗-subalgebra of Γ ⋉ C0(X). There is a canonical (dual) coaction of Γ on Γ ⋉ C0(X) given by uγf 7→ uγf ⊗ λγ, where λγ are the generators of the group C ∗-algebra of Γ. For each multiplicative homomorphism + that can be interpreted as a regular action σN of +, there is a quotient coaction of R∗ N : Γ → R∗ the additive group R, viewed as the dual of R∗ +; it is determined by σN t (uγf ) = N (γ)ituγf. This dynamics on C ∗(Γ× X) = Γ ⋉ C0(X) is the one induced by the 1-cocycle c on Γ× X given by c(γ, x) := log N (γ), (γ, x) ∈ Γ × X. Suppose now that Y is a clopen subset of X such that ΓY = X and form the reduction of Γ× X by Y . By definition, this reduction is the groupoid Γ ⊠ Y := {(γ, y) ∈ Γ × Y γy ∈ Y } in which the product is the obvious restriction of that on Γ × X. The C ∗-algebra of this restricted groupoid is canonically isomorphic to the corner 1Y (Γ⋉C0(X))1Y of the crossed product Γ⋉C0(X). At the level of C ∗(Γ ⊠ Y ), the restricted dynamics is determined by the restricted cocycle, σc t (f )(γ, y) = N (γ)itf (γ, y), f ∈ Cc(Γ ⊠ Y ) and (γ, y) ∈ Γ ⊠ Y. The first step towards the explicit description of the σc-ground states of C ∗(Γ ⊠ Y ) is to identify the boundary groupoid in the present situation. Theorem 1.9. Suppose Γ is a countable discrete group acting on a locally compact second countable space X, and Y is a clopen subset of X such that ΓY = X. Assume N : Γ → (0,∞) is a multiplicative homomorphism and let c(γ, y) = log N (γ) for γ ∈ Γ and y ∈ Y be the associated 1-cocycle on Γ ⊠ Y . Then (1) the boundary set of c is Y0 := Y \(cid:16) [γ : N (γ)>1 γY(cid:17); (2) the boundary groupoid Γ ⊠ Y0 coincides with (ker N ) ⊠ Y0; (3) both c−1(0) = (ker N ) ⊠ Y and (ker N ) ⊠ Y0 are ´etale groupoids; (4) if Y0 6= ∅, then the map ψ 7→ ϕψ defined by ϕψ(f ) = ψ(f(ker N )⊠Y0) for f ∈ Cc(Γ ⊠ Y ) as in Theorem 1.4 is an affine isomorphism of the state space of C ∗((ker N ) ⊠ Y0) onto the σc-ground state space of C ∗(Γ ⊠ Y ); (5) if Y0 = ∅, there are no σc-ground states on C ∗(Γ ⊠ Y ). Proof. First notice that, by definition, the boundary set of c is {y ∈ Y : if y = γz ∈ Y for some γ ∈ Γ and z ∈ Y, then N (γ) ≤ 1}, If γY0 ∩ Y0 6= ∅ we also have Y0 ∩ γ−1Y0 6= ∅ hence N (γ) ≥ 1 and also so part (1) is clear. N (γ−1) ≥ 1 so N (γ) = 1. Thus the reduction of Γ ⊠ Y by Y0 is (ker N ) ⊠ Y0, proving part (2). The groupoid c−1(0) is ´etale because Y is clopen and Γ is discrete, and the rest of part (3) follows 8 from Lemma 1.3. Since the boundary groupoid of c is (ker N ) ⊠ Y0, parts (4) and (5) follow from Theorem 1.4. (cid:3) Lemma 1.10. In the situation of Theorem 1.9 let S0 := {γ ∈ Γ γY0 ∩ Y0 6= ∅}. Then S0 ⊂ S ∩ S−1 ⊂ ker N and N (γ) ≥ 1 for every γ ∈ S. If in addition the series S := {γ ∈ Γ : γY0 ∩ Y 6= ∅} and ζS(β) :=Xs∈S N (s)−β (1.6) has a finite abscissa of convergence, then (ker N ) ∩ S is a finite set. Proof. The chain of inclusions is part of the proof of Theorem 1.9, the rest follows from elementary considerations about Dirichlet series. (cid:3) Remark 1.11. Clearly we have the equality (ker N ) ⊠ Y0 = S0 ⊠ Y0 of sets. Therefore, if S0 is finite, then the orbit and the isotropy group of every y0 ∈ Y0 under the action of the groupoid (ker N ) ⊠ Y0 has at most S0 elements each, and thus C ∗(Γ ⊠ Y0) = C ∗((ker N ) ⊠ Y0) is a subhomogeneous C ∗- algebra. Next we consider a set-up in which the characterization of KMSβ-states from Theorem 1.1 has an explicit interpretation in terms of traces on the C ∗-algebra of the boundary groupoid. Theorem 1.12 (cf. [18, Proposition 1.2]). Let X, Y , Γ, N , c and Y0 be as in Theorem 1.9. Suppose that Y0 6= ∅, let S := {γ ∈ Γ : γY0 ∩ Y 6= ∅} be defined as in Lemma 1.10 and suppose that the series ζS(β) has a finite abscissa of convergence β0 < ∞. Assume that there are Borel subsets Yn ⊂ Y and elements γn ∈ Γ for n = 1, 2, . . . such that (i) for every neighborhood U of Y0 we have Y \ SU ⊂ ∪∞ (ii) N (γn) 6= 1 and γnYn = Yn for all n ≥ 1. n=1Yn; Then, for each β ∈ (β0, +∞], there is an affine isomorphism of the simplex of tracial states on C ∗((ker N ) ⊠ Y0) onto the simplex of σc-KMSβ-states on C ∗(Γ ⊠ Y ). Proof. We consider first the case β ∈ (β0, +∞). By Theorem 1.1 we know that in order to describe the KMSβ-states we first of all have to describe all quasi-invariant Borel probability measures on Y with Radon -- Nikodym cocycle e−βc, that is, measures µ such that if A ⊂ Y is Borel and γA ⊂ Y for some γ ∈ Γ, then (1.7) We claim that for any such measure µ we have µ(Y0) > 0 and the map µ 7→ µ(Y0)−1µY0 is an affine isomorphism between the set of all such measures and the set of Borel probability measures on Y0 invariant under the partial action of ker N . µ(γA) = N (γ)−βµ(A). By assumption (ii) and property (1.7), we have µ(Yn) = 0 for all n ≥ 1. Then assumption (i) implies that SU ∩ Y is a subset of Y of full measure for any open neighborhood U of Y0 in Y . Hence 1 = µ(SU ∩ Y ) ≤Xs∈S µ(sU ∩ Y ) ≤Xs∈S N (s)−βµ(U ) = ζS(β)µ(U ). It follows that µ(Y0) ≥ ζS(β)−1, which proves the first part of our claim. case, the measure µ defined by Next, let us show that µ is concentrated on ΓY0 ∩ Y = SY0 ∩ Y . Indeed, if this were not the µ(A) = µ(Y \ ΓY0)−1µ(A \ ΓY0) 9 would be a quasi-invariant Borel probability measure with Radon -- Nikodym cocycle e−βc. By what we have just proved this would mean µ(Y0) > 0, contradicting µ(Y0) = 0, which holds by construction. It follows that µ is completely determined by its restriction to Y0, which is a measure invariant under the partial action of ker N . To finish the proof of the claim it remains to show that any (ker N )-invariant probability measure on Y0 arises from a probability measure µ satisfying (1.7). Let ν be such a measure on Y0. We can choose distinct elements γi ∈ Γ and Borel subsets Zi ⊂ Y0 such that ΓY0 is the disjoint union of the sets γiZi. Define a Borel measure νβ on X by As in the proof of [16, Lemma 2.2], one sees that this measure does not depend on the choice of γi and Zi, satisfies (1.7) for all Borel subsets A ⊂ X and all γ ∈ Γ, and νβY0 = ν. Furthermore, if γ−1 i Y ∩ Zi 6= ∅ for some i, then γi ∈ S, which implies that νβ(A) =Xi N (γi)−βν(γ−1 i A ∩ Zi). νβ(Y ) ≤ Xi : γi∈S N (γi)−β ≤ ζS(β). Therefore µβ = νβ(Y )−1 νβY is the required measure on Y . This completes the proof of the claim. Later in the proof we will need the following consequence of the above considerations. By the c−1(0)-invariance of the boundary set, an element γ ∈ Γ can map a point in Y0 into Y \ Y0 only if N (γ) > 1. It follows that νβ(Y \ Y0) ≤Ps∈S : N (s)>1 N (s)−β, whence −1 µβ(Y0) ≥ 1 + Xs∈S : N (s)>1 N (s)−β  . Hence µβ(Y0) → 1 as β → +∞, and therefore the measures µβ converge in norm to the measure ν on Y0 we started with (considered as a measure on Y ). Let us now complete the classification of KMSβ-states for β ∈ (β0, +∞). By Theorem 1.1, every such state ϕ is given by a probability measure µ on Y satisfying (1.7) and a µ-measurable field of tracial states τy on the C ∗-algebras of the stabilizers Γy ⊂ Γ of points y ∈ Y such that τy = τγy(uγ · u−1 γ ) (1.8) whenever y ∈ Y and γy ∈ Y . As we showed above, the measure µ is concentrated on ΓY0 ∩ Y and is completely determined by its restriction to Y0. Condition (1.8) shows then that, modulo a set of µ-measure zero, the field (τy)y∈Y is also determined by its restriction to Y0. By Theorem 1.1, the measure ν = µ(Y0)−1µY0 together with the field (τy)y∈Y0 define a tracial state on C ∗((ker N ) ⊠ Y0). We therefore get an injective affine map from the set of KMSβ-states on C ∗(Γ ⊠ Y ) into the set of tracial states on C ∗((ker N ) ⊠ Y0). Conversely, if we now start with a tracial state on C ∗((ker N ) ⊠ Y0), we see that, again by Theorem 1.1, it is given by a probability measure ν on Y0 invariant under the partial action of ker N and a field of tracial states τy on C ∗(Γy), y ∈ Y0, satisfying (1.8) whenever y ∈ Y0 and γy ∈ Y0. As we showed above, the measure ν arises from a probability measure µ on Y satisfying (1.7). The field (τy)y∈Y0 extends in a unique way to a field of tracial states τy, y ∈ ΓY0, satisfying (1.8). As µ(Y \ ΓY0) = 0, the pair (µ, (τy)y∈Y ∩ΓY0) defines a KMSβ-state of C ∗(Γ ⊠ Y ). This finishes the proof of the result in the case of finite β. Turning to the remaining case of KMS∞-states, take a tracial state τ on C ∗((ker N ) ⊠ Y0). For each β ∈ (β0, +∞), let ϕβ be the KMSβ-state corresponding to τ . As we observed above, the measures µβ on Y defined by ϕβC0(Y ) converge in norm as β → +∞ to the measure ν on Y0 defined by τC0(Y0). From the construction of ϕβ it follows then that the states ϕβ converge in 10 norm to a KMS∞-state ϕ such that ϕ(f ) = τ (f(ker N )⊠Y ) for f ∈ Cc(Γ ⊠ Y ). But by Corollary 1.8 we know that any KMS∞-state has this form for some tracial state τ . Hence we get the required affine isomorphism. (cid:3) 2. Hecke algebras of orientation preserving affine groups By a Hecke pair (G, Γ) we mean a group G with a subgroup Γ such that every double coset ΓgΓ contains finitely many left and right cosets of Γ. The Hecke algebra H(G, Γ) of such a pair is defined as the vector space of complex valued Γ-bi-invariant functions on G supported on finitely many double cosets, endowed with the product (f1 ∗ f2)(g) = Xh∈G/Γ f1(h)f2(h−1g). This construction of course also makes sense for functions with values in any field K, in which case we denote the Hecke algebra by HK (G, Γ). For g ∈ G, we denote by [ΓgΓ] ∈ HK (G, Γ) the characteristic function of the double coset ΓgΓ. When K = C, we in addition have an involution on H(G, Γ) defined by f ∗(g) = f (g−1). Then the formula (λ(f )ξ)(g) = Xh∈G/Γ f (h)ξ(h−1g) defines a ∗-representation λ of H(G, Γ) on ℓ2(Γ\G). The Hecke C ∗-algebra of (G, Γ) is defined as the norm closure of λ(H(G, Γ)) ⊂ B(ℓ2(Γ\G)) and is denoted by C ∗ r (G, Γ), cf. [1]. We remark that in general H(G, Γ) does not admit a universal enveloping C ∗-algebra. Consider now a number field K with ring of integers O. An element of K is called totally + be the multiplicative group of totally + the monoid of totally positive algebraic integers in K, and positive if it is positive in every real embedding of K. Let K ∗ positive elements, with O× O∗ + := O∗ ∩ K ∗ + the group of totally positive units in O. Following [1, 21] we consider the pair + := O× ∩ K ∗ O :=(cid:18) 1 O0 O∗ P + + (cid:19) ⊂ P + K :=(cid:18) 1 K 0 K ∗ + (cid:19) . That this is a Hecke pair of groups can be seen by embedding it into a topological pair, see [21, Section 2], and also by a direct approach that counts the left cosets in every double coset, along the lines of [1, 14]. Both P + O are groups of affine transformations of K that preserve the orientation in every real embedding, so we will refer to them as the orientation preserving Hecke pair of affine groups associated to K. The associated Hecke C ∗-algebra C ∗ O ) has a universal property with respect to ∗-representations of H(P + O ), and also a presentation in terms of generators and relations arising from two classes of double cosets which we discuss briefly next. K and P + K , P + K , P + r (P + Following [1] we consider distinguished elements in C ∗ r (P + K , P + O ) given by two specific collections of double cosets. For each a ∈ O× + let Na = O/aO be the absolute norm of a and define µa := √Na (cid:20)P + O (cid:18)1 0 0 a(cid:19) P + O(cid:21) . 1 (2.1) For each r ∈ K let R(r) be the number of right cosets in the double coset of (cid:0)1 0 er := 1 R(r)(cid:20)P + O (cid:18)1 r 0 1(cid:19) P + O(cid:21) . 11 r 1(cid:1), and define (2.2) +-orbit of r modulo O, or, equivalently, the index of the subgroup O∗ An argument similar to the proof of [14, Lemma 1.2] shows that R(r) is equal to the cardinality of the O∗ + : ur = r (mod O)} in O∗ Proposition 2.1. The elements µa and er satisfy the following relations: +; see also [17, Proposition 1.3]. +r := {u ∈ O∗ µw = 1, µ∗ aµa = 1, µaµb = µab, ewr+b = er, e0 = 1; e∗ r = e−r, 1 eres = R(r) w ∈ O∗ +; a ∈ O× +; a, b ∈ O× +; r ∈ K, w ∈ O∗ +, b ∈ O; r ∈ K; 1 R(s) Xu∈O∗ +/O∗ +r Xv∈O∗ +/O∗ +s eur+vs, r, s ∈ K, and µaerµ∗ a = 1 Na Xb∈O/aO e r+b a , a ∈ O× +, r ∈ K/O. (2.3) (2.4) (2.5) (2.6) (2.7) (2.8) (2.9) (2.10) Moreover, the above relations give a presentation of H(P + as a C ∗-algebra. K , P + O ) as a ∗-algebra, and of C ∗ r (P + K , P + O ) It is immediate from the relations that µa depends only on a modulo O∗ +-orbit of r + O ∈ K/O. Thus, µ gives a representation of O× The proof goes along the same lines as the proofs of Proposition 1.7 and Theorem 1.10 in [14], with the minor difference that since the unit −1 is not totally positive, the generators er are not selfadjoint (for r 6= 0) in contrast to the situation studied in [14]. +, while er depends only +/O∗ on the O∗ + by isometries and the multiplication rule of the er reflects the multiplication of O∗ +-averages of generating unitaries in C ∗(K/O). By (2.9), the linear span of {er : r ∈ K} is a commutative, unital ∗-subalgebra of H(P + O ), which happens to be universal for the relations (2.7)-(2.9) in the category of ∗- O ) is a commutative C ∗-algebra and algebras. The closure of span{er : r ∈ K} inside C ∗ is universal, for the same relations, in the category of C ∗-algebras, see [14, Proposition 1.6] for a similar argument. K , P + K , P + r (P + The usual time evolution on the C ∗-algebra of a Hecke pair is defined in terms of the modular homomorphism which for our Hecke pair is given explicitly by O \P + ∆(g) := P + O gP + P + O gP + O O /P + O , g ∈ P + K , ∆(cid:18)1 y 0 x(cid:19) = NK(x)−1, y ∈ K and x ∈ K ∗ +, where NK(x) is the absolute norm of the principal fractional ideal (x), so NK(a) = Na = O/aO if a ∈ O×. Accordingly, we define a time evolution on C ∗ O gP + O ] O ]) := ∆(g)−it[P + O gP + σt([P + r (P + (2.11) K , P + O ) by for g ∈ P + K . On generators this amounts to σt(µa) = N −it a µa and σt(er) = er. 12 r (P + K , P + The C ∗-algebra C ∗ O ) can be written as a semigroup crossed product, which, in turn, can be realized as the C ∗-algebra of a reduction of a transformation groupoid, see [21, Proposition 2.2]. This will allow us to compute KMS-states and ground states using the techniques from Section 1. + by injective endo- The right hand side of (2.10) defines an action α of the semigroup O× +/O∗ morphisms of C ∗(er : r ∈ K) given by αa(er) := 1 Na Xb∈O/aO e r+b a . r (P + K , P + +/O∗ + ⋉ C ∗(er : r ∈ K). That this is indeed an action by endomorphisms can be proved directly using the presentation of C ∗(er : r ∈ K), but is actually obvious if we use the left-hand side of (2.10). A standard argument using universal properties now shows that C ∗ O ) is canonically isomorphic to the semigroup crossed product O× The spectrum of C ∗(er : r ∈ K) is most naturally described in terms of the ring AK,f of finite adeles associated with K. We denote by O ⊂ AK,f the compact subring of integral adeles. Viewed as an abelian group, O can be identified with the Pontryagin dual of K/O, although there is no such + inside the integral ideles O∗. It is a compact canonical identification. Consider the closure O∗ group under multiplication, which also coincides with the natural profinite compactification of O∗ + obtained from its action on K/O. Then C ∗(er : r ∈ K) ∼= C( O/O∗ + denotes the orbit space for the action of O∗ +/O∗ +. Since the action of O∗ + on AK,f is not trivial, multiplication does not induce an action of Γ on AK,f , but Γ acts by multiplication on the orbit space X := AK,f /O∗ +, which is a totally disconnected, second countable, locally compact Hausdorff space. This gives rise to the transformation groupoid + of totally positive elements in K acts by multiplication on AK,f . Let Γ := K ∗ + on O by multiplication. +), where O/O∗ The group K ∗ + of O∗ Γ × X = K ∗ +/O∗ + × AK,f /O∗ +, and we will be interested in its reduction by the subset Y := O/O∗ ⊠ O/O∗ C ∗(er : r ∈ K). The C ∗-algebra of the groupoid K ∗ +)(cid:1) p O/O∗ where p O/O∗ this corner is also isomorphic to the semigroup crossed product O× +/O∗ algebra is isomorphic to denotes the characteristic function of O/O∗ + ⋉ C0(AK,f /O∗ +(cid:0)K ∗ +/O∗ +/O∗ p O/O∗ + + + , +. By the dilation/extension results of [13], +). The last C ∗- + ⋉ C( O/O∗ +, which is the spectrum of + coincides with the corner It will be useful to dig a little deeper into the construction of the above isomorphisms in order O× +/O∗ + α⋉ C ∗(er : r ∈ K) ∼= C ∗ r (P + K , P + O ). to identify explicitly the images of the canonical generators of the Hecke algebra. Proposition 2.2. Let χ : AK,f → T be a character implementing a self-duality of AK,f in which the annihilator of O equals O. There is an isomorphism C ∗ r (P + K , P + O ) ∼= p O/O∗ +/O∗ + ⋉ C0(AK,f /O∗ = C ∗(K ∗ +/O∗ + ⊠ O/O∗ +) (2.12) determined by and +)(cid:1)p O/O∗ + λ(x)p O/O∗ + , x ∈ O× +, +(cid:0)K ∗ + µx 7→ p O/O∗ R(y)ZO∗ 1 + 13 ey 7→ χ(· uy)du ∈ C( O/O∗ +), y ∈ K. + + λ(x)p O/O∗ + are mapped into p O/O∗ Proof. We refer to the proof of [21, Proposition 2.2]. By the arguments there and [17, Lemma 2.3], the elements µx for x ∈ O× . On the other hand, the images of the elements R(y)ey for y ∈ K can be computed by first mapping y into the corresponding generator of C ∗(K/O) = C ∗(AK,f / O), then using an isomorphism C ∗(K/O) ∼= C( O), and finally + ⊂ O∗. The isomorphism C ∗(K/O) ∼= C( O) depends by averaging the result by the action of O∗ on the choice of χ and maps the generator λy+O of C ∗(K/O) into the function χ(· y). Remark 2.3. Note that the integrals RO∗ +, so the image of R(y)ey is a normalized sum of finitely many characters of O of the form χ(·uy), u ∈ O∗ +. +) defined + → (0, +∞). Our next goal is to understand the boundary set Each prime ideal p determines a discrete valuation v = vp on K, defined to be the integer-valued function on the set of integral ideals giving the largest power of p that divides a given ideal. This can be extended to a valuation defined on O and on AK,f using a uniformizing element πp in place of the prime ideal p, and allowing finitely many finite negative powers. Clearly, discrete valuations factor through the quotient O/O∗ +. Let VK,f denote the set of all such discrete valuations. For each integral ideal a define K , P + O ) corresponds to the dynamics on C ∗(K ∗ +/O∗ +/O∗ χ(· uy)du are averages over finite quotients of O∗ by the homomorphism NK : K ∗ of the associated cocycle on K ∗ The dynamics σ on C ∗ ⊠ O/O∗ +. ⊠ O/O∗ +/O∗ r (P + (cid:3) + + + Ωa := {ω ∈ O/O∗ + : v(ω) = v(a) for all v ∈ VK,f}. +/O∗ Y0 := ( O/O∗ ⊠ O/O∗ +) \ Denote by Cl+ K the narrow class group of K, the quotient of the group of fractional ideals by the These are mutually disjoint sets such that ⊔aΩa = (A∗ subgroup of principal fractional ideals with a totally positive generator. Proposition 2.4. For each narrow ideal class c ∈ Cl+ K , let ac,1, ac,2, . . . , ac,kc be the integral ideals in c that have minimal norm among all integral ideals in c. Then the boundary set Y0 of the cocycle determined by NK on K ∗ + decomposes as the disjoint union K,f ∩ O)/O∗ +. + Ωac,j . +: NK (x)>1 (x O/O∗ K G1≤j≤kc +) = Gc∈Cl+ Proof. Recall that we denote O/O∗ ideals (x) and (y). [x∈K ∗ + by Y . For x, y ∈ K ∗, denote by [x, y] the lcm of the fractional Let c be a narrow class and let ω ∈ Ωac,j for some 1 ≤ j ≤ kc. If ω ∈ Y ∩ xY for some x ∈ K ∗ +, then ω ∈ Ωb[1,x] for some integral ideal b, and since the sets Ωa are disjoint, we have b[1, x] = ac,j. But then ac,jx−1 = b[1, x]x−1 = b[x−1, 1] is an integral ideal in the same narrow class as ac,j, so NK(x) ≤ 1. Hence ω /∈ xY for every x with NK (x) > 1, and thus ω ∈ Y0. Conversely, suppose now ω ∈ Y0. We claim that the valuation vector of ω has finite support and finite values. Otherwise the superideal corresponding to those valuations would have infinitely many prime ideal factors (counted with multiplicity) in the same narrow class. The product of any subcollection of these with cardinality the narrow class number would produce a totally positive a ∈ O× + such that ω ∈ aY , which would contradict the assumption that ω ∈ Y0 because NK(a) > 1. This proves the claim and implies that ω ∈ Ωa for an integral ideal a. Let c be the narrow class of a and suppose b is an integral ideal in c. Then there exists x ∈ K ∗ such that a = xb, so we have ω ∈ xY . By assumption, ω ∈ Y0, so NK(x) ≤ 1. This implies that a is of minimal norm in its class, that is, a = ac,j for some j = 1, . . . , kc. Corollary 2.5. In the context of Proposition 2.4, the subset S0 ⊂ Γ from Lemma 1.10 is a finite subset (K ∗ +,1 is the group of totally positive elements of norm one. +, where K ∗ +)0 of K ∗ (cid:3) + +/O∗ +,1/O∗ 14 This set consists of principal ideals factoring as the ratio of two ideals of minimal norm in the same narrow class: (K ∗ +/O∗ +)0 = {ac,ia−1 c,j : c ∈ Cl+ K i, j = 1, 2, . . . , kc}. +. By Proposition 2.4, ω ∈ Ωac,j and xω ∈ Ωad,i K and i, j. But then xac,j = ad,i, so c = d and (x) = ac,ia−1 c,j . Proof. Suppose ω ∈ Y0 and xω ∈ Y0 for some x ∈ K ∗ for some c, d ∈ Cl+ elements of Ωac,i ⊂ Y0, so γ ∈ (K ∗ Corollary 2.6. In the context of Proposition 2.4 the boundary groupoid K ∗ isotropy and On the other hand, every γ ∈ K ∗ +/O∗ + that factors as ac,ia−1 +/O∗ +)0. c,j maps elements of Ωac,j ⊂ Y0 to (cid:3) +/O∗ + ⊠ Y0 has trivial C ∗(K ∗ +/O∗ + ⊠ Y0) ∼= Matkc(C)  Mc∈Cl+ K +).  ⊗ C( O∗/O∗ Proof. The groupoid K ∗ ⊠ Y0 has trivial isotropy, because none of the elements in Y0 have the valuation of a zero-divisor. Therefore this groupoid is an equivalence relation on Y0. By the proof of the previous corollary, the equivalence class of every point ω ∈ Ωac,j consists precisely of kc elements, with one point in each Ωac,i. It follows that +/O∗ + C ∗(K ∗ +/O∗ + ⊠ Y0) ∼= Mc∈Cl+ K Matkc(C(Ωac,1)) ∼= Matkc(C)  Mc∈Cl+ K +),  ⊗ C( O∗/O∗ where we used that the sets Ωa are all homeomorphic to O∗/O∗ +. Theorem 2.7. Let K be an algebraic number field and consider the time evolution σ on C ∗ defined by (2.11). Let (cid:3) r (P + K , P + O ) Y0 := ( O/O∗ +) \ be the set described in Proposition 2.4. Then [x∈K ∗ +: NK (x)>1 (x O/O∗ +) (1) there is an affine isomorphism of the ground state space of (C ∗ r (P + K , P + O ), σ) onto the state space of C ∗(K ∗ ⊠ Y0); + +/O∗ r (P + K , P + C ∗-algebra C ∗ +, and Theorem 1.9. (2) for each β ∈ (1, +∞], there is an affine isomorphism of the simplex of σ-KMSβ-states on the K × ( O∗/O∗ +). ⊠ Y ) O ) onto the simplex of Borel probability measures on Cl+ Proof. Part (1) is an immediate consequence of the isomorphism C ∗ from (2.12), where Y = O/O∗ Part (2) for β ∈ (1,∞) is already proved in [21, Theorem 2.5] (where the interval (0, 1] is also dealt with) using the classification of KMS-states on the full Bost-Connes system associated with K. Here we will give a more direct argument based on Theorem 1.12 and also consider the case β = ∞. + from Lemma 1.10 and to determine the region of convergence of its zeta function ζS(β). For this, suppose γ ∈ S; then there exists ω0 ∈ Y0 such that γω0 = ω ∈ Y . By Proposition 2.4, ω0 ∈ Ωac,j for some narrow class c and some j = 1, 2, . . . , kc, so we have v(ω0) = v(ac,j) and v(γ) = v(ω) − v(ac,j) ≥ −v(ac,j). Choose d ∈ O× + such that (d) ⊂ ∩c,jac,j, so that v(d) ≥ v(ac,j) for all v ∈ VK,f , proving that S ⊂ ( 1 +). As a consequence ζS(β) has the same abscissa of convergence as the Dedekind zeta function of K. The first step is to characterize the set S ⊂ Γ = K ∗ O ) ∼= C ∗(K ∗ +/O∗ +/O∗ r (P + K , P + +/O∗ dO× + The next step is to come up with a countable collection of Borel subsets of Y and associated group elements satisfying properties (i) and (ii) of Theorem 1.12. For each v ∈ VK,f , let Yv := {ω ∈ Y : v(ω) = ∞}, 15 and let xv be a totally positive element whose principal ideal is a power of the prime ideal p corresponding to the discrete valuation v, e.g., let h+ denote the narrow class number of K and choose xv to be a totally positive generator of the principal ideal ph+. Notice that Sv Yv is the image in Y of the set of zero divisors in O. We clearly have NK (xv) > 1 and xvYv = Yv for every v ∈ VK,f , so property (ii) in Theorem 1.12 is satisfied. It remains to prove that the collection {Yv}v also has property (i), that is, for every open set U containing Y0 and every ω ∈ Y \∪∞ n=1Yn there exists γ ∈ S such that ω ∈ γU . Suppose U is a basic open set containing Y0; we may assume that U is of the form Uc,j = [(c,j) {ω ∈ Y : v(ω) = v(ac,j) for v ∈ F}, UF = [(c,j) where F is a finite subset of VK,f containing every valuation that does not vanish on ac,j for some c and some j = 1, 2, . . . , kc. Yv. Then ev := v(ω) < ∞ for each v ∈ F , so b := Qv∈F pev Let ω ∈ Y \Sv∈VK,f is an integral ideal. Let ac,1 be an ideal of minimal norm in the narrow class of b, so there exists a totally positive element x such that (x)ac,1 = b. By assumption ac,1 is supported in F , and so is b, hence (x) is supported in F . For any ω0 ∈ Ωac,1 ⊂ Y0 we have xω0 ∈ Ωb ⊂ Y , so (x) ∈ S. We also have v(x−1ω) = v(ω) ≥ 0 for v ∈ VK,f \ F , because the support of (x) is contained in F , and v(x−1ω) = v(ac,1) − v(b) + v(ω) = v(ac,1) for v ∈ F . This shows that x−1ω ∈ UF , finishing the proof of property (i) in Theorem 1.12. r (P + O ) for β ∈ (1, +∞] are para- metrized by the tracial states on the groupoid C ∗-algebra of K ∗ ⊠ Y0. By Corollary 2.6 the simplex of such tracial states is affinely homeomorphic to the simplex of probability measures on Cl+ (cid:3) By Theorem 1.12 we conclude that the KMSβ-states of C ∗ K , P + +/O∗ + v K × ( O∗/O∗ +). Corollary 2.8. The extremal KMSβ-states of (C ∗ by Cl+ K × ( O∗/O∗ +). r (P + K , P + O ), σ) for β ∈ (1, +∞] are parametrized We point out that by class field theory the set Cl+ +) supports a free transitive action of the Galois group G(K ab/K) of the maximal abelian extension K ab of K, but, unlike the situation for Bost-Connes type systems, this does not seem to arise from an action of G(K ab/K) as symmetries of C ∗ K × ( O∗/O∗ r (P + K , P + O ). 3. Arithmetic subalgebra and fabulous states In addition to the time evolution σ, the C ∗-algebra C ∗ r (P + K , P + O ) carries several other natural actions. K , P + r (P + First, using the presentation of C ∗ O× +/O∗ O ) as a semigroup crossed product + α⋉ C ∗(er : r ∈ K), we have the dual action α of the Pontryagin dual of K ∗ +. At the level of generators it is given by α(η)(cid:18)(cid:20)P + O (cid:18)1 y 0 x(cid:19) P + O(cid:21)(cid:19) = η(x)(cid:20)P + O (cid:18)1 y 0 x(cid:19) P + O(cid:21) for characters η of K ∗ think of α as a coaction of K ∗ + and call it the dual coaction. +. In order to not confuse Pontryagin duals with adic completions, we will Second, the group O∗ acts by the automorphisms τ (u) defined by O (cid:18)1 u−1y O(cid:21)(cid:19) =(cid:20)P + τ (u)(cid:18)(cid:20)P + 0 x(cid:19) P + O (cid:18)1 y 0 x (cid:19) P + O(cid:21) , 16 (3.1) where the product u−1y is defined by approximating u−1 in O by elements v in O and using these elements in the above formula instead of u−1, the result being independent of v sufficiently close to u−1. Since O∗ is abelian, we could of course have used u instead of u−1 in the above definition, but our choice is preferable for consistency with the full Bost-Connes system, as we will see later. +, which has a class field theoretic interpreta- tion. Namely, we have a homomorphism O∗ → G(K ab/K) obtained by restricting the Artin map K → G(K ab/K) to O∗. Using basic properties of the Artin map it is easy to check that the rK : A∗ + and that its image is G(K ab/H+(K)), where H+(K) is the maximal kernel of this restriction is O∗ abelian extension of K that is unramified at all finite places, see, e.g., [21, Proposition 1.1]. The action τ factors through the group O∗/O∗ Motivated by [3] we now give the following definition. Definition 3.1. Let A be an O∗-invariant K-subalgebra of C ∗ r (P + C ∗ O ) is fabulous with respect to A if K ab is generated by φ(A) as a K-algebra and O ). We say that a state φ on K , P + K , P + r (P + φ(τ (u)(a)) = rK(u−1)(φ(a)) for all u ∈ O∗ and a ∈ A. The algebra A is called an arithmetic subalgebra if there is a fabulous state with respect to A and the C-algebra generated by A in C ∗ Remark 3.2. Since rK ( O∗) = G(K ab/H+(K)), it may be more natural to require that A is an H+(K)-algebra. This is anyway the case for the algebra A we define below. O ) is dense. K , P + r (P + r (P + K , P + For the full Bost-Connes system (AK , σK ), whose construction we recall below, it is shown in [32] that there is a canonical choice of an arithmetic subalgebra AK. By [21, Theorem 2.4] we know also that C ∗ O ) can be realized as a full corner pK AKpK of AK . Similarly to the isomorphism in Proposition 2.2, this realization, however, depends on the choice of a character χ of AK,f to implement the duality between K/O and O. Our main result about arithmetic subalgebras and fabulous ground states for the Hecke C ∗-algebra of the orientation preserving affine group of a number field is as follows. Theorem 3.3. Let AK be the arithmetic subalgebra of AK associated in [32] to the Bost-Connes system of a number field K; then the K-subalgebra A ⊂ C ∗ O ) corresponding to pKAKpK in the isomorphism of C ∗ O ) to pKAK pK has the following properties: (1) A is independent of the choice of a character χ used to identify C ∗ O ) with the corner pKAKpK of the full Bost-Connes algebra. Explicitly, A coincides with the fixed point algebra K , P + K , P + K , P + r (P + r (P + r (P + HK ab(P + K , P + O ) O∗ K , P + O ) over the field K ab. This action is defined by with respect to an action β of the group O∗ on the Hecke algebra HK ab(P + (P + β(u)(cid:18)a(cid:20)P + O (cid:18) 1 NKQ(u)u−1r O(cid:21)(cid:19) = rK(u)(a)(cid:20)P + 0 1 (cid:19) P + O (cid:18) 1 r (cid:19) P + 0 1 O(cid:21) for a ∈ K ab, K , P + O ) of the pair where NKQ is the norm map for the field extension K/Q, extended to a map A∗ and rK (u)(a) is simply the Galois action of rK (u) on a. K → A∗ Q, O ). Specifically, every extremal σ-ground state (2) A is an arithmetic subalgebra of C ∗ r (P + invariant under the dual coaction of K ∗ K , P + + is a fabulous state with respect to A. Before we embark on the proof of the theorem, we shall revisit the basic facts of the construction of the Bost-Connes system (AK , σK ) associated to the number field K, see [11, 18]. The multiplication action of O∗ on AK,f can be induced to an action of G(K ab/K) using the homomorphism O∗ → G(K ab/K) obtained from restricting the Artin map rK to the subgroup O∗. 17 This gives rise to the balanced product XK := G(K ab/K) × O∗ AK,f . The left action of O∗ on G(K ab/K) × AK,f that is responsible for the balancing is consistent with the action of A∗ K,f on G(K ab/K) × AK,f given by g(γ, x) = (γrK (g)−1, gx) K,f and x ∈ AK,f . There is therefore an action of the quotient group A∗ for g ∈ A∗ identified with the discrete group JK of fractional ideals in K, on XK . Let J + subsemigroup of integral ideals. Finally, restricting to the clopen subset K,f / O∗, naturally K ⊂ JK be the of XK and letting p denote the projection in JK ⋉ C0(XK ) corresponding to YK, we obtain the C ∗-algebra of the Bost-Connes system: YK := G(K ab/K) × O∗ O where the isomorphism is from Theorems 2.1 and 2.4 of [13]. AK := C ∗(YK ⊠ JK) = p(cid:0)JK ⋉ C0(XK )(cid:1)p ∼= J + K ⋉ C(YK), The dynamics σK on AK , and more generally on JK ⋉ C0(XK ), is defined using the absolute norm NK on JK: σK t (f λg) = NK(g)itf λg for f ∈ C0(XK ) and g ∈ JK . The Galois group G(K ab/K) acts on XK by γ(γ′, m) = (γγ′, m) for γ, γ′ ∈ G(K ab/K) and m ∈ AK,f . This gives an action τK of G(K ab/K) on AK and JK ⋉C0(XK ) such that (3.2) τK(γ)(f λg) = f (γ−1·)λg for f ∈ C(XK), γ ∈ G(K ab/K) and g ∈ JK. For the proof of Theorem 3.3 we will need the notation and a more explicit description of the isomorphism established in [21, Theorem 2.4] together with some of the details of its proof, so we summarize those results next. Theorem 3.4. (cf. compact open subset [21, Theorem 2.4]). Denote by pK the projection in AK associated to the ZH+(K) := G(K ab/H+(K)) × O∗ O + ∼= G(K ab/H+(K)). Then of YK and recall that the Artin map induces an isomorphism O∗/O∗ the obvious map O → G(K ab/H+(K)) × O given by a 7→ (e, a) gives a homeomorphism of O/O∗ + onto ZH+(K). Using this homeomorphism and viewing K ∗ + as a subgroup of JK we obtain an isomorphism +/O∗ p O/O∗ +(cid:0)K ∗ +/O∗ + ⋉ C0(AK,f /O∗ Composing this with the isomorphism C ∗ r (P + O ) ∼= p O/O∗ from Proposition 2.2 gives an isomorphism FK : C ∗ K , P + + ∼= pKAKpK = pK(cid:0)JK ⋉ C0(XK )(cid:1)pK. +)(cid:1)p O/O∗ +/O∗ +(cid:0)K ∗ + ⋉ C0(AK,f /O∗ +)(cid:1)p O/O∗ + r (P + K , P + O ) → pKAK pK. 18 βK(γ)(f pλgp) = γ(f (γ−1·))pλgp, for γ ∈ G(K ab/K), g ∈ JK and f ∈ C(YK, K ab). In order to understand what happens to the action βK under the isomorphism FK , recall an important observation from [31]. Denoting by ρ : G(K ab/K) → G(Qab/Q) the homomorphism given by restriction, we have (3.3) ρ ◦ rK = rQ ◦ NKQ, see [31, Corollary 1, page 246]. From this we get the following result. Lemma 3.5 (cf. [14, Theorem 4.4]). Under the isomorphism FK the action βKG(K ab/H+(K)) on pK(J + Proof. Since for each y ∈ K the image of K ⋉ C(YK, K ab))pK is transformed into the action β defined in Theorem 3.3. R(y)ey =(cid:20)P + O (cid:18)1 y 0 1(cid:19) P + O(cid:21) As an immediate corollary we see that FK intertwines the action τ of O∗ on C ∗ r (P + in (3.1) with the action (τK ◦ rK) O∗ on pK AKpK obtained from (3.2). On the algebra K , P + O ) defined p O/O∗ +(cid:0)K ∗ +/O∗ + ⋉ C0(AK,f /O∗ +)(cid:1)p O/O∗ + the same action is implemented by multiplication by the elements of O∗ on AK,f /O∗ +. As for the arithmetic subalgebra of AK , the K-subalgebra AK of AK is generated by the elements pλgp for g ∈ JK and the algebra of locally constant G(K ab/K)-equivariant functions YK → K ab, K ⋉ C(YK, K ab) the K ab-subalgebra of see [32, Section 10]. AK = J + K ⋉C(YK) generated by the elements pλgp for g ∈ JK and the algebra C(YK, K ab) of locally constant K ab-valued function on YK, then AK is the fixed point K-subalgebra of J + K ⋉ C(YK, K ab) under the action βK of G(K ab/K) defined by In other words, if we denote by J + in C( O/O∗ acters of O of the form χ(·vy), v ∈ O∗ rK(u)(χ(u−1z)) = χ(NKQ(u)u−1z) for all z ∈ AK,f . Equivalently, we must show that +) under the isomorphism from Proposition 2.2 is an average of finitely many char- +, it suffices to show that, for every u ∈ O∗, we have rK (u)(χ(z)) = χ(NKQ(u)z) for all z ∈ AK,f . r (P + K , P + K , P + But this follows from (3.3) and the fact that rQ(w)(χQ(1)) = χQ(w) for any character χQ of AQ,f and w ∈ Z∗. (cid:3) Proof of Theorem 3.3. From the explicit form of the isomorphism FK given in Theorem 3.4 and Proposition 2.2 we see that the preimage of pK(J + O ) coincides with the Hecke algebra HK ab(P + O ), since the functions χ(·y) for y ∈ O span over Qab the space of locally constant Qab-valued functions on O, hence they span over K ab the space of locally constant K ab-valued functions on O. Part (1) of the theorem now follows from Lemma 3.5. In order to prove the second part of Theorem 3.3 it remains to show that every extremal σ-ground state invariant under the dual coaction of K ∗ + is a fabulous state. By Theorem 2.7 the ground states correspond to the states of C ∗(K ∗ ⊠ Y0), among which the ones that are invariant under the dual coaction of K ∗ + correspond to the probability measures on Y0. The ground state corresponding to such a measure is extremal if and only if the measure is a point mass. So the space of extremal ground states that are invariant under the dual coaction can be identified with Y0. K ⋉ C(YK, K ab))pK in C ∗ Using our identification of O/O∗ + with ZH+(K) = G(K ab/H+(K)) × O∗ O we view Y0 as a subset If B is the K- of ZH+(K). Then in order to finish the proof it suffices to check the following. algebra of locally constant G(K ab/H+(K))-equivariant functions ZH+(K) → K ab, then for every +/O∗ + 19 point z ∈ Y0 ⊂ ZH+(K) the set {f (z) f ∈ B} generates K ab as a K-algebra. The proof of this is similar to the analogous property of AK in [32] and relies on the following two key points: the G(K ab/H+(K))-space ZH+(K) is a projective limit of finite G(K ab/H+(K))-spaces Zn and every point z ∈ Y0 has trivial stabilizer in G(K ab/H+(K)). To see this, take any a ∈ K ab and let H be the stabilizer of a in G(K ab/H+(K)). Since H is an open subgroup of the compact group G(K ab/H+(K)), there is a sufficiently large n such that the stabilizer of the image zn of z in Zn is contained in H. We can then define a G(K ab/H+(K))-equivariant function fn on Zn such that fn(zn) = a by letting fn(γzn) = γ(a) and f = 0 outside the orbit of zn. Then composing fn with the projection ZH+(K) → Zn we get a function f ∈ B such that f (z) = a. (cid:3) References [1] J.-B. Bost and A. Connes, Hecke algebras, type III factors and phase transitions with spontaneous symmetry breaking in number theory, Selecta Math. (N.S.) 1 (1995), 411 -- 457. [2] O. Bratteli and D.W. Robinson. Operator algebras and quantum statistical mechanics. 2. Equilibrium states. Models in quantum statistical mechanics. Second edition. Texts and Monographs in Physics. Springer-Verlag, Berlin, 1997. [3] A. Connes and M. Marcolli, From physics to number theory via noncommutative geometry, in: Frontiers in number theory, physics, and geometry. I, 269 -- 347, Springer, Berlin, 2006. [4] A. Connes, M. Marcolli and N. Ramachandran, KMS states and complex multiplication, Selecta Math., New Series, 11 (2005), 325 -- 347. [5] G. Cornelissen, B. de Smit, X. Li, M. Marcolli and H. Smit, Reconstructing global fields from Dirichlet L-series, preprint arXiv:1706.04515v1 [math.NT]. [6] G. Cornelissen, B. de Smit, X. Li, M. Marcolli and H. Smit, Reconstructing global fields from dynamics in the abelianized Galois group, preprint arXiv:1706.04517v1 [math.NT]. [7] J. Cuntz, C ∗-algebras associated with the ax + b-semigroup over N, K-theory and noncommutative geometry, EMS Ser. Congr. Rep., Zurich, 2008, pages 201 -- 25. [8] J. Cuntz, C. Deninger and M. Laca, C ∗-algebras of Toeplitz type associated to algebraic number fields, Math. Ann. 355 (2013), 1383 -- 1423. [9] J. Cuntz and X. Li, The regular C ∗-algebra of an integral domain, Quanta of Math, Clay. Math. Proc. 11, Amer. Math. Soc., Providence, RI, 2010, pages 149 -- 170. [10] J. Cuntz and X. Li, C ∗-algebras associated with integral domains and crossed products by actions on adele spaces, J. Noncommut. Geom. 5 (2011), no. 1, 1 -- 37. [11] E. Ha and F. Paugam, Bost-Connes-Marcolli systems for Shimura varieties. I. Definitions and formal analytic properties, IMRP Int. Math. Res. Pap. 5 (2005), 237 -- 286. [12] Y. Kubota and T. Takeishi, Reconstructing the Bost-Connes semigroup actions from K-theory, preprint arXiv:1709.03281v1[math.OA]. [13] M. Laca, From endomorphisms to automorphisms and back: dilations and full corners, J. London Math. Soc. 61 (2000), 893 -- 904. [14] M. Laca and M. van Frankenhuijsen, Phase transitions on Hecke C ∗-algebras and class-field theory over Q, J. Reine Angew. Math. 595 (2006), 25 -- 53. [15] M. Laca and M. van Frankenhuijsen, Phase transitions with spontaneous symmetry breaking on Hecke C ∗-algebras from number fields, in Noncommutative Geometry and Number Theory, Consani and Marcolli, eds. Aspects of Mathematics E37, Vieweg Verlag, (2006), 205 -- 216. [16] M. Laca, N.S. Larsen and S. Neshveyev, Phase transition in the Connes-Marcolli GL2-system, J. Noncommut. Geom. 1 (2007), no. 4, 397 -- 430. [17] M. Laca, N.S. Larsen and S. Neshveyev, Hecke algebras of semidirect products and the finite part of the Connes- Marcolli C ∗-algebra, Adv. Math. 217 (2008), 449 -- 488. [18] M. Laca, N.S. Larsen and S. Neshveyev, On Bost-Connes type systems for number fields, J. Number Theory, 129 (2009), 325 -- 338. [19] M. Laca and S. Neshveyev, KMS states of quasi-free dynamics on Pimsner algebras, J. Funct. Anal. 211 (2004), 457 -- 482. [20] M. Laca and S. Neshveyev, Type III1 equilibrium states of the Toeplitz algebra of the affine semigroup over the natural numbers, J. Funct. Anal. 261 (2011), 169 -- 187. 20 [21] M. Laca, S. Neshveyev and M. Trifkovic, Bost-Connes systems, Hecke algebras, and induction, J. Noncommut. Geom. 7(2013), 525 -- 546, DOI 10.4171/JNCG/125 [22] M. Laca and I. Raeburn, Phase transition on the Toeplitz algebra of the affine semigroup over the natural numbers, Adv. Math. 225 (2010), 643 -- 688. [23] X. Li, On K-theoretic invariants of semigroup C ∗-algebras attached to number fields, Adv. Math. 264(2014), 371 -- 395. [24] X. Li and W. Luck, K-theory for ring C ∗-algebras: the case of number fields with higher roots of unity, J. Topol. Anal., 4 (2012), 449 -- 479. [25] P.S. Muhly, J.N. Renault and D.P. Williams, Equivalence and isomorphism for groupoid C ∗-algebras, J. Operator Theory 17 (1987), no. 1, 3 -- 22. [26] S. Neshveyev, KMS states on the C ∗-algebras of non-principal groupoids, J. Operator Theory 70 (2013), no. 2, 513 -- 530. [27] J. Renault, A groupoid approach to C ∗-algebras, Lecture Notes in Mathematics, 793, Springer, Berlin, 1980. [28] J. Renault, Induced representations and hypergroupoids, SIGMA Symmetry Integrability Geom. Methods Appl. 10 (2014), Paper 057. [29] T. Takeishi, Primitive ideals and K-theoretic approach to Bost-Connes systems, Adv. Math. 302 (2016), 1069 -- 1079. [30] K. Thomsen, KMS weights on graph C ∗-algebras II. Factor types and ground states, preprint arXiv:1412.6762v2 [math.OA]. [31] A. Weil, Basic Number Theory, Springer-Verlag, New York Inc. 1967. [32] B. Yalkinoglu, On arithmetic models and functoriality of Bost-Connes systems. With an appendix by Sergey Neshveyev, Invent. Math. 191(2013), 383 -- 425. DOI 10.1007/s00222-012-0396-1. Department of Mathematics and Statistics, University of Victoria, P.O. Box 1700 STN CSC, Vic- toria, British Columbia, V8W 2Y2, Canada. E-mail address: [email protected] Department of Mathematics, University of Oslo, P.O. Box 1053 Blindern, N-0316 Oslo, Norway. E-mail address: [email protected] Department of Mathematics, University of Oslo, P.O. Box 1053 Blindern, N-0316 Oslo, Norway. E-mail address: [email protected] 21
1212.5756	1	1212	2012-12-23T03:09:43	Crossed products by Hecke pairs I: *-algebraic theory	[ "math.OA" ]	We develop a theory of crossed products by "actions" of Hecke pairs $(G, \Gamma)$, motivated by applications in non-abelian $C^$-duality. Our approach gives back the usual crossed product construction whenever $G / \Gamma$ is a group and retains many of the aspects of crossed products by groups. In this first of two articles we lay the $^$-algebraic foundations of these crossed products by Hecke pairs and we explore their representation theory.	math.OA	math	Crossed products by Hecke pairs I: ∗-algebraic theory Rui Palma Department of Mathematics, University of Oslo, P.O. Box 1053 Blindern, NO-0316 Oslo, Norway E-mail: [email protected] Abstract We develop a theory of crossed products by "actions" of Hecke pairs (G, Γ), motivated by applications in non-abelian C ∗-duality. Our ap- proach gives back the usual crossed product construction whenever G/Γ is a group and retains many of the aspects of crossed products by groups. In this first of two articles we lay the ∗-algebraic foundations of these crossed products by Hecke pairs and we explore their representation the- ory. Contents Introduction 1 Preliminaries 1.1 1.2 1.3 Hecke algebras 1.4 Fell bundles over discrete groupoids ∗-Algebras and (pre-)∗-representations . . . . . . . . . . . . . . . ∗-Algebraic multiplier algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Orbit space groupoids and Fell bundles 2.1 Group actions on Fell bundles . . . . . . . . . . . . . . . . . . . . 2.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 The algebra M (Cc(A)) . . . . . . . . . . . . . . . . . . . . . . . . 3 ∗-Algebraic crossed product by a Hecke pair 3.1 Definition of the crossed product and basic properties . . . . . . 3.2 Basic Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Representation theory . . . . . . . . . . . . . . . . . . . . . . . . 3.4 More on covariant pre-∗-representations . . . . . . . . . . . . . . 3.5 Crossed product in the case of free actions . . . . . . . . . . . . . Bibliography 2 4 5 7 14 17 19 19 28 29 37 37 54 57 69 71 75 Date: November 12, 2018 Research supported by the Research Council of Norway and the Nordforsk research net- work "Operator Algebra and Dynamics" 1 Introduction The present work is the first of two articles whose goal is the development of a theory of crossed products by Hecke pairs with a view towards application in non-abelian C∗-duality. A Hecke pair (G, Γ) consists of a group G and a subgroup Γ ⊆ G for which every double coset ΓgΓ is the union of finitely many left cosets. In this case Γ is also said to be a Hecke subgroup of G. Examples of Hecke subgroups include finite subgroups, finite-index subgroups and normal subgroups. It is in fact many times insightful to think of this definition as a generalization of the notion of normality of a subgroup. Given a Hecke pair (G, Γ) the Hecke algebra H(G, Γ) is a ∗-algebra of func- tions over the set of double cosets Γ\G/Γ, with a suitable convolution product and involution. It generalizes the definition of the group algebra C(G/Γ) of the quotient group when Γ is a normal subgroup. Heuristically, a crossed product of an algebra A by a Hecke pair (G, Γ) should be thought of as a crossed product (in the usual sense) of A by an "action" of G/Γ. The quest for a sound definition of crossed products by Hecke pairs may seem hopelessly flawed since G/Γ is not necessarily a group and thus it is unclear how it should "act" on A. It is the goal of this article and its coming sequel to show that in some circumstances such a definition can be given in a meaningful way, recovering the original one whenever G/Γ is a group. The term "crossed product by a Hecke pair" was first used by Tzanev [18] in order to give another perspective on the work of Connes and Marcolli [2]. This point of view was later formalized by Laca, Larsen and Neshveyev in [12], where they defined a C∗-algebra which can be interpreted as a reduced C∗-crossed product of a commutative C∗-algebra by a Hecke pair. It seems to be a very difficult task to define crossed products of any given algebra A by a Hecke pair, and for this reason we set as our goal to define a crossed product by a Hecke pair in a generality that will cover the following aspects: • existence of a canonical spanning set of elements in the crossed product; • possibility of defining covariant representations; • the Hecke algebra must be a trivial example of a crossed product by a Hecke pair; • the classical definition of a crossed product must be recovered whenever G/Γ is a group; • our construction should agree with that of Laca, Larsen and Neshveyev, whenever they are both definable; • our definition should be suitable for applications in non-abelian C∗-duality. In this first article on this subject we focus on defining such crossed products on a purely ∗-algebraic level and on developing their representation theory. The subjects of C∗-completions and relations with non-abelian C∗-duality will be further explored in the second article on the subject. We develop a theory of crossed products of certain algebras A by Hecke pairs which takes into account the above requirements. Our approach makes sense 2 when A is a certain algebra of sections of a Fell bundle over a discrete groupoid. To summarize our set up: we start with a Hecke pair (G, Γ), a Fell bundle A over a discrete groupoid X and an action α of G on A satisfying some "nice" properties. From this we naturally give the space A/Γ of Γ-orbits of A a Fell bundle structure over the orbit space X/Γ, which under our assumptions on the action α is in fact a groupoid. We can then define a ∗-algebra Cc(A/Γ) ×alg α G/Γ , which can be thought of as the crossed product of Cc(A/Γ) by the Hecke pair (G, Γ). We should point out that a crossed product for us is simply a ∗-algebra, which we can then complete with different C∗-norms or an L1-norm. Hence, and so that no confusion arises, the symbol ×alg will always be used when talking about the (uncompleted) ∗-algebraic crossed product. Our construction gives back the usual crossed product construction when Γ is a normal subgroup of G. Moreover, given any action of the group G/Γ on a Fell bundle B over a groupoid Y , the usual crossed product Cc(B) ×alg G/Γ can be obtained via our setup as a crossed product by the Hecke pair (G, Γ). Many of the features present in crossed products by discrete groups carry over to our setting. For instance, the role of the group G/Γ is played by the Hecke algebra H(G, Γ), which embeds in a natural way in the multiplier algebra of Cc(A/Γ) ×alg G/Γ. Additionally, just like a crossed product A × G by a discrete group is spanned by elements of the form a ∗ g, with a ∈ A and g ∈ G, our crossed products by Hecke pairs also admit a canonical spanning set of elements. The representation theory of crossed products by Hecke pairs also has many similarities with the group case, but some distinctive new features arise. For instance, as it is well-known in the group case, there is a bijective correspondence between nondegenerate representations of a crossed product A × G and the so- called covariant representations of A and G, which are certain pairs of unitary representations of G and representations of A. We will show that something completely analogous occurs for Hecke pairs, but in this case one is obliged to consider pre-representations of the Hecke algebra, i.e. representations of H(G, Γ) as (possibly) unbounded operators. This consideration was unnecessary in the group case because unitary operators are automatically bounded. As stated before, this theory of crossed products by Hecke pairs is intended for applications in non-abelian C∗-duality theory. One of the main motivations is the establishment of a Stone-von Neumann theorem for Hecke pairs that encompasses the work of an Huef, Kaliszewski and Raeburn [7] and expresses their results in the language of crossed products. Additionally, we envisage for future work a form of Katayama duality with respect to Echterhoff-Quigg's "crossed product" [4] (a terminology used in [7]). In a succinct and non-rigorous way this would mean that there is a canonical isomorphism of the type: A ×δ G/Γ ×bδ,ω G/Γ ∼= A ⊗ K(ℓ2(G/Γ)) , where A ×δ G/Γ is a crossed product by a coaction of the homogeneous space G/Γ, while the second crossed product should be by the "dual action" of the Hecke pair (G, Γ) in our sense. Such a result would bring insight into the emerging theory of crossed products by coactions of homogeneous spaces ([3], 3 [4]). We explain in Example 2.15 how our construction adapts well to the settings of [4]. This article is organized as follows. In Section 1 we set up the conventions and prelimanry results to be used in the rest of the article. Section 2 is dedicated to the development of the required set up for defining crossed products by Hecke pairs. Here we explain what type of actions are involved, how to define the orbit space groupoids X/H and the orbit bundles A/H out of A, and how all the algebras Cc(A/H) are related with each other for different subgroups H ⊆ G. Lastly, in Section 3 we introduce the notion of a crossed product by a Hecke pair, explore some of its algebraic aspects and develop its representation theory. In the last part of this section we show how many of the formulas become much simpler in the case of free actions. The present work is based on the author's Ph.D. thesis [13] written at the University of Oslo. There are a few differences between the present work and [13], notably the greater generality of the types of actions involved. This im- provement follows a suggestion of Dana Williams and John Quigg. The author would like to thank his advisor Nadia Larsen for the very helpful discussions, suggestions and comments during the elaboration of this work. A word of appreciation goes also to John Quigg, Dana Williams and Erik Bédos for some very helpful comments. 1 Preliminaries In this section we set up the conventions, notation, and background results which will be used throughout this work. We indicate the references where the reader can find more details, but we also provide proofs for those results which we could not find in the literature. Convention. The following convention for displayed equations will be used throughout this work: if a displayed formula starts with the equality sign, it should be read as a continuation of the previously displayed formula. A typical example takes the following form: (expression 1) = (expression 2) = (expression 3) . By Theorem A and Lemma B it then follows that = (expression 4) = (expression 5) . Under our convention starting with the equality sign in the second array of equations simply means that (expression 3) is equal to (expression 4). 4 1.1 ∗-Algebras and (pre-)∗-representations Let V be an inner product space over C. Recall that a function T : V → V is said to be adjointable if there exists a function T ∗ : V → V such that hT ξ , ηi = hξ , T ∗ηi , for all ξ, η ∈ V . Recall also that every adjointable operator T is necessarily linear and that T ∗ is unique and adjointable with T ∗∗ = T . We will use the following notation: • L(V ) denotes the ∗-algebra of all adjointable operators in V • B(V ) denotes the ∗-algebra of all bounded adjointable operators in V . Of course, we always have B(V ) ⊆ L(V ), with both ∗-algebras coinciding when V is a Hilbert space (see, for example, [16, Proposition 9.1.11]). Following [16, Def. 9.2.1], we define a pre-∗-representation of a ∗-algebra A on an inner product space V to be a ∗-homomorphism π : A → L(V ) and a ∗-representation of A on a Hilbert space H to be a ∗-homomorphism π : A → B(H ). As in [15, Def. 4.2.1], a pre-∗-representation π : A → L(V ) is said to be normed if π(A) ⊆ B(V ), i.e. if π(a) is a bounded operator for all a ∈ A. Definition 1.1 ([16], Def. 10.1.17). A ∗-algebra A is called a BG∗-algebra if all pre-∗-representations of A are normed. We now introduce our notion of an essential ideal. Our definition is not the usual one, but this choice of terminology will be justified in what follows. Definition 1.2. Let A be a ∗-algebra. An ideal I ⊆ A is said to be essential if aI 6= {0} for all a ∈ A \ {0}. The usual definition of an essential ideal states that I is essential if it has nonzero intersection with every other nonzero ideal. Our definition is stronger, but coincides with the usual one for the general class of semiprime ∗-algebras. We recall from [15, Definition 4.4.1] that a ∗-algebra is said to be semiprime if aAa = {0} implies a = 0, where a ∈ A. The class of semiprime ∗-algebras is quite large, containing all ∗-algebras that have a faithful ∗-representation on a Hilbert space (in particular, all C∗-algebras) and many other classes of ∗- algebras (see [16, Theorem 9.7.21]). Proposition 1.3. Let A be an algebra and I ⊆ A a nonzero ideal. We have i) If I is essential, then I has a nonzero intersection with every other nonzero ideal of A. ii) The converse of i) is true in case A is semiprime. 5 Proof: i) Let I be an essential ideal of A. Let J ⊆ A be a nonzero ideal and a ∈ J \ {0}. Since a is nonzero, then aI 6= {0}. Hence, J · I 6= {0}, and since J · I ⊆ J ∩ I, we have J ∩ I 6= {0}. ii) Suppose A is semiprime. Suppose also that I is not essential. Thus, there is a ∈ A \ {0} such that aI = {0}. Let Ja ⊆ A be the ideal generated by a. We have Ja · I = {0}. Since (Ja ∩ I)2 ⊆ Ja · I we have (Ja ∩ I)2 = {0}. Since A is semiprime this implies that Ja ∩ I = {0} (see [15, Theorem 4.4.3]). Hence, I has zero intersection with a nonzero ideal. For C∗-algebras the focus is mostly on closed ideals. In this setting we still see that our definition is equivalent to the usual one ([17, Definition 2.35]): Proposition 1.4. Let A be a C∗-algebra and I ⊆ A a closed ideal. The follow- ing are equivalent: i) I is essential. ii) I has nonzero intersection with every other nonzero ideal of A. iii) I has nonzero intersection with every other nonzero closed ideal of A. Proof: i) ⇐⇒ ii) This was established in Proposition 1.3, since C∗-algebras are automatically semiprime. ii) =⇒ iii) This is obvious. ii) ⇐= iii) Let J be a nonzero ideal of A and J its closure. From iii) we have I ∩ J 6= {0}. Since I and J are both closed, and A is a C∗-algebra, we have I · J = I ∩ J. Now, it is clear that I · J = {0} if and only if I · J = {0}. Hence, we necessarily have I · J 6= {0}, which implies I ∩ J 6= {0}. We now introduce the notion of an essential ∗-algebra. The class of essential ∗-algebras seems to be the appropriate class of ∗-algebras for which one can a define a multiplier algebra (as we shall see in Section 1.2). Definition 1.5. A ∗-algebra A is said to be essential if A is an essential ideal of itself, i.e. if aA 6= {0} for all a ∈ A \ {0}. Any unital ∗-algebra is obviously essential. Also, it is easy to see that a semiprime ∗-algebra is essential. The converse is false, so that essential ∗- algebras form a more general class than that of semiprime ∗-algebras: Example 1.6. Let C[X] be the polynomial algebra in one selfadjoint variable X. For any n ≥ 2 the algebra C[X]/hX ni is essential, because it is unital, but it is not semiprime because [X n−1](cid:16)C[X]/hX ni(cid:17)[X n−1] = {0}. 6 1.2 ∗-Algebraic multiplier algebras Every C∗-algebra can be embedded in a unital C∗-algebra in a "maximal" way. These maximal unitizations of C∗-algebras enjoy a number of useful properties and certain concrete realizations of these algebras are commonly referred to as multiplier algebras. The reader is referred to [17] for an account. The definition of a multiplier algebra is quite standard in C∗-algebra theory, but this notion is in fact more general and applicable for more general types of rings and algebras. For example, in [1, Section 1.1] it is explained how multiplier algebras can be defined for semiprime algebras. In this section we are going to generalize this notion to the context of es- sential ∗-algebras and derive their basic properties. We believe that essential ∗-algebras are the appropriate class of ∗-algebras for which one can define mul- tiplier algebras, since the property aA = {0} ⇒ a = 0, which characterizes an essential ∗-algebra, is constantly used in proofs. Multiplier algebras are many times defined via the so-called double central- izers (see for example [1]), but since we are only interested in algebras with an involution a slightly simpler and more convenient approach can be given, ana- logue to the Hilbert C∗-module approach to C∗-multiplier algebras (presented in [17, Section 2.3]). This is the approach we follow. Definition 1.7. Let C be a subclass of ∗-algebras. A ∗-algebra A ∈ C is said to have a maximal unitization in C if there exists a unital ∗-algebra B ∈ C (called the maximal unitization of A) and a ∗-embedding i : A ֒→ B for which i(A) is an essential ideal of B and such that for every other ∗-embedding j of A as an essential ideal of a unital ∗-algebra C ∈ C, there is a unique ∗-homomorphism φ : C → B such that B φ ?⑦ ⑦ ⑦ ⑦ i ⑦ ⑦ ⑦ ⑦ / C j commutes. A Lemma 1.8. In the above diagram the ∗-homomorphism φ is always injective (even if C was not unital). Proof: We have that j(A)∩Ker φ = {0}, because if j(a) ∈ j(A)∩Ker φ, then 0 = φ(j(a)) = i(a) and hence a = 0 and therefore j(a) = 0. Hence, since j(A) is an essential ideal of C, it follows from Proposition 1.3 i) that Ker φ = {0}. For C∗-algebras, one might expect to replace "ideal" by "closed ideal", in Definition 1.7. This condition, however, follows automatically since i(A) and j(A) are automatically closed. Hence, this definition encompasses the usual definition of a maximal unitization for a C∗-algebra. Definition 1.9. Let A be a ∗-algebra. By a right A-module we mean a vector space X together with a mapping X × A → X satisfying the usual consistency 7 ? / O O conditions. An A-linear mapping T : X → Y between A-modules is a linear mapping between the underlying vector spaces such that T (xa) = T (x)a, for all x ∈ X and a ∈ A. We will often use the notation T x, instead of T (x). Every ∗-algebra A is canonically a right A-module, with the action of right multiplication. This is the example we will use thoroughly in what follows. Let h·, ·iA : A × A → A be the function ha, biA := a∗b . The function h·, ·iA is an A-linear form, in the sense that the following properties are satisfied: a) ha , λ1b1 + λ2b2iA = λ1ha, b1iA + λ2ha, b2iA , b) hλ1a1 + λ2a2 , biA = λ1ha1, biA + λ2ha2, biA , c) ha, bciA = ha, biA c , d) hac, biA = c∗ha, biA , e) ha, bi∗ A = hb, aiA , for all a, a1, a2, b, b1, b2 ∈ A and λ1, λ2 ∈ C. If the ∗-algebra A is essential we also have: f) If ha , biA = 0 for all b ∈ A, then a = 0 . Definition 1.10. Let A be a ∗-algebra. A function T : A → A is called adjointable if there is a function T ∗ : A → A such that hT (a), biA = ha, T ∗(b)iA , for all a, b ∈ A. Proposition 1.11. If A is an essential ∗-algebra, then every adjointable map T : A → A is A-linear. Moreover, the adjoint T ∗ is unique and adjointable with T ∗∗ = T . Proof: Let T be an adjointable map in A and x1, x2, y ∈ A. We have hT (λ1x1 + λ2x2) , yiA = hλ1x1 + λ2x2 , T ∗(y)iA = λ1 hx1 , T ∗(y)iA + λ2 hx2 , T ∗(y)iA = λ1 hT (x1) , yiA + λ2 hT (x2) , yiA = hλ1T (x1) + λ2T (x2) , yiA . Hence, we have hT (λ1x1 + λ2x2) − λ1T (x1) + λ2T (x2) , yiA = 0. We can then conclude from f) that T (λ1x1 + λ2x2) − λ1T (x1) + λ2T (x2) = 0 , 8 i.e. T is a linear map. Let us now check that T is A-linear. For any x, y, a ∈ A we have hT (xa) , yiA = hxa , T ∗(y)iA = a∗hx , T ∗(y)iA = a∗hT (x) , yiA = hT (x)a , yiA . Hence, we have hT (xa) − T (x)a , yiA = 0. We can then conclude from f) that T (xa) − T (x)a = 0, i.e. T is A-linear. Let us now prove the uniqueness of the adjoint T ∗. Suppose there was a function S : A → A such that hx , T ∗(y)iA = hx , S(y)iA . for all x, y ∈ A. Then, hT ∗(y) − S(y) , xiA = 0. We can then conclude from f) that T ∗(y) − S(y) = 0, i.e. T ∗ = S. It remains to prove that T ∗ is adjointable with T ∗∗ = T . This follows easily from the equality hT ∗x , yiA = hy , T ∗xi∗ A = hT y , xi∗ A = hx, T yiA . Definition 1.12. Let A be an essential ∗-algebra. The set of all adjointable maps on A is called the multiplier algebra of A and is denoted by M (A). The multiplier algebra is in fact a ∗-algebra, and the proof of this fact is standard. Proposition 1.13. Let A be an essential ∗-algebra. The multiplier algebra of A is a unital ∗-algebra with the sum and multiplication given by pointwise sum and composition (respectively), and the involution given by the adjoint. Proposition 1.14. Let A be an essential ∗-algebra. There is a ∗-embedding L : A → M (A) of A as an essential ideal of M (A), given by where La : A → A is the left multiplication by a, i.e. La(b) := ab. a 7→ La Proof: It is easy to see that, for every a ∈ A, La is adjointable with adjoint La∗, thus the mapping L is well-defined. Also clear is the fact that L is a ∗- homomorphism. Let us prove that it is injective: suppose La = 0 for some a ∈ A. Then, for all b ∈ A we have ab = Lab = 0 and since A is essential this implies a = 0. Thus, L is injective. It remains to prove that L(A) is an essential ideal of M (A). Let us begin by proving that it is an ideal. Let T ∈ M (A). For every a, b ∈ A we have T La(b) = T (ab) = T (a)b = LT a(b) , 9 and also Hence we have LaT (b) = aT (b) = ha∗, T (b)i = hT ∗(a∗), bi = (T ∗(a∗))∗b = L(T ∗a∗)∗ (b) . T La = LT a and LaT = L(T ∗a∗)∗ , (1) from which it follows easily that L(A) is an ideal of M (A). Let us now prove that this ideal is essential. Let T ∈ M (A) be such that T L(A) = {0}. Then, in particular, T La = 0 for all a ∈ A, but as we have seen before T La = LT a, and since L is injective we must have T a = 0 for all a ∈ A, i.e T = 0. Remark 1.15. According to Proposition 1.14, an essential ∗-algebra A is canon- ically embedded in its multiplier algebra M (A). We will often make no distinc- tion of notation between A and its embedded image in M (A), i.e. we will often just write a to denote an element of A and to denote the element L(a) of M (A). No confusion will arise from this because the left equality in (1) simply means, in this notation, that T · a = T (a). Theorem 1.16. Let A be an essential ∗-algebra and L : A → M (A) the canon- ical ∗-embedding of A in M (A). If j : A → C is a ∗-embedding of A as an ideal of a ∗-algebra C, then there exists a unique ∗-homomorphism φ : C → M (A) such that the following diagram commutes M (A) <② φ / C ② L ② ② ② j ② ② ② A Moreover, if j(A) is essential then φ is injective. Proof: For simplicity of notation let us assume, without any loss of gener- ality, that A itself is an ideal of a ∗-algebra C, so that we avoid any reference to j (or its inverse). Let φ : C → M (A) be the function defined by φ(c) : A → A φ(c)a := ca , for every c ∈ C. It is a straightforward computation to check that φ(c) ∈ M (A) and that φ itself is a ∗-homomorphism. It is also easy to see that φ(a) = La, for every a ∈ A. Hence, φ ◦ j = L. Let us now prove the uniqueness of φ relatively to this property. Suppose eφ : C → M (A) is another ∗-homomorphism such that 10 < / O O eφ ◦ j = L. Then, for all c ∈ C and a ∈ A we have (cid:0)eφ(c) − φ(c)(cid:1)La = eφ(c)La − φ(c)La = eφ(c)eφ(a) − φ(c)φ(a) = eφ(ca) − φ(ca) = Lca − Lca = 0 . Since L(A) is an essential ideal of M (A) it follows thateφ(c) = φ(c) for all c ∈ C, i.e. eφ = φ. The last claim of the theorem, concerning injectivity of φ, was proven in Lemma 1.8. Corollary 1.17. The multiplier algebra M (A) is a maximal unitization of A in the class of: essential ∗-algebras, semiprime ∗-algebras and C∗-algebras. Proof: By Theorem 1.16 we only need to check that if A is an essential ∗-algebra (respectively, semiprime ∗-algebra or C∗-algebra), then the multiplier algebra has the same property. Suppose A is an essential ∗-algebra. Let T ∈ M (A) be such that T M (A) = {0}. Then, by the embedding of A in M (A) we have T a = 0 for all a ∈ A, i.e. T = 0. Hence, M (A) is also an essential ∗-algebra. Suppose A is a semiprime ∗-algebra. Let T ∈ M (A) be such that T M (A)T = {0}. Then, we also have that T LaM (A)T La = {0} for any a ∈ A, and therefore LT (a)M (A)LT (a) = {0}. Thus, in particular, LT (a)L(A)LT (a) = {0}, and since L is injective this means that T (a)AT (a) = {0}. Since A is semiprime we conclude that T (a) = 0, and therefore T = 0. Hence, M (A) is semiprime. It is well-known that M (A) is a C∗-algebra when A is a C∗-algebra. An important feature of C∗-multiplier algebras is that a nondegenerate ∗- representation of A extends uniquely to M (A). This result does not hold in general for essential ∗-algebras. Nevertheless we can still extend a nondegener- ate ∗-representation of A to a unique pre-∗-representation of M (A): Theorem 1.18. Let A be an essential ∗-algebra, π : A → B(H ) a nonde- generate ∗-representation of A on a Hilbert space H and V ⊆ H the dense subspace V := π(A)H = span {π(a)ξ : a ∈ A , ξ ∈ H } . Then there is a unique pre-∗-representation such that eπ(a) = π(a)V for every a ∈ A. eπ : M (A) → L(V ) 11 Proof: We define the pre-∗-representationeπ : M (A) → L(V ) by eπ(T )(cid:2) nXi=1 π(ai)ξi(cid:3) := for n ∈ N, a1, . . . , an ∈ A and ξ1, . . . , ξn ∈ H . Let us first check that eπ is i=1 π(ai)ξi =Pm well-defined. SupposePn π(T bi)ηj(cid:17) = π(z)(cid:16) nXi=1 mXj=1 j=1 π(bj)ηj. Then, for every z ∈ A we π(zT ai)ξi − nXi=1 mXj=1 π(T ai)ξi − π(zT bi)ηj π(T ai)ξi , have nXi=1 = π(zT )(cid:16) nXi=1 = 0 . π(ai)ξi − mXj=1 π(bi)ηj(cid:17) Since the ∗-representation π is nondegenerate we necessarily have π(ai)ξi , π(T ai)ξi − mXj=1 nXi=1 nXi=1 π(T bi)ηj = 0 , hπ(T ai)ξi , π(bj)ηj i the following equality mXj=1 which means thateπ(T ) is well-defined. Let us now check thateπ(T ) ∈ L(V ), i.e. thateπ(T ) is indeed an adjointable operator in V . We will in fact prove that eπ(T )∗ = eπ(T ∗), which follows from π(bi)ηjE = Deπ(T ) mXj=1 nXi=1 mXj=1 nXi=1 mXj=1 nXi=1 mXj=1 nXi=1 π(bi)ηjE . = D nXi=1 mXj=1 π(ai)ξi ,eπ(T ∗) It is straightforward to see thateπ is linear, multiplicative and, as we have seen, eπ(T ∗) =eπ(T )∗, henceeπ is a pre-∗-representation of M (A) on V . It is also clear that, for any a ∈ A,eπ(a) is just π(a) restricted to V , because hπ(ai)ξi , π(T ∗bj)ηj i i T ∗)π(bj )ηji of the equality i T ∗bj)ηji hξi , π(a∗ hξi , π(a∗ = = = π(ai)ξi = π(aai)ξi = π(a) π(ai)ξi . nXi=1 eπ(a) nXi=1 nXi=1 representation such that φ(a) = π(a)V . Then, for every z ∈ A and v ∈ V we Let us now prove the uniqueness of eπ. Suppose φ : M (A) → L(V ) is a pre-∗- 12 have π(z)(φ(T )v −eπ(T )v) = π(z)φ(T )v − π(z)eπ(T )v = φ(z)φ(T )v −eπ(z)eπ(T )v = φ(zT )v −eπ(zT )v = π(zT )v − π(zT )v = 0 . Since the ∗-representation π is nondegenerate, we necessarily have which means that φ(T ) =eπ(T ), i.e. φ =eπ. φ(T )v −eπ(T )v = 0 , Remark 1.19. Theorem 1.18 can be interpreted in the following way: every nondegenerate ∗-representation π : A → B(H ) can be extended to M (A) by possibly unbounded operators, defined on the dense subspace π(A)H . Definition 1.20. Let A be an essential ∗-algebra. We will denote by MB(A) the for all nondegenerate ∗-representations π : A → B(H ), where V := π(A)H subset of M (A) consisting of all the elements T ∈ M (A) such thateπ(T ) ∈ B(V ) andeπ is the unique pre-∗-representation extending π as in Proposition 1.18. As stated in the next result, MB(A) is a ∗-subalgebra of M (A). The advan- tage of working with MB(A) over M (A) is that nondegenerate ∗-representations of A always extend to ∗-representations of MB(A). Easy examples of elements of MB(A) that might not belong to A are the projections and unitaries of M (A). Proposition 1.21. Let A be an essential ∗-algebra. The set MB(A) is a ∗- subalgebra of M (A) containing A. Moreover, if π : A → B(H ) is a nonde- generate ∗-representation of A, then there exists a unique ∗-representation of MB(A) on H that extends π. Proof: Let T, S ∈ MB(A). Let π : A → B(H ) be any nondegenerate Let us now prove the last claim of this proposition. Let π : A → B(H ) ∗-representation of A and eπ its extension to L(V ), in the sense of Theorem 1.18, where V := π(A)H . By definition, eπ(T ),eπ(S) ∈ B(V ), and therefore eπ(T + S),eπ(T S),eπ(T ∗) ∈ B(V ), since B(V ) is a ∗-algebra. Hence, MB(A) is a ∗-subalgebra of M (A). Moreover, A ⊆ MB(A) sinceeπ(a) = π(a)V ∈ B(V ). be a nondegenerate ∗-representation and eπ : M (A) → L(V ) its extension as in Theorem 1.18. Then we obtain by restriction a pre-∗-representation eπ : MB(A) → L(V ). By definition of MB(A) we actually haveeπ(MB(A)) ⊆ B(V ). Hence eπ gives rise to a ∗-representationeπ : MB(A) → B(H ), since V is dense in H . Let us now prove the uniqueness claim. Suppose ϕ is another representation of MB(A) that extends π. For T ∈ MB(A), a ∈ A and ξ ∈ H we have ϕ(T ) π(a)ξ = ϕ(T ) ϕ(a)ξ = ϕ(T a)ξ = π(T a)ξ = eπ(T ) π(a)ξ . 13 By linearity and density it follows that ϕ(T ) =eπ(T ), i.e. ϕ =eπ. The above result is a generalization of the well-known result for C∗-algebras which states that any nondegenerate ∗-representation can be extended to the multiplier algebra (see for example [17, Corollary 2.51]), because M (A) = MB(A) for any C∗-algebra A. 1.3 Hecke algebras We start by establishing some notation and conventions concerning left coset spaces and double coset spaces and we prove two resuls which will be useful later on. Let G be a group, B, C subgroups of G and e ∈ G the identity element. The double coset space B\G/C is the set B\G/C := {BgC ⊆ G : g ∈ G} . (2) It is easy to see that the sets of the form BgC are either equal or disjoint, or in other words, we have an equivalence relation defined in G whose equivalence classes are precisely the sets BgC. The left coset space G/C is the set G/C := {e}\G/C = {gC ⊆ G : g ∈ G} . (3) Given an element g ∈ G and a double coset space B\G/C (which can in particular be a left coset space by taking B = {e}) we will denote by [g] the double coset BgC. Thus, [g] denotes the whole equivalence class for which g ∈ G is a representative. If A is a subset of G we define the double coset space B\A/C as the set of double cosets in B\G/C which have a representative in A, i.e. B\A/C := {BaC ⊆ G : a ∈ A} . (4) Proposition 1.22. Let A, B and C be subgroups of a group G. If C ⊆ A, then the following map is a bijective correspondence between the double coset spaces: B\A/C −→ (B ∩ A)\A/C [a] 7→ [a] . Similarly, if B ⊆ A, then the following map is a bijective correspondence: B\A/C −→ B\A/(A ∩ C) [a] 7→ [a] . (5) (6) 14 if Proof: We first need to show that the map (5) is well defined, i.e. Ba1C = Ba2C, for some a1, a2 ∈ A, then (B ∩ A)a1C = (B ∩ A)a2C. If Ba1C = Ba2C then there exist b ∈ B and c ∈ C such that a1 = ba2c, from which it follows that b = a1c−1a−1 2 . Since A is a subgroup and C ⊆ A, it follows readily that b ∈ B ∩ A, and therefore a1 ∈ (B ∩ A)a2C, i.e. (B ∩ A)a1C = (B ∩ A)a2C. The map defined in (5) is clearly surjective. It is also injective because if (B ∩ A)a1C = (B ∩ A)a2C, then clearly Ba1C = Ba2C. A completely analogous argument shows that map defined in (6) is a bijec- tion. Suppose a group G acts (on the right) on a set X and let x ∈ X. We will henceforward denote by Sx the stabilizer of the point x, i.e. Sx := {g ∈ G : xg = x} . (7) Given a subset Z ⊆ X and a subgroup H ⊆ G we denote by Z/H the set of H-orbits which have representatives in Z, i.e. Z/H := {zH : z ∈ Z} . Suppose now that H, K ⊆ G are subgroups and let x ∈ X be a point. The fol- lowing result establishes a correspondence between the set of H-orbits (xK)/H and the double coset space Sx\K/H: Proposition 1.23. Let G be a group which acts (on the right) on a set X. Let x ∈ X be a point and H, K ⊆ G be subgroups. We have a bijection (xK)/H −→ Sx\K/H , given by xgH 7→ SxgH, where g ∈ K. Proof: Let us first prove that the map xgH 7→ SxgH is well defined, i.e. if xg1H = xg2H, then Sxg1H = Sxg2H. If xg1H = xg2H, then there exists h ∈ h such that xg1 = xg2h, which implies that x = xg2hg−1 1 , from which it follows that g2hg−1 1 ∈ Sx. Thus we see that Sxg1H = Sxg2hg−1 1 g1H = Sxg2H . We conclude that the map is well-defined. The map is obviously surjective. It is also injective because if Sxg1H = Sxg2H, then there exists r ∈ Sx and h ∈ H such that g1 = rg2h, from which it follows that xg1H = xrg2hH = xg2H. We will mostly follow [10] and [8] in what regards Hecke pairs and Hecke algebras and refer to these references for more details. We start by establishing some notation which will be useful later on. Given a group G, a subgroup Γ ⊆ G and g ∈ G, we will denote by Γg the subgroup Γg := Γ ∩ gΓg−1 . (8) We now recall the definition of a Hecke pair: 15 Definition 1.24. Let G be a group and Γ a subgroup. The pair (G, Γ) is called a Hecke pair if every double coset ΓgΓ is the union of finitely many right (and left) cosets. In this case, Γ is also called a Hecke subgroup of G. Given a Hecke pair (G, Γ) we will denote by L and R, respectively, the left and right coset counting functions, i.e. L(g) := ΓgΓ/Γ = [Γ : Γg] < ∞ R(g) := Γ\ΓgΓ = [Γ : Γg−1 ] < ∞ . (9) (10) We recall that L and R are Γ-biinvariant functions which satisfy L(g) = R(g−1) for all g ∈ G. Moreover, the function ∆ : G → Q+ given by ∆(g) := L(g) R(g) , (11) is a group homomorphism, usually called the modular function of (G, Γ). Definition 1.25. Given a Hecke pair (G, Γ), the Hecke algebra H(G, Γ) is the ∗-algebra of finitely supported C-valued functions on the double coset space Γ\G/Γ with the product and involution defined by (f1 ∗ f2)(ΓgΓ) := XhΓ∈G/Γ f ∗(ΓgΓ) := ∆(g−1)f (Γg−1Γ) . f1(ΓhΓ)f2(Γh−1gΓ) , (12) (13) (14) (15) Equivalently, we can define H(G, Γ) as the ∗-algebra of finitely supported Γ- left invariant functions f : G/Γ → C with the product and involution operations given by (f1 ∗ f2)(gΓ) := XhΓ∈G/Γ f ∗(gΓ) := ∆(g−1)f (g−1Γ) . f1(hΓ)f2(h−1gΓ) , Remark 1.26. Some authors, including Krieg [10], do not include the factor ∆ in the involution. Here we adopt the convention of Kaliszewski, Landstad and Quigg [8] in doing so, as it gives rise to a more natural L1-norm. We note, nev- ertheless, that there is no loss (or gain) in doing so, because these two different involutions give rise to ∗-isomorphic Hecke algebras. The Hecke algebra has a natural basis, as a vector space, given by the charac- teristic functions of double cosets. We will henceforward identify a characteristic function of a double coset 1ΓgΓ with the double coset ΓgΓ itself. 16 1.4 Fell bundles over discrete groupoids Let X be a discrete groupoid. We will denote by X 0 the unit space of X and by s and r the source and range functions X → X 0, respectively. We will essentially follow [11] when it comes to Fell bundles over groupoids. All the groupoids in this work are assumed to be discrete, so that the theory of Fell bundles admits a few simplifications. Basically a Fell bundle over a discrete groupoid X consists of: • a space A together with a surjective map p : A → X, such that each fiber Ax := p−1(x) is a Banach space, for every x ∈ X,; • a multiplication operation between fibers over composable elements of the groupoid, which we suggestively write as Ax · Ay ⊆ Axy; • an involution a 7→ a∗ which takes Ax into Ax−1 . These operations and norms satisfy some consistency properties which are de- scribed in [11, Section 2]. As it is well-known, each fiber over a unit element is naturally a C∗-algebra. Standing Assumption 1.27. Given a Fell bundle A over a discrete groupoid X we will always assume that the fibers over units are non-trivial, i.e. Au 6= {0} for all u ∈ X 0. Assumption 1.27 is not very restrictive. In fact, removing from the groupoid X all the units u ∈ X 0 for which Au = {0} and also all the elements x ∈ X such that s(x) or r(x) is u, we obtain a subgroupoid Y for which the assumption holds (relatively to the restriction AY of A to Y ). Moreover, and this is the important fact, the algebras of finitely supported sections are canonically isomorphic, i.e. Cc(AY ) ∼= Cc(A). The reason for us to follow Assumption 1.27 is because it will make our theory slightly simpler. Since we are interested mostly in algebras of sections, this assumption does not reduce the generality of the work in any way, as we observed in the previous paragraph. Definition 1.28. Let A be a Fell bundle over a discrete groupoid X. An automorphism of A is a bijective map β : A → A which preserves the bundle structure, i.e. such that i) β takes any fiber onto another fiber; ii) β takes fibers over composable elements of X to fibers over composable elements; iii) As a map between (two) fibers, β is a linear map; iv) β(a · b) = β(a) · β(b), whenever multiplication is defined; v) β(a∗) = β(a)∗. 17 The set of all automorphisms of A forms a group under composition and will be denoted by Aut(A). It follows easily from i) and ii) above that every automorphism β of A en- tails a groupoid automorphism β0 of X such that β0(cid:0)p(a)(cid:1) = p(β(a)). We also note that, by being a groupoid automorphism, β0 takes units into units. Remark 1.29. The restricted map β : Ax → Aβ0(x) is an isometric linear map. Linearity was required in condition iii), but the fact that the map is an isometry follows from the other axioms. To see this we note that kβ(a)k = kβ(a)∗β(a)k 1 2 = kβ(a∗a)k 1 2 . Now a∗a ∈ As(x) and s(x) ∈ X 0. Thus, we also have β0(s(x)) ∈ X 0 and therefore both As(x) and Aβ0(s(x)) are C∗-algebras. It follows from iii), iv) and v) that the restricted map β : As(x) → Aβ0(s(x)) is a C∗-isomorphism and is therefore isometric. Hence we have kβ(a)k = kβ(a∗a)k 1 2 = ka∗ak 1 2 = kak , which shows that β : Ax → Aβ0(x) is an isometry. Given a Fell bundle A over a discrete groupoid X we will denote by Cc(A) its corresponding ∗-algebra of finitely supported sections. The following notation will be used throughout the rest of this work: for x ∈ X and a ∈ Ax the symbol ax will always denote the element of Cc(A) such that ax(y) :=(a , 0, if y = x otherwise . (16) According to the notation above we can then write any f ∈ Cc(A) uniquely as a sum of the form f = Xx∈X (f (x))x . (17) We recall that the operations of multiplication and involution in Cc(A) are determined by ax · by =((ab)xy , 0, (ax)∗ = (a∗)x−1 , if s(x) = r(y) otherwise , where x, y ∈ X and a ∈ Ax, b ∈ Ay. 18 2 Orbit space groupoids and Fell bundles In this section we present the basic set up which will enable us to define crossed products by Hecke pairs later in Section 3. Our construction of a (∗-algebraic) crossed product A ×alg G/Γ of an algebra A by a Hecke pair (G, Γ) will make sense when A is a certain algebra of sections of a Fell bundle over a discrete groupoid. In this section we show in detail what type of algebras A are involved in the crossed product and how they are obtained. 2.1 Group actions on Fell bundles Throughout this section G will denote a discrete group. One of our ingredients for defining crossed products by Hecke pairs consists of a group action on a Fell bundle over a groupoid (a concept we borrow from [9, Section 6]). Such actions always carry an associated action on the corresponding groupoid (by groupoid automorphisms). Since we are primarily interested in right actions on groupoids, we start by recalling what they are: Definition 2.1. Let X be a groupoid. A right action of G on X is a mapping X × G → X (x, g) 7→ xg , which is a right action of G on the underlying set of X, meaning that 1) xe = x, for all x ∈ X, 2) x(g1g2) = (xg1)g2, for all x ∈ X, g1, g2 ∈ G, which is compatible with the groupoid operations, meaning that 3) if x and y are composable in X, then so are xg and yg, for all g ∈ G, and moreover (xg)(yg) = (xy)g , 4) (xg)−1 = x−1g, for all x ∈ X and g ∈ G. In other words, a right action of G on X is a right action on the set X performed by groupoid automorphisms. Lemma 2.2. Let X be a groupoid endowed with a right G-action. For every x ∈ X and g ∈ G we have s(xg) = s(x)g and r(xg) = r(x)g . In particular, G restricts to an action on the unit space X 0. 19 Proof: It follows easily from the definition of a right G-action that s(x)g = (x−1x)g = (x−1g)(xg) = (xg)−1(xg) = s(xg) , and similarly for the range function. Remark 2.3. Given elements x, y in a groupoid X endowed with a right G- action and given g ∈ G, we will often drop the brackets in expressions like (xg)y and simply use the notation xgy. No confusion arises from this since G is only assumed to act on the right. On the other hand, we will never write an expres- sion like xyg without brackets, since it can be confusing on whether it means x(yg) or (xy)g. Definition 2.4. [9, Section 6] Let G be a group and A a Fell bundle over a discrete groupoid X. An action of G on A consists of a homomorphism α : G → Aut(A). As observed in Section 1.4, each automorphism of A carries with it an as- sociated automorphism of the underlying groupoid X. Hence, an action of a group G on A entails an action of G on X by groupoid automorphisms. Since we are interested only in right actions on groupoids, we just ensure that these associated actions are on the right simply by taking inverses. Moreover, even though we will typically denote by α the action of G on A, we will simply write (x, g) 7→ xg to denote its associated action on X and it will be always assumed that this action comes from α. To summarize what we have said so far: given an action α of G on a Fell bundle A over a groupoid X, there is an associated right G-action (x, g) 7→ xg on X such that p(αg(a)) = p(a)g−1 . (18) Remark 2.5. Typically one would require the mapping (a, g) 7→ αg(a) to be continuous, but this is not necessary here since both G and X are discrete. Proposition 2.6. Let α be an action of a group G on a Fell bundle A over a groupoid X. We have an associated action α : G → Aut(Cc(A)) of G on Cc(A) given by αg(f ) (x) := αg(f (xg)) , for g ∈ G, f ∈ Cc(A) and x ∈ X. Proof: Let us first prove that the action is well-defined, i.e. αg(f ) ∈ Cc(A). The fact that αg(f ) is finitely supported is obvious, so the only thing one needs to check is that αg(f ) is indeed a section of the bundle, i.e. αg(f (xg)) ∈ Ax for all x ∈ X, which is clear because αg(Ay) = Ayg−1 . 20 Let us now check that αg is indeed a ∗-homomorphism for all g ∈ G. Lin- earity of αg is obvious. Let f, f1, f2 ∈ Cc(A). We have αg(f1 · f2) (x) = αg((f1 · f2) (xg)) αg(f1(y))αg(f2(z)) αg(f1(yg))αg(f2(zg)) yz=xg (yg−1)(zg−1)=x αg(f1(y)f2(z)) = Xy,z∈X = Xy,z∈X = Xy,z∈X = Xy,z∈X = (cid:0)αg(f1) · αg(f2)(cid:1) (x) . yz=x yz=x αg(f1)(y)αg(f2)(z) Hence, αg(f1 · f2) = αg(f1) · αg(f2). Also, αg(f ∗) (x) = αg(f ∗(xg)) = αg(f (x−1g))∗ = (cid:0)αg(f ) (x−1)(cid:1)∗ =(cid:0)αg(f )(cid:1)∗ (x) . Hence, αg(f ∗) = (αg(f ))∗. The fact that αg1g2 = αg1 ◦ αg2 for every g1, g2 ∈ G is also easily checked. Definition 2.7. Let α be a group action of G on a Fell bundle A over a groupoid X and let H be a subgroup of G. We will say that the G-action is H-good if for any x ∈ X and h ∈ H we have s(x)h = s(x) =⇒ αh−1(a) = a ∀a ∈ Ax . (19) Also, a right G-action on a groupoid X is said to be H-good if for any x ∈ X and h ∈ H we have s(x)h = s(x) =⇒ xh = x . (20) It is clear from the definitions that if the action α of G on A is H-good, then its associated right G-action on the underlying groupoid X is also H-good. We will mostly use actions on Fell bundles. However, some of our results (namely Proposition 2.10) are about groupoids only, and this is the reason for defining H-good actions for groupoids as well. We now give equivalent definitions of a H-good action. For that we recall from (7) that given an action of G on a set X we denote by Sx the stabilizer of the point x ∈ X. We will also denote by S(Ax) the set S(Ax) := {g ∈ G : αg−1 (a) = a , ∀a ∈ Ax}. Proposition 2.8. Let α be an action of G on a Fell bundle A over a groupoid X. The following statements are equivalent: 21 i) The action α is H-good. ii) For every x ∈ X we have that Ss(x) ∩ H = S(Ax) ∩ H. iii) For any x ∈ X we have Ss(x) ∩ H = Sx ∩ H = Sr(x) ∩ H = (21) = S(As(x)) ∩ H = S(Ax) ∩ H = S(Ar(x)) ∩ H . iv) The stabilizers of the H-actions on X and on the fibers of A are the same on composable pairs, i.e. if x ∈ X and y ∈ Y are composable, then Sx ∩ H = Sy ∩ H = = S(Ax) ∩ H = S(Ay) ∩ H . Proof: i) =⇒ ii) Since the action is H-good we have, by definition, that Ss(x) ∩ H ⊆ S(Ax) ∩ H. Also, if h ∈ S(Ax) ∩ H, then we necessarily have xh = x, and therefore by Lemma 2.2 we get s(x) = s(xh) = s(x)h, from which we conclude that h ∈ Ss(x) ∩ H. Hence we have Ss(x) ∩ H = S(Ax) ∩ H. ii) =⇒ iii) Repeating a little bit of what we did above: if h ∈ S(Ax) ∩ H, then we necessarily have that xh = h, and therefore h ∈ Sx ∩ H. Moreover, if h ∈ Sx ∩ H, then it follows that by Lemma 2.2 that h ∈ Ss(x) ∩ H. Thus, we have that S(Ax) ∩ H = Sx ∩ H = Ss(x) ∩ H . Since s(s(x)) = s(x), we also have, directly by our assumption of ii), that Ss(x) ∩ H = S(As(x)) ∩ H. Since we have (xg)−1 = x−1g, it follows easily that Sx = Sx−1. Similarly, since αg(a)∗ = αg(a∗), it follows easily that S(Ax) = S(Ax−1 ). Observing that s(x−1) = r(x), equality (21) follows directly from what we proved above. iii) =⇒ iv) Suppose x ∈ X and y ∈ X are composable. Then, s(x) = r(y) and equality (21) immediately yelds that Sx ∩ H = Sy ∩ H = = S(Ax) ∩ H = S(Ay) ∩ H . iv) =⇒ i) Let h ∈ H and x ∈ X be such that s(x)h = s(x). From iv) it follows that h ∈ S(Ax) ∩ H. This means that the action is H-good. It is easy to see that any H-good action is also gHg−1-good for any conjugate gHg−1, and also K-good for any subgroup K ⊆ H. The following property will also be important for defining crossed products by Hecke pairs: Definition 2.9. Let X be a groupoid endowed with a right G-action and let H be a subgroup of G. We will say that the action has the H-intersection property if uH ∩ ugHg−1 = uH g , (22) 22 for every unit u ∈ X 0 and g ∈ G. An action of G on a Fell bundle A is said to have the H-intersection property if its associated right G-action on the underlying groupoid has the H-intersection property. We defer examples of H-good actions and actions with the H-intersection property for the next section. We now introduce one of the important ingredients for our definition of crossed products by Hecke pairs: the orbit space groupoid. Let G be a group, H ⊆ G a subgroup and X a groupoid endowed with a H-good right G-action. Then, the orbit space X/H becomes a groupoid in a canonical way which we will now describe. For that, and throughout this text, we will use the following notation: given elements x, y we define the set Hx,y := {h ∈ H : s(x)h = r(y)} . (23) The groupoid structure on X/H is described as follows: • A pair (xH, yH) ∈ (X/H)2 is composable if and only if Hx,y 6= ∅, or equivalently, r(y) ∈ s(x)H. This property is easily seen not to depend on the choice of representatives x, y from the orbits xH, yH respectively. • Given a composable pair (xH, yH) ∈ (X/H)2, their product is xH yH := xehyH , (24) where eh is any element of Hx,y. It will follow from the fact the action is H-good that xeh does not depend on the representativeeh chosen from Hx,y. The result of the product xH yH also does not depend on the choice of representatives x, y. We will prove this in the next result. • The inverse of the element xH is simply the element x−1H. It is also easy to see that this does not depend on the choice of representative x. Proposition 2.10. Let G be a group, H ⊆ G a subgroup and X a groupoid endowed with a H-good right G-action. The operations above give the orbit space X/H the structure of a groupoid. Moreover, the unit space (X/H)0 of this groupoid is X 0/H = {uH : u ∈ X 0}, where X 0 is the unit space of X, and the range and source functions satisfy s(xH) = s(x)H and r(xH) = r(x)H . Proof: Let us first prove that the product is well-defined. Let (xH, yH) ∈ (X/H)2 be a composable pair. The fact that xeh does not depend on the rep- resentativeeh chosen from Hx,y follows from the assumption that the action is H-good, since if h1, h2 ∈ Hx,y then we have s(x)h1 = r(y) = s(x)h2 , and therefore s(x)h1h−1 i.e. xh1 = xh2. 2 = s(x), and because the action is H-good xh1h−1 2 = x, 23 Hx,yfh2z = {h ∈ H : s(x)h = r(yfh2z)} = {h ∈ H : s(x)h = r(y)fh2} = {h ∈ H : s(x)hfh2 = Hx,yfh2 . −1 = r(y)} composable. Hence, since Hx,y 6= ∅ it follows that Hx,yfh2z 6= ∅, and therefore (xH, yfh2zH) is As we saw above Hxfh1y,z = Hy,z, and sincefh2 ∈ Hy,z, we can write (xHyH)zH = xfh1yHzH = (xfh1y)fh2zH = xfh1fh2yfh2zH . Let us now prove that X/H is a groupoid with the operations above. We check associativity first. Suppose xH, yH, zH ∈ X/H are such that (xH, yH) is composable and (yH, zH) is composable. We want to prove that (xHyH, zH) and (xH, yHzH) are also composable and moreover (xHyH)zH = xH(yHzH). We have by definition that xHyH = xfh1yH and yHzH = yfh2zH, wherefh1 is any element of Hx,y andfh2 is any element of Hy,z. We now notice that Hxfh1y,z = {h ∈ H : s(xfh1y)h = r(z)} = {h ∈ H : s(y)h = r(z)} = Hy,z . Since Hy,z 6= ∅ it follows that Hxfh1y,z 6= ∅, and therefore (xfh1yH, zH) is com- posable. Similarly, Also seen above, we have that Hx,yfh2z = Hx,yfh2, so that fh1fh2 ∈ Hx,yfh2z. Hence, we conclude that (xHyH)zH = xH(yHzH) . We now check that for any element xH ∈ X/H we have that (xH, x−1H) and (x−1H, xH) are composable pairs. We have that Hx,x−1 = {h ∈ H : s(x)h = r(x−1)} = {h ∈ H : s(x)h = s(x)} , and the identity element e obviously belongs to the latter set. Hence we conclude that Hx,x−1 6= ∅, and therefore (xH, x−1H) is composable. A similar observation shows that (x−1H, xH) is also composable. To prove that X/H is a groupoid it now remains to prove the inverse identi- ties xHyHy−1H = xH and y−1HyHxH = xH, in case (xH, yH) is composable (for the first identity) and (yH, xH) is composable (for the second identity). We first show that yHy−1H = r(y)H. We have that yHy−1H = yehy−1H for any elementeh ∈ Hy,y−1. Since, as we observed above, we always have e ∈ Hy,y−1, it follows that we can takeeh as e. Thus, we get yHy−1H = yy−1H = r(y)H . (25) From this it follows that xHyHy−1H = xH r(y)H = xfh1r(y)H , 24 wherefh1 is any element of Hx,r(y). By definition,fh1 is such that r(y) = s(x)fh1 = s(xfh1). Hence we have that xfh1r(y) = xfh1, and therefore xHyHy−1H = xfh1H = xH . The other identity y−1HyHxH = xH is proven in a similar fashion. Hence, we conclude that X/H is a groupoid. From equality (25) it follows easily that the units of X/H are precisely the elements of the form uH where u ∈ X 0, so that we can write (X/H)0 = X 0/H. Also from (25) it follows that the range function in X/H satisfies: r(xH) = r(x)H . The analogous result for the source function is proven in a similar fashion. Let α be an action of G on a Fell bundle A over a groupoid X. Assume that the action is H-good, where H is a subgroup of G. We will now define a new Fell bundle A/H over the groupoid X/H. First we set some notation. The set of H-orbits of the action α on A gives us a partition of A into equivalence classes. We will denote by [a] the equivalence class of the element a ∈ A, i.e. We define A/H as the set of all the H-orbits in A, i.e. [a] := {αh(a)}h∈H . A/H := {[a] : a ∈ A} , (26) which, as we will now see, is a Fell bundle over X/H in a natural way. Proposition 2.11. Let α be an action of a group G on a Fell bundle A over a groupoid X and H ⊆ G be a subgroup for which the G-action is H-good. The set of H-orbits A/H forms a Fell bundle over the groupoid X/H in the following way: • The associated projection pH : A/H → X/H is defined by pH([a]) := p(a)H, where p is the associated projection of the bundle A. • The vector space structure on each fiber (cid:0)A/H(cid:1)xH is defined in the fol- if a, b ∈ Ax then [a] + [b] := [a + b], and if λ ∈ C then lowing way: λ[a] := [λa]. • The norm on A/H is defined by k[a]k := kak. • The multiplication maps (cid:0)A/H(cid:1)xH ×(cid:0)A/H(cid:1)yH → (cid:0)A/H(cid:1)xH·yH , for a composable pair (xH, yH), are defined in the following way: and b ∈ Ay, then if a ∈ Ax [a][b] = [αeh−1(a)b] , (27) whereeh is any element of Hx,y. • The involution map is defined by [a]∗ := [a∗]. 25 Lemma 2.12. Let α be an action of G on a Fell bundle A over a groupoid X and H ⊆ G be a subgroup for which the G-action is H-good. Let x ∈ X and a ∈ Ax. Given any y ∈ xH there exists a unique representative b of [a] such that b ∈ Ay. Proof: Given an element y ∈ xH we have that y = xh for some h ∈ H. The element αh−1 (a) is then a representative of [a] such that αh−1(a) ∈ Axh = Ay, thus existence is established. The uniqueness claim follows from the fact the action is H-good. Suppose we have two representatives b and c of [a] such that both b and c belong to Ay. Being representatives of [a] means that there are elements h1, h2 ∈ H such that b = αh1 (a) and c = αh2 (a). Hence we have that αh2h−1 1 (b) = c , and therefore h2h−1 1 fore s(y)h1h−1 and therefore b = c. 2 = y and there- (b) = b, takes Ay into Ay. This means that yh1h−1 2 = s(y). Since the action is H-good it follows that αh2h−1 1 Proof of Proposition 2.11: First, it is clear that the vector space struc- ture on each fiber(cid:0)A/H(cid:1)xH is well-defined. By this we mean two things: first, given two elements [a], [b] ∈ (cid:0)A/H(cid:1)xH there exist unique representatives a, b second, the sum [a + b] still lies in(cid:0)A/H(cid:1)xH and does not depend on the choice such that a, b ∈ Ax for a given representative x of the orbit xH (Lemma 2.12); of representatives a and b (provided only that a and b are in the same fiber). The norm on A/H is also easily seen to be well-defined, i.e. independent of the choice of representative. This is true because any other representative of [a] is of the form αh(a) for some h ∈ H, and by Remark 1.29 we know that αh gives an isometry between fibers. It is also clear that each fiber(cid:0)A/H(cid:1)xH is a The multiplication map is also easily seen to be well-defined: using the fact the G-action on A is H-good we know that αeh−1 (a)b does not depend on [αeh−1 (a)b] ∈ AxehyH . The fact that the multiplication map does not depend on the chosen representatives of the orbits [a] and [b] is also easily checked. the choice of element eh ∈ Hx,y. Moreover, αeh−1(a)b ∈ Axehy and therefore It follows from a routine computation that map (cid:0)A/H(cid:1)xH ×(cid:0)A/H(cid:1)yH → (cid:0)A/H(cid:1)xH·yH is bilinear. Moreover, for [a] ∈ (cid:0)A/H(cid:1)xH and [b] ∈ (cid:0)A/H(cid:1)yH , where we assume without loss of generality that a ∈ Ax and b ∈ Ay, we have that Banach space under this norm. k[a][b]k = kαeh−1(a)bk ≤ kαeh−1(a)kkbk = kakkbk = k[a]kk[b]k . We will now check associativity of the multiplication maps. Let (xH, yH) and (yH, zH) be two composable pairs in X/H, and let [a] ∈ (A/H)xH, [b] ∈ (A/H)yH and [c] ∈ (A/H)zH , where we assume without loss of generality that a ∈ Ax, b ∈ Ay and c ∈ Az. By definition, we have [a][b] = [αfh1 −1 (a)b], where 26 fh1 is any element of Hx,y. Thus, we have (cid:0)[a][b](cid:1)[c] = [αfh1 = [αfh2 −1 (a)b][c] = [αfh2 −1 fh1 −1(a)αfh2 −1 (b)c] , −1(cid:0)αfh1 −1 (a)b(cid:1)c] Proposition 2.10 where this is done) that Hxfh1y,z = Hy,z and moreover that where fh2 is any element of Hxfh1y,z. One can easily check (or see the proof of fh1fh2 ∈ Hx,yfh2z. From this observations it follows that −1(b)c] −1 (a)αfh2 −1(b)c] = [a][αfh2 (cid:0)[a][b](cid:1)[c] = [αfh2 −1 fh1 = [a](cid:0)[b][c]) . Hence, the multiplication maps are associative. The involution on A/H is also easily seen not to depend on choice of repre- sentative of the orbit, since the maps αh preserve the involution of A. Morevoer, it is easily checked that: if [a] ∈(cid:0)A/H(cid:1)xH then [a]∗ ∈(cid:0)A/H(cid:1)x−1H , the associ- ated map(cid:0)A/H(cid:1)xH →(cid:0)A/H(cid:1)x−1H is conjugate linear, and [a]∗∗ = [a]. Let us now check that(cid:0)[a][b](cid:1)∗ = [b]∗[a]∗, whenever the multiplication is defined. Let us assume that a ∈ Ax and b ∈ Ay and that (xH, yH) is composable. We have that = [αeh−1(a)b]∗ = [b∗αeh−1 (a∗)] = [αeh(b∗)a∗] , (cid:0)[a][b](cid:1)∗ (cid:0)[a][b](cid:1)∗ whereeh is any element of Hx,y. It is easily seen thateh−1 ∈ Hy−1,x−1, so that = [αeh(b∗)a∗] = [b∗][a∗] = [b]∗[a]∗ . We also need to prove that k[a]∗[a]k = k[a]k2. This is also easy because k[a]∗[a]k = k[a∗][a]k = k[a∗a]k = ka∗ak = kak2 = k[a]k2 . The last thing we need to check is that if [a] ∈ (A/H)xH , then [a]∗[a] is a positive element of (A/H)s(x)H (seen as a C∗-algebra). We have that [a]∗[a] = [a∗a]. We can assume without loss of generality that a ∈ Ax, so that a∗a ∈ As(x). Since A is a Fell bundle we have that a∗a is a positive element of As(x) (seen as a C∗-algebra). Hence, there exists an element b ∈ As(x) such that a∗a = b∗b. Moreover, [b] ∈ (A/H)s(x) and it is now clear that [a]∗[a] = [a∗a] = [b∗b] = [b]∗[b] , i.e. [a]∗[a] is a positive element of (A/H)s(x). This finishes our proof that A/H is a Fell bundle. Convention. For simplicity we will henceforward make the following conven- tion. Given an orbit Fell bundle A/H as discribed in Proposition 2.11, if we write that an element [a] belongs to some fiber (A/H)xH , we will always assume that the representative a of [a] belongs to fiber over the representative x of xH. In other words, if we write that [a] ∈ (A/H)xH, then we are implicitly assuming that a ∈ Ax. This is possible and unambiguous by Lemma 2.12. 27 We apply this convention also for elements of Cc(A/H), meaning that a canonical element [a]xH ∈ Cc(A/H) is always assumed to be written in a way that a ∈ Ax. It is a strighforward fact that any function in Cc(X/H) can also be seen as a complex-valued (H-invariant) function in X. This funcion in X is in general no longer finitely supported, but it still makes sense as a function in C(X), the vector space of all complex-valued functions in X. We will now see that something analogous can be said for the elements of Cc(A/H). Given an element f ∈ Cc(A/H) we define a function ι(f ) ∈ C(A), where C(A) is the vector space of all sections of A, by the following rule: ι(f )(x) := Rx(f (xH)) , (28) where Rx(f (xH)) is the unique representative of f (xH) such that Rx(f (xH)) ∈ Ax, which is well-defined according to Lemma 2.12. It is then easy to see that the map ι is an injective linear map from Cc(A/H) to C(A). For ease of reading we will henceforward drop the symbol ι and use the same notation both for elements of Cc(A/H) and for their correspondents in C(A). It will then be clear from context which one we are using. Under this convention we can then write, for any f ∈ Cc(A/H) and x ∈ X, that [f (x)] = f (xH). Moreover, the decomposition (17) of f ∈ Cc(A/H) as a sum of elements of the form [a]xH can now be written as: f = XxH∈X/H(cid:0)f (xH)(cid:1)xH = XxH∈X/H [f (x)]xH . (29) 2.2 Examples In this section we give some examples of H-good actions and actions satisfying the H-intersection property. For the rest of the section we assume that A is a Fell bundle over a groupoid X where a group G acts and H ⊆ G denotes a subgroup. The first two examples (2.13 and 2.14) show that H-good actions that satisfy the H-intersection property are present in actions that have completely opposite behaviours, such as free actions and actions that fix every point. Example 2.13. If the restricted action of H on the unit space X 0 is free, then the action is H-good and satisfies the H-intersection property. Example 2.14. If the restricted action of H on A fixes every point, then the action is H-good and satisfies the H-intersection property. The following example is one of the examples that motivated the develop- ment of this theory of crossed products by Hecke pairs. This example, and the study of the crossed products associated to it, seems to be valuable for obtaining a form of Katayama duality with respect to crossed products by "coactions" of 28 discrete homogeneous spaces. Example 2.15. Suppose X is the transformation groupoid G × G. We recall that the multiplication and inversion operations on this groupoid are given by: (s, tr)(t, r) = (st, r) and (s, t)−1 = (s−1, st) . Recall also that the source and range functions on G × G are defined by s(s, t) = (e, t) and r(s, t) = (e, st) . We observe that there is a natural right G-action on G × G, given by (s, t)g := (s, tg) . (30) Let δ be a coaction of G on a C∗-algebra B and B the associated Fell bundle. Following [5, Section 3], we will denote by A := B × G the corresponding Fell bundle over the groupoid G × G. Elements of A have the form (bs, t), where bs ∈ Bs and s, t ∈ G. Any such element lies in the fiber A(s,t) over (s, t). It is easy to see that there is a canonical actin α of G on A, given by αg(bs, t) := (bs, tg−1) . This action of G on A entails the natural right action of G on G×G, as described in (30). This G-action on G × G is free and therefore the action α is H-good and satisfies the H-intersection property with respect to any subgroup H ⊆ G. The orbit space groupoid (G × G)/H can be canonically identified with the groupoid G × G/H of [4], whose operations are given by: (s, trH)(t, rH) = (st, rH) and (s, tH)−1 = (s−1, stH) . Moreover, the orbit Fell bundle A/H is canonically identified with the Fell bun- dle B × G/H over G × G/H defined in [4], and in this way Cc(A/H) is canon- ically isomorphic with the Echterhoff-Quigg algebra Cc(B × G/H), also from [4]. 2.3 The algebra M(Cc(A)) We will assume for the rest of this section that G is a group, H ⊆ G is a subgroup and A is a Fell bundle over a groupoid X endowed with a G-action α. We also assume that the action α is H-good. We recall that A/H stands for the orbit Fell bundle over the groupoid X/H, as defined in (26). For the purpose of defining crossed products by Hecke pairs it is convenient to have a "large" algebra which contains the algebras Cc(A/H) for different sub- groups H ⊆ G. In this way we are allowed to multiply elements of Cc(A/H) and Cc(A/K), for different subgroups H, K ⊆ G, in a meaningful way. This large algebra will be the multiplier algebra M (Cc(A)). This section is thus devoted to show how algebras such as Cc(A/H) and Cc(X 0/H) embed in M (Cc(A)) in a canonical way. Our first result shows that there is a natural inclusion Cc(A/H) ⊆ M (Cc(A)). 29 Theorem 2.16. There is an embedding ι of Cc(A/H) into M (Cc(A)) deter- mined by the following rule: for any x, y ∈ X, a ∈ Ax and b ∈ Ay we have ι([a]xH )by :=((αeh−1(a)b)xehy , 0, if Hx,y 6= ∅ otherwise, (31) whereeh is any element of Hx,y. Remark 2.17. The above result allows us to see Cc(A/H) as a ∗-subalgebra of M (Cc(A)). We shall henceforward drop the symbol ι and make no distinction of notation between an element of Cc(A/H) and its correspondent multiplier in M (Cc(A)). Proof of Theorem 2.16: Let us first show that expression (31) does indeed define an element of M (Cc(A)). For this it is enough to check that hι([a]xH )by, czi = hby, ι([a]∗ x−1H ) czi, for all b ∈ Ay and c ∈ Az, with y, z ∈ X. For ι([a]xH )by to be non-zero, we must necessarily have Hx,y 6= ∅, and in this case ι([a]xH )by = (αeh−1(a)b)xehy, whereeh ∈ Hx,y. Now, hι([a]xH )by, czi = h(αeh−1 (a)b)xehy, czi = (b∗αeh−1 (a)∗)y−1(x−1eh)cz = b∗ y−1αeh−1(a)∗ x−1eh cz x−1eh For αeh−1 (a)∗ cz to be non-zero we must necessarily have r(z) = s(x−1)eh, i.e. eh ∈ Hx−1,z. So, to summarize, for h[a]xH by, czi to be non-zero we must have Hx,y ∩ Hx−1,z 6= ∅ and in this case we obtain hι([a]xH )by, czi = b∗ cz , y−1αeh−1 (a)∗ x−1eh whereeh is any element of Hx,y∩Hx−1,z. A similar computation for hby, ι([a]∗ Recall from (17) that any f ∈ Cc(A/H) can be written as yelds the exact same result. x−1H ) czi f = XxH∈X/H(cid:0)f (xH)(cid:1)xH . From this we are able to define a multiplier ι(f ) ∈ M (Cc(A)), simply by ex- tending expression (31) by linearity. We want to show that ι is an injective ∗-homomorphism. First, we claim that given [a]xH, [b]yH ∈ Cc(A/H) we have ι([a]xH )ι([b]yH ) = ι([a]xH [b]yH ) . This amounts to proving that ι([a]xH )ι([b]yH ) =(ι([αeh−1 (a)b]xehyH ) , 0, if Hx,y 6= ∅ otherwise 30 have with eh being any element of Hx,y. To see this, let cz ∈ Az, with z ∈ X. We ι([a]xH )ι([b]yH )cz = (ι([a]xH )(αh−1 if Hy,z 6= ∅ otherwise (b)c)yh0z , 0, 0 = ((αh−1 0, 1 (a)αh−1 0 (b)c)xh1yh0z , if Hy,z 6= ∅ and Hx,yh0z 6= ∅ otherwise with h0 ∈ Hy,z and h1 ∈ Hx,yh0z. But Hx,yh0z = Hx,yh0 = Hx,y h0, hence the above can be written as 0 0, = ((αh−1 = ((cid:0)αh−1 0, 0 eh−1(a)αh−1 0 (b)c)xehh0yh0z , if Hy,z 6= ∅ and Hx,y 6= ∅ otherwise (αeh−1 (a)b)c(cid:1)(xehy)h0z , if Hy,z 6= ∅ and Hx,y 6= ∅ otherwise whereeh ∈ Hx,y. Also, Hy,z = Hxehy,z. Thus, we obtain 0 0, (αeh−1 (a)b)c(cid:1)(xehy)h0z , = ((cid:0)αh−1 = (ι([αeh−1 (a)b]xehyH) cz , 0, if Hx,y 6= ∅ otherwise if Hxehy,z 6= ∅ and Hx,y 6= ∅ otherwise Since ι is linear and multiplicative on the elements of the form [a]xH , it is necessarily an homomorphism. Now the fact that ι([a]xH )∗ = ι(([a]xH )∗) = ι([a]∗ x−1H ) follows directly from the computations in the beginning of this proof. Hence, ι is a ∗-homomorphism. Let us now prove injectivity of ι. Suppose f ∈ Cc(A/H) is such that ι(f ) = 0. Decomposing f as a sum of elements of the form [a]xH , following (29), we get 0 = ι(f ) = XxH∈X/H For any y ∈ X we then have ι(cid:0)[f (x)]xH(cid:1) . ι(cid:16)(cid:0)f (xH)(cid:1)xH(cid:17) = XxH∈X/H ι(cid:0)[f (x)]xH(cid:1)(f (y)∗)y−1 ι(cid:0)[f (x)]xH(cid:1)(f (y)∗)y−1 0 = XxH∈X/H = XxH∈X/H = XxH∈X/H s(y)∈s(x)H(cid:0)αfhx s(y)∈s(x)H −1 (f (x))f (y)∗(cid:1)xfhxy−1 , wherefhx is any element of Hx,y−1. Now the elements xfhxy−1 in the sum above are all different, because if we had x1ghx1y−1 = x2ghx2y−1, then we would have 31 x1ghx1 = x2ghx2 and therefore x1H = x2H. Therefore each of the summands in the above sum is zero, and in particular we must have −1 (f (y))f (y)∗(cid:1)yfhyy−1 0 = (cid:0)αfhy = (cid:0)f (y)f (y)∗(cid:1)r(y) , and therefore f (y)f (y)∗ = 0. Hence we get f (y) = 0, and since this is true for any y ∈ X, we have f = 0, i.e. ι is injective. Proposition 2.18. There is an embedding ι of Cb(X 0) into M (Cc(A)) defined by ι(f ) by := f (r(y))by . (32) for every f ∈ Cb(X 0), y ∈ X and b ∈ Ay. Remark 2.19. The above result allows us to see Cb(X 0) as a ∗-subalgebra of M (Cc(A)). We shall henceforward drop the symbol ι and make no distinction of notation between an element of Cb(X 0) and its correspondent multiplier in M (Cc(A)). Proof of Proposition 2.18 : It is easy to see that hι(f )by , czi = hby , ι(f ∗)czi for any y, z ∈ X, b ∈ Ay and c ∈ Az, so that the expression (32) does define an element of M (Cc(A)). Hence we get a linear map ι : Cb(X 0) → M (Cc(A)). Given two elements f1, f2 ∈ Cb(X 0), we have that ι(f1)ι(f2)by = f1(r(y))f2(r(y))by = ι(f1f2)by for any y ∈ X andb ∈ Ay, so that ι is a ∗-homomorphism. Hence, we only need to prove that ι is injective. This is not difficult to see: given f ∈ Cb(X 0) such that ι(f ) = 0 we have, for any unit u ∈ X 0 and b ∈ Au, that 0 = ι(f )bu = f (u)bu . Hence, f (u) = 0 because each fiber Au is non-zero by our assumption on Fell bundles (see Assumption 1.27). Since this is true for any u ∈ X 0 we get f = 0, i.e. ι is injective. Recall, from Lemma 2.2, that the action of G on X restricts to an action of G on the set X 0. Thus it makes sense to talk about the commutative ∗-algebra Cc(X 0/H) ⊆ Cb(X 0) . Since there is a canonical embedding, given by Proposition 2.18, of Cb(X 0) into M (Cc(A)), we have in particular an embedding of Cc(X 0/H) into M (Cc(A)) which identifies an element f ∈ Cc(X 0/H) with the multiplier f ∈ M (Cc(A)) given by: f by := f (r(y)H)by . 32 Moreover Proposition 2.18 applied to the groupoid X/H and the Fell bundle A/H shows that there is a canonical embedding of Cb(X 0/H) into M (Cc(A/H)), which identifies an element f ∈ Cb(X 0/H) with the multiplier f ∈ M (Cc(A/H)) given by f [b]yH := f (r(y)H)[b]yH . (33) Since both Cc(X 0/H) and Cc(A/H) are canonically embedded in M (Cc(A)), it is convenient to understand what happens (inside M (Cc(A))) when one multi- plies an element of Cc(X 0/H) by an element Cc(A/H). Perhaps unsurprisingly, this product is given exactly by expression (33), which models the action of Cc(X 0/H) on Cc(A/H) as multipliers of the latter algebra. In other words, it makes no difference to view Cc(X 0/H) inside M (Cc(A/H)) or inside M (Cc(A)) when it comes to multiplication by elements of Cc(A/H). We will now show how the multiplication of elements of Cc(A/H) by ele- ments of Cc(X 0) is determined (inside M (Cc(A))). Before we proceed we will first introduce some notation that will be used throughout this work: Given a set A ⊂ X 0 we will denote by 1A ∈ Cb(X 0) the characteristic function of A. In case A is a singleton {u} we will simply write 1u. Proposition 2.20. Inside M (Cc(A)) we have that, for x ∈ X, a ∈ Ax and u ∈ X 0, [a]xH1u =(αeh−1(a)xeh , 0, if Hx,u 6= ∅ otherwise, whereeh is any element of Hx,u. Proof: Let y ∈ X and b ∈ Ay. For the product [a]xH1u by to be non-zero we must necessarily have u = r(y) (from (32)), and in this case we obtain [a]xH1u by = [a]xH by = (αeh−1 (a)b)xehy = αeh−1 (a)xehby , whereeh is any element of Hx,y. Since u = r(y), we have Hx,y = Hx,u, and this concludes the proof. It will be of particular importance to know how to multiply, inside M (Cc(A)), elements of Cc(A/H) with elements of Cc(A/K) when K ⊆ H is an arbitrary subgroup. It turns out that the algebra Cc(A/K) is preserved by multiplication by elements of Cc(A/H), as we show in the next result: Proposition 2.21. Let K ⊆ H be any subgroup. We have that [a]xH [b]yK =([αeh−1 (a)b]xehyK , 0, if Hx,y 6= ∅ otherwise, (34) where x, y ∈ X, a ∈ Ax and b ∈ Ay. In particular Cc(A/K) is invariant under multiplication by elements of Cc(A/H). 33 Proof: First we observe that since the action is assumed to be H-good, it is automatically K-good, so that we can form the groupoid X/K and the Fell bundle A/K. Let z ∈ X and c ∈ Az. We have that [a]xH [b]yKcz = ([a]xH (αek−1 (b)c)yekz , 0, if Ky,z 6= ∅ otherwise, 0, = ((αeh−1 (a)αek−1 (b)c)xehyekz , = ((αeh−1 (a)αek−1 (b)c)(cid:0)xehek−1y(cid:1)ekz 0, if Hx,yekz 6= ∅ and Ky,z 6= ∅ otherwise, , if Hx,yekz 6= ∅ and Ky,z 6= ∅ otherwise, , Ky,z = Kxehek−1y,z, we conclude that where ek is any element of Ky,z and eh is any element of Hx,yekz. Now, since Hx,yekz = Hx,yek = Hx,yek, it follows that ehek−1 ∈ Hx,y, and moreover since = ((αek−1 (αekeh−1 (a)b)c)(cid:0)xehek−1y(cid:1)ekz = ([αekeh−1 (a)b]xehek−1yKcz , Thus (34) follows immediately (the elementeh in (34) is simply the element de- noted byehek−1 above). if Hx,y 6= ∅ and Kxehek−1y,z 6= ∅ otherwise, In case the subgroup K has finite index in H we can strengthen Proposition if Hx,y 6= ∅ otherwise. 0, 0, 2.21 in the following way: Proposition 2.22. Let K ⊆ H be a subgroup such that [H : K] < ∞. Inside M (Cc(A)) we have that [a]xH = X[h]∈Sx\H/K [αh−1 (a)]xhK , (35) for any x ∈ X and a ∈ Ax. Cc(A/H) is a ∗-subalgebra of Cc(A/K). In particular, inside M (Cc(A)) we have that Proof: First we notice that since [H : K] < ∞ we have that the right hand side of (35) is a finite sum and therefore does indeed define an element of Cc(A/K). To prove this result it suffices to show that [a]xH by = X[h]∈Sx\H/K [αh−1 (a)]xhKby , (36) for all y ∈ X and b ∈ Ay. First we notice that both the right and left hand sides of (36) are zero unless r(y) ∈ s(x)H. In case r(y) ∈ s(x)H we have [a]xH by = (αeh−1 (a)b)xehy , 34 Recall from Proposition 1.23 that there is a bijective correspondence between the set of K-orbits (xH)/K and the double coset space Sx\H/K. It is clear that whereeh is any element of Hx,y. [a]xehKby = (αeh−1 (a)b)xehy. Moreover, for all the classes [h] 6= [eh] in Sx\H/K we have r(y) /∈ s(x)hK, because r(y) ∈ s(x)ehK. Hence, for all the classes [h] 6= [eh] in Sx\H/K we have [αh−1(a)]xhKby = 0. We conclude that [αh−1(a)]xhKby = [αeh−1 (a)]xehKby = (αeh−1 (a)b)xehy , X[h]∈Sx\H/K and equality (36) is proven. Remark 2.23. In Proposition 2.22 the fact that [H : K] < ∞ was only used to ensure that the sum on the right hand side of (35) was finite. One could more generally just require that the sets Sx\H/K are finite for all x ∈ X, but this generality will not be used here. As we saw in Proposition 2.6 we have an action α of G on Cc(A). This action can be extended in a unique way to an action on M (Cc(A)), which we will still denote by α, by the following formula: where g ∈ G, T ∈ M (Cc(A)) and f ∈ Cc(A). We will now show what this action on M (Cc(A)) does to the algebras Cb(X 0), Cc(A/H) and Cc(X 0/H). αg(T )f := αg(cid:0)T αg−1 (f )(cid:1) , (37) Proposition 2.24. The extension of the action α to M (Cc(A)), also denoted by α, satisfies the following properties: (i) The restriction of α to Cb(X 0) is precisely the action that comes from the G-action on X 0. (ii) For any g ∈ G the automorphism αg takes Cc(X 0/H) to Cc(X 0/gHg−1), by αg(1xH) = 1(xg−1)(gHg−1) . (38) (iii) For any g ∈ G the automorphism αg takes Cc(A/H) to Cc(A/gHg−1), by αg([a]xH ) = [αg(a)](xg−1)(gHg−1) . (39) (iv) Both Cc(A/H) and Cc(X 0/H) are contained in M (Cc(A))H , the algebra of H-fixed points. Proof: (i) Let y ∈ X, b ∈ Ay and f ∈ Cb(X 0). For any g ∈ G let us denote by fg ∈ Cb(X 0) the function defined by fg(x) = f (xg). By definition of the extension of α to M (Cc(A)), we have αg(f ) by = αg(f · α−1 g (by)) = αg(f · αg−1 (b)yg) = αg(f (r(yg))αg−1 (b)yg) = αg(f (r(y)g)αg−1 (b)yg) = f (r(y)g)by = fg(r(y))by = fg · by . 35 Hence we conclude that αg(f ) = fg and therefore the action α on Cb(X 0) is just the action that comes from the G-action on X 0. (ii) This follows directly from (i). (iii) Let y ∈ X and b ∈ Ay. By definition of the extension of α to M (Cc(A)), we have αg([a]xH ) by = αg([a]xH α−1 g (by)) = αg([a]xH αg−1 (b)yg) . Also, we can see that αg([a]xHαg−1 (b)yg) = (αg(cid:0)(αeh−1 (a)αg−1 (b))xeh(yg)(cid:1) , 0, if Hx,yg 6= ∅ otherwise 0, = ((αgeh−1 (a)b)xehg−1y , = ((αgeh−1 (a)b)xg−1gehg−1y , 0, if Hx,yg 6= ∅ otherwise if Hx,yg 6= ∅ otherwise Hx,yg = g−1(cid:0)gHg−1(cid:1)xg−1,y g , whereeh ∈ Hx,yg. Now an easy computation shows that we have and thereby we obtain, for t ∈(cid:0)gHg−1(cid:1)xg−1,y, αg([a]xH) by = ((αgg−1t−1g(a)b)xg−1ty , = ((αt−1g(a)b)xg−1ty , otherwise otherwise if (cid:0)gHg−1(cid:1)xg−1,y 6= ∅ if (cid:0)gHg−1(cid:1)xg−1,y 6= ∅ 0, 0, = [αg(a)](xg−1)(gHg−1 ) by . (iv) This follows directly from (ii) and (iii). It is important to know how to multiply an element of Cc(A/H) with an element of Cc(X 0/gHg−1) inside M (Cc(A)). This is easy if we are under the assumption that G-action satisfies the H-intersection property. We recall from (8) that H g stands for the subgroup H ∩ gHg−1. Proposition 2.25. If the G-action moreover satisfies the H-intersection prop- erty, then for every x ∈ X and g ∈ G the following equality holds in M (Cc(A)): [a]xH 1s(x)gHg−1 = [a]xHg . 36 Proof: For any y ∈ X and b ∈ Ay we have [a]xH 1s(x)gHg−1 by = 0, = ([a]xH by , = ((αeh−1 (a)b)xehy , = ((αeh−1 (a)b)xehy , 0, 0, if r(y) ∈ s(x)gHg−1 otherwise if r(y) ∈ s(x)gHg−1 and r(y) ∈ s(x)H otherwise if r(y) ∈ s(x)H ∩ s(x)gHg−1 otherwise whereeh ∈ Hx,y. Now, by the H-intersection property, we obtain =((αeh−1 (a)b)xehy , 0, if r(y) ∈ s(x)H g otherwise . Of course, we have (H g)x,y ⊆ Hx,y, and hence we can chooseeh as an element of (H g)x,y, thereby obtaining = [a]xHg by , which finishes the proof. 3 ∗-Algebraic crossed product by a Hecke pair In this section we introduce our notion of a (∗-algebraic) crossed product by a Hecke pair and we explore its basic properties and its representation theory. Throughout the rest of this work we impose the following standing assumption, based on the tools developed in Section 2.1. Standing Assumption 3.1. We assume from now on that (G, Γ) is a Hecke pair, A is a Fell bundle over a groupoid X endowed with a Γ-good right G-action α satisfying the Γ-intersection property. 3.1 Definition of the crossed product and basic properties In this section we aim at defining the (∗-algebraic) crossed product of Cc(A/Γ) by the Hecke pair (G, Γ). For that we are going to define some sort of a bun- dle over G/Γ, where the fiber over each gΓ is precisely Cc(A/Γg). Recall that we denote by α the associated action of G on Cc(A) and also its extension to M (Cc(A)). Definition 3.2. Let B(A, G, Γ) be the vector space of finitely supported func- tions f : G/Γ → M (Cc(A)) satisfying the following compatibility condition f (γgΓ) = αγ(f (gΓ)) , (40) 37 for all γ ∈ Γ and gΓ ∈ G/Γ. Lemma 3.3. For every f ∈ B(A, G, Γ) and gΓ ∈ G/Γ we have f (gΓ) ∈ M (Cc(A))Γg . Proof : This follows directly from the compatibility condition (40), since for every γ ∈ Γg we have αγ(f (gΓ)) = f (γgΓ) = f (gΓ). Definition 3.4. The vector subspace of B(A, G, Γ) consisting of the functions f : G/Γ → M (Cc(A)) satisfying the compatibility condition (40) and the prop- erty f (gΓ) ∈ Cc(A/Γg) , (41) will be denoted by Cc(A/Γ) ×alg product of Cc(A/Γ) by the Hecke pair (G, Γ). α G/Γ and will be called the ∗-algebraic crossed It is relevant to point out that the definitions of the spaces B(A, G, Γ) and Cc(A/Γ) ×alg α G/Γ seem more suitable for Hecke pairs (G, Γ), as in general a function in B(A, G, Γ) could only have support on those elements gΓ ∈ G/Γ such that ΓgΓ/Γ < ∞. We now define a product and an involution in B(A, G, Γ) by: (f1 ∗ f2)(gΓ) := X[h]∈G/Γ (f ∗) (gΓ) := ∆(g−1)αg(f (g−1Γ))∗ . f1(hΓ) αh(f2(h−1gΓ)) , (42) (43) Proposition 3.5. B(A, G, Γ) becomes a unital ∗-algebra under the product and involution defined above, whose identity element is the function f such that f (Γ) = 1 and is zero in the remaining points of G/Γ. Proof: First, we claim that the expression for the product defined above is well-defined in B(A, G, Γ), i.e. for f1, f2 ∈ B(A, G, Γ) the expression (f1 ∗ f2)(gΓ) := X[h]∈G/Γ f1(hΓ) αh(f2(h−1gΓ)) is independent from the choice of the representatives [h] and also that it has finitely many summands. Independence from the choice of the representatives [h] ∈ G/Γ follows directly from the compatibility condition (40) and the fact that the sum is finite follows simply from the fact that f1 has finite support. Now we claim that f1 ∗ f2 has also finite support, for f1, f2 ∈ B(A, G, Γ). Let S1, S2 ⊆ G/Γ be the supports of the functions f1 and f2 respectively. We 38 will regard S1 and S2 as subsets of G (being finite unions of left cosets). It is easy to check that the function G × G → M (Cc(A)) (h, g) 7→ f1(hΓ)αh(f2(h−1gΓ)) has support contained in S1 × (S1 · S2). Since (G, Γ) is a Hecke pair, the product S1 · S2 is also a finite union of left cosets. Hence, f1 ∗ f2 has finite support. We also notice that f1 ∗ f2 satisfies the compatibility condition (40), thus defining an element of B(A, G, Γ), since for any γ ∈ Γ we have f1(hΓ) αh(f2(h−1γgΓ)) f1(γhΓ) αγh(f2(h−1gΓ)) (f1 ∗ f2)(γgΓ) = X[h]∈G/Γ = X[h]∈G/Γ = X[h]∈G/Γ = αγ(cid:0)(f1 ∗ f2)(gΓ)(cid:1) . αγ(f1(hΓ)) αγ ◦ αh(f2(h−1gΓ)) In a similar way we can see that the expression that defines the involution is well-defined in B(A, G, Γ). There are now a few things that need to be checked before we can say that B(A, G, Γ) is a ∗-algebra, namely that the product is associative and the involution is indeed an involution relatively to this product (the fact that the product is distributive and the properties concerning multi- plication by scalars are obvious). The proofs of these facts are essentially just a mimic of the corresponding proofs for "classical" crossed products by groups. Thus, we can say that B(A, G, Γ) is ∗-algebra under this product and involu- tion. Theorem 3.6. Cc(A/Γ) ×alg a ∗-algebra for the above operations. α G/Γ is a ∗-ideal of B(A, G, Γ). In particular it is Proof: It is easy to see that the space Cc(A/Γ) ×alg α G/Γ is invariant for the involution, i.e. f ∈ Cc(A/Γ) ×alg α G/Γ =⇒ f ∗ ∈ Cc(A/Γ) ×alg α G/Γ . Thus, to prove that Cc(A/Γ) ×alg α G/Γ is a (two-sided) ∗-ideal of B(A, G, Γ) if f1 ∈ B(A, G, Γ) and f2 ∈ it is enough to prove that is a right ideal, i.e. Cc(A/Γ) ×alg α G/Γ, because any right ∗-ideal α G/Γ then f1 ∗ f2 ∈ Cc(A/Γ) ×alg is automatically two-sided. Hence, all we need to prove is that (f1 ∗ f2)(gΓ) ∈ Cc(A/Γg), for every f1 ∈ B(A, G, Γ) and f2 ∈ Cc(A/Γ) ×alg α G/Γ. The proof of this fact will follow the following steps: 1) Prove that: given a subgroup H ⊆ G, f ∈ Cc(A/H) and a unit u ∈ X 0, we have f · 1u ∈ Cc(A). 2) Let T := (f1 ∗ f2)(gΓ) =P[h]∈G/Γ f1(hΓ)αh(f2(h−1gΓ)). Use 1) to show that T · 1u ∈ Cc(A) for any unit u ∈ X 0. 39 conclude that T 1uΓg ∈ Cc(A/Γg). 3) Fix a unit u ∈ X 0. By 2) we have T 1u =Pi(ai)xi , where the elements xi ∈ X are such that s(xi) = u. Show that T 1uΓg = Pi[ai]xiΓg , and T =Pn 4) Prove that there exists a finite set of units {u1, . . . , un} ⊆ X 0 such that i=1 T 1uiΓg . Conclude that T ∈ Cc(A/Γg). • Proof of 1) : This follows immediately from Proposition 2.20. • Proof of 2) : We know that f2(h−1gΓ) ∈ Cc(A/Γh−1g). Thus, from Proposition 2.24, we conclude that αh(f2(h−1g)) ∈ Cc(cid:0)A/hΓh−1∩gΓg−1(cid:1). Now, using 1), we see that αh(f2(h−1g)) 1u ∈ Cc(A) and consequently f1(hΓ)αh(f2(h−1g)) 1u ∈ Cc(A). Hence, T 1u ∈ Cc(A). • Proof of 3) : For any γ ∈ Γg we have, using Lemma 3.3, T 1uγ = αγ−1(T ) 1uγ = αγ−1(cid:0)T αγ(1uγ)(cid:1) αγ−1(ai)xiγ . = αγ−1(cid:0)T 1u(cid:1) =Xi Let y ∈ X and b ∈ Ay. We have T 1uΓg by = (T by , 0, if r(y) ∈ uΓg otherwise . then have Assume now that r(y) ∈ uΓg and leteγ ∈ Γg be such that r(y) = ueγ. We T by = T 1ueγ by = Xi αeγ−1(ai)xieγ by . Since s(xi) = u, we have s(xieγ) = ueγ = r(y). Hence, (αeγ−1(ai)b)xi eγy . T by = Xi We conclude that T 1uΓg by = (Pi(αeγ−1 (ai)b)xieγy , 0, if r(y) ∈ uΓg otherwise . = Xi [ai]xiΓg by . Thus, T 1uΓg =Pi[ai]xiΓg ∈ Cc(A/Γg). • Proof of 4) : For easiness of reading of this last part of the proof we introduce the following definition: given F ∈ M (Cc(A)) we define the support of F to be the set {u ∈ X 0 : F 1u 6= 0}. Notice in particular that the support of an element [a]xH , with a 6= 0, is the set s(x)H. 40 Since αh(f2(h−1gΓ)) ∈ Cc(A/hΓh−1 ∩gΓg−1), there exists a finite number of units v1, . . . , vk ∈ X 0 such that αh(f2(h−1gΓ)) has support in k[i=1 vi(cid:0)hΓh−1 ∩ gΓg−1(cid:1) ⊆ vigΓg−1 . k[i=1 Hence, there is a finite number of units w1, . . . , wl ∈ X 0 such that T has support contained in wigΓg−1 . l[i=1 Therefore, T has support contained in l[i=1 m[j=1 wiθjΓg , where θ1, . . . , θm are representatives of the classes of gΓg−1/Γg (being a finite number because (G, Γ) is a Hecke pair). Thus, we have proven that there is a finite number of units u1, . . . , un ∈ X 0 such that T has support i=1 uiΓg. Moreover, we can suppose we have chosen the units u1, . . . , un such that the corresponding orbits uiΓg are mutually disjoint. i=1 T 1uiΓg . Indeed, given y ∈ X inside Sn It is now easy to see that we have T =Pn and b ∈ Ay, if r(y) /∈Sn nXi=1 T by = T 1r(y) by = 0 = i=1 uiΓg, then T 1uiΓg by , i=1 uiΓg, then r(y) belongs to precisely one of the orbits, and if r(y) ∈Sn say ui0 Γg, and we have nXi=1 T 1uiΓg by = T 1ui0 Γg by = T by . Hence, we must have T = Pn T ∈ Cc(A/Γg). i=1 T 1uiΓg , and by 3) we conclude that As it is well-known, when working with crossed products A × G by discrete groups, one always has an embedded copy of A inside the crossed product. Something analogous happens in the case of crossed products by Hecke pairs, where Cc(A/Γ) is canonically embedded in Cc(A/Γ) ×alg α G/Γ, as is stated in the next result (whose proof amounts to routine verification). Proposition 3.7. There is a natural embedding of the ∗-algebra Cc(A/Γ) in Cc(A/Γ) ×alg α G/Γ, which identifies an element f ∈ Cc(A/Γ) with the function ι(f ) ∈ Cc(A/Γ) ×alg α G/Γ such that ι(f )(Γ) = f and ι(f ) is zero elsewhere . 41 Remark 3.8. The above result says that we can identify Cc(A/Γ) with the functions of Cc(A/Γ) ×alg α G/Γ with support in Γ. We shall, henceforward, make no distinctions in notation between an element of Cc(A/Γ) and its corre- spondent in Cc(A/Γ) ×alg α G/Γ. Theorem 3.9. Cc(A/Γ) ×alg particular, Cc(A/Γ) ×alg natural embeddings α G/Γ is an essential ∗-ideal of B(A, G, Γ). In α G/Γ is an essential ∗-algebra. Moreover, there are Cc(A/Γ) ×alg α G/Γ ֒→ B(A, G, Γ) ֒→ M (Cc(A/Γ) ×alg α G/Γ) , that make the following diagram commute M (Cc(A/Γ) ×alg α G/Γ) 4❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ L Cc(A/Γ) ×alg α G/Γ / B(A, G, Γ) . Proof: We have already proven that Cc(A/Γ) ×alg α G/Γ is a ∗-ideal of B(A, G, Γ), thus we only need to check that this ideal is in fact essential. Sup- pose f ∈ B(A, G, Γ) is such that f ∗(cid:0)Cc(A/Γ) ×alg α G/Γ(cid:1) = {0}. Then, in particular, using Proposition 3.7, we must have f ∗(cid:0)Cc(A/Γ)(cid:1) = {0}. Let g ∈ G and take [a]xΓ ∈ Cc(A/Γ), we then have 0 =(cid:16)f ∗ [a]xΓ(cid:17) (gΓ) = f (gΓ)αg([a]xΓ) = f (gΓ)[αg(a)]xg−1gΓg−1 . Thus, multiplying by 1s(x)g−1 ∈ M (Cc(A)) we get 0 = f (gΓ)[αg(a)]xg−1gΓg−1 1s(x)g−1 = f (gΓ)αg(a)xg−1 = f (gΓ)αg(ax) . Since this true for all a ∈ Ax and x ∈ X and given that α takes fibers of A bijectively into fibers of A, we must have f (gΓ)by = 0 for all b ∈ Ay and y ∈ X. Hence, we must have f (gΓ) = 0. Thus, f = 0 and we conclude that Cc(A/Γ) ×alg α G/Γ is an essential ∗-ideal of B(A, G, Γ). Since Cc(A/Γ) ×alg α G/Γ is a ∗-subalgebra of B(A, G, Γ), we immediately conclude that Cc(A/Γ) ×alg α G/Γ is an essential ∗-algebra. The embedding of B(A, G, Γ) in M (Cc(A/Γ) ×alg α G/Γ) then follows from the universal property of multiplier algebras, Theorem 1.16. In the theory of crossed products A × G by groups, one always has an em- bedded copy of the group algebra C(G) inside the multiplier algebra M (A × G). Something analogous happens in the case of crossed products by Hecke pairs, where the Hecke algebra H(G, Γ) is canonically embedded in the multiplier alge- bra M (Cc(A/Γ) ×alg α G/Γ), as is stated in the next result (whose proof amounts to routine verification). 42 4 / O O Proposition 3.10. The Hecke ∗-algebra H(G, Γ) embeds in B(A, G, Γ) in the following way: an element f ∈ H(G, Γ) is identified with the element ef ∈ B(A, G, Γ) given by where 1 is the unit of M (Cc(A)). ef (gΓ) := f (gΓ)1 , The next result does not typically play an essential role in the case of crossed products by groups, but will be extremely important for us in case of crossed products by Hecke pairs. The proof is also just routine verification. Proposition 3.11. The algebra Cc(X 0/Γ) embeds in B(A, G, Γ) in the fol- lowing way: an element f ∈ Cc(X 0/Γ) is identified with the function ι(f ) ∈ B(A, G, Γ) given by ι(f )(Γ) = f and ι(f ) is zero elsewhere . Remark 3.12. Propositions 3.10 and 3.11 allow us to view both the Hecke ∗- algebra H(G, Γ) and Cc(X 0/Γ) as ∗-subalgebras of B(A, G, Γ). We shall hence- forward make no distinctions in notation between an element of H(G, Γ) or Cc(X 0/Γ) and its correspondent in B(A, G, Γ). The purpose of the following diagram is to illustrate, in a more condensed form, all the canonical embeddings we have been considering so far: Cc(A/Γ) H(G, Γ) Cc(X 0/Γ) / Cc(A/Γ) ×alg α G/Γ ◗◗◗◗◗◗◗◗◗◗◗◗◗ 3❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢❢ / B(A, G, Γ) / M (Cc(A/Γ) ×alg α G/Γ)) Remark 3.13. The reason for considering the algebra B(A, G, Γ) is two-fold. On one side B(A, G, Γ) made it easier to make sure the convolution product (42) was well-defined in Cc(A/Γ) ×alg α G/Γ. On the other (perhaps more important) side, the fact that both H(G, Γ) and Cc(X 0/Γ) are canonically embedded in B(A, G, Γ) allows us to treat the elements of H(G, Γ) and Cc(X 0/Γ) both as multipliers in M (Cc(A/Γ) ×alg α G/Γ)), but also allows us to operate these ele- ments with the convolution product and involution expressions (42) and (43), as these are defined in B(A, G, Γ). 43 / ( ( / / 3 As it is well-known in the theory of crossed products by discrete groups, a (∗-algebraic) crossed product A × G is spanned by elements of the form a ∗ g, where a ∈ A and g ∈ G (here g is seen as an element of the group algebra C(G) ⊆ M (A × G)). We will now explore something analogous in the case of crossed products by Hecke pairs. It turns out that Cc(A/Γ)×alg α G/Γ is spanned by elements of the form [a]xΓ∗ΓgΓ∗1s(x)gΓ, where x ∈ X, a ∈ Ax and gΓ ∈ G/Γ, as we show in the next result. Theorem 3.14. For any f ∈ Cc(A/Γ) ×alg α G/Γ we have f = X[g]∈Γ\G/Γ XxΓg ∈X/Γghf (gΓ)(x)ixΓ ∗ ΓgΓ ∗ 1s(x)gΓ . (44) In particular, Cc(A/Γ) ×alg α G/Γ is spanned by elements of the form with x ∈ X, a ∈ Ax and gΓ ∈ G/Γ. [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ , The following lemma is needed in order to prove the above result: Lemma 3.15. Let x ∈ X, a ∈ Ax and gΓ ∈ G/Γ. We have [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ (hΓ) =([αγ(a)]xγ−1Γγg , 0, if hΓ = γgΓ, with γ ∈ Γ otherwise . In particular, [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ (gΓ) = [a]xΓg . Proof: An easy computation yields [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ (hΓ) = [a]xΓ · ΓgΓ(hΓ) · αh(1s(x)gΓ) , from which we conclude that [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ is supported in the double coset ΓgΓ. Now, evaluating at the point gΓ ∈ G/Γ we get [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ (gΓ) = [a]xΓ · ΓgΓ(gΓ) · αg(1s(x)gΓ) = [a]xΓ · αg(1s(x)gΓ) = [a]xΓ · 1s(x)gΓg−1 = [a]xΓg , where the last equality comes from Proposition 2.25. From the compatibility condition (40) and Proposition 2.24 it then follows that, for γ ∈ Γ, [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ (γgΓ) = αγ([a]xΓg ) = [αγ(a)]xγ−1Γγg . 44 Proof of Theorem 3.14: Let us first prove that the expression on the right hand side of (44) is well-defined. It is easy to see that for every g ∈ G, the expression XxΓg∈X/Γghf (gΓ)(x)ixΓ ∗ ΓgΓ ∗ 1s(x)gΓ does not depend on the choice of the representative x of xΓg. Now, let us see that it also does not depend on the choice of the representative g in ΓgΓ. Let γgθ, with γ, θ ∈ Γ, be any other representative. We have ∗ ΓγgθΓ ∗ 1s(x)γgθΓ = XxΓγgθ ∈X/Γγgθhf (γgθΓ)(x)ixΓ = XxΓγg ∈X/Γγghf (γgΓ)(x)ixΓ = XxΓγg ∈X/Γγghαγ(f (gΓ))(x)ixΓ = XxΓγg ∈X/Γγghαγ(f (gΓ)(xγ))ixΓ ∗ ΓgΓ ∗ 1s(x)γgΓ ∗ ΓgΓ ∗ 1s(x)γgΓ ∗ ΓgΓ ∗ 1s(x)γgΓ We notice that there is a well-defined bijective correspondence X/Γg → X/Γγg given by xΓg 7→ xγ−1Γγg. Thus, we get = XxΓg∈X/Γghαγ(f (gΓ)(x))ixγ−1Γ = XxΓg∈X/Γghf (gΓ)(x)ixΓ ∗ ΓgΓ ∗ 1s(x)gΓ . ∗ ΓgΓ ∗ 1s(xγ−1)γgΓ Hence, the expression in (44) is well-defined. Let us now prove the decomposi- tion in question. For any tΓ ∈ G/Γ we have X[g]∈Γ\G/Γ XxΓg∈X/Γghf (gΓ)(x)ixΓ = XxΓt∈X/Γthf (tΓ)(x)ixΓ ∗ ΓtΓ ∗ 1s(x)tΓ (tΓ) . ∗ ΓgΓ ∗ 1s(x)gΓ (tΓ) = By Lemma 3.15 it follows that = XxΓt∈X/Γthf (tΓ)(x)ixΓt = f (tΓ) , and this finishes the proof. In the following result we collect some useful equalities concerning products in Cc(A/Γ) ×alg α G/Γ, which will be useful later on. One should observe the sim- ilarities between the equalities (47) and (48) and the equalities obtained by an 45 Huef, Kaliszewski and Raeburn in [7, Lemma 1.3 (i) and (ii)] if in their setting one was allowed to somehow "drop" the representations. The similarity is more than a coincidence and will be addressed in the sequel of this article. Proposition 3.16. In Cc(A/Γ) ×alg α G/Γ the following equalities hold: (cid:0)[a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ(cid:1)∗ = ∆(g) [αg−1 (a∗)]x−1gΓ ∗ Γg−1Γ ∗ 1s(x−1)Γ , (45) 1r(x)Γ ∗ ΓgΓ ∗ [αg−1 (a)]xgΓ = [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ , [a]xΓ ∗ ΓgΓ ∗ 1s(x)γgΓ . (46) (47) [a]xΓ ∗ ΓgΓ = X[γ]∈Sx\Γ/Γg ΓgΓ ∗ [a]xΓ = X[γ]∈Sx\Γ/Γg−1 1r(x)γg−1Γ ∗ ΓgΓ ∗ [a]xΓ . (48) In particular, from (46) we see that Cc(A/Γ) ×alg elements of the form 1r(x)Γ ∗ ΓgΓ ∗ [a]xgΓ, with g ∈ G, x ∈ X and a ∈ Ax. α G/Γ is also spanned by all Proof: Let us first prove equality (45). First we notice that (cid:0)[a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ(cid:1)∗ = ∆(g) 1s(x)gΓ ∗ Γg−1Γ ∗ [a∗]x−1Γ , has support in the double coset Γg−1Γ. (g−1Γ) = Now evaluating this element on g−1Γ we get, which means that(cid:0)[a]xΓ ∗ ΓgΓ∗ 1s(x)gΓ(cid:1)∗ (cid:0)[a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ(cid:1)∗ = ∆(g) αg−1(cid:0)([a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ) (gΓ)(cid:1)∗ = ∆(g)(cid:0)[αg−1 (a∗)]x−1gΓ ∗ Γg−1Γ ∗ 1s(x−1)Γ(cid:1) (g−1Γ) . 1r(x)Γ ∗ ΓgΓ ∗ [αg−1 (a)]xgΓ = ∆(g)(cid:0)[αg−1 (a∗)]x−1gΓ ∗ Γg−1Γ ∗ 1r(x)Γ(cid:1)∗ = ∆(g) αg−1 ([a]xΓg )∗ = ∆(g) [αg−1 (a∗)]x−1gΓg−1 Let us now prove equality (46). We have = ∆(g)(cid:0)[αg−1 (a∗)]x−1gΓ ∗ Γg−1Γ ∗ 1s(x−1g)g−1Γ(cid:1)∗ , which together with (45) yields = ∆(g)∆(g−1) [a]xΓ ∗ ΓgΓ ∗ 1s(xg)Γ = [a]xΓ ∗ ΓgΓ ∗ 1s(xg)Γ . Let us now prove (47). An easy computation yields [a]xΓ ∗ ΓgΓ (hΓ) = [a]xΓ · ΓgΓ(hΓ) , 46 from which we conclude that [a]xΓ ∗ ΓgΓ has support in ΓgΓ. Evaluating this element on the point gΓ we get [a]xΓ ∗ ΓgΓ (gΓ) = [a]xΓ · ΓgΓ(gΓ) = [a]xΓ . From Proposition 2.22 one always has the following decomposition [a]xΓ = X[γ]∈Sx\Γ/Γg [αγ−1 (a)]xγΓg . Together with Lemma 3.15 we get [a]xΓ ∗ ΓgΓ (gΓ) = [a]xΓ = X[γ]∈Sx\Γ/Γg = X[γ]∈Sx\Γ/Γg = X[γ]∈Sx\Γ/Γg [αγ−1 (a)]xγΓg [αγ−1 (a)]xγΓ ∗ ΓgΓ ∗ 1s(x)γgΓ (gΓ) [a]xΓ ∗ ΓgΓ ∗ 1s(x)γgΓ (gΓ) , and equality (47) is proven. Equality (48) follows easily from (47) by taking the involution and using the fact that Sx = Sx−1 . The last claim of this proposition follows simply from (46) and Proposition 3.14. In the theory of crossed products A × G by discrete groups one has a "co- variance relation" of the form g ∗ a ∗ g−1 = αg(a). This relation is essential in the passage from covariant representations of the system (A, G, α) to represen- tations of the crossed product. More generally, the following relation holds in A × G: g ∗ a ∗ h = αg(a) ∗ gh . We will now explore how this generalizes to the setting of crossed products by Hecke pairs. What we are aiming for is a description of how products of the form ΓgΓ∗[a]xΓ∗ΓsΓ can be expressed by the canonical spanning set of elements of the form [b]yΓ ∗ ΓhΓ ∗ 1s(x)hΓ (according to Theorem 3.14). This will be achieved in Corollary 3.19 below and will play an important role in the representation theory of crossed products by Hecke pairs, particularly in the definition of covariant representations. One should observe the similarities between the expressions we obtain both in Theorem 3.17 and Corollary 3.19 and the expression provided by an Huef, Kaliszewski and Raeburn in [7, Definition 1.1] (if one "forgets" the representations in their setting). Once again, this is more than a coincidence as we will see in in the sequel to this article. In fact, an Huef, Kaliszewski and Raeburn's definition served as a guiding line for our results below and for the definition of a covariant representation (Definition 3.24) which we shall present in the next section. Before we establish the results we are aiming for we need to establish some notation, which will be used throughout this work. For w, v ∈ G and a unit 47 y ∈ X 0 we define the sets ny dy w,v :=(cid:8)[r] ∈ ΓwΓ/Γ : r−1wvΓ ⊆ ΓvΓ and yw−1 ∈ yΓr−1(cid:9) , w,v :=(cid:8)[r] ∈ ΓwΓ/Γ : r−1wvΓ ⊆ ΓvΓ and yw−1 ∈ yΓr−1Γwv(cid:9) . and the numbers ny w,v := # ny dy w,v := # dy w,v , w,v , N y w,v := ny w,v dy w,v . We will also denote by Ey u,v the double coset space Ey u,v := Sy\Γ/(uΓu−1 ∩ vΓv−1) . Theorem 3.17. Let g, s ∈ G and y ∈ X 0. We have that (49) (50) (51) (52) (53) (54) ΓgΓ ∗ 1yΓ ∗ ΓsΓ = X[w]∈ΓgΓ/Γ w−1,v [v]∈ΓsΓ/Γ X[γ]∈Ey = X[v]∈ΓsΓ/Γ X[γ]∈Ey [v]∈ΓsΓ/Γ X[γ]∈Ey = X[u]∈Γg−1Γ/Γ g−1,v u,v N yγ w,v L(g)N yγ g,v L(gv) L(wv) (cid:0)1yγw−1Γ ∗ ΓwvΓ ∗ 1yγvΓ(cid:1) (cid:0)1yγg−1Γ ∗ ΓgvΓ ∗ 1yγvΓ(cid:1) (cid:0)1yγuΓ ∗ Γu−1vΓ ∗ 1yγvΓ(cid:1) . ∆(g)N yγ L(u−1v) u−1,v In order to prove the above result we will need the following lemma, which gives some properties of the numbers ny w,v and dy w,v. Lemma 3.18. Let w, v, ∈ G, θ ∈ Γ and y ∈ X 0. The numbers ny satisfy the following properties: w,v and dy w,v i) ny w,vθ = ny w,v ii) ny θw,v = ny w,v i′) dy w,vθ = dy w,v ii′) dy θw,v = dy w,v iii) nyθ w,θ−1v = ny wθ−1,v iii′) dyθ w,θ−1v = dy wθ−1,v More generally, if ew,ev ∈ G and ey ∈ X 0 are such that ΓewΓ = ΓwΓ, ΓevΓ = ΓvΓ, eyΓ = yΓ, ewevΓ = wvΓ and eyew−1Γwv = yw−1Γwv, then w,v = ney w,v = dey iv′) dy iv) ny ew,ev ew,ev 48 Proof: Assertions i) and i′) are obvious. Assertion ii) follows from the observation that [r] 7→ [θ−1r] establishes a bijection between the sets ny w,v and n y θw,v. Assertion ii′) is proven in a similar fashion as assertion ii). n ey To prove assertion iv), let θ ∈ Γwv be such that eyew−1 = yw−1θ. We have ew,ev = (cid:8)[r] ∈ ΓewΓ/Γ : r−1ewevΓ ⊆ ΓevΓ and eyew−1 ∈eyΓr−1(cid:9) = (cid:8)[r] ∈ ΓwΓ/Γ : r−1wvΓ ⊆ ΓvΓ and yw−1θ ∈ yΓr−1(cid:9) . = (cid:8)[r] ∈ Γθ−1wΓ/Γ : r−1θ−1wvΓ ⊆ ΓvΓ and yw−1θ ∈ yΓr−1(cid:9) Since θ ∈ Γwv we have θwvΓ = wvΓ, so that y θ−1w,v . = n Now, from assertion ii), it follows that ney ew,ev = ny θ−1w,v = ny w,v. notice that d ey As for assertion iv′), taking θ ∈ Γwv again as such that eyew−1 = yw−1θ, we ew,ev = (cid:8)[r] ∈ ΓewΓ/Γ : r−1ewevΓ ⊆ ΓevΓ and eyew−1 ∈eyΓr−1Γ ewev(cid:9) = (cid:8)[r] ∈ ΓwΓ/Γ : r−1wvΓ ⊆ ΓvΓ and yw−1θ ∈ yΓr−1Γwv(cid:9) = (cid:8)[r] ∈ ΓwΓ/Γ : r−1wvΓ ⊆ ΓvΓ and yw−1 ∈ yΓr−1Γwv(cid:9) = dy w,v . Assertions iii) and iii′) are a direct consequence of iv) and iv′). Proof of Theorem 3.17: We have ΓgΓ(wΓ) αw(cid:0)(1yΓ ∗ ΓsΓ) (w−1tΓ)(cid:1) αw(cid:0)(1yΓ ∗ ΓsΓ) (w−1tΓ)(cid:1) αw(cid:0)1yΓ · ΓsΓ(w−1tΓ)(cid:1) ΓgΓ ∗ 1yΓ ∗ ΓsΓ (tΓ) = X[w]∈G/Γ = X[w]∈ΓgΓ/Γ = X[w]∈ΓgΓ/Γ = X[w]∈ΓgΓ/Γ = X[w]∈ΓgΓ/Γ w−1tΓ⊆ΓsΓ w−1tΓ⊆ΓsΓ αw(1yΓ) 1yΓw−1 We now claim that X[w]∈ΓgΓ/Γ w−1tΓ⊆ΓsΓ 1yΓw−1 = X[w]∈ΓgΓ/Γ w−1tΓ⊆ΓsΓ X[γ]∈Ey w−1,w−1 t N yγ w,w−1t 1yγw−1Γt . (55) To see this, we will evaluate both the right and left expressions above on all points x ∈ X 0 and see that we obtain the same value. First, we note that 49 if x ∈ X 0 is not of the form yθew−1, for some θ ∈ Γ and ew ∈ ΓgΓ such that ew−1tΓ ⊆ ΓsΓ, then both expressions are zero. Suppose now that x = yθew−1 for some ew ∈ ΓgΓ such that ew−1tΓ ⊆ ΓsΓ. Evaluating the left expression we get X[w]∈ΓgΓ/Γ As for the right expression, first we observe that if yθew−1 ∈ yγw−1Γt, then by Lemma 3.18 iv) and iv′) we have N yθ right expression we get 1yΓw−1(yθew−1) = nyθ 1yΓw−1(yθew−1) = w,w−1t. Thus, evaluating the X[w]∈Γ ewΓ/Γ ew, ew−1t = N yγ w−1 ew ew−1tΓ⊆Γ ew−1tΓ ew, ew−1t . w−1tΓ⊆ΓsΓ w−1,w−1t X[w]∈ΓgΓ/Γ w−1tΓ⊆ΓsΓ X[γ]∈Ey w−1tΓ⊆ΓsΓ X[γ]∈Ey = X[w]∈ΓgΓ/Γ ew, ew−1t X[w]∈ΓgΓ/Γ = N yθ w−1,w−1t w−1tΓ⊆ΓsΓ X[γ]∈Ey N yγ N yθ w,w−1t 1yγw−1Γt (yθew−1) = ew, ew−1t 1yγw−1Γt (yθew−1) 1yγw−1Γt (yθew−1) w−1,w−1t Using Proposition 1.23 we notice that X[γ]∈Ey w−1,w−1 t 1yγw−1Γt = X[γ]∈Ey w−1,w−1t = 1yΓw−1Γt , from which we obtain that, 1yγ(w−1Γw∩w−1tΓt−1w)w−1 N yθ w−1tΓ⊆ΓsΓ X[γ]∈Ey ew, ew−1t X[w]∈ΓgΓ/Γ ew, ew−1t X[w]∈ΓgΓ/Γ w−1tΓ⊆ΓsΓ = N yθ w−1,w−1t 1yΓw−1Γt (yθew−1) 1yγw−1Γt (yθew−1) = = N yθ ew, ew−1t X[w]∈Γ ewΓ/Γ w−1 ew ew−1tΓ⊆Γ ew−1tΓ 1yΓw−1Γt (yθew−1) = N yθ = nyθ ew, ew−1t dyθ ew, ew−1t . ew, ew−1t So, equality (55) is established. Now, by Proposition 3.15, we see that w−1tΓ⊆ΓsΓ X[γ]∈Ey X[w]∈ΓgΓ/Γ w−1tΓ⊆ΓsΓ X[γ]∈Ey = X[w]∈ΓgΓ/Γ w−1,w−1t w−1,w−1 t N yγ w,w−1t 1yγw−1Γt = N yγ w,w−1t(cid:0)1yγw−1Γ ∗ ΓtΓ ∗ 1yγw−1tΓ(cid:1)(tΓ) 50 Now, using the fact that condition w−1tΓ ⊆ ΓsΓ means that there exists a (necessarily unique) element [v] ∈ ΓsΓ/Γ such that w−1tΓ = vΓ, or equivalently, tΓ = wvΓ, we obtain = X[w]∈ΓgΓ/Γ [v]∈ΓsΓ/Γ wvΓ=tΓ X[γ]∈Ey = X[w]∈ΓgΓ/Γ [v]∈ΓsΓ/Γ N yγ w,w−1t(cid:0)1yγw−1Γ ∗ ΓtΓ ∗ 1yγw−1tΓ(cid:1)(tΓ) w,v(cid:0)1yγw−1Γ ∗ ΓwvΓ ∗ 1yγvΓ(cid:1)(tΓ) . N yγ X[γ]∈Ey w−1,v w−1,w−1t wvΓ=tΓ We now claim that [v]∈ΓsΓ/Γ X[γ]∈Ey X[w]∈ΓgΓ/Γ [v]∈ΓsΓ/Γ X[γ]∈Ey = X[w]∈ΓgΓ/Γ wvΓ=tΓ w−1,v w−1,v N yγ w,v(cid:0)1yγw−1Γ ∗ ΓwvΓ ∗ 1yγvΓ(cid:1)(tΓ) = L(wv) (cid:0)1yγw−1Γ ∗ ΓwvΓ ∗ 1yγvΓ(cid:1)(tΓ) N yγ w,v To prove this we note that, given any [w] ∈ ΓgΓ/Γ and [v] ∈ ΓsΓ/Γ, the element (cid:0)1yγw−1Γ ∗ ΓwvΓ ∗ 1yγvΓ(cid:1)(tΓ) is nonzero if and only if ΓtΓ = ΓwvΓ, so that we can write w−1,v w−1,v [v]∈ΓsΓ/Γ wvΓ⊆ΓtΓ [v]∈ΓsΓ/Γ X[γ]∈Ey X[w]∈ΓgΓ/Γ X[γ]∈Ey = X[w]∈ΓgΓ/Γ = X[θ]∈Γ/Γt X[w]∈ΓgΓ/Γ = X[θ]∈Γ/Γt X[w]∈ΓgΓ/Γ = X[θ]∈Γ/Γt X[w]∈ΓgΓ/Γ [v]∈ΓsΓ/Γ θwvΓ=θtΓ [v]∈ΓsΓ/Γ wvΓ=θtΓ [v]∈ΓsΓ/Γ N yγ w,v N yγ w,v N yγ w,v L(wv) (cid:0)1yγw−1Γ ∗ ΓwvΓ ∗ 1yγvΓ(cid:1)(tΓ) = L(wv) (cid:0)1yγw−1Γ ∗ ΓwvΓ ∗ 1yγvΓ(cid:1)(tΓ) X[γ]∈Ey X[γ]∈Ey X[γ]∈Ey L(wv) (cid:0)1yγw−1Γ ∗ ΓwvΓ ∗ 1yγvΓ(cid:1)(tΓ) L(θwv) (cid:0)1yγw−1θ−1Γ ∗ ΓθwvΓ ∗ 1yγvΓ(cid:1)(tΓ) L(wv) (cid:0)1yγw−1Γ ∗ ΓwvΓ ∗ 1yγvΓ(cid:1)(tΓ) N yγ N yγ w−1θ−1,v w−1,v w−1,v θw,v θw,v wvΓ=tΓ By Lemma 3.18 ii) and ii′) we know that N yγ θw,v = N yγ w,v, hence 51 wvΓ=tΓ [v]∈ΓsΓ/Γ = X[θ]∈Γ/Γt X[w]∈ΓgΓ/Γ = L(t) X[w]∈ΓgΓ/Γ X[γ]∈Ey = X[w]∈ΓgΓ/Γ [v]∈ΓsΓ/Γ wvΓ=tΓ [v]∈ΓsΓ/Γ X[γ]∈Ey X[γ]∈Ey w−1,v N yγ w,v w−1,v N yγ w,v L(wv) (cid:0)1yγw−1Γ ∗ ΓwvΓ ∗ 1yγvΓ(cid:1)(tΓ) L(wv) (cid:0)1yγw−1Γ ∗ ΓwvΓ ∗ 1yγvΓ(cid:1)(tΓ) w,v(cid:0)1yγw−1Γ ∗ ΓwvΓ ∗ 1yγvΓ(cid:1)(tΓ) . N yγ w−1,v wvΓ=tΓ Hence, we have proven that ΓgΓ ∗ 1yΓ ∗ ΓsΓ = X[w]∈ΓgΓ/Γ [v]∈ΓsΓ/Γ X[γ]∈Ey w−1,v N yγ w,v L(wv) (cid:0)1yγw−1Γ ∗ ΓwvΓ ∗ 1yγvΓ(cid:1) . Also, w−1,v [v]∈ΓsΓ/Γ X[γ]∈Ey X[w]∈ΓgΓ/Γ [v]∈ΓsΓ/Γ X[γ]∈Ey = X[θ]∈Γ/Γg [v]∈ΓsΓ/Γ X[γ]∈Ey = X[θ]∈Γ/Γg = X[v]∈ΓsΓ/Γ X[γ]∈Ey g−1,v g−1,v g−1θ−1 ,v N yγ w,v N yγ θg,v L(wv) (cid:0)1yγw−1Γ ∗ ΓwvΓ ∗ 1yγvΓ(cid:1) L(θgv) (cid:0)1yγg−1θ−1Γ ∗ ΓθgvΓ ∗ 1yγvΓ(cid:1) L(gv) (cid:0)1yγg−1Γ ∗ ΓgvΓ ∗ 1yγvΓ(cid:1) N yγ g,v L(g)N yγ g,v L(gv) (cid:0)1yγg−1Γ ∗ ΓgvΓ ∗ 1yγvΓ(cid:1) . Moreover, we also have L(g)N yγ g,v L(gv) (cid:0)1yγg−1Γ ∗ ΓgvΓ ∗ 1yγvΓ(cid:1) ∆(g)N yγ g,v L(gv) (cid:0)1yγg−1Γ ∗ ΓgvΓ ∗ 1yγvΓ(cid:1) (cid:0)1yγg−1Γ ∗ ΓgvΓ ∗ 1yγvΓ(cid:1) g−1,v X[v]∈ΓsΓ/Γ X[γ]∈Ey = L(g−1) X[v]∈ΓsΓ/Γ X[γ]∈Ey = X[θ]∈Γ/Γg−1 = X[θ]∈Γ/Γg−1 X[γ]∈Ey X[γ]∈Ey [v]∈ΓsΓ/Γ g−1,θ−1 v g−1,v [v]∈ΓsΓ/Γ g−1,v ∆(g)N yγ g,v L(gv) ∆(g)N yγ g,θ−1v L(gθ−1v) (cid:0)1yγg−1Γ ∗ Γgθ−1vΓ ∗ 1yγθ−1vΓ(cid:1) , 52 but since there is a well-defined bijection Ey [γθ], we obtain θg−1,v → Ey g−1,θ−1v given by [γ] 7→ = X[θ]∈Γ/Γg−1 [v]∈ΓsΓ/Γ X[γ]∈Ey θg−1,v ∆(g)N yγθ L(gθ−1v) g,θ−1v (cid:0)1yγθg−1Γ ∗ Γgθ−1vΓ ∗ 1yγθθ−1vΓ(cid:1) and from Lemma 3.18 we get N yγθ g,θ−1v = N yγ gθ−1,v, thus [v]∈ΓsΓ/Γ = X[θ]∈Γ/Γg−1 = X[u]∈Γg−1Γ/Γ X[γ]∈Ey [v]∈ΓsΓ/Γ X[γ]∈Ey u,v θg−1,v ∆(g)N yγ gθ−1,v L(gθ−1v) ∆(g)N yγ L(u−1v) u−1,v (cid:0)1yγθg−1Γ ∗ Γgθ−1vΓ ∗ 1yγvΓ(cid:1) (cid:0)1yγuΓ ∗ Γu−1vΓ ∗ 1yγvΓ(cid:1) . Corollary 3.19. Similarly, for a ∈ Ax with x ∈ X, we have ΓgΓ ∗ [a]xΓ ∗ ΓsΓ = w−1,v [v]∈ΓsΓ/Γ X[γ]∈E s(x) = X[w]∈ΓgΓ/Γ = X[v]∈ΓsΓ/Γ X[γ]∈E s(x) [v]∈ΓsΓ/Γ X[γ]∈E s(x) = X[u]∈Γg−1Γ/Γ g−1,v u,v N s(x)γ w,v L(g)N s(x)γ g,v L(gv) L(wv) (cid:0)[αwγ−1(a)]xγw−1Γ ∗ ΓwvΓ ∗ 1s(x)γvΓ(cid:1) (cid:0)[αgγ−1 (a)]xγg−1Γ ∗ ΓgvΓ ∗ 1s(x)γvΓ(cid:1) (cid:0)[αu−1γ−1(a)]xγuΓ ∗ Γu−1vΓ ∗ 1s(x)γvΓ(cid:1) . ∆(g)N s(x)γ u−1,v L(u−1v) Proof: According to equality (48) in Proposition 3.16 we have ΓgΓ ∗ [a]xΓ ∗ ΓsΓ = = = X[θ]∈Sx\Γ/Γg−1 X[θ]∈Sx\Γ/Γg−1 1r(x)θg−1Γ ∗ ΓgΓ ∗ [a]xΓ ∗ ΓsΓ 1r(x)θg−1Γ ∗ ΓgΓ ∗ [αg−1 (αgθ−1 (a))]xθg−1gΓ ∗ ΓsΓ and by (46) in the same proposition we get = X[θ]∈Sx\Γ/Γg−1 [αgθ−1 (a)]xθg−1Γ ∗ ΓgΓ ∗ 1s(x)Γ ∗ ΓsΓ , 53 and by Theorem 3.17 we obtain w,v N s(x)γ L(wv) [αgθ−1 (a)]xθg−1Γ ∗ 1s(x)γw−1Γ ∗ ΓwvΓ ∗ 1s(x)γvΓ . = X[θ]∈Sx\Γ/Γg−1 [w]∈ΓgΓ/Γ [v]∈ΓsΓ/Γ [γ]∈E s(x) w−1,v For each fixed w, v and γ all the summands in the expression X[θ]∈Sx\Γ/Γg−1 w,v N s(x)γ L(wv) [αgθ−1(a)]xθg−1Γ ∗ 1s(x)γw−1Γ ∗ ΓwvΓ ∗ 1s(x)γvΓ , are zero except precisely for one summand and we have X[θ]∈Sx\Γ/Γg−1 w,v N s(x)γ L(wv) [αgθ−1 (a)]xθg−1Γ ∗ 1s(x)γw−1Γ ∗ ΓwvΓ ∗ 1s(x)γvΓ = w,v N s(x)γ L(wv) Hence we obtain [αwγ−1(a)]xγw−1Γ ∗ ΓwvΓ ∗ 1s(x)γvΓ . ΓgΓ ∗ axΓ ∗ ΓsΓ = [v]∈ΓsΓ/Γ X[γ]∈E s(x) = X[w]∈ΓgΓ/Γ w−1,v w,v N s(x)γ L(wv) [αwγ−1(a)]xγw−1Γ ∗ ΓwvΓ ∗ 1s(x)γvΓ . The remaining equalities in the statement of this corollary are proven in a sim- ilar fashion. 3.2 Basic Examples Example 3.20. We will now show that when Γ is a normal subgroup of G our notion of a crossed product by the Hecke pair (G, Γ) is precisely the usual crossed product by the group G/Γ. Normality of the subgroup Γ implies that the G-action α on M (Cc(A)) gives rise to an action of G/Γ on Cc(A/Γ). Moreover, we have Γg = Γ for all g ∈ G, and it follows easily from the definitions that Cc(A/Γ) ×alg G/Γ is nothing but the usual crossed product by the action of the group G/Γ. It is also interesting to observe that any usual crossed product Cc(B)×algG/Γ coming from an action of the group G/Γ on a Fell bundle B over a groupoid Y is actually a crossed by the Hecke pair (G, Γ) in our sense. To see this we note that the action of G/Γ on B lifts to an action of G on B. In this lifted action the subgroup Γ acts trivially, so that the action is Γ-good. Moreover, since Γ is normal in G, the Γ-intersection property is also trivially satisfied. It is clear that Y /Γ is just Y and B/Γ coincides with B. Thus, forming the crossed product by the Hecke pair (G, Γ) will give nothing but the usual crossed product by G/Γ, i.e. Cc(B/Γ) ×alg G/Γ ∼= Cc(B) ×alg G/Γ. 54 Example 3.21. We will now explain how the Hecke algebra H(G, Γ) is an example of a crossed product by a Hecke pair, namely H(G, Γ) ∼= C ×alg G/Γ, just like group algebras are examples of crossed products by groups. We start with a groupoid X consisting of only one element, i.e. X = {∗}, and we take C as the Fell A bundle over X, i.e. A∗ = C. We take also the trivial G-action α on A. Since the G-action fixes every element of A, it is indeed Γ-good and in this case we have X/Γ = X = {∗}. For the orbit bundle we have that A/Γ = A, and moreover Cc(A/Γ) ∼= Cc(X/Γ) ∼= Cc(X) ∼= C . Hence, we are in the conditions of the Standing Assumption 3.1 and we can form the crossed product Cc(A/Γ) ×alg α G/Γ, which we will simply write as C ×alg α G/Γ. Since C is unital the definitions of B(A, G, Γ) and C ×alg α G/Γ coincide in this case. Moreover Definition 3.4 reads that C ×alg α G/Γ is the set of functions f : G/Γ → C satisfying the compatibility condition (40). Since the action α is trivial, the compatibility condition simply says that C ×alg α G/Γ consists of all the functions f : G/Γ → C which are left Γ-invariant. Morever, the product and involution expressions become respectively (f1 ∗ f2)(gΓ) := X[h]∈G/Γ (f ∗) (gΓ) := ∆(g−1) f (g−1Γ) . f1(hΓ) f2(h−1gΓ) , Hence, it is clear that C ×alg α G/Γ is nothing but the Hecke algebra H(G, Γ). It follows from this that the product ΓgΓ ∗ 1∗Γ ∗ ΓsΓ is just the product of the double cosets ΓgΓ and ΓsΓ inside the Hecke algebra, since 1∗Γ is the identity element. It is interesting to note in this regard that the expression for this product described in Theorem 3.17 is a familiar expression for the product ΓgΓ∗ΓsΓ in H(G, Γ). To see this, we note that the stabilizer S∗ of ∗ is the whole group G, and therefore E∗ u,v consists only of the class [e]. Moreover, the numbers u−1,v and d∗ n∗ u−1,v = 1. Thus, the expression described in Theorem 3.17 is just the usual expression u−1,v, defined in (51) and (52), are equal, so that N ∗ ΓgΓ ∗ ΓsΓ = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ ∆(g) L(u−1v) Γu−1vΓ . Example 3.22. As a generalization of Example 3.21 we will now show that if the G-action fixes every element of the bundle A, then Cc(A/Γ) ×alg α G/Γ is isomorphic to the ∗-algebraic tensor product of Cc(A/Γ) and H(G, Γ). This result also has a known analogue in the theory of crossed products by groups. Proposition 3.23. If the G-action fixes every element of A, then we have Cc(A/Γ) ×alg α G/Γ ∼= Cc(A/Γ) ⊙ H(G, Γ) , where ⊙ is the symbol that denotes the ∗-algebraic tensor product. 55 Proof: Given that we have canonical embeddings of Cc(A/Γ) and H(G, Γ) α G/Γ) we have a canonical linear map from Cc(A/Γ) ⊙ into M (Cc(A/Γ) ×alg H(G, Γ) to M (Cc(A/Γ) ×alg α G/Γ) determined by f1 ⊗ f2 7→ f1 ∗ f2 , (56) where f1 ∈ Cc(A/Γ) and f2 ∈ H(G, Γ). Standard arguments can be used to show that this mapping is injective (since the mappings from both Cc(A/Γ) and H(G, Γ) into the multiplier algebra of the crossed product are injections). It is also clear that the image of the map determined by (56) is contained in Cc(A/Γ) ×alg α G/Γ. Let us now check that this mapping is surjective. First we will show that the elements of Cc(A/Γ) commute with elements of H(G, Γ) inside M (Cc(A/Γ) ×alg α G/Γ). It follows from expressions (47) and (46) that [a]xΓ ∗ ΓgΓ = X[γ]∈Sx\Γ/Γg = X[γ]∈Sx\Γ/Γg [a]xΓ ∗ ΓgΓ ∗ 1s(x)γgΓ 1r(x)Γ ∗ ΓgΓ ∗ [αg−1γ−1(a)]xγgΓ . Since every point of X is fixed by the associated G-action on X, we have that Sx = G, and therefore Sx\Γ/Γg consists only of the class [e], so that we can write = 1r(x)Γ ∗ ΓgΓ ∗ [αg−1 (a)]xgΓ . Moreover, since the G-actions on A and X are trivial we can furthermore write = 1r(x)g−1Γ ∗ ΓgΓ ∗ [a]xΓ . Now, by the same reasoning as above and using expression (48) we have = X[γ]∈Sx\Γ/Γg−1 = ΓgΓ ∗ [a]xΓ . 1r(x)γg−1Γ ∗ ΓgΓ ∗ [a]xΓ Thus we conclude that [a]xΓ ∗ ΓgΓ = ΓgΓ ∗ [a]xΓ. By Theorem 3.14 we know that elements of the form [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ span Cc(A/Γ) ×alg α G/Γ, and from the commutation relation we just proved it follows that [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ = ΓgΓ ∗ [a]xΓ ∗ 1s(x)gΓ = ΓgΓ ∗ [a]xΓ ∗ 1s(x)Γ = ΓgΓ ∗ [a]xΓ = [a]xΓ ∗ ΓgΓ , so that Cc(A/Γ) ×alg now conclude that the image of the map (56) is the whole Cc(A/Γ) ×alg α G/Γ is spanned by elements of the form axΓ ∗ ΓgΓ. We α G/Γ. The fact that this map is a ∗-homomorphism also follows directly from the commutation relation proved above. 56 3.3 Representation theory In this section we develop the representation theory of crossed products by Hecke pairs. We will introduce the notion of a covariant pre-representation and show that there is a bijective correspondence between covariant pre-representations and representations of the crossed product, in a similar fashion to the theory of crossed products by groups. Recall from Proposition 1.21 that every nondegenerate ∗-representation π : Cc(A/Γ) → B(H ) extends uniquely to a ∗-representation eπ : MB(Cc(A/Γ)) → B(H ) . We will use the notation eπ to denote this extension throughout this section, many times without any reference. Since Cc(X 0/Γ) is a BG∗-algebra (it is spanned by projections) we naturally have Cc(X 0/Γ) ⊆ MB(Cc(A/Γ)). Definition 3.24. Let π be a nondegenerate ∗-representation of Cc(A/Γ) on a Hilbert space H andeπ its unique extension to a ∗-representation of MB(Cc(A/Γ)). Let µ be a unital pre-∗-representation of H(G, Γ) on the inner product space W := π(Cc(A/Γ))H . We say that (π, µ) is a covariant pre-∗-representation if the following equality µ(ΓgΓ)π([a]xΓ)µ(ΓsΓ) = (57) [v]∈ΓsΓ/Γ X[γ]∈E s(x) = X[u]∈Γg−1Γ/Γ u,v ∆(g)N s(x)γ u−1,v L(u−1v) eπ([αu−1γ−1(a)]xγuΓ) µ(Γu−1vΓ)eπ(1s(x)γvΓ) , holds on L(W ), for all g, s ∈ G and x ∈ X. Condition (57) simply says that the pair (π, µ) must preserve the structure of products of the form ΓgΓ ∗ [a]xΓ ∗ ΓsΓ, when expressed in terms of the canonical spanning set of elements of the form [b]yΓ ∗ ΓdΓ ∗ 1s(y)dΓ, as explicitly described in Corollary 3.19. The reader should note the similarity between our definition of a covariant pre-∗-representation and the covariant pairs of an Huef, Kaliszewski and Rae- burn in [7, Definition 1.1]. Their notion of covariant pairs served as a motivation for us and is actually a particular case of our Definition 3.24, as we shall see in the sequel to this article. The operatorseπ([αu−1γ−1(a)]xγuΓ) µ(Γu−1vΓ)eπ(1s(x)γvΓ) in expression (57) are all bounded, as we will now show, and are therefore defined in the whole Hilbert space H . Theorem 3.25. Let π : C(A/Γ) → B(H ) be a nondegenerate ∗-representation and µ : H(G, Γ) → L(W ) a pre-∗-representation on the inner product space W := π(Cc(A/Γ)). Every element of the form is a bounded operator on W and therefore extends uniquely to the whole Hilbert space H . π([a]xΓ)µ(ΓgΓ)eπ(1s(x)gΓ) , 57 We will need some preliminary facts and lemmas in order to prove Theorem 3.25. These auxiliary results will also be useful later in this section. Let π : Cc(A/Γ) → B(H ) be a nondegenerate ∗-representation and eπ its extension to MB(Cc(A/Γ)). For any unit u ∈ X 0 the operatoreπ(1uΓ) ∈ B(H ) is a projection, and thereforeeπ(1uΓ)H is a Hilbert subspace. The fiber (A/Γ)uΓ is a C∗-algebra which we can naturally identify with the ∗-subalgebra {[a]uΓ ∈ Cc(A/Γ) : [a] ∈ (A/Γ)uΓ} ⊆ Cc(A/Γ) , under the identification given by (A/Γ)uΓ ∋ [a] ←→ [a]uΓ ∈ Cc(A/Γ) . The ∗-representationeπ when restricted to (A/Γ)uΓ, under the above identifica- tion, leaves the subspaceeπ(1uΓ)H invariant, because eπ([a]uΓ)eπ(1uΓ)ξ =eπ([a]uΓ)ξ =eπ(1uΓ)eπ([a]uΓ)ξ . The following lemma assures that this restriction is nondegenerate. Lemma 3.26. Let π : Cc(A/Γ) → B(H ) be a nondegenerate ∗-representation and eπ its unique extension to MB(Cc(A/Γ)). The ∗-representation of (A/Γ)uΓ on the Hilbert space eπ(1uΓ)H , as above, is nondegenerate. Proof: Leteπ(1uΓ)ξ be an element ofeπ(1uΓ)H such that for all [a] ∈ (A/Γ)uΓ. We want to prove thateπ(1uΓ)ξ = 0. To see this, let x ∈ X and [b] ∈ (A/Γ)xΓ. We have two alternatives: either s(x)Γ 6= uΓ or s(x)Γ = uΓ. In the first case we see that eπ([a]uΓ)eπ(1uΓ)ξ = 0 , whereas for the second we see that eπ([b]xΓ)eπ(1uΓ)ξ =eπ([b]xΓ · 1uΓ)ξ = 0 , keπ([b]xΓ)eπ(1uΓ)ξk2 = heπ([b]xΓ)eπ(1uΓ)ξ ,eπ([b]xΓ)eπ(1uΓ)ξi = heπ([b∗b]s(x)Γ)eπ(1uΓ)ξ ,eπ(1uΓ)ξi = heπ([b∗b]uΓ)eπ(1uΓ)ξ ,eπ(1uΓ)ξi = 0 , we wanted to prove. by assumption. Thus, in any case we have eπ([b]xΓ)eπ(1uΓ)ξ = 0 for all x ∈ X and [b] ∈ (A/Γ)xΓ. By nondegeneracy of π, this implies that eπ(1uΓ)ξ = 0, as Hilbert space H . We have that π(Cc(A/Γ))H =eπ(Cc(X 0/Γ))H . Lemma 3.27. Let π be a nondegenerate ∗-representation of Cc(A/Γ) on a 58 ment of the form [a]xΓ in Cc(A/Γ) and ξ ∈ H we have π([a]xΓ)ξ = π(1r(x)Γ[a]xΓ)ξ = eπ(1uΓ)ξ = π([c]uΓ)η , have, by the general version of Cohen's factorization theorem ([15, Theorem and ξ ∈ H . We know, by Lemma 3.26, that π gives a nondegenerate ∗- Proof: It is clear that π(Cc(A/Γ))H ⊆eπ(Cc(X 0/Γ))H since for any ele- eπ(1r(x)Γ)π([a]xΓ)ξ. Let us now prove that eπ(Cc(X 0/Γ))H ⊆ π(Cc(A/Γ))H . Let uΓ ∈ X 0/Γ representation of (A/Γ)uΓ on eπ(1uΓ)H . Since (A/Γ)uΓ is a C∗-algebra we 5.2.2]), that there exists [c] ∈ (A/Γ)uΓ and η ∈eπ(1uΓ)H such that which means thateπ(1uΓ)ξ ∈ π(Cc(A/Γ))H . This finishes the proof. Proof of Theorem 3.25: The operator π([a]xΓ)µ(ΓgΓ)eπ(1s(x)gΓ) is clearly is then defined on the spaceeπ(Cc(X 0/Γ))H . Since π([a]xΓ)µ(ΓgΓ)eπ(1s(x)gΓ) = π([a]xΓ)µ(ΓgΓ)eπ(1s(x)gΓ)eπ(1s(x)gΓ) , it follows that the operator π([a]xΓ)µ(ΓgΓ)eπ(1s(x)gΓ) is actually defined in the A similar argument shows thateπ(1s(x)gΓ)µ((ΓgΓ)∗)π([a∗]x−1Γ) is also defined in the whole Hilbert space H and it is easy to see that π([a]xΓ)µ(ΓgΓ)eπ(1s(x)gΓ) is an adjointable operator on H , whose adjoint iseπ(1s(x)gΓ)µ((ΓgΓ)∗)π([a∗]x−1Γ). Proposition 9.1.11]), it follows that π([a]xΓ)µ(ΓgΓ)eπ(1s(x)gΓ) is a bounded op- defined on the inner product space π(Cc(A/Γ))H . By Lemma 3.27 this operator whole Hilbert space H (or in other words, it extends canonically to H ). Since adjointable operators on a Hilbert space are necessarily bounded (see [16, erator. The striking feature that we actually have to consider pre-representations of H(G, Γ), and not just representations, was not present in the theory of crossed products by groups because a group algebra C(G) of a discrete group is al- ways a BG∗-algebra and therefore all of its pre-representations come from true representations (see further Remark 3.31). It will be useful to distinguish between covariant pre-∗-representations and covariant ∗-representations, so we will treat them in separate definitions. As will be discussed below we will see covariant ∗-representations as a particular type of covariant pre-∗-representations. Definition 3.28. Let π be a nondegenerate ∗-representation of Cc(A/Γ) on a Hilbert space H and µ a unital ∗-representation of H(G, Γ) on H . We say that (π, µ) is a covariant ∗-representation if equality (57) holds in B(H ) for all g, s ∈ G and x ∈ X. Lemma 3.29. Let (π, µ) be a covariant ∗-representation on a Hilbert space H . Then µ leaves the subspace W := π(Cc(A/Γ))H invariant. Proof: Consider elements of the form π([a]xΓ)ξ, whose span gives W . Using 59 the fact that µ is unital and the covariance relation (57) we see that µ(ΓgΓ)π([a]xΓ)ξ = = µ(ΓgΓ)π([a]xΓ)µ(Γ)ξ = X[u]∈Γg−1Γ/Γ X[γ]∈E s(x) u,e ∆(g)N s(x)γ u−1,e L(u−1) eπ([αu−1γ−1(a)]xγuΓ) µ(Γu−1Γ)eπ(1s(x)γΓ)ξ . Hence, µ(ΓgΓ)π([a]xΓ)ξ ∈ W , and consequently µ(ΓgΓ) leaves W invariant. This finishes the proof. From a covariant ∗-representation (π, µ) one can obtain canonically a co- variant pre-∗-representation (π, µ), just by restricting µ to the dense subspace W := π(Cc(A/Γ))H (which is an invariant subspace by Lemma 3.29). So we can regard covariant ∗-representations as a special kind of covariant pre-∗- representations: they are exactly those for which µ is normed. As we shall see later in Example 3.41, there are covariant pre-∗-representations which are not covariant ∗-representations, thus in general the latter form a proper subclass of the former. We shall also see examples where they actually coincide. Remark 3.30. Equivalently, one could define covariant (pre-)∗-representation using any other of the equalities in Corollary 3.19 and substituting with the appropriate (pre-)∗-representations. It is easy to see, using completely analo- gous arguments as in the proof of Corollary 3.19 or Theorem 3.17, that all three expressions yield the same result. Remark 3.31. Even though it might not be entirely clear from the start, when Γ is a normal subgroup of G the definition of a covariant pre-representation is nothing but the usual definition of covariant representation of the system (Cc(A/Γ), G/Γ). We recall that a covariant representation of (Cc(A/Γ), G/Γ) is a pair (π, U ) consisting of a nondegenerate ∗-representation π of Cc(A/Γ) and a unitary representation U of G/Γ satisfying the relation π(αgΓ(f )) = UgΓπ(f )Ug−1Γ , for all f ∈ Cc(A/Γ) and gΓ ∈ G/Γ. Now, as it is well known, every unitary rep- resentation U of G/Γ is associated in a canonical way to a unital ∗-representation µ of the group algebra C(G/Γ), so that we can write the covariance condition as π(αgΓ(f )) = µ(gΓ)π(f )µ(g−1Γ). As a consequence we have that for any gΓ, sΓ ∈ G/Γ, x ∈ X and a ∈ Ax: µ(gΓ)π([a]xΓ)µ(sΓ) = π([αg(a)]xg−1Γ)µ(g−1sΓ) . We want to check that covariant representations of the system (Cc(A/Γ), G/Γ) are the same as covariant pre-∗-representations as in Definition 3.24. Given a covariant pre-∗-representation (π, µ) on some Hilbert space H in the sense of Definition 3.24, we have that µ is a pre-∗-representation of C(G/Γ), which is normed since any group algebra of a discrete group is a BG∗-algebra, 60 and thus we can see µ as a true ∗-representation on H . We then have that µ(gΓ)π([a]xΓ)µ(g−1Γ) = µ(ΓgΓ)π([a]xΓ)µ(Γg−1Γ) [v]∈Γg−1Γ/Γ = X[u]∈Γg−1Γ/Γ = X[γ]∈E s(x) = X[γ]∈E s(x) = X[γ]∈E s(x) g−1,g−1 g−1,g−1 g−1,g−1 eπ([αu−1γ−1(a)]xγuΓ) µ(Γu−1vΓ)eπ(1s(x)γvΓ) u,v N s(x)γ ∆(g)N s(x)γ u−1,v L(u−1v) X[γ]∈E s(x) g,g−1 eπ([αg(a)]xg−1Γ) µ(gg−1Γ)eπ(1s(x)g−1Γ) g,g−1 eπ([αg(a)]xg−1Γ · 1s(x)g−1Γ) N s(x)γ N s(x)γ g,g−1 π([αg(a)]xg−1Γ) . It is clear from the normality of Γ that E s(x) and moreover N s(x) g,g−1 = 1, so that g−1,g−1 consists only of the class [e] µ(gΓ)π([a]xΓ)µ(g−1Γ) = π([αg(a)]xg−1Γ) . By linearity it follows that µ(gΓ)π(f )µ(g−1Γ) = π(αgΓ(f )) for any f ∈ Cc(A/Γ). Thus, with U being the unitary representation of G/Γ associated to µ, we see that (π, U ) is covariant representation of the system (Cc(A/Γ), G/Γ). For the other direction, let (π, U ) be a covariant representation of the system (Cc(A/Γ), G/Γ) and let µ be the ∗-representation of C(G/Γ) associated to U , which we restrict to the inner product space π(Cc(A/Γ))H . We want to prove that (π, µ) is a covariant pre-∗-representation in the sense of Definition 3.24. We have µ(gΓ)π([a]xΓ)µ(sΓ) = µ(gΓ)π([a]xΓ)µ(g−1Γ)µ(gsΓ) = π([αg(a)]xg−1Γ)µ(gsΓ) [v]∈ΓsΓ/Γ X[γ]∈E s(x) = X[u]∈Γg−1Γ/Γ u,v ∆(g)N s(x)γ u−1,v L(u−1v) eπ([αu−1γ−1(a)]xγuΓ) µ(Γu−1vΓ)eπ(1s(x)γvΓ) , where the last equality is obtained following analogous computations as those above. Thus, (π, µ) is a covariant pre-∗-representation in the sense of Definition 3.24. The following result makes it clear that some of the relations we have inside the crossed product (see Proposition 3.16) are preserved upon taking covariant pre-∗-representations. This is expected since, as we stated before, we will prove that covariant pre-representations give rise to representations of the crossed product, and this result is the first step in that direction: 61 Proposition 3.32. Let (π, µ) be a covariant pre-∗-representation. The follow- ing two equalities hold: eπ(1r(x)Γ)µ(ΓgΓ)eπ([αg−1 (a)]xgΓ) = eπ([a]xΓ)µ(ΓgΓ)eπ(1s(x)gΓ) . µ(ΓgΓ)eπ([a]xΓ) = X[γ]∈E s(x) g−1,eeπ(1r(x)γg−1Γ)µ(ΓgΓ)eπ([a]xΓ) . (58) (59) Proof: Since (π, µ) is a covariant pre-∗-representation we have µ(ΓgΓ)eπ([a]xΓ) = µ(ΓgΓ)eπ([a]xΓ)µ(Γ) N s(x)γ g,e = X[γ]∈E s(x) = X[γ]∈E s(x) g−1,e eπ([αgγ−1 (a)]xγg−1Γ)µ(ΓgΓ)eπ(1s(x)γΓ) g−1,eeπ([αgγ−1 (a)]xγg−1Γ)µ(ΓgΓ)eπ(1s(x)Γ) , where the last equality comes from the fact that ns(x)γ N s(x)γ g,e = 1. From this it follows that g,e = 1 = ds(x)γ g,e , and thus eπ(1r(x)g−1Γ)µ(ΓgΓ)eπ([a]xΓ) = = X[γ]∈E s(x) = X[γ]∈E s(x) g−1,eeπ(1r(x)g−1Γ)eπ([αgγ−1 (a)]xγg−1Γ)µ(ΓgΓ)eπ(1s(x)Γ) g−1,eeπ(1r(x)g−1Γ · [αgγ−1 (a)]xγg−1Γ)µ(ΓgΓ)eπ(1s(x)Γ) . Now the product 1r(x)g−1Γ · [αgγ−1 (a)]xγg−1Γ is nonzero only when r(x)g−1Γ = r(x)γg−1Γ, from which one readily concludes that r(x)γ ∈ r(x)g−1Γg. Since one trivially has r(x)γ ∈ r(x)Γ we conclude that r(x)γ ∈ r(x)Γ ∩ r(x)g−1Γg , and by the Γ-intersection property we have r(x)γ ∈ r(x)Γg−1 . From Proposition g−1,e. We recall that E r(x) 1.23 this means that [γ] = [e] in E r(x) g−1,e = Sr(x)\Γ/Γg−1 , and since Γg−1 ⊆ Γ we have by Proposition 1.22 that [γ] → [γ] defines a canon- ical bijection between E r(x) . Since the G-action is Γ-good we necessarily have Ss(x) ∩ Γ = Sx ∩ Γ = Sr(x) ∩ Γ, and therefore using Proposition 1.22 one more time we can say that E r(x) g−1,e. Hence, we can say that [γ] = [e] in E s(x) g−1,e and (Sr(x) ∩ Γ)\Γ/Γg−1 g−1,e = E s(x) g−1,e. We conclude that eπ(1r(x)g−1Γ)µ(ΓgΓ)eπ([a]xΓ) = eπ(1r(x)g−1Γ · [αg(a)]xg−1Γ)µ(ΓgΓ)eπ(1s(x)Γ) = eπ([αg(a)]xg−1Γ)µ(ΓgΓ)eπ(1s(x)Γ) . 62 Since the last expression is valid for any x ∈ X and [a] ∈ (A/Γ)xΓ, if we take x to be xg and [a] to be [αg−1 (a)] we obtain the desired equality (58): eπ(1r(x)Γ)µ(ΓgΓ)eπ([αg−1 (a)]xgΓ) =eπ([a]xΓ)µ(ΓgΓ)eπ(1s(x)gΓ) . Let us now prove equality (59). Using the equality in beginning of this proof and equality (58) which we have just proven, we get precisely µ(ΓgΓ)eπ([a]xΓ) = X[γ]∈E s(x) = X[γ]∈E s(x) = X[γ]∈E s(x) = X[γ]∈E s(x) g−1,eeπ([αgγ−1 (a)]xγg−1Γ)µ(ΓgΓ)eπ(1s(x)Γ) g−1,eeπ([αgγ−1 (a)]xγg−1Γ)µ(ΓgΓ)eπ(1s(xγg−1)gΓ) g−1,eeπ(1r(x)γg−1Γ)µ(ΓgΓ)eπ([αγ−1 (a)]xγΓ) g−1,eeπ(1r(x)γg−1Γ)µ(ΓgΓ)eπ([a]xΓ) . This finishes the proof. The passage from a covariant pre-representation (π, µ) to a representation α G/Γ is done via the so-called integrated form π × µ, which we of Cc(A/Γ) ×alg now describe: Definition 3.33. Let (π, µ) be a covariant pre-∗-representation on a Hilbert space H . We define the integrated form of (π, µ) as the function π × µ : Cc(A/Γ) ×alg α G/Γ → B(H ) defined by [π × µ](f ) := X[g]∈Γ\G/Γ XxΓg∈X/Γgeπ(cid:16)(cid:2)f (gΓ)(x)(cid:3)xΓ(cid:17) µ(ΓgΓ)eπ(1s(x)gΓ) . Remark 3.34. For f of the form f = axΓ ∗ ΓgΓ ∗ 1s(x)gΓ we have Moreover, from equality (58), for f ′ of the form f ′ = 1r(x)Γ ∗ ΓgΓ ∗ [αg−1 (a)]xgΓ we have [π × µ](f ) =eπ([a]xΓ) µ(ΓgΓ)eπ(1s(x)gΓ) . [π × µ](f ′) =eπ(1r(x)Γ) µ(ΓgΓ)eπ([αg−1 (a)]xgΓ) . Proposition 3.35. The integrated form π×µ of a covariant pre-∗-representation (π, µ) is a well-defined nondegenerate ∗-representation. Proof: First we need to check that the expression that defines [π × µ](f ) α G/Γ is well-defined. This is proven in an entirely for a given f ∈ Cc(A/Γ) ×alg 63 analogous way as in the proof that the expression (44) in Proposition 3.14 is well-defined. Secondly, we need to show that [π × µ](f ) makes sense as an element of B(H ). From Theorem 3.25 we have that eπ(cid:16)(cid:2)f (gΓ)(x)(cid:3)xΓ(cid:17)µ(ΓgΓ)eπ(1s(x)gΓ) ∈ B(W ) , thus, it follows that [π × µ](f ) ∈ B(W ), and therefore [π × µ](f ) admits a unique extension to B(H ). Now, it is obvious that π × µ is a linear transformation. Let us check that it preserves the involution. It is then enough to check it for elements of the form f = [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ. Since (π, µ) is a covariant pre-∗-representation we have, by Propositions 3.32 and 3.16, (cid:0)[π × µ](f )(cid:1)∗ = ∆(g)eπ(1s(x)gΓ) µ(Γg−1Γ)eπ([a∗]x−1Γ) = ∆(g)eπ(1r(x−1)gΓ) µ(Γg−1Γ)eπ([a∗]x−1gg−1Γ) = ∆(g)eπ([αg−1 (a∗)]x−1gΓ) µ(Γg−1Γ)eπ(1s(x−1)gg−1Γ) = ∆(g)eπ([αg−1 (a∗)]x−1gΓ) µ(Γg−1Γ)eπ(1s(x−1)Γ) = [π × µ] (∆(g) [αg−1 (a∗)]x−1gΓ ∗ Γg−1Γ ∗ 1s(x−1)Γ) = [π × µ] (f ∗) . Let us now prove that π × µ preserves products. We will start by proving that [π × µ](f1 ∗ f2) = [π × µ](f1) [π × µ](f2) , (60) for f1 := [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ and f2 := [b]yΓ ∗ ΓsΓ ∗ 1s(y)sΓ. Let us compute the expression on the left side of (60). First, we notice that for the product f1 ∗ f2 to be non-zero one must have r(y) ∈ s(x)gΓ, and in this case we obtain f1 ∗ f2 = [a]xΓ ∗ ΓgΓ ∗ [b]yΓ ∗ ΓsΓ ∗ 1s(y)sΓ which by Corollary 3.19 gives ∆(g)N s(y)γ u−1,v L(u−1v) = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ [γ]∈E s(y) u,v [a]xΓ ∗ [αu−1γ−1(b)]yγuΓ ∗ Γu−1vΓ ∗ 1s(y)γvΓ ∗ 1s(y)sΓ ∆(g)N s(y)γ u−1,v L(u−1v) = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ [γ]∈E s(y) u,v s(y)sΓ=s(y)γvΓ ∆(g)N s(y)γ u−1,v L(u−1v) = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ [γ]∈E s(y) u,v s(y)sΓ=s(y)γvΓ [a]xΓ ∗ [αu−1γ−1(b)]yγuΓ ∗ Γu−1vΓ ∗ 1s(y)γvΓ [a]xΓ ∗ [αu−1γ−1(b)]yγuΓ ∗ Γu−1vΓ ∗ 1s(yγu)u−1vΓ 64 The product [a]xΓ∗[αu−1γ−1(b)]yγuΓ is always either zero or of the form [c](xθ)(yγu)Γ, for some θ ∈ Γ and c ∈ A(xθ)(yγu). The point is that s(cid:0)(xθ)(yγu)(cid:1) = s(yγu), so that each non-zero summand in the last sum above is actually of the form [c]zΓ ∗ ΓdΓ ∗ 1s(z)dΓ , for appropriate [c] ∈ (A/Γ)zΓ, z ∈ X and d ∈ G. Thus, by linearity of π × µ and Remark 3.34 we obtain [π × µ](f1 ∗ f2) = ∆(g)N s(y)γ u−1,v L(u−1v) = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ [γ]∈E s(y) u,v s(y)sΓ=s(y)γvΓ eπ([a]xΓ · [αu−1γ−1(b)]yγuΓ) µ(Γu−1vΓ)eπ(1s(y)γvΓ) . Let us now compute the expression on the right side of (60). We have [π × µ](f1) [π × µ](f2) = eπ([a]xΓ) µ(ΓgΓ)eπ(1s(x)gΓ)eπ([b]yΓ) µ(ΓsΓ)eπ(1s(y)sΓ) . For 1s(x)gΓ · [b]yΓ to be non-zero we must have r(y) ∈ s(x)gΓ, and in this case we obtain, using the definition of a covariant pre-∗-representation, [π × µ](f1) [π × µ](f2) = = eπ([a]xΓ) µ(ΓgΓ)eπ([b]yΓ) µ(ΓsΓ)eπ(1s(y)sΓ) = X[u]∈Γg−1Γ/Γ ∆(g)N s(y)γ u−1,v L(u−1v) eπ([a]xΓ[αu−1γ−1(b)]yγuΓ)µ(Γu−1vΓ)eπ(1s(y)γvΓ)eπ(1s(y)sΓ) L(u−1v) eπ([a]xΓ · [αu−1γ−1(b)]yγuΓ) µ(Γu−1vΓ)eπ(1s(y)γvΓ) . ∆(g)N s(y)γ u−1,v [v]∈ΓsΓ/Γ [γ]∈E s(y) u,v = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ [γ]∈E s(y) u,v s(y)sΓ=s(y)γvΓ Hence, we have proven equality (60) for the special case of f1 and f2 being f1 := [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ and f2 := [b]yΓ ∗ ΓsΓ ∗ 1s(y)sΓ. Using this we will now show that equality (60) holds for any f1, f2 ∈ Cc(A/Γ) ×alg α G/Γ. In fact, by Proposition 3.14, f1 and f2 can be written as sums f1 =Xi vi , f1 =Xj wj , where each vi and wj is of the form [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ, for some gΓ ∈ G/Γ, x ∈ X and a ∈ Ax. Since π × µ is a linear mapping we have [π × µ](f1 ∗ f2) = [π × µ](cid:16)(cid:0)Xi = [π × µ](cid:16)Xi,j = Xi,j wj(cid:1)(cid:17) vi(cid:1) ∗(cid:0)Xj vi ∗ wj(cid:17) [π × µ](vi ∗ wj) , 65 and by the special case of equality (60) we have just proven we get [π × µ](vi)[π × µ](wj ) [π × µ](f1 ∗ f2) = Xi,j = (cid:16)Xi = [π × µ](cid:0)Xi [π × µ](vi)(cid:17)(cid:16)Xj [π × µ](wj)(cid:17) wj(cid:1) vi(cid:1)[π × µ](cid:0)Xj = [π × µ](f1)[π × µ](f2) . Hence, π × µ is a ∗-representation. To finish the proof we now only need to show that π × µ is nondegenerate. The restriction of π × µ to the ∗-subalgebra Cc(A/Γ) is precisely the representation π. Since π is assumed to be nondegen- erate it follows that π × µ must be nondegenerate as well. The next result shows how from a representation of the crossed product one can naturally form a covariant pre-representation. Proposition 3.36. Let Φ : Cc(A/Γ) ×alg ∗-representation. Consider the pair (Φ, ωΦ) defined by α G/Γ → B(H ) be a nondegenerate • Φ is the restriction of Φ to Cc(A/Γ). 1.18) of M (Cc(A/Γ)×alg • Let eΦ be the extension of Φ to a pre-∗-representation (via Proposition G/Γ)H . We define ωΦ to be the restriction of eΦ to H(G, Γ). α G/Γ) on the inner product space Φ(Cc(A/Γ)×alg α The pair (Φ, ωΦ) is a covariant pre-∗-representation. We will need some preliminary lemmas in order to prove Proposition 3.36. Lemma 3.37. If Φ : Cc(A/Γ) ×alg representation, then its restriction to Cc(A/Γ) is also nondegenerate. α G/Γ → B(H ) is a nondegenerate ∗- Proof: Let ξ ∈ H be such that Φ(Cc(A/Γ)) ξ = {0}. We want to show that ξ = 0. Since Φ is nondegenerate, it is then enough to prove that Φ(Cc(A/Γ)×alg α G/Γ) ξ = {0}. Thus, by virtue of Proposition 3.16, it suffices to prove that Φ(1r(x)Γ ∗ ΓgΓ ∗ [αg−1 (a)]xgΓ)ξ = 0 for all g ∈ G, x ∈ X, a ∈ Ax. We have kΦ(1r(x)Γ ∗ ΓgΓ ∗ [αg−1 (a)]xgΓ)ξk2 = = ∆(g)hΦ([αg−1 (a∗)]x−1gΓ ∗ Γg−1Γ ∗ 1r(x)Γ ∗ 1r(x)Γ ∗ ΓgΓ ∗ [αg−1 (a)]xgΓ)ξ , ξi = ∆(g)hΦ([αg−1 (a∗)]x−1gΓ)Φ(Γg−1Γ ∗ 1r(x)Γ ∗ ΓgΓ ∗ [αg−1 (a)]xgΓ)ξ , ξi = ∆(g)hΦ(Γg−1Γ ∗ 1r(x)Γ ∗ ΓgΓ ∗ [αg−1 (a)]xgΓ)ξ , Φ([αg−1 (a)]xgΓ)ξi = 0 . Hence ξ = 0 and therefore Φ restricted to Cc(A/Γ) is nondegenerate. 66 Lemma 3.38. Let Φ : Cc(A/Γ) ×alg representation andeΦ its unique extension to MB(Cc(A/Γ)×alg sition 1.21). Let Φ be the restriction of Φ to Cc(A/Γ) andfΦ its unique exten- sion to MB(Cc(A/Γ)). We have that α G/Γ → B(H ) be a nondegenerate ∗- α G/Γ) (via Propo- eΦ(f ) =fΦ(f ) , for all f ∈ Cc(X 0/Γ). In other words, the two ∗-representations eΦ and fΦ are the same in Cc(X 0/Γ). Proof: By Lemma 3.37 the subspace Φ(Cc(A/Γ))H is dense in H , so that it is enough to check that eΦ(f )Φ(f2)ξ =fΦ(f )Φ(f2)ξ, for all f2 ∈ Cc(A/Γ) and ξ ∈ H . By definition of the extension eΦ (see Proposition 1.21) we have where f ∗ f2 is the product of f and f2, which lies inside Cc(A/Γ) ×alg α G/Γ. Since both f and f2 are elements of B(A, G, Γ) we see the product f ∗ f2 as taking place in B(A, G, Γ). By definition of the embeddings of Cc(X 0/Γ) and Cc(A/Γ) in B(A, G, Γ) we have that f ∗ f2 is nothing but the element f · f2, where the product is just the product of f and f2 inside M (Cc(A)). As we observed in Section 2.3, this product is exactly same as the product of f and f2 in M (Cc(A/Γ)). Thus, the following computation makes sense: eΦ(f )Φ(f2)ξ = Φ(f ∗ f2)ξ , eΦ(f )Φ(f2)ξ = Φ(f ∗ f2)ξ = Φ(f · f2)ξ = Φ(f · f2)ξ = fΦ(f )Φ(f2)ξ . This finishes the proof. Lemma 3.39. Let Φ : Cc(A/Γ) ×alg representation. We have that α G/Γ → B(H ) be a nondegenerate ∗- Φ(Cc(A/Γ))H = Φ(Cc(A/Γ) ×alg α G/Γ)H . Proof: The inclusion Φ(Cc(A/Γ))H ⊆ Φ(Cc(A/Γ)×alg α G/Γ)H is obvious. To check the converse inclusion it is enough to prove that Φ([a]xΓ ∗ ΓgΓ ∗ 1s(x)Γ)ξ ∈ Φ(Cc(A/Γ))H , for all x ∈ X, a ∈ Ax, g ∈ G and ξ ∈ H . Let eΦ : MB(Cc(A/Γ) ×alg α G/Γ) → B(H ) be the unique extension of Φ to a ∗-representation of MB(Cc(A/Γ) ×alg α G/Γ), as in Proposition 1.21. We then get Φ([a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ)ξ = Φ(1r(x)Γ ∗ [a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ)ξ Denoting by Φ the restriction of Φ to Cc(A/Γ) we have, by Lemma 3.38, that = eΦ(1r(x)Γ)Φ([a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ)ξ . = fΦ(1r(x)Γ)Φ([a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ)ξ , 67 i.e. Φ([a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ)ξ ∈fΦ(Cc(X/Γ))H . By Lemma 3.27 it then follows that Φ([a]xΓ ∗ ΓgΓ ∗ 1s(x)gΓ)ξ ∈ Φ(Cc(A/Γ))H . Proof of Proposition 3.36: First of all, by Lemma 3.37, Φ is indeed a nondegenerate ∗-representation of Cc(A/Γ). Secondly, from Lemma 3.39, we have Φ(Cc(A/Γ))H = Φ(Cc(A/Γ) ×alg α G/Γ)H . Thus, ωΦ is a pre-∗-representation of H(G, Γ) on W := Φ(Cc(A/Γ))H . We now only need to check covariance. We have ωΦ(ΓgΓ)Φ([a]xΓ)ωΦ(ΓsΓ) = u,v ∆(g)N s(x)γ u−1,v L(u−1v) ∆(g)N s(x)γ u−1,v L(u−1v) [αu−1γ−1(a)]xγuΓ ∗ Γu−1vΓ ∗ 1s(x)γvΓ(cid:17) = eΦ(ΓgΓ)eΦ([a]xΓ)eΦ(ΓsΓ) = eΦ(ΓgΓ ∗ [a]xΓ ∗ ΓsΓ) = eΦ(cid:16) X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ X[γ]∈E s(x) = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ X[γ]∈E s(x) eΦ([αu−1γ−1(a)]xγuΓ)eΦ(Γu−1vΓ)eΦ(1s(x)γvΓ) . Denoting byfΦ the unique extension of Φ to MB(Cc(A/Γ)) we have, by Lemma [v]∈ΓsΓ/Γ X[γ]∈E s(x) = X[u]∈Γg−1Γ/Γ L(u−1v) fΦ([αu−1γ−1(a)]xγuΓ)ωΦ(Γu−1vΓ)fΦ(1s(x)γvΓ) . u,v u,v 3.38, that ∆(g)N s(x)γ u−1,v This finishes the proof. Theorem 3.40. There is a bijective correspondence between nondegenerate ∗- α G/Γ and covariant pre-∗-representations. This representations of Cc(A/Γ) ×alg bijection is given by (π, ω) 7−→ π × µ, with inverse given by Φ 7−→ (Φ, ωΦ). Proof: We have to prove that the composition of these maps, in both orders, is the identity. Let (π, µ) be a covariant pre-∗-representation and π × µ its integrated form. We want to show that (cid:0)(π × µ), ωπ×µ(cid:1) = (π, µ) . By definition of the integrated form we readily have (π × µ) = π. This also implies, via Lemma 3.37, that the inner product spaces on which µ and ωπ×µ are defined are actually the same. Thus, it remains to be checked that ωπ×µ = µ. Let π([a]xΓ)ξ be one of the generators of π(Cc(A/Γ))H . We have ωπ×µ(ΓgΓ) π([a]xΓ)ξ = = ^[π × µ](ΓgΓ) π([a]xΓ)ξ = [π × µ](ΓgΓ ∗ [a]xΓ)ξ , 68 and using Proposition 3.16, Remark 3.34 and Proposition 3.32 we obtain = [π × µ](cid:0) X[γ]∈E s(x) = X[γ]∈E s(x) 1r(x)γgΓ ∗ ΓgΓ ∗ [a]xΓ(cid:1)ξ g−1,eeπ(1r(x)γgΓ)µ(ΓgΓ)eπ([a]xΓ)ξ g−1,e = µ(ΓgΓ) π([a]xΓ)ξ Hence, we conclude that ωπ×µ = µ. Now let Φ be a ∗-representation of Cc(A/Γ) ×alg α G/Γ and (Φ, ωΦ) its cor- responding covariant pre-∗-representation. We want to prove that Φ × ωΦ = Φ . Let 1r(x)Γ ∗ ΓgΓ ∗ [αg−1 (a)]xgΓ be one of the spanning elements of Cc(A/Γ) ×alg α G/Γ and ξ ∈ H . We have [Φ × ωΦ] (1r(x)Γ ∗ ΓgΓ ∗ [αg−1 (a)]xgΓ) ξ =fΦ(1r(x)Γ)ωΦ(ΓgΓ)fΦ([αg−1 (a)]xgΓ) ξ , which by Lemma 3.38 gives that = eΦ(1r(x)Γ)eΦ(ΓgΓ)eΦ([αg−1 (a)]xgΓ) ξ = Φ(1r(x)Γ ∗ ΓgΓ ∗ [αg−1 (a)]xgΓ) ξ . Thus, Φ × ωΦ = Φ. 3.4 More on covariant pre-∗-representations In the previous section we introduced the notion of covariant pre-∗-representations of Cc(A/Γ) ×alg α G/Γ (Definition 3.24) and a particular instance of these which we called covariant ∗-representations (Definition 3.28). In this section we will see that the class of covariant pre-∗-representations is in general strictly larger than the class of covariant ∗-representations. It is thus unavoidable, in general, to consider pre-representations of the Hecke alge- bra in the representation theory of crossed products by Hecke pairs. We shall also see, nevertheless, that in many interesting situations every covariant pre-∗- representation is actually a covariant ∗-representation. Example 3.41. Let (G, Γ) be a Hecke pair such that its corresponding Hecke algebra H(G, Γ) does not have an enveloping C∗-algebra (it is well known that such pairs exist, as for example (G, Γ) = (SL2(Qp), SL2(Zp)) as discussed in [6]). The fact that the Hecke algebra does not have an enveloping C∗-algebra implies that there is a sequence of ∗-representations {µn}n∈N of H(G, Γ) on Hilbert spaces {Hn}n∈N and an element f ∈ H(G, Γ) such that kµn(f )k → ∞. Hn and µ : H(G, Γ) → L(V ) the diagonal pre-∗-representation Let V be the inner product space V :=Ln∈N µ :=Mn∈N µn , 69 which of course is not normed. Let X = {x1, x2, . . . } be an infinite countable set, with the trivial groupoid structure, i.e. X is just a set. We consider the Fell bundle A over X whose fibers are the complex numbers, i.e. Ax = C for every x ∈ X, and we consider the trivial action of G on A, i.e. the action that fixes every element of A. Thus, the action is Γ-good and has the Γ-intersection property. We also have that Cc(A/Γ) = Cc(X) = Cc(X 0/Γ) . Let π : Cc(X) → B(V ) be the ∗-representation on the Hilbert space com- pletion V of V such that π(1xn) is the projection onto the subspace Hn. We claim that (π, µ) is a covariant pre-∗-representation of Cc(X) ×alg α G/Γ. To see this, first we notice that π is obviously nondegenerate and moreover π(Cc(X))V = V , which is the inner product space where µ is defined. Next we notice that for every xn ∈ X and g ∈ G, the operators π(1xn) and µ(ΓgΓ) commute. Moreover, we have π(1xn )µ(ΓgΓ)π(1xn) = µn(ΓgΓ) , on the subspace Hn. Also we have µ(ΓgΓ)π(1xn)µ(ΓsΓ) = = µ(ΓgΓ)µ(ΓsΓ)π(1xn ) [v]∈ΓsΓ/Γ = X[u]∈Γg−1Γ/Γ = X[u]∈Γg−1Γ/Γ = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ [v]∈ΓsΓ/Γ X[γ]∈Exn u,v ∆(g) L(u−1v) ∆(g) L(u−1v) µ(Γu−1vΓ) π(1xn ) π(1xn )µ(Γu−1vΓ) π(1xn ) ∆(g)N xnγ u−1,v L(u−1v) π(1xnγu)µ(Γu−1vΓ) π(1xnγv) , where the last equality comes from the fact that since Sxn = G we must have that Exn u−1,v = 1 and also that 1xnγu = 1xn = 1xnγv. u,v consists only of the class [e], N xn So we have established that (π, µ) is indeed a covariant pre-∗-representation. Nevertheless, µ is not normed, so that (π, µ) is not a covariant ∗-representation. It is worth noting that here we are in the conditions of Example 3.22, so that Cc(X) ×alg α G/Γ ∼= Cc(X) ⊙ H(G, Γ). Example 3.41 shows that there can be more covariant pre-∗-representations than covariant ∗-representations. Nevertheless, the two classes actually coincide in many cases. One such case is when Cc(A/Γ) has an identity element: Proposition 3.42. If the crossed product Cc(A/Γ) ×alg α G/Γ has an identity element (equivalently, if Cc(A/Γ) has an identity element), then every covariant pre-∗-representation is a covariant ∗-representation. 70 Proof: Let us assume that Cc(A/Γ)×alg alently, Cc(A/Γ) has an identity element). α G/Γ has an identity element (equiv- Let (π, µ) be a covariant pre-∗-representation. As it was shown in Theo- rem 3.40, the integrated form π × µ is a ∗-representation of Cc(A/Γ) ×alg α G/Γ such that µ = ωπ×µ, where ωπ×µ is the pre-∗-representation which is obtained by extending π × µ to the multiplier algebra M (Cc(A/Γ) ×alg α G/Γ) and then restricting it to H(G, Γ). Since the crossed product Cc(A/Γ) ×alg α G/Γ has an identity element, we have M (Cc(A/Γ) ×alg α G/Γ) = Cc(A/Γ) ×alg α G/Γ , and therefore ωπ×µ is just the restriction of π × µ to the the Hecke algebra H(G, Γ). Hence, µ = ωπ×µ is a true ∗-representation. Another interesting situation where covariant pre-∗-representations coincide with covariant ∗-representations is when H(G, Γ) is a BG∗-algebra. This is known to be the case for many classes of Hecke pairs (G, Γ) as we proved in [14]. Actually, most of the classes of Hecke pairs for which a full Hecke C∗- algebra is known to exist are such that H(G, Γ) is BG∗-algebra. Proposition 3.43. If H(G, Γ) is a BG∗-algebra, then every covariant pre-∗- representation is a covariant ∗-representation. Proof: If H(G, Γ) is a BG∗-algebra, then every pre-∗-representation of H(G, Γ) is automatically normed and hence arises from a true ∗-representation. 3.5 Crossed product in the case of free actions In this section we will see that when the associated G-action on X is free the expressions for the products of the form ΓgΓ∗ [a]xΓ ∗ ΓsΓ, described in Corollary 3.19, as well as the definition of a covariant pre-∗-representation become much simpler and even more similar to the notion of covariant pairs of [7]. Theorem 3.44. If the action of G on X is free, then ΓgΓ ∗ 1yΓ ∗ ΓsΓ = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ 1yuΓ ∗ Γu−1vΓ ∗ 1yvΓ (61) and similarly, ΓgΓ ∗ [a]xΓ ∗ ΓsΓ = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ [αu−1 (a)]xuΓ ∗ Γu−1vΓ ∗ 1s(x)vΓ . (62) 71 Lemma 3.45. If the action of G on X is free, then ny w,v = 1 and dy w,v = [Γwv : Γwv ∩ wΓw−1] . Proof: We have ny w,v = (cid:8)[r] ∈ ΓwΓ/Γ : r−1wvΓ ⊆ ΓvΓ and yw−1 ∈ yΓr−1(cid:9) = (cid:8)[r] ∈ ΓwΓ/Γ : r−1wvΓ ⊆ ΓvΓ and w−1 ∈ Γr−1(cid:9) = (cid:8)[r] ∈ ΓwΓ/Γ : r−1wvΓ ⊆ ΓvΓ and rΓ = wΓ(cid:9) = {wΓ} . Thus, ny w,v = 1. Also, dy w,v = (cid:8)[r] ∈ ΓwΓ/Γ : r−1wvΓ ⊆ ΓvΓ and yw−1 ∈ yΓr−1Γwv(cid:9) = (cid:8)[r] ∈ ΓwΓ/Γ : r−1wvΓ ⊆ ΓvΓ and w−1 ∈ Γr−1Γwv(cid:9) . Now we notice that in the above set the condition r−1wvΓ ⊆ ΓvΓ is automat- ically satisfied from the second condition w−1 ∈ Γr−1Γwv, because the latter means that r−1 = θ1w−1θ2 for some θ1 ∈ Γ and θ2 ∈ Γwv. Thus, we get Thus, we obtain dy Proof of Theorem 3.44: We have seen in Theorem 3.17 that dy = ΓwvwΓ/Γ . w,v = (cid:8)[r] ∈ ΓwΓ/Γ : w−1 ∈ Γr−1Γwv(cid:9) = (cid:8)[r] ∈ ΓwΓ/Γ : r ∈ ΓwvwΓ(cid:9) w,v =(cid:12)(cid:12)ΓwvwΓ/Γ(cid:12)(cid:12) = [Γwv : Γwv ∩ wΓw−1]. [v]∈ΓsΓ/Γ X[γ]∈Ey ∆(g)N yγ L(u−1v) u−1,v ΓgΓ ∗ 1yΓ ∗ ΓsΓ = X[u]∈Γg−1Γ/Γ u,v (cid:0)1yγuΓ ∗ Γu−1vΓ ∗ 1yγvΓ(cid:1) It follows from Lemma 3.45 that N yγ u−1,v = 1 [Γu−1v : Γu−1v ∩ u−1Γu] . Moreover, freeness of the action also implies that Ey u,v = Sy\Γ/(vΓv−1 ∩ uΓu−1) = Γ/(vΓv−1 ∩ uΓu−1) . Now, we have the following well-defined bijective correspondence Γ/(Γu ∩ Γv) −→ Γ/(vΓv−1 ∩ uΓu−1) [θ] 7→ [θ] , given by Proposition 1.22. Note that Γu ∩ Γv is simply the subgroup uΓu−1 ∩ vΓv−1 ∩ Γ, but in the following we will take preference on the notation Γu ∩ Γv for being shorter. 72 Consider now the action of Γ on G/Γ×G/Γ by left multiplication and denote by Oh1,h2 the orbit of the element (h1Γ, h2Γ) ∈ G/Γ × G/Γ. It is easy to see that the map Γ/(Γh1 ∩ Γh2 ) −→ Oh1,h2 [θ] 7→ (θh1Γ, θh2Γ) is also well-defined and is a bijection. We will denote by C the set of all orbits contained in Γg−1Γ/Γ × ΓsΓ/Γ (note that this set is Γ-invariant, so that it is a union of orbits). We then have ΓgΓ ∗ 1yΓ ∗ ΓsΓ = [v]∈ΓsΓ/Γ X[γ]∈Ey u,v ∆(g)N yγ L(u−1v) u−1,v (cid:0)1yγuΓ ∗ Γu−1vΓ ∗ 1yγvΓ(cid:1) = X[u]∈Γg−1Γ/Γ = X[u]∈Γg−1Γ/Γ = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ [v]∈ΓsΓ/Γ X[γ]∈Γ/(Γu∩Γv) X[γ]∈Γ/(Γu∩Γv) ∆(g)N yγ L(u−1v) u−1,v (cid:0)1yγuΓ ∗ Γu−1vΓ ∗ 1yγvΓ(cid:1) ∆(g)N y u−1γ−1,γv L(u−1γ−1γv) (cid:0)1yγuΓ ∗ Γu−1γ−1γvΓ ∗ 1yγvΓ(cid:1) where the last equality comes from the fact that N yγ u−1γ−1,γv, which is a consequence of Lemma 3.18 iii), or simply by Lemma 3.45. Using now the bijection between Γ/(Γu ∩ Γv) and the orbit space Ou,v as described above, we obtain u−1,v = N y = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ X([r],[t])∈Ou,v = XO∈C X([u],[v])∈O X([r],[t])∈Ou,v = XO∈C X([u],[v])∈O X([r],[t])∈O = XO∈C X([r],[t])∈O #O ∆(g)N y L(r−1t) r−1,t (cid:0)1yrΓ ∗ Γr−1tΓ ∗ 1ytΓ(cid:1) (cid:0)1yrΓ ∗ Γr−1tΓ ∗ 1ytΓ(cid:1) (cid:0)1yrΓ ∗ Γr−1tΓ ∗ 1ytΓ(cid:1) (cid:0)1yrΓ ∗ Γr−1tΓ ∗ 1ytΓ(cid:1) , r−1,t ∆(g)N y r−1,t L(r−1t) ∆(g)N y L(r−1t) ∆(g)N y L(r−1t) r−1,t where #O denotes the total number of elements of the given orbit O. Changing the names of the variables (r to u and t to v) we get = XO∈C X([u],[v])∈O = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ #O ∆(g)N y L(u−1v) u−1,v #Ou,v ∆(g)N y L(u−1v) u−1,v (cid:0)1yuΓ ∗ Γu−1vΓ ∗ 1yvΓ(cid:1) (cid:0)1yuΓ ∗ Γu−1vΓ ∗ 1yvΓ(cid:1) . 73 We are now going to prove that the coefficients satisfy #Ou,v ∆(g)N y L(u−1v) u−1,v = 1 . This follows from the following computation: #Ou,v L(u−1v) N y u−1,v ∆(g) = = = = = = [Γ : Γu ∩ Γv] [Γ : Γu−1v] [Γ : Γu ∩ Γv] [Γ : Γu−1 · ] 1 [Γu−1v : Γu−1v ∩ u−1Γu] [Γ : Γu−1 [Γ : Γu] ] · [Γ : Γu−1v ∩ u−1Γu][Γ : Γu] [Γu : Γu ∩ Γv] [Γ : Γu−1 [Γ : Γu−1v ∩ u−1Γu] ] [Γu : Γu ∩ Γv] [uΓu−1 : Γu] [Γ : Γu−1v ∩ u−1Γu] [uΓu−1 : Γu ∩ Γv] [Γ : Γu−1v ∩ u−1Γu] [uΓu−1 : Γu ∩ Γv] [uΓu−1 : Γu ∩ Γv] = 1 . This finishes the first claim of the theorem. The second claim, concerning the product ΓgΓ ∗ [a]xΓ ∗ ΓsΓ, is proven in a completely similar fashion. Proposition 3.46. Let π : Cc(A/Γ) → B(H ) be a nondegenerate ∗-representation, µ : H(G, Γ) → L(π(Cc(A/Γ)H ) a unital pre-∗-representation, and let us as- sume that the associated G-action on X is free. The pair (π, µ) is a covariant pre-∗-representation if and only if the following equality µ(ΓgΓ)π([a]xΓ)µ(ΓsΓ) = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ π([αu−1 (a)]xuΓ)µ(Γu−1vΓ)eπ(1s(x)vΓ) . (63) holds for all g, s ∈ G, x ∈ X and a ∈ Ax. Proof: (=⇒) Assume that (π, µ) is a covariant pre-∗-representation. Then we have µ(ΓgΓ)π([a]xΓ)µ(ΓsΓ) = [π × µ](ΓgΓ ∗ [a]xΓ ∗ ΓsΓ) = [π × µ](cid:16) X[u]∈Γg−1Γ/Γ = X[u]∈Γg−1Γ/Γ [αu−1(a)]xuΓ ∗ Γu−1vΓ ∗ 1s(x)vΓ(cid:17) π([αu−1 (a)]xuΓ)µ(Γu−1vΓ)eπ(1s(x)vΓ) . [v]∈ΓsΓ/Γ [v]∈ΓsΓ/Γ 74 (⇐=) In order to prove equality (57) one just needs to show that ∆(g)N s(x)γ u−1,v L(u−1v) [v]∈ΓsΓ/Γ X[γ]∈E s(x) X[u]∈Γg−1Γ/Γ = X[u]∈Γg−1Γ/Γ [v]∈ΓsΓ/Γ eπ([αu−1 (a)]xuΓ)µ(Γu−1vΓ)eπ(1s(x)vΓ) , u,v eπ([αu−1 (a)]xγuΓ) µ(Γu−1vΓ)eπ(1s(x)γvΓ) and this is proven in a completely analogous way as in the proof of Theorem 3.44. References [1] P. Ara, M. Mathieu, Local Multipliers of C-Algebras, Springer Monogr. Math., Springer-Verlag, Berlin (2003). [2] A. Connes, M. Marcolli, From physics to number theory via noncommuta- tive geometry, Part I: Quatum statistical mechanics of Q-lattices, in "Fron- tiers in Number Theory, Physics, and Geometry, I", Springer Verlag (2006), 269-350. [3] S. Echterhoff, S. Kaliszewski, I. Raeburn, Crossed products by dual coac- tions of groups and homogeneous spaces, J. Operator Theory 39 (1998), 151-176. [4] S. Echterhoff, J. Quigg, Full duality for coactions of discrete groups, Math. Scand. 90 (2002), 267-288. [5] S. Echterhoff, J. Quigg, Induced coactions of discrete groups on C∗-algebras, Canad. J. Math. 51 (1999), 745-770. [6] R. W. Hall, Hecke C∗-algebras, Ph.D. thesis, The Pennsylvania State Uni- versity, December 1999. [7] A. an Huef, S. Kaliszewski, I. Raeburn, Covariant representations of Hecke algebras and imprimitivity for crossed products by homogeneous spaces, J. Pure Appl. Algebra 212 (2008), 2344-2357. [8] S. Kaliszewski, M. B. Landstad and J. Quigg, Hecke C∗-algebras, Schlicht- ing completions, and Morita equivalence, Proc. Edinb. Math. Soc. (Series 2) 51 (2008), no. 3, 657-695. [9] S. Kaliszewski, P. S. Muhly, J. Quigg, D. P. Williams, Coactions and Fell bundles, New York J. Math 16 (2010), 315-359. [10] A. Krieg, Hecke algebras, Mem. Amer. Math. Soc. 87 (1990), No. 435. [11] A. Kumjian, Fell bundles over groupoids, Proc. Amer. Math. Soc. 126 (1998), 1115-1125. 75 [12] M. Laca, N. S. Larsen, S. Neshveyev, Phase transition in the Connes- Marcolli GL2-system, J. Noncommut. Geom. 1 (2007), 397-430. [13] R. Palma, C∗-completions of Hecke algebras and crossed products by Hecke pairs, Ph.D. thesis, University of Oslo (2012). [14] R. Palma, On enveloping C∗-algebras of Hecke algebras, pre-print (2012), arXiv: 1201.0709 . [15] T. W. Palmer, Banach algebras and the general theory of ∗-algebras. Vol. I Algebras and Banach Algebras, Encyclopedia of Mathematics and its Ap- plications, 49. Cambridge University Press, Cambridge (1994). [16] T. W. Palmer, Banach algebras and the general theory of ∗-algebras. Vol. II ∗-Algebras, Encyclopedia of Mathematics and its Applications, 79. Cam- bridge University Press, Cambridge (2001). [17] I. Raeburn, D. P. Williams, Morita equivalence and continuous-trace C- algebras, Mathematical Surveys and Monographs, 60, American Mathe- matical Society, Providence, RI, (1998). [18] K. Tzanev, Cross product by Hecke pairs, talk at "Workshop on noncom- mutative geometry and number theory, II" MPIM, Bonn June 14-18, 2004. 76
1407.8328	2	1407	2016-06-14T07:37:53	Algebraically irreducible representations and structure space of the Banach algebra associated with a topological dynamical system	[ "math.OA", "math.FA" ]	If $X$ is a compact Hausdorff space and $\sigma$ is a homeomorphism of $X$, then a Banach algebra $\ell^1(\Sigma)$ of crossed product type is naturally associated with this topological dynamical system $\Sigma=(X,\sigma)$. If $X$ consists of one point, then $\ell^1(\Sigma)$ is the group algebra of the integers. We study the algebraically irreducible representations of $\ell^1(\Sigma)$ on complex vector spaces, its primitive ideals and its structure space. The finite dimensional algebraically irreducible representations are determined up to algebraic equivalence, and a sufficiently rich family of infinite dimensional algebraically irreducible representations is constructed to be able to conclude that $\ell^1(\Sigma)$ is semisimple. All primitive ideals of $\ell^1(\Sigma)$ are selfadjoint, and $\ell^1(\Sigma)$ is Hermitian if there are only periodic points in $X$. If $X$ is metrisable or all points are periodic, then all primitive ideals arise as in our construction. A part of the structure space of $\ell^1(\Sigma)$ is conditionally shown to be homeomorphic to the product of a space of finite orbits and $\mathbb T$. If $X$ is a finite set, then the structure space is the topological disjoint union of a number of tori, one for each orbit in $X$. If all points of $X$ have the same finite period, then it is the product of the orbit space $X/\mathbb Z$ and $\mathbb T$. For rational rotations of $\mathbb T$, this implies that the structure space is homeomorphic to $\mathbb T^2$.	math.OA	math	ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE OF THE BANACH ALGEBRA ASSOCIATED WITH A TOPOLOGICAL DYNAMICAL SYSTEM MARCEL DE JEU AND JUN TOMIYAMA Abstract. If X is a compact Hausdorff space and σ is a homeomorphism of X, then a Banach algebra ℓ1(Σ) of crossed product type is naturally associated with this topological dynamical system Σ = (X, σ). If X consists of one point, then ℓ1(Σ) is the group algebra of the integers. We study the algebraically irreducible representations of ℓ1(Σ) on complex vector spaces, its primitive ideals, and its structure space. The finite dimen- sional algebraically irreducible representations are determined up to algebraic equivalence, and a sufficiently rich family of infinite dimensional algebraically irreducible representations is constructed to be able to conclude that ℓ1(Σ) is semisimple. All primitive ideals of ℓ1(Σ) are selfadjoint, and ℓ1(Σ) is Hermit- ian if there are only periodic points in X. If X is metrizable or all points are periodic, then all primitive ideals arise as in our construction. A part of the structure space of ℓ1(Σ) is conditionally shown to be homeomorphic to the product of a space of finite orbits and T. If X is a finite set, then the structure space is the topological disjoint union of a number of tori, one for each orbit in X. If all points of X have the same finite period, then it is the product of the orbit space X/Z and T. For rational rotations of T, this implies that the structure space is homeomorphic to T2. 1. Introduction and overview If X is a compact Hausdorff space and σ is a homeomorphism of X, then there is a Banach algebra ℓ1(Σ) of crossed product type associated with the dynamical system Σ = (X, σ). It is an involutive algebra, and there is a significant amount of literature on the relation between the properties of the enveloping C∗-algebra C∗(Σ) of ℓ1(Σ) and those of the dynamical system. The algebra ℓ1(Σ) itself, however, is far less well studied, even though it is arguably more naturally associated with Σ than C∗(Σ), the construction of which takes one extra step. The investigation of ℓ1(Σ), which is an algebra with a more complicated structure than C∗(Σ), has been taken up in [10] and has been continued in [11] and [12]. The present paper is a further step in the study of ℓ1(Σ). It fits into what seems to be an emerging line of research where Banach algebras of crossed prod- uct (or related) type are considered that are associated with (abstract) dynamical systems, but that are not C∗-algebras or closed subalgebras of C∗-algebras. We refer to [3, 4, 5, 6, 8, 9, 18, 19, 20] as examples of this development. These new algebras present an extra challenge compared to C∗-algebras, because the latter with their rigidity properties are still reasonably manageable, and have a relatively 2010 Mathematics Subject Classification. Primary 46H15; Secondary 46H10, 47L65, 54H20. Key words and phrases. Banach algebra, algebraically irreducible representation, primitive ideal, structure space, crossed product, topological dynamical system. 1 2 MARCEL DE JEU AND JUN TOMIYAMA uncomplicated -- though still far from trivial -- structure. As an example, the al- gebra ℓ1(Z) -- our algebra ℓ1(Σ) reduces to this algebra when X consists of one point -- has closed non-selfadjoint ideals, whereas this is of course no longer true for its enveloping C∗-algebra C(T). In this paper, we concentrate on the algebraically irreducible representations of ℓ1(Σ) on complex vector spaces, the primitive ideals, and the structure space of ℓ1(Σ). Hence there is no topology on the representation space involved, although the fact that a topology can always be brought into play (see Theorem 2.1) will play an important part in the proofs. This is contrary to what sometimes seems to have become the main objective (in particular for involutive Banach algebras), namely to study topologically irreducible (∗-)representations. In many basic papers, including [21], this is done without further comment, and also the present authors have used this definition of an irreducible representation in [12]. The fact that for C∗(Σ)-algebras there is no difference (see Theorem 2.2) will have encouraged this tendency. In the present paper, however, we return to the purely algebraic viewpoint. This gives the correct definition of a primitive ideal that enables one to introduce the hull-kernel topology on the set of primitive ideals. The reader who is familiar with the formulas used to define representations of ℓ1(Σ) in [10, 11, 12] will notice a clear similarity with the formulas in the present paper. Although these formulas have certainly been an inspiration, the similarity does not go much further than that, because in a purely algebraic context we need other techniques than in the previous papers. For example, it is not so difficult to determine the finite dimensional algebraically irreducible ∗-representations of ℓ1(Σ), since there is a theory of states available in this Hilbert space context, but to show that these actually exhaust the finite dimensional algebraically irreducible repre- sentations on complex vector spaces up to equivalence is another matter. Likewise, with every aperiodic point we shall associate a representation of ℓ1(Σ) on ℓp(Z) for every p ∈ [1, ∞]. It is easy to show that this representation is topologically irreducible if p ∈ (0, ∞) and not topologically irreducible if p = ∞ (see Proposi- tion 3.15), but proving that it is algebraically irreducible if p = 1 is more demanding (see Proposition 3.13). This paper is organised as follows. In Section 2, we establish notation, collect non-trivial key material on alge- braically irreducible representations of Banach algebras, introduce our Banach al- gebra ℓ1(Σ), and establish basic results on dynamical systems that are induced by non-zero homomorphisms from ℓ1(Σ) into a normed algebra. Section 3 contains the description of all finite dimensional algebraically irre- ducible representations of ℓ1(Σ) up to algebraic equivalence (see Theorem 3.5). For p ∈ [1, ∞], the representations of ℓ1(Σ) on ℓp(Z) associated with aperiodic points are introduced and investigated from the viewpoint of equivalence (see Proposi- tions 3.8 and 3.10) and algebraic and topological irreducibility (see Theorem 3.16). With the primary algebraic goal of this paper in mind, we could have restricted ourselves to the algebraically irreducible representations on ℓ1(Z) and their alge- braic equivalences, but it seemed less than satisfactory not to present the complete picture. Section 4 combines the algebraically irreducible representations from Section 3 with the technique of induced dynamical systems from Section 2. It is shown that, even though we do not generally know them all, every primitive ideal of ℓ1(Σ) is ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 3 selfadjoint (see Theorem 4.3). Furthermore, ℓ1(Σ) is a Hermitian Banach algebra if all points of X are periodic (see Theorem 4.7). If X is metrizable, then, even though we do not generally know all infinite dimensional algebraically irreducible representations, we can still show that all primitive ideals can be obtained from Section 2 (see Theorem 4.18). This section also contains -- partly as a prelude to Section 5 -- a more detailed investigation of the primitive ideals originating from Section 3. There are already enough of these to conclude that ℓ1(Σ) is semisimple (see Theorem 4.11). The final Section 5 concentrates on the structure space of ℓ1(Σ), i.e. on the set of primitive ideals of ℓ1(Σ) in the hull-kernel topology. Several results of Section 4 can be interpreted in this context (see Theorem 5.1). The main goal of this section is the description -- under conditions -- of parts of this structure space as topological products of spaces of finite orbits and T (see Theorem 5.10). If X is a finite set, then the structure space is the topological disjoint union of a number of tori, one for each orbit in X. If all points of X are periodic with the same period, then it is homeomorphic to the product of the orbit space X/Z and T. For rational rotations of T, this implies that the structure space is homeomorphic to T2. The methods in Section 5 could conceivably be adapted to yield similar results for the structure space of C∗(Σ), but further research is needed to explore this perspective. 2. Preliminaries This section contains the necessary preliminary definitions and results. We start by introducing some conventions and terminology for representations. The latter, which we give in detail since there are various different terminologies in use, is consistent with that in [17], with the exception that -- to prevent any misunderstanding -- we write 'algebraically irreducible' where [17] uses 'irreducible'. All vector spaces in this paper are complex. Algebras are not necessarily unital. Ideals of an algebra are two-sided; ideals of normed algebras are not necessarily closed. A representation of an algebra A on a vector space E is a not necessarily unital homomorphism π : A → L(E) into the linear operators L(E) on E. It is algebraically irreducible if π(A)(E) 6= {0} and the only invariant subspaces of E are {0} and E. An algebraically irreducible representation of a unital algebra is neces- sarily unital. If E is a normed space, then a normed representation of an algebra A on E is a representation π of A on E such that π(A) ⊂ B(E), where B(E) denotes the bounded operators on E. A normed representation is topologically irreducible if π(A)(E) 6= {0} and the only closed invariant subspaces of E are {0} and E. A topologically irreducible normed representation of a unital algebra is necessarily unital. A normed representation of a normed algebra A on a normed space E is continuous if π : A → B(E) is continuous, and it is contractive if π : A → B(E) is contractive. The notions ∗-representation and algebraic equivalence, topological equivalence, isometric equivalence, and unitary equivalence of representations are self-explanatory. A ∗-representation of a Banach algebra with isometric involution on a Hilbert space is automatically contractive. Next, we collect some non-trivial facts on algebraically irreducible representa- tions of Banach algebras. Theorem 2.1. Let A be a Banach algebra and let E be a vector space. Suppose that π : A → L(E) is an algebraically irreducible representation. Then: 4 MARCEL DE JEU AND JUN TOMIYAMA (1) The algebra of intertwining operators on E consists of the complex multiples of the identity. (2) E has a unique Banach space topology relative to which π is normed. There exists a norm inducing this topology such that π is a contractive represen- tation. (3) An algebraic equivalence between two algebraically irreducible normed rep- resentations of A on Banach spaces is a topological equivalence. Part (1) follows from [2, Corollary 25.3.(i) and Theorem 14.2], where we use the fact that we are working over C. The first part of (2) is [17, Corollary 4.2.16.(a)], and the second part follows from an inspection of the proof of [2, Lemma 25.2], or as a special case of [17, Theorem 4.2.7]. Part (3) is [17, Corollary 4.2.16.(b)]. The fact that every algebraically irreducible representation of a Banach algebra can be viewed as a continuous (even contractive) representation, as asserted in part (2), will be used repeatedly. This possibility is a consequence of the fact that maximal modular left ideals of a Banach algebra are closed. The next result may well underlie the fact that in parts of the literature the 'irre- ducibility' of a normed representation stands for what is 'topological irreducibility' in our terminology. The 'if' part is Kadison's result (see [13] or [7, Corollary 2.8.4]); the 'only if' part follows from [7, Corollary 2.9.6.(i)]. Theorem 2.2. Let A be a C∗-algebra and let π be a representation of A on a vector space E. Then π is algebraically irreducible if and only if E can be supplied with the structure of a Hilbert space such that π is a topologically irreducible ∗-representation of A on E. We now turn to the dynamical system and its associated Banach algebra. Throughout this paper, X is a non-empty compact Hausdorff space and σ : X → X is a homeomorphism. Hence Z acts on X, and we write Σ = (X, σ) for this topological dynamical system. We let Aper(σ) and Per(σ) denote the aperiodic and the periodic points of σ, respectively. We say that (X, σ) is topologically free if Aper(σ) is dense in X, and that it is free if the Z-action is free, i.e. if Per(σ) = ∅. It is topologically transitive ifSn∈Z σn(V ) is dense in X for every non-empty open subset V of X. For every integer p ≥ 1, let Perp(σ) be the set of points with an orbit of p elements. A subset S of X is invariant if it is invariant under the Z-action, i.e. if σ(S) = S. If S is invariant, then so are its closure and interior. The sets Aper(σ), Per(σ) are invariant, as are the sets Perp(σ) for every integer p ≥ 1. We shall write o for a general finite or infinite orbit, and o for the closure of an orbit. The involutive algebra of continuous (complex-valued) functions on X is denoted by C(X), and we write α for the involutive automorphism of C(X) induced by σ, defined by α(f ) = f ◦ σ−1 for f ∈ C(X). Via n 7→ αn, Z acts on C(X). With k · k denoting the supremum norm on C(X), we let ℓ1(Σ) = ℓ1(Z, C(X)) =(a : Z → C(X) : kak :=Xn∈Z ka(n)k < ∞) . We supply ℓ1(Σ) with the usual twisted convolution as multiplication, defined by (aa′)(n) =Xk∈Z a(k) · αk(a′(n − k)) ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 5 for n ∈ Z and a, a′ ∈ ℓ1(Σ), and define an involution on ℓ1(Σ) by a∗(n) = αn(a(−n)) for n ∈ Z and a ∈ ℓ1(Σ). Thus ℓ1(Σ) becomes a unital Banach ∗-algebra with isometric involution, and we call ℓ1(Σ) the Banach algebra associated with Σ. If X consists of one point, then ℓ1(Σ) is the group algebra ℓ1(Z) of the integers. A convenient way to work with ℓ1(Σ) is provided by the following. For n, m ∈ Z, let χ{n}(m) =(1 if m = n; 0 if m 6= n, where the constants denote the corresponding constant functions in C(X). Then χ{0} is the identity element of ℓ1(Σ). Let δ = χ{1}; then χ{−1} = δ−1 = δ∗. If we put δ0 = χ{0}, then δn = χ{n} for all n ∈ Z. We may view C(X) as a closed abelian ∗-subalgebra of ℓ1(Σ), namely as {a0δ0 : a0 ∈ C(X)}. If a ∈ ℓ1(Σ), and if we write fn for a(n) as a more intuitive notation, then a = Pn∈Z fnδn and kak =Pn∈Z kfnk < ∞. In the rest of this paper, we shall constantly use this series representation a = Pn∈Z fnδn of an arbitrary element a ∈ ℓ1(Σ), with uniquely determined fn ∈ C(X) for n ∈ Z. Thus ℓ1(Σ) is generated, as a unital Banach algebra, by an isometrically isomorphic copy of C(X) and the elements δ and δ−1, subject to the relation δf δ−1 = α(f ) = f ◦ σ−1 for f ∈ C(X). The isometric involution is determined by f ∗ = f for f ∈ C(X) and by δ∗ = δ−1. We continue our preparations by including some material on Banach algebras associated with non-empty invariant closed subsets and -- this is the actual pur- pose -- with homomorphisms. For a non-empty closed invariant subset S of X, let K (S) =(Xn∈Z fnδn ∈ ℓ1(Σ) : fn↾S = 0 for all n ∈ Z) be the closed ideal of ℓ1(Σ) that is generated by {f ∈ C(X) : f ↾S = 0}. It is proper and selfadjoint. Since S is invariant, ΣS := (S, σ↾S) is a topological dynamical system in its own right; hence there is an associated Banach algebra ℓ1(ΣS). Elements of this algebra can be written as absolutely convergent series Pn∈Z gnδn S, where gn ∈ C(S) for n ∈ Z. Lemma 2.3. Let S ⊂ X be a non-empty invariant closed subset, and define R : ℓ1(Σ) → ℓ1(ΣS) by RS Xn∈Z forPn∈Z fnδn ∈ ℓ1(Σ). Then: fnδn! =Xn∈Z fn↾S δn S (1) RS is a surjective unital contractive ∗-homomorphism; (2) Ker (RS) = K (S) is a proper closed selfadjoint ideal of ℓ1(Σ); (3) If QS : ℓ1(Σ) → ℓ1(Σ)/K (S) denotes the unital quotient ∗-homomorphism, S ◦ QS = RS S : ℓ1(Σ)/K (S) → ℓ1(ΣS) such that R′ then the induced map R′ is an isometric ∗-isomorphism. Proof. As a consequence of Tietze's extension theorem, the canonical map from C(X)/{f ∈ C(X) : f ↾S = 0} to C(S) is a ∗-isomorphism of C∗-algebras. Hence 6 MARCEL DE JEU AND JUN TOMIYAMA it is isometric. Therefore, if g ∈ C(S) and ε > 0 are given, there exists f ∈ C(X) such that f ↾S = g and kf k < kgk + ε. This implies that RS is surjective. The rest in (1) is clear, as is (2), and the isometric nature of the induced map R′ S in (3) follows again from the above extension property of elements of C(S). (cid:3) The above construction can be put to good use when studying homomorphisms, as follows. Let π : ℓ1(Σ) → A be a non-zero not necessarily unital continuous homomorphism from ℓ1(Σ) into a not necessarily unital normed algebra A. If π(C(X)) = {0}, then π(1) = 0, implying that π = 0. Since this is not the case, {f ∈ C(X) : π(f ) = 0} is a proper α-invariant closed ideal of C(X). There exists a unique closed subset Xπ of X such that, for f ∈ C(X), π(f ) = 0 if and only if f ↾Xπ = 0; we see that Xπ is non-empty and σ-invariant. The dynamical system Σπ := (Xπ, σXπ ) is called the dynamical system induced by π. The continuity of π implies that K (Xπ) ⊂ Ker (π); hence we have a continuous homomorphism π : ℓ1(Σ)/K (Xπ) → A such that π = QXπ ◦ π, where QXπ : ℓ1(Σ) → ℓ1(Σ)/K (Xπ) is the quotient map. Then kπk = kπk. On the other hand, Lemma 2.3 shows that the map R′ Xπ ◦ QXπ = RXπ is an , then π′ : ℓ1(Σπ) → A is isometric ∗-isomorphism. continuous, kπ′k = kπk = kπk, and π = π′ ◦ R. The following is now clear. Xπ : ℓ1(Σ)/K (Xπ) → ℓ1(Σπ) such that R′ If we let π′ = π ◦(cid:0)R′ Xπ(cid:1)−1 Proposition 2.4. Let π : ℓ1(Σ) → A be a non-zero not necessarily unital contin- uous homomorphism from ℓ1(Σ) into a not necessarily unital normed algebra A. Then {f ∈ C(X) : π(f ) = 0} = {f ∈ C(X) : f ↾Xπ = 0} for a unique closed subset Xπ of X. This Xπ is non-empty and invariant; hence it yields a dynamical system Σπ = (Xπ, σ↾Xπ ). Define RXπ : ℓ1(Σ) → ℓ1(Σπ) by RXπ Xn∈Z forPn∈Z fnδn ∈ ℓ1(Σ). Then: fnδn! =Xn∈Z fn↾Xπ δn Xπ (1) RXπ is a surjective unital contractive ∗-homomorphism; (2) There exists a unique map π′ : ℓ1(Σπ) → A such that π = π′ ◦ RXπ . This π′ is a continuous homomorphism and kπ′k = kπk. If A is unital, then π is unital precisely when π′ is unital. If A is involutive, then π is involutive precisely when π′ is involutive; (3) π′ is injective on C(Xπ); (4) If A = B(E) for a normed space E, then π is an algebraically (resp. topo- logically) irreducible representation of ℓ1(Σ) on E if and only if π′ is an algebraically (resp. topologically) irreducible representation of ℓ1(Σπ) on E. Remark 2.5. If π : ℓ1(Σ) → L(E) is an algebraically irreducible representation of ℓ1(Σ) on a vector space E, then, as noted in Theorem 2.1, we may assume that E is a normed space (even that it is a Banach space) and that π : A → B(E) is continuous (even that it is contractive). Hence we can assign an induced dynamical system Σπ and corresponding Banach algebra ℓ1(Σπ) to every algebraically irreducible representation of ℓ1(Σ), and Proposition 2.4 applies. We shall exploit this several times in Section 4. The following result is the cornerstone when showing that the primitive ideals corresponding to the infinite dimensional algebraically irreducible representations of ℓ1(Σ) are selfadjoint (see the proof of Theorem 4.3). It relies on one of the main ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 7 results of [10]: the commutant C(X)′ of C(X) in ℓ1(Σ) has non-zero intersection with every non-zero closed ideal of ℓ1(Σ) (see [10, Theorem 3.7]). Proposition 2.6. Let π be a non-zero not necessarily unital continuous homo- morphism from ℓ1(Σ) into a not necessarily unital normed algebra A such that its induced dynamical system Σπ is topologically free. As in Proposition 2.4, let π′ : ℓ1(Σπ) → A be the continuous homomorphism such that π = π′ ◦ RXπ . Then π′ is injective on ℓ1(Σπ). Consequently, Ker (π) = Ker (RXπ ) = K (Xπ) is the closed ideal of ℓ1(Σ) that is generated by {f ∈ C(X) : f ↾Xπ = 0}. In particular, Ker (π) is selfadjoint. Proof. To see that π′ is injective, assume that Ker (π′) 6= {0}. Then the result men- tioned preceding the theorem, applied to ℓ1(Σπ), implies that Ker (π′) ∩ C(Xπ)′ 6= {0}, where C(Xπ)′ is the commutant of C(Xπ) in ℓ1(Σπ). However, since Σ is topologically free, [10, Proposition 3.1] implies that C(Xπ)′ = C(Xπ). Hence Ker (π′)∩C(Xπ) 6= {0}. But this contradicts the injectivity of π′ on C(Xπ) in Propo- sition 2.4. Therefore we must have Ker (π′) = {0}, and hence Ker (π) = Ker (RXπ ). The rest is clear. (cid:3) We conclude our preparations with a few elementary topological results for which we are not aware of a reference. They are needed in the proof of Theorems 5.9 and 5.10. Lemma 2.7. Let S be a topological space. i=1 Si is the finite disjoint union of subsets Si, where each Si is a closed subset of S that is a compact Hausdorff space in the induced topology. Then the topological space S is a compact Hausdorff space, i=1 Si of the topological spaces Si carrying their (1) Suppose that S =Sn and it is the disjoint unionFn (2) Suppose that S =Fi∈I Si is the arbitrary disjoint union of topological spaces Si, and that T is a topological space. Then S × T = Fi∈I (Si × T ) as a (3) Suppose that S =Fi∈I Si is the arbitrary disjoint union of topological spaces in S is entirely included in one (unique) Si. Then S/ ∼=Fi∈I (Si/ ∼) as Si. Let ∼ be an equivalence relation on S such that each equivalence class induced topologies. disjoint union of topological spaces. a disjoint union of topological spaces. Proof. For part (1), we note that each Si is open. Combining this with the fact that the Si are Hausdorff, one sees that S is Hausdorff. It is now already clear that S is a compact Hausdorff space. Since the canonical map from the disjoint union of i=1 Si to S is a continuous bijection between a compact space and a Hausdorff space, it is a homeomorphism. This proves (1). Parts (2) and (3) are completely elementary. (cid:3) topological spacesFn 3. Representations associated with points In this section, we study representations of ℓ1(Σ) that are naturally associated with the points of X, and their algebraic and (when applicable) topological equiv- alence. To each periodic point corresponds a family (parameterized by T) of finite dimensional algebraically irreducible representations. Each of the pertinent rep- resentation spaces can be supplied with a Hilbert space structure such that the 8 MARCEL DE JEU AND JUN TOMIYAMA representation is then a ∗-representations, and, taken together, these representa- tions exhaust the algebraically irreducible representations up to algebraic equiva- lence (see Theorem 3.5). The representations associated with infinite orbits have ℓp(Z) for p ∈ [1, ∞] as representation spaces, and also for these representations we can resolve the algebraic equivalence and the algebraic or topological irreducibility questions (see Proposition 3.8 and Theorem 3.16). 3.1. Preparations. In order not to interrupt the main exposition, we formulate two preparatory results on representations of ℓ1(Σ) in this subsection that will be used a number of times. Part (1) of the following result follows from more general results for covariant representations of Banach algebra dynamical systems (see [8, Theorem 5.20]), ap- plied to the case (C(X), Z, α) at hand. However, once one notes that δn = 1 for all n ∈ Z, both part (1) and (2) can also be derived by elementary arguments for the discrete group Z; the proof is left to the reader. Lemma 3.1. (1) Let π : ℓ1(Σ) → B(E) be a unital continuous representation of ℓ1(Σ) on a normed space E. Put T = π(δ). Then the restriction ρ := π↾C(X) : C(X) → B(E) of π to C(X) is a unital continuous representation of C(X) on E, T is invertible in B(E), ρ(α(f )) = T ρ(f )T −1 for f ∈ C(X), and there exists M ≥ 0 such that T n ≤ M for all n ∈ Z. If E is a Banach space, then, conversely, if ρ and T are given satisfying these four properties, then there is a unique unital continuous representation π of ℓ1(Σ) on E such that its restriction to C(X) is ρ and T = π(δ). (2) Let π : ℓ1(Σ) → B(H) be a unital ∗-representation of ℓ1(Σ) on a Hilbert space H. Put T = π(δ). Then the restriction ρ := π↾C(X) : C(X) → B(H) of π to C(X) is a unital ∗-representation of C(X) on H, T is unitary, and ρ(α(f )) = T ρ(f )T −1 for f ∈ C(X). Conversely, if ρ and T are given sat- isfying these three properties, then there is a unique unital ∗-representation π of ℓ1(Σ) on H such that its restriction to C(X) is ρ and T = π(δ). We shall need the following result when studying the common eigenspaces for C(X) that are associated with representations of ℓ1(Σ). Lemma 3.2. Let π : ℓ1(Σ) → L(E) be a unital representation of ℓ1(Σ) on a vector space E. For x ∈ X, let (3.1) Then: Ex = {e ∈ E : π(f )e = f (x)e for all f ∈ C(X)}. (1) π(δn)Ex = Eσnx for all n ∈ Z; (2) If {x} is an open subset of X and χx ∈ C(X) is its characteristic function, then π(χx) is a projection, and π(χx)E = Ex. (3) If x1, . . . , xk ∈ X are k different points, then the sumPk sum. i=1 Exi is a direct Proof. If f ∈ C(X) and e ∈ Ex, then π(f )π(δ)e = π(δ)π(α−1(f ))e = (α−1(f ))(x)π(δ)e = f (σx)π(δe). Hence π(δ)Ex ⊂ Eσx. Likewise, π(δ)−1Ex ⊂ Eσ−1x; hence Eσx ⊂ π(δ)Eσ−1σx = π(δ)Ex. We conclude that π(δ)Ex = Eσx for all x ∈ X, and this easily implies (1). ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 9 For (2), suppose that e ∈ π(χx)E. If f ∈ C(X), then π(f )e = π(f )π(χx)e = π(f χx)e = π(f (x)χx)e = f (x)π(χe)e = f (x)e. Hence π(χx)E ⊂ Ex. Conversely, if e ∈ Ex, then in particular π(χx)e = χx(x)e = e, i=1 ei = 0, where ei ∈ Exi for i = 1, . . . , k. For each i0 such that 1 ≤ i0 ≤ k, there exists fi0 ∈ C(X) such that f (xi0 ) = 1 and f (xi) = 0 for i 6= i0. Letting fi0 act shows that ei0 = 0. so that Ex ⊂ π(χx)E. Turning to (3), suppose that Pk (cid:3) 3.2. Finite dimensional representations associated with periodic points. The algebraic equivalence classes of finite dimensional algebraically irreducible rep- resentations of ℓ1(Σ) can be described explicitly in terms of the space of finite orbits and T (see Theorem 3.5). We shall now proceed towards this result. One can associate a ∗-representation with x ∈ Per(σ) and a unimodular complex number λ ∈ T, as follows. Let p be the period of x, and let Hx,λ be a Hilbert space with orthonormal basis {e0, . . . , ep−1}. Let Tλ ∈ B(H) be represented with respect to this basis by the matrix For f ∈ C(X), let ρx(f ) be represented with respect to this basis by the matrix 0 0 1 0 0 1 ... ... 0 0 . . . 0 λ 0 . . . 0 0 . . . 0 ... ... . . . 0 . . . 1  .  .  f (x) 0 0 ... 0 f (σx) ... 0 . . . . . . . . . . . . 0 0 ... f (σp−1x)  It is easily checked that ρ and T meet the boundedness and covariance requirements in the second part of Lemma 3.1; hence there exists a unique ∗-representation πx,λ : ℓ1(Σ) → B(Hx,λ) such that πx,λ↾C(X) = ρx and π(δ) = Tλ. Proposition 3.3 (Irreducibility and equivalences). Let x ∈ Per(σ) and let λ ∈ T. Then the ∗-representation πx,λ on Hx,λ is algebraically irreducible; the dimension of Hx,λ is the cardinality of the orbit of x. If x, y ∈ Per(σ) and λ, µ ∈ T, then the following are equivalent: (1) πx,λ and πy,µ are algebraically equivalent; (2) πx,λ and πy,µ are topologically equivalent; (3) πx,λ and πy,µ are unitarily equivalent; (4) The orbits of x and y coincide, and λ = µ. Proof. Suppose that x has period p. subspace of Hx,λ, choose h = Pp If L ⊂ Hx,λ is a non-zero ℓ1(Σ)-invariant i=1 ξiei ∈ L with ξi0 6= 0 for some i0 such that 1 ≤ i0 ≤ p. Since the points x, . . . , σp−1x are different, one can choose f ∈ C(X) such that f (σi0 x) = 1 and f vanishes at the other points of the orbit of x. Letting f act on h, we see that ei0 ∈ L, and then the action of δ implies that L = H. Hence πx,λ is algebraically irreducible. We turn to the equivalences. 10 MARCEL DE JEU AND JUN TOMIYAMA Suppose that πx,λ and πy,µ are algebraically equivalent, where x has period p and y has period n. Since the dimensions of the representation spaces are then equal, we must have n = p. Furthermore, since (πx,λ(δ))p = λ and (πx,µ(δ))n = µ, we can then also conclude that λ = µ. Since Hx,λ is the direct sum of common eigenspaces of the elements of π(C(X)), with each summand corresponding to a point of the orbit of x, this must also be the case for Hy,µ. Since there is an analo- gous decomposition of Hy,µ in terms of the orbit of y, the third part of Lemma 3.2 implies that the orbit of x is included in the orbit of y. The converse inclusion follows likewise, and hence the orbits are equal. This shows that (1) implies (4). In order to show that (4) implies (3), it is sufficient to prove that πx,λ and πσx,λ are unitarily equivalent for λ ∈ T and x ∈ Perp(σ) with p ≥ 2. For this, let e0, . . . , ep−1 be the orthonormal basis of Hx,λ as described in the construction of the representation πx,λ on Hx,λ, and let e′ p−1 be the orthonormal basis of Hσx,λ as described in the construction of the representations πσx,λ on Hσx,λ. Define U : Hx,λ → Hσx,λ by U e0 = e′ j−1 for j such that 1 ≤ j ≤ p − 1. Then it is easily checked that U implements a unitary equivalence between πx,λ and πσx,λ. Hence (4) implies (3). p−1 and by U ej = λe′ 0, . . . , e′ It is trivial that (3) implies (2), and that (2) implies (1). (cid:3) The picture for the finite dimensional algebraically irreducible representation of ℓ1(Σ) is completed by the following result. We use the notation as in (3.1) in its proof, that uses the second statement in Theorem 2.1.(2) in an essential way. Proposition 3.4 (Exhaustion). Let π : ℓ1(Σ) → L(E) be an algebraically irre- ducible representation of ℓ1(Σ) on a finite dimensional vector space E. Then there exist x ∈ Per(σ) and λ ∈ T such that π and πx,λ are algebraically equivalent. Eσix(cid:17) = dim(cid:16)Ldim(E) i=0 Eσix(cid:17) ≥ dim(E) + 1. Proof. Since π(C(X)) is a commuting family of linear maps on the finite dimen- sional complex vector space E, there exists a common eigenvector e0 for this fam- ily. Consequently, there exists a point x ∈ X such that Ex 6= 0. If the points x, . . . , σdim(E)x were all different, then the first and third parts of Lemma 3.2 would imply that dim(E) ≥ dim(cid:16)Pdim(E) i=0 This contradiction implies that x is a periodic point. Let p ≥ 1 be such that x ∈ Perp(σ). Then π(δ)pEx = Eσpx = Ex; hence there exist a non-zero e0 ∈ Ex and λ ∈ C such that π(δ)pe0 = λe0. Since π is automatically unital, π(δ) is in- vertible, and hence λ 6= 0. The second part of Theorem 2.1 allows us to introduce a norm on E such that π is a continuous representation, and then Lemma 3.1 im- plies that π(δ) is two-sided power bounded. In particular, kπ(δ)kpe0k = λkke0k is bounded as k ranges over Z. Hence λ ∈ T. To conclude the proof, we let ej = π(δ)je0 ∈ Eσj x for j = 0, . . . , p − 1. It follows from Lemma 3.2 that the ej are linearly independent. Furthermore, it is easy to see that they span a non-zero subspace of E that is invariant under π(δ), π(δ−1), and π(C(X)). Since it is closed in our topology and π is continuous, it is in fact invariant under ℓ1(Σ). Therefore it equals E, and we conclude that the ej form a basis of E. It is immediate that, with respect to this basis, the matrix of π(δ) and, for every f ∈ C(X), the matrix of π(f ) are as in the construction of πx,λ outlined above. The uniqueness statement in the first part of Lemma 3.1 then implies that π and πx,λ are algebraically equivalent representations of ℓ1(Σ). (cid:3) ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 11 The following description of the algebraic equivalence classes of finite dimensional algebraically irreducible representations of ℓ1(Σ) in terms of an orbit space is now clear from Propositions 3.3 and 3.4. Theorem 3.5 (Finite dimensional representations). The Banach algebra ℓ1(Σ) has finite dimensional algebraically irreducible representations if and only if Per(σ) 6= ∅. In that case, let Per(σ)/Z be the space of finite orbits. If o ∈ Per(σ)/Z and λ ∈ T, choose x ∈ o, and let Ξo,λ denote the algebraic equivalence class of πx,λ. Then this is well defined, and Ξ is a bijection between Per(σ)/Z×T and the collection of algebraic equivalence classes of finite dimensional algebraically irreducible representations of ℓ1(Σ). Each of the representation spaces can be supplied with a Hilbert space structure such that the representation is a ∗-representations. In particular, all corresponding primitive ideals are selfadjoint. We conclude by showing that the algebraically irreducible representations of ℓ1(Σ) are all finite dimensional when X is a finite set (see Theorem 3.7). This will be used in the proof of Corollary 4.2, which, in turn, is necessary to establish Proposition 4.5. The latter is a more general result than Theorem 3.7 on which it partly builds. It is based on the following result, valid for general X. Proposition 3.6. Let π : ℓ1(Σ) → L(E) be an algebraically irreducible represen- tation of ℓ1(Σ) on the vector space E. Suppose that x ∈ Per(σ) is such that {x} is open, and that π(χx) 6= 0, where χx ∈ C(X) denotes the characteristic function of {x}. Then there exists λ ∈ T such that π is algebraically equivalent to the al- gebraically irreducible representation πx,λ associated with the periodic point x and λ ∈ T. In particular, E is finite dimensional. denote the characteristic function of the orbit of x. It follows from χ =Pp−1 and the second and third parts of Lemma 3.2 that π(χ)E = Lp−1 Proof. By the second part of Theorem 2.1, we may assume that E is a normed space and that π is a continuous representation. Let p be the period of x, and let χ ∈ C(X) j=0 χσj x j=0 Eσj x. Then clearly π(χ)E is invariant under C(X), and the first part of Lemma 3.2 implies that it is invariant under δ and δ−1. Since π(χ) is a continuous projection, its range is closed, and we can now conclude from the continuity of π that π(χ)E is invariant under π(ℓ1(Σ)). Since π(χx) 6= 0, we have π(χ)E 6= {0}, and hence E = π(χ)E =Lp−1 j=0 Eσj x. We note that π(δ)p leaves each Eσj x invariant, as a consequence of the first part of Lemma 3.2. This implies that π(δ)p commutes with π(C(X)). Since it obviously commutes with π(δ) and π(δ)−1, it commutes with π(ℓ1(Σ)) by continuity. Hence the first part of Theorem 2.1 implies that π(δ)p = λ for some λ ∈ C. Since π(δ)p is double-sided power bounded, we must have λ ∈ T. If we take e ∈ Ex to be non-zero, then, by Lemma 3.2, {e, π(δ)e, . . . , π(δ)p−1e} is independent. Its closed linear span is clearly invariant under π(C(X)), π(δ), and π(δ)−1; by the continuity of π it is then invariant under π(ℓ1(Σ)). Therefore it equals E. It is now clear from the uniqueness statement in the first part of Lemma 3.1 that π and πx,λ are algebraically equivalent. (cid:3) Now assume that X is finite and that π is an algebraically irreducible represen- tation of ℓ1(Σ) on a vector space E. Since 1 = Px∈X χx, and an algebraically irreducible representation is non-zero and automatically unital, there exists x ∈ X 12 MARCEL DE JEU AND JUN TOMIYAMA such that π(χx) 6= 0. Hence Proposition 3.6 applies. In conclusion, we have the following, as a precursor to and stepping stone for Proposition 4.5. Theorem 3.7. Suppose that X is a finite set. Then all algebraically irreducible representations of ℓ1(Σ) are finite dimensional, and the collection of algebraic equiv- alence classes of algebraically irreducible representations of ℓ1(Σ) can be identified with X/Z × T as in Theorem 3.5. Each of the representation spaces can be supplied with a Hilbert space structure such that the representation is a ∗-representations. In particular, all primitive ideals of ℓ1(Σ) are selfadjoint. 3.3. Infinite dimensional representations associated with aperiodic points. In this section, starting from an aperiodic point, we shall define representations of ℓ1(Σ) on the two-sided ℓp(Z)-spaces for all p ∈ [1, ∞]. We shall determine the algebraic, topological and isometrical equivalences (see Propositions 3.8 and 3.10), and decide the algebraic and topological irreducibility (see Theorem 3.16). The kernels of these representations are closed ideals that turn out to be selfadjoint. The representations on ℓ1(Z) are algebraically irreducible. In contrast to the finite dimensional case, we do not know all infinite dimensional algebraically irre- ducible representations up to algebraic equivalence. Later, in Theorem 4.18, we shall see that, if X is metrizable, we nevertheless know all primitive ideals corre- sponding to such representations. To construct these representations on ℓp(Z) for x ∈ Aper(σ), we first let p ∈ [1, ∞) and, for k ∈ Z, we let ek denote the element of ℓp(Z) with 1 in the kth coordinate and zero elsewhere. Let S ∈ B(ℓp(Z)) be the right shift, determined x(f ) ∈ B(ℓp(Z)) be determined by by Sek = ek+1 for k ∈ Z. For f ∈ C(X), let πp x(f )ek = f (σkx)ek for all k ∈ Z. One can then easily see that πp πp x and S satisfy the requirements of Lemma 3.1. Hence there exists a unique unital continuous representation πp x : ℓ1(Σ) → B(ℓp(Z)) such that πp x(f )Sn (3.2) fnδn! =Xn∈Z for Pn∈Z fnδn ∈ ℓ1(Σ). Furthermore, πp x Xn∈Z πp x is contractive and, for p = 2, it is a unital ∗-representation on the Hilbert space ℓ2(Z). For p = ∞, the construction is similar. We let S denote the right shift on ℓ∞(Z), and, for f ∈ C(X), we let the operator π∞ x (f ) act on a two-sided sequence via multiplication with f (σkx) in the kth coordinate for all k ∈ Z. Again Lemma 3.1 implies that there exists a unique unital continuous representation π∞ x : ℓ1(Σ) → B(ℓp(Z)) such that (3.2) holds for p = ∞. Then π∞ x is contractive. The question concerning equivalence of these representations for fixed p is easily answered. Proposition 3.8 (Equivalences). Let x, y ∈ Aper(σ) and let p ∈ [1, ∞]. Then the following are equivalent: x and πp x and πp x and πp (1) πp (2) πp (3) πp (4) The orbits of x and y coincide. y are algebraically equivalent; y are topologically equivalent; y are isometrically equivalent; Proof. We prove that (1) implies (4). If π : C(X) → L(E) is a representation of C(X) on a vector space E, then a common eigenspace for the action of C(X) on ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 13 E is equal to Ez = {e ∈ E : π(f )e = f (z)e} for a uniquely determined z ∈ X. x, these subspaces of ℓp(Z) are It is routine to check that, for the representation πp non-zero -- and in fact one dimensional and spanned by a standard vector ek -- if and only if z is in the orbit of x. Therefore, if πx,λ and πy,µ are algebraically equivalent, then the orbits of x and y must coincide. If (4) holds, say y = σjx for some unique j ∈ Z, then S−j is an isometric equivalence between πp x and πp y. Hence (4) implies (3). It is trivial that (3) implies (2), and that (2) implies (1). (cid:3) Remark 3.9. Let p ∈ [1, ∞]. Using the one dimensionality of the common eigenspaces for the action of C(X), and considering the action of δ and δ−1, one sees easily that, for any T ∈ L(ℓp(Z)) commuting with the action of ℓ1(Σ), there exists λ ∈ C such that T ek = λek for all k ∈ Z. If p ∈ [1, ∞), this implies that the bounded intertwining operators for πp x are the multiples of the identity. It is natural to ask, in addition, which algebraic equivalences exist between πp x and πr y if p and r are such that 1 ≤ p < r ≤ ∞. As in the proof of Proposition 3.8, the orbits of x and y must then coincide. Furthermore, as will become obvious from Theorem 3.16, we must then have 1 < p < r ≤ ∞, but additional information is not available at this moment. For topological equivalence, however, the picture is clear. Proposition 3.10. Let x, y ∈ Aper(σ) and let p, r ∈ [1, ∞]. Then the following are equivalent: x and πr x and πr (1) πp (2) πp (3) The orbits of x and y coincide, and p = r. y are topologically equivalent; y are isometrically equivalent; x and πr Proof. If (1) holds, then πp y are algebraically equivalent and, as remarked preceding the proposition, the orbits of x and y must then coincide. Also, if (1) holds, then ℓp(Z) and ℓr(Z) are topologically isomorphic. Since each of these spaces is clearly topologically isomorphic to its one-sided version, it then follows from [1, Corollary 2.1.6] and the non-reflexivity and non-separability of ℓ∞(N) that we must have p = r. Hence (1) implies (3). From Proposition 3.8, we see that (3) implies (2), and trivially (2) implies (1). (cid:3) We shall now investigate the algebraic and topological irreducibility of the rep- x for x ∈ Aper(σ) and p ∈ [1, ∞]. Theorem 3.16 gives a complete resentations πp answer. The hardest part is the algebraic irreducibility of π1 x, which we now take up. For the proof we need two lemmas, where the second builds on the first. λ0 = 1. If τ ∈ ℓ1(Z) and ε > 0 are given, then there exists a ∈ ℓ1(Σ) such that x(a)ρ(cid:13)(cid:13) < ε and kak ≤ kτ k. Lemma 3.11. Let x ∈ Aper(σ), and suppose that ρ = Pn∈Z λnen ∈ ℓ1(Z) with (cid:13)(cid:13)τ − π1 Proof. Let τ =Pn∈Z µnen. For every integer N ≥ 0, let ρN =Pn≤N λnen and let τN =Pn≤N µnen. If ε > 0 is given, choose N1, N2 ≥ 0 such that kτ kkρ − ρN1k + kτ − τN2 k < ε. Next, choose f ∈ C(X) such that kf k = 1, f (x) = 1, and f (σjx) = 0 for all j such that 0 < j ≤ N1. Since the points σjx for j = 0, . . . , N1 are all 14 MARCEL DE JEU AND JUN TOMIYAMA different, this is indeed possible. We then have λnf (σnx)en = e0. π1 x(f )ρN1 = Xn≤N1 a = Xn≤N2 µnδn · f ; x Xn≤N2 µnδn · f ρN1 = π1 µnδn(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) kak ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) kf k = Xn≤N2 Xn≤N2 Let then x(a)ρN1 = π1 π1 Furthermore, and x Xn≤N2 µnδn e0 = τN2. µn ≤ kτ k, kπ1 x(a)ρ − τ k ≤ kπ1 x(a)(ρ − ρN1)k + kπ1 x(a)ρN1 − τN2k + kτN2 − τ k ≤ kakkρ − ρN1k + kτN2 − τ k ≤ kτ kkρ − ρN1k + kτN2 − τ k < ǫ. Hence a meets all requirements. (cid:3) Lemma 3.12. Let x ∈ Aper(σ), and suppose that ρ = Pn∈Z λnen ∈ ℓ1(Z) with λ0 = 1. If τ ∈ ℓ1(Z) and ε > 0 are given, then there exists a ∈ ℓ1(Σ) such that π1 x(a)ρ = τ and kak ≤ (1 + ε)kτ k. Proof. We may assume that τ 6= 0. Fix γ such that 0 < γ < 1. We claim that there exist a1, a2, . . . ∈ ℓ1(Σ) such that (3.3) τ − π1 < γN kτ k (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) x NXj=1 aj ρ(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) kτ − π1 x(a1)ρk < γkτ k for all N ≥ 1 and kajk ≤ γj−1kτ k for all j ≥ 1. To see this, we use an inductive construction. First apply Lemma 3.11 with ε = γkτ k > 0 to find a1 ∈ ℓ1(Σ) such that and ka1k ≤ kτ k. Let N ≥ 1, and assume that a1, a2, . . . , aN ∈ ℓ1(Σ) have already been chosen such that for all n such that 1 ≤ n ≤ N , and kajk ≤ γj−1kτ k for all j such that 1 ≤ j ≤ N . We apply Lemma 3.11 again with ε = γN +1kτ k > 0, and find aN +1 ∈ ℓ1(Σ) such that τ − π1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) x τ − π1 NXj=1 < γnkτ k aj ρ(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) x nXj=1 x(aN +1)ρ(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) aj ρ − π1 < γN +1kτ k ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 15 This completes the inductive step in the construction and establishes our claim. τ − π1 < γN kτ k. x NXj=1 aj ρ(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) γj−1kτ k = 1 1 − γ kτ k. and kaN +1k ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ∞Xj=1 kak ≤ j=1 aj ∈ ℓ1(Σ), then If we let a =P∞ Furthermore, if we let N → ∞ in (3.3), we see that π1 x(a)ρ = τ. Since this can be done for any γ such that 0 < γ < 1, the proof is complete. (cid:3) Proposition 3.13. Let x ∈ Aper(σ). Then the representation π1 is algebraically irreducible. In fact, for any ρ =Pn∈Z λnen ∈ ℓ1(Z) such that ρ 6= 0, τ ∈ ℓ1(Z), and ε > 0, there exists a ∈ ℓ1(Σ) such that πx(a)ρ = τ and x of ℓ1(Σ) on ℓ1(Z) kak ≤ (1 + ε) maxn λn kτ k. Proof. Suppose that λn0 = maxn λn 6= 0. Then π1 dition for ρ in Lemma 3.12. Hence there exists a′ ∈ ℓ1(Σ) such that x(λ−1 n0 δ−n0 )ρ satisfies the con- and π1 x(a′)(cid:0)π1 x(λ−1 n0 δ−n0 )(cid:1) ρ = τ Then a = a′ · λ−1 ka′k ≤ (1 + ε)kτ k. n0 δ−n0 has the required properties. (cid:3) The algebraic irreducibility for other cases is easily dealt with. Lemma 3.14. Let x ∈ Aper(σ) and let p ∈ (1, ∞]. Then the representation πp x of ℓ1(Σ) on ℓp(Z) is not algebraically irreducible. In fact, πp x(ℓ1(Σ))e0 is a proper invariant subspace. Proof. A moment's thought shows that πp a subspace of ℓp(Z). For p ∈ (1, ∞] this is indeed a proper subspace. x(ℓ1(Σ))e0 = ℓ1(Z), viewed canonically as (cid:3) Topological irreducibility is settled as follows. Proposition 3.15. Let x ∈ Aper(σ). For p ∈ [1, ∞), the representation πp x of ℓ1(Σ) on ℓp(Z) is topologically irreducible. For p = ∞, it is not topologically irreducible. In fact, π∞ x (ℓ1(Σ))e0 is a proper closed invariant subspace of ℓ∞(Z). Proof. Assume that p ∈ [1, ∞) and that L is a non-zero closed invariant subspace x(δ) or its inverse a number of times, followed by scaling, we may assume that λ0 = 1. Let ε > 0, and of ℓp(Z). Let ρ =Pn∈Z λnen ∈ L be non-zero. After applying πp choose N such thatPn>N λnp < εp. Since the 2N + 1 points σjx for j ≤ N are 16 MARCEL DE JEU AND JUN TOMIYAMA all different, we can choose f ∈ C(X) such that kf k = 1, f (x) = 1, and f (σj x) = 0 for all j such that 1 ≤ j ≤ N . Then kπp x(f )ρ − e0k =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xn>N = Xn>N ≤ Xn>N f (σnx)λnen(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) f (σnx)λnp λnp < ε. 1/p 1/p Since L is closed, this implies that e0 ∈ L. But then also πp already observed earlier, πp Since this copy is clearly dense in ℓp(Z) for p ∈ [1, ∞), we must have L = ℓp(Z). x(ℓ1(Σ))e0 ⊂ L. As x(ℓ1(Σ))e0 equals the canonical copy of ℓ1(Z) in ℓp(Z). If p = ∞, then π∞ x (ℓ1(Σ))e0 is not dense in ℓ∞(Z). Indeed, this is the image of ℓ1(Z) under the canonical inclusion map; since this inclusion is continuous and ℓ1(Z) is separable, the density of the image of ℓ1(Z) would then imply that ℓ∞(Z) is separable. (cid:3) We summarise our results on irreducibility in the following theorem, including Remark 3.9. Theorem 3.16 (Irreducibility). Let x ∈ Aper(σ) and let p ∈ [1, ∞]. Then the unital contractive representation πp x : ℓ1(Σ) → B(ℓp(Z)) is: (1) Algebraically irreducible if p = 1; (2) Not algebraically irreducible, but still topologically irreducible, if p ∈ (1, ∞); (3) Not topologically irreducible if p = ∞. If p ∈ [1, ∞), then the commutant of πp the identity operator. x(ℓ1(Σ)) in B(ℓp(Z)) equals the multiples of Remark 3.17. (1) In the algebraically irreducible case p = 1, the statement on bounded in- tertwining operators is also clear from the first part of Theorem 2.1, since then even the commutant of π1 x(ℓ1(Σ)) in L(ℓ1(Z)) consists of the scalars. For p = 2 it follows from topological irreducibility and Schur's Lemma, but for p such that 1 < p < 2 or 2 < p < ∞ there do not seem to be alternative general approaches. (2) As a consequence of (2) and (3) in Theorem 2.1, every algebraically irre- ducible representation of a Banach algebra comes naturally with the topo- logical isomorphism class of all Banach spaces in which the representation can be realised as a normed representation. Therefore, even though the corresponding primitive ideal is selfadjoint according to Lemma 4.9 below, the algebraically irreducible representation π1 x for x ∈ Aper(σ) is not al- gebraically equivalent to a normed representation on a Hilbert space -- let alone to a ∗-representation on such a space -- since a reflexive space is not topologically isomorphic to ℓ1(Z). ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 17 Remark 3.18. The representations as associated with aperiodic points in this section are members of a more general class of (usually) infinite dimensional repre- sentations of ℓ1(Σ), namely the representations associated with invariant measures. If µ is a σ-invariant Borel measure on X, then, for p ∈ [1, ∞], there is a natural representation πp µ of ℓ1(Σ) on Lp(X, µ), in which C(X) acts as multiplication oper- ators, and in which (πp µ(δ)f )(x) = f (σ−1x) for f ∈ Lp(X, µ) and x ∈ X. If µ is the counting measure corresponding to the orbit of an aperiodic point x ∈ X, then, for p ∈ [1, ∞], the representation πp µ is isometrically equivalent to the representation πp x. Quite in contrast to Theorem 3.16, it is presently unknown under which nec- essary and sufficient conditions these general representations πp µ are topologically or algebraically irreducible. If µ is finite and πp µ is topologically irreducible, then µ must be ergodic, but the converse is still open. The representations π1 x in the present section are the only presently known examples of algebraically irreducible representations of ℓ1(Σ). For metrizable X, a thorough study of topologically irreducible ∗-representations of ℓ1(Σ) on Hilbert spaces and their algebraic (ir)reducibility is made in [14], where these representations are related to ergodic σ-quasi-invariant probability measures on X. Here is a sample result: If µ is an ergodic σ-invariant non-atomic probability measure on X, then π2 µ is a ∗-representation of ℓ1(Σ) on L2(X, µ) that is topologi- cally irreducible, but not algebraically irreducible. The latter statement is a special case of [14, Theorem 3.4]. We refer the reader to [14] for further material in this direction. 4. Algebraically irreducible representations and primitive ideals Section 3 provides a basic stockpile of algebraically irreducible representations and corresponding primitive ideals, all related to the points of X. In this section, we combine this information with the technique of induced dynamical systems from Section 2, and obtain various results on general algebraically irreducible represen- tations and primitive ideals of ℓ1(Σ). As a side result, we show that ℓ1(Σ) is a Hermitian Banach algebra if all points are periodic. We also have a closer look at the primitive ideals associated with the algebraically irreducible representations from Section 3, and show how they can be parameterized (see Proposition 4.16). Using these primitive ideals, it already follows that ℓ1(Σ) is semisimple. As is to be expected, the primitive ideals are more accessible than the in general more numerous (algebraic equivalence classes of) algebraically irreducible represen- tations. The fact that, even though we have no explicit description, all primitive ideals of ℓ1(Σ) can be shown to be selfadjoint (see Theorem 4.3) is an example of this. Theorem 4.18 is an even clearer illustration: if X is metrizable, then we know all primitive ideals -- they all originate from Section 3 -- even though we do not know all algebraically irreducible representations. In some results, the enveloping C∗-algebra C∗(Σ) of ℓ1(Σ) makes an appearance. The results for that algebra are not used to obtain results for ℓ1(Σ). They are simply cited, and are included to show that some properties are in fact preserved when making the non-trivial passage from ℓ1(Σ) to C∗(Σ). Our first main task is to show that all primitive ideals are selfadjoint. For this, we shall use the properties of induced dynamical systems as in Proposition 2.4, Remark 2.5, Proposition 3.4, and the notation therein. We start with a lemma. Recall from Section 2 that a topological dynamical system Σ = (X, σ) is said to 18 MARCEL DE JEU AND JUN TOMIYAMA be topologically transitive ifSn∈Z σn(V ) is dense in X for every non-empty open subset V of X. Lemma 4.1. Let π : ℓ1(Σ) → B(E) be a topologically irreducible continuous repre- sentation of ℓ1(Σ) on a normed space E. Then the induced dynamical system Σπ is topologically transitive. that there exists a non-empty open subset U such that S :=Sn∈Z σn Proof. The short proof of [25, Proposition 4.4] in a Hilbert context also works in general, and we include it for the convenience of the reader. Suppose to the contrary (U ) 6= Xπ, so that S is a proper closed invariant subset of Xπ. Hence, if we let ker(S) = {f ∈ C(Xπ) : f ↾S = 0}, then ker(S) 6= {0}. Since π′ is injective by the third part of Proposition 2.4, we have π′(ker(S))E 6= {0}. The invariance of S implies that π′(ker(S))E is not only invariant under π′(C(Xπ)), but also under π′(δXπ ) and π′(δ−1 ). Since π′ is continuous, the non-zero closed subspace π′(ker(S))E is Xπ invariant under π′(ℓ1(Σπ)). We conclude that π′(ker(S))E = E. Now take a non- zero f ∈ C(Xπ) such that supp (f ) ⊂ U . Then f ker(S) = {0} since U ⊂ S, and this implies that π′(f ) = 0. But this contradicts the injectivity of π′ on C(Xπ). (cid:3) Xπ Before stating a consequence of the previous result, we recall from Section 2 that a topological dynamical system Σ = (X, σ) is said to be topologically free if the subset Aper(σ) of aperiodic points of σ is dense in X. Corollary 4.2. Let π : ℓ1(Σ) → L(E) be an algebraically irreducible representation of ℓ1(Σ) on a vector space E. Then the induced dynamical system Σπ is topologically If E is infinite dimensional, then Xπ is an infinite set, and Σπ is transitive. topologically free. Proof. For the first part, we note that, by the second part of Theorem 2.1, we can assume that E is normed and that π is continuous. Then π is certainly topologically irreducible; hence the first statement follows from Lemma 4.1. For the second part, we know from the fourth part of Proposition 2.4 that the induced representation π′ of ℓ1(Σπ) on E is algebraically irreducible. If Xπ were finite, then Theorem 3.7 would imply that E is finite dimensional. This is not the case, so Xπ is infinite. Since the proof of [27, Proposition 3.4] contains a proof of the statement that a topologically transitive dynamical system on an infinite compact Hausdorff space is topologically free -- a fact that can also be deduced from [11, Corollary A.2] -- the proof is complete. (cid:3) We can now reach our desired conclusion. Theorem 4.3. Every primitive ideal of ℓ1(Σ) is selfadjoint. Proof. Let π be an algebraically irreducible representation of ℓ1(Σ). If π is finite dimensional, then Ker (π) is selfadjoint by Theorem 3.5. If π is infinite dimensional, then Corollary 4.2 shows that the induced system Σπ is topologically free. Using once more that, by the second part of Theorem 2.1, we can assume that E is normed and that π is continuous, Proposition 2.6 then implies that Ker (π) is selfadjoint. (cid:3) With Theorem 4.3 available, we can now establish the following result on spectral synthesis. In order not to interrupt the line of argumentation, we shall already use the fact that ℓ1(Σ) is semisimple (see Theorem 4.11). That result will follow by direct computation on the algebraically irreducible representations of ℓ1(Σ), as ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 19 constructed in Section 3, and does not depend on any other results in the present section. Theorem 4.4. The following are equivalent: (1) Every closed ideal of ℓ1(Σ) is the intersection of primitive ideals; (2) Σ is free; (3) Every closed ideal of ℓ1(Σ) is selfadjoint. Proof. The equivalence of (2) and (3) is the statement of [10, Theorem 4.4]. Since we know from Theorem 4.3 that all primitive ideals of ℓ1(Σ) are selfadjoint, (1) implies (3). We show that (2) implies (1). Let I be a closed ideal of ℓ1(Σ); we may assume that I is proper. Consider the non-zero continuous quotient homomorphism Q : ℓ1(Σ) → ℓ1(Σ)/I. If ΣQ is its induced dynamical system, then ΣQ is free, since even Σ as a whole is free. Hence ΣQ is certainly topologically free, and then Proposition 2.6 shows that I = Ker (Q) = Ker (RXπ ), where RXπ : ℓ1(Σ) → ℓ1(Σπ) is the surjective homomorphism in Proposition 2.4. Since ℓ1(Σπ) is semisimple by Theorem 4.11, there is a collection of algebraically irreducible representations of ℓ1(Σπ) that separates the points of ℓ1(Σπ). Pulling these back to ℓ1(Σ) via the surjective homomorphism RXπ , one obtains a collection of algebraically irreducible representations of ℓ1(Σ) such that the intersection of the corresponding primitive ideals equals Ker (RXπ ), i.e. such that this intersection equals I. (cid:3) Next, we have the following two exhaustion results at the level of algebraically irreducible representations. The first one was already announced preceding Propo- sition 3.6 and Theorem 3.7. Proposition 4.5. The following are equivalent: (1) Every algebraically irreducible representation of ℓ1(Σ) on a vector space is finite dimensional; (2) X = Per(σ); (3) Every algebraically irreducible representation of C∗(Σ) is finite dimensional. Proof. If x ∈ Aper(σ), then the infinite dimensional algebraically irreducible repre- sentations π1 x in Section 3 exist. This shows that (1) implies (2). If π is an infinite dimensional algebraically irreducible representation of ℓ1(Σ), then Corollary 4.2 shows that its induced system Σπ is topologically free. In particular, there is at least one aperiodic point in Xπ, and we see that (2) implies (1). The equivalence of (2) and (3) follows from [25, Theorem 4.6.(1)], taken together (cid:3) with Theorem 2.2. Its counterpart is the following. It is more elaborate than Proposition 4.5, be- cause for finite dimensional spaces the notions of algebraic irreducibility and topo- logical irreducibility coincide. Proposition 4.6. The following are equivalent: (1) Every algebraically irreducible representation of ℓ1(Σ) on a vector space is infinite dimensional; (2) Every topologically irreducible continuous representation of ℓ1(Σ) on a Ba- nach space is infinite dimensional; (3) Every topologically irreducible representation of ℓ1(Σ) on a normed space is infinite dimensional; (4) X = Aper(σ); 20 MARCEL DE JEU AND JUN TOMIYAMA (5) Every algebraically irreducible representation of C∗(Σ) is infinite dimen- sional. Proof. If x ∈ Per(σ), then the finite dimensional algebraically irreducible represen- tations πx,λ for λ ∈ T from Section 3 exist. This shows that (1) implies (4). Since a finite dimensional topologically irreducible representation on a normed space is a finite dimensional algebraically irreducible representation, the existence of such a representation would imply the existence of a periodic point by Theorem 3.5. Hence (4) implies (3), and trivially (3) implies (2). Since, by the second part of Theorem 2.1, every algebraically irreducible representation yields an algebraically (in particular: topologically) irreducible continuous representation on a Banach space, (2) implies (1). The equivalence of (4) and (5) follows from [25, Proposition 4.5], taken together (cid:3) with Theorem 2.2. Before proceeding with the main line, we give an application of what has been obtained so far. In [10, Theorem 4.6], it was established that ℓ1(Σ) is a Hermitian algebra (i.e. that the spectrum of every selfadjoint element is a subset of the real numbers) whenever X is a finite set. We can now improve this. Theorem 4.7. If all points of X are periodic, then ℓ1(Σ) is a Hermitian algebra. Proof. As a consequence of [16], ℓ1(Σ) being Hermitian is equivalent to the following: If π : ℓ1(Σ) → L(E) is an algebraically irreducible representation of ℓ1(Σ) on a vector space E and e ∈ E is non-zero, then there exist a topologically irreducible ∗-representation π′ : ℓ1(Σ) → B(H) on a Hilbert space and a non-zero h ∈ H such that {a ∈ ℓ1(Σ) : π(a)e = 0} = {a ∈ ℓ1(Σ) : π′(a)h = 0}. But this is obvious, since Proposition 4.5 shows that every algebraically irreducible representation is finite dimensional, and, by Theorem 3.5, every such representation can be realised as a topologically irreducible ∗-representation on a Hilbert space. (cid:3) As an example, we know from Theorem 4.7 that ℓ1(Σ) is Hermitian for rational rotations of T. Later, we shall also determine the structure space of these algebras (see Example 5.13). After this sidestep, we return to main line. In order to get further, we need to have more detailed information about the primitive ideals originating from Section 3. While presenting this, we also introduce some terminology and notation that will be even more prominent in Section 5. The algebraically irreducible representations in Section 3 are associated with points, but algebraically equivalent representations yield the same primitive ideal. We shall now take this into account, and describe the primitive ideals originating from Section 3 in a bijective fashion (see Proposition 4.16). It turns out to be more natural to work with closures of orbits than with orbits, so, to establish notation, we let O denote the set of all closures of orbits in X. Note that, as irrational rotations of T show, it may well be that different orbits have equal closure. The subset of (closures of) orbits of periodic points is denoted by OPer(σ), and the subset of closures of orbits of aperiodic points by OAper(σ). Then O = OPer(σ) ⊔ OAper(σ) is a disjoint union. Let x ∈ Per(σ) and let λ ∈ T. If y is in the orbit of x, then, by Proposition 3.3, the algebraically irreducible representations πx,λ and πy,λ are algebraically equivalent. Hence we can associate a well defined primitive ideal Po,λ with λ ∈ T and the closure ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 21 o of the orbit o of x, by putting Po,λ = Ker (πx,λ). Of course, o = o here, but, when we consider primitive ideals originating from aperiodic points, we shall see why the closure of an orbit is more natural than the orbit itself, and for consistency and ease of argumentation we employ this notation for finite orbits as well. We have the following description of such Po,λ. Proposition 4.8. Let o ∈ OPer(σ) be (the closure of ) an orbit consisting of p- periodic points. If a =Pn∈Z fnδn ∈ ℓ1(Σ), then for the primitive ideals associated with o we have: (1) If λ ∈ T, then a ∈ Po,λ if and only if λlflp+j(x) = 0 Xl∈Z for all j = 0, . . . , p − 1 and all x ∈ o. (2) a ∈Tλ∈T Po,λ if and only if fn↾o = 0 for all n ∈ Z. For the proof we refer to [12, Proposition 2.10]. It involves a straightforward computation for the first part, and then the second part follows easily from the injectivity of the Fourier transform on ℓ1(Z). We now turn to the closure of an infinite orbit. The following is immediately x from Section 3.3 x, as asserted by Proposition 3.13. For the clear from the description of the corresponding representations πp and the algebraic irreducibility of π1 selfadjointness one need not resort to Theorem 4.3, as this is clear by inspection. Lemma 4.9. Let x ∈ Aper(σ), let p ∈ [1, ∞], and let a = Pn∈Z fnδn ∈ ℓ1(Σ). x(a) = 0 if and only fn vanishes on the closure of the orbit of x for all x) does not depend on p, and it is a selfadjoint primitive ideal Then πp n ∈ Z. Hence Ker (πp of ℓ1(Σ). The description of this common kernel in terms of the coefficients fn was already noted for π2 x in [12, Proposition 2.10]. As a consequence of Lemma 4.9, if o ∈ OAper(σ), then we can take any x ∈ X such that Z · x = o, and obtain a well defined primitive ideal by putting Po = Ker (π1 x). We include the above description in this notation for reference purposes. Proposition 4.10. Let o ∈ OAper(σ) be the closure of an infinite orbit. Then for the primitive ideal associated with o we have Po =(Xn∈Z fnδn ∈ ℓ1(Σ) : fn↾o = 0 for all n ∈ Z) . Since X is the union of all orbit closures, the following result (already used in the proof of Theorem 4.4) is now clear from Proposition 4.8 and Proposition 4.10. Theorem 4.11. \o∈OPer(σ) Po,λ \o∈OAper(σ) λ∈T Po = {0}. In particular, the Banach algebra ℓ1(Σ) is semisimple. We can now establish two separation results at the level of primitive ideals. 22 MARCEL DE JEU AND JUN TOMIYAMA Proposition 4.12. The following are equivalent: (1) The algebraically irreducible representations of ℓ1(Σ) on finite dimensional vector spaces separate the elements of ℓ1(Σ); (2) The algebraically irreducible representations of ℓ1(Σ) on finite dimensional vector spaces separate the elements of C(X); (3) P er(σ) is dense in X; (4) The algebraically irreducible representations of C∗(Σ) on finite dimensional vector spaces separate the elements of C∗(Σ). In view of the second part of Theorem 2.1, one could also have considered the topologically irreducible continuous (or even contractive) representations on finite dimensional Banach spaces in (1) and (2). Proof. From Theorem 3.5, we know that the algebraically irreducible finite dimen- sional representations are precisely the πx,λ from Section 3.2, and, from Proposi- tion 4.8, we haveTo∈OPer(σ),λ∈T Po,λ = {Pn∈Z fnδn ∈ ℓ1(Σ) : fn↾Per(σ) = 0 for all n ∈ Z}. It is now clear that (1), (2), and (3) are equivalent. The equivalence of (3) and (4) follows from [25, Theorem 4.6.(2)], taken together with Theorem 2.2. Proposition 4.13. The following are equivalent: (cid:3) (1) The algebraically irreducible representations {π1 x : x ∈ Aper(σ)} of ℓ1(Σ) on infinite dimensional vector spaces separate the elements of ℓ1(Σ); (2) The algebraically irreducible representations of ℓ1(Σ) on infinite dimen- sional vector spaces separate the elements of ℓ1(Σ); (3) The algebraically irreducible representations {π1 x : x ∈ Aper(σ)} of ℓ1(Σ) on infinite dimensional vector spaces separate the elements of C(X); (4) The algebraically irreducible representations of ℓ1(Σ) on infinite dimen- sional vector spaces separate the elements of C(X); (5) Σ is topologically free; (6) The algebraically irreducible representations of C∗(Σ) on infinite dimen- sional vector spaces separate the elements of C∗(Σ). Po = {Pn∈Z fnδn ∈ ℓ1(Σ) : Proof. From Proposition 4.10, we see thatTo∈OAper(σ) fn↾Aper(σ) = 0 for all n ∈ Z}. Hence (5) implies (1). Trivially (1) implies (2), and (2) implies (4). Assume that (4) holds, and that {πi}i∈I is a family of infinite dimensional algebraically irreducible representations that separates the points of and an associated C(X). Each of these has an induced dynamical system Xπi,σπi representation π′ ) for f ∈ C(X). The is dense in X. Since the aperiodic points are dense in each Xπi by Corollary 4.2, we can now conclude that Aper(σ) is dense in X, i.e. that (5) holds. It is trivial that (1) implies (3) and that (3) implies (4). separation property of {πi}i∈I implies that Si∈I Xπi i of ℓ1(Σπi ) such that πi(f ) = π′ i(f ↾Xπi The equivalence of (3) and (4) follows from [26, Proposition 4], taken together (cid:3) with Theorem 2.2. Part (1), (3), and (4) of the following result on inclusions between primitive ideals are from [12, Proposition 2.15]; we include the short proofs for the convenience of the reader. ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 23 Lemma 4.14. (1) Let o1, o2 ∈ OPer(σ) and let λ1, λ2 ∈ T. If Po1,λ1 ⊂ Po2,λ2 , then o1 = o2 and λ1 = λ2. In particular, Po,λ1 = Po,λ2 . (2) If o ∈ OPer(σ), λ ∈ T, x ∈ o, and π is an algebraically irreducible represen- tation of ℓ1(Σ) such that Po,λ ⊂ Ker (π), then π is algebraically equivalent to πx,λ. In particular, Po,λ = Ker (π). (3) Let o1, o2 ∈ OAper(σ). Then Po1 ⊂ Po2 if and only if o1 ⊃ o2. (4) Let o1 ∈ OAper(σ), let o2 ∈ OPer(σ), and let λ ∈ T. Then Po1 ⊂ Po2,λ if and only if o1 ⊃ o2. Proof. For (1), we recall that Po,λ = Ker (πx,λ), where x ∈ X is such that o = Z · x. If Po1,λ1 ⊂ Po2,λ2 , then in particular Ker (πx1,λ1 ↾C(X)) ⊂ Ker (πx2,λ2 ↾C(X)). This implies that (and is equivalent to) o1 ⊃ o2; hence o1 = o2. Then also the representation spaces have the same dimension p. Since πx1,λ1(1 − δp/λ1) = 0, we have πx2,λ2 (1 − δp/λ1) = 0, and therefore λ1 = λ2. Turning to (2), we first note that ℓ1(Σ)/Ker (π) is a quotient of ℓ1(Σ)/Po,λ. Hence we have an algebraically irreducible representation of ℓ1(Σ)/Po,λ on the representation space of π. Since ℓ1(Σ)/Po,λ is a finite dimensional algebra, π is finite dimensional. Proposition 3.4 then shows that π is algebraically equivalent to πy,µ for some y ∈ Per(σ) and µ ∈ T. An appeal to part (1) and Proposition 3.3 then concludes the proof of part (2). Part (3) is clear from Proposition 4.10. Turning to part (4), if o1 ⊃ o2, then Proposition 4.8 and Proposition 4.10 show that Po1 ⊂ Po2,λ. If o1 6⊃ o2, then o1 ∩ o2 = ∅, so that there exists f ∈ C(X) such that f ↾o1 = 0 and f ↾o2 6= 0. Then f ∈ Po1, but f /∈ Po2,λ; hence Po1 6⊂ Po2,λ. (cid:3) Remark 4.15. The argument used in the proof of the second part of Lemma 4.14 shows that the kernel of a finite dimensional algebraically irreducible representation of ℓ1(Σ) (or any other associative algebra) is never contained in (and, in particu- lar, never equal to) the kernel of an infinite dimensional algebraically irreducible representation. This elementary observation is needed in the proof of Theorem 5.1. We let Πℓ1(Σ) denote the set of all primitive ideals of ℓ1(Σ). The following is now clear from Lemma 4.14. Proposition 4.16. (1) Consider the natural map Ψ :(cid:0)OPer(σ) × T(cid:1) ⊔ OAper(σ) → Πℓ1(Σ), defined by Ψ((o, λ)) = Po,λ for o ∈ OPer(σ) and λ ∈ T, and by Ψ(o) = Po for o ∈ OAper(σ). Then Ψ is injective. (2) For all o ∈ OPer(σ) and λ ∈ T, the primitive ideal Po,λ is a maximal primitive ideal. (3) For all o ∈ OAper(σ), the primitive ideal Po is a maximal element of the range of Ψ if and only if o is a minimal orbit closure. Thus Ψ parameterizes the part of Πℓ1(Σ) that is naturally associated with the points of X, and we shall denote this part by Πℓ1(Σ) (X). It is always large in Πℓ1(Σ) in the sense that, as a consequence of Theorem 4.11, it is dense in the hull-kernel topology of Πℓ1(Σ) to be studied in Section 5. The subset of 24 MARCEL DE JEU AND JUN TOMIYAMA Πℓ1(Σ) (X) that consists of the primitive ideals Po,λ for all o ∈ OPer(σ) and λ ∈ T is denoted by Πℓ1(Σ) (Per(σ)), and the subset that consists of the primitive ideals Po for all o ∈ OAper(σ) is denoted by Πℓ1(Σ) (Aper(σ)), so that Πℓ1(Σ) (X) = Πℓ1(Σ) (Per(σ)) ⊔ Πℓ1(Σ) (Aper(σ)) is a disjoint union. We let Πℓ1(Σ) (∞) denote the subset of Πℓ1(Σ) consisting of the kernels of all infinite dimensional algebraically irreducible representations, so that Πℓ1(Σ) = Πℓ1(Σ) (Per(σ)) ⊔ Πℓ1(Σ) (∞) is a dis- joint union. It is remarkable that we can prove that Πℓ1(Σ) (X) actually equals Πℓ1(Σ) if X is metrizable, even though we do not generally know all equivalence classes of algebraically irreducible representations. This is a consequence of the following result, the proof of which once more illustrates the relevance of the technique of induced dynamical systems. Proposition 4.17. Let π be an algebraically irreducible representation of ℓ1(Σ) on an infinite dimensional vector space. If the topological space Xπ is metrizable, then there exists an aperiodic point x ∈ Xπ such that its orbit is dense in Xπ. Consequently, Ker (π) = Po, where o is the closure of the orbit of x. Proof. Let Σπ be the induced dynamical system. Since π is infinite dimensional, Corollary 4.2 shows that Σπ is topologically free, and hence Proposition 2.6 implies that Ker (π) is the closed ideal generated by {f ∈ C(X) : f ↾Xπ = 0}. Since Corollary 4.2 asserts that Xπ is topologically transitive, the metrizability of Xπ then implies, by [24, Theorem 1.1.3], that there exists a point x ∈ Xπ such that its orbit is dense in Xπ. Since Corollary 4.2 also asserts that Xπ is an infinite set, x must be aperiodic. If we note that Xπ is closed, we see that Xπ is the closure in X of the orbit of x. The rest is clear. (cid:3) When observing that finite orbits are clearly minimal orbit closures, the following is now obvious from Propositions 4.16 and 4.17; note that Proposition 4.5 asserts that there are no infinite dimensional algebraically irreducible representations of ℓ1(Σ) at all if X = Per(σ). Theorem 4.18. If, for every infinite dimensional algebraically irreducible repre- sentation π of ℓ1(Σ), the closed subset Xπ of X is metrizable (this is certainly the case if X is metrizable or X = Per(σ)), then: (1) Πℓ1(Σ) (∞) = Πℓ1(Σ) (Aper(σ)), and hence Πℓ1(Σ) = Πℓ1(Σ) (X); (2) A primitive ideal is a maximal primitive ideal if and only if it is associated with a minimal orbit closure. Remark 4.19. To each algebraic equivalence class of algebraically irreducible rep- resentations of ℓ1(Σ) one can assign the primitive ideal that is the common kernel of these representations. It follows from part (1) of Lemma 4.14 and Proposition 3.3 that this map is injective on the collection of algebraic equivalence classes of finite dimensional algebraically irreducible representations of ℓ1(Σ). On the collection of algebraic equivalence classes of infinite dimensional alge- braically irreducible representations, however, this map can be very far from injec- tive. For example, for the irrational rotations of T, where every orbit is infinite and dense, Πℓ1(Σ) consists only of the zero ideal. Even though Theorem 3.8 shows that different orbits provide different equivalence classes of algebraically irreducible representations, the associated primitive ideals are always equal to the zero ideal. Proposition 5.2 below describes when this happens for general systems. ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 25 5. The structure space We shall now consider the structure space of ℓ1(Σ), i.e. Πℓ1(Σ) in its hull-kernel topology. The main goal is Theorem 5.10, asserting that, under suitable conditions, parts of Πℓ1(Σ) are homeomorphic to products of spaces of finite orbits and T. We recall from [2, §26] that, for E ⊂ Πℓ1(Σ), the closure of E in this topology is the hull hk (E) of the kernel k (E), where k (E) =TP ∈E P , and hk (E) = {P ∈ Jacobson radicalTP ∈Π Πℓ1(Σ) : P ⊃ k (E)}. Clearly, E is dense in Πℓ1(Σ) if and only if k (E) equals the P of ℓ1(Σ). Since this is the zero ideal by Theorem 4.11, E is dense if and only if k (E) = {0}. Thus, for example, if there exists an aperiodic point with dense orbit, then the associated singleton {Po} = {{0}} is a dense subset of Πℓ1(Σ). ℓ1 (Σ) We shall now rephrase some of the results in Section 4 in terms of Πℓ1(Σ) and its topology. Here, and elsewhere, subsets of Πℓ1(Σ) are supplied with the induced topologies from Πℓ1(Σ) unless otherwise stated. Proposition 5.6 contains two more results in this vein. Theorem 5.1. (1) Πℓ1(Σ) is compact. (2) Πℓ1(Σ) (X) is dense in Πℓ1(Σ). (3) The following are equivalent: (a) Πℓ1(Σ) (Per(σ)) = Πℓ1(Σ); (b) Per(σ) = X. (4) The following are equivalent: (a) Πℓ1(Σ) (Per(σ)) is dense in Πℓ1(Σ); (b) Per(σ) is dense in X. (5) The following are equivalent: (a) Πℓ1(Σ) (∞) = Πℓ1(Σ); (b) Aper(σ) = X. (6) The following are equivalent: (a) Πℓ1(Σ) (Aper(σ)) is dense in Πℓ1(Σ); (b) Πℓ1(Σ) (∞) is dense in Πℓ1(Σ); (c) Aper(σ) is dense in X. (a) For all o ∈ OPer(σ) and λ ∈ T, the singleton {Po,λ} is closed in Πℓ1(Σ). (b) For all o ∈ OAper(σ), the singleton {Po} is closed in Πℓ1(Σ) (X) if and (7) only if o is a minimal orbit closure. (8) If, for every infinite dimensional algebraically irreducible representation π of ℓ1(Σ), the closed subset Xπ of X is metrizable (this is certainly the case if X is metrizable or X = Per(σ)), then: (a) Πℓ1(Σ) (X) = Πℓ1(Σ), and Πℓ1(Σ) (X) is compact; (b) If P ∈ Πℓ1(Σ), then the singleton {P } is closed in Πℓ1(Σ) if and only if P is associated with a minimal orbit closure. Proof. Part (1) follows from the fact that ℓ1(Σ) is a unital algebra (see [2, Corol- lary 26.5]). Part (2) through (7) follow from Theorem 4.11, Proposition 4.5, Propo- sition 4.12, Proposition 4.6, Proposition 4.13, Proposition 4.16, and Theorem 4.18, respectively, taking Remark 4.15 and Theorem 3.5 into account where necessary. Part (8)(a) follows from Theorem 4.18.(1) and part (1). Noting that finite orbits are 26 MARCEL DE JEU AND JUN TOMIYAMA minimal orbit closures, part (8)(b) follows from the fact that Πℓ1(Σ) (X) = Πℓ1(Σ), combined with part (7). (cid:3) Before we proceed, we note that we can determine when Πℓ1(Σ) (X) degenerates to its minimal size of one point: this occurs precisely for infinite minimal systems. Note that, for metrizable X, this is equivalent to the degeneracy of the whole primitive ideal space Πℓ1(Σ) = Πℓ1(Σ) (X). Proposition 5.2. The following are equivalent: (1) Πℓ1(Σ) (X) consists of one point; (2) Πℓ1(Σ) (X) = {{0}}; (3) X = Aper(σ), and every orbit is dense. Proof. It follows from Theorem 4.11 that (1) implies (2). If (2) holds, then all points must be aperiodic, since a periodic point yields non-zero primitive ideals; it is then clear that every orbit must be dense. Hence (2) implies (3), and it is obvious that (3) implies (1). (cid:3) Continuing with the main line, we consider natural subsets of Πℓ1(Σ) that are associated with more general invariant subsets of X than X, Per(σ), or Aper(σ), as we have done so far. With notation consistent with that already introduced, we define, for invariant S ⊂ X, Πℓ1(Σ) (S) = {Po,λ : o ∈ OPer(σ), λ ∈ T, o ⊂ S} ∪ {Po : o ∈ OAper(σ), o ⊂ S}. Hence Πℓ1(Σ) (S) consists of all primitive ideals associated with the closures of all orbits contained in S; note that these closures themselves need not be contained in S. For further investigation of such subsets of Πℓ1(Σ), the following lemma is conve- nient. Lemma 5.3. Let S ⊂ X be invariant. Then: Z} ⊂ Po,λ if and only if o ⊂ S. (1) If o ∈ OPer(σ) and λ ∈ T, then {Pn∈Z fnδn ∈ ℓ1(Σ) : fn↾S = 0 for all n ∈ (2) If o ∈ OAper(σ), then {Pn∈Z fnδn ∈ ℓ1(Σ) : fn↾S = 0 for all n ∈ Z} ⊂ Po if and only if o ⊂ S. Proof. For (1), if o 6⊂ S, then the invariance of S and the finiteness of o imply that o ∩ S = ∅. Hence there exists f ∈ C(X) such that f ↾S = 0 and f ↾o 6= 0. Then f ∈ {Pn∈Z fnδn ∈ ℓ1(Σ) : fn↾S = 0 for all n ∈ Z}, but f /∈ Po,λ. Hence {Pn∈Z fnδn ∈ ℓ1(Σ) : fn↾S = 0 for all n ∈ Z} 6⊂ Po. The converse implication in (1) is clear. Part (2) is obvious. (cid:3) For invariant S ⊂ X, we can now describe the closure of Πℓ1(Σ) (S) in Πℓ1(Σ) (X). Proposition 5.4. (1) Let S ⊂ X be invariant. Then the closure of Πℓ1(Σ) (S) in Πℓ1(Σ) (X) is (2) Let S1 ⊂ S2 ⊂ Per(σ) be two invariant subsets. Then: Πℓ1(Σ)(cid:0)S(cid:1). (a) The closure of Πℓ1(Σ) (S1) in Πℓ1(Σ) (S2) is Πℓ1(Σ)(cid:16)S1 is the closure of S1 in S2. S2(cid:17), where S1 S2 (b) Πℓ1(Σ) (S1) is closed in Πℓ1(Σ) (S2) if and only if S1 is closed in S2. ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 27 Proof. For the first statement, we note that it follows from the second part of Proposition 4.8, Proposition 4.10, and continuity, that fnδn ∈ ℓ1(Σ) : fn↾So⊂S k (Πℓ1(Σ) (S)) = k (cid:0)(cid:8)Po,λ : o ∈ OPer(σ), λ ∈ T, o ⊂ S(cid:9) ∪(cid:8)Po : o ∈ OAper(σ), o ⊂ S(cid:9)(cid:1) ButSo⊂S =(Xn∈Z hk (Πℓ1(Σ) (S)) ∩ Πℓ1(Σ) (X) =(cid:8)Po,λ : o ∈ OPer(σ), λ ∈ T, o ⊂ S(cid:9) ∪(cid:8)Po : o ∈ OAper(σ), o ⊂ S(cid:9) . Since S is closed, an orbit is contained in S precisely when its closure is contained o = 0 for all n ∈ Z) . o = S, and hence Lemma 5.3 shows that in S; hence the right hand side equals Πℓ1(Σ)(cid:0)S(cid:1), as claimed. For the second part, we note that, if S1 ⊂ S2 ⊂ Per(σ) are two invariant subsets, then hk (Πℓ1(Σ) (S1)) ∩ Πℓ1(Σ) (S2) = hk (Πℓ1(Σ) (S1)) ∩ Πℓ1(Σ) (X) ∩ Πℓ1(Σ) (S2) =(cid:8)Po,λ : o ∈ OPer(σ), λ ∈ T, o ⊂ S1(cid:9) ∩ Πℓ1(Σ) (S2) =(cid:8)Po,λ : o ∈ OPer(σ), λ ∈ T, o ⊂ S1(cid:9) ∩(cid:8)Po,λ : o ∈ OPer(σ), λ ∈ T, o ⊂ S2(cid:9) =(cid:8)Po,λ : o ∈ OPer(σ), λ ∈ T, o ⊂ S1 ∩ S2(cid:9) = Πℓ1(Σ)(cid:16)S1 S2(cid:17) . This proves the first statement in the second part. For the second statement we need then merely note that the map S 7→ Πℓ1(Σ) (S) is injective on the collection of invariant subsets of Per(σ), as a direct consequence of the first part of Lemma 4.14. (Note, for the sake of completeness, that -- as irrational rotations of T show -- this is not generally true for the collection of invariant subsets of Aper(σ).) (cid:3) Corollary 5.5. Let S ⊂ Per(σ) be invariant. Then the following are equivalent: (1) Πℓ1(Σ) (S) is closed in Πℓ1(Σ) (X); (2) S is closed in X. Proof. Assume that (1) holds. If S is not closed in X, then there is an orbit o in S such that o 6⊂ S. If this is an infinite orbit, then trivially Po /∈ Πℓ1(Σ) (S). The first part of Proposition 5.4, however, implies that Po is in the closure of Πℓ1(Σ) (S) in Πℓ1(Σ) (X). This contradicts the fact that Πℓ1(Σ) (S) is closed in Πℓ1(Σ) (X). If o is a finite orbit, then o = o 6⊂ S, and hence, by the first part of Lemma 4.14, Po,λ /∈ Πℓ1(Σ) (S) for all λ ∈ T. The first part of Proposition 5.4 implies again that Po,λ is in the closure of Πℓ1(Σ) (S) in Πℓ1(Σ) (X), which is again a contradiction. Hence S must be closed in S, and (1) implies (2). It is immediate from the first part of Proposition 5.4 that (2) implies (1). (cid:3) We can now establish the following two additions to the results listed in Theo- rem 5.1. Proposition 5.6. (1) The following are equivalent: (a) Πℓ1(Σ) (Per(σ)) is closed in Πℓ1(Σ) (X); 28 MARCEL DE JEU AND JUN TOMIYAMA (b) Per(σ) is closed in X. (2) The following are equivalent: (a) Πℓ1(Σ) (Aper(σ)) is closed in Πℓ1(Σ) (X); (b) Aper(σ) is closed in X. Proof. Part (1) is a special case of Corollary 5.5. For part (2), it is clear from the first part of Proposition 5.4 that (b) implies (a). If (a) holds, but Aper(σ) is not closed in X, choose a finite orbit o in Per(σ) ∩ Aper(σ). Trivially, for λ ∈ T, Po,λ /∈ Πℓ1(Σ) (Aper(σ)), but the first part of Proposition 5.4 shows that, for λ ∈ T, Po,λ is in the closure of Πℓ1(Σ) (Aper(σ)) in Πℓ1(Σ) (X) . This contradicts the fact that Πℓ1(Σ) (Aper(σ)) is closed in Πℓ1(Σ) (X). (cid:3) If S ⊂ Per(σ) is invariant, we let S/Z be the associated orbit space, supplied with the quotient topology. For the remainder of this section, we shall concentrate, for suitable invariant subsets S of Per(σ), on describing the topology of Πℓ1(Σ) (S) in terms of the topological product S/Z × T (see Theorem 5.10). If o = o ⊂ S is an orbit, we shall write o for the subset of S as well as for the corresponding element of S/Z. The following result on a restriction of the inverse of Ψ in Proposition 4.16 (that can be defined on the whole of Πℓ1(Σ) (X)) relies on a non-trivial result in Fourier analysis. Lemma 5.7. Let p ≥ 1 be an integer, and suppose that S ⊂ Perp(σ) is invariant. Then the restricted inverse map Ψ−1 : Πℓ1(Σ) (S) → S/Z × T, sending Po,λ ∈ Πℓ1(Σ) (S) to (o, λ) ∈ S/Z × T, is a continuous bijection. Proof. It is clear that the map is a bijection, and it remains to show that it is continuous. Let q1, q2 be the canonical projections from S/Z × T onto the first and second factor, respectively. We are to show that q1 ◦ Ψ−1 and q2 ◦ Ψ−1 are continuous. If F ⊂ S/Z is closed, then there exists an invariant SF ⊂ S that is closed in S, and such that F = {o : o ⊂ SF }. Hence (q1 ◦ Ψ−1)−1(F ) = {Po,λ : λ ∈ T, o ⊂ S} = Πℓ1(Σ) (SF ) . Since SF is closed in S, part (2)(b) of Proposition 5.4 implies that Πℓ1(Σ) (SF ) is closed in Πℓ1(Σ) (S). Hence q1 ◦ Ψ−1 is continuous. If F ′ ⊂ T, then(cid:0)q2 ◦ Ψ−1(cid:1)−1 (F ′) = {Po,λ : λ ∈ F ′, o ⊂ S}. If F ′ = ∅ or F ′ = T, then this equals ∅ or Πℓ1(Σ) (S), respectively; hence it is closed in Πℓ1(Σ) (S). If F ′ ⊂ T is a non-trivial closed subset, we argue as follows. Proposition 4.8 shows that k ({Po,λ : λ ∈ F ′, o ⊂ S}) = =(Xn∈Z fnδn ∈ ℓ1(Σ) :Xl∈Z λlflp+j (x) = 0 for j = 0, . . . , p − 1, x ∈ S, and λ ∈ F ′) . Suppose that λ0 /∈ F ′. Then, as a consequence of [15, Theorem 7.1.2.(iii)], the regularity of the Banach algebra ℓ1(Z), and the well known fact that the maxi- mal ideal space of ℓ1(Z) is T with the usual topology, there exists (cn) ∈ ℓ1(Z) 0 cn = 1. Let alp+j = such that Pn∈Z λncn = 0 for all λ ∈ F ′ and Pn∈Z λn cl for j = 0, . . . , p − 1 and l ∈ Z, and put a = Pn∈Z anδn ∈ ℓ1(Σ). Then ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 29 a ∈ k ({Po,λ : λ ∈ F ′, o ⊂ S}), but a /∈ Po,λ0 for all o ⊂ Perp(σ), and in partic- ular for all o ⊂ S. Hence k ({Po,λ : λ ∈ F ′, o ⊂ S}) 6⊂ Po,λ0 for all o ⊂ S, i.e. Po,λ0 /∈ hk ({Po,λ : λ ∈ F ′, o ⊂ S}) for all o ⊂ S. Since λ0 /∈ F ′ was arbitrary, (F ′), i.e. that (cid:3) (F ′)) ∩ Πℓ1(Σ) (S) =(cid:0)q2 ◦ Ψ−1(cid:1)−1 (F ′) is closed in Πℓ1(Σ) (S). Hence q2 ◦ Ψ−1 is continuous. we conclude that hk ((cid:0)q2 ◦ Ψ−1(cid:1)−1 (cid:0)q2 ◦ Ψ−1(cid:1)−1 Corollary 5.8. Let p ≥ 1 be an integer. Suppose that S ⊂ Perp(σ) is invariant and that Πℓ1(Σ) (S) is compact. Then the restricted map Ψ−1 : Πℓ1(Σ) (S) → S/Z × T, sending Po,λ ∈ Πℓ1(Σ) (S) to (o, λ) ∈ S/Z×T, is a homeomorphism between compact Hausdorff spaces. Proof. The space S/Z is Hausdorff. Indeed, if o1, o2 are two different orbits in S, then there exist disjoint (relatively) open subsets U1 and U2 of S such that U1 ⊃ o1 n=0 σn(U2) are two disjoint invariant open subsets of S separating o1 and o2. The images under the (open) quotient map then provide two open disjoint subsets of S/Z separating o1 and o2, as required. and U2 ⊃ o2. ThenTn∈Z σn(U1) =Tp−1 n=0 σn(U1) andTn∈Z σn(U2) =Tp−1 By Lemma 5.7, we see that Ψ−1 : Πℓ1(Σ) (S) → S/Z × T is a continuous bijection between a compact space and a Hausdorff space. Hence it is a homeomorphism. (cid:3) We can now describe the topology of Πℓ1(Σ) (S) for certain invariant S ⊂ Per(σ). sets. i=1 Si, so that Πℓ1(Σ) (S) = Sn Theorem 5.9. Suppose that S1, . . . , Sn ⊂ Per(σ) are mutually disjoint invariant subsets, such that Si ⊂ Perpi (σ) for not necessarily different integers p1, . . . , pn ≥ 1. i=1 Πℓ1(Σ) (Si) as a disjoint union of If each Si is closed in S, and Πℓ1(Σ) (S) is compact, then the map Ψ : i=1 (Si/Z × T) → Πℓ1(Σ) (S), sending (o, λ) to Po,λ, is a homeomorphism between i=1 (Si/Z × T) is the disjoint union of the topo- Let S = Sn Fn compact Hausdorff spaces. Here Fn logical spaces Si/Z × T. Proof. Since each Si is closed in S, part (2)(b) of Proposition 5.4 shows that each Πℓ1(Σ) (Si) is closed in Πℓ1(Σ) (S). Since the latter space is compact by assumption, each Πℓ1(Σ) (Si) is compact. Hence Corollary 5.8 applies, and it shows that each Πℓ1(Σ) (Si) is a compact Hausdorff space. We are now in the situation of part (1) of i=1 Πℓ1(Σ) (Si) is the finite disjoint union of the subsets Πℓ1(Σ) (Si) of Πℓ1(Σ) (S), and each Πℓ1(Σ) (Si) is a closed subset of Πℓ1(Σ) (S) that is a compact Hausdorff space in the induced topology. Hence Πℓ1(Σ) (S) is a compact Hausdorff space, and it is the disjoint union of the topological spaces Πℓ1(Σ) (Si). Since Corollary 5.8 shows that each Πℓ1(Σ) (Si) is homeomorphic to Si/Z × T, the proof is complete. (cid:3) Lemma 2.7, where Πℓ1(Σ) (S) =Sn Πℓ1(Σ) (S) is homeomorphic to (Fn An application of the second and then the third part of Lemma 2.7 shows that i=1 Si is the topological disjoint union of the Si. This will be used in the proof of the following main result on the topological structure of a part of Πℓ1(Σ). i=1 Si) /Z × T, where Fn Theorem 5.10. Suppose that Πℓ1(Σ) (X) is compact; this is certainly the case if X is metrizable or X = Per(σ). Furthermore, assume that S1, . . . , Sn are mutually disjoint invariant closed subsets of X such that Si ⊂ Perpi(σ) for not necessarily i=1 Si. Then the map Ψ : S/Z × T → Πℓ1(Σ) (S), sending (o, λ) to Po,λ, is a homeomorphism between compact Hausdorff different integers p1, . . . , pn ≥ 1. Let S =Sn 30 MARCEL DE JEU AND JUN TOMIYAMA spaces. As a topological space, the codomain is also homeomorphic to the disjoint union of the topological spaces Πℓ1(Σ) (Si), and each such space is homeomorphic to Si/Z × T. Proof. The sufficiency of the conditions in the first sentence for Πℓ1(Σ) (X) to be compact follows from part (8) of Theorem 5.1. We turn to the remaining state- ments. Since the Si are now closed, S is closed, and then Corollary 5.5 shows that Πℓ1(Σ) (S) is closed in Πℓ1(Σ) (X). The latter space is compact by assumption, so that Πℓ1(Σ) (S) is compact. Therefore Theorem 5.9 applies. Combining this with the remark following that theorem, we see that Πℓ1(Σ) (S) is homeomorphic i=1 Si is the topological disjoint union of the Si. An i=1 Si is homeomorphic to (cid:3) to (Fn application of the first part of Lemma 2.7 shows thatFn i=1 Si) /Z × T, whereFn S. This completes the proof. We conclude with two special cases in which there are homeomorphisms as in Theorem 5.10. With Remark 4.19 in mind, the first one can be regarded as an improved version (with topology added) of part of Theorem 3.7. Corollary 5.11. Suppose that X is a finite set. Then the structure space Πℓ1(Σ) of ℓ1(Σ) is homeomorphic to the topological disjoint union of copies of T, one for each orbit in X. Corollary 5.12. Suppose that all orbits in X are of the same finite order. Then the structure space Πℓ1(Σ) of ℓ1(Σ) is homeomorphic to X/Z × T. Example 5.13 (Rotations of T). Let X = T and let σ be the rotation by 2πp/q, where p, q are integers such that q 6= 0 and having greatest common divisor equal to 1. Corollary 5.12 shows that Πℓ1(Σ) is homeomorphic to T/Z × T. For each z0 ∈ T, the orbit of z0 consists of all z ∈ T such that zq = zq 0. The latter implies that the map z 7→ zq from T onto T induces a homeomorphism between T/Z and T. We conclude that Πℓ1(Σ) is homeomorphic to T2. For irrational rotations we had already seen in Remark 4.19 that Πℓ1(Σ) = {{0}}. Remark 5.14. We are not aware of results for the structure space of C∗(Σ) that are the analogues of those for ℓ1(Σ) in the present section. The algebra ℓ1(Σ) is very concretely given, and this makes it more accessible to explicit computations than C∗(Σ). For C∗(Σ), one could conceivably use the generalized Fourier coefficients of its elements as substitutes for the coefficients of the elements of ℓ1(Σ) to work with. As first evidence that such an approach might be successful, we mention that, for rational rotations of T, the structure space of C∗(Σ) is known to be homeomorphic to T2, as would also follow from the C∗(Σ)-analogue of Corollary 5.12. Indeed, C∗(Σ) is strongly Morita equivalent to C(T2) (see [23]), and therefore its structure space is homeomorphic to that of the latter algebra by [22, Corollary 3.30], i.e. to T2. Furthermore, the product of a space of orbits and T occurs in the following con- text. Let p ≥ 1 be an integer, and let Irrp(Σ) be the set of all irreducible unitary representations of C∗(Σ) on a fixed Hilbert space of dimension p, supplied with equivalence classes of irreducible unitary representations of C∗(Σ) of dimension p, supplied with the quotient topology originating from Irrp(Σ). Then, as in Theo- the topology of pointwise strong convergence. Let bAp(Σ) be the set of unitary rem 3.5, there is natural bijection Ξp between Perp(σ)/Z× T and bAp(Σ). According ALGEBRAICALLY IRREDUCIBLE REPRESENTATIONS AND STRUCTURE SPACE 31 to [24, Theorem 4.2.1], this map is a homeomorphism. This result is in the same spirit as Theorem 5.10, but it does not involve the hull-kernel topology of a part of the primitive ideal space as such. We leave the hull-kernel topology on the primitive ideal space of C∗(Σ) for further research. Acknowledgements. The authors are indebted to Kazunori Kodaka for pointing out the result on rotation algebras in Remark 5.14, and to the referee for the very precise reading of the manuscript. This work was partially supported by the Netherlands Organisation for Scientific Research (NWO). References [1] F. Albiac and N.J. Kalton, Topics in Banach space theory, Springer, New York, 2006. [2] F.F. Bonsall and J. Duncan, Complete normed algebras, Springer Verlag, New York- Heidelberg, 1973. [3] K.R. Davidson and E.G. Katsoulis, Isomorphisms between topological conjugacy algebras, J. Reine Angew. Math. 621 (2008), 29 -- 51. [4] K.R. Davidson and E.G. Katsoulis, Nonself-adjoint operator algebras for dynamical systems, in Operator Structures and Dynamical Systems, edited by M. de Jeu, S. Silvestrov, C. Skau, and J. Tomiyama, Contemporary Mathematics 503 (2009), 39-51. [5] K.R. Davidson and E.G. Katsoulis, Operator algebras for multivariable dynamics, Mem. Amer. Math. Soc. 209 (2011), no. 982. [6] S. Dirksen, M. de Jeu, M. Wortel, Crossed products of Banach algebras. I, 2011, to appear in Dissertationes Math., arXiv:1104.5151. [7] J. Dixmier, C ∗-algebras, North-Holland Publishing Co., Amsterdam-New York-Oxford, 1977. [8] M. de Jeu, M. Messerschmidt and M. Wortel, Crossed products of Banach algebras. II, 2013, to appear in Dissertationes Math., arXiv:1305.2304. [9] M. de Jeu and M. Messerschmidt, Crossed products of Banach algebras. III, 2013, to appear in Dissertationes Math., arXiv:1306.6290. [10] M. de Jeu, C. Svensson and J. Tomiyama, On the Banach ∗-algebra crossed product associated with a topological dynamical system, J. Funct. Anal. 262 (2012), 4746 -- 4765. [11] M. de Jeu and J. Tomiyama, Maximal abelian subalgebras and projections in two Banach algebras associated with a topological dynamical system, Studia Math. 208 (2012), 47 -- 75. [12] M. de Jeu and J. Tomiyama, Noncommutative spectral synthesis for the involutive Banach algebra associated with a topological dynamical system, Banach J. Math. Anal. 7 (2013), 103 -- 135. [13] R.V. Kadison, Irreducible operator algebras, Proc. Nat. Acad. Sci. U.S.A. 43 (1957), 273 -- 276. [14] A. Kishimoto and J. Tomiyama, Topologically irreducible representations of the Banach ∗- algebra associated with a dynamical system, 2016, arXiv:1604.02539v1. [15] R. Larsen, Banach algebras. An introduction, Marcel Dekker, New York, 1973. [16] T.W. Palmer, Hermitian Banach ∗-algebras, Bull. Amer. Math. Soc. 78 (1972), 522 -- 524. [17] T.W. Palmer, Banach algebras and the general theory of ∗-algebras. Vol. I. Algebras and Banach algebras, Cambridge University Press, Cambridge, 1994. [18] N.C. Phillips, Analogs of Cuntz algebras on Lp-spaces, 2012, arXiv:1201.4196. [19] N.C. Phillips, Simplicity of UHF and Cuntz algebras on Lp-spaces, 2013, arXiv:1309.0115. [20] N.C. Phillips, Crossed products of Lp operator algebras and the K-theory of Cuntz algebras on Lp spaces, 2013, arXiv:1309.6406. [21] V. Pt´ak, Banach algebras with involution, Manuscripta Math. 6 (1972), 245 -- 290. [22] I. Raeburn and D.P. Williams, Morita equivalence and continuous-trace C ∗-algebras, Math. Surveys and Monographs, vol. 60, Amer. Math. Soc., Providence, 1998. [23] M.A. Rieffel, The cancellation theorem for projective modules over irrational rotation C ∗- algebras, Proc. London Math. Soc. 47 (1983), 285 -- 302. [24] J. Tomiyama, Invitation to C ∗-algebras and topological dynamical systems, World Scientific Publishing Co., Singapore, 1987. 32 MARCEL DE JEU AND JUN TOMIYAMA [25] J. Tomiyama, The interplay between topological dynamics and theory of C ∗-algebras, Lecture Notes Series, Vol. 2, Seoul National University Research Institute of Mathematics Global Analysis Research Center, Seoul, 1992. [26] J. Tomiyama, Structure of ideals and isomorphisms of C ∗-crossed products by single home- omorphisms, Tokyo J. Math. 23 (2000), 1 -- 13. [27] J. Tomiyama, Hulls and kernels from topological dynamical systems and their applications to homeomorphism C ∗-algebras, J. Math. Soc. Japan 56 (2004), 349 -- 364. Marcel de Jeu, Mathematical Institute, Leiden University, P.O. Box 9512, 2300 RA Leiden, the Netherlands E-mail address: [email protected] Jun Tomiyama, Department of Mathematics, Tokyo Metropolitan University, Minami- Osawa, Hachioji City, Japan E-mail address: [email protected]
1612.02733	2	1612	2017-02-17T22:43:36	Regular Dilation on Graph Products of $\mathbb{N}$	[ "math.OA" ]	We extended the definition of regular dilation to graph products of $\mathbb{N}$, which is an important class of quasi-lattice ordered semigroups. Two important results in dilation theory are unified under our result: namely, Brehmer's regular dilation on $\mathbb{N}^k$ and Frazho-Bunce-Popescu's dilation of row contractions. We further show that a representation of a graph product has an isometric Nica-covariant dilation if and only if it is $\ast$-regular. A special case of our result was considered by Popescu, and we studied the connection with Popescu's work.	math.OA	math	REGULAR DILATION ON GRAPH PRODUCTS OF N BOYU LI Abstract. We extended the definition of regular dilation to graph products of N, which is an important class of quasi-lattice ordered semi- groups. Two important results in dilation theory are unified under our result: namely, Brehmer's regular dilation on Nk and Frazho-Bunce- Popescu's dilation of row contractions. We further show that a repre- sentation of a graph product has an isometric Nica-covariant dilation if and only if it is ∗-regular. A special case of our result was considered by Popescu, and we studied the connection with Popescu's work. 1. Introduction Since the celebrated Sz.Nagy dilation theorem that showed a contraction has an isometric dilation, people have been trying to generalize this beau- tiful result to several variables. Ando showed that a pair of commuting contractions has a commuting isometric dilation. However, a well-known example due to Parrott shows a triple of commuting contractions may fail to have a commuting isometric dilation. Many studies consider what kind of extra conditions we need to guarantee a Sz.Nagy type dilation. There are two seemingly opposite directions this paper aims to unify. On one direction, Brehmer [2] showed that if we put some extra condi- tions on the family of commuting contractions, then not only do they have an isometric dilation, but the isometric dilation actually satisfies a stronger condition known as regularity. Brehmer's result has been recently general- ized to representations of any lattice ordered semigroup by the author [14]. In another direction, we may consider non-commutative variables. Sup- pose T1, · · · , Tn are contractions that are not necessarily commuting. It is observed by Frazho-Bunce-Popescu [8, 3, 19] that if T1, · · · , Tn forms a row i ≤ I, then there exists an isometric i=1 TiT ∗ dilation for Ti that is a row contraction. contraction in the sense that Pn These two directions are seemingly unrelated. Indeed, one requires the commuting contractions to satisfy a stronger condition, whereas the other deals with non-commutative contractions. However, in this paper, I will show both results are special cases of ∗-regular dilation on graph products of N. Date: July 11, 2018. 2010 Mathematics Subject Classification. 43A35 ,47A20 ,20F36 . Key words and phrases. regular dilation, graph product, right-angled Artin semigroup, Nica covariant. 1 2 BOYU LI Graph products of N are considered as important examples of quasi-lattice ordered semigroups in [4]. Isometric Nica-covariant representations on a quasi-lattice ordered group have been intensively studied in the past decade. However, contractive Nica-covariant representations have only recently been defined in the lattice ordered group case (see [9, 6]). Lattice ordered groups are quite restrictive compared to quasi-lattice ordered groups, and many in- teresting quasi-lattice ordered semigroups (e.g. the free semigroup F+ n ) are not lattice ordered. This leads to the question for which type of representa- tion on a quasi-lattice ordered group has a minimal isometric Nica-covariant dilation. We answer this question for this special class of quasi-lattice or- dered semigroups of graph products of N and establish that having a minimal isometric Nica-covariant dilation is equivalent to being ∗-regular. Popescu [21] showed that having a minimal isometric Nica-covariant di- lation for a special class operators is equivalent to a property for which he calls the property (P). We extend Popescu's property (P) to the larger class of operators that correspond to representations of graph products of N, and show that the property (P) holds whenever the representation is ∗-regular. 2. Background Throughout this paper, an operator T is understood as a bounded linear operator on a complex Hilbert space H. It is called a contraction if its operator norm kT k ≤ 1, and an isometry if T ∗T = I. If there is a larger Hilbert space K ⊇ H, we say that an operator W ∈ B(K) is a dilation of T if PHW n(cid:12)(cid:12)H = T n for all n ≥ 1. A familiar result due to Sarason [25] states that in such case, K decomposes as H− ⊕ H ⊕ H+, and with respect to this decomposition, W =  0 ∗ 0 ∗ T 0 ∗ ∗ ∗  . In particular, we say that W is an extension of T if H is invariant. In other words, H+ = {0}, and with respect to K = H− ⊕ H, Dually, W is called a co-extension for T if H⊥ is invariant for W . In other 0 W =(cid:20)∗ ∗ T(cid:21) . words, H− = {0}, and with respect to K = H ⊕ H+, ∗(cid:21) . W =(cid:20)T 0 ∗ The celebrated Sz.Nagy dilation states that a contraction T ∈ B(H) has an isometric co-extension. Moreover, the isometric co-extension W can be chosen to be minimal in the sense that K = span{W kh : k ≥ 0, h ∈ H}. We call W a minimal isometric dilation for T . REGULAR DILATION ON GRAPH PRODUCTS OF N 3 There are several attempts to generalize Sz.Nagy's result to the multivari- ate context. Ando [1] proved that a pair of commuting contractions T1, T2 has commuting isometric dilation. However, the generalization to three com- muting contractions fails as Parrott [18] gave a counter-example where three commuting contractions fails to have commuting isometric dilations. Given n contractions T1, · · · , Tn, when can we find certain isometric dilation for them? There are two approaches to this question I would like to discuss in this paper. The first one requires some extra conditions of Ti. Brehmer [2] first considered the question when Ti has a stronger version of isometric dilation, now known as regular dilation. It has since been studied by many authors [12, 30, 10]. It has been generalized to product systems [29, 26], and more recently, to any lattice ordered semigroup [14]. For n = (n1, · · · , nk) ∈ k ) and n− = k ), where x+ = max{0, x} and x− = max{0, −x}. It is clear that Zk, denote T n = Qk . Also denote n+ = (n+ (n− 1 , · · · , n− n = n+ − n−. i=1 T ni i 1 , · · · , n+ Definition 2.1. An isometric dilation (Wi) for (Ti) is called regular if it has an additional property that for any n ∈ Zk, Dually, it is called ∗-regular if for any n ∈ Zk, T ∗n− T n+ = PHW ∗n− T n+ T ∗n− = PHW ∗n− W n+(cid:12)(cid:12)H. W n+(cid:12)(cid:12)H. For every subset J ⊂ {1, · · · , n}, we denote TJ = Qj∈J Tj. Brehmer shows a necessary and sufficient condition for Ti to have a regular (or ∗- regular) dilation: Theorem 2.2 (Brehmer [2]). A commuting n-tuple of contractions T1, · · · , Tn has a regular dilation if and only if for every V ⊂ {1, 2, · · · , k}, we have, (2.1) (−1)U T ∗ U TU ≥ 0 XU ⊂V Here U denotes the cardinality of U . Dually, Ti has a ∗-regular dilation if and only if for every V ⊂ {1, 2, · · · , k}, we have, (2.2) (−1)U TU T ∗ U ≥ 0 XU ⊂V Remark 2.3. Due to Theorem 2.2, the family (Ti) has a regular dilation if and only if (T ∗ i ) has a ∗-regular dilation. To make the notation more consistent with the row contraction condition in the Frazho-Popescu-Bunce dilation, we shall mostly consider the ∗-regular dilation from now on. 4 BOYU LI Condition (2.1) is a much stronger condition than the usual contractive condition that we require for an isometric dilation. For example, given two commuting contractions T1, T2, Ando's theorem always give an isometric dilation for this pair. However, to have a ∗-regular dilation, Brehmer's condition is equivalent of saying I − T1T ∗ 1 ≥ 0. This can be false for many pairs of commuting contractions. 2 + T1T2T ∗ 1 − T2T ∗ 2 T ∗ The other approach to generalize Sz.Nagy's dilation is kind of the opposite to Brehmer's approach. Instead of adding more conditions to commutative Ti, we can replace the commutative condition by the row contractive condi- i ≤ I. Here, Ti are no longer required to be commuting. It is now known as the Frazho-Bunce-Popescu dilation that if T1, · · · , Tn is row contractive, then they can be dilated to row contractive isometries V1, · · · , Vn. tion, which states that Pn i=1 TiT ∗ Row contractive isometries (often called row isometries) is a well-studied subject in the field of operator algebra. For example, the C ∗-algebra gener- ated by row isometry is the well-known Cuntz-Toeplitz algebra. If we require i = I, its C ∗-algebra is the renowned Cuntz-algebra. The W OT -closed non-self-adjoint algebra generated by a row isometry is a free semigroup algebra. further that Pn i=1 ViV ∗ At first glance, Brehmer's result is seemingly unrelated to the Frazho- Bunce-Popescu dilation. Indeed, Brehmer's result is related to represen- tations of the commutative semigroup Nk while the Frazho-Bunce-Popescu dilation is related to representations of the non-commutative free semigroup F+ k . Though regular dilation has been generalized to a larger class of semi- groups called lattice ordered semigroups, the lattice order is a very restrictive requirement. For example, the free semigroup F+ n , and most quasi-lattice ordered semigroups, are not lattice ordered. This paper makes a first attempt to define regular and ∗-regular dilations on a larger class of semigroups, known as graph products of N. It is also called the graph semigroup, or the right-angled Artin monoid (see [4, 7]). It is an important class of quasi-lattice ordered semigroups, recently studied in [4] for its Nica-covariant covariant representations. The main result of this paper generalizes both Brehmer's theorem and the Frazho-Bunce-Popescu dilation to this context. Throughout this paper, we let Γ denote a countable simple graph with In other words, the vertex set Λ is a vertex set Λ and edge set E(Γ). countable set, and every edge e ∈ E(Γ) corresponds to two distinct vertices i, j ∈ Λ. Every edge is undirected, and we say that i is adjacent to j if there is an edge e = (i, j) ∈ E(Γ). The graph product of N, PΓ = Γi∈ΛN, is defined to be the unital semigroup generated by {ei}i∈Λ, with additional rules that ei, ej commutes whenever i is adjacent to j. A representation T : PΓ → B(H) is uniquely determined by its value on the set of generators Ti = T (ei) that satisfies TiTj = TjTi whenever i is adjacent to j. We extend the definition of regularity (see Definition 4.7) to REGULAR DILATION ON GRAPH PRODUCTS OF N 5 representations on PΓ and show that T is ∗-regular if and only if for every finite subset W ⊆ Λ: (2.3) (−1)U TU T ∗ U ≥ 0. XU ⊆W U is a clique Here, a set U ⊆ Λ is called a clique if vertices in U are pairwise ad- jacent to one another. This unifies Brehmer's regular dilation on Nk and Frazho-Bunce-Popescu's dilation of row contractions. Indeed, when Γ is the complete graph, every subset U ⊆ W is a clique. In such case, Condition (2.3) is the same as the ∗-regular Condition (2.2) in the Brehmer's result. When Γ contains no edge, the only cliques in Γ are the singletons. In such i ≤ I and thus a row case, Condition (2.3) is equivalent of saying Pi∈W TiT ∗ contraction as in the Frazho-Bunce-Popescu's dilation. We further consider the question of when a representation of the graph product semigroup PΓ has a minimal isometric Nica-covariant dilation. Ac- cording to [10], a pair of commuting contractions has isometric Nica-covariant dilation if and only if they are ∗-regular. Frazho-Bunce-Popescu's result also shows a row contraction can be dilated to isometries with pair-wise orthogo- nal range, which corresponds to an isometric Nica-covariant dilation on the free semigroup. We show that the ∗-regular condition is equivalent of having an isometric Nica-covariant dilation. To summarize, we establishs the following equivalence: Theorem 2.4. Let T : PΓ → B(H) be a representation. Then the following are equivalent: (1) T is ∗-regular, (2) T has a minimal isometric Nica-covariant dilation, (3) T satisfies Condition (2.3) for every finite subset W ⊆ Λ. 3. Graph Products Fix a simple graph Γ with a countable vertex set Λ. Recall that a graph product of N is a unital semigroup PΓ = Γi∈ΛN, generated by generators {ei}i∈Λ where ei, ej commute whenever i, j are adjacent in Γ. We also call PΓ the graph semigroup or the right-angled Artin monoid. It is also closely related to the Cartier-Foata monoid [13] where ei, ej commute whenever i, j are not adjacent. We can similarly define the graph product of Z, GΓ = Γi∈ΛZ. It is defined to be the free product of Z modulo the rule that elements in the i-th and j-th copies of Z commute whenever (i, j) is an edge of Γ. GΓ is a group, which is also called the graph group or the right-angled Artin group. GΓ together with PΓ is an important example of a quasi-lattice ordered group that is studied by Crisp and Laca [4]. Example 3.1. [Examples of Graph Products] 6 BOYU LI (1) Consider the complete graph Γ that contains every possible edge (i, j) i 6= j. The graph product Γi∈ΛN is equal to the abelian semi- group N+ k , since any two generators ei, ej commute. (2) Consider the graph Γ that contains no edges. The graph product PΓ = Γi∈ΛN is equal to the free product F+ k . (3) Consider the following graph product associated with the graph in Figure 1. 1 • 4 • • 2 • 3 Figure 1. A simple graph of 4 vertices The graph product semigroup is a unital semigroup generated by 4 generators e1, · · · , e4, where the commutation relation is dictated by the edges of the graph. In this example, ei, ej pairwise commute except for the pair e1, e3. A typical element of PΓ is equal to x = x1x2 · · · xn, where each xi belongs to a certain copy of N. We often call x an element of the semigroup, and x1x2 · · · xn an expression of x. Each xi is called a syllable of this expression. We shall denote by I(xi) the index of N to which xi belongs. There might be many equivalent forms for x. First of all, x may be able to be rewritten using fewer syllables: if I(xi) = I(xi+1), we may simply replace xixi+1 = x′ i, and write x using x1 · · · xi−1x′ ixi+2 · · · xn. In particular, if xi = e, we can always treat this e as the identity for I(xi+1) (or I(xi−1)), and thus I(xi) = I(xi+1). This process of merging two adjacent syllable from the same copy of N is called an amalgamation. If I(xi) is adjacent to I(xi+1) in the graph Γ, the commutative rules implies xixi+1 = xi+1xi, and thus x can also be written as x1 · · · xi−1xi+1xixi+2 · · · xn. This process of switching two adjacent syllables when their corresponding copies of N commutes is called a shuffle. We call two elements x = x1 · · · xn and y = y1 · · · ym shuffle equivalent if we can obtain y by repeatedly shuffling x. An expression x = x1 · · · xn is called reduced if it cannot be shuffled to another expression x′ which admits an amalgamation. Similar definitions of amalgamation and shuffle can be made for the graph group Γi∈ΛZ. Lemma 3.2. An expression x = x1 · · · xn is reduced (x can be in either PΓ or GΓ) if and only if for all i < j such that I(xi) = I(xj), there exists an i < t < j so that I(xt) is not adjacent to I(xi). REGULAR DILATION ON GRAPH PRODUCTS OF N 7 The idea is that when I(xi) = I(xj), as long as everything between xi and xj commute with xi and xj, we can shuffle xj to be adjacent to xi and amalgamate the two. It is observed in [11] that reduced expressions are shuffle equivalent: Theorem 3.3 (Green [11]). If x = x1 · · · xn = x′ m are two reduced expressions for x ∈ GΓ (or PΓ). Then two expressions are shuffle equivalent. In particular m = n. 1 · · · x′ This allows us to define the length of an element x to be ℓ(x) = n, when x has a reduced expression x1 · · · xn. Given a reduced expression x = x1 · · · xn, a syllable xi is called an initial syllable if x can be shuffled as x = xix1 · · · xi−1xi+1 · · · xn. Equivalently, it means the vertex I(xi) is adjacent to any previous vertices I(xj), j < i. The vertex I(xi) of an initial syllable is called an initial vertex. The following lemma is partially taken from [4, Lemma 2.3]. Lemma 3.4. Let x = x1 · · · xn be a reduced expression. Then, (1) If i 6= j and xi, xj are two initial syllables, then I(xi) 6= I(xj). (2) The initial vertices of X are pairwise adjacent. (3) Let J = {i : xi is an initial syllable}. Then x = Qj∈J xjQj /∈J xj, where the second product is taken in the same order as in the original expression. Proof. If I(xi) = I(xj) in a reduced expression, by Lemma 3.2, there has to be an index i < t < j so that I(xt) is not adjacent to I(xi) = I(xj). Therefore, it is impossible to shuffle xj to the front. Therefore, any two initial syllables have different vertices. If xi, xj are two initial syllables where i < j. Then to shuffle xj to the front, it must be the case that xj can commute with xi, and thus I(xi) is adjacent with I(xj). This shows initial vertices are pairwise adjacent. Now let J = {1 < j1 < j2 < · · · < jm} be all i where xi is an initial syllable. Then, we can recursively shift each xjs to the front. The result is that we can shuffle all the initial vertices to the front as Qj∈J xj, while all the other syllables are multiplied subsequently in the original order. (cid:3) Lemma 3.4 shows that the initial vertices are pairwise adjacent and thus form a clique of the graph Γ. Lemma 3.4 allows us to further divide a reduced expression of x into blocks. Given a reduced expression x = x1 · · · xn, we define the first block b1 of x to be the product of all initial syllables. Since any two initial syllables commute, there is no ambiguity in the order of this product. We simply denote I1(x) = {i : xi is an initial syllable}, and b1 = Qj∈I1(x) xj. Since x1 is always an initial syllable, I1(x) 6= ∅ and b1 6= e. Now x = b1x(1), where x(1) has strictly shorter length compared to x. We can define the second block b2 of x to be the first block of x(1) when x(1) 6= e. Of course, if x(1) = e, we are finished since x = b1. Repeat this process, 8 BOYU LI and let each x(t) = bt+1x(t+1), where bt+1 is the first block of x(t). Since the length of x(t) is always strictly decreasing, we eventually reach a state when x(m−1) = bmx(m) and x(m) = e. In such case, x is written as a product of m blocks x = b1b2 · · · bm. Here, each bj is the first block of bjbj+1 · · · bm. We call this a block representation of x. We shall denote It(x) be the vertex of all syllables in the t-th block bt. Since any two reduced expressions are shuffle equivalent, it is easy to see this block representation is unique. Lemma 3.5. Let a reduced expression x = x1 · · · xn have a block represen- tation b1 · · · bm (1) Two adjacent It(x), It+1(x) are disjoint. (2) For any vertex λ2 ∈ It+1(x), there exists another vertex λ1 ∈ It(x) so that λ1, λ2 are not adjacent. Proof. For (1), if It(x), It+1(x) share some common vertex δ, then the syl- lable corresponding to δ in the (t + 1)-th block can be shuffled to the front of the (t + 1)-th block, and since δ ∈ It(x), this syllable commutes with all syllable in the t-th block. Therefore, it can be amalgamated into the t-th block, leading to a contradiction that the expression is reduced. For (2), if otherwise, we can pick a vertex λ2 ∈ It+1(x) that is adjacent to every vertex in It(x). The syllable corresponding to λ2 can be shuffled to the front of (t + 1)-th block, and commutes with everything in the t-th block. Therefore, it must be an initial syllable for btbt+1 · · · bm. But in such case, δ ∈ It(x) and cannot be in It+1(x) by (1). (cid:3) Studying regular dilations often requires a deep understanding of elements of the form x−1y for x, y from the semigroup. Lemma 3.6. Let x, y ∈ PΓ. Then, there exists u, v ∈ PΓ with x−1y = u−1v, and I1(u) is disjoint from I1(v). Moreover, u, v are unique. Proof. Suppose that there exists a vertex λ ∈ I1(x)T I1(y). Then we can find initial syllables em1 without loss of generality assume that x1 = em1 from reduced expressions of x, y. We may λ and y1 = em2 λ . λ and em2 λ Set u1 = e− min{m1,m2} y. We have the relation u−1 1 v1 = x−1y. Notice that at least one of x1 and y1 is removed in this process, and thus the total length ℓ(u1)+ℓ(v1) is strictly less than ℓ(x)+ℓ(y). x and v1 = e− min{m1,m2} λ λ Repeat this process whenever I1(uj)T I1(vj) 6= ∅, and recursively define uj+1, vj+1 in the same manner to keep u−1 j+1vj+1. Since the total length uj, vj is strictly decreasing in the process, we eventually stop in a state when I1(uj) is disjoint from I1(vj). This gives a desired u = uj, v = vj. j vj = u−1 REGULAR DILATION ON GRAPH PRODUCTS OF N 9 Suppose that u−1v = s−1t for some other s, t ∈ PΓ with I1(s)T I1(t) = ∅. Let reduced expressions for u, v, s, t be, u = u1 · · · um v = v1 · · · vn s = s1 · · · sl t = t1 · · · tr l · · · s−1 m · · · u−1 We first show u−1v = u−1 1 v1 · · · vn is a reduced expression in GΓ, and so is s−1t = s−1 1 t1 · · · tr. Assume otherwise, by Lemma 3.2, there exists two syllables from the same vertex that commute with everything in between. These two syllables must have one from u and the other from v, since u1 · · · um and v1 · · · vn are both reduced. Let ui, vj be two such syllables that come from the same vertex that commutes with everything in between. In that case, by Lemma 3.4, ui, vj are both initial syllables for u, v. But u, v have no common initial syllables, this leads to a contradiction. · · · s−1 l m · · · u−1 Therefore, u−1 1 v1 · · · vn = s−1 1 t1 · · · tr are both reduced ex- pressions for u−1v = s−1t, and thus by Theorem 3.3 are shuffle equivalent. Notice each individual syllable ui, vi, si, ti is from the graph semigroup. To shuffle from u−1 i must be some u−1 , and ti must be some vj. Therefore, v1 · · · vn must be a shuffle of t1 · · · tr, j and also u1 · · · um is a shuffle of s1 · · · sl. Hence, s = u, t = v. (cid:3) 1 t1 · · · tr, each s−1 1 v1 · · · vn to s−1 m · · · u−1 · · · s−1 l Lemma 3.7. Suppose u, v ∈ Γi∈ΛN. Then the following are equivalent: (1) u, v commute. (2) Every syllable vj of v commutes with u. Proof. (2)=⇒(1) is trivial. Assuming (1) and let v = v1 · · · vm. Consider the first syllable v1 of v. Since uv = vu, v1 is a initial syllable of uv. Therefore, v1 commutes with u. By canceling v1, one can observe that v2 · · · vm also commutes with u, and recursively each vj commutes with u. (cid:3) Lemma 3.8. Suppose p ∈ PΓ, λ ∈ Λ so that λ /∈ I1(p) and eλ does not commute with p. Let x, y ∈ PΓ and apply the procedure in the Lemma 3.6 to repeatedly remove common initial vertex of eλx and py until (eλx)−1py = u−1v with I1(u)T I1(v) = ∅. Then u, v do not commute. Proof. Let p = p1 · · · pn be a reduced expression of p. By Lemma 3.7, there exists a smallest i so that eλ does not commute with pi. We first observe that none of p1, · · · , pi−1 come from the vertex λ. Otherwise, if some ps comes from the vertex λ, it must commute with every p1, · · · , pi−1 as eλ does. Therefore, ps is an initial syllable and λ ∈ I1(p), which contradicts to our assumption. Let pi be a syllable corresponding to vertex λ′, where λ′ is certainly not adjacent to λ. Consider the procedure of removing a common initial vertex for u0 = eλx and v0 = py. At each step, we removed a common initial vertex λi for ui, vi 10 BOYU LI i+1vi+1 = u−1 and obtained u−1 no common initial vertex. It is clear that λ /∈ I1(v0) and λ′ /∈ I1(u0). i vi, until we reach um = u, vm = v that shares Observe that λ0 6= λ′ since λ ∈ I1(eλx) and λ′ cannot be an initial vertex of eλx. Therefore, the syllable pi remains in u1 after the first elimination step, while no syllable before pi belongs to the vertex λ. Hence, λ /∈ I1(v1) and λ′ /∈ I1(u1). Inductively, λ /∈ I1(vj) and λ′ /∈ I1(uj), and thus eλ is still an initial syllable of u and pi is still a syllable of v. Therefore, u, v do not commute. (cid:3) 4. Completely Positive Definite Kernels The problem of finding an isometric dilation turns out to be equivalent to showing that a certain kernel satisfies a so-called completely positive definite condition. Structures of completely positive definite kernels are studied in [20, 22], and we shall restate some of the results to our context. Let P be a unital semigroup sitting inside a group G so that P T P −1 = {e}. For our purpose, the unital semigroup is taken to be a graph product PΓ = Γi∈ΛN, which lives naturally inside GΓ = Γi∈ΛZ. A unital Toeplitz kernel on P is a map K : P ×P → B(H) with the property that K(e, e) = I, K(p, q) = K(q, p)∗, and K(ap, aq) = K(p, q) for all a, p, q ∈ P . We call such a kernel completely positive definite if for any p1, · · · , pn ∈ P and h1, · · · , hn ∈ H, we have n hK(pi, pj)hj, hii ≥ 0. Xi,j=1 Equivalently, this is saying that the n × n operator matrix [K(pi, pj)], viewed as an operator on Hn, is positive. Alternatively, each unital Toeplitz kernel K corresponds to a map K : P −1P → B(H), where K(p−1q) = K(p, q) and K(x−1) = K(x)∗. We shall abbreviate unital completely posi- tive definite Toeplitz kernel as completely positive definite kernel. Existence of a completely positive definite kernel is closely related to the existence of an isometric dilation. A classical result known as Naimark dilation theorem [15] can be restated as the following theorem ([22, Theorem 3.2]): Theorem 4.1. If K is a completely positive definite kernel on a unital semigroup P , then there exists a Hilbert space K ⊃ H and an isometric representation V : P → B(K) so that Moreover, V can be taken as minimal in the sense that K(p, q) = PHV (p)∗V (q)(cid:12)(cid:12)H for all p, q ∈ P. span{V (p)h : p ∈ P, h ∈ H} = K, The minimal isometric representation V is unique up to unitary equiva- lence. REGULAR DILATION ON GRAPH PRODUCTS OF N 11 Assume now that T : P → B(H) is a contractive representation. If we can find a completely positive definite kernel K so that K(e, q) = T (q) for all q ∈ P , then Theorem 4.1 gives us an isometric representation V so Notice that in Theorem 4.1, if we set p = e, we get K(e, q) = PHV (q)(cid:12)(cid:12)H. that T (q) = PHV (q)(cid:12)(cid:12)H. In other words, V is an isometric dilation for T . Corollary 4.2. Let T : P → B(H) be a contractive representation, for which there exists a completely positive definite kernel K so that K(e, q) = T (q). Then T has an isometric dilation V : P → B(K), which can be taken as minimal in the sense that Therefore, we reach the following conclusion: span{V (p)h : p ∈ P, h ∈ H} = K. In particular, each V (p) is a co-extension of T (p). Such a kernel K may not always exist. Indeed, if P = N3, let T send three generators to the three commuting contractions as in the Parrott's example [18]. Such T can never have an isometric dilation and thus there is no completely positive definite kernel K so that K(e, q) = T (q). Even when T has an isometric dilation, K may be extremely hard to define explicitly. Let us now turn our attention to contractive representations on a graph product PΓ = Γi∈ΛN. This semigroup is the free semigroup generated by e1, · · · , en with additional rules that eiej = ejei whenever (i, j) ∈ E(Γ). Therefore, a representation T of PΓ is uniquely determined by its values on generators Ti = T (ei), where they have to satisfy TiTj = TjTi whenever (i, j) ∈ E(Γ). Let us fix a contractive representation T : PΓ → B(H). We start by finding an appropriate completely positive definite kernel for T . Suppose an isometric regular dilation V for T exists, Theorem 4.1 implies that K(p, q) = V (p)∗V (q) = V (p′)∗V (ej )∗V (ej)V (q′) = V (p′)∗V (q′). starts with a syllable in the same copy of N. But since V is isometric, PHV (p)∗V (q)(cid:12)(cid:12)H. Therefore, if I1(p)T I1(q) 6= ∅, p = ejp′, q = ejq′ can both Therefore, it suffices to first consider the case that I1(p)T I1(q) = ∅. If otherwise, Lemma 3.6 gives u, v so that I1(u)T I1(v) = ∅ and u−1v = p−1q. In such case, we shall define K(p, q) = K(u, v). Definition 4.3. Given a contractive representation T of the graph product Γi∈ΛN, we define the Toeplitz kernel K associated with T using the following rules: (1) K(p, q) = T (q)T (p)∗ whenever I1(p)T I1(q) = ∅ and p, q commute. (2) K(p, q) = 0 whenever I1(p)T I1(q) = ∅ and p, q do not commute. (3) If there exists a vertex i = I1(p)T I1(q) 6= ∅. Let p = eip′, q = eiq′. Define K(p, q) = K(p′, q′). Remark 4.4. We may observe that since I1(e) = ∅, and e commutes with any q. K(e, q) = T (q) by (1). Therefore, if K is completely positive definite, the isometric Naimark dilation V will be a dilation for T . 12 BOYU LI Remark 4.5. It follows from Lemma 3.6 that one can recursively remove common initial vertices from p, q using (3), until we end up with unique u, v with u−1v = p−1q and I1(u)T I1(v) = ∅. Therefore, the Definition 4.3 is well-defined for all pairs of p, q. One can verify that the kernel K is indeed a Toeplitz kernel. In fact, it satisfies a stronger property. Lemma 4.6. If p, q, x, y ∈ PΓ satisfies p−1q = x−1y, then K(p, q) = K(x, y). Proof. Repeatedly removing common initial vertices for the pairs p, q and x, y using the procedure in Lemma 3.6, we end up with p−1q = u−1v, x−1y = s−1t, where u, v has no common initial vertex; s, t has no common initial vertex. Then, K(p, q) = K(u, v) and K(x, y) = K(s, t). By Lemma 3.6, u = s, t = v. Therefore, K(p, q) = K(x, y). (cid:3) Definition 4.7. We say that T is ∗-regular if the Toeplitz kernel K asso- ciated with T as defined in Definition 4.3 is completely positive definite. A Naimark dilation V for this kernel K is called a ∗-regular dilation for T . Dually, we say that T is regular if T ∗ is ∗-regular. Here, T ∗(ei) = T (ei)∗. Remark 4.8. Our definition of regular dilation is slightly different from that of Brehmer's. When the graph semigroup is the abelian semigroup Nk, Brehmer defined T to be regular if a kernel K ∗ is completely positive definite, where K ∗ is the Toeplitz kernel by replacing Condition (1) by K ∗(p, q) = T (p)∗T (q). In general, the kernel K ∗ is different from the kernel we defined in Definition 4.3. However, it turns out when the semigroup is the abelian semigroup Nk, our definition of regular dilation (Definition 4.7) coincides with Brehmer's definition (Definition 2.1). However, on a general graph semigroup, when the kernel K ∗ is completely positive definite is hard to characterize. For example, when the graph Γ contains no edge and the graph semigroup corresponds to the free semigroup, the only chance that p, q commute and I1(p)T I1(q) = ∅ is when at least one of p, q is e. Therefore, in such case, K ∗ = K and K ∗ is completelely positive definite whenever K is. Our definition of regular dilation implies there are isometric dilations for T ∗ i and thus co-isometric extensions for Ti. This coincides with the litera- ture on the dilation of row contractions: for example, dilations for column contractions considered by Bunce [3] can be thought as regular dilation on the free semigroup Fk +. The ∗-regular representations are precisely those with a certain minimal Naimark dilation due to Theorem 4.1. Theorem 4.9. T : PΓ → B(H) is ∗-regular if and only if it has a mini- mal isometric Naimark dilation V : PΓ → B(K) so that for all p, q ∈ PΓ, K(p, q) = PHV (p)∗V (q)(cid:12)(cid:12)H. REGULAR DILATION ON GRAPH PRODUCTS OF N 13 Remark 4.10. Given a representation T : PΓ → B(H), there might be kernels different from the kernel we defined in Definition 4.3 that are also completely positive definite. For example, it is pointed out in [17] that when Γ is acyclic, T always has a unitary dilation. By restricting to H, such a unitary dilation defines a completely positive definite kernel that is generally different from the kernel we defined. Popescu [22] has also considered many ways to construct completely positive definite kernels on the free semigroup. The goal of the next two sections is to provide a necessary condition for ∗- regularity of a contractive representation of a graph semigroup, which turns out to be also a sufficient condition. We draw our inspiration from two special cases where the graph is the complete graph and where the graph is the empty graph. Example 4.11. In the case when Γ is a complete graph on k vertices. The graph semigroup PΓ is simply the abelian semigroup Nk. It forms a lattice ordered semigroup. Each element in this semigroup can be written as a k tuple (a1, · · · , ak). Since this semigroup is abelian, the set of initial vertex is precisely {i : ai 6= 0}. Two elements p = (pi), q = (qi) have disjoint initial vertex sets if and only if at least one of pi, qi is zero for all i. In the terminology of the lattice order, this implies the greatest lower bound p ∧ q = e. As it is first defined in [2], a representation T : Nk → B(H) is called ∗-regular if the kernel K(p, q) is completely positive definite. Brehmer's result (Theorem 2.2) shows that K is completely positive def- inite if and only if for every subset V ⊆ {1, 2, · · · , k}, (−1)U TU T ∗ U ≥ 0. XU ⊆V Here U is the cardinality of U , and TU =Qi∈U T (ei) with the convention that T∅ = I. Example 4.12. In the case when Γ is a graph on k vertices with no edge. The graph semigroup Γi∈ΛN is simply the free semigroup F+ k . Fix a con- tractive representation T : F+ k → B(H), which is uniquely determined by its value on generators Ti = T (ei). The Toeplitz kernel associated with T defined in Definition 4.3 is the same as the kernel considered in [20, 22], where it is shown that K is completely positive definite if and only if T is row contractive in the sense that k I − TiT ∗ i ≥ 0. Xi=1 It turns out the minimal Naimark dilation for K in this case is also a row contraction, and thus proves the Frazho-Bunce-Popescu dilation. Inspired by both Example 4.11 and 4.12, our first main result unifies the Brehmer's dilation and the Frazho-Bunce-Popescu dilation. Recall that a 14 BOYU LI set of vertices U ⊆ Λ is called a clique if the subgraph induced on U is a complete subgraph. Theorem 4.13. Let T be a contractive representation of a graph semigroup PΓ. Then, T is ∗-regular if for every W ⊆ Λ, (4.1) (−1)U TU T ∗ U ≥ 0. XU ⊆W U is a clique Remark 4.14. Condition (4.1) coincides with the condition in both Exam- ple 4.11 and 4.12. Indeed, when Γ is a complete graph, any U ⊆ V is a clique. When Γ contains no edge, the only cliques in Γ are singletons {i}. 5. Technical Lemmas Since we are dealing with positive definiteness of operator matrices, the following lemma, taken from [5, Lemma 14.13], is extremely useful. Lemma 5.1. If an operator matrix (cid:20)A B∗ B C(cid:21) ∈ B(H1 ⊕ H2) is positive, then there exists an operator X : H1 → H2 so that B = XA1/2. Moreover, if B has this form, then the operator matrix is positive if and only if C ≥ XX∗. Lemma 5.2. Let X, L ∈ B(H) and X ≥ 0. Define an n × n operator matrix XL∗ XL∗2 XL∗ X An = X . . . Ln−1X Ln−2X · · · LX . . . X LX L2X ...   · · · XL∗(n−1) · · · XL∗(n−2) . . . . . . LX ... XL∗ X .   If LXL∗ ≤ X, then every An is positive. Proof. Assuming LXL∗ ≤ X, we shall inductively show each An is positive. Since the case when n = 1, A1 = X ≥ 0 is given. Suppose An ≥ 0, and rewrite An+1 as An+1 =   An LnX Ln−1X · · · · · · LX XL∗n XL∗(n−1) ... ... XL∗ X .   REGULAR DILATION ON GRAPH PRODUCTS OF N 15 Now notice that the row operator [LnX, · · · , LX] = [0, · · · , 0, L]An. There- fore, by Lemma 5.1, An+1 ≥ 0 if [0, · · · , 0, L]An  0 ... 0 L∗   ≤ X. Expand the left hand side gives LXL∗ ≤ X. (cid:3) Corollary 5.3. The matrix An defined in Lemma 5.2 is positive if and only if A0 = X ≥ 0 and A1 ≥ 0. X Proof. Indeed, A1 = (cid:20) (cid:21) ≥ 0 if and only if X ≥ 0 and (cid:0)LX 1/2(cid:1)(cid:0)X 1/2L(cid:1) = LXL∗ ≤ X by Lemma 5.1. This is sufficient for every An ≥ 0 by Lemma 5.2. LX 1/2X 1/2 X 1/2X 1/2L∗ X (cid:3) We now turn our attention to the contractive representation T of a graph semigroup PΓ = Γi∈ΛN. Throughout this section, we fix such a representa- tion T and its associated Toeplitz kernel K defined in Definition 4.3. For two finite subsets F1, F2 ⊂ PΓ, where F1 = {p1, · · · , pm} and F2 = {q1, · · · , qn}, we denote K[F1, F2] to be the m × n operator matrix, whose (i, j)-entry is equal to K(pi, qj). When F1 = F2, we simply write K[F1] = K[F1, F1]. Recall K is completely positive definite if and only if for all finite subsets F ⊆ PΓ, K[F ] ≥ 0. If F is a collection of elements that may contain dupli- cates, we may similarly define K[F ]. It turns out duplicated elements will not affect the positivity of K[F ]. Lemma 5.4. Let F = {p1, p1, p2, · · · , pm} and F1 = {p1, p2, · · · , pm}. Then K[F ] ≥ 0 if and only if K[F1] ≥ 0. Proof. Denote F2 = {p2, · · · , pm}. We have, K[F ] =  I I I I K[F2, p1] K[F2, p1] K[F2]   . K[p1, F2] K[p1, F2] Here, the lower right corner is K[F1]. By Lemma 5.1, K[F ] ≥ 0 if and only if K[F2, p1]K[p1F2] ≤ K[F2]. By (cid:3) Lemma 5.1 again, this happens if and only if K[F1] ≥ 0. Lemma 5.5. Let F1 = {p1, · · · , pm} and F2 = {q1, · · · , qn} and fix a vertex λ ∈ Λ so that λ is not an initial vertex for any of the pi. Let D(λ, F1) be a diagonal m × m operator matrix whose i-th diagonal entry is equal to T (eλ)m if eλ commutes with pi and 0 otherwise. Then, K[F1, em λ · F2] = D(λ, F1) · K[F1, F2]. Proof. This is essentially proving that K(pi, em commutes with pi and 0 otherwise. λ qj) = T (eλ)mK(pi, qj) if eλ 16 BOYU LI Assuming first that eλ commutes with pi. Then p−1 i qj. A key observation here is that when this happens, pi contains no syllable from the vertex λ. Since eλ commutes with every syllable of pi, if there is a syllable of pi from the vertex λ, it must be an initial syllable, which contradicts to our selection of pi. λ qj = em λ p−1 i em Repeatedly removing common initial vertices for pi, qj using Lemma 3.6, we end up with p−1 i qj = u−1v, where u, v have no common initial vertex. It follows from the Definition 4.3 that K(pi, qj) = K(u, v). Notice that I1(em λ v) includes λ and every vertex in I1(v) that is adjacent to λ. Moreover, we observed that λ /∈ I1(u). Therefore, we have I1(em Suppose u, v commute. Then p−1 i em λ qj) = K(u, em by Lemma 4.6, K(pi, em λ pj = em λ v). Hence, in this case, λ v)T I1(u) = ∅. λ vu−1 = u−1eλv. Therefore, K(u, eλv) = T (eλ)mT (v)T (u)∗ = T (eλ)mK(u, v). If u, v does not commute, em λ v also does not commute with u. Therefore, K(u, v) = K(u, eλv) = 0. Assume now that eλ does not commute with pi. Consider the procedure of removing common initial syllables in pi and em λ qj: since λ is not an initial vertex of pi, each step we have to cancel out a syllable from pi and qj that both commute with em λ . After each step of removing a common initial vertex, we removed some syllable from pi that commute with eλ. Since λ is not an initial vertex of pi, each step will not cancel out any em λ . Eventually, we always end up with p−1 λ v do not share any common initial vertex. λ v, where u, em i qj = u−1em By Lemma 3.7, some syllable in pi does not commute with eλ. Since all the syllables that got canceled commute with eλ, there has to be some syllable in the left over u that does not commute with eλ. Therefore, u and em λ v do not commute. Hence, K(u, em (cid:3) λ v) = 0. As an immediate corollary, Fm =Sm Corollary 5.6. Let F = {p1, · · · , pn} be a finite subset of PΓ, and λ ∈ Λ is a vertex that is not an initial vertex for any of pi. For every m ≥ 0, denote λ · F . Then K[Fm] ≥ 0 if and only if K[F ] ≥ 0 and K[F1] ≥ 0. j=0 ej Proof. For each i ≤ j, K[ei λ F ]. Let D = D(λ, F ) be the n × n diagonal operator matrix, whose (i, i)-entry is T (eλ) if eλ commutes with pi and 0 otherwise. λ F ] = Dj−iK[F ]. Similarly, for each i > j, It follows from Lemma 5.5 that K[F, ej−i λF ] = K[F, ej−i λF, ej K[ei λF, ej λF ] = K[ej λF, ei λF ]∗ = K[F ]D∗(i−j). REGULAR DILATION ON GRAPH PRODUCTS OF N 17 Therefore, K[Fm] = K[F ]D∗ K[F ]D∗2 K[F ]D∗ K[F ] DK[F ] K[F ] D2K[F ] DK[F ] ... . . . DmK[F ] Dm−1X K[F ] . . . · · ·   · · · · · · . . . . . . DK[F ] K[F ]D∗m K[F ]D∗(m−1) ... K[F ]D∗ K[F ] .   Corollary 5.3 can be applied so that K[Fm] ≥ 0 if and only if K[F ] ≥ 0 (cid:3) and K[F1] ≥ 0. Lemma 5.7. Let F1 = {p1, · · · , pn}, F2 = {q1, · · · , qm} be finite subsets of PΓ, and λ ∈ Λ is a vertex that is not an initial vertex for any of pi nor qj. Suppose that eλ commutes with every qj, but not with any pi. Denote, F0 = F1[ F2 F = eλ ·(cid:16)F1[ F2(cid:17)[(cid:16)F1[ F2(cid:17) = eλF0[ F0 F ′ = eλ · F2[ F1[ F2 Then, K[F ] ≥ 0 if and only if K[F ′] ≥ 0. Proof. Let D denote an m × m diagonal operator matrix whose diagonal entries are all T (eλ). Repeatedly apply Lemma 5.5, 0 0 0 K[F1] K[F2] 0 K[F2, F1] K[F1, F2] K[F1, F2]D∗ K[F ] =    Denote the upper left 2 × 2 corner by X = (cid:20) K[F1] K[F1, F2] DK[F2, F1] DK[F2] K[F2, F1] K[F2, F1] K[F2] (cid:21). It K[F2]D∗ K[F1, F2] is clear that X = K[F0]. Let L be a (n + m) × (n + m) diagonal operator matrix, whose first n diagonal entries are 0, and the rest m diagonal entries be T (eλ). Then, the lower left 2 × 2 corner can be written as LX, and K[F1] K[F2] . K[F ] =(cid:20) X XL∗ LX X (cid:21). Lemma 5.2 states that K[F ] ≥ 0 if and only if X = K[F0] ≥ 0 and LXL∗ ≤ X. Explicitly writing out X − LXL∗, we get, K[F1, F2] (5.1) (5.2) Now consider K[F ′]: K[F2, F1] K[F2] − DK[F2]D∗(cid:21) . X − LXL∗ =(cid:20) K[F1] K[F ′] = DK[F2] K[F2, F1] K[F2]   .  K[F2]D∗ K[F1] K[F1, F2] K[F2] 0 0 18 BOYU LI 0 only if K[F2] ≥ 0 and Notice here (cid:20) DK[F2](cid:21) = (cid:20) 0 (cid:20) 0 D(cid:21) K[F2](cid:2)0 D∗(cid:3) =(cid:20)0 D(cid:21) K[F2]. By Lemma 5.1, K[F ′] ≥ 0 if and 0 DK[F2]D∗(cid:21) ≤(cid:20) K[F1] K[F1, F2] K[F2, F1] K[F2] (cid:21) . 0 This is precisely the condition required in Condition (5.1). Therefore, combing the results from above, K[F ] ≥ 0 if and only if K[F ′] ≥ 0, K[F0] ≥ 0 and K[F2] ≥ 0. But notice F0, F2 are subset of F ′, the later condition is equivalent to K[F ′] ≥ 0. (cid:3) 6. Proof of The Main Result We prove the first main result (Theorem 4.13) in this section. The goal is to show that for every finite F = {p1, · · · , pn} ⊂ PΓ, K[F ] ≥ 0 where K is the Toeplitz kernel associated with a contractive representation T : PΓ → B(H) that satisfies Condition (4.1). The plan to prove the main result Theorem 4.13 is divided into 2 steps. In the first step, we define an order on finite subsets of PΓ, and show that for each F ⊂ PΓ, K[F ] ≥ 0 follows from K[F ′] ≥ 0 for some F ′ < F under this order. This allows us to make an induction along finite subsets of PΓ. The base case of the induction turns out to be the case when every element in F has precisely one block. The second step is to show for all such F , K[F ] ≥ 0. Inspired by [14, Section 6], we shall then use an argument to show such K[F ] can be decomposed as RR∗ for some operator matrix R explicitly. For the first step, we show that as long as F contains some element that has more than 1 block, one can find another finite subset F ′ ⊂ PΓ so that K[F ] ≥ 0 if K[F ′] ≥ 0. The key is then to show that this process of finding F ′ will terminate after finitely many steps. Definition 6.1. For each λ ∈ Λ, and p ∈ PΓ, define dλ(p) to be: (1) If p = en1 λ p′ ∈ F where eλ does not commute with p′, then dλ(p) = {p′}. (2) If p = en1 λ p′ ∈ F where eλ commutes with p′, then dλ(p) = {eλp′, p′}. (3) If λ is not an initial vertex of p and eλ does not commute with p, then dλ(p) = {p}. (4) If λ is not an initial vertex of p and eλ commutes with p,, then dλ(p) = {eλp, p}. For any finite set F ⊆ PΓ, denote dλ(F ) =Sp∈F dλ(p). Lemma 6.2. Let F = {p1, · · · , pn} ⊂ PΓ with some pi containing at least 2 blocks. Pick a λ that is an initial vertex for some pi, but eλ does not commute with pi. Then K[F ] ≥ 0 if K[dλ(F )] ≥ 0. REGULAR DILATION ON GRAPH PRODUCTS OF N 19 Proof. Without loss of generality, assume p1 has at least two blocks. First of all, by Lemma 3.5, there exists an initial vertex λ of p1 that is not adjacent to some vertex λ′ in the second block of p1. Therefore, eλ does not commute with p1. We fix this vertex λ, and reorder p1, · · · , pn so that λ is an initial vertex for p1, · · · , pm but not pm+1, · · · , pn. Write pi = eni λ p′ i for all 1 ≤ i ≤ m. Denote F0 = {p′ 1, · · · , p′ m, pm+1, · · · , pn}. None of elements in F0 has λ as an initial vertex. Let N = max{ni} and λ · F0. It is clear that F ⊆ FN , and thus K[F ] ≥ 0 if K[FN ] ≥ 0. By Corollary 5.6, K[FN ] ≥ 0 if and only if K[F1] ≥ 0 where j=0 ej denote FN = SN F1 = (eλ · F0)S F0. We may further split F0 into two subsets F0 = CS N , where C = {f ∈ Now apply Lemma 5.7, K[F1] ≥ 0 if and only if K[(eλ · C)S F0] ≥ 0. Denote F : f commutes with eλ} and N = {f ∈ F : f does not commute with eλ}. F ′ = (eλ · C)[ F0 = (eλ · C)[ C[ N. This proves that K[F ′] ≥ 0 implies K[F ] ≥ 0. To see F ′ = dλ(F ): fix an element pi ∈ F and consider 4 possibilities: (1) If pi = en1 i ∈ F where eλ does not commute with p′ i, then dλ(pi) = λ p′ {p′ i} is contained in N ⊆ F0 ⊆ F ′; (2) If pi = en1 λ p′ i ∈ F where eλ commutes with p′ i, then p′ i is an element of C and thus dλ(pi) = {eλp′ i, p′ (3) If λ is not an initial vertex of pi and eλ does not commute with pi, then pi is in the set N and dλ(pi) = {pi} is contained in N ⊆ F ′; i} is contained in (eλ · C)S C ⊆ F ′; (4) If λ is not an initial vertex of pi and eλ commutes with pi, then pi is in the set C and dλ(pi) = {eλpi, pi} is contained in (eλ · C)S C ⊆ F ′. One can now observe that F ′ = dλ(F ). This finishes the proof. (cid:3) Remark 6.3. One may observe that due to (2) and (4), the set F ′ might be a larger set compared to F . The idea here is we removed eλ where it does not commute with some later syllables, this should make syllables of each element in F ′ more commutative with one another. Therefore repeating this process will end up with an F ′ where every element has only one block. This motivates the Definition 6.4. Definition 6.4. For each element p ∈ PΓ with m blocks, we define the block-vertex sequence of p to be m sets of vertices B1(p), · · · , Bm(p), where B1(p) = {λ ∈ I1(p) : eλ does not commute with p}, and Bj(p) = Ij(p) for all 2 ≤ j ≤ m. In other words, the j-th set is equal to the vertex set of j-th block of p, except for the first block, where we only include any vertex that does not commutes with the rest of the blocks. We also define B0(p) = {λ ∈ I1(p) : eλ commutes with p}, the set of all initial vertices that are adjacent to every vertex that appears in p. Define the block-vertex length of p be c(p) =Pm j=1 Bj(p). 20 BOYU LI Remark 6.5. In the case that p has only one block, then every syllable is initial and thus commuting. In such case, B1(p) = ∅ and c(p) = 0. This is the only case when c(p) = 0. Also observe that for p = em1 λ1 , the power mi ≥ 1 does not affect the block-vertex sequence of p. The only thing that matters is what kind of vertex appears in each block. · · · emn λn In a reduced expression of p, each syllable uniquely corresponds to some j=0 Bj(p). The quantity c(p) = ℓ(p) − B0(p) counts the number of syllables that do not commute with the rest. vertex in one of B0(p), · · · , Bm(p). Therefore, the length ℓ(p) =Pm Lemma 6.6. Let p ∈ PΓ and λ ∈ Λ. (1) If λ ∈ B1(p), and p = en (2) If eλ commutes with p, then the block vertex sequence of any element in dλ(p) is the same as that of p. Here, dλ(p) is defined as in the Definition 6.1. λp′. Then c(p′) < c(p). (3) If eλ does not commute with p and λ is not an initial vertex of p, then the block vertex sequence of any element in dλ(p) is the same as that of p. Proof. For (1), every vertex in B0(p) is still in B0(p′). Since we removed the λ, ℓ(p′) ≤ ℓ(p) − 1, it is observed by Remark 6.5 that c(p′) < c(p). syllable en For (2), there are two cases: either λ ∈ B0(p) or not. In the first case, λp′ and dλ(p) = {p, p′}. Since we only removed an initial vertex write p = en that commutes with the rest of the word, p′ has the same block-vertex sequence as p. In the later case when λ /∈ B0(p), dλ(p) = {p, eλp}. Since eλ commutes with p, λ will be added to B0(eλp) and thus will not change the block-vertex sequence of eλp. In any case, the block vertex sequence of any element in dλ(p) is the same as that of p. For (3), dλ(p) = {p}, and it is clear. (cid:3) Lemma 6.7. If p1, p2 have the same block-vertex sequence, then so does every element of dλ(p1), dλ(p2). Proof. If λ ∈ B1(p1) = B1(p2), write pi = eni i}. Then i is pi with the syllable eni p′ λ removed, and since p1, p2 have the same block- vertex sequence, p′ 2 must also have the same block-vertex sequence. In any other case, by Lemma 6.6, every element in dλ(pi) has the same block- vertex sequence as pi. (cid:3) i and dλ(pi) = {p′ 1, p′ λ p′ Definition 6.8. Let F ⊂ PΓ be a finite set. Define c(F ) = P c(f ), where the summation is over all f ∈ F , but multiple elements with the same block-vertex sequence are only summed once. Lemma 6.9. c(dλ(F )) < c(F ). Proof. Without loss of generality, let f1, · · · , ft have distinct block-vertex sequences while ft+1, · · · , fn have the same block vertex sequence as some REGULAR DILATION ON GRAPH PRODUCTS OF N 21 fi, 1 ≤ i ≤ t, where f1 = p1 = en1 λ p′ 1 and eλ not commuting with p′ 1. Then i=1 c(pi). Now, from Lemma 6.2, λ ∈ B1(p1). Therefore, dλ(f1) = {p′ 1}, and c(p′ 1) < c(f1). Now apply Lemma 6.7, the block-vertex sequence of each dλ(ft+1), · · · , dλ(fn) is the same as that of some dλ(f1), · · · , dλ(ft). More- i=1 dλ(fi), c(F ) =Pt over, by Lemma 6.6, c(dλ(fi)) ≤ c(fi). Therefore, since dλ(F ) =Sn we have, c(dλ(F )) ≤ c(dλ(fi)) < c(fi) = c(F ). (cid:3) t Xi=1 t Xi=1 To summarize the first step towards the proof of the main theorem, Proposition 6.10. For every finite subset F ⊂ PΓ, there exists finite subset F ⊂ PΓ, where every element in F contains exactly one block, and K[F ] ≥ 0 if K[ F ] ≥ 0. Proof. We start with F = F0 and repeatedly apply Lemma 6.2 to ob- tain F1 = dλ(F ), F2 = dλ(F1), · · · . Lemma 6.2 proves that K[Fn] ≥ 0 if K[Fn+1] ≥ 0. Lemma 6.9 shows that c(Fn) is a strictly decreasing integral sequence, and thus must stop at some FN = F . If c( F ) 6= 0, some elements in F has at least 2 blocks and Lemma 6.2 can still be applied to obtain an- other set F ′ = dλ( F ) with c( F ′) < c( F ). Therefore, the last FN = F must have c(FN ) = 0, which is equivalent of saying every element in F contains exactly one block. It is also clear that K[F ] ≥ 0 if K[FN ] ≥ 0. (cid:3) Our second step shall prove that for every finite subset F where every element has exactly one block, K[F ] ≥ 0. Since F only containly finitely many syllables, we may consider only the case when Γ is a finite graph. If an element has exactly one block, then every syllable commutes with all other syllables, and thus their vertices corresponds to a clique in Λ. For a clique U , denote eU =Qλ∈J eλ. Since U is a clique, there is no ambiguity in the order of this product. One exception to the definition is that we shall consider the empty set as a clique as well, and denote e∅ = e. When Γ is a finite graph, there are only finitely many cliques. Denote Fc = {eU : U is a clique}. The first lemma shows that it suffices to prove K[Fc] ≥ 0. Lemma 6.11. If K[Fc] ≥ 0, then for any finite subset F of PΓ whose elements all have one block, K[F ] ≥ 0. λp′ with n ≥ 2, Proof. Suppose F = {p1, · · · , pn} contains an element en then reorder p1, · · · , pn so that λ is an initial vertex for p1, · · · , pm but not pm+1, · · · , pn. Let p′ i be the pi with the syllable corresponding to λ 1, · · · , p′ removed. Let F0 = {p′ m, pm+1, · · · , pn} and let C ⊆ F0 be all elements that commute with eλ. Lemma 6.2 proves that K[F0] ≥ 0 if K[F ′] = K[(eλ · C)S F0] ≥ 0. Since elements in F0 contain exactly one block, and elements in C commute with F0, we have every element in F ′ contains exactly one block. 22 BOYU LI Moreover, each syllable corresponding to the vertex λ is eλ. Repeat this process until we reach F where for all λ, all syllables corresponding to λ are eλ. In such case, every element has the form eU for some clique U . It is clear that F ⊂ Fc and thus if K[Fc] ≥ 0, then K[ F ] ≥ 0 and thus K[F ] ≥ 0. (cid:3) To show K[Fc] ≥ 0, it suffices to show K[Fc] can be decomposed as RcR∗ c . Following the technique outlined in [14, Section 6], we can explicitly find such Rc. Moreover, under a certain ordering, Rc can be chosen to be a lower triangular matrix, and can thus be viewed as a Cholesky decomposition of K[Fc]. This will be done in Proposition 6.14, where we shall see where the conditions in Condition (4.1) come from. From Condition (4.1), denote (6.1) ZV = XU ⊆V U is a clique (−1)U TU T ∗ U ≥ 0. Here, V is any subset of the vertex set Λ, and TU = T (eU ). Assuming Condition (4.1) holds true for a contractive representation T , each ZV ≥ 0 and we can thus take its square root Z 1/2 V ≥ 0. Definition 6.12. For a clique V , we define the neighborhood of V , denoted by NV , to be NV = {λ ∈ Λ : λ /∈ V, and λ is adjacent to every vertex in V }. In particular, we define N∅ = Λ. Lemma 6.13. Fix a clique F , then XF ⊆W W is a clique TW \F ZNW T ∗ W \F = I. Proof. Replace ZNW using Equation (6.1), XW ⊇F W is a clique TW \F ZNW T ∗ W \F XU ⊆NW U is a clique (−1)U TU T ∗ U T ∗ W \F   (−1)U T(U S W )\F T ∗ (U S W )\F = XW ⊇F W is a clique = XW ⊇F W is a clique TW \F   XU ⊆NW   U is a clique   Suppose U ⊆ NW is a clique, then every vertex of U is adjacent to every vertex in W , and vertices in U are adjacent to one another. Therefore, REGULAR DILATION ON GRAPH PRODUCTS OF N 23 US W is also a clique. The converse is true as well: if US W is a clique where UT W = ∅, then U ⊆ NW is a clique. Hence, we can rearrange the double summation so that we first sum over all possible cliques V = US W , and then sum over all possible U . For a fixed clique V = US W , the set W = V \U and the only requirement is that F ⊆ W . Therefore, we only sum those U so that U ⊆ V \F . Rewrite the double summation as: XV =U S W V is a clique (−1)U TV \F T ∗ V \F XU ⊆V \F U is a clique   .   summation over all clique U ⊆ V \F . For a fixed clique V = US W where UT W = ∅, consider the inner and V \F . Moreover, for a fixed size U = k, there are precisely (cid:0)V \F U (cid:1) possibilities for U where U ⊆ V \F with size k. U can take any value between 0 Therefore, the coefficient for TV \F T ∗ V \F where V is a clique containing F , is equal to V \F Xj=0 (cid:18)V \F j (cid:19)(−1)j This summation is equal to 1 if V = F and V \F = 0. Otherwise, this is equal to (1 − 1)V \F = 0. This proves the double summation is equal to TF \F T ∗ (cid:3) F \F = I. We are now ready to show K[Fc] ≥ 0. K[Fc] is a Fc × Fc oper- ator matrix, whose rows and columns are indexed by cliques U, V . Its (U, V )-entry is equal to K[eU , eV ]. Eliminating common initial vertices, K[eU , eV ] = K[eU \V , eV \U ]. Now eU \V commutes with eV \U if and only if all vertices in U \V are adjacent to all vertices in V \U . In other words, US V is a clique. Therefore, we have, K[eU , eV ] =(TV \U T ∗ (6.2) 0, otherwise. U \V , if US V is a clique; Let Rc be a Fc × Fc operator matrix, where (6.3) Rc[U, W ] =(TW \U Z 1/2 0, otherwise. NW , if U ⊆ W c . In particular, K[Fc] ≥ 0. Proposition 6.14. K[Fc] = Rc · R∗ Proof. The (U, V )-entry for Rc · R∗ If US V is not a clique, we cannot find a clique W that contains both U and V . Therefore, for every clique W , we cannot have both U, V contained in W . By Equation (6.3), this implies at least one of Rc[U, W ], Rc[V, W ] is 0. c is equal to PW Rc[U, W ]Rc[V, W ]∗. 24 BOYU LI Hence, the (U, V )-entry for Rc · R∗ of K[Fc] by Equation (6.2). c is 0, which agrees with the (U, V )-entry If US V is a clique, then Rc[U, W ]Rc[V, W ]∗ may be non-zero only when W is a clique containing both U, V . Therefore, in such case, Rc[U, W ]Rc[V, W ]∗ Rc[U, W ]Rc[V, W ]∗ TW \U ZNW T ∗ W \V XW = XU S V ⊆W = XU S V ⊆W =TV \U TW \(U S V )ZNW T ∗  XU S V ⊆W W \(U S V )  T ∗ U \V The summation in the middle is equal to I by Lemma 6.13, in which F c is is the fixed clique US V . This proves that the (U, V )-entry for Rc · R∗ U \V = K[eU , eV ] in this case. equal to TV \U T ∗ Therefore, we conclude that K[Fc] = Rc · R∗ c and K[Fc] ≥ 0. (cid:3) Remark 6.15. We can regard Rc as a Cholesky decomposition of K[Fc] by rearranging Rc as a lower triangular matrix. We first notice that whenever U contains more elements than W , Rc[U, W ] = 0. Moreover, when U = W , U ⊆ W is equivalent to U = W . Therefore, Rc[U, W ] = 0 whenever U ≤ W and U 6= W . Therefore, if we rearrange Fc according to the size of cliques (larger cliques come first), Rc becomes a lower triangular matrix. Example 6.16. Let us consider the graph product of N associated with the graph in Figure 2: 2 • 1 • • 3 Figure 2. A Simple Graph on 3 Vertices The graph semigroup is the unital semigroup generated by e1, e2, e3 where e1, e2 commute. There are 5 cliques in this graph: {1, 2}, {1}, {2}, {3}, and ∅. Under this ordering, K[Fc] = I T2 T1 0 T1T2   T ∗ 2 I T ∗ 2 T1 0 T1 T ∗ 1 T ∗ 1 T2 I 0 T2 0 0 0 I T3 2 T ∗ T ∗ 1 T ∗ 1 T ∗ 2 T ∗ 3 I .   REGULAR DILATION ON GRAPH PRODUCTS OF N 25 We can write out the matrix Rc using Equation (6.3): Rc =   I T2 T1 0 0 Z 1/2 2 0 0 T1T2 T1Z 1/2 2 0 0 Z 1/2 1 0 T2Z 1/2 1 0 0 0 0 0 0 I 0 T3 Z 1/2 {1,2,3} .   One can verify that K[Fc] = Rc · R∗ c . We are now ready to prove the main Theorem 4.13. Theorem 4.13. Let T be a contractive representation of a graph semigroup PΓ. Then, T is ∗-regular if for every finite W ⊆ Λ, (−1)U TU T ∗ U ≥ 0. XU ⊆W U is a clique Proof. To show T is ∗-regular given Condition (4.1), it suffices to prove that the Toeplitz kernel K in Definition 4.3 is completely positive definite. For any finite subset F ⊂ PΓ, it suffices to prove K[F ] ≥ 0. Proposition 6.10 shows that it suffices to prove K[ F ] ≥ 0 for some finite subset F ⊂ PΓ, where each element in F has precisely one block. Let Λ0 be all the vertices that appears in a syllable of some element of F , which is a finite set. Denote Fc = {eJ ∈ Λ0 : J is a clique}. By Lemma 6.11, K[ F ] ≥ 0 if K[Fc] ≥ 0. Finally, by Proposition 6.14, K[Fc] ≥ 0. (cid:3) Remark 6.17. The converse of Theorem 4.13 is also true (see Corollary 7.4). 7. Nica-Covariant Representation on Graph Products Isometric Nica-covariant representations on a quasi-lattice ordered groups are first studied in [16], and were soon found to be an important concept in the study of operator algebras. Isometric Nica-covariant representations on graph semigroups, in particular graph products of N, are intensively studied in [4]. It is observed in [4, Theorem 24] that an isometric representation V of the graph semigroup is isometric Nica-covariant if (1) for any two adjacent vertices i, j, Vi and Vj ∗-commute. (2) for any two non-adjacent vertices i, j, Vi and Vj have orthogonal ranges. In other words, V ∗ i Vj = 0. Contractive Nica-covariant representations on lattice ordered semigroups are first defined and studied in [9, 6]. However, lattice order is quite re- strictive compared to quasi-lattice order. For example, the free semigroup F+ m is quasi-lattice ordered, but not lattice ordered. In particular, the graph product PΓ is only lattice ordered when the graph Γ is the complete graph, which corresponds to the abelian semigroup Nk. This leads to a question of 26 BOYU LI which representations of the graph product PΓ have isometric Nica-covariant dilations. In [10], it is shown that a pair of commuting contractions has a ∗-regular dilation if and only if they have a ∗-commuting isometric dilation, which is an equivalent way of saying a Nica-covariant dilation. The contractive Nica-covariant representations defined in [9, 6, 14] are always ∗-regular. It turns out that ∗-regular is equivalent of having an isometric Nica-covariant dilation. Theorem 7.1. If T : PΓ → B(H) is ∗-regular, then it has a minimal Naimark dilation that is an isometric Nica-covariant representation of the graph semigroup. The minimal Naimark dilation in Theorem 4.1 can be constructed ex- plicitly. We loosely follow the construction in [22, Theorem 3.2]. Given a completely positive definite kernel K : P × P → B(H), define K0 = P ⊗ H with a semi-inner product defined by DX δp ⊗ hp,X δq ⊗ kqE =Xp,q hK(q, p)hp, kqi. The original Hilbert space H can be embedded into K0 as δe ⊗ H. The minimal Naimark dilation V of T acts on the K by V (p)δq ⊗ h = δpq ⊗ h, which are clearly isometries. Moreoever, for any h1, h2 ∈ H, hV (q)∗V (p)h1, h2i =hδp ⊗ h1, δq ⊗ h2i =hK(q, p)h1, h2i Therefore, PHV (q)∗V (p)(cid:12)(cid:12)H = K(q, p). Let N = {k ∈ K0 : hk, ki = 0}. One can show that N is invariant for all V (p), and thus we can let K = K0/N , which is a Hilbert space. V can be defined as isometries on K, and it turns out that it is a minimal Naimark dilation. For technical details, one may refer to [22, Theorem 3.2]. It is worth noting that H is coinvariant for the minimal Naimark dilation V , and thus invariant for V ∗. To prove Theorem 7.1, it suffices to prove that the minimal Naimark dilation is isometric Nica-covariant. Throughout the rest of this section, we fix a contractive representation T on PΓ that is ∗-regular, and let V : PΓ → B(K) be the minimal Naimark dilation for T described as above. Lemma 7.2. Suppose p ∈ PΓ, λ ∈ Λ so that λ /∈ I1(p) and eλ does not commute with p. Then V (eλ) and V (p) have orthogonal ranges. In other words, V (eλ)∗V (p) = 0. Proof. It suffices to prove for any h = Pi δxi ⊗ hi ∈ K0 = PΓ ⊗ H and h =Pj δyj ⊗ ki ∈ K0 = PΓ ⊗ H, hV (p)h, V (eλ)ki = 0. REGULAR DILATION ON GRAPH PRODUCTS OF N 27 By the definition of the pre-inner product on K0, hV (p)h, V (eλ)ki = hXi =Xi,j δp·xi ⊗ hi,Xj δeλ·yj ⊗ kii hK(eλ · yj, p · xi)hi, kj i Suppose (eλ · yj)−1p · xi = u−1v for some u, v ∈ PΓ, where u, v share no common initial vertices. By Lemma 3.8, u, v do not commute. Therefore, K(eλ · yj, p · xi) = 0 for all i, j. Hence, the inner product is equal to 0. (cid:3) Lemma 7.3. Let p ∈ PΓ and λ ∈ Λ be a vertex such that λ /∈ I1(p) and eλ commutes with p. Then V (eλ)∗V (p)(cid:12)(cid:12)H = V (p)V (eλ)∗(cid:12)(cid:12)H Proof. By the minimality of V , span{V (q)k : q ∈ PΓ, k ∈ H} is dense in K. Therefore, it suffices to prove for all q ∈ PΓ, h, k ∈ H, (7.1) hV (eλ)∗V (p)h, V (q)ki = hV (p)V (eλ)∗h, V (q)ki Starting from the left hand side of Equation (7.1), hV (eλ)∗V (p)h, V (q)ki =hV (eλq)∗V (p)h, ki =hK(eλq, p)h, ki =hK(q, p)T (eλ)∗h, ki Here we used Lemma 5.5 to show K(eλq, p) = K(q, p)T (eλ)∗. Now since V (eλ) = (cid:20)T (eλ) 0 ∗(cid:21) with respect to the decomposition K = H ⊕ H⊥, V (eλ)∗h = T (eλ)∗h ∈ H. Therefore, ∗ hK(q, p)T (eλ)∗h, ki =hK(q, p)V (eλ)∗h, ki =hV (q)∗V (p)V (eλ)∗h, ki =hV (p)V (eλ)∗h, V (q)ki This proves Equation (7.1). (cid:3) We now prove the main result of this section: Proof of Theorem 7.1. It suffices to pick any two vertices λ1, λ2 and consider two cases when they are adjacent or not. If λ1, λ2 are not adjacent, by Lemma 7.2, V (eλ1 ) and V (eλ2 ) are isometries with orthogonal ranges. If λ1, λ2 are adjacent, it suffices to prove for all p ∈ PΓ, Indeed, since span{V (p)h : p ∈ PΓ, h ∈ H} is dense in K, Equation (7.2) V (eλ1 )∗V (eλ2)V (p)(cid:12)(cid:12)H = V (eλ2 )V (eλ1 )∗V (p)(cid:12)(cid:12)H. (7.2) implies that V (eλ1 )∗V (eλ2 ) = V (eλ2 )V (eλ1)∗. There are now several possibilities: 28 BOYU LI If λ ∈ I1(p), we can write p = eλ1 p′, and thus V (p) = V (eλ1)V (p′). Since λ1, λ2 are adjacent, V (eλ1 ) commutes with V (eλ2 ). Hence, both sides of the Equation (7.2) are equal to V (eλ2 )V (p′)(cid:12)(cid:12)H. If λ /∈ I1(p) and eλ1 does not commute with p, then λ1 /∈ I1(eλ2 p) and eλ1 does not commute with eλ2p as well. Therefore, by Lemma 7.2, V (eλ1 ) and V (p) are isometries with orthogonal ranges, and V (eλ1)∗V (p) = 0. Similarly, V (eλ1 ) and V (eλ2p) are isometries with orthogonal ranges, and V (eλ1)∗V (eλ2 p) = 0. Both sides of the Equation (7.2) are 0. Lastly, if λ /∈ I1(p) and eλ1 commutes with p. Then eλ2p and p are both element in PΓ that commutes with eλ1 without λ1 as an initial vertex. By Lemma 7.3, for every h ∈ H, V (eλ1 )∗V (eλ2 )V (p)h =V (eλ2 )V (p)V (eλ1 )∗h =V (eλ2 )V (eλ1 )∗V (p)h This is precisely the Equation (7.2), and thus we finished the proof. (cid:3) Corollary 7.4. If T is has a minimal isometric Nica-covariant dilation, then, (−1)U TU T ∗ U ≥ 0. XU ⊆W U is a clique Proof. Let V : PΓ → B(K) be the minimal Naimark dilation for T . We have H is co-invariant for V , and thus with respect to the decomposition K = H ⊕ H⊥, V (p) =(cid:20)T (p) 0 ∗ ∗(cid:21). Therefore, for every clique U in Γ, U = PHV (eU )V (eU )∗(cid:12)(cid:12)H. TU T ∗ It suffices to show for every W ⊆ Λ, (7.3) (−1)U V (eU )V (eU )∗ ≥ 0. XU ⊆W U is a clique For each vertex i ∈ Λ, denote Pi = V (ei)V (ei)∗ the range projection of the isometry V (ei). Since V is Nica-covariant, Pi, Pj commutes and PiPj =(ViVjV ∗ 0, otherwise. j V ∗ i , if i is adjacent to j; For each U ⊆ W , denote PU = Qi∈U Pi and in particular let P∅ = I. If U ⊆ W is not a clique, then we can find two vertices i, j ∈ U that are not adjacent. Since PiPj = 0, it follows that PU = 0. If U ⊆ W is a clique, then it follows from that Nica-covariant condition that PU = V (eU )V (eU )∗. REGULAR DILATION ON GRAPH PRODUCTS OF N 29 Consider the projection R =Qi∈W (I − Pi): (I − Pi) R = Yi∈W = XU ⊆W = XU ⊆W = XU ⊆W U is a clique U is a clique (−1)U PU (−1)U PU (−1)U V (eU )V (eU )∗. Since R is a projection, R ≥ 0 and this proves the Condition (7.3). (cid:3) We have now established the equivalence among Condition (2.3), ∗-regular, and having a minimal isometric Nica-covariant dilation. Theorem 2.4. Let T : PΓ → B(H) be a representation. Then the following are equivalent: (1) T is ∗-regular, (2) T has a minimal isometric Nica-covariant dilation, (3) T satisfies Condition (2.3). Proof. (1) =⇒ (2) is established in Theorem 7.1. (1) =⇒ (2) is established in Corollary 7.4. Finally, (3) =⇒ (1) is established in Theorem 4.13. (cid:3) 8. The Property (P) Popescu [21] first studied the noncommutative Poisson transform asso- ciated to a certain class of operators that satisfies the property (P). The property (P) has recently been generalized to higher rank graphs [27, 28]. It turns out that the class of operators Popescu studied can be viewed as a representation of a graph product of N, and we thereby extend the Prop- erty (P) to representations of graph products of N. This section proves that ∗-regular condition implies the property (P), and they are equivalent under certain conditions. Throughout this section, we fix a finite simple graph Γ whose vertex set is denoted by Λ. Definition 8.1. A contractive representation T : PΓ → B(H) is said to have the Property (P) if there exists 0 ≤ ρ < 1 so that for all ρ ≤ r ≤ 1, (8.1) XU ⊆Λ U is a clique (−1)U rU T (eU )T (eU )∗ ≥ 0. 30 BOYU LI Example 8.2. Let Γ be a complete k-partite graph Kn1,n2,··· ,nk . In other words, denote Λ = {(i, j) : 1 ≤ i ≤ k, 1 ≤ j ≤ ni} be the vertex set, and (i1, j1) is adjacent to (i2, j2) in Γ if and only if i1 6= i2. A contrac- tive representation T of this graph semigroup PΓ is uniquely determined by Ti,j = T (ei,j). Here, for each i, Ti,1, · · · , Ti,ni are not necessarily commuting contractions. However, for each i1 6= i2, Ti1,j1 commutes with Ti2,j2. In [21], Popescu considered such class of operators {Ti,j} where for each i, {Ti,j}nj j=1 forms a row contraction in the sense that, Ti,jT ∗ i,j ≤ I. ni Xj=1 This family of operators is also considered in many subsequent papers on non-commutative polyballs (see also [23, 24]). For such family of operators, Popescu says it has the property (P) if the Condition (8.1) is satisfied. It is observed in [21] that the property (P) allows one to obtain a Poisson transform and subsequently a dilation of the family of operators {Ti,j}. One may observe that Definition 8.1 of the property (P) does not re- quire the row contractive condition. Instead, this paper mostly considers a contractive representation T of the graph product PΓ that satisfies the Con- dition (2.3) and thus has a ∗-regular dilation. The row contractive condition is embedded in the Condition (2.3). Our first result shows that if T satisfies the Condition (2.3), then it has the property (P). Let T : PΓ → B(H) be a representation that satisfies the Condition (2.3). By Theorem 2.4, it has a minimal isometric Nica-covariant dilation V : PΓ → B(K). Moreover, H is co-invariant for V , and thus Therefore, to show T has the property (P), it suffices to show V has the property (P). For r ∈ R, let us denote PHV (eU )V (eU )∗(cid:12)(cid:12)H = T (eU )T (eU )∗. f (r) = XU ⊆Λ U is a clique (−1)U rU V (eU )V (eU )∗. It follows from the proof of Corollary 7.4 that f (1) ≥ 0. In fact, f (1) is a projection onto the subspace that is orthogonal to all the ranges of V (ei). Following the notation we used in the proof of Corollary 7.4, for each vertex i ∈ Λ, denote Pi = ViV ∗ i . Since V is Nica-covariant, Pi, Pj commute, and PiPj =(ViVjV ∗ 0, otherwise. j V ∗ i , if i is adjacent to j; For each U ⊆ Λ, denote PU = Qi∈U Pi, the projection onto the intersec- tion of the ranges of all {Pi}i∈U . In particular, we let P∅ = I. Notice that if there are two vertices i, j ∈ U that are not adjacent, PiPj = 0 and thus REGULAR DILATION ON GRAPH PRODUCTS OF N 31 PU = 0. Therefore, PU 6= 0 only if U is a clique. The function f (r) can be rewritten as f (r) = XU ⊆Λ U is a clique (−1)U rU PU Λ = XU ⊆Λ   XU ⊆Λ Xk=0 = (−1)U rU PU (−1)kPU  rk U =k For each U ⊆ Λ, denote RU = PU ·Qi /∈U P ⊥ Pi where i /∈ U . In particular, R∅ =Qi∈Λ P ⊥ i . The range of RU are those vectors that are contained in the range of PU but orthogonal to the range of i , which is the projection onto those vectors that are orthogonal to the ranges of all Pi. It was observed in Corollary 7.4 that R∅ = XU ⊆Λ U is a clique (−1)U V (eU )V (eU )∗ = f (1). Finally, denote (8.2) Qm = XU ⊆Λ U =m RU . In particular, Q0 = R∅ = f (1). Notice that if two distinct subsets U1, U2 ⊆ Λ and U1 = U2 = m, then at least one vertex in U1 is not in U2 and vice versa. Therefore, RU1RU2 = 0 and thus RU1, RU2 are projec- tions onto orthogonal subspaces. Hence, Qm is a projection. Intuitively, the range of Qm are those vectors that are contained in the range of m of Pi and orthogonal to the range of all other Pi. Therefore, {Qm}Λ m=0 are pairwise orthogonal projections and Λ Xm=0 Qm = I. 32 BOYU LI We first obtain a Taylor expansion of f about r = 1. For each 1 ≤ m ≤ Λ, the m-th derivative of f is equal to: f (m)(r) = Λ Xk=m XU ⊆Λ U =k (−1)k k! (k − m)! rk−mPU = (−1)mm! Λ Xk=m XU ⊆Λ U =k (−1)k−m(cid:18) k m(cid:19)rk−mPU Lemma 8.3. f (m)(1) = (−1)mm! · Qm. Moreover, f has the Taylor series expansion f (r) = Λ Xm=0 (−1)m(r − 1)mQm. Proof. It suffices to prove Qm = Λ Xk=m XU ⊆Λ U =k (−1)k−m(cid:18) k m(cid:19)PU . Denote the right hand side of the summation Sm. It suffices to prove SmQi = QiSm =(Qi, if i = m; 0, if i 6= m. From Equation (8.2), Qm is the sum of all RW where W = m. Since {RW }W =m are pairwise orthogonal projections, it suffices to prove SmRW = RW Sm =(RW , if W = m; 0, if W 6= m. First of all, since {Pi}i∈Λ are commuting orthogonal projections, RW , Sm commute for all W ⊆ Λ and 0 ≤ m ≤ Λ. Fix W and consider SmRW . If W < m, then every U ≥ m contains some vertex not in W . There- fore, PU RW = 0, and hence SmRW = 0. If W ≥ m, then for each U ≥ m, PU RW =(RW , if U ⊆ W ; 0, otherwise. REGULAR DILATION ON GRAPH PRODUCTS OF N 33 Therefore, Λ SmRW = Xk=m XU ⊆Λ  Xk=m XU ⊆W W = U =k U =k (−1)k−m(cid:18) k (−1)k−m(cid:18) k · RW m(cid:19)PU  m(cid:19)RW W = = W Xk=m Xk=m =(cid:18)W m (cid:19) =(cid:18)W m (cid:19) (−1)j(cid:0)W −m j (−1)k−m(cid:18)W k (cid:19)(cid:18) k m(cid:19)(cid:18) k m(cid:19)RW (−1)k−m W ! k! k!(W − k)! m!(k − m)! RW W W −m k − m (cid:19)RW (cid:19)RW . (−1)k−m(cid:18)W − m Xk=m (−1)j(cid:18)W − m Xj=0 (cid:1) is equal to (1 − 1)W −m = 0 if W > m, and j Here, PW −m j=0 1 if W = m. Therefore, SmRW =(RW , if W = m; 0, otherwise. This proves Sm = Qm. Since the graph Γ is assumed to be a finite graph, f (r) is a finite operator-valued polynomial. Its Taylor series expansion about 1 is equal to: f (r) = = Λ Λ Xm=0 Xm=0 f (m)(1) m! (r − 1)m (−1)m(r − 1)mQm. (cid:3) Theorem 8.4. If a representation T : PΓ → B(H) is ∗-regular, then T satisfies the property (P). Moreover, the constant ρ in the property (P) can be chosen to be ρ = 0. 34 BOYU LI Proof. Let V : PΓ → B(K) be the minimal isometric ∗-regular dilation for T . By Lemma 8.3, for each 0 ≤ r ≤ 1, (−1)U rU PU f (r) = XU ⊆Λ U is a clique Λ (−1)m(r − 1)mQm = Xm=0 For 0 ≤ r ≤ 1, (−1)m(r − 1)m ≥ 0. Since each Qm is an orthogonal U , where VU = projection, f (r) ≥ 0. Notice when U is a clique, PU = VU V ∗ ∗(cid:21) with respect to K = H ⊕ H⊥. Therefore, by projecting onto the corner corresponding to H, we obtain that for all 0 ≤ r ≤ 1, (cid:20)TU 0 ∗ (−1)U rU TU T ∗ U ≥ 0. XU ⊆Λ U is a clique This implies T satisfies the property (P) with ρ = 0. (cid:3) It is not clear when the converse of Theorem 8.4 also holds. Popescu established in [21, Corollary 5.2] the converse for a special class of operators. Proposition 8.5 (Corollary 5.2, [21]). Let Γ = Kn1,··· ,nk be a complete k- multipartite graph. Let {Ti,j ∈ B(H) : 1 ≤ i ≤ k, 1 ≤ j ≤ ni} be a family of operators such that: (1) For each i, Pni j=1 Ti,jT ∗ i,j ≤ I, (2) The associated representation T : PΓ → B(H) has the property (P). Then the associated representation T has a minimal isometric Nica-covariant dilation. However, for a representation of an arbitrary graph semigroup, it is not clear how one can replace the Condition (1) in the Proposition 8.5. Example 8.6. Let us consider the special case when n1 = · · · = nk = 1 and the graph Γ is the complete graph on k-vertices. Let {Ti}k i=1 be a family of operators as in the Proposition 8.5. Notice the Condition (1) is simply saying that each Ti is a contraction. Proposition 8.5 states that such Ti has a minimal isometric Nica-covariant dilation, and thus by the Theorem 2.4, Ti has to satisfy the Condition (2.3). Note that in a complete graph, Condition 2.3 is the same as the Brehmer's Condition (2.1). In fact, we can derive the Condition (2.3) directly from the property (P), without invoking the minimal isometric Nica-covariant dilation. For any subset W ⊆ {1, 2, · · · , n}, denote ∆W (r) = XU ⊆W (−1)U rU TU T ∗ U . REGULAR DILATION ON GRAPH PRODUCTS OF N 35 The property (P) implies for some 0 ≤ ρ < 1 and all ρ ≤ r ≤ 1, ∆{1,2,··· ,n}(r) ≥ 0. For any 1 ≤ i ≤ n, let Wi = {1, · · · , i − 1, i + 1, · · · , n}. Notice that, ∆{1,2,··· ,n}(r) = ∆Wi(r) − rTi∆Wi(r)T ∗ i . We claim that ∆Wi(r) ≥ 0 for all ρ ≤ r < 1. If otherwise, since ∆Wi(r) is a self-adjoint operator, let −M = inf{h∆Wi(r)h, hi : khk = 1} < 0. Pick a unit vector h so that −M ≤ h∆Wi(r)h, hi < −M · r. Then, hrTi∆Wi(r)T ∗ i h, hi = r · h∆Wi(r)T ∗ i h, T ∗ i hi Therefore, ≥ −M · r. h∆{1,2,··· ,n}(r)h, hi = h∆Wi(r)h, hi − hrTi∆Wi(r)T ∗ i h, hi < −M · r + M · r = 0. This contradicts that ∆{1,2,··· ,n}(r) ≥ 0. Hence, we can conclude that ∆Wi(r) ≥ 0. In other words, {T1, · · · , Ti−1, Ti+1, · · · , Tn} satisfies the prop- erty (P). Similarly, by removing one element each time, we obtain that for any W ⊆ {1, 2, · · · , n}, ∆W (r) ≥ 0 for all ρ ≤ r < 1. In particular, let r → 1, we obtain that for every W ⊆ {1, 2, · · · , n}, (−1)U TU T ∗ U ≥ 0. XU ⊆W This is exactly the Condition (2.3) on the complete graph (equivalently, the Brehmer's Condition (2.1)). Remark 8.7. For an arbitrary graph Γ, it is not clear how we can replace the Condition (2) in the Proposition 8.5 to guarantee a minimal isometric Nica-covariant dilation for a representation T : PΓ → B(H). References [1] T. Ando. On a pair of commutative contractions. Acta Sci. Math. (Szeged), 24:88 -- 90, 1963. [2] S. Brehmer. Uber vetauschbare Kontraktionen des Hilbertschen Raumes. Acta Sci. Math. Szeged, 22:106 -- 111, 1961. [3] J. W. Bunce. Models for n-tuples of noncommuting operators. J. Funct. Anal., 57(1):21 -- 30, 1984. [4] J. Crisp and M. Laca. On the Toeplitz algebras of right-angled and finite-type Artin groups. J. Aust. Math. Soc., 72(2):223 -- 245, 2002. [5] K. R. Davidson. Nest algebras, volume 191 of Pitman Research Notes in Mathematics Series. Longman Scientific & Technical, Harlow; copublished in the United States with John Wiley & Sons, Inc., New York, 1988. Triangular forms for operator algebras on Hilbert space. [6] K. R. Davidson, A. Fuller, and E. Kakariadis. Semicrossed products of operator al- gebras by semigroups. arXiv preprint arXiv:1404.1906, 2014. [7] S. Eilers, X. Li, and E. Ruiz. The isomorphism problem for semigroup C ∗-algebras of right-angled Artin monoids. Doc. Math., 21:309 -- 343, 2016. 36 BOYU LI [8] A. E. Frazho. Models for noncommuting operators. J. Funct. Anal., 48(1):1 -- 11, 1982. [9] A. H. Fuller. Nonself-adjoint semicrossed products by abelian semigroups. Canad. J. Math., 65(4):768 -- 782, 2013. [10] D. Ga¸spar and N. Suciu. On the intertwinings of regular dilations. Ann. Polon. Math., 66:105 -- 121, 1997. Volume dedicated to the memory of W lodzimierz Mlak. [11] E. Green. Graph Products of Groups. PhD thesis, University of Leeds, 1990. [12] I. Halperin. Sz.-Nagy-Brehmer dilations. Acta Sci. Math. (Szeged), 23:279 -- 289, 1962. [13] C. Krattenthaler. The theory of heaps and the Cartier-Foata monoid. Appendix of the electronic edition of Problemes combinatoires de commutation et r´earrangements, 2006. [14] B. Li. Regular representations of lattice ordered semigroups. J. Operator Theory, 76(1):33 -- 56, 2016. [15] M. Neumark. Positive definite operator functions on a commutative group. Bull. Acad. Sci. URSS S´er. Math. [Izvestia Akad. Nauk SSSR], 7:237 -- 244, 1943. [16] A. Nica. C ∗-algebras generated by isometries and Wiener-Hopf operators. J. Operator Theory, 27(1):17 -- 52, 1992. [17] D. Opela. A generalization of Ando's theorem and Parrott's example. Proc. Amer. Math. Soc., 134(9):2703 -- 2710 (electronic), 2006. [18] S. Parrott. Unitary dilations for commuting contractions. Pacific J. Math., 34:481 -- 490, 1970. [19] G. Popescu. Isometric dilations for infinite sequences of noncommuting operators. Trans. Amer. Math. Soc., 316(2):523 -- 536, 1989. [20] G. Popescu. Positive-definite functions on free semigroups. Canad. J. Math., 48(4):887 -- 896, 1996. [21] G. Popescu. Poisson transforms on some C ∗-algebras generated by isometries. J. Funct. Anal., 161(1):27 -- 61, 1999. [22] G. Popescu. Positive definite kernels on free product semigroups and universal alge- bras. Math. Scand., 84(1):137 -- 160, 1999. [23] G. Popescu. Curvature invariant on noncommutative polyballs. Adv. Math., 279:104 -- 158, 2015. [24] G. Popescu. Holomorphic automorphisms of noncommutative polyballs. J. Operator Theory, 76(2):4357 -- 4416, 2016. [25] D. Sarason. Invariant subspaces and unstarred operator algebras. Pacific J. Math., 17:511 -- 517, 1966. [26] O. M. Shalit. Representing a product system representation as a contractive semi- group and applications to regular isometric dilations. Canad. Math. Bull., 53(3):550 -- 563, 2010. [27] A. Skalski. On isometric dilations of product systems of C ∗-correspondences and applications to families of contractions associated to higher-rank graphs. Indiana Univ. Math. J., 58(5):2227 -- 2252, 2009. [28] A. Skalski and J. Zacharias. Poisson transform for higher-rank graph algebras and its applications. J. Operator Theory, 63(2):425 -- 454, 2010. [29] B. Solel. Regular dilations of representations of product systems. Math. Proc. R. Ir. Acad., 108(1):89 -- 110, 2008. [30] B. Sz.-Nagy and C. Foia s. Harmonic analysis of operators on Hilbert space. Translated ' from the French and revised. North-Holland Publishing Co., Amsterdam-London; American Elsevier Publishing Co., Inc., New York; Akad´emiai Kiad´o, Budapest, 1970. Pure Mathematics Department, University of Waterloo, Waterloo, ON, Canada N2L -- 3G1 E-mail address: [email protected]
1809.05154	2	1809	2018-12-02T06:16:59	Compressions of compact tuples	[ "math.OA", "math.FA" ]	We study the matrix range of a tuple of compact operators on a Hilbert space and examine the notions of minimal, nonsingular, and fully compressed tuples. In this pursuit, we refine previous results by characterizing nonsingular compact tuples in terms of matrix extreme points of the matrix range. Further, we find that a compact tuple $A$ is fully compressed if and only if it is multiplicity-free and the Shilov ideal is trivial, which occurs if and only if $A$ is minimal and nonsingular. Fully compressed compact tuples are therefore uniquely determined up to unitary equivalence by their matrix ranges. We also produce a proof of this fact which does not depend on the concept of nonsingularity.	math.OA	math	COMPRESSIONS OF COMPACT TUPLES BENJAMIN PASSER AND ORR MOSHE SHALIT Abstract. We study the matrix range of a tuple of compact operators on a Hilbert space and examine the notions of minimal, nonsingular, and fully compressed tuples. In this pursuit, we refine previous results by characterizing nonsingular compact tuples in terms of matrix extreme points of the matrix range. Further, we find that a compact tuple A is fully compressed if and only if it is multiplicity-free and the Shilov ideal is trivial, which occurs if and only if A is minimal and nonsingular. Fully compressed compact tuples are therefore uniquely determined up to unitary equivalence by their matrix ranges. We also produce a proof of this fact which does not depend on the concept of nonsingularity. 1. Background and statement of main results Let A = (A1, . . . , Ad) ∈ B(H)d be a d-tuple of bounded operators on a Hilbert space H. We write SA for the operator system generated by A, and UCP(SA, Mn) for the space of all unital completely positive maps from SA into the algebra Mn = Mn(C) of n × n matrices. The matrix range of A is the matrix convex set W(A) = ⊔∞ n=1Wn(A) ⊆ ⊔∞ n, where n=1M d Wn(A) := {φ(A) : φ ∈ UCP(SA, Mn)} ⊆ M d n, and φ(A) := (φ(A1), . . . , φ(Ad)). The matrix range of a single operator was introduced by Arveson in [2, 3], and the matrix range of a d-tuple A is an organic extension of the concept. In particular, it determines the operator space SA up to unit preserving and completely isometric isomorphism (see [3, Theorem 2.4.2] or [9, Theorem 5.1]). The matrix range has been used and studied in recent works, in the contexts of the UCP interpolation problem [8] (following [17]), finite-dimensional/compact representability of operator systems [19, 20] (following [16]), and extremal problems in matrix convex sets [12] (following [14]). The main purpose of this note is to explore the extent to which the matrix range determines a d-tuple up to unitary equivalence, under suitable assumptions. We unify the treatment in [9] and [19] by considering the problem for tuples of compact operators, and separately for tuples of normal operators. It is clear that the matrix range does not detect multiplicity (e.g., W(A) = W(A ⊕ A)), thus one needs to impose some kind of minimality condition. Definition 1.1. A d-tuple A ∈ B(H)d is said to be minimal if there is no proper closed reducing subspace G ⊂ H such that W(PGAG) = W(A). This notion of minimality was used in [8, Section 6], and in the finite-dimensional case it corresponds precisely to the notion of minimal pencil used earlier in [17] (and to the notion of σ-minimal pencil used in [23]). Using matricial polar duality [11], one can show that 2010 Mathematics Subject Classification. 47A20, 47A13, 46L07, 47L25. Key words and phrases. Matrix convex set; matrix range; matrix extreme point; operator system; struc- ture of compact tuples. The first author was partially supported by a 2016-2018 Zuckerman Fellowship at the Technion. The second author was partially supported by Israel Science Foundation Grant no. 195/16. 1 the results of [17] imply that a minimal d-tuple of operators on a finite-dimensional space is determined up to unitary equivalence by its matrix range. In [8, Section 6], this problem was treated in the wider setting of compact tuples. It was claimed there, mistakenly, that if A and B are two minimal tuples of compact operators, then W(A) = W(B) if and only if A is unitarily equivalent to B. This is false, as [19, Example 3.14] shows. After the mistake was discovered, a corrected version of that paper appeared on the arXiv [9]. The correct result, [9, Theorem 6.9], shows that if two minimal tuples of compact operators have the same matrix range and also have a new property called nonsingularity, then they are unitarily equivalent. The property was used as a somewhat ad hoc fix, so one of our goals here is to present different conditions which allow us to avoid the nonsingularity assumption. In order define nonsingularity, we need to review some basic facts about representations of C-algebras of compact operators (see [7, Section I.10] for proofs of the facts stated in this paragraph). Let K(H) denote the algebra of compact operators on a Hilbert space H. If A is a C-subalgebra of K(H), then every representation of A is the direct sum of irreducible representations, and every nonzero irreducible representation of A is unitarily equivalent to a direct summand of the identity representation. It follows that if A ∈ K(H)d, then C ∗(A) (which is not assumed unital) is given as a direct sum C ∗(A) = ⊕i∈IAi, where for every i ∈ I, the algebra Ai is either unitarily equivalent to K(Hi) ⊗ IKi for some Hilbert spaces Hi and Ki, or Ai = 0. In particular, A is the direct sum of irreducible compact d-tuples, some of which may be zero. If there are no two irreducible summands that are unitarily equivalent, we say that A is multiplicity-free. Recall that SA denotes the operator system generated by A. Thus, the C-algebra C ∗(SA) generated by SA is just the unital C-algebra generated by A. Every nondegenerate represen- tation of C ∗(A) extends uniquely to a unital representation of C ∗(SA). When dim H < ∞, the irreducible representations of C ∗(SA) are precisely the unitizations of irreducible sub- representations of C ∗(A). When dim H = ∞, C ∗(SA) may have an additional kind of representation, the singular representation π0 : C ∗(SA) → C, determined by π0(I) = 1 and π0(Ai) = 0 for i ∈ {1, . . . , d}. The singular representation π0 may or may not be equivalent to a subrepresentation of the identity representation. π : C ∗(SA) → B(Hπ), such that π is the unique UCP extension of π(cid:12)(cid:12)SA We shall also require the theory of boundary representations and the C-envelope [1, 3, 4]. Recall that a boundary representation for SA in C ∗(SA) is an irreducible unital representation to C ∗(SA). An ideal J ⊳ C ∗(SA) is called a boundary ideal (for SA) if the quotient map C ∗(SA) → C ∗(SA)/J is completely isometric on SA. The Shilov ideal is the largest boundary ideal, and the C- envelope of SA is the quotient of C ∗(SA) by the Shilov ideal. The C-envelope of SA can also be identified with the image of C ∗(SA) under the sum of all boundary representations (see [5, Theorem 7.1] or [10, Theorem 3.4]). Let us write ∂A for the collection of irreducible subrepresentations of the identity representation of C ∗(SA) which are also boundary representations for SA in C ∗(SA). Thus, when H is infinite dimensional, we may write C ∗ e (SA) ∼= σ(C ∗(SA)) ⊕ Mπ∈∂A π(C ∗(SA)), where σ is π0 if π0 is a boundary representation, and σ is the nil representation otherwise. Some of the summands might be redundant, since π0 might be a boundary representation as well as a subrepresentation of the identity representation. That is, it is possible that π0 ∈ ∂A. 2 We can now finally give the definition of nonsingularity, and the corresponding uniqueness theorem. Definition 1.2. [9, Definition 6.3] A tuple A = (A1, ..., Ad) ∈ K(H)d is said to be nonsin- gular if either dim H < ∞, or dim H = ∞ and for every n and every matrix (sij) ∈ Mn(SA), (1.1) kπ0(sij)k ≤ sup {kπ(sij)k : π ∈ ∂A} . Otherwise A is said to be singular. Theorem 1.3. [9, Theorem 6.9] Let A and B be two nonsingular and minimal d-tuples of compact operators. Then W(A) = W(B) if and only if A is unitarily equivalent to B. We extend and simplify the results in [9] in two separate ways. First, we identify singular and nonsingular compact tuples by studying compressions, summands, and matrix extreme points of the matrix range, as in Proposition 3.4, Theorem 3.6, and Theorem 3.7. Further, our results allow us to conclude that the examples considered by Evert in [12] are nonsingular. We also consider a different minimality condition, opting to discuss arbitrary compressions instead of compressions to reducing subspaces. Definition 1.4. [19, Definition 3.20] A d-tuple A ∈ B(H)d is said to be fully compressed if there is no proper closed subspace G ⊂ H such that W(PGAG) = W(A). Proposition 3.21 of [19] gives an extremely restrictive uniqueness theorem for fully com- pressed compact tuples, as a result of direct computations. Namely, if T is a fully compressed d-tuple of compact operators, and W(T ) is a matrix convex set which is generated by its first level, then T is uniquely determined up to unitary equivalence. The question of whether the uniqueness result persists without the assumption about the first level was left open. In pursuit of this result, we characterize fully compressed compact tuples and nonsingular compact tuples in terms of the C ∗-envelope, extending the results of [9]. Our main theorem (Theorem 4.4) is as follows. Theorem 1.5. Let A ∈ K(H)d be a tuple of compact operators. Then the following are equivalent. (1) A is fully compressed. (2) A is multiplicity-free, and the Shilov ideal of SA in C ∗(SA) is trivial. (3) A is minimal and nonsingular. The equivalence of (2) and (3) is a direct improvement of [9, Proposition 6.7], which only provided a partial version of (2) =⇒ (3). We also immediately obtain the following corollary of Theorems 1.3 and 1.5, which appears as Corollary 4.3. Corollary 1.6. Let A and B be two fully compressed d-tuples of compact operators. Then W(A) = W(B) if and only if A is unitarily equivalent to B. We approach the above results from two points of view. First, just as the introduction of nonsingularity may be used to patch the errors in [8], so too may one consider fully com- pressed compact tuples instead of minimal ones. Thus, we provide proofs of the equivalence (1) ⇐⇒ (2) in Theorem 1.5, as well as of Corollary 1.6, which do not rely on the notion of nonsingularity. We believe this is of interest, as fully compressed tuples need not be compact (unlike nonsingular tuples), so the results are potentially open to generalization. However, 3 we also find that by using the concepts of fully compressed tuples and nonsingularity in tan- dem, we may prove all of Theorem 1.5 and unify previous results, thereby making it easier to detect tuples which meet the (equivalent) conditions. Finally, we close with some brief discussion of tuples which are not necessarily compact. In particular, in Theorem 4.10 we prove that a normal tuple is fully compressed if and only if it is minimal. Using an earlier result [19, Theorem 3.26], we are thus able to give a complete description of all fully compressed normal tuples. 2. Matrix convexity and extreme points The sets considered in the theory of matrix convexity are the "free sets". For fixed d ∈ Z+, we consider subsets of the form S = ⊔∞ n, where for every n the set Sn consists of d-tuples of n × n matrices. Below, we shall refer to Sn as the nth level of S. A free set S is said to be matrix convex if for every X ∈ Sm, Y ∈ Sn, n=1Sn contained in ⊔∞ n=1M d and in addition, for every φ ∈ UCP(Mm, Mk), X ⊕ Y ∈ Sm+n, φ(X) := (φ(X1), . . . , φ(Xd)) ∈ Sk. We say that a matrix convex set S is closed/bounded if every level Sn is closed/bounded, and we note that if S is bounded, there is actually a uniform norm bound that applies simultaneously to each Sn. If T ∈ B(H)d is a tuple of bounded operators, then the matrix range W(T ) is a closed and bounded matrix convex set, and in fact every closed and bounded matrix convex set arises this way [8, Section 2.2]. Matrix convexity is defined above in reference to UCP maps. From Choi's theorem (see [6]), n and V ∈ Mn,m, a concrete version immediately follows. First, for X = (X1, . . . , Xd) ∈ M d we write V ∗XV = (V ∗X1V, . . . , V ∗XdV ) ∈ M d m. If X (i) ∈ Sni and Vi ∈ Mni,n satisfy Pk i X (i)Vi is called a matrix convex combination of the X (i). The matrix convex combination is said to be proper if rank Vi = ni for all i, and weakly proper if Vi 6= 0 for all i. A free set S is matrix convex if and only if it is closed under matrix convex combinations. i Vi = In, then the sum Pk i=1 V ∗ i=1 V ∗ For compact convex sets K ⊆ Cd, the Krein-Milman theorem and Milman's converse show that K is the closed convex set generated by the set of extreme points, and that the set of extreme points is minimal with respect to this property. An analogous study of extreme points for matrix convex sets is more complicated, as there are multiple relevant notions of extreme point to consider. Definition 2.1. Let S be a matrix convex set. A point X ∈ S, say X ∈ Sn, is said to be (1) a Euclidean extreme point of S if X = tY + (1 − t)Z with t ∈ (0, 1), Y, Z ∈ Sn implies X = Y = Z; (2) a matrix extreme point (MEP) of S if whenever X is written as a proper matrix i X (i)Vi, then X (i) is unitarily equivalent to X for (3) an absolute extreme point (AEP) of S if whenever X is written as a weakly proper i X (i)Vi, then for all i, the tuple X (i) is i=1 V ∗ unitarily equivalent to X or to a direct sum X ⊕ Zi for some Zi ∈ S. all i; i=1 V ∗ convex combination X = Pk matrix convex combination X = Pk 4 For any convex set K, we use ext(K) to denote the set of extreme points of K. In particular, if S is matrix convex, then ext(Sn) consists of the Euclidean extreme points of S which lie in level n. We will also let MEP(S) and AEP(S), respectively, denote the set of matrix extreme and absolute extreme points of a matrix convex set S. In the first level S1 of a matrix convex set, there is no distinction between Euclidean extreme points and matrix extreme points: MEP(S) ∩ S1 = ext(S1). Further, Webster and Winkler proved a matricial Krein-Milman theorem [22, Theorem 4.3], which says that if S is a closed and bounded matrix convex set, then the closed matrix convex hull of MEP(S) is S. However, the set of matrix extreme points is not necessarily minimal. For closed and bounded real free spectrahedra (that is, matrix convex sets defined by a linear inequality), absolute extreme points are a minimal spanning set [13, Theorem 1.1]. On the other hand, there are closed and bounded matrix convex sets which have no absolute extreme points at all [12, Theorem 1.2]. Let K ⊂ Cd be compact and convex. If S is a matrix convex set with S1 = K, then S sits between two extremal sets, which we denote as in [8]. First, (2.1) W min(K) := {M ∈ ⊔∞ n=1M d n : M has a normal dilation with joint spectrum in K} is the matrix convex hull of the compact convex set K. We remind the reader that a tuple T ∈ B(H)d is called normal if T consists of commuting normal operators, and that [8, Corollary 4.4] shows that W min(K) is the matrix range of any normal tuple N whose joint spectrum satisfies conv(σ(N)) = K. Second, (2.2) W max(K) := {M ∈ ⊔∞ n=1M d n : W1(M) ⊆ K} is the largest matrix convex set whose first level is K. These two sets are equal precisely when K is a simplex by [20, Theorem 4.1] (see also [16, Theorem 4.7] for a similar result with the assumption that K is a polytope). In [18], Kriel considers the extremal matrix convex sets whose level n is specified, where n is any fixed positive integer. In particular, [18, Corollary 6.12] implies that if a closed and bounded matrix convex set S of self-adjoints is equal to the matrix convex hull of Sn for some n, then the absolute extreme points of S are a minimal spanning set. A crucial aspect of the proof is the fact that matrix extreme points are either absolute extreme points or admit nontrivial matrix extreme dilations. Theorem 2.2. [18, Lemma 6.11] Let S ⊂ ⊔∞ n)sa be a closed and bounded matrix convex set of self-adjoints, and let X ∈ Sm be a matrix extreme point of S. Then either X is an absolute extreme point of S, or there is a matrix extreme point of S which is of the form n=1(M d (cid:18)X b c(cid:19) for some b ∈ (Cm)d \ {0} and c ∈ (Rm)d. b∗ While the above result is stated for self-adjoints, we may easily obtain a corresponding result in the general case by breaking a tuple into real and imaginary parts. Alternatively, the result can also be obtained by a combination of [10, Lemma 2.3] and [15, Theorem B]. Below we show that under certain geometric conditions, Euclidean extreme points can automatically be absolute extreme points. Recall that from the classical Krein-Milman theorem and Milman's converse, it follows that if K ⊂ Cd is a compact convex set, then any point λ ∈ K satisfies (2.3) λ is an isolated extreme point of K ⇐⇒ λ 6∈ conv(ext(K) \ {λ}). 5 We write and note that any point λ ∈ IK is the vertex of some polytope which contains K. IK := {λ ∈ ext(K) : λ is isolated in ext(K)} Proposition 2.3. Suppose S is a closed and bounded matrix convex set with S1 = K. If there is a polytope P such that P contains K and λ ∈ K is a vertex of P , then λ is an absolute extreme point of S. Consequently, if λ is an isolated extreme point of K, then λ is an absolute extreme point of S. Proof. If λ ∈ K is a vertex of the polytope P , then λ is also a vertex of a simplex ∆ which contains P . It follows that λ is an absolute extreme point of W min(∆) (see [19, Lemma 3.9] and the commentary immediately thereafter). Since S ⊆ W max(K) ⊆ W max(∆) = W min(∆), it follows that λ ∈ S1 is an absolute extreme point of a set larger than S, so λ is also an absolute extreme point of S. Motivated by the equivalence (2.3), we define a collection of matrix extreme points which behave in a similar way. Definition 2.4. Let S be a closed and bounded matrix convex set, and let X ∈ S. Then we call X a crucial matrix extreme point if the collection C := {M ∈ S : M is a matrix extreme point of S and M is not unitarily equivalent to X} has the property that the closed matrix convex hull of C excludes X. Note that by Webster and Winkler's matricial Krein-Milman theorem, a crucial matrix extreme point of S is indeed a matrix extreme point of S. We also immediately reach the following from (2.3): S1 ∩ {X ∈ S : X is a crucial matrix extreme point of S} ⊆ IS1. That is, a crucial matrix extreme point of S which belongs to the first level S1 must be an isolated extreme point of S1. For sets of the form W min(K), the converse also holds. Proposition 2.5. Let K ⊂ Cd be compact and convex. Then (2.4) and (2.5) ext(K) = MEP(W min(K)) = AEP(W min(K)) IK = {X ∈ W min(K) : X is a crucial matrix extreme point of W min(K)}. Proof. For any X ∈ W min(K), X admits a normal matrix dilation N ∈ M d m with σ(N) ⊆ K by (2.1) and [8, Theorem 7.1]. The joint diagonalization of N shows that X can be written as a proper matrix convex combination of points λi in K = S1. If X is a matrix extreme point of W min(K), then all the λi are unitarily equivalent to X, so X is in level one. Since X is certainly still extreme, we conclude that MEP(W min(K)) ⊆ ext(K). From [19, Lemma 3.9], we have that ext(K) ⊆ AEP(W min(K)), and finally the containment AEP(W min(K)) ⊆ MEP(W min(K)) is trivial. If X is an isolated extreme point of K, then the set of matrix extreme points which are not unitarily equivalent to X is precisely C := ext(K) \ {X} by (2.4). The closed matrix 6 convex set generated by C is the matrix range of N := Lλ∈C W min(conv(C)). Since the first level conv(C) excludes X by (2.3), we see that X is a crucial matrix extreme point of W min(K). λ, which by [8, Corollary 4.4] is Suppose instead that X is a crucial matrix extreme point of W min(K). Since X is a matrix extreme point, (2.4) shows that X ∈ ext(K), and the fact that X is crucial implies that X 6∈ conv(ext(K) \ {X}). By (2.3), X is isolated as an extreme point of K. An immediate consequence of Proposition 2.5 is that a closed and bounded matrix convex set might have no crucial matrix extreme points. For example, consider W min(K) where K is the unit disk. The following result also shows that in general, crucial matrix extreme points must also be absolute extreme points. Proposition 2.6. Let S be a closed and bounded matrix convex set. Then every crucial matrix extreme point of S is an absolute extreme point. Proof. Suppose X is a matrix extreme point of S which is not an absolute extreme point. Theorem 2.2 shows that there is a nontrivial dilation Y of X which is also a matrix extreme point of S. Moreover, since X and Y have distinct finite dimension, Y cannot be unitarily equivalent to X. We conclude that since X is in the matrix convex hull of Y , X cannot be a crucial matrix extreme point of S by definition. In the next section, we extend the results of [9] by showing how the crucial matrix extreme points of the matrix range may be used to characterize when a tuple of compact operators is nonsingular. 3. Characterizations of Nonsingularity Let us recall some definitions surrounding nonsingular compact tuples. If A ∈ K(H)d is a tuple of compact operators on an infinite-dimensional space, then the operator system SA is not contained in the compacts, as it by definition includes the unit. Indeed, the unital C ∗-algebra generated by A, denoted C ∗(SA), admits a singular representation π0 which annihilates all compact operators and maps the identity operator to 1 ∈ C. It turns out that π0 may or may not be dominated by ∂A, the collection of irreducible subrepresentations of the identity representation of C ∗(SA) which are also boundary representations of SA in C ∗(SA). Definition 3.1. [9, Definition 6.3] A tuple A = (A1, ..., Ad) ∈ K(H)d is said to be nonsingu- lar if either dim H < ∞, or if dim H = ∞ and for every n and every matrix (sij) ∈ Mn(SA), (3.1) kπ0(sij)k ≤ sup {kπ(sij)k : π ∈ ∂A} . Otherwise A is said to be singular. The notion of nonsingularity behaves very well with respect to direct summands and multiplicity. Lemma 3.2. Let A and B be d-tuples of compact operators, and assume that B is a summand of A with W(B) = W(A). If B is nonsingular, then A is nonsingular. Similarly, if C ∈ K(H)d and NLi=1 C is nonsingular for some N ∈ Z+, then C is nonsingular. 7 Proof. The equality W(A) = W(B) implies that the map Ai 7→ Bi extends to a completely isometric isomorphism from SA to SB [8, Theorem 5.1], so it extends to a ∗-isomorphism C ∗ e (SB). Since B is nonsingular, e (SA) → C ∗ e (SA) ∼= C ∗ C ∗ e (SB) ∼= Mπ∈∂B π (C ∗(SB)) . Now, the compression of A to B extends to a ∗-homomorphism, so we can identify ∂B with a subset of ∂A (see [1, Theorem 2.1.2]). It follows that for any representation σ of C ∗(SA) and for any (sij) ∈ Mn(SA), kσ(sij)k ≤ sup π∈∂B kπ(sij)k ≤ sup π∈∂A kπ(sij)k. This shows that A must be nonsingular. Next, let C ∈ K(H)d and define D = C, so the map x 7→ x ⊗ IN is a ∗-isomorphism be- tween C ∗(SC) and C ∗(SD). The unitary equivalence classes of irreducible subrepresentations of the identity representations of C and D are the same, hence NLi=1 sup π∈∂C kπ(sij)k = sup σ∈∂D kσ(sij ⊗ IN )k for every (sij) ∈ Mn(SC). If dim H = ∞, then the corresponding singular representations satisfy π0(sij) = π0(sij ⊗ IN ), and we conclude that C is singular if and only if D is. Nonsingularity may be detected using any of the following conditions. Proposition 3.3. [9, Proposition 6.6] The following conditions are sufficient for a tuple A ∈ K(H)d to be nonsingular: (1) dim H < ∞, or (2) A contains 0 as a direct summand, or (3) 0 is not an isolated extreme point of W1(A). We will strengthen sufficient condition (2), after which we will present a separate theorem which characterizes nonsingularity completely. To accomplish the first goal, we use a notion crucial to the arguments in [12]. Proposition 3.4. Let A ∈ K(H)d be a tuple of compact operators. If 0 is a compression of A for some N ∈ Z+, then A is nonsingular. NL1=1 Proof. By Lemma 3.2, we need only prove nonsingularity of eA := pression of eA. If 0 is a direct summand of eA, then eA is nonsingular by Proposition 3.3, so we may assume otherwise. In particular, we have that eA is a nontrivial dilation of 0. It follows that there is some two-dimensional compression B of eA such that B is a nontrivial W(eA). Finally, Proposition 2.3 shows that 0 is not an isolated extreme point of W1(eA), and nonsingularity of eA follows from Proposition 3.3. dilation of 0. From [14, Lemma 3.5], we conclude that 0 is not an absolute extreme point of A. Let 0 be a com- NLi=1 8 Remark 3.5. Said differently, any tuple A ∈ K(H)d which satisfies the condition "0 is in the finite interior of the noncommutative convex hull" from [12, §1.2] is automatically nonsingular. Next, we show that the conditions of Proposition 3.3 may be adapted to form a charac- terization of nonsingularity, primarily by modifying the proof to consider extreme points in various levels. We will need the following observation: given a compact tuple A ∈ K(H)d, the set (3.2) C := MEP(W(A)) \ {0} has the property that any X ∈ C is the compression of a summand of A. Indeed, if X ∈ C is of size n × n, then [15, Theorem B] shows that there is a pure UCP map φ ∈ UCP(SA, Mn) with φ(A) = X. By [10, Theorem 2.4], φ is the compression of the restriction of a boundary representation ρ. Since X 6= 0, we must have ρ(A) 6= 0, and hence ρ is also (up to unitary equivalence) a subrepresentation of the identity representation -- that is, ρ ∈ ∂A. In other words, ρ(A) is an irreducible summand of A, and X is a compression of this summand. Further, if X ∈ C is actually an absolute extreme point, then we must have X = ρ(A) (see also [18, Corollary 6.27]). On the other hand, if A is infinite-dimensional, then the point 0 ∈ W(A) corresponds to the UCP map sending Ai 7→ 0, which is the restriction of the singular representation π0 to SA. Even if π0 is a boundary representation, it might not be a subrepresentation of the identity. Regardless, we may guarantee that A is nonsingular if sufficiently many nonzero matrix extreme points exist. Theorem 3.6. Let A ∈ K(H)d be a tuple of compact operators. Then A is nonsingular if and only if one of the following conditions holds. (1) There is a finite-dimensional summand B of A with W(A) = W(B), or (2) 0 is not a crucial matrix extreme point of W(A). Proof. If B is a finite-dimensional summand of A (perhaps equal to A), then certainly B is nonsingular. From Lemma 3.2, it follows that if W(B) = W(A), then A is also nonsingular. Next, suppose that A is infinite-dimensional and 0 is not a crucial matrix extreme point of W(A), so the collection C of (3.2) has the property that 0 can be approximated by matrix convex combinations of points in C. Since any point X ∈ C dilates to ρ(A) for some ρ ∈ ∂A, it follows that π0 may be dominated on Mn(SA) by the collection ∂A. That is, the inequality (3.1) in Definition 3.1 holds, and A is nonsingular. To prove the converse, suppose that A is nonsingular, and consider nonzero matrix extreme points Y ∈ C. Following the logic in the proof of [18, Corollary 6.12], use Theorem 2.2 to produce successive nontrivial dilations Y (1) ≺ Y (2) ≺ . . . of nonzero matrix extreme points. Such a sequence will terminate if and only if some Y (i) is an absolute extreme point of W(A), so we consider cases. Case I. Suppose that there is an infinite sequence Y (1) ≺ Y (2) ≺ . . . of nonzero matrix extreme points. We may then form an orthonormal sequence {ei}∞ i Aei = i Y (i)ei for each i. Since A is compact, it follows that 0 is in the closed matrix convex hull e∗ of C. That is, 0 is not a crucial matrix extreme point of W(A), and condition (2) holds. i=1 in H such that e∗ Case II. Suppose that any nonzero matrix extreme point of W(A) may be dilated to an absolute extreme point. 9 If there are finitely many (non unitarily-equivalent) nonzero absolute extreme points X (1), . . . , X (N ) of W(A), then one of the finite-dimensional tuples 0 ⊕ X (i) or NLi=1 X (i) NLi=1 has the same matrix range as A. It follows that there is a finite-dimensional (hence nonsin- gular) minimal tuple Y with W(Y ) = W(A). On the other hand, [9, Corollary 6.8] shows there is a summand B of A which is minimal and has W(B) = W(A). Further, the proof of the result produces a choice of B which is itself nonsingular. From Theorem 1.3, B is unitarily equivalent to Y , and condition (1) holds. If there are infinitely many (non unitarily-equivalent) nonzero absolute extreme points X (1), X (2), . . . of W(A), then since each X (i) is unitarily equivalent to a distinct summand of the compact tuple A, it follows that 0 is in the closed matrix hull of the X (i). As such, 0 is not a crucial matrix extreme point of W(A), and condition (2) holds. Of course, since nonsingular compact tuples have now been characterized, we immediately find that singular compact tuples are characterized using the negation. From revisiting the proof, one can also see that minimal singular compact tuples must be essentially of the form outlined in [19, Example 3.14 and Corollary 3.15]. Theorem 3.7. Let A ∈ K(H)d be a singular tuple of compact operators. Then there is an integer N ≥ 1 and a decomposition A ∼= NMi=1 X (i) ⊕ Y, where X (1), . . . , X (N ) are (up to unitary equivalence) all the nonzero absolute extreme points of W(A), Y is an infinite-dimensional compact tuple, and W(A) = W(cid:18) NLi=1 over, if A is minimal, then Y is irreducible. Xi ⊕ 0(cid:19). More- Proof. Suppose A is singular, so certainly A 6= 0 and W(A) has at least one nonzero matrix extreme point. From Theorem 3.6, we have that 0 is a crucial matrix extreme point of W(A). In particular, 0 is an isolated extreme point of W1(A) and hence an absolute extreme point of W(A) by Proposition 2.3. Using the same arguments as in the proof of Theorem 3.6, we conclude that since 0 is a crucial matrix extreme point, it holds that any matrix extreme point X 6= 0 is a compression of an absolute extreme point, and there can be only finitely many nonzero absolute extreme points X (1), . . . , X (N ). In particular, we note that N 6= 0: W(A) has at least one nonzero matrix extreme point, which dilates to an absolute extreme point. Since each X (i) must be unitarily equivalent to a summand of A, we may write A ∼= NMi=1 X (i) ⊕ Y for some compact tuple Y . However, A is necessarily infinite-dimensional, so we must have that Y is infinite-dimensional. It also holds that W(cid:18) NLi=1 X (i) ⊕ 0(cid:19) = W(A), as the left hand side includes every matrix extreme point of W(A). Finally, if A is minimal, then Y must be irreducible, as otherwise Y has an infinite-dimensional summand which detects 0. 10 In the next section, we examine the assumption that a compact tuple is fully compressed, aiming to prove a uniqueness theorem for the matrix range and relate the assumption to minimality and nonsingularity. 4. Compressions Recall that a compact tuple A ∈ K(H)d is called fully compressed if no compression of A to a proper closed subspace has the same matrix range as A. A tuple which is fully compressed is automatically minimal, but the converse need not hold, as most proper subspaces of H are not reducing subspaces of A. In what follows, we consider the following questions. • If A ∈ K(H)d is fully compressed, does W(A) uniquely determine A? • How does the assumption that A is fully compressed relate to previous conditions, like minimality and nonsingularity? Nonsingularity (Definition 1.2) is an assumption added to the correction [9] of [8], after counterexamples were found to the theorems therein. Particularly, a compact tuple which is minimal and nonsingular is uniquely determined by its matrix range. This assumption is useful in that it directly solves the problem at hand, but it is somewhat ad-hoc. Therefore, our approach in this section is two-pronged. First, we wish to show that "fully compressed" is a suitable replacement condition for "minimal and nonsingular", in that a uniqueness theorem follows by appropriately adapting the techniques of [8] in a natural way. That is, we may prove a uniqueness theorem for compact tuples by examining arbitrary compressions instead of nonsingular tuples. We choose this approach to begin with, primarily because nonsingularity is defined only in reference to compact tuples, whereas any tuple of operators may be fully compressed. Thus, this approach is open to potential generalization. Second, we see how the study of fully compressed tuples can help us better understand the condition of nonsingularity. For example, [9, Proposition 6.7] claims that a compact tuple A which is minimal and nonsingular must also have the following properties: • A is multiplicity-free • The Shilov ideal of SA in C ∗(SA) is trivial. However, only a partial converse is given. We will complete the converse in a round-robin proof, ultimately finding that for compact tuples, "fully compressed" means the same thing as "minimal and nonsingular". Therefore, examining arbitrary compressions in addition to nonsingularity leads to a somewhat better understanding of both conditions in the compact setting. We begin with a lemma considering finite-dimensional tuples. Lemma 4.1. If A ∈ K(H)d is minimal and 0 is a summand of A, then H is finite- dimensional. Moreover, if B is a d-tuple of operators on a finite-dimensional space, then B is minimal if and only if it is fully compressed. Proof. Suppose A ∈ K(H)d is such that A = B ⊕ 0 and A is infinite-dimensional. Then B is infinite-dimensional, 0 ∈ W(B), and W(A) = W(B). We conclude that A is not minimal. For the second statement, we need only prove that minimal tuples of finite-dimensional operators are fully compressed. If B is minimal, then because B acts on a space of finite dimension, there is a compression C of B such that W(B) = W(C) and C is fully compressed. 11 Both B and C are minimal, so by Theorem 1.3, they are unitarily equivalent. It follows that C cannot be a proper compression, and hence B is fully compressed. Note that while Theorem 1.3 (i.e., [9, Theorem 6.9]) concerns nonsingular tuples, we have only used it for tuples of finite-dimensional operators in the above proof. The uniqueness result we need may therefore ultimately be deduced from results in free spectrahedra, as studied in [17] and [23]. Namely, so long as one applies a shift to ensure 0 is present in the matrix range W(A) of a matrix tuple A ∈ M d n, it follows that W(A) is the polar dual of the free spectrahedron determined by A (as in [9, Proposition 3.3 and Lemma 3.4]). Lemma 4.2. If A ∈ K(H)d is a tuple of compact operators, then A is fully compressed if and only if A is multiplicity-free and the Shilov ideal of SA in C ∗(SA) is trivial. Proof. If A is fully compressed, then it is clearly multiplicity-free. Therefore, we may find a Ki, where Ki = K(Hi) is the space of compact operators on Hi. If H is finite-dimensional, the presence of the unit is redundant. Ki for some I0 ⊆ I, so we define I1 = I \ I0 Hi with C ∗(SA) = CIH +Li∈I decomposition H = Li∈I The Shilov ideal J ⊳ C ∗(SA) has the form Li∈I0 and Gk = Li∈Ik e (SA) = C ∗(SA)/J is ∗-isomorphic to CIG1 + Li∈I1 Hi for k ∈ {0, 1}. If G0 is finite-dimensional, or if G0 and G1 are both infinite-dimensional, then the quotient Ki, where the sum is taken in B(G1). It C ∗ follows that the compression onto G1 is completely isometric on SA, and hence W(A) = W(PG1AG1). Since A is fully compressed, we conclude that G1 = H, and J is trivial. Suppose instead that G0 is infinite-dimensional and G1 is finite-dimensional. In this case, (4.1) C ∗ e (SA) = C ∗(SA)/J ∼= CIH +Mi∈I1 Ki ∼= C ⊕Mi∈I1 Ki, where the first sum is in B(H) and the second sum is an external direct sum. Define X = PG1AG1, so that the image of A through (4.1) is 0 ⊕ X, and hence W(A) = W(0 ⊕ X). If F is an infinite-dimensional proper subspace of G0, and B is the compression of A to the subspace F ⊕ G1 ⊂ H, then there is a UCP map sending B 7→ 0 ⊕ X. We conclude that W(0 ⊕ X) ⊆ W(B) ⊆ W(A) = W(0 ⊕ X), and hence W(B) = W(A). This contradicts the assumption that A is fully compressed, so this case does not occur. Next, we consider the converse. If A is multiplicity-free and the Shilov ideal is trivial, then [9, Proposition 6.7] implies that A is minimal. (Note that the portion of [9, Proposition 6.7] we are actually using is a carry-over from [8] and does not require nonsingularity). From Lemma 4.1, it follows that if A acts on a finite-dimensional space, which must happen if 0 is a summand of A, then A is fully compressed. Assume that A acts on an infinite-dimensional space, so 0 is not a summand of A. More- over, note that triviality of the Shilov ideal directly implies that C ∗(SA) = C ∗ e (SA). Suppose that B is a compression of A to a closed subspace G ⊆ H, with W(A) = W(B). Since SA and SB are then completely isometrically isomorphic, the universal property of C ∗ e (SA) gives rise to a surjective ∗-homomorphism π : C ∗(SB) → C ∗(SA) that extends the complete isometry B 7→ A. By the representation theory of algebras of compact operators, and keeping in mind 12 that A does not have 0 as a direct summand, the map π is a direct sum of representations unitarily equivalent to subrepresentations of the identity representation of C ∗(SB). Since A is minimal, every equivalence class of a subrepresentation of the identity representation of C ∗(SB) appears at most once. We therefore find that A is unitarily equivalent to a com- pression of B to a reducing subspace F ⊆ G. Since the compression of A to F is a unitary equivalence, it extends to a ∗-representation of C ∗(SA). By Sarason's lemma [21, Lemma 0], F is semi-invariant for C ∗(SA). But a semi-invariant subspace for a C-algebra is a reducing subspace, so the compression of A to F is actually a direct summand with the same matrix range. Since A is minimal, we conclude that F = H, and hence G = H. That is, A is fully compressed. We may now show that fully compressed compact tuples are uniquely determined by their matrix ranges, without using nonsingularity. That is, we may import the original proof technique (sans flaws) from [8]. Corollary 4.3. Let A ∈ K(H1)d and B ∈ K(H2)d be fully compressed tuples of compact operators satisfying W(A) = W(B). Then A and B are unitarily equivalent. Proof. Suppose that W(A) = W(B). Then A 7→ B extends to a unital completely isometric isomorphism of SA onto SB, and therefore it extends to a ∗-isomorphism of the corresponding C-envelopes. By Lemma 4.2, C ∗(SA) = C ∗ e (SB). We therefore have a ∗-isomorphism π : C ∗(SA) → C ∗(SB), which must be a direct sum of subrepresentations of the identity representation and perhaps the singular representation. e (SA) and C ∗(SB) = C ∗ Let us first assume that 0 is not a direct summand of either tuple. Then π is the sum of subrepresentations of the identity representation. Each one of these representations appears at most once in the sum because B is fully compressed and consequently multiplicity-free. On the other hand, since A does not have 0 as a direct summand, each subrepresentation of the identity must appear at least once, since π is injective. We see that π must be implemented by a unitary equivalence, as required. Next, assume that 0 is a direct summand of one of the tuples, say A. By Lemma 4.1, A then acts on a finite dimensional space, and it follows that B also acts on a finite dimensional space. Arguing as above, we find that π must be implemented by a unitary equivalence. Alternatively, recall that [9, Proposition 6.7] relates the same two conditions used above, multiplicity-free and trivial Shilov ideal, to minimality and nonsingularity. However, that result is not an equivalence. We may now extend both that result and Lemma 4.2 in the following theorem. Theorem 4.4. Let A ∈ K(H)d be a tuple of compact operators. Then the following are equivalent. (1) A is fully compressed. (2) A is multiplicity-free, and the Shilov ideal of SA in C ∗(SA) is trivial. (3) A is minimal and nonsingular. Proof. (1) =⇒ (3): If A is fully compressed, it is certainly minimal, so we need only show that A is nonsingular. Theorem 3.7 shows that no singular compact tuple is fully compressed, as we may form a compression A′ = X (i) ⊕ Y ′ with the same matrix range, where Y ′ is an infinite-dimensional compression of Y . NLi=1 13 (3) =⇒ (2): This is part of [9, Proposition 6.7]. (2) =⇒ (1): This is part of Lemma 4.2. Remark 4.5. Corollary 4.3 can alternatively be deduced from Theorems 1.3 and 4.4. It follows from Theorem 4.4 that the tuples considered in [12] admit summands which are fully compressed. Example 4.6. Evert proves in [12, Theorem 1.2] that if d ≥ 2, T ∈ K(H)d sa has no finite- dimensional reducing subspaces, and 0 is in the "finite interior" of the noncommutative convex hull KT , then KT is a closed matrix convex set which has no absolute extreme points. Since KT is also dense in W(T ) (see e.g. the explanation given in [12, §1.3.1]), it immediately follows that KT = W(T ) in this case. The assumption that 0 is in the finite interior means that 0 is a compression of T for some positive integer N, so from Proposition 3.4, we have NL1=i that T is nonsingular. T need not be minimal, but [9, Corollary 6.8] shows that T admits a summand T ′ which is minimal and has W(T ′) = W(T ), and the proof produces T ′ which is also nonsingular. We conclude from Theorem 4.4 that T ′ is fully compressed. Of course, the simplest examples which meet Evert's conditions are irreducible, and one may produce many such examples by modifying the coefficients used in [12, Proposition 4.1]. Generally speaking, the assumption that two compact tuples be irreducible is very different from the assumption that they be fully compressed, so it is interesting that irreducible compact tuples are indeed always fully compressed (in particular, they are automatically minimal, and they cannot be singular as they do not fit into the mold of Theorem 3.7). It should be noted that irreducible compact operators are easily determined uniquely by their matrix ranges, as this was part of the study initiated by Arveson. Indeed, in [3, Theorem 2.4.3], Arveson showed that two irreducible GCR operators which have trivial Shilov ideals are unitarily equivalent if and only if their matrix ranges are equal. For compact operators, the Shilov ideal requirement is automatically satisfied (and the uniqueness result for compact operators is stated explicitly in [2]). From Arveson's theorem, and using the structure theorem for compact operators, one obtains a general classification theorem for Bj are two compact operators written as direct sums compacts: if A = Li∈I Ai and B = Lj∈J of irreducibles, then A is unitarily equivalent to B if and only if up to a bijection one has I = J and W(Ai) = W(Bi) for all i ∈ I. A similar statement can be made regarding GCR operators that are decomposed into a direct integrals, but one needs to throw in the assumption that every constituent of the integral has trivial Shilov boundary. Recent classification theorems are somewhat different in spirit, as they use the matrix range of a tuple as a single entity. At the bottom of page 304 in [3], Arveson points out that the matrix range is not a complete invariant for irreducible operators in general, and he provides examples of non- unitarily equivalent, irreducible GCR operators having the same matrix range. In fact, he observes that if S, T ∈ B(H) are contractions, both of which have spectrum that contains the unit circle, then W(S) = W(T ). By considering compressions of the unilateral shift to suitable subspaces [1, p. 207], one obtains such operators that are also irreducible (and even GCR). One can also readily see that all these examples satisfy W(T ) = W min(D). In the 14 following example, we present another large family of non unitarily-equivalent irreducible tuples with the same matrix ranges. Example 4.7. Let e1, e2, . . . be the standard basis of ℓ2(Z+), and let w1, w2, . . . be another orthonormal basis satisfying the following technical conditions. • For every j, hw1, eji 6= 0. • lim j→∞ hwj, eji = 1. • The sum R(i) := • The sum C(j) := hwi, eji is finite for every i, and lim i→∞ R(i) = 1. hwi, eji is finite for every j, and lim j→∞ C(j) = 1. ∞Pj=1 ∞Pi=1 Let T = (T1, . . . , Td) be such that T1 is diagonal with respect to the basis {ej}∞ T2, . . . , Td are diagonal with respect to the basis {wi}∞ such a way that j=1 and i=1. We may select the eigenvalues in • T is irreducible, • Ti ≥ 0 for each i, • T1 + . . . + Td ≤ I, and • 0 and the standard basis vectors belong to W1(T ) (but are not compressions of T ). It follows that W1(T ) is precisely equal to the standard simplex ∆d in Rd, and hence W(T ) is the unique matrix convex set W min(∆d) over the simplex. There is enough freedom left in selecting the eigenvalues that we may produce uncountably many choices of T which are not unitarily equivalent. The matrix range considered above is also a set of the form W min(K), similar to Arveson's examples. It is not clear precisely which K can be used in such constructions. Regardless, we note that none of the examples considered are fully compressed, and we are led to the following questions. Question 4.8. For tuples A, B ∈ B(H)d which are not necessarily compact, if A and B are fully compressed and W(A) = W(B), does it follow that A is unitarily equivalent to B? Question 4.9. Let S be a closed and bounded matrix convex set. When can S be written as the matrix range of a fully compressed tuple T ∈ B(H)d? One may immediately apply [19, Theorem 3.26] to see that fully compressed normal tu- ples, which are necessarily also minimal, are uniquely determined by their matrix ranges. Moreover, normal tuples are minimal if and only if they are fully compressed. Theorem 4.10. Let N ∈ B(H)d be a normal tuple. Then N is fully compressed if and only if it is minimal. In particular, this occurs if and only if there is a compact convex set K such that the set of isolated extreme points IK has ext(K) = IK and N is unitarily equivalent to Lλ∈IK λ. Proof. We need only prove that minimal normal tuples are fully compressed. If N ∈ B(H)d is normal and minimal for its matrix range, then from [19, Theorem 3.26], we have that N is a multiplicity-free diagonal operator, the eigenvalues of N lie at the set IK of isolated 15 extreme points of some compact convex set K, and K satisfies IK = ext(K). Given a joint eigenvalue λ ∈ IK, let vλ denote a corresponding eigenvector. Let G ⊆ H be a closed subspace with W(PGNG) = W(N), so in particular K = W1(PGNG). Fix λ ∈ IK, and fix an R-affine transformation Φ : Cd → R and a num- ber 0 < r < 1 such that Φ(λ) = 1 but −r ≤ Φ(γ) ≤ r for any γ ∈ IK \ {λ}. Next, let A := Φ(N) ∈ B(H)sa, so that A is a diagonal self-adjoint operator. In particular, A has a joint eigenvalue 1 at vλ and eigenvalues of magnitude at most r at vγ, γ 6= λ. Since λ ∈ W1(PGNG), we have that 1 ∈ W1(PGAG). Keeping in mind that PGAG is a self-adjoint contraction, it follows that we may find unit vectors gn ∈ G such that hAgn, gni → 1. However, if one applies the decomposition H = Cvλ ⊕ (Cvλ)⊥ (where both subspaces are reducing for A) to obtain gn = bn + cn, it follows that hAgn, gni = hAbn, bni + hAcn, cni ≤ bn2 + rcn2. Since r < 1 and bn2 + cn2 = 1, we must have that bn → 1 and cn → 0. That is, after a unimodular rescaling, gn converges to vλ, and hence vλ ∈ G. This applies to each eigenvector vλ, so G = H, and finally, N is fully compressed. A fully compressed normal tuple is the direct sum of the isolated extreme points of the first level of its matrix range W min(K), and in particular, these points are the crucial matrix extreme points of W min(K). In general, however, a fully compressed tuple need not be the direct sum of crucial matrix extreme points, as seen in Example 4.6. That is, the boundary representations of the corresponding operator system might be exclusively infinite- dimensional. Acknowledgement. We thank the anonymous referee for insightful comments. References 1. W. Arveson, Subalgebras of C ∗-algebras, Acta Math. 123 (1969), 141 -- 224. 2, 8, 14 2. W. Arveson, Unitary invariants for compact operators, Bull. Amer. Math. Soc. 76 (1970), 88 -- 91. 1, 14 3. W. Arveson, Subalgebras of C-algebras. II., Acta Math. 128 (1972), 271 -- 308. 1, 2, 14 unpublished note 4. W. Arveson, Notes extension property, on the unique (2003). Link: http://users.uoa.gr/ akatavol/newtexfil/arveson/unExt.pdf . 2 5. W. Arveson, The noncommutative Choquet boundary, J. Amer. Math. Soc. 21 (2008), 1065 -- 1084. 2 6. M.D. Choi, Completely Positive Linear Maps on Complex Matrices Linear Algebra Appl. 10 (1975), 285 -- 290. 4 7. K.R. Davidson, C*-Algebras by Example, American Mathematical Society, 1996. 2 8. K.R. Davidson, A. Dor-On, O.M. Shalit and B. Solel, Dilations, inclusions of matrix convex sets, and completely positive maps, Int. Math. Res. Not. 2017 (2017), 4069 -- 4130. 1, 2, 3, 4, 5, 6, 7, 8, 11, 12, 13 9. K.R. Davidson, A. Dor-On, O.M. Shalit and B. Solel, Dilations, inclusions of matrix convex sets, and completely positive maps, updated and corrected arXiv version: arXiv:1601.07993v3 [math.OA] (2018). 1, 2, 3, 7, 8, 10, 11, 12, 13, 14 10. K.R. Davidson and M. Kennedy, The Choquet boundary of an operator system, Duke J. Math. 164 (2015), 2989 -- 3004. 2, 5, 9 11. E.G. Effros and S. Winkler, Matrix convexity: Operator analogues of the bipolar and Hahn-Banach theorems, J. Funct. Anal. 144 (1997), 117 -- 152. 1 12. E. Evert, Matrix convex sets without absolute extreme points, Linear Algebra Appl. 537 (2018), 287 -- 301. 1, 3, 5, 8, 14 13. E. Evert and J. William Helton, Arveson extreme points span free spectrahedra, preprint, arXiv:1806.09053. 5 16 14. E. Evert, J.W. Helton, I. Klep and S. McCullough, Extreme points of matrix convex sets, free spectra- hedra, and dilation theory, J. Geom. Anal. 28 (2018) 1373 -- 1408. 1, 8 15. D.R. Farenick, Extremal matrix states on operator systems, J. London Math. Soc. 61 (2000), 885 -- 892. 5, 9 16. T. Fritz, T. Netzer and A. Thom Spectrahedral containment and operator systems with finite-dimensional realization, preprint, arXiv:1609.07908. 1, 5 17. J.W. Helton, I. Klep and S. McCullough, The matricial relaxation of a linear matrix inequality, Math. Program. 138 (2013), 401 -- 445. 1, 2, 12 18. T.-L. Kriel, An introduction to matrix convex sets and free spectrahedra, preprint, arXiv:1611.03103v6. 5, 9 19. B. Passer, Shape, scale, and minimality of matrix ranges, preprint. To appear in Trans. Amer. Math. Soc. 1, 2, 3, 4, 6, 10, 15 20. B. Passer, O.M. Shalit and B. Solel, Minimal and maximal matrix convex sets, J. Funct. Anal. 274 (2018), 3197 -- 3253. 1, 5 21. D. Sarason, On spectral sets having connected complement, Acta Sci. Math. (Szeged) 26 (1965), 289 -- 299. 13 22. C. Webster and S. Winkler, The Krein-Milman theorem in operator convexity, Trans. Amer. Math. Soc. 351 (1999): 307 -- 322. 5 23. A. Zalar. Operator positivestellensatze for noncommutative polynomials positive on matrix convex sets, J. Math. Anal. Appl. 445 (2017): 32 -- 80. 1, 12 Department of Pure Mathematics, University of Waterloo, Waterloo, ON, Canada E-mail address: [email protected] Faculty of Mathematics, Technion - Israel Institute of Technology, Haifa 3200003, Is- rael E-mail address: [email protected] 17
1705.08981	1	1705	2017-05-24T22:19:47	H${}^2$ Spaces of Non-Commutative Functions	[ "math.OA", "math.FA" ]	We define the Hardy spaces of free noncommutative functions on the noncommutative polydisc and the noncommutative ball and study their basic properties. Our technique combines the general methods of noncommutative function theory and asymptotic formulae for integration over the unitary group. The results are the first step in developing the general theory of free noncommutative bounded symmetric domains on the one hand and in studying the asymptotic free noncommutative analogues of classical spaces of analytic functions on the other.	math.OA	math	H2 SPACES OF NON-COMMUTATIVE FUNCTIONS MIHAI POPA AND VICTOR VINNIKOV Abstract. We define the Hardy spaces of free noncommutative functions on the noncommutative polydisc and the noncommutative ball and study their basic properties. Our technique combines the general methods of noncommu- tative function theory and asymptotic formulae for integration over the unitary group. The results are the first step in developing the general theory of free noncommutative bounded symmetric domains on the one hand and in studying the asymptotic free noncommutative analogues of classical spaces of analytic functions on the other. 1. Introduction There emerged, over the years, a general paradigm of passing from the commu- tative setting to the free noncommutative setting: we replace a vector space by the disjoint union of square matrices of all sizes over this vector space. Some instances of this paradigm are Amitsur's theory of rational identities [15, Chapter 8], oper- ator space theory [5, 13], free probability -- when viewed asymptotically [21, 16], and free noncommutative algebraic and semialgebraic geometry [7]. Another in- stance is free noncommutative function theory that originated with the work of J. L. Taylor on noncommutative spectral theory [18, 19] and was further developed by Voiculescu [23, 24] and Kaliuzhnyi-Verbovetskyi -- Vinnikov [8] (we refer to [8] for a historical account and further references). The aim of this paper is to take first steps towards the theory of Hardy spaces on noncommutative domains, within the framework of noncommutative function theory. More specifically we introduce and study the Hardy space H 2 of noncom- mutative functions on the noncommutative polydisc (Dm)nc = ∞an=1 {(X1, . . . , Xm) ∈ (Cn×n)m : kXjk < 1, j = 1, . . . , m} ∞an=1 {(X1, . . . , Xm) ∈ (Cn×n)m : X∗i Xi < In}. mXi=1 and the noncommutative ball (Bm)nc = Our technique combines the methods of noncommutative function theory, especially the Taylor -- Taylor noncommutative power series expansions, see [8, Chapter 7], and asymptotic formulae for integration over the unitary group coming from random matrix theory and free probability, see [22, 17] and [3, 4]. The resulting Hardy spaces have some of the same basic properties as their commutative counterparts; the most striking difference is that they are not complete, though their completions This work was partially supported by a grant of the Romanian National Authority for Sci- entific Research, CNCS UEFISCDI, project number PN-II-ID-PCE-2011-3-0119 and Simmons Foundation Grant No. 360242. 1 2 MIHAI POPA AND VICTOR VINNIKOV can be identified as Hilbert spaces of noncommutative functions (in fact, noncom- mutative reproducing kernel Hilbert spaces) on a certain noncommutative set. This is related to the different (as opposed to the commutative case) convergence pat- terns for noncommutative power series, see [8, Section 8.3]. A general theory of noncommutative reproducing kernel Hilbert spaces and their multipliers is devel- oped in the forthcoming paper [2]. The paper is organized as follows. Section 2 discusses the preliminaries on the asymptotic integration over the unitary group, noncommutative function theory, and the noncommutative unit balls of two operator space structures on Cm, namely the noncommutative polydisc and the noncommutative ball mentioned above, their distiniguished boundaries and the invariant measures thereupon. Section 3 contains the main results on the definition and the basic properties of the Hardy spaces. In the case of bounded noncommutative functions, asymptotic integral formulae in terms of tracial integrals over the distinguished boundary with respect to the invariant measure have been obtained by Voiculescu [24, Chapters 14 -- 16] for the noncommutative polydisc and the square noncommutative matrix ball, i.e., the noncommutative unit ball of the noncommutative space over Cm×m with its obvious operator space structure. The noncommutative polydisc and the noncommutative ball are of course non- commutative analogues of the usual polydisc and ball in Cm. In a forthcoming paper we will consider noncommutative analogues of other matrix balls and of ir- reducible bounded symmetric domains of type II and III, and develop a general theory of noncommutative Jordan triples. It would be also interesting to consider in this context the unit ball in the OH operator space norm on Cm [13]. In a different direction, the results here point towards a study of asymptotically defined spaces of noncommutative functions as analogues of various classical spaces of ana- lytic functions. In particular, it would be interesting to study the noncommutative Bargmann -- Fock space in relation to free stochastic processes [1]. 2. Preliminaries 2.1. Haar Unitaries and Free Independence. Let N be a positive integer and U(N ) be the compact group of the N × N unitary matrices with complex entries. The Haar measure on U(N ) will be denoted with dUN . For each i, j ∈ {1, 2, . . . , N} we define the maps ui,j : U(N ) −→ C giving the i, j-th entry of each element from U(N ). As shown in [3], the maps ui,j are in L∞(U(N ), dUN ). Let Sn be the symmetric group of order n; for σ ∈ Sn denote by #(σ) the number of cycles in a minimal decomposition of the permutation σ. The following result is shown in [4, Corollary 2.4]: Theorem 2.1. There exists a map Wg : Z+ × ∪∞n=1Sn −→ R such that: (1) The function Wg(·,·) is analytic at ∞ in the first variable and, for each (2) For any indices ik, i′k, jk, j′k ∈ {1, 2, . . . , N}, where 1 ≤ k ≤ n, we have that τ (k)! ZU (N ) n dUN = Xσ,τ∈Sn Wg(N, τ σ−1) · ui1,j1 ··· uin,jn ui′ nYk=1 1,j′ 1 ··· ui′ n,j′ δik ,i′ σ(k) δjk,j′ σ ∈ Sn, the limit lim N→∞ Wg(N, σ) N 2n−#(σ) exists and is finite. H2 SPACES OF NON-COMMUTATIVE FUNCTIONS 3 Moreover, if m 6= n , then ZU (N ) ui1,j1 ··· uin,jn ui′ 1,j′ 1 ··· ui′ m,j′ mdUN = 0. An immediate consequence of the result above is the following: Corollary 2.2. Let U : U(N ) −→ CN×N , U = [ui,j]N integers α, i,j=1. Then, for all non-zero ZU (N ) Tr(U α)dUN = 0, where Tr denotes the non-normalized trace. Proof. Suppose that α > 0. Then ZU (N ) Tr(U α)dUN = X1≤i1,...,iα≤NZU (N ) ui1,i2 ··· uiα−1,iα uiα,i1 dUN . From the last part of Theorem 2.1, all the terms in the above summation are zero, hence the conclusion. Since U−1 = U∗, the case α < 0 is similar. (cid:3) When studying the joint asymptotic behavior of several large random matrices with independent entries, an important tool is the notion of free independence (see, for example, [22], [11]). As shown in the extensive literature on the subject (see [21], [16], [11], [12]), this is in fact the natural relation of independence in a non-commutative framework. The precise definition for the version of the notion free independence that will be used in the present work is presented below. Suppose that A is a unital C∗-algebra and φ : A −→ C is a positive conditional expectation. A family {Aj}j∈J of unital C∗-subalgebras of A is said to be free if any alternating product of centered (with respect to φ) elements from {Aj}j∈J is centered, i.e., for any n > 0, any ǫ(k) ∈ J (1 ≤ k ≤ n) such that ǫ(k) 6= ǫ(k + 1) and any ak ∈ Aǫ(k) such that φ(ak) = 0 we have that φ(a1a2 ··· an) = 0. Subsets M1, M2, . . . , Mn of A are said to be free or free independent if the unital C∗-algebras generated by the elements of each of them form a free family. The following result, Theorem 2.3, is proved in [22] and, in a more general frame- work, in [4], [17]. We first introduce some notation. Let A = {Aj,N}j∈J,N≥1 be an ensemble of matrices such that Aj,N ∈ CN×N for all j ∈ J. The ensemble A is said to have limit distribution if for any m ∈ Z+ and j1, . . . , jm ∈ J the limit Tr(Aj1,N ··· Ajm,N ) exists and is finite. Also, let j1, . . . , js ∈ J; a polyno- lim N→∞ mial in s non-commutative variables, p ∈ Chxj1 , . . . , xjsi is said to be asymptotically centered in A if Tr(p(Aj1 ,N , . . . , Ajs,N )) = 0. 1 N 1 N lim N→∞ i,j : U(N ) −→ C such that u(k) Theorem 2.3. Let m be a positive integer; for 1 ≤ k ≤ m and 1 ≤ i, j ≤ N con- sider the random variables u(k) i,j and ui,j are identically distributed for each i, j, k and {u(k) i,j=1 are independent. Finally, for each k and N, consider the matrix Uk,N ∈ L∞(U(N ), dUN )N×N , having the entries u(k) i,j . Suppose that A = {Aj,N}j∈J,N≥1 is an ensemble of complex matrices that has limit distribution. Then the ensembles of random matrices {U1,N , U∗1,N}N≥1, {U2,N , U∗2,N}N≥1, . . . , {Um,N , U∗m,N}N≥1 and A are asymptotically free with respect i,j }N 4 MIHAI POPA AND VICTOR VINNIKOV to the functional ZU (N ) 1 N Tr(·)dUN , in the following sense: N→∞ZU (N ) lim 1 N Tr(p1 · p2 ··· pk)dUN = 0 for any p1, . . . , pk either centered polynomials in {Ul,N , U∗l,N}N≥1 for some l , or asymptotically centered polynomials in elements of A, such that ps and ps+1 are polynomials in elements of different ensembles for all s. Remark 2.4. In the framework of Theorem 2.3 above, suppose that the joint dis- tribution of elements of A does not depend on N , i.e. for any j1, j2, . . . , js there exists a complex constant c(j1, j2, . . . , js) such that Then 1 N Tr(Aj1 ,N Aj2,N ··· Ajs,N ) = c(j1, j2, . . . , js). N→∞ZU (N ) lim Tr(p1 · p2 ··· pk)dUN < ∞. expression ZU (N ) Proof. First note that the condition on A implies that the polynomials p1, p2, . . . , pk in elements of A are centered for any N . Then using the analyticity in the first variable at ∞ of the function Wg(·,·) from Theorem 2.1(i), it follows that the Tr(p1 · p2 ··· pk)dUN expands at ∞ as a Laurent series in N , with coefficients depending on the polynomials p1, p2, . . . , pk. Theorem 2.3 from above implies that all the coefficients of monomials N q with q ≥ 0 are null, hence the conclusion. 1 N (cid:3) Throughout the paper, Fm will denote the free monoid with m generators {1, . . . , m}. The elements of Fm are arbitrary words w = w1 ··· wl−1wl; the length of the word w will be denoted by w = l. We will also use the notation F [l] m for the set of all words from Fm of length l. 2.1 and 2.3 above: In the next section we will utilize the following consequences of the Theorems Corollary 2.5. For y = y1 ··· yt−1yt a word in Fm and U1,N , . . . , Um,N as in Theorem 2.3, we will denote U y N = Uy1,N ··· Uyt−1,N Uyt,N . With this notation, for any w, v ∈ Fm we have that: Tr ((U w (i) if v 6= w, thenZU (N ) N→∞ZU (N ) (ii) if v = w but v 6= w, then lim (iii) if v = w, then lim 1 N N→∞ZU (N ) Tr ((U w Tr ((U w 1 √N N )∗U v N ) dUN = 1 N )∗U v N ) dUN = 0, for any positive integer N ; N )∗U v N ) dUN = 0; Proof. Suppose that w = w1 ··· wt−1wt and v = v1 ··· vs−1vs (here w = t and v = s). For part (i), let L = {l = (l−v, l−v+1, . . . , l0, l1, . . . , lw) ∈ {1, . . . , N}v+w+1 : l−v = lw} H2 SPACES OF NON-COMMUTATIVE FUNCTIONS 5 and Vk = {j ∈ {1, . . . ,v} : vj = k}, respectively Wk = {j ∈ {1, . . . ,w} : wj = k}. Then, from the independence of the families {u(k) N )dUN =Xl∈LZU (N ) wYk=1 (ZU (N ) Yk∈Wr mYr=1 =Xl∈L vYk=1 lk−1,lk · Yk∈Vr ZU (N ) u(wk) lk−1,lk · u(vk) l−k+1,l−k u(vk) l−k+1,l−k dUN ). i,j }N Tr ((U w N )∗U v dUN u(wk) i,j=1, From Theorem 2.1, if card(Wr) 6= card(Vr), then the coresponding factor in the above product vanishes, hence the conclusion. For (ii), it suffices to consider the case when v1 6= w1 (since U∗k,N Uk,N = IdN ). From Corollary 2.2,ZU (N ) Tr ([Uk,N ]p)dUN = 0, for all integers p, all N > 1 and all 1 ≤ k ≤ m, and the conclusion follows now from Remark 2.4. 1 N Finally, for (iii), if w = v, then (U w N = IdN , and the assertion is trivial. (cid:3) N )∗ U v An analogous result for the matricial block entries of a Haar unitary is given below. Corollary 2.6. Fix m a positive integer and suppose that U = [ui,j]mN i,j=1, with the functions ui,j : U(mN ) −→ C as defined above. For 1 ≤ k ≤ m , consider Uk ∈ L∞(U(mN ), dUmN )N×N given by Uk = [ui,(k−1)N +j]N i,j=1. (I. e., U1, . . . , Um are the N × N matricial block entries of the first N × mN matricial row of U ). Then: Let v, w ∈ Fm , and, for y = y1 ··· ys−1ys ∈ Fm, denote U y = Uy1 ··· Uys−1Uys. (i) If v 6= w,ZU (mN ) (ii) If v = w, but v 6= w, then lim (iii) If v = w, then lim N→∞ZU (mN ) Tr ((U w)∗U v) dUmN = 0 for any positive integer N ; Tr ((U w)∗U v) dUmN = 0; N→∞ZU (mN ) Tr ((U w)∗U v) dUmN = 1 √N 1 mv . 1 N A similar statement holds for the first mN × N matricial column of U . Proof. Parts (i), respectively (iii) are immediate consequences of Theorem 2.1, re- spectively of the identity U∗k Uk = IdN . For part (ii), let us suppose that w = w1 ··· wt−1wt and v = v1 ··· vs−1vs, with w1 6= v1. Let ei,j be the m × m matrix with the i, j entry 1 and all other entries 0 and Ei,j = ei,j ⊗ IdN ∈ CmN×mN . Then for all 1 ≤ k ≤ m, we have that fUk = e1,1 ⊗ Uk = E1,1U Ek,1, hence Tr ((U w)∗U v) = Tr(cid:16)gUwt ∗ ···gUw1 = Tr(cid:0)E1,wt U∗E1,wt−1 U∗ ··· E1,w1 U∗E1,1U Ev1,1U ··· Evs−1,1U Evs,1(cid:1) . ∗gUv1 ···gUvs(cid:17) To simplify the notation, we shall write (1) (2) E0 i,j = Ei,j − δi,j 1 m IdmN . Note that Tr(E0 i,j ) = 0, and that the ensemble {ei,j ⊗ IdN} has the property from Remark 2.4, therefore for all non-zero integers α0, . . . , αn and all indices 6 MIHAI POPA AND VICTOR VINNIKOV i, j, k, l, kr, lr ∈ {1, . . . , m} we have that (3) lim N→∞ 1 √N ZU (mN ) m√N ZU (mN ) 1 hence (4) lim N→∞ k1,l1 U α1E0 Tr(cid:0)Ei,j U α0E0 Tr(cid:0)E1,wt U∗ ··· E1,w1 U∗E0 k2,l2 ··· E0 kn,ln U αn Ek,l(cid:1) dUmN = 0, 1,1U Ev1,1U ··· Evs,1(cid:1) dUmN = 0, because, using (2) for E1,w1 , . . . , E1,wt−1 and Ev1,1, . . . , Evs−1,1, the integrand from (4) is a finite linear combination of integrands from (3). (cid:3) 2.2. Non-Commutative Functions and Taylor -- Taylor Expansions. We de- fine non-commutative functions following [8], see also [9] and [14]. For V a (complex) linear space, we will denote by Vnc the set`∞n=1 V n×n. For a subset Ω of Vnc, we denote Ωn = Ω ∩ V n×n; Ω is said to be a non-commutative set if for all positive integers m, n and all X ∈ Ωm and Y ∈ Ωm we have that X ⊕ Y ∈ Ωm+n, where X ⊕ Y is the block diagonal matrix from V (m+n)×(m+n) with X and Y the block entries of the main diagonal and all other entries zero. If V and W are two linear spaces and Ω a non-commutative subset of Vnc, a mapping f : Ω −→ Wnc is said to a non-commutative function if it satisfies the following conditions: • f (Ωn) ⊂ W n×n for all positive integers n; • f (X ⊕ Y ) = f (X) ⊕ f (Y ) for all X, Y ∈ Ω; • if X ∈ Ωn and T ∈ Cn×n is invertible with T XT −1 ∈ Ω, then f (T XT −1) = T f (X)T −1. Non-commutative functions have strong regularity properties -- for an introduc- tion to the basic theory see [8]. Below we will mention only a particular form of the Taylor -- Taylor expansion, as established in [8, Chapter 7], that will be extensively utilized in Section 3 of the present work. Let V be a finite dimensional vector space with basis e1, . . . , ed. For X ∈ V N×N , there exist unique X1, . . . , Xd ∈ CN×N such that X = X1e1 + . . . + Xded. If w = w1 ··· wt ∈ Fd, we write X w = Xi1 ··· Xit . Suppose that Ω ⊆ Vnc is a non-commutative set such that for all N , the set ΩN = Ω∩V N×N is open, let W be a Banach space, and suppose that f : Ω −→ Wnc is a non-commutative function locally bounded on slices separately in every matrix dimension, that is for all positive integers N , all X ∈ ΩN , and all Y ∈ V N×N , there exists ε > 0 such that the function t 7→ f (X + tY ) is bounded for t < ε. Let b ∈ Ω1, and for N a positive integer, define the set Υ(IdN b) = {X ∈ ΩN : IdN b + t(X − IdN ·b) ∈ ΩN for all t ∈ C such that t ≤ 1} (this is the maximal subset of ΩN that is complete circular around IdN b). Then [8, Theorem 7.2] (see also Theorems 7.8 and 7.10 there) states that for all X ∈ Υ(IdN b) (5) f (X) = ∞Xl=0 Xw=l (X − IdN b)w ⊗ fw , where the series converges absolutely and uniformly (in fact, normally) on com- pacta of Υ(IdN b). (The Taylor -- Taylor coefficients fw ∈ W are given by fw = H2 SPACES OF NON-COMMUTATIVE FUNCTIONS 7 R is the higher order partial difference -- differential oper- R f (b, . . . , b), where ∆w⊤ ∆w⊤ ator corresponding to w ∈ Fd.) 2.3. Operator space structures on Cm. An operator space structure on a linear space V is given (see [5, Proposition 2.3.6]) by a family of norms {k · kn}n>0, such that each k·kn is a norm on V n×n and, for all X ∈ V n×n, Y ∈ V m×m, T, S ∈ Cn×n, we have that: complex matrices. • kX ⊕ Y kn+m = max{kXkn,kY km}; • kT XSkn ≤ kTkkXknkSk, where k · k denotes the usual operator norm of We will consider the operator spaces structures on Cm given by the k·k∞, k·kcol, and k · krow, where, for X = (X1, . . . , Xm) ∈ (CN×N )m ≃ (Cm)N×N and k · k the usual operator norm in CN×N kXk∞ = max{kX1k, . . . ,kXmk}, kXkcol = k 1 2 , mXi=1 X∗i Xik mXi=1 XiX∗i k 1 2 . kXkrow = k For the norm k·k∞, the non-commutative unit ball is the non-commutative polydisc (Dm)nc = ∞aN =1 {(X1, . . . , Xm) ∈ (CN×N )m : kXjk < 1, j = 1, . . . , m}. For the norms k·kcol, respectively k·krow, the non-commutative unit balls are given by (Bm)nc = respectively by (Bm row)nc = ∞aN =1 {(X1, . . . , Xm) ∈ (CN×N )m : ∞aN =1 {(X1, . . . , Xm) ∈ (CN×N )m : mXi=1 mXi=1 X∗i Xi < IN}, XiX∗i < IN}. Identifying the components from (CN×N )m of (Dm)nc, (Bm)nc, respectively (Bm with the corresponding subsets of CmN 2 , the Shilov boundaries for the commutative algebras of complex analytic functions in mN 2 variables, as shown in [20, Exam- ple 1.5.51], are U(N )m in the case of (Dm)nc, respectively the set of all isometries and coisometries of CmN×N for (Bm)nc, respectively of CN×mN for (Bm row)nc. Since CmN×N does not have any coisometries, it follows that for the case of (Bm)nc the above Shilov boundary is row)nc (∂S(Bm)nc)N = {(X1, . . . , Xm) ∈ (CN×N )m : X∗i Xi = IN} mXi=1 = {(X1, . . . , Xm) ∈ (CN×N )m : there exists some U ∈ U(mN ) such that U [IN 0 . . . 0]T = [X1 . . . Xm]T}, where, for A ∈ Cn×m, the notation AT stand for the matrix transpose of A. 8 MIHAI POPA AND VICTOR VINNIKOV Similarly, since CN×mN does not have any isometries, the Shilov boundary for the case of (Bm row)nc is (∂S(Bm row)nc)N = {(X1, . . . , Xm) ∈ (CN×N )m : XiX∗i = IN}. mXi=1 To simplify the writing in the next section, we will denote U(N )m by (∂S(Dm)nc)N . The natural measure on (∂S(Dm)nc)N is the m-fold product measure µN of the Haar measure on U(N ). Corollary 2.5 then yields that for any v, w ∈ Fm we have: (6) Z(∂S (Dm)nc)N N→∞Z(∂S (Dm)nc)N Tr(cid:0)(X w)∗ X v(cid:1) dµN = 0 if v 6= w Tr(cid:0)(X w)∗ X v(cid:1) dµN = δv,w. 1 N lim For the case of (Bm)nc, note that the group U(mN ) acts transitively on (∂S(Bm)nc)N via [X1 . . . Xm]T 7→ U · [X1 . . . Xm]T . Moreover, denoting (7) H(m, N ) = {In ⊕ U : U ∈ U((m − 1)N )}, (8)ZU (mN ) we have that H(m, N ) is a compact subgroup of U(mN ) which is the stabilizer of [IN 0 . . . 0]T ∈ (∂S(Bm)nc)N . Hence (∂S(Bm)nc)N is isomorphic to U(mN )/H(m, N ) and (see [6, Theorem 2.49]) there exists a unique Radon measure νN of mass 1 on (∂S(Bm)nc)N invariant under the action of U(mN ) and for any continuous function f : U(mN ) −→ C we have that f (U )dUmN (U ) =Z(∂S (Bm)nc)NZH(m,N ) f (U V )dU(m−1)N (V )dνN (U H(m, N )). For 1 ≤ i ≤ mN and 1 ≤ j ≤ N and ui,j : U(mN ) → C as defined in Section 2.1, a simple verification gives that for all U ∈ U(mN ) and V ∈ H(m, N ) (9) Fix now f ∈ Alg{ui,j, ui,j : 1 ≤ i ≤ mN, 1 ≤ j ≤ N}. For all U ∈ U(mn), equation (9) implies ui,j(U ) = ui,j(U · V ). (10) (11) (13) Define bf on (∂S(Bm)nc)N via bf (U H(m, N )) = f (U ). From (9), bf is well-defined. Moreover, equations (8) and (10) gives f (U V )dU(m−1)N (V ) =ZH(m,N ) ZH(m,N ) Z(∂S (Bm)nc)N bf (U H(m, N ))dνN (U H(m, N )) =ZU (mN ) Z(∂S (Bm)nc)N Tr(cid:0)(X w)∗ X v(cid:1) dνN (X) = 0 if v 6= w N→∞Z(∂S (Bm)nc)N Tr(cid:0)(X w)∗ X v(cid:1) dνN (X) = δv,w 1 N lim 1 mv . Hence, Corollary 2.6 (more precisely, its analogue for the first mN × N matricial column) implies that for all v, w ∈ Fm we have: (12) For the case of (Bm a unique Radon measure of mass 1 on (∂S(Bm row)nc, a similar argument as above gives that there exists ν′N , row)nc)N invariant under the action of f (U )dU(m−1)N (V ) = f (U ). f (U )dUmN (U ). H2 SPACES OF NON-COMMUTATIVE FUNCTIONS 9 U(mN ), and that the pair ((∂S(Bm (13). row)nc)N , ν′N ) also satisfies the equalities (12) and 3. Main results The present section will address properties of certain H 2 Hardy spaces associated to the non-commutative unit balls for the operator norms k·k∞ and k·kcol on Cm. row)nc)n, ν′n) satisfy (12) and (13), similar Since both ((∂S(Bm)nc)n , νn) and ((∂S(Bm results to the case of k · kcol can be stated for the setting of k · krow. For Ω either (Bm)nc or (Dm)nc, consider the algebras AΩ ={f : Ω −→ Cnc : f is a non-commutative function, locally bounded on slices separately in every matrix dimension}. Equation (5) gives that for all f ∈ AΩ there exists a family of complex numbers {fw}w∈Fm with f∅ = f (0), such that for all X ∈ Ω (14) f (X) = ∞Xl=0(cid:0) Xw∈F [l] m X wfw(cid:1), where, for any positive integer N , the series converges absolutely and uniformly on compacta of ΩN = Ω ∩ CN×N . For f ∈ AΩ as above and X ∈ (Cm)nc, we will denote f [l](X) = Xw∈F [l] X wfw. Remark that f [l](rX) = rl · f [l](X) for any real r; also, if l 6= p, then m Theorem 3.1. (i) If f ∈ ADm and r ∈ (0, 1), then: Z∂(Ω,N ) 1 1 N Tr(cid:16)(f [l](X))∗f [p](X)(cid:17) dωN = 0. rwZ(∂S (Dm)nc)N N−→∞Z(∂S (Dm)nc)N rw · mwZ(∂S (Bm)nc)N mwZ(∂S (Bm)nc)N 1 N 1 N 1 N lim lim N−→∞ 1 Tr ((X w)∗f (rX)) dµN Tr ((X w)∗f (rX)) dµN . Tr ((X w)∗f (rX)) dνN Tr ((X w)∗f (rX)) dνN . fw = lim N−→∞ = lim r−→1− fw = lim N−→∞ = lim r−→1− (ii) If f ∈ ABm and r ∈ (0, 1), then Proof. For any positive integer N , relations (6) and (14) give (notice that we can interchange the integral and the infinite sum since the convergence is uniform on r (∂S(Dm)nc)N ) Z(∂S (Dm)nc)N Tr ((X w)∗f (rX)) dµN =  Xv∈F [l] mZ(∂S (Dm)nc)N ∞Xl=0 rvfv ·Z(∂S (Dm)nc)N = Xv∈Fm v=w Tr ((X w)∗X v) · rvfvdµN Tr ((X w)∗X v) dµN , 10 MIHAI POPA AND VICTOR VINNIKOV and the equalities from (i) follow from equation (7). The argument for (ii) is similar, using equations (12), (14) and (13). (cid:3) Definition 3.2. For (Ω, dωN ) either ((Bm)nc, dνN ) or ((Dm)nc, dµN ), we define Tr (f (rX)∗f (rX)) dωN < ∞}. H 2(Ω) = {f ∈ AΩ : S(f ) = sup 1 N sup N Theorem 3.3. (i) If f ∈ H 2((Dm)nc), then (ii) If f ∈ H 2((Bm)nc), then fw2) < ∞. (cid:3) ( Xw∈F [l] m fw2) < ∞. r<1Z(∂S Ω)N ∞Xl=0 ml Xw∈F [l] 1 ( m ∞Xl=0 Proof. For (i), note first that if r ∈ (0, 1), f ∈ H 2((Dm)nc) and X ∈ (∂S(Dm)nc)N , rlf [l](X) and the series is then, as in equation (14), we have that f (rX) = ∞Xl=0 r2lZ(∂S (Dm)nc)N ∞Xl=0 1 N Tr(f [l](X)∗f [l](X))dµN absolutely convergent. Therefore equation (6) implies that Z(∂S (Dm)nc)N 1 N Tr(f (rX)∗f (rX))dµN = = r2l[al + bl,N ] ∞Xl=0 where al = Xw=l fw2, and bl,N = Xv=w=l fwfv ·Z(∂S (Dm)nc)N ∞Xl=0 1 N Tr((X w)∗X v)dµN − al. Since f ∈ H 2((Dm)nc), according to Definition 3.2, we have that sup N sup r<1 r2l · (al + bl,N ) = S(f ) < ∞. For each X and each l, N , the matrix f [l](X)∗f [l](X) is positive, henceforth al + bl,N ≥ 0 thus, for each positive integer L and each r ∈ (0, 1), LXl=0 r2l(al + bl,N ) ≤ S(f ). But relation (7) implies that for each l, each L and each r, henceforth sup r<1 bl,N = 0, thus lim N→∞ r2lal < ∞, and, since al ≥ 0, we obtain r2lal ≤ S(f ) for LXl=0 ∞Xl=0 ∞Xl=0 al < ∞. The argument for part (ii) is analogous, utilizing equations (12) and (13). (cid:3) H2 SPACES OF NON-COMMUTATIVE FUNCTIONS 11 Proposition 3.4. For f, g ∈ H 2(Ω) and r > 0, define ϕf,g(r) = lim Tr (g(rX)∗f (rX)) dωN . N−→∞Z(∂S Ω)N 1 N Then ϕf,g exists for r small and equals ∞Xl=0 r2l(cid:0) Xw∈F [l] gwfw(cid:1) for Ω = (Dm)nc, gwfw(cid:1) for Ω = (Bm)nc; in both cases the latest m ∞Xl=0 r2l(cid:0) Xw∈F [l] m extends analytically on (0, 1) and continuously on [0, 1] to a function that we will respectively equals Proof. Since f, g ∈ H 2(Ω), equations (6) and (12) give that (15) denote by gϕf,g. Z(∂S Ω)N ∞Xl=0 Tr(cid:16)g[l](rX)∗ f [l](rX)(cid:17) ≤ kg[l](rX)k · kf [l](rX)k Tr (g(rX)∗f (rX)) dωN =Z(∂S Ω)N On the other hand 1 N 1 N 1 N ≤(rl Xv∈F [l] m ≤(m · r)2l( gv · kX vk) · (rl Xw∈F [l] ml Xv∈cF ml gv) · ( 1 m 1 ml Xw∈F [l] m fw) fw · kX wk) Tr(cid:16)g[l](rX)∗f [l](rX)(cid:17) dωN . ≤(m · r)2l sup v∈F [l] fw ≤ (m · r)2l( sup v∈F [l] m , respectively if Ω = m2 the sum from the right-hand side of equation (15) is absolutely Theorem 3.3 implies then that if Ω = (Dm)nc and r < 1 (Bm)nc and r < 1 convergent, uniform in N and X, hence gw2 + sup w∈F [l] gw · sup w∈F [l] fw2). m m m m with the right-hand side absolutely convergent uniform in N . From equations (7) and (13), Z(∂S Ω)N 1 N N→∞Z∂(Ω,N ) lim 1 N Tr (g(rX)∗f (rX)) dωN = ∞Xl=0Z(∂S Ω)N Tr(cid:16)g[l](rX)∗f [l](rX)(cid:17) dωN = and the conclusion follows from Theorem 3.3. 1 N Tr(cid:16)g[l](rX)∗f [l](rX)(cid:17) dωN , r2l Xw∈F [l] r2l Xw∈F [l] m m gwfw if f, g ∈ H 2((Dm)nc) 1 ml gwfw if f, g ∈ H 2((Bm)nc) (cid:3) An immediate consequence of the Proposition above is the following Theorem. Theorem 3.5. (i) With the notations from Definition 3.2, H 2(Ωnc) are inner-product spaces, with 12 MIHAI POPA AND VICTOR VINNIKOV the inner product given by hf, gi = lim r−→1− gϕf,g(r) = Xw∈Fm Xw∈Fm gwfw if f, g ∈ H 2((Dm)nc) 1 ml gwfw if f, g ∈ H 2((Bm)nc) (ii) {X w}w∈Fm is a complete orthonormal system in H 2((Dm)nc) and, for all f ∈ H 2((Dm)nc), we have that fw = hf, X wi and f = Xw∈Fm all f ∈ H 2((Bm)nc), we have that fw = hf, mwX wi and f = Xw∈Fm 2 X w}w∈Fm is a complete orthonormal system in H 2((Bm)nc) and, for fwX w in fwX w in H 2((Dm)nc). (iii) {m w H 2((Bm)nc). Proof. Part (i) follows from Proposition 3.4 and since the sums are convergent from Theorem 3.3. The parts (ii) and (iii) are simple consequences of Theorem 3.1 and equations (cid:3) (7) and (13). Remark 3.6. The limit over N in Theorem 3.5(i) is not the supremum. For example, if m = 2 and f (X) = X1X2 + X2X1, then Z(∂S (Dm)nc)N 1 N Tr (f (rX)∗f (rX)) dµN = 2r2(1 + 1 N 2 ) Definition 3.7. As before, Ω will denote either Dm or Bm. For X ∈ Ωnc, define the map EX Let BΩ,N = {X ∈ Ωnc ∩ CN×N : EX Ω : H 2(Ωnc) −→ CN×N via EX Ω is a bounded map} and BΩ = Ω (f ) = f (X). BΩ,N . ∞aN =1 For p > 0, we define the Hilbert space l2 p(Fm) = {{fw}w∈Fm : fw ∈ C,k{fw}w∈Fmk2,p = ∞Xl=0 ( Xw∈F [l] m 1 pl fwfw) < ∞}. Proposition 3.8. With the above notations, we have that (i) BDm = {X ∈ (Dm)nc : the series sequence {fw}w∈Fm ∈ l2(Fm)} (ii) BBm = {X ∈ (Bm)nc : the series m ∞Xl=0 ( Xw∈F [l] ∞Xl=0 ( Xw∈F [l] m fwX w) converges for any fwX w) converges for any sequence {fw}w∈Fm ∈ l2 m(Fm)} ∞Xl=0 Proof. Suppose that X ∈ Dm ∩ CN×N is such that all {fw} ∈ l2(Fm) and consider the linear map gEX ( Xw∈F [l] Dm({fw}w∈Fm) = ∞Xl=0 m gEX ( Xw∈F [l] m fwX w). fwX w) converges for Dm : l2(Fm) −→ CN×N , given by H2 SPACES OF NON-COMMUTATIVE FUNCTIONS 13 For every l, define also EX,l Dm ({fw}) = lXs=0 ( Xw∈F [l] m fwX w). Dm is bounded. Dm}l>0. Each EX,l Dm is the pointwise limit of {EX,l From the initial assertion, gEX Dm is a bounded linear operator from l2(Fm) to CN×N , so Banach-Steinhaus Theorem gives that gEX Take now f ∈ H 2((Dm)nc). From Theorem 3.3, the sequence {fw}w∈Fm of its Taylor-Taylor coefficients is in l2(Fm) and its norm coincides to the norm of f in H 2((Dm)nc), hence the operator EX For the converse, fix {fw}w ∈ l2(Fm) and, for all l > 0 consider the functions fwX w. The sums are finite, αl : ((Dm)nc)N −→ CN×N given by αl(X) = Xw≤l therefore αl ∈ H 2((Dm)nc), hence, if X ∈ Dm ∩ CN×N , then Dmk ≤ kgEX Dm is bounded and kEX Dmk. kαl+s(X) − αl(X)k ≤ kEX ≤ kEX Dmk · kαl+s − αlkH2((Dm)nc) fw2). Since the sequence {Pw≤l fw2}l≥0 is Cauchy, it follows that {αl}l Cauchy sequence, therefore the series fwX w) converges. is also a The argument for part (ii) is similar, replacing l2(Fm) to l2 second parts of Theorem 3.3 and of relation (??). m(Fm) and using Dmk · ( Xl<w≤l+s ∞Xl=0 ( Xw∈F [l] m Theorem 3.9. For p > 0 , define Υm p = {X ∈ (Cm)nc : Xw∈Fm (cid:3) pw(X w)∗X wconverges}. Then BDm = (Dm)nc ∩ Υm Moreover, if X ∈ Υm 1 and BBm = (Bm)nc ∩ Υm m. p ∩ CN×N , then {X w}w∈Fm ∈ l2 1 p (Fm) ⊗ CN×N . m series ∞Xl=0 Proof. Suppose that X ∈ BDm,N . Since, according to Proposition 3.8(i), the series fwX w) converges for any {fw}w∈Fm from l2(Fm), it follows that the ( Xw∈F [l] ∞Xl=0 fwe∗X wee) also converges for any e,ee ∈ CN . The Riesz Repre- sentation Theorem gives that {e∗X wee}w∈Fm ∈ l2(Fm), therefore also the series ∞Xl=0 Xw∈F [l] Taking e,ee from the cannonical basis of CN , we get thatPw∈Fm ( Xw∈F [l] e∗(X w)∗eeX w converges for all e,ee ∈ CN . verges on each entry, therefore in CN×N . The argument for (Bm)nc is similar. (X w)∗X w con- m m 14 MIHAI POPA AND VICTOR VINNIKOV Suppose now than X ∈ Υm pw(X w)∗X w also converges entrywise, and, since the (j, j)-entry of the series equals pw( Xw∈Fm p ∩ CN×N . Then Xw∈Fm NXl=1 ( Xw∈Fm (Fm) ⊗ CN×N . NXl=1 x(w) l,j x(w) l,j ) = 1 p pwx(w) l,j x(w) l,j ) l,j where x(w) l, j. In particular {X w}w∈Fm ∈ l2 is the (l, j)-entry of X w, it follows that {x(w) l,j }w∈Fm ∈ l2 1 p (Fm) for all (cid:3) Remark 3.10. (i) Υm 1 6⊂ (Dm)nc and Υm m 6⊂ (Bm)nc. (ii) If X = (X1, X2, . . . , Xm) ∈ (CN×N )m is such that X∗X1 + X∗2 X2 + ··· + X∗mXm < 1 p p . In particular, then X ∈ Υm Proof. For part (i), it suffices to take X = (X1, 0, . . . , 0) with X1 nilpotent with norm larger than 1. Then X ∈ Υm p for any p > 0, but X 6∈ (Bm)nc, (Dm)nc. 1√m (Dm)nc ⊂ BDm and 1√m (Bm)nc ⊂ BBm. For part (ii), suppose that X∗X1 + X∗2 X2 +···+ X∗mXm < θ p for some 0 < θ < 1. Denote by X [l] = pl Xw∈F [l] m (X w)∗X w. Then 0 ≤ X [l+1] = Xw∈F [l] m pl(X w)∗ p mXk=1 X∗k Xk! X w < θX [l] hence Xw∈Fm (X w)∗X w < 1 1 − θ . Definition 3.11. For p > 0, we will consider the sets Kp = {(X, Y ) ∈ (Cm)nc × (Cm)nc : and the maps Kp : Kp −→ Cnc, given by Kp(X, Y ) = ∞Xl=0 [ Xw∈F [l] m ∞Xl=0 [ Xw∈F [l] m plX w ⊗ (Y w)∗] converges} plX w ⊗ (Y w)∗]. Theorem 3.9 implies the following result: Remark 3.12. (Υm p × Υm p ) ⊂ Kp. Also, note that from the second part of Theorem 3.9, any sequence f = {fw}w∈Fm from l2 p(Fm) can be identified with a nc-function on Υm fwX w). f (X) = p via ∞Xl=o ( Xw∈F [l] m Next, we will consider the following spaces of nc-functions: (cid:3) (cid:3) H2 SPACES OF NON-COMMUTATIVE FUNCTIONS 15 Definition 3.13. For p > 0 define H 2 m,p as follows: m,p = {f : Υm H 2 p −→ Cnc : f is nc-function such that there exists some sequence {fw}w∈Fm ∈ l2 p(Fm) such that f (X) = ∞Xl=o Xw∈F [l] m fwX w}. Note that H 2 p are Hilbert spaces with the inner-products inherited from l2 p(Fm). Proposition 3.14 below shows that in fact they are reproducing kernel Hilbert spaces with respect to Kp. Proposition 3.14. Fix Y ∈ Υm (i) The map Kp(·, Y ) : Υm p ∩ CM×M . With the notations above, we have that: p −→ (CM×M )nc is a non-commutative function that belongs to H 2 p,m ⊗ CM×M . (ii) for any e1, e2 ∈ CM and any f ∈ H 2 p,m , hf, e∗1Kp(·, Y )e2il2 p(Fm) = e∗2f (Y )e1. Proof. For part (i), first note that for any w ∈ Fm, the map X 7→ pwX w ⊗ (Y w)∗ is a noncommutative function from Cnc to (CM×M )nc, hence it suffices to prove the convergence in H 2 p,m. p(Fm) for all i, j, hence y(w) i,j X w is a Cnc -valued non-commutative function from i,j }w∈Fm are in l2 H 2 p,m. i,j be the (i, j)-entry of Y w. Let y(w) From Theorem 3.9, the sequences {y(w) the map X 7→ Pw∈Fm For part (ii), let {fw}w∈Fm ∈ l2(Fm) such that f (X) = Xw∈Fm p(Fm) = h{fw}w∈Fm,{e∗1(Y w)∗e2}w∈Fmil2 = Xw∈Fm hf, e∗1Kp(·, Y )e2il2 e∗2fwY we1 = e∗2f (Y )e1. fwX w. Then p(Fm) Proposition 3.15. Suppose that f is a non-commutative function locally bounded on slices separately in every matrix dimension around 0 and (cid:3) Φ(r) = lim Ψ(r) = lim N−→∞Z(∂S (Dm)nc)N N−→∞Z(∂S (Bm)nc)N 1 N 1 N Tr(f (rX)∗f (rX))dµN , Tr(f (rX)∗f (rX))dνN . Then f extends to a function in H 2 m,m, if and only if Φ(r), respectively Ψ(r), exists for all small r (in which case Φ, respectively Ψ, are also analytic at 0 ) and it extends analytically to (0, 1) and continously to [0, 1] . 1,m, respectively in H 2 Moreover, lim r−→1− Φ(r) ≥ kfkH2 1,m , respectively lim r−→1− Ψ(r) ≥ kfkH2 m,m . 16 MIHAI POPA AND VICTOR VINNIKOV such that Proof. Suppose first that f extends to ef ∈ H 2 ∞Xl=0 ( Xw∈F [l] f (X) = (16) m fwX w) 1,m, that is there exists {fw}w∈Fm 1 ; in particular, Remark 3.10(ii) gives that the expansion (16) holds As before, consider the non-commutative functions f [l] : (Cm)nc −→ Cnc given fwX w. Then, for X ∈ (∂S(Dm)nc)N , we have that Dm. for all X ∈ Υm for all X ∈ 1√m by f [l](X) =Pw∈F [l] kf [l]( 1 m m X)k ≤ Xw∈F [l] m 1 ml sup w∈F [l] m (fw · kX wk) ≤ sup w∈F [l] m fw, therefore, for r ∈ (0, 1 m ), Z(∂S (Dm)nc)N 1 N Tr(cid:16)f [l](rX)∗f [l](rX)(cid:17) dµN ≤ sup w∈F [l] m fw2. hence, expansion (16) and Corollary 2.5(ii) give that Z(∂S (Dm)nc)N 1 N Tr (f (rX)∗ f (rX)) dµN = ∞Xl=0Z(∂S (Dm)nc)N 1 N Tr(cid:16)f [l](rX)∗f [l](rX)(cid:17) dµN ≤ kfk2 l2(Fm). Therefore, using Corollary 2.5(i), we have that for r ∈ (0, 1 m ), Φ(r) = ∞Xl=0 r2l( Xw∈F [l] m fw2) and Φ extends analytically to (0, 1) and continously to [0, 1]. The proof for Ψ is similar, using Remark 3.10 and Corollary 2.6. For the converse, suppose that there exists δ > 0 such that Φ(r) exists for r < δ and extends analitically to (0, 1). In particular there exists some N0 such that the integral from the definition of Φ(·) is finite if N > N0. Fix now N > N0; equation fwX w converges absolutely for X ∈ αDm, particularly {( α (5) gives that there exists some α > 0 such that the seriesPw∈Fm Let R = min{δ, α m}. Then Corollary 2.5 gives that, for r ∈ (0, R), m )wfw}w∈Fm ∈ l1(Fm) ⊂ l2(Fm). Φ(r) = ∞Xl=0 r2l( Xw∈F [l] m fw2) and the conclusion follows since Φ(·) extends analytically to (0, 1). As before, the proof for Ψ(·) is similar, using equation (5) and Corollary 2.6. (cid:3) H2 SPACES OF NON-COMMUTATIVE FUNCTIONS 17 ϕf,Y (r) = lim Proposition 3.16. For f ∈ H 2 and Y ′ ∈ Υm m ∩ CM×M , we have that N−→∞Z(∂S (Dm)nc)N N−→∞Z(∂S (Bm)nc)N 1 N 1 N 1,m and g ∈ H 2 m,m, respectively Y ∈ Υm 1 ∩ CM×M Tr ⊗ IdCM ×M (f (rX)K1(rX, Y ∗)∗) dµN (X) ψg,Y ′(r) = lim Tr ⊗ IdCM ×M (g(rX)Km(rX, (Y ′)∗)∗) dνN (X) are analytic functions of r for r small and they extend analytically to (0, 1) and ψg,Y ′(r) = g(Y ′). continously to [0, 1]. Moreover, ϕf,Y (r) = f (Y ) and lim r−→1− Proof. Let p ≥ 1 and r ∈ (0, (2m2p)−1), let {fw}w∈Fm ∈ l2 X ∈ CN×N , Y ∈ CM×M such that sup lim r−→1− p(Fm), and consider w∈Fm kX wk ≤ 1. r2wfwpwY w) converges asolutely, since w∈Fm kZ wk = m(Z) < ∞ and sup ( Xw∈F [l] m First, note that the series ∞Xl=0 k{(rp)2w}wkl2(Fm) < 2 and ∞Xl=0 ( Xw∈F [l] m kr2wfwpwY wk) ≤ m(Y ) · ≤m(Y ) · k{fwkwk Also, we have that k{(rmp)w}wkl2(Fm) < 2 and ( Xv∈F [l] ( Xw∈F [k] ∞Xk=0 [ m ∞Xl=0 (rp)2w · p−wfw ( Xw∈F [l] p(Fm) · k{(rp)2w}wk 1 2 l2 m 1 2 l2(Fm) ≤ [ kfw · rwX wk)] · [ ∞Xk=0 ∞Xk=0 ∞Xl=0 mkrk( Xw∈F [k] (r · mp)k · ( Xw∈F [k] ≤ [ m m fw)] · ( m pl · rlkX v ⊗ (Y v)∗k)] ∞Xl=0 p−wfw)] · 2m(Y ) plrlml · m(Y )) ≤ k{fw}wk p(Fm) · k{(rmp)w}wk 1 2 l2 1 2 l2(Fm) · 2m(Y ). Therefore, for p = 1, if f and Y are as in the statement of 3.16, we have that 1 N Z(∂S (Dm)nc)N =Z(∂S (Dm)nc)N Tr ⊗ IdCM ×M (f (rX)K1(rX, Y ∗)∗) dµN (X) 1 N Tr ⊗ IdCM ×M ( fw · rkX w · (X v)∗ ⊗ Y v)dµN (X). Since 1 N Tr ⊗ IdCM ×M is a bounded linear map, using Corollary 2.5, the right hand fwY w). and the conclusion for ϕY,f side of the equation above equals m Xv∈F [l] m ∞Xk,l=0 Xw∈F [k] ∞Xl=0 ( Xw∈F [l] m follows. The proof for ψg,Y ′ is analogous letting p = m and using Corollary 2.6. (cid:3) 18 MIHAI POPA AND VICTOR VINNIKOV Definition 3.17. For Ω either Bm or Dm, define H∞(Ωnc) = {f : Ωnc −→ Cnc : f is a non-commutative function and sup H∞(BΩ) = {f : BΩ −→ Cnc : f is a non-commutative function and sup Obviously, H∞(Ωnc) ⊂ H∞(BΩ), since BΩ ⊂ Ωnc. We will further detail this Z∈Ωkf (Z)k < ∞} Z∈BΩ kf (Z)k < ∞}. inclusion below. Definition 3.18. For Ω either Bm or Dm, define M(Ωnc) = {f : Ωnc −→ Cnc : f is a non-commutative function which is also a bounded left multiplier for H 2(Ωnc)} M(BΩ) = {f : BΩ −→ Cnc : f is a non-commutative function which is also a bounded left multiplier for H 2 1,m, if Ω = Bm, respectively H 2 m,m if Ω = Dm}. where the multiplier norms are the natural ones. Proposition 3.19. With the notations above, we have that H∞(Ωnc) ⊆ M(Ωnc) ⊆ M(BΩ) ⊆ H∞(BΩ). Proof. From the consideration above, we only need to prove the last inclusion. Consider g ∈ M(BΩ), denote Mg the left multiplier with g and take X ∈ BΩ ∩ p ∩ CN×N . From Poposition 3.14, for any e1, e2 ∈ CM and f1, f2 ∈ CM×M , Y ∈ Υm CN , we have that h(Mg)∗e∗1K(·, X)e2,f∗1 K(·, Y )f2i = he∗1K(·, X)e2, Mgf∗1 K(·, Y )f2i =hg(·)f∗1 K(·, Y )f2, e∗1K(·, X)e2i∗ =(e∗2g(X)f∗1 K(X, Y )f2e1)∗, hence (Mg)∗K(·, X) = K(·, X)g(X)∗ and since k(Mg)∗K(·, X)k ≤ kMgk·kK(·, X)k and K(·,·) is a reproducing kernel, it follows that kg(X)k ≤ kMgk. (cid:3) References [1] D. Alpay, P. Jorgensen and G. Salomon. On free stochastic processes and their derivatives. Stochastic Processes and their Applications, vol. 214 (2014), 3392 -- 3411. [2] J. A. Ball, G. Marx, and V. Vinnikov. Noncommutative Reproducing Kernel Hilbert Spaces, arXiv:1602.00760 [3] B. Collins Moments and cumulants of polynomial random variables on unitary groups, the Itzykson-Zuber integral, and free probability, Int. Math. Res. Not. 2003, no. 17, 953 -- 982 [4] B. Collins, P. Sniady Integration with respect to the Haar measure on unitary, orthogonal and symplectic group, Comm. Math. Phys. 264 (2006), no. 3, 773 -- 795 [5] E. G. Effors, Z.-J. Ruan Operator spaces, London Math. Soc. Monographs, New Series 23, Oxford University Press, New York 2000 [6] G. Folland A Course in Abstract Harmonic Analysis, CRC Press, Boca Raton, FL, 1995 [7] J. W. Helton, I. Klep, and S. McCullough. Free Convex Algebraic Geometry. In: "Semidefinite Optimization and Convex Algebraic Geometry" edited by G. Blekherman, P. Parrilo, R. Thomas, pp. 341 -- 405, SIAM, 2013. [8] D. S. Kaliuzhnyi-Verbovetskyi and V. Vinnikov, Foundations of Noncommutative Function Theory, Mathematical Surveys and Monographs 199, AMS, 2014. [9] D. S. Kaliuzhnyi-Verbovetskyi and V. Vinnikov, Noncommutative rational functions, their difference-differential calculus and realizations, Multidimens. Syst. Signal Process. 23 (2012), no.1-2, 49 -- 77 H2 SPACES OF NON-COMMUTATIVE FUNCTIONS 19 [10] A. Koranyi, The Poisson integral for generalized half-planes and bounded symmetric domains, Ann. Math. 82 (1965), 332 -- 350 [11] J.A. Mingo and M. Popa, Real Second Order Freeness and Haar Orthogonal Matrices, J. Math. Phy. 54, 051701 (2013) [12] J.A. Mingo and M. Popa, Freeness and The Transposes of Unitarily Invariant Random Matrices, J. Func. Anal. 271(4) (2016), 883 -- 921 [13] G. Pisier, Introduction to Operator Space Theory, Cambridge University Press, 2003 [14] M. Popa and V. Vinnikov, Non-commutative functions and the non-commutative free L´evy- Hincin formula, Adv. Math. 236 (2013), 131 -- 157 [15] L. H. Rowen. Polynomial identities in ring theory, vol. 84 of Pure and Applied Mathematics. Academic Press Inc., New York, 1980. [16] R. Speicher, A. Nica Lectures on combinatorics of free probability, London Math Soc Lecture Note Series 335, Cambridge Univ. Press 2006 [17] R. Speicher, J. Mingo , P. Sniady Second order freeness and fluctuations of random matrices. II. Unitary random matrices, Adv. Math. 209 (2007), no. 1, 212 -- 240 [18] J. L. Taylor. A general framework for a multi-operator functional calculus. Advances in Math., 9:183 -- 252, 1972. [19] J. L. Taylor. Functions of several noncommuting variables. Bull. Amer. Math. Soc., 79:1 -- 34, 1973. [20] H. Upmeier, Toeplitz Operators and Index Theory in Several Complex Variables (Operator Theory: Advances and Applications, Vol. 81), Birkhauser, 1996 [21] D-V. Voiculescu, K. Dykema, A. Nica Free Random Variables CRM Monograph Series, 1, American Mathematical Society, 1992 [22] D. V. Voiculescu, Limit laws for random matrices and free products. Invent. Math. 104 (1991), no. 1, 201 -- 220 [23] D.-V. Voiculescu. Free Analysis Questions I: Duality Transform for the Coalgebra of ∂X :B International Math. Res. Notices 16:793 -- 822, 2004. [24] D.-V. Voiculescu. Free analysis questions. II: The Grassmannian completion and the series expansion at the origin. J. Reine Angew. Math. 645 (2010), 155 -- 236. Department of Mathematics, University of Texas at San Antonio, One UTSA Circle San Antonio, Texas 78249, USA, Institute of Mathematics 'Simion Stoilow' of the Romanian Academy, P.O. Box 1-764, Bucharest, RO-70700, Romania E-mail address: [email protected] Department of Mathematics, Ben Gurion University of Negev, Be'er Sheva 84105, Israel E-mail address: [email protected]
1804.09277	3	1804	2018-08-10T18:48:31	Compact Group Actions On Operator Algebras and Their Spectra	[ "math.OA", "math.FA" ]	We consider a class of dynamical systems with compact non abelian groups that include C-, W- and multiplier dynamical systems. We prove results that relate the algebraic properties such as simplicity or primeness of the fixed point algebras as defned in Section 3., to the spectral properties of the action, including the Connes and strong Connes spectra.	math.OA	math	Compact Group Actions On Operator Algebras and Their Spectra Costel Peligrad Department of Mathematical Sciences, University of Cincinnati, 4508 French Hall West, Cincinnati, OH 45221-0025, United States. E-mail address: cos- [email protected] Abstract. We consider a class of dynamical systems with compact non abelian groups that include C-, W- and multiplier dynamical systems. We prove results that relate the algebraic properties such as simplicity or primeness of the fixed point algebras as defined in Section 3., to the spectral properties of the action, including the Connes and strong Connes spectra. Keywords: dynamical system, compact group, simple C-algebra, prime C- algebra, von Neumann factor, Connes spectrum, strong Connes spectrum. 2010 Mathematics Subject Classifications: Primary 46L05, 46L10, 46L55, Secondary 46L40, 37B99. 1 Introduction In [2], Connes introduced the invariant Γ(U ) known as the Connes spectrum of the action U of a locally compact abelian group on a von Neumann algebra and used it in his seminal classification of type III von Neumann factors. Soon after, Olesen [10] defined the Connes spectrum of an action of a locally compact abelian group on a C-algebra. In [11], using the definition of the Connes spectrum in [10], it is proven an analog of a result of Connes and Takesaki [3, Chapter III, Corollary 3.4.] regarding the significance of the Connes spectrum of a locally compact abelian group action on a C-algebra for the ideal structure of the crossed product. In particular, in [11] is discussed a spectral characterization for the crossed product to be a prime C-algebra. This definition of the Connes spectrum in [10] cannot be used to prove similar results for the simplicity of the crossed product, unless the group is discrete [11]. Kishimoto [8] defined the strong Connes spectrum for C-dynamical systems with locally compact abelian groups that coincides with the Connes spectrum for the W-dynamical system and with the Connes spectrum defined by Olesen for discrete abelian group actions on C-algebras and he proved the Connes-Takesaki result for simple crossed products. In [2] Connes obtained results that relate the spectral properties of the von Neumann algebra with the algebraic properties of the fixed point algebra. These results were extended in [12] to C-algebras and compact abelian groups. In [6], [14] we considered the problems of simplicity 1 and primeness of the crossed product by compact, non abelian group actions. In particular, in [6] we have defined the Connes and strong Connes spectra for such actions that coincide with Connes spectra [2], [10], respectively with the strong Connes spectra [8] for compact abelian groups. Further, in [15] we have considered the case of one-parameter F -dynamical systems that include the C- the W- and the multiplier one-parameter dynamical systems. In particular, we have obtained extensions of some results in [2], [12] for W ∗−, respectively C∗− dynamical systems to the case of F -dynamical systems with compact abelian groups [15, Theorems 3.2 and 3.4.]. In this paper we will prove results for F - dynamical systems with compact non abelian groups. Our results contain and extend to the case of compact non abelian groups the following: [2 Proposition 2.2.2. b) and Theorem 2.4.1], [12, Theorem 2], [13, Theorem 8.10.4] and [15, Theorems 3.2. and 3.4.]. In Section 2. we will set up the framework and state some results that will be used in the rest of the paper. In Section 3. we discuss the connection between the strong Connes spectrum, eΓF (α), of the action and the F -simplicity of the fixed point algebras (X ⊗ B(Hπ))α⊗adπ. In Section 4. we will get similar results about the connection between the F -primeness of the fixed point algebras and the Connes spectrum, ΓF (α), of the action. 2 Notations and preliminary results This section contains the definitions of the basic concepts used in the rest of the paper, the notations and some preliminary results. 2.1. Definition. ([1], [16 ]) A dual pair of Banach spaces is, by definition, a pair (X, F ) of Banach spaces with the following properties: a) F is a Banach subspace of the dual X ∗ of X. b) kxk = sup {ϕ(x) : ϕ ∈ F , kϕk ≤ 1} , x ∈ X. c) kϕk = sup {ϕ(x) : x ∈ X, kxk ≤ 1} , ϕ ∈ F . d) The convex hull of every relativel y F -compact subset of X is relatively F -compact. e) The convex hull of every relatively X-compact subset of F is relatively X-compact. In the rest of the paper X will be assumed to be a C-algebra with the addi- tional property f ) The involution of X is F -continuous and the multiplication in X is sep- arately F -continuous. The property d) implies the existence of the weak integrals of continuous functions defined on a locally compact measure space, (S, µ) with values in X endowed with the F -topology: If f is such a function, we will denote by Z f (s)dµ 2 the unique element y of X such that ϕ(y) =ZS ϕ(f (s))dµ for every ϕ ∈ F [1, Proposition 1.2.]. The propery e) was used by Arveson [1, Proposition 1.4.] to prove the continuity in the F -topology of some linear mappings on X (in particular the mappings Pα(π) and (Pα)ij (π) defined below). 2.2. Examples. a) [1] If X is a C-algebra and and F = X ∗, conditions 1)-5) are satisfied. b) [1] If X is a W-algebra and F = X∗ is its predual then conditions 1)-5) are satisfied. c) [4] If X = M (Y ) is the multiplier algebra of Y and F = Y ∗ then conditions 1)-5) are satisfied. In addition, in this case, the F -topology on X is compatible with the strict toplogy on X = M (Y ). Let (X, F ) be a dual pair of Banach spaces G a compact group and α : G → Aut(X) a homeomorphism of G into the group of ∗−automorphisms of X. We say that (X, G, α) is an F -dynamical system if the mapping is continuous for every x ∈ X and ϕ ∈ F . g → ϕ(αg(x)) 2.3. Examples. a) If F = X ∗, the dual of X then, by [7 p. 306] the above condition is equivalent to the continuity of the mapping g → αg(x) from G to X endowed with the norm topology for every x ∈ X, so, in this case (X, G, α) is a C-dynamical system. b) If X is a von Neumann algebra and F = X∗, the predual of X then (X, G, α) is a W-dynamical system. c) If X = M (Y ) is the multiplier algebra of Y and F = Y ∗, then (X, G, α) is said to be a multiplier dynamical system. the set of unitary equivalence classes of irreducible representations of G. For Let (X, G, α) be an F -dynamical system with G compact. Denote by bG each π ∈ bG denote also by π a fixed representative of that class. If χπ(g) = i=1 πii(g−1) = dπP πii(g) is the character of π, denote by dπPdπ Pα(π)(x) =ZG χπ(g)αg(x)dg. Then Pα(π) is a projection of X onto the spectral subspace X1(π) = {x ∈ X : Pα(π)(x)} . where the integral is taken in the weak sense defined in (1) above. As in [14] one can also define for every 1 ≤ i, j ≤ dπ (Pα)ij (π)(x) =ZG πji(g)αg(x)dg. 3 where dπ is the dimension of the Hilbert space Hπ of π and show that (Pα)ij (π)(X) ⊂ X1(π). Using [1, Proposition 1.4.] it follows that Pα(π), (Pα)ij(π) are F -continuous. If π is the identity one dimensional representation ι of X, we will denote and Pα(ι) = Pα. X1(ι) = X α. is the fixed point algebra of the action. σ = X, where Pπ∈ bG X1(π) σ 2.4. Remark. Pπ∈ bG X1(π) closure of Pπ∈ bG X1(π) in the F -topology of X. π ∈ bG. Since, as noticed above, (Pα)ij (π)(X) ⊂ X1(π), it follows that Proof. Suppose that there exists ϕ ∈ F such that ϕ(X1(π)) = {0} for every denotes the ZG πij (g)ϕ(αg(x))dg = 0. for every x ∈ X and every π ∈ bG. Since nπij (g) : π ∈ bG, 1 ≤ i, j ≤ dπo is an orthogonal basis of L2(G), and ϕ(αg(x)) is a continuous function of g, for every x ∈ X, it follows that ϕ(x) = 0 for every x ∈ X so ϕ = 0 and we are done. In ([9], [14], [6]) it is pointed out that the spectral subspaces X2(π) = {a ∈ X ⊗ B(Hπ) : (αg ⊗ ι)(a) = a(1 ⊗ πg)} . where ι is the identity automorphism of B(Hπ) are, in some respects more useful. In [14] it is shown that X2(π) consists of all matrices {a = [(Pα)ij (π)(x)] (= [aij]) ∈ X ⊗ B(Hπ) : x ∈ X, 1 ≤ i, j ≤ dπ} . It is straightforward to prove that, if a ∈ X2(π) and x = Pi aii, then aij = (Pα)ij(π)(x). In what follows, if b ∈ X ⊗ B(Hπ) we will denote tr(b) =X bii which is an F -continuous linear mapping from X ⊗ B(Hπ) to X. The following lemma is proven for compact non abelian group actions on C-algebras in [6, Lemma 2.3.] and for compact abelian F -dynamical systems in [15]. Since the proof is very similar with the proof of [6, Lemma 2.3.] we will state it without proof 4 2.5. Lemma. Let ( X, G, α) be an F -dynamical system with G compact and J a two sided ideal of X α. Then σ (XJ X )α = F -closed linear span of ntr(X2(π)J X2(π)∗) : π ∈ bGo . where, if a = [akl] ∈ X ⊗ B(Hπ) and j ∈ X, by ja we mean the matrix [jakl] and the multiplications XJ X , X2(π)J X2(π)∗ are defined in 2.6. below. σ We will use the following notations 2.6. Notation. Let ( X, F ) be a dual pair of Banach spaces with X a C-algebra satisfying conditions 1)-6). If Y, Z are subsets of X denote: a) lin {Y } is the linear span of Y . b) Y ∗ = {y∗ : y ∈ Y } . c) Y Z = lin {yz : y ∈ Y, z ∈ Z} . d) Y = F −closure of Y in X. σ kk w = norm closure of Y. = the w∗-closure of Y in F ∗. e) Y f ) Y If ( X, G, α) is an F −dynamical system denote g) Hα σ (X) the set of all non-zero globally α−invariant F −closed hereditary C* −subalgebras of X. Notice that if ( X, G, α) is an F −dynamical system and if X2(π) is the spectral subspace defined above, then X2(π)X2(π)∗ is a two sided ideal of X α ⊗ B(Hπ) and X2(π)∗X2(π) is a two sided ideal of (X ⊗B(Hπ))α⊗adπ where α⊗adπ is the action (αg ⊗ adπg)(a) = (1 ⊗ πg)[αg(aij )](1 ⊗ πg−1 ). on X ⊗ B(Hπ). σ σ Corresponding to the above Arveson type spectra b) and c) we define two Connes type spectra d) ΓF (α) = ∩ {spF (αY ) : Y ∈ Hα 2.7. Definition. a) sp(α) =nπ ∈ bG : X1(π) 6= {0}o . b) spF (α) =nπ ∈ bG : X2(π)∗X2(π) is essential in (X ⊗ B(Hπ))α⊗adπo . c) espF (α) =nπ ∈ bG : X2(π)∗X2(π) = (X ⊗ B(Hπ))α⊗adπo . e) eΓF (α) = ∩ {espF (αY ) : Y ∈ Hα Clearly, espF (α) ⊂ spF (α) ⊂ sp(α), so eΓF (α) ⊂ ΓF (α). The definition of eΓF (α) is a direct generalization of the strong Connes spectrum of Kishimoto to compact non abelian groups. Our motivation for the definition of ΓF (α) above (and Γ(α) for C-dynamical systems in [6]) is the following observation σ (X)} . σ (X)} . 5 2.8. Remark a) If ( X, G, α) is an F −dynamical system with G compact abelian, then ∩ {sp(αY ) : Y ∈ Hα σ (X)} = ∩ {spF (αY ) : Y ∈ Hα σ (X)} . and the left hand side of the above equality is the Connes spectrum for W −as well as for C-dynamical systems. b) If G is not abelian, the equality in part a) is not true. Proof. a) We have to prove only one inclusion, the opposite one being obvious. Let γ ∈ ∩ {sp(αY ) : Y ∈ Hα γ Yγ = {0} γ = {0} . Therefore, if we denote Z = aY a∗σ for some a ∈ Y α, a 6= 0. Then aY ∗ , σ (X) and Z ∗ it follows that Z ∈ Hα γ = {0} which is in contradiction with the hypothesis that γ ∈ γ ∈ ∩ {sp(αY ) : Y ∈ Hα σ (X). Suppose that aY ∗ σ (X)} and Y ∈ Hα σ (X)} ⊂ sp(αZ ). b) In [14, Example 3.9.] we provided an example of an action of an action of G = S3 the permutation group on three elements on the algebra X of 2 × 2 matrices such that sp(α) = bG, H α σ = {X} , so ∩ {sp(αY ) : Y ∈ Hα σ (X)} = sp(α) and we have shown that there exists π ∈ bG such that (X ⊗ B(Hπ))α⊗adπ has nontrivial center and, therefore, it is not a prime C-algebra. By [6, Thm. 2.2.], it follows that ∩ {sp(αY ) : Y ∈ Hα σ (X)} = Γ(α). σ (X)} 6= ∩ {spF (αY ) : Y ∈ Hα 3 F − simple fixed point algebras Let (X, G, α) be an F -dynamical system with G compact. In the rest of this paper we will study how the F -simplicity (respectively F -primeness) as defined below, of the fixed point algebras (X ⊗ B(Hπ))α⊗adπ is reflected in the spectral properties of the action. 3.1. Definition. Let (B, F ) be a dual pair of Banach spaces with B a C-algebra. a) B is called F -simple if every non zero two sided ideal of B is F -dense in B. b) B is called F -prime if the annihilator of every non zero two sided ideal of B is trivial, or, equivalently, every non zero two sided ideal of B is an essential ideal (using Definition 2.1. f ) it is easy to see that X is F −prime if and only if X is prime as a C-algebra). Let ( X, G, α) be an F -dynamical system. c) X is called α−simple if every non zero α−invariant two sided ideal of X is F −dense in X. 6 d) X is called α−prime if every non zero α−invariant two sided ideal of X is an essential ideal. In the particular case when B is a C-algebra and F = B∗ is its dual, then, clearly, the concepts of F −simple, (respectively F −prime) in the above Definition 3.1. a) (respectively b)) coincide with the usual concepts of simple (respectively prime) C-algebras. Similarly, if ( X, G, α) is a C-dynamical sys- tem, that is if X is a C-algebra and F = X ∗ is its dual, then the notions of α−simple and α−prime coincide with the usual ones for C-dynamical systems. If B is a von Neumann algebra and F = B∗ is its predual, then, since the weak closure of every essential ideal equals B, it follows that B is F -simple if and only if B is F −prime, so, if and only if B is a factor. It is also obvious that if ( X, G, α) is a W-dynamical system, that is if X is a von Neumann algebra and F = X∗ is its predual, then X is α−simpl e if and only if it is α−prime, and this holds if and only if α acts ergodically on the center of X (i.e. every fixed element in the center of X is a scalar). The above observations and the next Remark show that for W-dynamical systems, (X, G, α) with G compact, the results in the current Section 3 and Section 4 are equivalent. 3.2. Remark. Let ( X, G, α) be a W-dynamical system, that is, an F - dynamical system with X a von Neumann algebra and F = X∗ its predual. Then eΓF (α) = ΓF (α). Proof. This follows from the fact that if X is a von Neumann algebra, p ∈ X α an α-invariant projection and pX2(π)p∗X2(π)p is essential in (pXp ⊗ B(Hπ))α⊗adπ, then pX2(π)p∗X2(π)p =(pXp ⊗ B(Hπ))α⊗adπ. σ σ The next lemma will be used in the proofs of the main results of the current Section 3 and the next Section. 3.3. Lemma. Let (B, G, α) be an F -dynamical system with G compact. Then a) If {eλ} is an approximate identity of Bα in the norm topology, then (norm) lim λ eλx = (norm) lim λ xeλ = (norm) lim λ eλxeλ = x. for every x ∈Pπ∈ bG B1(π) kk . b) If b ∈ B is such that BαbBα = {0} then b = 0. c) BαBBα d) BαB1(π) = BαB σ = B1(π)Bα = BBα = B. σ σ σ σ = B1(π), π ∈ bG. Proof. a) This follows from the proof of [5, Lemma 2.7] in the more general case of compact quantum group actions. 7 b) If {eλ} is an approximate identity of Bα, then eλbeλ = 0 implies eλPα(πij )(b)eλ = Pα(eλbeλ) = 0. for every π ∈ bG, 1 ≤ i, j ≤ dπ, so, by a), Pα(πij )(b) = 0 Therefore, πji(g)ϕ(αg(b))dg = 0. ϕ(Pα(πij )(b)) =ZG for every ϕ ∈ F , π ∈ bG, 1 ≤ i, j ≤ dπ. Sincenπij (g) : π ∈ bG, 1 ≤ i, j ≤ dπo form an orthogonal basis of L2(G), and ϕ(αg(b)) is continuous on G, it follows that ϕ(αg(b)) = 0 for every g ∈ G, ϕ ∈ F , so b = 0. c) We will prove only that BαBBα = B, the proofs of the other equalities σ being similar. Let {eλ} be an approximate identity of Bα. By a), (norm) lim λ eλxeλ = xf oreveryx ∈ Xπ∈ bG kk B1(π) . Therefore kk B1(π) ⊂ BαBBα kk ⊂ BαBBα σ . Xπ∈ bG Since, by Remark 2.4., the F -closure of Pπ∈ bG B1(π) kk B = Xπ∈ bG σ B1(π) ⊂ BαBBα σ . so BαBBα σ = B. d) The proof is similar with the proof of part c). equals B it follows that Theorem 3.4. below is an extension of [2, Proposition 2.2.2. b)] to the case of F -dynamical systems with compact groups, not neccessarily abelian, for the strong Connes spectrum, eΓF (α). 3.4. Theorem. Let (X, G, α) be an F -dynamical system with G compact. Then eΓF (α) = ∩ {espF (αJXJ σ ) : J ⊂ X α, F -closed two sided ideal } Proof. Clearly, since J XJ σ ∈ Hα σ (X), eΓF (α) ⊂ ∩ {espF (αJXJ σ ) : J ⊂ X α, F -closed two sided ideal} . so Y α ∈ Hσ(X α). We will prove that Y2(π)∗Y2(π) σ ) : J ⊂ X α, F -closed two sided ideal} and Y ∈ Hα σ σ (X), = (Y ⊗ B(Hπ))α⊗adπ and Let π ∈ ∩ {espF (αJXJ 8 thus π ∈ espF (αY ). Since Y ∈ Hα Denote by J the following ideal of X α σ (X) is arbitrary, it will follow that π ∈ eΓF (α). J = X αY αX α σ . It is clear that J = J X αJ holds without the closure, but we do not need this fact). Also (actually it is quite easy to show that this equality σ Y αJ Y α σ = Y αX αY αX αY α σ = (Y αX αY α)(Y αX αY α) σ = (1) = Y αY α σ = Y α. Denote Z = J XJ σ . Notice that, since Y ∈ Hα σ (X), we have Y αXY α σ = Y, so Z = X αY αX α σ XX αY αX α σ = X αY αX αXX αY αX α σ X αY α(X αXX α)Y αX α σ = X αY αXY αX α σ = X αY X α = σ . (2) σ ) : J ⊂ X α, F -closed two sided ideal}, it follows that Since π ∈ ∩ {espF (αJXJ π ∈ espF (αZ ), so Z2(π)∗Z2(π) σ = (Z ⊗ B(Hπ))α⊗adπ. (3) Using the equalities (2) above, the fact that Y is a hereditary C-subalgebra of X, and the obvious equality Pij (π)(xyz) = xPij (π)(y)z for every x, z ∈ X α, y ∈ X, and 1 ≤ i, j ≤ dim Hπ, the relation (3) becomes X αY2(π)∗Y2(π)X α σ = X α(Y ⊗ B(Hπ))α⊗adπX α σ . (4) where, for x ∈ X α and a ∈ X ⊗ B(Hπ), a = [akl] , by xa we mean the matrix whose kl entry is xakl. Therefore, by applying Lemma 3.3. d) to B = Y, we get X αY αY2(π)∗Y2(π)Y αX α σ = X αY α(Y ⊗ B(Hπ))α⊗adπY αX α σ . (5) By multiplying (5) on the right and on the left by Y α and taking into account = Y α, it that, by Lemma 3.3. c) Y αY Y α follows that = Y and consequently, Y αX αY α σ σ σ Y2(π)∗Y2(π) = (Y ⊗ B(Hπ))α⊗adπ. Therefore, π ∈ espF (αY ) and the proof is complete. In the next Lemma and the rest of the paper, a subalgebra of X ⊗ B(Hπ) will be called F −simple (respectively F −prime) if it is F ⊗ B(Hπ)∗−simple (respectively F ⊗ B(Hπ)∗−prime) where B(Hπ)∗ denotes the dual of B(Hπ). Clearly, a subalgebra of X ⊗ B(Hπ) is F −prime if and only if it is a prime C-algebra. The similar statement for the F −simple case is not true. 9 3.5. Lemma. Let (X, G, α) be an F -dynamical system with G compact. Then, if X α is F -simple, it follows that ( X ⊗ B(Hπ))α⊗adπ is F −simple. Proof. Let π ∈ espF (α) ⊂ sp(α). Since X α is F -simple, so X α ⊗ B(Hπ) is also F -simple and X2(π)X2(π)∗ is an ideal of X α ⊗ B(Hπ), it follows that = X α ⊗ B(Hπ). To prove that (X ⊗ B(Hπ))α⊗adπ is simple, let X2(π)X2(π)∗ I ⊂ (X ⊗ B(Hπ))α⊗adπ be a non-zero ideal. Then it can be easily verified that σ σ J = lin {yy∗ : y ∈ X2(π)I} = = X2(π)IX2(π)∗ σ . is an ideal of X α⊗B(Hπ) and, since the latter algebra is F -simple, it follows that = (X ⊗ = X2(π), J = X α ⊗ B(Hπ). Therefore, since π ∈ espF (α), we have X2(π)∗X2(π) σ B(Hπ))α⊗adπ and consequently, since, by Lemma 3.3. d) X αX2(π) we have σ (X ⊗ B(Hπ))α⊗adπ = X2(π)∗J X2(π) σ ⊂ X2(π)∗X2(π)IX2(π)∗X2(π) σ ⊂ I. Thus I = (X ⊗ B(Hπ))α⊗adπ and we are done. The following result extends [2, Th´eor`eme 2.4.1], [12, Theorem 2. i)⇔ ii)]] and [15, Theorem 3.4.] to the more general case of F -dynamical systems and non abelian compact groups G. 3.6. Theorem. Let (X, G, α) be an F -dynamical system with G compact. The following conditions are equivalent: i) ( X ⊗ B(Hπ))α⊗adπ is F -simple for all π ∈ sp(α). π ∈ sp(α) be arbitrary. Since, in particular, X α is F -simple, so it has no Proof. i) ⇒ ii) Suppose that (X ⊗ B(Hπ))α⊗adπ is F -simple for all π ∈ sp(α). ii) X is α-simple and sp(α) = eΓF (α). Then, it follows immediately from the definitions that sp(α) = espF (α). Let non-trivial F -closed ideals, from Theorem 3.4. it follows that π ∈ eΓF (α), so sp(α) = eΓF (α). Let us prove that X is α-simple. If I is an F -closed α-invariant ideal of X, then I α is an F -closed ideal of X α, so I α = X α. By Lemma 3.3. c) applied to B = I, and to B = X it follows that I αII α = X, so = I and X αXX α σ σ X = X αXX α = I αXI α ⊂ I. σ σ Therefore, I = X, hence X is α-simple. ii) ⇒ i). Suppose that X is α-simple and sp(α) = eΓF (α). We will prove first that X α is F -simple. Let J ⊂ X α be a non zero ideal and π ∈ eΓF (α). Since ∈ Hα σ (X), and π ∈ eΓF (α), it follows that J XJ σ σ J X2(π)∗J X2(π)J = J(X ⊗ B(Hπ))α⊗adπJ σ . (6) 10 where, for j ∈ J ⊂ X α and a ∈ X ⊗ B(Hπ), a = [akl] , by ja we mean the matrix whose kl entry is jakl. By multiplying the above relation on the left by X2(π) and on the right by X2(π)∗, we get X2(π)J X2(π)∗J X2(π)J X2(π)∗ σ = X2(π)J(X ⊗ B(Hπ))α⊗adπJ X2(π)∗ σ . (7) From the above relations (6) and (7) it follows that X2(π)J X2(π)∗ = X2(π)J X2(π)∗ σ ⊂ X2(π)J(X ⊗ B(Hπ))α⊗adπJ X2(π)∗ σ = = X2(π)J X2(π)∗J X2(π)J X2(π)∗ σ ⊂ (X α ⊗ B(Hπ))J(X α ⊗ B(Hπ)) σ ⊂ J ⊗ B(Hπ) It follows that tr(X2(π)J X2(π)∗) ⊂ J. From Lemma 2.5. it follows that (XJ X = X, so J = X α and therefore, X α is J. Since X is α-simple, we have XJ X F -simple. Applying lemma 3.5. it follows that (X ⊗ B(Hπ))α⊗adπ is F -simple σ σ )α ⊂ for all π ∈ sp(α) = eΓF (α). 4 F -prime fixed point algebras This section is concerned with the relationship between the F -primeness of the fixed point algebras and the spectral properties, involving the Connes spectrum ΓF (α) of the F -dynamical system (X, G, α). Theorem 4.1. below is an extension of [2, Proposition 2.2.2. b)] to the case of F -dynamical systems with compact groups, not neccessarily abelian, for the Connes spectrum, ΓF (α). By Remark 3.2. and the discussion preceding it, if (X, G, α) is a W-dynamical system (that is X is a von Neumann algebra and F = X∗ its predual), then the next Theorem 4.1. is equivalent with Theorem 3.4. 4.1. Theorem. Let (X, G, α) be an F -dynamical system. Then ΓF (α) = ∩ {spF (αJXJ σ ) : J ⊂ X α, a non-zero F -closed two sided ideal } . Proof. Since J XJ J ⊂ X α, we have, σ ∈ Hα σ (X) for every non-zero F -closed two sided ideal ΓF (α) ⊂ ∩ {spF (αJXJ σ ) : J ⊂ X α, F -closed two sided ideal} . Now let π ∈ ∩ {spF (αJXJ Hα (Y ⊗B(Hπ))α⊗adπ. As in the proof of Theorem 3.4., let J = X αY αX α σ ) : J ⊂ X α, F -closed two sided ideal} and Y ∈ is essential in and Z = σ (X), so Y α ∈ Hσ(X α). We will prove that Y2(π)∗Y2(π) σ σ 11 σ ∈ H α σ (X). Since π ∈ ∩ {spF (αJXJ J XJ we have that π ∈ spF (αZ ). Therefore, Z2(π)∗Z2(π) B(Hπ))α⊗adπ. As noticed in the proof of Theorem 3.4., σ ) : J ⊂ X α, F -closed two sided ideal} , σ is essential in (Z ⊗ Z2(π)∗Z2(π) σ = X αY2(π)∗Y2(π)X α σ . and (Z ⊗ B(Hπ))α⊗adπ = X α(Y ⊗ B(Hπ))α⊗adπX α σ . Let a ∈ (Y ⊗ B(Hπ))α⊗adπ be such that aY2(π)∗Y2(π) σ = {0} . Then, by Lemma 3.3. c) Y = Y αY Y2(π)∗ = Y αY2(π)∗ . Then, it follows that σ σ = Y αX αY αY σ and, by Lemma 3.3. d), σ aY2(π)∗Y2(π) = aY αY2(π)∗Y2(π) σ = aY αX αY αY2(π)∗Y2(π) σ = {0} . Therefore Y αaY αX αY αY2(π)∗Y2(π)X α σ σ σ = {0} . so, since Z2(π)∗Z2(π) we have Y αaY α = 0 and therefore, by Lemma 3.3. b) applied to B = Y, it fol- lows that a = 0. = X αY2(π)∗Y2(π)X α is essential in X α(Y ⊗ B(Hπ))α⊗adπX α σ , 4.2. Lemma. Let (X, G, α) be an F -dynamical system with G compact. Then, if X α is F -prime, it follows that ( X ⊗ B(Hπ))α⊗adπ is F −prime for every π ∈ spF (α). Proof. Since X α is F -prime. it follows that X α ⊗ B(Hπ) is F -prime for every π ∈ bG. Since X2(π)X2(π)∗ is a non zero ideal of X α ⊗ B(Hπ), it follows that X2(π)X2(π)∗is an essential ideal. To prove that (X ⊗ B(Hπ))α⊗adπ is F -prime, let I ⊂ (X ⊗ B(Hπ))α⊗adπ be a non-zero ideal. Then, as in the proof of Lemma 3.5., consider the following ideal of X α ⊗ B(Hπ) σ J = lin {yy∗ : y ∈ X2(π)I} = = X2(π)IX2(π)∗ σ . Since X α ⊗ B(Hπ) is F -prime, it follows that J is essential in X α ⊗ B(Hπ). Therefore, if a ∈ (X ⊗ B(Hπ))α⊗adπ and aI = {0} . we have aX2(π)∗J X2(π) ⊂ aX2(π)∗X2(π)IX2(π)∗X2(π) σ σ ⊂ aI = {0} . so (X2(π)aX2(π)∗J)X2(π)X2(π)∗ = {0} Thus, since X2(π)X2(π)∗ is essential in X α⊗B(Hπ), it follows that X2(π)aX2(π)∗J = {0} . Since X2(π)aX2(π)∗ ⊂ X α ⊗ B(Hπ), J is essential in X α ⊗ B(Hπ) and π ∈ spF (α), it follows that a = 0. 12 The next result extends [2, Th´eor`eme 2.4.1.], and [13, Theorem 8.10.4.] to the case of F -dynamical systems with compact non abelian groups. 4.3. Theorem. Let (X, G, α) be an F -dynamical system with G compact. The following conditions are equivalent: i) ( X ⊗ B(Hπ))α⊗adπ is F -prime for all π ∈ sp(α). ii) X is α-prime and sp(α) = ΓF (α). Proof. i) ⇒ ii) Suppose that (X ⊗ B(Hπ))α⊗adπ is F -prime for all π ∈ sp(α). Then, it follows immediately from i) and the definitions that sp(α) = spF (α). Let π ∈ sp(α) be arbitrary. We will use Theorem 4.1. to show that π ∈ ΓF (α). Indeed, let J be a non-trivial ideal of X α and Z = J XJ σ (X). We will show that π ∈ sp(αZ ), that is Z2(π)∗Z2(π) is.esential in (Z ⊗ B(Hπ))α⊗adπ. Notice that ∈ H α σ so Z2(π) = J X2(π)J. Z2(π)∗Z2(π) = J X2(π)∗J X2(π)J. Since, in particular, X α is prime, and J is an essential ideal of X α, we have Z2(π) 6= {0} . Indeed as observed after 2.6., X2(π)X2(π)∗ is an ideal of X α ⊗ B(Hπ), so, as X α is F -prime it follows that X α ⊗ B(Hπ) is a prime C-algebra and therefore J X2(π)X2(π)∗ 6= {0} , hence J X2(π) 6= {0} and X2(π)∗J 6= {0} , so, X2(π)∗J X2(π) 6= {0} . Using the hypothesis that ( X ⊗ B(Hπ))α⊗adπ is F −prime and the fact that J is a non-trivial ideal of X α, it follows that X2(π)∗J X2(π)J 6= {0} , so, Z2(π) 6= {0} . As noticed above, X2(π)∗J X2(π) is a non-trivial ideal of (X ⊗ B(Hπ))α⊗adπ. If a ∈(Z ⊗ B(Hπ))α⊗adπ, a ≥ 0 is such that aZ2(π)∗Z2(π) = {0} , then Hence aJ X2(π)∗J X2(π)J = {0} . J aJ X2(π)∗J X2(π)J = {0} . Since, as noticed above, X2(π)∗J X2(π) is non-trivial and (X ⊗ B(Hπ))α⊗adπ is F -prime it follows that J aJ = {0} . so J a = {0} . Hence J tr(a) = {0} . Since X α is F -prime, we deduce that tr(a) = 0, so a = 0 because a was assumed to be non negative. Therefore, π ∈ ΓF (α), so sp(α) = ΓF (α). It remains to prove that X is F -prime. Let I ⊂ X be an α-invariant non-trivial ideal and x ∈ X, x ≥ 0 be such that xI = {0} . Then, in particular, xI α = {0} , so P (x)I α = {0} . Since X α is F -prime and I α is a non trivial ideal of X α we have P (x) = 0 so, since P is faithful, x = 0. ii)⇒ i) Suppose that X is α-prime and sp(α) = ΓF (α). We will prove first that X α is F -prime. Let J ⊂ X α be a non zero ideal and a ∈ X α, a ≥ 0, a 6= 0 such 13 that J a = {0} . Since X is α−prime, and XJ X is a non zero α-invariant ideal of X, it follows that XJ Xa 6= {0} so J Xa 6= {0} . Therefore, since by Remark 2.4., X = Xπ∈sp(α) σ X1(π) . σ σ (X). there exists π ∈ sp(α) such that J X1(π)a 6= {0} . Denote Z = aXa Then, since π ∈ sp(α) = ΓF (α), Z2(π)∗Z2(π) is essential in (Z ⊗ B(Hπ))α⊗adπ. But ∈ H α Z2(π)∗Z2(π) = aX2(π)∗a2X2(π)a σ σ . and (Z ⊗ B(Hπ))α⊗adπ = a((X ⊗ B(Hπ))α⊗adπ)a σ . Taking into account that X2(π)a2X2(π)∗ ⊂ X α ⊗ B(Hπ) and J a = {0} we immediately get that J X2(π)a2X2(π)∗ ⊂ J ⊗ B(Hπ). Hence Therefore It follows that (aX2(π)∗J)X2(π)a2X2(π)∗(a2X2(π)a) σ = {0} . (aX2(π)∗J X2(π)a)(aX2(π)∗a2X2(π)a) σ = {0} . (aX2(π)∗J X2(π)a)(Z2(π)∗Z2(π)) = {0} . Since Z2(π)∗Z2(π) is essential in (Z⊗B(Hπ))α⊗adπ and obviously aX2(π)∗J X2(π)∗a ⊂ (Z ⊗ B(Hπ))α⊗adπ it follows that J X2(π)a = {0} and hence J X1(π)a = {0} , but this is in contradiction with our choice of π in sp(α) so X α is F -prime. it follows that (X ⊗ B(Hπ))α⊗adπ is F -prime for all π ∈ From Lemma 4.2. ΓF (α) = spF (α) = sp(α) and we are done. References [1] W.B. Arveson, On groups of automorphisms of operator algebras, J. Funct. Anal. 15 (1974) 217-243. [2] A. Connes, Une classification des facteurs de type III, Ann. Sci. ´Ec. Norm. Sup´er. 6 (1973) 133-252. [3] A. Connes, M. Takesaki, The flow of weights on factors of type III, Tohoku Math. J. 29 (1977) 473-575. [4] C. D'Antoni, L. Zsido, Groups of linear isometries on multiplier C- algebras, Pacific J. Math. 193 (2000) 279-306. [5] R. Dumitru, C. Peligrad, Compact quantum group actions and invariant derivations, Proc. Amer. Math. Soc. 135 (2007) 3977-3984. 14 [6] E. C. Gootman, A. J. Lazar, C. Peligrad, Spectra for compact group actions, J. Operator Theory 31 (1994) 381-399. [7] E. Hille, R. Phillips, Functional Analysis and Semi-Groups, AMS, 1957. [8] A. Kishimoto, Simple crossed products of C-algebras by locally compact abelian groups, Yokohama Math. J. 28 (1980) 69-85. [9] M. B. Landstad, Algebras of spherical functions associated with covariant systems over a compact group, Math. Scand. 47 (1980), 137-149. [10] D. Olesen, Inner ∗-automorphisms of simple C-algebras, Comm. Math. Phys..44 (1975) 175-190. [11] D. Olesen, G.K. Pedersen, Applications of the Connes spectrum to C- dynamical systems, J. Funct. Anal. 30 (1978) 179-197. [12] D. Olesen, G.K. Pedersen, E. Stormer, Compact abelian groups of au- tomorphisms of simple C-algebras, Invent. Math. 39 (1977) 55-64. [13] G.K. Pedersen, C-algebras and Their Automorphism Groups, Aca- demic Press, 1979. [14] C. Peligrad, Locally compact group actions and compact subgroups, J. Funct. Anal. 76 (1988) 126-139. [15] C. Peligrad, A solution of the maximality problem for one-parameter dynamical systems, Adv. Math. 329 (2018) 742-780. [16] L. Zsido, Spectral and ergodic properties of the analytic generators, J. Approx. Theory 20 (1977) 77-138. 15
1303.4371	2	1303	2013-07-04T14:48:59	Decomposition rank of UHF-absorbing C*-algebras	[ "math.OA" ]	Let A be a unital separable simple C*-algebra with a unique tracial state. We prove that if A is nuclear and quasidiagonal, then A tensored with the universal UHF-algebra has decomposition rank at most one. Then it is proved that A is nuclear, quasidiagonal and has strict comparison if and only if A has finite decomposition rank. For such A, we also give a direct proof that A tensored with a UHF-algebra has tracial rank zero. Applying this characterization, we obtain a counter-example to the Powers-Sakai conjecture.	math.OA	math	Decomposition rank of UHF-absorbing C∗-algebras Hiroki Matui Yasuhiko Sato Graduate School of Science Graduate School of Science Chiba University Kyoto University Inage-ku, Chiba 263-8522, Japan Sakyo-ku, Kyoto 606-8502, Japan Abstract Let A be a unital separable simple C∗-algebra with a unique tracial state. We prove that if A is nuclear and quasidiagonal, then A tensored with the universal UHF- algebra has decomposition rank at most one. Then it is proved that A is nuclear, quasidiagonal and has strict comparison if and only if A has finite decomposition rank. For such A, we also give a direct proof that A tensored with a UHF-algebra has tracial rank zero. Applying this characterization, we obtain a counter-example to the Powers-Sakai conjecture. 1 Introduction E. Kirchberg and W. Winter [13] introduced the notion of decomposition rank for nuclear C∗-algebras as a generalization of topological dimension. Among others, they showed that finite decomposition rank implies quasidiagonality. The concept of decomposition rank is particularly relevant for Elliott's program to classify nuclear C∗-algebras by K-theoretic data. Indeed, all nuclear stably finite C∗-algebras classified so far by their Elliott invariants have in fact finite decomposition rank. Also, this property is expected to be equivalent to other regularity properties. W. Winter [33] proved that any unital separable simple infinite-dimensional C∗-algebra A with finite decomposition rank is Z-absorbing, i.e. A absorbs the Jiang-Su algebra Z tensorially. M. Rørdam [27] showed that Z-absorption implies strict comparison for unital simple exact C∗-algebras. In the present paper we provide the converse direction of these results under the assumption that the algebra has a unique tracial state. Theorem 1.1. Let A be a unital separable simple C∗-algebra with a unique tracial state. If A is nuclear, quasidiagonal and has strict comparison, then the decomposition rank of A is at most three. The notion of strict comparison was first introduced by B. Blackadar [2] for projections, which was called FCQ2. Later this definition is extended to positive elements in the following sense. Throughout this paper, for a C∗-algebra A we denote by T (A) the set of tracial states on A. We say that A has strict comparison if for two positive elements a, b in Mk(A) with limn→∞ τ (a1/n) < limn→∞ τ (b1/n) for any τ ∈ T (A) there exist rn ∈ Mk(A), n ∈ N such that r∗ nbrn → a, where τ is regarded as an unnormalized trace on Mk(A). The precise definition of decomposition rank is given in Definition 4.1 and the proof of the main theorem is given in Section 5. 1 In 2008, A. S. Toms and W. Winter conjectured that properties of finite decomposi- tion rank, Z-absorption, and strict comparison are equivalent for unital separable simple infinite-dimensional finite nuclear C∗-algebras (see [33, Conjecture 0.1] for example). The theorem above, together with the results in [13, 27, 33, 20], gives a partial affirmative answer to this conjecture in the following sense. Corollary 1.2. Let A be a unital separable simple infinite-dimensional nuclear C∗-algebra with a unique tracial state. Then the following are equivalent. (i) A has finite decomposition rank. (ii) A is quasidiagonal and is Z-absorbing. (iii) A is quasidiagonal and has strict comparison. The technique that we develop in proving Theorem 1.1 also allows us to bound other ranks of nuclear C∗-algebras. In Section 6, for A as in Theorem 1.1, we show that A tensored with a UHF-algebra has tracial rank zero. This enables us to remove the technical condition from H. Lin's result [16, Corollary 5.10] for nuclear C∗-algebras, and to provide a partial answer to his conjecture mentioned in the introduction of [16]. As an application, we construct a counter-example to the Powers-Sakai conjecture. In Section 7, we show that any Kirchberg algebra has nuclear dimension at most three. We remark that W. Winter and J. Zacharias [34, Theorem 7.5] proved that if A is a Kirchberg algebra satisfying the Universal Coefficient Theorem (UCT), then it has finite nuclear dimension. 2 Murray-von Neumann equivalence A celebrated result of A. Connes says that injective factors with separable predual are approximately finite dimensional (AFD), see [8]. The proof relies on his deep study of automorphisms of factors [6, 7]. By now, two alternative proofs which do not use auto- morphisms are known. One is due to U. Haagerup [9]. His proof uses trace preserving completely positive maps instead of automorphisms. S. Popa gave another short proof by using excision of amenable traces [23]. In the present paper, we focus on the proofs by Connes and Haagerup, in which the Murray-von Neumann equivalence for projections plays an essential role. They constructed approximation by finite dimensional subalgebras by using partial isometries inducing the Murray-von Neumann equivalence for certain projections. In this section we establish a generalization of the Murray-von Neumann equivalence (Lemma 2.1). In our setting, the 'equivalence' is given by two contractions. These two contractions will become involved in the estimate of decomposition rank of UHF-absorbing C∗-algebras in Section 4. We say that a C∗-algebra A has strict comparison for projections if for any two pro- jections p and q in Mk(A) satisfying τ (p) < τ (q) for any tracial state τ of A, there exists a partial isometry v in Mk(A) such that v∗v = p and vv∗ ≤ q. Lemma 2.1. Let A be a unital C∗-algebra with strict comparison for projections. Let p, q be projections in A and n ∈ N. If τ (p) = τ (q) for any tracial state τ of A, then there exist 2 two contractions vi in A ⊗ Mn, i = 0, 1 such that (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) and Xi=0,1 v∗ i vi − p ⊗ 1n(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ≤ 4 n , (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=0,1 ≤ 4 n , viv∗ i − q ⊗ 1n(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2 n dist(v∗ i vi, {p ⊗ x x ∈ Mn}) ≤ for i = 0, 1. Proof. Let ej, j = 1, 2, . . . , n be mutually orthogonal minimal projections of Mn. When n = 1, 2, or 3 the statement is trivial, thus we may assume that n ≥ 4. Since A has strict comparison for projections, there exists a partial isometry r1 ∈ A ⊗ Mn such that r∗ 1r1 = p ⊗ e1, r1r∗ 1 ≤ q ⊗ (e1 + e2). By the same reason, we obtain a partial isometry r2 ∈ A ⊗ Mn such that r∗ 2r2 ≤ p ⊗ (e2 + e3), r2r∗ 2 = q ⊗ (e1 + e2) − r1r∗ 1. In the same way, there exists a partial isometry r3 ∈ A ⊗ Mn such that r∗ 3r3 = p ⊗ (e2 + e3) − r∗ 2r2, r3r∗ 3 ≤ q ⊗ (e3 + e4). Repeating this argument we have partial isometries rj ∈ A ⊗ Mn, j = 1, 2, . . . , n − 1. Set Since {rjr∗ projections, it follows that j j = 1, 2, . . . , n−1} and {r∗ j rjj = 1, 2, . . . , n−1} are sets of mutually orthogonal n ej, f = n − j nXj=1 0v0 − p ⊗ f k =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) n−1Xj=1 0 − q ⊗ f k =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) n−1Xj=1 kv∗ kv0v∗ v0 = n − j n n − j n rj. n n−1Xj=1r n − j j rj − p ⊗ f(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) j − q ⊗ f(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) r∗ rjr∗ ≤ 2 n ≤ 2 n . and Similarly we can construct a contraction v1 ∈ A ⊗ Mn such that and thereby completing the proof. kv∗ 1v1 − p ⊗ (1n − f )k ≤ kv1v∗ 1 − q ⊗ (1n − f )k ≤ 2 n 2 n , 3 3 Relative Property (SI) In our previous work [21, Proposition 4.8], we showed certain strict comparison of central sequence algebras by using a technical condition named property (SI). The aim of this section is to introduce a relative version of property (SI) and to show a variant of [21, Proposition 4.8]. Until the end of this section, we let A and B be unital separable simple nuclear infinite- dimensional C∗-algebras and suppose that B has strict comparison. First, we recall the definitions of central sequence algebras. Let ω be a free ultrafilter on N, and let kbk2 = max τ ∈T (B) τ (b∗b)1/2 for b ∈ B, cω(B) =n(bn)n ∈ ℓ∞(N, B) J =n(bn)n ∈ ℓ∞(N, B) lim n→ω kbnk = 0o . kbnk2 = 0o lim n→ω We denote by Bω the ultraproduct C∗-algebra ℓ∞(N, B)/cω(B). Set and M = ℓ∞(N, B)/J. Since cω(B) ⊂ J we regard M as a quotient algebra of Bω, and denote by Φ the quotient map from Bω to M . Note that if B has finitely many extreme traces then M is isomorphic to a finite direct sum of II1-factors. Let π be a unital embedding of A into Bω. We consider the relative commutants B = Bω ∩ π(A)′ and M = M ∩ Φ(π(A))′. The Choi-Effros lifting theorem allows us to get a sequence of unital completely positive maps πn : A → B such that (πn(a))n = π(a) in Bω for any a ∈ A. The following theorem is based on the technique of Kirchberg and Rørdam [12, The- orem 3.3]. Combining this result and a generalized property (SI) (Lemma 3.2), we shall show the strict comparison of B for projections in Proposition 3.3. Theorem 3.1 (Kirchberg-Rørdam). The restriction ΦB : B → M is surjective. Proof. Let x be a contraction in M, and (bn)n ∈ Bω be such that Φ((bn)n) = x. Set D = C ∗(π(A) ∪ {(bn)n}) ⊂ Bω and I = ker(ΦD) ⊂ D. Since x ∈ Φ(π(A))′ it follows that [π(a), (bn)n] ∈ I for any a ∈ A. Let eλ ∈ I, λ ∈ Λ be a quasicentral approximate unit of I. Then we have 0 = lim λ→∞ = lim λ→∞ k(1 − eλ)[(bn)n, π(a)](1 − eλ)k k[(1 − eλ)(bn)n(1 − eλ), π(a)]k for any a ∈ A. This implies that for any finite subset F of A and ε > 0 there exists a sequence eF,ε,n, n ∈ N of positive contractions in B such that lim n→ω keF,ε,nk2 = 0, lim n→ω k[(1 − eF,ε,n)bn(1 − eF,ε,n), πn(a)]k < ε, for all a ∈ F . 4 Let Fn, n ∈ N be an increasing sequence of finite subsets of A whose union is dense in A and εn > 0, n ∈ N a decreasing sequence which converges to 0. For any n ∈ N we inductively obtain a subset Ωn ∈ ω such that Ωn+1 ⊂ Ωn,Tn∈N Ωn = ∅ and k(1 − eFn,εn,m)bm(1 − eFn,εn,m), πm(a)]k < εn, keFn,εn,mk2 < εn, for any m ∈ Ωn and a ∈ Fn. We define em = 0 for m ∈ N \ Ω1 and em = eFn,εn,m for m ∈ Ωn \ Ωn+1. Then it follows that lim n→ω kenk2 = 0, lim n→ω k[(1 − en)bn(1 − en), πn(a)]k = 0, for any a ∈ [n∈N Fn. This means that Φ(((1 − en)bn(1 − en))n) = x, ((1 − en)bn(1 − en))n ∈ B. Lemma 3.2. Suppose that T (B) has a compact extreme boundary with finite topological dimension. Then B has property (SI) relative to π(A) in the following sense: For any sequences (en)n and (fn)n of positive contractions in B satisfying (en)n, (fn)n ∈ B, (en)n ∈ J, lim m→∞ lim n→ω min τ ∈T (B) τ (f m n ) > 0, there exists a sequence (sn)n in B such that (sn)n ∈ B, lim n→ω ks∗ nsn − enk = 0, lim n→ω kfnsn − snk = 0. Proof. One can prove this lemma in a similar fashion to the proof of [20, Theorem 1.1]. We sketch a proof for the reader's convenience. In the proof we write x ≈ε y if kx − yk < ε for x, y in a C∗-algebra. Let F be a finite subset of contractions in A and ε > 0. Since A is nuclear there are two completely positive contractions ρ : A → MN and σ : MN → A such that σ ◦ ρ(x) ≈ε/2 x for x ∈ F . Since A is unital simple infinite-dimensional, applying Voiculescu's theorem to ρ we can obtain di ∈ A, i = 1, 2, . . . , N such that ρ(x) ≈ε/2 [ϕ(d∗ i xdj)]i,j for some pure state ϕ of A and for all x ∈ F . Identifying σ with a positive element of MN (A) we have cl,i ∈ A, l, i = 1, 2, . . . , N such that σ(ei,j) = {ei,j}i,j of MN . Thus it follows that NXl=1 c∗ l,icl,j for the standard matrix units NXl=1 NXi,j=1 ϕ(d∗ i xdj)π(c∗ l,icl,j) ≈ε π(x) for x ∈ F. Due to the Akemann-Anderson-Pedersen excision theorem, there exists a positive con- i xdj)a2 for x ∈ F and ϕ(a) = 1. Because traction a ∈ A such that ad∗ T (B) has compact extreme boundary with finite topological dimension, by [29, Theorem 1.1] (see also [12, 31]), we obtain a unital embedding of MN into M ∩ B′. By Theorem i xdja ≈ε/N 2 ϕ(d∗ 5 3.1 we obtain a sequence of completely positive order zero maps ψn : MN → B such that minτ ∈T (B) τ (ψn(1)m) → 1 for any m ∈ N and k[ψn(x), b]k → 0 as n → ω for any x ∈ MN and b ∈ B. Since A is separable, we may further assume lim n→ω k[ψn(x), πn(y)]k = 0, lim n→ω k[ψn(x), fn]k = 0 for any x ∈ MN and y ∈ A. Then a similar argument as in the proof of [20, Lemma 3.4] implies that there exist sequences (fl,n)n, l = 1, 2, . . . , N of positive contractions in B such that (fl,n)n ∈ B, (fl,nfn)n = (fl,n)n, (fl,nfl′,n)n = 0 for l 6= l′ in Bω, and lim n→ω min τ ∈T (B) τ (f m l,n) = 1 N lim n→ω min τ ∈T (B) τ (f m n ) for any m ∈ N. Since A is unital and simple, there exist vi ∈ A, i = 1, 2, . . . , k such that Set α =(cid:16)Pk i=1 kvik2(cid:17)−1 > 0. By (fl,n)n ∈ B we see that for any m ∈ N lim n→ω min τ ∈T (B) τ (f m/2 l,n πn(a)2f m/2 l,n ) ≥ α lim n→ω min τ ∈T (B) τ (f m l,n) > 0. v∗ i a2vi = 1. kXi=1 As lim m→∞ lim n→ω min τ ∈T (B) τ (f m l,n) > 0, kak = 1 and B has strict comparison, there exist sequences rl,n ∈ B, l = 1, 2, . . . , N , n ∈ N such that l,nf 1/2 l,n πn(a)2f 1/2 lim n→ω krl,nk = lim n→ω kenk1/2 ≤ 1. lim n→ω(cid:13)(cid:13)(cid:13)r∗ Define l,n rl,n − en(cid:13)(cid:13)(cid:13) = 0, NXi=1 πn(dia)f 1/2 sn = NXl=1 l,n rl,nπn(cl,i) ∈ B, for n ∈ N. Then we have (fnsn)n = (sn)n in Bω. And there exists a neighborhood Ω ∈ ω such that for x ∈ F and n ∈ Ω s∗ nπn(x)sn ≈ πn(c∗ l,i)r∗ l,nf 1/2 l,n πn(ad∗ i xdja)f 1/2 l,n rl,nπn(cl,j) ϕ(d∗ i xdj)πn(c∗ l,i)r∗ l,nf 1/2 l,n πn(a2)f 1/2 l,n rl,nπn(cl,j) ϕ(d∗ i xdj)πn(c∗ l,icl,j)en ≈ε πn(x)en. NXi,j=1 NXl=1 ≈εXl Xi,j ≈Xl Xi,j Since F ⊂ A and ε > 0 are arbitrary, A is separable, and π is unital, we can find nsn)n = (en)n in Bω and another sequence sn, n ∈ N of B such that (fnsn)n = (sn)n, (s∗ (sn)n ∈ B. Proposition 3.3. Suppose that B has finitely many extremal tracial states. Then the following holds. (i) The map T (M) → T (B) induced by Φ : B → M is surjective. 6 (ii) B has strict comparison for projections. Proof of (i). It suffices to show τ (x) = 0 for any τ ∈ T (B) and x ∈ B ∩ ker Φ. We may assume that x is a positive contraction. Let (xn)n be a representative sequence of x consisting of positive contractions. We have lim n→ω min τ ∈T (B) τ ((1 − xn)m) = 1 for any m ∈ N. By Lemma 3.2, B has property (SI) relative to π(A). It follows that there exists s1 ∈ B ∩ ker Φ such that s∗ 1 is a positive contraction in B ∩ ker Φ. Hence, in the same way we obtain s2 ∈ B ∩ ker Φ such that s∗ 2s2 = x and (1 − x − s1s∗ 2 is again a positive contraction in B ∩ ker Φ. Repeating this argument, one can find si ∈ B ∩ ker Φ, i ∈ N satisfying 1s1 = x and (1 − x)s1 = s1. Clearly x + s1s∗ 1)s2 = s2. Then x + s1s∗ 1 + s2s∗ s∗ i si = x and x + s1s∗ 1 + s2s∗ 2 + · · · + sms∗ m ≤ 1 for every m ∈ N. Hence τ (x) = 0 for all τ ∈ T (B). Proof of (ii). Let π(n) be the amplification map of π defined by π(n)(a) = π(a) ⊗ 1n ∈ Bω ⊗ Mn ∼= (B ⊗ Mn)ω, n ∈ N, a ∈ A. By the canonical identification we have B ⊗ Mn ∼= (B ⊗ Mn)ω ∩ π(n)(A)′ for any n ∈ N. Then it suffices to consider two projections in B. Assume that two projections p and q in B satisfy τ (p) < τ (q) for any τ ∈ T (B). This implies τ ◦Φ(p) < τ ◦Φ(q) for any τ ∈ T (M). Because M is a finite von Neumann algebra, there exists r ∈ M such that r∗r = Φ(p) and rr∗ ≤ Φ(q). By Theorem 3.1 we obtain r ∈ B such that Φ(r) = r and krk ≤ 1. Set w = qrp ∈ B and let qn, wn ∈ B, n ∈ N are representatives of q and w. Since q is a projection and q ≥ ww∗, we may assume that qn, n ∈ N are projections and qn ≥ wnw∗ n for any n ∈ N. Since τ (q − ww∗) ≥ τ (q − p) > 0 for any τ ∈ T (B) it follows that lim n→ω τ ′(qn − wnw∗ n) > 0 for any τ ′ ∈ T (B). Because Φ(q − ww∗)m = Φ(q − ww∗) for any m ∈ N and B has finitely many extremal traces, we have lim n→ω min τ ′∈T (B) τ ′((qn − wnw∗ n)m) = lim n→ω min τ ′∈T (B) τ ′(qn − wnw∗ n) > 0, for any m ∈ N. By Lemma 3.2 and p − w∗w ∈ B ∩ ker Φ, there exists s ∈ B ∩ ker Φ such that s∗s = p − w∗w and (q − ww∗)s = s. We define v = w + s, then we conclude that v∗v = w∗w + s∗s = p and qv = v. This completes the proof. 4 Decomposition rank We denote by U the universal UHF-algebra. In this section, we show that certain U - absorbing C∗-algebras have finite decomposition rank. First let us recall the definition of decomposition rank. 7 Definition 4.1 (Kirchberg-Winter [13, 33]). A separable C∗-algebra A is said to have decomposition rank at most d ∈ N ∪ {0} if there exist completely positive contractions i=0 Fi,n, n ∈ N and ψi,n : Fi,n → A, i = 0, 1, . . . , d, n ∈ N, where Fi,n are finite-dimensional C∗-algebras, such that ψi,n are order zero (disjointness preserving), ϕn : A → Ld Pd i=0 ψi,n :Ld i=0 Fi,n → A is contractive, and lim ψi,n! ◦ ϕn(a) − a(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) n→∞(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) dXi=0 i=0 xi) =Pd i=0 ψi,n(Ld where we simply writePd = 0 for any a ∈ A, i=0 ψi,n(xi) for xi ∈ Fi,n. The proof of the following theorem is based on the crucial technique which was devel- oped by A. Connes and U. Haagerup in the proof of [8, Theorem 3.1, (d)⇒(a)] and [9, Theorem 4.2]. Theorem 4.2. Let A be a unital separable simple nuclear C∗-algebra with a unique tracial state. If A is quasidiagonal then the decomposition rank of A ⊗ U is at most one. Lemma 4.3. Let A be a unital simple stably finite C∗-algebra which absorbs a UHF- algebra tensorially, A1 a C∗-subalgebra of A, and ω a free ultrafilter on N. Suppose that vn, n ∈ N are contractions in A such that (vn)n ∈ Aω 1 ⊂ Aω. Then there exists a sequence (vn)n of contractions in A such that vn ∈ A1 for any n ∈ N and (vn)n = (vn)n in Aω. Proof. Let εn > 0, n ∈ N be a decreasing sequence which converges to 0, and let fn(t) = max{0, t − εn} for t ∈ [0, ∞), n ∈ N. M. Rørdam showed that any unital simple stably finite UHF-absorbing C∗-algebra has stable rank one [26, Corollary 6.6]. Then, by G. K. Pedersen's polar decomposition theorem [22, Theorem 5], there exists a sequence un, n ∈ N of unitaries in A such that kunfn(vn) − vnk ≤ εn. By (vn)n ∈ Aω that (hn)n = (vn)n = (fn(vn))n in Aω (vn)n = (unfn(vn))n = (vn)n in Aω. 1 there exists a sequence hn, n ∈ N of positive contractions in A1 such 1 . Define vn = unhn. Then vn = hn ∈ A1 and In what follows, we let Ad a(b) = aba∗ for a, b in a C∗-algebra and denote by kak2,τ the 2-norm of a associated with a tracial state τ , i.e. kak2,τ = τ (a∗a)1/2. Proof of Theorem 4.2. Let ι be the canonical unital embedding of A into A ⊗ U . Since A ⊗ Mn also satisfies the assumptions in the theorem for any n ∈ N, it suffices to construct completely positive contractions eρn : A → Mkn ⊕ Mkn, n ∈ N and eσi,n : Mkn → A ⊗ U , i = 0, 1, n ∈ N such that eσi,n are order zero, Pi=0,1eσi,n : Mkn ⊕ Mkn → A ⊗ U is contractive, and (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=0,1eσi,n ◦eρn(a) − ι(a)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) → 0, n → ∞, for any a ∈ A. 8 We let Ui, i = 0, 1 be commuting unital C∗-subalgebras of U such that U0 ∼= U1 ∼= U and C∗(U0 ∪ U1) = U . We denote by τA, τUi, i = 0, 1, and τU the unique tracial states of A, Ui, and U . Let ω be a free ultrafilter on N. Since A is quasidiagonal, there are unital completely positive maps ρn : A → Mkn, n ∈ N such that lim n→∞ kρn(a)ρn(b) − ρn(ab)k = 0 and lim n→∞ kρn(a)k = kak , for any a, b ∈ A (see [5, Lemma 7.1.4] for example). Let σn : Mkn → U0, n ∈ N be a sequence of unital embeddings. We define unital completely positive maps ϕn : A → U0, n ∈ N and ϕ : A → U ω 0 by ϕn(a) = σn ◦ ρn(a) and ϕ(a) = (ϕn(a))n, for any n ∈ N and a ∈ A. Note that ϕ is a unital embedding because ρn are asymptotically multiplicative. In what follows, we identify A⊗U0 with the set of constant sequences in (A⊗U0)ω and U0 with the unital C∗-subalgebra 1A⊗U0 of A⊗U0. By this canonical unital embedding we also 0 with the unital C∗-subalgebra of (A⊗U0)ω. From these identifications it follows identify U ω that [ι(a), ϕ(b)] = 0 for a, b ∈ A. We define the product map ι × ϕ : A ⊙ A → (A ⊗ U0)ω by (ι × ϕ)(a ⊗ b) = ι(a) · ϕ(b) for a, b ∈ A, where ⊙ means the algebraic tensor product. By the nuclearity of A we can extend ι × ϕ to A ⊗ A. Since τA is the unique tracial state of A, A ⊗ A also has the unique tracial state τA ⊗ τA. Then the strong closure of A ⊗ A in the GNS-representation, associated with τA ⊗ τA, is the injective II1-factor, and the flip automorphism on A ⊗ A is approximately inner in the strong topology [8], i.e., there exists a sequence wn, n ∈ N of unitaries in A ⊗ A such that lim n→∞ kwn(a ⊗ b)w∗ n − b ⊗ ak2,τA⊗τA = 0, for any a, b ∈ A. Set Un = (ι × ϕ)(wn) ∈ (A ⊗ U0)ω, n ∈ N. We define the tracial state on (A ⊗ U0)ω by (τA ⊗ τU0)ω((xn)n) = limn→ω(τA ⊗ τU0)(xn) for (xn)n ∈ (A ⊗ U0)ω. Since τA⊗τA is a unique tracial state of A⊗A it follows that (τA⊗τU0)ω ◦(ι×ϕ)(x) = (τA⊗τA)(x) for x ∈ A ⊗ A, which implies that lim n→∞ kUnϕ(a)U ∗ n − ι(a)k2,(τA⊗τU0 )ω = 0, for any a ∈ A. Since A is separable, by the standard reindexation trick we obtain a sequence un, n ∈ N of unitaries in A ⊗ U0 such that lim n→ω kunϕn(a)u∗ n − ι(a)k2,τA⊗τU0 = 0 for any a ∈ A. Set A = (A ⊗ U ⊗ M2)ω, A0 = (A ⊗ U0 ⊗ M2)ω ⊂ A, J0 = {(xn)n ∈ ℓ∞(A ⊗ U0 ⊗ M2) M0 = ℓ∞(A ⊗ U0 ⊗ M2)/J0, 9 lim n→ω kxnk2,τA⊗τU0 ⊗tr2 = 0}, and let Φ0 be the quotient map from A0 to M0. For any a ∈ A, we define π(a) =(cid:18)(cid:20) ϕn(a) 0 0 ι(a) (cid:21)(cid:19)n ∈ A0, u =(cid:18)(cid:20) 0 un 0 (cid:21)(cid:19)n 0 ∈ A0. Note that π is a unital embedding of A into A0. Because of limn→ω kunϕn(a) − ι(a)unk2,τA⊗τU0 = 0 for any a ∈ A, we have Φ0(u) ∈ M0 ∩ Φ0(π(A))′. We also define p =(cid:20) 1A⊗U0 0 0 0 (cid:21) ∈ A0, q =(cid:20) 0 0 1A⊗U0 (cid:21) ∈ A0. 0 Then p and q belong to A0 ∩ π(A)′. Since un are unitaries it follows that As Φ0(u), Φ0(p), and Φ0(q) are in M0 ∩ Φ0(π(A))′ we have u∗u = p and uu∗ = q in A0. τ ◦ Φ0(p) = τ ◦ Φ0(q), for any tracial state τ of M0 ∩Φ0(π(A))′. Applying Proposition 3.3 (i) to B = A⊗U0 ⊗M2, we have τ (p) = τ (q), for any tracial state τ of A0 ∩ π(A)′. By (ii) of Proposition 3.3, A0 ∩ π(A)′ has strict comparison for projections. Then by ≤ ≤ 4 n , 4 n , such that for i = 0, 1. We may think of Mn as a unital subalgebra of U1, and so for each n ∈ N we get Lemma 2.1, for each n ∈ N we obtain two contractions ewi,n, i = 0, 1 in (A0 ∩ π(A)′) ⊗ Mn i,n − q ⊗ 1n(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) i,newi,n − p ⊗ 1n(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=0,1ewi,new∗ Xi=0,1ew∗ dist(cid:0)ew∗ i,newi,n, {p ⊗ x x ∈ Mn}(cid:1) ≤ i,n − q(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) i,newi,n − p(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=0,1ewi,new∗ Xi=0,1ew∗ dist(cid:0)ew∗ i,newi,n, p(1A ⊗ U1 ⊗ 12)ω(cid:1) ≤ Therefore one can find two sequences (evi,n)n, i = 0, 1 of contractions in A ⊗ U ⊗ M2 such that (evi,n)n ∈ A ∩ π(A)′ for i = 0, 1, Xi=0,1ev∗ i,nn Xi=0,1evi,nev∗ i,nevi,nn for i = 0, 1. 4 n , = q in A, ≤ 4 n , 2 n 2 n and and ≤ = p, 10 Hence there exist two sequences (vi,n)n, i = 0, 1 of contractions in A ⊗ U such that 0 and and (evi,n)n ∈ p(1A ⊗ U1 ⊗ 12)ω ⊂ pAp, = (evi,n)n ∈ A, i,nn (cid:18)(cid:20) 0 vi,n 0 (cid:21)(cid:19)n i,nvi,nn =Xi=0,1 Because of (evi,n)n ∈ A ∩ π(A)′ we have Xi=0,1 (vi,nϕn(a))n = (ι(a)vi,n)n vi,nv∗ v∗ in (A ⊗ U )ω, for i = 0, 1. i = 0, 1, = 1A⊗U in (A ⊗ U )ω. for any a ∈ A. It follows from these conditions that for any a ∈ A (vi,nϕn(a)v∗ i,n)n ι(a) = Xi=0,1 = Xi=0,1 (Ad vi,n ◦ σn ◦ ρn(a))n in (A ⊗ U )ω. Since ι and ρn are unital it follows that Lemma 4.3 we may assume that vi,n is in 1A ⊗ U1 for every n ∈ N. For i = 0, 1 and From (evi,n)n ∈ p(1A ⊗ U1 ⊗ 12)ω we have (vi,n)n ∈ (1A ⊗ U1)ω ⊂ (A ⊗ U )ω. By n ∈ N, we define an order zero completely positive contractioneσi,n : Mkn → A ⊗ U , n ∈ N byeσi,n = Ad vi,n ◦ σn and simply writePi=0,1eσi,n(x0 ⊕ x1) =Pi=0,1eσ(xi) for xi ∈ Mkn. n→ω(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=0,1eσi,n(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) By multiplying eσi,n by suitable scalars, we may further assume that Pi=0,1eσi,n are con- tractive. Define completely positive contractions eρn, n ∈ N from A to Mkn ⊕ Mkn by eρn(a) = ρn(a) ⊕ ρn(a). Then we conclude that ι(a) =Xi=0,1eσi,n ◦eρn(a)n in (A ⊗ U )ω, = 1. lim for any a ∈ A. This completes the proof. 5 Z-absorbing C∗-algebras In this section, we prove Theorem 1.1. Proof of Theorem 1.1. By Theorem [20, Theorem 1.1], A is Z-absorbing, i.e. A ∼= A ⊗ Z. Let F be a finite subset of contractions in A and let ε > 0. By Theorem 4.2, decomposition rank of A ⊗ U is at most one, and so there exist completely positive contractions ϕ : A → 11 E0 ⊕ E1 and ψi : Ei → A ⊗ U , i = 0, 1 such that Ei are finite dimensional C∗-algebras, ψi are order zero,Pi=0,1 ψi : E0 ⊕ E1 → A ⊗ U is contractive and ψi ◦ ϕ(a) − a ⊗ 1U(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=0,1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) < ε/2 for any a ∈ F . Any order zero completely positive map from a finite dimensional C∗- algebra can be identified with a ∗-homomorphism from the cone over the algebra (see [33, Proposition 1.2] for instance). A cone over a finite dimensional C∗-algebra is projective [18]. Therefore we may assume that ψi, i = 0, 1 satisfy ψi(Ei) ⊂ A ⊗ Ml for some unital matrix subalgebra Ml of U . Choose m ∈ N so that (m−1)−1 < ε/2. Let e1, e2, . . . , elm ∈ Mlm and f0, f1, . . . , flm ∈ Mlm+1 be mutually orthogonal minimal projections. For j = 1, 2, . . . , m, we put Ij = {n ∈ N (j − 1)l + 1 ≤ n ≤ jl}. Let αj : Ml → Mlm, βj : Ml → Mlm+1 and γj : Ml → Mlm+1 be embeddings such that αj(1l) = Xn∈Ij en, βj(1l) = Xn∈Ij fn, γj(1l) = Xn∈Ij fn−1. Take a unitary w ∈ Ml ⊗ Ml such that w(x ⊗ y)w∗ = y ⊗ x for any x, y ∈ Ml. Let u : [0, 1] → Ml ⊗ Ml be a continuous path of unitaries such that u(0) = 1l and u(1) = w. Define a continuous path of unitaries u : [0, 1] → Mlm ⊗ Mlm+1 by u(t) = 1lm ⊗ f0 + (αj ⊗ βk)(u(t)). mXj,k=1 For i = 0, 1, we define order zero completely positive maps λi : Ml → C([0, 1]) ⊗ Mlm ⊗ Mlm+1 by λ0(x)(t) = u(t) and mXj=1 (1 − t)αj(x) ⊗ f0 + mXj,k=1(cid:18)1 − (m−k)t m−1 (cid:19) αj(x) ⊗ βk(1l) u(t)∗ λ1(x)(t) = It is straightforward to check (m−k)t m−1 mXk=1 1lm ⊗ γk(x). 1 − t m−1 ≤ λ0(1l)(t) + λ1(1l)(t) ≤ 1 for any t ∈ [0, 1]. This means that λ0 + λ1 is contractive. Moreover one has λ0(x)(0) = mXj=1 αj(x) ⊗ 1lm+1, λ0(x)(1) = mXk=1(cid:18)1 − m−k m−1(cid:19) 1lm ⊗ βk(x) 12 and λ1(x)(0) = 0, λ1(x)(1) = m−k m−1 1lm ⊗ γk(x). mXk=1 Hence we may regard λi as a map from Ml to the dimension drop algebra Z = {g ∈ C([0, 1]) ⊗ Mlm ⊗ Mlm+1 g(0) ∈ Mlm ⊗ C, g(1) ∈ C ⊗ Mlm+1}. Then (idA ⊗λi) ◦ ψj is an order zero map from Ej to A ⊗ Z for each i, j = 0, 1. We also obtain (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (idA ⊗λi) ◦ ψj ◦ (ϕ ⊕ ϕ)(a) − a ⊗ 1Z(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi,j idA ⊗λi(a ⊗ 1l) − a ⊗ 1Z(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) <(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=0,1 < ε ε 2 ≤ + 1 m−1 + ε 2 for every a ∈ F . Since there exists a unital embedding Z → Z, we can conclude that A ∼= A ⊗ Z has decomposition rank at most three. 6 C∗-algebras with tracial rank zero In [15, 16], H. Lin introduced the notion of tracial rank, which is based on ideas of S. Popa [24]. W. Winter [32, 33] proved that if a unital separable simple C∗-algebra A has (locally) finite decomposition rank and has real rank zero, then A has tracial rank zero. In this section, using the same strategy as in the proof of Theorem 4.2 we give a direct proof that A has tracial rank zero under the assumption that the C∗-algebra absorbs a UHF-algebra tensorially and has a unique tracial state. Theorem 6.1. Let A be a unital separable simple nuclear quasidiagonal C∗-algebra with a unique tracial state. Then the tracial rank of A ⊗ B is zero for any UHF-algebra B. Proof. Because of [19, Lemma 2.4] it is enough to show that A ⊗ U has tracial rank zero. We use the same notations as in the proof of Theorem 4.2. Let ε > 0. Since A ⊗ Mn also satisfies the same assumption in the theorem for any n ∈ N, it suffices to construct a sequence of finite dimensional C∗-subalgebras Bn, n ∈ N of A ⊗ U such that and τA⊗U (1A⊗U − 1Bn) < ε, lim n→ω k[1Bn, ι(a)]k = 0, lim n→ω dist(ι(a)1Bn , Bn) = 0, for any a ∈ A. Recall that ϕn : A → U0 is the composition of the unital completely positive map ρn : A → Mkn and the unital embedding σn : Mkn → U0. Clearly we have [ϕn(a), 1A ⊗ x] = 0 for any n ∈ N, a ∈ A, and x ∈ U1. For any τ ∈ T (A ∩ π(A)′) we have seen that 13 τ (p) = τ (q) = 1/2 in the proof of Theorem 4.2 . Thus for any τ ∈ T (A ∩ π(A)′), the functional gives the unique tracial state τU1 of U1. Let e0 be a projection in U1 such that 1 − ε < τU1(e0) < 1. Set 0 U1 ∋ x 7→ 2τ(cid:18)(cid:18)(cid:20)1A ⊗ x 0 e =(cid:18)(cid:20) 1A ⊗ e0 0 0 (cid:21)(cid:19)n 0 0(cid:21)(cid:19)n(cid:19) ∈ C ∈ A ∩ π(A)′. Then we have τ (e) = 1 2 τU1(e0) < 1 2 = τ (p) = τ (q) for any τ ∈ T (A ∩ π(A)′). Applying (ii) of Proposition 3.3 to B = A ⊗ U ⊗ M2, we see that A ∩ π(A)′ has strict comparison for projections. We obtain a partial isometryev ∈ A ∩ π(A)′ such that and By the definition of p and q there exists a sequence vn, n ∈ N of contractions in A ⊗ U such that ev∗ev = e qev =ev. (cid:18)(cid:20) 0 vn 0 (cid:21)(cid:19)n 0 =ev in A, and (v∗ nvn)n = (1A ⊗ e0)n in (A ⊗ U )ω. By perturbing vn slightly, we may further assume v∗ we have nvn = 1A ⊗ e0. Because of ev ∈ π(A)′ in (A ⊗ U )ω for any a ∈ A. (vnϕn(a))n = (ι(a)vn)n Then it follows that [(vnv∗ dimensional C∗-subalgebras Bn ⊂ A ⊗ U for n ∈ N by n)n , ι(a)] = 0 in (A ⊗ U )ω for any a ∈ A. We define finite One can verify that these Bn satisfy the desired properties. Bn = Ad vn (σn(Mkn)) . The following corollary is a direct consequence of [17, Theorem 5.4], [20, Theorem 1.1] and the theorem above. Corollary 6.2. Let A be a unital separable simple nuclear quasidiagonal C∗-algebra with a unique tracial state. Suppose that A has strict comparison and satisfies the UCT. Then A is classifiable by the Elliott invariants and is isomorphic to a unital simple approximately subhomogeneous algebra. In the rest of this section, we would like to discuss an application of the corollary above. R. T. Powers and S. Sakai [25] conjectured that every strongly continuous action α : R y A on a UHF algebra A is approximately inner. An action α : R y A is said to be strongly continuous if the map t 7→ αt(a) is continuous for every a ∈ A. An action α : R y A is said to be approximately inner if there exists a sequence (hn)n of self-adjoint elements in A such that keithn ae−ithn − αt(a)k → 0 max t≤1 14 as n → ∞ for all a ∈ A. A. Kishimoto [14] gave a counter-example to the AF version of the Powers-Sakai conjecture. Namely, he constructed a unital simple AF algebra A and a strongly continuous action α : R y A which is not approximately inner. Here we shall present a counter-example to the Powers-Sakai conjecture by using our main result. In what follows, an action of T = R/Z is identified with a periodic action of R. Theorem 6.3. Let A be a unital simple infinite-dimensional AF algebra with a unique tracial state τ . There exists a strongly continuous action α : T y A such that the crossed product A ⋊α T is simple. In particular, α is not approximately inner as an R-action. Proof. Let G =LZ K0(A) be the infinite direct sum of K0(A) over Z. Define G+ ⊂ G by G+ =((xn)n ∈ G Xn∈Z τ∗(xn) > 0) ∪ {0}. Since τ∗(K0(A)) is dence in R, (G, G+) is a simple dimension group. Let B be a simple stable AF algebra whose K0-group is isomorphic to (G, G+). The AF algebra B admits a densely defined lower semi-continuous trace, unique up to constant multiples. Choose a projection e ∈ B whose K0-class is equal to (. . . , 0, 0, [1A], 0, 0, . . . ) ∈ G, where [1A] is in the 0-th summand. There exists an automorphism σ ∈ Aut(B) such that K0(σ) is the shift on G, that is, K0(σ)((xn)n) = (xn+1)n. The trace on B is invariant under σ. Let γ ∈ Aut(Z) be an automorphism of Z which has the weak Rohlin property ([30, Definition 1.1]). We consider the crossed product C = (B ⊗ Z ⊗ Z) ⋊σ⊗γ⊗id Z = ((B ⊗ Z) ⋊σ⊗γ Z) ⊗ Z. Clearly C is Z-stable. Since the trace on B ⊗ Z ⊗ Z is preserved by the automorphism, C admits a densely defined lower semi-continuous trace. Moreover, this trace is unique up to scalar multiples, because γ has the weak Rohlin property (one can prove this in the same way as [19, Remark 2.8]). By [4], C is AF embeddable, and hence is quasidiagonal. By the Pimsner-Voiculescu exact sequence, we have K0(C) ∼= K0(A) and K1(C) = 0. Set D = (e ⊗ 1 ⊗ 1)C(e ⊗ 1 ⊗ 1). Then D is a unital separable simple nuclear C ∗-algebra, which is Z-stable, has a unique tracial state and is quasidiagonal. Besides, (K0(D), K0(D)+, [1D], K1(D)) ∼= (K0(A), K0(A)+, [1A], K1(A)). It follows from the corol- lary above that D is isomorphic to A. Let α : T y C be the dual action of σ ⊗ γ ⊗ id. The projection e ⊗ 1 ⊗ 1 is invariant under α, and so we can restrict α to D. The crossed product D ⋊α T is stably isomorphic to C ⋊α T, which is isomorphic to B ⊗Z ⊗Z ⊗K by the duality theorem. Therefore D ⋊α T is simple. In particular, α on D is not approximately inner by [25, Theorem 2.3] and [3, Remark 3.5] 7 Kirchberg algebras A separable simple nuclear purely infinite C∗-algebra is called a Kirchberg algebra. In this section we prove the following theorem. See [34, Definition 2.1] for the definition of nuclear dimension. Theorem 7.1. Any Kirchberg algebra has nuclear dimension at most three. 15 Proof. By [34, Corollary 2.8], it suffices to show that a unital Kirchberg algebra A has nuclear dimension at most three. Because of the Kirchberg-Phillips theorem, A is isomor- phic to A ⊗ O∞. We let Bi, i = 0, 1 be commuting unital C∗-subalgebras of O∞ such that B0 ∼= B1 ∼= O∞ and C∗(B0 ∪ B1) = O∞. Let s ∈ B0 be a non-unitary isometry and put q = ss∗, p = 1B0 − q. Define ι : A → A ⊗ B0 by ι(a) = a ⊗ q. The Kirchberg-Phillips theorem also tells us that there exists a unital embedding ρ : A → O2. Since pB0p contains a unital copy of O2, there exists a unital embedding σ : O2 → 1A ⊗ pB0p ⊂ A ⊗ B0. Let ϕ = σ ◦ ρ. We would like to apply the same argument as the proof of Theorem 4.2 to ϕ and ι. Set A = (A ⊗ O∞)ω, A0 = (A ⊗ B0)ω ⊂ A. Define a unital homomorphism π : A → A0 by π(a) = ϕ(a) + ι(a) for a ∈ A. By [11, Corollary 4], A0 ∩ π(A)′ is purely infinite simple. In particular, for any non-zero projections e, f ∈ (A0 ∩ π(A)′) ⊗ Mk, e is Murray-von Neumann equivalent to a subprojection of f . Hence, for any n ∈ N, we can apply (the proof of) Lemma 2.1 and obtain contractions wi,n ∈ (A0 ∩ π(A)′) ⊗ (Mn ⊕ Mn+1), i = 0, 1, such that wi,nw∗ Xi=0,1 i,n − 1A ⊗ q ⊗ (1n ⊕ 1n+1)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ≤ 4 n , i,nwi,n, {1A ⊗ p ⊗ x x ∈ Mn ⊕ Mn+1}) ≤ 2 n . Since Mn ⊕ Mn+1 can be unitally embedded in B1 ∼= O∞ and n ∈ N is arbitrary, there i = 1A ⊗ q, and w∗ ≤ 4 n , and dist(w∗ Xi=0,1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) i,nwi,n − 1A ⊗ p ⊗ (1n ⊕ 1n+1)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) exist vi ∈ A ∩ π(A)′, i = 0, 1 such that Pi=0,1 v∗ = (1A ⊗ s∗)Xi=0,1 = Xi=0,1 v∗ i vi ∈ (1A ⊗ pB1)ω, i = 0, 1. a ⊗ 1O∞ = (1A ⊗ s∗)ι(a)(1A ⊗ s) In the same way as the proof of Theorem 4.2, one has i vi = 1A ⊗ p, Pi=1,0 viv∗ i (1A ⊗ s) viϕ(a)v∗ Ad(1A ⊗ s∗) ◦ Ad vi ◦ σ ◦ ρ(a) in A, for every a ∈ A. For i = 0, 1, we define an order zero completely positive contraction σi : O2 → A by σi = Ad(1A ⊗ s∗) ◦ Ad vi ◦ σ. Let F ⊂ A be a finite subset and let ε > 0. By [34, Theorem 7.4], the nuclear dimension of O2 is one. It follows that there exist completely positive contractions α : O2 → E0 ⊕ E1 16 and βj : Ej → O2, j = 0, 1, such that Ej are finite dimensional C∗-algebras, βj are order zero and for any a ∈ F . Then σi ◦ βj is an order zero map from Ej to A for each i, j = 0, 1. Define a completely positive contraction ρ : A → E0 ⊕ E1 ⊕ E0 ⊕ E1 by ρ(a) = α(ρ(a)) ⊕ α(ρ(a)). We have < ε (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) βj ◦ α(ρ(a)) − ρ(a)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xj=0,1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) σi ◦ βj ◦ ρ(a) − a ⊗ 1O∞(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)  Xi,j=0,1 < ε for any a ∈ F . Since the cone over Ej is projective, σi ◦ βj lifts to an order zero map to ℓ∞(N, A ⊗ O∞). The proof is completed. References [1] C. A. Akemann, J. Anderson, and G. K. Pedersen, Excising states of C∗-algebras, Canad. J. Math. 38 (1986), 1239 -- 1260. [2] B. Blackadar, Comparison theory for simple C∗-algebras, Operator algebras and applications, London Math. Soc. Lecture Note Ser., 135, Cambridge Univ. Press, Cambridge, (1988), 21 -- 54. [3] O. Bratteli, G. Elliott and R. Herman, On the possible temperatures of a dynamical system, Comm. Math. Phys. 74 (1980), 281 -- 295. [4] N. Brown, AF embeddability of crossed products of AF algebras by the integers, J. Funct. Anal. 160 (1998), 150 -- 175. [5] N. P. Brown and N. Ozawa, C∗-algebras and finite-dimensional approximations, Graduate Studies in Mathematics, 88. Amer. Math. Soc., Providence, RI, 2008. [6] A. Connes, Periodic automorphisms of the hyperfinite factor of type II1, Acta Sci. Math. 39 (1977), no. 1-2, 39 -- 66. [7] A. Connes, Outer conjugacy classes of automorphisms of factors, Ann. Sci. ´Ecole Norm. Sup. (4) 8 (1975), no. 3, 383 -- 419. [8] A. Connes, Classification of injective factors: cases II1, II∞, IIIλ, λ 6= 1, Ann. of Math. 104 (1976), 73 -- 115. [9] U. Haagerup, A new proof of the equivalence of injectivity and hyperfiniteness for factors on a separable Hilbert space, J. Funct. Anal. 62 (1985), no. 2, 160 -- 201. [10] X. Jiang and H. Su, On a simple unital projectionless C∗-algebra, Amer. J. Math. 121 (1999), 359 -- 413. [11] E. Kirchberg, The classification of purely infinite C ∗-algebras using Kasparov's theory, preprint, 1994. 17 [12] E. Kirchberg and M. Rørdam, Central sequence C∗-algebras and tensorial absorption of the Jiang-Su algebra, To appear in J. Reine Angew. Math. [13] E. Kirchberg and W. Winter, Covering dimension and quasidiagonality, Internat. J. Math. 15 (2004), no. 1, 63 -- 85. [14] A. Kishimoto, Non-commutative shifts and crossed products, J. Funct. Anal. 200 (2003), 281 -- 300. [15] H. Lin, The tracial topological rank of C∗-algebras, Proc. London Math. Soc. (3) 83 (2001), no. 1, 199 -- 234. [16] H. Lin, Tracially AF C∗-algebras, Trans. Amer. Math. Soc. 353 (2001), no. 2, 693 -- 722. [17] H. Lin and Z. Niu, Lifting KK-elements, asymptotical unitary equivalence and clas- sification of simple C∗-algebras, Adv. Math. 219 (2008), 1729 -- 1769. [18] T. Loring, Lifting solutions to perturbing problems in C∗-algebras, Fields Institute monographs 8, AMS, Providence, Rhode Island (1997). [19] H. Matui and Y. Sato, Z-stability of crossed products by strongly outer actions, Comm. Math. Phys. 314 (2012), no. 1, 193 -- 228. [20] H. Matui and Y. Sato, Strict comparison and Z-absorption of nuclear C∗-algebras, Acta Math. 209 (2012), no. 1, 179 -- 196. [21] H. Matui and Y. Sato, Z-stability of crossed products by strongly outer actions II, preprint, arXiv:1205.1590. [22] G. K. Pedersen, Unitary extensions and polar decompositions in a C∗-algebra, J. Operator Theory 17 (1987), no. 2, 357-364. [23] S. Popa, A short proof of "injectivity implies hyperfiniteness" for finite von Neumann algebras, J. Operator Theory 16 (1986), no. 2, 261 -- 272. [24] S. Popa, On local finite-dimensional approximation of C∗-algebras, Pacific J. Math. 181 (1997), no. 1, 141 -- 158. [25] R. T. Powers and S. Sakai, Existence of ground states and KMS states for approxi- mately inner dynamics, Comm. Math. Phys. 39 (1975), 273 -- 288. [26] M. Rørdam, On the structure of simple C∗-algebras tensored with a UHF-algebra, J. Funct. Anal. 100 (1991), no. 1, 1-17. [27] M. Rørdam, The stable and the real rank of Z-absorbing C∗-algebras, Internat. J. Math. 15 (2004), no. 10, 1065 -- 1084. arXiv:math/0408020. [28] M. Rørdam and W. Winter, The Jiang-Su algebra revisited, J. Reine. Angew. Math. 642 (2010), 129 -- 155. arXiv:0801.2259. [29] Y. Sato, Trace spaces of simple nuclear C∗-algebras with finite-dimensional extreme boundary, preprint, arXiv:1209.3000. 18 [30] Y. Sato, The Rohlin property for automorphisms of the Jiang-Su algebra, J. Funct. Anal. 259 (2010), 453 -- 476. [31] A. S. Toms, S. White, W. Winter, Z-stability and finite dimensional tracial bound- aries , preprint, arXiv:1209.3292. [32] W. Winter, Simple C∗-algebras with locally finite decomposition rank, J. Funct. Anal. 243 (2007), no. 2, 394 -- 425. [33] W. Winter, Decomposition rank and Z-stability Invent. Math. 179 (2010), no. 2, 229- 301. [34] W. Winter and J. Zacharias, The nuclear dimension of C∗-algebras, Adv. Math. 224 (2010), no. 2, 461 -- 498, arXiv:0903.4914. 19
0704.1533	3	0704	2012-04-16T02:24:58	Concrete Classification and Centralizers of Certain $\mathbb{Z}^2 \rtimes {\rm SL}(2,\mathbb{Z})$-actions	[ "math.OA" ]	We introduce a new class of actions of the group $\G$ on finite von Neumann algebras and call them twisted Bernoulli shift actions. We classify these actions up to conjugacy and give an explicit description of their centralizers. We also distinguish many of those actions on the AFD $\mathrm{II}_1$ factor in view of outer conjugacy.	math.OA	math	CONCRETE CLASSIFICATION AND CENTRALIZERS OF CERTAIN Z2 ⋊ SL(2, Z)-ACTIONS HIROKI SAKO Abstract. We introduce a new class of actions of the group Z2 ⋊ SL(2, Z) on finite von Neumann algebras and call them twisted Bernoulli shift actions. We classify these actions up to conjugacy and give an explicit description of their centralizers. We also distinguish many of those actions on the AFD II1 factor in view of outer conjugacy. 1. Introduction We consider the classification of Z2 ⋊ SL(2, Z)-actions on finite von Neumann algebras in this paper. Mainly, we concentrate on the case that the finite von Neumann algebra is the AFD factor of type II1 or non-atomic abelian. There are two difficulties for analyzing discrete group actions on operator algebras. The first is that we do not have various ways to construct actions. The second is that we can not analyze them by concrete calculation in most cases. To give many examples of actions which admit concrete analysis, we introduce a class of trace preserving Z2 ⋊ SL(2, Z)-actions on finite von Neumann algebras and call them twisted Bernoulli shift actions. We classify those actions up to conjugacy and study them up to outer conjugacy. An action β(H, µ, χ) in the class is defined for a triplet (H, µ, χ), where H is an abelian countable discrete group, µ is a normalized scalar 2-cocycle of H and χ is a character of H. We obtain the action by restricting the so-called generalized Bernoulli shift action to a subalgebra N(H, µ) and "twisting" it by the character χ. The process of restriction has a vital role in concrete analysis of these actions. A ∗-isomorphism which gives conjugacy between two twisted Bernoulli shift ac- tions β(Ha, µa, χa) and β(Hb, µb, χb) must be induced from an isomorphism between the two abelian groups Ha and Hb. We prove this by concrete calculation (Section 4). It turns out that there exist continuously many, non-conjugate Z2 ⋊ SL(2, Z)- actions on the AFD factor of type II1 (Section 5). By using the same technique, we describe the centralizers of all twisted Bernoulli shift actions. Here we should men- tion that the present work was motivated by the previous ones [Ch], [NPS], where similar studies were carried out in the case of SL(n, Z). In Section 6, we distinguish many twisted Bernoulli shift actions in view of outer conjugacy. The classification for actions of discrete amenable groups on the AFD factor of type II1 was given by Ocneanu [Oc]. Outer actions of countable amenable groups are outer conjugate. In the contrast to this, V. F. R. Jones [Jon] proved that any discrete non-amenable group has at least two non outer conjugate actions on 2000 Mathematics Subject Classification. Primary 46L40; Secondary 46L10. Key words and phrases. von Neumann algebras; automorphisms. 1 2 HIROKI SAKO the AFD factor of type II1. S. Popa ([Po3], [Po4], [PoSa], etc.) used the malleabil- ity/deformation arguments for the Bernoulli shift actions to study (weak) 1-cocycles for the actions. For some of twisted Bernoulli shift actions, which we introduce in this paper, it is shown that (weak) 1-cocycles are represented in simple forms under some assumption on the (weak) 1-cocycles. We prove that there exist continuously many twisted Bernoulli shift actions which are mutually non outer conjugate. This strengthens the above mentioned result due to Jones in the Z2 ⋊ SL(2, Z) cases. 2. Preparations 2.1. Functions det and gcd. For the definition of twisted Bernoulli shift actions in Section 3, we define two Z-valued functions det and gcd. The function det is given by the following equation: det(cid:18)(cid:18) q r (cid:19) ,(cid:18) q0 r0 (cid:19)(cid:19) = qr0 − rq0, (cid:18) q r (cid:19) ,(cid:18) q0 r0 (cid:19) ∈ Z2. The value of the function gcd at k ∈ Z2 is the greatest common divisor of the two entries. For 0 ∈ Z2, let the value of gcd be 0. Lemma 2.1. (1) The action of SL(2, Z) on Z2 preserves the functions det and gcd, that is, det(k, k0) = det(γ · k, γ · k0), gcd(k) = gcd(γ · k), (2) The following equation holds true: k, k0 ∈ Z2, γ ∈ SL(2, Z). det(k, k0) = gcd(k) + gcd(k0) − gcd(k + k0) mod 2, k, k0 ∈ Z2. Proof. The claim (1) is a well-known fact, so we prove the claim (2). For the function gcd, we get gcd(cid:18)(cid:18) q r (cid:19)(cid:19) =(cid:26) 1 mod 2, 0 mod 2, (either q or r is odd), (both q and r are even). Since the action of SL(2, Z) on Z2 preserves the functions det and gcd, it suffices to show the desired equation against the following four pairs: (k, k0) = (cid:18)(cid:18) 0 (cid:18)(cid:18) 1 0 (cid:19) ,(cid:18) 0 0 (cid:19) ,(cid:18) 1 0 (cid:19)(cid:19) ,(cid:18)(cid:18) 1 0 (cid:19)(cid:19) ,(cid:18)(cid:18) 1 0 (cid:19) ,(cid:18) 0 0 (cid:19) ,(cid:18) 0 0 (cid:19)(cid:19) , 1 (cid:19)(cid:19) mod 2. (cid:3) 2.2. Scalar 2-cocycles for abelian groups. We fix some notations for countable abelian groups and their scalar 2-cocycles. For the rest of this paper, let H be an abelian countable discrete group and suppose that any scalar 2-cocycle µ : H × H → T = {z ∈ C z = 1} is normalized, that is, µ(g, 0) = 1 = µ(0, g) for g ∈ H. We denote by µ∗ the 2-cocycle for H given by µ∗(g, h) = µ(h, g), g, h ∈ H. Let µ∗µ be the function on H × H defined by µ∗µ(g, h) = µ(h, g) µ(g, h), g, h ∈ H. CLASSIFICATION AND CENTRALIZERS 3 This is a bi-character, that is, µ∗µ(g,·) and µ∗µ(·, h) are characters of H. By using this function, we can describe the cohomology class of µ. See [OPT] for the proof of the following Proposition: Proposition 2.2. Two scalar 2-cocycles µ1 and µ2 of H are cohomologous if and only if µ∗ 1µ1 = µ∗ 2µ2. Let Cµ(H) be the twisted group algebra of H with respect to the 2-cocycle µ. We denote by {uh h ∈ H} the standard basis for Cµ(H) as C-linear space. We recall that the C-algebra Cµ(H) has a structure of ∗-algebra defined by g, h ∈ H. u∗ g = µ(g,−g) u−g, ug uh = µ(g, h) ug+h, (Eq1) µ(λ1(k), λ2(k)), group defined by λ1, λ2 ∈ ⊕Z2H. Leteµ be the T-valued function on ⊕Z2H × ⊕Z2H defined by eµ(λ1, λ2) = Yk∈Z2 The functioneµ is a normalized scalar 2-cocycle for ⊕Z2H. Let Λ(H) be the abelian Λ(H) =(λ : Z2 → H (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) finitely supported and Xk∈Z2 Its additive rule is defined by pointwise addition. λ(k) = 0) . 2.3. Definition of a Z2 ⋊ SL(2, Z)-action on Λ(H). The group SL(2, Z) acts on Z2 as matrix-multiplication and the group Z2 also does on Z2 by addition. These two actions define the action of Z2 ⋊ SL(2, Z) on Z2 which is explicitly described as for all (cid:18)(cid:18) q r (cid:19) ,(cid:18) x y (cid:18)(cid:18) q r (cid:19) ,(cid:18) x y r0 (cid:19) =(cid:18) q + xq0 + yr0 r + zq0 + wr0 (cid:19) , r0 (cid:19) ∈ Z2. z w (cid:19)(cid:19) · (cid:18) q0 z w (cid:19)(cid:19) ∈ Z2 ⋊ SL(2, Z), (cid:18) q0 k ∈ Z2, (γ · λ)(k) = λ(γ−1 · k), We define an action of Z2 ⋊ SL(2, Z) on ⊕Z2H as for γ ∈ Z2 ⋊ SL(2, Z) and λ ∈ ⊕Z2H. 2.4. On the relative property (T) of Kazhdan. We give the definition of the relative property (T) of Kazhdan for a pair of discrete groups. Definition 2.3. Let G ⊂ Γ be an inclusion of discrete groups. We say that the pair (Γ, G) has the relative property (T) if the following condition holds: There exist a finite subset F of Γ and δ > 0 such that if π : Γ → U(H) is a unitary representation of Γ on a Hilbert space H with a unit vector ξ ∈ H satisfying kπ(g)ξ − ξk < δ for g ∈ F , then there exists a non-zero vector η ∈ H such that π(h)η = η for h ∈ G. Instead of this original definition, we use the following condition. 4 HIROKI SAKO Proposition 2.4. ([Jol]) Let G ⊂ Γ be an inclusion of discrete groups. The pair (Γ, G) has the relative property (T) if and only if the following condition holds: For any ǫ > 0, there exist a finite subset F of Γ and δ > 0 such that if π : Γ → U(H) is a unitary representation of Γ on a Hilbert space H with a unit vector ξ ∈ H satisfying kπ(g)ξ − ξk < δ for g ∈ F , then kπ(h)ξ − ξk < ǫ for h ∈ G. The pair (Z2⋊SL(2, Z), Z2) is a typical example of group with the relative property (T). See [Bu] or [Sh] for the proof. 2.5. Weakly mixing actions. An action of a countable discrete group G on a von Neumann algebra N is said to be ergodic if any G-invariant element of N is a scalar multiple of 1. The weak mixing property is a stronger notion of ergodicity. Definition 2.5. Let N be a von Neumann algebra with a faithful normal state φ. A state preserving action (ρg)g∈G of a countable discrete group G on N is said to be weakly mixing if for every finite subset {a1, a2, . . . , an} ⊂ N and ǫ > 0, there exists g ∈ G such that φ(aiρg(aj)) − φ(ai)φ(aj) < ǫ, i, j = 1, . . . , n. The following is a basic characterization of the weak mixing property. Between two von Neumann algebra N and M, N ⊗ M stands for the tensor product von Neumann algebra. Proposition 2.6 (Proposition D.2 in [Vaes]). Let a countable discrete group G act on a finite von Neumann algebra (N, tr) by trace preserving automorphisms (ρg)g∈G. The following statements are equivalent: (1) The action (ρg) is weakly mixing. (2) The only finite-dimensional invariant subspace of N is C1. (3) For any action (αg) of G on a finite von Neumann algebra (M, τ ), we have (N ⊗ M)ρ⊗α = 1 ⊗ M α, where (N ⊗ M)ρ⊗α and M α are the fixed point subalgebras. 2.6. A remark on group von Neumann algebras. Let Γ be a discrete group and let µ be a scalar 2-cocycle of a countable group Γ. A group Γ acts on the Hilbert space ℓ2Γ by the following two ways; uγ(δg) = µ(γ, g)δγg, ργ(δg) = µ(g, γ−1)δgγ−1, γ, g ∈ Γ. These two representations commute with each other. The von Neumann algebra Lµ(Γ) generated by the image of u is called the group von Neumann algebra of Γ twisted by µ. The normal state h·δe, δei is a trace on Lµ(Γ). The vector δe is separating for Lµ(Γ). For any element a ∈ Lµ(Γ), we define the square summable function a(·) on Γ by aδe = P a(g)δg. The function a(·) is called the Fourier coefficient of a. We write a = Pg∈Γ a(g)ug and call this the Fourier expansion of a. The Fourier expansion of a∗ is given by a∗ =Pg∈Γ µ(g, g−1)a(g−1)ug, since the Fourier coefficient a∗(g) = ha∗δe, δgi is described as hδe, aρg−1δei =*ρ∗ g−1δe,Xg a(g)δg+ = µ(g, g−1)a(g−1). CLASSIFICATION AND CENTRALIZERS 5 Here we used the equation ρ∗ tion. For two elements a, b, the Fourier coefficient of ab is given by g−1 = µ(g, g−1)ρg, which is verified by direct computa- µ(g−1, gγ)a(g−1)b(gγ). This equation allows us to calculate the Fourier coefficient algebraically, that is, ab(γ) = hbδe, ργ−1a∗δei =Xg ab =Xγ Xg a∗(g)µ(g, γ)b(gγ) =Xg µ(g−1, gγ)a(g−1)b(gγ)! uγ =Xγ Xgh=γ a(g)b(h)! uγ. For a subgroup Λ ⊂ Γ, the subalgebra {uλ λ ∈ Λ}′′ ⊂ Lµ(Γ) is isomorphic to Lµ(Λ). We sometimes identify them. An element a ∈ Lµ(Γ) is in the subalgebra Lµ(Λ) if and only if the Fourier expansion a(·) : Γ → C is supported on Λ, since the trace h·δe, δei preserving conditional expectation E from Lµ(Γ) onto Lµ(Λ) is described as E(a) =Pλ∈Λ a(λ)uλ. 3. Definition of twisted Bernoulli shift actions In this section, we introduce twisted Bernoulli shift actions of Z2 ⋊ SL(2, Z) on finite von Neumann algebras. The action is defined for a triplet i = (H, µ, χ), where H 6= {0} is an abelian countable discrete group, µ is a normalized scalar 2-cocycle of H and χ is a character of H. The finite von Neumann algebra, on which the group Z2 ⋊ SL(2, Z) acts, is defined by the pair (H, µ). (c1, k, γ1)(c2, l, γ2) =(cid:0)c1c2χdet(k,γ1·l), k + γ1 · l, γ1γ2(cid:1) We introduce a group structure on the set Γ0 = bH × Z2 × SL(2, Z) as for any c1, c2 ∈ bH, k, l ∈ Z2, γ1, γ2 ∈ SL(2, Z). The associativity is verified by Lemma 2.1. It turns out that the subsets bH = bH × {0} × {e} and G0 = bH × Z2 × {e} are subgroups in Γ0. It is easy to see that G0 is a normal subgroup of Γ0 and that bH is a normal subgroup of G0 and Γ0. We get a normal inclusion of groups G0/bH ⊂ Γ0/bH and this is isomorphic to Z2 ⊂ Z2 ⋊ SL(2, Z). Before stating the definition of the twisted Bernoulli shift action, we define a Γ0- action ρ on the von Neumann algebra Leµ(⊕Z2H). We denote by u(λ) ∈ Leµ(⊕Z2H) the unitary corresponding to λ ∈ ⊕Z2H. We define a faithful normal trace tr of the linear transformations on Ceµ(⊕Z2H) given by, Leµ(⊕Z2H) in the usual way. For c ∈ bH, k ∈ Z2, γ ∈ SL(2, Z), let ρ(c), ρ(k), ρ(γ) be ρ(c)(u(λ)) = Yl∈Z2 ρ(k)(u(λ)) = Ym∈Z2 c(λ(l))! u(λ), χ(λ(m))det(k,m)! u(k · λ), ρ(γ)(u(λ)) = u(γ · λ), λ ∈ Λ(H), These maps are compatible with the multiplication rule and the ∗-operation of Ceµ(⊕Z2H) . Since these maps preserve the trace, they extend to ∗-automorphisms 6 HIROKI SAKO on Leµ(⊕Z2H). It is immediate to see that ρ(c) commutes with ρ(k) and ρ(γ). For k, l ∈ Z2, we have the following relation: χ(λ(m))det(l,m)ρ(k)(u(l · λ)) χ(λ(m))det(l,m) Ym∈Z2 χ(λ(m))det(l,m)χ(λ(m))det(k,m+l)u((k + l) · λ). χ((l · λ)(m))det(k,m)u(k · (l · λ)) ρ(k) ◦ ρ(l)(u(λ)) = Ym∈Z2 = Ym∈Z2 = Ym∈Z2 χ(λ(m))!det(k,l) Ym∈Z2 Ym∈Z2 = ρ(χdet(k,l)) ◦ ρ(k + l)(u(λ)). By det(l, m) + det(k, m + l) = det(k, l) + det(k + l, m), this equals to χ(λ(m))det(k+l,m)u((k + l) · λ) Since det is SL(2, Z)-invariant (Lemma 2.1), for k ∈ Z2, γ ∈ SL(2, Z), we get ρ(γ · k) ◦ ρ(γ)(u(λ)) = ρ(γ · k)(u(γ · λ)) = Yl∈Z2 = Yl∈Z2 = Yl∈Z2 χ((γ · λ)(l))det(γ·k,l)u((γ · k) · (γ · λ)) χ(λ(l))det(γ·k,γ·l)u(γ · (k · λ)) χ(λ(l))det(k,l)ρ(γ)(u(k · λ)) = ρ(γ) ◦ ρ(k)(u(λ)), λ ∈ Λ(H). By using the above two equations, ρ satisfies the following formula: (ρ(c1) ◦ ρ(k) ◦ ρ(γ1)) ◦ (ρ(c2) ◦ ρ(l) ◦ ρ(γ2)) = ρ(c1) ◦ ρ(c2) ◦ ρ(k) ◦ ρ(γ1) ◦ ρ(l) ◦ ρ(γ2) = ρ(c1) ◦ ρ(c2) ◦ ρ(k) ◦ ρ(γ1 · l) ◦ ρ(γ1) ◦ ρ(γ2) = ρ(c1) ◦ ρ(c2) ◦ ρ(χdet(k,γ1·l)) ◦ ρ(k + (γ1 · l)) ◦ ρ(γ1γ2) = ρ(c1c2χdet(k,γ1·l)) ◦ ρ(k + (γ1 · l)) ◦ ρ(γ1γ2). With ρ(c, k, γ) = ρ(c) ◦ ρ(k) ◦ ρ(γ), ρ gives a Γ0-action on Leµ(⊕Z2H). We define the finite von Neumann algebra N(H, µ) as the group von Neumann algebra Leµ(Λ(H)). By using Fourier coefficients, we can prove that N(H, µ) is the fixed point algebra under the bH-action ρ(bH, 0, e) on Leµ(⊕Z2H). We get a Z2 ⋊ SL(2, Z)-action on N(H, µ) by β(k, γ)(x) = ρ(1, k, γ)(x), k ∈ Z2, γ ∈ SL(2, Z), x ∈ N(H, µ). This is the definition of the twisted Bernoulli shift action β = β(H, µ, χ) on N(H, µ). We obtained the actions β(H, µ, χ) not only by twisting generalized Bernoulli shift actions but also restricting to subalgebras N(H, µ) ⊂ Lbµ(⊕Z2H) = NZ2 Lµ(H). This restriction allows us to classify the actions up to conjugacy in the next section. CLASSIFICATION AND CENTRALIZERS 7 In order to give a variety of the actions, we twisted the shift actions by the character χ of the abelian group H. Remark 3.1. The action βZ2 = β(H, µ, χ)Z2 has the weak mixing property. In definition 2.5, we may assume that the Fourier coefficients of ai (i = 1, 2,· · · , n) are finitely supported, by approximating in the L2-norm. Then for appropriate k ∈ Z2, we get tr(aiβ(k)(aj)) = tr(ai)tr(aj), i, j = 1, 2,· · · , n. 4. Classification up to conjugacy In this section, we classify the twisted Bernoulli shift actions {β(H, µ, χ)} up to conjugacy (Theorem 4.1). We prove that an isomorphism which gives conjugacy between two twisted Bernoulli shift actions is of a very special form. In fact it comes from an isomorphism in the level of base groups H. We also determine the centralizer of the Z2 ⋊ SL(2, Z)-action β(H, µ, χ) on N(H, µ) (Theorem 4.4). We fix some notations for the proofs. We define 0, e1, e2 ∈ Z2 as 0 =(cid:18) 0 0 (cid:19) , e1 =(cid:18) 1 0 (cid:19) , e2 =(cid:18) 0 1 (cid:19) . Let ξ be the element of Z2 ⋊ SL(2, Z) satisfying ξ · 0 = e1, ξ · e1 = e2, ξ · e2 = 0. The elements ξ and ξ2 are explicitly described as 0 (cid:19) (cid:19) , −1 −1 (cid:19) (cid:19) . The order of ξ is 3. Let η, δ ∈ SL(2, Z) be given by η =(cid:18) −1 ξ = (cid:18) e1, (cid:18) −1 −1 ξ2 = (cid:18) e2, (cid:18) 0  1 1 D =(cid:26) (cid:18) n 0 (cid:19) n ∈ Z (cid:27) . Let D be the subset of all elements of Z2 fixed under the action of δ, that is, 0 0 −1 (cid:19), δ =(cid:18) 1 1 0 1 (cid:19). Then we get ξ · D =(cid:26) (cid:18) 1 − n n (cid:19) n ∈ Z (cid:27) , ξ2 · D =(cid:26) (cid:18) 0 1 − n (cid:19) n ∈ Z (cid:27) . We define the subgroup ΛD(H) of Λ(H) by ΛD(H) = {λ ∈ Λ(H) λ : Z2 → H is supported on D}. Let (Ha, µa, χa) and (Hb, µb, χb) be triplets of countable abelian groups, their normalized 2-cocycles and characters. For h ∈ Ha, we define λh ∈ ΛD(Ha) as λh(k) = h (k = e1), −h (k = 0), 0 (k 6= e1, 0). 8 HIROKI SAKO For g ∈ Hb, we define σg ∈ ΛD(Hb) as σg(k) = g (k = e1), −g (k = 0), 0 (k 6= e1, 0). We denote by v(σ) ∈ N(Hb, µb) the unitary corresponding to σ ∈ Λ(Hb). Theorem 4.1. If π : N(Ha, µa) → N(Hb, µb) is a ∗-isomorphism giving conjugacy between βa = β(Ha, µa, χa) and βb = β(Hb, µb, χb), then there exists a group isomor- phism φ = φπ : Ha → Hb satisfying (1) π(u(λ)) = v(φ ◦ λ) mod T for λ ∈ Λ(Ha), (2) the 2-cocycles µa(·,·) and µb(φ(·), φ(·)) of Ha are cohomologous, (3) χ2 a = (χb ◦ φ)2. Conversely, given a group isomorphism φ : Ha → Hb satisfying (2) and (3), there exists a ∗-isomorphism π = πφ : N(Ha, µa) → N(Hb, µb) which satisfies condition (1) and gives conjugacy between βa, βb. We note that by Proposition 2.2 condition (2) for φ is equivalent to (2)′ µ∗ aµa(g, h) = µ∗ bµb(φ(g), φ(h)), g, h ∈ Ha. Proof for the first half of Theorem 4.1. Suppose that there exists a (not necessarily trace preserving) ∗-isomorphism π from N(Ha, µa) onto N(Hb, µb) such that π ◦ βa(γ) = βb(γ) ◦ π, γ ∈ Z2 ⋊ SL(2, Z). We prove that for every h ∈ Ha there exists φ(h) ∈ Hb satisfying π(u(λh)) = v(σφ(h)) mod T. Let Uh denote the unitary in N(Hb, µb) Uh = π(cid:16)µa(h,−h) u(λh)(cid:17) , h ∈ Ha. which is canonically isomorphic to Leµ (⊕Z2H). The preimage π−1(Uh) can be written h ⊗ uh. Here uh is the unitary corresponding to h ∈ Ha and placed on 1 ∈ Z2 as u∗ h is placed on 0 ∈ Z2. We describe Uh as the Fourier expansion and the unitary u∗ We identify N(H, µ) with the subalgebra of the infinite tensor productNZ2 Lµ(H), Uh =Pσ∈Λ(Hb) c(σ)v(σ). Since e1 and 0 are fixed under the action of δ, one has It follows that the Fourier expansion Uh = Pσ∈Λ(Hb) c(σ)v(σ) must satisfy that Σc(σ)2 = 1 < +∞, so that Uh =Pσ∈ΛD (Hb) c(σ)v(σ). c(σ) = c(δ−n · σ) for every σ ∈ Λ(Hb) and n ∈ Z. For σ ∈ Λ(Hb) \ ΛD(Hb), the orbit of σ under the action of δ−1 is an infinite set, since the support supp(σ) ⊂ Z2 is not included in D. It turns out that c(σ) = 0 for all σ ∈ Λ(Hb) \ ΛD(Hb) due to βb(δ)n(Uh) = π ◦ βa(δ)n(π−1(Uh)) = Uh. The unitary χa(h)U ∗ h is also fixed under the action of δ and can be written as χa(h)U ∗ h = π(cid:16)χa(h)µa(h,−h)u(−λh)(cid:17) = π(cid:16)χa(h)µa(h,−h)u(ξ · λh)(cid:17) π(cid:16)µa(h,−h)u(ξ2 · λh)(cid:17) = βb(ξ)(Uh) βb(ξ2)(Uh). CLASSIFICATION AND CENTRALIZERS 9 Letting ne1 = (n, 0)T ∈ Z2, we get βb(ξ)(Uh) = βb(ξ)(cid:16)X c(σ) v(σ)(cid:17) = Xσ∈ΛD(Hb) βb(ξ2)(Uh) = βb(ξ2)(cid:16)X c(σ) v(σ)(cid:17) = Xσ∈ΛD(Hb) c(σ) v(ξ · σ)Yn∈Z c(σ) v(ξ2 · σ). χb(σ(ne1))n, Since Fourier expansion admits algebraical calculation as in subsection 2.6, the ex- pansion of χa(h)U ∗ h is χb(σ1(ne1))n χb(σ1(ne1))n v(ξ · σ1 + ξ2 · σ2). χa(h)U ∗ h = βb(ξ)(Uh) βb(ξ2)(Uh) = Xσ1,σ2∈ΛD(Hb) = Xσ1,σ2∈ΛD(Hb) c(σ1) c(σ2) v(ξ · σ1) v(ξ2 · σ2)Yn∈Z c(σ1) c(σ2)eµb(ξ · σ1, ξ2 · σ2)Yn∈Z and σ1(e1) = −Pk∈D\{e1} σ1(k). Here we used the condition P σ1(k) = 0. The The map ΛD(Hb)×ΛD(Hb) ∋ (σ1, σ2) 7→ ξ·σ1+ξ2·σ2 ∈ Λ(Hb) is injective. Indeed, σ1 is uniquely determined by ξ·σ1+ξ2·σ2, since σ1(k) = (ξ·σ1+ξ2·σ2)(ξ·k), k ∈ D\{e1} element σ2 is also determined by ξ · σ1 + ξ2 · σ2. Thus the index (σ1, σ2) uniquely determines ξ · σ1 + ξ2 · σ2. We take arbitrary elements σ1, σ2 ∈ ΛD(Hb) and suppose that c(σ1) 6= 0, c(σ2) 6= 0. Since the unitary χa(h)U ∗ h is invariant under the action of δ and the coefficient of ξ· σ1 + ξ2· σ2 is not zero, ξ· σ1 + ξ2· σ2 is supported on D. It follows that the elements σ1 and σ2 can be written as σ1 = σφ(h) = σ2, by some φ(h) ∈ Hb. Indeed, since the subsets D \{0, e1}, ξD\{e1, e2} and ξ2D \{e2, 0} are mutually disjoint, the element ξ · σ1 must be supported on {e1, e2} and the element ξ2 · σ2 must be supported on Then using the fact that (ξ · σ1 + ξ2 · σ2)(e2) = σ1(e1) + σ2(0) = 0, we get that σ1 = σφ(h) = σ2 for some h ∈ Hb. This means that there exists only one σ ∈ Λ(Hb) such that c(σ) 6= 0 and that it is of the form σ = σφ(h). Then the unitary Uh satisfies Uh = π(u(λh)) = v(σφ(h)) mod T. We claim that the map φ = φπ : Ha → Hb is a group isomorphism. For all {e2, 0}. By the assumptionPk∈Z2 σi(k) = 0 (i = 1, 2), σi can be written as σφ(hi). h1, h2 ∈ Ha, we get π(u(λh1+h2)) = π(u(λh1)) π(u(λh2)) = v(σφ(h1)) v(σφ(h2)) = v(σφ(h1) + σφ(h2)) = v(σφ(h1)+φ(h2)) mod T. On the other hand, we get π(u(λh1+h2)) = v(σφ(h1+h2)) mod T. Since {v(σ)} are linearly independent, we get σφ(h1+h2) = σφ(h1)+φ(h2), and hence φ(h1 + h2) = φ(h1) + φ(h2). This means that the map φ is a group homomorphism. The bijectivity of the ∗- isomorphism π leads to that of the group homomorphism φ = φπ. Since {γ · λh γ ∈ Z2 ⋊ SL(2, Z), h ∈ Ha} ⊂ Λ(Ha) generates Λ(Ha), we get π(u(λ)) = v(φ◦ λ) mod T for λ ∈ Λ(Ha). 10 HIROKI SAKO We prove that the group isomorphism φ = φπ satisfies conditions (2) and (3) in the theorem. For all h ∈ Ha, there exists c(h) ∈ T satisfying Since (e1, η) ∈ Z2 ⋊ SL(2, Z) acts on Z2 as (e1, η) · e1 = 0, (e1, η) · 0 = e1, we get Uh = π(cid:16)µa(h,−h) u(λh)(cid:17) = c(h) µb(φ(h),−φ(h)) v(σφ(h)). Uh βb(e1, η)(Uh) = π(cid:16)µa(h,−h) u(λh) µa(h,−h) u(−λh)(cid:17) 2 = µa(h,−h) The following equation also holds: Uh βb(e1, η)(Uh) = c(h) µb(φ(h),−φ(h)) v(σφ(h)) c(h) µb(φ(h),−φ(h)) v(σ−φ(h)) fµa(λh,−λh) = 1. eµb(σφ(h),−σφ(h)) = c(h)2. 2 = c(h)2 µb(φ(h),−φ(h)) Thus we have c(h) ∈ {1,−1} for h ∈ Ha. Since ξ · e1 = e2, ξ · e2 = 0 and ξ · 0 = e1, we have Uh βb(ξ)(Uh) βb(ξ2)(Uh) = π(cid:16)µa(h,−h) u(λh)(cid:17) π(cid:16)χa(h) µa(h,−h) u(ξ · λh)(cid:17) π(cid:16)µa(h,−h) u(ξ2 · λh)(cid:17) = χa(h). On the other hand, we have the following: Uh βb(ξ)(Uh) βb(ξ2)(Uh) = c(h) µb(φ(h),−φ(h)) v(σφ(h)) c(h) χb(φ(h)) µb(φ(h),−φ(h)) v(ξ · σφ(h)) c(h) µb(φ(h),−φ(h)) v(ξ2 · σφ(h)) = c(h)3 χb(φ(h)) = c(h) χb(φ(h)). It follows that (Eq2) and χb(φ(h))2 = χa(h)2, for all h ∈ Ha. c(h) = χb(φ(h)) χa(h) tensor product NZ2 Lµa(Ha). The unitary π−1(Uh) ∈ N(Ha, µa) ⊂ NZ2 Lµa(Ha) We recall that the algebra Lfµa(⊕Z2Ha) is canonically identified wit the infinite can be written as 1 ⊗ u∗ h is placed on 0 and uh is placed on e1. Since η ∈ Z2 ⋊ SL(2, Z) acts on Z2 as η · e1 = −e1, η · 0 = 0, the unitary π−1(βb(η)(Ug)) can be written as ug ⊗ u∗ g ⊗ 1. We have the following equation: h ⊗ uh, where 1 is placed on −e1 ∈ Z2, u∗ Ug βb(η)(Uh) U ∗ g βb(η)(Uh)∗ = π((1 ⊗ u∗ g ⊗ ug)(uh ⊗ u∗ h ⊗ 1)(1 ⊗ u∗ The unitary Uh can be written as c(h)(1 ⊗ v∗ Here we write vφ(h) for the unitary in Lµb(Hb) corresponding to φ(h). The unitary βb(η)(Ug) can be written as c(g)(vφ(g) ⊗ v∗ g ⊗ ug)∗(uh ⊗ u∗ φ(h) ⊗ vφ(h)) ∈ N(Hb, µb) ⊂N Lµb(Hb). φ(g) ⊗ 1). Then we get h ⊗ 1)∗) = µ∗ aµa(g, h). Ug βb(η)(Uh) U ∗ g βb(η)(U ∗ h) φ(g) ⊗ vφ(g))(vφ(h) ⊗ v∗ bµb(φ(g), φ(h)). = (1 ⊗ v∗ = µ∗ φ(h) ⊗ 1)(1 ⊗ v∗ φ(g) ⊗ vφ(g))∗(vφ(h) ⊗ v∗ φ(h) ⊗ 1)∗ CLASSIFICATION AND CENTRALIZERS 11 Thus we get µ∗ group isomorphism φ = φπ satisfies conditions (1), (2) and (3). bµb(φ(g), φ(h)), for all g, h ∈ Ha. We proved that the aµa(g, h) = µ∗ (cid:3) λ ∈ Λ(H), bµ(λ) = µ j−1Xi=0 λ(ki), λ(kj)! , From a group homomorphism which satisfies conditions (2) and (3), we construct a ∗-homomorphism from N(Ha, µa) to N(Hb, µb) with condition (1). In the con- Z2 = {k0, k1, k2,· · ·} throughout the rest of this section. For a scalar 2-cocycle µ of where λ is supported on {k0, k1, k2,· · · , kn}. This definition depends on the choice relation in Cµ(H): struction, the function bµ on Λ(H) given below is useful. We fix an index for Z2 as H, we define the functionbµ by nYj=1 of an order on Z2. SincePi λ(ki) = 0, the functionbµ is also given by the following If µ is a coboundary, then the definition of bµ does not depend on the order on Z2, Lemma 4.2. Let µ0 be another normalized scalar 2-cocycle for H. Let eµ0 be the tion 2.2 and let bµ0 be the function on Λ(H) constructed from µ0 in the above manner. scalar 2-cocycle on Λ(H)×Λ(H) given in the same way as equation (Eq1) in subsec- If the scalar 2-cocycles µ and µ0 are cohomologous, then for all λ1, λ2 ∈ Λ(H), we have the equation bµ(λ)1 = uλ(k0)uλ(k1)uλ(k2) · · · uλ(kn), since Cµ(H) is commutative. λ ∈ Λ(H). Proof. We denote by {ν(g, h)} the scalar 2-cocycle {µ0(g, h)µ(g, h)} of H. Since ν is a 2-coboundary, there exists {c(g)}g∈H ⊂ T satisfying ν(g, h) = b(g)b(h)b(g + h). eµ(λ1, λ2)bµ(λ1)bµ(λ2)bµ(λ1 + λ2) = eµ0(λ1, λ2) bµ0(λ1)bµ0(λ2) bµ0(λ1 + λ2). Then the mapbν becomesbν(λ) =Qi b(λ(ki)). Since b(λ1(ki)) b(λ2(ki)), bν(λ1)bν(λ2) = Yi eν(λ1, λ2) = Yi bν(λ1 + λ2) = Yi bν(λ) =bµ(λ) bµ0(λ), b(λ1(ki)) b(λ2(ki)) b(λ1(ki) + λ2(ki)), b(λ1(ki) + λ2(ki)), we getbν(λ1)bν(λ2) =eν(λ1, λ2)bν(λ1 + λ2). By the definitions of eµ, eµ0,bµ and bµ0, the mapsbν andeν are given by Thus the desired equality immediately follows. eν(λ1, λ2) =eµ(λ1, λ2) eµ0(λ1, λ2), Proof for the second half of Theorem 4.1. Suppose that there exists a group isomorphism φ satisfying conditions (2) and (3) in the theorem. We prove that there exists a ∗-isomorphism π = πφ from N(Ha, µa) onto N(Hb, µb) preserving the Z2 ⋊ SL(2, Z)-actions with condition (1). (cid:3) We define a group homomorphism cφ from Ha to {1,−1} ⊂ T by cφ(h) = χb(φ(h))χa(h), h ∈ Ha. 12 HIROKI SAKO We define a linear map π from the group algebra Ceµa(Λ(Ha)) onto Ceµb(Λ(Hb)) by λ ∈ Λ(Ha). gcd(k) , λ ∈ Λ(Ha). χa(λ(k))gcd(k)χb(φ(λ(k))) On the other hand, we have the following equation: By direct computations, for all λ1, λ2 ∈ Λ(Ha), we get Let ecφ be the group homomorphism from Λ(Ha) to {1,−1} ⊂ T given by ecφ(λ) = Yk∈Z2 cφ(λ(k))gcd(k) = Yk∈Z2 π(cid:16)cµa(λ) u(λ)(cid:17) = ecφ(λ)bµb(φ ◦ λ) v(φ ◦ λ), π(cid:16)cµa(λ1) u(λ1)(cid:17) π(cid:16)cµa(λ2) u(λ2)(cid:17) = ecφ(λ1)ecφ(λ2)bµb(φ ◦ λ1)bµb(φ ◦ λ2) v(φ ◦ λ1) v(φ ◦ λ2) = ecφ(λ1 + λ2)bµb(φ ◦ λ1)bµb(φ ◦ λ2)eµb(φ ◦ λ1, φ ◦ λ2) v(φ ◦ (λ1 + λ2)). π(cid:16)cµa(λ1)cµa(λ2) u(λ1) u(λ2)(cid:17) = π(cid:16)cµa(λ1)cµa(λ2)fµa(λ1, λ2) u(λ1 + λ2)(cid:17) = cµa(λ1)cµa(λ2)fµa(λ1, λ2)cµa(λ1 + λ2) π(cid:16)cµa(λ1 + λ2) u(λ1 + λ2)(cid:17) = ecφ(λ1 + λ2)cµa(λ1)cµa(λ2)fµa(λ1, λ2)cµa(λ1 + λ2) bµb(φ ◦ (λ1 + λ2)) v(φ ◦ (λ1 + λ2)). bµb(φ ◦ λ1)bµb(φ ◦ λ2)eµb(φ ◦ λ1, φ ◦ λ2) = cµa(λ1)cµa(λ2)fµa(λ1, λ2)cµa(λ1 + λ2)bµb(φ ◦ (λ1) + φ ◦ (λ2)). By Lemma 4.2 and condition (2) for the group isomorphism φ in the theorem, we have that Therefore we get π(u(λ1)) π(u(λ2)) = π(u(λ1) u(λ2)). The linear map π also pre- serves the ∗-operation. As a consequence, π is a ∗-isomorphism from Ceµa(Λ(Ha)) onto Ceµb(Λ(Hb)) and this preserves the trace. The map π = πφ is extended to a normal ∗-isomorphism from N(Ha, µa) onto N(Hb, µb). We next prove that this π preserves the Z2 ⋊ SL(2, Z)-actions. The group ho- by Lemma 2.1 (1). The scalar 2-cocycle ν(g, h) = µa(g, h)µb(φ(h), φ(g)) satisfies momorphism ecφ from Λ(Ha) to {1,−1} is invariant under the action of SL(2, Z), ν(g, h) = ν(h, g) by condition (2), so the function bν(·) = cµa(·)bµb(φ ◦ ·) on Λ(Ha) does not depend on the order on Z2 chosen before. Since π ◦ βa(γ)(u(λ)) = π(u(γ · λ)) = ecφ(γ · λ)cµa(γ · λ)bµb(φ ◦ (γ · λ))v(φ ◦ (γ · λ)) = ecφ(λ)cµa(λ)bµb(φ ◦ λ)v(γ · (φ ◦ λ)) = βb(γ)(cid:16)ecφ(λ)cµa(λ)bµb(φ ◦ λ)v(φ ◦ λ)(cid:17) γ ∈ SL(2, Z), λ ∈ Λ(Ha), = βb(γ) ◦ π(u(λ)), it turns out that the ∗-isomorphism π preserves the SL(2, Z)-action. π ◦ βa(k)(u(λ)) = π Yl∈Z2 cφ((k · λ)(l))gcd(l)Yl∈Z2 = Yl∈Z2 χa(λ(l))det(k,l)u(k · λ)! χa(λ(l))det(k,l)cµa(k · λ)bµb(φ ◦ (k · λ)) v(φ ◦ (k · λ)). Since cφ(h)det(k,l)cφ(h)gcd(k+l) = cφ(h)gcd(k)cφ(h)gcd(l), by Lemma 2.1 (2), the unitary π ◦ βa(k)(u(λ)) equals to cφ(λ(l))gcd(k+l)Yl∈Z2(cid:0)cφ(λ(l))det(k,l)χb(φ ◦ λ(l))det(k,l)(cid:1) Yl∈Z2 cµa(k · λ)bµb(φ ◦ (k · λ)) v(φ ◦ (k · λ)) cφ(λ(l))gcd(l)Yl∈Z2 cφ(λ(l))gcd(k)Yl∈Z2 = Yl∈Z2 cµa(k · λ)bµb(φ ◦ (k · λ)) v(k · (φ ◦ λ)) = ecφ(λ)cµa(k · λ)bµb(k · (φ ◦ λ)) βb(k)(v(φ · λ)) = βb(k) ◦ π(u(λ)). χb(φ ◦ λ(l))det(k,l) CLASSIFICATION AND CENTRALIZERS 13 For all λ ∈ Λ(Ha) and k ∈ Z2, we have This means that the ∗-isomorphism π preserves the Z2-actions. between βa and βb. We get the ∗-isomorphism π = πφ from N(Ha, µa) onto N(Hb, µb) giving conjugacy (cid:3) Remark 4.3. The proof of the first half of Theorem 4.1 shows that any isomorphism π giving conjugacy between βa and βb is of the form πφ. This means that an isomorphism which gives conjugacy between two twisted Bernoulli shift actions must be trace preserving. This proof shows that an isomorphism giving conjugacy between the two actions β(Ha, µa, χa), β(Hb, µb, χb) is of a very special form derived from a group isomor- phism between Ha and Hb. Taking notice of this fact, we can describe the centralizer of a twisted Bernoulli shift action. We define two topological groups before we state Theorem 4.4. Let β be a trace preserving action of some group Γ on a separable finite von Neumann algebra (N, tr). We denote by Aut(N, β) the group of all automorphisms which commute with the action β, that is, {α ∈ Aut(N) β(γ) ◦ α = α ◦ β(γ), γ ∈ Γ}. We regard the group Aut(N, β) as a topological group equipped with the pointwise- strong topology. When β is a twisted Bernoulli shift action on N, an automorphism α commuting with β is necessarily trace preserving by Remark 4.3. We consider that Aut(N, β) is equipped with the pointwise-2-norm topology. Let Aut(H, µ, χ) be the group of all automorphisms of an abelian group H which preserve its 2-cocycle µ and character χ, that is, {φ ∈ Aut(H) µ(g, h) = µ(φ(g), φ(h)), χ(g) = χ(φ(g)), g, h ∈ H}. We define the topology of Aut(H, µ, χ) by pointwise convergence. 14 HIROKI SAKO Theorem 4.4. For π ∈ Aut(N(H, µ), β(H, µ, χ)), there exists a unique element φ = φπ ∈ Aut(H, µ∗µ, χ2) satisfying π(u(λ)) = u(φ ◦ λ) mod T for λ ∈ Λ(H). The map π 7→ φπ gives an isomorphism between two topological groups Aut(N(H, µ), β(H, µ, χ)) ∼= Aut(H, µ∗µ, χ2). Proof. We use the notations in the proof of the previous theorem letting Ha = Hb = H, µa = µb = µ and χa = χb = χ. Denote N = N(H, µ) and β = β(H, µ, χ). We have already have shown the first claim. Let Aut(H, µ∗µ, χ2) ∋ φ 7→ πφ ∈ Aut(N, β), be the map given as in the proof of Theorem 4.1, that is, where πφ(cid:16)bµ(λ) u(λ)(cid:17) = ecφ(λ)bµ(φ ◦ λ) u(φ ◦ λ), χ(λ(k))gcd(k) χ(φ ◦ λ(k)) λ ∈ Λ(H), gcd(k) . ecφ(λ) = Yk∈Z2 It is easy to prove that φπφ = φ by the definition. Thus the map π 7→ φπ is surjective. This map is also injective. Let φ be an element of Aut(H, µ∗µ, χ2). Suppose that π is an arbitrary element of Aut(N, β) satisfying φ = φπ. The set {β(γ)(u(λh)) h ∈ H, γ ∈ Z2 ⋊ SL(2, Z)} generates N, so we have only to prove the uniqueness of c(h) ∈ T satisfying π(cid:16)µ(h,−h) u(λh)(cid:17) = c(h) µ(φ(h),−φ(h)) u(λφ(h)), for all h ∈ H. In the proof of the first half of the previous theorem (equation (Eq2)), we have already shown that c(h) = χ(h)χ(φ(h)). Thus the ∗-isomorphism π is uniquely determined and the map π 7→ φπ is injective. We prove the two maps φ 7→ πφ and π 7→ φπ are continuous. Let (φi) be a net in Aut(H, µ∗µ, χ2) converging to φ. For all h ∈ H, we have The right side of the equation converges to πφi(cid:16)χ(h) µ(h,−h) u(λh)(cid:17) = χ(φi(h)) µ(φi(h),−φi(h)) u(λφi(h)). χ(φ(h)) µ(φ(h),−φ(h)) u(λφ(h)) = πφ(cid:16)χ(h) µ(h,−h) u(λh)(cid:17) . Conversely, This proves that πφi converges to πφ in pointwise 2-norm topology on the generating set {β(γ)(u(λh)) h ∈ H, γ ∈ Z2 ⋊ SL(2, Z)} of N. Thus πφi converges to πφ on N. let (πi) be a net in Aut(N, β) converging to π. For all h ∈ H, we get πi(u(λh)) = u(λφπi (h)) mod T. The left side of the equation converges to π(u(λh)) = u(λφπ(h)). If φπi(h) 6= φπ(h), then the distance between Tu(λφπ(h)) and Tu(λφπi (h)) is √2 in the 2-norm. Thus φπi(h) = φπ(h) for large enough i. This means that (φπi) converges to φπ. As a consequence, the two maps φ 7→ πφ and π 7→ φπ are continuous group homomorphisms and inverse maps of each other. (cid:3) CLASSIFICATION AND CENTRALIZERS 15 5. Examples 5.1. Twisted Bernoulli shift actions on L∞(X). In this subsection, we consider the case of µ = 1 and H 6= {0}. Then the algebra N(H, 1) is abelian and has a faithful normal state, so it is isomorphic to L∞(X), where X is a standard probability space. The measure of X is determined by the trace on N(H, 1). Furthermore, X is non-atomic, since N(H, 1) is infinite dimensional and the action β(H, 1, χ) is ergodic. As corollaries of Theorems 4.1 and 4.4, we get trace preserving Z2 ⋊ SL(2, Z)-actions on L∞(X) whose centralizers are isomorphic to some prescribed groups. Remark 5.1. The Z2 ⋊ SL(2, Z)-action on X defined by β = β(H, 1, χ) is free. An automorphism α ∈ Aut(L∞(X), β) is free or the identity map for any twisted Bernoulli shift action β on L∞(X). This is proved as follows. We identify β(γ) (γ ∈ Z2 ⋊ SL(2, Z)) and α with measure preserving Borel isomorphisms on X here. Suppose that there exists a non-null Borel subset Y ⊂ X whose elements are fixed under α. All elements in eY = ∪{β(γ)(Y ) γ ∈ Z2 ⋊ SL(2, Z)} are fixed under α. By the ergodicity of β, the measure of eY is 1. Then α is the identity map of L∞(X). Corollary 5.2. For any abelian countable discrete group H 6= {0}, there exists a trace preserving essentially free ergodic action β of Z2⋊SL(2, Z) on L∞(X) satisfying Aut(L∞(X), β) ∼= Aut(H). Proof. When we define β = β(H, 1, 1), we have the above relation by Theorem 4.4. (cid:3) In the next corollary we use the effect of twisting by a character χ. Corollary 5.3. For every abelian countable discrete group H 6= {0}, there exist continuously many trace preserving essentially free ergodic actions {βc} of Z2 ⋊ SL(2, Z) on L∞(X) which are mutually non-conjugate and satisfy Aut(L∞(X), βc) ∼= H ⋊ Aut(H). Here the topology of H ⋊ Aut(H) is the product of the discrete topology on H and the pointwise convergence topology on Aut(H). Proof. Let c ∈ {eiπt t ∈ (0, 1/2) \ Q}. We put βc = β(H ⊕ Z, 1, 1 × χc), where the character χc of Z is defined as χc(n) = cn. By Theorem 4.4, we get Aut(L∞(X), βc) ∼= Aut(H ⊕ Z, 1, 1 × χ2 c). c is injective, a group automorphism α ∈ Aut(H ⊕ Z, 1, 1× χ2 c) c), there exist φα ∈ Aut(H) Since the character χ2 preserves the second entry. For all α ∈ Aut(H⊕Z, 1, 1×χ2 and hα ∈ H satisfying α(h, n) = (φα(h) + nhα, n), (h, n) ∈ H ⊕ Z. The map Aut(H ⊕ Z, 1, 1 × χ2 group isomorphism. c) ∋ α 7→ (hα, φα) ∈ H ⋊ Aut(H) is a homeomorphic If c1, c2 ∈ {eiπt t ∈ (0, 1/2) \ Q} and c1 6= c2, then there exists no isomorphism c2 is equal to 1 × χ2 c1. (cid:3) from H ⊕ Z to H ⊕ Z whose pull back of the character 1 × χ2 The two actions βc1 and βc2 are not conjugate by Theorem 4.1. 16 HIROKI SAKO Corollary 5.4. There exist continuously many trace preserving essentially free er- godic actions {βc} of Z2 ⋊ SL(2, Z) on L∞(X) which are mutually non-conjugate and have the trivial centralizer Aut(L∞(X), βc) = {idL∞X}. Proof. Let {χc c = eiπt, t ∈ (0, 1/2)} be characters of Z such that χc(m) = cm. Since χ2 c(1) is in the upper half plane, the identity map is the only automorphism of Z preserving χ2 c. By Theorem 4.4, we get Aut(β(Z, 1, χc)) = {id}. c2 is χ2 If β(Z, 1, χc1), β(Z, 1, χc2) are conjugate, then there exists a group isomorphism c1 by Theorem 4.1. This means c1 = c2. Thus the (cid:3) on Z whose pull back of χ2 actions {β(Z, 1, χc)} are mutually non-conjugate. 5.2. Twisted Bernoulli shift actions on the AFD factor of type II1. Firstly, we find a condition that the finite von Neumann algebra N(H, µ) is the AFD factor of type II1. Lemma 5.5. For an abelian countable discrete group H 6= {0} and its normalized scalar 2-cocycle µ, the following statements are equivalent: (1) The algebra N(H, µ) is the AFD factor of type II1. (2) The group von Neumann algebra Lµ(H) twisted by the scalar 2-cocycle µ is a factor (of type II1 or In). (3) For all g ∈ H \ {0}, there exists h ∈ H such that µ(g, h) 6= µ(h, g). Proof. The amenability of the group Λ(H) leads the injectivity for N(H, µ). The injectivity for N(H, µ) implies that N(H, µ) is approximately finite dimensional ([Co]). We have only to show the equivalence of conditions (2), (3) and (1)′ The algebra N(H, µ) is a factor. By using Fourier expansion it is easy to see that condition (2) holds true if and only if for any g ∈ H \ {0} there exists h ∈ H satisfying uguh 6= uhug. This is equivalent to condition (3). Similarly, condition (1)′ is equivalent to (1)′′ For any λ1 ∈ Λ(H) \ {0}, there exists λ2 ∈ Λ(H) satisfying eµ(λ2, λ1)eµ(λ1, λ2) 6= 1. Suppose condition (3). For any λ1, choose element k, l ∈ Z2 so that k ∈ supp(λ1) and l /∈ supp(λ1). By condition (3), there exists h ∈ H satisfying µ∗µ(λ1(k), h) 6= 1. Let λ2 be the element in Λ(H) which takes h at k, −h at l and 0 for the other condition (1)′′. The implication from (1)′′ to (3) is easily shown. places. The element λ2 satisfieseµ(λ2, λ1)eµ(λ1, λ2) = µ∗µ(λ1(k), h) 6= 1. Here we get Remark 5.6. The twisted Bernoulli shift action β = β(H, µ, χ) is an outer action of Z2 ⋊ SL(2, Z). Any non-trivial automorphism in Aut(R, β(H, µ, χ)) is also outer. This is proved by the weak mixing property of the action β(H, µ, χ) as follows. If α ∈ Aut(R, β(H, µ, χ)) is an inner automorphism Ad(u), then we have (cid:3) Ad(β(γ)(u))(x) = β(γ)(uβ(γ)−1(x)u∗) = β(γ) ◦ α ◦ β(γ)−1(x) = α(x) = Ad(u)(x), for all x ∈ R and γ ∈ Z2 ⋊ SL(2, Z). Since Ad(β(γ)(u)u∗) = id, Cu ⊂ R is an invariant subspace of the action β. The only subspace invariant under the weakly mixing action β is C1 (Proposition 2.6), thus we get α = id. CLASSIFICATION AND CENTRALIZERS 17 Using Theorems 4.1 and 4.4, we give continuously many actions of Z2 ⋊ SL(2, Z) on R such that there exists no commuting automorphism except for trivial one. Corollary 5.7. There exist continuously many ergodic outer actions {βc} of Z2 ⋊ SL(2, Z) on the AFD factor R of type II1 which are mutually non-conjugate and have the trivial centralizer Aut(R, βc) = {idR}. Proof. We can choose and fix a character χ on Z2 such that χ2 is injective. Let {µc c = eiπt, t ∈ (0, 1/2) \ Q} be scalar 2-cocycles for Z2 defined by µc(cid:18)(cid:18) s1 t1 (cid:19) ,(cid:18) s2 t2 (cid:19)(cid:19) = cs1t2−t1s2, s1, t1, s2, t2 ∈ Z. We put βc = β(Z2, µc, χ). The 2-cocycle µc satisfies condition (3) in Lemma 5.5. Thus βc defines a Z2 ⋊ SL(2, Z)-action on R. By Theorem 4.4, we get the following isomorphism between topological groups: Aut(R, βc) ∼= Aut(Z2, µ∗ cµc, χ2) = Aut(Z2, µc2, χ2). Since the character χ2 of Z2 is injective, so the group of the right side is {idZ2}. This means that the action βc has trivial centralizers. Finally, we prove that the actions {βc c = eiπt, t ∈ (0, 1/2) \ Q} are mutually non-conjugate. Suppose that actions βc1 and βc2 are conjugate. By Theorem 4.1, there exists a group isomorphism φ of Z2 satisfying µc2 1 (g, h) = µc2 2 (φ(g), φ(h)), g, h ∈ Z2. A group isomorphism of Z2 is given by an element of GL(2, Z). If the automorphism φ is given by an element of SL(2, Z), we get c2 1. If φ is given by an element of GL(2, Z) \ SL(2, Z), then we get c2 1. Since both c1 and c2 have the form eiπt, t ∈ (0, 1/2), we get c1 = c2. (cid:3) 2 = −c2 2 = c2 Any cyclic group of an odd order can be realized as the centralizer of a twisted Bernoulli shift actions on R. Corollary 5.8. Let q be an odd natural number ≥ 3 and denote by Hq the abelian group (Z/qZ)2. We define the 2-cocycle µq and the character χq on Hq as µq(cid:18)(cid:18) s1 t1 (cid:19) ,(cid:18) s2 t2 (cid:19)(cid:19) = exp (2πis1t2/q), χq(cid:18)(cid:18) s1 t1 (cid:19)(cid:19) = exp (2πis1/q). Then the algebra N(Hq, µq) is the AFD factor R of type II1 and the centralizer of the twisted Bernoulli shift action βa = β(Hq, µq, χq) is isomorphic to Z/qZ. Let φ be in Aut(Hq, µ∗ Proof. By Lemma 5.5, it is shown that the algebra N(HQ, µQ) is the AFD factor of q) ∼= Z/qZ. type II1. Using Theorem 4.4, we have only to prove that Aut(Hq, µ∗ q). The automorphism φ of Hq is given by a 2 × 2 qµq, the determinant of A must be 1. Since q determines the first entry of (Z/qZ)2 and φ preserves χ2 q. matrix A of Z/qZ. Since φ preserves µ∗ q is odd, the value of χ2 The matrix A is of the form qµq, χ2 qµq, χ2 0 (cid:18) 1 tφ 1 (cid:19) , tφ ∈ Z/qZ. In turn, if the matrix A is of this form, it (cid:3) The map φ 7→ tφ is an isomorphism. qµq, χ2 defines an element in Aut(Hq, µ∗ q). µQ((sq), (tq)) = Yq∈Q χQ((sq)) = Yq∈Q µq(sq, tq), χq(sq), (sq), (tq) ∈ HQ, sq, tq ∈ Hq. Using Theorem 4.4, we have only to prove QµQ, χ2 Aut(HQ, µ∗ Q) ∼=Yq∈Q Z/qZ. 18 HIROKI SAKO Corollary 5.9. For a set Q consisting of odd prime numbers, let βQ be the tensor product Nq∈Q βq of the actions βq on the AFD factor of type II1. The centralizer of βQ is isomorphic toQq∈Q Z/qZ. Proof. The action βQ is the twisted Bernoulli shift action β(HQ, µQ, χQ), where HQ is the abelian group ⊕q∈QHq and the scalar 2-cocycle µq and a character χQ on HQ are given by A group automorphism φ of HQ = ⊕q∈QHq has a form φ((kq)) = (φq(kq)), for some {φq ∈ Aut(Hq)}. Thus we get Aut(HQ, µ∗ Aut(Hq, µ∗ QµQ, χ2 qµq, χ2 q). Q) ∼=Yq∈Q Together with the previous corollary, we get the conclusion. (cid:3) Remark 5.10. If Q1 6= Q2, then the two groupsQq∈Q1 Z/qZ are not isomorphic. The continuously many outer actions {βQ} are distinguished in view of conjugacy only by using the centralizers {Aut(R, βQ)}. Z/qZ andQq∈Q2 6. Malleability and rigidity arguments In this section, we give malleability and rigidity type arguments invented by S. Popa, in order to examine weak 1-cocycles for actions. See Popa [Po2], [Po3], [Po4] and Popa -- Sasyk [PoSa] for the references. S. Popa in [Po3] showed that every 1-cocycle for a Connes-Størmer Bernoulli shift by property (T) group (or w-rigid group like Z2 ⋊ SL(2, Z)) vanishes modulo scalars. As a consequence, two such actions are cocycle conjugate if and only if they are conjugate. In our case, 1- cocycles do not vanish modulo scalars but they are still in the situation that cocycle (outer) conjugacy implies conjugacy. We need the following notion to examine outer conjugacy of two group actions. Definition 6.1. Let α be an action of discrete group Γ on a von Neumann algebra M. A weak 1-cocycle for α is a map w : Γ → U(M) satisfying wgh = wgαg(wh) mod T, g, h ∈ Γ. The weak 1-cocycle w is called a weak 1-coboundary if there exists a unitary v ∈ U(M) satisfying wg = vαg(v)∗ mod T. Two weak 1-cocycles w and w′ are said to be equivalent when w′ g = vwgαg(v)∗ mod T for some v ∈ U(M). Let N be a finite von Neumann algebra with a faithful normal trace. The following is directly obtained by combining Lemmas 2.4 and 2.5 in [PoSa], although these Lemmas were proved for Bernoulli shift actions on standard probability space. The CLASSIFICATION AND CENTRALIZERS 19 following can be also regarded as a weak 1-cocycle version of Proposition 3.2 in [Po4]. g ∈ G. Proposition 6.2. Let G be a countable discrete group. Let β be a trace preserving weakly mixing action of G on N. A weak 1-cocycle {wg}g∈G ⊂ N for β is a weak 1-coboundary if only if there exists a non-zero element ex0 ∈ N ⊗ N satisfying The following is a weak 1-cocycle version of Proposition 3.6.3◦ in [Po4]. (wg ⊗ 1)(βg ⊗ βg)(ex0)(1 ⊗ w∗ g) = ex0, Proposition 6.3. Let Γ be a countable discrete group and G be a normal subgroup of Γ. The group Γ acts on a finite von Neumann algebra N in a trace-preserving way by β. Suppose that the restriction of β to G is weakly mixing. Let {wγ}γ∈Γ be a weak 1-cocycle for β. If wG is a weak 1-coboundary, then w is a weak 1-coboundary for the Γ-action. Proof. Suppose that wG is a weak 1-coboundary, that is, there exists a unitary element v in N such that wg = vβg(v∗) mod T for g ∈ G. It suffices to show that {w′ γ} = {v∗wγβγ(v)} is in T for all γ ∈ Γ. Take arbitrary γ ∈ Γ, g ∈ G. Write h = γ−1gγ ∈ G. Let πγ be the unitary on L2(N) induced from βγ. Since w′ g ∈ T, we get h, w′ w′ γπgw′ γ ∗ = (w′ γπγ)(w′ hπh)(w′ γπγ)∗ = w′ γhγ−1πγhγ−1 = πg mod T, By applying these operators to 1 ∈ bN ⊂ L2(N), it follows that w′ Since the G-action is weakly mixing, we have w′ γ ∈ T. By using the above propositions, we will "untwist" some weak 1-cocycles later. ∗) ∈ T. (cid:3) γβg(w′ γ We require some ergodicity assumption on the weak 1-cocycles. Definition 6.4. Let Γ be a discrete group and G be a subgroup of Γ. Suppose that its restriction to G is ergodic. Let β be a trace preserving action of Γ on N. A weak 1-cocycle w = {wg}g∈Γ for β is said to be ergodic on G, if the action βw of G is still ergodic, where βw is defined by βw g = Adwg ◦ βg, g ∈ G. Let β be a Γ-action on N. Suppose that the diagonal action β⊗β on (N⊗N, tr⊗tr) G ⊂ Γ, we get the following: Proposition 6.5. Let {wγ}γ∈Γ ⊂ N be a weak 1-cocycle for β. Let α be a trace has an extension eβ on a finite von Neumann algebra (eN , τ ). The algebra eN is not necessarily identical with N⊗N. When the actioneβ is ergodic on a normal subgroup preserving continuous action of R on eN satisfying the following properties: Suppose that the weak 1-cocycle {wγ ⊗ 1} ⊂ eN is ergodic for the G-action eβG. If the non-zero element ex0 ∈ eN so that (wg ⊗ 1)eβg(ex0)(1 ⊗ w∗ for all t ∈ R, γ ∈ Γ and ex ∈ eN. g) = ex0, g ∈ G. This is proved in the same way for Bernoulli shift actions on the infinite tensor product of abelian von Neumann algebras ([PoSa], Lemma 3.5). Since we are in- terested in actions on the AFD II1 factor, we require the ergodicity assumption on group inclusion G ⊂ Γ has the relative property (T) of Kazhdan, then there exists a • αt ◦eβ(γ)(ex) = eβ(γ) ◦ αt(ex), • α1(x ⊗ 1) = 1 ⊗ x, for all x ∈ N. 20 HIROKI SAKO weak 1-cocycle {wγ ⊗ 1}. For the self-containedness and in order to make it clear where the ergodicity assumption works, we write down a complete proof. Proof. For t ∈ (0, 1], let Kt be the convex weak closure of and ext ∈ Kt be the unique element whose 2-norm is minimum in Kt. Since g ⊗ 1) g ⊗ 1) g ∈ G} ⊂ eN {(wg ⊗ 1)αt(w∗ (wg ⊗ 1)eβg((wg1 ⊗ 1)αt(w∗ = (wgβg(wg1) ⊗ 1)αt(βg(w∗ = (wgg1 ⊗ 1)αt(w∗ gg1 ⊗ 1), g1 ⊗ 1))αt(w∗ g1)w∗ g ⊗ 1) g, g1 ∈ G, (Eq3) lation ∗)(w∗ g ∈ G. g ⊗ 1) = ext, we have (wg ⊗ 1)eβg(Kt)αt(w∗ g ⊗ 1) = Kt, for g ∈ G. By the uniqueness of ext, we get By the assumption, the action (Ad(wg ⊗ 1) ◦eβg)g∈G is ergodic on eN . By the calcu- (wg ⊗ 1)eβg(ext)αt(w∗ (wg ⊗ 1)eβg(extext = (wg ⊗ 1)eβg(ext)αt(w∗ = extext ∗ ∈ C1. The element ext is a scalar multiple of a unitary in eN . we get extext We shall next prove that gx1/n is not zero for some positive integer n. The pair (Γ, G) has the relative property (T) of Kazhdan. By proposition 2.4, we can find a positive number δ and a finite subset F ⊂ Γ satisfying the following condition: If a unitary representation (π,H) of Γ and a unit vector ξ of H satisfy kπ(γ)ξ − ξk ≤ δ (γ ∈ F ), then kπ(g)ξ − ξk ≤ 1/2 (g ∈ G). By the continuity of the action α, there exists n such that g ⊗ 1)αt(wg ⊗ 1)eβg(ext g ∈ G, g ⊗ 1) g ⊗ 1) ∗)(w∗ ∗, The actions β and (αl γ ⊗ 1) − 1ktr,2 ≤ δ, k(wγ ⊗ 1)α1/n(w∗ 1/n)l∈Z on eN give a Γ × Z action on eN . Let P be the crossed product von Neumann algebra P = eN ⋊ (Γ × Z). Let (Uγ)γ∈Γ and W be the implementing unitaries in P for Γ and 1 ∈ Z respectively. We put Vγ = (wγ ⊗ 1)Uγ, γ ∈ Γ. We regard AdV· as a unitary representation of Γ on L2(P ). Since γ ∈ F, kAdVγ(W ) − WkL2(P ) = k(wγ ⊗ 1)W (w∗ = k(wγ ⊗ 1)α1/n(w∗ γ ⊗ 1)W ∗ − 1kL2(P ) γ ⊗ 1) − 1kL2( eN ) ≤ δ, γ ∈ F, we have the following inequality: g ⊗ 1) − 1kL2( eN ), 1/2 ≥ kAdVg(W ) − WkL2(P ) = k(wg ⊗ 1)α1/n(w∗ We get 1/2 ≥ kgx1/n − 1kL2( eN ) and gx1/n 6= 0. Let gu1/n be the unitary of eN given by a scalar multiple of gx1/n. By equation (wg ⊗ 1)eβg(gu1/n)α1/n(w∗ g ⊗ 1) = gu1/n, (Eq3), the unitary satisfies g ∈ G. g ∈ G. CLASSIFICATION AND CENTRALIZERS 21 By direct computations, we have the following desired equality: Let ex0 be the unitary defined by (wg ⊗ 1)eβg(ex0)(1 ⊗ w∗ ex0 = gu1/nα1/n(gu1/n)α2/n(gu1/n) . . . α(n−1)/n(gu1/n). g ⊗ 1) = ex0, g) = (wg ⊗ 1)eβg(ex0)α1(w∗ g ∈ G. (cid:3) Theorem 6.6. Let β = β(H, µ, χ) be a twisted Bernoulli shift action on N(H, µ). Suppose that N(H, µ) is the AFD factor of type II1 and that there exists a continuous R-action (α(0) t )t∈R on Lµ(H) ⊗ Lµ(H) satisfying the following properties: • For any x ∈ Lµ(H), α(0) • The automorphism α(0) t 1 (x ⊗ 1) = 1 ⊗ x, Let β(1) be another action of Z2 ⋊ SL(2, Z) on the AFD factor N (1) of type II1 and suppose that its restriction to Z2 is ergodic. The action β(1) is outer conjugate to β, if and only if β(1) is conjugate to β. commutes with the diagonal action of bH. Proof. We deduce from outer conjugacy to conjugacy in the above situation. Let θ be a ∗-isomorphism from N (1) onto N(H, µ) which gives the outer conjugacy of the action β(1) and β = β(H, µ, χ). There exists a weak 1-cocycle {wγ}γ∈Z2⋊SL(2,Z) for β satisfying θ ◦ β(1)(γ) = Adwγ ◦ β(γ) ◦ θ, γ ∈ Z2 ⋊ SL(2, Z). Since the action β(1) is ergodic on Z2, the weak 1-cocycle w is ergodic on Z2. eρ(γ0)(a ⊗ b) = ρ(γ0)(a) ⊗ ρ(γ0)(b). We use the notations Γ0, G0 given in Section 3. Leteρ be the diagonal action ρ⊗ ρ of Γ0 on the tensor product algebra fM = Leµ(⊕Z2H) ⊗ Leµ(⊕Z2H): The fixed point algebra eN ⊂ fM of the diagonal bH-action contains N(H, µ) ⊗ N(H, µ). Since Z2 ⋊SL(2, Z) = Γ0/bH, the actioneρ gives a Z2 ⋊SL(2, Z)-actioneβ on eN . The actioneβ is the extension of the diagonal action β⊗β on N(H, µ)⊗N(H, µ). We denote by αt the action onfM ∼=NZ2(Lµ(H)⊗Lµ(H)) given by the infinite tensor with the action eρ. It follows that the subalgebra eN is globally invariant under αt. The set of unitary {Wγ = wγ ⊗ 1}γ∈Z2⋊SL(2,Z) ⊂ eN is a weak 1-cocycle for eβ. We shall prove that this weak 1-cocycle is ergodic on Z2. Let a be an element in eN fixed under eβZ2. The element a can be written as a =Pλ∈⊕Z2 H aλ ⊗ u(λ) in L2fM, where aλ ⊗ 1 = EM ⊗C(a(1 ⊗ u(λ))∗). Since a is fixed under the action of Z2, we have product of the R-action α(0) . By the assumption on α(0) , the R-action αt commutes t t a = eβW (k)(a) = Xλ∈⊕Z2 H = Xλ∈⊕Z2 H Adwk ◦ ρ(1, k)(aλ) ⊗Yl∈Z2 Adwk ◦ ρ(1, k)(aλ) ⊗ ρ(1, k)(u(λ)) χ(λ(l))det(k,l)u(k · λ). Since Adwk ◦ ρ(1, k) preserves the 2-norm, we get kaλk2 = kak−1·λk2. Since kak2 2 = 2 < ∞ and the set {ak−1·λ k ∈ Z2} is infinite for λ 6= 0, it turns out that Pkaλk2 aλ = 0 for λ 6= 0 and thus a ∈ eN ∩ (M ⊗ C) = N(H, µ) ⊗ C. By the ergodicity of 22 HIROKI SAKO the Z2-action {Adwk ◦ β(k)}, we get a ∈ C. We conclude that the weak 1-cocycle By the relative property (T) for the inclusion Z2 ⊂ Z2 ⋊ SL(2, Z) and Proposition {Wγ} ⊂ eN is ergodic on Z2. 6.5, there exists a non-zero element ex0 ∈ eN satisfying The element ex0 can be written as the following Fourier expansion: ex0 =X c(λ1, λ2)u(λ1) ⊗ u(λ2) ∈ eN ⊂ Leµ(⊕Z2H) ⊗ Leµ(⊕Z2H). satisfyingPk∈Z2(λ1(k) + λ2(k)) = 0. Choose and fix a pair (λ1, λ2) satisfying Here c(λ1, λ2) is a complex number and (λ1, λ2) ∈ (⊕Z2H)2 runs through all pairs (wk ⊗ 1)eβ(k)(ex0)(1 ⊗ w∗ k) = ex0, k ∈ Z2. λ2(k), c(λ1, λ2) 6= 0. −Xk∈Z2 λ1(k) = h = Xk∈Z2 h ∈ M be the unitary written as v′ Let v′ placed on 0 ∈ Z2. The following unitaries {w′ for β: h = uh ⊗ 1 ⊗ 1 ⊗ · · · , where uh ∈ Lµ(H) is γ} ⊂ N(H, µ) give a weak 1-cocycle w′ (k,γ0) = v′ hw(k,γ0)ρ(1, k, γ0)(v′ h ∗), (k, γ0) ∈ Z2 ⋊ SL(2, Z). Applying the trace preserving conditional expectation E = EN (H,µ)⊗N (H,µ), we get Lettingey = (v′ h ⊗ 1)ex0(1 ⊗ v′ ey = (w′ E(ey) = (w′ = (w′ k ∗), h)∗ ∈ fM , we get k ⊗ 1)eβ(k)(ey)(1 ⊗ w′ k ⊗ 1)E(eβ(k)(ey))(1 ⊗ w′ k ⊗ 1)eβ(k)(E(ey))(1 ⊗ w′ ∗) ∗), k k k ∈ Z2. k ∈ Z2. Since the Fourier coefficient of ex0 at (λ1, λ2) ∈ (⊕Z2H)2 is not zero, that of E(ey) at (λ1 + δh,0, λ2 − δh,0) ∈ Λ(H)2 is also non-zero, where δh,0 ∈ ⊕Z2H is zero on Z2 \ {0} and is h on 0 ∈ Z2. By Proposition 6.2, it follows that the weak 1-cocycle (k,e)}k∈Z2 ⊂ N(H, µ) is a weak 1-coboundary of βZ2. Since the Z2-action βZ2 is {w′ weakly mixing, w′ is a weak 1-coboundary on Z2 ⋊ SL(2, Z), by Proposition 6.3. In other words, there exists v ∈ N(H, µ) satisfying γ = vβ(γ)(v∗) mod T, w′ wγ = v′ h ∗vρ(1, γ)(v∗v′ h) mod T, γ ∈ Z2 ⋊ SL(2, Z). Noting that u = v∗v′ h ∈ M is a normalizer of N(H, µ), we get (Ad(u) ◦ θ) ◦ β(1)(γ) = Ad(u) ◦ Ad(wγ) ◦ β(γ) ◦ θ = Ad(ρ(1, γ)(u)) ◦ β(γ) ◦ θ = ρ(1, γ) ◦ Ad(u) ◦ θ = β(γ) ◦ (Ad(u) ◦ θ), γ ∈ Z2 ⋊ SL(2, Z). Thus we get the conjugacy of two Z2 ⋊ SL(2, Z)-actions β(0) and β. (cid:3) We can always apply Theorem 6.6 if H is finite. CLASSIFICATION AND CENTRALIZERS 23 Corollary 6.7. Let H be a finite abelian group and let β = β(H, µ, χ) be a twisted Bernoulli shift action on N(H, µ). Suppose that N(H, µ) is the AFD factor of type II1. Let β(1) be an action of Z2 ⋊ SL(2, Z) on the AFD factor N (1) of type II1 and suppose that its restriction to Z2 is ergodic. The action β(1) is outer conjugate to β, if and only if β(1) is conjugate to β. Proof. We have only to construct an R-action on Lµ(H) ⊗ Lµ(H) satisfying the properties in Theorem 6.6. Let U be an element of Lµ(H) ⊗ Lµ(H) defined by U = 1 H1/2Xh∈H uh ⊗ u∗ h. We note that µ∗µ(g,·) is a character of H and that it is not identically 1 provided g 6= 0 by Lemma 5.5. The element U is self-adjoint and unitary, since µ(h,−h)u−h ⊗ µ(h,−h)u∗ u∗ h ⊗ uh = −h = U, U ∗ = 1 µ∗µ(g, h)ug+h ⊗ u∗ g+h 1 H1/2Xh∈H H Xg,h∈H HXg∈H Xh∈H 1 1 1 H1/2Xh∈H H Xg,h∈H µ∗µ(g, h − g)! ug ⊗ u∗ gu∗ h = uguh ⊗ u∗ g = 1. U 2 = = The operator U is a fixed point under the action of bH, so the projections P1 = (1 + U)/2 and P−1 = (1 − U)/2 are also fixed points. Thus the R-action α(0) Ad(P1+exp (iπt)P−1) commutes with the bH-action. The automorphism α(0) α(0) 1 (ug ⊗ 1) = U(ug ⊗ 1)U ∗ = (1 ⊗ ug)UU ∗ = 1 ⊗ ug, t = satisfies g ∈ H. 1 This verifies the first condition for α(0). (cid:3) Corollary 6.8. Let Q be a set consisting of odd prime numbers and βQ be the twisted Bernoulli shift action defined in Corollary 5.9. Let β be a Z2 ⋊ SL(2, Z)-action on the AFD factor of type II1 whose restriction to Z2 is ergodic. The actions βQ and β are outer conjugate if and only if they are conjugate. In particular, {βQ} is an uncountable family of Z2 ⋊SL(2, Z)-actions which are mutually non outer conjugate. Proof. We will use the notation given in Corollary 5.8 and 5.9. Let α(q) t be the R- action on Lµq (Hq)⊗ Lµq (Hq) constructed as in the previous corollary. We define the R-action α(Q) on LµQ(HQ) ⊗ LµQ(HQ) by α(Q) t (xq), where xq ∈ Lµq (Hq) ⊗ Lµq (Hq) and xq 6= 1 only for finitely many q. The R-action satisfies the conditions in Theorem 6.6. By Corollary 5.9, {βQ} are mutually non conjugate and their restriction to Z2 is ergodic. Thus they are mutually non outer conjugate. (cid:3) (⊗q∈Qxq) = ⊗q∈Qα(q) t Acknowledgment . The author would like to thank Professor Yasuyuki Kawahigashi for helpful conversations. He thanks the referee for careful reading and numerous detailed comments. He is supported by JSPS Research Fellowships for Young Sci- entists. 24 HIROKI SAKO References [Bu] M. Burger, Kazhdan constants for SL3(Z), J. Reine Angew. Math., 413 (1991), 36 -- 67. [Ch] M. Choda, A continuum of non-conjugate property T actions of SL(n,Z) on the hyperfinite II1-factor, Math. Japon., 30 (1985), 133 -- 150. [Co] A. Connes, Classification of injective factors. Cases II1, II∞, IIIλ, λ 6= 1. Ann. of Math. (2) [Jol] P. Jolissaint, On property (T) for pairs of topological groups. Enseign. Math., (2) 51 (2005), 104 (1976), no. 1, 73 -- 115. no. 1-2, 31E5. [Jon] V. F. R. Jones, A converse to Ocneanu's theorem, J. Operator Theory, 10 (1983), no. 1, 61 -- 63. [Ka] Y. Kawahigashi, Cohomology of actions of discrete groups on factors of type II1, Pac. J. Math., 149 (1991), no. 1, 303 -- 317. [NPS] R. Nicoara, S. Popa, R. Sasyk, On II1 factors arising from 2-cocycles of w-rigid groups. J. Funct. Anal. 242 (2007),no. 1, 230 -- 246. [Oc] A. Ocneanu, Actions of discrete amenable groups on von Neumann algebras, Lecture Notes in Mathematics, 1138. Springer-Verlag, Berlin, 1985. [OPT] D. Olesen, G. K. Pedersen, M. Takesaki, Ergodic actions of compact abelian groups, J. Operator Theory, 3 (1980), no. 2, 237 -- 269. [Po1] S. Popa, On a class of type II1 factors with Betti numbers invariants, Ann. of Math., (2) 163 (2006), no. 3, 809 -- 899. [Po2] S. Popa, Some computations of 1-cohomology groups and construction of non orbit equivalent actions, J. Inst. Math. Jussieu, 5 (2006), 309 -- 332. [Po3] S. Popa, Some rigidity results for non-commutative Bernoulli shifts, J. Funct. Anal., 230 (2006), no. 2, 273 -- 328. [Po4] S. Popa, Cocycle and orbit equivalence superrigidity for malleable actions of w-rigid groups, Invent. Math. 170 (2007), no. 2, 243 -- 295. [PoSa] S. Popa, R. Sasyk, On the cohomology of Bernoulli actions, Erg. Theory Dyn. Sys., 26 (2007), 1 -- 11. [Sh] Y. Shalom, Bounded generation and Kazhdan's Property (T). Publ. Math. IHES, 90 (1999), 145 -- 168. [Vaes] S. Vaes, Rigidity results for Bernoulli actions and their von Neumann algebras (after Sorin Popa), Sminaire Bourbaki, exp. no. 961, Astrisque 311 (2007) 237 -- 294. Department of Mathematical Sciences, University of Tokyo, Komaba, Tokyo, 153-8914, Japan E-mail address: [email protected]
1212.6750	1	1212	2012-12-30T17:26:11	Classification of graph C*-algebras with no more than four primitive ideals	[ "math.OA" ]	We describe the status quo of the classification problem of graph C*-algebras with four primitive ideals or less.	math.OA	math	CLASSIFICATION OF GRAPH C∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ Abstract. We describe the status quo of the classification problem of graph C ∗-algebras with four primitive ideals or less. 1. Introduction The class of graph C∗-algebras (cf. [Rae05] and the references therein) has proven to be an important and interesting venue for classification theory by K-theoretical invariants; in particular with respect to C∗-algebras with finitely many ideals, and in 2009, the authors formulated the following working conjecture: Conjecture 1.1. Graph C∗-algebras C∗(E) with finitely many ideals are classified up to stable isomorphism by their filtered, ordered K-theory FK+ Prim(C ∗(E)) (C∗(E)). Here, the filtered, ordered K-theory is simply the collection of all K0- and K1- groups of subquotients of the C∗-algebra in question, taking into account all the natural transformations among them (details will be given below). The conjecture addresses the possibility of a classification result which is not strong (cf. [Ell10]) in the sense that we do not expect every possible isomorphism at the level of the invariant to lift to the C∗-algebras. The conjecture remains open and we are forthwith optimistic about its veracity, although some of the results which have been obtained, as we shall see, seem to indicate that an added condition of finitely generated K-theory could be needed. In the present paper we will discuss the status of this conjecture for graph algebras with four or fewer primitive ideals; if the number is three or fewer we can present a complete classification under the condition of finitely generated K-theory, but for the number four there are many cases still eluding our methods. Adding, in some cases, the condition of finitely generated K-theory -- or even stronger, that the graph algebra is unital -- we may solve 103 of the 125 cases, leaving less than one fifth of the cases open. Our main contribution in the present paper concerns the class of fan spaces which has not been accessible through the methods we have used earlier, but we will also go through those results in our two papers [ERRa] and [ERR09] which apply here. 1.1. Tempered primitive ideal spaces. Invoking an idea from [ERS11] we orga- nize our overview using a tempered ideal space of the C∗-algebra in question. This is defined for any C∗-algebra with only finitely many ideals as the pair (Prim(A), τ ) Date: September 5, 2018. 1 2 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ where τ : Prim(A) → {0, 1} is defined as τ (I) =(0 K0(I/I0)+ 6= K0(I/I0) 1 K0(I/I0)+ = K0(I/I0) with I0 the maximal proper ideal of I (this exists by the fact that I is prime and contains only finitely many ideals). We set X(cid:3) = {x ∈ X τ (x) = 0} X(cid:4) = {x ∈ X τ (x) = 1} To be able to work systematically with these objects, we now give them a combi- natorial description. Definition 1.2. Let A be a C∗-algebra. We let Prim(A) denote the primitive ideal space of A, equipped with the usual hull-kernel topology, also called the Jacobson topology. We always identify the open sets of Prim(A), O(Prim(A)), and the lattice of ideals of A, I(A), using the lattice isomorphism U 7→ \p∈Prim(A)\U p. When U is an open set we write A(U ) for the corresponding ideal of A. When U ⊃ V are both open, so that U \ V is locally closed, we write A(U \ V ) for the subquotient A(U )/A(V ). Note that whenever X(cid:3) or X(cid:4) are locally closed, standard results in graph C∗- algebra theory give that A(X(cid:3)) and A(X(cid:4)) are AF algebras and O∞-absorbing algebras, respectively. Definition 1.3. Let X be a topological space. The specialisation preorder ≺ on X is defined by x ≺ y if and only if x ∈ {y}. A topological space satisfies the T0 separation axiom if and only if its speciali- sation preorder is a partial order. Definition 1.4. A subset H of a preordered set (X,≤) is called hereditary if x ≤ y ∈ H implies x ∈ H. Definition 1.5. Let (X,≤) be a preordered set. The Alexandrov topology of X is the topology with the closed sets being the hereditary sets. A topological set is called an Alexandrov space if it carries the Alexandrov topol- ogy of some preordered set. The preorder is necessarily the specialisation preorder. A topological space is an Alexandrov space if and only if arbitrary intersections of open sets are open. Since we are dealing with C∗-algebras with finite primitive ideal spaces, these are all Alexandrov spaces satisfying the T0 separation axiom. Consequently, we can equivalently consider all partial orders on finite sets. The tempered primitive ideal space for a C∗-algebra with n primitive ideals may hence be uniquely described using a partial order on {1, . . . , n} and a map in {0, 1}{1,...,n}. The transitive reduction of a relation R on a set X is a minimal relation S on X having the same transitive closure as R. In general neither existence nor uniqueness are guaranteed, but if the transitive closure of R is antisymmetric and finite, there is a unique transitive reduction. We will illustrate our (finite) topological spaces with graphs of the transitive reduction of the specialisation order, where we write t1 a1,2 t2 A =  a2,3 . . . a1,n−1 . . . tn−1 a1,n a2,n an−1,n tn  GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 3 an arrow x → y if and only if x is less than y in the transitive reduction of the specialisation order (similar to the Hasse diagram). The value of τ will be indicated by colors of the vertices of the graph; white for 0 and black for 1. We obtain a unique signature for each tempered ideal space as follows. Consider the adjacency matrix of the graph of the specialisation order and recall that (by transitivity and antisymmetry) we can always permute the vertices so that the ad- jacency matrix becomes an upper triangular matrix. Since the relation is reflexive, we will have ones in the diagonal, so without loss of information we may write the values of τ there. To each such upper triangular matrix we associate two binary numbers and a = a1,2a1,3 ··· a1,na2,3a2,4 ··· a2,n ··· an−1,n t = t1 ··· tn In general, there are several such binary numbers associated with a specialisation order by means of permuting the vertices. We choose the order of the vertices to obtain the smallest possible pair (a, t) ordered lexicographically as the unique identifier for this specific tempered ideal structure. In the interest of conserving space we write hexadecimal expansion of the numbers when referring to a certain structure. We write n.a.t and n.a to indicate signatures and tempered signatures, respectively, defined this way (where n and a are numbers written in decimal ex- pansions and t is a number written in hexadecimal expansion). If a primitive ideal space is disconnected, we may classify the C∗-algebras asso- ciated to each component individually. We will hence assume throughout that the C∗-algebras have connected primitive ideal space (when considering graph algebras, a necessary, but not sufficient, condition for this is that the underlying graphs are connected considered as undirected graphs). Determining the number of connected T0-spaces with n points is hard for most n; the number has been computed up to n = 16 in [BM02]. But for small n even the number of tempered ideal spaces can readily be found by naive enumeration, by first counting all spaces and then performing inverse Euler transform to obtain those that are connected: Prim(A) Number of spaces Number of connected spaces Number of tempered spaces Number of connected tempered spaces 3 5 3 4 16 10 1 1 1 2 10 62 510 5292 69364 2 20 125 1058 11549 5 63 44 2 2 1 4 6 318 238 We will restrict our attention to Prim(A) ≤ 4 and hence have 15 (connected) primitive ideal spaces1 which may be given temperatures in a total of 151 different ways to concern ourselves with: 1The space 4.E was forgotten on page 230 of [MNb] 4 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ 1.0 2.1 3.7 4.E 4.F 4.39 4.3F [A],[F] [A],[F] [F] [F] [L],[A] [L],[A] [L],[A] [A] [A] [A] 3.3 3.6 4.A [L],[A] 4.38 4.1F 4.3E 4.1E 4.3B [Y] [Y] [O] [O] where ❏ just indicates that it is either (cid:3) or (cid:4). We call a finite T0 space linear ([L]) if its partial order is total. Following [BK] we call it an accordion space ([A]) if the symmetrization of the space is the symmetrization of a linear space. We call it a fan space ([F]) when there is a smallest or largest element in the preorder, so that when this is removed, what remains is a disjoint union of linear spaces. The remaining spaces we organize as [Y]-spaces and [O]-spaces as indicated. In Section 6 below we summarize our results subject to this organization. 1.2. The invariant. Let A be a C∗-algebra with finitely many ideals and set X = Prim(A). Note that for any locally closed subset Y = U \ V of X, we have two groups K0(A(Y )) and K1(A(Y )). Moreover, for any three open subsets U ⊆ V ⊆ W of X, we have a six term exact sequence K0(A(Y1)) ι0 / K0(A(Y2)) π0 / K0(A(Y3)) ∂1 ∂0 K1(A(Y3)) π1 K1(A(Y2)) ι1 K1(A(Y1)) where Y1 = V \ U , Y2 = W \ U , and Y3 = W \ V . The filtered, ordered K- theoryFK+ X (A) of A is the collection of all K-groups thus occurring, equipped with order on K0 and the natural transformations {ι∗, π∗, ∂∗}. locally closed subset Y of X, there exist group isomorphisms Consequently, if also Prim(B) = X, we write FK+ X (A) ∼= FK+ X (B) if for each αY ∗ : K∗(A(Y )) → K∗(B(Y )) ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ ❏ / / O O o o o o GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 5 preserving all natural transformations in such a way that all αY 0 are also order isomorphisms. All components of this invariant are readily computable ([CET]), and often, much of it is redundant. We will not pursue that issue here. The filtered K-theoryFKX(A) of A is defined analogously by disregarding the order structure on K0. The filtered (ordered) K-theory over a finite T0-space X can also be used for C∗-algebras over X without being tight.2 1.3. Graph C∗-algebra. A graph (E0, E1, r, s) consists of a countable set E0 of vertices, a countable set E1 of edges, and maps r : E1 → E0 and s : E1 → E0 identifying the range and source of each edge. If E is a graph, the graph C∗-algebra C∗(E) is the universal C∗-algebra generated by mutually orthogonal projections {pv : v ∈ E0} and partial isometries {se : e ∈ E1} with mutually orthogonal ranges satisfying (1) s∗ (2) ses∗ ese = pr(e) e ≤ ps(e) for all e ∈ E1 for all e ∈ E1 (3) pv =P{e∈E 1:s(e)=v} ses∗ e for all v with 0 < s−1(v) < ∞. The countability hypothesis ensures that all our graph C∗-algebras are separable, which is a necessary hypothesis for many of the classification results. We will be mainly interested in graph C∗-algebras with real rank zero. For a graph E, we have that the real rank of C∗(E) is zero if and only if E is satisfying Condition (K), i.e., no vertex of E is the base point of exactly one simple cycle (see Theorem 3.5 of [JJ]). Moreover, by Proposition 3.3 of [JJ], every graph C∗-algebra with finitely many ideals has real rank zero. Thus, every graph C∗-algebra with finitely many ideals has a norm-full projection, and by [Bro77], every graph C∗-algebra with finitely many ideals is stably isomorphic to a unital C∗-algebra. Throughout the paper we will use the following facts about graph C∗-algebras without further mention. Theorem 1.6. Let C∗(E) be a unital graph C∗-algebra satisfying Condition (K). (1) Every ideal of C∗(E) is stably isomorphic to a unital graph C∗-algebra. (2) Every sub-quotient of C∗(E) is stably isomorphic to a unital graph C∗-algebra. (3) The K-groups of every sub-quotient of C∗(E) is finitely generated. (4) Every non-unital simple sub-quotient of C∗(E) that is an AF-algebra is iso- morphic to K. Proof. As in the proof of Theorem 5.7 (4) of [MT07] (see also [BHRS, Proposi- tion 3.4]), every ideal of a graph C∗-algebra satisfying Condition (K) is Morita equivalent to C∗(F ), where F 0 ⊆ E0. Hence, (1) holds since a graph C∗-algebra C∗(E) is unital if and only if E0 is finite. (2) follows from (1) and [BHRS, Corol- lary 3.5]. (3) follows from (2) and [DT02, Theorem 3.1]. Suppose C∗(F ) is a simple unital AF-algebra. Then F has no cycles. Since C∗(F ) is unital, F 0 is finite. Therefore, F has a sink. By [DT05, Corollary 2.15], every singular vertex must be reached by any other vertex since C∗(F ) is simple. Thus, F must be a finite graph. Hence, C∗(F ) ∼= Mn. From this observation, (4) follows from (1) and (2) since any non-unital simple C∗-algebra stably isomorphic to K is isomorphic to K. (cid:3) 2Although this is not exactly the same definition as the filtrated K-theory in [MNa], it is known to be the same for all the cases where we have a UCT. For more on this invariant and C ∗-algebras over X the reader is referred to [MNa] and the references therein. 6 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ See [Rae05] and the references therein for more on graph C∗-algebras. 2. General theory We first describe the situations in which the graph algebras can be classified using widely applicable results. 2.1. The AF case. The AF case corresponds to temperatures that are constantly 0. We incur these at the tempered signatures 1.0.0, 2.1.0, 3.3.0, 3.6.0, 3.7.0, 4.A.0, 4.E.0, 4.F.0, 4.1E.0, 4.1F.0, 4.38.0, 4.39.0, 4.3B.0, 4.3E.0, and 4.3F.0. Of course the classification question is resolved by Elliott's theorem: Theorem 2.1 ([Ell76]). AF algebras are classified up to stable isomorphism by their ordered K0-group. 2.2. The purely infinite case. Recall that there are three notions of pure in- finiteness for non-simple C∗-algebras, namely pure infiniteness, strong pure infinite- ness, and O∞-absorption, introduced by E. Kirchberg and M. Rørdam; cf. [KR00] and [KR02]. Corollary 2.2. For each nuclear, separable C∗-algebra A with finite primitive ideal space, the following are equivalent: (a) A is purely infinite, (b) A is strongly purely infinite, (c) A is O∞-absorbing, i.e., A ⊗ O∞ ∼= A. Proof. It follows from Theorem 9.1 and Corollary 9.2 of [KR02] that (c) implies (b), that (b) implies (a), and that the three coincide in the simple case. It follows from Proposition 3.5 of [KR02], that pure infiniteness passes to ideals and subquotients. Thus it follows from [TW07] that (a) implies (c). (cid:3) The purely infinite case (the O∞-absorbing case) corresponds to temperatures that are constantly 1. We incur these at the tempered signatures 1.0.1, 2.1.3, 3.3.7, 3.6.7, 3.7.7, 4.A.F, 4.E.F, 4.F.F, 4.1E.F, 4.1F.F, 4.38.F, 4.39.F, 4.3B.F, 4.3E.F, and 4.3F.F. As we will outline below, all but the case 4.1E.F are resolved through the recent work of many hands. The isomorphism result of Kirchberg (cf. [Kir94] and [Kir00]) reduces the classi- fication problem of nuclear and strongly purely infinite C∗-algebras which are also in the bootstrap class to an isomorphism problem in ideal-related KK -theory. Since all purely infinite graph C∗-algebras fall in this class we may hence confirm Con- jecture 1.1 in the purely infinite case by providing a universal coefficient theorem which allows the lifting of isomorphisms at the level of filtered K-theory to invert- ible KK X -classes. This, however, is not known to be possible in general. Indeed, Meyer and Nest in [MNa] showed that there are purely infinite C∗-algebras over the space 4.A which fails to have this property, but since the examples provided there cannot possibly come from graph algebras, the question remains open in general. The work of Bentmann and Kohler established that general UCTs are available precisely when the space X is an accordion space, and Arklint with the second and third named authors provided UCTs for other spaces, including 4.A, under the added assumption that the C∗-algebra has real rank zero which is automatic here. Specializing even further, Arklint, Bentmann and Katsura provided a UCT which applies for our space 4.3B under the added assumption that the C∗-algebra has GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 7 real rank zero and that the K1 groups of all subquotients are free, which also is automatic here. The space 4.1E remains open. In conclusion: Theorem 2.3. Purely infinite, separable, nuclear C∗-algebras A with finite prim- itive ideal space X in the bootstrap class of Meyer and Nest (i.e., all simple sub- quotients are in the bootstrap class of Rosenberg and Schochet) are classified up to stable isomorphism by their filtered K-theory FKX (−) in the cases (i) X is an accordion space [1.0, 2.1, 3.3, 3.6, 3.7, 4.E, 4.F, 4.39, 4.3F] ([Kir94], [NCP00], [Ror97], [Res08], [MNa], [BK], [Kir00]) (ii) X is one of the spaces 4.A, 4.38, 4.1F, 4.3E and rr(A) = 0. ([ARR12]) (iii) X is the space 4.3B, rr(A) = 0, and K1(J/I) is free for any I ⊳ J E A ([ABK]) 2.3. The separated case. The classification problem for the two mixed cases with Prim(A) = 2 not covered by the results mentioned above -- the tempered signatures 2.1.1 and 2.1.2 -- were resolved in [ET10] drawing heavily on [ERR09]. In [ERRa], we generalized this to more complicated cases having the separation property which is automatic in the two-point case, as detailed below. The idea is to find an ideal I such that I is AF and A/I is O∞-absorbing, or vice versa. We do not know in general how to prove classification in this case, but under certain added assumptions related to the notion of fullness, this leads to results that may be used to resolve the cases of tempered signature 3.7.1, 3.7.3, 4.F.1, 4.1F.1, 4.1F.3, 4.3B.1, 4.3F.1, 4.3F.3, 4.3F.7 by Proposition 2.5 below and 3.7.4, 3.7.6, 4.39.8, 4.3B.8, 4.3E.8, 4.3E.C, 4.3F.8, 4.3F.C, 4.3F.E by Proposition 2.6. Definition 2.4. Let n > 1 be a given integer. Then we let Xn denote the partially ordered set (actually totally ordered) Xn = {1, 2, . . . , n} with the usual order. For a, b ∈ Xn with a ≤ b, we let [a, b] denote the set {x ∈ Xn : a ≤ x ≤ b}. Proposition 2.5. Let A1 and A2 be separable, nuclear, C∗-algebras over Xn in the bootstrap class of Meyer and Nest (i.e., every simple subquotient is in the bootstrap class of Rosenberg and Schochet). Suppose Ai({1}) is an AF algebra and Ai([2, n]) is a tight stable O∞-absorbing C∗-algebra over [2, n], and Ai({2}) is an essential ideal of Ai([1, 2]). Then A1⊗ K ∼= A2⊗ K if and only if there exists an isomorphism α : FKXn (A1) → FKXn (A2) such that α{1} is positive. Proposition 2.6. Let A1 and A2 be graph C∗-algebras satisfying Condition (K). Suppose Ai is a C∗-algebra over Xn such that Ai({n}) is an AF algebra, for every ideal I of Ai we have that I ⊆ Ai({n}) or Ai({n}) ⊆ I, and Ai([1, n− 1]) is a tight, O∞-absorbing C∗-algebra over [1, n− 1]. Then A1⊗ K ∼= A2⊗ K if and only if there exists an isomorphism α : FKXn (A1) → FKXn (A2) such that α{n} is positive. 3. Fan spaces In this section, we develop methods to deal mainly with the spaces 3.3, 3.6, 4.A, 4.38. We observe the following in [ERRa] Lemma 3.1. Let E be a graph such that C∗(E) has finitely many ideals and assume that I ⊳ J E C∗(E) are ideals. Then (i) C∗(E) ⊗ K has the corona factorization property (ii) (J/I) ⊗ K is of the form C∗(F ) ⊗ K for some graph F (iii) (J/I) ⊗ K has the corona factorization property 8 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ The graph F above can be chosen as a subgraph of the Drinen-Tomforde desin- gularization of E ([DT05]). Definition 3.2. For each C∗-algebra A, we let M(A) and Q(A) denote the multi- plier algebra and the corona algebra of A, respectively. For each extension e : 0 → B → E → A → 0, we let ηe : A → Q(B) denote the Busby map of the extension. Moreover, for each surjective (or, more generally, proper) ∗-homomorphism ϕ : extension to the multiplier algebras and the induced ∗-homomorphism between the corona algebras, respectively (cf. §2.1 of [ELP99]). Lemma 3.3. Let (Bi)i∈I be a family of C∗-algebras (small enough for direct sums A → B, we let eϕ : M(A) → M(B) and ϕ : Q(A) → Q(B) denote the unique and products to exist). Let πj : Li∈I Bi → Bj denote the canonical projection, for each j ∈ I. Then there is a canonical isomorphism Qi∈I eπi : M(Li∈I Bi) → Qi∈I M(Bi) which has the unique extension eπj : M(Li∈I Bi) → M(Bj) of πj as Consequently, if I is finite, there is an induced isomorphismQi∈I πi : Q(Li∈I Bi) → Qi∈I Q(Bi), and it induces homomorphisms πj : Q(Li∈I Bi) → Q(Bj) as the j'th (cf. pp. 39 and 81-82 in [Mur90]). Let (ρ1, ρ2) be a double centralizer onLi∈I Bi (i.e., an arbitrary element of M(Li∈I Bi)). Using an approximate unit, it is easy to ∗-homomorphism from M(Li∈I Bi) to M(Bj ). By the universal property of the direct product, we get a ∗-homomorphism ϕ from M(Li∈I Bi) to Qi∈I M(Bi), bras, and hence it is the extension eπj of πj . Clearly, ϕ is injective. It is also easy see that ρ1 and ρ2 restricted to Bj map into Bj itself. In this way we get a canonical to show that ϕ is surjective by constructing the preimage. Therefore, if I is finite, the direct product of the short exact sequences where the j'th coordinate map clearly is an extension of πj to the multiplier alge- the j'th coordinate map. coordinate map. In this case, the direct product coincides with the direct sum. Proof. Here we view the multiplier algebras as the algebras of double centralisers 0 / Bj / M(Bj ) / Q(Bj ) / 0 is canonically isomorphic to 0 /Li∈I Bi / M(Li∈I Bi) / Q(Li∈I Bi) / 0 (cid:3) 3.1. Primitive ideal space with n maximal elements. Assumption 3.4. For this subsection, let n > 1 be a fixed integer, and let Xi = Xli for i = 1, 2, . . . , n, where l1, l2, . . . , ln are fixed positive integers. Let, moreover, X = {m} ⊔ X1 ⊔ X2 ⊔ ··· ⊔ Xn and define a partial order on X as follows. The element m is the least element of X, and for each i = 1, 2, . . . , n, if x, y ∈ Xi then x ≤ y in X if and only if x ≤ y in Xi. No other relations exist between the elements of X. / / / / / / / / GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 9 Lemma 3.5. Let A be a tight C∗-algebra over X and let k ∈ {1, 2, . . . , n} be given. Consider the extensions and e : 0 → A(X \ {m}) → A → A({m}) → 0 e · πk : 0 → A(Xk) → A(Xk ∪ {m}) → A({m}) → 0, where πk : A(X \ {m}) → A(Xk) is the canonical quotient ∗-homomorphism. Then ηe·πk = πk ◦ ηe, and πk ◦ ηe is injective. Proof. Note that the diagram e : 0 / A(X \ {m}) A A({m}) e · πk : 0 πk / A(Xk) / A(Xk ∪ {m}) / A({m}) / 0 / 0 is commutative. Since πk is surjective, by Theorem 2.2 of [ELP99], πk ◦ ηe = ηe·πk. Also note, that Corollary 4.3 of [ELP99] justifies the notation e· πk. Suppose πk ◦ ηe is not injective, then πk ◦ ηe = 0 since A({m}) is a simple C∗-algebra. Hence, A(Xk ∪ {m}) ∼= A(Xk) ⊕ A({m}). Since A(Xk ∪ {m}) ∼= A/A(X \ (Xk ∪ {m})), then there exist proper ideals I and J of A such that I + J = A and I ∩ J = A(X \ (Xk ∪ {m})). But this contradicts the fact that A is a tight C∗-algebra over X. Hence, πk ◦ ηe is injective. Lemma 3.6. Let A be a tight C∗-algebra over X. Then (cid:3) e : 0 → A(X \ {m}) → A → A({m}) → 0 is full if and only if e · πk is full for all k = 1, 2, . . . , n. Proof. By Lemma 3.5, ηe·πk = πk ◦ ηe. Thus, if e is a full extension, then e · πk is a full extension since πk is surjective. Suppose e · πk is a full extension for all j=1 A(Xj) and thus from Lemma 3.3 i=1 πi) ◦ ηe is exactly πj ◦ ηe = ηe·πj i=1 πi is an isomorphism and since e · πk is a full extension for all k = 1, 2, . . . , n, we have that e is a full extension. That this direct sum of full extensions is again full can easily be shown by first cutting down to each coordinate. (cid:3) k = 1, 2, . . . , n. Note that A(X \ {m}) is Ln it follows that the j'th coordinate map of (Ln (according to Lemma 3.5). Since Ln The signatures 3.6.1, 3.6.5, 4.39.1, 4.39.3, 4.39.4, 4.39.5, 4.39.7, 4.38.1, 4.38.3, 4.38.7 are covered by the following theorem. X (A) → FK+ Theorem 3.7. Let A and B be graph C∗-algebras that are tight C∗-algebras over X. Assume that there exists an isomorphism α : FK+ X (B). Assume, moreover, that A({m}) is an AF algebra and that X(cid:3) is hereditary. Then A⊗ K ∼= B ⊗ K. Proof. We may assume that A and B are stable C∗-algebras. Note that for each x ∈ X, A({x}) is an AF algebra if and only if since B({x}) is an AF algebra, and A({x}) is O∞-absorbing if and only if B({x}) is O∞-absorbing (since there exists a positive isomorphism from K0(A({x})) to K0(B({x}))). Specifically, B({m}) is an AF algebra. First we assume that X(cid:4) 6= ∅ and X(cid:3) \ {m} 6= ∅. Note that A(X(cid:3)) and B(X(cid:3)) are AF algebras. Since αX(cid:3) : K0(A(X(cid:3))) → K0(B(X(cid:3))) is a positive isomorphism, there exists an isomorphism β : A(X(cid:3)) → / / / / / / / / / / 10 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ B(X(cid:3)) such that K0(β) = αX(cid:3) (by Elliott's classification result [Ell76]). Since A(X(cid:3)) and B(X(cid:3)) are AF algebras and β is an X(cid:3)-equivariant isomorphism, we have that K0(βY ) = αY for all Y ∈ LC(X) such that Y ⊆ X(cid:3). In particular, K0(β{x}) = α{x} for all x ∈ X(cid:3). (cid:4) be the set of minimal elements of X(cid:4), and for each a, b ∈ X let Let X min [a,∞) = {x ∈ X : a ≤ x} , [a, b) = {x ∈ X : a ≤ x < b} . for all x ∈ X min (cid:4) , and ψx ◦ ηeA·πix = ηeB·πix ◦ β{m}, βXj ◦ ηeA·πj = ηeB·πj ◦ β{m}, and (cid:4) be given. Let ix ∈ {1, 2, . . . , n} be the unique number such that Let x ∈ X min x ∈ Xix. Note that Xix ⊔{m} = [m, x)∪ [x,∞), which we will denote by eXix. Let, moreover, eA x : 0 → A([x,∞)) → A(eXix) → A([m, x)) → 0. : 0 → B([x,∞)) → B(eXix ) → B([m, x)) → 0. eB x : FK+ X (A) → FK+ X (B) is an isomorphism, we also have an isomorphism Since α : FK+ α eXix Kirchberg [Kir00], and Theorem 3.3 of [ERRa], there exists an isomorphism ϕx : A([x,∞)) → B([x,∞)) such that K∗(ϕx) = α[x,∞), and eXix(cid:16)B(eXix )(cid:17). So by Theorem 4.14 of [MNa], eXix(cid:16)A(eXix)(cid:17) → FK+ x ◦ β[m,x)(cid:3) =(cid:2)ϕx ◦ ηeA (cid:2)ηeB x(cid:3) in KK 1(A([m, x)), B([x,∞))), since KK (β[m,x)) is the unique lifting of α[m,x). As in the proof of Proposition 6.3 of [ERRa], Corollary 5.3 of [ERRa] implies are full extensions, and thus also the extensions with Busby that ηeA maps ηeB are full. Since the extensions are non-unital and B([x,∞)) satisfies the corona factorization property, there exists a unitary ux ∈ M(B([x,∞))) such that and ηeB x ◦ β[m,x) and ϕx ◦ ηeA x x x x ◦ β[m,x) = Ad(ux) ◦ ϕx ◦ ηeA ηeB x where ux is the image of ux in the corona algebra (this follows from [EK01] and [KN06]). Hence, by Theorem 2.2 of [ELP99], there exists an isomorphism ηx : x to A(eXix) → B(eXix ) such that (Ad(ux)◦ ϕx, ηx, β[m,x)) is an isomorphism from eA eB x . Let and eA : 0 → A(X \ {m}) → A → A({m}) → 0, eB : 0 → B(X \ {m}) → B → B({m}) → 0. Since A(eXix ) and B(eXix ) have linear ideal lattices, this induces an isomorphism / A(Xix) / 0, 0 / A({m}) eA · πix : ψx eB · πix : 0 / B(Xix ) So now by construction, β{m} / B({m}) / 0. / A(eXix) / B(eXix ) / / / / / / / / GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 11 for all j = 1, 2, . . . , n satisfying that A(Xj) is an AF algebra. Now we define an isomorphism θ from A(X \ {m}) to B(X \ {m}) as the direct sum of the ψx's and βXj 's. We get that (from Lemma 3.3 and Lemma 3.5) ηeA·πj = θ ◦ ηeA = θ ◦ nMj=1 ηeB·πj ◦ β{m} = ηeB ◦ β{m}, ηeB·πj ◦ β{m} = nMj=1 θj ◦ ηeA·πj = nMj=1 nMj=1 (cid:3) where the θj's denote the corresponding ψx's and βXj 's. Hence, by Theorem 2.2 of [ELP99], A ∼= B. If X(cid:4) = ∅ the result is due to Elliott's classification result [Ell76], and if X(cid:3) = {m} the theorem follows easily by making modifications to the above proof. Remark 3.8. Let A and B be graph C∗-algebras that are C∗-algebras over X, so that A(Xi) and B(Xi) are tight C∗-algebras over Xi, for i = 1, 2, . . . , n. Assume that 0 → A(Xi)/A(Xi \ {xi}) → A(Xi ∪ {m})/A(Xi \ {xi}) → A(Xi ∪ {m})/A(Xi) → 0 is essential whenever A(Xi) is O∞-absorbing, where xi is the greatest element of Xi. Assume that there exists an isomorphism α : FK+ X (B). Assume moreover, that A({m}) is an AF algebra and that the set of x ∈ X for which A({x}) is an AF algebra is hereditary. Then A ⊗ K ∼= B ⊗ K. This follows from the proof above. The above extensions are essential, e.g., if A({xi}) is the least ideal of A({xi, m}), for all i = 1, 2, . . . , n, and the remark applies to the cases3 (a) 4.E.1, where we view the algebra A that is tight over the space 4.E as a C∗-al- X (A) → FK+ (b) 4.1E.1 and 4.1E.3, where we view the algebra A that is tight over the space gebra over a ← b → c as indicated by the assignment b → a ← b → c. 4.1E as a C∗-algebra over a ← b → c as indicated by the assignment C∗-algebra over a ← b → c as indicated by the assignment The following proposition follows from the results in [ET10]. (c) 4.3E.1, where we view the algebra A that is tight over the space 4.3E as a Proposition 3.9. Let A be a graph C∗-algebra with exactly one nontrivial ideal I. If A is not an AF algebra, then 0 → I ⊗ K → A ⊗ K → A/I ⊗ K → 0 is a full extension. Using the UCT for accordion spaces (see [MNa] and [BK]) and for many other four-point spaces under the added assumption of real rank zero as described in [ARR12], the cases 3.6.2, 3.6.3, 4.38.8, 4.38.9, 4.38.B, can be classified using the following theorem. Theorem 3.10. Let A and B be graph C∗-algebras that are tight C∗-algebras over X, with Xi being a singleton, for each i = 1, 2, . . . , n. Suppose there ex- ists an isomorphism α : FK+ X (A) → FK+ X (B) which lifts to an invertible element in KK (X; A, B). Then A ⊗ K ∼= B ⊗ K. 3Here we specify how we view the algebras as algebras over a ← b → c by providing a continuous map from the primitive ideal space to {a, b, c} 12 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ Proof. If A({m}) is an AF algebra, the result follows from Theorem 3.7. Suppose A({m}) is an O∞-absorbing simple C∗-algebra and that A and B are stable C∗-al- gebras. Then by Lemma 3.5 and Proposition 3.9, πi ◦ ηeA : A({m}) → Q(A(Xi)) and πi ◦ ηeB : B({m}) → Q(B(Xi)) are full extensions, for all i = 1, 2, . . . , n. Hence, by Lemma 3.6, ηeA and ηeB are full extensions. The theorem now follows from the results of [ERRa]. (cid:3) 3.2. Primitive ideal space with n minimal elements. Assumption 3.11. For this subsection, let n > 1 be a fixed integer, and let Xi = Xli for i = 1, 2, . . . , n, where l1, l2, . . . , ln are fixed positive integers. Let, moreover, X = {M} ⊔ X1 ⊔ X2 ⊔ ··· ⊔ Xn and define a partial order on X as follows. The element M is the greatest element of X, and for each i = 1, 2, . . . , n, if x, y ∈ Xi then x ≤ y in X if and only if x ≤ y in Xi. No other relations are between the elements of X. Lemma 3.12. Let A be a tight C∗-algebra over X and let Y ∈ O(X(cid:3) \ {M}) be given. Consider the extensions and e : 0 → A({M}) → A → A(X \ {M}) → 0 ιA,Y · e : 0 → A({M}) → A(Y ∪ {M}) → A(Y ) → 0 where ιA,Y : A(Y ) → A(X \{M}) is the usual embedding. Then ηιA,Y ·e = ηe ◦ ιA,Y . Proof. Note that the diagram 0 / A({M}) / A(Y ∪ {M}) A(Y ) 0 ιA,Y 0 / A / A({M}) / A(X \ {M}) commutes. Hence, by Theorem 2.2 of [ELP99], ηιA,Y ·e = ηe ◦ ιA,Y . Lemma 3.13. Suppose the following diagram of C∗-algebras with short exact rows is commutative / 0 (cid:3) 0 0 / B / B ι1 ι2 / E1 π1 A1 ϕ1 ϕ2 / E2 π2 / A2 0 / 0. (1) If ϕ2(A1) is a hereditary sub-C∗-algebra of A2, then ϕ1(E1) is a hereditary sub-C∗-algebra of E2. (2) If ϕ2(A1) is full in A2, then ϕ1(E1) is full in E2. Proof. We first prove (1). Let x ∈ E1 and y ∈ E2 such that 0 ≤ y ≤ ϕ1(x). Since ϕ2(A1) is a hereditary sub-C∗-algebra of A2, we have that there exists z ∈ ϕ1(E1) such that π2(y) = π2(z). Thus, y − z ∈ B. Since the map on the ideals is the identity, we have that y − z ∈ ϕ1(E1). Hence, y ∈ ϕ1(E1). Therefore, ϕ1(E1) is a hereditary sub-C∗-algebra of E2. We now prove (2). Let x ∈ E2. Since ϕ2(A1) is full in A2, there exists y in the ideal of E2 generated by ϕ1(E1) such that x− y ∈ B. Since the map on the ideals is / / / / / / / / / / / / / / / / / / / / GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 13 the identity, we have that y − z ∈ ϕ1(E1). Hence, x is in the ideal of E2 generated by ϕ1(E1). (cid:3) full. k=1 Ak → 0 be an extension and let k=1 Ak be the inclusion. Suppose ηe ◦ ιk is full for each k. Then ηe is Lemma 3.14. Let e : 0 → I → A → Ln ιk : Ak →Ln Proof. Let (a1, a2, . . . , an) be a nonzero positive element inLn k=1 Ak. Without loss of generality, we may assume that a1 6= 0. Note that ideal in Q(I) generated by ηe(a1, . . . , an) contains the ideal in Q(I) generated by ηe ◦ ι1(a1). Since ηe ◦ ιk is full, we have that the ideal in Q(I) generated by ηe◦ ι1(a1) is Q(I). Thus, the ideal in Q(I) generated by ηe(a1, . . . , an) is Q(I). (cid:3) The following result applies to the cases 3.3.1, 3.3.5, 4.F.6, 4.F.8, 4.F.E, 4.A.2, 4.A.6, 4.A.E. X (A) → FK+ Theorem 3.15. Let A and B be graph C∗-algebras that are tight C∗-algebras over X such that each of A(Xi), B(Xi) are either AF algebras or O∞-absorbing. Sup- pose there exists an isomorphism α : FK+ X (B) and A({M}) is an AF algebra. Then A ⊗ K ∼= B ⊗ K. Proof. We may assume that A and B are stable C∗-algebras. Note that for each x ∈ X, A({x}) is an AF algebra if and only if B({x}) is an AF algebra, and A({x}) is O∞-absorbing if and only if B({x}) is O∞-absorbing (since there exists a positive isomorphism from K0(A({x}) to K0(B({x})). Specifically, B({M}) is an AF algebra. First we assume that X(cid:4) 6= ∅ and X(cid:3) \ {M} 6= ∅. Note that A(X(cid:3)) and B(X(cid:3)) are AF algebras. Since αX(cid:3) : K0(A(X(cid:3))) → K0(B(X(cid:3))) is a positive isomorphism, there exists an isomorphism β : A(X(cid:3)) → B(X(cid:3)) such that K0(β) = αX(cid:3) (by Elliott's classification result [Ell76]). Since A(X(cid:3)) and B(X(cid:3)) are AF algebras and β is an X(cid:3)-equivariant isomorphism, we have that K0(βY ) = αY for all Y ∈ LC(X) such that Y ⊆ X(cid:3). In particular, K0(β{x}) = α{x} for all x ∈ X(cid:3). Let and eA : 0 → A({M}) → A → A(X \ {M}) → 0, eB : 0 → B({M}) → B → B(X \ {M}) → 0. Since β is an X(cid:3)-equivariant isomorphism, by Lemma 3.12 above and Theorem 2.2 of [ELP99], for Y ∈ O(X(cid:3) \ {M}) β{M} ◦ ηeA ◦ ιA,Y = ηeB ◦ ιB,Y ◦ βY for all Y ∈ O(X(cid:3) \ {M}), where ιA,Y : A(Y ) → A(X \ {M}) and ιB,Y : B(Y ) → B(X \ {M}) are the canonical embeddings. Since α induces an isomorphism reaching from FK+ X(cid:4)∪{M} (A(X(cid:4) ∪ {M})) to FK+ X(cid:4)∪{M} (B(X(cid:4) ∪ {M})), by Lemma 3.12, Theorem 2.3 of [ERR09], Theorem 4.14 of [MNa], Kirchberg [Kir00], and Theorem 3.3 of [ERRa]), there exists an X(cid:4)- equivariant isomorphism ψ : A(X(cid:4)) → B(X(cid:4)) such that K∗(ψ) = αX(cid:4) and [β{M} ◦ ηeA ◦ ιA,X(cid:4) ] = [ηeB ◦ ιB,X(cid:4) ◦ ψ] in KK 1(A(X(cid:4)), B({M})). By Corollary 5.6 of [ERRa], ηeA ◦ ιA,Xi and ηeB ◦ ιB,Xi are full extensions for each i = 1, 2, . . . , n with Xi being O∞-absorbing 14 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ A(X(cid:4)) = Li∈{1,2,...,n},Xi⊆X(cid:4) (i.e., Xi ⊆ X(cid:4)). Thus, ηeA ◦ ιA,X(cid:4) and ηeB ◦ ιB,X(cid:4) are full extensions since B(Xi). A(Xi) and B(X(cid:4)) = Li∈{1,2,...,n},Xi⊆X(cid:4) Hence, β{M} ◦ ηeA ◦ ιA,X(cid:4) and ηeB ◦ ιB,X(cid:4) ◦ ψ are full extensions. Let πA,X(cid:3)\{M} : A(X \ {M}) → A(X(cid:3) \ {M}), πA,X(cid:4) : A(X \ {M}) → A(X(cid:4)), πB,X(cid:3)\{M} : B(X \ {M}) → B(X(cid:3) \ {M}), πB,X(cid:4) : B(X \ {M}) → B(X(cid:4)) be the canonical projections. Note that the range of ηeA ◦ ιA,X(cid:3)\{M} and the range of ηeA ◦ ιA,X(cid:4) are orthogonal and the range of ηeB ◦ ιB,X(cid:3)\{M} and the range of ηeB ◦ ιB,X(cid:4) are orthogonal. Moreover, ηeA = ηeA ◦ ιA,X(cid:3) \{M} ◦ πA,X(cid:3)\{M} + ηeA ◦ ιA,X(cid:4) ◦ πA,X(cid:4) ηeB = ηeB ◦ ιB,X(cid:3)\{M} ◦ πB,X(cid:3)\{M} + ηeB ◦ ιB,X(cid:4) ◦ πB,X(cid:4) . We claim that there exist full hereditary sub-C∗-algebras E1 and E2 of A and B, respectively, such that E1 ∼= E2. Then by Theorem 2.8 of [Bro77], A ⊗ K ∼= B ⊗ K. Choose full projections p1, q1 ∈ A(X(cid:4)) and p2, q2 ∈ A(X(cid:3) \ {M}) such that p1 + p2 is orthogonal to q1 + q2 in A(X \ {M}) (to do this, we use stability, and that graph algebras with finitely many ideals satisfies Condition (K) and hence are of real rank zero). Therefore, ηeA(p1 + p2) 6= 1Q(A({M})) since ηeA(p1 + p2) is orthogonal to ηeA(q1 + q2). Set e1 = ψ(p1), e2 = βX(cid:3)\{M}(p2), f1 = ψ(q1), and f2 = βX(cid:3)\{M}(q2). Then e1 + e2 and f1 + f2 are nonzero orthogonal projections. So, ηeB(e1 + e2) 6= 1Q(B({M})). Set e = β{M} ◦ ηeA ◦ ιA,X(cid:3)\{M}(p2) = ηeB ◦ ιB,X(cid:3)\{M} ◦ βX(cid:3)\{M}(p2) and set f = (1Q(B({M})) − e). Let j(cid:4) : p1A(X(cid:4))p1 → A(X(cid:4)) and j(cid:3) : p2A(X(cid:3) \{M})p2 → A(X(cid:3) \ {M}) be the usual embeddings. Note that eβ{M} ◦ ηeA ◦ ιA,(cid:4) ◦ j(cid:4)(x) = β{M} ◦ ηeA ◦ ιA,(cid:4) ◦ j(cid:4)(x)e = 0 and as well as e(cid:0)ηeB ◦ ιB,X(cid:4) ◦ ψ ◦ j(cid:4)(x)(cid:1) = (cid:0)ηeB ◦ ιB,X(cid:4) ◦ ψ ◦ j(cid:4)(x)(cid:1) e = (cid:0)ηeB ◦ ιB,X(cid:3)\{M} ◦ βX(cid:3)\{M}(p2)(cid:1) ·(cid:0)ηeB ◦ ιB,X(cid:4) ◦ ψ ◦ j(cid:4)(x)(cid:1) = 0 (cid:0)ηeB ◦ ιB,X(cid:4) ◦ ψ ◦ j(cid:4)(x)(cid:1) ·(cid:0)ηeB ◦ ιB,X(cid:3)\{M} ◦ βX(cid:3)\{M}(p2)(cid:1) = 0 for all x ∈ p1A(X(cid:4))p1. Hence, we have injective homomorphisms β{M}◦ ηeA ◦ ιA,(cid:4)◦ j(cid:4) and ηeB ◦ ιB,X(cid:4) ◦ ψ ◦ j(cid:4) from p1A(X(cid:4))p1 to fQ(B({M}))f . Since B({M}) is an AF algebra, by Corollary 2.11 of [Zha91] f lifts to a projection f ′ in M(B({M})). Note that there exists an isomorphism γ from f ′M(B({M}))f ′ to M(f ′B({M})f ′) which is the identity on f ′B({M})f ′ (see II.7.3.14, pp. 147 of [BB]). Thus, we have an isomorphism γ from fQ(B({M}))f to Q(f ′B({M})f ′) such that the diagram 0 0 / f ′B({M})f ′ / f ′M(B({M}))f ′ γ fQ(B({M}))f γ / f ′B({M})f ′ / M(f ′B({M})f ′) / Q(f ′B({M})f ′) 0 / 0 is commutative. By Corollary 5.6 of [ERRa], ηeA ◦ ιA,Xi and ηeB ◦ ιB,Xi are full extensions for each i = 1, 2, . . . , n with Xi being O∞-absorbing (i.e., Xi ⊆ / / / / / / / / / / GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 15 A(X(cid:4)) = Li∈{1,2,...,n},Xi⊆X(cid:4) A(Xi) and B(X(cid:4)) = Li∈{1,2,...,n},Xi⊆X(cid:4) X(cid:4)). Thus, by Lemma 3.14, ηeA ◦ ιA,X(cid:4) and ηeB ◦ ιB,X(cid:4) are full extensions since B(Xi). Hence, β{M}◦ ηeA ◦ ιA,X(cid:4) and ηeB ◦ ιB,X(cid:4) ◦ ψ are full extensions. Thus, β{M}◦ ηeA ◦ ιA,X(cid:4) (p1) is a norm-full projection in Q(B({M})). Since β{M}◦ηeA◦ιA,X(cid:4)(p1) ≤ f , we have that f is a norm-full projection in Q(B({M})). By Lemma 3.3 of [ERRb], we have that f ′ is a norm-full projection in M(B({M})) since B({M}) has an approximate identity consisting of projections. Since B({M}) is an AF algebra, by Lemma 3.10 of [ERR09], B({M}) has the corona factorization property. Thus, f ′ is Murray-von Neumann equivalent to 1M(B({M})). Thus, f ′B({M})f ′ ∼= B({M}) which implies that f ′B({M})f ′ is a stable C∗-algebra since B({M}) is a stable C∗-algebra. Let ι be the embedding of f ′B({M})f ′ into B({M}), eι be the embedding of f ′M(B({M}))f ′ into M(B({M})), and ι be the embedding of fQ(B({M}))f into Q(B({M})). Note that the following diagram 0 0 / f ′B({M})f ′ f ′M(B({M}))f ′ fQ(B({M}))f ι / B({M}) eι ι / M(B({M})) / Q(B({M})) 0 / 0 is commutative. Note that the range of ηeB ◦ ιB,X(cid:4) ◦ ψ ◦ j(cid:4) and the range of β{M} ◦ ηeA ◦ ιA,X(cid:4) ◦ j(cid:4) are contained in fQ(B({M}))f . Let e1 be the extension defined by γ ◦ ι−1 ◦ β{M} ◦ ηeA ◦ ιA,X(cid:4) ◦ j(cid:4) and let e2 be the extension defined by γ ◦ ι−1 ◦ ηeB ◦ ιB,X(cid:4) ◦ ψ ◦ j(cid:4). Then ι ◦ γ−1 ◦ ηe1 = β{M} ◦ ηeA ◦ ιA,X(cid:4) ◦ j(cid:4) and ι ◦ γ−1 ◦ ηe2 = ηeB ◦ ιB,X(cid:4) ◦ ψ ◦ j(cid:4) Since ηeA(p1 + p2) 6= 1Q(A({M})) and ηeB(e1 + e2) 6= 1Q(B({M})) and since β{M} and ψ are isomorphisms, we have that β{M} ◦ ηeA ◦ ιA,X(cid:4) ◦ j(cid:4)(p1) 6= f and ηeB ◦ ιB,X(cid:4) ◦ ψ ◦ j(cid:4)(p1) 6= f . Thus, ηe1(p1) and ηe2 (p1) are not equal to 1Q(f ′B({M})f ′). Therefore, e1 and e2 are non-unital full extensions. Since [β{M} ◦ ηeA ◦ ιA,X(cid:4) ] = [ηeB ◦ ιB,X(cid:4) ◦ ψ] in KK 1(A(X(cid:4)), B({M})), since ι induces an element in KK (f ′B({M})f ′, B({M})) which is invertible, and since γ is an isomorphism, we have that [ηe1 ] = [ηe2] in KK 1(p1A(X(cid:4))p1, f ′B({M})f ′). Since f ′B({M})f ′ ∼= B({M}), we have that f ′B({M})f ′ has the corona factorization property. Thus, there exists a unitary u′ in M(f ′B({M})f ′) such that Ad(u′) ◦ ηe1 = ηe2, where u′ is the image of u′ in Q(f ′B({M})f ′). Let u =eι ◦ γ−1(u′). Then u is a partial isometry in M(B({M})) such that u∗u = f ′ = uu∗ and Ad(u) ◦ β{M} ◦ ηeA ◦ ιA,X(cid:4) ◦ j(cid:4) = ηeB ◦ ιB,X(cid:4) ◦ ψ ◦ j(cid:4) where u is the image of u in Q(B({M})). Set v = u + 1M(B({M})) − f ′ and let v be the image of v in Q(B({M})). Note that v = u + e and Ad(v) ◦ β{M} ◦ ηeA ◦ ιA,X(cid:3) \{M} ◦ j(cid:3) = β{M} ◦ ηeA ◦ ιA,X(cid:3) \{M} ◦ j(cid:3) Ad(v) ◦ β{M} ◦ ηeA ◦ ιA,X(cid:4) ◦ j(cid:4) = ηeB ◦ ιB,X(cid:4) ◦ ψ ◦ j(cid:4). / / / / / / / / / / / 16 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ Let a1 ∈ p1A(X(cid:4))p1 and a2 ∈ p2A(X(cid:3) \ {M})p2. Then v(cid:16)β{M} ◦ ηeA ◦ ιA,X(cid:4) ◦ j(cid:4)(a1) + β{M} ◦ ηeA ◦ ιA,X(cid:3)\{M} ◦ j(cid:3)(a2)(cid:17) v∗ = ηeB ◦ ιB,X(cid:4) ◦ ψ ◦ j(cid:4)(a1) + β{M} ◦ ηeA ◦ ιA,X(cid:3) \{M} ◦ j(cid:3)(a2) = ηeB ◦ ιB,X(cid:4) ◦ ψ ◦ j(cid:4)(a1) + ηeB ◦ ιB,X(cid:3)\{M} ◦ βX(cid:3)\{M} ◦ j(cid:3)(a2) = ηeB ◦ (ψ ◦ j(cid:4)(a1) + βX(cid:3)\{M} ◦ j(cid:3)(a2)). Hence, (1) Ad(v)◦β{M}◦ηeA◦(ιA,X(cid:4)◦j(cid:4) +ιA,X(cid:3)\{M}◦j(cid:3)) = ηeB◦(ψ◦j(cid:4) +βX(cid:3)\{M}◦j(cid:3)). Note that the Busby invariant of the extension 0 → A({M}) → E1 → (p1 + p2) (A(X(cid:4)) ⊕ A(X(cid:3) \ {M})) (p1 + p2) → 0 is given by ηeA ◦ (ιA,X(cid:4) ◦ j(cid:4) + ιA,X(cid:3) \{M} ◦ j(cid:3)) and the Busby invariant of the extension 0 → B({M}) → E2 → (e1 + e2) (B(X(cid:4)) ⊕ B(X(cid:3) \ {M})) (e1 + e2) → 0 is given by ηeB ◦ (κ(cid:4) + κ(cid:3)), where κ(cid:4) : e1B(X(cid:4))e1 → B(X(cid:4)) and κ(cid:3) : e2B(X(cid:3) \ {M})e2 → B(X(cid:3) \ {M}) are the natural embeddings. Hence, by Equation (1), Theorem 2.2 of [ELP99], and the five lemma, E1 ∼= E2. By Lemma 3.13, E1 is isomorphic to a full hereditary sub-C∗-algebra of A and E2 is isomorphic to a full hereditary sub-C∗-algebra of B. We have just proved the claim. If X(cid:4) = ∅ the result is due to Elliott's classification result [Ell76], and if X(cid:3) \ {M} = ∅ the theorem follows easily by making modifications to the above proof. (cid:3) Remark 3.16. Let A and B be graph C∗-algebras satisfying Condition (K) that are C∗-algebras over X such that each of A(Xi), B(Xi) are either AF algebras or O∞- absorbing and such that A(Xi) and B(Xi) are tight C∗-algebras over Xi, whenever A(Xi) and B(Xi) are O∞-absorbing. Assume that there exists an isomorphism α : FK+ X (B). Assume moreover, that A({M}) is an AF algebra and that for every ideal I of A, we have that I ⊆ A({M}) or A({M}) ⊆ I. Then A ⊗ K ∼= B ⊗ K. This follows from the proof above together with Corollary 5.6 of [ERRa] and applies to the cases4 (a) 4.1E.4 and 4.1E.C, where we view the algebra A that is tight over the space X (A) → FK+ (b) 4.1F.4 and 4.1F.C, where we view the algebra A that is tight over the space 4.1E as a C∗-algebra over a → b ← c as indicated by the assignment 4.1F as a C∗-algebra over a → b ← c as indicated by the assignment The following result resolves the cases 3.3.2, 3.3.3, 4.A.1, 4.A.3, 4.A.7. Theorem 3.17. Let A and B be graph C∗-algebras that are tight C∗-algebras over X, with Xi being a singleton, for each i = 1, 2, . . . , n. Suppose there exists an isomorphism α : FK+ X (B) such that α lifts to an invertible element in KK (X; A, B). Then A ⊗ K ∼= B ⊗ K. Proof. Note that we may assume that A and B are stable C∗-algebras. If A({M}) is an AF algebra, then the theorem follows from Theorem 3.15. Suppose A({M}) X (A) → FK+ 4Here we specify how we view the algebras as algebras over a → b ← c by providing a continuous map from the primitive ideal space to {a, b, c} GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 17 is O∞-absorbing. Then B({M}) is O∞-absorbing. Hence, by Proposition 3.9 and Lemma 3.14, the extensions 0 → A({M}) → A → A(X \ {M}) → 0, 0 → B({M}) → B → B(X \ {M}) → 0 are full extensions. The theorem now follows from the results of Theorem 4.6 of [ERRa]. (cid:3) 4. A pullback technique The main idea of this section is to write the algebra as a pullback of extensions we can classify coherently. The problem is, that classification usually does not give us unique isomorphisms on the algebra level. But when the quotient is an AF algebra we can in certain cases use that the KK -class of the isomorphism is unique. The main idea here is similar to the main idea of Section 3. Lemma 4.1. For each i = 1, 2, let there be given C∗-algebras Ai, Bi, and Ci together with ∗-homomorphisms αi : Ai → Ci and βi : Bi → Ci. Let Pi denote the pullback of Ai and Bi along αi and βi, for each i = 1, 2. Assume that there are isomorphisms ϕA : A1 → A2, ϕB : B1 → B2 and ϕC : C1 → C2, such that the following diagram commutes: B1 / C1 A1 α1 β1 ϕA ϕC ϕB A2 α2 / C2 β2 B2. Then we get a canonically induced isomorphism from P1 to P2. Proof. The existence of the ∗-homomorphism from P1 to P2 follows from the uni- versal property of the pullback. That this ∗-homomorphism is an isomorphism also follows from the universal property. Lemma 4.2. Let I and J be ideals of a C∗-algebra A satisfying I ∩ J = 0. Then A is the pullback of A/J and A/I along the quotient maps A/J → A/(I + J) and A/I → A/(I + J). Proof. This follows from Proposition 3.1 of [Ped99] by noting that we have a com- muting diagram (cid:3) I I / A / A/I / A/J / A/(I + J) with short exact rows. (cid:3) The signatures 4.E.4 and 4.E.5 are covered by the following theorem. Theorem 4.3. Let A and B be graph C∗-algebras that are tight over X, where X is some finite T0 space. Assume that there exists an isomorphism α : FK+ X (A) → FK+ X (B). Assume, moreover, that we have disjoint open subsets O0 and O1 of X. Let Y0 = X \ O1, Y1 = X \ O0, and Z = X \ (O0 ∪ O1). / o o / o o / / / / 18 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ Assume also Z 6= ∅ and that A(Z) is an AF algebra. (a) There exist two disjoint clopen subsets Y 1 For each i = 0, 1, if A(Oi) is O∞-absorbing, then we assume that: topology) such that Yi = Y 1 i and Y 2 i ∪ Y 2 i and Oi ⊆ Y 1 i . (b) The ideal lattice of A(Oi) is linear, i.e., Oi ∼= Xj for some j. (c) A(Oi) is an essential ideal of A(Y 1 i ) (d) A({mi}) is essential in A({mi} ∪ (Y 1 i \ Oi)), where mi is the least element of i of Yi (with the subspace Oi. Then A ⊗ K ∼= B ⊗ K. Proof. We may assume that A and B are stable C∗-algebras. Note that for each locally closed subset Y of X, A(Y ) is an AF algebra if and only if B(Y ) is an AF algebra, and A(Y ) is O∞-absorbing if and only if B(Y ) is O∞-absorbing (since there exists a positive isomorphism from K0(A(Y )) to K0(B(Y ))). Specifically B(X \ (O0 ∪ O1)) is an AF algebra. Note that the diagram 0 A(O1) A(O1) A(O0) / A / A(Y1) A(O0) / A(Y0) / A(Z) is commutative with short exact rows and columns, analogously for B. If both A(O0) and A(O1) are AF algebras, then it follows from the permanence properties of AF algebras that A is an AF algebra, and thus also B. In this case the theorem follows from Elliott's classification result [Ell76]. 1 and Z 2 1 . Then Z 1 1 and Z 2 Now assume that A(O0) is an AF algebra and that A(O1) is O∞-absorbing. Let 1 = Z \ Y 2 Z 1 1 = Y 2 1 and Z 2 1 are locally closed subsets of X, and Z is the disjoint union of Z 1 1 . Since A(Y0) and B(Y0) are extensions of AF algebras, these are themselves AF algebras. Since αY0 : K0(A(Y0)) → K0(B(Y0)) is a positive isomorphism, there exists an isomorphism β : A(Y0) → B(Y0) such that K0(β) = αY0 (by Elliott's classification result [Ell76]). Since A(Y0) and B(Y0) are AF algebras and β is an Y0-equivariant isomorphism, we have that K0(βY ) = αY for all Y ∈ LC(X) such that Y ⊆ Y0. Let and eA : 0 → A(O1) → A(Y 1 1 ) → A(Z 1) → 0. eB : 0 → B(O1) → B(Y 1 1 ) → B(Z 1) → 0. : FK+ Y 1 Since α : FK+ X (A) → FK+ X (B) is an isomorphism, we also have an isomorphism (A) → FK+ αY 1 (B). So by Theorem 4.14 of of [MNa], Kirchberg [Kir00], and Theorem 3.3 of [ERRa], there exists an isomorphism ϕ : A(O1) → B(O1) such that K∗(ϕ) = αO1 , and Y 1 1 1 1 in KK 1(A(Z 1), B(O1)), since KK (βZ 1 ) is the unique lifting of αZ 1 . (cid:2)ηeB ◦ βOZ 1(cid:3) = [ϕ ◦ ηeA] / / / / / / GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 19 As in the proof of Proposition 6.3 of [ERRa], Corollary 5.3 of [ERRa] implies that ηeA and ηeB are full extensions, and thus also the extensions with Busby maps ηeB ◦ βZ 1 and ϕ ◦ ηeA are full. Since the extensions are non-unital and B(O1) satisfies the corona factorization property, there exists a unitary u ∈ M(B(O1)) such that ηeB ◦ βZ 1 = Ad(u) ◦ ϕ ◦ ηeA where u is the image of u in the corona algebra (this follows from [EK01] and [KN06]). Hence, by Theorem 2.2 of [ELP99], there exists an isomorphism η : A(Y 1 ) is an isomorphism from eA to eB. 1 ) → B(Y 1 Since the extension 1 ) such that (Ad(u) ◦ ϕ, η, βZ 1 1 is the direct sum of the extensions 0 → A(O1) → A(Y1) → A(Z) → 0 0 → A(O1) → A(Y 1 1 ) → A(Z 1) → 0 and 0 → 0 → A(Z 2) → A(Z 2) → 0 and analogously for B, we get an isomorphism from 0 → A(O1) → A(Y1) → A(Z) → 0 to 0 → B(O1) → B(Y1) → B(Z) → 0, which is equal to βZ on the quotient. Now the theorem follows from Lemma 4.2 and Lemma 4.1. Now assume instead that both I and J are O∞-absorbing. The proof is similar to the case above. Instead of lifting αY0 : K0(A(Y0)) → K0(B(Y0)) to β : A(Y0) → B(Y0) we just lift αZ : K0(A(Z)) → K0(B(Z)) to β : A(Z) → B(Z). Then we do as above first for the extensions corresponding to the relative open subset O0 of Y0 and then for the extensions corresponding to the relative open subset O1 of Y1. As above, the theorem then follows from Lemma 4.2 and Lemma 4.1. (cid:3) 5. Ad hoc methods In this section we present arguments which resolve the classification question for some examples of tempered ideal spaces which are not covered by the general results above. Most of the results are based on knowing strong classification for smaller ideal spaces, as explained below. Our results of this nature, presented in [ERRc], are of a rather limited scope, and require restrictions on the K-theory, requiring the K-groups to be finitely generated, or even for the graph C∗-algebra to be unital. We will see this idea in use in a very clear form in the two open cases for three primitive ideals (cf. Section 5.1) and in more complicated four-point cases. Our starting point is Theorem 5.1. Let A1 and A2 be graph C∗-algebras that are tight C∗-algebras over a finite T0-space X and let U ∈ O(X) be non-empty. Let ei be the extension 0 → Ai(U ) ⊗ K → Ai ⊗ K → Ai(X \ U ) ⊗ K → 0. Suppose (1) ei is a full extension; (2) there exists an invertible element α ∈ KK (X; A1, A2); and (3) the induced invertible element αY ∈ KK (A1(Y ) ⊗ K, A2(Y ) ⊗ K) lifts to an Then A1 ⊗ K ∼= A2 ⊗ K. isomorphism from A1(Y ) ⊗ K to A2(Y ) ⊗ K for Y = U and Y = X \ U . 20 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ Proof. By (3), there exists an isomorphism ϕY : A1(Y ) ⊗ K → A2(Y ) ⊗ K for Y = U and Y = X \ U such that KK (ϕY ) = αY . It follows from (1) that ei are essential, so by [ERRa, Theorem 3.3], αX\U × [ηe2] = [ηe1] × αU . Therefore, KK (ϕX\U ) × [ηe2 ] = [ηe1 ] × KK (ϕU ). Hence, by [ERRa, Proposition 6.1 and Lemma 4.5], we have that A1 ⊗ K ∼= A2 ⊗ K. Definition 5.2. For a T0 topological space X, we will consider classes CX of sep- arable, nuclear C∗-algebras in the bootstrap category of Rosenberg and Schochet N such that (1) any element in CX is a C∗-algebra over X; (2) if A and B are in CX and there exists an invertible element α in KK (X; A, B) which induces an isomorphism from FK+ X (B), then there exists an isomorphism ϕ : A → B such that KK (ϕ) = αX , where αX is the element in KK (A, B) induced by α. X (A) to FK+ (cid:3) Remark 5.3. Let X be a finite T0-space, let U be an open subset of X, and let CU and CX\U be classes of C∗-algebras satisfying the conditions of Definition 5.2. If A1 and A2 are separable C∗-algebras such that A1(U ), A2(U ) ∈ CU and A1(X \ U ), A2(X \ U ) ∈ CX\U , then (3) in Theorem 5.1 holds. Let CX and CY be classes of C∗-algebras satisfy the conditions in Definition 5.2. Let CX⊔Y be the classes of C∗-algebras consisting of elements A ⊕ B with A ∈ CX and B ∈ CY . Then CX⊔Y satisfies the conditions in Definition 5.2. Remark 5.4. Here we will provide some examples of classes satisfying the conditions in Definition 5.2. (1) By [Kir00], the class all stable, nuclear, separable, O∞-absorbing C∗-algebras that are tight over a finite T0-space satisfy the conditions in Definition 5.2. By [ERRc, Corollary 3.10 and Theorem 3.13] and by the results of [EK], the following classes of C∗-algebras satisfies the conditions in Definition 5.2. (2) Let CXn be the class of nuclear, separable, tight C∗-algebras A over Xn such that A is stable, A({n}) is a Kirchberg algebra, A([1, n − 1]) is an AF-algebra, and Ki(A[Y ]) is finitely generated for all Y ∈ LC(Xn). X2 be the class of unital graph C∗-algebras with exactly one non-trivial ideal with the ideal being an AF algebra and the quotient O∞-absorbing, simple C∗-algebras. Let CX2 be the class of C∗-algebras A such that A ∼= B ⊗ K for some B ∈ C′ By [Ell76], the following class of C∗-algebras satisfy the conditions in Defini- (3) Let C′ X2 . tion 5.2. (4) Let CX be the class of stable AF-algebras over X. 5.1. Linear spaces. This case is solved in [ERRc], and the reader is referred there for details. However, since this is the most basic case in which our approach via Theorem 5.1 is applied, we will explain the methods for the benefit of the reader. Lemma 5.5. Let A be a graph C∗-algebra such that A is a tight C∗-algebra over Xn. (i) If A({n}) and A({1}) are O∞-absorbing and A([2, n − 1]) is an AF-algebra, then e : 0 → A([2, n]) ⊗ K → A ⊗ K → A({1}) ⊗ K → 0 GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 21 is a full extension. (ii) If A([k, n]) and A([1, k− 2]) are AF-algebras and A({k− 1}) is O∞-absorbing, then e : 0 → A([k, n]) ⊗ K → A ⊗ K → A([1, k − 1]) ⊗ K → 0 is a full extension. (iii) If A([k, n]) and A([1, k− 2]) are AF-algebras and A({k− 1}) is O∞-absorbing, then e : 0 → A([k − 1, n]) ⊗ K → A ⊗ K → A([1, k − 2]) ⊗ K → 0 is a full extension. Proof. In [ERRc], we prove (i) and (ii). We now prove (iii). Note that 0 → A({k − 1}) ⊗ K → A([k − 2, k − 1]) ⊗ K → A({k − 2}) ⊗ K → 0 is full since this is an essential extension and A({k − 1}) is O∞-absorbing. Since A([k, n]) is the largest AF -ideal of A([k− 1, n]) and A([k− 1, n])/A([k, n]) = A({k− 1}) is O∞-absorbing, by [ET10, Proposition 3.10] and [ERR09, Lemma 1.5], 0 → A([k, n]) ⊗ K → A([k − 1, n]) ⊗ K → A({k − 1}) ⊗ K → 0 is full. By [ERR10, Proposition 3.2], 0 → A([k− 1, n])⊗ K → A([k− 2, n])⊗ K → A({k− 2})⊗ K → 0 is full. Since A({k − 2}) = A([k − 2, n])/A([k− 1, n]) is an essential of A/A([k − 1, n]), the extension in (iii) is full by [ERRa, Proposition 5.4]. (cid:3) To solve the cases 3.7.5 and 4.3F.9, we now argue as follows: Theorem 5.6. Let A1 and A2 be graph C∗-algebras that are tight C∗-algebras over Xn. Suppose (i) Ai({n}) and Ai({1}) are O∞-absorbing; (ii) Ai([2, n − 1]) is an AF-algebra; and (iii) the K-groups of Ai are finitely generated. Then A1 ⊗ K ∼= A2 ⊗ K if and only if FK+ Proof. Let ei be the extension (A1 ⊗ K) ∼= FK+ (A2 ⊗ K). Xn Xn 0 → Ai([2, n]) ⊗ K → Ai ⊗ K → Ai({1}) ⊗ K → 0. Xn (A1 ⊗ K) → FK+ Note now that x induces invertible elements r[2,n] By Lemma 5.5(i), ei is a full extension. Thus, Assumption (1) of Theorem 5.1 holds. Suppose α : FK+ (A2 ⊗ K) is an isomorphism. Lift α Xn to an invertible element x ∈ KK (Xn; A1 ⊗ K, A2 ⊗ K), such a lifting exists by Theorem 4.14 of [MNa]. Therefore, Assumption (2) of Theorem 5.1 holds. (x) in KK ([2, n]; A1([2, n]) ⊗ K, A2([2, n])⊗ K) and r[1] (x) in KK (A({1})⊗ K, A2({1})⊗ K). Note that Ai([2, n]) has a smallest ideal Ai({n}) which is O∞-absorbing and the quotient Ai([2, n − 1]) is an AF algebra. By Theorem 3.9 of [ERRc], there exists an isomorphism ϕ : A1([2, n]) ⊗ K → A2([2, n]) ⊗ K such that KL(ϕ) is the invertible element in KL(A1([2, n]), A2([2, n])) induced by x. Since the K-theory of A1 is finitely generated, KL(A1([2, n]), A2([2, n])) = KK (A1([2, n]), A2([2, n])). Thus, KK (ϕ) is the invertible element in KK (A1([2, n]), A2([2, n])) induced by x. By the Kirchberg- Phillips classification, there exists an isomorphism ψ : A1({1}) ⊗ K → A2({1})⊗ K lifting r[1] Xn (x). We have just shown that Assumption (3) of Theorem 5.1 holds. Xn Xn By Theorem 5.1, we can conclude that A1 ⊗ K ∼= A2 ⊗ K. (cid:3) 22 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ Similarly, one solves 3.7.2, 4.3F.2, and 4.3F.4 using Theorem 5.7. Let A1 and A2 be graph C∗-algebras that are tight C∗-algebras over Xn. Suppose (i) Ai([k, n]) and Ai([1, k − 2]) are AF algebras; (ii) Ai({k − 1}) is O∞-absorbing; and (iii) the K-groups of Ai are finitely generated. Then A1 ⊗ K ∼= A2 ⊗ K if and only if FK+ (A1 ⊗ K) ∼= FK+ (A2 ⊗ K). Xn Xn A proof is given in [ERRc]. 5.2. Accordion spaces. Lemma 5.8. Let A be a graph C∗-algebra with signature 4.F.x, and let I be the smallest ideal of A. (1) When x = 3, 5, 7, 9, A, B, D, then the extension 0 → I⊗K → A⊗K → A/I⊗K → (2) When x = 2, 4, C, then the extension 0 → I ⊗ K → A ⊗ K → A/I ⊗ K → 0 is 0 is full. full provided that A is unital. Proof. First note that the extension 0 → I⊗K → A⊗K → A/I⊗K → 0 is essential. Hence, in the case 4.F.x for x = 3, 5, 7, 9, B, D the extension is full since I ⊗ K is a simple, purely infinite, stable C∗-algebra, which implies that Q(I ⊗ K) is simple. If A is unital and Y is the space 4.F.x for x = 2, 4, and C, then the extension is full since in this case I ∼= K and Q(K) is simple. We are left with showing the extension is full for the case 4.F.A. This case follows from [ERRa, Proposition 5.4 and Corollary 5.6]. (cid:3) Lemma 5.9. Let A be a graph C∗-algebra with tempered signature 4.3F.x for x = 5, 6, A, D. Then the ideal lattice of A is 0 E I1 E I2 E I3 E A and the extension 0 → I2 ⊗ K → A ⊗ K → A/I2 ⊗ K → 0 is full. Proof. We will for show that e : 0 → I2 ⊗ K → I3 ⊗ K → I3/I2 ⊗ K → 0 is a full extension. By Lemma 5.5, e is a full extension for x = 5, A, D. Consider the case x = 6. Note that I2 and I3/I1 are isomorphic to non-AF graph C∗-algebras with exactly one nontrivial ideal. Therefore, by Proposition 3.9, 0 → I1 ⊗ K → I2 ⊗ K → I2/I1 ⊗ K → 0 0 → I2/I1 ⊗ K → I3/I1 ⊗ K → I3/I2 ⊗ K → 0 are full extensions. By [ERR10, Proposition 3.2], e is a full extension. The lemma now follows from [ERRa, Proposition 5.4]. (cid:3) Lemma 5.10. Let A be a graph C∗-algebra with tempered signature 4.39.x for x = 2, 6, 9, A, B, C, D, or E. Let I be the greatest proper ideal of A. (1) If A is unital, then the extension 0 → I ⊗ K → A ⊗ K → A/I ⊗ K → 0 is full. (2) When x = 9, B, C, D, the extension 0 → I⊗ K → A⊗ K → A/I⊗ K → 0 is full. Proof. Suppose A is unital. Using the general theory of graph C∗-algebras with this specific ideal structure, we have that I is stable. Since A/I is simple and unital, the conclusion now follows from [ERR09, Lemma 1.5 and Proposition 1.6]. We now prove the extension 0 → I ⊗ K → A ⊗ K → A/I ⊗ K → 0 is always full for the spaces 4.39.x with x = 9, B, C, D. Note that I = I1 ⊕ I2 with I1 simple and GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 23 I2 a tight C∗-algebra over X2. By [ERRc, Lemma 4.5] and [ERRa, Corollary 5.3 and Corollary 5.6], we have 0 → I2 ⊗ K → A/I1 ⊗ K → (A/I) ⊗ K → 0 is full. Since A/I2 ⊗ K is a non-AF graph C∗-algebra with exactly one nontrivial ideal, the extension 0 → I1 ⊗ K → A/I2 ⊗ K → A/I ⊗ K → 0 is a full extension (cf. Proposition 3.9). Thus, by Lemma 3.6, 0 → I ⊗ K → A ⊗ K → A/I ⊗ K → 0 is full. (cid:3) Using the above lemmas and the Universal Coefficient Theorem of Bentmann and Kohler [BK], we get the following cases: Corollary 5.11. Let A and B be graph C∗-algebras that are tight over a finite accordion space X. Assume that there exists an isomorphism from FK+ X (A) to FK+ X (B). If (1) A and B both have tempered signature 4.F.7, 4.F.9, 4.39.B, 4.39.C, or (2) A and B both have finitely generated K-theory and have tempered signature 4.F.3, 4.F.A, 4.F.B, 4.39.9, 4.39.D, 4.3F.5, 4.3F.D, or (3) A and B both are unital and have tempered signature 4.F.2, 4.F.4, 4.F.5, 4.F.C, 4.F.D, 4.39.2, 4.39.6, 4.39.A, 4.39.E, 4.3F.6, 4.3F.A, then A ⊗ K ∼= B ⊗ K. Proof. By the above lemmas, all the extensions are full. Note that the specified ideal and quotient for each space belongs to classes of C∗-algebras satisfying the conditions in Definition 5.2. Hence, the result now follows from Theorem 5.1 and the UCT for accordion spaces. (cid:3) 5.3. Y -shaped spaces. Lemma 5.12. Let A be a graph C∗-algebra with tempered signature 4.1F.x for x = 2, 5, 6, 7, or D, and let I1 be the smallest ideal of A and let I2 be the ideal of A containing I1 such that I2/I1 is simple. (1) When x = 2, 6, 7, or D, the extension 0 → I2 ⊗ K → A ⊗ K → A/I2 ⊗ K → 0 (2) When x = 5, the extension 0 → I2 ⊗ K → A ⊗ K → A/I2 ⊗ K → 0 is full if A is full. is unital. Proof. Let J1 and J2 be the maximal ideals of A containing I2. Suppose x = 2, 6, 7, or D. Then, by Lemma 5.5, [ERR10, Proposition 3.2], and [ERRa, Corollary 5.3 and Corollary 5.6], 0 → I2 ⊗ K → Jℓ ⊗ K → Jℓ/I2 ⊗ K → 0 is full. Hence, by Lemma 3.14, 0 → I2 ⊗ K → A ⊗ K → A/I2 ⊗ K → 0 is full. Suppose that the signature is 4.1F.5 and A is unital. Assume that J1/I2 is an AF-algebra and J2/I2 is purely infinite. By Lemma 5.5, 0 → I2 ⊗ K → J2 ⊗ K → J2/I2⊗K → 0 is full. Since A is a unital graph C∗-algebra, we have that I2/I1 ∼= K. Therefore, 0 → I2/I1 ⊗ K → J1/I1 ⊗ K → J1/I2 ⊗ K → 0 is full. Since I2 is a stably isomorphic to a non-AF graph C∗-algebra with exactly one nontrivial ideal, by Proposition 3.9, 0 → I1 ⊗ K → I2 ⊗ K → I2/I1 ⊗ K → 0 is full. By [ERR10, Proposition 3.2], 0 → I2 ⊗ K → J1 ⊗ K → J1/I2 ⊗ K → 0 is full. Hence, by Lemma 3.14, 0 → I2 ⊗ K → A ⊗ K → A/I2 ⊗ K → 0 is full. (cid:3) 24 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ Lemma 5.13. Let A be a graph C∗-algebra with tempered signature 4.3E.x for x = 3, 4, 5, 9, B, or D, and let I1 and I2 be the minimal ideals of A. (1) When x = 3, 4, 5, B, D, the extension 0 → (I1 ⊕ I2) ⊗ K → A ⊗ K → A/(I1 ⊕ (2) When x = 9, and A is unital, then 0 → (I1 ⊕ I2) ⊗ K → A ⊗ K → A/(I1 ⊕ I2) ⊗ K → 0 is a full extension. I2) ⊗ K → 0 is a full extension. Proof. Suppose x = 4, 5, B, or D. Let I be the ideal of A containing (I1 ⊕ I2) such that I/(I1 ⊕ I2) is simple. Note that push forward extension of the extension 0 → (I1 ⊕ I2) ⊗ K → I ⊗ K → I/(I1 ⊕ I2) ⊗ K → 0 via the coordinate projection (I1⊕I2) → Ii is a full extension since its isomorphic to a non-AF graph C∗-algebras with exactly one nontrivial ideal. Therefore, by Lemma 3.6, 0 → (I1 ⊕ I2) ⊗ K → I ⊗ K → I/(I1 ⊕ I2) ⊗ K → 0 is a full extension. By [ERRa, Proposition 5.4], 0 → (I1 ⊕ I2) ⊗ K → A ⊗ K → A/(I1 ⊕ I2) ⊗ K → 0 is a full extension since I/(I1 ⊕ I2) ⊗ K is an essential ideal of A/(I1 ⊕ I2) ⊗ K. We now prove the extension is full for the case x = 3. Note that in this case I1 ⊗ K and I2 ⊗ K are purely infinite, simple C∗-algebras. Let I be the ideal of A containing (I1⊕I2) such that I/(I1⊕I2) is simple. By Lemma 3.5 and Lemma 3.6, 0 → (I1⊕I2)⊗K → I⊗K → I/(I1⊕I2)⊗K → 0 is a full extension. The conclusion now follows from [ERRa, Proposition 5.4] since I/(I1 ⊕ I2)⊗ K is an essential ideal of A/(I1 ⊕ I2) ⊗ K. Suppose x = 9 and A is unital. Then Ii is either K or a stable, purely infinite, simple C∗-algebra. Let I be the ideal containing I1 ⊕ I2 such that I/(I1 ⊕ I2) is simple. Note that the signature of I is 3.6. By Lemma 3.5, the push forward extension of the extension 0 → (I1 ⊕ I2) ⊗ K → I ⊗ K → I/(I1 ⊕ I2) ⊗ K → 0 via the coordinate projection (I1 ⊕ I2) ⊗ K → Ii ⊗ K is essential, and hence full since Q(Ii ⊗ K) is simple. Thus, by Lemma 3.6, 0 → (I1 ⊕ I2) ⊗ K → I ⊗ K → I/(I1 ⊕ I2) ⊗ K → 0 is full. By [ERRa, Proposition 5.4], 0 → (I1 ⊕ I2) ⊗ K → A ⊗ K → A/(I1 ⊕ I2) ⊗ K → 0 is a full extension since I/(I1 ⊕ I2) is an essential ideal of A/(I1 ⊕ I2). Lemma 5.14. Let A be a graph C∗-algebra with tempered signature 4.3E.7. Let I be the ideal of A such that A/I is simple. Then 0 → I⊗ K → A⊗ K → A/I⊗ K → 0 is a full extension. (cid:3) Proof. Let I1 and I2 be the minimal ideals of A which is contained in I. Since I/(I1 + I2) is a non-unital, purely infinite, simple C∗-algebra, we have that 0 → I/(I1 + I2) ⊗ K → A/(I1 + I2) ⊗ K → A/I ⊗ K → 0 is a full extension. The conclusion of the lemma now follows from Corollary 5.3 of [ERRa]. (cid:3) Lemma 5.15. Let A be a graph C∗-algebra with tempered signature 4.1F.E. Let I be the smallest ideal of A. Then 0 → I ⊗ K → A ⊗ K → A/I ⊗ K → 0 is a full extension. Proof. Let I1 be the ideal of A such that I1 contains I and I1/I is simple. Since I1 is stably isomorphic to a non-AF graph C∗-algebra with exactly one nontrivial ideal, we have that 0 → I ⊗ K → I1 ⊗ K → I1/I ⊗ K → 0 is full. Since I1/I is an essential ideal of A/I, the conclusion of the lemma follows from Proposition 5.4 of [ERRa]. (cid:3) Using the above lemmas and the results of [ARR12], we get the following: GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 25 Corollary 5.16. Let A and B be graph C∗-algebras with signature either 4.1F or 4.3E, and assume that there exists an isomorphism from FK+ X (B). If (1) A and B both have tempered signature 4.1F.7, 4.1F.E, 4.3E.3, 4.3E.7, or 4.3E.D, X (A) to FK+ or (2) A and B both have finitely generated K-theory and have tempered signature 4.1F.D, 4.3E.4 or 4.3E.5, or (3) A and B both are unital and have tempered signature 4.1F.2, 4.1F.5, 4.1F.6, 4.3E.9 or 4.3E.B, then A ⊗ K ∼= B ⊗ K. Proof. By the above lemmas, all the extensions are full. Note that the specified ideal and quotient for each space belongs to classes of C∗-algebras satisfying the conditions in Definition 5.2. Hence, the result now follows from Theorem 5.1. (cid:3) 5.4. O-shaped spaces. Lemma 5.17. Let A be a graph C∗-algebra that is a tight C∗-algebra over the O- shaped space 4.3B.7. Let I be the smallest ideal of A and let I1 and I2 be the ideals of A which contain I and Ik/I is simple. Then 0 → (I1 + I2) ⊗ K → A ⊗ K → A/(I1 + I2) ⊗ K → 0 is a full extension. Proof. Note that A/I is a tight C∗-algebra over the space 3.6.5. Then by Lemma 3.6, 0 → (I1 + I2)/I ⊗ K → A/I ⊗ K → A/(I1 + I2) ⊗ K → 0 is a full extension since I1/I and I2/I are purely infinite, simple C∗-algebras. Also, since I is an essential ideal of I1 + I2 and since I is a purely infinite, simple C∗-algebra, we have that 0 → I⊗ K → (I1 + I2)⊗ K → (I1 + I2)/I⊗ K → 0 is a full extension. The conclu- sion of the lemma now follows from Proposition 3.2 of [ERR10] since A/(I1 + I2) is simple. (cid:3) Lemma 5.18. Let A be a graph C∗-algebra that is a tight C∗-algebra over the O- shaped space 4.3B.E. Let I be the smallest ideal of A. Then 0 → I⊗ K → A ⊗ K → A/I ⊗ K → 0 is a full extension. Proof. Let I1 and I2 be the ideals of A which contain I and Ik/I is simple. Since Ik ⊗ K is isomorphic to a graph C∗-algebra with exactly one non-trivial ideal and Ik ⊗ K is not an AF algebra, by Proposition 3.9, we have that 0 → I ⊗ K → Ik ⊗ K → Ik/I ⊗ K → 0 is a full extension. By Lemma 3.14, 0 → I ⊗ K → (I1 + I2) ⊗ K → (I1 + I2)/I ⊗ K → 0 is a full extension. The conclusion of the lemma now follows from Proposition 5.4 of [ERR09] since (I1 + I2)/I ⊗ K is an essential ideal of A/I. (cid:3) Using the above lemmas and the results of [ABK], we get the following cases: Corollary 5.19. Let A and B be graph C∗-algebras that are tight over a O-shaped space X. Assume that there exists an isomorphism from FK+ X (B). If A and B both have tempered signature 4.3B.7 or 4.3B.E, then A ⊗ K ∼= B ⊗ K. Proof. By the above lemmas, all the extensions are full. Note that the specified ideal and quotient for each space belongs to classes of C∗-algebras satisfying the conditions in Definition 5.2. Hence, the result now follows from Theorem 5.1. (cid:3) X (A) to FK+ 26 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ 6. Summary of results In this final section, we index our results. Cases that open are indicated by "?". Cases that are solved in general are marked by "√", and if we need to impose conditions of finitely generated K-theory or unitality, this is indicated by "√f.g." or "√1", respectively. 6.1. One point spaces. Having nothing new to add, we include the simple case only for completeness. 1.0.x 0 (cid:3) √ Theorem 2.1 1 (cid:4) √ Theorem 2.3 6.2. Two point spaces. This case was solved in [ET10], so again we include it only for completeness. 0 (cid:3) 1 (cid:3) 2 (cid:4) 3 (cid:4) 2.1.x (cid:3) √ Theorem 2.1 (cid:4) √ Proposition 2.5 (cid:3) √ Proposition 2.6 (cid:4) √ Theorem 2.3 6.3. Three point spaces. We resolve the case of three primitive ideal spaces here, up to a condition of finite generation which must be imposed in the cases of signature 3.7.2 and 3.7.5. We do not know if this condition is necessary. 0 (cid:3) 1 (cid:3) 2 (cid:3) 3 (cid:3) 5 (cid:4) 7 (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) 3.3.x (cid:3) √ Theorem 2.1 (cid:4) √ Theorem 3.15 (cid:3) √ Theorem 3.17 (cid:4) √ Theorem 3.17 (cid:4) √ Theorem 3.15 (cid:4) √ Theorem 2.3 3.7.x 0 (cid:3) 1 (cid:3) 2 (cid:3) 3 (cid:3) 5 (cid:4) 7 (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) 3.6.x (cid:3) √ Theorem 2.1 (cid:4) √ Theorem 3.7 (cid:3) √ Theorem 3.10 (cid:4) √ Theorem 3.10 (cid:4) √ Theorem 3.7 (cid:4) √ Theorem 2.3 0 (cid:3) 1 (cid:3) 2 (cid:3) 3 (cid:3) 4 (cid:4) 5 (cid:4) 6 (cid:4) 7 (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) (cid:4) √ √ (cid:3) √f.g. √ √ (cid:4) √f.g. √ √ (cid:3) (cid:3) (cid:4) Theorem 2.1 Proposition 2.5 Theorem 5.7 Proposition 2.5 Proposition 2.6 Theorem 5.6 Proposition 2.6 Theorem 2.3 GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 27 6.4. Four point spaces. In this section, we present our results for the case of four primitive ideals. As will be obvious below, the strength of our results varies dramatically with the nature of the spaces. In general, we can say quite a lot about all spaces apart from 4.E, 4.1E, and 4.3B. It may be interesting to note what makes these spaces difficult to handle; indeed the case 4.E is an accordion space in which a general UCT is know to hold, but it differs from the other accordion spaces by having poor separation properties when it comes to establishing fullness. The O- shaped spaces are also hard to separate fully, but have the added difficulty that no general UCT is known for them. 0 (cid:3) 1 (cid:3) 2 (cid:3) 3 (cid:3) 4 (cid:3) 5 (cid:3) 6 (cid:3) 7 (cid:3) 8 (cid:4) 9 (cid:4) A (cid:4) B (cid:4) C (cid:4) D (cid:4) E (cid:4) F (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) 4.E.x (cid:3) √ Theorem 2.1 (cid:3) √ Remark 3.8 (cid:3) ? (cid:3) ? (cid:4) √ Theorem 4.3 (cid:4) √ Theorem 4.3 (cid:4) ? (cid:4) ? (cid:3) ? (cid:3) ? (cid:3) ? (cid:3) ? (cid:4) ? (cid:4) ? (cid:4) ? (cid:4) √ Theorem 2.3 (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) 0 (cid:3) 1 (cid:3) 2 (cid:3) 3 (cid:3) 4 (cid:3) 5 (cid:3) 6 (cid:3) 7 (cid:3) 8 (cid:4) 9 (cid:4) A (cid:4) B (cid:4) C (cid:4) D (cid:4) E (cid:4) F (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) 4.F.x (cid:3) (cid:3) Theorem 2.1 Proposition 2.5 (cid:3) Corollary 5.11 (cid:3) √f.g. Corollary 5.11 Corollary 5.11 (cid:4) Corollary 5.11 Theorem 3.15 Corollary 5.11 Theorem 3.15 Corollary 5.11 (cid:3) √f.g. Corollary 5.11 (cid:3) √f.g. Corollary 5.11 Corollary 5.11 (cid:4) Corollary 5.11 Theorem 3.15 Theorem 2.3 √ √ √1 √1 √1 √ √ √ √ √1 √1 √ √ (cid:4) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:4) 2 8 S Ø R E N E I L E R S , G U N N A R R E S T O R F F , A N D E F R E N R U I Z 4.39.x 4.3F.x 0 (cid:3) 1 (cid:3) 2 (cid:3) 3 (cid:3) 4 (cid:4) 5 (cid:4) 6 (cid:4) 7 (cid:4) 8 (cid:3) 9 (cid:3) A (cid:3) B (cid:3) C (cid:4) D (cid:4) E (cid:4) F (cid:4) (cid:3) (cid:3) (cid:3) (cid:3) (cid:3) (cid:3) (cid:3) (cid:3) (cid:4) (cid:4) (cid:4) (cid:4) (cid:4) (cid:4) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) √ √ √1 √ √ √ √1 √ √ √1 √ √ √1 √ Theorem 2.1 Theorem 3.7 Corollary 5.11 Theorem 3.7 Theorem 3.7 Theorem 3.7 Corollary 5.11 Theorem 3.7 Proposition 2.6 (cid:4) √f.g. Corollary 5.11 Corollary 5.11 (cid:3) Corollary 5.11 Corollary 5.11 (cid:4) √f.g. Corollary 5.11 Corollary 5.11 (cid:3) Theorem 2.3 0 (cid:3) 1 (cid:3) 2 (cid:3) 3 (cid:3) 4 (cid:3) 5 (cid:3) 6 (cid:3) 7 (cid:3) 8 (cid:4) 9 (cid:4) A (cid:4) B (cid:4) C (cid:4) D (cid:4) E (cid:4) F (cid:4) (cid:3) (cid:3) (cid:3) (cid:3) (cid:4) (cid:4) (cid:4) (cid:4) (cid:3) (cid:3) (cid:3) (cid:3) (cid:4) (cid:4) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:4) (cid:3) (cid:4) Proposition 2.5 Theorem 5.7 Theorem 2.1 √ √ (cid:3) √f.g. √ (cid:3) √f.g. Theorem 5.7 (cid:4) √f.g. Corollary 5.11 Corollary 5.11 Proposition 2.5 Proposition 2.5 Proposition 2.6 Theorem 5.6 Corollary 5.11 (cid:3) (cid:4) (cid:3) √1 √ √ (cid:4) √f.g. √1 ? √ (cid:3) (cid:4) (cid:3) Proposition 2.6 (cid:4) √f.g. Corollary 5.11 Proposition 2.6 (cid:3) (cid:4) Theorem 2.3 √ √ G R A P H C ∗ - A L G E B R A S W I T H N O M O R E T H A N F O U R I P R M I T I V E I D E A L S 2 9 30 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ 4.A.x 4.38.x (cid:3) (cid:3) (cid:3) (cid:3) (cid:3) (cid:3) (cid:4) (cid:4) 0 1 2 3 6 7 E F (cid:3) (cid:3) (cid:3) (cid:4) (cid:3) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:4) (cid:4) (cid:4) (cid:4) (cid:4) (cid:4) √ Theorem 2.1 0 (cid:3) √ Theorem 3.17 1 (cid:3) √ Theorem 3.15 3 (cid:3) √ Theorem 3.17 7 (cid:4) √ Theorem 3.15 8 (cid:3) √ Theorem 3.17 9 (cid:3) √ Theorem 3.15 B (cid:3) √ Theorem 2.3 F (cid:4) (cid:3) (cid:3) (cid:3) (cid:3) (cid:3) (cid:4) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) √ Theorem 2.1 √ Theorem 3.7 √ Theorem 3.7 √ Theorem 3.7 √ Theorem 3.10 √ Theorem 3.10 √ Theorem 3.10 √ Theorem 2.3 GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 31 4.1F.x 4.3E.x (cid:3) (cid:3) (cid:3) (cid:3) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:3) (cid:3) (cid:3) (cid:3) (cid:4) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) (cid:4) (cid:3) (cid:3) (cid:3) (cid:4) (cid:3) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) (cid:3) (cid:4) (cid:4) 0 1 2 3 4 5 6 7 C D E F √ √ √1 √ √ √1 √1 √ √ Theorem 2.1 Proposition 2.5 Corollary 5.16 Proposition 2.5 Remark 3.16 Corollary 5.16 Corollary 5.16 Corollary 5.16 Remark 3.16 √f.g. Corollary 5.16 √ √ Corollary 5.16 Theorem 2.3 (cid:3) (cid:3) (cid:3) (cid:3) (cid:3) (cid:4) (cid:3) (cid:3) (cid:3) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:3) (cid:4) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) (cid:4) (cid:3) (cid:3) (cid:3) (cid:4) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) (cid:4) 0 1 3 4 5 7 8 9 B C D F √ √ √ Theorem 2.1 Remark 3.8 Corollary 5.16 √f.g. Corollary 5.16 √f.g. Corollary 5.16 √ √ √1 √1 √ √ √ Corollary 5.16 Proposition 2.6 Corollary 5.16 Corollary 5.16 Proposition 2.6 Corollary 5.16 Theorem 2.3 32 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ (cid:3) (cid:3) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:3) (cid:3) (cid:4) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) (cid:4) (cid:3) (cid:3) (cid:3) (cid:3) (cid:4) (cid:3) (cid:3) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) 0 1 3 4 5 7 C D F 4.1E.x √ Theorem 2.1 √ Remark 3.8 √ Remark 3.8 √ Remark 3.16 ? ? √ Remark 3.16 ? ? (cid:3) (cid:3) (cid:3) (cid:3) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) (cid:3) (cid:4) (cid:4) (cid:4) (cid:4) (cid:4) (cid:4) (cid:4) (cid:4) (cid:3) (cid:3) (cid:3) (cid:4) (cid:3) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) (cid:4) (cid:3) (cid:3) (cid:3) (cid:4) (cid:3) (cid:3) (cid:3) (cid:4) (cid:4) (cid:3) (cid:4) (cid:4) 0 1 2 3 6 7 8 9 A B E F 4.3B.x √ Theorem 2.1 √ Proposition 2.5 ? ? ? √ Corollary 5.19 √ Proposition 2.6 ? ? ? √ Corollary 5.19 √ Theorem 2.3 GRAPH C ∗-ALGEBRAS WITH NO MORE THAN FOUR PRIMITIVE IDEALS 33 References [ABK] S. Arklint, R. Bentmann, and T. Katsura, Reduction of filtered K-theory and a charac- terization of Cuntz-Krieger algebras, in preparation. [ARR12] S. Arklint, G. Restorff, and E. Ruiz, Filtrated K-theory for real rank zero C ∗-algebras, Int. J. Math. 23 (2012). [BHRS] T. Bates, J. Hong, I. Raeburn, and W. Szyma´nski, The ideal structure of the C ∗-algebras [BK] of infinite graphs, Illinois J. Math., 46 (2002), 1159 -- 1176. R. Bentmann and M. Kohler, Universal coefficient theorems for C ∗-algebras over finite topological spaces, preprint arXiv:1101.5702. B. Blackadar, Operator algebras, Springer-Verlag, Berlin, 2006. [BB] [BM02] G. Brinkmann and B.D. McKay, Posets on up to 16 points, Order 19 (2002), no. 2, 147 -- 179. [Bro77] L.G. Brown, Stable isomorphism of hereditary subalgebras of C ∗-algebras, Pacific J. [CET] Math. 71 (1977), no. 2, 335 -- 348. T.M. Carlsen, S. Eilers, and M. Tomforde, Index maps in the K-theory of graph algebras, J. K-Theory 9 (2012), 385 -- 406. [DT02] D. Drinen and M. Tomforde, Computing K-theory and Ext for graph C ∗-algebras, Illi- nois J. Math.46 (2002), 81 -- 91. [DT05] D. Drinen and M. Tomforde, The C ∗-algebras of arbitrary graphs, Rocky Mountain [EK] J. Math. 35 (2005), 105 -- 135. S. Eilers and T. Katsura, Semiprojectivity and properly infinite projections in graph C ∗-algebra, in preparation. [EK01] G.A. Elliott and D. Kucerovsky, An abstract Voiculescu-Brown-Douglas-Fillmore ab- sorption theorem, Pacific J. Math. 198 (2001), no. 2, 385 -- 409. [Ell76] G.A. Elliott, On the classification of inductive limits of sequences of semisimple finite- dimensional algebras, J. Algebra 38 (1976), no. 1, 29 -- 44. [Ell10] [ELP99] S. Eilers, T.A. Loring, and G.K. Pedersen, Morphisms of extensions of C ∗-algebras: , Towards a theory of classification, Adv. Math. 223 (2010), no. 1, 30 -- 48. pushing forward the Busby invariant, Adv. Math. 147 (1999), no. 1, 74 -- 109. [ERRa] S. Eilers, G. Restorff, and E. Ruiz, Classifying C ∗-algebras with both finite and infinite subquotients, preprint arXiv:1009.4778. [ERRb] [ERRc] [ERR09] , The ordered K-theory of a full extension, preprint arXiv: 1106.1551. , Strong classification of extensions of classifiable C ∗-algebras, in preparation. , Classification of extensions of classifiable C ∗-algebras, Adv. Math. 222 (2009), 2153 -- 2172. [ERR10] , On graph C ∗-algebras with a linear ideal lattice, Bull. Malays. Math. Sci. Soc. 33 (2010), no. 2, 233 -- 241. [ERS11] S. Eilers, E. Ruiz, and A.P.W. Sørensen, Amplified graph C ∗-algebras, to appear in [ET10] [JJ] Munster J. Math, 2011. S. Eilers and M. Tomforde, On the classification of nonsimple graph algebras, Math. Ann. 346 (2010), 393 -- 418. JA. Jeong, Real rank of C ∗-algebras associated with graphs, J. Aust. Math. Soc., 77 (2004), 141 -- 147. [Kir94] E. Kirchberg, The classification of purely infinite C ∗-algebras using Kasparov's theory, preprint, third draft, 1994. [Kir00] E. Kirchberg, Das nicht-kommutative Michael-Auswahlprinzip und die Klassifikation nicht-einfacher Algebren, C ∗-algebras (Munster, 1999), Springer, Berlin, 2000, pp. 92 -- 141. [KR00] E. Kirchberg and M. Rørdam, Non-simple purely infinite C ∗-algebras, Amer. J. Math., 122 (2000), 637 -- 666. [KR02] E. Kirchberg and M. Rørdam, Infinite non-simple C ∗-algebras: absorbing the Cuntz algebras O∞, Adv. Math., 167 (2002), 195 -- 264. [KN06] D. Kucerovsky and P. W. Ng, The corona factorization property and approximate uni- [MNa] tary equivalence, Houston J. Math. 32 (2006), no. 2, 531 -- 550 (electronic). R. Meyer and R. Nest, C ∗-algebras over topological spaces: Filtrated K-theory, Canad. J. Math. 64 (2012), no. 2, 368 -- 408. 34 SØREN EILERS, GUNNAR RESTORFF, AND EFREN RUIZ [MNb] , C ∗-algebras over topological spaces: The bootstrap class, Munster J. Math. 2 (2009), 215 -- 252. [Mur90] G.J. Murphy, C ∗-algebras and operator theory, Academic Press, San Diego, 1990. [Ped99] G.K. Pedersen, Pullback and pushout constructions in C ∗-algebra theory, J. Funct. Anal. 167 (1999), 243 -- 344. [NCP00] N.C. Phillips, A classification theorem for nuclear purely infinite simple C ∗-algebras, [Rae05] Doc. Math. 5 (2000), 49 -- 114. I. Raeburn, Graph algebras, CBMS Regional Conference Series in Mathematics, vol. 103, Published for the Conference Board of the Mathematical Sciences, Washington, DC, 2005. [Res08] G. Restorff, Classification of non-simple C ∗-algebras, Ph.D. partment http://www.math.ku.dk/∼restorff/papers/thesis.pdf. of Mathematical Sciences, University of Copenhagen, thesis, De- 2008, [Ror97] M. Rørdam, Classification of extensions of certain C ∗-algebras by their six term exact sequences in K-theory, Math. Ann. 308 (1997), no. 1, 93 -- 117. [MT07] M. Tomforde, Uniqueness theorems and ideal structure for Leavitt path algebras, J. Algebra, 318 (2007), 270 -- 299. [TW07] A. Toms and W. Winter, Strongly self-absorbing C ∗-algebras, Trans. Amer. Math. Soc., 359 (2007), 3999 -- 4029. [Zha91] S. Zhang, K1-groups, quasidiagonality, and interpolation by multiplier projections, Trans. Amer. Math. Soc. 325 (1991), no. 2, 793 -- 818. Department of Mathematical Sciences, University of Copenhagen, Universitetsparken 5, DK-2100 Copenhagen, Denmark E-mail address: [email protected] Department of Science and Technology, University of the Faroe Islands, N´oat´un 3, FO-100 T´orshavn, Faroe Islands E-mail address: [email protected] Department of Mathematics, University of Hawaii, Hilo, 200 W. Kawili St., Hilo, Hawaii, 96720-4091 USA E-mail address: [email protected]
1109.5171	2	1109	2012-02-08T17:48:23	Open projections in operator algebras I: Comparison theory	[ "math.OA", "math-ph", "math.FA", "math-ph" ]	We begin a program of generalizing basic elements of the theory of comparison, equivalence, and subequivalence, of elements in C*-algebras, to the setting of more general algebras. In particular, we follow the recent lead of Lin, Ortega, Rordam, and Thiel of studying these equivalences, etc., in terms of open projections or module isomorphisms. We also define and characterize a new class of inner ideals in operator algebras, and develop a matching theory of open partial isometries in operator ideals which simultaneously generalize the open projections in operator algebras (in the sense of the authors and Hay), and the open partial isometries (tripotents) introduced by the authors.	math.OA	math	OPEN PROJECTIONS IN OPERATOR ALGEBRAS I: COMPARISON THEORY DAVID P. BLECHER AND MATTHEW NEAL Abstract. We begin a program of generalizing basic elements of the theory of comparison, equivalence, and subequivalence, of elements in C ∗-algebras, to the setting of more general algebras. In particular, we follow the recent lead of Lin, Ortega, Rørdam, and Thiel of studying these equivalences, etc., in terms of open projections or module isomorphisms. We also define and characterize a new class of inner ideals in operator algebras, and develop a matching theory of open partial isometries in operator ideals which simultaneously generalize the open projections in operator algebras (in the sense of the authors and Hay), and the open partial isometries (tripotents) introduced by the authors. 1. Introduction and notation Inspired by a recent paper of Ortega, Rørdam, and Thiel [31], we begin a pro- gram of generalizing basic elements of the theory of comparison, equivalence, and subequivalence, of elements in C∗-algebras, to the setting of more general algebras. To do this we establish several technical results and tools, such as a new class of inner ideals in operator algebras, and a matching theory of 'open partial isometries' which simultaneously generalize the open projections in operator algebras (in the sense of the authors and Hay [9]), and (essentially) the open partial isometries or tripotents introduced in [12]. We begin by considering a relation between two elements a and b which one may define in any monoid or algebra A: namely that there exists x, y ∈ A with a = xy, b = yx. If A is a group, then this defines an equivalence relation. In an algebra this is not an equivalence relation in general. In fact this fails to be an equivalence relation even for the case A = Mn, the n by n matrices. Indeed even in this simple case the characterization of when two matrices a and b can be related in this way is quite subtle [27], and does not match any kind of equivalence relation on a C∗-algebra that has been important in C∗-algebra theory as far as we know. The question arises of how to fix this problem in an operator algebra, by which we will mean a (not necessarily selfadjoint) subalgebra of B(H), for a Hilbert space H. In a C∗-algebra A the 'fix' that works, on a large subset of A, is to insist that y = x∗ above; and then this defines an equivalence relation ∼ on the positive cone A+ of A. This is sometimes called Pedersen equivalence (see e.g. [33]). We will 2010 Mathematics Subject Classification. Primary 46L85, 46H10, 46L07, 47L30; Secondary 06F25, 17C65, 46L08, 47L07. Key words and phrases. TRO's, JB-triples, Hilbert C-modules, nonselfadjoint operator al- gebra, open projection, Cuntz semigroup, comparison theory, equivalence relations on an operator algebra, hereditary subalgebra, ideals, partial isometry. February 7, 2012. 1 2 D. P. BLECHER AND MATTHEW NEAL expand this 'fix' to a larger set than A+, and to more general operator algebras than C∗-algebras. In C∗-algebra theory, coarser equivalence relations than Pedersen equivalence, and matching notions of subequivalence or comparison, are becoming increasingly important [6]. For example, recently the study of Cuntz equivalence and subequiv- alence has become one of the most important areas of C∗-algebra theory (see e.g. [4, 31]). And comparison theory in C∗-algebras, where one considers coarser notions of ordering of elements in a C∗-algebra than the usual ≤, generalizing in some sense the crucially important comparison theory of projections in a von Neumann algebra (which is often most effectively done using traces), is obviously important even in its own right. Studying these equivalence relations and comparisons involves mar- shaling a formidable array of tools, results, and perspectives. Some of this is simply impossible for more general operator algebras, due for example to scarcity of pro- jections and hereditary subalgebras in general. A classification program along these lines would be misguided. Our goal in this project is simply to transfer some small portion of these tools, results, and perspectives to more general operator algebras than C∗-algebras. Along the way we were also led to develop some aspects of 'non- commutative topology and noncommutative function theory relative to an operator algebra'. In the present paper we discuss generalizations of Pedersen, Blackadar, and Peligrad-Zsid´o equivalence, using the paper of Ortega, Rørdam, and Thiel [31] as our guide. The main point is that the authors of [31] recast the equivalences and subequivalences mentioned above, and also Cuntz equivalence and subequivalence, in terms of open projections (see also [29]). In view of our generalizations of open projections in earlier projects [9, 12, 15], it was tempting to follow their approach in a more general setting. Our variants of Pedersen, Blackadar, and Peligrad-Zsid´o equivalence lead us to introduce in Section 3 the notion of a ∗-open tripotent, which may be viewed as a generalization of Akemann's open projections, and a generalization of our version of the latter for general operator algebras from [9]. They also are essentially a 'non-selfadjoint variant' of the open tripotents which we introduced in [12]. These ∗-open tripotents play a central role in our equivalence and comparison theory. One characterization of ∗-open tripotents is in terms of hereditary bimodules, which encapsulate the algebraic structure of aAb when a and b are Pedersen or Blackadar equivalent. We develop the theory of hereditary bimodules in Section 4. For ex- ample we show using results from [15] that every separable hereditary bimodule is of the form aAb where a and b are Pedersen equivalent; and that a general heredi- tary bimodule is the closure of an increasing union of subspaces of the same form. These sections are mostly of technical interest, for later use in the development of the theory, however applications are given there in the form of rephrasings of our variant of Pedersen equivalence in the language of ∗-open tripotents or hereditary bimodules. In Section 5 we study a variant of 'Blackadar equivalence and subequiv- alence' of elements in our algebra A, and prove the analogue of some results of Lin and Ortega-Rørdam-Thiel [29, 31]. For example, elements a and b of the type we consider are Blackadar equivalent in our sense iff aA ∼= bA completely isometrically and as right A-modules. It is also equivalent to aAb being a "principal' hereditary bimodule, or to the support projections of a and b being equivalent in an appro- priate variant of the sense of Peligrad and Zsid´o. In an appendix we prove TRO versions of a few of our earlier results. OPEN PROJECTIONS IN OPERATOR ALGEBRAS I 3 We remark that much of our theory depends on the existence of nth roots of elements satisfying k1 − ak ≤ 1. Thus we would expect that a certain portion of this theory generalizes to bigger classes of elements which have roots, and a smaller portion may generalize further still to a bigger class of Banach algebras using the facts about roots in Banach algebras (see e.g. [28]). Blanket convention: Throughout this paper, A is a fixed operator algebra, and B is a C∗-algebra that contains it. Sometimes B will be generated by A; that is there is no proper C∗-subalgebra of B containing A. The diagonal ∆(A) = {a ∈ A : a∗ ∈ A} is a C∗-algebra and a subalgebra of A. Concerning notation and background, which we will discuss next, it will be helpful for the reader to have easy access to several of the references, particularly [10, 9, 12, 15, 11], for example for more detail beyond what is presented here. For us a projection is always an orthogonal projection. We recall that by a theorem due to Ralf Meyer, every operator algebra A has a unique unitization A1 (see e.g. [10, Section 2.1]). Below 1 always refers to the identity of A1 if A has no identity. We are mostly interested in operator algebras with contractive approximate identities (cai's). We also call these approximately unital operator algebras. If A is a nonunital operator algebra represented (completely) isometrically on a Hilbert space H then one may identify the unitization A1 with A + C IH . The second dual A∗∗ is also an operator algebra with its (unique) Arens product, this is also the product inherited from the von Neumann algebra B∗∗ if A is a subalgebra of a C∗-algebra B. Meets and joins in B∗∗ of projections in A∗∗ remain in A∗∗, since these meets and joins may be computed in the biggest von Neumann algebra contained inside A∗∗. Note that A has a cai iff A∗∗ has an identity 1A∗∗ of norm 1, and then A1 is sometimes identified with A + C 1A∗∗. For sets X and Y we write XY for the closure of the sum of products of the form xy for x ∈ X, y ∈ Y , and similarly for a product of three sets. If a ∈ A then we write Aa for aAa. An inner ideal in A is a subspace D with DAD ⊂ D. A hereditary subalgebra (HSA) of A is an inner ideal which has a cai. For the theory of HSA's in general operator algebras see [9]. These objects are in an order preserving, bijective correspondence with the open projections p ∈ A∗∗, by which we mean that there is a net xt ∈ A with xt = pxtp → p weak. These are also the open projections p in the sense of Akemann [1, 2] in B∗∗, where B is a C∗-algebra containing A, such that p ∈ A⊥⊥. Indeed the weak limit of a cai for a HSA is an open projection, and is called the support projection of the HSA. Conversely, if p is an open projection in A∗∗, then pA∗∗p ∩ A is a HSA in A. We write pA∗∗p ∩ A as pAp, or simply as Ap. Similarly, if q is another projection, then pAq = {a ∈ A : a = paq}. We recall that a closed projection is the 'perp' of an open projection. Suprema (resp. infima) of open (resp. closed) projections in A∗∗, remain in A∗∗, by the fact mentioned two paragraphs earlier about meets and joins, together with the C∗-algebraic case of these facts [1, 2]. In [34], Peligrad and Zsid´o introduce a notion of equivalence for open projec- tions p and q in the bidual of a C∗-algebra B: We say p and q are Peligrad-Zsid´o equivalent, and write p ∼PZ q if there is a partial isometry v ∈ B∗∗ such that p = v∗v, q = vv∗, vBp ⊂ B, v∗Bq ⊂ B. Peligrad and Zsid´o prove in [34, Lemma 1.3] that in fact q being open and v∗Bq ⊂ B are implied by the other conditions above. We will need some background from [12]. A ternary ring of operators (or TRO for short), is a closed subspace Z of a C-algebra A such that ZZ∗Z ⊂ Z. A 4 D. P. BLECHER AND MATTHEW NEAL tripotent is an element u ∈ Z such that uu∗u = u. This is clearly simply a partial isometry in Z. We order tripotents by u ≤ v if and only if uv∗u = u. This turns out to be equivalent to u = vu∗u, or to u = uu∗v, and implies that u∗u ≤ v∗v and uu∗ ≤ vv∗. The linking C-algebra L(Z) of a TRO Z has 'four corners' ZZ∗, Z, Z∗, and Z∗Z. Here ZZ∗ is the closure of the linear span of products zw∗ with z, w ∈ Z, and similarly for Z∗Z. The second dual of a TRO Z is a TRO, which is studied in terms of the von Neumann algebra which is the second dual of L(Z). An inner ideal (resp. ternary ideal) of a TRO Z is defined to be a closed subspace J with JZ∗J ⊂ J (resp. JZ∗Z ⊂ J and ZZ∗J ⊂ J). In most of our paper, although we will be using notation and concepts from the theory of TRO's and JB-triples, in fact the TRO concerned will simply be the C∗-algebra B. Thus we will write the discussion in the next few paragraphs in terms of B, although it all makes sense even if B were a TRO. The Peirce 2-space of a tripotent u in B is the subset B2(u) = {z ∈ B : z = uu∗zu∗u} = uu∗Bu∗u. Clearly B2(u) is an inner ideal of B in the TRO sense. There is a natural product (the Peirce product x · y = xu∗y) and involution (namely x♯ = ux∗u) on B2(u) making the latter space into a unital C∗-algebra. The identity element is u. It is easy to check that u∗B2(u) is a C-subalgebra of B (this will be B∗B if B is a TRO), and the map z 7→ u∗z is a ∗-isomorphism from B2(u), with the product and involution above, onto this C-subalgebra. We will write B(u) for B∗∗2 (u) ∩ B = {z ∈ B : z = uu∗zu∗u}. Here B∗∗2 (u) is the Peirce 2-space of u in B∗∗. We will also occasionally use other facts from Sections 2 and 3 of [12], like those concerning the range tripotent r(x) of an element x ∈ B. Since this is simply the partial isometry occurring in the polar decomposition of x, so x = r(x)x, the facts concerning range tripotents are fairly simple and essentially well known. For example, r(x) ∈ B∗∗ and x = x∗r(x). In [12] we defined a tripotent u in the second dual of a TRO to be open, if when we consider the Peirce 2-space for u as a W-algebra in the way described above, then u is the weak* limit in the second dual, of an increasing net of positive elements from B(u) (positive with respect to the C∗-algebra structure determined by the Pierce product; and B(u) is replaced by Z(u) in the case of a general TRO Z rather than B). Beware that this definition differs from the one given in [20]. For example, all unitaries are open in the sense of that paper. See [22, 23, 18] for a recent JB-triple generalization of our notion. There are several equivalent definitions of open tripotents in [12, Theorem 2.10, Corollary 3.4]. For example, a tripotent u in the second dual is open iff it is a weak limit of an increasing net of range tripotents (defined above), or if the projection u = 1 2 (cid:20) uu∗ u∗ u u∗u (cid:21) is an open projection in Akemann's sense (discussed earlier). Another important characterization of open tripotents, as the 'support tripotent' of certain inner ideals, will be discussed in Section 4. In Proposition 6.1, we show that our open tripotents are (at least in the case that the TRO is a C∗-algebra) the partial isometries occurring in the Peligrad and Zsid´o equivalence of open projections defined above. A special case of this was proven as the equivalence of (i) and (vi) in [16]. In our work [12, 16] we were unaware of [34], which in retrospect clearly overlaps in small OPEN PROJECTIONS IN OPERATOR ALGEBRAS I 5 part with [12]. Namely, as we have just seen, [34] implicitly contains the notion of an open tripotent; and a couple of the results in [34] may thus be read as results about open tripotents (these results were not repeated in [12]). In what follows we use the notation SA for either {a ∈ A : k1 − 2ak ≤ 1}, or {a ∈ Ball(A) : k1 − ak ≤ 1}. Of course the first of these two sets is contained in the second. However the reader should choose which of these two they prefer, and stick with this choice for the rest of the paper. Note that SA is convex, and closed under multiplication by scalars in [0, 1]. We write cA for the cone R+ SA. This cone will, just as in [15], play a role for us very much akin to the role of the positive cone in a C∗-algebra. Indeed an underlying philosophy in [15] and the present paper and its sequel [13] is to use this cone to generalize important facts and theories for C∗-algebras which employ the positive cone, to more general operator algebras. The set cA is 'large': for example in a unital operator algebra it includes R+(1 + Ball(A)), which spans A. On the other hand, if A has a cai then that cai may be chosen in this cone [35], and A = Span(SA). To see the latter, note that by [15, Lemma 8.1] and the unital case just discussed, the weak* closure E of Span(SA) equals A∗∗, hence by the bipolar theorem A = Span(SA). 1 We will use throughout the fact from [15] that elements in SA have nth roots for all n ∈ N, which are again in SA. There is an explicit formula for these roots, and indeed for xt for all 0 < t ≤ 1 and k1 − xk ≤ 1 (the series in Lemma 1.2 below). If a ∈ SA, then (a n ) converges weak* to an open projection which is written as pa or s(a), and this is both the left and the right support projection of a (see [15, Section 2]). If a ∈ cA, its support projection s(a) = pa is s( a ). In this case aAa is a HSA kak of A, and the support projection of this HSA is s(a), so aAa = Apa = paA∗∗pa ∩ A. We will need some simple properties of the roots mentioned above: Lemma 1.1. If a ∈ SA then ars = (ar)s if r, s ∈ (0, 1], and aras = ar+s if r, s, r + s ∈ (0, 1]. Proof. This follows from the disk algebra functional calculus used to define these powers in [15]. (cid:3) Lemma 1.2. If a ∈ SA then a has a unique square root in SA. Indeed for n ∈ N, a has a unique nth root in SA whose numerical range is contained in the set of k(cid:1)(−1)k(1 − a)k, where t = 1 numbers reiθ with θ ≤ π n , and this is a norm limit of polynomials in a with no constant term. n . This nth root is P∞k=0 (cid:0) t Proof. This follows from a well known operator theoretic fact concerning roots (see e.g. [28, Theorem 0.1]), together with [15, Proposition 2.3]. We remark that an early version of [15] contained this explicit formula, the published version referred to P∞k=0 ck(1 − x)k without explicitly stating the values of ck. However that ck = (cid:0)t k(cid:1)(−1)k is obvious. Remark. A clarification, for our operators a, one cannot expect (a 2 = a. In 2 does not satisfy f (a∗a) = f (a∗)f (a). To other words, our 'square root' f (a) = a see this note that if a is any invertible matrix in SMn, then b = a 2 is invertible, and so if b∗bb∗b = a∗a = (b∗)2b2 then bb∗ = b∗b. So b and hence a is normal. But this is false in general. Lemma 1.3. If S ∈ B(H) with kIH−Sk ≤ 1, and if T is a contraction in B(K, H), then kIK − T ∗STk ≤ 1. (cid:3) 1 1 1 2 )∗a 1 6 D. P. BLECHER AND MATTHEW NEAL 1 Proof. IK −T ∗ST = V ∗DV , where V ∗ = [(IK −T ∗T ) S). Since V is an isometry the result is clear. 2 T ] and D = diag(IK , IH − (cid:3) We note that if a ∈ cA and n ∈ N, then it is clear that the left and right supports of an equals s(a). Lemma 1.4. If a ∈ SA and v is a partial isometry in any containing C∗-algebra B with v∗v = s(a), then vav∗ ∈ SB and (vav∗)r = varv∗ if r ∈ (0, 1) ∪ N. Proof. This is clear if r = k ∈ N. Also v(v∗v− a)kv∗ = (vv∗− vav∗)k. For r ∈ (0, 1) we have varv∗ equal to ∞ Xk=0 (cid:18)r k(cid:19)(−1)kv(1 − a)kv∗ = = ∞ Xk=0 Xk=0 ∞ (cid:18)r k(cid:19)(−1)k(vv∗ − vav∗)k (cid:18)r k(cid:19)(−1)k(1 − vav∗)k, which equals (vav∗)r, using Lemma 1.3. (cid:3) The following result will not be used here, but is of independent interest. It may be generalized to give the variant of [29, Lemma 3.12] appropriate to the setting of quotients of 'rigged modules' in the sense of [7]. Proposition 1.5. If J is a right ideal in A with a left cai (et) in SA, for example if J = aA for some a ∈ SA, and if x ∈ A then the norm of x + J in A/J equals limt k(1 − et)xk. Proof. Just as in the proof of [29, Lemma 3.12]. (cid:3) We now consider a, b ∈ SA, and are interested in aAb, an inner ideal in A. Proposition 1.6. If a, b ∈ SA, and p = pa, q = pb, then aAb = pAq = {x ∈ A : x = paq} = aA ∩ Ab = aA Ab = AaAAb. Proof. Clearly aAb ⊂ pAq. On the other hand if x ∈ pAq = pA ∩ Aq = aA ∩ Ab then a n xb 1 1 n → x. So x ∈ aAb. So aAb = pAq = {x ∈ A : x = paq} = aA ∩ Ab ⊃ aA Ab. 1 1 But of course aAb = (aAb aAb, and so AaAAb = aAb since a ∈ Aa, b ∈ Ab. 2 ⊂ aA Ab, and so aAb = aA Ab. Clearly AaAAb ⊂ (cid:3) 2 )b 2. A Pedersen type equivalence in operator algebras In order to fix the problem with the relation considered in the second paragraph of our paper, it is tempting to define a relation a ∼c b if there exist x, y ∈ Ball(B) with a = xy, yx = b. As we will see momentarily this is still not what one would want, but nonetheless this notation ∼c will be useful. Indeed ∼c is not an equiva- lence relation on SA. However it is almost an equivalence relation in the following way: a ∼c b and b ∼c d implies that a2 ∼c d2. Indeed if a = xy, yx = b = wz, d = zw then a2 = xyxy = xby = xwzy and zyxw = zbw = zwzw = d2. The relation ∼c is certainly not a definition that we will consider seriously, even if one insists on extra conditions on the support projections for x and y. Indeed even in the simple case A = M2 this definition would be wrong in our context. OPEN PROJECTIONS IN OPERATOR ALGEBRAS I 7 It differs completely from our simplest equivalence relation, Pedersen equivalence, , 1+3K√2 }, which in this case is just unitary equivalence. To see this, let x = diag{ 1√2 for some small positive K to be determined, and let y be the 2 by 2 matrix with rows 1+K√2 . Then xy is the 2 by 2 matrix with rows 1+K 2 , K and 0, 1+3K 2 , (1 + 3K)K and 0, 1+3K . For K small enough both xy and yx are in SA (indeed x and y are in SA too). Simple formulae for the norm of an upper triangular 2 by 2 matrix show that xy and yx have different norms in general, hence are not unitarily equivalent. ,√2K and 0, 1√2 , whereas yx is the 2 by 2 matrix with rows 1+K 2 2 The following seems to be a much more promising relation: 1 1 Definition 2.1. We say that a and b are root equivalent, and write a ∼r b, if n for all n ∈ N. This is not the same as a ∼c b on SA, even in the case n ∼c b a A = M2. This may be seen to be an equivalence relation on SA using the fact at the end of the first paragraph of this section. If a, b ∈ SA, with a ∼r b, consider the four-tuple or 'context' Remark. (1) 3n = yx for some x, y ∈ Ball(A), then letting (aAa, bAb, aAb, bAa). If a x′ = xyx, y′ = yxy, we have a n = y′x′, and x′ ∈ aAb and y′ ∈ bAa. Since these roots of a and b constitute cai's for Aa and Ab respectively, (aAa, bAb, aAb, bAa) is a Morita context in the sense of [11]. n = x′y′ and b 3n = xy, b 1 1 1 1 1 1 1 (2) Suppose that a n . Then n )n = xb, and similarly amx = xbm a n x = xyx = xb for all m ∈ N. So f (a)x = xf (b) for any function f in the disk algebra, and even for more general functions. In particular, pax = xpb. Similar assertions apply to y. n for some n ∈ N, so a n . Thus ax = (a n = xy, yx = b n )nx = x(b n ∼c b 1 1 1 1 1 Parts of the following theorem use operations in the containing C∗-algebra B. However it will be seen that it does not matter which containing C∗-algebra we use, since other parts such as (iii) do not use B at all. The careful reader will also notice apparent redundancy in some of the following statements and proofs; however this is necessary to obtain the first part of the last assertion of the theorem (about being equivalent with the same x and y). Theorem 2.2. Suppose that a, b ∈ SA, and let c = a 1 2 and d = b 1 2 . TFAE: (i) There exist w, y ∈ A with a = wy, b = yw and y = c. (ii) There exist x, y ∈ A with a = xy, b = yx and y = c, and x = xpb. (ii)′ There exist x, y ∈ A with a = xy, b = yx and y = c, and x∗ = c∗. (ii)′′ There exist x, y ∈ A with a = xy, b = yx and y = c,y∗ = d∗,x = d,x∗ = c∗. (ii)′′′ There exist x, y ∈ A with a = xy, b = yx and x = cR and y = Sc for some contractions R, S in A∗∗ (or if one prefers, in B∗∗). (iii) For all n ∈ N, there exist xn, yn ∈ Ball(A) with a n = ynxn, (iv) There exists a partial isometry v ∈ B∗∗ with pa = v∗v, pb = vv∗, and (iv)′ There exists v ∈ ∆(A∗∗) with pa = v∗v, and va ∈ A, and b = vav∗. and the sequence (yna) has a norm convergent subsequence. va, av∗ ∈ A and b = vav∗. n = xnyn, b 1 1 These imply that kak = kbk. The partial isometry v in (iv) above may be chosen to be open and in ∆(A⊥⊥). Also, the clauses (ii),(ii)′,(ii)′′, (ii)′′′ are equivalent to each other with the same x and y, and these imply that x ∈ aAb, y ∈ bAa. 8 D. P. BLECHER AND MATTHEW NEAL Proof. Note that (iv) clearly implies that kak = kbk. (ii)′′′ ⇒ (ii)′ Given (ii)′′′, a = xy = cRSc, we may assume that S = Spa and paR = R. So cRSa = ca and so cRS = cRSpa = c. Thus RS = paRS = pa. It follows by e.g. [30] that S is a partial isometry and R = S∗, so that y∗y = c∗S∗Sc = c∗pac = c∗c. Similarly, xx∗ = cc∗. (ii)′ ⇒ (ii)′′′ Obvious. (ii) ⇒ (ii)′′ Assume that there exist x, y ∈ A with a = xy, b = yx and y = c. Then y = r(y)c = vc, where v = r(y)r(c)∗. Hence y = ypa, so that r(y)∗r(y) ≤ pa. However r(y)∗r(y) ≥ pa since a = xy, so r(y)∗r(y) = pa. Similarly, r(y)r(y)∗ ≥ pb. Note that v∗v = r(c)par(c)∗ = pa. So v is a tripotent. Also, vv∗ = r(y)par(y)∗ = r(y)r(y)∗ ≥ pb, and we will see momentarily that this is an equality. Note that xy = xvc = c2 implies xva = ca, and hence xv = xvpa = cpa = c. Since y = vc we have va = vc2 = yc = yxv = bv. Hence y = vc = dv. It is now clear that pby = y, so that pb ≥ r(y)r(y)∗, giving pb = r(y)r(y)∗ = vv∗. Also yy∗ = dvv∗d∗ = dpbd∗ = dd∗. We now see that x = cv∗, assuming that x = xpb (since then cv∗ = xvv∗ = xpb = x). Thus xx∗ = cc∗. By symmetry to the case for y, we must have x∗x = c∗c. (ii) ⇒ (iv) We use facts from the proof above, noting that va = vc2 = yc ∈ A, av∗ = c2v∗ = cx ∈ A, and va = vc2 = d2v = bv, so that b = vav∗. We now prove the assertions at the end of the theorem, starting from (ii), and using the notation and facts already established above. To see that v is open, consider Rr(c)∗ , right multiplication by r(c)∗, on the TRO hyi generated by y. This is a one-to-one ternary morphism, since yr(c)∗r(c) = ypa = y. We show it also maps n ∈ yr(c)∗aA ⊂ B. into B: note yr(c)∗c = yc ∈ B, so yr(c)∗a ∈ B, hence yr(c)∗a n → c∗ by 2.1.6 However yr(c)∗ = r(y)cr(c)∗ = r(y)c∗. Now c∗(a in [10]. Hence yr(c)∗ = r(y)c∗ = limn r(y)c∗a n ∈ B. So Rr(c)∗(hyi) ⊂ B. Hence v, the image under this ternary morphism of r(y), is open, since r(y) is open. As we saw two paragraphs back, va ∈ A and av∗ ∈ A. By Lemma 1.2 we n v∗ ∈ A and so n ∈ A. In the weak* limit, v = vpa ∈ A∗∗. Similarly, a n )∗ → c∗, so c∗a 1 1 1 1 1 1 have va v∗ = pav∗ ∈ A∗∗. So v ∈ ∆(A∗∗). By Proposition 1.6, x ∈ pa Apb = aAb. Similarly, y ∈ bAa. (ii)′ ⇒ (ii) As in the proof above, we have v = r(y)r(c)∗ is a tripotent with y = vc, v∗v = pa, vv∗ = r(y)r(y)∗ = pb. By symmetry, w = r(c)∗r(x) is a tripotent with x = cw, ww∗ = pa, w∗w = r(x)∗r(x) = pb. Now a = xy = cwvc, and by the idea in the proof of (ii)′′′ ⇒ (ii)′ above, it follows that w = v∗, so that x = cv∗ = cv∗pb = xpb. (iv) ⇒ (ii)′ Set x = cv∗, y = vc. These are in A by Lemma 1.2. The rest is easy 2 2. The proof that (i) ⇒ (ii)′ Let x = wyw, y′ = ywy. Then y′2 = a∗y∗ya = a (ii) implies (ii)′′ and (iv) above, with a, b replaced by a3, b3, yields that x = a 2 v, y′ = 2 for a partial isometry v satisfying v∗v = s(a) = s(a3), v∗v = s(b) = s(b3), and va va3 = b3v, which implies that vc = dv. Since c ∈ aA and aA = ccA ⊂ caA ⊂ c3A, from facts in the proof of the last two implications. 3 3 3 vc ∈ vaA ⊂ vc3A ⊂ vc3A = y′A ⊂ A. Similarly, cv∗ ∈ A. Then a = cv∗vc, b = dvv∗d = vc2v∗ as desired in (ii)′. 2n v∗, then 2n , xn = a n . As we said above, (ii) ⇒ (i) Trivial. (ii) ⇒ (iii) Using the notation and proof above, if yn = va these are in A by considerations used above, and xnyn = a 1 1 1 OPEN PROJECTIONS IN OPERATOR ALGEBRAS I 9 va = bv, hence van = bnv, and so va Lemma 1.2. Hence ynxn = b 2n vv∗b 1 1 n = b n . 1 2n = b 1 1 n v by considering the power series in 1 1 3 n = ynxnyn = b (iii) ⇒ (iv)′ Note that yna n yn so that yna = byn. Similarly, axn = xnb. Replacing yn by ynxnyn and xn by xnynxn as before, but still calling n = xnyn where xn = paxn, yn = ynpa, and we still have them yn and xn, we have a yna = byn and axn = xnb, and (yna) still has a convergent subsequence. Suppose that xnk → S and ynk → T weak, and ynk a → T a in norm. We have S = paS and T = T pa. Then xnk ynk a → ST a = paa = a. As in the proof of (ii)′′′ ⇒ (ii)′, this yields ST = ST pa = pa, and S∗ = T . Thus T is a partial isometry v ∈ ∆(A∗∗), with v∗v = pa. Since ynk a → T a in norm, va ∈ A. Since yna = byn and axn = xnb, it follows that bynk xnk = ynk axnk → vav∗, so that b = vav∗. Hence pb ≤ vv∗, and in fact they are equal because pbvv∗ = vv∗ since pbyn = yn. Thus we have (iv)′. (iv)′ ⇒ (iv) By Lemma 1.4, b 1 n = (vav∗) 1 pb = lim n b 1 n = limn va 1 n = va n v∗. Then n v∗ = vpav∗ = vv∗, 1 where these are weak limits. If va ∈ A then vBp = vaBa ⊂ B, where p = pa. Hence Bpv∗ ⊂ B, so that av∗ ∈ B ∩ A∗∗ = A. (cid:3) Definition 2.3. Let a, b ∈ SA. We say that a is Pedersen equivalent to b in A, and write a ∼A b, if the equivalent conditions in the last theorem hold. We will see some other characterizations of this variant of Pedersen equivalence later in our paper. Sometimes we write a ∼A b via x and y, if Theorem 2.2 (ii)′′ (or equivalently (ii)) holds. Similarly if n ∈ N we write an ∼A bn if (ii)′′ in Theorem 2.2 holds with a and b replaced by an, bn. We will see that this happens iff a ∼A b. If a, b ∈ cA with kak = kbk, then we say that a ∼A b if some positive scalar multiple of a is Pedersen equivalent in the sense above to the same multiple of b. It should be admitted at the outset that our Pedersen equivalence (which gener- alizes the C∗-algebra variant -- see Remark (2) below) happens more rarely in general operator algebras than it does in C∗-algebras. However this is similar, for example, to the fact that right ideals with left contractive approximate identities are rarer in general operator algebras than in C∗-algebras, but are still interesting (they are related to important topics, and have a compelling theory -- see e.g. [9, 15]. While some algebras may have none, in other algebras they can be key for some pur- poses). Results such as Proposition 3.9, Theorem 4.9, and Theorem 5.1 show that our Pedersen equivalence happens quite naturally. Remark. (1) Of course if a, b ∈ Ball(A+) in the case A is a C∗-algebra, then a ∼A b iff there exists x ∈ A with a = xx∗, b = x∗x (the usual Pedersen equivalence). This is quite obvious; for example if a ∼A b, with notation as in Theorem 2.2 (ii)′′, then y∗y = b If A is a C∗-algebra then unitary equivalence of elements a, b ∈ SA implies Indeed if b = uau∗ for a unitary u in the multiplier algebra of A (or a ∼A b. simply a unitary in A∗∗ such that au∗, ua ∈ A), then Ran(b) = uRan(a). Thus upa = pbupa. If we set x = a 2 = a. Also, yx = uau∗ = b, and xx∗ = a 2 = b and yy∗ = a 2 u∗ and y = ua 2 then xy = a 2 u∗u(a 2 u∗ua 2 = a. 2 a 2 b 1 1 1 1 1 1 1 1 1 (2) From Theorem 2.2 (iii) we see that in a finite dimensional operator algebra 1 2 )∗. So a ∼A b. A, a ∼A b iff a ∼r b. Here a, b ∈ SA. We do not know if this is true in general. (3) The relation ∼A in the case that A = Mn is exactly unitary equivalence of elements of SMn . The one direction of this follows from (1). For the other 10 D. P. BLECHER AND MATTHEW NEAL direction, if a ∼A b then we can get b = w∗aw for a tripotent w that the reader can easily check may be replaced by a unitary. Unitary equivalence and similarity are the same for positive matrices, but this is false for SMn (even for 2 × 2 upper triangular matrices). (4) We mention three other interesting ways to state Pedersen equivalence: (a) Theorem 2.2 (ii)′′ is equivalent with the same x and y to the same condition, but with x∗ = c∗ replaced by x = d. Clearly this is implied by (ii)′′, and conversely it implies (ii), since x = r(x)x = r(x)d = r(x)r(d)∗d, which gives x = xpb. (b) a ∼A b iff there exist x, y ∈ A with a = xy, b = yx, and for each n ∈ N there exist contractions xn, yn with x = c1− 1 n . The xn, yn can be chosen in A if one wishes. To see that this is equivalent to (ii)′′′, in the one direction simply let xn = c n . For the other direction, if subnets xnk → R and ynk → S weak, then (ii)′′′ holds. This is because by the continuity of the disk algebra functional calculus defining powers in [15, Section 2], it is easy to argue that c1− 1 n xn, y = ync1− 1 n R, yn = Sc 1 1 (c) a ∼A b iff a ∼r b with some kind of 'compatibility' between the x and y elements for different roots. That is, for all n ∈ N one may write a n = r + 1 ynxn, where xn, yn satisfy some conditions like ykxℓ = yrxs whenever 1 s ; or like ykxℓym = yrxsyt whenever 1 t ≤ 1. To see for example that a ∼A b is implied by the last condition, together with a couple of other reasonable conditions such as xn, yn are contractions and yn ∈ bAa: Note 4 . It follows that y2nx2ny2 = b (yna) converges, so Theorem 2.2 (iii) holds. Conversely, if Theorem 2.2 (ii) holds, let xn = a 2n . As in an argument in the proof of Theorem 2.2, these are in A, xnyn = a 2n y2 converges to y2, but it also equals ynx4y4 = yna n = xnyn, b ℓ = 1 2n v∗, yn = va n , ynxn = b m = 1 ℓ + 1 r + 1 s + 1 k + 1 n , and k + 1 1 1 1 1 1 1 1 1 n → c in norm. yrxsyt = va 1 1 1 2r a ℓ + 1 m = 1 2s v∗va r + 1 2t = va s + 1 t ≤ 1. 1 2 ( 1 r + 1 s + 1 t ). Thus ykxℓym = yrxsyt if 1 k + 1 (5) Note that pa and pb are the left and right support projections of the element x in Theorem 2.2 (ii) (and the variants of (ii)). The choice of the square root, as opposed to a different root, in (i) and the variants of (ii) in Theorem 2.2 is not essential; one can instead take the rth power and the (1 − r)th power, for any r ∈ (0, 1). For example: Proposition 2.4. If a, b ∈ SA then a ∼A b iff there exist x, y ∈ A with a = xy, b = yx and y = ar for some r ∈ (0, 1) and x = xpb; or iff there exist x, y ∈ A with a = xy, b = yx and y = ar and x∗ = (a1−r)∗ for some r ∈ (0, 1). Proof. Exactly as in the proof of Theorem 2.2, but replacing c by ar. For example, given (iv) in Theorem 2.2, let y = var, x = a1−rv∗, then it is easy to see, by the same arguments as in the proof of Theorem 2.2 that x, y ∈ A with a = xy, b = yx, and y∗y = (ar)∗v∗var = (ar)∗paar = ar2, so that y = ar, and x = xpb. Conversely, if x, y ∈ A with a = xy, b = yx and y = ar, and x = xpb, then all of the arguments in the first paragraph of the proof that (ii) ⇒ (ii)′′ go through with c replaced by ar. In the second paragraph of that proof, one sees that xvar = a1−rar, hence xva = a1−ra, so that xv = a1−r. Also va = bv, since va = vara1−r = ya1−r = yxv = bv. One can modify the lines that follow in the proof we are mimicking, to get y = var = brv, pb = vv∗, and yy∗ = ar(ar)∗. Thus a1−rv∗ = xvv∗ = x, and so OPEN PROJECTIONS IN OPERATOR ALGEBRAS I 11 xx∗ = a1−r(a1−r)∗. This all is more than enough to obtain (iv) in Theorem 2.2, indeed va = vara1−r = ya1−r ∈ A, av∗ = ara1−rv∗ = arx ∈ A. Since va = bv, we have b = vav∗. (cid:3) Proposition 2.5. Suppose that a, b ∈ SA and r ∈ (0, 1) ∪ N. Then a ∼A b iff ar ∼A br. Proof. If a ∼A b, then using notation from the proof of Theorem 2.2 above, let y′ = vcr, x′ = crv∗. Note that y′ = vcr ∈ vaAa ⊂ A since y = vc ∈ A. Similarly x′ ∈ A. Then x′y′ = c2r = ar, and y′x′ = vc2rv∗ = varv∗ = br. The latter follows by Lemma 1.4 and Lemma 1.3. So ar ∼A br. 2 , so that br = varv∗. Let x = a The converse is very similar, suppose that ar ∼A br via a 2 . Then x, y ∈ A and xy = a as before, and 2 v∗, y = va 2 v∗, va 1 1 r r 1 r = b, yx = vav∗ = ((vav∗)r) r = (varv∗) r = (br) 1 1 as can be easily argued following the idea in the last paragraph. Corollary 2.6. ∼A is an equivalence relation on SA. Proof. We leave this as an exercise using e.g. Theorem 2.2 (iv). (cid:3) (cid:3) Theorem 2.7. Suppose that a ∈ SA, and let c = a a = xy, and y = c and x∗ = c∗, then b = yx is in SA, and a ∼A b. Proof. Using the notation and proof of (ii)′ ⇒ (ii) in Theorem 2.2 above, we have y = vc, and hence vv∗b = b, and x = cw and bw∗w = b. Following that proof, we see that w = v∗ again, so that b = vv∗bvv∗. So 2 . If there exists x, y ∈ A with 1 k1 − 2bk = kvv∗ − 2vv∗bvv∗k ≤ k1 − 2v∗yxvk = k1 − 2c2k = k1 − 2ak. Similarly, k1 − bk ≤ k1 − ak. So b = yx is in SA, and a ∼A b. (cid:3) 3. Open tripotents for operator algebras and Peligrad-Zsid´o equivalence The partial isometry or tripotent v that occurs in Theorem 2.2 is central to our study of the above variant of Pedersen equivalence. It also happens to be what we will call a ∗-open tripotent. In the present section we define this notion and develop the associated theory. Later we will apply this theory back to Pedersen equivalence, and to coarser notions of equivalence and subequivalence. In the introduction we discussed the notion from [12] of an open tripotent in B∗∗, and gave several equivalent definitions of these objects. We will also use below the following characterization, which is proved below in the Appendix (Proposition 6.1): A partial isometry v in B∗∗ is an open tripotent in the sense of [12] iff p = v∗v and q = vv∗ are open projections in the sense of Akemann, and vBp ⊂ B and v∗Bq ⊂ B. As we said in the introduction, first, p and q are said to be Peligrad- Zsid´o equivalent in this case, and second, the conditions involving q in the last line are automatic and are stated for the sake of symmetry. We define a ∗-open tripotent (or ∗-open partial isometry) in A∗∗, for an operator algebra A, to be an open tripotent (in the sense of [12]) in B∗∗, which lies in ∆(A⊥⊥). Here B is a C∗-algebra generated by A. Lemma 3.1. If v is an open tripotent in B∗∗ then so is v∗. Hence if v is a ∗-open tripotent in A∗∗ then so is v∗. 12 D. P. BLECHER AND MATTHEW NEAL Proof. We leave this as an exercise. (cid:3) Being a ∗-open tripotent makes sense independently of the particular C∗-algebra B generated by A, since one may define '∗-openness' in A∗∗ purely in terms of A: u is a ∗-open tripotent in A∗∗ iff u is a tripotent in the C∗-algebra ∆(A∗∗), and the projection u = 1 2 (cid:20) uu∗ u∗ u u∗u (cid:21) in M2(∆(A∗∗)), is an open projection in M2(A)∗∗ in the sense of [9] (that is, there exists a net at ∈ M2(A) such that at = uat = at u → u weak in M2(A)∗∗). This is also equivalent to v being open in M2(A)∗∗ in the sense of [9], where v = u∗. Remark. (1) Open projections in A∗∗ in the sense of [9], are clearly ∗-open tripotents in A∗∗. This follows from the definition above since open projections in B∗∗ are open tripotents in the sense of [12]. (2) It is easy to see how to define a ∗-open tripotent in the second dual of a strong Morita equivalence bimodule X in the sense of [11], and a matching Peligrad-Zsid´o equivalence for open projections in the second duals of the algebras acting on the left and right of X. Most of the results below about ∗-open tripotents and Peligrad- Zsid´o equivalence will be almost identical in this more general setting, with almost identical proofs, and we leave this to the reader. This will then simultaneously generalize much of the theory in the present paper, and the theory of open tripotents in [12]. We have chosen not to present this here, since it is often better to present a more concrete theory than to present the most general abstract formulation. We do remark however that many results in this more general theory are derivable from the case we consider by taking A = L(X), the linking algebra of X. We say that two projections p, q ∈ A∗∗ are Peligrad-Zsid´o equivalent in A∗∗, and write p ∼P Z,A q if there exists an open tripotent (in the sense of [12]) v ∈ B∗∗, which lies in ∆(A⊥⊥) (hence is ∗-open), implementing a Peligrad-Zsid´o equivalence of p and q in B∗∗. That is, v∗v = p, vv∗ = q (since v is open this forces p, q to be open projections [12]). Remark. If p ∼B,P Z q it does not follow that p ∼P Z,A q necessarily. A simple counterexample is the diagonal matrix units in the case A is the upper triangular 2 by 2 matrices. To connect our definition to Pelegrad and Zsid´o's original equivalence, as we did in the selfadjoint case (see Proposition 6.1), we have: Lemma 3.2. Two projections p, q ∈ A∗∗ are Peligrad-Zsid´o equivalent in A∗∗ iff p, q are open projections and there exists v ∈ B∗∗ such that v∗v = p, vv∗ = q, and vAp ⊂ A and v∗Aq ⊂ A. This is also equivalent to: p is open and there exists v ∈ ∆(A∗∗) such that v∗v = p, vv∗ = q, and vAp ⊂ A. Proof. If p ∼P Z,A q via v then as we said above, p, q are open. Also by Proposition 6.1 from the Appendix, vAp ⊂ vBp ⊂ B ∩ A⊥⊥ = A, and similarly v∗Aq ⊂ A. Conversely, suppose that p, q are open and that vAp ⊂ A and v∗Aq ⊂ A. Then vet ∈ A if (et) is a cai for Ap. Taking a weak* limit, vp = v ∈ A⊥⊥. Similarly, v∗ ∈ A⊥⊥, so v ∈ ∆(A⊥⊥). Now (et) is also a cai for Bp by [9, Theorem 2.9], so it follows that for any c ∈ Bp we have vc = limt vetc ∈ AC ⊂ B. That is, vBp ⊂ B. Similarly, v∗Bq ⊂ B, so by Proposition 6.1 from the Appendix, v implements a OPEN PROJECTIONS IN OPERATOR ALGEBRAS I 13 Peligrad-Zsid´o equivalence of pa and pb in B∗∗. If p is open and vAp ⊂ A then the argument above gives vBp ⊂ B. So by [34, Lemma 1.3], q is open and v∗Bq ⊂ B. Thus if v∗ ∈ A⊥⊥ then v∗Aq ⊂ B ∩ A⊥⊥ = A. (cid:3) Remark. Under the equivalent conditions of the Lemma, we also have Aqv ⊂ Bqv = (v∗Bq)∗ ⊂ B, so Aqv ⊂ B ∩ A⊥⊥ = A. Similarly, Apv∗ ⊂ A. Theorem 3.3. Let u be a tripotent in A⊥⊥, and let Au = {a ∈ A : a = uu∗au∗u}. Suppose that u is in the weak* closure of Au, and that Auu∗ ⊂ A. Then u is an open tripotent in B∗∗. If, further, u∗ ∈ A⊥⊥ then u is a ∗-open tripotent in A∗∗, u∗Au ⊂ A, and the weak* closure of Au equals {η ∈ A∗∗ : η = uu∗ηu∗u} = uu∗A∗∗u∗u. Conversely, if u is a ∗-open tripotent in A∗∗ then u is in the weak* closure of Au, and Auu∗ ⊂ A. Proof. We have uA∗uAu ⊂ B. As explained in the introduction, B∗∗2 (u) is a C∗- algebra with product xu∗y and involution ux∗u. In this product and involution the C∗-algebra generated by uA∗uAu is contained in B. The increasing cai for this C∗-algebra converges to a projection in the C∗-algebra B∗∗2 (u), hence to a tripotent w ∈ B∗∗. Since wu∗w = w, we have w ≤ u in the ordering of tripotents. It is an exercise to check that w satisfies the definition from [12] of an open tripotent in B∗∗. If x, y ∈ Au we have ux∗y = wu∗ux∗y = ww∗ux∗y = wx∗y. Replacing x with xt ∈ Au such that xt → u weak, we have uu∗y = wu∗y = ww∗y. Replacing y with xt in the limit we have u = uu∗u = ww∗u = w. So u is open in B∗∗. If further u∗ ∈ A⊥⊥ then u∗Au ⊂ u∗Bu ⊂ B ∩ A∗∗ = A. Clearly u is ∗-open in A∗∗. The weak closure of Au is contained in uu∗A∗∗u∗u clearly. Conversely, if p = u∗u, which is open in A∗∗ by [12], and η ∈ uu∗A∗∗u∗u, then u∗η ∈ pA∗∗p, so there is a net (at) in pAp converging weak* to u∗η. Then uat → η weak, and uat ∈ A, indeed uat ∈ Au. So the weak closure of Au is uu∗A∗∗u∗u. For the converse, if u is a ∗-open tripotent in A∗∗ then by the fact above the theorem there exists a net xt ∈ Au with xt → u weak. If x ∈ Au then x ∈ Bu. In the notation of [12], Buu∗ = B(u)u∗ ⊂ B by facts in [12]. Then xu∗ ∈ Buu∗∩A∗∗ ⊂ B ∩ A∗∗ = A. (cid:3) In the following, SAu is computed with respect to operator algebra Au with the Peirce product. That is, SAu = {a ∈ Au : ku − Kak ≤ 1}, where K = 1 or 2. Lemma 3.4. If u is a ∗-open tripotent in A∗∗ then Au is a subalgebra of the C∗- algebra B(u) with its 'Peirce product', and it has a cai with respect to this product. If u, w are ∗-open tripotents in A∗∗, then TFAE: (i) u ≤ w in the ordering of tripotents, (ii) Au is a subalgebra (indeed is a HSA) of Aw in the product above, (iii) SAu ⊂ SAw as sets, (iv) u ≤ w. Indeed a tripotent in A∗∗ which is dominated by w in the ordering of tripotents, is ∗-open in A∗∗ iff it is an open projection in the operator algebra A∗∗w (with its Peirce product). Moreover, u = w iff Au = Aw as sets and as algebras (with the product above). 14 D. P. BLECHER AND MATTHEW NEAL Proof. That Au is a subalgebra of B(u) follows from the result Auu∗ ⊂ A above. Since the second dual of Au is its weak closure uu∗A∗∗u∗u, which contains u which is an identity in the 'Peirce product', it follows from e.g. [10, Proposition 2.5.8] that Au has a cai. Thus uw∗u = u, so u ≤ w. (iii) ⇒ (i) Taking second duals, and using [15, Lemma 8.1] we have SA∗∗ If u ≤ w then it is clear that Au ⊂ Aw as sets. (i) ⇒ (ii) If x, y ∈ Au then xw∗y = xu∗uw∗y = xu∗y, so Au is a subalgebra of Aw. We leave it as an exercise that it is a HSA. If Au is a subalgebra of Aw then (Au)∗∗ is a subalgebra of (Aw)∗∗. (ii) ⇒ (i) (i) ⇔ (iv) is in [12, Proposition 2.7]. (ii) ⇒ (iii) If (ii) holds then by (i) u may be viewed as a projection in (Aw)∗∗ dominated by the identity w. Let x ∈ Au ⊆ Aw with u − 2x ≤ 1. Then w − 2x = w − u + u − 2x ≤ max{1,u − 2x} ≤ 1. w . u ⊂ SA∗∗ By the fact mentioned in the introduction that cA densely spans A, we have that A∗∗u ⊂ A∗∗w . In particular, kw − 2uk ≤ 1 since ku − 2uk ≤ 1. We wish to show that uw∗u = u, or equivalently that u is a projection in the C∗-algebra C = B∗∗2 (w). Working inside the latter C∗-algebra we have w = 1. Thus it suffices to prove that any partial isometry V in a unital C∗-algebra C, with k1− 2V k ≤ 1, is a projection. Since V ∈ SC, its left support and right supports agree, so V ∗V = V V ∗ equals a projection p = s(V ). If D = pCp then V is a unitary in D with kp − 2V k ≤ 1. By spectral theory it follows that V = p. This completes the proof of (i). Given an open projection p ∈ (Aw)⊥⊥, we recall that Aw is a subalgebra of B(w) ⊂ B∗∗2 (w). So by [9, Theorem 2.4], p is open in B∗∗2 (w). By [12, Corollary 2.11], p is an open tripotent u in B∗∗. We also have that u∗ = p∗ = w∗pw∗ ∈ A∗∗. So u is a ∗-open tripotent in A∗∗. If u ≤ w, then u ∈ (Aw)⊥⊥ ⊂ B∗∗2 (w). By [12, Corollary 2.11], u is open as a projection in B∗∗2 (w). Since Aw is a subalgebra of B(w) and u ∈ (Aw)⊥⊥, u is open as a projection in (Aw)⊥⊥ by [9, Theorem 2.4]. (cid:3) Conversely, any ∗-open tripotent u ∈ A∗∗ is open as a tripotent in B∗∗. Remark. If Au = Aw as subsets of A, it need not follow that u = w. This may be seen by considering for example any two distinct unitaries in K(H)∗∗ = B(H). Indeed Au = Aw iff u∗u = w∗w and uu∗ = ww∗. To see the harder direction of this, if Au = Aw then in the second duals we see that u ∈ B∗∗2 (w), so that u∗u ≤ w∗w and uu∗ ≤ ww∗. Similarly, w∗w ≤ u∗u and ww∗ ≤ uu∗. From Sections 2 and 3 of [12] one can write down a host of properties and behaviors of open tripotents in A∗∗. For example: Lemma 3.5. An increasing net of ∗-open tripotents in A∗∗ has a least upper bound tripotent in A∗∗, namely its weak* limit, and this limit tripotent is ∗-open in A∗∗. Proof. Clear from [12, Proposition 2.12] and the characterization of ∗-open tripo- tents in A∗∗ as open tripotents in B∗∗ which are also in the von Neumann algebra ∆(A∗∗). (cid:3) Lemma 3.6. If u and v are ∗-open tripotents in A∗∗, which are both dominated by some tripotent in B∗∗, then there exists a smallest tripotent u ∨ v in B∗∗ which dominates both u and v, and this tripotent is a ∗-open tripotent in A∗∗. OPEN PROJECTIONS IN OPERATOR ALGEBRAS I 15 Proof. Of course u and v are open tripotents in B∗∗. That there exists a smallest tripotent u ∨ v in B∗∗ which dominates both u and v is well known (see e.g. [12, Lemma 3.7]). By [12, Lemma 3.7], [u ∨ v = u ∨ v. However u and v are open projections in M2(∆(A∗∗)), so we deduce that [u ∨ v is an open projection and is in M2(∆(A∗∗)), so that u ∨ v is a ∗-open tripotent in A∗∗. (cid:3) Lemma 3.7. A family {ui} of ∗-open tripotents in A∗∗, which are all dominated by a tripotent in B∗∗, has a least upper bound amongst the tripotents in B∗∗, and this is a ∗-open tripotent in A∗∗. Proof. Of course the ui are open tripotents in B∗∗, so by [12, Proposition 3.8] the family has a least upper bound v amongst the tripotents in B∗∗, and this is an open tripotent in B∗∗. The supremum of any finite subfamily of {ui} is a ∗-open tripotent by Lemma 3.6, and such supremums form an increasing net with least upper bound v. so v is a ∗-open tripotent in A∗∗ by Lemma 3.5. (cid:3) 1 We now rephrase Pedersen equivalence in terms of ∗-open tripotents. We can rephrase a special case of Lemma 3.2 as follows. Let v ∈ B∗∗ with v∗v = pa for some a ∈ SA with va ∈ A. Then vBpa = vaBa ⊂ B, and so v is an open tripotent in B∗∗, and v = limn va n ∈ A∗∗. In this case, v is a ∗-open tripotent in A∗∗ iff v∗ ∈ A∗∗, and iff av∗ ∈ A. Proposition 3.8. If a, b ∈ SA then a ∼A b iff there exists a ∗-open tripotent v with v∗v = pa, vv∗ = pb and b = vav∗. Also, in this case the x and y in Theorem 2.2 (ii) are in bijective correspondence with such ∗-open tripotents v satisfying all the conditions in the last line. The bijection here takes v to y = vc (and x = cv∗), and its inverse takes y to r(y)r(c)∗. Here c = a 2 . 1 Proof. The first assertion is clear from Theorem 2.2 (iv) and Lemma 3.2 (and the fact that va ∈ A implies that v aAa ⊂ A). To finish we show that the If y is as in Theorem 2.2 (ii) two maps described are inverses of each other. we have r(y)r(c)∗c = r(y)c = r(y)y = y. Conversely, if v is a ∗-open tripo- tent satisfying all the conditions in the first line of the theorem statement, then vc = vr(c)c = vr(c)vc. Now vr(c) is a tripotent, with left support vs(c)v∗ = vv∗ and right support r(c)∗v∗vr(c) = pa. By the 'uniqueness of the polar decomposi- tion' we deduce that vr(c) = r(vc). Thus r(vc)r(c)∗ = vr(c)∗r(c) = vpa = v. This completes the proof. (cid:3) Proposition 3.9. Let v be a ∗-open tripotent in A∗∗, and suppose that p = v∗v is the support projection for some a ∈ SA. Then q = vv∗ is the support projection for some b ∈ SA, and a ∼A b. Proof. Let b = vav∗. As in [15], aAa = Ap since p = s(a). Note that b = 2 v∗ ∈ vApApv∗ ⊂ Aq, using the Remark after Lemma 3.2. Also, b ∈ SA va and (vav∗) n v∗ by Lemma 1.4. As at the end of the proof of Theorem 2.2, pb = vv∗ = q. That a ∼A b now follows from Proposition 3.8. (cid:3) Remark. A similar result: suppose that a ∈ SA, and that for all n ∈ N, there exist xn, yn ∈ Ball(A) with a n = xnyn, and the sequence (yna) has a norm convergent subsequence with limit va say, where v is a weak* cluster point of (yn). Then v is a ∗-open tripotent, v ∈ ∆(A∗∗), and b = vav∗ ∈ SA with a ∼A b. To prove this, suppose that xnk → S and ynk → v weak, and ynk a → va in norm. As usual n = va 2 a 1 1 1 1 1 16 D. P. BLECHER AND MATTHEW NEAL we may assume that yn = ynpa, xn = paxn, and S = paS and v = vpa. Then xnk ynk a → Sva = paa = a, so that Sv = Svpa = pa. This forces S∗ = v, and v is a tripotent in ∆(A∗∗). Since ynk a → va in norm, va ∈ A. As in Lemma 3.2 above, vBp ⊂ B, where p = pa. By [34, Lemma 1.3], if q = vv∗ then q is open and v∗Bq ⊂ B. So v is open, and a 2 ∈ A. So by Proposition 3.9 we have b = vav∗ ∈ SA and a ∼A b. 4. Hereditary bimodules 2 v∗ ∈ Bpv∗ ∩ A⊥⊥ = A, and similarly va 1 1 Hereditary bimodules, defined below, as well as being interesting in their own right, will be important for us for two main reasons. First, they are useful in understanding Pedersen and Blackadar equivalence: if one examines the algebraic structure of aAb when a and b are equivalent in these ways, then one is led naturally to hereditary bimodules. (Indeed we give several characterizations of these equiv- alences in terms of hereditary bimodules in this section and the next.) Secondly, hereditary bimodules will turn out to be much the same thing as ∗-open tripotents u, via the correspondence u 7→ Au, with Au defined as in the previous section. One important definition of open tripotents from [12] is as the 'support tripotents' of what was called an inner C∗-ideal. Hereditary bimodules are the analogue of inner C∗-ideals in our setting, and we shall see in Proposition 4.1 below that ∗-open tripotents are the same as 'support tripotents' of hereditary bimodules. We define a hereditary bimodule to be the 1-2 corner of a HSA in M2(A), where that HSA is isomorphic to M2(C) for an approximately unital operator algebra C, via a 'corner preserving' completely isometric homomorphism ρ : M2(C) → M2(A). Thus we have subalgebras D, E of A, and subspaces X, Y of A, such that the subalgebra L of M2(A) with rows C, X and Y, D, is a HSA in M2(A), and there is a completely isometric isomorphism ρ from M2(C) onto L taking each of the corners D, E, X, Y of L onto the copy of C in the matching corner of M2(C). In this case we say that X is a hereditary bimodule, or a hereditary D-E-bimodule, and we call L the linking algebra of the hereditary bimodule. It then follows easily that (D, E, X, Y ) is a strong Morita equivalence context in the sense of [11], that Y is the 'dual module' X considered in that theory, X = Y , and that L is the linking algebra in the sense of [11] for this strong Morita equivalence. Conversely, given a strong Morita equivalence context (D, E, X, Y ) whose linking algebra L is a HSA in M2(A), then saying that L is isomorphic to M2(C) in the way described above, is, by [8, Corollary 10.4], the same as saying that the 1-2 corner X is linearly completely isometric to an approximately unital operator algebra. That is, given a strong Morita equivalence context (D, E, X, Y ) whose linking algebra L is a HSA in M2(A), then X is a hereditary bimodule iff X is linearly completely isometric to an approximately unital operator algebra. We point out that D = XY and E = Y X above. These are HSA's in A. But also X = DAE and Y = EAD, and these are inner ideals in A (that is, XAX ⊂ X and similarly for Y ). So X and Y are retrievable from D and E. To see this, note that clearly X = DXE ⊂ DAE, and conversely DAE = XY AY X ⊂ X since X is an inner ideal. So X = DAE, and similarly Y = EAD. In the notation above, note that a hereditary bimodule X is an approximately unital operator algebra in the product transferred from C via ρ. The weak limit in A∗∗ of the cai for X will be called a support tripotent of X (in the proof of the next result it will be seen to be a tripotent). This support tripotent is not unique, as we OPEN PROJECTIONS IN OPERATOR ALGEBRAS I 17 discussed for example in the remark before Lemma 3.5, but can be easily be made so similarly to the case in [12] by keeping track of one further peice of structure on a hereditary bimodule, such as its 'product' (or one of the other items in Lemma 3.4). We define a principal hereditary Aa-Ab-bimodule, or principal hereditary bimodule for short, to be a hereditary Aa-Ab-bimodule X for some a, b ∈ SA. We have X = aAb by Proposition 1.6. The linking algebra L of the bimodule is the subalgebra of M2(A) with rows aAa, aAb and bAa, bAb. Note that this L is certainly a HSA in M2(A). Remark. For any a, b ∈ SA, the subalgebra of M2(A) with rows aAa, aAb and bAa, bAb, is a HSA in M2(A) clearly. However it will not in general correspond to a strong Morita equivalence. Indeed, if X = aAb, there may be no HSA in M2(A) with X as its 1-2 corner which corresponds to a strong Morita equivalence. Obvious counterexamples exist in the case A is the upper triangular 2 by 2 matrices. Proposition 4.1. If u is a ∗-open tripotent in A∗∗, then Au is a hereditary bimod- ule. Conversely, any hereditary bimodule equals Au for a ∗-open tripotent u; and u is a support tripotent of this hereditary bimodule. Proof. If u is a ∗-open tripotent in A∗∗ then by Lemma 3.1 so is v = u∗. Let p = v∗v, q = vv∗. The subalgebra L of M2(A) with rows pAp, pAq and qAp, qAq, is a HSA of M2(A). Note pAq = Au. Since Au is an approximately unital operator algebra in the 'Peirce product', it is a hereditary bimodule by the definition of the latter. Explicitly, if S = diag(1, v), then ℓ 7→ S∗ℓS is a completely contractive homomorphism from L to M2(pAp), whose completely contractive inverse is k 7→ SkS∗. For the converse, we use the notation in our definition and discussion of 'hered- itary bimodule' at the start of Section 4. Let e be the identity of C∗∗. Note that ρ∗∗ : M2(C∗∗) → M2(A∗∗) is a completely isometric corner preserving homorphism too, and it restricts to an injective ∗-homomorphism from M2(C e) into ∆(L∗∗). Note that the identity of L∗∗ is p + q where p is the support projection of D and q is the support projection of E. Also p and q correspond via ρ∗∗ to e ⊕ 0 and 0 ⊕ e in M2(C∗∗). Let u be the image in A∗∗ of e under the 1-2 corner of ρ∗∗. Then v = u∗ is the image in A∗∗ of e under the 2-1 corner of ρ∗∗. Also, u is a tripotent in X⊥⊥, and vv∗ = q, v∗v = p (since the matching facts are true in M2(C e)). Also, uAp ⊂ A, since the matching products are in C, and ρ(C) ⊂ M2(A). So u is a ∗-open tripotent in A∗∗ by Lemma 3.2 (one could also use Theorem 3.3 to see this). Since a cai for C converges weak* to e, it is clear that u is a support tripotent in the sense above. (cid:3) The following is an intrinsic characterization of hereditary bimodules: Theorem 4.2. An inner ideal X in A is a hereditary bimodule, or equivalently is of the form Au for a ∗-open tripotent u, if and only if there are nets of contractions cn ∈ X and dn ∈ A, such that (1) cndnx → x and xdncn → x in norm for all x ∈ X, (2) (dncm) and (cmdn) are norm convergent with n for fixed m, and are norm convergent with m for fixed n. If these hold then a subnet of the cn converge weak* to a support tripotent u for X, so that X = Au. 18 D. P. BLECHER AND MATTHEW NEAL Proof. (⇒) Suppose that u is a ∗-open tripotent, so that Au is a hereditary bimod- ule, isomorphic to an approximately unital operator algebra C. Then since these conditions clearly hold in C, by applying the isomorphism ρ in the definition of 'hereditary bimodule' we see that they will hold too in M2(A). (⇐) Replacing by subnets, we may suppose that cn, dn, cndn, and dncn, con- verge in the weak* topology to say elements u, w, p, q, respectively. Then cndncm converges weak* to uwcm = cm = pcm. Taking the limit over m we see that uwu = u = pu, so (uw)2 = uw = puw. Since uw is a contraction, it is a projection. Also, cndncmdm converges weak* to uwcmdm = cmdm = pcmdm. Taking the limit over m we see that uwp = p = p2. So p is a projection too, and now we see that p = uw. Also u = pu, and similarly wp = limn dnp = w. It follows that u = w∗, and this is a partial isometry, and is in ∆(A∗∗). Similarly, q = u∗u. Condition (1) implies in the limit that px = x = xq for x ∈ X, so X ⊂ pAq. Hence u is in the weak* closure of Au. Conversely, if y ∈ pAq then u∗yu∗ ∈ A⊥⊥, and so since X⊥⊥ is an inner ideal we see that y = uu∗yu∗u ∈ X⊥⊥ ∩ A = X. So X = pAq = Au. Since (xdmcmdn) converges with n for fixed m, for x ∈ X, we have xdmcmu∗ ∈ A. Taking the limit over m shows that xu∗ ∈ A. By Theorem 3.3 we see that u is a ∗-open tripotent. (cid:3) Remark. Hereditary bimodules generalize the hereditary subalgebras of [9] (called HSA's). Indeed, HSA's are just the case that cn = dn in the characterization above. If cn = dn in the characterization above, then u = u∗, so that p = q in the proof above, and X = pAp is a HSA in A. Conversely given a HSA with cai (en), the conditions above are met with cn = dn = en. It is clear in the definition of a hereditary bimodule X at the start of this section that D ∼= E completely isometrically as operator algebras, and D ∼= X ∼= Y canonically as operator algebras. In fact these isomorphisms are easy to write down in terms of the support tripotent u for X: Proposition 4.3. Let X be a hereditary bimodule with a support tripotent u. In the notation at the start of this section we have D ∼= E completely isometrically via the homomorphism a 7→ u∗au, and X ∼= Y completely isometrically via the linear isometry x 7→ u∗xu∗ def= x♯. Similarly, D ∼= X via right multiplication by u, and D ∼= Y via left multiplication by u∗. Proof. That D = pAp ∼= qAq = E via this map was done in the first paragraph of the proof of Proposition 4.1; as were the D ∼= X and D ∼= Y assertions. Similar arguments prove the other assertion: If X = Au for a ∗-open tripotent u ∈ ∆(A∗∗), then the 'dual module' Y = X (see the definition of a hereditary bimodule and the accompanying discussion there), is simply vXv where v = u∗. Indeed as we saw in the proof of Proposition 4.1, Xu∗ = Auu∗ = Auu∗ , and similarly u∗Auu∗ = Au∗ = Y . Thus we can define a complete isometry from X onto Y by x 7→ x♯ = vxv. The inverse of this is of course the map y 7→ uyu. (cid:3) Remark. (1) One should beware though in the last result: the isomorphisms mentioned there are not unique, but depend on the particular support tripotent u chosen for X. They are, however, determined for example by the cone R+ SAu in Lemma 3.4 (iii). (The nonuniqueness of the support tripotent is discussed a few paragraphs before Proposition 4.1.) OPEN PROJECTIONS IN OPERATOR ALGEBRAS I 19 (2) Consider the bijection between ∗-open tripotents v and the elements x and y in the variants of item (ii) in Theorem 2.2, described in Proposition 3.8. One may go further by saying that the x and y in Theorem 2.2 (ii) and its variants, are nothing but the 'square root of a with respect to the tripotent' v. That is, for example, via the canonical isomorphisms Aa ∼= aAb and Aa ∼= bAa from Proposition 4.3 (but with u = v∗), x and y both correspond to a 2 . Of course y = x♯ in the language of Proposition 4.3. 1 We can now give several characterizations of Pedersen equivalence in terms of hereditary bimodules: Corollary 4.4. If a, b ∈ SA then a ∼A b if and only if aAb is a hereditary bimodule and there exists x ∈ aAb such that a = x x♯, b = x♯ x. Here x♯ = u∗xu∗ is as in Proposition 4.3, where u is a ∗-open tripotent with aAb = Au. If these hold then aAb is indeed a principal hereditary Aa-Ab-bimodule. Proof. (⇒) If a ∼A b, and u is as in Proposition 3.8, and v = u∗, then Au = aAb by Proposition 1.6. By Proposition 4.1 this is a hereditary bimodule. The rest is clear from Remark (2) above. (⇐) Follows from Proposition 3.8, since if v = u∗ then vav∗ = vxvxvv∗ = x♯ x = b. If these hold, then as we said aAb is a hereditary bimodule. We know from [15] that Aa = pAp and Ab = qAq, where p = pa, q = pb. Also, aAb = pAq by Proposition 1.6, and similarly, bAa = qAp. So aAb is a principal hereditary Aa-Ab- bimodule. (cid:3) Corollary 4.5. If a, b ∈ SA and aAb is a hereditary bimodule, with nets cn, dn as in Theorem 4.2, then a ∼A b if and only if we can write a = xy, b = yx, where y ∈ bAa and x ∈ aAb, and ycn − dnx → 0 in norm. Proof. For the one direction, if u is in the proof of Theorem 4.2, then yu = u∗x because ycn − dnx → 0. So y = x♯ and the rest follows from Corollary 4.4. For the other direction, if a ∼A b then a = xy, b = yx where x = a1/2v∗ = v∗b1/2 and y = va1/2 = b1/2v for open tripotents v, v∗ ∈ A⊥⊥. One can let cn = a1/2nv∗, dn = va1/2n, and the rest is easy. (cid:3) Remark. (1) If A = Mn then saying that aAb is a hereditary bimodule is simply saying that rank(a) = rank(b). So we can view aAb being a hereditary bimodule as a generalization of this rank condition. If we reduce the conditions in Corollary 4.5 to say: (a = xy, b = yx where aAb is a hereditary bimodule, x ∈ aAb, and y ∈ bAa), then one obtains a relation that is in general strictly stronger (by Corollary 5.2) than the (variant of) 'Blackadar equivalence' which we study later, and is strictly weaker in general than Pedersen equivalence. (2) Another characterization of Pedersen equivalence: If a, b ∈ SA then a ∼A b iff there exists a principal hereditary Aa-Ab-bimodule, such that a ⊕ 0 and 0 ⊕ b in the linking algebra of the bimodule correspond via the map ρ mentioned in the first few paragraphs of Section 4, to c⊕ 0 and 0⊕ c for some element c ∈ C. For the one direction of this, in Corollary 4.4 we saw aAb is a principal hereditary bimodule. If v = u∗ are as in the proof of Corollary 4.4, with v∗bv = a, and if ρ−1 is the map in the first paragraph of the proof of Proposition 4.1, then ρ−1(1 ⊕ b) = 1 ⊕ (v∗bv) = 1 ⊕ a. For the other direction, by the proof of Proposition 4.1, aAb = Au for a 20 D. P. BLECHER AND MATTHEW NEAL ∗-open tripotent with uu∗ = pa, u∗u = pb. If ρ is as in the second paragraph of that proof, and if ρ(c ⊕ 0) = a ⊕ 0, then by hypothesis 0 ⊕ b = ρ(0 ⊕ c). However, ρ(0 ⊕ c) = ρ∗∗(E21(c ⊕ 0)E12) = ρ∗∗(E21)(a ⊕ 0)ρ∗∗(E12) = 0 ⊕ vav∗. Thus b = vav∗. So (iv) in Theorem 2.2 holds. Definition 4.6. If u is a ∗-open tripotent in A∗∗ and x ∈ cAu, where the latter denotes R+ SAv , then the nth roots of x (with respect to the Peirce product on Au), converge weak* to an open projection su(x) ∈ B∗∗2 (u). By Lemma 3.4, su(x) is a ∗-open tripotent in A∗∗, and su(x) ≤ u in the ordering on tripotents. We call su(x) the u-support tripotent of x. We also call su(x) a local support tripotent. Lemma 4.7. Let u be a ∗-open tripotent in A∗∗ and x ∈ cAu. The hereditary bimodule As corresponding to the local support tripotent s = su(x) is a principal hereditary bimodule, indeed may be written as aAb for some a, b ∈ SA with a ∼A b. Proof. By [15, Corollary 2.6], working within the algebra Au, and letting s = su(x) and v = u∗, we have xvAuvx = {x ∈ Au : x = svxvs = ss∗xs∗s} = {x ∈ A : x = ss∗xs∗s} = As. Also xv = xvsv = xvss∗sv = xs∗sv = xs∗, and similarly vx = s∗x. Let a = xs∗, b = s∗x = s∗as. Since s is ∗-open in A∗∗, if p = ss∗ then right multiplication by s∗ is a completely isometric homomorphism from As onto pAp, which extends to a completely isometric homomorphism from (As)⊥⊥ onto pA∗∗p. It follows that pa = ss∗. Similarly pb = s∗s. Then a and b are in SA, since for example k1 − ak = kuu∗ − xu∗k ≤ ku − xk ≤ 1. Clearly as = x ∈ A, so a ∼A b by Theorem 2.2 (iv)′. Clearly, xvAuvx = aAub ⊂ aAb. To see the converse, first note that aA = a2A. Indeed, squaring the power series in Lemma 1.1 shows that a ∈ a2A and the statement is thus clear. Similarly Ab = Ab2. So the latter since uu∗x = x and xu∗u = x. So xu∗Auu∗x = aAb, and we are done. (cid:3) aAb = aAAb = a2AAb2 ⊂ aAub, Remark. It is easy to see that aAb in the last proof, actually equals aAsu(x), and also equals Asu(x)b. Theorem 4.8. (1) Every hereditary bimodule in A is the closure of an increas- ing net of principal hereditary bimodules of the form aAb, where a, b ∈ SA with a ∼A b. (2) Any ∗-open tripotent u in A∗∗ is the limit of an increasing net of local support tripotents. Proof. (1) Viewing Au as an operator algebra with the Peirce product, by [15, Theorem 2.15] Au is the closure of an increasing net (Dt) of 'peak principal' HSA's, each of the form xtAuxt (with respect to the Peirce product in Au), for some xt ∈ SAu. As we saw in Lemma 4.7, we may write each Dt as aAb for some a, b ∈ SA with a ∼A b. OPEN PROJECTIONS IN OPERATOR ALGEBRAS I 21 (2) In the above, su(xt) forms an increasing net of open projections in (Au)⊥⊥ which converge weak* to u. Hence by Lemma 3.4 they are local support tripotents, and converging weak* to u. (cid:3) Theorem 4.9. Any separable hereditary bimodule in A is a principal hereditary bimodule, that is of the form aAb, where a, b ∈ SA with a ∼A b. Proof. If Au is separable, then viewing it as an operator algebra with respect to the Peirce product, we have by [15, Corollary 2.18] that Au = xAux with respect to the Peirce product, for some x ∈ SAu. By Lemma 4.7, Au = aAb for some a, b ∈ SA with a ∼A b. (cid:3) Remark. It follows as in [15, Corollary 2.18] that the cai for a separable heredi- tary bimodule may be chosen to be countable, and consist of mutually commuting elements in the Peirce product. We are not sure what implications this may have. Corollary 4.10. If A is a separable operator algebra, then every ∗-open tripotent in A∗∗ is a local support tripotent su(x) for some x ∈ SA. Proof. In the last proof, we have Au = Asu(x), with su(x) a tripotent dominated by the ∗-open tripotent u. By Lemma 3.4, u = su(x). (cid:3) We say that a hereditary bimodule D in A is a hereditary subbimodule of another hereditary bimodule D′ in A, if D ⊂ D′, if the Peirce product of D coincides with the restriction of the Peirce product of D′, and DD′D ⊂ D in the Peirce product of D′. Proposition 4.11. If a hereditary bimodule D in A is generated by a countable collection of principal hereditary subbimodules Dn of D, (so in the Peirce product on D there is no proper HSA containing all the Dn), then D is a principal hereditary bimodule in A. Proof. This follows from [15, Theorem 2.16 (2)] in the same way that the last several results followed from matching results in [15]. (cid:3) For interests sake we give a generalization of Theorem 4.9 to a slightly more general class of inner ideals in A. Similarly to our first definition of a hereditary bimodule at the start of this section, suppose that we have subalgebras D, E of A, and subspaces X, Y of A, such that (D, E, X, Y ) is a strong Morita equivalence context in the sense of [11], and the linking algebra of this context, namely the subalgebra L of M2(A) with rows C, X and Y, D, is a HSA in M2(A). Then we say that X is an inner equivalence bimodule, or an inner equivalence D-E-bimodule. Note that again D = XY, E = Y X, X = DAE, Y = EAD, and D and E are HSA's and X and Y are inner ideals in A. These follow by the same reasons as before. A principal inner equivalence bimodule, or a principal inner equivalence Aa- Ab-bimodule, is the case when the context is (Aa, Ab, aAb, bAa) for some a, b ∈ SA. Remark (1) before Theorem 2.2 essentially says that if a ∼r b then aAb is a principal inner equivalence Aa-Ab-bimodule. Theorem 4.12. Any inner equivalence bimodule in A which is also separable, is a principal inner equivalence bimodule. Proof. Let (D, E, X, Y ) is as above, and suppose that the inner equivalence bi- module X is separable. Then the ternary envelope Z = T (X) is separable (see e.g. 22 D. P. BLECHER AND MATTHEW NEAL [8, 10] for a discussion of the ternary envelope, sometimes called the triple envelope). By [8, Lemma 10.2], the TRO Z is isomorphic to a canonical C∗-algebraic Morita equivalence bimodule Z containing X. By facts in e.g. p. 407 of [7], Z∗ contains Y , ZZ∗ contains D, and Z∗Z contains E. So D, E, Y are each separable. By [15, Theorem 2.16 (1)], D = Aa, E = Ab for some a, b ∈ SA. Thus X = DAE ⊂ aAb, and conversely aAb ⊂ DAE = X since a ∈ D, b ∈ E. So X = aAb, and similarly Y = bAa. (cid:3) 5. Blackadar equivalence is equivalent to aA = bA, and also equivalent to aAa = bAb. For a, b ∈ SA we define a ∼= b if s(a) = s(b). By the theory in [9, Section 2] this If a, b ∈ cA we define a ∼s b if there exist a′, b′ ∈ SA, with a ∼= a′, a′ ∼A b′, and b′ ∼= b. It will follow from the next result that this is an equivalence relation. We say that a and b are Blackadar equivalent in A if a ∼s b. Theorem 5.1. (Cf. [29], [31, Proposition 4.3].) For a, b ∈ cA TFAE: (i) a ∼s b. (ii) aA ∼= bA completely isometrically via a right A-module map. (iii) pa ∼P Z,A pb. (iv) There exists b′ ∈ cA, with a ∼A b′ and b′ ∼= b. (v) There exists a′ ∈ cA, with a ∼= a′ ∼A b. (vi) aAb is a principal hereditary Aa-Ab-bimodule. Proof. We may assume that a, b ∈ SA. aAf is a principal hereditary bimodule, but f Af = bAb. (iv) ⇒ (vi) There exists f ∈ cA with a ∼A f and f ∼= b. By Proposition 4.4, (vi) ⇒ (iii) By Proposition 4.1 we have X = aAb = Au for a ∗-open tripotent u ∈ A⊥⊥. Note that pax = x for all x ∈ X, so that pau = u. Hence pauu∗ = uu∗, so uu∗ ≤ pa. Similarly, u∗u ≤ pb. However since a ∈ Aa = XY , we have uu∗a = a, so that uu∗ ≤ pa. Hence uu∗ = pa and similarly u∗u = pb. Of course ua ∈ A by definition of a ∗-open tripotent. (iv) ⇒ (i) Trivial. (ii) ⇒ (iv) If Φ : aA → bA is a surjective completely isometric right A-module map, then by [7, Corollary 3.7] there exists a surjective completely isometric left A-module map Ψ : Aa ∼= Ab such that Ψ(y)Φ(x) = yx for all x ∈ aA, y ∈ Aa. We obtain using [7, Theorem 6.8], a C∗-module isomorphism B ⊗hA Aa → B ⊗hA Ab. Now the multiplication map m : B ⊗hA Aa → Ba has an asymptotic contractive left inverse θn : z 7→ z ⊗ a n . Indeed 1 1 1 θn(m(b ⊗ x)) = bx ⊗ a n = b ⊗ xa n → b ⊗ x. 1 1 for all b ∈ B, x ∈ Aa. So m is isometric, and so Ψ extends to a C∗-module map Ba → Bb which is unitary onto its range. It follows from e.g. [10, Corollary 8.1.8] that y∗y = (a 2 ), 2 ) = a. By Theorem 2.7, b′ = yx ∈ cA and a ∼A b′. and xy = Ψ(a Finally, b′A = yxA ⊂ yA. On the other hand, y = limn ya n ∈ yaA ⊂ yxA, since a = xy. So 2 ). Similarly, xx∗ = a 2 , where y = Φ(a 2 )∗ if x = Ψ(a 2 )∗a 2 )Φ(a 1 2 ) = a 1 2 a 1 2 (a 1 1 1 1 1 1 b′A = yA = Φ(a 1 2 )A = Φ(a 1 2 A) = bA. OPEN PROJECTIONS IN OPERATOR ALGEBRAS I 23 (i) ⇒ (iii) Thus a ∼A b′ ∼= b. If a ∼A b and v is as in the proof of Theorem 2.2 above, then v, v∗ ∈ A⊥⊥ as we said in Theorem 2.2. Thus pa ∼P Z,A pb via v (for example, v∗b = v∗dd = xd ∈ A so v∗Bq = v∗bBb ⊂ B). Since pa = pa′ if a ∼= a′ (iii) is now clear. (iii) ⇒ (ii) If pa ∼PZ,A pb via a partial isometry v, let Φ(x) = vx and Ψ(x) = xv∗. These are the desired completely isometric module maps in (ii) (note that Φ(aA) = vaA ⊂ vApA ∈ A ∩ qB∗∗ = bA, and similarly Ψ(bA) ⊂ aA). with (iv). The proof of the equivalence with (v) is similar to the proof of the equivalence (cid:3) Remark. For algebras without any nontrivial r-ideals all elements of cA are Black- adar equivalent to each other of course. Such algebras are discussed in [15]. Corollary 5.2. If a, b ∈ cA, and X = aAb and Y = bAa then TFAE: (i) pa ∼P Z,A pb. (ii) There are nets of contractions cn ∈ X, and dn ∈ A satisfying (2) of Theo- rem 4.2, and also cndna → a and bdncn → b in norm. (iii) X is a hereditary bimodule and also a ∈ XY, b ∈ Y X. Proof. (ii) ⇒ (i) These conditions guarantee by Theorem 4.2 that X is a hereditary bimodule, and X = Au for a ∗-open tripotent u ∈ A⊥⊥. As in the proof of Theorem 5.1, uu∗ ≤ pa and u∗u ≤ pb. However, the last condition in (ii), together with a part of the proof of Theorem 4.2, guarantees that uu∗a = a and bu∗u = b. So uu∗ = pa and u∗u = pb. (i) ⇒ (iii) Follows from Theorem 5.1. (iii) ⇒ (ii) Follows from Theorem 4.2. We now give some other conditions that imply Blackadar equivalence in a C∗- (cid:3) 1 1 n x = xb algebra. Theorem 5.3. Suppose that a, b ∈ cA for an operator algebra A. If a = xy and b = yx for some x, y ∈ A, then pa ∼P Z pb in B∗∗. Proof. Without loss of generality, by dividing by a suitable scalar, we may assume n for all that a, b ∈ SA. Under these conditions ax = xyx = xb, and hence a n ∈ N (as in Remark (2) before Theorem 2.2). In the limit, px = xq and x∗p = qx∗, where p = pa and q = pb. It follows that pxx∗x = xx∗xq, and similarly for fivefold and sevenfold products of this type, etc. Thus pr(x) = r(x)q, where r(x) is the range tripotent of x in Z∗∗, because r(x) is a weak* limit of linear combinations of such 'odd products' (see e.g. the proof of [12, Lemma 3.3]). Since p is a limit of functions of xy, clearly r(x)r(x)∗p = p. Thus w = p r(x) is a tripotent and it is easy to see that ww∗ = p and w∗w = q. Note that w∗a = r(x)∗xy = xy ∈ B. So w∗ pBp = w∗aBa ⊂ B. By [34, Lemma 1.3] we have pa ∼P Z pb. (cid:3) It is not necessarily true that if a = xy and b = yx for some x, y ∈ A, then a and b are Blackadar equivalent in A, if A is nonselfadjoint. For a counterexample, let R be an invertible operator on a Hilbert space H, and let A be the span in M2(B(H)) of a = IH ⊕ 0, b = 0 ⊕ IH , x = E12 ⊗ R, y = E21 ⊗ R−1. Then a = xy and b = yx, but a and b are not Blackadar equivalent in A if R is chosen appropriately. This counterexample also shows that the conditions considered in (2) of the next result also do not characterize Blackadar equivalence in A if A is nonselfadjoint. 24 D. P. BLECHER AND MATTHEW NEAL Proposition 5.4. Let A be an approximately unital operator algebra, suppose that a, b ∈ cA, and let B be a C∗-algebra containing A as usual. Then b are Blackadar equivalent in B. (1) Suppose that for some x ∈ A we have aA = xA, and Ab = Ax. Then a and (2) If a and b are Blackadar equivalent in A then there exists an x ∈ A such (3) a and b are Blackadar equivalent in B if and only if there exists an x ∈ B that aA = xA, and Ab = Ax. such that aA = xA, and Ab = Ax. If these conditions hold then x ∈ aA ∩ Ab, so Proof. (1) Let p = pa, q = pb. px = pxq = xq. Since a ∈ xA we have r(x)r(x)∗p = p, and the proof of (1) then follows as in the previous theorem. (2) These follow for example by the same reasoning used in the proof of Theorem 5.1 to show that yA = b′A = bA. (3) This is now obvious. (cid:3) Remark. (1) If A is a C∗-algebra then the last result gives an alternative charac- terization of Peligrad-Zsid´o/Blackadar equivalence. One may ask what is the least restrictive extra condition we can add to these to characterize our variant of Black- adar equivalence in a more general operator algebra. For example, one such extra condition is that aAb is a hereditary bimodule (see Corollary 5.2 (iii)). (2) Another relation equivalent to the one considered in the last result, is that Aa x = x Ab for some x ∈ A, and a ∈ xA and b ∈ Bx. This also characterizes Peligrad-Zsid´o/Blackadar equivalence in a C∗-algebra. In the remainder of this section, we turn to the topic of subequivalence. For any a, b ∈ cA we have as in 4.4 in [31], that a ∈ Ab iff Aa ⊂ Ab iff aA ⊂ bA iff pa ≤ pb. Also (a ∈ Ab and b ∈ Aa) iff a ∼= b iff aA = bA iff pa = pb. Lemma 5.5. Suppose that a ∈ cA, and that Φ, Ψ are a pair of completely con- tractive A-module maps from aA and Aa into A, such that Ψ(x)Φ(y) = xy for all x ∈ Aa, y ∈ aA. Then Φ, Ψ are completely isometric, and there exists b ∈ cA with a ∼A b and Ran(Φ) = bA and Ran(Ψ) = Ab. Proof. Let F, E be the ranges of these two maps, which are respectively a right and a left ideal in A. By applying the ⊗hA tensor product with B as in the proof that (ii) ⇒ (iv) in Theorem 5.1 above, we may extend Φ, Ψ to contractive B-module maps 2 , then y∗y ≤ c∗c by from aB and Ba into B. If Φ(c) = y, Ψ(c) = x, where c = a e.g. 8.1.5 in [10], and similarly xx∗ ≤ cc∗. By basic operator theory, y = Sc, x = cR for contractions S, R. By the proof that (ii)′′′ implies (ii)′ in Theorem 2.2, we actually have y = c,x∗ = c∗. Thus Φ(cz)∗Φ(cw) = z∗y∗yw = (cz)∗(cw), for all z, w ∈ A, which forces Φ to be a complete isometry. Similarly Ψ is a complete isometry. Setting b = yx, then b ∈ cA by Theorem 2.7, and a ∼A b. Also, bA = yxA ⊂ yA, and conversely, ya n ∈ yaA ⊂ bxA ⊂ bA, so y ∈ bA and yA ⊂ bA. Thus yA = bA. Hence F = Φ(aA) = Φ(cA) = yA = bA. Similarly, E = Ab. (cid:3) If a, b ∈ cA, we define a -s b if there exists b′ ∈ Ab such that a ∼s b′. This is clearly equivalent to: there exists b′ ∈ Ab such that a ∼A b′. We will call this Blackadar comparison in A. If p, q are open projections in A∗∗ we say that p -P Z,A q if there is an open projection q′ ≤ q in A∗∗ with p ∼P Z,A q′. We will call this Peligrad-Zsid´o subequivalence in A∗∗. 1 1 OPEN PROJECTIONS IN OPERATOR ALGEBRAS I 25 The following is the version of [31, Proposition 4.6] in our setting (see also [29]): Proposition 5.6. If a, b ∈ cA, TFAE: (i) a -s b. (ii) pa -P Z,A pb. (iii) There exist a pair of completely contractive A-module maps Φ : aA → bA and Ψ : Aa → Ab, such that Ψ(x)Φ(y) = xy for all x ∈ Aa, y ∈ aA. Proof. (iii) ⇒ (i) By Lemma 5.5, (iii) implies that there exist x, y ∈ Ball(A) with a = xy, b′ = yx ∈ cA. In our case, b′ = yx ∈ Ab too. (i) ⇒ (ii) By Theorem 5.1, pa ∼P Z,A pb′, and clearly pb′ ≤ pb. (ii) ⇒ (iii) If we have an open projection q′ ≤ pb in A∗∗ with pa ∼P Z,A q′ via a tripotent v, then by Proposition 3.9 we have q′ = vv∗ = pb′ where b′ = vav∗. Left multiplication by v is a completely isometric module map from aA into A. Note that pbva = pbvv∗va = pbpb′ va = pb′ va = va. So the map maps into pb A = bA. Similarly, right multiplication by v∗ is a completely isometric module map from Aa into Ab. The rest is clear. (cid:3) As in [31, Proposition 4.6], these give x, y ∈ Ball(A) such that aA = xyA and yxA ⊂ bA, and so on. Here yx ∈ cA. Remark. In contrast to Theorem 5.1 (ii), one needs the statement about Ψ in (iii) of Proposition 5.6. It is not automatic, as one may see by considering for example multiplication by z on the disk algebra. 6. Appendix: Some related results for TROs We include the following results about TROs for the reason that, as we claimed in Remark (2) at the start of Section 3, many of our results generalize easily to the case that the operator algebra A is replaced by a strong Morita equivalence bimodule in the sense of [11]. The results below show how this can proceed in the case of TROs. If v is a tripotent in the second dual of a TRO Z such that p = v∗v and q = vv∗ are open projections with respect to Z∗Z and ZZ∗ respectively, and if v(ZZ∗)p ⊂ Z, v∗(ZZ∗)q ⊂ Z∗, then we say that v implements a TRO Peligrad- Zsid´o equivalence of open projections. This is of course equivalent to v, thought of as in the second dual of the linking C∗-algebra of Z, implementing a Peligrad-Zsid´o equivalence in the original sense between 0⊕ p and q ⊕ 0. So again we need not say v∗(ZZ∗)q ⊂ Z∗ in this definition. The following proof shows that the two 'inclusion conditions' are equivalent to: v∗ (qZp) ⊂ Z∗Z and v (pZ∗q ) ⊂ ZZ∗. The following (which we should have seen in [12]), is the appropriate variant of the equivalence (i) ⇔ (vi) in [16]: Proposition 6.1. If v is a tripotent in the second dual of a TRO Z, then v im- plements a TRO Peligrad-Zsid´o equivalence of open projections iff v is an open tripotent in the sense of [12]. Proof. (⇒) Under these conditions, note that D = Z(v) = Z∗∗2 (v) ∩ Z = {b ∈ Z : qbp = b} 26 D. P. BLECHER AND MATTHEW NEAL is an inner ideal in Z. Here p = v∗v, q = vv∗. It is also a C∗-subalgebra of Z∗∗(v). Indeed and v∗D = v∗DD∗D ⊂ Z∗D ⊂ Z∗Z vD∗ = vD∗DD∗ ⊂ ZD∗ ⊂ ZZ∗, hence Dv∗D ⊂ Z ∩ Z∗∗(v) = D, and vD∗v = vD∗DD∗v ⊂ ZZ∗DZ∗Z ⊂ Z, so Dv∗D ⊂ Z ∩ Z∗∗(v) = D. If B = Z∗Z then Bp = {b ∈ B : pbp = b} is a HSA with support projection p, and v∗D = Bp (clearly v∗D ⊂ Bp and conversely if b ∈ Bp then vb ⊂ Z ∩ Z∗∗2 (v) = D, so b ∈ v∗D). So left multiplication Lv∗ by v∗ is a linear complete isometry from D onto Bp, and is a ternary isomorphism, and indeed it is a ∗-isomorphism with respect to the v-product on D. It follows that D∗D = Bp, and similarly DD∗ = (ZZ∗)q, although these are clear directly: D∗D = D∗vv∗D = B∗pBp = Bp. Suppose that xt is a positive cai in the C∗-algebra D with the v-product, with weak* limit w. This will be an open tripotent in the sense of [12] dominated by v. Then v∗xt is a positive cai in Bp, so that in the limit we have v∗w = p. Since wp = w and vp = v, it follows from a basic operator theory fact that v = w. So v is open in the sense of [12]. (⇐) If v is open, then by the remark after Corollary 2.11 in [12], p and q If x ∈ Z(v) then v∗vx∗x = x∗x ⊂ Z∗Z. Since vx∗x is the generic are open. positive element in the C∗-algebra Z(v), it follows that v∗Z(v) ⊂ Z∗Z. Similarly, Z(v)v∗ ⊂ ZZ∗. We note that the HSA Z(v)Z(v)∗ of ZZ∗ has support projection q, by the first part of the proof, and similarly Z(v)∗Z(v) has support projection p. So v∗ (ZZ∗)q = v∗Z(v)Z(v)∗ ⊂ Z∗Z Z(v)∗ ⊂ Z∗, and similarly v (ZZ∗)p ⊂ Z. (cid:3) Proposition 6.2. Let Z be a TRO, set A = ZZ∗, B = Z∗Z, and let v be an open tripotent in Z∗∗, and suppose that p = v∗v is the support projection for some b ∈ SB. Then q = vv∗ is the support projection for some a ∈ SA, and in this case there exist x, y ∈ Ball(Z) with x ∈ qZp, y ∈ pZq, and xy = a, yx = b. Moreover, aAa and bBb are ∗-isomorphic C∗-algebras. They are also strongly Morita equivalent via the equivalence bimodule aZb = bZ∗a∗, and this bimodule is ternary isomorphic to bBb. Proof. As in Proposition 3.3 of [31], let a = vbv∗. Note that a = vb vBp(vBp)∗ ⊂ Aq. Also, a ∈ SA by Lemma 1.3, and (vbv∗) 1.4. Then n = vb 1 2 b 2 v∗ ∈ n v∗ by Lemma 1 1 1 pa = lim n (vbv∗) 1 n = lim n vb 1 n v∗ = vpbv∗ = q, where these are weak* limits. Hence q is the support projection of a. If x = vb and y = b 2 v∗, then xy = a, yx = b since b 1 1 1 2 2 ∈ Bp. Continuing as in [15], bBb = Bp since p = s(b), and similarly aAa = Aq. These are C∗-algebras in this case. Also aZb ⊂ qZp, and, conversely, we have qZp = Aq qZpBp ⊂ aZb. So aZb = qZp, and similarly bZ∗a = pZ∗q . So we have a C∗- algebraic Morita equivalence between aAa and bBb, implemented by the equivalence bimodule aZb = bZ∗a∗. The last assertions follow from the first paragraph of Proposition 6.1. Note that z 7→ vzv∗ is a ∗-isomorphism from Ap onto Bq, with inverse the map z 7→ v∗zv. (cid:3) OPEN PROJECTIONS IN OPERATOR ALGEBRAS I 27 Proposition 6.3. Any separable inner ideal D in a TRO Z is of the form aZb for a ∈ (ZZ∗)+, b ∈ (Z∗Z)+. If D is also ternary isomorphic to a C∗-algebra then this can be done with pa ∼P Z pb. Proof. Let A = ZZ∗, B = Z∗Z. If a ∈ A+, b ∈ B+ then aZb is an inner ideal. Conversely, if D is an inner ideal in Z, then DD∗ is a HSA in A, which is separable if D is separable. Hence DD∗ = aAa for some a ∈ A+. Similarly, D∗D = bBb for some b ∈ B+. Thus D = DD∗DD∗D = aZb (since aZb ⊂ D(D∗ZD∗)D ⊂ D because D is an inner ideal). If D is also ternary isomorphic to a C∗-algebra C then the identity in C∗∗ corresponds to an open tripotent u ∈ D∗∗ by [12, Proposition 3.5], and it is easy to argue that uu∗ (resp. u∗u) is the support projection of DD∗ (resp. D∗D). So uu∗ (resp. u∗u) is the support projection of a (resp. b). (cid:3) Acknowledgements: The second author was supported by Denison University. References [1] C. A. Akemann, The general Stone-Weierstrass problem, J. Funct. Anal. 4 (1969), 277 -- 294. [2] C. A. Akemann, Left ideal structure of C ∗-algebras, J. Funct. Anal. 6 (1970), 305 -- 317. [3] C. A. Akemann and G. K. Pedersen, Facial structure in operator algebra theory, Proc. London Math. Soc. 64 (1992), 418 -- 448. [4] P. Ara, F. Perera, and A. S. Toms, K-Theory for operator algebras. Classification of C∗- algebras, in: Aspects of operator algebras and applications, 171, Contemp. Math., 534, Amer. Math. Soc., Providence, RI, 2011. [5] W. B. Arveson, Subalgebras of C ∗-algebras, Acta Math. 123(1969), 141 -- 224. [6] B. Blackadar, Operator algebras. Theory of C ∗-algebras and von Neumann algebras, En- cyclopaedia of Mathematical Sciences Vol. 122, Operator Algebras and Non-commutative Geometry, III, Springer-Verlag, Berlin, 2006. [7] D. P. Blecher, A generalization of Hilbert modules, J. Funct. Anal. 136 (1996), 365 -- 421. [8] D. P. Blecher, Multipliers, C ∗-modules, and algebraic structure in spaces of Hilbert space operators, in: Operator algebras, quantization, and noncommutative geometry: A centennial celebration honoring J. von Neumann and M. H. Stone, Contemp. Math., 365, Amer. Math. Soc., 2004. [9] D. P. Blecher, D. M. Hay, and M. Neal, Hereditary subalgebras of operator algebras, J. Operator Theory 59 (2008), 333 -- 357. [10] D. P. Blecher and C. Le Merdy, Operator algebras and their modules -- an operator space approach, Oxford Univ. Press, Oxford (2004). [11] D. P. Blecher, P. S. Muhly, and V. I. Paulsen, Categories of operator modules (Morita equiv- alence and projective modules), Mem. Amer. Math. Soc. 681 (2000). [12] D. P. Blecher and M. Neal, Open partial isometries and positivity in operator spaces, Studia Math. 182 (2007), 227 -- 262. [13] D. P. Blecher and M. Neal, Open projections in operator algebras II: Compact projections, Preprint 2010. [14] D. P. Blecher, K. Kirkpatrick, M. Neal, and W.Werner, Ordered involutive operator spaces, Positivity 11 (2007), 497 -- 510. [15] D. P. Blecher and C. J. Read, Operator algebras with contractive approximate identities, J. Funct. Anal. 261 (2011), 188 -- 217. [16] D. P. Blecher and W.Werner, Ordered C ∗-modules, Proc. London Math. Soc. 92 (2006), 682 -- 712. [17] K. T. Coward, G. A. Elliott, C. Ivanescu, The Cuntz semigroup as an invariant for C- algebras, J. Reine Angew. Math. 623 (2008), 161193. [18] C. M. Edwards, F. J. Fernandez-Polo, C. S. Hoskin, and A. M. Peralta, On the facial structure of the unit ball in a JB∗-triple, J. Reine. Angew. Math. 641 (2010), 123 -- 144. [19] C. M. Edwards, K. McCrimmon, and G. T. Ruttimann, The Range of a Structural Projection, J. Funct. Anal. 139 (1996), 196 -- 224. [20] C. M. Edwards and G. T. Ruttimann, Inner ideals in C-algebras, Math. Ann. 290 (1991), 621 -- 628. 28 D. P. BLECHER AND MATTHEW NEAL [21] C. M. Edwards and G. T. Ruttimann, Exposed faces of the unit ball in a JBW-triple, Math. Scand. 82 (1998), no. 2, 287 -- 304. [22] F. J. Fernandez-Polo and A. M. Peralta, Non-commutative generalisations of Urysohn's lemma and hereditary inner ideals, J. Funct. Anal. 259 (2010), 343 -- 358. [23] F. J. Fernandez-Polo and A. M. Peralta, On the facial structure of the unit ball of the dual space of a JB∗-triple, Math. Ann. 348 (2010), 1019 -- 1032. [24] T. W. Gamelin, Uniform Algebras, Second edition, Chelsea, New York, 1984. [25] D. M. Hay, Closed projections and peak interpolation for operator algebras, Integral Equations Operator Theory 57 (2007), 491 -- 512. [26] D. M. Hay, Multipliers and hereditary subalgebras of operator algebras, Studia Math. 205 (2011), 31 -- 40. [27] C. R. Johnson and E. A. Schreiner, The relationship between AB and BA, Amer. Math. Monthly 103 (1996), 578 -- 582. [28] C-K. Li, L. Rodman, and I. M. Spitkovsky, On numerical ranges and roots, J. Math. Anal. Appl. 282 (2003), 329 -- 340. [29] H. Lin, Cuntz semigroups of C*-algebras of stable rank one and projective Hilbert modules, Preprint (2010), arXiv:1001.4558. [30] M. Mbekhta, Partial isometries and generalized inverses, Acta Sci. Math. (Szeged) 70 (2004), 767-781. [31] E. Ortega, M. Rørdam, and H. Thiel, The Cuntz semigroup and comparison of open projec- tions, J. Funct. Anal. 260 (2011), 3474 -- 3493. [32] G. K. Pedersen, C ∗-algebras and their automorphism groups, Academic Press, London (1979). [33] G. K. Pedersen, Factorization in C ∗-algebras, Exposition. Math. 16 (1998), 145 -- 156. [34] C. Peligrad and L. Zsid´o, Open projections of C ∗-algebras: comparison and regularity, Op- erator theoretical methods (Timisoara, 1998), 285-300, Theta Found., Bucharest, 2000. [35] C. J. Read, On the quest for positivity in operator algebras, J. Math. Anal. Appl. 381 (2011), 202 -- 214. Department of Mathematics, University of Houston, Houston, TX 77204-3008 E-mail address, David P. Blecher: [email protected] Department of Mathematics, Denison University, Granville, OH 43023 E-mail address: [email protected]
1508.01917	1	1508	2015-08-08T15:30:10	Two roads to noncommutative causality	[ "math.OA", "gr-qc" ]	We review the physical motivations and the mathematical results obtained so far in the isocone-based approach to noncommutative causality. We also give a briefer account of the alternative framework of Franco and Eckstein which is based on Lorentzian spectral triples. We compare the two theories on the simple example of the product geometry of the Minkowski plane by the finite noncommutative space with algebra $M_2({\mathbb C})$.	math.OA	math	Two roads to noncommutative causality Fabien Besnard∗ July 3, 2018 Abstract We review the physical motivations and the mathematical results ob- tained so far in the isocone-based approach to noncommutative causality. We also give a briefer account of the alternative framework of Franco and Eckstein which is based on Lorentzian spectral triples. We compare the two theories on the simple example of the product geometry of the Minkowski plane by the finite noncommutative space with algebra M2(C). 1 Introduction The properties of the electromagnetic field led the early 20th century physicists to a dramatic change in their notions of space and time. Yet more upheaval came from Einstein's theory of the gravitational field. Today, known fields other than gravitation are described together in the Standard Model of particle physics. The derivation of this very complex piece of machinery from noncommutative geometric ingredients is a major breakthrough [1]. While it sheds light on some of the most contrived aspects of the Standard Model, such as the Higgs mechanism, it also opens the door to a renewed understanding of the structure of spacetime. However two hard problems remain. The first is the passage from the euclidean signature to the physical lorentzian one, which is still elusive for noncommutative manifolds (in spite of recent progress in some particular cases, [2]). The second is the quantization of the noncommutative scheme [3]. In fact, there is a third difficulty, which we believe is the key to the elucida- tion of the first two: the physical interpretation of the theory. In noncommuta- tive geometry, a C ∗-algebra is interpreted as an algebra of would-be functions on some virtual space. In the case of the spectral Standard Model the algebra A is composed of true functions on space with values in C ⊕ H ⊕ M3(C). On a dif- ferent stance, coordinates are expected to become noncommutative observables in quantum gravity [4]. We think that it is a fruitful point of view to consider the elements of A already as quantum observables, which take into account the quantum nature of the "inner" space unraveled by the Standard Model, but not yet the quantum nature of the spacetime manifold. With this in mind, we try a causal approach to the problem of the lorentzian signature. More precisely, we observe that causality is a property which is specific to lorentzian manifolds, and that moreover the causal order relation ∗Ple de recherche M.L. Paris, EPF, 3 bis rue Lakanal, F-92330 Sceaux. [email protected] 1 encapsulates the metric information up to a conformal factor. This is hardly a new observation: it is a the root of the Causal Sets approach to quantum gravity [5]. Thus, we want to characterize, among the general elements of the algebra of noncommutative functions A, those which play the role of causal functions. The subset of these distinguished elements is called an isocone. The isocone program has been started in [6] and is still in its early stages of developments. In particular it does not yet make contact with the Dirac operator, which is the central tool of noncommutative geometry. In other words it is not yet known how to distinguish in the noncommutative case true causality, which comes from a metric, from general order relations. However there is another theory, started by Franco and Eckstein in [7], which aims at defining causality out of a Dirac operator. Of course this top- down approach depends on a correct formulation of noncommutative geometry in the lorentzian signature in the first place. In this paper we will review the isocone-based approach to causality, start- ing with general physical motivations in section 2. This will lead us to a set of four fundamental hypotheses, which we will transform in section 3 into a math- ematically more tractable set of axioms that defines isocones in the C ∗-algebra category. In section 4 we will review general properties of isocones, and in sec- tion 5 some classification results for them. In section 6 we briefly review the Franco-Eckstein theory, and compare the two approaches on a simple example in section 7. We will then conclude with a few remarks. In all the text "compact" means "compact and Hausdorff" and the order relation used on R is always the natural one. 2 Time, clocks and causality We start with a rather philosophical discussion, where "philosophical" is here an elegant substitute for "vague". This vagueness will perhaps be forgiven since the question we want, to address "what is time ?", is certainly one of the most difficult of all. It seems reasonable to look for an answer in our two best theories of natural phaenomena to date, namely General Relativity and Quantum theory. The former theory carefully distinguishes between chronology, duration and causality, and puts strong structural constraints on them. The latter theory says nothing at all: time in quantum theory is an external parameter. So the punchline appears to be this: time, in any of its three facets mentioned above, is structurally defined by GR and has no quantum aspect at all. Note that by "structurally defined" we mean that there is no more explicit definition of time in GR than there is any explicit definition of sets in axiomatic set theory, but that the two theories implicitly define their objects by specifying how they behave. That the definition of time is a structural one is not really surprising, the only alternative being that it is an emergent phenomenon that can be defined in term of something else. However it is unsatisfying that quantum theory should have no word on this matter, and it is rather unlikely that we have here the definitive answer. There are indeed several arguments in favor of an influence of quantum theory on the question of time, and we refer to [8] for some of them. Here we quote only two. The first and most obvious is that if the gravitational field has to be quantized, then the structural constraints on time observables will have to be 2 modified accordingly. The second is that the so-called clock hypothesis, which is a postulate of GR, states that a clock measures the proper time of an observer who carries it. But clocks being made of quantum fields, like everything else, cannot in principle be properly modelized without the help of quantum theory. In other words there should be quantum observables associated to them. A possible objection to this point of view would be that clocks are necessarily macroscopic objects. Once again this would amount to say that time is an emergent phenomenon. The characterization of quantum observables associated to measures of du- ration poses a real challenge, even in the context of quantum mechanics on a non-relativistic background [9]. However, we will argue that if we fall back to the cruder notion of causality, a tentative answer to this problem can be given. Let us now explain more precisely what we propose to do. From General Relativity we borrow the concept of causality. A spacetime M , i.e. a connected time-orientable Lorentzian manifold, is said to be causal if the following relation p (cid:22) q ⇔ there exists a causal curve from p to q (1) is a partial order. Not all metrics define causal spacetimes, even those which solve the Einstein's equations can have pathological causal properties. However, there always exists a neighbourhood U of a given event p such that (cid:22) is a partial order on U . In fact much more is true: one can find a neighbourhood with the best possible causality property, namely global hyperbolicity (see [10], th 2.14). The causal ordering permits to define causal functions, that is functions f : M → R such that x (cid:22) y ⇒ f (x) ≤ f (y). These functions are classical observables which correspond to measurement devices that we call generalized clocks. A generalized clock, if compared to a genuine clock, can arbitrarily change pace or even stop for a while. The only thing it cannot do is run backwards. We now want to associate quantum observables to these generalized clocks. For this we introduce a quite general setting for quantum physics [11]. 1. For a given physical system, there exists a set of states S, a set of observ- ables O, and a duality map S × O → R, which is separating for S and O. 2. S is a convex set. 3. O is a real vector space. 4. O is stable under a 7→ f (a) where f is a real-valued function. Note that we do not assume for the moment that the duality function gives the expectation value of the observable when the system is in some state: this is a matter of interpretation of the formalism and we will deal with that later on. There are many good reasons to believe in these axioms, and we refer to [11] for justifications. Let us only comment on the last of them, which will be crucial for our purpose. This axiom follows from the following intuitive interpretation: the observable f (a) corresponds to an observation of a followed by the mathematical operation of applying the function f to any result of such an observation. A neat way of fulfilling all these requirements is to suppose that O is the self- adjoint part of a C ∗-algebra and that S is its state space, this is the so-called 3 C ∗-paradigm. However it is not quite the most general solution: a Jordan algebra would suffice. So we will start by assuming that O is a Jordan algebra. We postpone to the next section the specification of the type of Jordan algebra that we will use in order to make sense of our hypotheses. We now call C the set of causal observables corresponding to generalized clocks. Our first hypothesis, justified above, is that C is a subset of a Jordan algebra. Hypothesis 0 (Jordan paradigm) C is a subset of a Jordan al- gebra O. In some classical limit C should correspond to the set of causal functions on M , but it should be stressed that the correspondence is in the direction causal observable → causal functions, in particular we do not assume the existence of a quantization map which would go in the other direction. In fact it is well- known that quantization maps cannot commute with functional calculus and be linear at the same time [12]. What are the properties of C ? Consider o ∈ C, c a generalized clock corresponding to o and f a non-decreasing real-valued function. Then the device c′ which is formed by c followed by a computation of the function f on the result obtained by c is still a generalized clock. Hence f (o) ∈ C. This is our hypothesis number 1 for the set C. Hypothesis 1 (functional stability) If o ∈ C and f is a real non-decreasing function, then f (o) ∈ C. Let us remark here that this implies that the algebra is unital and that C contains all the real multiples of the unit. Let us dissipate any worry that could arise at this point: it does not mean that our approach only applies to compact spacetimes. If we start with some noncompact spacetime and use the approach advocated here with a unital commutative algebra, we will recover some com- pactification of the spacetime we started from, depending on the algebra used. Of course these objects have boundaries. It is indeed a well-known theorem that compact causal spacetimes always have boundaries (or to put it another way, that compact closed causal spacetimes do not exist). We now introduce a separation hypothesis. Hypothesis 2 (separation) C separates the pure states of O. The origin of this hypothesis is the classical situation. In any sufficiently small given region of spacetime one can introduce causal coordinates, for in- stance GPS coordinates [13], which obviously separates events (i.e. pure states). If one considers the spacetime manifolds globally, there can be too few causal functions to separate the events (in fact, there can be none at all, except the con- stants). However the hypothesis will be fulfilled if one assumes strong causality, which despite its name, is a weaker causality condition than global hyperbolicity [10]. Hypothesis 2 can thus be viewed as an exportation to the quantum regime of the idea that causality plays a fundamental role in the process of identification of what we call "different spacetime events". As such it cannot be fully justified, it is an act of faith. As soon as we have adopted it, this hypothesis grants us a partial ordering on the set P (O) of pure states of O, which is so defined: 4 ∀p1, p2 ∈ P (O), p1 (cid:22) p2 ⇔ p1(c) ≤ p2(c) for all c ∈ C (2) In the classical situation, if we started from the set of causal functions, we recover in this way the causal order on spacetime events. Mind that this is not a mere tautology: it could well be that the order defined through (2) turn out to be stronger than the causal order (see the definition of toposets in section 4). But as is now familiar, the equality between causal order and the order defined through the causal functions always holds locally and will hold globally under some causality condition [6]. In the present situation we started from a set of observables that we did not know at first to be associated with some partial order. But we know now from (2) that each observable of our set C is non-decreasing for (cid:22). In the classical case the causal functions are exactly those which are non-decreasing for (cid:22), but even there, we may very well start from a smaller set of causal functions which is still large enough to fulfill the hypothesis 1 and 2 and end up with the same causal order. Hence we see that we need a "saturation" hypothesis to ensure that we started from the whole set of causal functions and not a smaller subset. The most natural is the following: Hypothesis 3 (saturation) Every observable o which have a non- decreasing Gelfand transform with respect to (cid:22), i.e. p1 (cid:22) p2 ⇒ p1(o) ≤ p2(o), is in C. As above, the leap leading to this hypothesis cannot be fully justified. How- ever we will give some arguments in the next section that we hope will give some credence to it. 3 From the fundamental hypotheses to a set of working axioms In the previous section we put forward four hypotheses, which we qualify as fundamental, about the set C of causal observables. However, they are not very handy from a mathematical point of view. This is particularly true of the saturation hypothesis. In this section, we will see that under some very reasonable conjectures, they can be turned into axioms which are easier to work with. First, we must specify the kind of Jordan algebra we use. In order to make sense of hypothesis 1 we must have a functional calculus. We choose JB-algebras, which possess continuous functional calculus. By doing so we subrepticely intro- duce a (noncommutative) topology, and the question of the compatibility with the causal structure arises. We will see in the next section that a very natural answer to this question can be given thanks to Gelfand theory for toposets. Let us now use the saturation hypothesis. If c1, c2 ∈ C, then the Gelfand transforms c1 : p 7→ p(c1) and c2 : p 7→ p(c2) defined on pure states is non- decreasing, hence their sum is also. Therefore c1 + c2 ∈ C by the saturation hypothesis. Similarly, if cn is a sequence of elements of C converging in the norm to c, then each cn is non-decreasing, hence the limit is, hence c ∈ C. Consider now the following set of axioms on C, which can be derived from the fundamental hypotheses, as we have seen: 5 1. C ⊂ O, a JB-algebra. 2. C separates the pure states of O. 3. C is stable by non-decreasing continuous functional calculus. 4. C is norm closed. 5. C is stable by sum. Though it is a priori a weaker set of axioms than the fundamental hypotheses, we conjecture that it is in fact equivalent. We call this the saturation conjecture. Whatever the status of this conjecture is, we will work in this paper with this maybe weaker set of axioms. In each case we will encounter, it is possible to show that the set C is indeed saturated, that is, contains all observables with non-decreasing Gelfand transform. But first we want to strengthen axiom (2) by asking instead that: (ii') C separates the states of O. We will show below that axiom (ii) can be replaced with axiom (ii') without loss of generality provided the following property is true for O: Strong Stone-Weierstrass property Let O be a JB-algebra and B be a subalgebra of O. If B separates P (O) ∪ {0}. Then B = O. The strong Stone-Weierstrass conjecture for JB-algebras states that all JB- algebras satisfy this property. This is known to be true if P (O) is closed and also if O is post-liminal [14]. All algebras considered in this paper will be post- liminal. Proposition 1 If O satisfies the Strong Stone-Weierstrass property, then the hypotheses (i) to (v) imply (ii'). The proof is easy. Let B be the strong closure of the set of differences C −C. First notice that C − C = C+ − C+, where C+ denotes the set of positive elements in C. Indeed, we can always rewrite a difference a − b in the form a + k.1 − (b + k.1) where k is a large enough real and 1 is the unit of O. (Remember that C contains the real multiples of 1, as noticed in the previous section.) Next, consider a, b ∈ C+. Then a + b ∈ C+. Moreover, if f is the continuous non- decreasing function such that f (t) = 0 for t < 0 and f (t) = t2 for t ≥ 0, then f (a + b) = (a + b)2, f (a) = a2 and f (b) = b2 all belong to C+. Thus the Jordan product a ◦ b = 1 2 ((a + b)2 − a2 − b2) ∈ B. By linearity and continuity of the Jordan product we easily see that B is indeed a subalgebra of O. Now by hypothesis B separates P (O), and since it contains a unit it separates 0 from the other pure states. Thus B = O. Since C − C = O it is now obvious that C separates the states of O. 4 Isocones and I ∗-algebras From now on we concentrate on JB-algebras which are the self-adjoint part of C ∗-algebras. In this context the structure of the set of causal observables becomes (under the saturation and Stone-Weierstrass conjectures) that of an isocone, the definition of which we recall below: 6 Definition 1 Let A be a unital C ∗-algebra, Re(A) be its self-adjoint part. An isocone of A is a non-empty set I such that: 1. I ⊂ Re(A), 2. ∀φ : R → R continuous and non-decreasing, a ∈ I ⇒ φ(a) ∈ I, 3. ∀a, b ∈ I, a + b ∈ I, 4. I is closed, 5. I separates the states of A, If I is an isocone of A, the couple (I, A) is called an I ∗-algebra. Isocones were first defined in [6] but with slight differences, so let us make some comments on them to help make the connection. First, instead of axiom (5) we required that the set of differences I − I be dense in Re(A). This is seemingly a stronger condition, but in fact they are equivalent. Indeed, suppose I separates S(A) and let V = I − I. Note that V is a real vector space by axioms (2) and (3). Suppose there exists a ∈ Re(A) such that a /∈ V and use Hahn-Banach theorem to find a continuous linear form f such that f (V ) = 0 and f (a) 6= 0. We can extend f as an hermitian form on A and by [18] there exist two states s1, s2 on A and two nonnegative numbers α1, α2 such that f = α1s1 − α2s2, and kf k = α1 + α2. Since 1 ∈ V , we find that α1 = α2 = kf k/2. Hence f = k(s1 − s2) with k = kf k/2. By hypothesis f (V ) = 0, hence s1 and s2 are equal on I, and since I separates the states, they are equal, which contradicts f (a) 6= 0. Moreover, in [6] we did not put forward axiom (2), but instead asked that if a, b ∈ I and ab = ba, then sup(a, b) and inf(a, b) must also be in I. This axiom was motivated by the hypotheses of the Kakutani-Stone theorem, thanks to which one can prove an analog of the Gelfand-Naimark theorem for commutative I ∗-algebras (see below). However, this mathematically motivated axiom was later shown to be equivalent to the more physically meaningful stability under non-decreasing continuous functional calculus in [15]. As alluded to above, the first step of the theory is to prove an analog of the Gelfand-Naimark theorem for commutative I ∗-algebras. Let us introduce some terminology in order to state this theorem. Let (M, (cid:22)) and (N, ≤) be two partially ordered topological spaces. A map f : M → N which preserves the order relations is said to be isotone or to be an isotony. The set of continuous isotonies from M to R, where R is equipped with the natural order and topology, is denoted by I(M ). A toposet is a topological ordered space which has "enough" isotonies, in the following sense: ∀x, y ∈ M, x (cid:22) y ⇔ (∀f ∈ I(M ), f (x) ≤ f (y)) (3) This is the compatibility condition between topology and order that we were wondering about in the previous section. A toposet is also called a completely separated topological ordered space. Note that if M is compact, it is a toposet if and only if the relation (cid:22) is closed as a subset of M × M [16]. A spacetime is a toposet if and only if it is causally simple [17]. It means that it does not contain closed causal curves and that the future and past of any event are closed. It is the second strongest causality condition after global hyperbolicity. 7 Theorem 1 Let M be a compact toposet. Then the couple (I(M ), C(M )) is a commutative I ∗-algebra. Conversely, every commutative I ∗-algebra (I, A) is canonically isomorphic to one of this form, where M is the set of pure states of A. As we explained before, there cannot be any such duality theorem in the locally compact case where one uses the algebra C0(M ) of continuous functions vanishing at infinity since this does not contain any non-trivial isotone function in general. Hence one is forced to use a preferred compactification in this case. The proof of theorem 1 can be found in [6]. Let us just recall that the order on M = P (A) is recovered from I by the formula ∀φ, ψ ∈ P (A), φ (cid:22)I ψ ⇔ (∀a ∈ I, φ(a) ≤ ψ(a)) (4) which is just (2) adapted to the present situation. As explained in the previous section, this formula makes sense in any I ∗-algebra, commutative or not, and we can see now that it endows P (A) with the structure of a toposet. Before proceeding let us give a few examples of noncommutative I ∗-algebras. if I is the whole set of self-adjoint elements of A, all The first is trivial: axioms are evidently satisfied. This is called the trivial isocone, and it induces on P (A) the equality ordering φ (cid:22) ψ ⇔ φ = ψ. For the second example we consider A = M2(C). In this case it is easily seen that any closed convex cone in ReA which contains the constants and has a non-empty interior is an isocone. We can give a geometric interpretation of the order induced by I on P (A) in the following way. First let us identify P (A) with 2-sphere of rank one projections which we call S. Then K = I ∩ S is compact, has non-empty interior and is either equal to S or to a geodesically closed subset of a hemisphere. Then if we write ω[ξ] for the pure vector state defined by the unit vector ξ and p[ξ] for the projection on the line generated by ξ, we have ω[ξ] (cid:22)I ω[η] ⇔ (∀x ∈ K, d(x, p[ξ]) ≥ d(x, p[η])) (5) where d is the geodesic distance on S. Clearly we can also use the identifi- cation of P (A) with CP 1 and the Fubini-Study metric (see [6] for details). The next example is infinite-dimensional. If we consider any compact toposet M with a regular Borel measure, we can faithfully represent the elements of C(M ) as multiplicative operators on L2(M ). Consider then the set K of compact operators on L2(M ) and I = I(M ) + Re(K), that is the set of perturbations of isotonies of M by self-adjoint compact operators. Then I is an isocone of A + K [15]. In this case the pure state space of A can be written P (A) = X ` Y where X ≈ M and Y is the set of vector states. Then the order induced by I is the original order on M extended trivially (no further relations on distinct elements) to P (A). The intuitive idea behind this example is that compact operators correspond to infinitesimals, and that a non-decreasing function stays non-decreasing when one adds an infinitesimal to it. To build more examples, one can use the two following constructions. Theorem 2 Let (P, (cid:22)) be a finite poset and for each x ∈ P let (Ix, Ax) be an I ∗-algebra. We set I = Lx∈P Ix, A = Lx∈P Ax, and we write elements of A 8 in the form (ax)x∈P . We define Lex x∈P Ix = {a ∈ I∀x, y ∈ P, x ≺ y ⇒ max σ(ax) ≤ min σ(ay)} where σ(a) denotes the spectrum of a. Then Lexx∈P Ix is an isocone of A. The isocone L := Lexx∈P Ix induces the lexicographic sum of the orderings (cid:22)x defined by Ix on P (Ax). More precisely put, we have P (A) = ` P (Ax) and the pure states are ordered by L in the following way: let φ ∈ P (Ax) and ψ ∈ P (Ay) then φ (cid:22)L ψ ⇔ (x 6= y and x (cid:22) y in P ) or (x = y and φ (cid:22)Ix ψ in P (Ax)) (6) This is why we call L the lexicographic sum over P of the isocones (Ix)x∈P . If the order relation on P is trivial, this is just a direct sum. There is a gener- alization of this construction to infinite posets, but one needs to add conditions on the order and the map x 7→ Ix. To simplify matters we also suppose that the algebra Ax = A is the same for all x. Theorem 3 Let (P, (cid:22)) be a metrizable compact poset and for each x ∈ P let Ix be an isocone of A. Let A be the C ∗-algebra C(P, A) and Lex x∈P Ix := {f ∈ C(A, P )∀x ∈ P, f (x) ∈ Ix and ∀x, y ∈ P, x ≺ y ⇒ max σ(ax) ≤ min σ(ay)} where σ(a) denotes the spectrum of a. hemicontinuous, then Lexx∈P Ix is an isocone of A. If ≺ is closed and x 7→ Ix is lower Remember that a multi-valued function ϕ : X −→ 2Y is said to be lower hemi-continuous if and only if the set {x ∈ Xϕ(x) ∩ V 6= ∅} is open for all open V ⊂ Y . The condition that ≺ is closed as a subset of the product P × P is equivalent to the following one: for each x ∈ P there exists a neighbourhood U of x such that no element of U \ {x} is comparable to x. It is of course satisfied if P is discrete (hence finite by compactness) but this is not necessary. For instance the ordering on P = [0; 1] defined by the relations x (cid:22) 1 for all x ≤ 1/2 and x (cid:22) x for all x ∈ [0; 1] satisfies it. Note that if ≺ is closed and P is compact then (cid:22) is closed hence P is a toposet. The other construction which will play an important role is the "pushfor- ward" by surjective ∗-morphisms. Proposition 2 Let A, B be C ∗-algebras, π : A → B a surjective ∗-morphism, I an isocone of A. Then π(I) is an isocone of B. Note that π(I) is automatically closed if B is finite-dimensional. 5 Isocones in finite-dimensional and almost-commutative algebras The first task is to classify isocones in matrix algebras. We will say that an algebra is egalitarian if it contains only the trivial isocone (the justification for the terminology is that the only isocone-induced order on its pure state space will be equality). In [15] we proved the following: 9 Theorem 4 The matrix algebra Mn(C) is egalitarian for all n except n = 2. To put this result in perspective let us move away for a moment from the category of C ∗-algebras. First we can consider real C ∗-algebras, which are important in the context of the spectral standard model. Then there are two new cases of non-egalitarian algebras which are M2(R) and M2(H). However in the more natural arena of JB-algebras, the exceptional character of these cases disappear, as they all fit into the familly of spin factors. Remember that a spin factor, or Jordan algebra of Clifford type is the direct sum N ⊕ R, where N is a real Hilbert space, with Jordan product (u, x) ◦ (v, y) = (xu + yv, xy + hu, vi). The Jordan algebras of self-adjoint real, complex and quaternionic 2×2 matrices are isomorphic to spin factors with N = R2, R3 and R5 respectively. It will be proven elsewhere that spin factors are the only simple finite-dimensional non- egalitarian JB-algebras (with the possible exception of the Albert algebra, which is the only case still under investigation). Returning to the case of C ∗-algebras, we now give a classification result. Theorem 5 Let I be an isocone in a finite-dimensional C ∗-algebra AF . Then there exists a finite poset P and positive integers (nx)x∈P such that up to a ∗-isomorphism one has A = Lx∈P Mnx(C) and I = Lexx∈P Ix. Since Mnx is egalitarian if nx 6= 2 and isocones in M2(C) are easily char- acterized, as we explained above, this theorem classifies all finite-dimensional I ∗-algebras. It shows that the generalization of partial orders given by I ∗- algebras is quite limited in the finite-dimensional case: partial order seems to be essentially a commutative phenomenon. However, the new freedom offered by the exceptional n = 2 case could prove to be important in the context of the Standard Model, possibly in relation with T-symmetry breaking. Let us make some comments on the physical interpretation of the above theorem. Recall that we see the elements of I as observables corresponding to generalized clocks. If we admit the usual interpretation rule of quantum physics that eigenvalues of observables corresponds to values that can be possibly mea- sured, then the order on P can be characterized as follows: x (cid:22) y if and only if no generalized clock ever give a result at x which is strictly larger than its result at y. On a more mathematical stance, let us note that what the theorem tells us is that the partial order naturally defined on the pure state space by I descends on the structure space, which is P . This was not guaranteed, and is proved by a combinatorial/analytical argument. If there is a conceptual reason for this, we are not aware of it. We now turn to the important case of almost-commutative algebras, that is, C ∗-algebras of the form A = C(M, AF ), where AF is a finite-dimensional algebra. We fix A an almost-commutative algebra with M compact. For sim- plicity of exposition, we suppose that AF = Mn(C) with n ≥ 2. The case of a general finite-dimensional algebra AF is not essentially different (we refer to [17] for details). Note that the pure state space of A is M × P(Cn). Theorem 6 Let I be an isocone of A. Then the order induced by I on M × P(Cn) is lexicographic: there exists a partial ordering (cid:22)M on M and for each 10 x ∈ M there exists an isocone Ix of Mn(C) such that : (x, ξ) (cid:22)I (y, η) ⇔ x ≺M y or x = y and ξ (cid:22)Ix η   The key to prove this theorem is the above classification of finite-dimensional I ∗-algebras and the pushforward construction of section 4. Note that the order- ing on M is defined from I through the following formula: x (cid:22)I y ↔ ∀a ∈ I, max(σ(ax)) ≤ min(σ(ay)) (7) Conversely, given a partial ordering (cid:22)M and a map x 7→ Ix into the isocones of Mn(C), we can state necessary and sufficient conditions for the existence of an isocone of A inducing them. Theorem 7 If M is metrizable and (cid:22) is a lexicographic order on M × P(Cn) associated to a partial order (cid:22)M on M and local isocones Ix ⊂ Mn(C), then there exists an isocone I in A inducing (cid:22) iff • ≺M is closed. • L : x 7→ Ix is lower hemi-continuous. In that case, I is the lexicographic sum of the isocones Ix over x ∈ M . This theorem would provide a classification of almost-commutative I ∗-algebras (with M metrizable) provided one could prove that the isocone inducing a given order is unique. This would of course follow from the saturation conjecture. Note that it is not difficult to prove that the lexicographic sum I = Lexx∈M Ix is saturated, that is, it is exactly the set of self-adjoint elements of A whose Gelfand transform are isotone for (cid:22). Hence, in the almost-commutative case the saturation conjecture amounts to say that I does not strictly contain any isocone inducing the same ordering on M × P(Cn). Let us end this section by remarking that the closedness condition on ≺M is really imposed by the noncommutativity of Mn(C) for n ≥ 2. If AF is a more general finite-dimensional C ∗-algebra then its structure space will have several sheets. If we consider C ⊕ M2(C) for instance, there are two sheets M1 and M2. Then, the order induced on M1 which is associated to the commutative algebra C will not have the closedness restriction. 6 Lorentzian spectral triple and causal cones Let us now summarize the approach initiated by Franco and Eckstein to the question of causality in noncommutative geometry. We refer to [7] and [19] for details, however we sometimes give indications of proofs which are different than those given there, and might be in some cases easier to generalize. One first defines Lorentzian spectral "triples", which in fact contains five pieces of information. Definition 2 A Lorentzian spectral triple is given by the following data: a pre- C ∗-algebra without unit A, a unitization A of A, a Hilbert space H on which A and A are faithfully represented, an unbounded operator D with dense domain such that for all a ∈ A 11 • [D, a] is bounded, • a(1 + hDi2)−1/2 is compact, where hDi2 = 1 2 (DD∗ + D∗D). and j is a bounded operator such that j2 = 1, j∗ = j, [j, A] = 0, D∗ = −jDj. Moreover one asks that j = −[D, T ] where T is an unbounded operator such that (1 + T 2)−1/2 ∈ A. The role of j is to turn H into a Krein space, and the requirement that j can in fact be written j = −[D, T ] is the way Lorentzian signature is single out among arbitrary pseudo-riemannian ones. The first example is the following: consider a complete and globally hy- perbolic Lorentz manifold M , S its spinor bundle, H = L2(S), D = −iγµ∇S µ the usual Dirac operator. The algebras A and A are respectively S(M ), the Schwartz functions on M , and the algebra generated by causal bounded func- tions with bounded derivatives. The role of the operator T is played by a global time function on M , called x0, so that j = iγ(dx0). On a coordinate chart where x0 is the first coordinate one can write j = iγ0. Given a lorentzian spectral triple, one can define a causal cone. Definition 3 A causal cone C is a cone in the self-adjoint part of A which contains 1 and such that • spanC(C) = A • ∀a ∈ C, j[D, a] ≤ 0 (is a negative semidefinite operator) We can then use a causal cone to define a partial order on the space of pure states P ( A), as one does with an isocone. We can see that we recover the usual partial order on M (which is a subset of the compact space P ( A)) through the following result ([7], theorem 11): Theorem 8 If (A, A, H, D, j) is the Lorentzian spectral triple built out of a globally hyperbolic Lorentzian manifold M , then f ∈ A is causal if and only if j[D, f ] ≤ 0. This can be most easily proved using a pseudo-orthonormal moving frame (vierbein). We can then define the Lorentz gradient of f by df (v) = g(v, ∇f ), and decompose ∇f in the form −(∂0f )e0 + (~∇f )iei in the moving frame. We have iγ0[D, f ] = −∂0f Id + b, with b = X1≤i≤3 ∂if γ0γi. Noting that b∗ = b and b2 = k~∇f k2Id, we see that the spectrum of b is {±k~∇f k}. Hence iγ0[D, f ] is negative semidefinite iff ∇fx lies inside the past-cone at every x ∈ M , which is easily seen to be a necessary and sufficient condition for f to be causal. Now since globally hyperbolic manifolds are toposets, we see that the order defined by x ≤M y ⇔ (∀f ∈ A, j[D, f ] ≤ 0 ⇒ f (x) ≤ f (y)) (8) is exactly the original causal order of the manifold. 12 7 Comparison of the two approaches We see that the two approaches agree as far as (globally hyperbolic) Lorentzian manifolds are concerned: a causal cone is just the smooth part of the isocone of causal functions. To see what happens in the noncommutative case, we will focus on a simple though enlightening example. We take M = R1,1 to be the 2- dimensional Minkowski spacetime. We consider the finite-dimensional spectral triple (AF , HF , DF ) with AF = M2(C), HF = C2, DF = diag(d1, d2). A product lorentzian spectral triple is defined by A = S(M ) ⊗ AF , H = L2(S) ⊗ HF , where S is the spinor bundle which here is trivial with fiber C2, D = DM ⊗ 1+γ0γ1⊗DF , with DM = −iγµ∂µ, and j = iγ0⊗1. The algebra AM is defined to be the linear span of smooth causal bounded functions with bounded derivatives, and A = AM ⊗ AF . This in fact corresponds to the Penrose compactification of the Minkowski plane which we briefly recall. Let (x0, x1) be the canonical coordinates on R1,1, and write u = x0 + x1, v = x0 − x1. The map φ : M → P =] − π; π[2 defined by φ(u, v) = (µ, ν) := (2 tan−1(u), 2 tan−1(v)) is a smooth diffeomorphism (the factor of 2 is inserted to stick with usual conventions). Moreover φ sends the causal order ≤M on M , which in the coordinates (u, v) is just the product order on R2, to the product order ≤P on P . Since the derivatives of φ are all bounded, the push-forward φ∗ : f 7→ f ◦ φ−1 sends AM to the linear span of bounded causal (for ≤P ) functions on P with bounded derivatives. Such functions can be extended by continuity on ¯P = [−π; π]2. To see it notice that causal functions on P can be expressed in the form g(µ) + h(ν) with g, h non-decreasing on R. Using the Stone-Weierstrass theorem we then immediately see that the linear span of bounded causal functions on P with bounded derivatives are dense in C( ¯P ). Hence the pure state space of AM is homeomorphic to the Penrose compactification of M . Let us look for the elements of the causal cone. A hermitian-matrix-valued function α is in C if j[D, α] = γ0γµ⊗∂µα−γ1⊗i[DF , α] ≤ 0. With γ0 = (cid:18) 0 γ1 = (cid:18) 0 −i 0 (cid:19) this condition can be written in block-matrix form: i i 0(cid:19), i (cid:18) 2∂uα −[DF , α] [DF , α] 2∂vα (cid:19) ≥ 0 In particular this condition implies that ∂uα ≥ 0, ∂vα ≥ 0 (9) (10) Integrating (10) we see that if x ≤ y (where ≤ denotes the usual causal order on Minkowski space), then α(x) ≤ α(y) as operators. Conversely, if α(x) ≤ α(y) for all α ∈ C, then using α =diag(t1, t2), where t1, t2 are two time functions, one sees that x ≤ y. Hence we have proved that x ≤M y ⇔ ∀α ∈ C, α(x) ≤ α(y) (11) This formula can be put in contrast with (7) (where, we recall, (cid:22)M can- In particular we note that if λ1(x) ≤ λ2(x) are the not be equal to ≤M ). eigenvalues of α(x), then (10) shows that x 7→ λi(x) are causal functions. It is however perfectly possible that λ2(x) > λ1(y) with x ≤M y. If we keep the eigenvalue/observed value correspondance and interpret the elements of C as 13 causal observables, we have here a generalized clock which can assign a later date to x than to y, which is inconsistent with the observations of section 5. This shows that a causal cone is not an isocone in the noncommutative case. The mathematical origin of the difference between causal cones and isocones is that the former are not stable under non-decreasing functional calculus (this can be traced back to the fact that monotone functions are not necessarily operator-monotone). Finally we recall the main result of [19], which is the characterization of the order induced on the pure states of the form (x, ω[ξ]), with x ∈ M and ω[ξ] is the vector state a 7→ hξ, aξi on M2(C). Let us fix some notations before stating the theorem: dN C (p, q) is Connes' noncommutative distance between two pure states on M2(C) induced by DF , and ℓ(x, y) is the Lorentz distance between x and y, that is to say the supremum of Rγ ds where γ runs over all future-directed causal curves from x to y (it is set to 0 if y is not in the future of x). In the case at hand, ℓ(x, y) is the proper time elapsed from x to y measured by an inertial observer. Theorem 9 (x, ω[ξ]) (cid:22)C (y, ω[η]) iff x ≤M y and ℓ(x, y) ≥ dN C (ω[ξ], ω[η]). Note that the inequality implies that dN C(ω[ξ], ω[η]) is finite. We recall that it is the case iff ω[ξ] and ω[η] are at the same latitude on the 2-sphere, where the north and south poles are given by the eigenstates of DF (see [20]). Let us explain briefly why the condition is necessary. First we can suppose without loss of generality that x = (0, 0) and y = (s, 0). Then ℓ(x, y) = s. Using matrix- valued functions of the form α(t, r) = tI2 + A, where A is a constant hermitian matrix, we easily obtain from (9) that α ∈ C iff k[DF , A]k ≤ 1, which gives the result. We now turn our attention to I ∗-algebras. We work directly with the Penrose compactification, so the algebra is A = C( ¯P ) ⊗ M2(C). The other datum we need is an isocone I of A. Note the difference with the previous approach: the usual causal order was encapsulated in the choice of D and j, wheras here it is induced by the isocone, and as such cannot be the usual causal order on ¯P , since the noncommutative nature of the algebra A imposes that the causal relations locally vanish. What we ask then, is that the isocone-induced order (cid:22)Λ on ¯P depends on a constant Λ and tends to the usual causal order when Λ → 0. If we moreover ask (cid:22)Λ to be Lorentz-covariant, it will be easily constructed. We first define (cid:22)Λ on R1,1 and then we will send it to ¯P . On R1,1 we set: (x0, x1) (cid:22)Λ (x′0, x′1) ⇔ x0 ≤ x′0 and (x0 − x′0)2 − (x1 − x′1)2 ≥ Λ2 (12) That is, we simply replace forward light-cones with forward hyperbolae of mass Λ. Changing to the coordinates (u, v), and using the notations ∆u = u−u′, ∆v = v − v′, we easily see that this definition is equivalent to (u, v) (cid:22)Λ (u′, v′) ⇔ ∆u ≥ 0, ∆u∆v ≥ Λ2 (13) Hence in P , using the same notation to denotes the partial order we obtain (µ, ν) (cid:22)Λ (µ′, ν′) ⇔ ∆µ ≥ 0, (∆ tan µ 2 )(∆ tan ν 2 ) ≥ Λ2 (14) 14 with obvious notations, which is extended to ¯P by (µ, ν) (cid:22)Λ (µ′, ν′) ⇔ ∆µ ≥ 0, ∆ν ≥ 0 anytime (µ, ν) or (µ′, ν′) belongs to the boundary. To complete the data we have to choose a lower-hemicontinuous isocone map (µ, ν) 7→ I(µ, ν) ⊂ M2(C). For instance we can take a constant map with value a non-trivial isocone in M2(C) defined by a closed and geodesically convex subset K of the sphere S2. We finally define I = Lex(µ,ν)∈ ¯P I(µ, ν), with Ix = I =constant, and ¯P ordered by (cid:22)Λ. We illustrate the two situations in figure 1. 8 Conclusion It is evident from the above examples that there is much more freedom in the isocone approach, and it is not surprising since isocones generalize partial orders which can be arbitrary apart from possesing the toposet property. In partic- ular we do not have any axiom which would ensure the compatibility with a given metric, incarnated in the noncommutative setting by the Dirac operator. This is clearly an important ingredient to add in the future development of the theory. On the other hand we believe that the approach is sufficiently general and physically well motivated to be trusted on its qualitative conclusions, in particular on the necessary vanishing of the causal relations at small scale (with no indication on the smallness scale, though). As a matter of fact, equivalent conclusions have been reached in [21] and [22] by studying the properties of the spectral action at high energy. The agreement between the isocone approach and the properties of the spectral action is all the more striking that the lat- ter is formulated in the euclidean formalism. However, this action reproduces, under some cut-off energy scale, the one of the Standard Model which we know is correct on experimental ground. An optimistic explanation of the agreement between the disappearance of causality at small scale/high energy, as required by the isocone approach, and the non-propagation of high-energy bosons im- plied by the spectral action formalism could therefore be the following one: 1) the fundamental hypotheses laid up in the first section can be trusted (i.e. Na- ture really behaves like this), 2) the asymptotic developments of the spectral action can be trusted, 3) Nature is consistent with itself. To temperate this enthousiasm, let us observe that the local disappearance of causality implied by the isocone approach might be more soberly viewed as just another theorem of the noncommutativity-implies-discreteness kind that we are familiar with since the advent of quantum theory. However, as we have seen, causal cone theory does not predict such a dis- creteness. Quite the contrary, it is fully compatible with usual causality at all scale. Does it mean that it is wrong-headed ? Not necessarily. In fact it is interesting to note that the bottom-up isocone approach and the top-bottom causal cone theory each has what the other lacks. The first has, we believe, a clear physical interpretation and is able to derive sensible qualitative conclu- sions at large energies, the second has a direct contact with the usual apparatus of noncommutative geometry, providing in particular a very transparent con- nection between the Dirac operator and the causal order. We think that it is not only possible, but necessary, that the two approaches merge. It is all too obvious that the culprit of their present incompatibility is Minkowski spacetime, whose smoothness cannot be trusted at arbitrary large energy scale. In replac- 15 ing smooth spacetime with a discrete structure we do not expect to encounter any difficulty on the isocone side, but we will have to discretize the Dirac oper- ator along the lines of [23]. Whether this permits to equate causal cones with isocones is under investigations. References [1] Chamseddine AH, Connes A, Marcolli M 2007, Gravity and the Standard Model with neutrino mixing, Adv. Theor. Math. Phys. 11 991 [2] van den Dungen K, Paschke M, Rennie A 2013, Pseudo-Riemannian spectral triples and the harmonic oscillator, J. Geom. Phys. 73 37 [3] Besnard F 2007, Canonical quantization and the spectral action, a nice ex- ample, J. Geom. Phys. 57 1757 [4] Doplicher S, Fredenhagen K, Roberts JE 1995, The quantum structure of spacetime at the Planck scale and quantum fields, Commun. Math. Phys., 172 187 [5] Bombelli L, Lee J, Meyer D, Sorkin RD 1987, Space-time as a causal set, Phys. Rev. Lett. 59, 521 [6] Besnard F 2009, A noncommutative view on topology and order, J. Geom. Phys. 59 861 [7] Franco N, Eckstein M 2013, An algebraic formulation of causality for non- commutative geometry, Class. Quant. Grav. 30 135007 [8] Besnard F 2011, Time of philosophers, time of physicists, time of mathe- maticians, Preprint arXiv:1104.4551 [9] Muga JG, Sala Mayato R, Egusquiza IL (Eds.) 2002, Time in quantum mechanics, (Berlin: Springer) [10] Minguzzi E, S`anchez M 2008, The causal hierarchy of spacetimes, Recent developments in pseudo-Riemannian geometry, ESI Lect. Math. Phys 299 [11] Emch GG 1984, Mathematical and Conceptual Foundations of 20-th Cen- tury Physics, (Amsterdam: Elsevier) [12] S. T. Ali, M. Englis, Quantization methods: a guide for physicists and analysts, Rev. Math. Phys. 17 (2005) 391 math-ph/0405065 [13] Rovelli C 2002, GPS observables in General Relativity, Phys. Rev. D 65 044017 [14] Sheppard B 2001, A Stone-Weierstrass theorem for postliminal JB-algebras. Q. J. Math. 52 507 [15] Besnard F 2013, Noncommutative ordered space: examples and counterex- amples, to appear in Class. Quantum Gravity, arXiv:1312.2442 [16] Nachbin L 1965, Topology and Order, (Princeton: D. Van Nostrand Com- pany) 16 [17] Bizi N, Besnard F 2014, The disappearance of causality at small scale in almost-commutative manifolds, Preprint arXiv:1411.0878 [18] Kadison R, Ringrose J 1983, Fundamentals of the Theory of Operator Al- gebras, (New York: Academic Press) [19] Franco N, Eckstein M 2014, Exploring the Causal Structures of Almost Commutative Geometries, SIGMA 10 010 [20] Iochum B, Krajewski T, Martinetti P 2001, Distances in Finite Spaces from Noncommutative Geometry, J. Geom. Phys. 37, 100 [21] Kurkov MA, Lizzi F, Vassilevich D 2014, High energy bosons do not prop- agate, Phys. Lett. B 731, 311 [22] Alkofer N, Saueressig F, Zanusso O 2014, Spectral dimensions from the spectral action, Preprint arXiv: 1410.7999 [23] Marcolli M, van Suijlekom WD 2014, Gauge networks in noncommutative geometry, J. Geom. Phys. 75, 71 17 b a n n p K p b a K s axS² s bxS² axS² bxS² Figure 1: This figure shows the set of pure states which are larger than (a, p), where a is an event in the Penrose compactification of the Minkowski plane, and p is a pure state in P (M2(C)) ≃ S2, for the two order relations described in the text: on the left side the order induced by a causal cone, and on the right from an isocone. Each point in the Penrose square represents a 2-sphere. It is depicted in grey if the sphere contains at least a pure state larger than (a, p). Under the square the two spheres at a and b are shown. On the left, only the states lying at the same latitude than p can be in the future of (a, p). The North and South poles are given by the eigenprojectors of the Dirac operator. In the sphere b × S2 one sees that the states in grey are those whose geodesic distance from p is less that the Lorentz distance from a to b. On the right we can see a non-trivial "inner" future light-cone in the sphere a × S2. It is easily seen to be the intersection of S2 with p + K ◦, where K ◦ is the dual cone of K, which is here taken to be a spherical cap. At b or any other point in the future of a and not equal to a, the sphere is entirely in the future of (a, p). 18 - p p - p p - p p - p p
1307.1120	1	1307	2013-07-03T19:51:27	Graphs, groups and self-similarity	[ "math.OA", "math.DS", "math.GR" ]	We study a family of C*-algebras generalizing both Katsura algebras and certain algebras introduced by Nekrashevych in terms of self-similar groups.	math.OA	math	GRAPHS, GROUPS AND SELF-SIMILARITY Ruy Exel and Enrique Pardo We study a family of C-algebras generalizing both Katsura algebras and certain alge- bras introduced by Nekrashevych in terms of self-similar groups. 1. Introduction. The purpose of this paper is to give a unified treatment to two classes of C-algebras which have been studied in the past few years from rather different points of view, namely Katsura's algebras [7], and certain algebras constructed by Nekrashevych [10], [11] from self-similar groups. The realization that these classes are indeed closely related, as well as the fact that they could be given a unified treatment, came to our mind as a result of our earlier attempt [3] to understand Katsura's algebras OA,B from the point of view of inverse semigroups. The fact, proven by Katsura in [7], that all Kirchberg algebras may be described in terms of his OA,B was, in turn, a strong motivation for that endeavor. While studying OA,B, it slowly became clear to us that the two matricial parameters A and B play very different roles. The reader acquainted with Katsura's work will easily recognize that the matrix A is destined to be viewed as the edge matrix of a graph, but it took us much longer to realize that B should be thought of as providing parameters for an action of the group Z on the graph given by A. In trying to understand these different roles, some interesting arithmetic popped up sparking a connection with the work done by Nekrashevych [11] on the C-algebra O(G,X) associated to a self-similar group (G, X). While Nekrashevych's algebras contain a Cuntz algebra, Katsura's algebras contain a graph C-algebra. This fact alone ought to be considered as a hint that self-similar groups lie in a much bigger class, where the group action takes place on the path space of a graph, rather than on a rooted tree (which, incidentally, is the path space of a bouquet of circles). One of the first important applications of the idea of self-similarity in group theory is in constructing groups with exotic properties [5], [6]. Many of these are defined as subgroups of the group of all automorphisms of a tree. Having been born from automorphisms, it is natural that the theory of self-similar groups generally assumes that the group acts faithfully on its tree (see, e.g. [11: Definition 2.1]). However, based on the example provided by Katsura's algebras, we decided that perhaps it is best to view the group on its own, the action being an extra ingredient. Date: 4 July 2013. 2010 Mathematics Subject Classification: 46L05, 46L55. Key words and phrases: Kirchberg algebra, Katsura algebra, tight representation, inverse semigroup, groupoid, groupoid C-algebra. The first-named author was partially supported by CNPq. The second-named author was partially supported by PAI III grants FQM-298 and P11-FQM-7156 of the Junta de Andaluc´ıa, by the DGI-MICINN and European Regional Development Fund, jointly, through Project MTM2011-28992-C02-02 and by 2009 SGR 1389 grant of the Comissionat per Universitats i Recerca de la Generalitat de Catalunya. 2 r. exel and e. pardo The main idea behind self-similar groups, namely the equation appearing in [11: Definition 2.1], and the subsequent notion of restriction, namely g(xw) = yh(w) (1.1) gx := h, depend on faithfulness, since otherwise the group element h appearing in (1.1) would not be unique and therefore will not be well defined as a function of g and x. Working with non-faithful group actions we were forced to postulate a functional dependence h = ϕ(g, x), and we were surprised to find that the natural properties expected of ϕ are that of a group cocycle. To be precise, the ingredients needed in our generalization of self-similar groups are: a countable discrete group G, an action of G on a finite graph E = (E0, E1, r, d), and a one-cocycle G × E → E ϕ : G × E1 → G for the action of G on the edges of E. Starting with this data (satisfying a few other natural axioms) we construct an action of G on the space of finite paths E∗ which satisfy the "self-similarity" equation g(αβ) = (gα)(cid:0)ϕ(g, α)β(cid:1), ∀ g ∈ E, ∀ α, β ∈ E∗. Adopting a philosophy similar to that embraced by Katsura and Nekrashevych, we define a C-algebra, denoted OG,E, in terms of generators and relations inspired by the above group action. The study of OG,E is, thus, the purpose of this paper. Given a self-similar group (G, X), if we consider X as the set of edges of a graph with a single vertex, and if we define ϕ(g, x) = gx, then our OG,E coincides with Nekrashevych's O(G,X). On the other hand, if we are given two integer N × N matrices A and B, we may form a graph E with vertex set E0 = {1, 2, . . . , N } and with Ai,j edges from vertex i to vertex j. We may then use B to define an action of Z on E, by fixing all vertexes and acting on the set of edges as follows: denote the edges in E from i to j by ei,j,n, where 0 ≤ n < Ai,j. Given m ∈ Z, we perform the Euclidean division of mBi,j + n by Ai,j, say mBi,j + n = kAi,j + n graphs, groups and self-similarity 3 with 0 ≤ n < Ai,j, and put σm(ei,j,n) = ei,j,n, so that the group element m permutes the Ai,j edges from i to j in the same way that addition by mBi,j, modulo Ai,j, permutes the integers {0, 1, . . . , Ai,j − 1}. The quotient k on the above Euclidean division also plays an important role, being used in the definition of the cocycle ϕ(m, ei,j,n) = k. In possession of the graph, the action of Z, and the cocycle ϕ, we apply our construc- tion and find that OG,E is isomorphic to Katsura's OA,B. So, both Nekrashevych's and Katsura's algebras become special cases of our construc- tion. We therefore believe that the project of studying such group actions on path spaces as well as the corresponding algebras is of great importance. Taking the first few steps we have been able to describe OG,E as the C-algebra of an ´etale groupoid GG,E, whose construction is remarkably similar to the groupoid associated to the relation of "tail equivalence with lag" on the path space, as described by Kumjian, Pask, Raeburn and Renault in [8]. The first similarity is that our groupoid GG,E has the exact same unit space as the corresponding graph groupoid, namely the infinite path space. The second, and most surprising similarity is that GG,E is also described by a lag function, except that the values of the lag are not integer numbers, as in [8], but lie in a slightly more complicated group, the semi-direct product of the corona group of G by the right shift automorphism (see below for precise definitions). The techniques we use to give OG,E a groupoid model bear heavily on the theory of tight representations of inverse semigroups developed by the first named author in [2]. In particular, from our initial data we construct an abstract inverse semigroup SG,E and show that OG,E is the universal C-algebra for tight representations of SG,E. As a second step we again take inspiration from Nekrashevych [10] and give a descrip- tion of OG,E as a Cuntz Pimnsner algebra for a very natural correspondence M over the algebra C(E0) ⋊ G. As a result we are able to prove that OG,E is nuclear when the G is amenable. We would like to stress that, like Nekrashevych [11: Theorem 5.1], our groupoid GG,E is constructed as a groupoid of germs. However, departing from Nekrashevych's techniques, we use Patterson's [12] notion of "germs", rather than the one employed in [11: Section 5]. While agreeing in many cases, the former has a much better chance of producing Hausdorff groupoids and, in our case, we may give a precise characterization of Hausdorffness in terms of a property we call residual freeness (see below for the precise definition). Part of this work was done during a visit of the second named author to the Departa- mento de Matem´atica da Universidade Federal de Santa Catarina (Florian´opolis, Brasil) and he would like to express his thanks to the host center for its warm hospitality. 4 r. exel and e. pardo 2. Groups acting on graphs. Let E = (E0, E1, r, d) be a directed graph, where E0 denotes the set of vertexes, E1 is the set of edges, r is the range map, and d is the source, or domain map. By definition, a source in E is a vertex x ∈ E0, for which r−1(x) = ∅. Thus, when we say that a graph has no sources, we mean that r−1(x) 6= ∅, for all x ∈ E0. By an automorphism of E we shall mean a bijective map σ : E0 ∪ E1 → E0 ∪ E1 such that σ(Ei) ⊆ Ei, for i = 0, 1, and moreover such that r ◦ σ = σ ◦ r, and d ◦ σ = σ ◦ d, on E1. It is evident that the collection of all automorphisms of E forms a group under composition. By an action of a group G on a graph E we shall mean a group homomorphism from G to the group of all automorphisms of E. If X is any set, and if σ is an action of a group G on X, we shall say that a map ϕ : G × X → G is a one-cocycle for σ, when for all g, h ∈ G, and all x ∈ X. Plugging g = h = 1 above we see that ϕ(gh, x) = ϕ(cid:0)g, σh(x)(cid:1)ϕ(h, x), for every x. ϕ(1, x) = 1, (2.1) (2.2) 2.3. Standing Hypothesis. Throughout this work we shall let G be a countable discrete group, E be a finite graph with no sources, σ be an action of G on E, and ϕ : G × E1 → G be a one-cocycle for the restriction of σ to E1, which moreover satisfies σϕ(g,e)(x) = σg(x), ∀ g ∈ G, ∀ e ∈ E1, ∀ x ∈ E0. (2.3.1) The assumptions that E is finite and has no sources will in fact only be used in the next section and it could probably be removed by using well known graph C-algebra techniques. By a path in E of length n ≥ 1 we shall mean any finite sequence of the form α = α1α2 . . . αn, where αi ∈ E1, and d(αi) = r(αi+1), for all i (this is the usual convention when treating graphs from a categorical point of view, in which functions compose from right to left). The range of α is defined by r(α) = r(α1), graphs, groups and self-similarity 5 while the source of α is defined by d(α) = d(αn). A vertex x ∈ E0 is considered to be a path of length zero, in which case we set r(x) = d(x) = x. For every integer n ≥ 0 we denote by En the set of all paths in E of length n (this being consistent with the already introduced notations for E0 and E1). Finally, we denote by E∗ the sets of all finite paths, and by E≤n the set of all paths of length at most n, namely E∗ = [k≥0 Ek, and E≤n = Ek. n[k=0 We will often employ the operation of concatenation of paths. That is, if (and only if) α and β are paths such that d(α) = r(β), we will denote by αβ the path obtained by juxtaposing α and β. In the special case in which α is a path of length zero, the concatenation αβ is allowed if and only if α = r(β), in which case we set αβ = β. Similarly, when β = 0, then αβ is defined iff d(α) = β, and then αβ = α. We would now like to describe a certain extension of σ and ϕ to finite paths. g (En) ⊆ En, g = σg, on E≤1, 2.4. Proposition. Under the assumptions of (2.3) there exists a unique pair (σ∗, ϕ∗), formed by an action σ∗ of G on E∗ (viewed simply as a set), and a one-cocycle ϕ∗ for σ∗, such that, for every n ≥ 0, every g ∈ G, and every x ∈ E0, one has that: (i) σ∗ (ii) ϕ∗(g, x) = g, (iii) ϕ∗ = ϕ, on G × E1, (iv) σ∗ (v) r ◦ σ∗ (vi) d ◦ σ∗ (vii) σϕ∗(g,α)(x) = σg(x), for all α ∈ En, (viii) σ∗ (ix) σ∗ ϕ∗(g,α)(β), provided α and β are finite paths with αβ ∈ En, 1 is the identity 1 on En, g (αβ) = σ∗ g = σg ◦ r, on En, g = σg ◦ d, on En, (x) ϕ∗(g, αβ) = ϕ∗(cid:0)ϕ∗(g, α), β(cid:1), provided α and β are finite paths with αβ ∈ En. Proof. Initially notice that, once (v), (vi) and (vii) are proved, the concatenation of the paths "σ∗ ϕ∗(g,α)(β)", appearing in (ix), is permitted because g (α)" and "σ∗ g (α) σ∗ r(cid:0)σ∗ ϕ∗(g,α)(β)(cid:1) (v) = σϕ∗(g,α)(r(β)) (vii) = σg(r(β)) = σg(cid:0)d(α)(cid:1) (vi) = d(cid:0)σ∗ g (α)(cid:1). 1 This is evidently already included in the statement that σ∗ is an action, but we repeat it here to aid our proof by induction. 6 r. exel and e. pardo For every g in G, define σ∗ g on E≤1 to coincide with σg. Also, define ϕ∗ on G × E≤1 by (ii) and (iii). It is then clear that (i -- iii) hold and it is easy to see that the remaining properties (iv -- x) hold for all n ≤ 1. We shall complete the definitions of σ∗ and ϕ∗ by induction, so we assume that m ≥ 1, that is defined for all g in G, that σ∗ g : E≤m → E≤m ϕ∗ : G × E≤m → G, is defined, and that (i -- x) hold for all n ≤ m. We then define for all g in G, and σ∗ g : Em+1 → Em+1 ϕ∗ : G × Em+1 → G, by induction as follows. Given α ∈ Em+1, write α = α′α′′, with α′ ∈ E1, and α′′ ∈ Em, and put σ∗ g (α) = σg(α′)σ∗ ϕ(g,α′)(α′′), and ϕ∗(g, α) = ϕ∗(cid:0)ϕ(g, α′), α′′(cid:1). (2.4.1) A quick analysis, as done in the first paragraph of this proof, shows that the concatenation of "σg(α′)" and "σ∗ ϕ(g,α′)(α′′)", appearing above, is permitted. We next verify (iv -- x), substituting m + 1 for n. We have that the length of σ∗ With respect to (v) we have that g (α), as defined above, is clearly 1 + m, thus proving (iv). r(cid:0)σ∗ g (α)(cid:1) = r(cid:0)σg(α′)(cid:1) = σg(cid:0)r(α′)(cid:1) = σg(cid:0)r(α)(cid:1). As for (vi), notice that d(cid:0)σ∗ g (α)(cid:1) = d(cid:0)σ∗ ϕ(g,α′)(α′′)(cid:1) = σϕ(g,α′)(cid:0)d(α′′)(cid:1) = σg(cid:0)d(α′′)(cid:1) = σg(cid:0)d(α)(cid:1). Given x ∈ E0, we have that σϕ∗(g,α)(x) = σϕ∗(ϕ(g,α′),α′′)(x) = σϕ(g,α′)(x) = σg(x), taking care of (vii). The verification of (viii) is done as follows: for α = α′α′′, as in (2.4.1), one has σ∗ 1(α) = σ∗ 1(α′α′′) = σ1(α′)σ∗ ϕ(1,α′)(α′′) (2.2) = σ1(α′)σ∗ 1(α′′) = α′α′′ = α. In order to prove (ix), pick paths α in Ek and β in El, where k + l = m + 1, and such that d(α) = r(β). graphs, groups and self-similarity 7 We leave it for the reader to verify (ix) in the easy case in which k = 0, that is, when g given in α is a vertex. The case k = 1 is also easy as it is nothing but the definition of σ∗ (2.4.1). So we may assume that k ≥ 2. Writing α = α′α′′, with α′ ∈ E1, and α′′ ∈ Ek−1, we then have that αβ = α′α′′β, and hence, by definition, σ∗ g (αβ) = σg(α′)σ∗ ϕ(g,α′)(α′′β) = σg(α′)σ∗ ϕ(g,α′)(α′′) σ∗ ϕ∗(ϕ(g,α′),α′′)(β) = = σ∗ g (α′α′′) σ∗ ϕ∗(g,α′α′′)(β). We remark that, in last step above, one should use the induction hypothesis in case k ≤ m, and the definitions of σ∗ and ϕ∗, when k = m + 1. To verify (x) we again pick paths α in Ek and β in El, where k + l = m + 1, and such that d(α) = r(β). We once more leave the easy case k = 0 to the reader and observe that the case k = 1 follows from the definition of ϕ∗. We may then suppose that k ≥ 2, so we write α = α′α′′, with α′ ∈ E1, and α′′ ∈ Ek−1. Then ϕ∗(g, αβ) = ϕ∗(g, α′α′′β) = ϕ∗(cid:0)ϕ(g, α′), α′′β(cid:1) = ϕ∗(cid:16)ϕ∗(cid:0)ϕ(g, α′), α′′(cid:1), β(cid:17) = = ϕ∗(cid:16)ϕ∗(g, α′α′′(cid:1), β(cid:17) = ϕ∗(cid:16)ϕ∗(g, α(cid:1), β(cid:17). σ∗ g σ∗ Let us now prove that σ∗ is in fact an action of G on En. We begin by proving that h = σ∗ This follows immediately from the hypothesis for n ≤ 1, so let us assume that n ≥ 2. gh on En, for every g and h in G, which we do by induction on n. Given α ∈ En, write α = α′α′′, with α′ ∈ E1, and α′′ ∈ En−1. Then σ∗ g(cid:0)σ∗ h(α)(cid:1) = σ∗ g(cid:0)σ∗ h(α′α′′)(cid:1) = σ∗ ϕ(g,σh(α′))(cid:0)σϕ(h,α′)(α′′)(cid:1) = σgh(α′)σ∗ = σg(cid:0)σh(α′)(cid:1)σ∗ g(cid:0)σh(α′)σϕ(h,α′)(α′′)(cid:1) = ϕ(gh,α′)(α′′) = σ∗ gh(α′α′′) = σ∗ = σgh(α′)σ∗ gh(α). ϕ(g,σh(α′))ϕ(h,α′)(α′′) = That α∗ g is bijective on each En then follows2 from (viii), so α∗ is indeed an action of G on En. Finally, let us show that ϕ∗ is a cocycle for σ∗ on En. For this fix g and h in G and let α ∈ En. Then, with α = α′α′′, as before, ϕ∗(gh, α) = ϕ∗(gh, α′α′′) = ϕ∗(ϕ(gh, α′), α′′) = ϕ∗(cid:16)ϕ(cid:0)g, σh(α′)(cid:1)ϕ(h, α′), α′′(cid:17) = which coincides with (⋆) above. This concludes the proof. (cid:3) 2 This is why it is useful to include (viii) as a separate statement, since we may now use it to prove bijectivity. On the other hand, focusing on the right-hand-side of (2.1), notice that ϕ∗(g, σ∗ = ϕ∗(cid:16)ϕ(cid:0)g, σh(α′)(cid:1), σ∗ ϕ(h,α′)(α′′)(cid:17)ϕ∗(cid:0)ϕ(h, α′), α′′(cid:1) =: (⋆). h(α′α′′)(cid:1)ϕ∗(h, α′α′′) = ϕ(h,α′)(α′′)(cid:1)ϕ∗(cid:0)ϕ(h, α′), α′′(cid:1) = ϕ(h,α′)(α′′)(cid:17)ϕ∗(cid:0)ϕ(h, α′), α′′(cid:1), h(α))ϕ∗(h, α) = ϕ∗(cid:0)g, σ∗ = ϕ∗(cid:0)g, σh(α′)σ∗ = ϕ∗(cid:16)ϕ(cid:0)g, σh(α′)(cid:1), σ∗ 8 r. exel and e. pardo The only action of G on E∗ to be considered in this paper is σ∗ so, from now on, we will adopt the shorthand notation gα = σ∗ g (α). Moreover, since ϕ∗ extends ϕ, we will drop the star decoration and denote ϕ∗ simply as ϕ. The group law, the cocycle condition, and properties (ii, v, vi, vii, ix, x) of (2.4) may then be rewritten as follows: 2.5. Equations. For every g and h in G, for every x ∈ E0, and for every α and β in E∗ such that d(α) = r(β), one has that (a) (gh)α = g(hα), (b) ϕ(gh, α) = ϕ(cid:0)g, hα(cid:1)ϕ(h, α), (ii) ϕ(g, x) = g, (v) r(gα) = gr(α), (vi) d(gα) = gd(α), (vii) ϕ(g, α)x = gx, (ix) g(αβ) = (gα) ϕ(g, α)β, (x) ϕ(g, αβ) = ϕ(cid:0)ϕ(g, α), β(cid:1). It might be worth noticing that if ϕ(g, α) = 1, then (2.5.ix) reads "g(αβ) = (gα)β", which may be viewed as an associativity property. However associativity does not hold in general as ϕ is not always trivial, and hence parentheses must be used. On the other hand parentheses are unnecessary in expressions of the form αgβ, when α, β ∈ E∗, and g ∈ G, since the only possible interpretation for this expression is the concatenation of α with gβ. Another useful property of ϕ is in order. 2.6. Proposition. For every g ∈ G, and every α ∈ E∗, one has that ϕ(g−1, α) = ϕ(g, g−1α)−1. Proof. We have 1 = ϕ(1, α) = ϕ(gg−1, α) = ϕ(g, g−1α)ϕ(g−1, α), from where the conclusion follows. (cid:3) 3. The universal C-algebra OG,E. As in the above section we fix a graph E, an action of a group G on E, and a one-cocycle ϕ satisfying (2.3). It is our next goal to build a C-algebra from this data but first let us recall the following notion from [15]: graphs, groups and self-similarity 9 3.1. Definition. A Cuntz-Krieger E-family consists of a set {px : x ∈ E0} of mutually orthogonal projections and a set {se : e ∈ E1} of partial isometries satisfying (i) s∗ ese = pd(e), for every e ∈ E1. (ii) px = Xe∈r−1(x) ses∗ e, for every x ∈ E0 for which r−1(x) is finite and nonempty. 3.2. Definition. We define OG,E to be the universal unital C-algebra generated by a set {px : x ∈ E0} ∪ {se : e ∈ E1} ∪ {ug : g ∈ G}, subject to the following relations: (a) {px : x ∈ E0} ∪ {se : e ∈ E1} is a Cuntz-Krieger E-family, (b) the map u : G → OG,E, defined by the rule g 7→ ug, is a unitary representation of G, ( c ) ugse = sgeuϕ(g,e), for every g ∈ G, and e ∈ E1, (d) ugpx = pgxug, for every g ∈ G, and x ∈ E0. Observe that, under our standing assumptions (2.3), for every x ∈ E0 we have that r−1(x) is finite and nonempty. So (3.1.ii) and (3.2.a) imply that ugpxu∗ ugses∗ g = Xr(e)=x = Xr(e)=x eu∗ g = Xr(e)=x ge = Xr(f )=gx sges∗ sgeuϕ(g,e)u∗ ϕ(g,e)s∗ ge = sf s∗ f = pgx, which says that (3.2.d) follows from the other conditions. We have nevertheless included it in (3.2) in the belief that our theory may be generalized to graphs with sources. Our construction generalizes some well known constructions in the literature as we would now like to mention. 3.3. Example. Let (G, X) be a self similar group as in [11: Definition 2.1]. We may then consider a graph E having only one vertex and such that E1 = X. If we define ϕ(g, x) = gx, where, in the terminology of [11], gx is the restriction (or section) of g at x, then the triple (G, E, ϕ) satisfies (2.3) and one may show that OG,E is isomorphic to the algebra O(G,X) introduced by Nekrashevych in [11]. 10 r. exel and e. pardo 3.4. Example. As in [7], given a positive integer N , let A ∈ MN (Z and let B ∈ MN (Z) be such that +) without zero rows Consider the graph E with Ai,j = 0 ⇒ Bi,j = 0. E0 = {1, 2, . . . , N }, and whose adjacency matrix is A. For each pair of vertexes i, j ∈ E0, such that Ai,j 6= 0, denote the set of edges with range i and source j by {ei,j,n : 0 ≤ n < Ai,j}. Define an action σ of Z on E, which is the identity on E0, and which acts on edges as follows: given m ∈ Z, and ei,j,n ∈ E1, let (k, n) be the unique pair of integers such that mBi,j + n = kAi,j + n, and 0 ≤ n < Ai,j. Thus, k is the quotient and n is the remainder of the division of mBi,j + n by Ai,j. We then put σm(ei,j,n) = ei,j,n. In other words, σm corresponds to the addition of mBi,j to the variable "n" of "ei,j,n", taken modulo Ai,j. In turn, the one-cocycle is defined by ϕ(m, ei,j,n) = k. It may be proved without much difficulty that OG,E is isomorphic to Katsura's [7] algebra OA,B, under an isomorphism sending each um to the mth power of the unitary u := NXi=1 ui in OA,B, and sending sei,j,n to si,j,n. When N = 1, the relevant graph for Katsura's algebras is the same as the one we used above in the description of Nekrashevych's example. However the former is not a special case of the latter because, contrary to what is required in [11], the group action might not be faithful. We now return to the general case of a triple (G, E, ϕ) satisfying (2.3). We initially recall the usual extension of the notation "se" to allow for paths of arbitrary length. 3.5. Definition. Given a finite path α in E∗, we shall let sα denote the element of OG,E given by: (i) when α = x ∈ E0, we let sα = px, (ii) when α ∈ E1, then sα is already defined above, (iii) when α ∈ En, with n > 1, write α = α′α′′, with α′ ∈ E1, and α′′ ∈ En−1, and set sα = sα′sα′′, by recurrence. graphs, groups and self-similarity 11 Commutation relation (3.2.c) may then be generalized to finite paths as follows: 3.6. Lemma. Given α ∈ E∗ and g ∈ G, one has that ugsα = sgαuϕ(g,α). Proof. Let n be the length of α. When n = 0, 1, this follows from (3.2.d&c), respectively. When n > 1, write α = α′α′′, with α′ ∈ E1, and α′′ ∈ En−1. Using induction, we then have ugsα = ugsα′ sα′′ = sgα′uϕ(g,α′)sα′′ = sgα′sϕ(g,α′)α′′ uϕ(ϕ(g,α′),α′′) = = s(gα′)ϕ(g,α′)α′′ uϕ(g,α′α′′) = sg(α′α′′)uϕ(g,α′α′′) = sgαuϕ(g,α). (cid:3) Our next result provides a spanning set for OG,E. 3.7. Proposition. Let S =(cid:8)sαugs∗ β : α, β ∈ E∗, g ∈ G, d(α) = gd(β)(cid:9) ∪ {0}. Then S is closed under multiplication and adjoints and its closed linear span coincides with OG,E. Proof. That S is closed under adjoints is clear. With respect to closure under multiplica- tion, let sαugs∗ δ be elements of S. β and sγuhs∗ From (3.2.a) we know that s∗ βsγ = 0, unless either γ = βε, or β = γε, for some ε ∈ E∗. If γ = βε, then and hence s∗ βsγ = s∗ βsβε = s∗ βsβsε = sε, (sαugs∗ β)(sγuhs∗ δ) = sαugsεuhs∗ δ = sαsgεuϕ(g,ε)uhs∗ δ = sαgεuϕ(g,ε)hs∗ δ. (3.7.1) Moreover, since d(αgε) = d(gε) = gd(ε) = ϕ(g, ε)d(ε) = ϕ(g, ε)d(γ) = ϕ(g, ε)hd(δ), we deduce that the element appearing in the right-hand-side of (3.7.1) indeed belongs to S. In the second case, namely if β = γε, then the adjoint of the term appearing in the left-hand-side of (3.7.1) is (sδuh−1s∗ γ)(sβug−1s∗ α), and the case already dealt with implies that this belongs to S. The result then follows from the fact that S is self-adjoint. In order to prove that OG,E coincides with the closed linear span of S, let A denote the latter. Given that S is self-adjoint and closed under multiplication, we see that A is a closed -subalgebra of OG,E. Since A evidently contains sα for every α in E≤1, and that it also contains ug for every g in G, we deduce that A = OG,E. (cid:3) 12 r. exel and e. pardo 4. The inverse semigroup SG,E. As before, we keep (2.3) in force. In this section we will give an abstract description of the set S appearing in (3.7) as well as its multiplication and adjoint operation. The goal is to construct an inverse semigroup from which we will later recover OG,E. 4.1. Definition. Over the set SG,E =(cid:8)(α, g, β) ∈ E∗ × G × E∗ : d(α) = gd(β)(cid:9) ∪ {0}, consider a binary multiplication operation defined by (α, g, β)(γ, h, δ) = (αgε, ϕ(g, ε)h, δ), if γ = βε, (α, gϕ(h−1, ε)−1, δh−1ε), if β = γε, 0, otherwise,   and a unary adjoint operation defined by (α, g, β)∗ := (β, g−1, α). The subset of SG,E formed by all elements (α, g, β), with g = 1, will be denoted by SE. It is easy to see that SE is closed under the above operations, and that it is isomorphic to the inverse semigroup generated by the canonical partial isometries in the graph C- algebra of E. Let us begin with a simple, but useful result: 4.2. Lemma. Given (α, g, β) and (γ, h, δ) in SG,E, one has β = γ ⇒ (α, g, β)(γ, h, δ) = (α, gh, δ). Proof. Focusing on the first clause of (4.1), write γ = βε, with ε = d(β). Then (α, g, β)(γ, h, δ) = (αgd(β), ϕ(cid:0)g, d(β)(cid:1)h, δ) = (αd(α), gh, δ) = (α, gh, δ). 4.3. Proposition. SG,E is an inverse semigroup with zero. (cid:3) Proof. We leave it for the reader to prove that the above operations are well defined and associative. In order to prove the statement it then suffices [9: Theorem 1.1.3] to show that, for all y, z ∈ SG,E, one has that (i) yy∗y = y, and (ii) yy∗ commutes with zz∗. Given y = (α, g, β) ∈ SG,E, we have by the above Lemma that yy∗y = (α, g, β)(β, g−1, α)(α, g, β) = (α, 1, α)(α, g, β) = (α, g, β) = y, proving (i). Notice also that is an element of the idempotent semi-lattice of SE, which is a commutative set because SE is an inverse semigroup. Point (ii) above then follows immediately, concluding the proof. (cid:3) yy∗ = (α, 1, α) (4.3.1) graphs, groups and self-similarity 13 As seen in (4.3.1), the idempotent semi-lattice of SG,E, henceforth denoted by E, is given by E =(cid:8)(α, 1, α) : α ∈ E∗(cid:9) ∪ {0}. Evidently E is also the idempotent semi-lattice of SE. For simplicity, from now on we will adopt the short-hand notation eα = (α, 1, α), ∀ α ∈ E∗. The following is a standard fact in the theory of graph C-algebras: 4.6. Proposition. If α, β ∈ E∗, then (4.4) (4.5) eαeβ =  eα, eβ, if there exists γ such that α = γβ, if there exists γ such that αγ = β, 0, otherwise. Recall that if α and β are in E∗, we say that α (cid:22) β, if α is a prefix of β, i.e. if there exists γ ∈ E∗, such that αγ = β. It therefore follows from (4.6) that eα ≤ eβ ⇐⇒ β (cid:22) α. (4.7) Another easy consequence of (4.6) is that, for any two elements e, f ∈ E, one has that either e ⊥ f , or e and f are comparable. In other words e ⋓ f ⇒ e ≤ f , or f ≤ e. (4.8) Recall that, according to [2: Definition 11.1], we say that e intersects f , in symbols e ⋓ f , when ef 6= 0. 5. Residual freeness and E-unitarity. Again working under (2.3), suppose we are given g in G and α in E∗ such that gα = α, and ϕ(g, α) = 1. Then, by (2.5.ix), we have that g(αβ) = αβ, whenever d(α) = r(β). Such an element g therefore acts trivially on a large set of finite words. Occasionally this will be an annoying feature which we would rather avoid, so we make the following: 5.1. Definition. We will say that (G, E, ϕ) is residually free if, whenever (g, e) ∈ G×E1, is such that ge = e, and ϕ(g, e) = 1, then g = 1. This property may be generalized to finite paths: 5.2. Proposition. Suppose that (G, E, ϕ) is residually free and that (g, α) ∈ G × E∗, is such that gα = α, and ϕ(g, α) = 1, then g = 1. 14 r. exel and e. pardo Proof. Assume that there is a counter-example (g, α) to the statement, which we assume is minimal in the sense that α is as small as possible. To be sure, to say that (g, α) is a counter-example is to say that gα = α, ϕ(g, α) = 1, and yet g 6= 1. By (2.5.ii), α can't be a vertex, and neither can it be an edge, by hypothesis. So α ≥ 2, and we may then write α = βγ, with β, γ ∈ E∗, and β, γ < α. Then βγ = α = gα = g(βγ) = (gβ)ϕ(g, β)γ, whence β = gβ, and γ = ϕ(g, β)γ, by length considerations. Should ϕ(g, β) = 1, the pair (g, β) would be a smaller counter-example to the statement, violating the minimality of α. So we have that ϕ(g, β) 6= 1. In addition, ϕ(cid:0)ϕ(g, β), γ(cid:1) = ϕ(g, βγ) = ϕ(g, α) = 1. It follows that(cid:0)ϕ(g, β), γ(cid:1) is a counter-example to the statement, violating the minimality of α. This is a contradiction and hence no counter-example exists whatsoever, concluding the proof. (cid:3) An apparently stronger version of residual freeness is in order. 5.3. Proposition. Suppose that (G, E, ϕ) is residually free. Then, for all g1, g2 ∈ G, and α ∈ E∗, one has that g1α = g2α ∧ ϕ(g1, α) = ϕ(g2, α) ⇒ g1 = g2. Proof. Defining g = g−1 2 g1, observe that gα = α, and we claim that ϕ(g, α) = 1. In fact, ϕ(g, α) = ϕ(cid:0)g−1 2 g1, α(cid:1) = ϕ(cid:0)g−1 2 , g1α(cid:1)ϕ(cid:0)g1, α(cid:1) (2.6) = ϕ(cid:0)g2, g−1 2 g1α(cid:1)−1 ϕ(g1, α) = = ϕ(g2, α)−1ϕ(g1, α) = 1, so it follows that g = 1, which is to say that g1 = g2. (cid:3) Recall that an inverse semigroup S with zero is called E-unitary, or sometimes 0-E- unitary [9: Chapter 9] if, whenever an element s ∈ S dominates a nonzero idempotent e, meaning that se = e, then s is necessarily also idempotent. Residual freeness is very closely related to E-unitary inverse semigroups, as we would now like to show. 5.4. Proposition. (G, E, ϕ) is residually free if and only if SG,E is an E-unitary inverse semigroup. Proof. Assuming that (G, E, ϕ) is residually free, let s = (α, g, β) be in SG,E, and let eγ = (γ, 1, γ) be a nonzero idempotent in E, such that eγ ≤ s. It follows that also eγ ≤ s∗s = (β, 1, β), graphs, groups and self-similarity 15 so γ = βε, for some ε ∈ E∗, by (4.6). The relation "eγ = seγ" translates into (γ, 1, γ) = (α, g, β)(γ, 1, γ) =(cid:0)αgε, ϕ(g, ε), γ(cid:1), (5.4.1) so βε = γ = αgε, which implies that ε = gε, and β = α. If we further notice that (5.4.1) gives ϕ(g, ε) = 1, we may conclude from (5.2) that g = 1, and hence that s = (α, g, β) = (β, 1, β) is an idempotent element. This shows that SG,E is E-unitary. In order to prove the converse, let (g, e) ∈ G×E1 be such that ge = e, and ϕ(g, e) = 1. Then (1, g, 1)(e, 1, e) = (ge, ϕ(g, e), e) = (e, 1, e). Therefore the nonzero element (1, g, 1) dominates the idempotent (e, 1, e) and, assuming that SG,E is E-unitary, we conclude that (1, g, 1) is idempotent, which is to say that g = 1. This proves that (G, E, ϕ) is residually free. (cid:3) 6. Tight representations of SG,E. As before, we keep (2.3) in force. It is the main goal of this section to show that OG,E is the universal C-algebra for tight representations of SG,E. Recall from (4.6) that eα ≤ ed(α), for every α ∈ E∗, so we see that the set {ex : x ∈ E0} (6.1) is a cover [2: Definition 11.5] for E. 6.2. Proposition. The map defined by π(0) = 0, and π : SG,E → OG,E, π(α, g, β) = sαugs∗ β, is a tight [2: Definition 13.1] representation. Proof. We leave it for the reader to show that π is in fact multiplicative and that it preserves adjoints. In order to prove that π is tight, we shall use the characterization given in [2: Propo- sition 11.8], observing that π satisfies condition (i) of [2: Proposition 11.7] because, with respect to the cover (6.1), we have that _x∈E0 π(ex) = Xx∈E0 π(ex) = Xx∈E0 px = 1, 16 r. exel and e. pardo by (3.2.a). So we assume that {eα1 , . . . , eαn } is a cover for a given eβ, where α1, . . . αn, β ∈ E∗, and we need to show that n_i=1 π(eαi) ≥ π(eβ). (6.2.1) In particular, for each i, we have that eαi ≤ eβ, which says that there exists γi ∈ E∗ such that αi = βγi. We shall prove (6.2.1) by induction on the variable L = min 1≤i≤n γi. If L = 0, we may pick i such that γi = 0, and then necessarily γi = d(β), in which case αi = β, and (6.2.1) is trivially true. Assuming that L ≥ 1, one sees that x := d(β) is not a source, meaning that r−1(x) is nonempty. Write and observe that r−1(x) = {e1, . . . , ek}, π(eβ) = sβs∗ β = sβpxs∗ β (3.2.a) = kXj=1 sβsej s∗ ej s∗ β = kXj=1 π(eβej ). In order to prove (6.2.1) it is therefore enough to show that n_i=1 π(eαi) ≥ π(eβej ), (6.2.2) for all j = 1, . . . , k. Fixing j we claim that eβej is covered by the set Z =(cid:8)eαi : 1 ≤ i ≤ n, eαi ≤ eβej(cid:9). In order to see this let x ∈ E be a nonzero element such that x ≤ eβej . Then x ≤ eβ, and so x ⋓ eαi for some i. Thus, to prove the claim it is enough to check that eαi lies in Z. Observe that xeβej eαi = xeαi 6= 0, which implies that eβej ⋓ eαi . By (4.8) we have that eβej and eαi are comparable, so either βej (cid:22) αi or αi (cid:22) βej , by (4.7). Since we are under the hypothesis that L ≥ 1, and hence that αi = βi + γi ≥ β + 1 = βej, we must have that βej (cid:22) αi, from where we deduce that eαi ≤ eβej , proving our claim. Employing the induction hypothesis we then deduce that _z∈Z π(z) ≥ π(eβej ), verifying (6.2.2), and thus concluding the proof. (cid:3) graphs, groups and self-similarity 17 We would now like to prove that the representation π above is in fact the universal tight representation of SG,E. 6.3. Theorem. Let A be a unital C-algebra and let ρ : SG,E → A be a tight represen- tation. Then there exists a unique unital -homomorphism ψ : OG,E → A, such that the diagram SG,E π −→ OG,E ................................................................................................................ .......... ρ yψ A commutes. Proof. We will initially prove that the elements px := ρ(x, 1, x), ∀x ∈ E0, ∀e ∈ E1, se := ρ(cid:0)e, 1, d(e)(cid:1), ug := Xx∈E0 ρ(x, g, g−1x), ∀g ∈ G, satisfy relations (3.2.a -- d). Since the ex (defined in (4.5)) are mutually orthogonal idempo- tents in SG,E, it is clear that the px are mutually orthogonal projections. Evidently the se are partial isometries so, in order to check (3.2.a), we must only verify (3.1.i) and (3.1.ii). With respect to the former, let e ∈ E1. Then s∗ e se = ρ(cid:0)(d(e), 1, e)(e, 1, d(e)(cid:1) = ρ(cid:0)d(e), 1, d(e)(cid:1) = pd(e), proving (3.1.i). In order to prove (3.1.ii), let x be a vertex such that r−1(x) is nonempty and write Putting qi = (ei, 1, ei), we then claim that the set r−1(x) =(cid:8)e1, . . . , en(cid:9). is a cover for q := (x, 1, x). idempotent f is dominated by q, then f ⋓ qi for some i. In order to prove this we must show that, if the nonzero (cid:8)q1, . . . , qn(cid:9) Let f = (α, 1, α) by (4.4) and notice that 0 6= f = f q = (α, 1, α)(x, 1, x). So α and x are comparable, and this can only happen when x = r(α). If α = 0 then necessarily α = x, so f = q, and it is clear that f ⋓ qi for all i. On the other hand, if α ≥ 1, we write α = α′α′′, 18 r. exel and e. pardo with α′ ∈ E1, so that r(α′) = r(α) = x, and hence α′ = ei, for some i. Therefore f qi = (α, 1, α)(ei, 1, ei) = (α, 1, α)(α′, 1, α′) = (α, 1, α) 6= 0, so f ⋓ qi, proving the claim. Since ρ is a tight representation, we deduce that ρ(q) = n_i=1 ρ(qi), but since the qi are easily seen to be pairwise orthogonal, their supremum coincides with their sum, whence px = ρ(q) = ρ(qi) = ρ(ei, 1, ei) = nXi=1 nXi=1 = nXi=1 ρ(cid:0)(ei, 1, d(ei)) (d(ei), 1, ei)(cid:1) = nXi=1 sei s∗ ei, thus verifying (3.1.ii), and hence proving (3.2.a). With respect to (3.2.b), let us first prove that u1 = 1. Considering the subsets of E given by X = ∅, Y = ∅, and Z =(cid:8)(x, 1, x) : x ∈ E0(cid:9), notice that, according to [2: Definition 11.4], one has that E X,Y = E, and that Z is a cover for E X,Y , as seen in (6.1). By the tightness condition [2: Definition 11.6] we have _z∈Z ρ(z) ≥ ^x∈X ρ(x) ∧ ^y∈Y ¬ρ(y). As explained in the discussion following [2: Definition 11.6], the right-hand-side above must be interpreted as 1 because X and Y are empty. On the other hand, since the ρ(z) are pairwise orthogonal, the supremum in the left-hand-side above becomes a sum, so 1 = Xz∈Z ρ(z) = Xx∈E0 ρ(x, 1, x) = u1. In order to prove that u is multiplicative, let g and h be in G. Then ug uh = Xx,y∈E0 ρ(cid:0)(x, g, g−1x)(y, h, h−1y)(cid:1) = = Xx∈E0 ρ(cid:0)(x, g, g−1x)(g−1x, h, h−1g−1x)(cid:1) = Xx∈E0 ρ(cid:0)x, gh, (gh)−1x(cid:1) = ugh. graphs, groups and self-similarity 19 We next claim that u∗ g = ug−1, for all g in G. To prove it we compute u∗ g = Xx∈E0 ρ(x, g, g−1x)∗ = Xx∈E0 ρ(g−1x, g−1, x) = · · · which, upon the change of variables y = g−1x, becomes · · · = Xy∈E0 ρ(y, g−1, gy) = ug−1. This shows that u is a unitary representation, verifying (3.2.b). Turning now our attention to (3.2.c), let g ∈ G and e ∈ E. Then ug se = Xx∈E0 ρ(x, g, g−1x) ρ(cid:0)e, 1, d(e)(cid:1) = ρ(cid:0)gr(e), g, r(e)(cid:1) ρ(cid:0)e, 1, d(e)(cid:1) = On the other hand = ρ(cid:0)r(ge)ge, ϕ(g, e), d(e)(cid:1) = ρ(cid:0)ge, ϕ(g, e), d(e)(cid:1) = (⋆). sge uϕ(g,e) = ρ(cid:0)ge, 1, d(ge)(cid:1) Xx∈E0 ρ(cid:0)x, ϕ(g, e), ϕ(g, e)−1x(cid:1) = = ρ(cid:0)ge, 1, d(ge)(cid:1) ρ(cid:0)d(ge), ϕ(g, e), g−1d(ge)(cid:1) = = ρ(cid:0)ge, 1, d(ge)(cid:1) ρ(cid:0)d(ge), ϕ(g, e), d(e)(cid:1) = ρ(cid:0)ge, ϕ(g, e), d(e)(cid:1), which coincides with (⋆) and hence proves (3.2.c). We leave the proof of (3.2.d) to the reader after which the universal property of OG,E intervenes to provide us with a - homomorphism ψ : OG,E → A sending px 7→ px, se 7→ se, and ug 7→ ug. Now we must show that ψ(cid:0)π(γ)(cid:1) = ρ(γ), ∀ γ ∈ SG,E. (6.3.1) We will first do so for the following special cases: (i) γ = (x, 1, x), for x ∈ E0, (ii) γ =(cid:0)e, 1, d(e)(cid:1), for e ∈ E1, (iii) γ = (x, g, g−1x), for x ∈ E0, and g ∈ G. In case (i) we have ψ(π(γ)) = ψ(π(x, 1, x)) = ψ(px) = px = ρ(x, 1, x) = ρ(γ). 20 As for (ii), Under (iii), r. exel and e. pardo ψ(π(γ)) = ψ(cid:0)π(cid:0)e, 1, d(e)(cid:1)(cid:1) = ψ(se) = se = ρ(cid:0)e, 1, d(e)(cid:1) = ρ(γ). ψ(π(γ)) = ψ(cid:0)π(x, g, g−1x)(cid:1) = ψ(pxugpg−1x) = ψ(pxug) = px ug = = ρ(x, 1, x) Xy∈E0 ρ(y, g, g−1y) = Xy∈E0 ρ(cid:0)(x, 1, x)(y, g, g−1y)(cid:1) = ρ(x, g, g−1x) = ρ(γ). In order to prove (6.3.1), it is now clearly enough to check that the -sub-semigroup of SG,E generated by the elements mentioned in (i -- iii), above, coincides with SG,E. Denoting this -sub-semigroup by T , we will first show that (cid:0)α, 1, d(α)(cid:1) is in T , for every α ∈ E∗. This is evident for α ≤ 1, so we suppose that α = α′α′′, with α′ ∈ E1, and r(α′′) = d(α′). We then have by induction that T ∋(cid:0)α′, 1, d(α′)(cid:1)(cid:0)α′′, 1, d(α′′)(cid:1) =(cid:0)αα′′, 1, d(α′′)(cid:1) =(cid:0)α, 1, d(α)(cid:1). Considering a general element (α, g, β) ∈ SG,E, let x = d(α), so that g−1x = d(β), and notice that T ∋(cid:0)α, 1, d(α)(cid:1)(x, g, g−1x)(cid:0)β, 1, d(β)(cid:1)∗ =(cid:0)α, 1, d(α)(cid:1)(cid:0)d(α), g, d(β)(cid:1)(cid:0)d(β), 1, β(cid:1) = (α, g, β), = which proves that T = SG,E, and hence that (6.3.1) holds. To conclude we observe that the uniqueness of ψ follows from the fact that OG,E is (cid:3) generated by the px, the se, and the ug. Given an inverse semigroup S with zero, recall from [2: Theorem 13.3] that Gtight(S) (denoted simply as Gtight in [2]) is the groupoid of germs for the natural action of S on the space of tight filters over its idempotent semi-lattice. Moreover the C-algebra of Gtight(S) is universal for tight representations of S. 6.4. Corollary. Under the assumptions of (2.3) one has that OG,E is isomorphic to the C-algebra of the groupoid Gtight(SG,E). Proof. Follows from [2: Theorem 13.3] and the uniqueness of universal C-algebras. (cid:3) We should notice that our requirement that G be countable in (2.3) is only used in the above proof, where the application of [2: Theorem 13.3] depends on the countability of SG,E. We conclude by analyzing the question of Hausdorffness for Gtight(SG,E). 6.5. Proposition. When (G, E, ϕ) is residually free, one has that Gtight(SG,E) is a Haus- dorff groupoid. Proof. By (5.4) we have that SG,E is E-unitary. The result then follows from [2: Propo- sitions 6.2 & 6.4]. (cid:3) graphs, groups and self-similarity 21 7. Corona Groups. It is our next goal to give a concrete description of Gtight(SG,E), similar to the description given to the groupoid associated to a row-finite graph in [8: Definition 2.3]. The crucial ingredient there is the notion of tail equivalence with lag. In this section we will construct a group where our generalized lag function will take values. Let G be a group. Within the infinite cartesian product3 consider the infinite direct sum G G∞ = Yn∈ G(∞) = Mn∈ G formed by the elements g = (gn)n∈ ∈ G∞ which are eventually trivial, that is, for which there exists n0 such that gn = 1, for all n ≥ n0. It is clear that G(∞) is a normal subgroup of G∞. 7.1. Definition. Given a group G, the corona of G is the quotient group G = G∞/G(∞). Consider the left and right shift endomorphisms of G∞ λ, ρ : G∞ → G∞ given for every g = (gn)n∈ ∈ G∞, by and λ(g)n = gn+1, ∀ n ∈ , ρ(g)n =(cid:26) 1, gn−1, if n = 0, if n ≥ 1. It is readily seen that G(∞) is invariant under both λ and ρ, so these pass to the quotient providing endomorphisms λ, ρ : G → G. For every g = (gn)n∈ ∈ G∞, we have that λ(ρ(g)) = g, and ρ(λ(g)) = (1, g2, g3, . . .) ≡ g, where we use "≡" to refer to the equivalence relation determined by the normal subgroup G(∞). Therefore both λρ and ρλ coincide with the identity, and hence λ and ρ are each other's inverse. In particular, they are both automorphisms of G. Iterating ρ therefore gives an action of Z on G. 7.4. Definition. Given any countable discrete group G, the lag group associated to G is the semi-direct product group G ⋊ρ Z. The reason we call this the "lag group" is that it will play a very important role in the next section, as the co-domain for our lag function. 3 For the purposes of this construction we adopt the convention that = {1, 2, 3, . . .}. (7.2) (7.3) 22 r. exel and e. pardo 8. The tight groupoid of SG,E. We would now like to give a detailed description of the groupoid Gtight(SG,E). As already mentioned this is the groupoid of germs for the natural action of SG,E on the space of tight filters over the idempotent semi-lattice E of SG,E. See [2: Section 4] for more details. Given an infinite word ξ = ξ1ξ2 . . . ∈ E∞, and an integer n ≥ 0, denote by ξn the finite word of length n given by ξn =  ξ1ξ2 . . . ξn, if n ≥ 1, r(ξ1), if n = 0. 8.1. Proposition. There is a unique action (g, ξ) ∈ G × E∞ 7→ gξ ∈ E∞ of G on E∞ such that, (gξ)n = g(ξn), for every g ∈ G, ξ ∈ E∞, and n ∈ . Proof. Left to the reader. (cid:3) Recall from (4.5) that, for any finite word α ∈ E∗, we denote by eα the idempotent element (α, 1, α) in E. Thus, given an infinite word ξ ∈ E∞, we may look at the subset which turns out to be an ultra-filter [2: Definition] over E. Denoting the set of all ultra- Fξ = {eξn : n ∈ } ⊆ E, that the correspondence ξ ∈ E∞ 7→ Fξ ∈ bE∞ may be proven to be a homeomorphism if E∞ is equipped with the product topology. filters over E by bE∞, as in [2: Definition 12.8], one may also show [2: Proposition 19.11] is bijective, and we will use it to identify E∞ and bE∞. Furthermore, this correspondence Since E is finite, E∞ is compact by Tychonov's Theorem, and consequently so is bE∞. Being the closure of bE∞ within bE [2: Theorem 12.9], the space bEtight formed by the tight filters therefore necessarily coincides with bE∞. Identifying bEtight with E∞, as above, we may transfer the canonical action of SG,E from the former to the latter resulting in the following: to each element (α, g, β) ∈ SG,E, we associate the partial homeomorphism of E∞ whose domain is the cylinder Z(β) := {η ∈ E∞ : η = βξ, for some ξ ∈ E∞}, and which sends each η = βξ ∈ Z(β) to αgξ, where the meaning of "gξ" is as in (8.1). graphs, groups and self-similarity 23 As before we will not use any special symbol to indicate this action, using module notation instead: (α, g, β)η = αgξ. Before we proceed let us at least check that αgξ is in fact an element of E∞, which is to say that d(α) = r(gξ). Firstly, for every element (α, g, β) ∈ SG,E, we have that d(α) = gd(β). Secondly, if η = βξ ∈ E∞, then d(β) = r(ξ). Therefore r(gξ) = gr(ξ) = gd(β) = d(α). This leads to a first, more or less concrete description of Gtight(SG,E), namely the groupoid of germs for the above action of SG,E on E∞. Our aim is nevertheless a much more precise description of it. Recall from [2: Definition 4.6] that the germ of an element s ∈ SG,E at a point ξ in the domain of s is denoted by [s, ξ]. If s = (α, g, β), this would lead to the somewhat awkward notation [(α, g, β), ξ], which from now on will be simply written as Thus the groupoid Gtight(SG,E), consisting of all germs for the action of SG,E on E∞, (cid:2)α, g, β; ξ(cid:3). is given by Gtight(SG,E) =n(cid:2)α, g, β; ξ(cid:3) : (α, g, β) ∈ SG,E, ξ ∈ Z(β)o. Let us now prove a criterion for equality of germs. (8.2) 8.3. Proposition. Suppose that (G, E, ϕ) is residually free and let us be given elements (α1, g1, β1) and (α2, g2, β2) in SG,E, with β1 ≤ β2, as well as infinite paths η1 in Z(β1), and η2 in Z(β2). Then (cid:2)α1, g1, β1; η1(cid:3) =(cid:2)α2, g2, β2; η2(cid:3) if and only if there is a finite path γ ∈ E∗ and an infinite path ξ ∈ E∞, such that (i) α2 = α1g1γ, (ii) g2 = ϕ(g1, γ), (iii) β2 = β1γ, (iv) η1 = η2 = β1γξ. Proof. Assuming that the germs are equal, we have by [2: Definition 4.6] that and there is an idempotent (δ, 1, δ) ∈ E, such that η ∈ Z(δ), and η1 = η2 =: η, (α1, g1, β1)(δ, 1, δ) = (α2, g2, β2)(δ, 1, δ). (8.3.1) It follows that η = δζ, for some ζ ∈ E∞. Upon replacing δ by a longer prefix of η, we may assume that δ is as large as we want. Furthermore the element of SG,E represented 24 r. exel and e. pardo by the two sides of the equation displayed above is evidently nonzero because the partial homeomorphism associated to it under our action has η in its domain. So, focusing on (4.1), we see that β1 and δ are comparable, and so are β2 and δ. Assuming that δ exceeds both β1 and β2, we may then write δ = β1ε1 = β2ε2, for suitable ε1 and ε2 in E∗. But since β1 ≤ β2, this in turn implies that β2 = β1γ, for some γ ∈ E∗, hence proving (iii). Therefore δ = β1γε2, so η = δζ = β1γε2ζ, and (iv) follows once we choose ξ = ε2ζ. Moreover, equation (8.3.1) reads (α1, g1, β1)(β1γε2, 1, β1γε2) = (α2, g2, β1γ)(β1γε2, 1, β1γε2). Computing the products according to (4.1), we get (cid:0)α1g1(γε2), ϕ(g1, γε2), β1γε2(cid:1) =(cid:0)α2g2ε2, ϕ(g2, ε2), β1γε2(cid:1), from where we obtain α2g2ε2 = α1g1(γε2) = α1(g1γ)ϕ(g1, γ)ε2, and ϕ(g2, ε2) = ϕ(g1, γε2) = ϕ(cid:0)ϕ(g1, γ), ε2(cid:1). Since g2ε2 = ε2 = ϕ(g1, γ)ε2, we deduce from (8.3.2) that and hence also that g2ε2 = ϕ(g1, γ)ε2, α2 = α1g1γ, proving (i). Defining g = g−1 2 ϕ(g1, γ), we claim that gε2 = ε2, and ϕ(g, ε2) = 1. (8.3.2) (8.3.3) (8.3.4) In view of (8.3.3) and (8.3.4), point (ii) follows from (5.3). Conversely, assume (i -- iv) and let us prove equality of the above germs. Setting δ = β1γ, we have by (iv) that η := η1 = η2 ∈ Z(δ), so it suffices to verify (8.3.1), which the reader could do without any difficulty. (cid:3) The above result then says that the typical situation in which an equality of germs takes place is (cid:2)α, g, β; βγξ(cid:3) =(cid:2)αgγ, ϕ(g, γ), βγ; βγξ(cid:3). Our next two results are designed to offer convenient representatives of germs. graphs, groups and self-similarity 25 8.4. Proposition. Given any germ u, there exists an integer n0, such that for every n ≥ n0, (i) there is a representation of u of the form u =(cid:2)α1, g1, β1; β1ξ1(cid:3), with α1 = n. (ii) there is a representation of u of the form u =(cid:2)α2, g2, β2; β2ξ2(cid:3), with β2 = n. Proof. Write u =(cid:2)α, g, β; η(cid:3), and choose any n0 ≥ max{α, β}. Then, for every n ≥ n0 we may write η = βγξ, with γ ∈ E∗, ξ ∈ E∞, and such that γ = n − α (resp. γ = n − β). Therefore u =(cid:2)α, g, β; βγξ(cid:3) =(cid:2)αgγ, ϕ(g, γ), βγ; βγξ(cid:3), and we have αgγ = α + gγ = α + γ = n (resp. βγ = β + γ = n). (cid:3) 8.5. Corollary. Given u1 and u2 in Gtight(SG,E), such that (u1, u2) ∈ Gtight(SG,E)(2) (that is, such that the multiplication u1u2 is allowed or, equivalently, such that d(u1) = r(u2)), there are representations of u1 and u2 of the form u1 =(cid:2)α1, g1, α2; α2g2ξ(cid:3), and u2 =(cid:2)α2, g2, β; βξ(cid:3), and in this case Proof. Using (8.4), write u1u2 =(cid:2)α1, g1g2, β; βξ(cid:3). ui =(cid:2)αi, gi, βi; βiξi(cid:3), with β1 = α2. By virtue of (u1, u2) lying in Gtight(SG,E)(2), we have that β1ξ1 = (α2, g2, β2)(β2ξ2) = α2g2ξ2, so in fact β1 = α2, and ξ1 = g2ξ2. Then u1 =(cid:2)α1, g1, β1; β1ξ1(cid:3) =(cid:2)α1, g1, α2; α2g2ξ2(cid:3), and it suffices to put ξ = ξ2, and β = β2. With respect to the last assertion we have that u1u2 = [s; βξ], where s is the element of SG,E given by concluding the proof. s = (α1, g1, α2)(α2, g2, β) (4.2) = (α1, g1g2, β), (cid:3) Having extended the action of G to the set of infinite words in (8.1), one may ask whether it is possible to do the same for the cocycle ϕ. The following is an attempt at this which however produces a map taking values in the infinite product G∞, rather than in G. 26 r. exel and e. pardo 8.6. Definition. We will denote by Φ, the map defined by the rule for g ∈ G, ξ ∈ E∞, and n ≥ 1. Φ : G × E∞ → G∞ Φ(g, ξ)n = ϕ(g, ξn−1), We wish to view Φ as some sort of cocycle but, unfortunately, property (2.5.x) does not hold quite as stated. On the fortunate side, a suitable modification of Φ, involving the left shift endomorphism λ of G∞, works nicely: 8.7. Proposition. Let α be a finite word in E∗ and let ξ be an infinite word in E∞ such that d(α) = r(ξ). Then, for every g in G, one has that Proof. For all n ≥ 1, we have Φ(cid:0)ϕ(g, α), ξ(cid:1) = λα(cid:0)Φ(cid:0)g, αξ)(cid:1) Φ(cid:0)ϕ(g, α), ξ(cid:1)n = ϕ(cid:0)ϕ(g, α), ξn−1(cid:1) = ϕ(cid:0)g, α(ξn−1)(cid:1) = = ϕ(cid:0)g, (αξ)n−1+α(cid:1) = λα(cid:0)Φ(g, αξ)(cid:1)n. Another reason to think of Φ as a cocycle is the following version of the cocycle identity (2.5.b): 8.8. Proposition. For every ξ ∈ E∞, and every g, h ∈ G, we have that (cid:3) Φ(gh, ξ) = Φ(cid:0)g, hξ(cid:1)Φ(h, ξ). Proof. We have for all n ∈ , that Φ(gh, ξ)n = ϕ(gh, ξn−1) (2.5.b) = ϕ(cid:0)g, h(ξn−1)(cid:1)ϕ(h, ξn−1) = ϕ(cid:0)g, (hξ)n−1(cid:1)ϕ(h, ξn−1) = Φ(cid:0)g, hξ(cid:1)nΦ(h, ξ)n. (8.1) = (cid:3) The following elementary fact might perhaps justify the choice of "n − 1" in the definition of Φ. 8.9. Proposition. Given g ∈ G, and ξ ∈ E∞, one has that (gξ)n = Φ(g, ξ)n ξn. Proof. By (8.1) we have that (gξ)n = g(ξn), so the nth letter of gξ is also the nth letter of g(ξn). In addition we have that g(ξn) = g(ξn−1ξn) (2.5.ix) = g(ξn−1)ϕ(g, ξn−1)ξn, so We now wish to define a homomorphism (also called a one-cocycle) from Gtight(SG,E) (gξ)n = ϕ(g, ξn−1)ξn = Φ(g, ξ)n ξn. (cid:3) to the lag group G ⋊ρ Z, by means of the rule (cid:2)α, g, β; βξ(cid:3) 7→(cid:16)ρα(cid:0)Φ(g, ξ)(cid:1), α − β(cid:17). As it is often the case for maps defined on groupoid of germs, the above tentative definition uses a representative of the germ, so some work is necessary to prove that the definition does not depend on the choice of the representative. The technical part of this task is the content of our next result. graphs, groups and self-similarity 27 8.10. Lemma. Suppose that (G, E, ϕ) is residually free. For each i = 1, 2, let us be given (αi, gi, βi) in SG,E, as well as ηi = βiξi ∈ Z(βi). If then modulo G(∞). (cid:2)α1, g1, β1; η1(cid:3) =(cid:2)α2, g2, β2; η2(cid:3), ρα1(cid:0)Φ(g1, ξ1)(cid:1) ≡ ρα2(cid:0)Φ(g2, ξ2)(cid:1) Proof. Assuming without loss of generality that β1 ≤ β2, we may use (8.3) to write α2 = α1g1γ, g2 = ϕ(g1, γ), β2 = β1γ, and η1 = η2 = β1γξ, for suitable γ ∈ E∗ and ξ ∈ E∞. Then necessarily ξ1 = γξ, and ξ2 = ξ, and ρα2(cid:0)Φ(g2, ξ2)(cid:1) = ρα1+γ(cid:16)Φ(cid:0)ϕ(g1, γ1), ξ(cid:1)(cid:17) (8.7) ≡ ρα1(cid:0)Φ(cid:0)g1, ξ1)(cid:1). = ρα1ργλγ(cid:0)Φ(cid:0)g1, γξ)(cid:1) (7.3) = (cid:3) ◮ Due to our reliance on (8.3) and (8.10), from now on and until the end of this section we will assume, in addition to (2.3), that (G, E, ϕ) is residually free. If g is in G∞, we will denote by g its class in the quotient group G. Likewise we will denote by Φ the composition of Φ with the quotient map Φ G × E∞ −→ G∞ −→ G ................................................................................................................................................................................................................................... .......... Φ from G∞ to G. It then follows from the above Lemma that the correspondence (cid:2)α, g, β; βξ(cid:3) ∈ Gtight(SG,E) 7→ ρα(cid:0) Φ(g, ξ)(cid:1) ∈ G is a well defined map. This is an important part of the one-cocycle we are about to introduce. 8.11. Proposition. The correspondence ℓ :(cid:2)α, g, β; βξ(cid:3) 7→(cid:16)ρα(cid:0) Φ(g, ξ)(cid:1), α − β(cid:17) gives a well defined map ℓ : Gtight(SG,E) → G ⋊ρ Z, which is moreover a one-cocycle. From now on ℓ will be called the lag function. 28 r. exel and e. pardo Proof. By the discussion above we have that the first coordinate of the above pair is well defined. On the other hand, in the context of (8.3) one easily sees that α1 − β1 = α2 − β2, so the second coordinate is also well defined. In order to show that ℓ is multiplicative, pick (u1, u2) ∈ Gtight(SG,E)(2). We may then use (8.5) to write u1 =(cid:2)α1, g1, α2; α2g2ξ(cid:3), and u2 =(cid:2)α2, g2, β; βξ(cid:3). So ℓ(u1)ℓ(u2) =(cid:16)ρα1(cid:0)Φ(g1, g2ξ)(cid:1), α1 − α2(cid:17)(cid:16)ρα2(cid:0)Φ(g2, ξ)(cid:1), α2 − β(cid:17) = α1 − α2 + α2 − β(cid:17) = =(cid:16)ρα1(cid:0)Φ(g1, g2ξ)(cid:1) ρα1(cid:0)Φ(g2, ξ)(cid:1), = (cid:16)ρα1(cid:0)Φ(g1g2, ξ)(cid:1), α1 − β(cid:17) = =(cid:16)ρα1(cid:16)Φ(g1, g2ξ)Φ(g2, ξ)(cid:17), α1 − β(cid:17) (8.8) = ℓ(cid:0)(cid:2)α1, g1g2, β; βξ(cid:3)(cid:1) (8.5) = ℓ(u1u2). (cid:3) The main relevance of this one-cocycle is that it essentially describes the elements of Gtight(SG,E), as we would like to show now. 8.12. Proposition. Given u1, u2 ∈ Gtight(SG,E), one has that d(u1) = d(u2) ∧ r(u1) = r(u2) ∧ ℓ(u1) = ℓ(u2) ⇒ u1 = u2. Proof. Using (8.4), write ui =(cid:2)αi, gi, βi; βiξi(cid:3), for i = 1, 2, with β1 = β2. Since β1ξ1 = d(u1) = d(u2) = β2ξ2, we conclude that β1 = β2, and ξ1 = ξ2 =: ξ. By focusing on the second coordinate of ℓ(ui), we see that α1 − β1 = α2 − β2, and hence α1 = α2. Moreover, since α1g1ξ = α1g1ξ1 = r(u1) = r(u2) = α2g2ξ2 = α2g2ξ, we see that α1 = α2, and g1ξ = g2ξ. (8.12.1) The fact that ℓ(u1) = ℓ(u2) also implies that ρα1(cid:0) Φ(g1, ξ)(cid:1) = ρα2(cid:0) Φ(g2, ξ)(cid:1), and since α1 = α2, we conclude that Φ(g1, ξ) = Φ(g2, ξ), and hence that there exists an integer n0 such that By (8.12.1) we also have that g1(ξn) = g2(ξn), so (5.3) gives g1 = g2, whence u1 = u2. (cid:3) ϕ(g1, ξn) = ϕ(g2, ξn), ∀ n ≥ n0. graphs, groups and self-similarity 29 As a consequence of the above result we see that the map F : Gtight(SG,E) → E∞ × ( G ⋊ρ Z) × E∞ defined by the rule is one-to-one. F (u) =(cid:0)r(u), ℓ(u), d(u)(cid:1), (8.13) Observe that the co-domain of F has a natural groupoid structure, being the cartesian product of the lag group G ⋊ρ Z by the graph of the transitive equivalence relation on E∞. Putting together (8.11) and (8.12) we may now easily prove: 8.14. Corollary. F is a groupoid homomorphism (functor), hence establishing an iso- morphism from Gtight(SG,E) to the range of F . The range of F is then the concrete model of Gtight(SG,E) we are after. But, before giving a detailed description of it, let us make a remark concerning notation: since the co-domain of F is a mixture of cartesian and semi-direct products, the standard notation for its elements would be something like (cid:0)η, (u, p), ζ(cid:1), for η, ζ ∈ E∞, u ∈ G, and p ∈ Z. As part of our effort to avoid heavy notation we will instead denote such an element by (cid:0)η; u, p; ζ(cid:1). 8.15. Proposition. The range of F is precisely the subset of E∞ × ( G ⋊ρ Z) × E∞, formed by the elements (η; g, p − q; ζ), where η, ζ ∈ E∞, g ∈ G∞, and p, q ∈ , are such that, for all n ≥ 1, (i) gn+p+1 = ϕ(gn+p, ζn+q), (ii) ηn+p = gn+pζn+q. (8.15.1) Proof. Pick a general element (cid:2)α, g, β; βξ(cid:3) ∈ Gtight(SG,E) and, recalling that F ((cid:2)α, g, β; βξ(cid:3)) =(cid:16)αgξ; ρα(cid:0) Φ(g, ξ)(cid:1), α − β; βξ(cid:17), let η = αgξ, g = ρα(cid:0)Φ(g, ξ)(cid:1), p = α, q = β, and ζ = βξ, so that the element depicted in (8.15.1) becomes (η; g, p − q; ζ), and we must now verify (i) and (ii). For all n ≥ 1, one has that gn+α = Φ(g, ξ)n = ϕ(g, ξn−1), so ηn+p = (αgξ)n+α = (gξ)n (8.9) = ϕ(g, ξn−1)ξn = gn+α(βξ)n+β = gn+pζn+q, proving (ii). Also, gn+p+1 = gn+α+1 = ϕ(cid:0)g, ξn(cid:1) = ϕ(g, ξn−1ξn) = ϕ(cid:0)ϕ(g, ξn−1), ξn(cid:1) = = ϕ(cid:0)gn+α, (βξ)n+β(cid:1) = ϕ(cid:0)gn+p, ζn+q(cid:1), 30 r. exel and e. pardo proving (i) and hence showing that the range of F is a subset of the set described in the statement. Conversely, pick η, ζ ∈ E∞, g ∈ G∞, and p, q ∈ satisfying (i) and (ii), and let us show that the element (η; g, p − q; ζ) lies in the range of F . Let g = gp+1, α = ηp, and β = ζq, so ζ = βξ for a unique ξ ∈ E∞. We then claim that (cid:2)α, g, β; βξ(cid:3) lies in Gtight(SG,E). In order to see this notice that gd(β) = gd(ζq) = gr(ζq+1) = r(gp+1ζq+1) (ii) = r(ηp+1) = d(ηp) = d(α), so (α, g, β) ∈ SG,E, and therefore (cid:2)α, g, β; βξ(cid:3) is indeed a member of Gtight(SG,E). The proof will then be concluded once we show that F ((cid:2)α, g, β; βξ(cid:3)) = (η; g, p − q; ζ), which in turn is equivalent to showing that (a) αgξ = η, (b) ρα(cid:0) Φ(g, ξ)(cid:1) = g, ( c ) α − β = p − q, (d) βξ = ζ. Before proving these points we will show that ϕ(gp+1, ξn) = gn+p+1, ∀ n ≥ 0. (†) This is obvious for n = 0. Assuming that n ≥ 1 and using induction, we have ϕ(gp+1, ξn) = ϕ(gp+1, ξn−1ξn) = ϕ(cid:0)ϕ(gp+1, ξn−1), ξn(cid:1) = = ϕ(gn+p, ζn+q) (i) = gn+p+1, verifying (†). Addressing (a) we have to prove that (αgξ)k = ηk, for all k ≥ 1, but given that α is defined to be ηp, this is trivially true for k ≤ p. On the other hand, for k = n + p, with n ≥ 1, we have (αgξ)k = (αgξ)n+p = (gξ)n (8.9) = ϕ(g, ξn−1)ξn = = ϕ(gp+1, ξn−1)ξn (†) = gn+pζn+q (ii) = ηn+p = ηk, proving (a). Focusing on (b) we have for all n ≥ 1 that ρα(cid:0)Φ(g, ξ)(cid:1)p+n = Φ(g, ξ)n = ϕ(gp+1, ξn−1) (†) = gn+p, proving that ρα(cid:0)Φ(g, ξ)(cid:1) ≡ g, modulo G(∞), hence taking care of (b). The last two points, namely (c) and (d) are immediate and so the proof is concluded. (cid:3) graphs, groups and self-similarity 31 As an immediate consequence we get a very precise description of the algebraic struc- ture of Gtight(SG,E): 8.16. Theorem. Suppose that (G, E, ϕ) satisfies the conditions of (2.3) and is moreover residually free. Then Gtight(SG,E) is isomorphic to the sub-groupoid of E∞×( G⋊ρ Z)×E∞ given by GG,E =   (η; g, p − q; ζ) ∈ E∞ × ( G ⋊ρ Z) × E∞ : g ∈ G∞, p, q ∈ , gn+p+1 = ϕ(gn+p, ζn+q), ηn+p = gn+pζn+q, for all n ≥ 1 .   Recall from [8] that the C-algebra of every graph is a groupoid C-algebra for a certain groupoid constructed from the graph, and informally called the groupoid for the tail equivalence with lag. Viewed through the above perspective, our groupoid may also deserve such a denom- ination, except that the lag is not just an integer as in [8], but an element of the lag group G ⋊ρ Z precisely described by the lag function ℓ introduced in (8.11). 9. The topology of Gtight(SG,E). It is now time we look at the topological aspects of Gtight(SG,E). In fact what we will do is simply transfer the topology of Gtight(SG,E) over to GG,E via F . Not surprisingly F will turn out to be an isomorphism of topological groupoids. Recall from [2: Proposition 4.14] that, if S is an inverse semigroup acting on a locally compact Hausdorff topological space X, then the corresponding groupoid of germs, say G, is topologized by means of the basis consisting of sets of the form where s ∈ S, and U is an open subset of X, contained in the domain of the partial homeomorphism attached to s by the given action. Each Θ(s, U ) is in turn defined by Θ(s, U ), Θ(s, U ) =n[s, x] ∈ G : x ∈ Uo. See [2: 4.12] for more details. If we restrict the choice of the U 's above to a predefined basis of open sets of X, e.g. the collection of all cylinders in E∞ in the present case, we evidently get the same topology on the groupoid of germs. Therefore, referring to the model of Gtight(SG,E) presented in (8.2), we see that a basis for its topology consists of the sets of the form Θ(α, g, β; γ) :=n(cid:2)α, g, β; ξ(cid:3) ∈ Gtight(SG,E) : ξ ∈ Z(γ)o, (9.1) where (α, g, β) ∈ SG,E, and γ ∈ E∗. We may clearly suppose that γ ≥ β and, since Θ(α, g, β; γ) = ∅, unless β is a prefix of γ, we may also assume that γ = βε, for some ε ∈ E∗. 32 r. exel and e. pardo In this case, given any (cid:2)α, g, β; ξ(cid:3) ∈ Θ(α, g, β; γ), notice that ξ ∈ Z(γ), and (α, g, β)(γ, 1, γ) =(cid:0)αgε, ϕ(g, ε), γ(cid:1), from where one concludes that (cid:2)α, g, β; ξ(cid:3) =(cid:2)αgε, ϕ(g, ε), γ; ξ(cid:3), for all ξ ∈ Z(γ), and hence also that Θ(α, g, β; γ) = Θ(cid:0)αgε, ϕ(g, ε), γ; γ(cid:1). This shows that any set of the form (9.1) coincides with another such set for which β = γ. We may therefore do away with this repetition and redefine Θ(α, g, β) :=n(cid:2)α, g, β; ξ(cid:3) ∈ Gtight(SG,E) : ξ ∈ Z(β)o. (9.2) We have therefore shown: 9.3. Proposition. The collection of all sets of the form Θ(α, g, β), where (α, g, β) range in SG,E, is a basis for the topology of Gtight(SG,E). We may now give a precise description of the topology of Gtight(SG,E), once it is viewed from the alternative point of view of (8.16): 9.4. Proposition. For each (α, g, β) in SG,E, the image of Θ(α, g, β) under F coincides with the set Ω(α, g, β) :=   (η; g, k; ζ) ∈ GG,E : η ∈ Z(α), g ∈ G∞, k = α − β, ζ ∈ Z(β), g1+α = g, gn+α+1 = ϕ(gn+α, ζn+β), ηn+α = gn+αζn+β, for all n ≥ 1 ,   and hence the collection of all such sets form the basis for a topology on GG,E, with respect to which the latter is isomorphic to Gtight(SG,E) as topological groupoids. Proof. Left for the reader. (cid:3) We may now summarize the main results obtained so far: 9.5. Theorem. Suppose that (G, E, ϕ) satisfies the conditions of (2.3) and is moreover residually free. Then OG,E is -isomorphic to the C-algebra of the groupoid GG,E de- scribed in (8.16), once the latter is equipped with the topology generated by the basis of open sets Ω(α, g, β) described above, for all (α, g, β) in SG,E. graphs, groups and self-similarity 33 10. OG,E as a Cuntz-Pimsner algebra. Inspired by Nekrashevych's paper [10], we will now give a description of OG,E as a Cuntz- Pimsner algebra [14]. With this we will also be able to prove that OG,E is nuclear and that Gtight(SG,E) is amenable when G is an amenable group. As before, we will work under the conditions of (2.3). We begin by introducing the algebra of coefficients over which the relevant Hilbert bimodule, also known as a correspondence, will later be constructed. Since the action of G on E preserves length (2.4.iv), we see that the set of vertexes of E is G-invariant, so we get an action of G on E0 by restriction. By dualization G acts on the algebra C(E0) of complex valued functions4 on E0. We may therefore form the crossed-product C-algebra A = C(E0) ⋊ G. Since C(E0) is a unital algebra, there is a canonical unitary representation of G in the crossed product, which we will denote by {vg}g∈G. On the other hand, C(E0) is also canonically isomorphic to a subalgebra of A and we will therefore identify these two algebras without further warnings. For each x in E0, we will denote the characteristic function of the singleton {x} by qx, so that {qx : x ∈ E0} is the canonical basis of C(E0), and thus A coincides with the closed linear span of the set (cid:8)qxvg : x ∈ E0, g ∈ G(cid:9). (10.1) For later reference, notice that the covariance condition in the crossed product reads vgqx = qgxvg, ∀ x ∈ E0, ∀ g ∈ G. (10.2) Our next step is to construct a correspondence over A. In preparation for this we denote by Ae the right ideal of A generated by qd(e), for each e ∈ E1. In technical terms Ae = qd(e)A. With the obvious right A-module structure, and the inner product defined by hy, zi = y∗z, ∀ y, z ∈ Ae, one has that Ae is a right Hilbert A-module. Notice that this in not necessarily a full Hilbert module since hAe, Aei is the two-sided ideal of A generated by qd(e), which might be a proper ideal in some cases. As already seen in (10.1), A is spanned by the elements of the form qxvg. Therefore Ae is spanned by the elements of the form qd(e)qxvg, but, since the q's are mutually orthogonal, this is either zero or equal to qd(e)vg. Therefore we see that Ae = span{qd(e)vg : g ∈ G}. 4 Notice that, since E0 is a finite set, C(E0) is nothing but C E0. 34 r. exel and e. pardo Introducing the right Hilbert A-module which will later be given the structure of a correspondence over A, we define M = Me∈E1 Ae. Observe that if x is a vertex which is the source of many edges, say then d−1(x) = {e1, e2, . . . , en}, Aei = qd(ei)A = qxA, for all i, so that qxA appears many times as a direct summand of M . However these copies of qxA should be suitably distinguished, according to which edge ei is being considered. On the other hand, notice that if d−1(x) = ∅, then qxA does not appear among the summands of M , at all. Addressing the fullness of M , observe that hM, M i = Xx∈E0 d−1(x)6=∅ AqxA, so, when E has no sinks, that is, when d−1(x) is nonempty for every x, one has that M is full. Given e ∈ E1, the element qd(e), when viewed as an element of Ae ⊆ M , will play a very special role in what follows, so we will give it a special notation, namely (10.3) There is a small risk of confusion here in the sense that, if e1, e2 ∈ E1 are such that te := qd(e). x := d(e1) = d(e2), then (10.3) assigns qx to both te1 and te2. However the coordinate in which qx appears in tei is determined by the corresponding ei, so if e1 6= e2, then te1 6= te2 . In order to completely dispel any confusion, here is the technical definition: where We should notice that te = (mf )f ∈E1 , mf =(cid:26) qd(e), 0, if f = e, otherwise. teqd(e) = te, and that any element of M may be written uniquely as Xe∈E1 teye, (10.4) (10.5) where each ye ∈ Ae. As the next step in constructing a correspondence over A, we would now like to define a certain -homomorphism from A to the algebra L(M ) of adjointable linear operators on M . Since A is a crossed product algebra, this will be accomplished once we produce a covariant representation (ψ, V ) of the C-dynamical system (cid:0)C(E0), G(cid:1). We begin with the group representation V . graphs, groups and self-similarity 35 10.6. Definition. For each g ∈ G, let Vg be the linear operator on M given by Vg(cid:16) Xe∈E1 teye(cid:17) = Xe∈E1 tgevϕ(g,e)ye, whenever ye ∈ Ae, for each e in E1. By the uniqueness in (10.5), it is clear that Vg is well defined. 10.7. Proposition. Each Vg is a unitary operator in L(M ). Moreover, the correspon- dence g 7→ Vg is a unitary representation of G. Proof. Let g ∈ G. We begin by claiming that the expression defining Vg above holds true whenever the ye are in A, and not necessarily restricted to Ae. Since Vg is clearly additive, we only need to check that Observing that te = teqd(e), we have Vg(tey) = tgevϕ(g,e)y, ∀ y ∈ A. Vg(tey) = Vg(teqd(e)y) = tgevϕ(g,e)qd(e)y = = tgeqd(ϕ(g,e)e)vϕ(g,e)y (2.5.vii) = tgeqd(ge)vϕ(g,e)y = tgevϕ(g,e)y, proving the claim. One therefore concludes that Vg is right-A-linear. We next claim that, for all e, f ∈ E1, one has We have hVg(te), tf i = hte, Vg−1(tf )i. (10.7.1) hVg(te), tf i = htgevϕ(g,e), tf i = v∗ ϕ(g,e)htge, tf i = [ge=f] v−1 ϕ(g,e)qd(ge) = = [ge=f] qd(ϕ(g,e)−1ge)v−1 ϕ(g,e) (2.5.vii) = [ge=f] qd(e)v−1 ϕ(g,e) = (⋆). Starting from the right-hand-side of (10.7.1), we have hte, Vg−1(tf )i = hte, tg−1f vϕ(g−1,f )i = [e=g−1f] qd(e)vϕ(g−1,f ) = = [ge=f] qd(e)vϕ(g,g−1f )−1 = [ge=f] qd(e)v−1 ϕ(g,e), which agrees with (⋆) above, and hence proves claim (10.7.1). If y, z ∈ A, we then have that hVg(tey), tf zi = y∗hVg(te), tf iz = y∗hte, Vg−1(tf )iz = htey, Vg−1(tf z)i, from where one sees that hVg(ξ), ηi = hξ, Vg−1(η)i, for all ξ, η ∈ M , hence proving that Vg is an adjointable operator with V ∗ g = Vg−1. Let us next prove that VgVh = Vgh, ∀ g, h ∈ G. By A-linearity it is enough to prove that these operators coincide on the set formed by the te's, which is a generating set for M . We thus compute Vg(cid:0)Vh(te)(cid:1) = Vg(cid:0)thevϕ(h,e)(cid:1) = Vg(the)vϕ(h,e) = = tghevϕ(g,he)vϕ(h,e) = tghevϕ(gh,e) = Vgh(te). Since it is evident that V1 is the identity operator on M we obtain, as a consequence, (cid:3) g , so each Vg is unitary and the proof is concluded. that V −1 g = Vg−1 = V ∗ 36 r. exel and e. pardo In order to complete our covariant pair we must now construct a -homomorphism from C(E0) to L(M ). With this in mind we give the following: 10.8. Definition. For every x in E0, let Mx = Me∈r−1(x) Ae, which we view as a complemented sub-module of M . orthogonal projection from M to Mx, so that In addition, we let Qx be the Qx(tey) = [r(e)=x] tey, ∀ e ∈ E1, ∀ y ∈ A. (10.8.1) Observe that the Qx are pairwise orthogonal projections and that Px∈E0 Qx = 1. 10.9. Definition. Let ψ : C(E0) → L(M ) be the unique unital -homomorphism such that ψ(qx) = Qx, ∀ x ∈ E0. From our working hypothesis that E has no sources, we see that for every x in E0, there is some e ∈ E1 such that r(e) = x. So whence Qx 6= 0. Consequently ψ is injective. Qx(te) = te, 10.10. Proposition. The pair (ψ, V ) is a covariant representation of the C-dynamical system (cid:0)C(E0), G(cid:1) in L(M ). Proof. All we must do is check the covariance condition Vgψ(y) = ψ(cid:0)σg(y)(cid:1)Vg, ∀ g ∈ G, ∀ y ∈ C(E0), where σ is the name we temporarily give to the action of G on C(E0). Since C(E0) is spanned by the qx, it suffices to consider y = qx, in which case the above identity becomes VgQx = QgxVg. (10.10.1) Furthermore M is generated, as an A-module, by the te, for e ∈ E1, so we only need to verify this on the te. We have while Vg(cid:0)Qx(te)(cid:1) = [r(e)=x] Vg(te) = [r(e)=x] tgevϕ(g,e), Qgx(cid:0)Vg(te)(cid:1) = Qgx(cid:0)tgevϕ(g,e)(cid:1) = [r(ge)=gx] tgevϕ(g,e), verifying (10.10.1) and concluding the proof. (cid:3) graphs, groups and self-similarity 37 It follows from [13: Proposition 7.6.4 and Theorem 7.6.6] that there exists a -homo- morphism such that and Ψ : C(E0) ⋊ G → L(M ), Ψ(qx) = Qx, ∀ x ∈ E0, Ψ(vg) = Vg, ∀ g ∈ G. Equipped with the left-A-module structure provided by Ψ, we then have that M is a correspondence over A. For later reference we record here a few useful calculations involving the left-module structure of M . 10.11. Proposition. Let g ∈ G, e ∈ E1, and x ∈ E0. Then (i) vgte = tgevϕ(g,e), (ii) qxvgte = [r(ge)=x] tgevϕ(g,e). Proof. We have proving (a). Also vgte = Ψ(vg)te = Vg(te) = tgevϕ(g,e), qxvgte = Ψ(qx)(vgte) = Qx(tgevϕ(g,e)) = [r(ge)=x] tgevϕ(g,e). (cid:3) It is our next goal to prove that OG,E is naturally isomorphic to the Cuntz-Pimsner algebra associated to the correspondence M , which we denote by OM . As a first step, we identify a certain Cuntz-Krieger E-family. 10.12. Proposition. The following relations hold within OM . (a) For every x ∈ E0, one has that Pe∈r−1(x) tet∗ (b) Pe∈E1 tet∗ e = 1. e = qx. ( c ) The set {qx : x ∈ E0} ∪ {te : e ∈ E1} is a Cuntz-Krieger E-family. Proof. We first claim that, for every x ∈ E0, and every m ∈ M , one has that Xe∈r−1(x) tet∗ em = qxm. To prove it, it is enough to consider the case in which m = tf , for f ∈ E1, since these generate M . In this case we have Xe∈r−1(x) tet∗ etf = [r(f )=x] tf t∗ f tf = [r(f )=x] tf (10.8.1) = Qx(tf ) = qxtf , 38 r. exel and e. pardo proving the claim. This says that the pair(cid:0)qx,Pe∈r−1(x) tet∗ the terminology of [14], that the generalized compact operator e(cid:1) is a redundancy or, adopting Xe∈r−1(x) Ωte,te is mapped to Ψ(qx) via Ψ(1). Therefore qx = Xe∈r−1(x) tet∗ e, in OM , proving (a). Point (b) then follows from the fact that Px∈E0 qx = 1. Focusing now on (c), it is evident that {qx : x ∈ E0} is a family of mutually orthogonal projections. Moreover, for each e ∈ E1, we have t∗ ete = hte, tei = qd(e), proving (3.1.i) and also that te is a partial isometry. Property (3.1.ii) also holds in view of (a), so the proof is concluded. (cid:3) 10.13. Proposition. There exists a unique surjective -homomorphism Λ : OG,E → OM such that Λ(px) = qx, Λ(se) = te, and Λ(ug) = vg. Proof. By the universal property of OG,E, in order to prove the existence of Λ it is enough to check that the qx, te, and vg satisfy the conditions of (3.2). Condition (3.2.a) has already been proved above while (3.2.b) is evidently true since v is a representation of G in C(E0) ⋊ G ⊆ OM . Condition (3.2.c) is precisely (10.11.i), while (3.2.d) was taken care of in (10.2). Since A is spanned by the qx and the vg by (10.1), and since M is generated over A by the te, we see that OM is spanned by the set {qx, te, vg : x ∈ E0, e ∈ E1, g ∈ G}, so Λ is surjective. (cid:3) Let us now prove that Λ is invertible by providing an inverse to it. Since A is the crossed product C-algebra C(E0) ⋊ G, one sees that (3.2.a&d) guarantees the existence of a -homomorphism θA : A → OG,E, sending the qx to the px, and the vg to the ug. For each e in E1, consider the linear mapping θM : M → OG,E, graphs, groups and self-similarity 39 given, for every m = (me)e∈E1 ∈ M , by θM (m) = Xe∈E1 seθA(me) ∈ OG,E. Notice that θM (te) = se, for all e ∈ E1, because θM (te) = seθA(qd(e)) = sepd(e) = se. 10.14. Lemma. The pair (θA, θM ) is a representation of the correspondence M in the sense of [14: Theorem 3.4], meaning that for all y ∈ A and all ξ, ξ′ ∈ M , (i) θM (ξ)θA(y) = θM (ξy), (ii) θA(y)θM (ξ) = θM (yξ), (iii) θM (ξ)∗θM (ξ′) = θA(hξ, ξ′i). Proof. Considering the various spanning sets at our disposal, we may assume that y = qxvg, that ξ = tez, and ξ′ = te′ z′, with x ∈ E0, g ∈ G, e, e′ ∈ E1, z ∈ qd(e)A, and z′ ∈ qd(e′)A. Then (10.1) θM (ξ)θA(y) = θM (tez)θA(y) = seθA(z)θA(y) = seθA(zy) = θM (tezy) = θM (ξy), proving (i). As for (ii), we have θA(y)θM (ξ) = θA(qxvg)θM (tez) = pxugseθA(z) = pxsgeuϕ(g,e)θA(z) = = [r(ge)=x] sgeθA(vϕ(g,e)z) = [r(ge)=x] θM (tgevϕ(g,e)z(cid:1) (10.11.ii) proving (ii). Focusing now on (iii), we have = θM (qxvgtez) = θM (yξ), θM (ξ)∗θM (ξ′) = (seθA(z)(cid:1)∗ se′θA(z′) = [e=e′] θA(z)∗pd(e)θA(z′) = = [e=e′] θA(z∗qd(e)z′) = θA(hξ, ξ′i). (cid:3) It is well known [14: Theorem 3.4] that the Toeplitz algebra for the correspondence M , usually denoted TM , is universal for representations of M , so there exists a -homo- morphism coinciding with θA on A and with θM on M . Θ0 : TM → OG,E, 10.15. Theorem. The map Θ0, defined above, factors through OM , providing a - isomorphism Θ : OM → OG,E, such that Θ(qx) = px, Θ(te) = se, and Θ(vg) = ug, for all x ∈ E0, e ∈ E1, and g ∈ G. 40 r. exel and e. pardo Proof. The factorization property follows immediately from (10.12.b) and an easy modifi- cation of [4: Proposition 7.1] to Cuntz-Pimsner algebras. In order to prove that Θ is an isomorphism, observe that Θ ◦ Λ coincides with the identity map on the generators of OG,E, by (10.13), and hence Θ ◦ Λ = id. The result then follows from the fact that Λ is surjective. (cid:3) 10.16. Corollary. If G is amenable then OG,E is nuclear. Proof. The amenability of G ensures that C(E0) ⋊ G is nuclear. The result then follows from (10.15), the fact that Toeplitz-Pimsner algebras over nuclear coefficient algebras is nuclear [1: Theorem 4.6.25], and so are quotients of nuclear algebras [1: Theorem 9.4.4]. (cid:3) 10.17. Remark. Since E0 is finite, the nuclearity of C(E0) ⋊ G is equivalent to the amenability of G. However, if the present construction is generalized for infinite graphs, one could produce examples of non amenable groups acting amenably on E0, in which case C(E0) ⋊ G would be amenable. The proof of (10.16) could then be adapted to prove that OG,E is nuclear. 10.18. Corollary. If G is amenable and (G, E, ϕ) is residually free then Gtight(SG,E) and its sibling GG,E are amenable groupoids. Proof. Follows from (10.16), (9.5), and [1: Theorem 5.6.18]. (cid:3) Nekrashevych has proven in [11: Theorem 5.6], that a certain groupoid of germs, denoted DG, constructed in the context of self-similar groups, is amenable under the hy- pothesis that the group is contracting and self-replicating. Even though there are numerous differences between DG and GG,E, including a different notion of germs and Nekrashevych's requirement that group actions be faithful, we believe it should be interesting to try to generalize Nekrashevych's result to our context. References [1] N. P. Brown and N. Ozawa, "C-algebras and finite-dimensional approximations", Graduate Studies in Mathematics, 88, American Mathematical Society, 2008. [2] R. Exel, "Inverse semigroups and combinatorial C-algebras", Bull. Braz. Math. Soc. 39 (2008), no. 2, 191 -- 313. [3] R. Exel and E. Pardo, "Representing Kirchberg algebras as inverse semigroup crossed products", preprint, 2013. [4] R. Exel and A. Vershik, "C-algebras of irreversible dynamical systems", Canadian Mathematical Journal, 58 (2006), 39 -- 63. [5] R. I. Grigorchuk, "On Burnsides problem on periodic groups", Funct. Anal. Appl., 14 (1980), 41 -- 43. [6] N. D. Gupta and S. N. Sidki, "On the Burnside problem for periodic groups", Math. Z., 182 (1983), 385 -- 388. [7] T. Katsura, "A construction of actions on Kirchberg algebras which induce given actions on their K-groups", J. reine angew. Math., 617 (2008), 27 -- 65. [8] A. Kumjian, D. Pask, I. Raeburn and J. Renault, "Graphs, groupoids, and Cuntz-Krieger algebras", J. Funct. Anal., 144 (1997), 505 -- 541. [9] M. V. Lawson, "Inverse semigroups, the theory of partial symmetries", World Scientific, 1998. [10] V. Nekrashevych, "Cuntz-Pimsner algebras of group actions", J. Operator Theory, 52 (2004), 223 -- 249. [11] V. Nekrashevych, "C-algebras and self-similar groups", J. reine angew. Math., 630 (2009), 59 -- 123. graphs, groups and self-similarity 41 [12] A. L. T. Paterson, "Groupoids, inverse semigroups, and their operator algebras", Birkhauser, 1999. [13] G. K. Pedersen, "C-algebras and Their Automorphism Groups", Academic Press, 1979. [14] M. V. Pimsner, "A class of C*-algebras generalizing both Cuntz-Krieger algebras and crossed products by Z", Fields Inst. Commun., 12 (1997), 189 -- 212. [15] I. Raeburn, "Graph algebras", CBMS Regional Conference Series in Mathematics, 103 (2005), pp. vi+113. Departamento de Matem´atica; Universidade Federal de Santa Catarina; 88010-970 Florian´opolis SC; Brazil ([email protected]) Departamento de Matem´aticas, Facultad de Ciencias; Universidad de C´adiz, Campus de Puerto Real; 11510 Puerto Real (C´adiz); Spain ([email protected])
1906.05477	3	1906	2019-06-19T17:24:44	Polish groups of unitaries	[ "math.OA", "math.GR", "math.RT" ]	We study the question of which Polish groups can be realized as subgroups of the unitary group of a separable infinite-dimensional Hilbert space. We also show that for a separable unital C$^*$-algebra $A$, the identity component $\mathcal{U}_0(A)$ of its unitary group has property (OB) of Rosendal (hence it also has property (FH)) if and only if the algebra has finite exponential length (e.g. if it has real rank zero), while in many cases the unitary group $\mathcal{U}(A)$ does not have property (T). On the other hand, the $p$-unitary group $\mathcal{U}_p(M,\tau)$ where $M$ is a properly infinite semifinite von Neumann algbera with separable predual, does not have property (FH) for any $1\le p<\infty$. This in particular solves a problem left unanswered in the work of Pestov \cite{Pestov18}.	math.OA	math	Polish groups of unitaries Hiroshi Ando Yasumichi Matsuzawa June 20, 2019 Abstract We study the question of which Polish groups can be realized as subgroups of the unitary group of a separable infinite-dimensional Hilbert space. We also show that for a separable unital C∗-algebra A, the identity component U0(A) of its unitary group has property (OB) of Rosendal (hence it also has property (FH)) if and only if the algebra has finite exponential length (e.g. if it has real rank zero), while in many cases the unitary group U(A) does not have property (T). On the other hand, the p-unitary group Up(M, τ ) where M is a properly infinite semifinite von Neumann algbera with separable predual, does not have property (FH) for any 1 ≤ p < ∞. This in particular solves a problem left unanswered in the work of Pestov [Pe18]. Keywords. Polish groups, unitary representability, properties (T), (OB) and (FH). Mathematics Subject Classification (2010) 22A25, 43A65, 57S99. 1 Introduction and Main Results Unitary representation theory of topological groups has a long history and has applications to diverse fields including mathematical physics, ergodic theory and operator algebra theory. For locally compact groups, the existence of the Haar measure guarantees the existence of sufficiently many unitary representations which enables us to study fine structures of these groups. The class of Polish groups is a natural and rich family of topological groups which include all second countable locally compact groups and many others. It is of interest to study which Polish groups admit abundance of unitary representations and in particular, which groups arise as groups of unitaries on some Hilbert space. It is known that there exist exotic Polish groups, namely those which do not have nontrivial strongly continuous unitary representations at all. The first such example is given by Herer -- Christensen [HC75]. See also the work of Megrelishvili [Me08] and Carderi -- Thom [CT18] for more exotic groups with surprising properties. For abelian exotic groups, see also Banaszczyk's book [Ba]. If a Polish group in question does admit a realization as a group of unitaries, there are more structural questions one may ask. The question of amenability or the property (T) of Kazhdan [Ka67] are such examples (see §2.3 for the definition of these and related properties of topological groups). Here, a (not necessarily locally compact) topological group G is said to be amenable if there exists a left-invariant mean on the space LUCB(G) of bounded left uniformly continuous functions on G. This is equivalent to the standard definition of the amenability if G is locally compact. The situation becomes drastically different for non-locally compact groups. For example, it is well-known that a locally compact group which is amenable and has property (T) is necessarily compact. On the other hand, the unitary group U(ℓ2) equipped with the strong operator topology is amenable by de la Harpe's result [dlH73] (see also [dlH78]) and at the same time has property (T) by Bekka's result [Be03], although it is far from being compact. In fact, U(ℓ2) is known to be extremely amenable by Gromov -- Milman's work [GM83], which means that whenever it acts continuously on a compact Hausdorff space, there is a fixed point. This is another remarkable phenomenon which cannot occur in the locally compact setting. As another example, the equivalence of the property (FH) and property (T), which hold for σ-compact locally compact groups, fails for non-locally compact Polish groups (the implication (T)⇒(FH) is valid for any topological group by Delorme's work [De77], and the implication (FH)⇒(T) in the σ-compact locally compact setting is due to Guichardet [Gu72]). See Bekka -- de la Harpe -- Valette' book [BHV08, §2.12] for more details). A recent intensive work by Pestov [Pe18] shows that nevertheless, the amenability and the property (T) are in some sense still opposite properties to each other for SIN Polish groups. For example, if a norm-closed subgroup of U(ℓ2) is amenable and has property (T), then it is maximally almost periodic [Pe18, Theorem 1.4], meaning that finite-dimensional strongly continuous unitary representations separate points. On the other hand, Rosendal [Ro09, Ro13, Ro18] has clarified 1 that it is profitable to study large scale geometric structures of non-locally compact Polish groups. One of the central concepts in his study of the coarse geometry of topological groups is the notion of coarsely boundedness (it is called the property (OB) in Rosendal [Ro09]), which is related to the boundedness in the sense of Hejcman [He58]. Here, a Polish group has property (OB) if whenever it acts continuously by isometries on a metric space, all orbits are bounded. The properties (OB), (FH) and (T) are thus closely related, and the distinction of them provides better understanding of non-locally compact Polish groups from the large scale viewpoint. The purpose of this paper is to study some of the above mentioned themes, focusing on the two different topologies on the unitary groups, namely the norm topology and the strong operator topology. We also take advantage of operator algebra theory for the study of Polish groups of unitaries. In order to explain the results of this article, let us introduce some terminology. We say that a Hausdorff topological group G is (i) strongly unitarily representable (SUR) if G is isomorphic as a topological group to a closed subgroup of U(H) for some Hilbert space H equipped with the strong operator topology. (ii) norm unitarily representable (NUR) if G is isomorphic as a topological group to a closed subgroup of U(H) for some Hilbert space H equipped with the norm topology. In §3 we study the unitary representability of some Polish groups consisting of unitary operators. Our work in this section is inspired by the works of Megrelishvili [Me08] and Galindo [Ga09] about SUR. If G is a subgroup of U(ℓ2), we denote by Gu (resp. Gs) the group G equipped with the norm (resp. the strong) topology. It is obvious that if M is a von Neumann algebra, then the unitary group U(M )s is SUR. On the other hand, the group U(M )s is NUR if and only if M is finite-dimensional (Corollary 3.12). It is also clear that if A is a separable unital C∗-algebra, then U(A)u is NUR. On the other hand, U(A)u is SUR if and only if A is finite-dimensional (Theorem 3.3). These results show the incompatibility of the two topologies for the non-locally compact case. Although the result is to be expected from the fact that the only von Neumann algebras which are separable in norm are finite-dimensional ones, the proof is entirely different. Theorem 3.3 is used to prove that the Fredholm unitary group U∞(ℓ2) (Definition 2.2) with the norm topology, which is evidently NUR, is not SUR. More interesting example of Polish groups of unitaries are the so-called Schatten p-unitary groups Up(ℓ2) (1 ≤ p < ∞) (Definition 2.2). Although there have been several intensive works on the (projective) unitary representations of Up(H) for p = 1, 2, not much is known about unitary representability of Up(ℓ2). We know that U2(ℓ2) is SUR (see [AM12-2, Theorem 3.2]; in fact it is of finite type, meaning that it embeds into the strongly closed subgroup of a II1 factor with separable predual), but it is unclear whether U2(ℓ2) is NUR. Also it is unknown whether Up(ℓ2) is SUR or NUR for p 6= 1, 2,∞ (if it is SUR, then by [AM12-2, Theorem 2.22], it is actually of finite type). In contrast, we show in §3.2 that U1(ℓ2) is NUR (Theorem 3.16). We do not know if it is SUR. This might suggest that the family Up(ℓ2) shares mixed features of both the NUR groups and the SUR groups. In §4, we study whether the following properties hold for a specific class of Polish groups of unitaries: boundedness, properties (FH), (OB) and (T). If A is a unital abelian C∗-algebra, we show that U(A)u is bounded if and only if its spectrum is totally disconnected, which is known to be equivalent to A having real rank zero. For general A, U(A)u is bounded if and only if the identity component U0(A)u is bounded and has finite index in U(A)u. Thus we focus on U0(A)u. We show that U0(A)u is bounded if and only if it has finite exponential length (Definition 4.10) introduced by Ringrose [Ri91] (Theorem 4.13). Phillips' survey [Ph93] is a very readable account of this and the related notion of the exponential rank. The notion of exponential length has been studied by experts, and it was one of the important ingredients in Phillips' proof [Ph00] of the celebrated Kirchberg -- Phillips' classification Theorem for purely infinite simple nuclear C∗-algebras in the UCT class. Thanks to Lin's result [Li93] that a C∗-algebra A having real rank zero has property weak (FU) of Phillips [Ph91], meaning that any u ∈ U0(A) is the norm limit of unitaries with finite spectrum, it follows that such an algebra has finite exponential length. But there exist unital C∗-algebras with finite exponential lengths which do not have real rank zero (see [Ph92] and Remark 4.15). Thus, the real rank zero property is not necessary for U0(A)u to be bounded. In [Pe18, Example 6.7], Pestov shows that the U2(ℓ2) does not have property (T), and it was left unanswered (see [Pe18, §3.5]) whether U2(ℓ2) has property (FH). We show that it does not. Actually, we prove a more general result that if M is a properly infinite semifinite von Neumann algebra with a faithful normal semifinite trace τ , then the Lp-unitary group Up(M, τ ) (see Definition 2.2) does not have property (FH) for 1 ≤ p < ∞. In fact, we prove that any closed subgroup of Up(M, τ ) which has property (FH) is in fact of finite type (Theorem 4.17, and Theorem 4.28). In particular, any closed subgroup of Up(ℓ2) (1 ≤ p < ∞) with property (FH) must be compact (Corollary 4.20 and Corollary 4.33). For technical reasons, we need two different proofs depending on whether 1 ≤ p ≤ 2 or 2 < p < ∞. This comes from the fact that only 2 for 1 ≤ p ≤ 2, there is a well-defined affine isometric action of Up(M, τ ) on L2(M, τ ) (regarded as a real Hilbert space by taking the real part of the inner product) given by u · x = ux + (u − 1), u ∈ Up(M, τ ), x ∈ L2(M, τ ). It might be interesting to note that the absence of property (FH) for Up(M, τ ) is valid even for M of the form N⊗B(ℓ2), where N is a type II1 factor with property (T) in the sense of [CJ85]. Finally, we show that if A is a unital separable C∗-algebra admitting a representation π which generates a type II1 factor with separable predual having property Gamma, then U(A)u does not have property (T) (Theorem 4.34). This is an immediate consequence of the definition itself, but shows that there exist many C∗-algebras whose unitary groups do not have property (T), while at the same time many of them do have property (FH) (or even (OB)). In particular, the unitary group of the UHF algebra M2∞ of type 2∞ does not have property (T) but has property (OB). 2 Preliminaries We consider complex Hilbert spaces unless explicitly stated otherwise. For a Hilbert space H, we denote by U(H) (resp. K(H)) the multiplicative (resp. additive) group of unitary operators (resp. compact operators) of H. σ(a) denotes the spectrum of a closed, densely defined operator a on H. For 1 ≤ p < ∞, Sp(H) is the additive group of Schatten p-class operators on H. We use the following abbreviations for the operator topologies: SOT for the strong operator topology, SRT for the strong resolvent topology (see e.g., [AM15, §2]). We sometimes use k·k∞ to denote the operator norm, if it is appropriate to distinguish it from other norms, such as the Schatten p-norm k · kp. For a topological group G, we denote by N (G) the set of all open neighborhoods of the identity in G. If G is a subgroup of U(H), we often denote by Gu or (G,k · k∞) (resp. Gs or (G, SOT)) the group G equipped with the norm (resp. the strong operator) topology. For a unital C∗-algebra A, its unitary group (resp. the set of self-adjoint elements in A) is denoted by U(A) (resp. Asa). U0(A) is the identity component of U(A) in the norm topology. Definition 2.1. Let M be a semifinite von Neuman algebra with separable predual and τ a normal faithful semifinite trace on M . The noncommutative Lp-space associated with M is denoted by Lp(M, τ ). We denote by k · kp the p-norm induced by τ . If x is a closed, densely defined operator affiliated with M 0 λ de(λ) is the spectral resolution of x, then and x =R ∞ kxkp = Z[0,∞) λp dτ (e(λ))! 1 p . Details about non-commutative integration theory can be found e.g., in [Tak, Chapter IX.2]. Definition 2.2 ([AM12-2]). Let (M, τ ) be as in Definition 2.1 and 1 ≤ p < ∞. The p-unitary group of M , denoted by Up(M, τ ) is the group of all unitaries u ∈ U(M ) such that u − 1 ∈ Lp(M, τ ). The group is endowed with the metric topology given by the following bi-invariant metric d(u, v) = ku − vkp, u, v ∈ Up(M ). In the case M = B(H) for a Hilbert space H and τ is the usual trace, the group is called the Shcatten p-unitary group and is denoted by Up(H). The group U∞(H) = {u ∈ U(ℓ2) u− 1 ∈ K(H)} endowed with the norm topology is sometimes called the Fredholm unitary group. If M is finite, then Up(M, τ ) = U(M )s is independent of the choice of τ and p. Definition 2.3. Let G be a topological group, π : G → U(H) be a strongly continuous unitary represen- tation. (i) For (∅ 6=)Q ⊂ G and ε > 0, a vector ξ ∈ H is (Q, ε)-invariant if the following inequality holds: x∈Qkπ(x)ξ − ξk < εkξk. sup (ii) We say that π almost has invariant vectors, if it has a (Q, ε)-invariant vector for every compact subset Q of G and every ε > 0. We say that 3 (iii) G has Property (T) if every strongly continuous unitary representation of G which almost has invariant vectors admits a nonzero invariant vector. (iv) G has Property (FH) if every continuous affine isometric action of G on a real Hilbert space has a fixed point. (v) G has Property (OB) if whenever G acts continuously on a metric space by isometries, all orbits are bounded. (vi) G is bounded if for every V ∈ N (G), there exists n ∈ N and a finite set F ⊂ G such that G = V nF . The following results help us understand the implications among the above properties. Theorem 2.4 ([Ro13, Theorem 1.11, Proposition 1.14]). Let G be a Polish group. Then (i) The following three conditions are equivalent. (a) G has property (OB). (b) For every symmetric V ∈ N (G), there exist a finite subset F ⊂ G and k ∈ N such that G = (F V )k, . (c) Every compatible left-invariant metric on G is bounded. (ii) The following three conditions are equivalent. (a) G is bounded. (b) Every left uniformly continuous function on G is bounded. (c) G has property (OB) and every open subgroup is of finite index. Here, a map f from a topological group G to a metric space (X, d) is called left uniformly continuous, if for every ε > 0, there exists U ∈ N (G) such that supx∈G d(f (ux), f (x)) < ε holds (this is sometimes called the right uniform continuity in the literature). Also, a topological group G has property (FH) if and only if for every continuous affine isometric action of G on a real Hilbert space, there is a bounded orbit (see e.g., [BHV08, Proposition 2.2.9]). Therefore, the following implications hold: bounded ⇒ (OB) ⇒ (FH) It is known that none of the above two implications can be reversed in general. Also, (T)⇒(FH) hold for any topological group [De77] (see also [BHV08, Theorem 2.12.4]). In general, there is no implication between (T) and (OB). As pointed out in [Pe18], any countable discrete group with property (T) does not have property (OB) (consider the isometric action on its Cayley graph). On the other hand, U∞(ℓ2) is bounded [At89] hence has property (FH), but it fails to have property (T) [Pe18, Example 6.6] (see also Remark 4.36). For more details about the above mentioned properties, we refer the reader to [BHV08] for property (T) and (FH), [Ro09, Ro13] for property (OB) and boundedness (see also [He58]). 3 Unitary Representability In this section we study a class of Polish groups which arise as closed subgroups of the unitary group. Definition 3.1. We say that a Hausdorff topological group G is (i) strongly unitarily representable (SUR) if G is isomorphic as a topological group to a closed sub- group of U(H) for some Hilbert space H equipped with the strong operator topology. (ii) norm unitarily representable (NUR) if G is isomorphic as a topological group to a closed subgroup of U(H) for some Hilbert space H equipped with the norm topology. (iii) of finite type, if G is isomorphic as a topological group to a closed subgroup of U(M )s for a finite type von Neumann algebra M . (iv) SIN (Small Invariant Neighborhoods) if for every U ∈ N (G), there exists V ∈ N (G) such that V ⊂ U and gV g−1 = V for every g ∈ G. 4 If G is a metrizable topological group, then G is SIN, if and only if G admits a compatible two-sided invariant metric. Thus, every NUR group is SIN (the metric is given by the norm). It is also clear that a finite type Polish group is SIN and SUR. The converse need not be true [AMTT17]. However, we do not know whether there exists a connected SUR SIN Polish group which is not of finite type. Recall also that a topological group G has small subgroups if for every neighborhood U of the identity in G, there exists a non-trivial subgroup of G contained in it. Remark 3.2. Every locally compact Polish groups is SUR via the left regular representation (and it is also NUR if the group is discrete). The solution to the Hilbert's 5th problem (see e.g., [Tao]) asserts that a compact Polish group does not have small subgroups if and only if it is a compact Lie group, in which case the group can be realized as a closed subgroup of U(n) for some n ∈ N, which is NUR. Thus for compact Polish groups, NUR is equivalent to the absence of small subgroups. If a locally compact Polish group is totally disconnected but perfect, then by van Dantzig's Theorem (see. e.g., [Tao]), it has small subgroups and therefore is not NUR (see Proposition 3.11 below). 3.1 Strong Unitary Representability (SUR) We start by showing the next theorem. Theorem 3.3. Let A be a unital C∗-algebra. Then the following conditions are equivalent. (i) The additive group A with the norm topology is SUR. (ii) The additive group Asa with the norm topology is SUR. (iii) U(A)u is SUR. (iv) A is finite-dimensional. The proof is inspired by the works of Megrelishvili [Me08] and Galindo [Ga09]. Recall the notion of cotype. We refer the reader to [AK06] for details. Definition 3.4. A Banach space X has cotype q ∈ [2,∞) if there exists a constant Cq > 0 such that for every finite set of vectors x1, . . . , xn ∈ X the following inequality holds: where (εn)∞ 1) = P(εn = −1) = 1 cotype. n=1 is a Rademacher sequence, i.e., a sequence of i.i.d.'s on a probability space with P(εn = 2 for all n ∈ N. If X has cotype q for some q ∈ [2,∞), we say that X has nontrivial We will also use the following characterization of the SUR, which has been known to experts (we follow [Ga05, Theorem 2.1]). Theorem 3.5 (Folklore). Let G be a Polish group with the identity 1G. The following conditions are equivalent. (i) G is SUR. (ii) G is isomorphic as a topological group to a closed subgroup of U(ℓ2)s. (iii) There exists a family F of continuous positive definite functions on G which generates a neigh- borhood basis of 1G in the sense that for every V ∈ N (G), there exist f1, . . . , fn ∈ F and open sets O1, . . . , On in C such that 1G ∈ f −1 i (Oi) ⊂ V. n\i=1 (iv) There exists a family F of continuous positive definite functions on G which separates 1G and closed sets not containing 1G in the sense that whenever A is a closed set in G with 1G /∈ A, there exists an f ∈ F such that the following inequality holds: g∈Af (g) < f (1G). sup 5 nXi=1 q kxikq! 1 ≤ Cq E(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) nXi=1 εixi(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 1 q , (1) We prepare several lemmas. Lemma 3.6. If a Banach space embeds as a topological vector space into L0(Ω, µ) equipped with the measure topology for some probability space (Ω, µ), then it has cotype 2. Proof. See [BL00, Corollary 8.17]. Lemma 3.7. Let X be a separable real Banach space. If X as an additive topological group is SUR, then there exists a probability space (Ω, µ) such that X embeds isomorphically into L0(Ω, µ) equipped with the topology of convergence in measure. Proof. Let π be a topological group isomorhism of X onto an SOT-closed subgroup of U(ℓ2). The image π(X) generates an abelian von Neumann algebra A with separable predual. Then we may identify A with L∞(Ω, µ) where (Ω, µ) is a probability space. For each a ∈ X, the map R ∋ t 7→ π(ta) ∈ U(A) is a strongly continuous one-parameter unitary group. By Stone Theorem, there exists a unique self-adjoint operator T (a) affiliated with A such that eitT (a) = π(ta) (t ∈ R). Since X is an abelian group, it is easy to see that T : X → L0(Ω, µ) is a real linear map. We show that T is a topological group isomorphism from X onto its image. Since X, L0(Ω, µ) are both metrizable, it suffices to show that for a sequence (an)∞ in X, kank n→∞→ 0 if and only if T (an) n→∞→ T (0) = 0 in measure. Now we use [RS81, Theorem VIII.21] and the fact that the convergence in measure in L0(Ω, µ) is the same as the strong resolvent convergence by [AM12-1, Lemma 3.12]. Then n=1 kank n→∞→ 0 ⇔ ∀t ∈ R [π(tan) = eitT (an) n→∞→ 1 (SOT)] ⇔ T (an) n→∞→ 0 (SRT) ⇔ T (an) n→∞→ 0 (in measure). This finishes the proof, because T (X) is then a Polish subgroup of the Polish group L0(Ω, µ), which therefore must be closed in L0(Ω, µ). Lemma 3.8. Let (an)∞ norm resolvent topology. Then limn→∞ kank = 0. n=1 be a sequence of bounded self-adjoint operators on H converging to 0 in the Proof. Let an =Rσ(a) λ dEan (λ) be the spectral resolution of an (n ∈ N). First, we show that supn∈N kank < n=1 is unbounded. Then for each k ∈ N, there exists nk ∈ N (R \ [−k, k]) 6= 0. Let ξk be a unit vector contained in ∞. Assume by contradiction that (an)∞ with n1 < n2 < ··· < nk, such that Eank ran Eank (R \ [−k, k]). Then k(ank − i)−1ξk − (0 − i)−1ξkk2 =Zλ>k ≥ k2 k2+1 k→∞ 6→ 0, λ2 λ2 + 1 dkEank (λ)ξkk2 which contradicts the assumption. Therefore C = supn∈N kank < ∞, and kank = k(an − i)[(0 − i)−1 − (an − i)−1](0 − i)k ≤ (C + 1)k(an − i)−1 − (0 − i)−1k n→∞→ 0. Proposition 3.9. Let A be a separable unital abelian C∗-algebra such that U(A)u is SUR. Then the additive group Asa with the norm topology is SUR. Proof. Since U(A)u is an SUR Polish group, there exists a topological group isomorphism π of U(A)u onto an SOT-closed subgroup of U(H) for some separable Hilbert space H. Define for each n ∈ N a continuous function ψn : Asa → C by ψn(a) =Z ∞ 0 e−thπ(e−iat)ξn, ξni dt, a ∈ A, where {ξn}∞ n=1 is a dense subset of the unit ball of H. Since A is abelian, the map Asa ∋ a 7→ hπ(e−iat)ξn, ξni ∈ C is a continuous positive definite function for all t ∈ R. This implies that ψn is 6 a continuous positive definite function as well. By Stone Theorem, for each a ∈ Asa there exists a self- adjoint operator T (a) on H such that π(eita) = eitT (a) (t ∈ R) holds. Note that ψn(a) is then expressed as (2) ψn(a) = −ih(T (a) − i)−1ξn, ξni. We show that the family {ψn}∞ n=1 generates a neighborhood basis of 0 in Asa. Let τ be the norm topology on Asa and τ ′ be the weakest topology on Asa making the functions ψn (n ∈ N) continuous. Then τ ′ has a countable basis B = J\j=1 ψ−1 nj (Umj )(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) J ∈ N, nj, mj ∈ N (1 ≤ j ≤ J) , where {Um}∞ τ ′ ⊂ τ holds. Note that because A is abelian, it holds that for a net (aj )j∈J in Asa and a ∈ Asa, m=1 is a countable basis for R. Thus both τ and τ ′ are second countable, and by definition aj j→∞ → a (in τ ′) ⇔ ∀n ∈ N ψn(aj) j→∞ → ψn(a) (∗) ⇔ ∀n ∈ N h(T (aj) − i)−1ξn, ξni ⇔ ∀ξ, η ∈ N h(T (aj) − i)−1ξ, ηi ⇔ T (aj) SRT→ T (a) (∗∗) j→∞ → h(a − i)−1ξn, ξni → h(a − i)−1ξ, ηi j→∞ Here, we used in (∗) the density of {ξn}∞ n=1 and the polarization identity, in (∗∗) the fact that the weak resolvent convergence implies the strong resolvent convergence for self-adjoint operators. Therefore, thanks to the fact that SRT is a vector space topology on the set T (Asa) of strongly commuting self- adjoint operators [AM12-1, Lemma 3.12], it follows that τ ′ is a vector space topology on Asa. We then show that τ ′ ⊃ τ , thus τ ′ = τ . By the first countability of both τ and τ ′, it suffices to show that the notion of sequential convergence to 0 in both topologies agree. Let (an)∞ n=1 be a sequence in Asa converging to 0 in τ ′. Then T (an) n→∞→ T (0) (SRT), whence by [RS81, Theorem VIII.21], we have ∀t ∈ R [eitT (a) n→∞→ eitT (0) (SOT)], so that because π is a topological embedding, the last condition is equivalent to ∀t ∈ R [keitan − 1k n→∞→ 0], which implies that (an − 0)−1 n→∞→ (0 − i)−1 in norm. Indeed, because each an is bounded, the map t 7→ eitan is norm-continuous. Thanks to the dominated convergence theorem, it then follows that k(an − i)−1 − (0 − i)−1k =(cid:13)(cid:13)(cid:13)(cid:13)iZ ∞ ≤Z ∞ 0 0 e−t(e−itan − 1) dt(cid:13)(cid:13)(cid:13)(cid:13) e−tke−itan − 1k dt n→∞→ 0. Then by Lemma 3.8, n→∞kank = 0 holds. Thus τ = τ ′, and Asa is SUR by Theorem 3.5. lim Proof of Theorem 3.3. Since A is isomorphic as an additive group to Asa × Asa, (i)⇔(ii) holds. (ii)⇒(iv): Assume that dim A = ∞. Let B ⊂ A be a maximal abelian ∗-subalgebra of A. Then by Proposition 4.40, dim B = ∞ holds. Then Bsa as a Banach space does not have cotype 2 by Lemma 4.38, so that Bsa is not SUR by Lemma 3.6. Therefore Asa(⊃ Bsa) is not SUR either. (iii)⇒(iv) We show the contrapositive. Assume that dim A = ∞. Let B be a maximal abelian C∗- subalgebra of A. Then dim B = ∞ by Proposition 4.40. By Lemma 4.39, B contains a separable infinite-dimensional unital C∗-subalgebra C. If U(C)u were SUR, then so would be Csa by Proposition 3.9. By Lemma 3.7, it follows that Csa would be isomorphic as a topological vector space to a closed subspace of L0(Ω, µ), which implies that Csa has cotype 2 by Lemma 3.6. However, this contradicts Lemma 4.38. Therefore U(C)u, hence U(A)u is not SUR. (iv)⇒(iii): By dim A < ∞, U(A)u = U(A)s is compact hence SUR. (iv)⇒(ii) Since Asa is a finite-dimensional Banach space, it is isomorphic as an additive Polish group to some Rn (n ∈ R), which is SUR. In [Pe18, The paragraph after Lemma 5.1] (see also §3.4 of the cited paper), Pestov implicitly men- tioned that the Fredholm unitary group U∞(ℓ2) (equipped with the norm topology) is SUR (recall by Theorem 3.5 that SUR is equivalent to having a family of continuous positive definite functions separating points and closed sets). We remark that this is not the case. We show that it is not the case. More generally, we show the following (take A = B(ℓ2), I = K(ℓ2)). 7 Corollary 3.10. Let A be a unital C∗-algebra, I be a separable two-sided closed ideal of A. Then the group UI (A) = {u ∈ U(A) u − 1 ∈ I} equipped with the norm topology is SUR if and only if I is finite-dimensional. 0u0 + zu∗ Proof. If 1 ∈ I, then UI (A) = U(I), which is SUR if and only if dim I < ∞ by Theorem 3.3. Thus we assume that I is proper and UI (A) is SUR. We will show that dim I < ∞. Let eI be the unitization of I, which can be identified with I + C1A ⊂ A. First, we observe that the map π : UI (A) ∋ u 7→ [u] ∈ U(eI)/T π is continuous. Let u = u0 + z1 ∈ U(eI), where u0 ∈ I and z ∈ C. Since u∗u = (u∗ is an isomorphism of topological groups, where [u] is the class of u ∈ UI (A) in the quotient group. Clearly 0 + z1)(u0 + z1) = (u∗ 0 + zu0) + z21 = 1, z = 1 holds. Therefore zu ∈ U(A) and zu − 1 = zu0 ∈ I. This shows that zu ∈ UI (A) and π(zu) = [u], whence π is surjective. For u ∈ UI (A), if [u] = 1, then there exists w ∈ T such that wu = 1, while u − 1 = w − 1 ∈ I. Since 1 /∈ I, this shows that w = 1 and u = 1 holds. Therefore π is injective. To see that π is open, it suffices (since all the topological groups involved are n→∞→ 1. Polish) to show that whenever a sequence (un)∞ Fix such (un)∞ n=1 and write un = vn + 1 (vn ∈ I, n ∈ N). Then for each n ∈ N, there exists zn ∈ T such that limn kznun − 1k = limn kznvn − (1 − zn)k = 0. If zn − 1 does not tend to 0 as n → ∞, then there exists a subsequence (znk )∞ n=1 in UI (A) satisfies [un] n→∞→ [1], we have un k=1 and z0 6= 1 such that znk − z0 k→∞→ 0. For each k, ℓ ∈ N, kznk vnk − znℓvnℓk ≤ kznk vnk − (1 − znk )k + k(1 − znk ) − (1 − znℓ)k + k(1 − znℓ) − znℓvnℓk k,ℓ→∞ → 0. Therefore the limit v = limk znk vnk ∈ I exists. This implies that 0 = lim k→∞ kznk vnk − (1 − znk )k = kv − (1 − z0)k. Then because v ∈ I and 1 − z0 ∈ C1, we have z0 = 1, a contradiction. Therefore π is open and consequently, U(eI)/T is SUR. Since the map U(eI) ∋ u = u0 + z1 7→ ([u], z) ∈ U(eI)/T× T is an embedding of the Polish group U(eI) onto a closed subgroup of U(eI)/T × T, which is SUR, it follows that U(eI) is SUR, whence eI is finite-dimensional by Theorem 3.3. 3.2 Norm Unitary Representability (NUR) Next, we turn our attention to NUR groups. First, the following result follows from a well-known fact that Banach Lie groups do not have small subgroups (see e.g., [Om97, Theorem III.4.1]). We include a simple proof for the reader's convenience. Proposition 3.11. Let H be a Hilbert space and G be a topological group. If G has small subgroups, then there is no norm-continuous injective homomorphism from G into U(H). Proof. Assume by contradiction that π : G → U(H) is a norm-continuous injective homomorphism. Since G has small subgroups, for each U ∈ N (G), there exists a subgroup GU 6= {1G} contained in U . In particular, there exists an element gU ∈ GU \{1G}. Since π is injective, its spectrum σ(π(gU )) contains an element λU ∈ T \ {1}. Then there exists kU ∈ N such that Re(λkU U ) ≤ 0 (we may assume that λU = eiθU with 0 < θU < π θU ). Since GU is a subgroup of G, we have gkU U ) − 1k = 0. However, for each U ∈ N (G), we have 2 . Then we can choose kU ∈ N such that U ∈ GU ⊂ U . This shows that 2θU ≤ kU ≤ π gkU U = 1G, whence U∈N (G)kπ(gkU U∈N (G) lim lim π kπ(gkU U ) − 1k ≥ λkU U − 1 ≥ 1, which is a contradiction. The above proposition shows in particular that the Polish groups of the formQ∞ n=1 Gn, where each Gn is a non-trivial SUR Polish group, are not NUR, although they are SUR (since SUR passes to countable products). Another immediate consequence is that for an infinite-dimensional von Neumann algebra M , the group U(M )s is never NUR. Corollary 3.12. For a von Neumann algebra M with separable predual, the Polish group U(M )s is NUR, if and only if M is finite-dimensional. 8 Proof. If dim M < ∞, then U(M )s = U(M )u is compact, so it is NUR. Conversely, if M is infinite- dimensional, let A be a maximal abelian subalgebra of M which is an infinite-dimensional abelian von Neumann algebra with separable predual (cf. Proposition 4.40). Then A has a direct summand A0 isomorphic to either ℓ∞ or L∞([0, 1]). Let Jn = (2−n, 2−n+1] (n ∈ N). Then the set of all f ∈ L∞([0, 1]) which are essentially constant on each Jn (n ∈ N) forms a von Neumann subalgebra isomorphic to ℓ∞ and U(ℓ∞)s = TN has small subgroups. Thus, in both cases the group U(A0)s, hence U(M )s, has small subgroups. The conclusion now follows from Proposition 3.11. Proposition 3.13. Let C([0, 1], R) be the Banach space of all continuous real valued functions on [0, 1]. Then C([0, 1], R) as an additive Polish group is NUR. Proof. Let G = C([0, 1], R). Let m : G → B(L2([0, 1])) be the map given by [m(f )g](x) = f (x)g(x), f ∈ n=1 L2([0, 1]) C([0, 1], R), g ∈ L2([0, 1]) and x ∈ [0, 1]. Fix an enumeration (tn)∞ and π : G → U(H) be the unitary representation given by π(f ) = (e2πitnm(f ))∞ n=1, f ∈ G. Then π is a topological group isomorphism of G onto π(G) endowed with the norm topology. To see this, let (fn)∞ be a sequence in G. If kfnk∞ n=1 of Q∩[−1, 1]. Let H =L∞ n=1 n→∞→ 0, then n∈Nke2πitnm(fk) − 1k = sup t≤1 kπ(fk) − 1k = sup max x∈[0,1]e2πitfk(x) − 1 k→∞→ 0. Also if kπ(fk) − 1k k→∞→ 0, then for each t ∈ [−1, 1], one has ke2πitm(fk) − 1k k→∞→ 0, and by the group property, the same convergence holds for all t ∈ R. Moreover, because the map t 7→ e−te−itm(fk) is norm-continuous, the dominated convergence theorem implies that (m(fk) − i)−1 − (0 − i)−1 = iZ ∞ Then by Lemma 3.8, we have lim e−t(e−itm(fk) − 1) dt k→∞→ 0 (norm). k→∞ kfkk∞ = 0. Therefore π is a topological group isomorphism of the Polish group G onto π(G) endowed with the norm topology. In particular, π(G) is norm-separable whence π(G) is a Polish group in the norm topology, of which π(G) is a Polish subgroup. Therefore π(G) is norm-closed. k→∞ km(fk)k = lim 0 Corollary 3.14. Every separable Banach space is NUR. Proof. Since C([0, 1]) is isomorphic as a topological group to C([0, 1], R) × C([0, 1], R), C([0, 1]) is NUR. Since every separable Banach space admits a linear isometric embedding into C([0, 1]) by Banach -- Mazur Theorem (see e.g., [AK06, Theorem 1.4.3]), the claim follows from Proposition 3.13. Remark 3.15. Corollary 3.14 implies that there exist non-locally compact Polish groups which are both SUR and NUR. Namely, ℓp (1 ≤ p ≤ 2) as an additive group is such an example (the fact that they are SUR is a classical result of Shoenberg [Sc38]). We do not know if there exists a Polsih group which is both NUR and SUR but fails to be of finite type (cf. [AMTT17]). Next, we show a more interesting example: U1(ℓ2) is NUR. To state the theorem precisely, we need the theory of Fermion Fock space. We briefly recall some basic notations and facts about it and refer the reader to [Ar18, Chapter 6] for details. Let H be a Hilbert space. For each natural number p ≥ 1, let Sp be the symmetric group of degree p ≥ 1, which acts on the p-fold tensor product Hilbert space ⊗pH by U (σ)ξ1 ⊗ ··· ⊗ ξp = ξσ(1) ⊗ ··· ⊗ ξσ(p), σ ∈ Sp, ξ1,··· , ξp ∈ H. Set Ap = 1 p! Xσ∈Sp sgn(σ)U (σ). Then Ap is an orthogonal projection, and its range ∧pH = Ap ⊗p H is called the p-fold anti-symmetric tensor product of H. Put ∧0H = C. The Fermion Fock space over H is defined by F = Ff (H) = ∞Mp=0 ∧p(H). The vector Ω = (1, 0, 0,··· ) ∈ F is called the Fock vacuum. 9 We next define the annihilation and creation operators. For each vector ξ ∈ H, we define the creation operator a†(ξ) as a bounded linear operator on F so that a†(ξ)ψ =pp + 1Ap+1(ξ ⊗ ψ), p ≥ 0, ψ ∈ ⊗pH. The annihilation operator a(ξ) is defined by a(ξ) = a†(ξ)∗, and hence a(ξ)∗ = a†(ξ). They satisfy the canonical anti-commutation relations (CARs): {a(ξ), a(η)∗} = hξ, ηi, {a(ξ), a(η)} = {a(ξ)∗, a(η)∗} = 0 for all ξ, η ∈ H. Here, for two bounded linear operators A, B, the anti-commutator {A, B} = AB + BA It is known that a(ξ)2 = (a(ξ)∗)2 = 0 and ka(ξ)k∞ = kξkH. Moreover, if H is separable is used. infinite-dimensional and {ξn}∞ n=1 is an orthogonal basis for H, then {a(ξn1 )∗ ··· a(ξnℓ )∗Ω ℓ ≥ 0, n1 < n2 < ··· < nℓ} is an orthogonal basis for F . Let CAR(H) be the C∗-algebra generated by {a(ξ) ξ ∈ H}, which is called the CAR algebra. It is known that if H is separable infinite-dimensional, then CAR(H) is ∗-isomorphic to the UHF algebra M2∞ of type 2∞. For any u ∈ U(H) and p ≥ 1, we define a unitary operator ⊗pu on ⊗pH by ξ1,··· , ξp ∈ H. (⊗pu)ξ1 ⊗ ··· ⊗ ξp = uξ1 ⊗ ··· ⊗ uξp, Then ∧pH reduces ⊗pu. We denote by ∧pu its reduced part. Set ∧0u = 1. The second quantization Γ(u) of u is defined by Γ(u) = ∞Mp=0 ∧pu. Note that Γ(u) is a unitary on F , and thus the map Γ : U(H)s → U(F )s is a continuous unitary representation. It also follows that Γ(u)a(ξ)Γ(u)∗ = a(uξ), u ∈ U(H), ξ ∈ H. We can now state the theorem. Theorem 3.16. U1(ℓ2) is isomorphic to a closed subgroup of U(CAR(ℓ2))u via the second quantization map Γ. In particular, U1(ℓ2) is NUR. We need the following three lemmas. Lemma 3.17. If a sequence {xn}∞ limit X = lim N→∞ exists in the operator norm topology and that the inequality n=1 kxnk∞ < ∞, then the (1 + xn) = lim N→∞ (1 + x1)(1 + x2)··· (1 + xN ) n=1 of bounded operators on H satisfies P∞ NYn=1 kX − 1k∞ ≤ exp ∞Xn=1 kxnk∞! − 1 holds. Proof. For natural numbers M > N , we have (1 + xn) − (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) MYn=1 (1 + xn)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)∞ NYn=1 ≤(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (1 + xn)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)∞(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (1 + xn) − 1(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)∞ NYn=1 MYn=N +1 ≤( NYn=1 (1 + kxnk∞))( MYn=N +1 (1 + kxnk∞) − 1) → 0 (M, N → ∞). 10 Hence the limit X = limN→∞QN n=1(1 + xn) exists in the operator norm topology. Similarly, we obtain kX − 1k∞ ≤ (1 + kxnk∞) − 1 ≤ ∞Yn=1 exp (kxnk∞) − 1 ≤ exp ∞Xn=1 ∞Yn=1 kxnk∞! − 1 This completes the proof. Lemma 3.18. For any u, v ∈ U(ℓ2), we have ku − vk∞ ≤ kΓ(u) − Γ(v)k∞. Proof. A direct computation shows that ku − vk∞ = sup kξk=1kuξ − vξkℓ2 = sup kξk=1k{Γ(u) − Γ(v)} a(ξ)∗ΩkF ≤ kΓ(u) − Γ(v)k∞, which is the desired result. Before going to the last lemma, recall that U∞(ℓ2) is a closed subgroup of U(ℓ2) with respect to the norm topology. Lemma 3.19. Let u ∈ U∞(ℓ2), and let 0 < ε < π/2. If kΓ(u) − 1k∞ ≤ (2 − 2 cos ε)1/2, then u ∈ U1(ℓ2) and ku − 1k1 ≤ 2ε hold. Proof. Let u = ∞Xn=1 λnhξn,·iξn, λn = 1, {ξn}∞ n=1 is an orthonormal basis for ℓ2 be the spectral resolution of u ∈ U∞(ℓ2). By Lemma 3.18, we have ku − 1k∞ ≤ (2 − 2 cos ε)1/2, whence we can write λn as λn = eiθn with some real number θn ≤ ε. Put D+ = {n ∈ N θn ≥ 0} and D− = {n ∈ N θn < 0}. θn ≤ ε. To show the claim, we may assume that ♯D+ = ∞, and let {θnℓ}∞ be the subsequence of {θn}∞ n=1 consisting of those θn for which n ∈ D+ holds. We show, by induction on ℓ=1 θnℓ ≤ ε holds. Once we prove this, the claim follows. The case L = 1 follows from the definition of θn. We next assume that the inequality holds for some L ∈ N. Since We claim thatPn∈D+ L ∈ N, that for every L ∈ N, the inequalityPL ℓ=1 (2 − 2 cos ε) we get 1 2 ≥ kΓ(u) − 1k∞ ≥(cid:13)(cid:13)(Γ(u) − 1)a(ξn1 )∗ ··· a(ξnL+1 )∗Ω(cid:13)(cid:13)F =(cid:13)(cid:13)a(uξn1)∗ ··· a(uξnL+1)∗Ω − a(ξn1 )∗ ··· a(ξnL+1 )∗Ω(cid:13)(cid:13)F =(cid:13)(cid:13)a(exp (iθn1)ξn1 )∗ ··· a(exp (iθnL+1)ξnL+1)∗Ω − a(ξn1 )∗ ··· a(ξnL+1 )∗Ω(cid:13)(cid:13)F =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) exp i θnℓ! − 1(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) L+1Xℓ=1 , for some integer m. But we know that 2πm − ε ≤ L+1Xℓ=1 θnℓ ≤ 2πm + ε θnℓ ≤ ε, 0 ≤ θnL+1 ≤ ε, 0 < ε < π 2 . ℓ=1 θnℓ ≤ ε holds, which is the case L + 1. This finishes the proof of the claim. Recall the elementary inequality 0 ≤ 2 − 2 cos x ≤ x2 (x ∈ R). Then we obtain θn ≤ ε. This together with the claim implies thatP∞ eiθn − 1 = (2 − 2 cos θn) θn ≤ 2ε, 2 ≤ 1 ∞Xn=1 ∞Xn=1 n=1 θn ≤ 2ε. 0 ≤ LXℓ=1 HencePL+1 Similarly, we can show thatPn∈D− ∞Xn=1 ku − 1k1 = so that u ∈ U1(ℓ2) and ku − 1k1 ≤ 2ε hold. 11 Proof of Theorem 3.16. We split the proof into two steps. Step 1. For any u ∈ U1(ℓ2), we have Γ(u) ∈ CAR(ℓ2) and kΓ(u) − 1k∞ ≤ eku−1k1 − 1 holds. In particular, the map (cid:16)U1(ℓ2),k · k1(cid:17) →(cid:16)U(CAR(ℓ2)),k · k∞(cid:17), u 7→ Γ(u) is continuous. To see this, let u = ∞Xn=1 λnhξn,·iξn, λn = 1, {ξn}∞ n=1 is an orthonormal basis for ℓ2 be the spectral resolution of u ∈ U1(ℓ2). Since ∞Xn=1 k(λn − 1)a(ξn)∗a(ξn)k∞ ≤ ∞Xn=1 λn − 1 = ku − 1k1 < ∞, we can apply Lemma 3.17 to the sequence ((λn − 1)a(ξn)∗a(ξn))∞ that the limit n=1 of bounded operators on F to get Xu = lim N→∞ NYn=1 {1 + (λn − 1)a(ξn)∗a(ξn)} ∈ CAR(ℓ2) exists in the operator norm topology, and the inequality kXu − 1k∞ ≤ eku−1k1 − 1 holds. Thus it is enough to show that Γ(u) = Xu. For this, let us consider Γ(u)a(ξℓ1 )∗ ··· a(ξℓj )∗Ω = a(uξℓ1)∗ ··· a(uξℓj )∗Ω for any natural numbers ℓ1 < ℓ2 < ··· < ℓj with j ≥ 1. Note that, by the CARs, if n 6= ℓ, we have a(ξn)∗a(ξn) · a(ξℓ)∗ = −a(ξn)∗a(ξℓ)∗a(ξn) = a(ξℓ)∗ · a(ξn)∗a(ξn), and if n = ℓ, we have a(ξn)∗a(ξn) · a(ξℓ)∗ = −a(ξn)∗a(ξℓ)∗a(ξn) + a(ξn)∗ = a(ξn)∗. Hence for all N > ℓj, we obtain NYn=1 {1 + (λn − 1)a(ξn)∗a(ξn)}·a(ξℓ1)∗ ··· a(ξℓj )∗Ω = λℓ1 a(ξℓ1 )∗ · λℓ2 a(ξℓ2 )∗ ··· λℓj a(ξℓj )∗Ω = a(uξℓ1 )∗ ··· a(uξℓj )∗Ω = Γ(u) · a(ξℓ1 )∗ ··· a(ξℓj )∗Ω. Therefore Γ(u) = Xu follows. Step 2. the image Γ(cid:16)U1(ℓ2)(cid:17) is closed and the map Γ is a homeomorphism. Tk n→∞→ 0. It is enough to show that there exists u ∈ U1(ℓ2) such that T = Γ(u) and kun − uk1 n=1 of U1(ℓ2) and any bounded operator T on F satisfying kΓ(un)− n→∞→ 0. To see this, take any sequence (un)∞ By Lemma 3.18, we have kum − unk∞ ≤ kΓ(um) − Γ(un)k∞ n,m→∞ −→ 0, thus there exists u ∈ U∞(ℓ2) such that kun − uk∞ continuous, Γ(un) n→∞→ Γ(u) (SOT). Thus T = Γ(u) and n→∞→ 0. Since the map Γ : U(ℓ2)s → U(F )s is kΓ(u∗un) − 1k∞ = kΓ(un) − Γ(u)k∞ n→∞→ 0. This, together with Lemma 3.19 implies that there exists n0 ∈ N such that for any n ≥ n0, we have u∗un ∈ n→∞→ 0. U1(ℓ2) and ku∗un − 1k1 This finishes the proof. Remark 3.20. We do not know if U1(ℓ2) is SUR. We also remark that Boyer [Bo80, Bo88] studied a family of strongly continuous unitary representations of U1(ℓ2) and U2(ℓ2) . n→∞→ 0. Since un ∈ U1(ℓ2), we conclude that u ∈ U1(ℓ2) and kun − uk1 12 4 Boundedness, Property (OB), (T) and (FH) In this section, we consider structural questions about specific groups of unitaries. As is explained in §2, (T)⇒(FH) and bounded⇒(OB)⇒(FH) hold in general. 4.1 Boundedness and Property (OB) let A be a unital C∗-algebra. Atkin [At91] has shown that U(ℓ2)u We consider the following question: (hence also U(ℓ2)s) is bounded. When is U(A)u bounded? By a simple argument using spectral theory one can show that in fact U(M )u is bounded for every von Neumann algebra and for every unital AF-algebra. This is a folklore result (see e.g., [Neeb14, Proposition 1.3]), but because the proof can be used to prove the boundedness for more general unitary groups, we provide a proof. Proposition 4.1 (Folklore). Let A be a unital AF algebra. Then the group U(A)u is bounded. Remark 4.2. Dowerk [Do18] has shown that the unitary group of a II1 factor or a properly infinite von Neumann algebra has a much stronger property called the Bergman's property. Namely, whenever W1 ⊂ W2 ⊂ ··· is an exhaustive increasing sequence of proper symmetric subsets of U(M ) (no topological condition is imposed on Wn's), there exist n, k ∈ N such that W k n = U(M ) holds. The property (OB) is considered to be (and initially introduced as such by Rosendal [Ro09]) a topological version of the Bergman's property. We need a preparation. Definition 4.3. For 0 < r < 2, let θ(r) be the unique θ ∈ (0, π) satisfying cos θ = 1 − r2 Ir = [0, θ(r)) ∪ (2π − θ(r), 2π), Jr = [θ(r), 2π − θ(r)] = [0, 2π) \ Ir and n0(r) =h 2(π−θ(r)) The following Lemma is also well-known (see e.g., [Neeb14, Proposition 1.3]). θ(r) 2 . We also define i + 1 ∈ N.. r , where Vr = {u ∈ U(M ); ku − 1k < r}. Lemma 4.4. Let M be a von Neumann algebra on a Hilbert space H. Then U(M )u is bounded. More precisely, for any 0 < r < 2, there exists n = n(r) ∈ N (which depends only on r, not on M ) such that U(M ) = V n(r) Proof. Write G = U(M )u. Fix 0 < r < 2. Observe that for θ ∈ [0, 2π), we have eiθ − 12 = 2(1 − cos θ), so that eiθ − 1 < r ⇔ θ ∈ Ir. Let u ∈ G, and let u = R 2π 0 eiθ dE(θ) be the spectral resolution of u. Then we have 2π−θ(r) n0(r) ∈ Ir by the choice of n0(r). Define u1, u2 ∈ G by eiθ dE(θ) + E(Jr), u2 =ZJr n0(r) < θ(r). Note that for θ ∈ Jr, we have u1 =ZIr n0(r) dE(θ) + E(Ir). ei θ θ r 2 ∈ V n0(r)+1 . This shows that U(M ) = V n0(r)+1 Then u1, u2 ∈ Vr, and u = u1un0(r) Proof of Proposition 4.1. Fix a faithful representation A ⊂ B(H) of A on a Hilbert space H. Write G = U(A)u. Let V ∈ N (G). Then there exists 0 < r < 2 such that Vr = {u ∈ G;ku − 1k < r} ⊂ V holds. Since A is AF, it is well-known that there exists a finite-dimensional unital ∗-subalgebra B ⊂ A and v ∈ U(B) such that ku − vk < r. We apply Lemma 4.4 to U(B) to find n(r) ∈ N (depending only on r, independent of B) such that U(B) = (Vr ∩ B)n(r). Also, k1− uv∗k = ku− vk < r implies that uv∗ ∈ Vr. Therefore u = (uv∗)v ∈ Vr(Vr ∩ B)n(r) ⊂ V n(r)+1 . This shows that G = V n(r)+1 and G is bounded. . r r r On the other hand, not every such group is bounded: Example 4.5. Let A = C(T). Then G = U(A)u is not bounded. Proof. Let V = {u ∈ G;ku − 1k∞ < 2}. Then there exists θ0 ∈ (0, π) such that σ(u) ∩ {eiθ; θ ∈ [θ0, 2π − θ0]} = ∅ for all u ∈ V . Assume by contradiction that G = F V n for some F = {f1, . . . , fk} ⊂ G and n ∈ N. Recall that π1(T, 1) = Z is generated by ι(z) = z. We denote by [f ] ∈ π1(T) represented by a loop f : T → T with f (1) = 1 (we identify f ∈ G such that f (1) = 1, with the loop cf : [0, 1] → T = {z ∈ C; z = 1} given by cf (t) = f (e2πit) (t ∈ [0, 1])). Recall also that since T is a topological group, the product [f ] + [g] in π1(T) of loops f, g : T → T(f (1) = g(1) = 1) corresponds to the class [f · g] represented by the product loop z 7→ f (z)g(z). Now, let g ∈ V with g(1) = 1. Then regarding g as a loop in T, we show that [g] = 0. Indeed, if there 13 were n ∈ Z \ {0} for which [g] = n[ι] = [ιn], there would exist a continuous map g : [0, 1] → R such that exp(2πig(t)) = g(e2πit) (t ∈ [0, 1)) and g(1) = n 6= 0. By considering g∗ instead of g if necessary, we may assume that n ≥ 1. Since g is continuous, there exists by mean value theorem a t0 ∈ R such that g(t0) = 1 2 . Thus g(e2πit0 ) = exp(2πig(t0)) = −1. But this contradicts the assumption that g ∈ V . Therefore [g] = 0. Let fi(1) = zi (1 ≤ i ≤ k) and let ni ∈ Z be given by [ fi] = ni[ι], where fi(z) = fi(z)zi. Fix N ∈ Z\{n1, . . . , nk}. We show that f = ιN /∈ F V n. Assume by contradiction that f ∈ F V n. Choose i ∈ {1, . . . , k} and g1, . . . , gn ∈ V such that f (z) = fi(z)g1(z)··· gn(z) (z ∈ T). Let z0 = f (1), wi = gi(1), f (z) = f (z)z0, and gj(z) = gj(z)wj(z ∈ T) (1 ≤ j ≤ n). Then z0 = ziQn j=1 gj is the product of loops based at 1. Therefore j=1 wj, so that f = fiQn [ f ] = N [ι] = [ fi] + [gj] = ni[ι], nXj=1 which contradicts N 6= ni. Therefore G = U(C(T))u is not bounded. Next we characterize abelian unital C∗-algebras having bounded unitary groups. Proposition 4.6. Let X be a compact Hausdorff space. Then the following conditions are equivalent. (i) U(C(X))u is bounded (ii) X is totally disconnected. Remark 4.7. In view of Theorem 4.13 below, Proposition 4.6 follows also from Phillips' work [Ph95, Corollary 3.3], where he shows that cel(C(X)⊗Mn(C)) < ∞ if and only if X is totally disconnected (here, cel stands for the exponential length (see Definition 4.10 below). Note that if X is totally disconnected, then C(X)⊗Mn(C) is a unital AF algebra and therefore its unitary group U(C(X)⊗Mn(C)) is connected. Proof of Proposition 4.6. Let G = U(C(X)). (ii)⇒(i) Assume that X is totally disconnected. Then C(X) is a unital AF-algebra. Thus G is bounded by Proposition 4.1. (i)⇒(ii) We show the contrapositive. Assume that X is not totally disconnected. Then there exists a connected component X0 ⊂ X with X0 ≥ 2. Let V = {u ∈ G; ku − 1k < √2} ∈ N (G). Let F = {g1, . . . , gk} ⊂ G be a finite set and n ∈ N. We show that G 6= F V n. Define D = {z ∈ T; arg(z) ∈ (− π 2 ) is a continuous function. For each g ∈ V , we have that gX0 ∈ C(X0, D). Therefore we can define the map θ ∈ C(X0, R) by θg(x) = log(g(x)) (x ∈ X0) such that g(x) = eiθg(x) (x ∈ X0). We call g ∈ G liftable on X0 if there exists θ ∈ C(X0, R) such that g(x) = eiθ(x) (x ∈ X0), and denote by G1 the set of all g ∈ G which are liftable on X0. Then V ⊂ G1. Note that G1 is a (possibly proper) subgroup of G. Define 2 )}. Then −i log : D ∋ eit 7→ t ∈ (− π 2 , π 2 , π d(g) = max x∈X0 θ(x) − min x∈X0 θ(x), g = eiθ ∈ G1. For g ∈ G1 if there are two θ1, θ2 ∈ C(X0, R) such that g(x) = eiθ1(x) = eiθ2(x) (x ∈ X0), then X0 ∋ x 7→ θ1(x) − θ2(x) ∈ 2πZ is constant, because X0 is connected and θ1, θ2 are continuous. Thus d(g) is independent of the choice of a lifting θ. Moreover, we have d(g1g2) ≤ d(g1) + d(g2) for all g1, g2 ∈ G1. Also observe that if g ∈ V , then d(g) < π (use the canonical lift θg). This shows that d(g) < nπ for all g ∈ V n. Let C = max1≤i≤k d(gi) + nπ. Since X0 ≥ 2, pick x0, x1 ∈ X0 with x0 6= x1. By Urysohn's Lemma, there exists θ ∈ C(X, R) with 0 ≤ θ(x) ≤ 2C (x ∈ X) such that θ(x0) = 0, θ(x1) = 2C. Define g ∈ G1 by g(x) = eiθ(x) (x ∈ X). Then d(g) ≥ 2C. We show that g /∈ F V n. Indeed, if g ∈ F V n, then g = gih1 ··· hn for some i ∈ {1, . . . , k} and hj ∈ V (1 ≤ j ≤ n). Since g, hj ∈ G1 (1 ≤ j ≤ n) and G1 is a j=1 d(hj) ≤ C, contradicting d(g) ≥ 2C. This shows that g /∈ F V n. Therefore G 6= F V n. Since F ⊂ G and n are arbitrary, G is not bounded. subgroup of G, we have gi ∈ G1 as well, so that d(g) ≤ d(gi) +Pn We note the following basic facts. Lemma 4.8. Let G be a locally connected topological group. Then G is bounded if and only if the identity component G0 is bounded and G has finitely many connected components. Proof. See [Ro13, Proposition 1.14] (G has finitely many connected components, if and only if every open subgroup of G has finite index). 14 Corollary 4.9. Let A be a unital C∗-algebra. Then U(A)u is locally connected. Consequently, U(A)u is bounded if and only if U(A)u has finitely many connected components and U0(A)u is bounded. Proof. Note that if u ∈ U(A) satisfies σ(u) 6= T, then u ∈ U0(A). In particular, Vr = {u ∈ U(A) ku − 1k < r} ⊂ U0(A) for every 0 < r < 2. Thus U(A)u is locally connected. Then use Lemma 4.8 to get the latter conclusion. Thus, in view of Corollary 4.9, it is natural to pay attention to the identity component of the unitary group. Then we characterize a unital C∗-algebra for which the connected component of the unitary group is bounded. We need the notion of exponential length introduced by Ringrose (see [Ph93] for more details about exponential length and a closely related notion of exponential rank). Definition 4.10 ([Ri91]). Let A be a unital C∗-algebra. The exponential length cel(u) of u ∈ U0(A) is given as follows: cel(u) = inf( nXk=1 khkk(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) We then define the exponential length of A by u = exp(ih1)··· exp(ihn), h1, . . . , hn ∈ Asa) . cel(A) = sup{cel(u) u ∈ U0(A)}. The exponential length can be defined for non-unital C∗-algebra, but we do not need it here. We need the following results due to Ringrose (see [Ri91, Corollary 2.5, Theorem 2.6 (i)(ii), Corollary 2.7, Proposition 2.9]) and the result due to Lin [Li93] (see also the work by Phillips [Ph94] for the purely infinite simple case). Theorem 4.11 ([Ri91]). Let A be a unital C∗-algebra and n ∈ N. (i) If h ∈ Asa is an element with khk ≤ π, then cel(eih) = khk. (ii) If u ∈ U0(A) satisfies cel(u) < kπ, then there exist h1, . . . , hk ∈ Asa such that khik < π (1 ≤ i ≤ k) and u = eih1 ··· eihk . In particular, cer(u) ≤ k. Theorem 4.12 ([Li93]). Let A be a C∗-algebra of real rank zero. Then A has property weak (FU): The set {eia; a ∈ Asa has finite spectrum} is norm-dense in U( A)0, where A is the unitization of A. Theorem 4.13. Let A be a unital C∗-algebra. Then the following conditions are equivalent. (i) U0(A) is bounded. (ii) cel(A) < ∞. In particular, U0(A) is bounded if A is a unital C∗-algebra of real rank zero. Proof. (ii)⇒(i) Fix the smallest d ∈ N such that cel(A) < dπ. Let u ∈ U0(A). We show that for every 0 < r < 2, there exists n(r, d) ∈ N such that V n(r,d) = U0(A). Choose θ(r), Ir as in Definition 4.3. Let n0(r) =h π n0(r) < θ(r). By Theorem 4.11(ii), there exist h1, . . . , hd ∈ Asa with khjk < π such that u = eih1 ··· eihd . Then σ( hk u = exp(i h1 θ(r)i + 1, so that n0(r) ) ∈ Vr for 1 ≤ k ≤ d. Thus n0(r) ) ⊂ Ir, so that exp(i hk n0(r) )n0(r) ··· exp(i hd n0(r) )n0(r) ∈ V dn0(r) r . π r Thus n(r, d) = dn0(r) works. (i)⇒(ii) Assume that U0(A) is bounded, and fix 0 < r < 2. Then there exist a finite set F = {v1, . . . , vj} ⊂ U0(A) and m = m(r) ∈ N such that U0(A) = F V m r . Because U0(A) is connected and Vr is an open neighborhood of 1 in U0(A), for each 1 ≤ i ≤ j, there exists mj such that vj ∈ V mj , so that U0(A) = V n r , where n = max1≤i≤j mj + m. Then if u ∈ U0(A), there exists u1, . . . , un ∈ Vr such that u = un ··· u1. Let 1 ≤ k ≤ n. Since σ(uk) ⊂ {eiθ; θ ∈ Ir}, there exists hk ∈ Asa with khkk < π such that eihk = uk. By cel(vw) ≤ cel(v) + cel(w) (v, w ∈ U0(A)) and Theorem 4.11(i), we obtain cel(u) < nπ. Since u is arbitrary, cel(A) ≤ nπ holds. Finally, if A moreover has real rank zero, then by Lin's Theorem 4.12, cel(A) = π holds (see also [Ph95, Theorem 3.5]). r 15 Remark 4.14. Let X be a compact Hausdorff space, then the conditions (i) (ii) in Proposition 4.6 are also equivalent to the condition (ii) U0(C(X))u is bounded. Indeed, (i)⇒(iii) follows from Corollary 4.9. For (iii)⇒(i), U0(C(X)) is bounded, then by Theorem 4.13, cel(C(X)) < ∞ holds. Then by [Ph95, Corollary 3.3], X is totally disconnected. In this case, C(X) is a unital AF algebra and therefore U(C(X)) = U0(C(X)) is bounded. Example 4.15. Although U0(C(T))u is not bounded (Example 4.5 and Remark 4.14), Phillips [Ph93, [Ph92]) showed that for A = C(T) ⊗ B(ℓ2), cel(A) = 2. Therefore U0(A)u is Theorem 6.5 (2)] (cf. bounded by Theorem 4.13. Note that A does not have real rank zero (a more general result can be found in Kodaka -- Osaka's work [KO95]). This shows that the real rank zero property is not a necessary property for the identity component of the unitary group of a unital C∗-algebra to be bounded. Finally we make another remark about the relationship between the property (OB) introduced and studied by Rosendal [Ro09] and boundedness in the sense of Hejcman [He58] for groups of the form U(A). In general, boundedness implies property (OB), but the converse may fail. However, if A has real rank zero (here, A is said to have real rank zero, if invertible self-adjoint elements are dense in Asa), then they are equivalent (cf. Remark 4.15): Corollary 4.16. Let A be a unital separable C∗-algebra of real rank zero. Then the following three conditions are equivalent. (i) U(A)u is bounded. (ii) U(A)u has property (OB). (iii) U(A)u has finitely many connected components. Proof. (i)⇒(ii) is clear. (ii)⇒(iii) Assume (iii) does not hold, so Γ = U(A)/U0(A) is a countable infinite discrete group (U(A)0,u is an open subgroup because U(A)u is locally connected). Let q : U(A) → Γ be the quotient map. If Γ has property (T), then Γ is finitely generated. Then Γ acts by left multiplication on its Cayley graph with respect to a finite generating set, which has unbounded orbit. If Γ does not have property (T), then it has some affine isometric action on a real Hilbert space with an unbounded orbit. In either case, we have an isometric action α : Γ y X on a metric space with an unbounded orbit. Then β(u) = α(q(u)) (u ∈ U(A)) defines an isometric action β : U(A) y X with an unbounded orbit. Thus U(A) does not have property (OB). (iii)⇒(i) By Theorem 4.13, U0(A) is bounded. Then (i) follows from Lemma 4.8. 4.2 Property (FH) In [Pe18, Example 6.7], Pestov shows that the group U2(ℓ2) does not have property (T), and it was left unanswered (see [Pe18, §3.5]) whether U2(ℓ2) has property (FH). In this section, we show that it does not have property (FH). Actually we show stronger results. Let M be a properly infinite von Neumann algebra with a normal faithful semifinite trace τ . We will show that the group Up(M, τ ) (1 ≤ p < ∞) does not have property (FH), hence they do not have property (T) either. We need two different proofs depending on whether 1 ≤ p ≤ 2 or 2 < p < ∞. 4.3 Case 1 ≤ p ≤ 2 For a projection e ∈ M , we set Me = {exran (e) x ∈ M}. It is known that Me is a von Neumann algebra acting on the Hilbert space ran (e). We say that e is τ -finite, if τ (e) < ∞. Note that in general the τ -finiteness implies that e is a finite projection, but the converse need not hold. Theorem 4.17. Let M be a semifinite von Neumann algebra with separable predual and τ be a faithful normal semifinite trace on M . Let 1 ≤ p ≤ 2, and let G be a closed subgroup of Up(M, τ ) with property (FH). Then there exists a τ -finite projection e in M such that e reduces all elements in G, that u = 1 on ran (e)⊥, and that the map G → U(Me), u 7→ uran (e) 16 defines an isomorphism from G onto a closed subgroup of U(Me). In particular, G is a Polish group of finite type. We need the following well-known Lemma. Lemma 4.18. Let M be a finite von Neumann algebra with a normal faithful tracial state τ . Then on bounded subsets of M , the topology given by the p-norm k · kp agrees with the SOT for any 1 ≤ p < ∞. Proof. It is well-known that k · k2 agrees with the SOT on bounded subsets of M . Let (xn)∞ n=1 be a bounded sequence in M converging to 0 in SOT. Then because M is of finite type, the ∗-operation is n→∞→ 0 (SOT). Hence (use Stone -- Weierstrass SOT-continuous on bounded subsets of M . Thus, x∗ n→∞→ 0 (SOT), whence kxn n→∞→ 0. Converesely, assume that Theorem) xn n→∞→ 0 (SOT). Therefore n→∞→ 0. Then xn kxnkp 2 k2 p = kxn n→∞→ 0, so that xn kxnk2 = kxnk2 Proof of Theorem 4.17. We regard L2(M, τ ) as a real Hilbert space by taking the real part of the inner product. Define an affine isometric action of G on L2(M, τ ) by n→∞→ 0 (SOT), whence xn = (xn 2 k2 n→∞→ 0 (SOT). 2 = τ (xnp) = kxnkp nxn 2 ) p 2 p 2 p p 2 p p 2 p u · x = ux + (u − 1), This action is well-defined because the inequality u ∈ G, x ∈ L2(M, τ ). τ (u − 12) = τ (u − 12−p · u − 1p) ≤(cid:13)(cid:13)u − 12−p(cid:13)(cid:13)∞ τ (u − 1p) < ∞ ku − 1k2 ≤(cid:13)(cid:13)u − 12−p(cid:13)(cid:13)1/2 implies u − 1 ∈ L2(M, τ ). Note that the above argument shows that we have p ≤ 21−p/2ku − 1kp/2 , n→∞→ u ∈ G and L2(M, τ ) ∋ xn To prove the continuity of the action, let G ∋ un ∞ ku − 1kp/2 u ∈ G. p We see that n→∞→ x ∈ L2(M, τ ). kun · xn − u · xk2 = k[unxn + (un − 1)] − [ux + (u − 1)]k2 ≤ kunxn − uxk2 + kun − uk2 ≤ kunxn − unxk2 + kunx − uxk2 + kun − uk2 ≤ kxn − xk2 + kun − uk2kxk∞ + kun − uk2 ≤ kxn − xk2 + 21−p/2kun − ukp/2 n→∞→ 0, p kxk∞ + 21−p/2kun − ukp/2 p whence the action is continuous. Since G has property (FH), there is a fixed point x0 ∈ L2(M, τ ). Then we have (u− 1)(x0 + 1) = 0 for 0 + 1) reduces all elements in G. We show that 0 + 1), e is in M . It is clear that x∗ 0e = −e, any u ∈ G. Thus the projection e of L2(M, τ ) onto ker (x∗ 0 + 1) = ker (x∗ e is a τ -finite projection in M . Since ker (x∗ thus ex0 = −e, which means that ex0x∗ 0e = e. Then τ (e) = τ (ex0x∗ 0e) = τ (x∗ 0ex0) ≤ τ (x∗ 0x0) < ∞, whence e is a τ -finite projection. The map G → U(Me)s, u 7→ uran (e) is a topological group isomorphism onto a subgroup of U(eM e). Here, we use the fact that for u ∈ G, the identity ku − 1kp = k(u − 1)ekp holds, and that on U(eM e), the p-norm agrees with the SOT by Lemma 4.18. Since G and U(eM e) are Polish, the image of the map is closed. This finishes the proof. Corollary 4.19. Let 1 ≤ p ≤ 2. Then Up(M, τ ) has property (FH) if and only if M is a finite von Neumann algebra. Proof. If M is a finite von Neumann algebra, then by Lemma 4.4, Up(M, τ ) = U(M )s is bounded, whence it has property (FH). Conversely, assume that Up(M, τ ) has property (FH). We apply Theorem 4.17 to G = Up(M, τ ) to find a τ -finite projection e ∈ M such that Up(M, τ ) ∋ u 7→ uran(e) ∈ U(Me)s is an isomorphism onto its image. Since Up(M, τ ) generates M , the projection e commutes with all elements in M . Thus M is a von Neumann subalgebra of a finite von Neumann algebra Me ⊕ Ce⊥. Hence M is a finite von Neumann algebra as well. 17 Corollary 4.20. Let 1 ≤ p ≤ 2, and let G be a closed subgroup of Up(ℓ2) with property (FH). Then G is a compact Lie group. Proof. Since every finite projection e in B(ℓ2) is of finite rank, the group U(B(ℓ2)e) is isomorphic to the unitary group of degree dim ran (e). Then by Theorem 4.17, G is isomorphic to a closed subgroup of a finite-dimensional unitary group, so it is a compact Lie group. 4.4 Case 2 < p < ∞ For the case 2 < p < ∞, we give two different proofs of the fact that Up(M, τ ) does not have property (FH) if M is a properly infinite semifinite von Neumann algebra with separable predual. The first proof (Theorem 4.21), though the argument is somewhat involved, provides a byproduct that for the unitary u ∈ U(M ), a noncommutative analogue of Leach's result [Le56] on a converse of the Holder's inequality holds with some additional estimate on the element z paired with u − 1 by the duality. See Proposition 4.23. Hopefully this and some of the other lemmas used in the first proof might be useful in other purposes. The second proof (Theorem 4.28) is simpler and in a sense a refinement of the argument in Theorem 4.17. Below, ∗SOT stands for the ∗-strong operator topology. Theorem 4.21. Let 2 < p < ∞. Let M be a properly infinite semifinite von Neuman algebra with separable predual and τ be a normal faithful semifinite trace on M . Let G be a closed subgroup of Up(M, τ ) 6⊂ Up(M, τ ). Then G does not have property (FH). In particular, Up(M, τ ) does not have such that G property (FH). ∗SOT Lemma 4.22. Let 2 < p < ∞ and M be a semifinite von Neuman algebra with separable predual and τ a normal faithful semifinite trace on M . Let un, u ∈ Up(M, τ ) (n ∈ N) be such that limn→∞ kun − ukp = 0. Then for every ξ ∈ L2(M, τ ), limn→∞ kunξ − uξk2 = 0 holds. Proof. Since supn∈N kun − uk∞ < ∞ and M ∩ L2(M, τ ) is k·k2-dense in L2(M, τ ), it suffices to show that limn→∞ k(un − u)xk2 = 0 for every x ∈ M ∩ L2(M, τ ). Choose 0 < r < 2 such that 1 r . Then by the generalized noncommutative Holder's inequality, p = 1 2 + 1 k(un − u)xk2 2 = τ ((un − u)x ≤ k(un − u)xk2−r ≤ 22−rkxk2−r r 2(un − u)x2−r(un − u)x ∞ k(un − u)xkr r r 2 ) ∞ kun − ukr pkxkr 2 n→∞→ 0. This finishes the proof. Proposition 4.23. Let 2 < p < ∞, M be a properly infinite semifinite von Neumann algebra with separable predual, τ be a faithful normal semifinite trace on M and let u ∈ U(M ) \ Up(M, τ ). Let 2 < q < ∞ be such that 1 2 . Then there exists z ∈ Lq(M, τ )+ such that 1[ε,∞)(z)z ∈ L2(M, τ ) for every ε > 0 and (u − 1)z /∈ L2(M, τ ). q = 1 p + 1 We need the following classical result due to Leach. Theorem 4.24 ([Le56]). Let 1 < p, q < ∞ be such that 1 following two conditions are equivalent. p + 1 q = 1. Let (X, µ) be a measure space. The (i) For every measurable subset E ⊂ X with µ(E) = ∞, there exists a measurable subset F ⊂ E with 0 < µ(F ) < ∞. (ii) The converse of the Holder's inequality holds, i.e., whenever a measurable function f : X → C has the property that f g ∈ L1(X, µ) for every g ∈ Lq(X, µ), then f ∈ Lp(X, µ) holds. Proof of Proposition 4.23. By assumption, M is of the form M =Mj∈J Qj ⊕ Q0 (3) where J ⊂ N is a subset and each Qj (j ∈ N) is a type I∞ factor (if J 6= ∅. The case J = ∅ is allowed, in which case (3) means M = Q0 is diffuse) and Q0 is either {0} (in which case (3) means J 6= ∅ and treat the case J 6= ∅ and Q0 6= {0}. The other cases are easier to handle. Set J0 = J ∪ {0} and we write M =Lj∈J Qj is completely atomic) or a diffuse properly infinite semifinite von Neumann algebra. We 18 p =Pj∈J0 kxjkp p (whether or not the both x ∈ M as x = (xj )j∈J0 , where xj ∈ Qj (j ∈ J0), so that kxkp sides are finite). 1 3 2 holds. Let A∞ = {λ ∈ T λ− 1 ≥ 2 Let u = RT λ de(λ) be the spectral resolution of u. Then for each j ∈ J0, uj = RT λ dej(λ) is the spectral resolution of uj, where u = (uj)j∈J0 and ej(·) = 1Qj e(·) = e(·)1Qj . Let λ∞ = λ∞+1 = ei π 2 . For 2 ] so that λn− 1 = 2−n+ each n ∈ N, define λn = eiθn , θn ∈ (0, π 2} and An = {eiθ θn+1 ≤ θ < θn} (n ∈ N). Also let pn = e(An), pj,n = ej(An) for each n ∈ N ∪ {∞} and j ∈ J0. Note that T = {1} ⊔Fk∈N∪{∞} Ak is a partition of T. By u − 1 /∈ Lp(M, τ ), we havePj∈J0 kuj − 1Qjkp p = ∞. For each j ∈ J, define cj = τ (qj ), where qj is a minimal projection in the type I∞ factor Qj, which is independent of the choice of qj. We also set c0 = 1. For each J ⊂ J0, we define c J = inf j∈ J cj. For each j ∈ J, the spectral resolution of uj takes the form uj =P∞ Case 1. There exists J ⊂ J such that c J > 0 andPj∈ J kuj − 1Qjkp Define a σ-finite measure space (X, µ) = ( J × N,P(j,k)∈X cjδj,k), which satisfies the condition (i) of k=1 is a sequence of mutually orthogonal minimal projections in Qj withP∞ Theorem 4.24. Define f : X → C by f (j, k) = λj (k) − 12 ((j, k) ∈ X). Then and (ej(k))∞ further divide the situation into two cases. k=1 λj(k)ej(k), where λj(k) ∈ T (k ∈ N) k=1 ej(k) = 1. We p = ∞. p 2 p 2 kfk = X(j,k)∈X λj(k) − 1pcj =Xj∈ J kuj − 1Qjkp p = ∞, 2 (X, µ) holds. By Theorem 4.24, there exists g ∈ L q 2 (X, µ) such that f g /∈ L1(X, µ). Define p and f /∈ L Then g(j, k) 1 2 ej(k). g(j, k) q 2 cj = kgk q 2 q 2 < ∞. z = X(j,k)∈X q = X(j,k)∈X kzkq εqc J ≤ X(j,k)∈Xε X(j,k)∈Xε Thus z ∈ Lq(M, τ )+. Let ε > 0 and Xε = {(j, k) ∈ X g(j, k) 1 holds. 2 cj ≤ kgk g(j, k) q q 2 q 2 < ∞. 2 ≥ ε}. Then the following inequality By c J > 0, this implies that Xε is a finite set, so that k1[ε,∞)(z)zk2 2 = X(j,k)∈Xε g(j, k)cj < ∞. Therefore 1[ε,∞)(z)z ∈ L2(M, τ ) holds. Moreover, k(u − 1)zk2 2 = X(j,k)∈X λj (k) − 12 g(j, k)cj = kf gk1 = ∞, whence (u − 1)z /∈ L2(M, τ ) holds. Case 2. There is no J ⊂ J satisfying the condition stated in Case 1. In this case, for any J ⊂ J, the implication c J > 0 ⇒ Pj∈ J kuj − 1Qjkp J cj ≤ 1} and J0 = J ∪ {0}. Then obviously cJ\ J ≥ 1 > 0, so that Pj∈J\ J kuj − 1Qjkp Pj∈J0 kuj − 1Qjkp Let pk =Pj∈ J0 p = ∞, we also havePj∈ J0 kuj − 1Qjkp we have the following inequality: pj,k (k ∈ N∪{∞}). Then by λ−1 ≤ λk−1, (λ ∈ Ak, k ∈ N) and λ−1 ≤ 2 (λ ∈ A∞), p < ∞ holds. Let J = {j ∈ p < ∞. By p = ∞. ∞ = Xj∈ J0 kuj − 1Qjkp p ≤ 2 p 2 τ (p∞) + ∞Xk=1 λk − 1pτ (pk), 19 we need to consider two cases. Case 2-1. τ (pk) = ∞ for some k ∈ N ∪ {∞}. Assume first that τ (pj,k) < ∞ for all j ∈ J0. In this case, we may find a sequence (en)∞ orthogonal projections in M , such that for each n ∈ N, en is of the form en =Pj∈ Jn fj,n, where ( Jn)∞ is a sequence of (possibly non-disjoint) subsets of J0, and for each j ∈ Jn, fj,n is a subprojection of pj,k ∈ Qj with 1 ≤ τ (en) ≤ 2 (the upper bound τ (en) ≤ 2 can be made possible thanks to the condition that cj ≤ 1 (j ∈ J) and the diffuseness of Q0). Define ∞Xn=1 n=1 of mutually n− 1 2 en. z = n=1 2 τ (en) < ∞ by q > 2 and τ (en) ≤ 2 (n ∈ N). Therefore z ∈ Lq(M, τ )+. For each Then kzkq ε > 0, n=1 n− q q =P∞ k1[ε,∞)(z)zk2 n−1τ (en) < ∞ 2 = Xn≤ε−2 because the right hand side is a finite sum. Thus 1[ε,∞)(z)z ∈ L2(M, τ ) holds. Moreover, thanks to λ − 1 ≥ λk+1 − 1 (λ ∈ Ak) and τ (en) ≥ 1 (n ∈ N), we see that k(u − 1)zk2 2 2 = n− 1 ∞Xn=1 (uj − 1Qj ) · 2 Xj∈ Jn k(uj − 1Qj )fj,nk2 fj,n(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xj∈J0 ∞Xn=1 n−1 Xj∈ Jn n−1 Xj∈ JnZAk λ − 12 dτ (ej (λ)fj,n) ∞Xn=1 ≥ λk+1 − 12 n−1τ (en) = ∞. = 2 ∞Xn=1 Therefore (u − 1)z /∈ L2(M, τ ). Next, assume that τ (pj,k) = ∞ for some j ∈ J0. Then by using the assumption that Qj is either a type I∞ factor or diffuse, we can find a sequence (en)∞ n=1 of mutually orthogonal projections in Qj with τ (en) = cj (n ∈ N), such thatP∞ 2 cj < ∞, whence z ∈ Lq(M, τ )+. By a similar argument as above, we have 1[ε,∞)(z)z ∈ L2(M, τ ) and q =P∞ 2 en. Then kzkq n=1 n− q n=1 n− 1 n=1 en = pj,k. Set z =P∞ ∞Xn=1 2 ≥ λk+1 − 12 k(u − 1)zk2 n−1cj = ∞. Thus (u − 1)z /∈ L2(M, τ ) holds. Case 2-2 τ (pk) < ∞ for all k ∈ N ∪ {∞}. In this case,P∞ 1 (k ∈ N). In view of λk − 1 = 2−k+ 3 k=1 λk − 1pτ (pk) = ∞, henceP∞ 2 (k ∈ N), we have k=1 λk+1 − 1pτ (pk) = ∞ holds by λk+1 − 1 = 2−1λk − Xk:τ ( pk)<1 λk+1 − 1pτ (pk) < ∞. of I. Consider the σ-finite measure space (X, µ) = (N,P∞ This implies that the set I = {k ∈ N 1 ≤ τ (pk)(< ∞)} is infinite. Let k1 < k2 < ··· be the enumeration j=1 ajδj), where aj = τ (pkj ) (j ∈ N), which satisfies the condition (i) of Theorem 4.24. Define f : X → C by f (j) = λkj +1 − 12 (j ∈ N). Then by 2 (X, µ) such that f g /∈ L1(X, µ). assumption, f /∈ L Define 2 (X, µ). Then by Theorem 4.24, there exists g ∈ L p q z = ∞Xj=1 1 2 pkj . g(j) 20 Since kzkq Then by z ∈ Lq(M, τ ), we have (use aj ≥ 1 (j ∈ N)) 2 dµ < ∞, z ∈ Lq(M, τ )+ holds. For each ε > 0, consider Jε = {j ∈ N g(j) 1 q 2 ≥ ε}. q =RX g Xj∈Jε εq ≤ Xj∈Jε ∞Xj=1 g(j) q 2 aj ≤ g(j) q 2 aj = kzkq q < ∞. Thus Jε is finite and consequently k1[ε,∞)(z)zk2 1 (λ ∈ Ak, k ∈ N), we obtain 2 =Pj∈Jε g(j)aj < ∞. Moreover, by λ − 1 ≥ λk+1 − ∞Xj=1 f (j)g(j)aj = ∞ k(u − 1)zk2 2 ≥ by f g /∈ L1(X, µ). Therefore (u − 1)z /∈ L2(M, τ ) holds. Lemma 4.25. Let 2 < p < ∞ and M be a semifinite von Neumann algebra with a faithful normal semifinite trace τ . Let (un)∞ n=1 be a sequence in Up(M, τ ) converging to u ∈ U(M ) in SOT. Let z be a positive self-adjoint operator on L2(M, τ ) affiliated with M , such that 1[ε,∞)(z)z ∈ L2(M, τ ) for every ε > 0. If supn∈N k(un − 1)zk2 < ∞, then (u − 1)z ∈ L2(M, τ ) holds. Proof. By assumption, the sequence ((un − 1)z)∞ L2(M, τ ), so there exists a subsequence ((unk − 1)z)∞ n=1 is a bounded sequence in the separable Hilbert space k=1 converging weakly to some y ∈ L2(M, τ ). Let m ,∞)(z) ∈ M . By k→∞→ u (SOT) assumption, emz = zem ∈ L2(M, τ ) holds. Let x ∈ L2(M, τ ). Then for each k, m ∈ N, unk implies that 0 λ de(λ) be the spectral resolution of z. For each m ∈ N, define em = 1[ 1 z = R ∞ On the other hand, by using the trace property, we also have h(unk − 1)emz, xi k→∞→ h(u − 1)emz, xi. h(unk − 1)emz, xi = h(unk − 1)zem, xi = h(unk − 1)z, xemi k→∞→ hy, xemi ≤ kyk2kxemk2 ≤ kyk2kxk2. This shows that Since x ∈ L2(M, τ ) is arbitrary, this implies that k(u− 1)emzk2 = k(u− 1)zemk2 ≤ kyk2. for every m ∈ N. By the normality of the trace and (u − 1)zem(u − 1)z ր (u − 1)z2 (m → ∞) (SOT), we obtain h(u − 1)emz, xi ≤ kyk2kxk2, m ∈ N. k(u − 1)zemk2 2 = τ (em(u − 1)z2em) = τ ((u − 1)zem(u − 1)z) ր τ ((u − 1)z2) (m → ∞). Thus k(u − 1)zk2 ≤ kyk2 holds. This shows that (u − 1)z ∈ L2(M, τ ). Proof of Theorem 4.21. Take q > 2 such that 1 such that 1[ε,∞)(z)z ∈ L2(M, τ ) for every ε > 0 and (u − 1)z /∈ L2(M, τ ) holds. Define an affine isometric action of G on L2(M, τ ) by q = 1 p + 1 2 . By Proposition 4.23, there exists z ∈ Lq(M, τ )+ u · x = ux + (u − 1)z, u ∈ G, x ∈ L2(M, τ ) (4) Thanks to the noncommutative Holder's inequality, the action is well-defined (by (u − 1)z ∈ L2(M, τ )). We show that the action is continuous. Let (vn)∞ n=1 be a sequence in G converging to v ∈ G and (xn)∞ be a sequence in L2(M, τ ) converging to x ∈ L2(M, τ ). Then by Lemma 4.22, we have n=1 kvn · xn − v · xk2 = kvnxn + (vn − 1)z − vx − (v − 1)zk2 ≤ kvn(xn − x)k2 + k(vn − v)xk2 + k(vn − v)zk2 ≤ kxn − xk2 + k(vn − v)xk2 + kvn − vkpkzkq n→∞→ 0. This shows that the action is continuous. If G had property (FH), there would exist a G-fixed point x0 ∈ L2(M, τ ). By G n=1 in G 6⊂ Up(M, τ ), there exists u ∈ U(M ) \ Up(M, τ ) and a sequence (un)∞ ∗SOT 21 such that un In particular, n→∞→ u (∗SOT). Then for each n ∈ N, we have un· x0 = x0, whence (un− 1)x0 = −(un− 1)z. n∈Nk(un − 1)zk2 = sup sup n∈Nk(un − 1)x0k2 ≤ 2kx0k2 < ∞. (FH). For each n ∈ N, define pn =P∞ However, this is impossible by Lemma 4.25 and (u− 1)z /∈ L2(M, τ ). Therefore G does not have property k=1 is a sequence of n→∞→ −1 mutually orthogonal projections in M such that τ (ek) = 1 (k ∈ N). Then un ∈ Up(M, τ ), and un (SOT). Therefore, by −1 /∈ Up(M, τ ), Up(M, τ ) does not have property (FH). Corollary 4.26. Let 2 < p < M and M be a semifinite von Neumann algebra with separable predual equipped with a normal faithful semifinite trace τ . Then Up(M, τ ) has property (FH), if and only if M is of finite type. k=n+1 ek and un = pn − p⊥ n (n ∈ N), where (ek)∞ Proof. If M is of finite type, then Up(M, τ ) = U(M ) is bounded by Lemma 4.4. Therefore it has property (FH). Conversely, if M is not of finite type, then we may write M = M1 ⊕ M2, where M1 is a finite von Neumann algebra (possibly {0}) and M2 is a properly infinite semifinite von Neumann algebra. If Up(M, τ ) = U(M1, τ1)×U(M2, τ2) had property (FH), then so would Up(M2, τ2), where τi = τMi (i = 1, 2). But this contradicts Theorem 4.21. Thus, Up(M, τ ) does not have property (FH). Remark 4.27. If M is a II∞ factor with separable predual, then the fact that Up(M, τ ) (1 ≤ p ≤ 2) does not have property (T) can be seen from the work of Enomoto -- Izumi [EI16]. Indeed, for each t > 0, let (πt, Ht, ξt) be the cyclic representation of U(M, τ )2 associated with the character ϕt(u) = e−tku−1k2 2 (u ∈ Up(M, τ )). Define π =L∞ . Then one can easily show that the trivial representation π0 is weakly contained in π, but because each πt is a II1 factor representation [EI16, Theorem 1.6], π does not contain π0. Therefore Up(M, τ ) does not have property (T). n=1 π 1 n Next, we give the second proof of the absence of property (FH) for Up(M, τ ). This may be considered to be an analogue of Theorem 4.17. Theorem 4.28. Let 2 < p < ∞, M be a properly infinite semifinite von Neumann algebra with separable predual equipped with a normal faithful semifinite trace τ . Let G be a closed subgroup of Up(M, τ ) with property (FH). Then there exists a τ -finite projection e in M such that e reduces all elements in G, that u = 1 on ran (e)⊥, and that the map defines an isomorphism from G onto a closed subgroup of U(eM e). In particular, G is a Polish group of finite type. G → U(Me), u 7→ uran (e) ΨG(x) = ((u − 1)x)u∈G, x ∈ L2(M, τ ). q = 1 2 . In the rest of this section, we fix (M, τ ) and G satisfying the hypothesis of Theorem 4.28. We represent M on L2(M, τ ) by the standard representation given by τ . Consider a map ΨG : L2(M, τ ) → Qu∈G L2(M, τ ) given by Also fix 2 < q < ∞ such that 1 Lemma 4.29. ker(ΨG) is a closed subspace of L2(M, τ ). Let e be the projection of L2(M, τ ) onto ker(ΨG)⊥. Then e ∈ M and ue = eu for every u ∈ G. Proof. It is clear that ker(ΨG) is a closed subspace of L2(M, τ ). If v ∈ M ′, then for every u ∈ G and x ∈ ker(ΨG), (u − 1)vx = v(u − 1)x = 0. Thus ker(ΨG) is invariant under M ′, whence e ∈ M holds. Moreover, if u, v ∈ G and x ∈ ker(ΨG), then (v − 1)ux = (vu − 1)x− (u − 1)x = 0, whence ux ∈ ker(ΨG). Thus ker(ΨG) is invariant under G. This shows that eu = ue for every u ∈ G. p + 1 Let X = Lq(M, τ ) ∩ M ∩ ker(ΨG)⊥ k·kq (use M ∩ L2(M, τ ) ⊂ Lq(M, τ )). Lemma 4.30. If z ∈ X, u ∈ G and x ∈ ker(ΨG)⊥, then ux + (u − 1)z ∈ ker(ΨG)⊥ holds. n→∞→ 0. Then for each u ∈ G and Proof. Let (zn)∞ n ∈ N, ux+(u−1)zn ∈ ker(ΨG)⊥ by Lemma 4.29. Moreover, by the generalized noncommutative Holder's n→∞→ 0. Thus ux + (u − 1)z ∈ ker(ΨG)⊥ inequality, we see that k(u − 1)(z − zn)k2 ≤ ku − 1kpkz − znkq holds. n=1 be a sequence in M ∩ ker(ΨG)⊥ such that kzn − zkq 22 By Lemma 4.30, for each z ∈ X, one can define a continuous affine isometric action αz of G on ker(ΨG)⊥ (regarded as a real Hilbert space by taking the real part of the inner product) by αz(u)x = ux + (u − 1)z, u ∈ G, x ∈ ker(ΨG)⊥. Because G is assumed to have property (FH), there exists an αz-fixed point in ker(ΨG)⊥. Let x, y ∈ ker(ΨG)⊥ be two αz-fixed points. Then for every u ∈ G, (u − 1)x = (u − 1)y, whence x − y belongs to ker(ΨG), which implies that x = y. Thus, there exists a unique αz-fixed point, which we denote by f (z) ∈ ker(ΨG)⊥. Lemma 4.31. The map f : X ∋ z 7→ f (z) ∈ ker(ΨG)⊥ is an Lq-L2 continuous linear map. Proof. By the uniqueness of the fixed point, it follows that f is a linear map. Because X (resp. ker(ΨG)⊥) is a closed subspace of the Banach space Lq(M, τ ) (resp. L2(M, τ )), thanks to the closed graph theorem, we only have to prove that f is closed. Suppose (zn)∞ n=1 is a sequence in X converging to z ∈ X and that kf (zn) − yk2 n→∞→ 0 for some y ∈ ker(ΨG)⊥. Then for each u ∈ G, we have n→∞→ 0, k(u − 1)(y − f (zn))k2 ≤ 2ky − f (zn)k2 while k(u − 1)(zn − z)k2 ≤ ku − 1kpkzn − zkq n→∞→ 0. Thus, it follows that in the L2-topology, we have (u − 1)y = lim n→∞ (u − 1)f (zn) = lim n→∞−(u − 1)zn = −(u − 1)z. This shows that αz(u)y = y. Then by the uniqueness of the αz-fixed point, y = f (z) holds. This finishes the proof. Lemma 4.32. e ∈ M is a τ -finite projection. Proof. Assume by contradiction that τ (e) = ∞. As in the proof of Proposition 4.23, we write M = cj = τ (qj ), where qj is a minimal projection and c0 = 1. Write e = (ej)j∈J0 . We treat the case J 6= ∅ and Lj∈J Qj ⊕ Q0 as the direct sum of type I∞ factors and a diffuse part, and for each j ∈ J, define Q0 6= {0}. By assumption,Pj∈J0 τ (ej) = ∞ holds. Case 1. There exists j ∈ J0 such that τ (ej) = ∞. Then by using either the diffuseness of Q0 or the fact that Qj (j ∈ J) is a type I∞ factor, there exists a sequence (ej,n)∞ n=1 of mutually orthogonal projections in Qj such that τ (ej,n) = cj for every n ∈ N and Then for each n ∈ N, zn ∈ M ∩ L2(M, τ ) ⊂ Lq(M, τ ) holds. Moreover, because e is the projection onto ker(ΨG)⊥ and ej,n ≤ en ≤ e, we see that (we regard ej,n ∈ L2(M, τ )) zn ∈ M ∩ ker(ΨG)⊥, hence zn ∈ X holds. Thus, because zn ∈ ker(ΨG)⊥, the uniqueness of the αzn -fixed point implies that f (zn) = −zn. On the other hand, by q > 2, for each n ∈ N, we have kznkq q = nXk=1 k− q 2 cj ≤ ∞Xk=1 k− q 2 cj (< ∞), whence supn∈N kznkq < ∞. On the other hand, kf (zn)k2 2 = kznk2 2 = nXk=1 k−1cj n→∞→ ∞. This contradicts the continuity of f by Lemma 4.31. Therefore, τ (e) < ∞ holds. Case 2. τ (ej) < ∞ for all j ∈ J0. Discarding all j ∈ J0 for which τ (ej) = 0, we may assume that τ (ej) 6= 0 for every j ∈ J0. In this case, p=1 is a family of mutually p=1 ej,p holds, where (ej,p)dj orthogonal minimal projections in Qj. for each j ∈ J, there exists dj ∈ N such that ej =Pdj Case 2-1. Pj: cj <1 τ (ej) < ∞. 23 ej =P∞ n=1 ej,n. Define zn := k− 1 2 ej,k, n ∈ N. nXk=1 In this case,Pj: cj ≥1 τ (ej ) = ∞ holds. Let j1 < j2 < . . . be the enumeration of the set J = {j ∈ J cj ≥ 1}. Define m(1, p) = p (1 ≤ p ≤ dj1 ) and m(k, p) = m(k − 1, djk−1 ) + p (1 ≤ p ≤ djk ) for k ≥ 2, and set zn = nXk=1 djkXp=1 m(k, p)− 1 2 c − 1 q jk ej,p. Then zn ∈ X (n ∈ N) as in Case 1, and kznkq q = = djkXp=1 nXk=1 m(n,djn )Xk=1 m(k, p)− q 2 c−1 jk cjk k− 2 q ≤ ∞Xk=1 k− 2 q < ∞, whence supn∈N kznkq < ∞. On the other hand, kznk2 2 = = djkXp=1 nXk=1 m(n,djn )Xk=1 m(k, p)−1c − 2 q jk cjk k−1c 1− 2 q jk ≥ nXk=1 k−1 n→∞→ ∞ by q > 2 and cjk ≥ 1 (k ∈ N). Thus, as in Case 1, we get a contradiction. may find a sequence (en)∞ Case 2-2.Pj: cj<1 τ (ej ) = ∞. Let j1 < j2 < . . . be the enumeration of the set J = {j ∈ J cj < 1}. We n=1 of mutually orthogonal projections in M such that for each n ∈ N, en is of the formPj∈Jn fj,n, where Jn ⊂ J and fj,n ≤ ej (it is possible that Jn ∩ Jm 6= ∅ for some n 6= m) and that 1 ≤ τ (en) ≤ 2 for every n ∈ N. Then define zn =Pn 2 ek ∈ X. By 1 ≤ τ (en) ≤ 2, we have k=1 k− 1 kznkq q = k− q 2 τ (ek) ≤ 2 and kznk2 2 = k−1τ (ek) ≥ then as in Case 1, we get a contradiction. nXk=1 nXk=1 k− q 2 < ∞, ∞Xk=1 nXk=1 k−1 n→∞→ ∞ Proof of Theorem 4.28. By Lemma 4.29 and Lemma 4.32, e is a τ -finite projection in M such that the map G ∋ u 7→ uran(e) ∈ U(eM e) defines a topological group isomorphism of G onto a subgroup of U(eM e)(cf. Lemma 4.18). Since G and U(eM e) are Polish, the image of this isomorphism is closed. Corollary 4.33. Let 2 < p < ∞, and let G be a closed subgroup of Up(ℓ2) with property (FH). Then G is a compact Lie group. Proof. By Theorem 4.28, the proof follows in exactly the same way as Corollary 4.20. 4.5 Property (T) We have seen that there are many examples of unital separable C∗-algebras whose unitary group with the norm topology are bounded as Polish groups. Consequently, such groups have property (FH). It is interesting to know whether they also have property (T). Pestov [Pe18, Example 6.6] showed that the Fredholm unitary group U∞(ℓ2) does not have property (T), while it has property (OB) by Atkin's result [At89, At91]. The notion of property (T) for C∗-algebras has been introduced by Bekka [Be06] and recently Bekka and Ng [BN19] have shown that if G is a locally compact group, then it has property (T) if and only if the full group C∗-algebra C∗(G) has property (T), and if moreover G is an [IN]-group, then this condition is also equivalent to the strong property (T) (in the sense of Ng [Ng14]) for the reduced group C∗-algebra C∗ r (G). Still, the relationship between the property (T) for the unitary group and the C∗-algebra itself is not well understood. We show here that for many A, its unitary group fails to have 24 property (T). This is an immediate consequence of the notion of the property Gamma of Murray and von Neumann. Recall that a type II1 factor M with normal faithful tracial state τ has property Gamma, if for each x1, . . . , xn ∈ M and ε > 0, there exists u ∈ U(M ) such that kuxi − xiuk2 < ε and τ (u) = 0. Theorem 4.34. Let A be a unital separable C∗-algebra admitting a ∗-representation π such that M = π(A)′′ is a type II1 factor with separable predual having property Gamma. Then U(A)u does not have property (T). In particular, this is the case if A is a unital separable simple infinte-dimensional nuclear C∗-algebra with a tracial state. Proof. Let τ be the unique faithful normal tracial state on M = π(A)′′ and we represent M on L2(M, τ ) by the GNS representation associated with τ . We write a (a ∈ M ) when we view a as an element in L2(M, τ ). Recall that any ϕ ∈ N + ∗ is represented as ϕ = h· ξϕ, ξϕi for a unique element ξϕ ∈ L2(M, τ )+. L2(M, τ ) is equipped with the canonical M -M bimodule structure given by x· ξ· y = xJy∗Jξ (x, y ∈ M, ξ ∈ L2(M, τ ) and J x =cx∗ (x ∈ M ). Let H = L2(M, τ ) ⊖ C1. Define a unitary representation ρ : U(A)u → U(H)s by Clearly ρ is strongly continuous. Let Q be a nonempty compact subset of U(A)u and ε > 0. By compactness, there exist u1, . . . um ∈ Q such that Q ⊂ Sm 3 ), where B(u; r) = {v ∈ U(A) ku − vk < r}. Let u ∈ Q. Because M has property Gamma, there exists v ∈ U(M ) such that kvπ(uj) − π(uj)vk2 < ε (1 ≤ j ≤ m) and τ (v) = 0. Then ξ = v ∈ H is a unit vector satisfying kρ(uj)ξ − ξk2 < ε (1 ≤ j ≤ m). Let u ∈ Q and choose 1 ≤ j ≤ m such that ku − ujk < ε ρ(u)ξ = π(u)ξπ(u)∗, ξ ∈ H, u ∈ U(A). j=1 B(uj; ε 3 . Then kρ(u)ξ − ξk ≤ kρ(u)ξ − ρ(uj)ξk + kρ(uj)ξ − ξk ≤ kπ(u − uj)ξπ(u)∗k + kπ(uj)ξπ(u − uj)∗k + ε < ε. 3 This shows that ρ almost has invariant vectors. Assume that η ∈ H is a ρ-invariant vector. Let η = vη be the polar decomposition of η regarded as a closed and densely defined operator on L2(M, τ ) affiliated with M . Then v ∈ M and η ∈ L2(M, τ )+. By the ρ-invariance, we have π(u)η = ηπ(u) for every u ∈ U(A). Since π(A)′ = M ′, this implies that for every u ∈ U(M ), uηu∗ = η holds. Therefore (uvu∗)(uηu∗) = vη, u ∈ U(M ). By the uniqueness of the polar decomposition, it follows that v ∈ M ′ = JM J and uηu∗ = η (u ∈ U(M )). Write v = Jw∗J for w ∈ U(M ). By the latter identity, the functional h·η,ηi is tracial on M . Therefore by the uniqueness of the representing vector in the positive cone, there exists a nonnegative real λ such that η = λ1. Then η = Jw∗Jλ1 = λ w. But for each u ∈ U(M ), the condition uηu∗ = η implies λ\uwu∗ = λ w. Thus w ∈ Z(M ) = C1. This shows that η ∈ C1. But η is orthogonal to 1 by hypothesis, whence η = 0. This shows that ρ does not admit a nonzero invariant vector. This shows that U(A)u does not have property (T). The last assertion is then immediate, because if τ is an extremal point in the tracial state space of A, then τ is faithful (by the simplicity of A) and its associated GNS representation πτ generates a hyperfinite type II1 factor, which has property Gamma. Remark 4.35. The same argument shows that U(M ) as a discrete group does not have property (T) if M is a type II1 factor with separable predual and with property Gamma. This generalizes Pestov's result [Pe18, Example 5.5] for U(R), where R is the hyperfinite type II1 factor. It is also immediate by using Connes -- Jones' notion [CJ85] of property (T) for von Neumann algebras, that for a type II1 factor M with separable predual, if U(M ) as a discrete group has property (T), then M has property (T). The converse implication is however unclear. Remark 4.36. As we mentioned in the beginning of this section, Pestov [Pe18, Example 6.6] showed that the Fredholm unitary group U∞(ℓ2) does not have property (T). His proof is based on his result [Pe18, Theorem 6.3] that a topological group which is both NUR and amenable must be maximally almost periodic, and the fact that U∞(ℓ2) is amenable (in fact, it is extremely amenable by Gromov -- Milman's result [GM83]). We remark that the same type of argument as the one in Theorem 4.34 provides another elementary proof of the absence of property (T) for U∞(ℓ2). This is likely to be known or implicit in the literature, but we include the proof for the reader's convenience. Consider the strongly continuous unitary representation π : U∞(ℓ2) → U(S2(ℓ2)) given by π(u)x = uxu∗, u ∈ U∞(ℓ2), x ∈ S2(ℓ2). 25 Then π almost has invariant vectors (e.g. projection pn onto Cξn (n ∈ N) satisfies kπ(u)pn − pnk2 see that there is no nonzero π-invariant vector. if (ξn)∞ n=1 is an orthonormal basis for ℓ2, then the rank one n→∞→ 0 (u ∈ U∞(ℓ2)), but it is straightforward to Let M2∞ be the UHF algebra of type 2∞ (the CAR algebra). Then U(M2∞ )u is bounded by Propo- sition 4.1, hence it has property (FH). We know that U(M2∞ )u does not have property (T) by Theorem 4.34. The absence of the property (T) can also be seen from the existence of certain type III factor representations. The reason for presenting an alternative proof is our hope that this method could be used to show the absence of property (T) for C∗-algebras without tracial states. We will not pursue this approach further in this paper. Proposition 4.37. The group U(M2∞ )u does not have property (T). 0 hλ = 1 1+λ(cid:18)1 For the proof, we recall the construction of the Powers factors Rλ (0 < λ < 1). For 0 < λ < 1, let 0 λ(cid:19) and ψλ = Tr(hλ · ), which is a normal faithful state on M2(C). Let ϕλ =N∞ n=1 ψλ be the infinite tensor product state on M2∞ and let (Hλ, πϕλ, ξϕλ) be the associated GNS representation of M2∞. Then Rλ = πϕλ (M2∞)′′ is called the Powers factor (of type IIIλ). Let Jψλ (resp. ∆ψλ ) be the modular conjugation operator (resp. the modular operator) associated with ψλ defined on its GNS representation. Similarly we let Jϕλ and ∆ϕλ be the corresponding modular objects associated with ϕλ. Then Jϕλ = Jψλ, ∆ϕλ = ∆ψλ . ∞On=1 ∞On=1 We use the notation Λψλ : M2(C) → Hψλ and Λϕλ : M2∞ → Hϕλ for the canonical dense embedding. For each n ∈ N, let λn = 1 − 1 n+1 . Set ϕn = ϕλn , ψn = ψλn , Hn = Hϕn , πn = πϕn . Then for each a, b, c, d ∈ C, the following formulas hold: ∆ 1 2 ψλ Λψλ(cid:18)(cid:18)a b JψλΛψλ(cid:18)(cid:18)a b d(cid:19)(cid:19) = Λψλ(cid:18)(cid:18) a d(cid:19)(cid:19) = Λψλ(cid:18)(cid:18) a λ λ c c 1 2 c 1 2 b λ− 1 2 b d (cid:19)(cid:19) , d (cid:19)(cid:19) . λ− 1 2 c n+1 (n ∈ N) and consider the GNS representation πn : A → B(Hn) of the Powers state ϕλn . Then Mn = πn(A)′′ is the Powers factor n=1 Hn, n=1 πn : A → B(H) and ξn = (0, . . . , 0, ξn, 0, . . . ) ∈ H. Each Hn is endowed with the standard Proof of Proposition 4.37. Let A = M2∞ ⊃ A0 =Nalg M2(C). Let λn = 1 − 1 of type IIIλn . Let ξn ∈ Hn be the GNS vector of ϕλn . Let ∆n = ∆ϕλn , Jn = Jϕλn , H = L∞ π = L∞ Mn − Mn bimodule structure given by a · ξ · b = aJnb∗Jnξ, Define a unitary representation ρ : U(A) → U(H) by a, b ∈ Mn, ξn ∈ Hn. u ∈ U(A), (ηn)∞ We show that ρ almost has invariant vectors. To this end, we first show that n=1 = (πn(u)ηnπn(u)∗)∞ ρ(u)(ηn)∞ n=1, n=1 ∈ H. n→∞kπn(a)ξn − ξnπn(a)k = 0, lim a ∈ A0. (5) (6) Since A0 is spanned by elementary tensors, it suffices to show (6) for elements of the form a = a1 ⊗ ··· ak ⊗ 1 ··· ∈ A0, where a1, . . . , ak ∈ M2(C) and k ∈ N. By construction, we may identify Hn as the m=1(M2(C), ξψλn ), and ξn = ξ⊗k j=k+1 ξψλn . We have infinite tensor product Hilbert spaceN∞ kOj=1 kOj=1 πn(a)ξn = ξnπn(a) = ψλn ⊗N∞ Λψλn (aj) ⊗ ∞Oj=k+1 ∆ 1 2 ψλn πn(aj)ξψλn ⊗ 26 ξψλn , ∞Oj=k+1 ξψλn . Thus kOj=1 Because the map H n ∋ (η1, . . . , ηn) 7→ η1 ⊗ ··· ⊗ ηn ∈ H ⊗n is continuous, in order to prove (6) for πψλn (b)ξψλnk n→∞→ 0 for all b ∈ M2(C). a = a1 ⊗ ··· ⊗ ak, it suffices to show that kπψλn (b)ξψλn − ∆ πψλn (aj)ξψλn − 1 2 ψλn 1 2 ψλn ∆ . kπn(a)ξn − ξnπn(a)k =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) kOj=1 s(cid:19) ∈ M2(C). Then by λn πn(aj)ξψλn(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Let b =(cid:18)p q r n→∞→ 1, we have kπψλn (b)ξψλn − ∆ 1 2 ψλn πψλn (b)ξψλnk2 = 1 1 + λn {(1 − λ n )2r2 + λn(1 − λ 1 2 − 1 2 n )2q2} n→∞→ 0. This shows (6). Let K ⊂ U(A) be a nonempty norm-compact subset and ε > 0. Because K is totally bounded, there exist u1, . . . , um ∈ K such that K ⊂Sm 3 ), where B(u; r) = {v ∈ U(A); kv− uk < r} (u ∈ U(A)). Since A0 is norm-dense in A, U(A0) is norm-dense in U(A) (well-known). For each i = 1, . . . m, there exists vi ∈ U(A0) such that kui − vik < ε 3 ε). By (6), for each i = 1, . . . , m, there exists ni ∈ N such that kπn(vi)ξn − ξnπn(vi)k = kπn(vi)ξnπn(vi)∗ − ξnk < ε 3 holds for all n ≥ ni. Let n = max1≤i≤m ni. Then for each i = 1, . . . , m, 3 . Then K ⊂Sm i=1 B(ui; ε i=1 B(vi; 2 If u ∈ K, then ku − vik < 2 kρ(vi) ξn − ξnk = kπn(vi)ξnπn(vi)∗ − ξnk < ε 3 . 3 ε for some 1 ≤ i ≤ m. Thus kρ(u) ξn − ξnk ≤ k(ρ(u) − ρ(vi)) ξnk + kρ(vi) ξn − ξnk ≤ kπn(u − vi) ξnπn(u)∗k + kπn(vi) ξn(πn(u∗ − v∗ ≤ 2ku − vik + ε 3 < ε. i )k + ε 3 Since u ∈ K is arbitrary, ρ almost has invariant vectors. Next, we show that ρ does not have a nonzero invariant vector. Assume by contradition that η = (ηn)∞ n=1 ∈ H is a nonzero ρ-invariant vector. Then there exists n ∈ N such that ηn 6= 0, and πn(u)ηnπn(u)∗ = ηn for all u ∈ U(A). We may assume that kηnk = 1. Let τn = h · ηn, ηni ∈ (Mn)+ ∗ . Then for each u, v ∈ U(A), τn(πn(u)πn(v)) = hπn(u)πn(v)ηn, ηni = hπn(u)ηnπn(v), ηni = hπn(u)ηn, ηnπn(v)∗i = hπn(u)ηn, πn(v)∗ηni = hπn(v)πn(u)ηn, ηni = τn(πn(v)πn(u)). Since span(U(A)) = A, we have τn(ab) = τn(ba) for all a, b ∈ πn(A), whence for all a, b ∈ Mn by the normality of τn. This shows that τn is a tracial state. This contradicts the fact that Mn is of type IIIλn . Therefore ρ does not have a nonzero invariant vector. Hence U(A) does not have property (T). Appendix For the reader's convenience, we include proofs of the three well-known results we need in §3.1. Lemma 4.38. Let X be an infinite compact Hausdorff space. Then then the Banach space C(X) does not have nontrivial cotype. Proof. We consider two cases separately. Case 1. X has a connected component X0 consisting of more than two points, say x, x′ ∈ X0 (x 6= x′), then since X0 is a normal space, by Urysohn's Lemma, there exists a continuous function f : X0 → [0, 1] satisfying f (x) = 0, f (x′) = 1. Because X0 is connected, this shows that f (X0) = [0, 1]. By Tietze's extension Theorem, f can be extended to a continuous function on X, which we still denote by f . Then the dual map f ∗ : C[0, 1] → C(X) is an isometric embedding. Since C[0, 1] does not have nontrivial cotype, C(X) does not either. 27 nXi=1 q kxikq! 1 = n 1 q 1 q , E(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) nXi=1 εixi(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) = 1. Case 2. X is totally disconnected. Since X is compact, there exist at most finitely many isolated points in X. Thus X has a connected component X0 which is totally disconnected and perfect. Since X0 is compact Hausdorff, X0 is therefore zero-dimensional, meaning that it has a neighborhood basis consisting of clopen subsets. Then because X0 is infinite, for each n ∈ N, there exists disjoint clopen sets U1, . . . , Un ⊂ X0. Define xi = 1Ui ∈ C(X) (1 ≤ i ≤ n). Let (εn)∞ n=1 be a Rademacher sequence and let q ∈ [2,∞). Then Since n ∈ N is arbitrary, this shows that C(X) cannot have cotype q, which shows the lemma. Lemma 4.39. Let A be an infinite-dimensional unital abelian C∗-algebra. Then there exists a separable infinite-dimensional unital C∗-subalgebra B of A. Proof. Let X be the spectrum of A and identify A = C(X). X is then an infinite compact Hausdorff space. If X is not totally disconnected, then as in the proof of Lemma 4.38, C(X) contains an isometric copy of C[0, 1]. If X is totally disconnected, then because X is compact, there are at most finitely many isolated points, so that there exists an infinite closed totally disconnected perfect subset X1 of X. Then X1 is zero-dimensional, and there exists a countable disjoint family {Un}∞ n=1 of nonempty clopen subsets n=1 2−n1Un ∈ C(X) has infinite spectrum. Thus B = C∗(x, 1) is a separable infinite-dimensional C∗-subalgebra of A. Proposition 4.40. Let A be an infinite-dimensional unital C∗-algebra and B be a maximal abelian ∗- subalgebra of A. Then B is infinite-dimensional. of X1. Then the self-adjoint element x =P∞ Proof. This can be found e.g., in [KRI, Exercise 4.6.12]. We show the contrapositive. Assume that B is finite-dimensional. Let I = {1, 2, . . . , n}, X be the spectrum of B, which is a finite set, say {x1, . . . , xn}. i=1 pi = 1. Since B is maximal abelian, we have piApi = Cpi for every 1 ≤ i ≤ n. In particular, there exists a state ϕi on A such that pixpi = ϕi(x)pi for all x ∈ A, i ∈ I. Define a relation ∼ on I by Let pi = 1{xi} ∈ B. Then {p1, . . . , pn} is an orthogonal family of projections in B andPn i ∼ j ⇔ pjApi 6= {0}. Then ∼ is an equivalence relation. Only the transitivity is not entirely obvious. If i ∼ j and j ∼ k, then pjxpi 6= 0 and pkypj 6= 0 for some x, y ∈ A. Note that for a ∈ A, (pjapi)∗(pjapi) = pia∗pjapi = ϕi(a∗pja)pi 6= 0, (pjapi)(pjapi)∗ = ϕj (apia∗)pj. Therefore pjapi 6= 0 if and only if ϕi(a∗pja) 6= 0, if and only if ϕj (apia∗) 6= 0. Then z = pkypjxpi ∈ pkApi satisfies z∗z = pix∗pjy∗pkypjxpi = pix∗ϕj(y∗pky)pjxpi = ϕi(x∗pjx)ϕj (y∗pky)pi 6= 0 by pjxpi 6= 0 6= pkypj. Thus i ∼ k. Next we show Claim. If pjApi 6= {0}, then dim(pjApi) = 1. Fix y ∈ A such that pjypi 6= 0. Then for every x ∈ A, we have whence (by ϕi(y∗pjy) 6= 0) (pjxpiy∗pj)ypi = pjx(piy∗pjypi) ⇔ ϕj (xpiy∗)pjypi = ϕi(y∗pjy)pjxpi, pjxpi = ϕi(y∗pjy)−1ϕj(xpiy∗)pjypi ∈ Cpjypi. Let I = I1 ⊔ ··· ⊔ Id be the decomposition of I into ∼ equivalence classes. For each 1 ≤ k ≤ d and i, j ∈ Ik, fix an element xi,j ∈ A with pjxi,j pi 6= 0. Then pjxi,j pi = ϕi(x∗ 2 pi, so that pjxi,jpi = uijpixi,j pi gives the polar decomposition of pjxi,j pi, where 2 pjxpi ∈ A. i,jpjxi,j )− 1 ui,j = ϕi(x∗ i,j pjxi,j ) 1 28 Then pjApi = Cui,j if i ∼ j and pjApi = {0} otherwise. have {ui,j; i, j ∈ Is} is a system of matrix units in qsAqs with qs =Pi∈Is It is also straightforward to check that pi. Therefore for every x ∈ A we x = Xi,j∈I dXs=1 Xi,j∈Is pjxpi = ai,j (x)ui,j s=1 MIs(C) is finite-dimensional. for some ai,j(x) ∈ C, and A ∼=Ld Acknowledgments HA is supported by JSPS KAKENHI 16K17608 and Grant for Basic Science Research Projects from The Sumitomo Foundation. YM is supported by KAKENHI 26800055 and 26350231. References [AK06] F. Albiac and N. Kalton, Topics in Banach space theory, Graduate Texts in Mathematics 233, Springer (2006). [AM12-1] H. Ando and Y. Matsuzawa, Lie Group-Lie Algebra Correspondences of Unitary Groups in Finte von Neumann Algebras, Hokkaido J. Math. 41 (2012), no. 1, 31 -- 99. [AM12-2] H. Ando and Y. Matsuzawa, On Polish groups of finite type, Publ. Res. Inst. Math. Sci. 48 (2012), no. 2, 389-408. [AM15] H. Ando and Y. Matsuzawa, The Weylvon Neumann theorem and Borel complexity of unitary equivalence modulo compacts of self-adjoint operators, Proc. Roy. Soc. Edinburgh Sect. A 145 (2015), no. 6, 1115-1144. [AMTT17] H. Ando, Y. Matsuzawa, A. Thom and A. Tornquist, Unitarizability, Maurey -- Nikishin fac- torization and Polish groups of finite type, to appear in J. Reine Angew. Math.. DOI:10.1515/crelle- 2017-0047 [Ar18] A. Arai, Analysis on Fock spaces and mathematical theory of quantum fields, World Scientific (2018). [At89] C. J. Atkin, The Finsler geometry of certain covering groups of operator groups Hokkaido Math. J. 18 (1989), no. 1, 45-77. [At91] C. J. Atkin, Boundedness in uniform spaces, topological groups, and homogeneous spaces, Acta Math. Hungar. 57 (1991), no. 3-4, 213 -- 232. [Ba] W. Banaszczyk, Additive subgroups of topological vector spaces, Lecture Notes in Mathematics, vol. 1466 (1991), Springer-Verlag. [Be03] B. Bekka, Kazhdan's Property (T) for the unitary group of a separable Hilbert space, [Be06] B. Bekka, Property (T) for C∗-algebras, Bull. London Math. Soc. 38 (2006), no. 5, 857867. [BN19] B. Bekka and C-K. Ng, Property (T) for locally compact groups and C-algebras, Proc. Amer. Math. Soc. 147 (2019), no. 5, 21592169. [BHV08] B. Bekka, P. de la Harpe, and A. Valette. Kazhdan's Property (T). Cambridge University Press, Cambridge, 2008. [BL00] Y. Benyamini and J. Lindenstrauss, Geometric nonlinear functional analysis Vol. 1, American Mathematical Society Colloquium Publications, 48. American Mathematical Society, Providence, RI (2000). [Bo80] R. P. Boyer, Representation theory of the Hilbert-Lie group U (H)2, Duke Math. J. 47 (1980), no. 2, 325 -- 344. [Bo88] R. P. Boyer, Representation theory of U1(H) in the symmetric tensors. J. Funct. Anal. 78 (1988), no. 1, 13 -- 23. 29 [CJ85] A. Connes and V. Jones, Property T for von Neumann algebras, Bull. London Math. Soc. 17 (1985), no. 1, 57 -- 62. [CT18] A. Carderi and A. Thom, An exotic group as limit of finite special linear groups, Ann. Inst. Fourier (Grenoble) 68 (2018), no. 1, 257-273 [De77] P. Delorme, 1-cohomologie des repr´esentations unitaires des groupes de Lie semi-simples et r´esolubles. Produits tensoriels continus de repr´esentations, Bull. Soc. Math. France 105 (1977), no. 3, 281-336. [Do18] P. Dowerk, The Bergman property for unitary groups of von Neumann algebras, arXiv:1804.09515. [EI16] T. Enomoto and M. Izumi, Indecomposable characters of infinite dimensional groups associated with operator algebras, J. Math. Soc. Japan 68 (2016), no. 3, 1231 -- 1270. [Ga05] S. Gao, Unitary group actions and Hilbertian Polish metric spaces. in Logic and its applica- tions, Contemp. Math. 380, Amer. Math. Soc. Providence, RI, (2005), 53 -- 72. [Ga09] J. Galindo, On unitary representability of topological groups, J. Math. Z. (2009), no. 1, 211 -- 220. [GM83] M. Gromov and V. D. Milman, A topological application of the isoperimetric inequality, Amer. J. Math. 105 (1983), no. 4, 843-854. [Gu72] A. Guichardet, Sur la cohomologie des groupes topologiques. II, Bull. Sci. Math. (2) 96 (1972), 305-332. [dlH73] P. de la Harpe, Moyennabilit´e de quelques groupes topologiques de dimension infinie, C. R. Acad. Sci. Paris S´er. A-B 277 (1973), A1037-A1040. [dlH78] P. de la Harpe, Moyennabilite du groupe unitaire et propriete P de Schwartz des algebres de von Neumann, Lecture Notes in Math., 725, Springer, (1979), 220 -- 227. [He58] J. Hejcman, Boundedness in uniform spaces and topological groups, Czechoslovak Math. J. 9 (84) (1959), 544 -- 563. [HC75] W. Here and J. P. R. Christensen, On the existence of pathological submeasures and the con- struction of exotic topological groups, Math. Ann. 213 (1975), 203-210. [KRI] R. Kadison, J. Ringrose, Fundamentals of the theory of operator algebras. Vol. I. Elementary theory, American Mathematical Society, Providence, RI (1997). [Ka67] D. A. Kazdan, On the connection of the dual space of a group with the structure of its closed subgroups (Russian), Funkcional. Anal. i Prilozen 1 (1967), 71 -- 74. [KO95] K. Kodaka and H. Osaka, Real rank of tensor products of C∗-algebras. Proc. Amer. Math. Soc. 123 (1995), no. 7, 2213 -- 2215. [Le56] E. Leach, On a converse of the Holder inequality, Proc. Amer. Math. Soc. 7 (1956), 607-608. [Li93] H. Lin, Exponential rank of C∗-algebras with real rank zero and the Brown-Pedersen conjectures, J. Funct. Anal. 114 (1993), no. 1, 1 -- 11. [Me08] M. Megrelishvili, Reflexively representable but not Hilbert representable compact flows and semi- topological semigroups, Colloq. Math. 110 (2008), no. 2, 383 -- 407. [Neeb14] K. H. Neeb, Unitary representations of unitary groups, Dev. Math., 37, Springer (2014), 197 -- 243. [Ng14] C-K. Ng, Property (T) for general C∗-algebras, Math. Proc. Cambridge Philos. Soc. 156 (2014), no. 2, 229239 [Om97] H. Omori, Infinite-Dimensional Lie Groups, Translations of Math. Monographs 158 Amer. Math. Soc. (1997). [Pe18] V. Pestov, Amenability versus property (T) for non locally compact topological groups, Trans. Amer. Math. Soc. 370 (2018), no. 10, 7417 -- 7436. 30 [Ph91] N. C. Phillips, Simple C-algebras with the property weak (FU), Math. Scand. 69 (1991), no. 1, 127-151. [Ph92] N. C. Phillips, The rectifiable metric on the space of projections in a C∗-algebra, Internat. J. Math. 3 (1992), no. 5, 679 -- 698. [Ph93] N. C. Phillips, A survey of exponential rank. Contemp. Math., 167, Amer. Math. Soc., Providence, RI, (1994), 352 -- 399. [Ph94] N. C. Phillips, Approximation by unitaries with finite spectrum in purely infinite C∗-algebras, J. Funct. Anal. 120 (1994), no. 1, 98-106. [Ph95] N. C. Phillips, Exponential length and traces, Proc. Roy. Soc. Edinburgh Sect. A 125 (1995), no. 1, 13 -- 29. [Ph00] N. C. Phillips, A classification theorem for nuclear purely infinite simple C∗-algebras, Doc. Math. 5 (2000), 49 -- 114. [Ph02] N. C. Phillips, Real rank and exponential length of tensor products with O∞, J. Operator Theory 47 (2002), no. 1, 117 -- 130. [Ri91] J. R. Ringrose, Exponential length and exponential rank in C∗-algebras, Current Topics in Oper- ator Algebras (H. Araki etc., ed.), World Scientific (1991), 395 -- 413. [Ri92] J. R. Ringrose, Exponential length and exponential rank in C∗-algebras, Proc. Edinburgh Math. Soc. A., (1992), no. 1 -- 2, 55 -- 71. [Ri71] J. R. Ringrose, Compact non-self-adjoint operators, Van Nostrand Reinhold Co (1971). [Ro09] C. Rosendal, A topological version of the Bergman property, Forum Math. 21 (2009), no. 2, 299 -- 332. [Ro13] C. Rosendal, Global and local boundedness of Polish groups, Indiana Univ. Math. J. 62 (2013), no. 5, 1621 -- 1678. [Ro18] C. Rosendal, Coarse geometry of topological groups, available at author's webpage. [RS81] M. Reed and B. Simon, Methods of modern mathematical physics, vol I: Functional Analysis, Academic Press (1981). [Sc38] I. J. Schoenberg, Metric spaces and positive definite functions, Trans. Amer. Math. Soc. 44 (1938), no. 3, 522 -- 536. [Tak] M. Takesaki, Theory of Operator Algebras. I (2002), II (2003) and III (2003). Springer-Verlag. [Tao] T. Tao, Hilbert's fifth problem and related topics, Graduate Studies in Mathematics, 153 (2014), American Mathematical Society. Hiroshi Ando Department of Mathematics and Informatics, Chiba University, 1-33 Yayoi-cho, Inage, Chiba, 263- 8522, Japan [email protected] Yasumichi Matsuzawa Department of Mathematics, Faculty of Education, Shinshu University, 6-Ro, Nishi-nagano, Nagano, 380- 8544, Japan [email protected] 31
1110.2758	1	1110	2011-10-12T18:57:07	Amplified graph C*-algebras	[ "math.OA" ]	We provide a complete invariant for graph C-algebras which are amplified in the sense that whenever there is an edge between two vertices, there are infinitely many. The invariant used is the standard primitive ideal space adorned with a map into {-1,0,1,2,...}, and we prove that the classification result is strong in the sense that isomorphisms at the level of the invariant always lift. We extend the classification result to cover more graphs, and give a range result for the invariant (in the vein of Effros-Handelman-Shen) which is further used to prove that extensions of graph C-algebras associated to amplified graphs are again graph C*-algebras of amplified graphs.	math.OA	math	AMPLIFIED GRAPH C∗-ALGEBRAS SØREN EILERS, EFREN RUIZ, AND ADAM P.W. SØRENSEN Abstract. We provide a complete invariant for graph C∗-algebras which are amplified in the sense that whenever there is an edge between two vertices, there are infinitely many. The invariant used is the standard primitive ideal space adorned with a map into {−1, 0, 1, 2, . . . }, and we prove that the classification result is strong in the sense that isomorphisms at the level of the invariant always lift. We extend the classification result to cover more graphs, and give a range result for the invariant (in the vein of Effros- Handelman-Shen) which is further used to prove that extensions of graph C∗-algebras associated to amplified graphs are again graph C∗-algebras of amplified graph. 1. Introduction 1 1 0 2 t c O 2 1 ] . A O h t a m [ 1 v 8 5 7 2 . 0 1 1 1 : v i X r a ∼= C2 ⇐⇒ F(C1) ∼= F(C2). C1 When classifying C∗-algebras we usually consider some subcategory, C say, of all C∗- algebras. We then hope to find a functor F from C or from the category of all C∗-algebras to some other category, D say, with the property that Hopefully it is easy to determine if two objects in D are isomorphic. If one somehow comes to be in possession of such a classifying functor, there are several natural questions to ask. Is our functor a strong classifying functor? That is, given some isomorphism φ : F(C1) → F(C2) can we find an isomorphism ψ : C1 → C2 such that F(ψ) = φ? What is the range of F? If the domain of F is all C∗-algebras, then we can ask under what conditions F(A) ∈ F(C) guarantees that A ∈ C. There are many examples of such functors. The best known is perhaps the one that sends a C∗-algebra A to its primitive ideal space, denoted Prim(A). Restricting to the category of commutative C∗-algebras we obtain a classifying functor, since in this case we may apply Gelfand duality, and if one restricts further to unital commutative C∗-algebras then this is even a strong classifying functor. It is also well known that the range of the functor is all locally compact Hausdorff spaces, but apart from the obvious fact that when Prim(A) is not Hausdorff, then A is not commutative, there is no obvious way to recognize the commutative C∗-algebras by looking at primitive ideal spaces. Among other well known classifying functors, we have the ordered K0 (with unit or scale) which classifies AF-algebras, and the graded K0 ⊕ K1, which classifies purely infinite simple C∗-algebras which are nuclear and fall in the UCT class. These are strong classification results, the ranges of the invariants are known, and in the former case quite a lot is known about how to recognize the classified C∗-algebra in larger classes by looking at their K- theory. In [14] it was boldly conjectured that the ideal related K-theory F K(−), is a classi- fying (up to stable isomorphism) functor for graph C∗-algebras with finitely many ideals. Supporting evidence for this conjecture can, for instance, be found in [15, 14, 10], where Date: July 4, 2018. 2000 Mathematics Subject Classification. Primary: 46L35, 37B10 Secondary: 46M15, 46M18. Key words and phrases. Classification, Extensions, Graph algebras. 1 2 SØREN EILERS, EFREN RUIZ, AND ADAM P.W. SØRENSEN graph C∗-algebras with small or otherwise special ideal lattices are classified using F K(−). Although we are apparantly still a long way away from resolving whether F K(−) is a clas- sifying (up to stable isomorphism) functor for graph C∗-algebras, the conjecture raises the following additional questions: A Is F K(−) a classifying functor? 1 Is it a strong classifying functor? 2 Can we achieve exact classification? 3 What is the range of F K? B Which relation on graphs is induced by stable isomorphism? C Are the graph algebras recognizable within larger classes of C*-algebras by F K(−)? 1 Is it possible to achieve permanence results for extensions of graph algebras? These questions and their answers affect one another. We have tried to capture these connections in the following diagram. In the class of graph C∗-algebras, results pertaining to A have been given in [15], [10], and [14], and more will appear in [12]. The issue A1 is the subject of [9] as well as [23], [12], and all of these papers along with [11] adress issue A2. A3 is the subject of [7] as well as forthcoming work in [1]. The question B is resolved for simple unital graph C∗-algebras in [25]. And results of relevance for C and C1 have been obtained in [7], [11] and [1]. In the present paper we resolve all the questions A, B and C for a special class of graph algebras – imposing this time, indirectly, a requirement on the involved K-groups instead of on the ideal lattice. Instead of working directly with F K(−), we introduce, for any C∗- algebra, the tempered primitive ideal space. It turns out to be a complete invariant for the algebras we wish to study. This invariant, which we denote by Primτ (A), consists of the standard primitive ideal space Prim(A) along with a map τ : Prim(A) → Z∪{−∞,∞} which describes the nature of the K0-groups of certain sub-quotients of A. A formal definition is given in Section 4. In terms of graphs, what we want to consider are graphs with finitely many vertices and the special property that if there is an edge between two vertices, then there are infinitely many edges between them. We call such graphs amplified. Naturally any vertex in such a graph is either an infinite emitter or a sink, and so the K0-group is easily computed. It is simply the free abelian group with as many generators as our graph has vertices. Furthermore such graphs always satisfies the technical condition (K) so the ideal structure of a graph algebra is readily understood from the path structure of the graph. An important concept for us is the transitive closure of a graph G, defined in the case of G by adding an edge e with s(e) = v and r(e) = w to the graph if no such edge exists, but there is a path from v to w in G. We denote this graph by tG. We also need the amplification of a graph G, defined by adding countably infinite number of edges from v to w if there exists an edge in G from v to w. We denote this graph by G. We can now state one of the key results of our paper. AAA2A3A1A1CBCAMPLIFIED GRAPH C∗-ALGEBRAS 3 Theorem 1.1. Let G and F be graphs with finitely many vertices. The following are equivalent. (i) tG ∼= tF . (ii) C∗(tG) ∼= C∗(tF ). (iii) C∗(G) ∼= C∗(F ). (iv) C∗(G) ⊗ K ∼= C∗(F ) ⊗ K. (v) Primτ (C∗(G)) ∼= Primτ (C∗(F )). (vi) F K(C∗(G)) ∼= F K(C∗(F )). Aside from showing that F K(−) is a classifying functor for amplified graph algebras, the result also gives a concrete geometric description of when two amplified graphs gives rise to isomorphic algebras. We provide various extensions of this result; we extend to cover classification of (some) graphs where all vertices are singularities and prove an Effros-Handelman-Shen type theorem for the range of the invariant. This is then used to show permanence properties for certain graph C∗-algebras, hence answering questions C and C1. 2. Graphs and their algebras Across the literature on graph algebras there is some inconsistency about how to turn the arrows when defining the graph algebra. We go with the notation from, for instance, [6]. Definition 2.1. Let G = (G0, G1, sG, rG) be a graph. A Cuntz-Krieger G-family is a set of mutually orthogonal projections {pv v ∈ G0} and a set {se e ∈ G1} of partial isometries satisfying the following conditions: esf = 0 if e, f ∈ G1 and e (cid:54)= f , (CK0) s∗ ese = prG(e) for all e ∈ G1, (CK1) s∗ e ≤ psG(e) for all e ∈ G1, and, (CK2) ses∗ (CK3) pv =(cid:80) e∈s G (v) ses∗ −1 e for all v ∈ G0 with 0 < s−1 G (v) < ∞. The graph algebra C∗(G) is defined as the universal C∗-algebra given by these generators and relations. We now define a few graph concepts. Definition 2.2. Let G be graph and let u, v be vertices in G. We write u ≥ v if there is a path from u to v in G or if u = v. Definition 2.3. Let G be a graph. A subset H ⊆ G0 is called hereditary if for all u ∈ H we have u ≥ v =⇒ v ∈ H. We denote by H(G) the lattice of hereditary sets in G0. Of particular importance to us is the amplification of a graph. = G0, Definition 2.4. Let G be a graph. The amplification of G, denoted by G, is defined by 0 G =(cid:8)e(v, w)n n ∈ N, v, w ∈ G0 and there exists an edge from v to w(cid:9) , 1 G and sG(e(v, w)n) = v, and rG(e(v, w)n) = w. If E = G for some graph G we say that E is an amplified graph. 4 SØREN EILERS, EFREN RUIZ, AND ADAM P.W. SØRENSEN The ideal structure of graph algebras is well understood. For amplified graphs it is especially nice. Since all sets of vertices automatically are saturated and the graphs always satisfy the technical condition (K), we have an isomorphism between H(G) and the ideals of C∗(G). See [2] for details. This connection between the path structure of G and the ideals of the associated algebra motivates our next definition. Definition 2.5. Let G = (G0, G1, rG, sG) be a graph. Define tG as follows: tG0 = G0, tG1 = G1 ∪ {e(v, w) there is a path but no edge from v to w} , with range and source maps that extend those of G and satisfies stG(e(v, w)) = v, rtG(e(v, w)) = w. The idea is that in tG the relations "there is a path between" and "there is an edge between" becomes the same. Of course adding one edge is fairly arbitrary. A more natural choice would perhaps be to add as many edges from u to v as there paths from u to v. But with that choice ttG might not be the same as tG. We will almost only use the transitive closure together with the amplification, thus this choice is irrelevant. There is one final class of graphs that will be important for us. Definition 2.6. A graph G is called singular if every vertex in G is either an infinite emitter or a sink. Remark 2.7. An amplified graph is obviously singular. And there are non-amplified sin- gular graphs, cf. Example 3.10. 3. A move on graphs In this section we will describe a simple way to alter an amplified graph without changing the isomorphism class of the associated algebra. Lemma 3.1. Let G be a graph, and let u ∈ G0 be some vertex that emits infinitely many edges to some finite emitter v in G0. Let E be the graph with vertex set G0, edge set E1 = G1 ∪ {f n n ∈ N, f ∈ s−1 G (v)}, and range and source maps that extend those of G and have rE(f n) = rG(f ) and sE(f n) = u. Then C∗(G) ∼= C∗(E). In the lemma if G looks like: ∞ / / v u ?~~~~~~~ ?~~~~~~~ @@@@@@@ • • ? ? AMPLIFIED GRAPH C∗-ALGEBRAS 5 then E will look like: ∞ ∞ / v u ∞ ?~~~~~~~ ?~~~~~~~ @@@@@@@ • • Proof. Let {pv, se v ∈ G0, e ∈ G1} be the Cuntz-Krieger G-family generating C∗(G). Let {e1, e2, . . .} = {e ∈ G1 sG(e) = u, rG(e) = v}. For each edge e ∈ G1 \ {en n ∈ N} we let te = se. For each en we let ten = se2n−1. For each f n we let tf n = se2nsf . We claim that {pv, te v ∈ E0, e ∈ E1} is a Cuntz-Krieger E-family in C∗(G). First we check that all the te, e ∈ E1, have orthogonal ranges. Let e, f ∈ G1. If neither e nor f has source u we have te = se and tf = sf , so they have orthogonal ranges. Suppose now that sG(e) = u and sG(f ) (cid:54)= u. Then we can write te = sgx where g is an edge with sG(g) = u and x is some element in C∗(G). Thus we have etf = x∗s∗ t∗ the last equility holds since sG(f ) (cid:54)= u so g (cid:54)= f . From this we also get t∗ etf )∗ = 0. We now consider the case where both e and f have source u. The only case which is different from before, is if e = gn and f = hm for some edges h, g ∈ s−1 G (v) and n, m ∈ N. In this case we have f te = (t∗ gsf = 0, etf = s∗ t∗ gs∗ e2n gpvs∗ se2m sh = δn,ms∗ f gsf = δn,mδg,f prG(g) = δn,ms∗ = δn,mδg,f prE (e), which gives the desired result. Note that by the above computations the relation t∗ ete = prE (e) holds for all e ∈ E1. The sum-relation (CK3) holds at all vertices, since the only vertex, where we changed the out going edges (and the corresponding partial isometries) is an infinite emitter. So it only e ≤ psE (e) for all e ∈ E1. This is easily seen to be true unless remains to verify that tet∗ e = f n for some f ∈ s−1 G (v) and n ∈ N. But even in this case we see that ≤ psG(e2n) = pu = psE (e). e = se2n sf s∗ ≤ se2n s∗ Hence {pv, te} is a Cuntz-Krieger E-family. By universality we get a ∗-homomorphism φ : C∗(E) → C∗(G). We claim that φ is an isomorphism. The only generators of C∗(G) that are not trivially in φ(C∗(E)) are se2n, n ∈ N. To see that these generators are in the image we fix some n ∈ N. For each f ∈ s−1 G (v) we have tet∗ f s∗ e2n e2n Since v is a finite emitter, we get (cid:88) f∈s −1 G (v) tf n t∗ f = se2n tf n t∗ f = se2nsf s∗ f .  (cid:88) f∈s −1 G (v)  = se2n pv = se2n . sf s∗ f So es2n is in the image of φ, and hence φ is surjective. / 0 0 . . ? ? 6 SØREN EILERS, EFREN RUIZ, AND ADAM P.W. SØRENSEN We now turn to injectivity. We will define a strongly continuous action α of T on C∗(G). Let H = G0 \ {e2n n ∈ N}. For each z ∈ T define αz(se) = and (cid:40) zse, se, if e ∈ H if e /∈ H , αz(pv) = pv, v ∈ G0. By universality this defines an action of T on C∗(G). A standard argument will show that α is strongly continuous. If γ is the standard gauge action on C∗(E) then we have φ ◦ γz = αz ◦ φ, for all z ∈ T. The gauge-invariant uniqueness theorem [2, Theorem 2.1] now implies that φ is injective. (cid:3) We do not need to add edges from u to all the the vertices v emits to. Corollary 3.2. Let G be a graph, u ∈ G0 an infinite emitter. Fix a finite emitter v that u emits infinitely to, and fix an edge f ∈ s−1 G (v). Let E be the graph with vertex set G0, edge set E1 = G1 ∪ {f n n ∈ N}, and range and source maps that extend those of G and have rE(f n) = rG(f ) and sE(f n) = u. Then C∗(E) ∼= C∗(F ). Proof. Applying Lemma 3.1 to both E and G yields isomorphic graphs. (cid:3) The requirement that v is a finite emitter seems somewhat artificial, as we are focusing on a single edge leaving v. We now wish to remove that requirement. To do that, we use the out-splittings of Bates and Pask [3]. For the convenience of the reader, we record a special case of [3, Theorem 3.2]: Theorem 3.3 (Out-split). Let G be a graph, and let v ∈ G0. Given an non-empty subset E1 of s−1 os, ros, sos) by G (v)\E1 is non-empty, we define a graph Gos = (G0 G (v) such that E0 = s−1 os, G1 os =(cid:0)G1 \ r−1 os = (G0 \ {v}) ∪ {v0, v1}, G0 G1 G (v)(cid:1) ∪ {e0, e1 e ∈ E1, rG(e) = v}. For e /∈ r−1 e /∈ s−1 G (v) we let ros(e) = rG(e), for e ∈ r−1 G (v) we let sos(v) = sG(e), for e ∈ s−1 If E1 is finite then C∗(G) (cid:39) C∗(Gos). We also explicitly write down how to go back. Note that these two theorems are two G (v) we let sos(e) = vi if e ∈ Ei, i = 1, 2. G (v) we let ros(ei) = vi, i = 0, 1. For ways of saying the same thing. Theorem 3.4 (Out-amalgamation). Let G be a graph, and let v0, v1 ∈ G0. We now define a new graph Goa = (G0 oa, roa, soa) by oa, G1 oa = (G0 \ {v0, v1}) ∪ {v}, G0 oa = G1 \ r−1 G1 G (v0) we let roa(e) = v, for e /∈ r−1 G (v1). G (v0) we let roa(e) = rG(e). For e ∈ s−1 G (vi) For e ∈ r−1 we let sos(e) = v, for e /∈ s−1 If v1 is a finite emitter then C∗(G) ∼= C∗(Goa). G (vi) we let soa(e) = sG(e), i = 1, 2. AMPLIFIED GRAPH C∗-ALGEBRAS 7 Remark 3.5. Out-splitting and out-amalgamating are inverse operations. More specifically, if we first out-split according to some partition of s−1(v), and then out-amalgamate v0 and v1 we get back to where we started. Lemma 3.6. Let G be a graph, u ∈ G0 an infinite emitter, and v a vertex that u emits infinitely to. Fix an edge f ∈ s−1 G (v). Let E be the graph with vertex set G0, edge set E1 = G1 ∪ {f n n ∈ N}, G (v) \ {f}. We out-split according to that partition E0,E1 of s−1 and range and source maps that extend those of G and have rE(f n) = rG(f ) and sE(f n) = u. Then C∗(E) ∼= C∗(F ). Proof. There is nothing to prove if f is a loop, since then E ∼= G. In the case v is a finite emitter, we can just appeal to Corollary 3.2. Let us consider the case where v is an infinite emitter and rG(f ) (cid:54)= v. Define E1 = {f} and E0 = s−1 G (v), to obtain a graph Gos that has vertex set (G0 \ {v}) ∪ {v0, v1}. By Theorem 3.3 C∗(G) ∼= C∗(Gos). Using Corollary 3.2 on u and v1 yields a graph F with C∗(Gos) ∼= C∗(F ). We now out-amalgamate v0, v1 in F to get a graph Foa. By Theorem 3.4 C∗(F ) ∼= ∼= E. They both have the same C∗(Foa) since v1 is a finite emitter. We claim that Foa vertex set as G. Given two vertices x, y such that x (cid:54)= u, there are the same number of edges from x to y in both E and Foa, as in both cases there is the same number of edges from x to y as there is in G. For any vertex y other than rG(f ) we must have that there are the same number of edges from u to y in both Foa and E as there is in G. But if y = rG(f ) then u has infinitely many edges to y in both E and Foa. Thus Foa ∼= E. In conclusion: C∗(G) ∼= C∗(Gos) ∼= C∗(F ) ∼= C∗(Foa) ∼= C∗(E). (cid:3) The next example illustrates the different graphs used in the above proof. Example 3.7. Let G be the graph: u ∞ / v ∞ ?~~~~~~~ @@@@@@@ • • The first step in the proof is to out-split G at v. This results in the graph Gos: ? ∞ u ∞ v1 v0 / • f ∞ ? @@@@@@@ • / ? / / / ? ? 8 SØREN EILERS, EFREN RUIZ, AND ADAM P.W. SØRENSEN Then Corollary 3.2 is applied, yielding F : ∞ ? ∞ u ∞ v1 v0 f ∞ ? @@@@@@@ / • • Which we finally out-amalgamate to Foa: ∞ u ∞ v We see that Foa ∼= E. ∞ ?~~~~~~~ @@@@@@@ • • We now can now present the final version of our move. Theorem 3.8. Let α = α1α2 ··· αn be a path in a graph G. Let E be the graph with vertex set G0, edge set E1 = G1 ∪ {αm m ∈ N}, and range and source maps that extend those of G and have rE(αm) = rG(α) and sE(αm) = sG(α). If then C∗(G) ∼= C∗(E). s−1 G (sG(α1)) ∩ r−1 G (rG(α1)) = ∞, Proof. Apply Lemma 3.6 a number of times (first adding edges, and then taking away the (cid:3) unwanted ones). Corollary 3.9. If G is a graph with G0 < ∞, then C∗(G) ∼= C∗(tG). Proof. For any path α in an amplified graph, we have s−1 G (sG(α1)) ∩ r−1 G (rG(α1)) = ∞, so Theorem 3.8 proves the desired result. We may ask ourselves: How important is the requirement G (rG(α1)) = ∞, s−1 G (sG(α1)) ∩ r−1 (cid:3) in Theorem 3.8; in particular, can we replace it by simply requiring that there are infinitely many paths from sG(α) to rG(α)? It turns out we cannot. Witness: Example 3.10. Consider the graph G: ∞o x u / y ∞ / / z . / / / ? ' ' ? / / ' ' ? o / AMPLIFIED GRAPH C∗-ALGEBRAS 9 There are infinitely many paths from u to z, so one might hope that C∗(G) is isomorphic to the algebra of the graph E: ∞o x u / y ∞ / ∞ / z . However, C∗(G) has seven ideals and C∗(E) only has six. The problem in the graph above is that u is a breaking vertex for {x} in G but not in E. Assuming that no vertices are breaking in this way, more general results are available, see Corollary 6.19. 4. The tempered primitive ideal space In this section we introduce the invariant with which we shall work. To state it in sufficient generality, we need a preliminary discussion about C∗-algebras over X. Most of the facts given about C∗-algebras over X are taken from [19]. Let X be a topological space and let O(X) be the set of open subsets of X, partially ordered by set inclusion ⊆. A subset Y of X is called locally closed if Y = U \ V where U, V ∈ O(X) and V ⊆ U . The set of all locally closed subsets of X will be denoted by LC(X). The set of all connected, non-empty locally closed subsets of X will be denoted by LC(X)∗. The partially ordered set (O(X),⊆) is a complete lattice, that is, any subset S of O(X) has both an infimum(cid:86) S and a supremum(cid:87) S. More precisely, for any subset S of O(X), U = U and U = U. For a C∗-algebra A the set of closed ideals of A, partially ordered by ⊆ is a complete lattice. More precisely, for any set S of ideals, (cid:94) U∈S (cid:33)◦ (cid:32)(cid:92) (cid:92) U∈S I∈S (cid:94) I∈S (cid:95) U∈S (cid:91) U∈S (cid:95) I∈S (cid:88) I∈S I = I and I = I. Definition 4.1. Let A be a C∗-algebra. Let Prim(A) denote the primitive ideal space of A, equipped with the usual hull-kernel topology, also called the Jacobson topology. Let X be a topological space. A C∗-algebra over X is a pair (A, ψ) consisting of a C∗- algebra A and a continuous map ψ : Prim(A) → X. A C∗-algebra over X, (A, ψ), is separable if A is a separable C∗-algebra. We say that (A, ψ) is tight if ψ is a homeomorphism. We always identify O(Prim(A)) with the ideals in A using the lattice isomorphism U (cid:55)→ (cid:92) p. p∈Prim(A)\U Let (A, ψ) be a C∗-algebra over X. Then we get a map ψ∗ : O(X) → O(Prim(A)) defined by U (cid:55)→ {p ∈ Prim(A) ψ(p) ∈ U} = A[U ]. For Y = U \ V ∈ LC(X), set A[Y ] = A[U ]/A[V ]. By Lemma 2.15 of [19], A[Y ] does not depend on U and V . Remark 4.2. By Example 2.16 of [19], if Prim(A) is finite, then for every x ∈ Prim(A), {x} ∈ LC(Prim(A)) and A[{x}] is simple. Moreover, every simple sub-quotient of A is of the form A[{x}] for some x ∈ Prim(A). To shorten the notation, we set A[x] = A[{x}] for each x ∈ Prim(A). o 7 7 / 10 SØREN EILERS, EFREN RUIZ, AND ADAM P.W. SØRENSEN Definition 4.3. Let A and B be C∗-algebras over X. A homomorphism φ : A → B is X-equivariant if φ(A[U ]) ⊆ B[U ] for all U ∈ O(X). Hence, for every Y = U \ V , φ induces a homomorphism φY : A[Y ] → B[Y ]. Let C∗-alg(X) be the category whose objects are C∗-algebras over X and whose morphisms are X-equivariant homomorphisms. We can now define what we mean by ideal related K-theory. This has been known as filtrated K-theory in [14, Definition 2.1]. Definition 4.4. Let A be a C∗-algebra over some finite set X. Whenever we have open sets U ⊆ V ⊆ W ⊆ X we have nested ideals A[U ] (cid:47) A[V ] (cid:47) A[W ] (cid:47) A. Hence we get the following six-term sequence in K-theory K0(A[V \ U ]) ι∗ / K0(A[W \ U ]) π∗ / K0(A[W \ V ]) δ δ K1(A[W \ V ]) K1(A[W \ U ]) ι∗ π∗ K1(A[V \ U ]) The ideal related K-theory F KX is the collection of all the K-groups (with K0 ordered) that arise in this way together with all the bounding maps {ι∗, π∗, δ}. Let B be another C∗-algebra over X and let ψ be a self-homeomorphism of X. We say that F KX (A) ∼= F KX (B) if there exists group isomorphism αU,V∗ : K∗(A[U \ V ]) → K∗(B[U\V ]), preserving all the bounding maps and such that αU,V is an order isomorphism. If X = Prim(A), then we get all K-groups of all sub-quotients. In this case we write F K 0 instead of F KPrim(A). In the case of a simple C∗-algebra A, F K(A) collapses to (K0(A), K + If A has precisely one ideal, J say, F K(A) is just the six-term exact sequence in K-theory coming from J (cid:26) A (cid:16) A/J. 0 (A), K1(A)). 4.1. The Alexandrov Topology. Definition 4.5. Let (X,≤) be a preordered set. A subset S ⊆ X is called Alexandrov-open if S (cid:51) x ≤ y implies y ∈ S. The Alexandrov-open subsets form a topology on X called the Alexandrov topology. A subset of X is closed in the Alexandrov topology if and only if S (cid:51) x and x ≥ y imply S (cid:51) y. It is locally closed if and only if it is convex, that is, x ≤ y ≤ z and x, z ∈ S imply y ∈ S. In particular, singletons are locally closed. The specialisation preorder for the Alexandrov topology is the given preorder. Moreover, a map (X,≤) → (Y,≤) is continuous for the Alexandrov topology if and only if it is mono- tone. Thus we have identified the category of preordered sets with monotone maps with a full subcategory of the category of topological spaces. It also follows that if a topological space carries an Alexandrov topology for some preorder, then this preorder must be the specialisation preorder. In this case, we call the space an Alexandrov space or a finitely generated space. The following theorem, [19, Corollary 2.33], provides an equivalent descriptions of Alexandrov spaces. Theorem 4.6. Any finite topological space is an Alexandrov space. Thus the construction of Alexandrov topologies and specialisation preorders provides a bijection between preorders and topologies on a finite set. Remark 4.7. A useful way to represent finite partially ordered sets is via finite directed acyclic graphs. / / O O o o o o AMPLIFIED GRAPH C∗-ALGEBRAS 11 To a partial order (cid:22) on X, we associate the finite acyclic graph with vertex set X and with an arrow x ← y if and only if x ≺ y and there is no z ∈ X with x ≺ z ≺ y. We can recover the partial order from this graph by letting x (cid:22) y if and only if the graph contains a directed path x ← x1 ← ··· ← xn ← y. 4.2. Invariant for C∗-algebras over a finite topological space. Let X and Y be topological spaces and let α : X → Y be a homeomorphism. Define (cid:101)α : LC(X) → LC(Y ) by (cid:101)α(U \ V ) = α(U ) \ α(V ), where U, V ∈ O(X) with V ⊆ U . Note that since α is a homeomorphism, (cid:101)α is a well-defined bijection. (cid:40)−rank(A[x]) (cid:40) (1) IX (A) is the ordered pair (X, τA), where τA : X → Z ∪ {−∞,∞} such that (2) IX,Σ(A) is the ordered triple (X, τA, σA), where the ordered pair (X, τA) = IX (A) Definition 4.8. Let A be a C∗-algebra over X. Then if K0(A[x])+ (cid:54)= K0(A[x]) if K0(A[x])+ = K0(A[x]) rank(A[x]) and σA : LC(X) → {0, 1} such that 1, 0, otherwise σA(Y ) = if A[Y ] is unital τA(x) = Σ(A). If X = Prim(A), we set IPrim(A)(A) = Primτ (A) and IPrim(A),Σ(A) = Primτ exists a homeomorphism α : X → Y such that τB ◦ α = τA. Let B be a C∗-algebra over Y . (i) We say that IX (A) and IY (B) are isomorphic, denoted by IX (A) ∼= IY (B), if there there exists a homeomorphism α : X → Y such that τB ◦ α = τA and σB ◦(cid:101)α = σA. (ii) We say that IX,Σ(A) and IY,Σ(B) are isomorphic, denoted by IX,Σ(A) ∼= IY,Σ(B), if If B is a C∗-algebra over X and α in (i) or (ii) is idX , then we write IX (A) ≡ IX (B) and IX,Σ(A) ≡ IX,Σ(B). 5. Classification of amplified graph C∗-algebras with finitely many vertices In order to use our invariant we will need to study simple sub-quotients of C∗(tG). We focus on a nice class of hereditary subsets. Definition 5.1. If G is a graph we define a map ιG : G0 → H(G) by ιG(v) = {w ∈ G0 v ≥ w}. H ⇒ H1 ⊆ H0. I.e. H contains a largest proper hereditary subset. By definition, we have u ≥ u for any vertex u ∈ G0. Hence u ∈ ιG(u) for all vertices u. We claim that the sets ιG(v) are "special" in the lattice of hereditary sets. More specifically: Lemma 5.2. The following are equivalent for a set H ∈ H(G): (i) There is a H0 ∈ H(G) such that H0 (cid:40) H and for every H1 ∈ H(G) we have H1 (cid:40) (ii) H = ιG(u) for some u ∈ G0. Proof. First we show (i) ⇒ (ii): Suppose that u ∈ H \ H0. We claim that for all v ∈ H, we must have u ≥ v. Since if u (cid:54)≥ v for some v ∈ H, then ιG(u) (cid:40) H, but since u /∈ H0, we have ιG(u) (cid:54)⊆ H0. That is a contradiction, so we must have u ≥ v for all v ∈ H. But then ιG(u) = H. We now show (ii) ⇒ (i): Suppose H = ιG(u). Put (cid:91) H0 = ιG(w). {w∈Hw(cid:54)≥u} 12 SØREN EILERS, EFREN RUIZ, AND ADAM P.W. SØRENSEN Note that H0 (cid:40) H. If H1 (cid:40) H, then no w ∈ H1 has a path to u. Thus (cid:91) ιG(w) ⊆ (cid:91) {w∈Hw(cid:54)≥u} H1 = w∈H1 ιG(w) = H0. (cid:3) Corollary 5.3. If Φ : H(G) → H(E) is a lattice isomorphism then Φ(ιG(G0)) = ιE(E0). Another important feature of the sets ιG(u) is that we can use them to get simple sub- quotients of C∗(tG). Definition 5.4. Let G = (G0, G1, sG, rG) be a graph. Put (cid:104)u(cid:105)G = {v ∈ G0 ιG(v) = ιG(u)}. Let Gu = ((cid:104)u(cid:105)G, s−1 G ((cid:104)u(cid:105)G) ∩ r−1 G ((cid:104)u(cid:105)G), sG(cid:104)u(cid:105)G, rG(cid:104)u(cid:105)G). Observe that if H0 is the largest proper hereditary subset of ιG(u), then (cid:104)u(cid:105) = {v ∈ G0 u ≥ v and v ≥ u} = ιG(u) \ (cid:91) = ιG(u) \ = ιG(u) \ H0 v(cid:54)≥u ιG(v) (cid:91) {v∈ιG(v)v(cid:54)≥u} ιG(v) From this we easily see that if u ∈ tG 0 then C∗(tGu) is simple. We also get that it is a simple sub-quotient (up to stable isomorphism). Lemma 5.5. Let G be a graph. Let u ∈ tG subset of H = ιtG(u). We have 0 In particular (cid:104)u(cid:105) = rank(K0(IH /IH0)). IH /IH0 ⊗ K ∼= C∗(tGu) ⊗ K. and let H0 be the largest proper hereditary We are now ready to classify amplified graph algebras. Proposition 5.6. Let G and E be graphs with finitely many vertices. If Primτ (C∗(tG)) ∼= Primτ (C∗(tE)) then tG ∼= tE. Proof. Suppose Primτ (C∗(tG)) ∼= Primτ (C∗(tE)). Then there exists a homeomorphism φ : Prim(C∗(tG)) → Prim(C∗(tE)) such that τC∗(tG) ◦ φ = τC∗(tE). The homeomorphism induces a lattice isomorphism Φ : H(tG) → H(tE). 0 → tE 0 0 We want to define a bijection ψ : tG . By Corollary 5.3 there is at least one vertex u ∈ tE such that Φ(ιtG(u)) = ιtE(φ(u)). Fix a vertex u ∈ tG such that Φ(ιtG(u)) = ιtE(u). Using the fact that τC∗(tG) ◦ φ = τC∗(tE), we will prove that the sets (cid:104)u(cid:105)tG and (cid:104)u(cid:105)tE are the same size. Indeed, considering first H = ιtG(u), we note that by Lemma 5.2 it contains a largest proper hereditary subset H0. Let xu ∈ Prim(C∗(tG)) be the unique element such that C∗(tG)[xu] = I tG . Note that by Lemma 5.5 0 H /I tG H0 τC∗(tG)(xu) = rank(K0(C∗(tG)[xu])) = rank(K0(I tG H /I tG H0 )) = (cid:104)u(cid:105)tG. AMPLIFIED GRAPH C∗-ALGEBRAS 13 Let F = ιtE(u) and let F0 be the largest proper hereditary subset of F . Note that since Φ is a lattice isomorphism and since H0 is the largest proper hereditary subset of H = ιtG(u), Φ(H0) is the largest proper hereditary subset of Φ(H) = ιtE((cid:101)u). That is Φ(H0) = F0. Therefore, I tE F /I tE F0 = C∗(tE)[φ(xu)]. Computing as before, we then get (cid:104)u(cid:105)tE = τC∗(tE)(φ(xu)) = τC∗(tG)(xu) = (cid:104)u(cid:105)tG. Thus we can define a bijection ψ : tG 0 → tE 0 that satisfies Φ(ιtG(u)) = ιtE(ψ(u)). We claim that ψ can be used to construct a graph isomorphism. Since we are dealing with amplifications of transitively closed graphs, all we need to show is that if u, v ∈ tG 0 then u ≥ v if and only if ψ(u) ≥ ψ(v), and that there is a simple loop based at u if and only if there is a simple loop based at ψ(u). Suppose u (cid:54)= v. Then u ≥ v ⇐⇒ ιtG(u) ⊇ ιtG(v) ⇐⇒ Φ(ιtG(u)) ⊇ Φ(ιtG(v)) ⇐⇒ ιtE(ψ(u)) ≥ ιtE(ψ(v)) ⇐⇒ ψ(u) ≥ ψ(v). 0 So we are left with checking the claim about simple loops. If (cid:104)u(cid:105) ≥ 2 then there is some vertex w ∈ tG such that w (cid:54)= u and u ≥ w ≥ u. So since tG is transitively closed, there is a simple loop based at u. Since (cid:104)u(cid:105) = (cid:104)ψ(u)(cid:105), the same argument shows that there is a simple loop based at ψ(u). We now consider the case (cid:104)u(cid:105) = 1. In this case C∗(tGu) is stably isomorphic to ei- ther O∞ or C depending on whether or not there is a simple loop based at u. Similarly for C∗(tEψ(u)). Let xu ∈ Prim(C∗(tG)) be the unique element such that C∗(tG)[xu] = I tG H /I tG H0 . As before we see that τC∗(tE)(φ(xu)) = τC∗(tG)(xu). Hence the simple sub-quotients are either both stably isomorphic to O∞ or both stably isomorphic to C (depending on the sign of τ (··· )). Therefore there is a simple loop based at u if and only if there is one based at ψ(u) if and only if C∗(tGu) is stably isomorphic to O∞. (cid:3) We can now extend ψ to an isomorphism between tG and tE. Theorem 5.7. Let G and F be graphs with finitely many vertices. The following are equivalent. (i) tG ∼= tF . (ii) C∗(tG) ∼= C∗(tF ). (iii) C∗(G) ∼= C∗(F ). (iv) C∗(G) ⊗ K ∼= C∗(F ) ⊗ K. (v) F K(C∗(G)) ∼= F K(C∗(F )). (vi) Primτ (C∗(G)) ∼= Primτ (C∗(F )). Proof. Clearly (i) =⇒ (ii). By Corollary 3.9, (ii) =⇒ (iii). The implications (iii) =⇒ (iv), (iv) =⇒ (v), and (v) =⇒ (vi) all hold for obvious reasons. Finally the implication (vi) =⇒ (i) follows by first appealing to Corollary 3.9 to see that Primτ (C∗(tG)) ∼= Primτ (C∗(tF )), 14 SØREN EILERS, EFREN RUIZ, AND ADAM P.W. SØRENSEN and then applying Proposition 5.6. We can even do strong classification. (cid:3) Lemma 5.8. Let G and F be graphs with finitely many vertices. If we are given an isomor- phism α : Primτ (C∗(tG)) → Primτ (C∗(tF )), then we can find an isomrphism φ : C∗(tG) → C∗(tF ) such that Primτ (φ) = α. Proof. Let β : H(tG) → H(tF ) denote the lattice isomorphism induced by α. Recall from the proof of Proposition 5.6, that we can find a graph isomorphism ψ : tG → tF , such that then u ≥ v if and only if ψ(u) ≥ ψ(v). Hence the lattice isomorphism from if u, v ∈ tG H(tG) to H(tF ) induced by ψ is the same as β. Therefore the C∗-isomorphism induced by (cid:3) ψ will induce α. 0 Proposition 5.9. Let G and F be graphs with finitely many vertices. If we are given an isomorphism α : Primτ (C∗(G)) → Primτ (C∗(F )), then we can find an isomorphism φ : C∗(G) → C∗(F ) such that Primτ (φ) = α. Proof. By Corollary 3.9 we can find ∗-isomorphisms φ : C∗(G) → C∗(tG) and ψ : C∗(F ) → C∗(tF ). Let β = Primτ (ψ) ◦ α ◦ Primτ (φ−1). Then β is an isomorphism from Primτ (C∗(tG)) to Primτ (C∗(tF )). So by Lemma 5.8 we can find a ∗-isomorphism χ : C∗(tG) → C∗(tF ) such that Primτ (χ) = β. We now put λ = ψ−1 ◦ χ ◦ φ. Note that λ is an isomorphism from C∗(G) to C∗(F ) and that Primτ (λ) = Primτ (ψ−1) ◦ Primτ (χ) ◦ Primτ (φ) = Primτ (ψ−1) ◦ β ◦ Primτ (φ) = Primτ (ψ−1) ◦ Primτ (ψ) ◦ α ◦ Primτ (φ−1) ◦ Primτ (φ) = α. (cid:3) Example 5.10. The graphs given by matrices  0 ∞ 0 0 ∞ 0 ∞ 0 0 0 0 0 0 0 ∞ 0 0 ∞ 0 0 ∞ 0 0 0 0    ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ ∞ 0 0 ∞ 0 0 0 ∞ 0 0 0 0 0 were considered and proved to give stably isomorphic C∗-algebras in [14]. We now know that their algebras are in fact isomorphic. Note also that the second graph is the transitive closure of the first; indeed examples of this type inspired the results in the present paper. 6. Classification of C∗-algebras over X with free K-theory In this section, we show that IX,Σ(−) is in fact a complete invariant for an a priori much larger class of C∗-algebras, containing the class of C∗-algebras associated to amplified graphs. We will use the generalized classification results in this section to prove permanence properties in the the next section. Definition 6.1. Let C be the class of separable, nuclear, simple, purely infinite C∗-algebras A satisfying the UCT such that K1(A) = 0 and K0(A) is free. Let Cfree be the class of C∗-algebras A such that (1) Prim(A) is finite; and AMPLIFIED GRAPH C∗-ALGEBRAS 15 (2) For each x ∈ Prim(A), A[x] is unital or stable with A[x] in C or stably isomorphic (cid:76) (3) For each x ∈ Prim(A), if A[x] is unital, then there exists an isomorphism K0(A[x]) ∼= to K. Z such that [1A[x]] is sent to (1, λ). n Remark 6.2. Let A be a C∗-algebra in Cfree. Let x ∈ Prim(A). Suppose A[x] is unital and not in C. Since A[x] is stably isomorphic to K, A[x] ∼= Mn for some n ∈ N. Condition (3) implies that A[x] = C. Proposition 6.3. Let A be a C∗-algebra with Prim(A) finite. (1) A in Cfree if and only if for each x ∈ Prim(A), A[x] is in Cfree. (2) If A in Cfree, then for each Y ∈ LC(Prim(A)), A[Y ] is in Cfree. (3) If A in Cfree, then A ⊗ K is in Cfree. Proof. (1) is clear from Definition 6.1. We now prove (2). Suppose A in Cfree. Set X = Prim(A). Let Y ∈ LC(X). Then Y = U \ V with U, V ∈ O(X) such that V ⊆ U . Note that B = A[Y ] is a tight C∗-algebra over Y . Since Y is finite, Prim(B) is finite. Since LC(Y ) ⊆ LC(X), we get that for each s ∈ Prim(B), B[s] ∼= A[xs] where xs ∈ Y . Since A[xs] ∈ Cfree, B[s] ∈ Cfree. Thus, by (1), B ∈ Cfree. For (3), first note that Prim(A) is homeomorphic to Prim(A ⊗ K). So, Prim(A ⊗ K) is finite. By Corollary 2.3 of [24], every quotient and ideal of a stable C∗-algebra is stable. Thus, for every x ∈ X, we have that (A ⊗ K)[x] is stable. Thus, (3) in Definition 6.1 is vacuously true. Note that for each x ∈ X, A[x] ⊗ K ∼= (A ⊗ K)[x]. Thus, Condition (2) in Definition 6.1 holds. We have just shown that A ⊗ K ∈ Cfree. (cid:3) 6.1. Singular graph C∗-algebras are in Cfree. We will now show that Cfree actually contains the algebras we are interested in. In fact we will show that it even contains the graph algebras of a larger class of graphs than the amplified, namely the singular graphs with no breaking vertices. This will be done in a sequence of small steps. Lemma 6.4. Let G be a singular graph. If C∗(G) is simple then either G0 = {v0} and G1 = ∅ or G contains a cycle. In the case that G contains a cycle, we have that G is strongly connected. Proof. Suppose G contains no cycle. Let H = {v ∈ G0 v0 ≥ v} \ {v0}. Since there are no cycles in G, H is hereditary. As C∗(G) is simple and H (cid:54)= G0 we must have that H is empty. Thus v0 is a sink. In particular the set {v0} is non-empty and hereditary. Using again that C∗(G) is simple we deduce that G0 = {v0}. if v, w ∈ G0, then v ≥ w and w ≥ v. Hence, G is strongly connected. Lemma 6.5. Let G be a singular graph such that C∗(G) is simple. Then C∗(G) ∈ Cfree. [5] and Theorem 2.2 of [26], K0(C∗(G)) ∼=(cid:76) Proof. By Lemma 6.4 and [6], C∗(G) ∼= C or C∗(G) is purely infinite. By Corollary 3.2 of to(cid:76) G0 Z via an isomorphism that maps [1C∗(G)] (cid:3) Suppose G contains a cycle. Since every vertex in G is singular, by Corollary 2.15 of [6], (cid:3) G0 1, and K1(C∗(G)) = 0. The following definition is useful when dealing with breaking vertices. Definition 6.6. Let G be a graph and let H be a hereditary subset of G0. Set FH = {α ∈ G∗ sG(α) /∈ H, rG(α) ∈ H, and rG(αi) /∈ H for i < α } Set H G0∅ = H ∪ FH and H G1∅ = s−1 G (H) ∪ {α α ∈ FH} . 16 SØREN EILERS, EFREN RUIZ, AND ADAM P.W. SØRENSEN Define (sH G∅ )s r(α). G (H), rH G∅ (α) = −1 G (H) = (sG)s −1 G (H), sH G∅ (α) = α, (rH G∅ )s −1 G (H) = (rG)s −1 Lemma 6.7. Let G be a singular graph with finitely many vertices and no breaking vertices. If H ⊆ G0 is hereditary then FH is either infinite or empty. Proof. Suppose FH is non-empty. Then there is some vertex v /∈ H such that r(s−1(v))∩ H is non-empty. If r(s−1(v))∩H is infinite then clearly FH is infinite. So suppose r(s−1(v))∩H is a finite set. Since v is an infinite emitter and G0 is finite there is some vertex u /∈ H that v emits infinitely to. Let K = {w ∈ G0 u ≥ w}. Clearly K is hereditary. As there are no breaking vertices in G we must have that r(s−1(v)) ∩ H ⊆ K. Since neither v nor u are in H there are infinitely many paths α such that s(α) = u, r(α) ∈ H, and r(αi) /∈ H for i < α. Hence FH is infinite. (cid:3) Lemma 6.8. Let G be a singular graph with finitely many vertices and no breaking vertices. Let H be a hereditary subset of G0 such that IH is simple. Put E = (H, s−1(H), sH , rH ). Either IH Proof. By Lemma 6.7 FH is either infinite or empty. If it is empty then H G∅ ∼= E and so IH ∼= C∗(E) or IH ∼= C∗(H G∅) ∼= C∗(E). If FH is infinite then since H is finite there must be some vertex in H G∅ that receives ∼= C∗(E) ⊗ K. edges from infinitely many other vertices. So by Lemma 6.3 of [15] IH is stable. Hence ∼= IH ⊗ K ∼= C∗(E) ⊗ K. IH (cid:3) Lemma 6.9. Let G be a singular graph with finitely many vertices and no breaking vertices. Let H ⊆ G0 be non-empty and hereditary. If IH is simple then IH ∈ Cfree Proof. Put E = (H, s−1(H), sH , rH ). Since G is singular so is E. Hence Lemma 6.5 implies that C∗(E) is in Cfree. By Proposition 6.3 so is C∗(E)⊗ K. Lemma 6.8 says that IH is isomorphic to either C∗(E) or C∗(E) ⊗ K, so it is in Cfree. (cid:3) Proposition 6.10. If G is a singular graph with finitely many vertices and no breaking vertices then C∗(G) ∈ Cfree. Proof. Since G0 is finite we have that Prim(C∗(G)) is finite. By Proposition 6.3 it suffices to show that every simple sub-quotient of C∗(G) is in Cfree. A simple sub-quotient is a simple ideal in a quotient. We have shown in Lemma 6.9 that all simple ideals of singular graph algebras are in Cfree. Hence if we can show that any quotient of C∗(G) is a singular graph algebra we will be done. Suppose that H is a hereditary subset of G0. Define a graph G/H by G/H = (G0 \ ∼= C∗(G/H). Since there are no breaking H, r−1(G0 \ H), r, s). We have that C∗(G)/IH vertices in G any vertex that maps infinitely into H will also map infinitely to G0 \ H, so (cid:3) G/H is singular and has no breaking vertices. 6.2. Classification of C∗-algebras in Cfree. We are now ready to prove that IX,Σ(−) is a complete invariant for C∗-algebras in Cfree, see Theorem 6.17. Although the proof of Theorem 6.17 is quite technical, the techniques are similar to that of the proof of Theorem 3.9 of [13]. The following definition is taken from [18, Definition 3.3]. Definition 6.11. Let A and B be C∗-algebras such that A is unital. Let H1(K0(A), K0(B)) be the subgroup of K0(B) consisting of all elements x ∈ K0(B) such that there exists a group homomorphism α : K0(A) → K0(B) with α([1A]) = x. AMPLIFIED GRAPH C∗-ALGEBRAS 17 Lemma 6.12. Let A be a unital C∗-algebra. If there exists a homomorphism β : K0(A) → Z such that β([1A]) = 1, then for any C∗-algebra B, H1(K0(A), K0(B)) = K0(B). Conse- quently, if K0(A) ∼= Z ⊕ G where the isomorphism sends [1A] to (1, x), then for any C∗- algebra B, H1(K0(A), K0(B)) = K0(B). Proof. Let x ∈ K0(B). Define γ : Z → K0(B) by γ(n) = nx. Then β ◦ γ is homomorphism from K0(A) to K0(B) with (β ◦ γ)([1A]) = x. (cid:3) The following lemma is one of the key technical lemma to prove our classification result. It provides a way to get an isomorphism between two extensions given that the ideals are isomorphic and quotients are isomorphic. Lemma 6.13. For i = 1, 2, let ei : 0 → Ii → Ei → Ai → 0 be a full extension. Suppose Ii is a stable C∗-algebra satisfying the corona factorization property. Suppose there exist an isomorphism φ0 : I1 → I2 and an isomorphism φ2 : A1 → A2 such that KK (φ2) × [τe2] = [τe1] × KK (φ0). (a) If e1 and e2 are non-unital extensions, then there exists an isomorphism φ1 : A1 → (b) If e1 and e2 are unital extensions and K0(A1) ∼= Z⊕ G with [1A1] mapping to (1, x), A2 such that π2 ◦ φ1 = φ2 ◦ π1. then there exists an isomorphism φ1 : A1 → A2 such that π2 ◦ φ1 = φ2 ◦ π1. Proof. Since [τe1·φ0] = [τe1] × KK (φ0) = KK (φ2) × [τe2 ] = [τφ2·e2] in KK 1(A1, I2), we have that [τe1·φ0 ] = [τφ2·e2 ]. Suppose e1 and e2 are non-unital extensions. Then τe1·φ0 and τφ2·e2 are non-unital full extensions. Since I2 satisfies the corona factorization property, by Theorem 3.2(2) of [20], there exists a unitary u in M(I2) such that Ad(π(u)) ◦ τe1·φ0 = τφ2·e2 . Suppose e1 and e2 are unital extensions with K0(A1) ∼= Z ⊕ G with [1A1] mapping to (1, x). Then τe1·φ0 and τφ2·e2 are unital full extensions. By Theorem 2.4 and Corollary 3.8 in [18] and Lemma 6.12, there exists a unitary u in M(I2) such that Ad(π(u))◦ τe1·φ0 = τφ2·e2 . In both cases, there exists u ∈ I2 such that Ad(π(u)) ◦ τe1·φ0 = τφ2·e2 . Hence, the triple (Ad(u), Ad(u), idA1) is an isomorphism between e1·φ0 and φ2·e2. Therefore, e1 is isomorphic (cid:3) to e2. Lemma 6.14. Let A and B be tight C∗-algebras over X. Let U ∈ O(X). Suppose there exist isomorphisms φ0 : A[U ] → B[U ], φ1 : A → B, and φ0 : A/A[U ] → B/B[U ] such that 0 0 / A[U ] ι1 π1 A A/A[U ] φ0 / B[U ] φ1 φ2 ι2 B π2 / B/B[U ] 0 0 commutes. ϕ1 is an X-equivariant isomorphism if and only if ϕ0 is a U -equivariant iso- morphism and ϕ2 is an X \ U -equivariant isomorphism. Proof. Suppose ϕ1 is an X-equivariant. Let V be an open subset of U . Then V ∈ O(X) since U ∈ O(X). Hence, ϕ0(A[V ]) = ϕ1(A[V ]) = B[V ]. Hence, ϕ0 is a U -equivariant isomorphism. Suppose Y ∈ O(X \ U ). Then Y = V \ U where V ∈ O(X) and U ⊆ V . Thus, ϕ2((A/A[U ])[Y ]) = ϕ2(A[V ]/A[U ]) = π2(B[V ]) = B[V ]/B[U ] = (B/B[U ])[Y ]. Hence, φ2 is an X \ U -equivariant isomorphism. / / / / / / / / / / / / / 18 SØREN EILERS, EFREN RUIZ, AND ADAM P.W. SØRENSEN Suppose ϕ0 is a U -equivariant isomorphism and ϕ2 is an X \ U -equivariant isomorphism. Let V ∈ O(X) such that V is not a subset of U . Then ϕ2(A[V ∪U ]/A[U ]) = B[V ∪U ]/B[U ]. Let V1 ∈ O(X) such that ϕ1(A[U ]) = B[V1]. Then B[(V1 ∪ U ) \ U ] = π2(B[V1]) = B[(V ∪ U ) \ U ]. Hence, Also, note that V1 \ U = (V1 ∪ U ) \ U = (V ∪ U ) \ U = V \ U B[V1 ∩ U ] = B[V1] ∩ B[U ] = φ1(A[V ] ∩ A[U ]) = φ1(A[V ∩ U ]) = B[V ∩ U ] Suppose V ∈ O(X) such that V ⊆ U . Then V ∈ O(U ). Hence, ϕ1(A[V ]) = ϕ0(A[V ]) = (cid:3) Z be given. If x1 = 1 then there exists an Since γ is a lower triangular matrix with 1's in the diagonal, we have that det(γ) = 1. Z is an isomorphism. Note that γ((1, 1, . . . , 1))i = xi for each i. Define δ = (δij) ∈ Mm(Z) as follows: i=1 i=1 Z such that α((1, 0, . . . )) = x. Proof. Suppose m ∈ N. Define γ = (γij) ∈ Mm(Z) as follows: if i = j, 1 ≤ i ≤ m if j = i − 1, 2 ≤ i ≤ m otherwise xi − 1, 0, γi,j = So, V1 ∩ U = V ∩ U . Therefore, V1 = V . i=1 i=1 i=1 B[V ]. Hence, ϕ1 is an X-equivariant isomorphism. Hence, γ is invertible in Mm(Z). Therefore, γ : (cid:76)m Lemma 6.15. Let m ∈ N ∪ {∞}. Let x ∈(cid:76)m isomorphism α : (cid:76)m Z →(cid:76)m 1, 1, Whence, δ is invertible in Mm(Z). So δ : (cid:76)m Set α = γ ◦ δ−1. Then α : (cid:76)m Z →(cid:76)m Z →(cid:76)n isomorphism αn : (cid:76)n Z →(cid:76)∞ β : (cid:76)∞ δ((1, 1, . . . , 1)) = (1, 0, . . . , 0). −1, 0, δi,j = i=1 i=1 i=1 i=1 i=1 Z by β((y1, y2, . . . )) = (α−1 Z →(cid:76)m if i = j, 1 ≤ i ≤ m if j = i − 1, 2 ≤ i ≤ m otherwise Z → (cid:76)m i=1 Since δ is a lower triangular matrix with 1's in the diagonal, we have that det(δ) = 1. Z is an isomorphism with Z is an isomorphism such that α((1, 0, . . . , 0)) = γ(δ−1((1, 0, . . . , 0))) = γ((1, 1, . . . , 1)) = x. i=1 i=1 Suppose m = ∞. Then x = (x1, x2, . . . , xn, 0, 0 . . . ). From the finite case, there exists an Z such that αn((1, 0, . . . , 0)) = (x1, x2, . . . , xn). Define n ((y1, . . . , yn)), yn+1, yn+2, . . . ) Since αn is an isomorphism, we have that β is an isomorphism. Moreover, β(x) = (α−1 Z →(cid:76)∞ n ((x1, . . . , xn)), 0, 0, . . . ) = (1, 0, 0 . . . ) Hence, α = β−1 : (cid:76)∞ (cid:3) Lemma 6.16. Suppose A and B are C∗-algebras in Cfree. Then Primτ,Σ(A) ∼= Primτ,Σ(B) implies that A[x] ∼= B[α(x)] for all x ∈ Prim(A), where α : Prim(A) → Prim(B) is the homeomorphism implementing the isomorphism Primτ,Σ(A) ∼= Primτ,Σ(B). Z is an isomorphism such that α((1, 0, . . .)) = x. i=1 i=1 AMPLIFIED GRAPH C∗-ALGEBRAS 19 B[α(x)]. we have that A[x] and B[α(x)] are either both stable or both unital. A[x] and B[α(x)] are either both AF algebras or both purely infinite. Since σA = σB ◦(cid:101)α, Proof. Let x ∈ Prim(A). Since τA = τB ◦ α, we have that K0(A[x]) ∼= K0(B[x]) and that Suppose A[x] and B[α(x)] are AF algebra. Then A[x] ∼= K ∼= B[α(x)] or A[x] ∼= C ∼= Suppose A[x] and B[α(x)] are purely infinite simple. Since K0(A[x]) ∼= K0(B[α(x)]), by [17] we have that A[x] ⊗ K ∼= B[α(x)] ⊗ K. Thus, if A[x] and B[α(x)] are stable, A[x] ∼= B[α(x)]. such that the isomorphism sends [1A[x]] to (1, yA) with yA ∈ ⊕n ⊕n Lemma 6.15 we get automorphisms α and β on ⊕n We now consider the case where A[x] and B[α(x)] are unital. Then K0(A[x]) ∼= ⊕n Z such that the isomorphism sends [1B[α(x)]] to (1, yB) with yB ∈ ⊕n Z and K0(B[α(x)]) ∼= Z such that Z. From k=1 k=2 Z k=1 k=2 α(1, 0, . . . , 0) = (1, yA), and k=1 β(1, 0, . . . , 0) = (1, yB). k=1 For the converse, we will induct on X. Suppose X has one point. Then A1 and A2 are Let γ = β◦ α−1. Then γ is an automorphism of ⊕n Z which takes (1, yA) to (1, yB). Thus, by [17], A[x] ∼= B[α(x)]. (cid:3) Theorem 6.17. Let X be a finite topological space and let A1 and A2 be tight C∗-algebras over X. Suppose A1 and A1 are in Cfree. There exists an X-equivariant isomorphism ϕ : A1 → A2 if and only if IX,Σ(A1) ≡ IX,Σ(A2). Consequently, there exists an X-equivariant isomorphism ϕ : A1 ⊗ K → A2 ⊗ K if and only if IX (A1) ≡ IX (A2). Proof. It is clear that if there exists an X-equivariant isomorphism ϕ : A1 → A2, then IX,Σ(A1) ≡ IX,Σ(A2). simple C∗-algebras. Suppose IX,Σ(A1) ≡ IX,Σ(A2). By Lemma 6.16, A1 Suppose the theorem is true for any finite topological space with less than or equal to m − 1 elements, and that X is a finite topological space with m elements. Suppose α : IX,Σ(A1) ≡ IX,Σ(A2). Note that if X is disconnected then A1 and A2 are isomorphic to the direct sum of C∗-algebras with primitive ideal space less than m. Hence, the result follows from our inductive hypothesis. So, we may assume that X is connected. We claim that for every V ∈ O(X) with V (cid:54)= ∅, there exists an (X \ V )-equivariant isomorphism from A1[X \ V ] to A2[X \ V ]. Let V ∈ O(X) with V (cid:54)= ∅. Set Y = X \ V ∈ LC(X). Then Z ∈ LC(Y ) if and only if Z ∈ LC(X) and Z ⊆ Y . Hence, α induces αY : IY,Σ(A1[Y ]) ≡ IY,Σ(A2[Y ]). Since Y ≤ m−1, there exists a Y -equivariant isomorphism from A1[Y ] to A2[Y ]. We have just proved the claim. j be the minimal ideals of Aj. Let ai ∈ X such that Ii j , . . . , In j , I2 j = Aj[ai]. Set j. Note that U = {a1, . . . , an} ∈ O(X) such that Ij = Aj[U ] . Since X is i=1 Ii connected, A1[ai] and A2[ai] are stable C∗-algebras and hence, A1[U ] and A2[U ] are stable C∗-algebras. Ij = (cid:80)n ∼= A2. Let I1 Let for j ∈ {1, 2} : 0 → Ij → Aj → Aj/Ij → 0 ej Suppose U = 1. Then e1 and e2 are full extensions. Since Aj/Ij are tight C∗-algebras over X \ U and X \ U = m − 1, by the above claim, there exists an X \ U -equivariant isomorphism β : A1/I1 → A2/I2. Also, there exists an isomorphism θ : I1 → I2. Since KK 1(A1/I1, I2) = 0, we have that KK (β) × [τe2] = [τe1 ] × KK (θ) = 0. Since σA1 = σA2 , we have that A1 and A2 are both unital or both non-unital. Hence, by Lemma 6.13, there exists an isomorphism φ : A1 → A2 such that π2◦ φ = β◦ π1. Since β is an X \ U -equivariant isomorphism and θ is a U -equivariant isomorphism, by Lemma 6.14, φ is an X-equivariant isomorphism. 20 SØREN EILERS, EFREN RUIZ, AND ADAM P.W. SØRENSEN j = (cid:80)k−1 j +(cid:80)n Suppose U ≥ 2. Set Ik Note that there exist extensions ej,k : 0 → Ik i=1 Ii projections. i=j+1 Ii j → Aj/Ik j. Let π1,k : Ij → Ik j be the natural j → Aj/Ij → 0 such that 0 0 / I1 πj,k / Ik j Aj Aj/Ij / Aj/ Ak j / Aj/Ij 0 0 By Theorem 2.2 of [8], πj,k ◦ τej = τej,k . We claim that there exists Uk ∈ O(X) such that U ⊆ Uk and Aj[Uk \ U ] = ker(τej,k ). Note that Aj/Ik j are in Cfree. 1 → Set Ak 2. Hence, by Theorem 2.2 of [8], ψ induces isomorphisms ψX\U : A1/I1 → A2/I2 and Ak ψ{ak} : Ik → Ik j are tight C∗-algebras over Yk = (X \ U ) ∪ {ak}. Moreover, Aj/Ik j . Since Yk ≤ m− 1, there exists a Yk-equivariant isomorphism ψ : Ak 2 such that the diagram j = Aj/Ik A1/I1 τe1,k / Q(Ik 1) ψX\U A2/I2 τe2,k ψ{ak} / Q(Ik 2) is commutative. Since the vertical maps are isomorphism, ψX\U (ker(τe1,k )) = ker(τe2,k ). Let Uk ∈ O(X) such that U ⊆ Uk and A1[Uk \ U ] = ker(τe1,k ). Since ψ is a Yk-equivariant isomorphism, ψX\U (A[Uk \ U ]) = A2[Uk \ U ]. Hence, ker(τe2,k ) = A2[Uk \ U ]. Note that A1/I1 and A2/I2 are tight C∗-algebras over X \ U . Moreover, A1/I1 and A2/I2 are in Cfree. Since X \ U ≤ m − 1, there exists an X \ U -equivariant isomorphism β : A1/I1 → A2/I2. Note that there exist injective homomorphisms τ ej,k : (Aj/Ij)/ ker(τej,k ) → Q(Ik j ) such that the diagrams Aj/Ij Q(Ik j ) τej,k 7oooooooooooo τ ej,k (Aj/Ij)/ ker(τej,k ) are commutative. Since β is an X \ U -equivariant isomorphism and since ker[τej,k ] = Aj[Uk \ U ], the map βX\Uk : (A1/I1)/ ker(τe1,k ) → (A2/I2)/ ker(τe2,k ) is an isomorphism. Note that there exists an isomorphism φk : Ik 2. Since KK 1((A1/I1)/ ker(τe1,k ), Ik 2) = 0, we have that 1 → Dk [τ e2,k ◦ βX\Uk ] = [φk ◦ τ e1,k ]. Since τ e2,k ◦ βX\Uk and φk ◦ τ e1,k are essential extensions, they are full extensions since Dk is isomorphic to K or a stable purely infinite simple C∗-algebra. Since A1 and A2 are in Cfree and since σA1 = σA2, we have that either τ e1,k and τ e2,k are both non-unital extensions or they are both unital extensions. In the unital extension case, Theorem 2.4 and Corollary 3.8 of [18] and Lemma 6.12, there exists a unitary uk ∈ M(Dk) K0((A/I)/ ker(τe1,k )) ∼= (cid:76)I Z such that [1(A/I)/ ker(τe1,k )] is mapped to (1, x). Hence, by / / / / / / / / / / / / / / / / 7 AMPLIFIED GRAPH C∗-ALGEBRAS 21 such that Ad(uk) ◦ φk ◦ τ e1,k = τ e2,k ◦ βX\Uk . Since β(ker(τe1,k )) = ker(τe2,k ) and since the diagram 0 0 / ker(τe1,k ) β A/I β (A/I)/ ker(τe1,k ) βX\Uk / ker(τe2,k ) / A2/D / (A2/D)/ ker(τe2,k ) 0 0 is commutative, Ad(uk) ◦ φk ◦ π1,k ◦ τe1 = Ad(uk) ◦ φk ◦ τe1,k Define θ : I → D by θ((cid:80)n i=1 xi) =(cid:80)n Dk ∩ D(cid:96) = 0, θ is an U -equivariant isomorphism such that = τe2,k ◦ β = π2,k ◦ τe2. ◦ β i=1 Ad(uk)◦ φk(xk). Since Ik ∩ I(cid:96) = 0 for k (cid:54)= (cid:96) and π2,k ◦ θ ◦ τe1 = Ad(uk) ◦ φk ◦ π1,k ◦ τe1 = π2,k ◦ τe2 ◦ β. Hence, θ◦ τe1 = τe2 ◦ β. By Theorem 2.2 of [8], there exists an isomorphism λ : A → A2 such that 0 0 I θ D A λ / A2 A/I β / A2/D 0 0 (cid:3) By Lemma 6.14, λ is an X-equivariant isomorphism. Corollary 6.18. Let A and B be in Cfree with finitely many ideals. Then (i) A ∼= B if and only if Primτ,Σ(A) ∼= Primτ,Σ(B). (ii) A ⊗ K ∼= B ⊗ K if and only if Primτ (A) ∼= Primτ (B). τA = τB ◦ α and σA = σB ◦(cid:101)α. Proof. Set X = Prim(A) and Y = Prim(B). We first prove (i). Suppose Primτ,Σ(A) ∼= Primτ,Σ(B), then there exists a homeomorphism α : X → Y such that Define a C∗-algebra C over X by C[Y ] = B[α(Y )] for each Y ∈ LC(X). Then C is a tight C∗-algebra over X. By construction, C ∼= B and IX,Σ(C) ≡ IX,Σ(A). Therefore, by Theorem 6.17, C ∼= A. Hence, A ∼= B. (cid:3) (ii) follows from (i) since σA = 0 = σB. Corollary 6.19. Let G be a singular graph with finitely many vertices and no breaking vertices. Then C∗(G) ∼= C∗(G). Proof. By Proposition 6.10 C∗(G) is in Cfree. One can easily check that since G has no breaking vertices the ideal spaces of C∗(G) and C∗(tG) are the same. Likewise for the K- groups. Therefore Primτ,Σ(C∗(G)) ∼= Primτ,Σ(C∗(G)) and so C∗(G) ∼= C∗(G) ∼= C∗(tG). (cid:3) It turns out that if we restrict our category to unital C∗-algebras in Cfree, then I(·) is a classification functor. To do this we need the following lemma. / / / / / / / / / / / / / / / / / / / / / / / / / / 22 SØREN EILERS, EFREN RUIZ, AND ADAM P.W. SØRENSEN Lemma 6.20. Let A and B be unital C∗-algebras such that Prim(A) and Prim(B) are finite. Suppose there exists a homeomorphism α : Prim(A) → Prim(B). Then for each Y ∈ LC(Prim(A)), A[Y ] is unital if and only if B[(cid:101)α(Y )] is unital. Proof. Set X = Prim(A) and Y = Prim(B). Let U, V ∈ O(X) such that U ⊆ V and β : X → Y be the homeomorphism given by α. Set W = V \ U , Z = X \ U , S = X \ V . Suppose A[W ] is unital. Since A[W ] is unital, W and S are both open and closed subsets of Z. Moreover, Z is homeomorphic to W (cid:116) S. Since α is a homeomorphism, α(W ) and α(S) are both open and closed subsets of α(Z) = Y \ α(U ) and α(Z) is homeomorphic to α(W ) (cid:116) α(S). Thus, A[W ] = A[Z \ S] and B[α(W )] = B[α(Z) \ α(S)]. Since B[α(Z)] is unital, B[α(Z) \ α(S)] is unital. Hence, B[α(W )] is unital. (cid:3) Theorem 6.21. Let A and B be unital C∗-algebras in Cfree, with finitely many ideals. Suppose I(A) ∼= I(B). Then A ∼= B. Proof. Let α : Prim(A) → Prim(B) be a homeomorphism such that A similar argument shows that if B[β(W )] is unital, then A[W ] is unital. Let x ∈ Prim(A). By Lemma 6.20, we have that A[Y ] is unital if and only if B[(cid:101)α(Y )] is unital for all Y ∈ LC(Prim(A)). Therefore, τA = τB ◦ α. τA = τB ◦(cid:101)α. Thus, we have that Primτ,Σ(A) ∼= Primτ,Σ(B). By Corollary 6.18, A ∼= B. (cid:3) 7. Range of the invariant and permanence properties We saw in Section 5 that Primτ (·) is a classification functor for the class of graph C∗- algebras associated to amplified graphs with finitely many vertices. In fact, we showed in Proposition 5.9 that Primτ (·) is a strong classification functor. We now determine the range of Primτ (·). Let G be a finite graph. By Proposition 6.10, C∗(G) ∈ Cfree. Hence, X = Prim(C∗(G)) is finite and for each x ∈ X, τC∗(G)(x) ∈ N when K0(C∗(G)[x])+ = K0(C∗(G)) and τC∗(G)(x) = −1 when K0(C∗(G)[x])+ (cid:54)= K0(C∗(G)). We will show in this section that this is the only obstruction for the range of Primτ (·). Lemma 7.1. Let (X,(cid:22)) be a finite partially ordered set and let F be the acyclic graph representing (X,(cid:22)) described in section 4.1. Let F op be the graph obtained from F by reversing the arrows of F . Then Prim(C∗(F op)) ∼= X. Proof. It is clear that H is a hereditary subset of F op0 F op is a singular graph with no breaking vertices, we have Prim(C∗(F op)) ∼= X. if and only if H is open in X. Since (cid:3) Lemma 7.2. Let G be a finitely generated free abelian group. Then there exists a strongly connected finite graph E such that C∗(E) is a purely infinite simple C∗-algebra, E 0 = rank(G), and (K0(C∗(E)), K0(C∗(E))+) ∼= (G, G) Proof. Set m = rank(G). Define E by E0 = {v1, v2, . . . , vm}, E1 = {e(v, w) v, w ∈ {v1, v2, . . . , vm}} , AMPLIFIED GRAPH C∗-ALGEBRAS 23 [26], K0(C∗(E)) ∼=(cid:76) sF (e(v, w)) = v and rF (e(v, w)) = w. It is clear that E is a strongly connected. Since E contains a cycle, C∗(E) is purely infinite and by Corollary 3.2 of [5] and Theorem 2.2 of v∈E0 Z ∼= G. Since C∗(E) is purely infinite, (K0(C∗(E)), K0(C∗(E))+) ∼= (G, G). (cid:3) Theorem 7.3. Let X be a finite topological space and let f : X → {−1} ∪ N be a function. Then there exist a finite graph G and a homeomorphism α : Prim(C∗(G)) → X such that f ◦ α = τC∗(G). 0 Proof. By Lemma 7.1, there exists a finite graph H0 such that Prim(C∗(H0)) ∼= X and = X. Set H = H0. Let v be H 0. If f (v) > 0, then by Lemma 7.2, there exists a H0 strongly connected singular graph Ev with finitely many vertices and no breaking vertices such that rank(K0(C∗(Ev))) = f (v) and C∗(Ev) is a purely infinite simple C∗-algebra. If f (v) < 0, set Ev = {v}. v . Define E as follows E0 =(cid:83) For each v ∈ H 0, let wv be an element of E0 v∈H 0 E0 v , ∪ {e(wv, wz)n n ∈ N, ∃ edge in H from v to w} (cid:32) (cid:91) (cid:33) E1 = E1 v v∈H 0 for some finite graph G. sEEv = sEv , sE(e(wv, wz)n) = wv, rEEv = rEv , and rE(e(wv, wz)n) = wz. Then E ∼= G a lattice isomorphism. Thus, β induces a homeomorphism (cid:101)β : X → Prim(C∗(E)) such that C∗(E)[(cid:101)β(v)] ∼= C∗(Ev). Set α = (cid:101)β−1. Then α is a homeomorphism. Let x ∈ Prim(C∗(E)) Define β : O(X) → H(E) by β(U ) = ∪v∈U E0 v . By the construction of E, we have that β is and let v = α(x). Thus, Hence, f ◦ α = τC∗(E) C∗(E)[x] = C∗(E)[(cid:101)β(v)] ∼= C∗(Ev). (cid:3) As a consequence of the above theorem and our general classification result (Proposition 6.10 and Theorem 6.21), we have that every unital C∗-algebra in Cfree with finitely generated K-theory is isomorphic to C∗(G) for some finite graph G. Corollary 7.4. Let A be a unital C∗-algebra in Cfree with K0(A) finitely generated. Then there exists a finite graph G such that A ∼= C∗(G). (cid:77) Proof. Note that Prim(A) is finite. Since A is finitely generated and K0(A) ∼= K0(A[x]), x∈Prim(A) K0(A[x]) is finitely generated for all x ∈ Prim(A). Thus τA(x) ∈ {−1} ∪ N. By Theorem 7.3, there exists a finite graph G such that I(A) ∼= I(C∗(G)). Hence, by Proposition 6.10 and Theorem 6.21, A ∼= C∗(G). (cid:3) Using our general classification result and our range result, we can achieve a permanence result for extensions of graph algebras associated to amplified graphs. Corollary 7.5. Let G1 and G2 be a finite graphs. If A is a unital C∗-algebra and A fits into the following exact sequence then A ∈ Cfree. Consequently, there exists a finite graph G such that A ∼= C∗(G). 0 → C∗(G1) ⊗ K → A → C∗(G2) → 0 24 SØREN EILERS, EFREN RUIZ, AND ADAM P.W. SØRENSEN Proof. Note that by Proposition 6.10 and Proposition 6.3, C∗(G1) ⊗ K and C∗(G2) are elements of Cfree. Set X = Prim(A). Let U be an open subset of X such that A[U ] ∼= C∗(G1) ⊗ K and A[X \ U ] ∼= C∗(G2). Set Y = X \ U . Since Prim(C∗(Gi)) is finite, U and Y are finite. Hence, X is finite. Since A[U ] is a tight C∗-algebra over U and A[Y ] is a tight C∗-algebra over X, there exist homeomorphisms βU : Prim(C∗(G1) ⊗ K) → U and βY : Prim(C∗(G2)) → Y . Let x ∈ Prim(A). Then x ∈ U or x ∈ X \ U . If x ∈ U , then A[x] ∼= (C∗(G1) ⊗ K)[β−1 U (x)] and if x ∈ X \ U , then A[x] ∼= C∗(G1)[β−1 Y (x)]. Hence, by Proposition 6.3, A[x] are elements of Cfree. Thus, by Proposition 6.3, A ∈ Cfree. The last part of the statement follows from Corollary 7.4 since K0(A) ∼= K0(C∗(G1)) ⊕ (cid:3) K0(C∗(G2)) which implies K0(A) is finitely generated. 8. Acknowledgement The second author wishes to thank the Department of Mathematical Sciences at the University of Copenhagen for their support and hospitality during his visit in the spring of 2010, during which some of this research were initiated. This research was supported by the Danish National Research Foundation (DNRF) through the Centre for Symmetry and Deformation. Support was also provided by the NordForsk Research Network "Operator Algebras and Dynamics" (grant #11580). References [1] S. Arklint, R. Bentmann and T. Katsura, The range of ideal related K-theory for graph algebras, [2] T. Bates, J. Hong, I. Raeburn and W. Szyma´nski, The ideal structure of the C∗-algebras of infinite in preparation. graphs, Illinois J. Math., 46 (2002), pp. 1159–1176. [3] T. Bates and D. Pask, Flow equivalence of graph algebras, Ergodic Theory Dynam. Systems, 24 (2004), pp. 367–382. [4] K. Deicke, J. H. Hong, and W. Szyma´nski, Stable rank of graph algebras. Type I graph algebras and [5] D. Drinen and M. Tomforde, Computing K-theory and Ext for graph C∗-algebras, Illinois J. Math., their limits, Indiana Univ. Math. J., 52 (2003), pp. 963–979. 46 (2002), pp. 81–91. nonsimple graph algebras. In preparation. , The C∗-algebras of arbitrary graphs, Rocky Mountain J. Math., 35 (2005), pp. 105–135. [6] [7] S. Eilers, T. Katsura, M. Tomforde, and J. West, The ranges of K-theoretical invariants for [8] S. Eilers, T. A. Loring, and G. K. Pedersen, Morphisms of extensions of C∗-algebras: pushing [9] S. Eilers and G. Restorff, On Rørdam's classification of certain C∗-algebras with one nontrivial ideal, in Operator algebras: The Abel symposium 2004, no. 1 in Abel Symposia, Springer-Verlag, 2006, pp. 87–96. [10] S. Eilers, G. Restorff, and E. Ruiz, Classifying C∗-algebras with both finite and infinite subquo- forward the Busby invariant, Adv. Math., 147 (1999), pp. 74–109. tients. Submitted. , The ordered K-theory of a full extension. Preprint, ArXiV: 1106.1551. [11] , Strong classification of extensions of classifiable C∗-algebras. In preparation. [12] , Classification of extensions of classifiable C∗-algebras, Adv. Math., 222 (2009), pp. 2153–2172. [13] , On graph C∗-algebras with a linear ideal lattice, Bull. Malays. Math. Sci. Soc. (2), 33 (2010), [14] [15] S. Eilers and M. Tomforde, On the classification of nonsimple graph C∗-algebras, Math. Ann., 346 pp. 233–241. (2010), pp. 393–418. [16] G. A. Elliott, On the classification of inductive limits of sequences of semisimple finite-dimensional [17] E. Kirchberg, Das nicht-kommutative Michael-Auswahlprinzip und die Klassifikation nicht-einfacher algebras, J. Algebra, 38 (1976), pp. 29–44. Algebren, in C∗-algebras (Munster, 1999), Springer, Berlin, 2000, pp. 92–141. AMPLIFIED GRAPH C∗-ALGEBRAS 25 [18] H. Lin, Unitary equivalences for essential extensions of C∗-algebras, Proc. Amer. Math. Soc., 137 [19] R. Meyer and R. Nest, C∗-algebras over topological spaces: the bootstrap class, Munster J. Math., 2 (2009), pp. 3407–3420. (2009), pp. 215–252. [20] P. W. Ng, The corona factorization property, in Operator theory, operator algebras, and applications, [21] N. C. Phillips, A classification theorem for nuclear purely infinite simple C∗-algebras, Doc. Math., 5 vol. 414 of Contemp. Math., Amer. Math. Soc., Providence, RI, 2006, pp. 97–110. (2000), pp. 49–114 (electronic). Math., 598 (2006), pp. 185–210. [22] G. Restorff, Classification of Cuntz-Krieger algebras up to stable isomorphism, J. Reine Angew. [23] G. Restorff and E. Ruiz, On Rørdam's classification of certain C∗-algebras with one nontrivial ideal [24] M. Rørdam, Stable C∗-algebras, in Operator algebras and applications, vol. 38 of Adv. Stud. Pure II, Math. Scand., 101 (2007), pp. 280–292. Math., Math. Soc. Japan, Tokyo, 2004, pp. 177–199. [25] A.P.W. Sørensen, Geometric classification of simple graph algebras, in preparation. [26] M. Tomforde, The ordered K0-group of a graph C∗-algebra, C. R. Math. Acad. Sci. Soc. R. Can., 25 (2003), pp. 19–25. Department of Mathematical Sciences, University of Copenhagen, Universitetsparken 5, DK- 2100 Copenhagen, Denmark E-mail address: [email protected] Department of Mathematics, University of Hawaii, Hilo, 200 W. Kawili St., Hilo, Hawaii, 96720-4091 USA E-mail address: [email protected] Department of Mathematical Sciences, University of Copenhagen, Universitetsparken 5, DK- 2100 Copenhagen, Denmark E-mail address: [email protected]
1503.02998	3	1503	2015-11-30T15:40:09	Spectral theory of von Neumann algebra valued differential operators over non-compact manifolds	[ "math.OA", "math.DG", "math.SP" ]	We provide criteria for self-adjointness and {\tau}-Fredhomness of first and second order differential operators acting on sections of infinite dimensional bundles, whose fibers are modules of finite type over a von Neumann algebra A endowed with a trace {\tau}. We extend the Callias-type index to operators acting on sections of such bundles and show that this index is stable under compact perturbations.	math.OA	math	SPECTRAL THEORY OF VON NEUMANN ALGEBRA VALUED DIFFERENTIAL OPERATORS OVER NON-COMPACT MANIFOLDS MAXIM BRAVERMAN† AND SIMONE CECCHINI Abstract. We provide criteria for self-adjointness and τ -Fredholmness of first and second order differential operators acting on sections of infinite dimensional bundles, whose fibers are modules of finite type over a von Neumann algebra A endowed with a trace τ . We extend the Callias-type index to operators acting on sections of such bundles and show that this index is stable under compact perturbations. 1. Introduction In this paper we extend some results about the spectral properties of first and second order differential operators on complete Riemannian manifolds (see [6, 10, 16, 18, 24, 27, 29, 30]) to operators acting on sections of certain infinite dimensional bundles, called A-Hilbert bundles of finite type, cf. Section 3.5. As a main example we consider lifts of operators acting on a finite dimensional vector bundle over a complete Riemannian manifold M to a Galois cover fM of M . Let A be a von Neumann algebra with a finite, faithful, normal trace τ . Let E+ and E− be A-Hilbert bundles of finite type over a complete Riemannian manifold M , cf. Section 3.5. In particular, this means that the fibers of E± are Hilbert spaces endowed with an action of A. We consider a first order differential operator D : C∞c (M, E+) → C∞c (M, E−) (here C∞c denotes the space of smooth compactly supported sections). We also fix an A-linear bundle map V : E+ → E−. We give a criterion for self-adjointness of the generalized Schrodinger operator HV := D∗D + V . Then we show that if there exists a constant C > 0 such that V (x) > C for all x outside of a compact subset of M , then the von Neumann spectral counting function Nτ (λ; HV ) of HV is finite for all λ < C. For operators acting on finite dimensional bundles this result is obtained in [1]. One of the motivations for our study is an attempt to extend Atiayh's L2-index theory for covering manifolds, [3,28], to coverings of non-compact manifolds. As a first application to index theory we develop in Section 7 a Callias-type index theory, [2, 5, 13, 15], for operators acting on A-Hilbert bundles over non-compact manifolds. More specifically, given a bundle map F : E+ → E− we provide a criterion for τ -Fredholmness of the operator DF := D + F . The τ -Fredholmness implies that the kernel of DF and D∗F have finite τ -dimension, where D∗F is the formal adjoint of DF . Hence, we can define the τ -index indτ DF . We prove that this index is stable under compact perturbations of F . Another possible application of our analysis, which will be discussed elsewhere, is to the equivariant index theory of operators acting on a non-compact manifold with a proper action of a Lie group G. A significant progress was achieved recently in developing the equivariant index theory of operators acting on sections of a finite dimensional bundle over such manifolds, [7 -- 9,19, 21 -- 23,26]. These results find applications in the study of geometric quantization, representations of reductive groups and other areas. This paper provides the analytic tools needed to extend these results to operators acting on A-Hilbert bundles. †Supported in part by the NSF grant DMS-1005888. 1 2 MAXIM BRAVERMAN† AND SIMONE CECCHINI The paper is organized as follows. In Section 2 we formulate our main results. In Section 3 we recall the basic properties of von Neumann algebras and A-Hilbert modules. In Sections 4 and 5 we prove criteria of self-adjointness for first and second order operators respectively. In Section 6 we prove criteria for finiteness of the spectral counting function Nτ (λ; HV ) and for Fredholmness of the operator DF := D + F . In Section 7 we introduce the Callias-type τ -index and prove its stability. Acknowledgment. We are very grateful to Ognjen Milatovic for very careful reading of the manuscript and providing a lot of useful remarks and corrections. In this section we formulate the main results of the paper. 2. The main results 2.1. An operator of order one. Let A be a von Neumann algebra with a finite, faithful, normal trace τ : A → C. Let E = E+ ⊕ E− be a Z2-graded A-Hilbert bundle of finite type over a complete Riemannian manifold M , cf. Section 3.5. In particular, this means that the fibers of E± are Hilbert spaces endowed with an action of A. We denote the Riemannian metric on M by gT M . Consider a first order differential operator D : C∞c (M, E+) → C∞c (M, E−) (here C∞c denotes the space of smooth compactly supported sections). We assume that the principal symbol σ(D) of D is injective so that D is elliptic. Assumption. Throughout the paper we assume that there exists a constant c > 0 such that 0 < kσ(D)(x, ξ)k ≤ cξ, for all x ∈ M, ξ ∈ T ∗x M\{0}. (2.1) Here ξ denotes the length of ξ defined by the Riemannian metric on M , σ(D)(x, ξ) : E+ is the leading symbol of D, and kσ(D)(x, ξ)k is its operator norm. x → E−x An interesting class of examples of operators satisfying (2.1) is given by Dirac-type operators on M paired with a connection on an A-Hilbert bundle, cf. [28, Section 7.4]. 2.2. Self-adjointness of a first order operator. Let dµ be a smooth measure on M . We don't assume that dµ is the measure defined by the Riemannian metric gT M . Let L2(M, E±) denote the space of square-integrable sections of E±. We also set E := E+ ⊕ E−. Then L2(M, E) = L2(M, E+) ⊕ L2(M, E−). Let D∗ denote the formal adjoint of D. Consider the operator D 0 (cid:19) . D := (cid:18) 0 D∗ (2.2) Then D is an unbounded operator on L2(M, E). Our first result is the following extension of [18, Th. 1.17]: Theorem 2.3. Suppose that D : C∞c (M, E+) → C∞c (M, E−) satisfies (2.1) and the Riemannian metric gT M is complete. Then D is essentially self-adjoint with initial domain C∞c (M, E) = C∞c (M, E+) ⊕ C∞c (M, E−). The proof is given is Section 4. VON NEUMANN ALGEBRA VALUED SCHR ODINGER OPERATORS 3 2.4. A Schrodinger-type operator. Consider the Schrodinger-type operator HV = D∗D + V, (2.3) where V (x) : E+ x → E+ x is an A-linear self-adjoint bundle endomorphism. We view HV as an unbounded operator on L2(M, E+) with initial domain C∞c (M, E+) and give a sufficient condition for self-adjointness of this operator. For simplicity we assume that the potential V is a measurable section which belongs to L∞loc.1 Our next result is the following extension of [10, Th. 2.7] to A-Hilbert bundles: Theorem 2.5. Suppose there exists a function q : M → R such that (1) V (x) ≥ −q(x) Id for all x ∈ M ; (2) q ≥ 1 and q−1/2 is globally Lipschitz, i.e. there exists a constant L > 0 such that, for every x1, x2 ∈ M , q−1/2(x1) − q−1/2(x2) ≤ Ld(x1, x2), (2.4) where d is the distance induced by the metric gT M ; (3) the metric g := q−1gT M on M is complete. Then HV is essentially self-adjoint on C∞c (M, E+). The proof is given in Section 5. Remark 2.6. We say that a curve γ : [a,∞) → M goes to infinity if for any compact set K ⊂ M there exists tK > 0 such that γ(t) 6∈ K, for all t ≥ tK. Condition (3) of the theorem is equivalent to the condition that the integral Rγ Corollary 2.7. If V (x) is bounded below, i.e. there exists a constant b > 0 such that V (x) ≥ −b, then the operator HV is essentially self-adjoint. ds√q = ∞ for every going to infinity curve γ. In the rest of this paper we denote by HV both the operator and its self-adjoint closure. 2.8. The spectral counting function of a Schrodinger-type operator. The trace τ : A → C on A extends to a (possibly infinite) trace Trτ on the space of bounded A-linear operators acting on L2(M, E±). For each λ ∈ R let Pλ denote the orthogonal projection onto the spectral subspace of the operator HV corresponding to the ray (−∞, λ] and define the spectral counting function Nτ (λ; HV ) := Trτ Pλ. Theorem 2.9. Suppose there exist a compact K ⊂ M and a constant C such that Then V (x) ≥ C, for all x 6∈ K. Nτ (λ; HV ) < ∞, for all λ < C. (2.5) (2.6) (2.7) Remark 2.10. Inequality (2.6) implies that V (x) is bounded below and, hence, the operator HV is self-adjoint by Corollary 2.7. 1Much more general potentials were considered in [10]. It would be interesting to extend the results we present in this paper to the type of potentials considered in [10]. 4 MAXIM BRAVERMAN† AND SIMONE CECCHINI 2.11. τ -Fredholmness. We first recall the notion of Fredholmness in the von Neumann setting, cf. [11, Section 3]. Let H1 and H2 be A-Hilbert spaces and T : Dom(T ) ⊂ H1 → H2 be a closed A-operator. Definition 2.12. We say that the operator T is τ -Fredholm if the following two conditions are satisfied: (i) Ker T has finite τ -dimension; (ii) there exists an A-Hilbert subspace L of H2 such that L ⊂ im T and dimτ L⊥ < ∞. Remark 2.13. If H1 and H2 are Hilbert spaces over C, an operator T : H1 → H2 is called Fredholm if its kernel and cokernel are finite dimensional. Equivalently, this means that the kernels of T and T ∗ are finite dimensional and the image of T is closed. The image of a τ - Fredholm operator does not need to be closed. Moreover, the image of T is not in general an A-Hilbert space and the τ -dimension of the cokernel of T is not defined. Because of this we need to replace the condition of finite-dimensionality of the cokernel by Condition (ii) of Definition 2.12. Consider the operator T = (cid:18) 0 T ∗ 0 (cid:19) : H1 ⊕ H2 −→ H1 ⊕ H2. T Lemma 2.14. The operator T is τ -Fredholm if and only if there exists λ > 0 such that Nτ (λ;T 2) < ∞. The lemma is proven in Section 6.8. Remark 2.15. The inequality (2.9) is equivalent to Nτ (λ; T ∗T ) < ∞, and Nτ (λ; T T ∗) < ∞. (2.8) (2.9) Remark 2.16. The condition (2.9) is often taken as a definition of τ -Fredholmness, cf. example [20, §2.1.1]. 2.17. τ -Fredholmness of a Callias-type operator. Let D be as in Section 2.1 and let F : E+ → E− be an A-linear bundle map. We set for 0 (cid:19) F := (cid:18) 0 F ∗ F and consider the operator DF := D + F = (cid:18) 0 D + F D∗ + F ∗ 0 (cid:19) . (2.10) This operator satisfies the conditions of Theorem 2.3 and, hence, is essentially self-adjoint. Its self-adjoint closure is denoted by the same symbol DF . Let {D,F} := D ◦ F + F ◦ D = (cid:18)D∗F + F ∗D 0 DF ∗ + F D∗(cid:19) 0 denote the anti-commutator of D and F. We also set DF := D + F . Definition 2.18. We say that the operators DF and DF are of Callias-type if (1) the leading symbol σ(D) anti-commutes with F so that the anti-commutator {D,F} is an A-linear bundle endomorphism of E = E+ ⊕ E−; VON NEUMANN ALGEBRA VALUED SCHR ODINGER OPERATORS 5 (2) there exist a compact set K ⊂ M and ǫ > 0 such that F 2(x) ≥ k{D,F} (x)k + ǫ for all x ∈ M\K. Here k{D,F} (x)k denotes the norm of the A-linear map {D,F} (x) : Ex → Ex. In Section 6 we get the following corollary of Theorem 2.9: Theorem 2.19. A Callias-type operator DF is τ -Fredholm. (2.11) 2.20. A Callias-type index. Suppose now that DF is an operator of Callias-type. By The- orem 2.19, dimτ Ker DF and dimτ Ker D∗F are finite. Hence, we can define the τ -index of DF by indτ DF := dimτ Ker DF − dimτ Ker D∗F . (2.12) For operators acting on sections of finite-dimensional bundles (i.e. when the von Neumann algebra A = C) this index was introduced by C. Callias [15] and was extended and studied in many papers including [2,5,13]. We refer to (2.12) as the Callias-type τ -index of the pair (D, F ). In Section 7 we prove the following stability property of the Callias-type τ -index: Theorem 2.21. Let F0 and F1 be two A-linear bundle maps E+ → E− satisfying conditions (1) and (2) of Definition 2.18. If there exists a compact set K ⊂ M such that F0(x) = F1(x) for all x 6∈ K, then indτ DF0 = indτ DF1 . (2.13) 3. A-Hilbert bundles In this section we recall some basic properties of von Neumann algebras, cf. [4], and recall the definitions of A-Hilbert space and A-Hilbert bundle, cf. [28]. 3.1. A Hilbert space completion. Let A be a von Neumann algebra endowed with a finite, faithful, normal trace τ : A → C, cf. [4, III.2.5]. We normalize the trace so that τ (Id) = 1. We define on A the inner product ha, biτ := τ (ab∗), a, b ∈ A, (3.1) and denote by l2(A) the Hilbert space completion of A with respect to this inner product. Notice that l2(A) is an Hilbert space endowed with an A-module structure. Example 3.2. Suppose Γ is a discrete group and denote by l2(Γ) the Hilbert space of complex valued square summable functions on Γ. The right regular representation of Γ is the unitary representation R : Γ → B(l2(Γ)) defined by (Rgu)(h) = u(hg), h, g ∈ Γ; u ∈ l2(Γ). The smallest subalgebra of B(l2(Γ)) which is weakly closed and contains all the operators Rg (g ∈ Γ) is called the von Neumann algebra of Γ and is denoted by N Γ. On N Γ we have the canonical faithful positive trace τ defined by: τ (f ) = hf (δe), δeil2(Γ), f ∈ N Γ, (3.2) where δe ∈ l2(Γ) is by definition the characteristic function of the unit element. to l2(Γ), cf. [28, Example 7.11]. The completion l2(N Γ) of N Γ with respect to this inner product is canonically isomorphic 6 MAXIM BRAVERMAN† AND SIMONE CECCHINI 3.3. A-Hilbert spaces. Let A be a von Neumnn algebra endowed with a trace τ satisfying the same properties as in section 3.1. Suppose A acts on a Hilbert space H. The action of A on H is said to be compatible with the inner product if (3.3) Notice, that the action of N Γ on l2(Γ) in Example 3.2 is compatible with the inner product on l2(Γ). a ∈ A; x, y ∈ H. hxa∗, yi = hx, yai, Definition 3.4. An A-Hilbert space is a Hilbert space H endowed with a compatible A-module structure such that there exists a separable Hilbert space VH and an isometric A-linear embed- ding We say that H is an A-Hilbert space of finite type if VH can be chosen finite dimensional. i : H ֒→ l2(A) ⊗ VH . Notice, that since the action of A is compatible with the scalar product, the orthogonal complement i(H)⊥ to i(H) in l2(A) ⊗ VH is also an A-module. In other words H is a projective A-module, cf. [4]. 3.5. A-Hilbert bundles. Definition 3.6. An A-Hilbert bundle E on a manifold M is a locally trivial bundle of A-Hilbert spaces, the transition functions being A-Hilbert space isomorphisms. If the fibers are A-Hilbert spaces of finite type, the bundle is called an A-Hilbert bundle of finite type. We denote by C∞c (M, E) the space of smooth compactly supported sections of E. If M is endowed with a smooth measure dµ we define the L2-scalar product s1, s2 ∈ C∞c (M, E), (s1, s2)2 := ZM hs1(x), s2(x)i dµ, and by L2(M, E) the completion of C∞c (M, E) with respect to this scalar product. We set s(x) := hs(x), s(x)i1/2 and denote by ksk := (cid:18)ZM s2 dµ(cid:19)1/2 the L2-norm of s. M with deck transformation group Γ. We let Γ act on the Hilbert space l2(Γ) by left and right convolution. Example 3.7. Let M be a smooth compact manifold and π : fM → M be the Galois cover of Let A = N Γ denote the von Neumann algebra of Γ. The right action of Γ extends to a right action of A on l2(Γ) commuting with the left Γ-action. Therefore, E := fM ×Γ l2(Γ) is an Notice that the space L2(M, E) coincides with the space L2(fM ) of square-integrable functions on fM . 4. Self-adjointness of first order differential operators A-Hilbert bundle of finite type on M . In this section we prove Theorem 2.3. Since the operator D is formally self-adjoint, to show that it is essentially self-adjoint we need to prove that its maximal and minimal extensions coincide. Since the domain Dom(Dmin) of the minimal extension is closed in the operator norm of D, it is enough to show that for every s ∈ Dom(Dmax) there exists a sequence {sk} in Dom(Dmin) such that sk = s, lim k→∞ lim k→∞Dsk = Ds, (4.1) VON NEUMANN ALGEBRA VALUED SCHR ODINGER OPERATORS 7 where the limits are in L2-norm topology. 4.1. The minimal extension. Recall that the Sobolev spaces of sections of an A-vector bundle In particular, if Ω ⊂ M is an open set with were defined by Miscenko and Fomenko [17]. compact closure, the Sobolev space H s 0(Ω, E) is defined as the closure in Sobolev norm of the space C∞0 (Ω, E) of smooth sections, having compact support in Ω. If T is a differential operator of order k, then T extends to a bounded operator T : H k 0 (Ω, E) → L2(M, E). Recall that the minimal domain Dom(Tmin) is the closure of C∞c (M, E) with respect to the graph norm of T . We conclude that H k 0 (Ω, E) ⊂ Dom(Tmin) (4.2) for any open set Ω whose closure is compact. 4.2. The maximal extension. Recall that the domain Dom(Tmax) of the maximal extension consists of all sections s ∈ L2(M, E) such that T s ∈ L2(M, E), where T s is understood in distributional sense. loc(M, E). It follows now from (4.2), Since D is a first order elliptic operator, Dom(Dmax) ⊂ H 1 that for any Lipschitz function φ ∈ C 0 c (M ) and any s ∈ Dom(Dmax) φ s ∈ Dom(Dmin). 4.3. Proof of Theorem 2.3. By Lemma 8.9 of [10] there exists a sequence {φk} of Lipschitz functions with compact support on M such that (i) 0 ≤ φk ≤ 1; (ii) ; 1 k dφk ≤ lim k→∞ (iii) φk(x) = 1, for al x ∈ M. (4.4) For any s ∈ Dom(Dmax) set sk = φks. Then sk ∈ Dom(Dmin) by (4.3) and limk→∞ sk = s. It remains to show that Dsk converges to Ds in the L2-norm. We have: (4.3) (4.5) (cid:3) Notice that is a bundle map. From (4.4) and (2.1) we conclude that Dsk = φkDs + [D, φk] s. [D, φk](x) = −iσ(D)(cid:0)x, dφk(x)(cid:1) (cid:13)(cid:13) [D, φk] s(cid:13)(cid:13) = (cid:13)(cid:13) σ(D)(x, dφk) s(cid:13)(cid:13) ≤ c k · ksk. Hence, limk→∞[D, φk] = 0. Since φkDs → Ds in L2-norm we obtain k→∞Dsk = Ds. The essential self-adjointness of the operator D is proved. lim 8 MAXIM BRAVERMAN† AND SIMONE CECCHINI 5. Self-adjointness of a Schrodinger-type operator In this section we prove Theorem 2.5. We use the notation of Section 2.4. We denote by HV,0 the restriction of the operator (2.3) to the space C∞c (M, E+) of smooth compactly supported sections of E+. Let H∗V,0 denote the operator adjoint to HV,0 and let Dom(H∗V,0) denote its domain. Since the operator HV,0 is symmetric, to show that its closure is self-adjoint it is enough to prove that (5.1) To prove (5.1) we need some information about the behavior of sections from Dom(H∗V,0) at (HV s1, s2) = (s1, HV s2), s1, s2 ∈ Dom(H∗V,0). infinity. This information is provided by the following Proposition 5.1. Suppose q : M → R is a function satisfying the assumptions of Theorem 2.5. If s ∈ Dom(H∗V,0), then q−1/2Ds is square integrable and kq−1/2Dsk2 ≤ 2(cid:16) (1 + 2L2)ksk2 + kskkHV sk(cid:17), (5.2) where L is the Lipschitz constant introduced in (2.4). Remark 5.2. For the Schrodinger operator on scalar valued functions on Rn an analogous lemma was established in [27]. The proof was adapted in [24, 25] to the case of a Riemannian manifold and to differential forms in [6]. The case of a general operator D and a singular potential V was considered in [10]. 5.3. Regularity of sections from Dom(H∗V,0). The theory of elliptic (pseudo)-differential operators on A-Hilbert bundles of finite type was developed in [14] (see also [17] for a similar theory for bundles of finitely generated A-modules). In particular, the Sobolev spaces of sections of such bundles are introduced in these papers and it is shown that any s ∈ Dom(H∗V,0) belongs to the Sobolev space H 2 loc. Hence, Ds ∈ L2 loc(M, E−), V s ∈ L2 loc(M, E+), for any s ∈ Dom(H∗V,0). (5.3) The new information provided by Proposition 5.1 is about the rate of decay of Ds at infinity. Remark 5.4. The equation (5.3) is the only place in the proof of Theorem 2.5 where we use the fact that the fibers of E± are modules of finite type. The rest of the proof of Theorem 2.5 follows the lines of [10, §9] with almost no changes. It is even simpler, since in [10] much more singular potentials are considered. Set bD = −iσ(D). Then Note that cD∗ = −(bD)∗. 5.5. Proof of Proposition 5.1. Let ψ be a Lipschitz function with compact support such that a 0 ≤ ψ ≤ q−1/2 ≤ 1. Set D(φs) = bD(dφ)s + φDs. Using (5.3) we obtain C = ess sup x∈M kbD(dψ)k. kψDsk2 = (D∗(ψ2Ds), s) = (ψ2D∗Ds, s) + 2(ψcD∗(dψ)Ds, s) = Re(ψ2D∗Ds, s) + 2Re(ψcD∗(dψ)Ds, s) ≤ Re(ψ2D∗Ds, s) + 2CkψDskksk = Re(ψ2HV s, s) − (ψ2V s, s) + 2CψDskksk. (5.4) VON NEUMANN ALGEBRA VALUED SCHR ODINGER OPERATORS 9 Since V ≥ −q Id, q ≥ 1 and qψ2 ≤ 1, we have (ψ2V s, s) = (V ψs, ψs) ≥ −(qψs, ψs) ≥ −ksk2. 2 a2 + 1 Using the inequality ab ≤ 1 2 b2 we obtain 1 2 kψDsk2 + 2C 2 ksk2. Combining (5.4),(5.5), and (5.6) and using that ψ2 ≤ q−1 ≤ 1 we get 2CkψDskksk ≤ and kψDsk2 ≤ kHV skksk + 1 2 kψDsk2 + (1 + 2C)ksk2, kψDsk2 ≤ 2(cid:16) (1 + 2C 2)ksk2 + kskkHV sk(cid:17). To prove (5.2) we now make a special choice of the function ψ. Let φk be as in (4.4) and set ψk := φk · q−1/2. Then 0 ≤ ψk ≤ q−1/2, and dψk ≤ dφk · q−1/2 + φkdq−1/2. k + L. Since ψk(x) → q−1/2(x) as k → ∞, the dominated convergence Therefore, dψk ≤ 1 theorem applied to (5.7) with ψ = ψk immediately implies (5.2). 5.6. Proof of Theorem 2.5. Let s1, s2 ∈ Dom(H∗V,0). Then (5.5) (5.6) (5.7) (cid:3) (φs1, D∗Ds2) = (D(φs1), Ds2) = (bD(dφ)s1, Ds2) + (φDs1, Ds2). (D∗Ds1, φs2) = (Ds1, bD(dφ)s2) + (φDs1, Ds2) (φs1, HV s2) − (HV s1, φs2) = (bD(dφ)s1, Ds2) − (Ds1, bD(dφ)s2). ess sup x∈M bD(dφ) ≤ c ess sup x∈M dφ(x), Similarly, Hence, By (2.1), Therefore, (φs1, HV s2) − (HV s1, φs2) ≤ c ess sup x∈M (cid:16)dφq1/2(cid:17) ·(cid:16)ks1kkq−1/2Ds2k + ks2kkq−1/2Ds1k(cid:17) . (5.8) Consider a metric g := q−1gT M . By condition (iii) of Theorem 2.5 this metric is complete. By Lemma 8.9 of [10] there exists a sequence {φk} of Lipschitz functions such that (1) 0 ≤ φk ≤ 1 and dφkg ≤ 1 k ; (2) limk→∞ φk(x) = 1, for all x ∈ M . Since dφkg = q1/2dφk, we conclude ess sup x∈M (dφkq1/2(x)) ≤ 1 k . Using (5.8), we obtain as k → ∞, (φku, HV v) − (HV u, φkv) ≤ where the convergence to 0 of the RHS of this inequality follows from Proposition 5.1. On the other side, by the dominated convergence theorem we have k(cid:16)ks1kkq−1/2Ds2k + kq−1/2Ds1kks2k(cid:17) → 0, c (φks1, HV s2) − (HV s1, φks2) −→ (s1, HV s2) − (HV s1, s2) as k → ∞. Thus (HV s1, s2) = (s1, HV s2), for all s1, s2 ∈ Dom(H∗V,0). Therefore, HV is essentially self-adjoint. (cid:3) 10 MAXIM BRAVERMAN† AND SIMONE CECCHINI In this section we prove Theorems 2.9 and 2.19. 6. Fredholmness 6.1. Variational principle. We make use of a variational principle in the von Neumann setting, stated in the following: Lemma 6.2. Let H be an A-Hilbert space and T be a self-adjoint operator commuting with the action of A. Then, for every λ ∈ R, Nτ (λ; T ) = sup L dimτ L, (6.1) where L varies among the A-Hilbert subspaces of H with L ⊂ Dom(T ) and satisfying: (T u, u) ≤ λ (u, u), u ∈ L. This lemma is well-known. For the proof we refer to [31, Lemma 2.4] where the case A = N Γ is treated (the case of a general finite von Neumann algebra A follows with minor modifications). 6.3. Restriction to a compact subset. Note, first, that for any A-linear self-adjoint operator T and any l, λ ∈ R we have: Nτ (λ + l; T + l) = Nτ (λ; T ). (6.2) Therefore, by replacing V (x) with V (x) + l in Theorem 2.9, we can assume that C, λ > 0 in (2.6) and (2.7) and V (x) ≥ 0 for all x ∈ M . Let the compact set K ⊂ M be as in (2.6). For any λ > 0 consider the set Then Mλ := {x ∈ M : V (x) ≥ λ}. is an open set. We denote its closure by ¯Ωλ. We now assume that 0 < λ < C and choose λ1 such that 0 < λ < λ1 < C. Then Ωλ := M\Mλ ¯Ωλ ⊂ Ωλ1, ¯Ωλ1 ⊂ ΩC , ¯ΩC ⊂ K. Let φ : M → [0, 1] be a smooth function such that φΩλ1 ≡ 1, φM\ΩC ≡ 0. Define the A-linear restriction map ρ : L2(M, E) → L2(ΩC , EΩC ) by the formula ρ(s) := φs. (6.3) Lemma 6.4. Let L ⊂ Dom(HV ) be an A-Hilbert subspace of L2(M, E) satisfying (HV s, s) ≤ λ (s, s), (6.4) Then ρ is injective when restricted to L and ρ(L) is a closed A-invariant subspace of L2(Ωλ, EΩλ ). Proof. The potential V satisfies the inequality V (x) ≥ 0 for any x ∈ M . Hence, the operator HV satisfies the conditions of Theorem 2.5 with q = const. It follows from Proposition 5.1 that for any s ∈ Dom(HV ) we have Ds ∈ L2(M, E). Hence, s ∈ L. (HV s, s) = kDsk2 + (V s, s) ≥ (V s, s). (6.5) In particular, (V s, s) < ∞. For any s ∈ L using (6.5) we obtain λksk2 ≥ (HV s, s) ≥ (V s, s) ≥ λ1 ZM\Ωλ1 s(x)2 dµ(x) VON NEUMANN ALGEBRA VALUED SCHR ODINGER OPERATORS 11 and Hence, Therefore, λ ZΩλ1 s(x)2 dµ(x) ≥ (λ1 − λ) ZM\Ωλ1 kρ(s)k2 ≥ ZΩλ1 ksk2 =ZΩλ1 λ1 − λ λ s(x)2 dµ(x) ≥ s(x)2 dµ(x) + ZM\Ωλ1 ≤ kρ(s)k2 + λ λ1 − λ kρ(s)k2 =(cid:18) λ1 s(x)2 dµ(x). s(x)2 dµ(x). ZM\Ωλ1 s(x)2 dµ(x) λ1 − λ(cid:19) kρ(s)k2. This inequality together with the fact that ρ is a bounded A-equivariant map proves the lemma. (cid:3) 6.5. Extension to a closed manifold. Choose a closed manifold cM containing ΩC as an open subset. Let bE = bE+ ⊕ bE− be a graded A-Hilbert bundle of finite type over cM extending EΩC . Let bD : C∞(bE+) → C∞(bE−) and bV : bE+ → bE+ be a first order elliptic differential operator and a positive bundle map which agree with D and V on ΩC. Set H bV := bD∗bD + bV . Then the restrictions of HV and H bV to ΩC coincide. We view ρ(L) ⊂ L2(ΩC , E+ΩC ) as a subspace of L2(cM , bE+). Lemma 6.6. Under the assumptions of Lemma 6.4, there exists a constant R > 0 such that for any u ∈ ρ(L) we have (6.6) (6.7) (6.8) (6.9) (cid:3) Proof. Set (cid:0) H bV u, u(cid:1) ≤ R (u, u). x∈M kσ(D)(x, dφ(x))k. a := max Notice that the maximum exists, since φ has compact support. For u ∈ ρ(L) there exists s ∈ L such that u = ρ(s) = φs. Then Also, since V (x) > 0 for all x ∈ M we conclude kbDuk2 = kD(φs)k2 = (cid:0)kφDsk + k[D, φ]sk(cid:1)2 ≤ (cid:0)kDsk + aksk(cid:1)2 ≤ 2kDsk2 + 2a2 ksk2. (bV u, u) = (φ2V s, s) ≤ (V s, s). By using (6.8) and (6.9) we obtain (cid:0) H bV u, u(cid:1) = kbDuk2 + (bV u, u) ≤ 2kDsk2 + 2a2ksk2 + (V s, s) ≤ 2(cid:0)kDsk2 + (V s, s)(cid:1) + 2a2ksk2 = 2 (HV s, s) + 2a2 ksk2. Using (6.4) and (6.6) we now conclude (cid:0) H bV u, u(cid:1) ≤ 2(λ + a2)ksk2 ≤ 2(λ + a2) (cid:18) λ1 λ1 − λ(cid:19) kuk2. Hence, (6.7) holds with R = 2(λ + a2) (cid:16) λ1 λ1−λ(cid:17). 12 MAXIM BRAVERMAN† AND SIMONE CECCHINI 6.7. Proof of Theorem 2.9. Since bV ≥ 0, we have Nτ (λ; H bV ) ≤ Nτ (λ; bD∗bD), It is shown in [14, §2.4] that the spectral counting function Nτ ( λ; bD∗bD) is finite. Hence, it follows from (6.10) and Lemmas 6.2 and 6.6 that λ ∈ R. (6.10) From Lemma 6.4 and the open mapping theorem we deduce that the map ρL : L → ρ(L) is an isomorphism of Hilbert A-spaces. By [20, Theorem 1.12 (2)] dimτ ρ(L) ≤ Nτ (R; H bV ) < ∞. Hence, by Lemma 6.2 we get Theorem 2.9 is proven. dimτ L = dimτ ρ(L) ≤ Nτ (R; H bV ). Nτ (λ; HV ) ≤ Nτ (R; H bV ) < ∞. 6.8. Proof of Lemma 2.14. Consider the bounded operator (6.11) (cid:3) (6.12) where T is the operator defined in (2.8). Observe that the operator Φ(T ) is τ -Fredholm if and only if T is τ -Fredholm, and Φ(T ) := T (I + T 2)− 1 2 , Nτ (λ;T 2) = Nτ(cid:18) λ 1 + λ ; Φ(T )2(cid:19) . Thus Lemma 2.14 is a direct consequence of the following: Lemma 6.9. Let S be a bounded self-adjoint A-linear operator on an A-Hilbert space H. Then S is τ -Fredholm if and only if there exists λ > 0 such that Nτ (λ; S2) < ∞. (6.13) Proof. Suppose Nτ (λ; S2) < ∞ for some λ > 0. We need to show that S is τ -Fredholm in the sense of Definition 2.12. Since the function Nτ (·; S2) is nondecreasing, we have: dimτ Ker S = Nτ (0; S2) ≤ Nτ (λ; S2) < ∞. Thus the condition (i) of Definition 2.12 is satisfied. Set L := im(I − Pλ(S2)), where Pλ(S2) denotes the orthogonal projection onto the spectral subspace of the operator S2 corresponding to [0, λ]. Then L is A-invariant, L ⊆ im S2 ⊆ im S and Hence the condition (ii) of Definition 2.12 is also satisfied. dimτ L⊥ = dimτ (Pλ(S)) = Nτ (λ; S2) < ∞. Suppose now that S is τ -Fredholm and let L be an A-Hilbert subspace of H such that L ⊆ im S and dimτ L⊥ < ∞. The map is one-to-one. Set S1 := S(Ker S)⊥ : (Ker S)⊥ → H (6.14) Then L1 is a closed A-invariant subspace of H and S : L1 → L is a bijection. It follows from the Open Mapping Theorem that there exists ǫ > 0 such that L1 := S−1 1 (L) ⊂ (Ker S)⊥. k Suk > ǫkuk, for any u ∈ L1. (6.15) VON NEUMANN ALGEBRA VALUED SCHR ODINGER OPERATORS 13 We finish the proof of the lemma by showing that any λ < ǫ2 satisfies (6.13). Recall that Pλ(S2) denotes the orthogonal projection on the spectral subspace of S2 corresponding to the interval [0, λ]. Thus (6.16) From (6.15) and (6.16) we now conclude that im Pλ(S2) ∩ L1 = {0}. Hence, it follows from for any u ∈ im Pλ(S2). k Suk ≤ √λ kuk < ǫkuk, [20, Theorem 1.12] that Nτ (λ; S2) ≤ dimτ L⊥1 . To finish the proof of the lemma it is now enough to show that dimτ L⊥1 < ∞. Let L2 be the orthogonal complement of L1 in (Ker S)⊥. Then dimτ L⊥1 = dimτ L2 + dimτ Ker S. (6.17) (6.18) Since S is τ -Fredholm, dimτ Ker S is finite and it suffices to prove that dimτ L2 is finite. Notice that, by (6.15), S(L2) is a closed subspace of H and S : L2 → S(L2) is a topological isomorphism. Hence, dimτ S(L2) = dimτ L2, (6.19) cf. [20, Lemma 1.13]. Since the map (6.14) is one-to-one, we conclude that S(L2) ∩ L = S1(L2) ∩ L = {0}. Let PL be the orthogonal projection onto the subspace L. Then I − PL : S(L2) → L⊥ is a one-to one map. Therefore, by [20, Theorem 1.12(2)], dimτ S(L2) ≤ dimτ L⊥ < ∞. From (6.17), (6.18), and (6.19) we now conclude that Nτ (λ; S2) < ∞. 6.10. Proof of Theorem 2.19. We are now ready to prove τ -Fredholmness of a Callias-type operator D + F . The operator (2.10) satisfies the conditions of Theorem 2.3 and, hence, is essentially self-adjoint. Hence (cid:3) Nτ (λ;DF ) = Nτ (λ2;D2 F ). (6.20) Moreover (6.21) where V = {D,F} + F 2. By the Callias condition (2.11), V (x) > ǫ for all x ∈ M\K. Hence, it follows from Theorem 2.9 that Nτ (ǫ;D2 F ) is finite. Theorem 2.19 follows now from Lemma 2.14. D2 F = D2 + {D,F} + F 2 = D2 + V, (cid:3) 7. Stability of the Callias-type τ -index In this section we prove Theorem 2.21. 7.1. Continuous perturbations. We start with an abstract result. Let H1 and H2 be A- Hilbert spaces and suppose that T : H1 → H2 is a closed A-linear operator. We denote by Dom(T ) the domain of T considered as a Hilbert space with the graph scalar product (x, y)Dom(T ) := (x, y)H1 + (Dx, Dy)H2 . Then T : Dom(T ) → H2 is a bounded A-linear operator. Assume in addition that T is τ -Fredholm, cf. Definition 2.12. 14 MAXIM BRAVERMAN† AND SIMONE CECCHINI Definition 7.2. A one parameter family {Tt}t∈[0,1] of τ -Fredholm operators Tt : H1 → H2 is called a continuous perturbation of T with fixed domain if T0 = T , Dom(Tt) = Dom(T ) for all t ∈ [0, 1], and the induced family Tt : Dom(T ) → H2 is continuous in operator norm (as a family of maps between the Hilbert spaces Dom(T ) and H2). The next lemma shows that the τ -index is stable with respect to continuous perturbations with fixed domain. Lemma 7.3. Let {Tt}t∈[0,1] be a continuous perturbation of T with fixed domain. Then indτ Tt = indτ T, for all t ∈ [0, 1]. (7.1) Proof. In general, note that the operator (I + T ∗T ) 1 2 : Dom(T ) −→ H1 is an isometric isomorphism of A-Hilbert spaces and consider the family of operators The operators Φ(Tt) form a continuous family of bounded τ -Fredholm operators and Φ(Tt) := Tt(I + T ∗T )− 1 2 : H1 −→ H2. indτ Φ(Tt) = indτ Tt. Hence, it is enough to prove the lemma for the case when Dom(T ) = H1 and Tt is a continuous family of bounded operators. In this case the lemma is proven in [12, Theorem 4]. (cid:3) 7.4. Proof of Theorem 2.21. For 0 ≤ t ≤ 1, we set: Dt := D + F0 + t (F1 − F0). Since the A-endomorphism F1 − F0 vanishes outside of the compact set K, all Dt satisfy the condition (2.11) and, hence, are Callias-type operators. It follows from Theorem 2.19 that all the operators Dt are τ -Fredholm. Since F1 − F0 : L2(M, E+) → L2(M, E−) is a bounded A-operator, the domain of Dt is independent of t. Finally, we have: kDs − Dtk = s − tkF1 − F0k , for all s, t ∈ [0, 1]. Thus the family {Dt} is continuous in operator norm. Theorem 2.19 follows now from Lemma 7.3. (cid:3) References [1] N. Anghel, An abstract index theorem on noncompact Riemannian manifolds, Houston J. Math. 19 (1993), no. 2, 223 -- 237. MR1225459 (94c:58193) [2] , On the index of Callias-type operators, Geom. Funct. Anal. 3 (1993), no. 5, 431 -- 438. MR1233861 (94m:58213) [3] M. F. Atiyah, Elliptic operators, discrete groups and von neumann algebras, Asr´erisque 32/33 (1976), 43 -- 72. [4] B. Blackadar, Operator algebras, Encyclopaedia of Mathematical Sciences, vol. 122, Springer-Verlag, Berlin, 2006. Theory of C ∗-algebras and von Neumann algebras, Operator Algebras and Non-commutative Geometry, III. MR2188261 (2006k:46082) [5] R. Bott and R. Seeley, Some remarks on the paper of Callias: "Axial anomalies and index theorems on open spaces" [Comm. Math. Phys. 62 (1978), no. 3, 213 -- 234; MR 80h:58045a], Comm. Math. Phys. 62 (1978), no. 3, 235 -- 245. MR507781 (80h:58045b) [6] M. Braverman, On self-adjointness of a Schrodinger operator on differential forms, Proc. Amer. Math. Soc. 126 (1998), 617 -- 623. [7] [8] , Index theorem for equivariant Dirac operators on noncompact manifolds, K-Theory 27 (2002), no. 1, 61 -- 101. , The index theory on non-compact manifolds with proper group action, J. Geom. Phys. 98 (2015), 275 -- 284. VON NEUMANN ALGEBRA VALUED SCHR ODINGER OPERATORS 15 [9] M. Braverman and L. Cano, Index theory for non-compact G-manifolds, Geometric, algebraic and topological methods for quantum field theory, 2014, pp. 60 -- 94. MR3204959 [10] M. Braverman, O. Milatovich, and M. Shubin, Essential selfadjointness of Schrodinger-type operators on manifolds, Russian Math. Surveys 57 (2002), 41 -- 692. [11] M. Breuer, Fredholm theories in von Neumann algebras. I, Math. Ann. 178 (1968), 243 -- 254. MR0234294 (38 #2611) [12] , Fredholm theories in von Neumann algebras. II, Math. Ann. 180 (1969), 313 -- 325. MR0264407 (41 #9002) [13] U. Bunke, A K-theoretic relative index theorem and Callias-type Dirac operators, Math. Ann. 303 (1995), no. 2, 241 -- 279. MR1348799 (96e:58148) [14] D. Burghelea, L. Friedlander, T. Kappeler, and P. McDonald, Analytic and Reidemeister torsion for repre- sentations in finite type Hilbert modules, Geom. Funct. Anal. (1996), 751 -- 859. [15] C. Callias, Axial anomalies and index theorems on open spaces, Comm. Math. Phys. 62 (1978), no. 3, 213 -- 234. MR507780 (80h:58045a) [16] P. Chernoff, Essential self-adjointness of powers of generators of hyperbolic equations, J. Functional Analysis 12 (1973), 401 -- 414. [17] A. T. Fomenko and A. S. Miscenko, The index of elliptic operators over C ∗-algebras, Math. USSR, Izv. 15 (1980), 87 -- 112. [18] M. Gromov and H. B. Lawson Jr., Positive scalar curvature and the Dirac operator on complete Riemannian manifolds, Inst. Hautes ´Etudes Sci. Publ. Math. 58 (1983), 83 -- 196 (1984). MR720933 (85g:58082) [19] P. Hochs and V. Mathai, Geometric quantization and families of inner products, 50 pages, [arXiv:1309.6760]. [20] W. Luck, L2-invariants: theory and applications to geometry and K-theory, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics], vol. 44, Springer-Verlag, Berlin, 2002. MR1926649 (2003m:58033) [21] X. Ma and W. Zhang, Geometric quantization for proper moment maps, 2008. arXiv:0812.3989. [22] , Geometric quantization for proper moment maps: the Vergne conjecture, Acta Math. 212 (2014), no. 1, 11 -- 57. MR3179607 [23] V. Mathai and W. Zhang, Geometric quantization for proper actions, Adv. Math. 225 (2010), 1224 -- 1247. With an appendix by Ulrich Bunke [arXiv:0806.3138]. [24] I.M. Oleinik, On the essential self-adjointness of the Schrodinger operator on a complete Riemannian mani- fold, Mathematical Notes 54 (1993), 934 -- 939. [25] , On the connection of the classical and quantum mechanical completeness of a potential at infinity on complete Riemannian manifolds, Mathematical Notes 55 (1994), 380 -- 386. [26] P.-´E. Paradan, Spinc-quantization and the K-multiplicities of the discrete series, Ann. Sci. ´Ecole Norm. Sup. (4) 36 (2003), 805 -- 845. [27] F.S. Rofe-Beketov, Self-adjointness conditions for the Schrodinger operator, Mat. Zametki 8 (1970), 741 -- 751. [28] T. Schick, L2-index theorems, KK-theory, and connections, New York J. Math. 11 (2005), 387 -- 443 (elec- tronic). MR2188248 (2006h:19007) [29] M. Shubin, Spectral theory of the Schrodinger operators on non-compact manifolds: qualitative results, Spec- tral theory and geometry (Edinburgh, 1998), 1999, pp. 226 -- 283. MR1736869 (2001d:58037) [30] [31] , Spectral theory of elliptic operators on non-compact manifolds, Ast´erisque 207 (1992), 37 -- 108. , Semiclassical asymptotics on covering manifolds and Morse inequalities, Geom. Funct. Anal. 6 (1996), 370 -- 409. Department of Mathematics, Northeastern University, Boston, MA 02115, USA E-mail address: [email protected] URL: www.math.neu.edu/~braverman/ Department of Mathematics, Northeastern University, Boston, MA 02115, USA E-mail address: [email protected]
1601.00613	3	1601	2016-04-26T14:22:49	Unitary dilation of freely independent contractions	[ "math.OA", "math.FA" ]	Inspired by the Sz.-Nagy-Foias dilation theorem we show that $n$ freely independent contractions dilate to $n$ freely independent unitaries.	math.OA	math	UNITARY DILATION OF FREELY INDEPENDENT CONTRACTIONS SCOTT ATKINSON AND CHRISTOPHER RAMSEY Abstract. Inspired by the Sz.-Nagy-Foias dilation theorem we show that n freely independent contractions dilate to n freely independent unitaries. 1. Introduction The Sz.-Nagy-Foias dilation theorem is a celebrated result in classical dilation theory. It says that n doubly commuting contractions can be simultaneously di- lated to n doubly commuting unitaries. This was the original multivariable dilation theory context proven by Brehmer and Sz.-Nagy [7, 25, 26] until Ando [1] proved that one can do this for just commuting and not doubly commuting contractions when n = 2. However, it was subsequently shown in [20] and [27] that there are three commuting contractions which do not dilate to three commuting unitaries. This obstruction spurred on dilation theories in other contexts [2, 8, 11, 14, 23] and many other generalizations. One recent usage of dilations of doubly commut- ing contractions is the dilation of Nica covariant representations of lattice-ordered semigroups [15, 17]. Doubly commuting is one of two ingredients in the notion of tensor independence (or classical independence). It is natural then to ask whether n tensor independent contractions can be dilated to n tensor independent unitaries. The answer is yes (Theorem 2.2) and begs the question whether this can be done with other notions of non-commutative probability, namely free probability. Stemming from the notion of reduced free product [3, 28] Voiculescu developed the theory of free probability in the 1980's with the goal of solving the free group factor problem. While this still remains unsolved, free probability has become a very important field of mathematical research. For further reading see [16, 19]. This paper culminates in Theorem 3.2, that n freely independent contractions do indeed dilate to n freely independent unitaries. In a dilation theory context this has been done by Boca in [5] where he gives the most general unitary dilation of n contractions. The only free probability dilation result we know of is the unitary dilation of L-free sets of contractions of Popa and Vaes [22]. Acknowledgements. We would like to thank David Sherman and Stuart White for some very helpful discussions during White's visit to the University of Virginia sponsored by the Institute of Mathematical Science. 2010 Mathematics Subject Classification. 47A20, 46L54, 46L09. Key words and phrases: Dilation, non-commutative probability, tensor independence, free independence, free product. 1 2 SCOTT ATKINSON AND CHRISTOPHER RAMSEY 2. Dilation theory of tensor independence We first turn to the classical setting for inspiration. There are many great proofs of the Sz.-Nagy-Foias dilation theorem and the following constructive method is probably quite old but the authors have only seen it written down in [10, Example 2.5.13]. Recall that two operators S, T ∈ B(H) doubly commute if ST = T S and S ∗T = T S ∗. This is equivalent to requiring that C ∗(1, S) and C ∗(1, T ) commute. Note that this does not require that S and T are normal. Theorem 2.1 (Sz.-Nagy-Foias). Let T1, . . . , Tn be doubly commuting contractions in B(H). Then there exists a Hilbert space K containing H and doubly commuting unitaries U1, . . . , Un ∈ B(K) such that T1(k1) · · · Tn(kn) = PHU k1 1 · · · U kn n H, where T (k) =(cid:26) T k, k ≥ 0 T ∗−k, k < 0 . Furthermore, this dilation is unique up to unitary equivalence when K is minimal, meaning that it is the smallest reducing subspace of U1, . . . , Un containing H. Proof. As mentioned above, this can also be found in [10, Example 2.5.13]. Let H1 = ℓ2(Z) ⊗ H and set T (1) 1 = and T (1) j = Iℓ2(Z) ⊗ Tj, 2 ≤ j ≤ n.   . . . . . . 0 I 0 DT ∗ −T ∗ T1 1 DT1 1 0 I 0 . . . . . .   1 Note that T (1) a unitary [24]. Since T1, . . . , Tn doubly commute then so do T (1) immediate after noticing that the defect operators DT1 = (I − T ∗ are in C ∗(1, T1). is the classic Schaffer form of the Sz.-Nagy dilation of T1 and so is , . . . , T (1) n , this is 1 T1)1/2 and DT ∗ 1 1 In the second step, let H2 = ℓ2(Z) ⊗ H1, T (2) 2 and T (2) of T (1) T 2 , . . . , T (2) 1 (2) n j = Iℓ2(Z) ⊗ T (1) are doubly commuting. j for j 6= 2. Then T (2) 1 be the Schaffer-Sz.-Nagy dilation are unitaries and and T (2) 2 Continuing in this way one arrives at the nth step with doubly commuting uni- n ∈ B(Hn) that are easily seen to satisfy the joint power dilation , . . . , T (n) taries T (n) condition. 1 Uniqueness when the dilation is minimal follows in the same way as in the one variable setting. It is proven by way of the uniqueness of the minimal Stinespring representation. (cid:3) This can be rephrased into a non-commutative probability context. Recall that a non-commutative C∗-probability space (A, ϕ) is a C∗-algebra A along with a state ϕ ∈ S(A). We say that the operators T1, . . . , Tn ∈ A are tensor (or classically) independent in (A, ϕ) (or with respect to ϕ) if C ∗(1, T1), . . . , C ∗(1, Tn) pairwise UNITARY DILATION OF FREELY INDEPENDENT CONTRACTIONS 3 commute and given ai ∈ C ∗(1, Ti) we have the following factorization ϕ n Yi=1 ai! = n Yi=1 ϕ(ai). Theorem 2.2. Let T1, . . . , Tn ∈ B(H) be tensor independent contractions in the non-commutative probability space (B(H), ϕ). Then there exists a Hilbert space K containing H and unitaries U1, . . . , Un ∈ B(K) that are tensor independent with respect to the state ψ = ϕ ◦ adPH such that T1(k1) · · · Tn(kn) = PHU k1 1 · · · U kn n H, where T (k) =(cid:26) T k, k ≥ 0 T ∗−k, k < 0 . Furthermore, this dilation is unique up to unitary equivalence when K is minimal, meaning that it is the smallest reducing subspace of U1, . . . , Un. Proof. All that needs to be shown is that the unitaries arising from Theorem 2.1 are tensor independent with respect to ψ. To this end first assume that K = Hn and Uj = T (n) as in the proof of Theorem 2.1. j Let ai ∈ C ∗(1, T (1) ), 1 ≤ i ≤ n. This implies that PHaiH ∈ C ∗(1, Ti) for 1 ≤ i ≤ n, and PH and ai commute for 2 ≤ i ≤ n since H is a reducing subspace for each C ∗(1, T (1) ), 2 ≤ i ≤ n. Hence, i i ϕ PH n Yi=1 aiPH! = ϕ n (PHaiPH)! Yi=1 Yi=1 ϕ(PHaiPH). = n 1 n , . . . , T (1) Thus, T (1) in this fashion one gets that T (n) ϕ ◦ adPH ◦ adPH1 1 ◦ · · · ◦ adPHn−1 are tensor independent with respect to ϕ ◦ adPH . Continuing , . . . , T (n) are tensor independent with respect to = ϕ ◦ adPH = ψ where the last copy of H is in Hn. (cid:3) Uniqueness of this dilation is given by Theorem 2.1. n 3. Dilation theory of free independence In this section we will prove a theorem very similar to Theorem 2.2 in another non-commutative probability context. Recall that the operators T1, . . . , Tn ∈ A are freely independent (or ∗-free) in (A, ϕ) if their C∗-algebras C ∗(1, T1), . . . , C ∗(1, Tn) are freely independent. That is, whenever aj ∈ C ∗(1, Tij ) such that ϕ(aj) = 0 for 1 ≤ ij ≤ n and ij 6= ij−1 for 1 < j ≤ m then ϕ(a1a2 · · · am) = 0. Another proof of the Sz.-Nagy-Foias Theorem (Theorem 2.1 above) can be found in Paulsen [21, Theorem 12.10]. Here one gets ucp maps θi : C(T) → C ∗(1, Ti) given by dilation theory. Now one can extend this to the ucp map θ1 ⊗ · · · ⊗ θn on C(T) ⊗max · · · ⊗max C(T) ≃ C(Tn). By taking the Stinespring representation of θ one gets the desired doubly commuting unitaries that jointly dilate T1, . . . , Tn. This provides a roadmap for an attempt to prove the free analogue of Theorem 2.2. Namely, by using the free product of ucp maps and then taking the Stinespring representation of this map it will be shown that one gets unitaries that jointly dilate 4 SCOTT ATKINSON AND CHRISTOPHER RAMSEY T1, . . . , Tn. One then hopes that these unitaries will be ∗-free with respect to a natural state. This approach, minus the free probability is exactly what Boca [5] uses to es- tablish his unitary dilation result, in fact proving a more general statement about normal rational dilations. Recall now, that the unital universal free product of unital C∗-algebras A1, . . . , An is the universal C∗-algebra amalgamated over C generated by A1, . . . , An and is denoted ∗n i=1Ai. In particular, whenever one has a unital C∗-algebra B and uni- tal ∗-homomorphisms πi : Ai → B then there exists a unital ∗-homomorphism π : ∗n i=1Ai → B. Suppose there are unital completely positive maps θi ϕi ∈ S(Ai). In [5], Boca proves that there exists a ucp map ∗n ∗n i=1Ai → B such that ∗n to the ϕi. Namely, when aj ∈ Aij with ϕij (aj) = 0 and ij 6= ij−1 then : Ai → B and states i=1θi = θ1 ∗ · · · ∗ θn : i=1θAi = θi. This is defined on reduced words with respect ∗n i=1θ(a1 · · · am) = θi1 (a1) · · · θim(am). This completely determines ∗n i=1θ as (reduced words + C1) is dense in ∗n i=1Ai. Lemma 3.1. Suppose T1, . . . , Tn ∈ B(H) and V1, . . . , Vn ∈ B(K) are contractions and θi : C ∗(1, Vi) → C ∗(1, Ti) such that p(Vi) 7→ p(Ti) are ucp maps (p a polyno- mial). If ψi ∈ S(C ∗(1, Vi)) then the free product ucp map with respect to the ψi, i=1θi, is a homomorphism on the subalgebra Alg{1, V1, . . . , Vn} of ∗n ∗n i=1C ∗(Vi). Proof. The result can be established by induction. By definition each θi is already a homomorphism on Alg{1, Vi} ⊆ C ∗(Vi). Now for m ≥ 1 assume that for all 1 ≤ k ≤ m and for any bj ∈ Alg{1, Vij }, 1 ≤ j ≤ k we have that ∗n Qk j=1 θij (bj). Suppose now we have aj ∈ Alg{1, Vij }, 1 ≤ j ≤ m + 1. If a pair of neighboring terms belongs to the same algebra, say ij = ij−1, by the inductive hypothesis and since θij is a homomorphism we have i=1θi(cid:16)Qk j=1 bj(cid:17) = ∗n i=1θi(a1 · · · aj−1aj · · · am+1) = θi1 (a1) · · · θij (aj−1aj) · · · θim+1 (am+1) = θi1 (a1) · · · θij−1 (aj−1)θij (aj) · · · θim+1 (am+1). Otherwise assume that ij−1 6= ij for 1 < j ≤ m + 1 and then we have ∗n i=1θi  m+1 Yj=1 aj  = ∗n m+1 m+1 i=1 θi Yj=1  i=1 θi Yj=2 ψ(a1) i=1 θi Yj=3 a1ψ(a2) (aj − ψ(aj ))  + (aj − ψ(aj))  + (aj − ψ(aj))  + · · · + m+1 ∗n ∗n ∗n i=1 θi (a1 · · · amψ(am+1)) m+1 θij (aj − ψ(aj)) + = Yj=1 UNITARY DILATION OF FREELY INDEPENDENT CONTRACTIONS 5 ψ(a1) m+1 Yj=2 θij (aj − ψ(aj)) + m+1 θi1 (a1)ψ(a2) θij (aj − ψ(aj)) + · · · + Yj=3 θi1 (a1) · · · θim (am)ψ(am+1) m+1 = θij (aj) Yj=1 by the definition of ∗n i=1θi and the inductive hypothesis. The result follows as it is a simple matter now to show that ∗n ∗n i=1θi(a) ∗n i=1 θi(b) for all a, b ∈ Alg{1, V1, . . . , Vn}. i=1θi(ab) = (cid:3) Theorem 3.2. Let T1, . . . , Tn ∈ B(H) be freely independent contractions in the non-commutative probability space (B(H), ϕ). Then there exists a Hilbert space K containing H and unitaries U1, . . . , Un ∈ B(K) that are freely independent with respect to ϕ ◦ ad PH such that = PHU k1 H, 1 ≤ ij ≤ n and kj ∈ N ∪ {0}. T k1 i1 · · · T kn im i1 · · · U kn im Furthermore, this dilation is unique up to unitary equivalence when K is minimal. Proof. For each 1 ≤ i ≤ n, let Vi ∈ B(Ki) with H ⊂ Ki be the minimal unitary dila- tion of Ti and θi : C ∗(Vi) → C ∗(1, Ti) given by θi = ad PH, a unital completely pos- itive map. Thus, ψi := ϕ◦θi is a state on C ∗(Vi) such that ψi(V n i ), ∀n ≥ 0. i=1C ∗(Vi) → C ∗(1, T1, . . . , Tn) i=1θi with i=1θi(a) = PHπ(a)H. Define unitaries Ui := π(Vi) and note relative to the states ψi. Let (π, K) be the Stinespring representation of ∗n H ⊂ K which gives ∗n that for m ≥ 1 and ki ≥ 0 then Consider now, the free product ucp map ∗n i ) = ϕ(T n i=1θi : ∗n PHU k1 i1 · · · U km im )H H = PHπ(V k1 i=1θi(V k1 i1 i1 · · · T km im = ∗n = T k1 i1 · · · V km im · · · V km im ) by Lemma 3.1. Furthermore, for aj ∈ C ∗(Uij ) with ϕ ◦ ad PH(aj) = 0 and ij 6= ij−1 we have that ∃bj ∈ C ∗(Vij ) such that π(bj) = aj, θij (bj) ∈ C ∗(1, Tij ) and 0 = ϕ ◦ ad PH(aj ) = ϕ ◦ ad PH ◦ π(bj ) = ϕ(θij (bj)). Hence, ϕ ◦ ad PH(a1 · · · am) = ϕ ◦ ad PH ◦ π(b1 · · · bm) i=1θi(b1 · · · bm) = ϕ ◦ ∗n = ϕ(θi1 (b1) · · · θim (bm)) = 0. Therefore, U1, . . . , Un are ∗-free with respect to ϕ ◦ ad PH. An argument to show that the minimal unitary dilation is unique is given in [5, Section 4] following from a classic remark of Durszt and Sz.-Nagy [12]. In particular, given two minimal unitary dilations U1, . . . , Un and U ′ n on K and K′ respectively, there is a unitary Θ : K → K′ fixing the subspace H such that 1, . . . , U ′ 6 SCOTT ATKINSON AND CHRISTOPHER RAMSEY ΘUi = U ′ of Θ since a state is completely determined by the values i Θ. From this it is easy to see that the states ϕ ◦ ad PH are equal by way ϕ ◦ ad PH(U k because U1, . . . , Un and U ′ 5.13]. i ) = ϕ(T k 1, . . . , U ′ i ) = ϕ ◦ ad PH((Ui)′)k), ∀k ∈ N n are collections of ∗-free unitaries [19, Lemma (cid:3) Remark 3.3. It should be noted, by [29, Proposition 2.5.3], ϕ ◦ ad PH is a tracial state on C ∗(U1, . . . , Un) since U1, . . . , Un are ∗-free and ϕ ◦ ad PH is a trace on each C ∗(Ui) (the algebra is commutative). One of the purposes of doing the previous arguments carefully is that we can now say what happens when the state ϕ is faithful. Lemma 3.4. Let T ∈ B(H) be a contraction and ϕ be a faithful state on C ∗(1, T ). If U ∈ B(K) with H ⊂ K is the minimal unitary dilation of T then ϕ ◦ ad PH is a faithful state on C ∗(U ). Proof. Let (π, K′, ξ) be the GNS representation of (C ∗(U ), ϕ◦ad PH). Then π(U ) is still a unitary, ϕ◦ ad PH(a) = hπ(a)ξ, ξi and h·ξ, ξi is a faithful state on π(C ∗(U )) = C ∗(π(U )). Suppose a is a positive element in ker π. Then 0 = hπ(a)ξ, ξi = ϕ ◦ ad PH(a) which implies that ad PH(a) = 0 since compression to H is a completely positive map and ϕ is faithful. Since ker π is a C ∗-algebra (so every element of ker π can be written as a linear combination of at most four positive elements from ker π), we can conclude that ker π ⊂ ker ad PH. Thus there exists a well-defined ucp map θ : C ∗(π(U )) → C ∗(1, T ) given by sending π(a) 7→ PHaH. Notably, we have θ(π(U )n) = T n, n ≥ 0. Let (π, K′′) with H ⊂ K′′ be the minimal Stinespring representation of θ. That is, π : C ∗(π(U )) → B(K′′) is a ∗-homomorphism (in fact a ∗-isomorphism) such that θ(a) = PH π(a)H and K′′ is the closed linear span of π(C ∗(π(U )))H by minimality. Define V := π(π(U )) a unitary and note that PHV nH = θ(π(U n)) = T n. Thus, because of this and the minimality of the Stinespring representation we have that V is a minimal unitary dilation of T . Consider now the state ψ := ϕ ◦ ad PH on C ∗(V ). Now ψ ◦ π n Xi=−n αiπ(U )n! = ψ n αiV n! Xi=−n = ϕ n αiT (n)! Xi=−n = ϕ ◦ ad PH n αiU n! Xi=−n =* n αiπ(U )n! ξ, ξ+ . Xi=−n Hence, ψ ◦ π(·) = h·ξ, ξi is a faithful state on C ∗(π(U )) and so ψ is a faithful state on C ∗(V ). By minimality of the dilations there exists a unitary W : K → K′′ such that W h = h for all h ∈ H and W U W ∗ = V . This implies that ϕ ◦ ad PH on C ∗(U ) UNITARY DILATION OF FREELY INDEPENDENT CONTRACTIONS 7 is equal to ψ ◦ ad W which is faithful. Therefore, ϕ ◦ ad PH was a faithful state on C ∗(U ) all along. (cid:3) One last ingredient before presenting Theorem 3.5 is the reduced free product of C∗-probability spaces (Ai, ϕi), denoted (A, ϕ) with A = ∗n i=1Ai. As mentioned in the introduction this was introduced by both Avitzour [3] and Voiculescu [28] in the 1980's. In correspondence with Boca's result on free products of ucp maps there is the reduced free product of ucp maps given by Choda-Blanchard-Dykema [9, 4] and the operator-valued conditionally free product of M lotkowski [18] based on the conditionally free product of Bozejko, Leinert and Speicher [6]. Now we can present the following theorem. Theorem 3.5. Let T1, . . . , Tn ∈ B(H) be freely independent contractions in the non-commutative probability space (B(H), ϕ) with ϕ faithful. If U1, . . . , Un ∈ B(K) is the minimal free unitary dilation then ϕ ◦ ad PH is faithful. In particular, when ϕ is faithful, the minimal free unitary dilation of T1, . . . , Tn, ϕ arises from the reduced free product of their minimal unitary dilations. Proof. Because of the faithfulness of ϕ, C ∗(1, T1, . . . , Tn) is in fact ∗-isomorphic to ∗n i=1(C ∗(1, Ti), ϕ) [13, Lemma 1.3]. Let Vi ∈ B(Ki) be the minimal unitary dilation of Ti and again let θi : C ∗(Vi) → C ∗(1, Ti) be the ucp map given by θi(V k i ) = T k i . By Lemma 3.4 ϕ ◦ θi is a faithful state. As mentioned, Choda [9] showed that there is a reduced free product of ucp maps, though with a gap in the proof that was filled by Blanchard-Dykema [4]. Thus, there exists a ucp map θ : (∗n i=1C ∗(Vi), ∗n i=1ϕ ◦ θi) → (∗n i=1C ∗(1, Ti), ϕ) extending each of the θi. By [19, Theorem 7.9] ∗n i=1ϕ ◦ θi = ϕ ◦ θ is a faithful state. Lemma 3.1 applies equally well in the reduced free product case to give that θ acts homomorphically on Alg{1, V1, . . . , Vn}. As in Theorem 3.2, let (π, K) be the minimal Stinespring dilation of θ and Ui = π(Vi). Then, U1, . . . , Un, ϕ ◦ ad PH is the minimal free unitary dilation of T1, . . . , Tn, ϕ with ϕ ◦ ad PH faithful. In fact, we can say a little more. Because ϕ ◦ θ is faithful then π, the Stinespring representation of θ, has to be faithful, i.e. injective. Hence, again by [13, Lemma 1.3] (C ∗(U1, . . . , Un), ϕ ◦ ad PH) ≃ ∗n i=1(C ∗(Vi), ϕ ◦ θi). Therefore, the minimal free unitary dilation of T1, . . . , Tn, ϕ when ϕ is faithful arises from the reduced free product of their minimal unitary dilations. (cid:3) References [1] T. Ando, On a pair of commutative contractions, Acta Sci. Math. (Szeged) 24 (1963), 88 -- 90. [2] W. Arveson, Subalgebras of C∗-algebras III, Acta Math. 181 (1998), 159 -- 228. [3] D. Avitzour, Free products of C∗-algebras, Trans. Amer. Math. Soc., 271 (1982), 423 -- 435. [4] E. Blanchard and K. Dykema, Embeddings of reduced free products of operator algebras, Pacific J. Math., 199 (2001), 1 -- 19. [5] F. Boca, Free products of completely positive maps and spectral sets, J. Funct. Anal. 97 (1991), 251 -- 263. [6] M. Bozejko, M. Leinert and R. Speicher, Convolution and limit theorems for conditionally free random variables, Pacific J. Math. 175 (1996), 357 -- 388. [7] S. Brehmer, Uber vetauschbare Kontraktionen des Hilbertschen Raumes (German), Acta Sci. Math. Szeged 22 (1961), 106 -- 111. 8 SCOTT ATKINSON AND CHRISTOPHER RAMSEY [8] J. Bunce, Models for n-tuples of non-commuting operators, J. Funct. Anal. 57 (1984), 21 -- 30. [9] M. Choda, Reduced free products of completely positive maps and entropy for free product of automorphisms, Publ. Res. Inst. Math. Sci., 32 (1996), 371 -- 382. [10] K. Davidson, A. Fuller and E. Kakariadis Semicrossed products of operator algebras by semi- groups, Mem. Amer. Math. Soc. to appear. [11] S. Drury, A generalization of von Neumann's inequality to the complex ball, Proc. Amer. Math. Soc., 68 (1978), 300 -- 304. [12] E. Durszt and B. Sz.-Nagy, Remark to a paper of A. E. Frazho: "Models for noncommuting operators," J. Funct. Anal. 52 (1983), 146 -- 147. [13] K. Dykema and M. Rørdam, Projections in free product C∗-algebras, Geom. Funct. Anal., 8 (1998), 1 -- 16; Erratum, idem., 10(4) (2000), 975. [14] A. Frazho, Models for non-commuting operators, J. Funct. Anal. 48 (1982), 1 -- 11. [15] A. Fuller, Nonself-adjoint semicrossed products by abelian semigroups, Canad. J. Math. 65 (2013), 768 -- 782. [16] F. Hiai and D. Petz, The Semicircle Law, Free Random Variables and Entropy, Mathematical Surveys and Monographs 77, American Math. Soc., 2000. [17] B. Li, Normal extensions of representations of abelian semigroups, preprint arXiv:1507.06698. [18] W. M lotkowski, Operator-valued version of conditionally free product, Studia Math. 153 (2002), 13 -- 30. [19] A. Nica and R. Speicher, Lectures on the combinatorics of free probability, London Math. Soc. Lecture Note Series 335, Cambridge University Press, 2006. [20] S. Parrott, Unitary dilations for commuting contractions, Pacific J. Math. 34 (1970), 481 -- 490. [21] V. Paulsen, Completely Bounded Maps and Operator Algebras, Cambridge Studies in Ad- vanced Mathematics 78, Cambridge University Press, 2002. [22] S. Popa and S. Vaes, On the optimal paving over MASAs in von Neumann algebras, to appear "Conf. Proc. Kadison's 90th" (2015). [23] G. Popescu, Isometric dilations for infinite sequences of noncommuting operators, Trans. Amer. Math. Soc. 316 (1989), no. 2, 523 -- 536. [24] J. Schaffer, On unitary dilations of contractions, Proc. Amer. Math. Soc. 6, (1955) 322. [25] B. Sz.-Nagy, Bemerkungen zur vorstehenden Arbeit des Herrn S. Brehmer (German), Acta Sci. Math. Szeged 22 (1961), 112 -- 114. [26] B. Sz.-Nagy, C. Foias, H. Bercovici and L. K´erchy, Harmonic analysis of operators on Hilbert space, 2nd edition, Universitext, Springer, 2010. [27] N. Th. Varopoulos, On an inequality of von Neumann and an application of the metric theory of tensor products to operators theory, J. Funct. Anal. 16 (1974), 83 -- 100. [28] D. Voiculescu, Symmetries of some reduced free product C∗-algebras, in Operator algebras and their connection with topology and ergodic theory, Lecture Notes in Math. 1132, Springer, 1985, 556 -- 588. [29] D. Voiculescu; K. Dykema and A. Nica, Free Random Variables, CRM Monograph Series 1, American Math. Soc., 1992. University of Virginia, Charlottesville, VA, USA E-mail address: [email protected] and [email protected]
1108.3020	1	1108	2011-08-15T15:34:18	Characterization of amenability by a factorization property of the group von Neumann algebra	[ "math.OA", "math.FA" ]	We show that the amenability of a locally compact group $G$ is equivalent to a factorization property of $VN(G)$ which is given by $ VN(G) = <VN(G)^*VN(G)>$. This answer partially two problems proposed by Z. Hu and M. Neufang in their article \textit{Distinguishing properties of Arens irregularity}, Proc. Amer. Math. Soc. 137 (2009), no. 5, 1753-1761.	math.OA	math	CHARACTERIZATION OF AMENABILITY BY A FACTORIZATION PROPERTY OF THE GROUP VON NEUMANN ALGEBRA DENIS POULIN Abstract. We show that the amenability of a locally compact group G is equivalent to a factorization property of V N (G) which is given by V N (G) = hV N (G)∗ V N (G)i. This answer partially two problems proposed by Z. Hu and M. Neufang in their article Distinguishing properties of Arens irregularity, Proc. Amer. Math. Soc. 137 (2009), no. 5, 1753 -- 1761 It is known that the amenability of a locally compact group G can be character- ized in many different ways. For example A(G) has a bounded approximate identity ([Lep]; [He, Theorem 6, p. 129]), A(G) factorizes weakly, i.e. A(G) is the linear span of A(G) · A(G) [Lo, Proposition 2, p. 138], or A(G) is closed in M (A(G), A(G)), its multiplier algebra [Lo2, Theorem 1]. It is also known that A(G) is unital if and only if G is compact which is equivalent to V N (G) = U CB( G). This last equality can be formulated for a Banach algebra A by A∗ = A∗A. It seems possible to believe that if we relax the factorization property of the dual, we could characterize the existence of a BAI for A(G), which is equivalent to the amenability of G by Leptin's theorem. More precisely, in this note, we prove that G is amenable if and only if the linear span V N (G)∗V N (G), denoted by hV N (G)∗V N (G)i, is equal to V N (G). With this characterization of amenability for locally compact groups, we give a partial answer to the following problems proposed by Z. Hu and M. Neufang in [HN]. Problem 1 : Does hV N (G)∗(cid:3)V N (G)i = V N (G) imply that G is amenable? In this article, we made a stronger assumption than only the norm density of hV N (G)∗(cid:3)V N (G)i in V N (G). However, with our assumption, we obtain a posi- tive answer to this problem. Another problem proposed in the same article of Z. Hu and M. Neufang is similar to the previous one. We stated here. Problem 2 : If V N (G) factors over Bρ(G), i.e., hBρ(G)(cid:3)V N (G)i = V N (G), is G amenable? By [Hu, Corollary 3.1], hBρ(G)(cid:3)V N (G)i = hV N (G)∗(cid:3)V N (G)i if G is discrete. Date: June 16, 2011 . 1991 Mathematics Subject Classification. 43A07, 46H05. Key words and phrases. Amenability, Locally compact groups, Fourier algebras, Banach alge- bras, Segal algebras . The author was supported by Carleton University. 1 2 DENIS POULIN Thus, with Theorem 3, if G is discrete such that hBρ(G)V N (G)i = hV N (G)∗(cid:3)V N (G)i and closed, then hBρ(G)(cid:3)V N (G)i = V N (G) implies that G is amenable. Our argument will use the theory of abstract Segal algebras. The reader is re- ferred to [Bu], [Bu2], [Le] and [Le2] to know more about this subject. For the definition and properties of the Fourier algebra A(G) and its dual V (G), the group von Neumann algebra, we refer to [E]. Also, to fix our notation, we recall the defi- nitions of the Arens products. Let A be a Banach algebra. For m, n ∈ A∗∗, f ∈ A∗, a, b ∈ A, hm(cid:3)n, f i = hm, n(cid:3)f i, where hn(cid:3)f, ai = hn, f (cid:3)ai and hf (cid:3)a, bi = hf, abi. Similarly, using the right A- module structure of A on A∗, we have that hm△n, f i = hn, f △mi, where hf △m, ai = hm, a△f i and ha△f, bi = hf, bai. For a Banach algebra A and a A-bimodule X, we denote by hXAi and hAXi the linear span of XA and AX respectively. We present the main tool of this note which is an improvement of [P, Theorem 3.3.3]. Theorem 1. Let A be a faithful Banach algebra. Let B be a proper right (left) abstract Segal algebra in A. Then B ∗ 6= hB ∗△B ∗∗i (B ∗ 6= hB ∗∗(cid:3)B ∗i ). Proof. We will proceed by contradiction. Let B ∗ = hB ∗△B ∗∗i. We first prove that each element of hB ∗△B ∗∗i as a unique extension on A. From there, we will conclude that the norm of A and B are equivalent on B which is a contradiction. Let mi ∈ B ∗∗, i = 1, . . . , n. By Goldstein's theorem, for each i, there exists a w∗ → mi in B ∗∗. Let fi ∈ B ∗, and b ∈ B. bounded net mi,α in B such that mi,α Observe that n X h i=1 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) fi△mi, bi (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = = = ≤ (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) n X i=1 n X i=1 n X i=1 n hmi△b, fii hmi(cid:3)b, fii (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) lim α hmi,αb, fii (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) X i=1 Ci sup α kmi,αkB kbkA kfikB∗ ≤ M kbkA This shows that Pn i=1 fi△mi is bounded on B with the norm of A. Since B is dense in A, then there exists a unique extension of Pn i=1 fi△mi on A. Let ι : B → A be the natural inclusion map. Then ι∗ is injective by the density of B, and for h ∈ A∗, ι∗(h) = hB. Let us prove that ι∗ is surjective. Let f ∈ B ∗. Since B ∗ = hB ∗B ∗∗i, then there are fi ∈ B ∗ and mi ∈ B ∗∗ such that f = Pn i=1 fi△mi. Now, if we apply ι∗ on the extension of f , we get f . From this, we deduce that ι∗ and ι∗∗ are AMENABILITY AND FACTORIZATION PROPERTY 3 isomorphisms. Hence, there exist non-zero constants D1 and D2 such that D1 kmkB∗∗ ≤ kι∗∗(m)kA∗∗ ≤ D2 kmkB∗∗ for all m ∈ B ∗∗. Since, ι∗∗(b) = ι(b) = b for all b ∈ B, we get from the previous inequalities that the norm of A and the norm of B are equivalent on B, which contradicts the fact that B is a proper abstract Segal algebra. (cid:3) Corollary 2. Let A be a right, respectively left, faithful Banach algebra. If A∗ = hA∗△A∗∗i, respectively A∗ = hA∗∗(cid:3)A∗i, then the norm of A is equivalent to its right, respectively left, multiplier algebra. Proof. We do a proof by contraposition. Suppose that the norm of A is not norm equivalent to the norm of its right multiplier algebra RM (A). Then A is a right abstract Segal algebra in its closure in RM (A). By Theorem 1, A∗ 6= hA∗△A∗∗i. (cid:3) We mention here that the equivalence of the norm of a Banach algebra A with its right or left multiplier algebra does not imply that A has a bounded approximate identity (see [W, Example 5]). Using Theorem 1, we give a positive answer of problem 1 of [HN] if hV N (G)∗(cid:3)V N (G)i is already closed and not only norm dense. Note that the commutativity of A(G) implies that V N (G)△V N (G)∗ = V N (G)∗(cid:3)V N (G). Theorem 3. Let G be a locally compact group. Then G is amenable if and only if V N (G) = hV N (G)∗(cid:3)V N (G)i. Proof. If G is amenable, then by Leptin's theorem, A(G) has a bounded approxi- mate identity and so V N (G) = V N (G)∗(cid:3)V N (G). If V N (G) = hV N (G)∗(cid:3)V N (G)i, then by Corollary 2, the norm of A(G) and its multiplier algebra are equivalents. Thus, by [Lo2, Theorem 1], G is amenable. (cid:3) It is interested to compare this result on factorization of V N (G) with the classical factorization property treated by A. Lau and A. Ulger in [LU], i.e., A∗ = A∗A for a Banach algebra A with a BAI. In the case of the Fourier algebra, for any locally compact group G, the equality V N (G) = V N (G)△A(G) = U CB( G) implies that A(G) is unital. This is not the case in general for Banach algebra. Take for example, A = K(X), where X is a non-reflexive Banach space such that X ∗ has the bounded approximation property and the Radon-Nikodym property. By [P, Corollary 4.2.12], A∗ = A∗A, but K(X) is not unital. However, Lau- Ulger asked a question after [LU, Theorem 2.6] which was made as a conjecture by the author [P, Conjecture 5.3.5], we state it here: Conjecture 4. There is no faithful infinite-dimensional non-unital weakly sequen- tially complete Banach algebra A with a BAI such that A∗ = A∗A. This conjecture is true for many Banach algebras like strongly Arens irregular Banach algebras, Arens regular Banach algebras, Banach algebras which are ideals in their biduals. To get a more complete list with proof of each case, see [P, Theorem 5.3.6]. Theorem 3 would suggest that for weakly sequentially complete Banach algebra, the factorization A∗ = hA∗△A∗∗i captures the existence of a BAI. We can reformulate this by saying that the factorization A∗ = hA∗△A∗∗i can characterize the co-amenability of a locally compact quantum group G. 4 DENIS POULIN References [Bu] J. T. Burnham, Closed ideals in subalgebras of Banach algebras. II. Ditkin's condition. Monatsh. Math., 73 (1974), 1-3 [Bu2] J. T. Burnham, Segal algebras and dense ideals in Banach algebras. Lecture Notes in Math., 399 (1974), 33-58 [D] H. G. Dales, Banach algebras and automatic continuity. London Mathematical Society Mono- graphs. New Series, 24. Oxford Science Publications. The Clarendon Press, Oxford University Press, New York, 2000. xviii+907 pp. [E] P. Eymard, L'alg`ebra de Fourier d'un groupe localement compact, Bull. Soc. Math. France 92 (1964), 181-236. [He] C. Herz, Harmonic synthesis for subgroups. Ann. Inst. Fourier (Grenoble) 23 (1973), 91-123. [Hu] Z. Hu, Open subgroups and the centre problem for the Fourier algebra. Proc. Amer. Math. Soc. 134 (2006), no. 10, 30853095 [HN] Z. Hu ; M. Neufang, Distinguishing properties of Arens irregularity., Proc. Amer. Math. Soc. 137 (2009), no. 5, 17531761 [L] A. T. Lau, The second conjugate algebra of the Fourier algebra of a locally compact group. Trans. Amer. Math. Soc. 267 (1981), no. 1, 5363. [LL3] A. T. Lau ; V. Losert, The C ∗-algebra generated by operators with compact support on a locally compact group. J. Funct. Anal. 112 (1993), no. 1, 130. [LU] A. T. Lau ;A. Ulger, Topological centers of certain dual algebras. Trans. Amer. Math. Soc. 348 (1996), no. 3, 11911212. [Le] M. Leinert, A contribution to Segal algebras., Manuscripta Math. 10 (1973), 297-306 [Le2] M. Leinert, Remarks on Segal algebras., Manuscripta Math. 16 (1975), no. 1, 1-9 [Lep] H. Leptin, Sur l'algebre de Fourier d'un groupe localement compact. C.R. Acad. Sci. Paris Ser. A-B 266 (1968), A1180-A1182. [Lo] V. Losert, , Some properties of groups without the property P1, Comment. Math. Helv. 54 (1979), 133-139. [Lo] V. Losert, The centre of the bidual of Fourier algebras (discrete groups). preprint. [Lo2] V. Losert, Properties of the Fourier algebra that are equivalent to amenability. Proc. Amer. Math. Soc. 92 (1984), no. 3, 347354. [NP1] M. Neufang ; D. Poulin, Characterization of the weak amenability of locally compact groups. preprint [NP2] M. Neufang ; D. Poulin, Strong topological center. preprint [NP3] M. Neufang ; D. Poulin, Arens regularity of the Fourier algebra. preprint [P] D. Poulin, The strong topological centre and dual factorization properties, PhD thesis, Car- leton University, in preparation. [W] G. Willis, Examples of factorization without bounded approximate units. Proc. London Math. Soc. (3) 64 (1992), no. 3, 602624. [Y] N. J. Young, The irregularity of multiplication in group algebras. Quart J. Math. Oxford Ser. (2) 24 (1973), 5962. Department of Mathematics, Carleton University, Ottawa, Ontario, Canada Current address: Department of Mathematics, 1125 Colonel By Drive, Ottawa, Ontario, K1S 5B6, Canada E-mail address: [email protected]
1111.5168	3	1111	2013-12-10T15:18:38	A note on weak amenability for reduced free products of discrete quantum groups	[ "math.OA" ]	We prove that the Cowling-Haagerup constant of a reduced free product of weakly amenable discrete quantum groups with Cowling-Haagerup constant equal to 1 is again equal to 1.	math.OA	math	A NOTE ON WEAK AMENABILITY FOR FREE PRODUCTS OF DISCRETE QUANTUM GROUPS AMAURY FRESLON Abstract. We prove that the Cowling-Haagerup constant of a reduced free product of weakly amenable discrete quantum groups with Cowling-Haagerup constant equal to 1 is again equal to 1. 1. Introduction In geometric group theory, weak amenability is an approximation property which is satisfied by a large class of groups (for example free or even hyperbolic groups [5]) but is strong enough to give interesting properties for the von Neumann algebras associated to these groups (for example related to deformation/rigidity techniques [6], [7]). The stability of this property under free products is still an open problem. However, using a very general version of the Khintchine inequality, E. Ricard and Q. Xu were able to prove in [8] that if (Gi) is a family of weakly amenable discrete groups with Cowling-Haagerup constant equal to 1, then their free product is again weakly amenable, and its Cowling-Haagerup constant is also equal to 1. The proof uses a classical characterization of the bounded functions on a group giving rise to completely bounded multipliers. This characterization having recently been generalized to arbitrary locally compact quantum groups by M. Daws in [3], we are able to prove an analogue of Ricard and Xu's result in the setting of discrete quantum groups. We are very grateful to P. Fima for many useful discussions and advice and to E. Blanchard and R. Vergnioux for their careful proof-reading of an early version. We also thank M. Brannan for pointing to us the paper [3]. 2. Definitions and notations In this section, we introduce the notations and results about discrete quantum groups which will be used in this paper. For two Hilbert spaces H and K, B(H, K) will denote the set of bounded linear maps from H to K and B(H) = B(H, H). In the same way, we use the notations K(H, K) and K(H) for compact linear maps. We will denote by B(H)∗ the predual of B(H), i.e. the Banach space of all normal linear forms on B(H). On any tensor product A ⊗ B, we define the flip operator Σ : x ⊗ y 7→ y ⊗ x. We will use the usual leg-numbering notations : for an operator X acting on a tensor product we set X12 := X ⊗ 1, X23 := 1 ⊗ X and X13 := (Σ⊗ 1)(1⊗ X)(Σ⊗ 1). For a subset B of a topological vector space C, spanB will denote the closed linear span of B in C. The symbol ⊗ will denote the minimal (or spatial) tensor product of C-algebras or the topological tensor product of Hilbert spaces. Let A be a C-algebra together with a distinguished state ϕ. Taking the completion of A with respect to the scalar product ha, bi = ϕ(a∗b) yields a Hilbert space which will be denoted L2(A, ϕ). If the scalar product is ha, biop = ϕ(ab∗), then the completion will be denoted L2(A, ϕ)op. Let G = (C(G), ∆)) be a compact quantum group in the sense of [11] and let bG = (C0(bG), b∆) be its dual discrete quantum group. The Haar state of G will be denoted h. We will always assume it to be faithful and identify C(G) with its image under the GNS map. We will denote by L2(G) the Hilbert space of the GNS construction, by ξh the cyclic separating vector and by W the unitary operator on L2(G) ⊗ L2(G) defined by W ∗(ξ ⊗ aξh) = ∆(a)(ξ ⊗ ξh) for ξ ∈ L2(G) and a ∈ C(G). Then, W is a multiplicative unitary in the sense of [1], i.e. W12W13W23 = W23W12. We have the following equalities : 3 1 0 2 c e D 0 1 ] . A O h t a m [ 3 v 8 6 1 5 . 1 1 1 1 : v i X r a C(G) = span(ı ⊗ B(L2(G))∗)(W ) ∆(x) = W ∗(1 ⊗ x)W C0(bG) = span(B(L2(G))∗ ⊗ id)(W ) b∆(x) = ΣW (x ⊗ 1)W ∗Σ. These faithful representations on L2(G) allow us to define the von Neumann algebras L∞(G) = C(G)′′ and ℓ∞(bG) = C0(bG)′′, and one can check that W ∈ L∞(G) ⊗ ℓ∞(bG). Finally, we set cW = ΣW ∗Σ. Let bG be a discrete quantum group and a ∈ ℓ∞(bG). The left multiplier associated to a is the map ma on (ı ⊗ B(L2(G))∗)(W ) defined by (ma ⊗ id)(W ) = (1 ⊗ a)W . A net (ai) of elements of ℓ∞(bG) is said to converge pointwise to a ∈ ℓ∞(bG) if aip → ap in B(L2(G)) for any minimal central projection p ∈ ℓ∞(bG). An element a ∈ ℓ∞(bG) is said to have finite support if ap vanishes for all but finitely many central projections p ∈ ℓ∞(bG). 3. Preliminaries 1 2 AMAURY FRESLON For a discrete group G, it is known that a bounded function ϕ : G → C gives rise to a completely bounded multiplier if and only if there exists a Hilbert space K and two families (ξs)s∈G and (ηt)t∈G of vectors in K such that ϕ(s) = hηt, ξsti (which is usually written ϕ(st−1) = hηt, ξsi). Moreover, the completely bounded norm is then inf k(ξs)k∞k(ηt)k∞ (see e.g. [2]). The following theorem [3, Prop 4.1 and Thm 4.2] gives the quantum analogue of this characterization. Theorem 3.1 (Daws [3]). Let bG be a discrete quantum group and a ∈ ℓ∞(bG). Then ma extends to a com- petely bounded multiplier on B(L2(G)) if and only if there exists a Hilbert space K and two maps α, β ∈ B(L2(G), L2(G) ⊗ K) such that (1 ⊗ β)∗cW ∗ 12(1 ⊗ α)cW = a ⊗ 1. Moreover, we then have ma(x) = β∗(x ⊗ 1)α and we can chose α and β to have norm equal to pkmakcb. Proof. We only give the construction of α and β since we will need their precise form later on. Assume ma to be completely contractive. By Wittstock's factorization theorem, there is a representation π : B(L2(G)) → B(K) and two isometries P, Q ∈ B(L2(G), K) such that for all x ∈ B(L2(G)), ma(x) = Q∗π(x)P . Then, setting α = (ı ⊗ π)(cW )(1 ⊗ P )cW (ı ⊗ ξh) and β = (ı ⊗ π)(cW )(1 ⊗ Q)cW (ı ⊗ ξh) yields the result. From this we easily deduce that a multiplier ma on a discrete quantum group bG has a completely bounded extension to B(L2(G)) if and only if it has a completely bounded extension to C(G) or to L∞(G) and that the completely bounded norms of these extensions are all equal. This justifies the following definition of weak amenability for discrete quantum groups. (cid:3) Definition 3.2. A discrete quantum group bG is said to be weakly amenable if there exists a net (ai) of elements of ℓ∞(bG) such that ai has finite support for all i, the net (ai) converges pointwise to 1 and the maps mai satisfy lim supkmaikcb < ∞. The infimum of lim supkmaikcb,B(L2(G)) for all nets satisfying the above conditions is denoted Λcb(bG) and called the Cowling-Haagerup constant of bG. By convention, Λcb(bG) = ∞ if bG is not weakly amenable. 4. Free products of weakly amenable discrete quantum groups Recall that given two discrete quantum groups bG and bH, there is a unique compact quantum group structure on the reduced free product C(G)∗ C(H) with respect to the Haar states which is compatible with the canonical inclusions (it is defined in [10]). By analogy with the classical case, the dual of this compact quantum group will be called the reduced free product of bG and bH and denoted bG∗bH. We will now prove that a reduced free product of weakly amenable discrete quantum groups with Cowling-Haagerup constant equal to 1 has Cowling-Haagerup constant equal to 1. This result has been proved in the classical case by E. Ricard and Q. Xu [8, Theorem 4.3] using the following result [8, Prop 4.11]. Theorem 4.1 (Ricard, Xu [8]). Let (Bi, ψi)i∈I be unital C-algebras with distinguished states (ψi) having faithful GNS constructions. Let Ai ⊂ Bi be unital C-subalgebras such that the states ϕi = ψiAi also have faithful GNS construction. Assume that for each i, there is a net of finite rank maps (Vi,j ) on Ai converging to the identity pointwise, preserving the state and such that lim supj kVi,jkcb = 1. Assume moreover that for each pair (i, j), there is a completely positive unital map Ui,j : Ai → Bi preserving the state and such that kVi,j − Ui,jkcb + kVi,j − Ui,jkB(L2(Ai,ϕi),L2(Bi,ψi)) + kVi,j − Ui,jkB(L2(Ai,ϕi)op,L2(Bi,ψi)op) → 0. Then, the reduced free product of the family (Ai, ϕi) has Cowling-Haagerup constant equal to 1. Theorem 4.2. Let (bGi)i∈I be a family of discrete quantum groups with Cowling-Haagerup constant equal to 1. Then Λcb(∗ibGi) = 1. lemmata, most of which are rather technical. We first introduce some general notations. Let bG be a discrete quantum group, let 0 < ǫ < 1 and let a ∈ ℓ∞(bG) be such that kmakcb 6 1 + ǫ. Let ξ, η : L2(G) → L2(G) ⊗ K denote the two maps given by Theorem 3.1 with kξk = kηk 6 √1 + ǫ and set γ = (ξ + η)/2 and δ = (ξ − η)/2. Lemma 4.3. Assume that a = bS(a)∗ and that ma(1) = 1. Then, there exists a u.c.p. map on B(L2(G)) approximating ma up to 6ǫ in completely bounded norm. Proof. We know from [4, Prop 2.6] that for any x ∈ C(G), ξ∗(x ⊗ 1)η = ma(x∗)∗ = m bS(a)∗(x). Thus, The proof of this theorem is quite involved. In order to make it more clear, we will divide it into several Obtaining an approximation by a unital completely positive map is the first step. ma(x) = 1 2 (ma(x) + m bS(a)∗(x)) = 1 2 ((η∗(x ⊗ 1)ξ + ξ∗(x ⊗ 1)η)) = Mγ(x) − Mδ(x), where Mγ(x) = γ∗(x ⊗ 1)γ and Mδ(x) = δ∗(x ⊗ 1)δ. The maps Mγ and Mδ are completely positive, thus kMγkcb = kγk2 6 1 + ǫ and evaluating at 1 gives k1 + δ∗δk 6 1 + ǫ, i.e. kMδkcb = kδ∗δk 6 ǫ. We now want to perturb Mγ into a unital completely positive map. To do this, first note that k1− γ∗γk = kδ∗δk 6 ǫ < 1, which A NOTE ON WEAK AMENABILITY FOR FREE PRODUCTS OF DISCRETE QUANTUM GROUPS 3 implies that γ∗γ is invertible, and set γ = γγ−1 where γ = (γ∗γ)1/2. Note that kγ − γk 6 ǫ. Thus, Mγ is a unital completely positive map and kMγ − Mγkcb = kMγ+(γ−γ) − Mγkcb 6 kγ − γkkγk + kγ − γkkγk + kγ − γkkγ − γk 6 ǫ(2 + 3ǫ) 6 5ǫ. This proves that Mγ is a unital completely positive map approximating ma on C(G) up to 6ǫ in completely bounded norm. (cid:3) Set D = B(L2(G)). We now want to prove that the previous approximation also works when the maps are seen as operators on L2(D, τ ) and L2(D, τ )op, where τ (x) = hξh, x(ξh)i. Let us start with a purely computational lemma. Lemma 4.4. For any ζ ∈ K, (ı ⊗ π)(cW )(ξh ⊗ ζ) = ξh ⊗ ζ. Proof. Note that since by definition W (ξ ⊗ ξh) = ξ ⊗ ξh for any ξ ∈ H, we also have cW (ξh ⊗ ξ) = ξh ⊗ ξ. For any θ1, θ2, ξ ∈ L2(G), Thus by density, we have hξ, (ı ⊗ ω)(cW )ξhi = ω(1)hξ, ξhi for any ω ∈ B(L2(G))∗. Now, let ζ1, ζ2 ∈ K and ξ ∈ L2(G), then hξ, (ı ⊗ ωθ1,θ2)(cW )ξhi = hξ ⊗ θ1,cW (ξh ⊗ θ2)i = hξ, ξhihθ1, θ2i = ωθ1,θ2(1)hξ, ξhi. hξ ⊗ ζ1, (ı ⊗ π)(cW )(ξh ⊗ ζ2)i = hξ, (ı ⊗ ωζ1,ζ2 ◦ π)(cW )ξhi = ωζ1,ζ2(π(1))hξ, ξhi = hξ ⊗ ζ1, ξh ⊗ ζ2i. This gives us a systematic way to investigate the L2-norm of some specific operators. Lemma 4.5. Let T be any bounded linear operator from L2(G) to K and set A(T ) = (ı ⊗ π)(cW )∗(1 ⊗ T )cW (ı ⊗ ξh) ∈ B(L2(G), L2(G) ⊗ K). and MA(T )(x) = A(T )∗(x ⊗ 1)A(T ). Then, τ (MA(T )(x∗x)) 6 kTk2τ (x∗x) and MA(T ) is a bounded operator on L2(D, τ ) of norm less than kTk2. If moreover A(T )∗A(T ) is invertible, then MA(T )A(T )−1 is τ -invariant. Proof. Let us compute (cid:3) A(T )ξh = (ı ⊗ π)(cW )∗(1 ⊗ T )cW (ξh ⊗ ξh) = (ı ⊗ π)(cW )∗(1 ⊗ T )(ξh ⊗ ξh) = (ı ⊗ π)(cW )∗(ξh ⊗ T (ξh)) = ξh ⊗ T (ξh). From this, we get hξh, A(T )∗(x ⊗ 1)A(T )ξhi = hA(T )ξh, (x ⊗ 1)A(T )(ξh)i = hξh, x(ξh)ikT (ξh)k2 and Kadison's inequality yields τ (MA(T )(x)∗MA(T )(x)) 6 kA(T )k2τ (MA(T )(x∗x)) 6 kTk2kA(T )k2τ (x∗x) 6 kTk4τ (x∗x). Let us now turn to A(T )∗A(T ). First, A(T )∗A(T )ξh = (ı ⊗ ξ∗ = (ı ⊗ ξ∗ = (ı ⊗ ξ∗ = (ı ⊗ ξ∗ = hξh, T ∗T (ξh)iξh = kT (ξh)k2ξh h)cW ∗(1 ⊗ T ∗)(ı ⊗ π)(cW )(ξh ⊗ T (ξh)) h)cW ∗(1 ⊗ T ∗)(ξh ⊗ T (ξh)) h)cW ∗(ξh ⊗ T ∗T (ξh)) h)(ξh ⊗ T ∗T (ξh)) 4 AMAURY FRESLON and ξh is an eigenvector for A(T )∗A(T ). If A(T )∗A(T ) is invertible, then (A(T )∗A(T ))−1/2ξh = kT (ξh)k−1ξh. Thus, A(T )A(T )−1ξh = ξh ⊗ kT (ξh)k−1T (ξh) and τ (MA(T )A(T )−1(x)) = hA(T )A(T )−1ξh, (x ⊗ 1)A(T )A(T )−1ξhi = hξh, x(ξh)i = τ (x). Applying Lemma 4.5 to Mδ = A([P − Q]/2), and setting kxk2 = τ (x∗x)1/2 for x ∈ D, we can compute k(ma − Mγ)(x)k2 2 = kMδ(x)k2 2 = τ (Mδ(x)∗Mδ(x)) 6 kδk4kxk2 2 6 ǫ4kxk2 2 i.e. k(ma − Mγ)(x)k2 6 ǫ2kxk2 and Mγ approximates ma up to ǫ2 in B(L2(D, τ )). We now only have to control kMγ − MγkB(L2(D,τ )). Lemma 4.6. τ ((Mγ(x) − Mγ(x))∗(Mγ(x) − Mγ(x)))1/2 6 5ǫτ (x∗x)1/2. Proof. We have γ = A((P + Q)/2). Thus, setting T = ((P + Q)/2) and observing that we can compute, again with Lemma 4.5, τ (Mγ−γ(x∗x)) 6 h(x∗x ⊗ 1)(γ − γ)ξh, (γ − γ)ξhi 1 T ξh(cid:19)(cid:13)(cid:13)(cid:13)(cid:13) = k(γ − γ)ξhk 6 kγ − γk 6 ǫ, T ξh(cid:19)(cid:29) T ξh(cid:19)(cid:19) , ξh ⊗(cid:18)T ξh − 1 kT ξhk (cid:18)T ξh − 1 kT ξhk (cid:13)(cid:13)(cid:13)(cid:13) 1 kT ξhk T ξh(cid:19)(cid:13)(cid:13)(cid:13)(cid:13) =(cid:13)(cid:13)(cid:13)(cid:13)ξh ⊗(cid:18)T ξh − = (cid:28)(x∗x ⊗ 1)(cid:18)ξh ⊗(cid:18)T ξh − = h(x∗x)ξh, ξhi(cid:13)(cid:13)(cid:13)(cid:13)T ξh − 1 kT ξhk 2 kT ξhk T ξh(cid:13)(cid:13)(cid:13)(cid:13) 6 ǫ2τ (x∗x) 6 ǫ2kxk2 2 (cid:3) (cid:3) and kMγ−γ(x)k2 2 = τ (Mγ−γ(x)∗Mγ−γ(x)) 6 kγ − γk2τ (Mγ−γ(x∗x)) 6 ǫ4kxk2 kMγ(x) − Mγ(x)k2 = kMγ(x) − Mγ+(γ−γ)(x)k2 2. Now, for all x ∈ D, we have 6 kMγ−γ(x)k2 + τ ((γ − γ)∗(x∗ ⊗ 1)γγ∗(x ⊗ 1)(γ − γ))1/2 + τ (γ∗(x∗ ⊗ 1)(γ − γ)(γ − γ)∗(x ⊗ 1)γ)1/2 6 ǫ2kxk2 + (1 + ǫ)τ ((γ − γ)∗(x∗ ⊗ 1)(x ⊗ 1)(γ − γ))1/2 + ǫτ (γ∗(x∗ ⊗ 1)(x ⊗ 1)γ)1/2 6 ǫ2kxk2 + (1 + ǫ)τ (Mγ−γ(x∗x))1/2 + ǫτ (Mγ(x∗x))1/2 6 ǫ2kxk2 + (1 + ǫ)ǫkxk2 + ǫ(1 + ǫ)kxk2 6 5ǫkxk2. Finally, k(ma − Mγ)(x)k2 6 6ǫkxk2. We have to chek that this approximation also works in B(L2(D, τ )op), but this is straightforward. Lemma 4.7. Mγ also approximates ma up to 6ǫ in B(L2(D, τ )op). Proof. To estimate the opposite L2-norm, one only needs to do all the previous computations exchanging P and Q. Since they play symmetric rôles, we get the same result. (cid:3) Thus, we are able to approximate ma by a unital completely positive map in all the required norms. There remains only to check the τ -invariance condition. Lemma 4.8. The maps Mγ and ma are τ -preserving. Proof. For Mγ, this comes from Lemma 4.5. For ma, this comes from the following computation τ (ma(x)) = hη∗(x ⊗ 1)ξ(ξh), ξhi = h(x ⊗ 1)ξ(ξh), η(ξh)i = h(x ⊗ 1)A(P )(ξh), A(Q)(ξh)i = h(x ⊗ 1)(ξh ⊗ P (ξh), ξh ⊗ Q(ξh)i = hx(ξh), ξhihP (ξh), Q(ξh)i = τ (x)τ (ma(1)) A NOTE ON WEAK AMENABILITY FOR FREE PRODUCTS OF DISCRETE QUANTUM GROUPS and the fact that we have assumed ma to be unital. We are now ready to prove the theorem. 5 (cid:3) Proof of Theorem 4.2. For each i, set Ai = Cred(Gi) and Bi = B(L2(Gi)). Consider a net (ai,t)t of finitely supported elements in ℓ∞(bGi) converging pointwise to the identity and such that lim supt kmai,tkcb = 1 and note that since bε(ai,t) → 1 (because of the pointwise convergence assumption), we can, up to extracting a suitable subsequence, assume it to be non-zero and divide by it so that mai,t becomes unital. For any 0 < ǫ < 1, there is a t(ǫ) such that kmai,t(ǫ)kcb 6 1 + ǫ (the same being automatically true for m bS(ai,t(ǫ))). Since bS ◦ ∗ ◦ bS ◦ ∗ = ı, we can replace ai,t by (ai,t +bS(ai,t))/2 so that all the hypothesis of Lemma 4.3 are satisfied. Then, by lemmata 4.3, 4.5, 4.6, 4.7 and 4.8 we get a unital completely positive approximation in completely bounded norm and in both L2-norms. Applying Theorem 4.1 proves that Λcb(∗iAi) = 1. Since the original maps were all multipliers, the resulting finite-dimensional approximation is also implemented by multipliers and Λcb(∗i∈IbGi) = 1. Example 1. Let (Gi) be any family of compact groups, then their duals in the sense of quantum groups are amenable (which implies that the Cowling-Haagerup constant is 1 thanks to [9, Thm 3.8]). Thus ∗i(C(Gi)) is the dual of a non-cocommutative discrete quantum group with Cowling-Haagerup constant equal to 1. Example 2. The free orthogonal quantum groups \Ao(F ) have Cowling-Haagerup constant equal to 1 for any F ∈ GL(2, C) such that F F = Id since they are amenable. Moreover, for any F ∈ GL(2, C) the free unitary quantum group \Au(F ) is a quantum subgroup of Z ∗ \Ao(F ), thus Λcb( \Au(F )) = 1. It follows that any reduced free product of some 2-dimensional free quantum groups with duals of compact groups has Cowling-Haagerup constant equal to 1. (cid:3) References 1. S. Baaj and G. Skandalis, Unitaires multiplicatifs et dualité pour les produits croisés de C-algèbres, Ann. Sci. École Norm. Sup. 26 (1993), no. 4, 425 -- 488. 2. Marek Bozejko and Gero Fendler, Herz-Schur multipliers and completely bounded multipliers of the Fourier algebra of a locally compact group, Boll. Un. Mat. Ital. A (6) 3 (1984), no. 2, 297 -- 302. 3. M. Daws, Multipliers of locally compact quantum groups via Hilbert C-modules, J. Lond. Math. Soc. 84 (2012), 385 -- 407. 4. J. Kraus and Z-J. Ruan, Approximation properties for Kac algebras, Indiana Univ. Math. J. 48 (1999), no. 2, 469 -- 535. 5. N. Ozawa, Weak amenability of hyperbolic groups, Groups Geom. Dyn. 2 (2008), 271 -- 280. 6. N. Ozawa and S. Popa, On a class of II1 factors with at most one Cartan subalgebra, Ann. of Math. 172 (2010), no. 1, 713 -- 749. 7. 8. E. Ricard and Q. Xu, Khintchine type inequalities for reduced free products and applications, J. Reine Angew. Math. 599 , On a class of II1 factors with at most one Cartan subalgebra II, American J. Math. 132 (2010), no. 3, 841 -- 866. (2006), 27 -- 59. 9. R. Tomatsu, Amenable discrete quantum groups, J. Math. Soc. Japan 58 (2006), no. 4, 949 -- 964. 10. S. Wang, Free products of compact quantum groups, Comm. Math. Phys. 167 (1995), no. 3, 671 -- 692. 11. S.L. Woronowicz, Compact quantum groups, Symétries quantiques (Les Houches, 1995) (1998), 845 -- 884. E-mail address: [email protected] URL: http://www.math.jussieu.fr/~freslon Univ. Paris Diderot, Paris Cité Sorbonne, IMJ, UMR 7586, 175 rue du Chevaleret, 75013, Paris, France
1703.01543	1	1703	2017-03-05T01:40:28	Isometries of perfect norm ideals of compact operators	[ "math.OA" ]	It is proved that for every surjective linear isometry $V$ on a perfect Banach symmetric ideal $\mathcal C_E\neq \mathcal C_2$ of compact operators, acting in a complex separable infnite-dimensional Hilbert space $\mathcal H$ there exist unitary operators $u$ and $v$ on $\mathcal H$ such that $V(x)=uxv$ or $V(x) = ux^tv$ for all $x\in \mathcal C_E$, where $x^t$ is the transpose of an operator $x$ with respect to a fixed orthonormal basis in $\mathcal H$. In addition, it is shown that any surjective 2-local isometry on a perfect Banach symmetric ideal $\mathcal C_E \neq \mathcal C_2$ is a linear isometry on $\mathcal C_E$.	math.OA	math	ISOMETRIES OF PERFECT NORM IDEALS OF COMPACT OPERATORS BEHZOD AMINOV AND VLADIMIR CHILIN Abstract. It is proved that for every surjective linear isometry V on a perfect Banach symmetric ideal CE 6= C2 of compact operators, acting in a complex separable infinite-dimensional Hilbert H there exist unitary operators u and v on H such that V (x) = uxv or V (x) = uxtv for all x ∈ CE , where xt is a transpose of an operator x with respect to a fixed orthonormal basis for H. In addition, it is shown that any surjective 2-local isometry on a perfect Banach symmetric ideal CE 6= C2 is a linear isometry on CE . 1. Introduction Let H be a complex separable infinite-dimensional Hilbert space. Let (E, k·kE) ⊂ c0 be a real Banach symmetric sequence space. Consider an ideal CE of compact linear operators in H, which is defined by the relations and x ∈ CE ⇐⇒ {sn(x)}∞ n=1 ∈ E kxkCE = k{sn(x)}∞ n=1kE, n=1 are the singular values of x (i.e. the eigenvalues of (x∗x)1/2 in where {sn(x)}∞ decreasing order). In the paper [10] it is shown that k · kCE is a Banach norm on CE. In addition, C1 := Cl1 ⊂ CE. In the case when (E, k · kE) is a separable Banach space, the Banach ideal (CE, k · kCE ) is a minimal Banach ideal in the terminology of Schatten [16], i.e. the set of finite rank operators is dense in CE. It is known [18] that for every surjective linear isometry V on a minimal Banach ideal CE 6= C2, where C2 := Cl2, there exist unitary operators u and v on H such that V (x) = uxv (or V (x) = uxtv) (1) for all x ∈ CE, where xt is the transpose of the operator x with respect to a fixed orthonormal basis in H. In the case of a Banach ideal C1, a description of surjective linear isometries of the form of (1) was obtained in [15], and for the ideals Cp := Clp , 1 ≤ p ≤ ∞, p 6= 2, in [2]. In this paper we show that, in the case a Banach symmetric sequence space E 6= l2 with Fatou property, every surjective linear isometry V on CE has the form (1). In addition, it is proved that in this case any 2-local surjective isometry on CE also is of the form (1). Date: March 4, 2017. 2010 Mathematics Subject Classification. 46L52(primary), 46B04 (secondary). Key words and phrases. Banach symmetric ideal of compact operators, Fatou property, Her- mitian operator, 2-local isometry. 1 2 BEHZOD AMINOV AND VLADIMIR CHILIN 2. Preliminaries Let l∞ (respectively, c0) be a Banach lattice of all bounded (respectively, con- n=1 of real numbers with respect to the norm n=1 ∈ l∞, ξn, where N is the set of natural numbers. If x = {ξn}∞ verging to zero) sequences {ξn}∞ k{ξn}k∞ = sup n∈N then a non-increasing rearrangement of x is defined by x∗ = {ξ∗ n}, where ξ∗ n := inf F <n sup n /∈F ξn, F is a finite subset of N. A non-zero linear subspace E ⊂ l∞ with a Banach norm k · kE is called Banach symmetric sequences space if the conditions y ∈ E, x ∈ l∞, x∗ ≤ y∗, imply that x ∈ E and kxkE ≤ kykE. In this case, the inequality kxk∞ ≤ kxkE follows for each x ∈ E. Let H be a complex separable infinite-dimensional Hilbert space and let B(H) (respectively, K(H), F (H)) be the ∗-algebra of all bounded (respectively, compact, finite rank) linear operators in H. It is well known that F (H) ⊂ I ⊂ K(H) for any proper two-sided ideal I in B(H) (see for example, [17, Proposition 2.1]). If (E, k · kE) ⊂ c0 is a Banach symmetric sequence space, then the set CE := {x ∈ K(H) : {sn(x)}∞ n=1 ∈ E} is a proper two-sided ideal in B(H) ([3]; see also [17, Theorem 2.5]). In addition, (CE, k · kCE )) is a Banach space with respect to the norm kxkCE = k{sn(x)}∞ n=1kE [10], and the norm k · kCE has the following properties 1) kxzykCE ≤ kxk∞kyk∞kzkCE for all x, y ∈ B(H) and z ∈ CE, where k · k∞ is the usual operator norm in B(H); 2) kxkCE = kxk∞ if x ∈ CE is of rank 1. In this case we say that (CE, k · kCE )) is a Banach symmetric ideal (cf. [9, Ch. It is clear that kuxvkCE = kxkCE for all unitary operators III], [17, Ch. 1, §1.7]). u, v ∈ B(H) and x ∈ CE. Besides, C1 ⊂ CE ⊂ K(H) and kxkCE ≤ kxkC1, kyk∞ ≤ kykCE for all x ∈ C1, y ∈ CE. A Banach symmetric sequence space (E, k·kE) (respectively, a Banach symmetric ideal (CE, k · kCE)) is said to have order continuous norm if kakkE ↓ 0 (respectively, kakkCE ↓ 0) whenever ak ∈ E (respectively, ak ∈ CE) and ak ↓ 0. It is known that (CE, k·kCE )) has order continuous norm if and only if (E, k·kE) has order continuous norm (see, for example, [5, Proposition 3.6]). Moreover, every Banach symmetric ideal with order continuous norm is a minimal norm ideal, i.e. the subspace F (H) is dense in CE. If (E, k · kE) is a Banach symmetric sequence space (respectively, (CE, k · kCE ) is E ) is defined a Banach symmetric ideal), then the Kothe dual E× (respectively, C× as E× = {ξ = {ξn}∞ n=1 ∈ l∞ : ξη ∈ l1 for all η ∈ E} (respectively, C× E = {x ∈ B(H) : xy ∈ C1 for all y ∈ CE}) and kξkE× = sup{ ∞ X n=1 ξnηn : η = {ηn}∞ n=1 ∈ E, kηkE ≤ 1} if ξ ∈ E× ISOMETRIES OF PERFECT NORM IDEALS OF COMPACT OPERATORS 3 (respectively, kxkC× E = sup{tr(xy) : y ∈ CE, kykCE ≤ 1} if x ∈ C× E ), where tr(·) is the standard trace on B(H). It is known that (E×, k · kE×) (respectively, (C× sequence space (respectively, a Banach symmetric ideal). In addition, C× and if CE 6= C1, then C× useful property [5, Theorem 5.6]: ) is a Banach symmetric 1 = B(H) E ⊂ K(H) [7, Proposition 7]. We also note the following E , k · kC× E (C× E , k · kC× ) = (CE×, k · kCE× ). E A Banach symmetric ideal CE is said to be perfect if CE = C×× is perfect if and only if E = E××. E . It is clear that CE A Banach symmetric sequence space (E, k·kE) (respectively, a Banach symmetric ideal (CE, k · kCE )) is said to possess Fatou property if the conditions 0 ≤ ak ≤ ak+1, ak ∈ E (respectively, ak ∈ CE) for all k ∈ N and sup k≥1 kakkE < ∞ (respectively, sup k≥1 kakkCE < ∞) imply that there exists a ∈ E (a ∈ CE) such that ak ↑ a and kakE = sup k≥1 kakkE (respectively, kakCE = sup k≥1 kakkCE ). It is known that (E, k · kE) (respectively, (CE, k · kCE )) has the Fatou property E [5, if and only if E = E×× [12, Vol. II, Ch. 1, Section a] (respectively, CE = C×× Theorem 5.14]). Therefore CE is perfect ⇔ CE = C×× E ⇔ E = E×× ⇔ E has the Fatou property ⇔ CE has Fatou property. If y ∈ C× = kykC× E E , where C∗ E , then a linear functional fy(x) = tr(xy) = tr(yx), x ∈ CE, is continuous E is the dual of the Banach space E and the linear E. Since E. Thus, the weak topology on CE. In addition, kfykC∗ (CE, k · kCE ) (see, for example, [7]). Identifying an element y ∈ C× functional fy, we may assume that C× F (H) ⊂ C× σ(CE, C× Proposition 1. [4, Proposition 2.8] A linear functional f ∈ C∗ respect to weak topology σ(CE, C× sequence xn ↓ 0, xn ∈ CE, n ∈ N. E is continuous with E ) if and only if f (xn) → 0 for every E is a closed linear subspace in C∗ E is a total subspace in C∗ E ) is a Hausdorff topology. E , it follows that C× E ) (i.e. f ∈ C× It is clear that Ch E = {x ∈ CE : x = x∗} is a real Banach space with respect to E and E, n ∈ N, then there exists E is a quasi-(o)-complete space [1]. Hence, by [1, Theorem E . If 0 ≤ xn ≤ xn+1 ≤ x, x, xn ∈ Ch E, i.e. Ch the norm k · kCE ; besides, the cone C+ E − C+ E = C+ Ch supn≥1 xn ∈ Ch 1], we have the following proposition. E : x ≥ 0} is closed in Ch E = {x ∈ Ch Proposition 2. Every linear functional f ∈ (Ch linear functionals from (Ch E)∗. E)∗ is a difference of two positive We need the following property of the weak topology σ(CE, C× E ). Proposition 3. If x ∈ CE, then there exists a sequence {xn} ⊂ F (H) such that σ(CE ,C× E ) −→ x. xn 4 BEHZOD AMINOV AND VLADIMIR CHILIN Proof. It suffices to establish the validity of Proposition 3 for 0 ≤ x = ∞ X j=1 λjpj ∈ CE \ F (H), where λj ≥ 0 are the eigenvalues of the compact operator x and pj ∈ F (H) are finite-dimensional projectors for all j ∈ N (the series converges with respect to the norm k · k∞). If yn = λjpj, then yn ↓ 0 and, by Proposition, 1 we have ∞ P j=n f (yn) → 0 for all f ∈ C× E . Therefore F (H) ∋ x − yn σ(CE ,C× E ) −→ x. (cid:3) Let T be a bounded linear operator acting in a Banach symmetric ideal (CE, k · kCE ), and let T ∗ be its adjoint operator. If T ∗(C× E ) ⊂ C× E , then f (T (xα)) = T ∗(f )(xα) → T ∗(f )(x) = f (T (x)) σ(CE ,C× E ) for every net xα continuous operator. −→ x, xα, x ∈ CE, and for all f ∈ C× E . Thus T is a σ(CE, C× E )- Inversely, if the operator T is a σ(CE, C× E )-continuous and f ∈ C× E , then the linear functional (T ∗f )(x) = f (T (x)) is also σ(CE, C× E )-continuous, i.e. T ∗(f ) ∈ C× E . Therefore, a linear bounded operator T : CE → CE is σ(CE, C× and only if T ∗(f ) ∈ C× operators T, S : CE → CE, continuous with respect to a σ(CE, C× operators T + S, T S and λT are σ(CE, C× E for every f ∈ C× E )-continuous if E . Thus, for any λ ∈ C and linear bounded E )-topology, the E )-continuous. Let B(CE) be the Banach space of bounded linear operators T acting in (CE, k · kCE ) with the norm kT kB(CE) = sup kxkCE ≤1 kT (x)kCE . Proposition 4. If Tn ∈ B(CE) are σ(CE, C× and kTn − T kB(CE) → 0 as n → ∞, then T is also σ(CE, C× E )-continuous operators, T ∈ B(CE), Proof. It suffices to show that T ∗(f ) ∈ C× Tn ∈ B(CE) are σ(CE, C× all f ∈ C× that E , n ∈ N. Considering the closed subspace C× E )-continuous operators, it follows that T ∗ E, k · kC∗ E in (C∗ E n(f ) ∈ C× E . Since E for ) and noting E )-continuous. E for any functional f ∈ C× kT ∗ n(f ) − T ∗(f )kC∗ E ≤ kT ∗ n − T ∗kB(C∗ E )kf kC∗ E we conclude that T ∗(f ) ∈ C× E . = kTn − T kB(CE)kf kC∗ E → 0, (cid:3) We also need the following well-known properties of perfect symmetrically normed ideals. Theorem 1. Let (CE, k · kCE ) ⊂ c0 be a Banach symmetric sequence space with Fatou property. Then (i). [5, Theorem 5.11]. kxkCE = sup E ,kyk y∈C× C × E tr(xy) for every x ∈ CE; ≤1 ISOMETRIES OF PERFECT NORM IDEALS OF COMPACT OPERATORS 5 (ii). [4, Theorem 3.5]. A Banach space CE is σ(CE, C× i.e. if for xn ∈ CE, n ∈ N, and for every f ∈ C× E )-sequentially complete, f (xn), then there E there is lim n→∞ exists x ∈ CE such that xn σ(CE ,C× E ) −→ x. 3. The ball topology in norm ideals of compact operators Let (X, k · kX) be a real Banach space and let bX be the ball topology in X, i.e. bX is the coarsest topology such that every closed ball B(x, ε) = {y ∈ X : ky − xkX ≤ ε}, ε > 0, is closed in bX [8]. The family X \ n [ i=1 B(xi, εi), xi ∈ X, kx0 − xkX > εi, i = 1, ..., n, n ∈ N. bX−→ x, xα, x ∈ is a base of neighborhoods of the point x0 ∈ X in bX . Therefore xα X, if and only if lim inf kxα − xkX ≥ kx − ykX for all y ∈ X [8]. In particular, every surjective isometry V in (X, k · kX ) is continuous with respect to the ball topology bX. Let us note that bX is not a Hausdorff topology. The following theorem provides a sufficient condition for T2-axiom of bX on subsets of X. Recall that A is a Rosenthal subset of (X, k · kX ) if every sequence in A has a weakly Cauchy subsequence. Theorem 2. [8, Theorem 3.3]. Let (X, k · kX) be a real Banach space, and let A be a bounded absolutely convex Rosenthal subset of X. Then (A, bX ) is a Hausdorff space. In the proof of required properties of the ball topology (see Theorem 3 below), we utilize the following well-known proposition. Proposition 5. [14, section 3, Ch. 1, §5]. Let pnk be real numbers, n, k ∈ N, such pnk = 1 for all n ∈ N, and let the limit lim n→∞ pnk = pk exist for every fixed that n P k=1 k ∈ N. Then the sequence sn = pn1r1 + pn2r2 + . . . + pnnrn converges for every convergent sequence {rn}. Theorem 3. Let (CE, k · kCE ) be a Banach symmetric ideal, xn ∈ C+ E , and let xn ↓ 0. Then the sequence {xn} can converge with respect to the topology bCE to no more than one element. Proof. Consider CE as a real Banach space. Let A be the absolutely convex hull of the sequence {0, x1, ..., xn, ...}. Since xn ↓ 0 and the norm k · kCE is monotone, it follows that A is a bounded subset of CE. Let us show that A is a Rosenthal subset. Using the inequality 0 ≤ xn+1 ≤ xn and a decomposition of linear functional E)∗ (see Proposition E)∗. E)∗ as the difference of two positive functionals from (Ch f (xn) = rn for every functional f ∈ (Ch f ∈ (Ch 2), we conclude that there exists lim n→∞ For any sequence {yn}∞ n=1 ⊂ A, we have yn = pn1x1 + pn2x2 + . . . + pnk(n)xk(n), where k(n) X i=1 pni = 1; 6 BEHZOD AMINOV AND VLADIMIR CHILIN in particular, pni ∈ [−1, 1] for all i ∈ 1, . . . , k(n). Consider a sequence qn = (pn1, pn2 . . . , pnk(n), 0, 0 . . .) ∈ ∞ Y i=1 [−1, 1]. ∞ Q i=1 By Tychonoff Theorem, a set ∞ Q i=1 [−1, 1] is compact with respect to the product topology. Moreover, by [11, ch. 4, theorem 17], the set [−1, 1] is a metrizable compact. Hence, the sequence {qn} has a convergent subsequence {qni}∞ particular, there are the limits In pnik for all k ∈ N. Besides, the sequence i=1. lim ni→∞ f (yni) = k P j=1 pnijf (xj), f ∈ (Ch E)∗, satisfies all conditions of Proposition 5, which implies its convergence. Therefore, the sequence {yn}∞ subsequence {yni}. This means that A is a Rosenthal subset of (Ch n=1 ⊂ A has a weakly Cauchy E, k · kCE ). By Theorem 2, the topological space (A, bCh ) is Hausdorff, hence the sequence n=1 ⊂ A could not have more than one limit with respect to the topology bCh . E is a closed subspace in (CE, k · kCE ) (we assume that CE is a real space), . Therefore, the sequence (cid:3) {xn}∞ Since Ch it follows that the restriction (bCE )Ch {xn}∞ n=1 can have no more than one limit with respect to the topology bCE . is finer than bCh E E E E Proposition 6. If E ⊂ c0 is a Banach symmetric sequence space with Fatou property, then bCE ≤ σ(CE, C× E ). Proof. It suffices to show that B(0, 1) = {x ∈ CE : kxkCE ≤ 1} is closed with respect to the weak topology σ(CE, C× −→ x ∈ CE. Assume that x /∈ B(0, 1), i.e. kxkCE = q > 1, and let ǫ > 0 be such that q − ǫ > 1. By Theorem 1 (i), there exists y ∈ C× ≤ 1 and q ≥ tr(xy) > q −ǫ. E ). Let xα ∈ B(0, 1) and xα σ(CE ,C× E ) E such that kykC× E On the other hand, xn −→ x implies that tr(xny) → tr(xy). Since xn ∈ B(0, 1), it follows that tr(xny) ≤ 1 and therefore tr(xy) ≤ 1, which is impossible. Thus B(0, 1) is a closed set in σ(CE, C× (cid:3) E ). σ(CE ,C× E ) 4. Weak continuity of Hermitian operators in the perfect ideals of compact operators In this section, we establish σ(CE, C× E )-continuity of the Hermitian operator act- ing in a perfect Banach symmetric ideal CE of compact operators. For that we need σ(CE, C× E )-continuity of a surjective isometry on CE. Recall that the series xn converges weakly unconditionally in a Banach space ∞ P n=1 ∞ P n=1 X if a numerical series §3]. f (xn) converges absolutely for every f ∈ X ∗ [19, Ch. 2, Proposition 7. [19, Ch. 2, §3]. Let (X, k · kX ) be a Banach space, xn ∈ X, n ∈ N. Then the following conditions are equivalent: (i) A series xn converges weakly unconditionally; ∞ P n=1 ISOMETRIES OF PERFECT NORM IDEALS OF COMPACT OPERATORS 7 (ii) There exists a constant C > 0, such that k sup N N X n=1 tnxnkX ≤ Ck{tn}∞ n=1k∞ for all {tn}∞ n=1 ∈ l∞. Proposition 7 implies the following. Corollary 4. If V is a surjective linear isometry on a Banach space X and a series xn converges weakly unconditionally in X, then the series V (xn) also ∞ P n=1 ∞ P n=1 converges weakly unconditionally in X. Now we can show that every surjective linear isometry of CE is σ(CE, C× E )- continuous. Proposition 8. Let CE be a perfect Banach symmetric ideal of compact operators and V a surjective linear isometry on CE. Then V is σ(CE, C× E )-continuous. Proof. It suffices to show that V ∗(f ) ∈ C× to Proposition 1, it should be established that xn ↓ 0, {xn}∞ f (V (xn)) = V ∗(f )(xn) → 0. E for each functional f ∈ C× n=1 ⊂ C+ E . According E , implies Let {xn}∞ n=1 ⊂ CE and xn ↓ 0. We will show that the series (xn − xn+1) ∞ P n=1 converges weakly unconditionally in CE. If f ∈ (Ch then E)∗ is a positive linear functional, m X n=1 f (xn − xn+1) = f ( (xn − xn+1)) = f (x1 − xm+1) ≤ f (x1) m X n=1 for all m ∈ N. Hence the numerical series f (xn − xn+1) converges absolutely. Since every functional f ∈ (Ch (Ch E)∗ (see Proposition 2), the series E)∗ is the difference of two positive functionals from (xn − xn+1) converges weakly uncondition- ∞ P n=1 ally in Ch E. Let f ∈ C∗ E. Denote u(x) = Ref (x) = f (x) + f (x) 2 , v(x) = Imf (x) = f (x) − f (x) 2i , x ∈ CE. It is clear that u, v ∈ (Ch E)∗. Therefore, the series ∞ P n=1 u(xn − xn+1) and ∞ P n=1 v(xn − xn+1) converge absolutely. Thus, the series f (xn − xn+1) also converges abso- ∞ P n=1 ∞ P n=1 lutely. This means that a series in CE. By Corollary 4, the series ∞ P n=1 (xn − xn+1) converges weakly unconditionally ∞ X n=1 (V xn − V xn+1) 8 BEHZOD AMINOV AND VLADIMIR CHILIN converges weakly unconditionally in CE. Hence the numerical series ∞ X n=1 f (V xn − V xn+1) = f (V x1) − lim n→∞ f (V xn) converges for every f ∈ C∗ is a σ(CE, C× f (V xn) exists. Since CE E )-sequentially complete set (see Theorem 1(ii)) and V is a bijection, E. In particular, the limit lim n→∞ it follows that there exists x0 ∈ CE such that V (xn) σ(CE ,C× E ) −→ V (x0) as n → ∞. It remains to show that V (x0) = 0. By Proposition 6, V (xn) bCE−→ V (x0). Since the isometry V −1 is continuous with respect to the topology bCE , it follows that xn bCE−→ x0. Now, taking into account that xn ↓ 0, we obtain xn −→ 0 bCE−→ 0, and Theorem 3 implies that σ(CE ,C× E ) (cid:3) (see Proposition 1). Then, by Proposition 6, xn x0 = 0. Let (X, k·kX) be a complex Banach space. A bounded linear operator T : X → X is called Hermitian, if the operator eitT = ∞ P n=0 X for all t ∈ R (see, for example, [6, Ch. 5, §2]). (itT )n n! is an isometry of the space We show that a Hermitian operator acting in a perfect Banach symmetric ideal CE is σ(CE, C× E )-continuous. Theorem 5. Let E ⊂ c0 be a Banach symmetric sequence space with Fatou prop- erty, and let T be a Hermitian operator acting in CE . Then T is σ(CE, C× E )- continuous. Proof. Consider the non-negative continuous function α(t) = keitT −IkB(CE), where t ∈ R. As α(0) = 0, it follows that there exists t0 ∈ R such that α(t0) < 1. Since the operator T is Hermitian, it follows that the operator V = eit0T (and hence V −1 = e−it0T ) is an isometry on CE. By Proposition 8, S = V − I is a σ(CE, C× E )- continuous operator. In addition, kSk = α(t0) < 1. Now, since ∞ X n=1 Proposition 4 implies that it0T is a σ(CE, C× is a σ(CE, C× E )-continuous operator. it0T = ln(I + S) = (−1)n−1Sn n , E )-continuous operator. Therefore, T (cid:3) If a, b are self-adjoint operators in B(H), x ∈ CE, and (2) T (x) = ax + xb for all x ∈ CE, then T is a Hermitian operator acting in CE [18]. In the following theorem, utilizing the method of proof of Theorem 1 in [18], we show that every Hermitian operators acting in a perfect norm ideal (CE, k · kCE ) 6= C2 has the form (2). Theorem 6. Let E ⊂ c0 be a Banach symmetric sequence space with Fatou prop- erty, E 6= l2, and let T be a Hermitian operator acting in a Banach symmetric ideal CE. Then there are self-adjoint operators a, b ∈ B(H) such that T (x) = ax+xb for all x ∈ CE. ISOMETRIES OF PERFECT NORM IDEALS OF COMPACT OPERATORS 9 Proof. As in the proof of Theorem 1 from [18], we have that there are self-adjoint operators a, b ∈ B(H) such that T (x) = ax + xb for all x ∈ F (H). Fix x ∈ CE. By −→ x. If y ∈ C× E , Proposition 3, there is a sequence {xn} ⊂ F (H) such that xn then σ(CE ,C× E ) tr(y(axn + xnb)) = tr((ya)xn) + tr((by)xn) → tr((ya)x) + tr((by)x) = = tr(y(ax) + tr(y(xb))) = tr(y(ax + xb)). Therefore axn + xnb σ(CE ,C× E ) −→ ax + xb. Since T is a Hermitian operator, it follows that T is σ(CE, C× Theorem 5). Therefore E )-continuous (see T (x) = σ(CE, C× E ) − lim n→∞ T (xn) = σ(CE, C× E ) − lim n→∞ (axn + xnb) = ax + xb. (cid:3) 5. Isometries of a Banach symmetric ideal In this section, we prove our main result, Theorem 7. The proof of Theorem 7 is similar to the proof of Theorem 2 in [18]. We use a version of Theorem 1 in [18] for a perfect Banach symmetric ideal CE (Theorem 6) as well as σ(CE, C× E )-continuity of every isometry on CE (Proposition 8) and σ(CE, C× E )-density of the space F (H) in CE (Proposition 3). Recall that xt stands for the transpose of an operator x ∈ K(H) with respect to a fixed orthonormal basis in H. Theorem 7. Let E ⊂ c0 be a Banach symmetric sequence space with Fatou prop- erty, E 6= l2, and let V be a surjective linear isometry on the Banach symmetric ideal CE. Then there are unitary operators u and v on H such that (3) for all x ∈ CE V (x) = uxv (or V (x) = uxtv) Note that each linear operator of the form (3) is an isometry on every Banach symmetric ideal CE. Proof. Let y ∈ B(H), and let ly(x) = yx (respectively, ry(x) = xy) for all x ∈ CE. It is clear that ly and ry are bounded linear operators acting in CE. Using Theorem 6 and repeating the proof of the Theorem 2 [18], we conclude that there are unitary operators u, v ∈ B(H) such that V lyV −1 = luyu∗ and V ryV −1 = rv∗yv for any y ∈ B(H). In the case V lyV −1 = luyu∗ , we define an isometry V0 on CE by the equation V0(x) = u∗V (x)v∗. As in the proof of Theorem 2 [18], we get that V0(x) = λx for every x ∈ F (H) and some λ ∈ C. Since the isometry V0 is σ(CE, C× E )-continuous (Proposition 8) and the space F (H) is σ(CE, C× E )- dense in CE (Proposition 3), it follows that V0(x) = λx for every x ∈ CE. Now, since V0 is an isometry on CE, we have λ = 1, i.e. V (x) = uxv for all x ∈ CE. In the case V lyV −1 = rv∗yv, as in the proof of Theorem 2 [18], we use the above to derive that there are unitary operators u, v ∈ B(H) such that V (x) = uxtv for all x ∈ CE. (cid:3) 10 BEHZOD AMINOV AND VLADIMIR CHILIN Corollary 8. Let E ⊂ c0 be a Banach symmetric sequence space with Fatou prop- erty, E 6= l2, and let V be a surjective linear isometry on CE. Then V (yx∗y) = V (y)(V (x))∗V (y) for all x, y ∈ CE. Proof. By Theorem 7, there are unitary operators u and v on H such that V (x) = uxv or V (x) = uxtv for all x ∈ CE. If V (x) = uxv, then V (y)(V (x))∗V (y) = (uyv)(uxv)∗(uyv) = uyx∗yv = V (yx∗y), x, y ∈ CE. Since (xt)∗ = (x∗)t, x ∈ B(H), in the case V (x) = uxtv, we have V (y)(V (x))∗V (y) = (uytv)(uxtv)∗(uytv) = = uyt(x∗)tytv = u(y(x∗)y)tv = V (yx∗y), x, y ∈ CE. (cid:3) Let (X, k · kX) be an arbitrary complex Banach space. A surjective (not nec- essarily linear) mapping T : X → X is called a surjective 2-local isometry [13], if for any x, y ∈ X there exists a surjective linear isometry Vx,y on X such that T (x) = Vx,y(x) and T (y) = Vx,y(y). It is clear that every surjective linear isometry on X is automatically a surjective 2-local isometry on X. In addition, T (λx) = Vx,λx(λx) = λVx,λx(x) = λT (x) for any x ∈ X and λ ∈ C. In particular, (4) T (0) = 0. Thus, in order to establish linearity of a 2-local isometry T , it is sufficient to show that T (x + y) = T (x) + T (y) for all x, y ∈ X. Note also that (5) kT (x) − T (y)kX = kVx,y(x) − Vx,y(y)kX = kx − ykX for any x, y ∈ X. Therefore, in the case a real Banach space X, from (4), (5) and Mazur-Ulam Theorem (see, for example, [6, Chapter I, §1.3, Theorem 1.3.5.]) it follows that every surjective 2-local isometry on X is a linear. For complex Banach spaces, this fact is not valid. Using the description of isometries on a minimal Banach symmetric ideal CE from [18], L. Molnar proved that every surjective 2-local isometry on a minimal Banach symmetric ideal CE is necessarily linear [13, Corallary 5]. The following Theorem is a version of Molnar's result for a perfect Banach sym- metric ideal. Theorem 9. Let E ⊂ c0 be a Banach symmetric sequence space with Fatou prop- erty, E 6= l2, and let V be a surjective 2-local isometry on a Banach symmetric ideal CE. Then V is a linear isometry on CE. Proof. Fix x, y ∈ F (H) and let Vx,y : CE → CE be a surjective isometry such that V (x) = Vx,y(x) and V (y) = Vx,y(y). By Theorem 7, there are unitary operators u and v on H such that V (x) = uxv or V (x) = uxtv (respectively, V (y) = uyv or V (y) = uytv). Then we have tr(V (x)(V (y))∗) = tr(Vx,y(x)(Vx,y(y))∗) = tr(xy∗) for every x, y ∈ F (H). In addition, V (x) ∈ F (H) and V is a bijective mapping on F (H). ISOMETRIES OF PERFECT NORM IDEALS OF COMPACT OPERATORS 11 If x, y, z ∈ F (H), then tr(V (x + y)(V (z))∗) = tr((x + y)z∗), tr(V (x)V (z)∗) = tr(xz∗), tr(V (y)V (z)∗) = tr(yz∗). Therefore tr((V (x + y) − V (x) − V (y))(V (z))∗) = 0 for all z ∈ F (H). Taking z = x + y, z = x and z = y, we obtain tr((V (x + y) − V (x) − V (y)((V (x + y) − V (x) − V (y))∗) = 0, that is, V (x + y) = V (x) + V (y) for all x, y ∈ F (H). Hence V is a bijective linear isometry on a normed space (F (H), k · kCE ). Let I be the closure of the subspace F (H) in the Banach space (CE, k · kCE ). It is clear that (I, k · kCE ) is a minimal Banach symmetric ideal. Since V is a surjective linear isometry on (F (H), k · kCE ), it follows that V is a surjective linear isometry on (I, k · kCE ). By Theorem 2 [18], there are unitary operators u1 and v1 on H such that V (x) = u1xv1 or V (x) = u1xtv1 for all x ∈ I. Repeating the ending of the proof of Theorem 1 in [13], we conclude that V is a linear isometry on CE. (cid:3) References [1] T. Ando. On fundamental properties of Banach space with a cone. Pacific J. Math. 12(1962), 1163-1169 [2] J. Arazy. The isometries of Cp. Israel J. Math. 22(1975), 247-256. [3] J. Calkin, Two-sided ideals and congruences in the ring of bounded operators in Hilbert space, Ann. of Math. 1941. V 42. No 2. P. 839-873. [4] P.G.Dodds, C.J.Lennard. Normality in trace ideals. I. Operator theory. V. 16. 1986. P. 127- 145. [5] P. G. Dodds, T. K. Dodds, and B. Pagter, Noncommutative Kothe duality, Trans. Amer. Math. Soc. 339 (2)(1993), 717-750. [6] R.J.Fleming, J.E.Jamison. Isometries on Banach spaces: function spaces. Chapman- Hall/CRC. 2003. [7] D.J.H.Garling. On ideals of operators in hilbert space. Proc. London Math. Soc. V. 17. 1967. P. 115-138. [8] G.Godefroy, N.J.Kalton, The ball topology and its applications. -- Contemp. Math. 85 (1989), 195 -- 237. [9] I.C. Gohberg , M.G.Krein. Introduction to the theory of linear nonselfadjoint operators . Translations of mathematical monographs. Volume 18. AMERICAN MATHEMATICAL SO- CIETY.Providence, Rhode Island 02904. 1969 [10] N.J. Kalton, F.A. Sukochev, Symmetric norms and spaces of operators, J. Reine Angew. Math., 2008. V. 621. P. 81-121. [11] L.Kelley, General Topology. Graduate Texts in Mathematics. Vol. 27. Springer-Verlag New York. 1975. [12] J.Lindenstrauss, L.Tzafriri. Classical Banach spaces. Springer-Verlag. Berlin-New York. 1996. [13] L.Molnar. 2-local isometries of some operator algebras. Proc.Eding.Math.Soc. V.45. 2002. P. 349-352. [14] G.Polya and G.Szego. Problems and theorems in analysis. Die Grundlehren der math. Wis- senschaften. Springer-Verlag. Berlin and New York. Vol. I. 1972. [15] B. Russo. Isometries of the trace class. Proc. Amer. Math. Soc. 23 (1969). 213. [16] R. Schatten, Norm Ideals of Completely Continuous Operators. Springer-Verlag. Berlin. 1960. [17] B. Simon, Trace ideals and their applications, Second edition. Mathematical Surveys and Monographs, 120. American Mathematical Society, Providence, RI, 2005. [18] A.Sourour, Isometries of norm ideals of compact operators. J.Funct.Anal. V. 43. 1981. P. 69-77 [19] P. Wojtaszczyk. Banach spaces for analysts. Cambridge Univercity Press. 1991. 12 BEHZOD AMINOV AND VLADIMIR CHILIN National University of Uzbekistan, Tashkent, 700174, Uzbekistan E-mail address: [email protected], [email protected] National University of Uzbekistan, Tashkent, 700174, Uzbekistan E-mail address: [email protected], [email protected]
1702.04112	2	1702	2017-03-08T09:05:41	The closure of ideals of $\boldsymbol{\ell^1(\Sigma)}$ in its enveloping $\boldsymbol{\mathrm{C}^\ast}$-algebra	[ "math.OA", "math.FA" ]	If $X$ is a compact Hausdorff space and $\sigma$ is a homeomorphism of $X$, then an involutive Banach algebra $\ell^1(\Sigma)$ of crossed product type is naturally associated with the topological dynamical system $\Sigma=(X,\sigma)$. We initiate the study of the relation between two-sided ideals of $\ell^1(\Sigma)$ and ${\mathrm C}^\ast(\Sigma)$, the enveloping $\mathrm{C}^\ast$-algebra ${\mathrm C}(X)\rtimes_\sigma \mathbb Z$ of $\ell^1(\Sigma)$. Among others, we prove that the closure of a proper two-sided ideal of $\ell^1(\Sigma)$ in ${\mathrm C}^\ast(\Sigma)$ is again a proper two-sided ideal of ${\mathrm C}^\ast(\Sigma)$.	math.OA	math	THE CLOSURE OF IDEALS OF ℓ1(Σ) IN ITS ENVELOPING C∗-ALGEBRA MARCEL DE JEU AND JUN TOMIYAMA AB STRACT. If X is a compact Hausdorff space and σ is a homeomorphism of X , then an involutive Banach algebra ℓ1(Σ) of crossed product type is naturally associated with the topological dynamical system Σ = (X , σ). We initiate the study of the relation between two-sided ideals of ℓ1(Σ) and C∗(Σ), the enveloping C∗-algebra C(X ) ⋊σ Z of ℓ1(Σ). Among others, we prove that the closure of a proper two-sided ideal of ℓ1(Σ) in C∗(Σ) is again a proper two-sided ideal of C∗(Σ). 1. INTRODUCTION AND OVERVIEW If X is a compact Hausdorff space and σ : X → X is a homeomorphism of X , then an involutive Banach algebra ℓ1(Σ) of crossed product type can be associated with this dy- namical system Σ = (X , σ); we shall recall the definition in Section 2. Its C∗-enveloping algebra, denoted by C∗(Σ), is the crossed product C∗-algebra C(X )⋊σ Z. Whereas C∗(Σ) is well-studied, the investigation of ℓ1(Σ) itself is of a more recent nature; this has been taken up in [2], [3], [4], and [5]. The algebra ℓ1(Σ) is more complicated than C∗(Σ). For example, it can occur that ℓ1(Σ) has a non-selfadjoint closed two-sided ideal (see [2, Theorem 4.4]), whereas this is, of course, never the case for C∗(Σ). The study of ℓ1(Σ) so far has proceeded without using what is known about C∗(Σ). Still, it has turned out that some analogous properties of ℓ1(Σ) and C∗(Σ) are equivalent. For example, these algebras are either both simple (i.e. have only trivial closed two- sided ideals), or both non-simple. The proof of this fact proceeds via the properties of Σ: for each of these algebras, this simplicity can be shown to X being an infinite set and Σ being minimal, i.e. to X being an infinite set that has only trivial invariant closed subsets. Hence the simplicity of one algebra also implies the simplicity of the other. It is desirable to have some basic results, formulated directly in terms of the algebras and not using the properties of the dynamical system, that can help transfer a property of the ideal structure of one algebra to an analogous property of the ideal structure of the other algebra. As the example of non-selfadjoint closed two-sided ideals makes clear, one cannot expect to be able to do this in all cases, but one would hope that something in this direction is still possible. The present paper contains the first steps in this direction. We show that the closure in C∗(Σ) of a proper not necessarily closed two-sided ideal of ℓ1(Σ) is still a proper two-sided ideal of C∗(Σ); see Theorem 3.7. We also investigate what the necessary and sufficient condition is so that all closed two-sided ideals of ℓ1(Σ) can be reconstructed from their closure in C∗(Σ) by taking the intersection with ℓ1(Σ) again. The latter is possible if and only if the Z-action on X is free, i.e. if and only if there are no periodic points of σ in X ; see Theorem 3.12. 2010 Mathematics Subject Classification. Primary 46K99; Secondary 46H10, 47L65, 54H20. Key words and phrases. Involutive Banach algebra, enveloping C∗-algebra, ideal, topological dynamical system. 1 2 MARCEL DE JEU AND JUN TOMIYAMA As a (now immediate) example of how such a relation between two-sided ideals of ℓ1(Σ) and C∗(Σ) can be exploited, we re-establish the fact that the minimality of Σ implies the simplicity of ℓ1(Σ), based on the knowledge that this is true for C∗(Σ); see Corollary 3.9 for a more elaborate statement. This paper is organized as follows. In Section 2, we introduce ℓ1(Σ) and its enveloping C∗-algebra C∗(Σ). In Section 3, we establish our main results. The key idea in this section is that there are some proper two-sided ideals of ℓ1(Σ) for which it is immediately obvious that their closure in C∗(Σ) is still a proper two-sided ideal of C∗(Σ), and that they can be retrieved as the intersection of their closure in C∗(Σ) with ℓ1(Σ). As is the case for all unital involutive Banach algebras that are a subalgebra of their enveloping C∗- algebra, all kernels of non-zero involutive representations of ℓ1(Σ) on Hilbert spaces have these two properties. The main technical result in this section is then to show that all primitive ideals (in the purely algebraic sense to be recalled in Section 3) of ℓ1(Σ) are such kernels. 2. TWO ALGEBRAS In this section, we introduce the two algebras playing a role in this paper, and we recall a few standard facts about their relation. We also introduce notation for the two closure operations figuring in this context, and state our convention concerning ideals of an algebra. Throughout this paper, X is a non-empty compact Hausdorff space and σ : X → X is a homeomorphism. Hence Z acts on X , and we write Σ for short for the topological dynamical system (X , σ). We let Aper(σ) and Per(σ) denote the aperiodic and the periodic points of σ, respectively. A subset S of X is invariant if it is invariant under the Z-action, i.e. if σ(S) = S. The involutive algebra of continuous (complex-valued) functions on X is denoted by C(X ), and we write α for the involutive automorphism of C(X ) induced by σ, defined by α( f ) = f ◦ σ−1 for f ∈ C(X ). Via n 7→ αn, Z acts on C(X ). With k · k denoting the supremum norm on C(X ), we let ℓ1(Σ) = ℓ1(Z, C(X )) = a : Z → C(X ) : kak :=Xn∈Z ka(n)k < ∞. We supply the Banach space ℓ1(Σ) with the usual twisted convolution as multiplication, defined by (aa′)(n) =Xk∈Z a(k) · αk(a′(n − k)) for n ∈ Z and a, a′ ∈ ℓ1(Σ), and define an involution on ℓ1(Σ) by a∗(n) = αn(a(−n)) for n ∈ Z and a ∈ ℓ1(Σ). Thus ℓ1(Σ) becomes a unital Banach ∗-algebra with isometric If X consists of one point, then ℓ1(Σ) is the group algebra ℓ1(Z) of the involution. integers. There is a convenient way to work with ℓ1(Σ), which we shall now explain. For n, m ∈ Z, let χ{ n }(m) =1 if m = n; 0 if m 6= n, CLOSURE OF IDEALS 3 where the constants denote the corresponding constant functions on X . Then χ{ 0 } is the identity element of ℓ1(Σ). Let δ = χ{ 1 }; then one sees easily that χ{ −1 } = δ−1 = δ∗. If we put δ0 = χ{ 0 }, then one computes that δn = χ{ n } for all n ∈ Z. We may view C(X ) as a closed abelian ∗-subalgebra of ℓ1(Σ), namely as { a0δ0 : a0 ∈ C(X ) }. If a ∈ ℓ1(Σ), fnδn and k fnk < ∞. In the rest of this paper, we shall constantly use this series fnδn of an arbitrary element a ∈ ℓ1(Σ), with uniquely de- termined fn ∈ C(X ) for n ∈ Z. Thus, all in all, ℓ1(Σ) is generated, as a unital Banach algebra, by an isometrically isomorphic copy of C(X ) and the elements δ and δ−1, sub- ject to the relation δ f δ−1 = α( f ) = f ◦ σ−1 for f ∈ C(X ). The isometric involution is determined by f ∗ = f for f ∈ C(X ) and by δ∗ = δ−1. and if we write fn for a(n) as a more intuitive notation, then a = Pn∈Z kak = Pn∈Z representation a = Pn∈Z Since ℓ1(Σ) is a unital Banach algebra with an isometric involution, it has an envelop- ing C∗-algebra as constructed in [6, Section 2.7]. We denote this enveloping C∗-algebra of ℓ1(Σ) by C∗(Σ). As in the general construction of crossed products of C∗-algebras (see [9]), the enveloping C∗-seminorm on ℓ1(Σ) is actually a norm; in this particular case, a somewhat shorter argument can also be used to see this (see [3, p. 51]). Hence ℓ1(Σ) can be viewed as a dense subalgebra of C∗(Σ), and the inclusion of ℓ1(Σ) into C∗(Σ) is continuous (even contractive). We infer from [6, Proposition 1.3.7] that a not necessarily unital involutive repres- entation of ℓ1(Σ) or C∗(Σ) as bounded operators on a Hilbert space is always con- tinuous (even contractive). As another standard fact, let us note that [6, Proposi- tion 2.7.4] shows that every not necessarily unital involutive representation of ℓ1(Σ) extends uniquely to an involutive representation of C∗(Σ), and that a bijection between the collections of involutive representations of the two algebras is thus obtained. If S ⊂ ℓ1(Σ), then Sℓ1 denotes its closure in ℓ1(Σ). If S ⊂ C∗(Σ), then SC∗ denotes its closure in C∗(Σ). Although we have written 'two-sided ideal' in the abstract and in Section 1 to avoid any possible misunderstanding, in the sequel of this paper, an ideal of an algebra is always a two-sided ideal, unless otherwise stated. It need not (if applicable) be closed. 3. RESULTS In this section, we take up the study of the relation between the ideals of ℓ1(Σ) and C∗(Σ). We shall show that the closure I C∗ of a proper ideal of ℓ1(Σ) is still a proper ideal of C∗(Σ); see Theorem 3.7. Furthermore, we shall investigate when closed ideals of ℓ1(Σ) can be recovered from their closure in C∗(Σ) in the most obvious fashion; see Corollary 3.11 and Theorem 3.12. The key technical result of this section is the fact that every primitive ideal of ℓ1(Σ) is the kernel of a non-zero involutive representation of ℓ1(Σ); see Theorem 3.6. The proof of Theorem 3.7 can be reduced to a particular case by the following purely algebraic argument. We recall that, if A is an algebra, an algebraically irreducible representation of A is a non-zero homomorphism into the linear operators on a vector space E over the per- tinent field such that E has only trivial invariant subspaces. A primitive ideal of A is the kernel of an algebraically irreducible representation of A. It was already noted by Jacobson (see [7]) that a maximal ideal M of a unital algebra A is a primitive ideal. Since it is this fact that allows us to make a reduction that is instrumental for the proof of Theorem 3.7, we recall the short argument, which is as follows. Since A is unital, the proper left ideal M is contained in a maximal left ideal I . The representation of 4 MARCEL DE JEU AND JUN TOMIYAMA A on A/I is then algebraically irreducible, so that Ker(π) is a primitive ideal. Since M is also a right ideal of A, one has M ⊂ Ker(π). By the maximality of M , we conclude that M = Ker(π). Hence M is a primitive ideal, as desired. Since in a unital algebra every proper ideal is contained in a maximal ideal, we see that in a unital algebra every proper ideal is contained in a primitive ideal. As a consequence of these well-known results, if we want to prove that I C∗ is a proper subset of C∗(Σ) for every proper ideal I of ℓ1(Σ), then it is sufficient to do this when I is a primitive ideal of ℓ1(Σ). In order to fully exploit this reduction, it is clearly important to have more information about the primitive ideals of ℓ1(Σ), and we shall now set out to collect the relevant facts from previous work. Before doing so, however, let us note that, in the literature on involutive Banach algebras, a primitive ideal is often defined as the kernel of a topologically irreducible (i.e. having only trivial closed invariant subspaces) non-zero involutive representation of the pertinent algebra on a Hilbert space. This was also the notion employed by the authors in [4], but in [5] the purely algebraic notion as in the present paper was used. For C∗-algebras, there is no difference (this follows from the combination of [6, Corollary 2.8.4] and [6, Corollary 2.9.6.(i)]), but otherwise a little care is in order when using results (including those of the authors) for primitive ideals of involutive Banach algebras as found in the literature. We start by describing a family of finite dimensional algebraically irreducible rep- resentations (hence also of primitive ideals) that are associated with periodic points of σ. When combined with the standard relation between the involutive representa- tions of ℓ1(Σ) and C∗(Σ) on Hilbert spaces, the existence of these representations also follows from general considerations for crossed products of C∗-algebras (see [9]), but the direct definition below suffices for our needs. More can be said about these rep- resentations of ℓ1(Σ) than we shall include here, and we refer to [5, Section 3.2] for further information. For x ∈ Per(σ) and λ ∈ T, we define a representation πx ,λ of ℓ1(Σ) as follows. Let p be the period of X . Let H x ,λ be a Hilbert space with orthonormal basis { e0, . . . , ep−1 } and bounded operators B(H x ,λ). We let Tλ ∈ B(H x ,λ) be the bounded linear operator on H x ,λ that is represented with respect to this basis by the matrix 0 0 . . . 0 λ 1 0 . . . 0 0 0 1 . . . 0 0 ... ... 0 0 . . . 1 0 .. . ... ...   .   For f ∈ C(X ), we let ρx ( f ) be the bounded linear operator on H x ,λ that is represented with respect to this basis by the matrix f (x) 0 0 ... 0 f (σx) ... 0 . . . . . . .. . . . .   0 0 ... f (σp−1 x) .   It is easily checked (we refer to [5, Lemma 3.1.2] for details), that there exists a unique involutive representation πx ,λ : ℓ1(Σ) → B(H x ,λ) such that πx ,λ ↾C(X )= ρx and π(δ) = Tλ. CLOSURE OF IDEALS 5 In [5, Theorem 3.5], the finite dimensional algebraically irreducible representations of ℓ1(Σ) are classified. Amongst others, this classifications shows, somewhat surpris- ingly, that the Hilbert space context is automatic for finite dimensional algebraically irreducible representations of ℓ1(Σ). This is the first part of the following result. The description of the intersection of primitive ideals in equation (3.1) follows from an explicit description of Ker(πx ,λ) and the injectivity of the Fourier transform on ℓ1(Z); see [4, Proposition 2.10]. Proposition 3.1. If π is a finite dimensional algebraically irreducible representation of ℓ1(Σ), then there exist x ∈ Per(σ) and λ ∈ T such that π and πx ,λ are algebraically equivalent. In particular, Ker(π) = Ker(πx ,λ). Hence every primitive ideal of ℓ1(Σ) that arises as the kernel of a finite dimensional algebraically irreducible representation of ℓ1(Σ) is also the kernel of a topologically irreducible non-zero involutive representation of ℓ1(Σ) on a Hilbert space, and it is a selfadjoint ideal. Furthermore, (3.1) \λ∈T Ker(πx ,λ) = Xn∈Z fnδn ∈ ℓ1(Σ) : fn(σk x) = 0 for all n, k ∈ Z. Remark 3.2. Given the algebraic irreducibility, the topological irreducibility of the fi- nite dimensional representations in Proposition 3.1 is, of course, immediate. We have nevertheless included it, since an analogous statement is also true when this is less obvious (see part (4) of Proposition 3.4). Although it is not relevant for the proofs of Theorem 3.7, Corollary 3.11, or Theorem 3.12, the presence of this topological irredu- cibility seems too remarkable not to include it; see also part (3) of Theorem 3.6. Now that all primitive ideals corresponding to finite dimensional algebraically ir- reducible representations have been described, and have been related to involutive representations, we turn to the infinite dimensional case. In [5] it was repeatedly used that the representation space of an algebraically irreducible representation of ℓ1(Σ) can always be normed in such a way that the algebra acts as continuous operators and that the representation is a bounded map. This is an immediate consequence (see e.g. [1, proof of Lemma 25.2]) of the fact that a maximal left ideal in a unital Banach algebra is closed. Combining this normability (which also shows that primitive ideals of ℓ1(Σ) are closed) with [5, Propositions 2.6, 4.2, and 4.17] yields the following. Proposition 3.3. If π is an infinite dimensional algebraically irreducible representation of ℓ1(Σ), then there exists an infinite invariant subset S of X such that (3.2) Ker(π) = Xn∈Z fnδn ∈ ℓ1(Σ) : fn ↾S= 0 for all n ∈ Z. Hence Ker(π) is a self-adjoint ideal. If X is metrizable, then there exists x ∈ Aper(σ) such that the description of I in equation (3.2) holds with S = { σn x : n ∈ Z }. In order to relate Ker(π) for an infinite dimensional algebraically irreducible rep- resentation π to an involutive representation, we shall use equation (3.2), the repres- entations πx ,λ as described before for periodic points x and λ ∈ T, and a family of infinite dimensional representations of ℓ1(Σ) that are associated with aperiodic points and that we shall now introduce. Here, again, more can be said than we shall include in the present paper, and we refer to [5, Section 3.3] for further information. 6 MARCEL DE JEU AND JUN TOMIYAMA Fix x ∈ Aper(σ), and let 1 ≤ p < ∞. We let B(ℓp(Z)) denote the bounded linear operators on ℓp(Z), and, for k ∈ Z, we let ek denote the element of ℓp(Z) with 1 in the kth coordinate and zero elsewhere. Let S ∈ B(ℓp(Z)) be the right shift, determined by Sek = ek+1 for k ∈ Z. For f ∈ C(X ), let πp x ( f )ek = f (σk x)ek for k ∈ Z. It is then easily seen (see [5, Lemma 3.1] for details) that there exists a unique unital continuous representation πp x ( f ) ∈ B(ℓp(Z)) be determined by πp x : ℓ1(Σ) → B(ℓp(Z)) such that (3.3) πp xXn∈Z fnδn =Xn∈Z πp x ( f )Sn fnδn ∈ ℓ1(Σ). Furthermore, if p = 2, then π2 x is an involutive representa- for all Pn∈Z tion of ℓ1(Σ) on the Hilbert space l 2(Z). We collect a few relevant facts about these infinite dimensional representations, the first three of which are taken from [5, Theorem 3.16 and Lemma 4.9]. Part (2) is not too hard to establish, and part (3) is rather obvious, but part (1) is considerably more intricate. Part (4) is immediate from the parts (1), (2), and (3). Proposition 3.4. Let x ∈ Aper(σ) and let 1 ≤ p < ∞. Then: (1) The representation π1 (2) For 1 < p < ∞, the representation πp cible, but not algebraically irreducible; x of ℓ1(Σ) on ℓ1(Z) is algebraically irreducible; x of ℓ1(Σ) on ℓp(Z) is topologically irredu- (3) The kernel of πp x is a selfadjoint ideal of ℓ1(Σ) that does not depend on p. In fact, (3.4) Ker(πp x ) =Xn∈Z fnδn ∈ ℓ1(Σ) : fn(σk x) = 0 for all n, k ∈ Z; (4) The primitive ideal Ker(π1 x ) of ℓ1(Σ) is also the kernel of the topologically irredu- cible involutive representation π2 x of ℓ1(Σ) on the Hilbert space ℓ2(Z). Remark 3.5. Before proceeding, let us note that, if X is metrizable, the combination of Propositions 3.1, 3.3, and 3.4 shows that we can describe the set of primitive ideals of ℓ1(Σ), even though we do not know all infinite dimensional algebraically irreducible representations. This set is { Ker(πx ,λ) : x ∈ Per(σ), λ ∈ T }∪{ Ker(π1 x ) : x ∈ Aper(σ) }. As a word of warning, let us note that there are multiple occurrences in this enu- ) if and meration. If x1, x2 ∈ Aper(σ), then only if the orbits of x1 and x2 coincide and λ1 = λ2. Ker(π1 ) if and only if the closures of the orbits of x1 and x2 coincide. Fur- x1 thermore, { Ker(πx ,λ) : x ∈ Per(σ), λ ∈ T } ∩ { Ker(π1 x ) : x ∈ Aper(σ) } = ;. We refer to [5, Lemma 4.14] for more details. If x1, x2 ∈ Per(σ) and λ1, λ2 ∈ T, then Ker(πx1,λ1 ) = Ker(πx2,λ2 ) = Ker(π1 x2 We collect the material on arbitrary primitive ideals in the next result. The crucial link with involutive representations is established in the parts (2) and (3). Theorem 3.6. Let I be a primitive ideal of ℓ1(Σ). (1) (a) If I is the kernel of a finite dimensional algebraically irreducible representa- tion of ℓ1(Σ), then I = Ker(πx ,λ) for some x ∈ Per(σ) and λ ∈ T. Furthermore, πx ,λ is a topologically irreducible involutive representation of ℓ1(Σ) on a finite dimensional Hilbert space. (b) If I is the kernel of an infinite dimensional algebraically irreducible repres- entation of ℓ1(Σ), then there exists an infinite invariant subset S of X such CLOSURE OF IDEALS 7 that I = \x∈Per(σ)∩S Ker(πx ,λ) ∩ \x∈Aper(σ)∩S λ∈T Ker(π2 x ). If X is metrizable, then there exists x ∈ Aper(σ) such that I = Ker(π2 Furthermore, all πx ,λ that occur are topologically irreducible involutive rep- resentations of ℓ1(Σ) on finite dimensional Hilbert spaces, and all π2 x that occur are topologically irreducible involutive representations of ℓ1(Σ) on the Hilbert space ℓ2(Z). x ). (2) There exists a unital involutive representation π of ℓ1(Σ) on a Hilbert space such that I = Ker(π). (3) If X is metrizable, then there exists a topologically irreducible involutive repres- entation π of ℓ1(Σ) on a Hilbert space such that I = Ker(π). Proof. Part (1.a) is contained in Proposition 3.1. The first statement in part (1.b) follows by combining equations (3.2), (3.3), and (3.4) for p = 2. The second statement in part (1.b) follows from the metrizable case in Proposition 3.3 and equation (3.4) for p = 2. Since part (1) shows that, in both cases, I is the simultaneous kernel of a suitable collection of involutive representations of ℓ1(Σ), it is also the kernel of the Hilbert direct sum of the representations in this collection; note that this sum can be defined, since the pertinent representations are all contractive. This establishes part (2). Part (3) is immediate from part (1). (cid:3) With part (2) of Theorem 3.6 available, we can now use the observation that was already mentioned in the introduction: kernels of involutive representations of ℓ1(Σ) are well-behaved when their relation with C∗(Σ) is concerned. Theorem 3.7. Let I be a not necessarily closed proper ideal of ℓ1(Σ). Then the closure I C∗ of I in C∗(Σ) is a proper closed ideal of C∗(Σ). Proof. It is elementary that I C∗ is an ideal of C∗(Σ); it remains to be shown that it is a proper subset of C∗(Σ). As explained earlier, we may assume that I is a primitive ideal of ℓ1(Σ). In that case, Theorem 3.6 shows that I = Ker(π) for some non-zero involutive representation of ℓ1(Σ) on a Hilbert space. If we let πe denote the extension of π to a non-zero involutive representation of C∗(Σ) on that Hilbert space, then Ker(πe) is a proper closed subset of C∗(Σ). Since I C∗ = Ker(πe), we see that I C∗ (cid:3) is also a proper subset of C∗(Σ). ⊂ Ker(πe)C∗ = Ker(π)C∗ Remark 3.8. (1) An inspection of the structure of the proof of Theorem 3.7 shows that [5, Pro- position 2.6] is used, which, in turn, is based on the so-called intersection property of the commutant of C(X ) in ℓ1(Σ); see [2, Theorem 3.7]. The latter result is the rather non-trivial key result in [2]. Thus, in spite of the simplicity of its formulation, Theorem 3.7 seems to be a reasonably deep fact about the relation between ℓ1(Σ) and C∗(Σ). (2) For a not necessarily closed proper ideal I of ℓ1(Σ), a necessary and sufficient condition for I C∗ to be a proper ideal of C∗(Σ) is given in [4, Proposition 4.12]. In effect, we have shown that this condition is always satisfied, with a proof that is in the spirit of the proof of [4, Proposition 4.12]. 8 MARCEL DE JEU AND JUN TOMIYAMA As an illustration how Theorem 3.7 can be used to deduce results about ℓ1(Σ) from results about C∗(Σ) without resorting to the properties of the dynamical system, we re-establish the following result. It can already be found as [2, Theorem 4.2]. Corollary 3.9. The following are equivalent: (1) The only closed ideals of ℓ1(Σ) are { 0 } and ℓ1(Σ); (2) The only closed selfadjoint ideals of ℓ1(Σ) are { 0 } and ℓ1(Σ); (3) X has an infinite number of points, and the only closed invariant subsets of X are ; and X . Proof. It is clear that (1) implies (2). If (3) does not hold, then it is not too difficult to construct a non-trivial closed selfadjoint ideal of ℓ1(Σ); we refer to [2, proof of Theorem 4.2] for details. Thus (2) implies (3). The hard part is to show that (3) implies (1), and it is here that Theorem 3.7 can be put to good use to be able to apply a result about C∗(Σ). Indeed, [8, Theorem 5.4] shows that, if (3) holds, then the algebra C∗(Σ) has only trivial closed ideals. Theorem 3.7 shows that the same is then true for ℓ1(Σ), which is (1). (cid:3) We shall now consider the relation between the closure operation in ℓ1(Σ) and in C∗(Σ). We start with the following observation, the first part of which was already alluded to in Section 1. Proposition 3.10. (1) If I is the kernel of an involutive representation of ℓ1(Σ) on a Hilbert space, then I = I C∗ ∩ ℓ1(Σ). (2) If { Iα : α ∈ A} is collection of closed ideals of ℓ1(Σ) such that Iα = Iα for all α ∈ A, then C∗ ∩ ℓ1(Σ) C∗ Iα ∩ ℓ1(Σ). \α∈A Iα = \α∈A Proof. Suppose that I = Ker(π) for an involutive representation π of ℓ1(Σ). As in the proof of Theorem 3.7, we let πe denote the extension of π to an involutive representa- tion of C∗(Σ). Then, again as in that proof, we have I C∗ = Ker(πe). Hence I C∗ ∩ ℓ1(Σ), part (1) has been established. ∩ ℓ1(Σ) ⊂ Ker(πe) ∩ ℓ1(Σ) = Ker(π) = I . Since obviously I ⊂ I C∗ = Ker(π)C∗ ⊂ Ker(πe)C∗ We turn to part (2). Using the properties of the Iα in the final step, we see that C∗ Iα ∩ ℓ1(Σ) ⊂\α∈A \α∈A Iα C∗ ∩ ℓ1(Σ) = \α∈AIα C∗ ∩ ℓ1(Σ) =\α Since the reverse inclusion is obvious, the proof is complete. Iα. (cid:3) Combining Proposition 3.10 and part (2) of Theorem 3.6, we have the following. Corollary 3.11. If I is an intersection of primitive ideals of ℓ1(Σ), then I = I C∗ ∩ ℓ1(Σ). We can now determine when all closed ideals of ℓ1(Σ) can be retrieved from their closure in C∗(Σ) as above. Theorem 3.12. The following are equivalent: = I C∗ (1) I ℓ1 (2) I = I C∗ (3) Every closed ideal of ℓ1(Σ) is an intersection of primitive ideals of ℓ1(Σ); ∩ ℓ1(Σ) for every not necessarily closed ideal of ℓ1(Σ); ∩ ℓ1(Σ) for every closed ideal of ℓ1(Σ); CLOSURE OF IDEALS 9 (4) Every closed ideal of ℓ1(Σ) is a selfadjoint ideal of ℓ1(Σ); (5) Every closed ideal of ℓ1(Σ) is the kernel of an involutive representation of ℓ1(Σ) on a Hilbert space; (6) There are no periodic points of σ in X . Proof. It is clear that (1) implies (2). Using the continuity of the inclusion of ℓ1(Σ) in C∗(Σ), an application of (2) to I ℓ1 shows that (2) implies (1). Certainly (2) implies (4), because all closed ideals of C∗(Σ) are selfadjoint. The equivalence of (3), (4), and (6) is the content of [5, Theorem 4.4]. We know from part (2) of Theorem 3.6 that every primitive ideal of ℓ1(Σ) is the kernel of an involutive representation of ℓ1(Σ) on a Hilbert space. Therefore, assuming (3), we see that (5) holds by taking a Hilbert direct sum. It is evident that (5) implies (4). The proof will be complete once we show that (3) implies (2), and this is immediate (cid:3) from Corollary 3.11. REFERENCES [1] F.F. Bonsall and J. Duncan, Complete normed algebras, Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 80, Springer-Verlag, New York-Heidelberg, 1973. [2] M. de Jeu, C. Svensson, and J. Tomiyama, On the Banach ∗-algebra crossed product associated with a topological dynamical system, J. Funct. Anal. 262 (2012), no. 11, 4746 -- 4765. [3] M. de Jeu and J. Tomiyama, Maximal abelian subalgebras and projections in two Banach algebras associated with a topological dynamical system, Studia Math. 208 (2012), no. 1, 47 -- 75. [4] M. de Jeu and J. Tomiyama, Noncommutative spectral synthesis for the involutive Banach algebra associated with a topological dynamical system, Banach J. Math. Anal. 7 (2013), no. 2, 103 -- 135. [5] M. de Jeu and J. Tomiyama, Algebraically irreducible representations and structure space of the Banach algebra associated with a topological dynamical system, Adv. Math. 301 (2016), 79 -- 115. [6] J. Dixmier, C∗-algebras, North-Holland Mathematical Library, vol. 15, North-Holland Publishing Co., Amsterdam-New York-Oxford, 1977. Translated from the French by Francis Jellett. [7] N. Jacobson, A topology for the set of primitive ideals in an arbitrary ring, Proc. Nat. Acad. Sci. U. S. A. 31 (1945), 333 -- 338. [8] J. Tomiyama, The interplay between topological dynamics and theory of C∗-algebras, Lecture Notes Series, vol. 2, Seoul National University, Research Institute of Mathematics, Global Analysis Research Center, Seoul, 1992. [9] D.P. Williams, Crossed products of C∗-algebras, Mathematical Surveys and Monographs, vol. 134, Amer- ican Mathematical Society, Providence, RI, 2007. MARCEL DE JEU, MATHEMATICAL INSTITUTE, LEIDEN UNIVERSITY, P.O. BOX 9512, 2300 RA LEIDEN, THE NETHERLANDS E-mail address: [email protected] JUN TOMIYAMA, DEPARTMENT OF MATHEMATICS, TOKYO METROPOLITAN UNIVERSITY, MINAMI-OSAWA, HA- CHIOJI CITY, JAPAN E-mail address: [email protected]
1209.6323	1	1209	2012-09-27T18:28:48	Subideals of operators II	[ "math.OA", "math.FA" ]	A subideal (also called a J-ideal) is an ideal of a B(H)-ideal J. This paper is the sequel to Subideals of operators where a complete characterization of principal and then finitely generated J-ideals were obtained by first generalizing the 1983 work of Fong and Radjavi who determined which principal K(H)-ideals are also B(H)-ideals. Here we determine which countably generated J-ideals are B(H)-ideals, and in the absence of the continuum hypothesis which J-ideals with generating sets of cardinality less than the continuum are B(H)-ideals. These and some other results herein are based on the dimension of a related quotient space. We use this to characterize these J-ideals and settle additional questions about subideals. A key property in our investigation turned out to be J-softness of a B(H)-ideal I inside J, that is, IJ = I, a generalization of a recent notion of softness of B(H)-ideals introduced by Kaftal-Weiss and earlier exploited for Banach spaces by Mityagin and Pietsch.	math.OA	math	SUBIDEALS OF OPERATORS II SASMITA PATNAIK AND GARY WEISS Abstract. A subideal (also called a J-ideal) is an ideal of a B(H)-ideal J. This paper is the sequel to Subideals of Operators where a complete characterization of principal and then finitely generated J-ideals were obtained by first generalizing the 1983 work of Fong and Radjavi who determined which principal K(H)-ideals are also B(H)-ideals. Here we determine which countably generated J-ideals are B(H)-ideals, and in the absence of the continuum hypothesis which J-ideals with generating sets of cardinality less than the continuum are B(H)-ideals. These and some other results herein are based on the dimension of a related quotient space. We use this to characterize these J-ideals and settle additional questions about subideals. A key property in our investigation turned out to be J-softness of a B(H)-ideal I inside J, that is, I J = I, a generalization of a recent notion of softness of B(H)-ideals introduced by Kaftal-Weiss and earlier exploited for Banach spaces by Mityagin and Pietsch. 1. Introduction In Subideals of Operators [8] we found three types of principal and finitely generated subideals (i.e., J- ideals): linear, real-linear and nonlinear subideals. Such types also carry over to general J-ideals. The linear K(H)-ideals, being the traditional ones, were studied in 1983 by Fong and Radjavi [3]. They found principal linear K(H)-ideals that are not B(H)-ideals. Herein we take all J-ideals I to be linear, but as proved in [8], we expect here also that most of the results and methods apply to the two other types of subideals (real-linear and nonlinear). Also H, as in [8], denotes a separable infinite-dimensional complex Hilbert space. One of our main contributions in [8] was to use a modern framework for B(H)-ideals to generalize [3, Theorem 2]. We generalized their result on principal K(H)-ideals to all principal J-ideals by proving that a principal and then a finitely generated J-ideal (S)J generated by the finite set S ⊂ J is also a B(H)-ideal if and only if (S) is J-soft, that is, (S) = (S)J where (S) is the B(H)-ideal generated by S. Then we used this to characterize the structure of (S)J . J-softness is a generalization of a recent notion of K(H)-softness of B(H)-ideals introduced by Kaftal-Weiss and earlier exploited for Banach spaces by Mityagin and Pietsch (see [8, Remark 2.6], [7], [9]). Here we further develop the subject by investigating J-ideals I = (S)J generated by arbitrary sets S of varying cardinality, their algebraic structure and when they are B(H)-ideals. To add perspective, the reader should keep in mind that all nonzero J-ideals have cardinality and Hamel dimension precisely cardinality c of the continuum (Remark 3.3), but questions on the cardinalities of their possible generating sets is another matter (Section 6, Questions 1-2), and this we shall see impacts questions on structure. After investigating the cases when S is countable or of cardinality less than c (absent the continuum hypothesis CH), we then consider general J-ideals I and questions on the possible cardinalities of their generating sets, observing that I is always a generating set for itself but may have generating sets of cardinality less than its cardinality c. When they do has special implications. We show (S)J is a B(H)-ideal if and only if (S) is J-soft for those (S)J generated as a J-ideal by countable sets and then when generated by sets of cardinality strictly less than cardinality c (Theorem 4.1). This will follow from sufficiency of the codimension condition on (S)0 J has Hamel dimension strictly less than c. (For (S)0 J see Definition 2.2.) We then investigate general J-ideals to provide an example where softness fails for a J-ideal I with codimension of I 0 in I equal to c (Example 4.5), thereby showing that J-ideals that are also B(H)-ideals need not be J-soft and as a consequence cannot be generated in J by sets of cardinality less than c. We also answer several questions on J-ideals posed in [8, Sections 6-7], provide some additional results and pose new questions. J in (S)J : (S)J /(S)0 Date: October 20, 2018. 1991 Mathematics Subject Classification. Primary: 47L20, 47B10, 47B07; Secondary: 47B47, 47B37, 13C05, 13C12. Key words and phrases. Ideals, operator ideals, principal ideals, subideals, lattices. 1 2 SASMITA PATNAIK AND GARY WEISS In summary the main theorems here are: For I 0 := span{IJ + JI} + J(I)J (see Definition 2.2) where (I) is the B(H)-ideal generated by J-ideal I, Theorem. (Theorem 4.1) The J-ideal (S)J generated by a set S of cardinality strictly less than c is a B(H)- ideal if and only if the B(H)-ideal (S) is J-soft (i.e., (S) = (S)J). Moreover, for a J-ideal I with Hamel dimension of I/I 0 strictly less than c, I is a B(H)-ideal if and only if (I) is J-soft. Structure Theorem. (Theorem 4.4) For (S)J when S < c, The algebraic structure of the J-ideal (S)J generated by a set S is given by (S)J = span{S + J S + SJ} + J(S)J J(S)J is a B(H)-ideal, span{J S + SJ} + J(S)J is a J-ideal, and J(S)J ⊂ span{J S + SJ} + J(S)J ⊂ (S)J . This inclusion collapses to J(S)J = (S)J if and only if (S) is J-soft. Remark. Although the methods in [3] are quite a bit more analytic, we found here and in [8] a more direct algebraic approach, albeit a key tool [4] used herein is essentially analytic. 2. Preliminaries Recall the following standard definitions from [8] with Definition 2.2 evolving from [8]. Definition 2.1. Let J be an ideal of B(H) (i.e., a B(H)-ideal) and S ∈ J. • The principal B(H)-ideal generated by the single operator S is given by • The principal J-ideal generated by S is given by (S) := T{I I is a B(H)-ideal containing S } (S)J := T{I I is a J-ideal containing S } • As above for principal J-ideals, likewise for an arbitrary subset S ⊂ J, (S) and (S)J respectively denote the smallest B(H)-ideal and the smallest J-ideal generated by the set S. In particular, (S) = (S)B(H). Denote (I) as the B(H)-ideal generated by the J-ideal I. Definition 2.2. For a J-ideal I, the algebraic J-interior of I is denoted by I 0 := span{IJ + JI} + J(I)J where IJ := {AB A ∈ I, B ∈ J}, JI is defined similarly, and the ideal product J(I)J is the B(H)-ideal given by J(I)J = {ES′F E, F ∈ J, S′ ∈ (I)} (single triple operator products) where equality follows from [2, Lemma 6.3]. Remark 2.3. For the J-ideal I = (S)J generated by a set S, I 0 has the simpler form: (S)0 J = span{SJ + J S} + J(S)J Notice also that (S)0 J ⊂ (S)J. Indeed, for S = {Sα}α∈A, (S)J = [{α1,··· ,αj }⊂A {(Sα1 )J + (Sα2 )J + · · · + (Sαj )J } where A is an index set. By definition, (S)0 J = span{(S)J J + J(S)J } + J((S)J )J, and because ((S)J ) = (S) one has the simplification: (S)0 J = span{(S)J J + J(S)J } + J(S)J. The algebraic structure for principal J-ideals yields (Sα)J = CSα + JSα + SαJ + J(Sα)J for each α ∈ A. So span{(Sα)J J + J(Sα)J } + J(Sα)J ⊂ SαJ + JSα + J(Sα)J for each α ∈ A. Therefore (S)0 J ⊂ span{SJ + J S} + J(S)J, and since the reverse inclusion is obvious, one has equality. We recall here the definition of J-softness of B(H)-ideals [8, Definition 2.5]. Definition 2.4. For B(H)-ideals I and J, the ideal I is called "J-soft" if IJ = I. Equivalently in the language of s-numbers: For every A ∈ I, sn(A) = O(sn(B)sn(C)) for some B ∈ I, C ∈ J, m ∈ N. Because IJ ⊂ J, only B(H)-ideals that are contained in J can be J-soft. SUBIDEALS OF OPERATORS II 3 Remark 2.5. Standard facts on B(H)-ideals from [8, Remark 2.2]. (i) [2, Sections 2.8, 4.3] (see also [6, Section 4]): If I, J are B(H)-ideals then the product IJ, which is both associative and commutative, is the B(H)-ideal given by the characteristic set Σ(IJ) = {ξ ∈ c∗ o ξ ≤ ηρ for some η ∈ Σ(I) and ρ ∈ Σ(J)}. In abstract rings, the ideal product is defined as the class of finite sums of aibi ai ∈ I, bi ∈ J}, but in B(H) the next lemma shows finite sums products of two elements, IJ := { Xf inite of operator products defining IJ can be reduced to single products. (ii) [2, Lemma 6.3] Let I and J be proper ideals of B(H). If A ∈ IJ, then A = XY for some X ∈ I and Y ∈ J. (iii) [6, Section 1] For T ∈ B(H), s(T ) denotes the sequence of s-numbers of T . Then A ∈ (T ) if and only if s(A) = O(Dm(s(T )) for some m ∈ N. Moreover, for a B(H)-ideal I, A ∈ I if and only if A∗ ∈ I if and only if A ∈ I (via the polar decomposition A = U A and U ∗A = A = (A∗A)1/2), with all equivalent to diag s(A) ∈ I. (iv) The lattice of B(H)-ideals forms a commutative semiring with multiplicative identity B(H). That is, the lattice is commutative and associative under ideal addition and multiplication (see [2, Section 2.8]) and it is distributive. Distributivity with multiplier K(H) is stated without proof in [6, Lemma 5.6-preceding comments]. The general proof is simple and is as follows. For B(H)-ideals I, J, K, one has I, J ⊂ I + J := {A + B A ∈ I, B ∈ J} and so IK, JK ⊂ (I + J)K, so one has IK + JK ⊂ (I + J)K. The reverse inclusion follows more simply if one invokes (ii) above: X ∈ (I + J)K if and only if X = (A + B)C for some A ∈ I, B ∈ J, C ∈ K. The lattice of B(H)-ideals is not a ring because, for instance, {0} is the only B(H)-ideal with an additive inverse, namely, {0} itself, so it is not an additive group. It is also clear that B(H) is the multiplicative identity but no B(H)-ideal has a multiplicative inverse. We summarize the main results of [8] generalizing the 1983 work of Fong and Radjavi and characterizing all finitely generated linear J-ideals. Though not needed here, we note that [8] provided similar results for real-linear and nonlinear J-ideals. Theorem 2.6. [8, Theorem 4.5] For S := {S1, · · · , SN } ⊆ J, the following are equivalent. (i) The finitely generated J-ideal (S)J is a B(H)-ideal. (ii) The B(H)-ideal (S) is J-soft, i.e., (S) = J(S) (equivalently, (S) = (S)J). (iii) For all 1 ≤ j ≤ N , Sj = AijkSkBijk for some Aijk, Bijk ∈ J, n(j, k) ∈ N. N Xk=1 n(j,k) Xi=1 (iv) For all 1 ≤ j ≤ N , s(Sj ) = O(Dm(s(S1 + · · · + SN ))s(T )) for some T ∈ J and m ∈ N. Theorem 2.7. [8, Theorem 4.6] (Structure theorem for finitely generated J-ideal (S)J for S = {S1, · · · , SN }) The algebraic structure of the finitely generated J-ideal (S)J is given by (S)J = CS1 + · · · + CSN + JS1 + · · · + JSN + S1J + · · · + SN J + J(S)J So one has J(S)J ⊆ JS1 + · · · + JSN + S1J + · · · + SN J + J(S)J ⊆ (S)J ⊆ (S) which first two, J(S)J and JS1 + · · · + JSN + S1J + · · · + SN J + J(S)J respectively, are a B(H)-ideal and a J-ideal. The inclusions collapse to merely if and only if the finitely generated B(H)-ideal (S) is J-soft. J(S)J = (S)J = (S) 4 SASMITA PATNAIK AND GARY WEISS 3. The Hamel dimension of ideals In this section we show that the Hamel dimension and the cardinality of every nonzero J-ideal is precisely c. This will impact the codimension of the algebraic J-interior I 0 in I and lead to questions on the possible generating sets for general J-ideals (see Question 4.3 and Section 6 -- Questions 1, 2, 5). It is straightforward to see that I 0 := span{IJ + JI} + J(I)J is an ideal of I and hence is a complex vector subspace of I. The quotient space I/I 0 is a complex vector space and therefore has a Hamel basis where the Hamel dimension is invariant over all Hamel bases. The key notion used in our results is the Hamel dimension of I/I 0 relative to its vector space structure. (I 0 being an ideal of I, the quotient space I/I 0 is also a ring but we will not exploit the ring structure.) Proposition 3.1. For the J-ideal (S)J generated by a set S and (S)0 dimension of the quotient space (S)J /(S)0 J is at most the cardinality of the generating set S. J = span{SJ + J S} + J(S)J, the Hamel Proof. From general ring theory, for S = {Sα}α∈A, (S)J = [{α1,··· ,αj }⊂A {(Sα1)J +(Sα2)J +· · ·+(Sαj )J } where A is an index set. Combining this with the algebraic structure for principal J-ideals implied by Theorem 2.7 (or for principal J-ideals in particular, see also [8, Proposition 4.2]), one obtains (S)J = span S + (S)0 J . For (S)0 J = span{[Sα]} where α ∈ A, Sα ∈ S and [Sα] denotes its quotient space coset. Therefore the Hamel dimension of the vector space (S)J /(S)0 (cid:3) J being a linear subspace of (S)J , one can show that the quotient space (S)J /(S)0 J is at most the cardinality of S. Finishing up our discussion on the Hamel dimension, the following proposition which we need in Example 4.5 is probably a well-known fact but we include it here for completeness. Proposition 3.2. The Hamel dimension of F (H), the B(H)-ideal of finite rank operators, is c when H is separable (at least c for H non-separable). Proof. Cardinal arithmetic applied to matrices when H is separable shows F (H) ≤ B(H) ≤ c, so the Hamel dimension of F (H) is at most c. Suppose the Hamel dimension of F (H) is strictly less than c. Let B := {Fα ∈ F (H) α ∈ A} be a Hamel basis for F (H) with cardinality A < c. Denote a finite Bα ⊂ H, one can extract a basis for the range of Fα by Bα. So [α∈A dimensional Hilbert space is at least c [4, Lemma 3.4], there is a 0 6= f ∈ H for which the set ES{f } forms a linearly independent set. Consider the rank one operator f N f . Since B is a Hamel basis for F (H), f N f = aiFαi for some ai ∈ C, n ∈ N. So, in particular, f N f (f ) = maximal linearly independent set E of cardinality at most A < c. Since the Hamel dimension of infinite- Bα = A and from the set [α∈A aiFαi (f ) which implies n n Xi=1 0 6= hf, f i f = aiFαi (f ) = Bα, hence f ∈ span E contradicting that ES{f } is a linearly independent set. In summary, the assumption that the Hamel dimension of F (H) is strictly less than c led to the existence of this f and hence to this contradiction. So the Hamel dimension of F (H) is precisely c, and consequently the cardinality B(H) = c. (cid:3) An unrelated and interesting question on Hamel bases appears in [1]. Remark 3.3. The Hamel dimension of every nonzero J-ideal I is precisely c. Indeed, F (H) ⊂ I (see [8, Section 6, ¶2]) so Proposition 3.2 implies the Hamel dimension of I is at least c. Also, since I ⊂ B(H) and since cardinality of B(H) is precisely c, the Hamel dimension of I is at most c. Hence the Hamel dimension of I is precisely c. Moreover, because F (H) ⊂ I ⊂ B(H), c = F (H) ≤ I ≤ B(H) = c so the cardinality of I is also precisely c. Xi=1 Xi=1 n m Xi=1 bβieβi where eβi ∈ E ⊂ [α∈A SUBIDEALS OF OPERATORS II 5 4. Main Results: Structure and Softness As mentioned earlier in Section 1, one of our main contributions in [8] was the generalization of Fong and Radjavi's result [3, Theorem 2] by showing that the principal J-ideals and the finitely generated J-ideals that are also B(H)-ideals must be J-soft. Here we show the same for J-ideals generated by countable sets and, absent CH, generated by sets of cardinality strictly less than c. We also show that if a J-ideal I is generated by sets of cardinality equal to c, then I being a B(H)-ideal does not necessarily imply that I is J-soft (Example 4.5). Our main softness theorem is: Theorem 4.1. The J-ideal (S)J generated by sets S of cardinality strictly less than c is a B(H)-ideal if and only if the B(H)-ideal (S) is J-soft. Moreover, for a J-ideal I with the Hamel dimension of I/I 0 strictly less than c, I is a B(H)-ideal if and only if (I) is J-soft, where (I) is the B(H)-ideal generated by I. Proof. ⇒: Since (S)J ⊂ (S), it suffices to show (S) ⊂ (S)J. Assume otherwise that there is some T ∈ (S) \ (S)J. We claim (S) = (S)J so that T ∈ (S)J . Since every B(H)-ideal is also a J-ideal, (S) is a J-ideal containing S. And (S)J being the smallest J-ideal containing S, one has (S)J ⊂ (S). The minimality of (S) as a B(H)-ideal containing S and (S)J assumed to be a B(H)-ideal imply (S) ⊂ (S)J . Hence (S)J = (S) and therefore T ∈ (S)J or, equivalently because when (S)J = (S) it is a B(H)-ideal, one has the equivalent condition diag s(T ) ∈ (S)J . Using the s-number sequence s(T ) we now construct a sequence of operators Dn as follows. For each n ≥ 1, let Dn be the diagonal operator with s2n−1(2k−1)(T ) at the 2n−1(2k − 1) scattered diagonal positions for k ≥ 1 and with zeros elsewhere. Every positive integer has this unique product decomposition 2n−1(2k − 1). Notice then that the diagonal sequences of the Dn's have pairwise disjoint support and the formal direct J is J is a vector space. We will use J . (In sum P⊕ Dn (which incidentally converges in the operator norm) is precisely diag s(T ). Recall that (S)0 a J-ideal and a complex vector subspace of (S)J so their quotient (S)J /(S)0 these Dn's to imbed isomorphic copies of ℓp (for every 1 ≤ p ≤ ∞) inside the quotient space (S)J /(S)0 fact, we imbed isometric isomorphic copies of ℓp inside the quotient space (S)J /(S)0 We next show that for each n ≥ 1, (Dn) = (T ) and that Dn /∈ (S)0 J . Clearly (Dn) ⊂ (T ) since Dn = P T for a suitable diagonal projection operator, so it remains to show (T ) ⊂ (Dn). For each n ≥ 1, Dn is explicitly given by diag ( 0, · · · , 0 0, · · · , 0 0, · · · , 0 0, · · · , 0 , · · · ) , s3·2n−1(T ), , s5·2n−1(T ), , s2n−1(T ), J .) so the 2n-fold ampliation of s(Dn) is (s2n−1 (T ), · · · , s2n−1(T ) , s3·2n−1(T ), · · · , s3·2n−1(T ) (2n−1)−times {z } {z } (2n−1)−times , · · · ) ∈ Σ((Dn)) (2n−1−1)−times {z } (2n−1)−times {z } {z 2n−times 2n−times {z (2n−1)−times {z } and since (s2n−1+1(T ), s2n−1+2(T ), · · · ) ≤ (s2n−1 (T ), · · · , s2n−1(T ) , s3·2n−1(T ), · · · , s3·2n−1(T ) , · · · ), so hs2n−1+k(T )i∞ k=1 ∈ Σ((Dn)) and hence ( 0, · · · , 0 , s2n−1+1(T ), s2n−1+2(T ), · · · ) ∈ Σ((Dn)). } 2n−times {z } } 2n−times {z } But also the finitely supported sequence (s1(T ), s2(T ), · · · , s2n−1(T ), 0, · · · ) ∈ Σ((Dn)). Adding both se- quences one obtains precisely s(T ). Then since Σ((Dn)) is additive, we have s(T ) ∈ Σ((Dn)) and hence diag s(T ) ∈ (Dn). Therefore (T ) ⊂ (Dn), and then from the reverse inclusion above one has (T ) = (Dn). To see that Dn /∈ (S)0 J ⊂ (S)J so T ∈ (Dn) ⊂ (S)J contradicting the assumption T ∈ (S) \ (S)J. So Dn /∈ (S)0 J for all n ≥ 1. (Other choices of the diagonal sequences for the Dn's are possible. Besides disjoint supports or "almost" disjoint supports, the only feature needed is bounded gaps between their nonzero entries.) J , assume otherwise that Dn ∈ (S)0 J . Then Dn ∈ (S)0 The set Xp := {P⊕ aiDi haii p < ∞, ai ∈ C} ⊂ (S)J for 1 ≤ p ≤ ∞. The inclusion is because the s-number sequence s(P⊕ aiDi) ≤ haii ∞ s(T ) and because (S)J contains diag s(T ) and (S)J , being assumed a B(H)-ideal, is hereditary. Clearly Xp, with its cannonical vector space structure, is a linear subspace of (S)J . Under the natural projection map, the set of cosets of elements of Xp in the quotient space (S)J /(S)0 J is given by the linear subspace X We first show that the map P⊕ aiDi → [P⊕ aiDi] is a one-to-one map, that is, each coset [P⊕ aiDi] has a unique element of the form P⊕ aiDi. Indeed, if [P⊕ aiDi] = [P⊕ a impliesP⊕(ai −a multiplyingP⊕(ai −a i)Di] = [0]. This i0 . Since (S)J is a B(H)-ideal, i0 )Di0 ∈ (S)J, and hence Di0 ∈ (S)J. But (T ) = (Di0 ) ⊂ (S)J implying T ∈ (S)J, again contradicting T ∈ (S) \ (S)J. i)Di by a suitable projection it follows that the diagonal operator (ai0 −a J ⊂ (S)J. Suppose there exist i0 such that ai0 6= a iDi], then [P⊕(ai − a i)Di ∈ (S)0 p := {[P⊕ aiDi] haii p < ∞}. ′ ′ ′ ′ ′ ′ ′ 6 SASMITA PATNAIK AND GARY WEISS Therefore ai = a ′ i for all i ≥ 1. So the mapP⊕ aiDi → [P⊕ aiDi] is a one-to-one map which is clearly linear, and therefore the map haii → [P⊕ aiDi] is an isomorphism from ℓp onto X to show that [P⊕ aiDi] := haii p is a well-defined complete norm on X linear map is an isometric isomorphism under this induced norm on X property.) ′ p. (In fact, it is straightforward p which establishes that this p, but we will not exploit this isometric ′ ′ ℓp being an infinite-dimensional Banach space over the complex numbers, the cardinality of a Hamel basis for ℓp is at least c ([4, Lemma 3.4]). Then likewise a Hamel basis B p is at least c, since isomorphisms p is a vector subspace of (S)J /(S)0 J , every Hamel basis of a subspace can be preserve Hamel bases. Since X extended to a Hamel basis of the full space and because the cardinality of all Hamel bases of a vector space J . Also since the generating set S for is invariant, it follows that B (S)J has cardinality strictly less than c, B < c by Proposition 3.1. Therefore c ≤ B ≤ B < c, a set theoretic contradiction. To sum up, this contradiction followed from assuming properness of the inclusion (S)J ( (S). Therefore (S)J = (S), that is, (S) is J-soft. ≤ B for B a Hamel basis of (S)J /(S)0 for X ′ ′ ′ ′ ′ Next we prove the first implication of the second assertion of this theorem, that is, if I is a B(H)-ideal, then (I) is J-soft. Following the same method as used above for I = (S)J , notice that the contradiction arose from assuming properness of the inclusion (I)J ( (I), that is, we showed there how the assumption I 0 without depending on of (I)J 6= (I) led to an imbedding of X cardinality of S, and yet still violating dim I p (an isometric isomorphic copy of ℓp) into I ′ I 0 < c. ⇐: From general ring theory, for S = {Sα}α∈A: (S)J = {(Sα1)J + (Sα2 )J + · · · + (Sαj )J } [{α1,··· ,αj }⊂A where A is an index set with A < c. Using the algebraic structure for principal J-ideals implied by Theorem 2.7 (or for principal J-ideals in particular, see also [8, Proposition 4.2]), one obtains (S)J = span S + (S)0 J . Using Remark 2.3 for (S)0 J , one obtains (S)J = span S + span{SJ + J S} + J(S)J. Then (S) ⊂ (S)J because (S) = (S)J and the commutativity of B(H)-ideal multiplication implies (S) = J(S)J [2, Sections 2.8, 4.3]. But the minimality of (S)J as a J-ideal containing S implies (S)J ⊂ (S), and so combining both inclusions, one obtains (S) = (S)J , that is, (S)J is a B(H)-ideal. Finally, we prove the second implication of the second assertion of this theorem, that is, if (I) is J-soft, then I is a B(H)-ideal. Notice that when (I) is J-soft, one obtains (I) = (I)J = J(I)J, hence I ⊂ J(I)J. And J(I)J ⊂ I because I is a J-ideal. Combining both inclusions one obtains I = J(I)J which is a B(H)-ideal. (cid:3) Remark 4.2. In Theorem 4.1 only one implication (i.e., (S)J is a B(H)-ideal ⇒ (S) is J-soft) requires the cardinality of S to be strictly less than the continuum. The reverse implication holds for arbitrary S. Question 4.3. Is the codimension condition equivalent to the J-ideal I possessing a generating set of cardinality less than c. (See also Section 6, Question 5.) As a consequence of Theorem 4.1 we have Theorem 4.4. (Structure theorem for (S)J when S < c) The algebraic structure of the J-ideal (S)J generated by the set S is given by (S)J = span{S + J S + SJ} + J(S)J J(S)J is a B(H)-ideal, span{J S + SJ} + J(S)J is a J-ideal, and J(S)J ⊂ span{J S + SJ} + J(S)J ⊂ (S)J . This inclusion collapse to J(S)J = (S)J if and only if (S) is J-soft. The fact that the cardinality of every nonzero J-ideal I is c (Remark 3.3) implies that generating sets for I have at most c elements. So in view of Theorem 4.1, the only J-softness cases left to investigate are: if I cannot be generated by fewer than c elements or (possibly more general, see Question 4.3) at least if the Hamel dimension of the quotient space I/I 0 is equal to c, does either of these imply I is J-soft? Indeed, we show in Example 4.5 that Theorem 4.1 is the best possible result of its type by giving an example of a J-ideal that is also a B(H)-ideal which is not J-soft. By the contrapositive of Theorem 4.1, this J-ideal has no generating sets of cardinality less than c and the Hamel dimension of its quotient I/I 0 is precisely c. Example 4.5. Consider J = K(H) and I = (diag(cid:10) 1 operator diag(cid:10) 1 n(cid:11)) the principal B(H)-ideal generated by the diagonal n(cid:11). Since every B(H)-ideal is a J-ideal, I is also a J-ideal. We will show that the Hamel SUBIDEALS OF OPERATORS II 7 dimension of the quotient space I/I 0 is precisely c, but yet I is not K(H)-soft. Indeed, I 0 = span{IK(H) + n(cid:11), imbed X ′ n(cid:11) /∈ I 0 [8, Example 3.3]. In the proof of Theorem p into I/I 0 for any 1 ≤ p ≤ ∞. So the Hamel dimension of the quotient space I/I 0 is at least c. Since I is a nonzero J-ideal, the Hamel dimension of I is equal to c (Remark 3.3), so the Hamel dimension of the quotient space I/I 0 is at most c. Therefore the Hamel dimension of I/I 0 is K(H)I} + K(H)IK(H) and one can show that diag(cid:10) 1 4.1 taking T = diag(cid:10) 1 n(cid:11)) is not K(H)-soft [8, Example 3.3], but for completeness we repeat equal to c. But we know that (diag(cid:10) 1 the proof here. If it were K(H)-soft, then (diag(cid:10) 1 n(cid:11) ∈ Σ((diag(cid:10) 1 n(cid:11))K(H)), i.e., (cid:10) 1 n(cid:11) = o(Dm(cid:10) 1 (cid:10) 1 n(cid:11)) = (diag(cid:10) 1 n(cid:11)) for some m ∈ N, contradicting (cid:18) h 1 n(cid:11))K(H) which further implies n i(cid:19)k n i Dmh 1 → 1 m mj+r = j+1 1 1 as k → ∞ where k = mj + r. 5. Questions and results on J -ideals In this section we address some of the questions posed in [8, Sections 6-7] and pose new questions. The algebraic structure of a principal J-ideal generated by S is (S)J = CS + SJ + JS + J(S)J. For idempotent B(H)-ideals J (i.e., J 2 = J), J(S)J = SJ + JS + J(S)J since J(S)J = (S)J 2 = (S)J and SJ, JS ⊂ (S)J = J(S). So the algebraic structure of (S)J simplifies to (S)J = CS + J(S)J. Is it necessary for J to be idempotent for J(S)J = SJ + JS + J(S)J to hold? (This is related to [8, Remark 6.4, Section 7-Question 6]: Find a necessary and sufficient condition(s) to make J(S)J = JS + SJ + J(S)J.) The following example shows that J need not be idempotent for the equality to hold. Example 5.1. For 0 < p < ∞, the Schatten p-ideal Cp is not idempotent because C2 Moreover, for S = diag(cid:10) 1 from the Cp-softness of (S). Indeed, for m > 2, recall that (Dm(s(S)))n = s⌈ n nn(cid:11) ∈ Cp we claim that Cp(S)Cp = SCp + CpS + Cp(S)Cp which will follow below p = Cp/2 6= Cp. sn(S) (Dm(s(S)))n = (j+1)j+1 (mj+r)mj+r ≤ (2j)2j (mj)mj < (mj)2j (mj)mj = m ⌉(S), so m(m−2)j j(m−2)j ≤ 1 jj , 1 where n = mj + r and the roof function ⌈ n which further implies that s(S) = s(S) of (S), so (S) = Cp(S)Cp. Because C2 SCp ⊂ Cp(S)Cp. Therefore SCp + CpS ⊂ Cp(S)Cp, hence Cp(S)Cp = SCp + CpS + Cp(S)Cp. Dm(s(S)) ∈ Σ(Cp) Dm(s(S)) Dm(s(S)) ∈ Σ(Cp(S)). Hence (S) = Cp(S) which is Cp-softness p (S)Cp ⊂ Cp(S)Cp and similarly m ⌉ = j. By the hereditary property of Σ(Cp), p ⊂ Cp, it follows that CpS ⊂ C2 s(S) Finishing this discussion on JS +SJ +J(S)J, recall that in the case of a principal J-ideal (S)J , JS +SJ + J(S)J is always a maximal J-ideal in (S)J [8, Remark 6.3]. The following example gives a partial answer to [8, Section 7, Question 3]: is JS + SJ + J(S)J always a principal J-ideal or is it always a non-principal J-ideal? We claim that this is not a principal K(H)-ideal. n(cid:11), Example 5.2. For J = K(H) and S = diag(cid:10) 1 n(cid:11) + diag(cid:10) 1 JS + SJ + J(S)J = K(H) diag(cid:10) 1 n(cid:11))K(H) and K(H) diag(cid:10) 1 n(cid:11))K(H) = (diag(cid:10) 1 K(H)(diag(cid:10) 1 fore (T )K(H) = (diag(cid:10) 1 ideal see also Theorem 1.2]. Since T ∈ (diag(cid:10) 1 Xl=1 Suppose JS + SJ + J(S)J = (T )K(H). Since B(H)-ideals commute and K(H)2 = K(H), one has n(cid:11))K(H). There- n(cid:11))K(H) hence (T )K(H) is the B(H)-ideal (T ) [8, Theorem 2.6, for principal J- n(cid:11) ρ) for some ρ = hρni ∈ c∗ n(cid:11))K(H), s(T ) = O(Dm(cid:10) 1 0 [6, Section 1, ¶1], and without loss of generality one can assume ρ1 = 1. Observe that the sequence ∞ n(cid:11) K(H) + K(H)(diag(cid:10) 1 n(cid:11) + diag(cid:10) 1 n(cid:11))K(H) n(cid:11) K(H) ⊂ (diag(cid:10) 1 norm limit of the sequences 0 (the cone of positive sequences decreasing to 0) because it is the c0- 1 2l (Dl(ρ1/l)) ∈ Σ(K(H)) = c∗ 1 Xl=1 2l (Dl(ρ1/l)) as n → ∞ and c0 (the sequence space of complex numbers n(cid:29))) = Σ((T )). n(cid:29) ∈ Σ(K(H)(diag(cid:28) 1 2l (Dl(ρ1/l)) · (cid:28) 1 tending to 0) is norm-closed. So in particular, ∞ 1 n Xl=1 8 So ∞ Xl=1 SASMITA PATNAIK AND GARY WEISS 1 2l (Dl(ρ1/l)) ·(cid:28) 1 n(cid:29) ∈ Σ((T )), and expressing this in terms of s-numbers, for some k, m ∈ N and all j = kmi + r for some i and some 0 ≤ r < km, ∞ Xl=1 1 2l (Dl(ρ1/l)) ·(cid:28) 1 n(cid:29) = O(Dks(T )) = O(Dk(Dm(cid:28) 1 n(cid:29) ρ)) = O(Dkm(cid:28) 1 n(cid:29) Dk(ρ)), thereby contradicting, after setting arbitrary j = kmi + r, for some i and 0 ≤ r ≤ km, n i h 1 ∞ Xl=1 1 2l (Dl(ρ1/l)) n iDk(ρ) Dkmh 1 1 j = ∞ Xl=1 1 2l (Dl(ρ1/l))j n iDk(ρ))j (Dkmh 1 1 j ∞ Xl=1 1 2l (Dl(ρ1/l))j n iDkm(ρ))j ≥ (Dkmh 1 ≥ 1 1 2km (Dkm(ρ j (Dkm( 1 1 km ))j n )Dkm(ρ))j = 1 (kmi+r) 2km 1 (i+1) ρ 1− 1 km i+1   =  j (i+1) 2km(kmi+r)ρ 1− 1 km i+1 which diverges to ∞ as j → ∞ since km > 1 and ρi → 0. For principal B(H)-ideals (S), (T ), (S) = (T ) if and only if s(S) = O(Dms(T )) and s(T ) = O(Dks(S)) for some m, k ∈ N. When s(S) or s(T ) satisfies the △1/2 condition, a simpler condition is: (S) = (T ) if and only if s(S) = O(s(T )) and s(T ) = O(s(S)). (See [5, Section 2, ¶1].) The following proposition gives a necessary and sufficient condition for two principal J-ideals to be equal [8, Section 7, Question 4]. We hope for a simpler condition. Proposition 5.3. For S, T ∈ J, (S)J = (T )J if and only if for some nonzero a, b ∈ C. aS + bT ∈ {SJ + JS + J(S)J}T{T J + JT + J(T )J} Proof. ⇒: Based on the algebraic structure of principal J-ideals implied by Theorem 2.7 (or for principal J-ideals in particular, see also [8, Proposition 2.3]), (S)J = (T )J if and only if S = αT + AT + T B + n Xi=1 AiT Bi and T = βS + CS + SD + m Xj=1 CjSDj for some α, β ∈ C, A, B, C, D, Ai, Bi, Ci, Di ∈ J. If α = 0 and β 6= 0, then substituting S from the first equation in the second equation yields T ∈ (T )J and then substituting T from the second equation in the first equation yields S ∈ (S)J. Therefore in this case (T ) = J(T )J = (T )J and (S) = J(S)J = (S)J implied by Theorem 2.6 (or for principal J-ideals in particular, see also [8, Theorem 1.2]). Since J(S)J = (S)J = (T )J = J(T )J, one has S, T ∈ J(S)JT J(T )J, and because J(S)J T J(T )J is a J- ideal, aS + bT ∈ {SJ + JS + J(S)J}T{T J + JT + J(T )J} for all a, b ∈ C. The cases α 6= 0 and β = 0 and α = 0 = β are handled similarly. Assume finally that α, β 6= 0. Substituting T from the second equation in the first equation one obtains S = αβS + X, where X ∈ (S)J. If αβ 6= 1, then S ∈ {SJ + JS + J(S)J}, and then, substituting S from the first equation in the second equation, one obtains T = αβT + Y where Y ∈ (T )J, so T ∈ {T J + JT + J(T )J}. This implies J-softness of (S) and (T ), hence J(S)J = (S)J = (T )J = J(T )J and then as above aS +bT ∈ {SJ +JS +J(S)J}T{T J +JT +J(T )J} for all n α , so substituting β in the second equation yields S−αT = AT +T B+ AiT Bi a, b ∈ C. When αβ = 1, β = 1 Xi=1 and S − αT = −α(CS + SD + which is the required condition. m Xj=1 Cj SDj). Therefore S − αT ∈ {SJ + JS + J(S)J}T{T J + JT + J(T )J} ⇐: Suppose aS + bT ∈ {SJ + JS + J(S)J}T{T J + JT + J(T )J} for some nonzero a, b ∈ C. Then m n aS + bT = AT + T B + AiT Bi and aS + bT = CS + SD + Cj SDj Xj=1 Xi=1 for some α, β ∈ C, A, B, C, D, Ai, Bi, Ci, Di ∈ J. From these equalities its clear that (S)J = (T )J . (cid:3) Unlike B(H)-ideals, J-ideals do not necessarily commute as given in the following example. SUBIDEALS OF OPERATORS II 9 Example 5.4. Consider J = K(H) and with respect to the standard basis take S to be the diagonal matrix S := diag (1, 0, 1/2, 0, 1/3, 0, ...) and T to be the weighted shift with this same weight sequence. We claim (S)K(H)(T )K(H) 6= (T )K(H)(S)K(H). Indeed, suppose (S)K(H)(T )K(H) = (T )K(H)(S)K(H). Then T S −ST ∈ n2(cid:11)). But n(cid:11)), the latter inclusion n2(cid:11)), a (S)K(H)(T ) = K(H)(diag(cid:10) 1 T ∈ K(H)(diag(cid:10) 1 contradicts diag(cid:10) 1 n2(cid:11)) if and only if diag(cid:10) 1 n(cid:11) /∈ K(H)(diag(cid:10) 1 contradiction. Therefore (S)K(H)(T )K(H) 6= (T )K(H)(S)K(H). n2(cid:11)). Since T S = T and ST = 0, one has T ∈ K(H)(diag(cid:10) 1 n(cid:11)) (Example 4.5 or [8, Example 3.3]), so T /∈ K(H)(diag(cid:10) 1 n(cid:11) ∈ K(H)(diag(cid:10) 1 n2(cid:11)) ⊂ K(H)(diag(cid:10) 1 1. Determine which J-ideals have generating sets of cardinality less than c? 2. Do J-ideals have minimal cardinality generating sets? QUESTIONS 3. Traces. B(H)-ideals are important in part because of the importance of their traces, unitarily invariant functionals on B(H)-ideals. Because B(H)-ideals J contain no unitaries this concept does not apply. Is there a modification for which one obtains a useful concept of traces for J-ideals? 4. The study of commutators in B(H)-ideals is directly related to their traces. What can be said about the commutator structure of the J-ideals that can be used to motivate notions of traces for J-ideals? 5. Is the dimension of I/I 0 the cardinality of some generating set for I? At least if dimension I/I 0 < c, must I possess a generating set of cardinality less than c? (See also Question 4.3.) [1] Blass, Andreas, Existence of bases implies the axiom of choice, Amer. Math. Soc., (1984), 31 -- 33. References [2] Dykema, K., Figiel, T., Weiss, G., and Wodzicki, M., The commutator structure of operator ideals, Adv. Math., 185 (1), (2004), 1 -- 79. [3] Fong, C.K. and Radjavi, H., On ideals and Lie Ideals of Compact Operators, Math. Ann. 262, (1983), 23 -- 28. [4] Halbeisen, Lorenz and Hungerbuhler, Norbert, The cardinality of Hamel bases of Banach spaces, East-West J. Math., (2000) 153 -- 159. [5] Kaftal, V and Weiss, G, B(H) lattices, density and arithmetic mean ideals, Houston J. Math., (2011), 233 -- 283. [6] Kaftal, V. and Weiss, G., Soft ideals and arithmetic mean ideals, Integral equations and Operator Theory, (2007), 363 -- 405. [7] Mityagin, B.S., Normed ideals of intermediate type, Amer. Math. Soc. Transl., (2) 63, (2000), 180-194. [8] Patnaik, S and Weiss, G, Subideals of Operators, Journal of Operator Theory, to appear. [9] Pietsch, Albrecht, Operator ideals, North-Holland Mathematical Library, North-Holland Publishing Co., 20 (1980). University of Cincinnati, Department of Mathematics, Cincinnati, OH, 45221-0025, USA E-mail address: sasmita [email protected] University of Cincinnati, Department of Mathematics, Cincinnati, OH, 45221-0025, USA E-mail address: [email protected]
0710.3460	2	0710	2013-03-12T16:31:31	Boundary $C^*$-algebras for acylindrical groups	[ "math.OA", "math.GR" ]	Let $\Delta$ be an infinite, locally finite tree with more than two ends. Let $\Gamma<\aut(\Delta)$ be an acylindrical uniform lattice. Then the boundary algebra $\cl A_\Gamma = C(\partial\Delta)\rtimes \Gamma$ is a simple Cuntz-Krieger algebra whose K-theory is determined explicitly.	math.OA	math	BOUNDARY C ∗-ALGEBRAS FOR ACYLINDRICAL GROUPS GUYAN ROBERTSON Abstract. Let ∆ be an infinite, locally finite tree with more than two ends. Let Γ < Aut(∆) be an acylindrical uniform lattice. Then the boundary algebra AΓ = C(∂∆) ⋊ Γ is a simple Cuntz-Krieger algebra whose K-theory is determined explicitly. 1. Introduction Let ∆ be an infinite, locally finite tree with more than two ends and with boundary ∂∆. Let k be a positive integer and let Γ be a group of automorphisms of ∆ without inversion and with no proper invariant subtree. Say that Γ is k-acylindrical if the stabilizer of any path of length k in ∆ is trivial [BP, Sel]. The group Γ is acylindrical if it is k-acylindrical for some integer k ≥ 1. The main result of this article is Theorem 1.1. Let Γ < Aut(∆) be an acylindrical uniform lattice. Then the boundary algebra AΓ = C(∂∆) ⋊ Γ is a simple Cuntz-Krieger algebra. The action of Γ on ∂∆ is amenable, so the maximal crossed product AΓ = C(∂∆) ⋊ Γ coincides with the reduced crossed product and is nuclear by Proposition 4.8 and Th´eor`eme 4.5 in [AD]. The algebra AΓ is described in Section 4 below. Cuntz-Krieger algebras were introduced in [CK] and are classified up to isomorphism by their K-theory [K]. A special case, where Γ is a free uniform tree lattice, was studied in [R2] by different methods. The K-groups of the boundary algebra AΓ are isomorphic to the Bowen-Franks invariants of flow equivalence for a certain subshift of finite type associated with the geodesic flow [C3]. This subshift was studied in [BP, 6.3]. The K-groups of the algebra AΓ may be computed explicitly. For example, if Γ = Zl+1 ∗ Zm+1 acts on its Bass-Serre tree, where l, m ≥ 1, then K0(AΓ) = Zlm−1 via an isomorphism sending the class of the ∼= Ml+1 ⊗ Olm, where identity idempotent to l + 1. It follows that AΓ 2000 Mathematics Subject Classification. Primary 20E08, 46L80. Key words and phrases. Acylindrical group, boundary, Cuntz-Krieger algebra. 1 2 GUYAN ROBERTSON On denotes the Cuntz algebra, which is generated by n isometries on a Hilbert space whose range projections sum to the identity operator [C1]. Remark 1.2. The algebra AΓ depends only on Γ. The group Γ and the tree ∆ are quasi-isometric; more precisely, for any base vertex in ∆, the natural mapping from Γ onto the orbit of this vertex is a quasi-isometry from Γ to ∆. This is a special case of the "Fundamental Observation of Geometric Group Theory" [Ha, Theorem IV.23]. The mapping from Γ to ∆ induces a Γ-equivariant homeomorphism of the boundaries of Γ and ∆. 2. Background The edges of the tree ∆ are directed, and each geometric edge of ∆ corresponds to two directed edges. Let ∆0 denote the set of vertices and ∆1 the set of directed edges of ∆. There is a distance function d defined on the geometric realization of ∆ which assigns unit length to each edge. Choose an orientation on the set of edges which is invariant under Γ. This orientation consists of a partition of ∆1 and a bijective involution e 7→ e : ∆1 → ∆1 which interchanges the two components of ∆1. Each directed edge e has an initial vertex o(e) and a terminal vertex t(e) such that o(e) = t(e). Let Γ be a group of automorphisms of ∆ without inversion and with no proper invariant subtree. Say that Γ is k-acylindrical, where k ≥ 1, if the stabilizer of any path of length k in ∆ is trivial [BP]. (In [Sel] such a group Γ is said to be (k−1)-acylindrical.) To say that Γ is 1-acylindrical is the same as saying that Γ acts freely on ∆1. For example, the action of a free product Γ = Γ1∗Γ2 on the associated Bass-Serre tree [Ser, I.4.1] is 1-acylindrical. If Γ = Γ1 ∗Γ0 Γ2 is a free product with amalgamation over Γ0, then the action of Γ on the associated Bass-Serre tree is 2- acylindrical if Γ0 is malnormal in each Γj, i.e. g−1Γ0g ∩ Γ0 = {1} for all g ∈ Γj − Γ0, j = 1, 2. Every small splitting of a torsion free hyperbolic group gives rise to a 3-acylindrical action [Sel]. The boundary ∂∆ is the set of equivalence classes of infinite semi- geodesics in ∆, where two semi-geodesics are said to be equivalent if they agree except on finitely many edges. For the rest of this article we make the following assumptions. Standing Hypotheses (1): ∆ is an infinite locally finite tree with more than two bound- ary points. BOUNDARY C ∗-ALGEBRAS FOR ACYLINDRICAL GROUPS 3 (2): Γ < Aut(∆) is a uniform tree lattice. (3): Γ acts without inversion and with no proper invariant sub- tree. (4): Γ is k-acylindrical, where k ≥ 1. Remark 2.1. The standing hypotheses imply that ∂∆ is uncountable. Remark 2.2. The assumption that Γ has no proper invariant subtree is part of the definition of "acylindrical" in [Sel] (but not in [BP], which is why it is emphasised separately here). It implies that the action of Γ on ∂∆ is minimal. The assumption could have been omitted here if ∂∆ were replaced throughout by the limit set ΛΓ. Remark 2.3. The assumption that Γ is a uniform lattice is natural and necessary in the context of the theory of [RS]. It would be interesting to study AΓ for non-uniform tree lattices. For example, [BP] provides an example of a 5-acylindrical action of the Nagao lattice PGL2(Fq[t]). 3. Cuntz-Krieger algebras It is convenient to use the approach to Cuntz-Krieger algebras de- veloped in [RS]. Choose a nonzero matrix M with entries in {0, 1}. For m ≥ 0, let Wm denote the set of all words of length m + 1 based on the alphabet A and the transition matrix M. A word w ∈ Wm is a formal product w = a0a1 . . . am, where aj ∈ A and M(aj+1, aj) = 1, 0 ≤ j ≤ m − 1. Define o(w) = a0 and t(w) = am. Fix a nonempty finite or countable set D (whose elements are "deco- rations") and a map δ : D → A. Let W m = {(d, w) ∈ D × Wm; o(w) = δ(d)}, the set of "decorated words" of length m+1, and identify D with W 0 via the map d 7→ (d, δ(d)). Let W = Sm Wm and W = Sm W m, the sets of all words and all decorated words respectively. Define o : W m → D and t : W m → A by o(d, w) = d and t(d, w) = t(w). Let u = a0a1 . . . am ∈ Wm and v = b0b1 . . . bn ∈ Wn. If t(u) = o(v), then there exists a unique product uv ∈ Wm+n defined by uv = a0a1 . . . amb1 . . . bn. If w = a0a1 . . . al ∈ Wl where l ≥ 0 and if p 6= 0, we say that w is p-periodic if aj+p = aj whenever both sides are defined. Assume that the nonzero {0, 1}-matrix M has been chosen so that the following conditions from [RS] hold. (H2): If a, b ∈ A then there exists w ∈ W such that o(w) = a and t(w) = b. (H3): For each nonzero integer p, there exists some w ∈ W which is not p-periodic. 4 GUYAN ROBERTSON Definition 3.1. [RS] The C ∗-algebra AD = A(A, D, M) is the univer- sal C ∗-algebra generated by a family of partial isometries {su,v; u, v ∈ W and t(u) = t(v)} satisfying the relations (3.1a) (3.1b) (3.1c) su,v ∗ = sv,u su,vsv,w = su,w su,v = Xw∈W1, o(w)=t(u)=t(v) suw,vw (3.1d) su,usv,v = 0, for u, v ∈ W 0, u 6= v. Remark 3.2. If D is finite (as is the case in this article), then AD is isomorphic to a simple Cuntz-Krieger algebra with identity element su,u [R1]. 1 = Xu∈W 0 Remark 3.3. Two decorations δ1 : D1 → A and δ2 : D2 → A are said to be equivalent [RS, Section 5] if there is a bijection η : D1 → D2 such that δ1 = δ2η. Equivalent decorations δ1, δ2 give rise to isomorphic algebras AD1, AD2. Remark 3.4. Denote by AA the Cuntz-Krieger algebra with decorating set A and with δ the identity map. The algebra AA is isomorphic to the algebra OM t generated by a set of partial isometries {Sa; a ∈ A} satisfying the relations S∗ b [RS, Remark 3.11]. If A contains n elements and M(b, a) = 1, for all a, b ∈ A, then AA is the Cuntz algebra On generated by n isometries whose range projections sum to the identity operator [C1]. aSa = Pb M(b, a)SbS∗ 4. The algebra associated with an acylindrical group The Standing Hypotheses (1) -- (4) are now in force. A geodesic γ in ∆ is a sequence (sj)∞ j=−∞ of vertices such that d(si, sj) = i − j. A directed segment σ of length n is a sequence (s0, s1, . . . , sn) of vertices such that d(si, sj) = i − j. Denote such a directed segment by [s0, sn] and let Sn be the set of directed segments of length n in ∆. Since the group Γ is k-acylindrical, Γ acts freely on the set Sk. The alphabet A is defined to be Γ\Sk+1, the set of Γ-orbits of di- rected segments of length k + 1 in ∆. Since Γ is a uniform lattice, A is finite. Define a matrix M with entries in {0, 1} as follows. If a, b ∈ A, we say that M(b, a) = 1 if and only if a = Γσ and b = Γτ , where σ = (s0, s1, . . . , sk+1), τ = (t0, t1, . . . , tk+1) are directed segments such that sj+1 = tj, 0 ≤ j ≤ k. The definition is illustrated in Figure 1. BOUNDARY C ∗-ALGEBRAS FOR ACYLINDRICAL GROUPS 5 t0 • • s0 • • • sk+1 tk+1 • Figure 1. The condition M(b, a) = 1. As in Section 3, Wm denotes the set of all words of length m+1 based on the alphabet A and transition matrix M. Let Wm = Γ\Sm+k+1 and let W =Sm defined by Wm. There is a map α : Wm → Wm α(Γ(s0, s1, . . . , sm+k+1)) = (Γ[s0, sk+1])(Γ[s1, sk+2]) . . . (Γ[sm, sm+k+1]). Lemma 4.1. The map α is a bijection from Wm onto Wm. Proof. Suppose that α(Γσ) = α(Γτ ), where σ = (s0, s1, . . . , sm+k+1) and τ = (t0, t1, . . . , tm+k+1). Then Γ[sj, sj+k+1] = Γ[tj, tj+k+1], 0 ≤ j ≤ m. For each j there exists gj ∈ Γ such that gj[sj, sj+k+1] = [tj, tj+k+1], and gj is uniquely determined, since Γ acts freely on the set of segments of length k. Now if 0 ≤ j ≤ m − 1, then [sj, sj+k+1] ∩ [sj+1, sj+k+2] = [sj+1, sj+k+1] is a segment of length k. Also gj[sj+1, sj+k+1] = [tj+1, tj+k+1] = gj+1[sj+1, sj+k+1]. Therefore gj = gj+1. It follows that g0σ = τ and so Γσ = Γτ . This proves injectivity. To prove surjectivity, suppose that w = a0a1 . . . am ∈ Wm. Then by definition aj = Γσj for 0 ≤ j ≤ m, where σj ∩ σj+1 is a segment of length k, for 1 ≤ j ≤ m − 1. It follows that there is a directed segment σ = (s0, s1, . . . , sm+k+1) ∈ Sm+k+1 such that σj = [sj, sj+k+1], 0 ≤ j ≤ m. Thus α(Γσ) = w. (cid:3) Now fix a vertex P ∈ ∆0. Let Wm denote the set of directed segments Wm. The decorating set is D = W0, and the decorating map δ : D → A is defined by δ(d) = Γd. Define α : W → W by of length m + k + 1 which begin at P and let W = Sm≥0 α(σ) = (o(σ), α(Γσ)) , where o(σ) = (s0, s1, . . . , sk+1) is the initial segment of length k + 1 of σ. Also define t(σ) to be the final segment of length k + 1 of σ. Lemma 4.2. The map α is a bijection from Wm onto W m, for each m ≥ 0. Proof. If α(σ1) = α(σ2), then o(σ1) = o(σ2); moreover Γσ1 = Γσ2 by Lemma 4.1. Since Γ acts freely on Sk, it follows that σ1 = σ2. Therefore α is injective. 6 GUYAN ROBERTSON To see that α is surjective, let w = (d, w) ∈ W m, where w ∈ Wm and d ∈ D. By Lemma 4.1, there exists σ ∈ Sm+k+1 such that α(Γσ) = w. Now Γd = δ(d) = o(w) = o(α(Γσ)) = Γo(σ). Replacing σ by gσ for suitable g ∈ Γ ensures that o(σ) = d. Then σ ∈ Wm and α(σ) = w. (cid:3) Recall that each ω ∈ ∂∆ is represented by a unique semi-geodesic If σ is a directed segment with [s0, ω) with initial vertex s0 ∈ ∆0. initial vertex t0, let Ω(σ) = {ω ∈ ∂∆ : [t0, ω) contains σ} . σ ................ ...................................... .......................................................................................................................................................................................................................................................................... •t0 ............................................................................................................................................................... ....................................................................................................................................................... • ....................................................................................................................................................... ······················ Ω(σ) The boundary ∂∆ has a natural compact totally disconnected topol- ogy generated by sets of the form Ω(σ) where σ ∈ W [Ser, I.2.2]. The group Γ acts on ∂∆, and one can form the crossed product C ∗-algebra C(∂∆) ⋊ Γ. This is the universal C ∗-algebra generated by the commu- tative C ∗-algebra C(∂∆) and the image of a unitary representation π of Γ, satisfying the covariance relation (4.1) f (g−1ω) = π(g) · f · π(g)−1(ω) for f ∈ C(∂∆), g ∈ Γ and ω ∈ ∂∆. It is convenient to denote π(g) simply by g. Equation (4.1) implies that for each clopen set E ⊂ ∂∆, (4.2) χgE = g · χE · g−1. The indicator function χE is continuous and is regarded as an element of the crossed product algebra via the embedding C(∂∆) ⊂ C(∂∆) ⋊Γ. The following is a more precise version of Theorem 1.1. Theorem 4.3. Let AΓ = C(∂∆) ⋊ Γ. Then AΓ is isomorphic to the Cuntz-Krieger algebra AD associated with the alphabet A, the decorating set D and the transition matrix M. Proof. The isomorphism φ : AD → C(∂∆) ⋊ Γ is defined as follows. Let wj = α(σj) ∈ W , j = 1, 2, with t(w1) = t(w2). By the definition of α, there is an element g ∈ Γ such that gt(σ1) = t(σ2). Recall that t(σ1) BOUNDARY C ∗-ALGEBRAS FOR ACYLINDRICAL GROUPS 7 and t(σ2) are segments of length k + 1. Therefore g is unique, since Γ acts freely on Sk. Define the homomorphism φ by φ(sw2,w1) = gχΩ(σ1) = χΩ(σ2)g. This equation defines a ∗-homomorphism of AΓ because the operators of the form φ(sw2,w1) are easily seen to satisfy the relations (3.1). Since the algebra AD is simple [RS, Theorem 5.9], φ is injective. Now χΩ(σ) = φ(sw,w), where σ ∈ W and w = α(σ). Since the sets Ω(σ), σ ∈ W, form a basis for the topology of Ω, the linear span of {χΩ(σ); σ ∈ W} is dense in C(∂∆). It follows that the range of φ contains C(∂∆). To show that φ is surjective, it therefore suffices to show that the range of φ contains Γ. Let g ∈ Γ and choose an integer m ≥ d(P, g−1P ). Let σ ∈ Wm. Then σ is a directed segment of length m + k + 1 with initial vertex P and final vertex Q, say. Let σ′′ be the directed segment with initial vertex g−1P and final vertex Q. Since m ≥ d(P, g−1P ), it follows that t(σ′′) = t(σ). g−1P P • • ................................................................................................................................. t(σ) ····························· • • Q Let σ′ = gσ′′. Then σ′ is a path beginning at P and t(σ′) = gt(σ). Let w1 = α(σ) and w2 = α(σ′). Then t(w1) = t(w2) and gχΩ(σ) = φ(sw2,w1) ∈ φ(AD). This holds for each σ ∈ Wm. Therefore gχΩ(σ) ∈ φ(AD). g = Xσ∈Wm This shows that the range of φ contains Γ, as required. (cid:3) We now prove that conditions (H2) and (H3) are satisfied. To verify condition (H2), it is enough to show that if a, b ∈ A, then there is a directed segment σ such that (4.3) a = Γo(σ), b = Γt(σ). Let a = Γσ1, b = Γτ2, where σ1, τ2 ∈ Sk+1. By [Ch, Proposition 1 (iii)], there is a Γ-periodic geodesic γ containing τ2. By definition, this means that there is a subgroup of Γ which leaves the geodesic γ invariant and acts upon it by translation. Choose ω ∈ ∂∆ to be the boundary point of γ with τ2 ⊂ [o(τ2), ω). Since the action of Γ on ∂∆ is minimal, there exists g ∈ Γ such that gω ∈ Ω(σ1). The geodesic gγ is Γ-periodic. Therefore the semi-geodesic 8 GUYAN ROBERTSON [go(τ2), gω) contains infinitely many directed segments σ2 which are Γ- translates of τ2. Choose such a segment σ2 far enough away from gτ2 so that σ2 ∈ Ω(σ1). Let σ be the directed segment with o(σ) = σ1 and t(σ) = σ2. Then (4.3) is satisfied. • go(τ2) gτ2 ······························ ...................................................................................................................................................................................... • ........................................................................................................................................................................................................................................................................................................................................................................... σ1 ····································· • • σ2 ··································· • • ................ ............................ gω To prove that condition (H3) holds, let p > 0. Since ∆ has more than two ends, there exist vertices of ∆ which have degree greater than 2. Let σ = (s0, s1, . . . , sp+k) ∈ Sp+k be a directed segment whose final vertex sp+k has degree greater than 2. Extend σ to two different segments: (s0, s1, . . . , sp+k, t), (s0, s1, . . . , sp+k, t′) ∈ Sp+k+1. s0 • • ..... ...... ..... ...... ...... ..... ...... ...... ..... ...... ...... ..... ..... ...... ...... sp+k+1 ...... ..... • ..... ........ ...... ..... ...... ...... ..... ...... ...... ..... ...... ...... ..... ...... ...... ...... • .. t t′ Let w = α(Γ(s0, . . . , sp+k, t)) and w′ = α(Γ(s0, . . . , sp+k, t′)). Then w, w′ ∈ Wp, o(w) = o(w′) and t(w) 6= t(w′) since Γ acts freely on Sk and α is injective. Therefore at least one of the words w, w′ is not p-periodic. 5. K-theory and examples The Standing Hypotheses (1) -- (4) remain in force. Thus the algebra AΓ is isomorphic to the Cuntz-Krieger algebra AD associated with the alphabet A, the set of decorations D and transition matrix M. The simple Cuntz-Krieger algebras AD are purely infinite, nuclear and satisfy the Universal Coefficient Theorem [RS, Remark 6.5]. They are therefore classified by their K-theory [K]. It is convenient to con- sider the related algebra AA which is stably isomorphic to AD. Recall from Remark 3.4 that AA is isomorphic to the algebra OM t. The groups K0(AA), K1(AA) are the Bowen-Franks invariants of flow equivalence for a certain subshift associated with (Γ, ∆). More precisely, according BOUNDARY C ∗-ALGEBRAS FOR ACYLINDRICAL GROUPS 9 to [C3, Proposition 3.1] the group K0(AA) is isomorphic to the abelian group GΓ =A (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) a =Xb∈A M(a, b)b, a ∈ A+ . Note that, as the notation suggests, GΓ depends only on Γ, by Remark 1.2. Also K1(AA) is the torsion free part of GΓ. Therefore AA is classified up to stable isomorphism by the group GΓ. Since the algebra AD is stably isomorphic to AA [RS, Corollary 5.15], we obtain the following result. Theorem 5.1. Under the Standing Hypotheses (1) -- (4), K0(AΓ) ∼= GΓ. To completely classify AA up to isomorphism, we need to identify the class [1] of the identity idempotent in K0(AA) [K]. By Remark 3.2, this class corresponds to the element (5.1) (5.2) a ∈ GΓ. ε =Xa∈A Here are explicit calculations in the case where Γ is a free product of finite cyclic groups. 5.1. Example: The group Γ = Zl+1 ∗ Zm+1 acts on its Bass-Serre tree [Ser, I.4] with an edge y = [P, Q] as its fundamental domain. The stabilizer ΓP of P is isomorphic to Zl+1 and the stabilizer ΓQ of Q is isomorphic to Zm+1. y ................. ..................... •P • Q By construction, Γ acts freely and transitively on the geometric edges of ∆. In other words, Γ is 1-acylindrical. The theory applies, with k = 1, and the alphabet A is the set of Γ-orbits of directed segments of length 2 in ∆. Let A1 = ΓP − {1} and A2 = ΓQ − {1}, so that A1 = l, A2 = m. Each directed segment of length 2 in ∆ lies in the Γ-orbit of one of the directed segments [P, a2P ], [Q, a1Q], for some a1 ∈ A1, a2 ∈ A2. Let a2 = Γ[P, a2P ], a1 = Γ[Q, a1Q] be the corresponding elements of A2, A1. P • y ................. ..................... Q • a2y ..................... ................. a2P • Q • y ..................... ................. P • a1y ................. ..................... a1Q • 10 GUYAN ROBERTSON The map a 7→ a is a bijection from A1 ∪ A2 onto A. The {0, 1}- matrix M is defined by M(a, b) = 1 ⇐⇒ either a ∈ A1, b ∈ A2 or b ∈ A1, a ∈ A2. P • Q = a2Q • a2P • a2a1Q • The condition M(a1, a2) = 1. By Theorem 5.1, (5.3) K0(AΓ) = A1 ∪ A2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) b, a = Xb∈A3−j a ∈ Aj, j = 1, 2+ . The relations on the right side of (5.3) show that all the generators of A1 are equal and all the generators of A2 are equal. Therefore K0(AΓ) = ha1, a2 a2 = la1, a1 = ma2i = ha1 a1 = lma1i = Zlm−1. Recall that the classical Cuntz algebra On is generated by n isome- tries whose range projections sum to the identity operator [C1]. Now K0(On) = Zn−1 [C2]. It follows from the classification theorem [K] that AΓ is stably isomorphic to Olm. In order to classify AΓ up to isomorphism, the class [1] of the identity idempotent in K0(AΓ) must be identified. Now [1] corresponds to the element (l + 1)a1 = a1 + a2 = (m + 1)a2. However it is known [C2] that This proves that AΓ (K0(Mk ⊗ On), [1]) ∼= (Zn−1, k). ∼= Mm+1 ⊗ Olm. ∼= Ml+1 ⊗ Olm 5.2. Example: More generally, a free product of finite groups Γ = Γ1 ∗ Γ2 ∗ · · · ∗ Γn acts on its Bass-Serre tree. The fundamental domain is a tree consisting of n edges yi with terminal vertex Qi emanating from a common vertex P . The stabilizer of P and of each of the edges yj is trivial, and the stabilizer of Qj is isomorphic to Γj. P • y1 y2 ................................................................................................................................................................................................................................................................................................................................ y3 • • Q1 • Q2 Q3 BOUNDARY C ∗-ALGEBRAS FOR ACYLINDRICAL GROUPS 11 The group Γ is 1-acylindrical and acts freely (but not transitively, in contrast to Example 5.1) on the set of edges of ∆, with finitely many orbits. Let Ai = Γi − {1} and γi = Ai, 1 ≤ i ≤ n. Each directed segment of length 2 in ∆ lies in the Γ-orbit of a directed segment of the form (P, Qi, aP ), a ∈ Ai, or (Qj, P, Qk), j 6= k, as illustrated below. P • yi Qi • ayi aP • Qj • yj P • yk Qk • Let Ai = {a = Γ(P, Qi, aP ) : a ∈ Ai}, and let B = {bjk = Γ(Qj, P, Qk) : j 6= k}. Then by Theorem 5.1 (5.4) K0(AΓ) =*[i bij (a ∈ Ai), a =Xj6=i a+ . bjk = Xa∈ Ak The relations on the right side of (5.4) show that bjk depends only on k. Therefore, for each i, all the generators in Ai are equal. It follows that Ai ∪ B (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (5.5) K0(AΓ) = hai ai =Xj6=i γjaji. It is easy to see from (5.5) that K0(AΓ) is a torsion group and therefore that K1(AΓ) = 0. In other words, the unitary group of AΓ is connected [C2]. If all the groups Γi have the same order, say γ + 1, and δ = γ(n − 1) − 1, then (5.5) simplifies to K0(AΓ) = Z(γ+1)δ ⊕ (Zγ+1)n−2 with canonical generators a1, a2 − a1, a3 − a1, . . . , an−1 − a1. References [AD] C. Anantharaman-Delaroche, Syst`emes dynamiques non commutatifs et moyennabilit´e, Math. Ann. 279 (1987), 297-315. [BP] A. Broise-Alamichel and F. Paulin, Sur le codage du flot g´eod´esique dans un arbre, Annales de la Facult´e des Sciences de Toulouse 16 (2007), 477 -- 527. [C1] J. Cuntz, Simple C ∗-algebras generated by isometries, Comm. Math. Phys. 57 (1977), 173 -- 185. [C2] J. Cuntz, K-theory for certain C ∗-algebras, Ann. of Math. 113 (1981), 181 -- 197. [C3] J. Cuntz, A class of C ∗-algebras and topological Markov chains: Reducible chains and the Ext-functor for C ∗-algebras, Invent. Math. 63 (1981), 23 -- 50. [CK] J. Cuntz and W. Krieger, A class of C ∗-algebras and topological Markov chains, Invent. Math. 56 (1980), 251-268. 12 GUYAN ROBERTSON [Ch] S-S. Chen, Limit sets of automorphism groups of a tree, Proc. Amer. Math. Soc. 83 (1981), 437 -- 441. [Ha] P. de la Harpe, Topics in Geometric Group Theory, University of Chicago Press, Chicago, 2000. [K] E. Kirchberg, Exact C ∗-algebras, tensor products, and the classification of purely infinite algebras, Proceedings of the International Congress of Mathemati- cians (Zurich, 1994), Vol. 2, 943 -- 954, Birkhauser, Basel, 1995. [R1] G. Robertson, Boundary actions for affine buildings and higher rank Cuntz- Krieger algebras. C ∗-algebras: Proceedings of the SFB Workshop on C ∗-algebras (Munster, March 8 -- 12, 1999), 182 -- 202, Springer-Verlag, 2000. [R2] G. Robertson, Boundary operator algebras for free uniform tree lattices, Hous- ton J. Math. 31 (2005), 913 -- 935. [RS] G. Robertson and T. Steger, Affine buildings, tiling systems and higher rank Cuntz-Krieger algebras, J. Reine Angew. Math. 513 (1999), 115 -- 144. [Sel] Z. Sela, Acylindrical accessibility for groups, Invent. Math. 129 (1997), 527 -- 565. [Ser] J-P. Serre, Trees, Springer-Verlag, Berlin, 1980. School of Mathematics and Statistics, University of Newcastle, NE1 7RU, U.K. E-mail address: [email protected]
1408.4242	2	1408	2015-07-22T14:06:46	C*-algebras associated to topological Ore semigroups	[ "math.OA" ]	Let $G$ be a locally compact group and $P \subset G$ be a closed Ore semigroup containing the identity element. Let $V: P \to B(\clh)$ be a representation such that for every $a \in P$, $V_{a}$ is an isometry and the final projections of $\{V_{a}: a \in P\}$ commute. In this article, we study the $C^{}$-algebra $\mathcal{W}_{V}(P,G)$, generated by $\{\int f(a)V_{a} da: f \in L^{1}(P)\}$. We show that there exists a universal $C^{}$-algebra, which admits a groupoid description, of which $\mathcal{W}_{V}(P,G)$ is a quotient. If $P=G$, then this universal algebra is just $C^{*}(G)$.	math.OA	math	C ∗-algebras associated to topological Ore semigroups 5 1 0 2 l u J 2 2 ] S. Sundar July 13, 2021 Abstract . A O h t a m [ 2 v 2 4 2 4 . 8 0 4 1 : v i X r a Let G be a locally compact group and P ⊂ G be a closed Ore semigroup containing the identity element. Let V : P → B(H) be an anti-homomorphism such that for every a ∈ P , Va is an isometry and the final projections of {Va : a ∈ P } commute. We study the C ∗-algebra generated by {R f (a)Vada : f ∈ L1(P )}. We show that there exists a groupoid C ∗-algebra which is universal for isometric representations with commuting range projections. AMS Classification No. : Primary 22A22; Secondary 54H20, 43A65, 46L55. Keywords. Wiener-Hopf algebras, Groupoids, Semigroups. 1 Introduction It is fair to say that C ∗-algebras of groups and their crossed products are the most studied C ∗-algebras in the theory of operator algebras. Several authors have tried to study C ∗-algebras associated to semigroups. For example, the Toeplitz algebra is the C ∗-algebra associated to the additive semigroup N. Recently, the theory of semigroup C ∗-algebras have received renewed attention. See for example [Cun08], [Li12], [Li13] and the references therein. The notion of crossed product by semigroups has also been studied by several authors most notably by Murphy in [Mur91], [Mur94], [Mur96b] and by Exel in [Exe03]. However much of the literature focusses on discrete semigroups. In the topological direction, upto the author's knowledege, the only example studied is the Wiener-Hopf C ∗-algebra. This was studied from the groupoid point of view first in [MR82] and then successively by Nica in [Nic87], [Nic90] and Hilgert and Neeb in [HN95]. Let G be a second countable locally compact group and P ⊂ G be a closed semigroup containing the identity element. We assume that Int(P ) is dense in P and P P −1 = G. 1 Let V : P → B(H) be an isometric representation on a Hilbert space H i.e. for a ∈ P , Va is an isometry and VaVb = Vba. For f ∈ L1(P ), let Wf :=Za∈P f (a)Vada. The semigroup C ∗-algebra or the Wiener-Hopf algebra, denoted WV (P, G), associated to the representation V is the C ∗-algebra generated by {Wf : f ∈ L1(P )}. If we consider the compression of the right regular representation of G on L2(G) onto L2(P ), then one obtains the usual Wiener-Hopf algebra studied in [MR82]. In general, it is much difficult to understand the structure of WV (P, G). However if we assume that the final projections a : a ∈ P } form a commuting family of projections then one can do better. {Ea := VaV ∗ Without this commutative assumption, the situation becomes much complicated even for the simplest case of P := N × N as is illustrated by Murphy in [Mur96a]. The results obtained and the organisation of the paper are described below. From now on, we assume that the range projections commute. For g = ab−1 ∈ G, let Wg := V ∗ It is shown in Section 3, that Wg is well-defined and {Eg : g ∈ G} forms a commuting family of projections. For f ∈ b Va and Eg be the final space of Wg. L1(G), let Wf :=R f (g)Wgdg. It is not difficult to show that WV (P, G) is generated by {R f (g)Egdg : f ∈ L1(G)}. The map C(Ω) ⋊ P ∋ (T, a) → V ∗ {Wf : f ∈ L1(G)}. Let Ω be the spectrum of the commutative C ∗-algebra generated by a T Va ∈ C(Ω) provides an action of P on Ω. In Section 4 and 5, we show that this action is injective. Let G := Ω ⋊ P := {(x, ab−1, y) ∈ Ω × G × Ω : xa = yb} be the Deaconu-Renault groupoid where the groupoid operations are given by (x, g, y)(y, h, z) = (x, gh, z) (x, g, y)−1 = (y, g−1, x). For f ∈ Cc(G), let ef ∈ Cc(G) be defined by ef (x, g, y) = f (g). We apply the results of [RS15] to show that Ω ⋊P has a Haar system. We also show that there exists a surjective representation λ : C ∗(G) → WV (P, G) such that for f ∈ Cc(G), λ(ef ) =Z f (g)∆(g)− 1 2 Wg−1dg. Here ∆ denotes the modular function of the group. This is achieved in Sections 4-6. For the Wiener-Hopf representation, the groupoid Ω ⋊ P is the groupoid considered in [MR82]. 2 We show in Section 7, that there exists a universal space Ωu on which P acts such that if V : P → B(H) is an isometric representation with commuting range projections then there exists a representation λ : C ∗(Ωu ⋊ P ) → B(H) such that for f ∈ Cc(G), λ(ef ) =Z f (g)∆(g)− 1 2 Wg−1dg. 2 Preliminaries For the convenience of the reader, we recall the essential facts from [RS15] that we need in this paper. The proofs can be found in [RS15]. Throughout this paper, G stands for a second countable, locally compact topological group and P ⊂ G for a closed subsemigroup containing the identity element e. We also assume the following. (C1) The group G = P P −1, and (C2) the interior of P in G, denoted Int(P ), is dense in P . Semigroups for which (C1) is satisfied are called Ore semigroups. consider only semigroups with identity for which (C1) and (C2) are satisfied. In this paper, we Let X be a compact Hausdorff space. A right action of P on X is a continuous map X × P ∋ (x, a) → xa ∈ X such that xe = x and (xa)b = x(ab) for x ∈ X and a, b ∈ P . Moreover we assume that the action is injective i.e. for every a ∈ P , the map X ∋ x → xa ∈ X is injective. Let X be a compact Hausdorff space on which P acts on the right injectively. Then the semi-direct product groupoid X ⋊ P is defined as follows: X ⋊ P := {(x, g, y) ∈ X × G × X : ∃ a, b ∈ P, such that g = ab−1, xa = yb}. The groupoid multiplication and the inversion are given by (x, g, y)(y, h, z) = (x, gh, z), (x, g, y)−1 = (y, g−1, x). The map X ⋊ P ∋ (x, g, y) → (x, g) ∈ X × G is injective. Thus X ⋊ P can be considered a subset of X × G which we do from now. Moreover X ⋊ P is a closed subset of X × G and when X ⋊ P is given the subspace topology, the groupoid X ⋊ P becomes a topological groupoid. We denote the range and source maps by r and s respectively. For x ∈ X, let Qx := {g ∈ G : (x, g) ∈ G}. Then r−1(x) = {x} × Qx. Note that for x ∈ X, Qx.P ⊂ Qx and Qx is closed. By Lemma 4.1 of [RS15], for x ∈ X, Int(Qx) is dense in Qx and the boundary of Qx has measure zero. 3 For x ∈ X, let λx be the measure on G defined as follows: For f ∈ Cc(G), Z f dλx =Z f (x, g)1Qx(g)dg. Here dg denotes the left Haar measure on G. In [RS15], it is shown that the groupoid G := X ⋊ P admits a Haar system if and only if the map X × Int(P ) ∋ (x, a) → xa ∈ X is open. In this case, the measures (λx)x∈X form a Haar system. We will use only this Haar system if X ⋊ P admits one. Suppose that G := X ⋊ P admits a Haar system. Then the action of P on X can be dilated to an action of G. That is there exists a locally compact Hausdorff space Y on which G acts on the right and a continuous P -equivariant injection i : X → Y such that 1. the set X0 := i(X)Int(P ) is open in Y , and 2. Y =Ta∈P i(X)a−1 =Ta∈Int(P ) X0a−1. Moreover the space Y is unique up to a G-equivariant homeomorphism. We will identiy X as a subspace of Y via the injection i and will suppress the notation i. Also the groupoid G is isomorphic to the reduction (Y ⋊ G)X. With this notation, note that for x ∈ X, 1Qx(g) = 1X(xg). Also we leave it to the reader to check that 1Int(Qx)(g) = 1X0(xg). For f ∈ Cc(G), let ef ∈ Cc(G) be defined by ef (x, g) = f (g). We also need the following proposition. The proof is a line by line imitation of that of Proposition 3.5 of [MR82]. Hence we omit the proof. See also [RS15] for some remarks concerning the proof. Let G := X ⋊ P and assume that it has a Haar system. Suppose that family { f : f ∈ Cc(G)} separates points of X. Then the ∗-algebra generated Proposition 2.1 For f ∈ Cc(G), let f ∈ C(X) be defined by f (x) =R f (g)1X(xg)dg. by {ef : f ∈ Cc(G)} is dense in Cc(G) where Cc(G) is given the inductive limit topology. As a consequence, C ∗(G) is generated by {ef : f ∈ Cc(G)}. 3 Isometric representations with commuting range projections Definition 3.1 A map V : P → B(H) is called an isometric representation of P on the Hilbert space H if (1) the maps P ∋ a → Va and P ∋ a → V ∗ a are strongly continuous, 4 (2) for a ∈ P , Va is an isometry, and (3) for a, b ∈ P , VaVb = Vba. For a ∈ P , let Ea := VaV ∗ say that V has commuting range projections. a . If {Ea : a ∈ P } is a commuting family of projections, we In the next example, we recall the Wiener-Hopf representation or the regular repre- sentation. The C ∗-algebra associated to the Wiener-Hopf representation has been studied by several authors. See the papers [MR82], [HN95] and the references therein. Example 3.2 Consider the Hilbert space L2(G) and consider L2(P ) as a closed subspace of L2(G). For ξ ∈ L2(P ) and a ∈ P , let Va(ξ) be defined as follows: Va(ξ)(x) := ( ξ(xa−1)∆(a) 0 −1 2 if x ∈ P a if x /∈ P a. Here ∆ denotes the modular function of the group G. Then the map a ∈ P → Va ∈ B(L2(P )) is an isometric representation with commuting range projections. Note that for a ∈ P , the range of Va is L2(P a). Till the end of Section 7, we fix an isometric representation V : P → B(H) with g . First commuting range projections. For g = ab−1, let Wg := V ∗ we show that that Wg is well defined and is a partial isometry. b Va and let Eg := WgW ∗ Proposition 3.3 Let V : P → B(H) be an isometric representation with commuting range projections. (1) For g ∈ G, Wg is well defined and is a partial isometry. (2) The family {Eg : g ∈ G} forms a commuting family of projections. (3) If g1g−1 2 ∈ P , then Eg1 ≤ Eg2. (4) The map G ∈ g → Wg ∈ B(H) is strongly continuous. (5) For g, h ∈ G, WgWh = EgWhg. Proof. Suppose g = a1b−1 α1, α2 ∈ P such that a−1 1 = a2b−1 1 a2 = α1α−1 2 . Then a−1 2 = b−1 1 a2 = b−1 1 b2. Since P P −1 = G, there exists 1 b2. Then a1α1 = a2α2 and b1α1 = b2α2. 5 Now observe that b1Va1 = V ∗ V ∗ = V ∗ α1Vα1Va1 b1V ∗ b1α1Va1α1 b2α2Va2α2 b2V ∗ b2Va2. = V ∗ = V ∗ = V ∗ α2Vα2Va2 This proves that Wg is well defined. Let Eg := WgW ∗ which is self adjoint. Now note that g . If g = ab−1, then Eg = V ∗ b EaVb E2 g = V ∗ = V ∗ b EaVb b EaVbV ∗ b EaEbEaVb b EbE2 b EaVb aVb = V ∗ = V ∗ ( Since Ea and Eb commute) = Eg. Thus Eg is a projection. This proves (1). Let g1, g2 ∈ G be given. Write g1 = a1b−1 α1, α2 ∈ P such that b1α1 = b2α2. Let a′ gi = a′ i)−1 for i = 1, 2. But now b′ 1 = b′ i(b′ 2. Thus 1 and g2 = a2b−1 i = aiαi and b′ 2 with ai, bi ∈ P . Choose i = biαi for i = 1, 2. Then Eg1Eg2 = V ∗ ′ b 1 Ea ′ 1 Vb ′ 1 V ∗ ′ b 1 Ea ′ 2 Vb ′ 1 = V ∗ ′ b 1 Ea ′ 2 Vb ′ 1 V ∗ ′ b 1 Ea ′ 1 Vb ′ 1 = Eg2Eg1. This proves (2). Suppose g1g−1 2 = a for some a ∈ P . Write g2 = bc−1. Then g1 = (ab)c−1. Then Eg1 = V ∗ = V ∗ ≤ V ∗ c VabV ∗ abVc c Vb(VaV ∗ c VbV ∗ b Vc a )V ∗ b Vc ≤ Eg2. This proves (3). Note that the map Int(P )×Int(P ) ∋ (a, b) → ab−1 ∈ G is surjective and open. Thus G is the quotient of Int(P ) × Int(P ). Since multiplication is strongly continuous on the 6 unit ball of B(H), it follows that the map Int(P ) × Int(P ) ∋ (a, b) → V ∗ b Va ∈ B(H) is strongly continuous. As a consequence, it follows that the map G ∋ g → Wg ∈ B(H) is strongly continuous. This proves (4). Let g, h ∈ G be given. Write g = ab−1 and h = cd−1 with a, b ∈ P . Choose α, β ∈ P such that dβ = aα. Now note that g = (aα)(bα)−1 and h = (cβ)(dβ)−1. Thus we can write g and h as g = a1b−1 1 . Now calculate as follows 1 and h = c1a−1 WgWh = V ∗ b1Va1V ∗ a1Vc1 b1Ea1Vb1V ∗ = V ∗ b1Vc1( Since Ea1 and Eb1 commute) = EgWhg This completes the proof. For f ∈ L1(G), the "Wiener-Hopf" operator with symbol f is defined as ✷ Wf :=Z f (g)Wgdg. We want to describe the C ∗-algebra, denoted WV (P, G), generated by {Wf : f ∈ L1(G)}. We suppress the subscript V and simply denote WV (P, G) by W(P, G) ( atleast till the end of Section 7.) Remark 3.4 One can show that W(P, G) is generated by {R f (a)Vada : f ∈ L1(P )}. The proof is similar to that of Proposition 2.2 of [RS15]. Hence we omit the proof. First we consider a related commutative C ∗-algebra. Note that by definition, for g is strongly continuous. g . Moreover the map G ∋ g → Eg = WgW ∗ g ∈ G, Wg−1 = W ∗ For f ∈ L1(G), let Let Ef :=Z f (g)Egdg. A := C ∗{Ef : f ∈ L1(G)}. Since {Eg : g ∈ G} forms a commuting family of projections, it follows that A is a commutative C ∗-subalgebra of B(H). Note that Eg = 1 if g ∈ P −1. If f ∈ L1(P −1), then Ef =R f (g)dg. Thus, it follows that A is a commutative unital C ∗ subalgebra of B(H). Denote the spectrum of A by Ω. Let Gn := G × G × · · · × G . For f ∈ Cc(Gn), let n times } {z Ef :=Z f (g1, g2, · · · , gn)Eg1Eg2 · · · Egndg1dg2 · · · dgn. 7 Let eA := ∞[n=1 {Ef : f ∈ Cc(Gn)}. Then eA forms a dense unital ∗-subalgebra of A. Also note that for every n, the map Cc(Gn) ∋ f → Ef ∈ A is continuous when Cc(Gn) is given the inductive limit topology and A is given the norm topology. a T Va. Clearly αe = id and αaαb = αab. For T ∈ B(H) and a ∈ P , let αa(T ) = V ∗ Observe that αa(V ∗ a V ∗ g ∈ G. Since the final projection VaV ∗ that αa(Eg1Eg2) = αa(Eg1)αa(Eg2). b EcVb) = V ∗ b EcVbVa = V ∗ abEcVab. Thus αa(Eg) = Ega−1 for a commutes with Eg for every g ∈ G, it follows Proposition 3.5 For a ∈ P , αa leaves A invariant and the map αa : A → A is a unital ∗-homomorphism. Moreover for T ∈ A, the map P ∋ a → αa(T ) ∈ A is norm continuous. Proof. For a ∈ P and f ∈ Cc(Gn), let efa ∈ Cc(Gn) be defined by efa(g1, g2, · · · , gn) = ∆(a)nf (g1a, g2a, · · · , gna). Then for f ∈ Cc(Gn), the map P ∋ a → efa ∈ Cc(Gn) is continuous if Cc(Gn) is given the inductive limit topology. Let a ∈ P and f ∈ Cc(Gn) be given. Then αa(Ef ) =Z f (g1, g2, · · · , gn)αa(Eg1Eg2 · · · Egn)dg1dg2 · · · dgn =Z f (g1, g2, · · · , gn)Eg1a−1Eg2a−1 . . . Egna−1dg1dg2 · · · dgn =Z f (g1a, g2a, · · · , gna)∆(a)nEg1Eg2 · · · Egndg1dg2 · · · dgn = E efa leaves A invariant. Observe that if Ea = VaV ∗ Thus αa leaves eA invariant. Since eA is dense in A and αa is bounded, it follows that αa a commutes with T, S ∈ B(H) then αa(T S) = αa(T )αa(S). By Proposition 3.3, if follows that Ea commutes with Ef for f ∈ Cc(Gn). Thus Ea commutes with every element of A. Hence αa : A → A is multiplicative. Clearly αa is unital and ∗-preserving. For a ∈ P , αa : A → A is contractive. Thus it is enough to show that for T ∈ eA, the map P ∋ a → αa(T ) ∈ A is continuous. Let T = Ef for some f ∈ Cc(Gn). Then αa(T ) = E efa is continuous as it is the composite . Hence the map P ∋ a → αa(T ) = E efa 8 of the continuous maps P ∋ a → efa ∈ Cc(Gn) and Cc(Gn) ∋ h → Eh where Cc(Gn) is given the inductive limit topology. This completes the proof. Since A = C(Ω), it follows that for every a ∈ P , there exists φa : Ω → Ω such that F ◦ φa = αa(F ) for F ∈ C(Ω). The condition αaαb = αab translates to φaφb = φba for a, b ∈ P . Also φe = id. Thus the map Ω × P ∋ (x, a) → φa(x) ∈ Ω defines a right action of P on Ω. We henceforth write φa(x) as xa for x ∈ Ω and a ∈ P . ✷ We claim that the map Ω×P ∋ (x, a) → xa ∈ Ω is continuous. Suppose (xn) → x and (an) → a. Let F ∈ C(Ω). By Proposition 3.5, it follows that αan(F ) converges uniformly to αa(F ). Since the convergence is uniform, it follows that αan(F )(xn) converges to αa(F )(x). In other words, for every F ∈ C(Ω), F (xnan) converges to F (xa). Hence xnan converges to xa. The goal of this paper is to prove the following statements. (1) The right action of P on Ω is injective. (2) The semidirect product groupoid G := Ω ⋊ P has a Haar system. (3) For f ∈ Cc(G), let ef ∈ Cc(G) be defined by ef (x, g) = f (g) for (x, g) ∈ G. There exists a surjective ∗-homomorphism π : C ∗(G) → W(P, G) such that π(ef ) = R ∆(g)− 1 2 f (g)Wg−1dg for f ∈ Cc(G). To prove the above statements, we need a better description of Ω which forms the content of the next section. We end this section with a lemma which is useful in showing that Ω ⋊ P has a Haar system. f (a)VaT V ∗ a da ∈ A. Lemma 3.6 Let f ∈ Cc(G) be such that supp(f ) ⊂ Int(P ). Then for T ∈ A, the integralZa∈P Proof. It is enough to prove the statement for T ∈ eA. Let T = Eφ for some φ ∈ Cc(Gn). a = EgaEa. Now calculate as follows a EgaVa = Eg. Hence VaEgV ∗ For a ∈ P and g ∈ G, V ∗ 9 to find that Za∈P f (a)VaEφV ∗ a da =Za∈P =Za∈P =Za∈P = Eψ ∈ eA f (a)φ(g1, g2, · · · , gn)VaEg1Eg2 · · · EgnV ∗ a da dg1 dg2 · · · dgn f (a)φ(g1, g2, · · · , gn)EaEg1aEg2a · · · Egna da dg1dg2 · · · dgn ∆(a)−nf (a)φ(g1a−1, g2a−1, · · · , gna−1)EaEg1Eg2 · · · Egn da dg1 dg2 · · · dgn where ψ ∈ Cc(Gn+1) is given by ψ(g, g1, g2, · · · , gn) = ∆(g)−nf (g)φ(g1g−1, g2g−1, · · · , gng−1). This completes the proof. ✷ 4 What is Ω ? We first discuss the case when G is discrete. The discrete semigroup C ∗-algebras are analysed in great detail in the papers [Li12] and [Li13]. Neverthless we discuss this case in the form that we need. This also motivates the topological case. Let G be a discrete group and P ⊂ G be a semigroup such that e ∈ P and P P −1 = G. In this case, the Wiener-Hopf C ∗-algebra WV (P, G) is simply the C ∗-algebra generated by {Wg : g ∈ G} and the commutative C ∗-algebra A is the C ∗-algebra generated by {Eg : g ∈ G}. Let χ be a character of A. Let us define the support of χ, denoted Aχ, as Aχ := {g ∈ G : χ(Eg) = 1}. Condition (3) of Proposition 3.3 implies that P −1Aχ ⊂ Aχ. Since Eg = 1 if g ∈ P −1, it follows that P −1 ⊂ Aχ. Let P(G) be the power set of G. Identify P(G) with {0, 1}G, via the map P(G) ∋ A → 1A ∈ {0, 1}G, and endow it with the product topology. The group G acts on P(G). The right action is given by : For g ∈ G and A ∈ P(G), Ag := {ag : a ∈ A}. Clearly the map Ω ∋ χ → Aχ ∈ {0, 1}G is continuous, injective and hence an embedding. We leave it to the reader to check that the above map is P -equivariant. From now, we view Ω as a subset of {0, 1}G. 10 Proposition 4.1 We have the following. (1) For A ∈ Ω and a ∈ P , Aa−1 ∈ Ω if and only if a ∈ A. (2) For A ∈ Ω and g ∈ G, Ag ∈ Ω if and only if g−1 ∈ A. (3) The action Ω × P → Ω is open. Proof. Let A ∈ Ω and a ∈ P be given. Suppose B := Aa−1 ∈ Ω. Since e ∈ B, it follows that a ∈ A. Now suppose a ∈ A. Let χ be the character corresponding to A. Since V ∗ a = EaEga. Thus the homomorphism B(H) ∋ T → VaT V ∗ a EgaVa = Eg, it follows that VaEgV ∗ a ∈ B(H) leaves A invariant. a ). Since χ(Ea) = 1, it follows Leteχ be the character on A defined byeχ(T ) = χ(VaT V ∗ that eχ is non-zero. Observe that for T ∈ A, (eχa)(T ) =eχ(V ∗ a T VaV ∗ a ) = χ(Ea)χ(T )χ(Ea) a T Va) = χ(VaV ∗ = 1A(a)χ(T )1A(a) = χ(T ). (1). Thus eχa = χ. Let B be the support of eχ. Then A = Ba. Thus Aa−1 ∈ Ω. This proves Now let A ∈ Ω and g = ab−1 ∈ G. Suppose g−1 = ba−1 ∈ A. Then b ∈ Aa ∈ Ω. By (1), it follows that Ag = Aab−1 ∈ Ω. Now suppose Ag ∈ Ω. Then A = (Ag)g−1. Since e ∈ Ag, it follows that g−1 ∈ A. This proves (2). When G is discrete, Int(P ) = P and ΩP = Ω. Thus, by Theorem 4.3 of [RS15], to prove that the action Ω × P → Ω is open, it is enough to show that Ωa is open in Ω for every a ∈ P . But note that by (1), for a ∈ P , Ωa = {A ∈ Ω : 1A(a) = 1} which is clearly open in Ω, as Ω has the subspace topology of {0, 1}G. This completes the proof. ✷ A consequence of Proposition 4.1 is that the semi-direct product groupoid Ω ⋊ P has a Haar system. For g ∈ G, let δg ∈ Cc(Ω × P ) be defined by δg(x, h) = 1 if h = g and δg(x, h) = 0 if h 6= g. Then it is not difficult to show that there exists a representation π : Cc(Ω × P ) → B(H) such that π(δg) = Wg−1 for every g ∈ G. We will prove this in the topological case. Now let us turn our attention to the topological case. Let χ be a character of the commutative C ∗-algebra A. The support of χ, denoted Aχ, is defined as follows: For 11 g ∈ G, g /∈ Aχ if and only if there exists an open set U of G containing g such that χ(R f (g)Egdg) = 0 for every f ∈ Cc(U). Here Cc(U) := {f ∈ Cc(G) : supp(f ) ⊂ U}. Note that Aχ is closed. Remark 4.2 Let χ be a character of A and A be its support. Then for g ∈ G, g ∈ A if and only if for every open set U containing g, there exists f ∈ Cc(U) such that f ≥ 0 and χ(R f (g)Egdg) > 0. Proposition 4.3 Let χ be a character of A and let A be its support. Then (1) P −1 ⊂ A and P −1A ⊂ A, (2) the interior Int(A) is dense in A, and (3) the boundary ∂(A) has measure zero. Proof. Let a ∈ Int(P ) and U be an open set containing a−1. Then U ∩Int(P )−1 is a non- empty open set containing a−1. Choose f ∈ Cc(G) such that supp(f ) ⊂ U ∩ Int(P )−1, f ≥ 0 and R f (g)dg = 1. Since Eg = 1 for g ∈ P −1, it follows that R f (g)Egdg = R f (g)dg = 1. Thus χ(R f (g)Egdg) = 1. This proves that a−1 ∈ A. As a consequence, Int(P )−1 ⊂ A. But Int(P )−1 is dense in P −1 and A is closed. Hence P −1 ⊂ A. For f ∈ Cc(G) and g ∈ G, let Lg(f ) ∈ Cc(G) be defined by Lg(f )(x) = f (g−1x). Let g ∈ A be given and a ∈ P . Let U be an open set containing a−1g. Then aU is open and contains g. Thus there exists f ∈ Cc(aU) such that f ≥ 0 and χ(R f (g)Egdg) > 0. Let ef = La−1f . Then ef ≥ 0 and supp(ef ) ⊂ U. Now Z ef (g)Egdg =Z f (ag)Egdg =Z f (g)Ea−1gdg ≥Z f (g)Egdg (By Proposition 3.3). Hence χ(R ef (g)Egdg) > 0. This implies that a−1g ∈ A. Thus P −1A ⊂ A. This proves (1). Statements (2) and (3) follow immediately from Lemma 4.1 of [RS15]. This completes the proof. ✷ Before proceeding further, let us review the Vietoris topology. Let X be a locally compact second countable Hausdorff space and let d be a metric on X inducing the topology. Let C(X) be the collection of closed subsets of X. Then C(X), endowed with 12 the Vietoris topology, is compact and metrisable. We recall here the convergence of sequences of elements in C(X). Let (An) be a sequence of closed subsets of X. Define lim inf An = {x ∈ X : lim sup d(x, An) = 0}, and lim sup An = {x ∈ X : lim inf d(x, An) = 0}. Then (An) converges in C(X) if and only if lim inf An = lim sup An. If lim inf An = lim sup An = A, then An converges to A. Observe that if U ⊂ X is closed then the subset {A ∈ C(X) : A ∩ U 6= ∅} is open in C(X). Consider C(G), the space of closed subsets of G, with the Vietoris topology. The group G acts on C(G) on the right. For A ∈ C(G) and g ∈ G, define Ag = {ag : a ∈ A}. Let Ωu := {A ∈ C(G) : P −1 ⊂ A and P −1A ⊂ A}. We leave it to the reader to verify that Ωu is a closed, and hence a compact, subset of C(G). Clearly Ωu is P -invariant. The space Ωu is first considered in [HN95]. Proposition 4.4 The action Ωu × Int(P ) → Ωu is open. Proof. Let a ∈ Int(P ). It is enough to show that ΩuInt(P )a is open in Ωu (See Theorem 4.3, [RS15]). We claim that ΩuInt(P )a = {A ∈ Ωu : A ∩ Int(P )a 6= ∅} which will imply that ΩuInt(P )a is open. Let A ∈ ΩuInt(P )a. Then A = Bba for some B ∈ Ωu and b ∈ Int(P ). Since e ∈ B, it follows that ba ∈ A. Hence A ∩ Int(P )a is non-empty. Now suppose A ∈ Ωu and A ∩ Int(P )a is non-empty. Choose b ∈ Int(P ) such that ba ∈ A. Since P −1A ⊂ A, it follows that P −1ba ⊂ A, equivalently P −1 ⊂ Aa−1b−1, and P −1Aa−1b−1 ⊂ Aa−1b−1. This proves that B = Aa−1b−1 ∈ Ωu. Then A = Bba ∈ ΩuInt(P )a. This completes the proof. ✷ We summarise a few facts regarding the space Ωu in the following remark. Remark 4.5 Note the following. (1) ΩuInt(P )a = {A ∈ Ωu : a ∈ Int(A)}. If A ∩ Int(P )a 6= ∅ then a ∈ Int(P )−1A which is open and contained in A. Thus a ∈ Int(A). Now suppose a ∈ Int(A) then Int(A) ∩ P a is non-empty. Since Int(P )a is dense in P a, it follows that Int(A) ∩ Int(P )a is non-empty and hence A ∩ Int(P )a is non-empty. 13 (2) If A ∈ Ωu then Int(A) is dense in A and the boundary ∂(A) has measure zero. This follows from Lemma 4.1 of [RS15] (3) Let A ∈ Ωu and g ∈ G. Then (A, g) ∈ Ωu ⋊ P if and only if Ag ∈ Ωu if and only if g−1 ∈ A. We leave this verification to the reader. (4) The map Ωu ∋ A → 1A ∈ L∞(G) is continuous and injective and hence an em- bedding. Here L∞(G) is given the weak ∗-topology. Let Gu := Ωu ⋊ P . Then Proposition 4.4 implies that Gu has a Haar system. Moreover a Haar system on Gu is given by (1QA(g)dg)A∈Ωu. For A ∈ Ωu, observe that QA := {g ∈ G : (A, g) ∈ Gu} is A−1. By the definition of a Haar system, it follows that for f ∈ Cc(Gu), Ωu ∋ A → R f (x, g)1A(g−1)dg is continuous. In particular, for f ∈ Cc(G), the function Ωu ∋ A →R f (g)1A(g−1)dg is continuous. As a consequence, the map Ωu ∋ A → 1A ∈ L∞(G) is continuous. Suppose A, B ∈ Ωu such that 1A = 1B in L∞(G). Then A\B and B\A has measure zero. If A\B is non-emtpy then Int(A)\B is non-empty since Int(A) is dense in A. But Int(A)\B is open and hence cannot have measure zero. Thus A\B = ∅. Similarly B\A = ∅. Hence A = B. This proves the map Ωu ∋ A → 1A ∈ L∞(G) is injective. Thus we can consider Ωu as a compact subset of L∞(G). Let V : P → B(H) be an isometric representation with commuting range projections. Proposition 4.6 Let χ be a character of A and let A be its support. Let f ∈ Cc(G). Then Denote the commutative C ∗-algebra generated by {R f (g)Egdg : f ∈ L1(G)} by A and let Ω be the spectrum of A. For f ∈ L1(G), let Ef :=R f (g)Egdg. (1) χ(cid:16)R f (g)1Ac(g)Egdg(cid:17) = 0. (2) if supp(f ) ⊂ Int(A) then χ(cid:16)R f (g)Egdg(cid:17) =R f (g)dg, and (3) we have the equality χ(cid:16)R f (g)Egdg(cid:17) =R f (g)1A(g)dg. Proof. First observe that if supp(f ) ⊂ Ac, where Ac denotes the complement of A, then χ(Ef ) = 0. This follows from the definition of A and by a partition of unity argument. Now write Ac = Kn with Kn compact and Kn increasing. This is possible as Ac is open. Choose φn ∈ Cc(G) such that 0 ≤ φn ≤ 1, φn = 1 on Kn and supp(φn) ⊂ Ac. ∞[n=1 14 Note that φn → 1Ac pointwise. Hence f φn converges to f 1Ac in L1(G). This implies that Ef φn converges to Ef 1Ac . Since χ(Ef φn) = 0, it follows that χ(Ef 1Ac ) = 0. This proves (1). Let f ∈ Cc(G) be such that supp(f ) ⊂ Int(A). Let g ∈ supp(f ). Then Int(A) ∩ P g is non-empty. Since Int(P ) is dense in P , it follows that Int(A) ∩ Int(P )g is non-empty. Let s ∈ Int(P ) be such that sg ∈ Int(A). Then (sg)g−1 ∈ Int(P ). Since Int(P ) is open, we can choose open sets U and V contained in Int(A), with compact closures, such that (g, sg) ∈ U × V ⊂ Int(A) × Int(A) and V U −1 ⊂ Int(P ). Then by Proposition 3.3, it follows that for g1 ∈ V and g2 ∈ U, Eg1Eg2 = Eg1. sets (Ui)n i=1 and (Vi)n i=1 Ui and ViU −1 Since supp(f ) is compact, it follows that there exists finitely many non-empty open i=1 with compact closures, contained in Int(A), such that supp(f ) ⊂ i ⊂ Int(P ). A partition of unity argument allows us to write f as f = Sn i=1 fi with supp(fi) ⊂ Ui. Thus to prove (2), it is enough to show χ(Efi) =R fi(g)dg. Pn Since Vi is a non-empty open set contained in A, by Remark 4.2, it follows that there exists φi ∈ Cc(G) such that supp(φi) ⊂ Vi and χ(Eφi) 6= 0. Observe the following φi(g1)fi(g2)Eg1dg1dg2 φi(g1)fi(g2)Eg1Eg2dg1dg2 EφiEfi =ZVi×Ui =ZVi×Ui =(cid:16)Z fi(g2)dg2(cid:17)Z φi(g1)Eg1dg1 =(cid:16)Z fi(g2)dg2(cid:17)Eφi. it follows that χ(Eφi)χ(Efi) = (cid:16)R fi(g)dg(cid:17)χ(Eφi). Now χ(Eφi) 6= 0. Hence χ(Efi) =R fi(g)dg. This proves (2). χ(Ef 1Int(A)). Write Int(A) = Sn Kn with Kn compact and Kn increasing. Choose Now let f ∈ Cc(G) be given. By (1), it follows that χ(Ef ) = χ(Ef 1A). But since the boundary of A has measure zero, it follows that 1Int(A) = 1A a.e. Thus χ(Ef ) = φn ∈ Cc(G) such that φn = 1 on Kn and supp(φn) ⊂ Int(A). Then φn → 1Int(A) pointwise and hence f φn converges to f 1IntA in L1(G). Note that supp(f φn) ⊂ Int(A). Since χ is multiplicative, 15 Now calculate, as follows, to find that χ(Ef ) = χ(Ef 1Int(A) ) χ(Ef φn) = lim n = lim n Z f (g)φn(g)dg (by (2)) =Z f (g)1Int(A)(g)dg =Z f (g)1A(g)dg ( Since 1A = 1Int(A) in L∞(G)). This proves (3). This completes the proof. ✷. Proposition 4.7 For χ ∈ Ω, let Aχ be its support. Then the map Ω ∋ χ → Aχ ∈ Ωu is one-one, continuous and P -equivariant. Consequently, the action of P on Ω is injective. Proof. By Proposition 4.3, it follows that Aχ ∈ Ωu if χ ∈ Ω. For f ∈ Cc(G), by Proposition 4.6, χ(Ef ) =R f (g)1Aχ(g)dg. Hence the map Ω ∋ χ → Aχ ∈ L∞(G) is one- one and continuous where L∞(G) is given the weak ∗-topology. By part (4) of Remark 4.5, it follows that Ω ∋ χ → Aχ ∈ Ωu is one-one and continuous. Let f ∈ Cc(G), a ∈ P and χ ∈ Ω and A be the support of χ. Observe that (χ.a)(Z f (g)Egdg) = χ(Z f (g)V ∗ a EgVa)dg = χ(Z f (g)Ega−1dg) = χ(Z f (ga)∆(a)Egdg) =Z f (ga)∆(a)1A(g)dg =Z f (g)1A(ga−1)dg =Z f (g)1Aa(g)dg. Hence the support of χ.a is Aa. Thus the map Ω ∋ χ → Aχ ∈ Ωu is a continuous P -equivariant embedding. This completes the proof. ✷. Thus we can and will consider Ω as a subset of Ωu with the subspace topology. 16 5 Haar system on Ω ⋊ P In this section, we show that the semi-direct product Ω ⋊ P admits a Haar system. We prove that the action Ω × Int(P ) → Ω is open. To prove this, we need an analogue of Proposition 4.1 in the topological setting. Proposition 5.1 Let a ∈ Int(P ) and A ∈ Ω. Then a ∈ A if and only if Aa−1 ∈ Ω. Proof. Let a ∈ Int(P ) and A ∈ Ω be given. Suppose B := Aa−1 ∈ Ω. Since e ∈ B, it follows that a ∈ A. Now suppose a ∈ A. In addition, assume that a ∈ Int(A). Let χ be the character defining A. Then for f ∈ Cc(G), χ(Z f (g)Egdg) =Z f (g)1A(g)dg. Choose a decreasing sequence of open sets (Un) in G such that (1) the intersectionT∞ n=1 Un = {a}, (2) if U is open in G and a ∈ U then there exists N such that Un ⊂ U for n ≥ N, and (3) for every n, Un ⊂ Int(P ) ∩ Int(A). This is possible, for we can choose a metric and let (Un) be the open balls containing a with diam(Un) → 0. For every n ∈ N, choose fn ∈ Cc(G) such that fn ≥ 0,R fn(g)dg = 1 and supp(fn) ⊂ Un. Note that χ(Efn) =R fn(g)1A(g)dg = 1 since supp(fn) ⊂ Int(A). Let φn be the linear functional on the commutative C ∗-algebra A defined by φn(T ) = χ(cid:0)Zb∈P fn(b)VbT V ∗ b db(cid:1). Note that φn is well defined by Lemma 3.6 and is clearly positive. Note that fn(b)Ebdb(cid:17) fn(b)1A(b)db φn(1) = χ(cid:16)Zb∈supp(fn) =Zb∈supp(fn) =Zb∈supp(fn) fn(b)db ( Since supp(fn) ⊂ Int(A)) = 1. Thus φn is a state for every n. But the set of states on a unital C ∗-algebra is weak∗- compact. By choosing a subsequence if necessary we can assume without loss of generality that (φn) converges in the weak∗-topology and let φ be its limit. 17 Recall that for a ∈ P , αa : A → A is given by αa(T ) = V ∗ a T Va. By Proposition 4.7, it follows that for every a ∈ A, αa is surjective. Claim: φ ◦ αa = χ. It is enough to show that (φ ◦ αa)(T ) = χ(T ) for T ∈ eA. Let T = Eψ for some φ ∈ Cc(Gm). Observe that for n ∈ N, φn(αa(Eψ)) = χ(Zb∈P = χ(Zb∈P = χ(Zb∈P = χ(Zb∈P fn(b)Vbαa(Eψ)V ∗ b db) fn(b)ψ(g1, g2, · · · , gm)VbV ∗ a Eg1 · · · EgmVaV ∗ b dbdg1 · · · dgm) fn(b)ψ(g1, g2, · · · , gm)EbEg1a−1b · · · Egma−1bdbdg1 · · · dgm) ∆(b−1a)mfn(b)ψ(g1b−1a, g2b−1a, · · · , gmb−1a)EbEg1Eg2 · · · Egmdbdg1dg2 · · · dgm). Let ǫ > 0 be given. Since ψ is continuous and compactly supported, it follows that there exists an open set U such that a ∈ U and for b ∈ U and g1, g2, · · · , gm ∈ G, Z ∆(b−1a)mψ(g1b−1a, g2b−1a, · · · , gmb−1a) − ψ(g1, g2, · · · , gm)dg1dg2 · · · dgm ≤ ǫ. Choose N ≥ 1 such that for n ≥ N, Un ⊂ U. Then for n ≥ N, supp(fn) ⊂ U. Note that for n ≥ N, φn(αa(Eψ)) − χ(Eψ) = φn(αa(Eψ)) − χ(EfnEψ) fn(b)(cid:16)Z ∆(b−1a)mψ(g1b−1a, · · · , gmb−1a) − ψ(g1, g2, · · · , gm)dg1 · · · dgm(cid:17)db ≤Zb∈Un ≤ ǫZb∈Un ≤ ǫ. fn(b) Thus it follows that φn(αa(Eψ)) → χ(Eψ) and hence φ ◦ αa = χ. This proves the claim. Since αa is surjective on A, it follows that φ is a character of A. Let B ∈ Ω be the support of φ. Then φ ◦ αa = χ translates to the equation Ba = A. Thus Aa−1 ∈ Ω. Now suppose a ∈ A. Let (sn) be a sequence in Int(P ) converging to the identity element e. Then s−1 n a → a and a ∈ Int(P ). But Int(P )−1A ⊂ Int(A). n a ∈ Int(P ) eventually, for s−1 18 Hence s−1 n a ∈ Int(A). By what we have proved, it follows that Aa−1sn ∈ Ω eventually. However Ω is a compact subset of Ωu and (Aa−1sn) converges to Aa−1. From this we conclude that Aa−1 ∈ Ω. This completes the proof. ✷ Just like in the discrete case, we have the following theorem. Proposition 5.2 Let A ∈ Ω and g ∈ G. Then Ag ∈ Ω if and only if g−1 ∈ A. Also the semi-direct product groupoid Ω ⋊ P has a Haar system. Proof. Let g ∈ G and A ∈ Ω be given. Suppose Ag ∈ Ω. Since e ∈ Ag, it follows that g−1 ∈ A. Now suppose g−1 ∈ A. As G = (Int(P ))(Int(P ))−1, write g = ab−1 with a, b ∈ Int(P ). Then ba−1 ∈ A or b ∈ Aa ∈ Ω. By Proposition 5.1, it follows that Aab−1 ∈ Ω. Hence Ag ∈ Ω. To prove that Ω⋊P has a Haar system, it is enough to show that the action Ω×P → Ω is open. By Theorem 4.3 of [RS15], it is enough to show that ΩInt(P )a is open in Ω for every a ∈ P . Let a ∈ P be given. Claim: ΩInt(P )a = {A ∈ Ω : A ∩ Int(P )a 6= ∅}. Suppose A ∩ Int(P )a is non-empty. Then there exists s ∈ Int(P ) such that sa ∈ A. By Proposition 5.1, B := Aa−1s−1 ∈ Ω. Thus A = Bsa ∈ ΩInt(P )a. Suppose A ∈ ΩInt(P )a. Then A = Bsa for some B ∈ Ω and s ∈ Int(P ). Since e ∈ B, it follows that sa ∈ A. Thus A ∩ Int(P )a is non empty. This proves the claim. The set {A ∈ C(G) : A ∩ Int(P )a 6= ∅} is open in C(G), when C(G) is given the Vietoris topology. This implies that ΩInt(P )a is open in Ω. This completes the proof. ✷ Remark 5.3 Consider the groupoid Ωu ⋊ P . Then by Proposition 5.2 and Statement (3) of Remark 4.5, it follows that Ω is an invariant subset of Ωu. Moreover the groupoid Ω ⋊ P is just the restriction Ωu ⋊ P Ω. We end this section by describing Ω in the case of the Wiener-Hopf representation. Recall that the Wiener-Hopf represention V : P → B(L2(P )) is given by the formula: For a ∈ P and ξ ∈ L2(P ), Va(ξ)(x) := ( ξ(xa−1)∆(a) 0 −1 2 if x ∈ P a if x /∈ P a. Here ∆ denotes the modular function of the group G. Note that for g ∈ G and ξ ∈ L2(P ), Wg is given by Wg(ξ)(x) := ( ξ(xg−1)∆(g) 0 −1 2 if xg−1 ∈ P if xg−1 /∈ P. 19 Let M : L∞(P ) → B(L2(P )) be the multiplication representation. Observe that for g = M(1P.g) where P.g := {xg : x ∈ P } ∩ P . Denote the algebra of g ∈ G, Eg = WgW ∗ bounded continuous functions on P by Cb(P ). Since Int(P ) = P , it follows that M is a faithful representation of Cb(P ). For f ∈ Cc(G), let 1P ∗ f ∈ Cb(P ) be defined by 1P ∗ f (t) =Z 1P (ts)f (s−1)ds =Z 1P (ts−1)f (s)∆(s)−1ds =Z 1P −1t(s)f (s)∆(s)−1ds Observe that given a ∈ P , there exists f ∈ Cc(G) such that (1P ∗ f )(a) = 1. For, if a ∈ P , choose f ∈ Cc(G) such that supp(f ) ⊂ Int(P )−1a andR f (s)∆(s)−1ds = 1. For such an f , (1P ∗ f )(a) = 1. Now let f ∈ Cc(G) and ξ ∈ L2(P ) be given. Calculate as below to find that < (Z f (g)Egdg)ξ, ξ > =Z f (g) < Egξ, ξ > dg =Z f (g)(cid:16)Zx∈P 1P g(x)ξ(x)2dx(cid:17)dg =Zx∈P(cid:16)Z f (g)1P g(x)dg(cid:17)ξ(x)2dx =Zx∈P(cid:16)Z f (g)1P (xg−1)dg(cid:17)ξ(x)2dx =Zx∈P(cid:16)Z 1P (xg)f (g−1)∆(g)−1dg(cid:17)ξ(x)2dx =Zx∈P (1P ∗ f )(x)ξ(x)2dx =< M(1P ∗ f )ξ, ξ > . where f (g) = f (g)∆(g). Thus the C ∗-algebra generated by {R f (g)Egdg : f ∈ Cc(G)}, is isomorphic to the C ∗- subalgebra of Cb(P ) generated by {1P ∗ f : f ∈ Cc(G)}. Thus for a ∈ P , there exists a character χa of A such that χa(Z f (g)Egdg) = (1P ∗ f )(a) =Z 1P −1a(s) f (s)∆(s)−1ds =Z 1P −1a(s)f (s)ds 20 This implies that the support of χa is P −1a. Also {χa : a ∈ P } separates the elements of Cb(P ) and hence those of A. This implies that {P −1a : a ∈ P } is dense in Ω. As a consequence, it follows that Ω is the closure of {P −1a : a ∈ P } in the space of closed subsets of G w.r.t. the Vietoris topology. The 'compactification' Ω of P is called the Wiener-Hopf compactification and is considered in [MR82] and in [RS15]. 6 Covariant representations In this section, let X be a compact Hausdorff space and assume that P acts on X on the right injectively. Let X0 := XInt(P ). We also assume that the semi-direct product G := X ⋊ P admits a Haar system. Let Y be a dilation of X, as explained in Section 1, on which the group G acts. For y ∈ Y , let Qy := {g ∈ G : y.g ∈ X}. Recall that for y ∈ Y and g ∈ G, 1Qy(g) = 1X(yg) and 1Int(Qy)(g) = 1X0(yg). Also note that for every y ∈ Y , Qy is closed and QyP ⊂ Qy. Thus by Lemma 4.1 of [RS15], it follows that for every y ∈ Y , the boundary of Qy has measure zero and Int(Qy) = Qy. To state the next lemma, we need to fix some notations. Let a ∈ P and let (Un) Un = {a} and if U is open and contains a then Un ⊂ U eventually. Note that for every n, Un ∩ P a is non-empty. Hence Un ∩ Int(P )a is non-empty for every n. Choose fn ∈ Cc(G) such that fn ≥ 0, be a decreasing sequence of open subsets of G such that ∞\n=1 R fn(g)dg = 1 and supp(fn) ⊂ Un ∩ Int(P )a. For x ∈ X, let Fn(x) =Z fn(g)1X(xg−1)dg. Then Fn ∈ C(X). The continuity of Fn follows from the fact that (1Qx(g)dg)x∈X is a Haar system on X ⋊ P . Observe that (Fn) is uniformly bounded. Lemma 6.1 The sequence Fn converges pointwise to 1X0a. Proof. Since for x ∈ X, 1Qx = 1Int(Qx) a.e., it follows that Fn is given by the equation Fn(x) =Z fn(g)1X0(xg−1) for x ∈ X. Let x ∈ X. From QxP ⊂ Qx, it is easily verifiable that a−1 ∈ Int(Qx) if and only if Int(Qx) ∩ a−1P −1 is non-empty. Suppose a−1 ∈ Int(Qx) i.e. xa−1 ∈ X0. Let U := {g ∈ G : xg−1 ∈ X0}. Then U is open and contains a. Thus there exists N such that n ≥ N implies supp(fn) ⊂ U eventually. Then for n ≥ N, Fn(x) =R fn(g)dg = 21 1. Now suppose xa−1 /∈ X0 i.e. a−1 /∈ Int(Qx). Then Int(Qx)−1 ∩ P a is empty. Thus for g ∈ P a, xg−1 /∈ X0. Since supp(fn) ⊂ P a, it follows that Fn(x) = 0. This proves that (Fn) converges pointwise to 1X0a. This completes the proof. ✷ Lemma 6.2 There exists a sequence (sn) in Int(P ) such that s−1 converges to the identity element e n+1sn ∈ Int(P ) and (sn) Proof. Let (Un) be a countable base (of open sets) at e. We can assume that Un is decreasing. Now U1 ∩ P contains e and is non-empty. Since Int(P ) is dense in P , it follows that U1 ∩ Int(P ) is non-empty. Pick s1 ∈ U1 ∩ Int(P ). Now suppose that s1, s2, · · · , sn are chosen such that sk ∈ Int(P ) ∩ Uk for 1 ≤ k ≤ n and s−1 k+1sk ∈ Int(P ) for 1 ≤ k ≤ n − 1. Since e ∈ sn(Int(P ))−1 ∩ Un+1, it follows that snInt(P )−1 ∩ Un+1 ∩ P is non-empty. But Int(P ) = P . Thus snInt(P )−1 ∩ Un+1 ∩ Int(P ) is non-empty. Let sn+1 ∈ snInt(P )−1 ∩ Un+1 ∩ Int(P ). s−1 n+1sn ∈ Int(P ) for every n. This completes the proof. Then it is clear that the sequence (sn) constructed as above converges to e and ✷. Consider a sequence (sn) as in Lemma 6.2 converging to the identiy e. Let a ∈ Int(P ) and set tn := s−1 n ∈ Int(P ). Since a ∈ Int(P ), Int(P ) is open and (tn) converges to a, we can assume without loss of generality that tn ∈ Int(P ) for every n. With this notation, we have the following lemma. n a. Then observe that tn+1t−1 Lemma 6.3 The sequence (1X0tn) decreases pointwise to 1Xa. Proof. Since tn+1t−1 n ∈ Int(P ), it follows that X0tn+1 ⊂ X0tn for every n. Since at−1 n ∈ Int(P ), it follows that Xa ⊂ X0tn for every n. Thus Xa ⊂ X0tn. Now suppose ∞\n=1 y ∈ X0tn for every n. Then yt−1 n ) → ya−1. Since the closure of X0 in Y is X, it follows that ya−1 ∈ X. Hence y ∈ Xa. This proves that n ⊂ X0 for every n. Note that (yt−1 Xa = X0tn. This completes the proof. ✷ Let B(X) be the space of bounded Borel measurable functions on X. For φ in B(X) ∞\n=1 and g ∈ G, let Rg(φ) be defined by Rg(φ)(x) :=( φ(x.g) 0 if x.g ∈ X if x.g /∈ X Then Rg(φ) ∈ B(X). 22 Definition 6.4 Let π : C(X) → B(H) be a unital ∗-representation and V : P → B(H) be an isometric representation with commuting range projections. Denote the extension of π to B(X), obtained via the Riesz representation theory, by π itself [See [Arv02]]. For g = ab−1, let Wg := V ∗ b Va. The pair (π, V ) is said to be a covariant representation of (X, P ) if for φ ∈ B(X), Wgπ(φ)W ∗ g = π(Rg−1(φ)). Remark 6.5 Since G = (Int(P ))(Int(P ))−1, it follows that (π, V ) is a covariant repre- a = π(Ra−1(φ)) for φ ∈ B(X) sentation if and only if V ∗ and a ∈ Int(P ). We leave this verification to the reader. a π(φ)Va = π(Ra(φ)) and Vaπ(φ)V ∗ We fix a few notations that will be useful for the rest of this section. Notations: Let Y be the dilation of X, as explained in Section 1, on which G acts. Then G := X ⋊ P is a closed subset of X × G and also of Y ⋊ G. For φ ∈ Cc(Y ), we let bφ ∈ C(X) be the restriction. Define πY : Cc(Y ) → B(H) by πY (φ) = π(bφ). For φ ∈ Cc(Y ) and g ∈ G, let Rg(φ) ∈ Cc(Y ) be given by Rg(φ)(y) = φ(y.g) for y ∈ Y . For ψ ∈ Cc(Y ⋊ G) and g ∈ G, let ψg ∈ Cc(Y ) be defined by ψg(y) = ψ(y, g). For χ ∈ Cc(G) and g ∈ G, let χg ∈ B(X) be defined by χg(x) = χ(x, g) if (x, g) ∈ G and χg(x) = 0 if (x, g) /∈ G. Let π : C(X) → B(H) be a unital ∗-representation. For ξ ∈ H, let dµξ,ξ be the probability measure on X such that Z φ(x)dµξ,ξ(x) =< π(φ)ξ, ξ > f or φ ∈ C(X). The same equality holds for φ ∈ B(X). Proposition 6.6 Let π : C(X) → B(H) be a unital ∗-representation and V : P → B(H) be an isometric representation with commuting range projections. Then the following are equivalent. (1) The pair (π, V ) is a covariant representation. (2) For a ∈ Int(P ) and φ ∈ C(X), V ∗ a π(φ)Va = π(Ra(φ)) and π(1X0a) = Ea. Proof. For a ∈ P , let σa : X → X be the map sending x → xa. Suppose (π, V ) is a covariant representation. Then the covariance relation implies that for g ∈ G, Eg := WgW ∗ g = π(1Xg∩X). 23 < π(F )ξ, ξ > =Z F (x)dµξ,ξ(x) =Z f (g)1Xg(x)dgdµξ,ξ(x) =Z f (g)(cid:16)Z 1Xg(x)dµξ,ξ(x)(cid:17)dg =Z f (g) < π(1Xg∩X)ξ, ξ > dg =Z f (g) < Egξ, ξ > dg. Let a ∈ Int(P ) be given. Choose a sequence (fn) as in Lemma 6.1 and Let Fn(x) := Thus π(F ) =R f (g)Eg. R fn(g)1Xg(x) for x ∈ X. Note that Fn is uniformly bounded. By Lemma 6.1, it follows that Fn converges pointwise to 1X0a. On the other hand, we have π(Fn) =R fn(g)Egdg. Since g → Eg is strongly continuous, it is easily verifiable that R fn(g)Egdg converges strongly to Ea. Hence π(1X0a) = Ea. Clearly by definition V ∗ proves (1) implies (2). a π(φ)Va = π(Ra(φ)). This Now assume (2). The equality V ∗ a π(φ)Va = π(Ra(φ)) for a ∈ P and φ ∈ C(X) translates to the fact that for a ∈ P and ξ ∈ H, the push-forward measure (σa)∗(µξ,ξ) = a π(φ)Va = π(Ra(φ)) for a ∈ P and φ ∈ B(X). Now by Remark 6.5, it is µVaξ,Vaξ. Hence V ∗ a = π(Ra−1(φ)) for a ∈ Int(P ) and φ ∈ B(X). Now let a ∈ enough to show that Vaπ(φ)V ∗ Int(P ) and φ ∈ B(X) be given. Then by assumption (2), we have V ∗ a π(Ra−1φ)Va = π(φ). a = Eaπ(Ra−1(φ))Ea. By the strong continuity of g → Eg, by assumption Hence Vaπ(φ)V ∗ (2) and Lemma 6.3, it follows that π(1Xa) = Ea. Hence Vaπ(φ)V ∗ a = π(1XaRa−1(φ)) = π(Ra−1(φ)). This completes the proof. ✷ Theorem 6.7 Let X be a compact Hausdorff space on which P acts injectively. Let G := X ⋊ P . Assume that G has a Haar system. For φ ∈ C(X) and f ∈ Cc(G), let φ ⊗ f ∈ Cc(G) be defined by the equation (φ ⊗ f )(x, g) = φ(x)f (g). We denote 1 ⊗ f by Let (π, V ) be a covariant representation of (X, P ) on a Hilbert space H. Then there Let f ∈ Cc(G). Set F (x) =R f (g)1X(xg−1)dg. Then F ∈ C(X). Now calculate to find that exists a representation λ : C ∗(G) → B(H) such that ef . (1) For f ∈ Cc(G), λ(ef ) =R ∆(g)− 1 the group G. 24 2 f (g)Wg−1dg. Here ∆ is the modular function of Proof of Theorem 6.7. Let φ ∈ Cc(G). We claim that G ∋ g → π(φg)Wg−1 ∈ B(H) is strongly continuous. Let Φ ∈ Cc(Y ⋊ G) be an extension of φ. Since (π, V ) is co- (2) For φ ∈ C(X) and f ∈ Cc(G), λ(φ ⊗ f ) = π(φ)λ(ef ). variant, it follows that Eg = π(1Xg∩X). Now observe that π(cΦg)Wg−1 = π(φg)Wg−1. For π(cΦg)Wg−1 = π(cΦg)Eg−1Wg−1 = π(cΦg1Xg−1)Wg−1 = π(φg)Wg−1. But g → cΦg ∈ C(X) is continuous. Hence G ∋ g → π(cΦg) ∈ B(H) is strongly continuous and consequently G ∋ g → π(φg)Wg−1 ∈ B(H) is strongly continuous. For φ ∈ Cc(G), let λ(φ) ∈ B(H) be Also we have shown that if Φ ∈ Cc(Y ⋊ G) is an extension of φ ∈ Cc(G), then 2 π(φg)Wg−1dg. 2 πY (Φg)Wg−1dg. For φ ∈ Cc(G), calculate as follows to find that λ(φ) :=Z ∆(g)− 1 λ(φ) =Z ∆(g)− 1 λ(φ)∗ =Z Wgπ(φg)∆(g)− 1 =Z Wgπ(φg)W ∗ =Z π(Rg−1(φg))Wg∆(g)− 1 =Z π(Rg(φg−1))Wg−1∆(g) =Z π((φ∗)g)Wg−1∆(g)− 1 g Wg∆(g)− 1 2 dg 2 dg = λ(φ∗). 2 dg 2 dg 1 2 ∆(g)−1dg Thus λ preserves the adjoint. Now let φ, ψ ∈ Cc(G) be given and let Φ, Ψ ∈ Cc(Y ⋊ G) be extensions of φ and ψ respectively. Consider the function on Y ⋊ G defined by the equation Φ ◦ Ψ(y, g) =Z Φ(y, h)Ψ(y.h, h−1g)1X(y.h)dh. A simple application of the dominated convergence theorem together with the fact that 1X(y.h) = 1X0(y.h) a.e. for every y implies that Φ ◦ Ψ is continuous. Clearly Φ ◦ Ψ is compactly supported and is an extension of φ ∗ ψ. 25 Let g ∈ G and ξ ∈ H be given. Then < πY ((Φ ◦ Ψ)g)ξ, ξ > =Z Φ ◦ Ψ(x, g)dµξ,ξ(x) =Z (cid:16)Z Φ(x, h)Ψ(x.h, h−1g)1X(xh)dh(cid:17)dµξ,ξ(x) =Z (cid:16)Z Φ(x, h)Ψ(x.h, h−1g)1Xh−1(x)dµξ,ξ(x)(cid:17)dh =Z < πY (Φh)πY (Rh(Ψh−1g))πY (1Xh−1)ξ, ξ > dh =Z < πY (Φh)πY (Rh(Ψh−1g))Eh−1ξ, ξ > dh find that 2 πY (Φg)W ∗ g πY (Ψh)WgWg−1Wh−1dgdh 2 πY (Φg)πY (Rg(Ψh))Wg−1Wh−1dgdh 2 πY (Φg)Wg−1πY (Ψh)Wh−1dgdh Thus for g ∈ G, πY ((Φ ◦ Ψ)g) = R πY (Φh)πY (Rh(Ψh−1g))Eh−1dh. Now calculate to λ(φ)λ(ψ) =Z ∆(gh)− 1 =Z ∆(gh)− 1 =Z ∆(gh)− 1 =Z ∆(gh)− 1 =Z (cid:16)Z ∆(k)− 1 =Z (cid:16)Z πY (Φg)πY (Rg(Ψg−1k)Eg−1dg(cid:17)∆(k)− 1 =Z ∆(k)− 1 2 πY (Φg)πY (Rg(Ψg−1k))Eg−1Wk−1dk(cid:17)dg 2 πY (Φg)πY (Rg(Ψh))Eg−1Wh−1g−1dgdh [by Proposition 3.3] 2 Wk−1dk 2 πY ((Φ ◦ Ψ)k)Wk−1 = λ(φ ∗ ψ). Hence λ preserves the multiplication. For φ ∈ B(X), one has π(φ) ≤ φ∞ where .∞ is the sup norm on B(X). Let 26 K be a compact subset of G. Then for φ ∈ Cc(G) with supp(φ) ⊂ X × K, observe that λ(φ) ≤Z ∆(g)− 1 2 π(φg)dg ≤Zg∈K ≤(cid:16) sup g∈K ∆(g)− 1 2 φgdg ∆(g)− 1 2(cid:17)φ∞Z 1K(g)dg. Thus it is clear that the map λ : Cc(G) → B(H) is continuous when Cc(G) is given the inductive limit topology and B(H) is given the norm toplogy. By Renault's disintegration theorem, one obtains a bonafide representation λ : C ∗(G) → B(H). Conditions (1) and (2) follows just from definitions. This completes the proof. ✷ 7 The main theorem Let V : P → B(H) be an isometric representation with commuting range projections. Let A and Ω be as in sections 3-5. Denote the open set ΩInt(P ) by Ω0. Let π : C(Ω) → B(H) be the representation induced by the inclusion A ⊂ B(H). Denote the extension to B(Ω) by π itself. First let us show that (π, V ) is a covariant representation. Lemma 7.1 The pair (π, V ) is a covariant representation. Proof. By definition, a π(φ)Va = π(Ra(φ)) for φ ∈ C(X). Now fix a ∈ P . Choose a sequence (fn) ∈ Cc(G) as in Lemma 6.1. Set Fn := it follows that for a ∈ P , V ∗ R fn(g)Egdg ∈ C(Ω). Then by the strong continuity of g → Eg, it is clear that Fn converges strongly to Ea. Now by definition, for A ∈ Ω, Fn(A) =R fn(g)1A(g)dg. By Proposition 5.2, it follows that Fn(A) =R fn(g)1Ω(Ag−1)dg for A ∈ Ω. By Lemma 6.1, it follows that Fn converges pointwise to 1Ω0a. Hence π(1Ω0a) = Ea. The proof follows now from Proposition 6.6. ✷ Proposition 7.2 Let H be a Hilbert space and V : P → B(H) be an isometric repre- sentation with commuting range projections. Let Ω be as in Sections 3-5. Then there exists a ∗-homomorphism λ : C ∗(Ω ⋊ P ) → B(H) such that for f ∈ Cc(G), λ(ef ) =Z ∆(g)− 1 2 f (g)Wg−1dg. Moreover the range of λ is generated by {R f (g)Wgdg : f ∈ Cc(G)}. 27 Proof. For f ∈ Cc(G), let bf ∈ C(Ω) be defined by By (4) of Remark 4.5 and the fact that Ω ⊂ Ωu, it follows that {bf : f ∈ Cc(G)} separates points of Ω. Thus by Proposition 2.1, the ∗-algebra generated by {ef : f ∈ Cc(G)} is bf (A) :=Z f (g)1Ω(Ag)dg =Z f (g)1A(g−1). dense in C ∗(Ω ⋊ P ). Now the proof follows directly from Lemma 7.1 and Proposition 6.7. ✷ Theorem 7.3 Let H be a Hilbert space and V : P → B(H) be an isometric representa- tion with commuting range projections. Let Ωu := {A ∈ C(G) : P −1 ⊂ A and P −1A ⊂ A} with the Vietoris topology. Consider the right action of P on Ωu by right multiplication. Then there exists ∗-homomorphism λ : C ∗(Ωu ⋊ P ) → B(H) such that for f ∈ Cc(G), λ(ef ) =Z ∆(g)− 1 2 f (g)Wg−1dg. Moreover the range of λ is generated by {R f (g)Wgdg : f ∈ Cc(G)}. Proof. By Remark 5.3, it follows that Ω ⋊P is isomorphic to the restriction Ωu ⋊P Ω and Ω is an invariant subset of Ωu. Consider the natural map res : Cc(Ωu ⋊ P ) → Cc(Ω ⋊ P ) which on Cc(Ωu ⋊ P ) is simply the restriction. Let eλ : C ∗(Ω ⋊ P ) → B(H) be the representation as in Propositon 7.2. Now one completes the proof by setting λ :=eλ◦ res. ✷ Remark 7.4 Proposition 7.3 says that the C ∗-algebra of the groupoid Ωu ⋊ P can be interpreted as the 'universal' C ∗-algebra which encodes the isometric representations with commuting range projections. However, the space Ωu is quite large to describe explicitly even for the simple example of the quarter plane [0, ∞) × [0, ∞) ⊂ R2. We end this article by considering two well-known results which are a part of folklore in operator algebras. Example 7.5 Let P := N and G := Z with the discrete topology. Consider the one- point compactification N∞ := N ∪ {∞}. The semigroup N acts on N∞ by translation with the convention that ∞ + n = ∞ for n ∈ N. It is easy to verify that the map N∞ ∋ n → (−∞, n] ∈ Ωu is an N-equivariant homeomorphism. Here (−∞, ∞] is just N. The groupoid N∞ ⋊ N is amenable and C ∗ red(N∞ ⋊ N) is just the Toeplitz-algebra. Now Theorem 7.3 is just the well-known Coburn's theorem. 28 Example 7.6 Let R+ = [0, ∞). Let P := R+ and G := R with the usual Euclidean topology and addition as the group operation. Consider the one-point compactification [0, ∞] := [0, ∞) ∪ {∞}. The semigroup [0, ∞) acts on R ∪ {∞} by translation with the convention that ∞ + x = ∞ for x ∈ [0, ∞). It is easily verifiable that the map [0, ∞] ∋ x → (−∞, x] ∈ Ωu is a R+-equivariant homeomorphism. The groupoid [0, ∞] ⋊ R+ is amenable and C ∗ red([0, ∞] ⋊ [0, ∞)) is the usual Wiener-Hopf algebra. [See [MR82]] : t ≥ 0} commutes. For if t = r + s then EtEr = VrVsV ∗ Observe that if V : R+ → B(H) is an isometric representation then the range projec- tions {Et := VtV ∗ r = Et. t Hence if t > r, then EtEr = Et. Now the claim follows from the fact that R+ is totally ordered. Thus if V : R+ → B(H) is an isometric representation, then there exists a representation π : W([0, ∞), R) → B(H) such that s V ∗ r VrV ∗ π(ef ) =Z ∞ 0 f (t)Vt +Z 0 −∞ f (t)V ∗ t for f ∈ Cc(R). References [Arv02] William Arveson, A short course on spectral theory, Graduate Texts in Math- ematics, vol. 209, Springer-Verlag, New York, 2002. [Cun08] Joachim Cuntz, C ∗-algebras associated with the ax + b-semigroup over N, K- theory and noncommutative geometry, EMS Ser. Congr. Rep., 2008, pp. 201 -- 215. [Exe03] Ruy Exel, A new look at the crossed-product of a C ∗-algebra by an endomor- phism, Ergodic Theory Dynam. Systems 23 (2003), no. 6, 1733 -- 1750. [HN95] Joachim Hilgert and Karl-Hermann Neeb, Wiener-Hopf operators on ordered homogeneous spaces. I, J. Funct. Anal. 132 (1995), no. 1, 86 -- 118. [Li12] Xin Li, Semigroup C∗-algebras and amenability of semigroups, J. Funct. Anal. 262 (2012), no. 10, 4302 -- 4340. [Li13] , Nuclearity of semigroup C ∗-algebras and the connection to amenabil- ity, Adv. Math. 244 (2013), 626 -- 662. [MR82] Paul S. Muhly and Jean N. Renault, C ∗-algebras of multivariable Wiener-Hopf operators, Trans. Amer. Math. Soc. 274 (1982), no. 1, 1 -- 44. 29 [Mur91] G. J. Murphy, Ordered groups and crossed products of C ∗-algebras, Pacific J. Math. 148 (1991), no. 2, 319 -- 349. [Mur94] Gerard J. Murphy, Crossed products of C ∗-algebras by semigroups of automor- phisms, Proc. London Math. Soc. (3) 68 (1994), no. 2, 423 -- 448. [Mur96a] , C ∗-algebras generated by commuting isometries, Rocky Mountain J. Math. 26 (1996), no. 1, 237 -- 267. [Mur96b] , Crossed products of C ∗-algebras by endomorphisms, Integral Equa- tions Operator Theory 24 (1996), no. 3, 298 -- 319. [Nic87] Alexandru Nica, Some remarks on the groupoid approach to Wiener-Hopf op- erators, J. Operator Theory 18 (1987), no. 1, 163 -- 198. [Nic90] , Wiener-Hopf operators on the positive semigroup of a Heisenberg group, Linear operators in function spaces (Timi¸soara, 1988), Oper. Theory Adv. Appl., vol. 43, Birkhauser, Basel, 1990, pp. 263 -- 278. [RS15] J Renault and S Sundar, Groupoids associated to Ore semigroup actions, to appear in J. Operator Theory (2015), arXiv:1402.2762v2. S. Sundar ([email protected]) Chennai Mathematical Institute, H1 Sipcot IT Park, Siruseri, Padur, 603103, Tamilnadu, INDIA. 30
1107.0500	2	1107	2012-05-20T18:27:45	Factorization of Matrices of Quaternions	[ "math.OA" ]	We review known factorization results in quaternion matrices. Specifically, we derive the Jordan canonical form, polar decomposition, singular value decomposition, the QR factorization. We prove there is a Schur factorization for commuting matrices, and from this derive the spectral theorem. We do not consider algorithms, but do point to some of the numerical literature. Rather than work directly with matrices of quaternions, we work with complex matrices with a specific symmetry based on the dual operation. We discuss related results regarding complex matrices that are self-dual or symmetric, but perhaps not Hermitian.	math.OA	math	FACTORIZATION OF MATRICES OF QUATERNIONS TERRY A. LORING Abstract. We review known factorization results in quaternion matrices. Specifically, we derive the Jordan canonical form, polar decomposition, singular value decomposition, the QR factorization. We prove there is a Schur factorization for commuting matrices, and from this derive the spectral theorem. We do not consider algorithms, but do point to some of the numerical literature. Rather than work directly with matrices of quaternions, we work with complex matrices with a specific symmetry based on the dual operation. We discuss related results regarding complex matrices that are self-dual or symmetric, but perhaps not Hermitian. 1. The quaternionic condition It is possible to prove many factorization results for matrices of qua- terions by deriving them from their familiar complex counterparts. The amount of additional work is surprisingly small. There are more abstract factorization theorems that apply to real C ∗-algebras, along the lines of the delightful paper by Pedersen on factorization in (complex) C ∗-algebras. We are dealing here with more basic questions, involving only finite-dimensional linear algebra. These are never (hardly ever?) addressed in basis linear algebra texts, creating the impression that linear algebra over the quaternions is more difficult than it really is. There are serious hazards within linear algebra over the quaternions. One can be lured to difficult questions of determinants and handedness of the spectrum, leaving perhaps victorious, but with the impression that all of linear algebra over the quaternions is going to be difficult. We mathematicians are well-advised, when upon such uneven ground, to seek guidance from physicists. Such guidance helped select the top- ics, and helped suggest notation. 1991 Mathematics Subject Classification. 15B33. Key words and phrases. Kramers pair, matrix decompositions, dual operation, quaternions. 1 FACTORIZATION OF MATRICES OF QUATERNIONS 2 One topic was included for reasons of elegance, the Jordan canonical form. For practical purposes, the Schur decomposition will generally suffice. We prove that as well. It is assumed the reader is familiar with such things as the spectral It is not theorem and the functional calculus for normal matrices. assumed that the reader knows about C ∗-algebras or physics, but these topics are mentioned incidentally. Let H denote the algebra of quaternions H =n a + bi + cj + dk(cid:12)(cid:12)(cid:12) a, b, c, d ∈ Ro . This is an algebra over R. The canonical embedding C ֒→ H sending 1 to 1 and i to i does not make H into an algebra over C. The trouble is that the embedding is not central. An additional algebraic operation is the involution = a − bi − cj − dk which satisfies several axioms, including the following. (cid:16)a + bi + cj + dk(cid:17)∗ α ∈ R, x ∈ H =⇒ (αx)∗ = αx∗ x∗x = 0 =⇒ x = 0 (xy)∗ = y∗x∗ Turning to matrices, the algebra MN (H) has the expected structure of a unital R-algebra, plus the involution [aij]∗ =(cid:2)a∗ ji(cid:3) of conjugate-transpose. We would like to know A∗A = 0 =⇒ A = 0 and AB = I =⇒ BA = I. We will quickly prove these after we consider a representation of MN (H) on C2N . We have an obvious representation of MN (H) on HN and quickly notice that since left and right scalar multiplication of HN disagree, we have both left-eigenvalues λ ∈ H solving Av = λv and right-eigenvalues µ ∈ H solving Av = vµ. A glance at the survey [13] reveals that many difficulties arise. FACTORIZATION OF MATRICES OF QUATERNIONS 3 Ponder the situation for real matrices, in MN (R). To the chagrin of undergraduates, it is most natural to consider the representation of MN (R) on CN . Given a real orthogonal matrix O we get the full picture of why it might not diagonalize in MN (R) when we look at the complex eigenvalues. It is at this point that we are implicitly letting MN (R) act on CN . We do well in the case of the quaternions to regard MN (H) as rep- resented on C2N , or in more modern terms that de-emphasizes the role of vectors, via a certain embedding χ : MN (H) → M2N (C). This is a very old trick. For N > 1 it appears to have been noticed first by H. C. Lee [7]. Definition 1.1. Given two complex N-by-N matrices A and B we set χ(cid:16)A + Bj(cid:17) =(cid:20) A B −B A (cid:21) . We generally study complex matrices in the image of χ and only at the end of our calculations do we draw conclusions about MN (H). This is in keeping with applications in quantum mechanics, where Hilbert space is always complex and time-reversal symmetry will often be incor- porated in the conjugate linear operation T . We define T : C2N → C2N by (1.1) T (cid:18)(cid:20) v v (cid:21) . w (cid:21)(cid:19) =(cid:20) −w See [9, §2.2] for details on when T is relevant to time reversal symmetry. A less generous description our our approach is we plan to study complex matrices with a certain symmetry that is useful in physics, and then sell the same results to pure mathematicians by re-branding them as theorems about matrices of quaternions. The operator T is relatively well behaved, despite being only conju- gate linear. It preserves orthogonality, and indeed (1.2) Also hT ξ,T ηi = hξ, ηi. (1.3) ξ ⊥ T ξ which is another one-line calculation. Lemma 1.2. The mapping χ is well-defined, is an R-algebra homo- morphism, is one-to-one and satisfies (χ (Y ))∗ = χ (Y ∗) . FACTORIZATION OF MATRICES OF QUATERNIONS 4 Proof. Notice that every quaternion q can be written as α + βj with α and β in C. This makes the map well-defined. It is clearly one-to-one and R-linear. Notice βj = j ¯β for complex number β and so Bj = jB. Therefore AD + BC χ(cid:16)A + Bj(cid:17) χ(cid:16)C + Dj(cid:17) =(cid:20) A B −B A (cid:21)(cid:20) C D −D C (cid:21) −BC − AD −BD + AC (cid:21) =(cid:20) AC − BD = χ(cid:16)(cid:0)AC − BD(cid:1) +(cid:0)AD + BC(cid:1) j(cid:17) = χ(cid:16)(cid:16)A + Bj(cid:17)(cid:16)C + Dj(cid:17)(cid:17) = χ(cid:16)A∗ − BTj(cid:17) = χ(cid:16)(cid:16)A + Bj(cid:17)∗(cid:17) . −B A (cid:21)∗ =(cid:20) A B (cid:16)χ(cid:16)A + Bj(cid:17)(cid:17)∗ and (cid:3) We can describe the image of χ in several useful ways. We will need several unary operations on complex matrices. For starters we need the transpose AT and the pointwise conjugate A and the conjugate- T transpose, or adjoint, A∗ = A = AT. Finally, we need a twisted transpose that is useful in physics. This is a "generalized involution" X ♯ called the dual operation that is defined only for X in M2N (C). Definition 1.3. For A, B, C and D all complex N-by-N matrices, define C D (cid:21)♯ (cid:20) A B =(cid:20) DT −BT −C T AT (cid:21) . Alternatively, we set X ♯ = −ZX TZ where −I 0 (cid:21) . Z =(cid:20) 0 I (1.4) Notice that Z could be replaced by a matrix with similar properties and define a variation on the dual operation. Indeed, the choice of Z is not standardized. Lemma 1.4. For X in M2N (C) the following are equivalent. (1) X is in the image of χ; (2) X ∗ = X ♯; (3) X = −T ◦ X ◦ T meaning Xξ = −T (XT (ξ)) for every vector ξ in C2N . −T ◦ X ◦ T (cid:20) v w (cid:21) = −T (cid:18)(cid:20) A B C D (cid:21)(cid:20) −w v (cid:21)(cid:19) −Cw + Dv (cid:21)(cid:19) = −T (cid:18)(cid:20) −Aw + Bv =(cid:20) −Cw + Dv Aw − Bv (cid:21) −B A (cid:21)(cid:20) v w (cid:21) =(cid:20) D −C FACTORIZATION OF MATRICES OF QUATERNIONS 5 Proof. Assume Then X =(cid:20) A B −B A (cid:21) . AT # = X ♯. −BT T T X ∗ =" A B Conversely X ∗ = X ♯ translates into block form as B∗ D∗ (cid:21) =(cid:20) DT −BT (cid:20) A∗ C ∗ −C T AT (cid:21) so we have proven (1) ⇐⇒ (2). We compute −T ◦ X ◦ T for X again in block form, and find so the matrix for the linear operator −T ◦ X ◦ T is (cid:0)X ♯(cid:1)∗. Therefore (2) ⇐⇒ (3). Definition 1.5. Let us call (cid:3) X ∗ = X ♯ the quaternionic condition and the matrix X we will call a quaternionic matrix. We now dispose of the implications and A∗A = 0 =⇒ A = 0 AB = I =⇒ BA = I for matrices of quaternions. These are true implications for complex matrices, and in particular for quaternionic matrices, and therefore true for matrices of quaternions. We pause to note some axioms of the dual operation. It behaves a lot like the transpose. It is linear, (X + αY )♯ = X + αY ♯ FACTORIZATION OF MATRICES OF QUATERNIONS 6 which is true even for complex α. Here X and Y are any 2N-by-2N complex matrices. The dual reverses multiplication, It undoes itself, (XY )♯ = Y ♯X ♯. and commutes with the adjoint = X = (X ∗)♯ . (cid:0)X ♯(cid:1)♯ (cid:0)X ♯(cid:1)∗ Lemma 1.6. Every matrix G in M2N (C) can we expressed in a unique way as with X and Y being quaternionic matrices. G = X + iY Proof. Given G we set and X = 1 2 G♯∗ + 1 2 G Y = i 2 G♯∗ − i 2 G. (cid:3) 2. Kramers degeneracy and Schur factorization Eigenvalue doubling is a key feature of self-dual, self-adjoint matri- ces. In physics this is called the Kramers degeneracy theorem, or the theory of Kramers pairs [6, 12, 9]. This generalizes in two ways, to give eigenvalue doubling given the symmetry X = X ♯ and conjugate-pairing of eigenvalues given the symmetry X ∗ = X ♯. Such a collection of paired eigenvectors will, in good situations, form If U is unitary matrix that satisfies the quater- a unitary matrix. nionic condition then it satisfies another symmetry making it symplec- tic, specifically U TZU = Z. Lemma 2.1. Suppose U is a unitary 2N -by-2N matrix. The following are equivalent: (1) U is symplectic; (2) U ∗ = U ♯; (3) ZU = U Z; (4) U ◦ T = T ◦ U ; (5) If v is column j of U for j ≤ N then column N + j of U is T (v). Proof. This follows easily from Lemma 1.4. (cid:3) FACTORIZATION OF MATRICES OF QUATERNIONS 7 We need to know that the group of symplectic unitary acts transi- tively on Cn. Lemma 2.2. If v is any unit vector in C2N then there is a symplectic unitary with U e1 = v. Proof. Let v1 = v. We use (1.2) and (1.3) to select in order vectors that are an orthonormal basis for C2N , but of the form If we reorder to v1,T v1, v2,T v2, . . . , vN ,T vN . we have the columns of the desired symplectic unitary. v1, v2, . . . , vN ,T v1,T v2, . . . ,T vN (cid:3) The most basic result in this realm is an ugly lemma that says that T maps left eigenvectors of X to right eigenvectors of X ♯, with the same eigenvalue. The second part of this lemma is more elegant, if less general. It specifices how every quaternionic matrix has a conjugate symmetry it its spectral decomposition. Lemma 2.3. Suppose X is in M2N (C). (1) If Xξ = λξ then (2) If X ∗ = X ♯ and Xξ = λξ then (T ξ)∗ X ♯ = λ (T ξ)∗ . Proof. (1) Starting with X (T ξ) = λ (T ξ) . and X =(cid:20) A B C D (cid:21) ξ =(cid:20) v w (cid:21) we find that Xξ = λξ translates to Av + Bw = λv Cv + Dw = λw and (T v)∗ X ♯ = λ (T v)∗ translates to −wTDT − vTC T = −λwT wTBT + vTAT = λvT so these are equivalent conditions. (2) follow from (1) by taking adjoints. (cid:3) FACTORIZATION OF MATRICES OF QUATERNIONS 8 Now we are able to extend Kramers degeneracy to a variety of situa- tions, starting with a block diagonalization for commuting quaternionic matrices. Part (2) of Theorem 2.4 appeared in [11]. Theorem 2.4. Suppose X1, . . . , Xk in M2N (C) commute pairwise and X ∗ j for all j. j = X ♯ (1) There is a single symplectic unitary U so that, for all j, U ∗XjU =(cid:20) Tj −Sj Tj (cid:21) . Sj with Tj upper-triangular and Sj strictly upper-triangular. (2) If, in addition, the Xj are normal then there is a single sym- plectic unitary U so that, for all j, U ∗XjU =(cid:20) Dj 0 Dj (cid:21) . 0 with Dj diagonal. (3) Every symplectic unitary has determinant one. Proof. (1) A finite set of commuting matrices will have a common eigen- vector, so let v be a unit vector so that Xjv = λjv. We know then that so T v is an eigenvalue for Xj with eigenvector λj. There is a symplectic unitary U1 so that U1e1 = v, and then U1eN +1 = T v by Lemma 2.1. Let Yj = U ∗ 1 XjU1. Then and Yje1 = λje1 YjeN +1 = λjeN +1. This means the column 1 and column N + 1 are all but zeroed-out, ∗ λj 0 Aj 0 ∗ λj 0 Bj ∗ 0 0 Cj ∗ 0 Dj Yj =  .   Up to a non-symplectic change of basis we are looking at block-upper triangular matrices that commute, so the lower-right corners in that basis commute. This means that the (2.1) Zj =(cid:20) Aj Cj Bj Dj (cid:21) all commute, and each must satisfy Z ∗ proven the first claim. j = Z ♯ j . By induction, we have FACTORIZATION OF MATRICES OF QUATERNIONS 9 For (2) we modify the proof just a little. Starting with the Xj normal, we find the Yj are also normal and so Yj =  .   0 λj 0 Aj 0 0 Bj 0 0 0 Cj 0 0 Dj 0 λj This, after an appropiate basis change, would be block-diagonal, and from this we can conclude that the matrix in (2.1) is normal. The induction proceeds as before, with the stronger conclusion that Bj = Cj = 0 and Aj = Dj is diagonal. (3) Applying (2) we find W = U(cid:20) D 0 0 D (cid:21) U ∗ where D is a diagonal unitary. Therefore det(W ) = det(D) det(D) = det(D)det(D) ≥ 0 and since a unitary has determinant on the unit circle, we are done. (cid:3) Corollary 2.5. Every matrix in MN (H) is unitarily equivalent to a upper-triangular matrix in MN (H) that has complex numbers on the diagonal. There is an algorithm [1] for the Schur decomposition of quaternionic matrices. Corollary 2.6. Every normal matrix in MN (H) is unitarily equivalent to a diagonal matrix in MN (C). Every Hermitian matrix in MN (H) is unitarily equivalent to a diagonal matrix in MN (R). Corollary 2.7. Every Hermitian self-dual matrix X in M2N (C) is of the form for some symplectic unitary U and a diagonal real matrix Dj. X = U(cid:20) Dj 0 Dj (cid:21) U ∗ 0 These corollaries cause us to reconsider the concept of left and right eigenvalues in H and focus on just those that are in C. For details on left eigenvalues and non-complex right eigenvalues, consult [4]. We put the complex right eigenvalues in a simple context with the following. Lemma 2.8. Suppose v, w ∈ CN and λ ∈ C and A, B ∈ MN (C). Then (cid:20) A B −B A (cid:21)(cid:20) v w (cid:21) = λ(cid:20) v w (cid:21) FACTORIZATION OF MATRICES OF QUATERNIONS 10 if and only if (cid:16)A + Bj(cid:17)(cid:16)v − jw(cid:17) =(cid:16)v − jw(cid:17) λ. Therefore λ ∈ C is a right eigenvalue of X ∈ MN (H) if and only if λ is an eigenvalue of χ (X). Proof. This a short, direct calculation. (cid:3) 3. Jordan canonical form Kramers degeneracy extends to the generalized eigenvectors used to find the Jordan canonical form. Lemma 3.1. Suppose X ♯ = X ∗. If (X − λI)r v = 0 and (cid:0)X − λI(cid:1)r Proof. For any Y we have then T v 6= 0. T v = 0 (X − λI)r−1 v 6= 0 (cid:0)X − λI(cid:1)r−1 kY vk = kY vk = kZY ZZvk and proving Since the result follows. kY vk = kY ♯∗T vk. (cid:16)(X − λI)k(cid:17)♯∗ =(cid:0)X − λI(cid:1)k (cid:3) We see the general idea of a proof [14] of the Jordan decomposition for a quaternionic matrix. When we build a Jordan basis we need to respect T in two ways. Whatever basis we pick for the subspace corresponding to λ with positive imaginary part, we apply T to get the basis for the subspace corresponding to λ. When λ is real we need to pick generalized eigenvectors in pairs. The Jordan form of a quaternionic matrix is not so elegant, as each Jordan blocks larger than 2-by-2 gets spread around to all four quad- rants of the matrix. We work directly with a Jordan basis and then make our final conclusion in terms of quaternions. Theorem 3.2. Suppose X ♯ = X ∗. There is a Jordan basis for X consisting of pairs of the form v,T v. FACTORIZATION OF MATRICES OF QUATERNIONS 11 Proof. Let Nλ denote the subspace of all generalized eigenvectors for λ toghether with the zero vector. Recall that just treating X as a complex matrix, the procedure to select a Jordan basis involves selecting, for each λ, a basis for Nλ with the following property: whenever b is in this basis, then (X − λI)b is either zero or back in this basis. For λ in the spectrum with positive imaginary part we make such a choice and then apply T to get a set of vectors in Nλ that has the correct number of elements to be a basis of Nλ. It will also be a linearly independent set since Z and conjugation both preserve linear independence. Since this basis of Nλ has the desired property. (X − λI)T b = T (X − λI)b For λ in the spectrum that is real we need to modify the procedure for selecting the basis of Nλ. A common procedure selects a basis br,1, . . . , br,mr for (3.1) ker (X − λ)r ∩(cid:0)ker (X − λ)r−1(cid:1)⊥ and constructs for Nλ the Jordan basis ∩ (im (X − λ))⊥ Since n(X − λ)j br,k(cid:12)(cid:12) 1 ≤ r ≤ rmax, 0 ≤ j ≤ r − 1, k = 1, . . . , mro T (X − λ)j br,k = (X − λ)j T br,k we get the desired structure if the subspaces (3.1) are T -invariant. This follows from the next two lemmas and the equality (im (X − λ))⊥ = ker(cid:0)X ∗ − λ(cid:1) . We can now assemble the bases of the various Nλ to get a Jordan basis built from the Kramers pairs. (cid:3) Lemma 3.3. If X ♯ = X ∗ then ker(X) is T -invariant. Proof. This is a restatement of the λ = 0 case of Lemma 2.3(2). Lemma 3.4. If a subspace H0 is T -invariant then (H0)⊥ is also T - (cid:3) invariant. Proof. If v is in (H0)⊥ then for every w ∈ H0 we have hT v, wi = −hT v,T T wi = −hv,T wi = 0. (cid:3) Corollary 3.5. Suppose X ∈ Mn(H). There is an invertible matrix S ∈ Mn(H) and a complex matrix J in Jordan Form such that X = S−1JS. FACTORIZATION OF MATRICES OF QUATERNIONS 12 4. Norms There are two operator norms to consider on X in MN (H), that induced by quaternionic Hilbert space and that induced by complex Hilbert space on χ (X). They end up identical. Theorem 4.1. Suppose X is in MN (H). Then using the norms on HN and on C2N we have kvk = N Xj=1 kwk = 2N Xj=1 kXvk kvk = sup w6=0 sup v6=0 v∗ j vj! 1 2 1 2 vjvj! kχ (X) wk kwk . Proof. Utilizing also the norm on C2N we calculate the four relevant norms: The result follows. 5. Singular value decomposition As in the complex case, a slick way to prove there is a singular value decomposition is to work out the polar decomposition and then use the spectral theorem on the positive part. In [13] it is stated that "a little more work is needed for the sin- gular case" when discussing the polar decomposition. The extra work involves padding out a quaternionic partial isometry to be a quater- nionic unitary (so symplectic unitary.) We remind the reader that U is a partial isometry when (U ∗U)2 = U ∗U, or equivalently U U ∗U = U or U ∗U U ∗ = U ∗ or (U U ∗)2 = U U ∗. 2 2 2 (cid:13)(cid:13)(cid:13)(cid:13) w (cid:21)(cid:13)(cid:13)(cid:13)(cid:13) (cid:20) v = kvk2 + kwk2 . v − jw(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13) = kvk2 + kwk2 . w (cid:21)(cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13) −B A (cid:21)(cid:20) v (cid:20) A B = kAv + Bwk2 +(cid:13)(cid:13)Aw − Bv(cid:13)(cid:13) Av + Bw + j(cid:0)Bv − Aw(cid:1)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:16)A + Bj(cid:17)(cid:16)v − jw(cid:17)(cid:13)(cid:13)(cid:13) =(cid:13)(cid:13)(cid:13) = kAv + Bwk2 +(cid:13)(cid:13)Bv − Aw(cid:13)(cid:13) 2 2 . 2 2 . (cid:3) FACTORIZATION OF MATRICES OF QUATERNIONS 13 If we restrict the domain and range of U we find it is an isometry from (ker(U))⊥ to (ker(U ∗))⊥. Lemma 5.1. Suppose U ∗ = U ♯and U is a partial isometry in M2N (C). There is a symplectic unitary W in MN (C) so that W ξ = U ξ for all ξ ⊥ ker(U). Proof. If v is in ker(U) then by Lemma 2.3, T v is also in ker(U). Since v and T v are orthogonal, we can show that ker(U) has even dimension 2m and that is has a basis for the form v1, . . . , vm,T v1, . . . ,T vm. We are working in finite dimensions so the dimension of ker(U ∗) is also 2m and we select for it a basis w1, . . . , wm,T w1, . . . ,T wm. We can define W to agree with U on (ker(U))⊥ and to send vj to wj and T vj to T wj and so get a unitary that commutes with T , which means it is symplectic. (cid:3) Lemma 5.2. Suppose X ∗ = X ♯ in M2N (C). Then there is a unitary U and a positive semidefinite P with U ∗ = U ♯and P ∗ = P ♯ and X = U P . Proof. Let f be a continuous function on the positive reals with f (0) = 0 and f (λ) = λ− 1 2 for every nonzero eigenvalue of X ∗X. Then let W = Xf (X ∗X). The usual calculations in functional calculus tell us W is a partial isom- etry and that X = W P for P = (X ∗X) 2 . Working with monomials, polynomials and then taking limits, we can show 1 for any positive operator Y and so (f (Y ))♯ = f(cid:0)Y ♯(cid:1) Also P ♯ = P = P ∗. W ♯ = f(cid:0)X ♯X ∗♯(cid:1) X ♯ = f (X ∗X) X ∗ = W ∗. We just showed that the matrices in the minimal polar decomposition (cid:3) are quaternionic. We use Lemma 5.1 to finish the argument. As expected, the polar decomposition leads to a singular value de- composition. Theorem 5.3. Suppose X ∗ = X ♯ in M2N (C). There are symplectic unitary matrices U and V and a diagonal matrix D with nonnegative real entries and D♯ = D∗ so that X = U DV . FACTORIZATION OF MATRICES OF QUATERNIONS 14 Proof. We take a quaternionic polar decomposition X = W P . Since P is positive, we apply Theorem 2.4 to get symplectic unitary matrices Q and V and diagonal matrix D so that P = QDV . The eigenvalues of P are nonnegative, so the same is true for the diagonal elements of D and we have the needed factorization X = (W Q)DV . (cid:3) 6. QR factorization It is easy to use Lemma 2.2 to get over H a QR factorization theorem. Notice that upper triangular matrices are sent by χ to matrices that are block upper-triangular. Theorem 6.1. If X ∗ = X ♯ in M2N (C) the there is a symplectic unitary Q and R of the form R =(cid:20) A B −B A (cid:21) with A and B upper triangular. If X ∈ MN (H) then there is a untiary Q and upper triangular matrix R in MN (H) so that X = QR. Proof. (2) follows directly from (1), so we prove (1). We apply Lemma 2.2 to the first column of X and find X = Q1R1 where R1 =(cid:20) A1 B1 −B1 A1 (cid:21) where A1 and B1 have zeros in their first columns, except perhaps in the top position. As we did earlier, we can proceed with a proof by induction. (cid:3) 7. Self-dual matrices If we study matrices with X ♯ = X then we are no longer working directly with quaternionic matrices, but as we discuss below, there is a connection. We discuss a Schur factorization and a structured polar decomposition for self-dual matrices. The latter is a bit tricky, so we warm up with a structured polar decomposition for symmetric complex matrices. For dealing with a single self-dual matrix, there is the efficient Paige / Van Loan algorithm [5, 10] to implement the following theorem. Theorem 7.1. Given X1, . . . , Xk in M2N (C) that commute pairwise and are self-dual, there is a single symplectic unitary U so that for all j, with Tj upper-triangular and the Cj skew-symmetric. U ∗XjU =(cid:20) Tj Cj 0 T T j (cid:21) . FACTORIZATION OF MATRICES OF QUATERNIONS 15 Proof. Let v be a nonzero unit vector so that Xjv = λjv for all j. By Lemma 2.3, X ∗ j (T v) = λj (T v) . There is a symplectic unitary U1 so that U1e1 = v and U1eN +1 = T v. Let Yj = U ∗ 1 XjU1. Then and Yje1 = λje1 Y ∗ j eN +1 = λjeN +1. Since eN +1 is real, we take adjoint and discover N +1Yj = λjeT eT N +1. This means the column 1 and row N + 1 are all but zeroed-out, .   ∗ ∗ ∗ λj ∗ Cj 0 Aj 0 0 0 λj ∗ Dj 0 Bj Yj =  Zj =(cid:20) Aj Cj Bj Dj (cid:21) Basic facts about block triangular matrices show that the are a commuting family of matrices, and since U1 was chosen to be symplectic, the Zj will be self-dual. As simple induction now finishes the proof. (cid:3) A promising numerical technique for the joint diagonalization of two commuting self-dual self-adjoint matrices H and K would be to form the normal self-dual matrix X = A + iB and apply Paige / Van Loan to reduce to block diagonal form. Then apply ordinary Schur decom- position. This technique was used in [5, §9] to diagonalize matrices that were exactly self-dual and approximately unitary. This idea is mentioned in [2], section 6.5. It is not hard to show that the minimal polar decomposition, the one that is unique and can involve a partial isometry, preserves in some way just about any symmetry thrown at it. This is because the functional calculus interacts well with the dual operation [8], as well as with the transpose. It is a bit harder to figure what happens for the maximal polar decomposition. By the minimal polar decomposition is meant the factorization that is unique and can involve a partial isometry. By the maximal polar decomposition is meant the factorization that involves a unitary. FACTORIZATION OF MATRICES OF QUATERNIONS 16 We begin with the easier result about the polar decomposition of complex symmetric matrices. Theorem 7.2. If X in M2n(C) satisfies X T = X then there is a unitary U so that U T = U and X = U X. Proof. Again chose f with f (0) = 0 and f (λ) = λ− 1 eigenvalue of X ∗X and let 2 for every nonzero W = Xf (X ∗X). As always, W is a partial isometry and X = W P for P = (X ∗X) X. Now we discover 1 2 = W T = f(cid:0)X TX(cid:1) X T = f (XX ∗) X = Xf (X ∗X) = W. To create a unitary U with X = U X we must extend W to map ker(W ) to ker (W ∗). We can arrange U T = U as follows. Let v1, . . . , vm be an orthonormal basis of ker(W ). Then v1, . . . , vm will be an or- thonormal basis of ker(W ∗) = ker(W ) and we define V to be zero on (ker(W ))⊥ and V vj = vj. Thus V ∗ will be zero on (ker(W ))⊥ and V ∗vj = vj, but the same can be said about V . That means V T = V and so U = W + V will be the required symmetric unitary. Notice there is no structure on X, but considered with X ∗ we get the formula (7.1) X ∗ = XT . (cid:3) For the self-dual situation, we shall see that a similar construction works so long as we respect Kramers degeneracy. Proposition 7.3. If a partial isometry W in M2N (C) is self-dual, then the initial space of W will have even dimension and T will map the initial space isometrically onto the final space of W . Proof. Applying Lemma 2.3 to the self-adjoint matrix W ∗W we find W ∗W v = v =⇒ (W ∗W )♯ T v = T v =⇒ (W W ∗)T v = T v FACTORIZATION OF MATRICES OF QUATERNIONS 17 That is, when v ∈ (ker W )⊥ we have T v ∈ (ker W ∗)⊥ and kW ∗T vk = kvk. This is useful because hv, W ∗T vi = hv,−W ∗Zvi =(cid:10)−W TZv, v(cid:11) = h−ZW v, vi = hv, W ∗Zvi = −hv, W ∗T vi which means v and W ∗T v are orthogonal, and W ∗T W ∗T v = W ∗ZW ∗Zv = W ∗ZW TZv = −W ∗W ♯v = −W ∗W v = −v. If we start with a unit vector v in (ker W )⊥ then we end up with an orthogonal pair of unit vectors v and w with W ∗T v = w and W ∗T w = −v. If q is a vector in (ker W )⊥ that is orthogonal to both v and w then W ∗T q will be orthogonal to both W ∗T v and W ∗T w since T preserves orthogonality everywhere and W ∗ preserves the orthogonality of vectors in (ker W ∗)⊥ . Thus we can create a basis of (ker W )⊥ out of pairs v1, W ∗T v1, . . . , vm, W ∗T vm. (cid:3) We need some examples of self-dual partial isometries. Treating vec- tors as 2N-by-1 matrices, if we set then this rank two (at most) matrix is self-dual since V = (T v) w∗ − (T w) v∗ V ♯ = −Z (−Zvw∗ + Zwv∗)T Z = −Z (wv∗Z − vw∗Z) Z = Zwv∗ − Zvw∗ = −T (w)v∗ + T (v)w∗ = V. FACTORIZATION OF MATRICES OF QUATERNIONS 18 If we start with v and w orthogonal, then V will be the rank-two partial isometry taking v to T w and w to T v. We record this as a lemma that avoids the ugly notation. Lemma 7.4. Suppose v and w are orthogonal unit vectors. Then the partial isometry from Cv + Cw to CT v + CT w that sends v to T w and w to −T v will be self-dual. Theorem 7.5. If X in M2n(C) satisfies X ♯ = X then there is a unitary U so that U ♯ = U and X = U X. Proof. Once more W = Xf (X ∗X) works to create a partial isometry W with X = W P for P = X and this time we have W ♯ = W . We need a partial isometry from ker(W ) to ker (W ∗) that is self-dual. The dimension of ker(W ) will be even, by Lemma 7.3. Moreover if is an orthogonal basis for ker(W ) then v1, w1, . . . , vm, wm T v1,T w1, . . . ,T vm,T wm will be an orthogonal basis for ker (W ∗). The needed self-dual partial isometry V will send vj to T wj and wj to T vj and the self-dual unitary we use will be U = W + V . (cid:3) 8. The odd particle causes even degeneracy We hope not to scare the mathematical reader with more discus- sion of Kramers degeneracy. What Kramers discovered was that for a certain systems involving a odd number of electrons, the Hamiltonian always had all eigenvalues with even multiplicity. A mathematical manifestation of this is that the tensor product of two dual operations is the transpose operation in disguise. In contrast to that, the tensor product of three dual operations is a large dual op- eration in disguise. Specifically if we implement an orthogonal change of basis, the dual operation becomes the transpose. This is essentially the same as the facts that is more familiar to mathematicians, that H ⊗R H ∼= M4(R) and H ⊗R H ⊗R H ∼= M4(H). In the following, we let ZN be as in (1.4), the matrix that specified the dual operation. Lemma 8.1. Consider U = 1 √2 (I ⊗ I − iZN ⊗ ZM ) . For all X ∈ M2N (C) and Y ∈ M2M (C), U ∗(cid:0)X ♯ ⊗ Y ♯(cid:1) U = (U ∗ (X ⊗ Y ) U)T . FACTORIZATION OF MATRICES OF QUATERNIONS 19 Proof. Since Z T K = −ZK we see U = U T. Also 1 2 1 U ∗U = = (I ⊗ I + iZN ⊗ ZM ) (I ⊗ I − iZN ⊗ ZM ) 2(cid:0)I ⊗ I + (ZN ⊗ ZM )2(cid:1) = I ⊗ I so U is a unitary and U = U −1. Since U 2 = −iZN ⊗ ZM we find U ∗(cid:0)X ♯ ⊗ Y ♯(cid:1) U = U(cid:0)X ♯ ⊗ Y ♯(cid:1) U = U (−iZN ⊗ ZM )(cid:0)X T ⊗ Y T(cid:1) (iZN ⊗ ZM ) U = U U 2(cid:0)X T ⊗ Y T(cid:1) U = (U ∗ (X ⊗ Y ) U)T . U 2 (cid:3) We have refrained from discussing C ∗-algebras, but here they add clarity. What the lemma is showing is (M2N (C) ⊗ M2M (C), ♯ ⊗ ♯) ∼=(cid:0)M2(M +N )(C), T(cid:1) . Lemma 8.2. For all X ∈ M2N (C) and Y ∈ M2M (C), X T ⊗ Y ♯ = (X ⊗ Y )♯ where the ♯ on the right is taken with respect to I ⊗ ZM . Proof. This is much simpler, as (X ⊗ Y )♯ = − (I ⊗ ZM ) (X ⊗ Y )T (I ⊗ ZM ) = − (I ⊗ ZM )(cid:0)X T ⊗ Y T(cid:1) (I ⊗ ZM ) = X T ⊗(cid:0)−ZM Y TZM(cid:1) . (cid:0)M2Nj (C), ♯(cid:1) leads to something isomorphic to (cid:0)M2(N1+N2+N3)(C), ♯(cid:1) . 9. Acknowledgements Together, these lemmas show us that a tensor product of three (cid:3) The author gratefully aknowleges the guidance and assistence re- ceived from mathematicians and physicists, in particular Vageli Cout- sias, Matthew Hastings and Adam Sørensen. This work was partially supported by a grant from the Simons Foun- dation (208723 to Loring), and by the Efroymson fund at the University of New Mexico. FACTORIZATION OF MATRICES OF QUATERNIONS 20 References [1] A. Bunse-Gerstner, R. Byers, and V. Mehrmann, A quaternion QR algorithm, Numerische Mathematik, 55 (1989), pp. 83 -- 95. [2] A. Bunse-Gerstner, R. Byers, and V. Mehrmann, Numerical meth- ods for simultaneous diagonalization, SIAM J. Matrix Anal. Appl., 14 (1993), pp. 927 -- 949. [3] G. K. Pedersen, Factorization in C ∗-algebras, Exposition. Math., 16 (1998), pp. 145 -- 156. [4] D. R. Farenick and B. A. F. Pidkowich, The spectral theorem in quater- nions, Linear Algebra Appl., 371 (2003), pp. 75 -- 102. [5] M. B. Hastings and T. A. Loring, Topological insulators and C ∗-algebras: Theory and numerical practice, Ann. Physics, 326 (2011), pp. 1699 -- 1759. [6] H. Kramers, Th´eorie g´en´erale de la rotation paramagn´etique dans les cristaux, Proc. Acad. Amst, 33 (1930), pp. 959 -- 972. [7] H. C. Lee, Eigenvalues and canonical forms of matrices with quaternion co- efficients, Proc. Roy. Irish Acad. Sect. A., 52 (1949), pp. 253 -- 260. [8] T. A. Loring and A. Sørensen, Almost commuting self-adjoint matrices -- the real and self-dual cases. arXiv:1012.3494. [9] M. Mehta, Random matrices, Academic press, 2004. [10] C. Paige and C. Van Loan, A Schur decomposition for Hamiltonian matri- ces, Linear Algebra Appl., 41 (1981), pp. 11 -- 32. [11] N. Wiegmann, Some theorems on matrices with real quaternion elements, Canad. J. Math, 7 (1955), pp. 191 -- 201. [12] E. Wigner, Uber die operation der zeitumkehr in der quantenmechanik, Nach. Ges. Wiss. Gdtt, 32 (1932), pp. 546 -- 559. [13] F. Zhang, Quaternions and matrices of quaternions, Linear Algebra Appl., 251 (1997), pp. 21 -- 57. [14] F. Zhang and Y. Wei, Jordan canonical form of a partitioned complex matrix and its application to real quaternion matrices, Communications in Algebra, 29 (2001), pp. 2363 -- 2375. University of New Mexico, Department of Mathematics and Statis- tics, Albuquerque, New Mexico, 87131, USA
1604.02173	4	1604	2017-09-02T13:27:27	Compact Group Actions on Topological and Noncommutative Joins	[ "math.OA", "math.AT" ]	We consider the Type 1 and Type 2 noncommutative Borsuk-Ulam conjectures of Baum, D$\k{a}$browski, and Hajac: there are no equivariant morphisms $A \to A \circledast_\delta H$ or $H \to A \circledast_\delta H$, respectively, when $H$ is a nontrivial compact quantum group acting freely on a unital $C^*$-algebra $A$. Here $A \circledast_\delta H$ denotes the equivariant noncommutative join of $A$ and $H$; this join procedure is a modification of the topological join that allows a free action of $H$ on $A$ to produce a free action of $H$ on $A \circledast_\delta H$. For the classical case $H = \mathcal{C}(G)$, $G$ a compact group, we present a reduction of the Type 1 conjecture and counterexamples to the Type 2 conjecture. We also present some examples and conditions under which the Type 2 conjecture does hold.	math.OA	math	Compact Group Actions on Topological and Noncommutative Joins Alexandru Chirvasitu∗, Benjamin Passer† Abstract We consider the Type 1 and Type 2 noncommutative Borsuk-Ulam conjectures of Baum, abrowski, and Hajac: there are no equivariant morphisms A → A ⊛δ H or H → A ⊛δ H, D ' respectively, when H is a nontrivial compact quantum group acting freely on a unital C ∗- algebra A. Here A ⊛δ H denotes the equivariant noncommutative join of A and H; this join procedure is a modification of the topological join that allows a free action of H on A to produce a free action of H on A ⊛δ H. For the classical case H = C(G), G a compact group, we present a reduction of the Type 1 conjecture and counterexamples to the Type 2 conjecture. We also present some examples and conditions under which the Type 2 conjecture does hold. Key words: join, compact group, Borsuk-Ulam theorem, compact quantum group, noncommutative topology MSC 2010: 20G42, 22C05, 46L85, 55S40 1 Introduction The join of two topological spaces X and Y is a quotient X ∗ Y = X × Y × [0, 1]/ ∼, where the equivalence relation makes identifications at the endpoints of [0, 1]. Namely, if x0 ∈ X is fixed, then all points (x0, y, 1) are identified, and if y0 ∈ Y is fixed, all points (x, y0, 0) are identified. In [3], the authors conjecture that the topological join, and a C ∗-algebraic variant thereof, may be used to greatly generalize the Borsuk-Ulam theorem. Their topological conjecture is as follows. Conjecture 1.1 Suppose G is a nontrivial compact group acting freely and continuously on a compact Hausdorff space X. Then there is no equivariant, continuous map from X ∗ G to X, where X ∗ G is equipped with the diagonal action [(x, g, t)] · h = [(x · h, gh, t)]. The conjecture generalizes the Borsuk-Ulam theorem in that if G = Z/2, X = Sk, and the group action is given by the antipodal map x 7→ −x, then the conjecture reads as the Borsuk-Ulam theorem itself: there is no odd function from Sk+1 to Sk. Of particular interest is the fact that the sphere Sk is an iterated join of k + 1 copies of Z/2, and the antipodal action is compatible with this join process. As remarked in [3], Conjecture 1.1 holds when X = G ∗ G due to non-contractibility of G. Further, [20, Corollary 3.1] shows that Conjecture 1.1 holds when G has a nontrivial torsion element (see also [16, Proposition 4.1] for a description of this proof). ∗University at Buffalo, [email protected] Partial support from NSF grant DMS 1565226 and the 2016 Simons Semester in Noncommutative Geometry through the Simons-Foundation grant 346300 and the Polish Gov- ernment MNiSW 2015-2019 matching fund. †Technion-Israel Institute of Technology, [email protected] Partial support from NSF grants DMS 1300280 and DMS 1363250, a Zuckerman Fellowship at the Technion, EU grant H2020-MSCA-RISE-2015-691246- QUANTUM DYNAMICS, and the 2016 Simons Semester in Noncommutative Geometry through the Simons- Foundation grant 346300 and the Polish Government MNiSW 2015-2019 matching fund. 1 If A and B are unital C ∗-algebras, the "classical" noncommutative join A ⊛ B = {f ∈ C([0, 1], A ⊗ B) f (0) ∈ C ⊗ B, f (1) ∈ C ⊗ A} (1) of [7] and [4] directly generalizes the join X ∗ Y of compact Hausdorff spaces. Here we have used ⊗ to refer to the minimal tensor product, as we do in the rest of the manuscript. However, if (H, ∆) is a compact quantum group with a free coaction δ : A → A ⊗ H, questions about this coaction are more suited to the equivariant join A ⊛δ H = {f ∈ C([0, 1], A ⊗ H) f (0) ∈ C ⊗ H, f (1) ∈ δ(A)}, (2) which admits a free coaction δ∆ generated by id ⊗ ∆ : C([0, 1], A) ⊗ H → C([0, 1], A) ⊗ H ⊗ H (see [3, Theorem 1.5]). Conjecture 2.3 of [3], which we repeat here, generalizes Conjecture 1.1 to the quantum setting. Conjecture 1.2 Suppose A is a unital C ∗-algebra with a free coaction of a nontrivial compact quantum group (H, ∆). Type 1. There does not exist a (δ, δ∆)-equivariant ∗-homomorphism A → A ⊛δ H. Type 2. There does not exist a (∆, δ∆)-equivariant ∗-homomorphism H → A ⊛δ H. Note that a coaction δ : A → A ⊗ H of (H, ∆) is free when it satisfies the following condition from [10], which is written succinctly as [3, (1.10)]: (Xfinite (ai ⊗ 1)δ(bi) : ai, bi ∈ A) = A ⊗ H. (3) Further, if A and B are C ∗-algebras with coactions δA, δB of (H, ∆), then a morphism φ : A → B is equivariant if A φ δA B δB B ⊗ H A ⊗ H φ⊗id (4) commutes. As a consequence of [17, Corollary 2.4], Conjecture 1.2 Type 1 holds when H = C(G) for G a compact group with a nontrivial torsion element. Specifically, [17, Corollary 2.4] is a reduction to the topological case in [20], and some alternative (but still very much related) proof strategies are described in the remainder of [17]. No counterexamples to Type 1 in its full generality are known, but in Section 2, we show that Type 2 counterexamples exist for the "quantum" group C(S1) acting on certain noncommutative C ∗-algebras, which can even be separable and nuclear. On the other hand, there are also some restrictive conditions, applicable in both the classical group and quantum group settings, under which the Type 2 conjecture holds. In Section 3, we deal exclusively with the classical case H = C(G) and present a reduction of the Type 1 conjecture in this setting. We also consider related questions regarding eigenfunctions in C(Z∗n p ), including the limit as n → ∞, based upon questions in [17] that attempt to use eigenfunctions to generalize a strategy in [20]. 2 Type 2: Counterexamples and Special Cases In this section, we address Conjecture 1.2 Type 2. For a coaction δ : A → A ⊗ H and a simple (hence finite-dimensional) H-comodule ρ : V → V ⊗ H, we let Aρ denote the ρ-isotypic subspace of A: Aρ = sum of the images of all H-equivariant maps V → A. 2 The sum of all Aρ (as ρ ranges over all simple H-comodules) is a dense ∗-subalgebra of A, referred to in [3, §1.2] as the Peter-Weyl subalgebra PH(A) of A. It can be recast, as in [3], as the preimage through ρ of A ⊗alg O(H), where the latter symbol denotes the unique dense Hopf ∗-subalgebra of H. Note that we have O(H) =Mρ Hρ, where H coacts on itself on the right via the comultiplication ∆ : H → H ⊗ H. Equivalent formulations of freeness were studied by the authors of [4]; below we present a special case of their results, with a proof for completeness. Proposition 2.1 (special case of [4, Theorem 0.4]) A coaction δ : A → A ⊗ H is free if and only if it is saturated, i.e. for every simple H-comodule ρ, the unit 1 ∈ A belongs to A∗ ρAρ. Proof (⇒) As in [4, Definition 0.1], freeness means that the linear span of elements of the form (x ⊗ 1)δ(y), x, y ∈ A is dense in A ⊗ H. It suffices to let x, y range over the dense subalgebra PH(A) =Lρ Aρ instead, and since we have δ(PH (A)) ⊂Lρ Aρ ⊗ Hρ, it follows that the linear span (5) A∗ ηAρ ⊗ Hρ Xη,ρ is dense in A ⊗ H. Since the subspaces Hρ and Hρ′ of H are orthogonal with respect to the Haar state h : H → C when ρ 6= ρ′, any element in A ⊗ Hρ can be approximated arbitrarily well by elements in A∗ ηAρ ⊗ Hρ Xη Pη A∗ alone (i.e., in (5) we fix ρ and allow only η to vary). ηAρ. The conclusion now follows from the fact that for η 6= ρ, the product A∗ In particular, 1 ∈ A is in the closure of ηAρ is contained in the sum of all Aµ, µ 6= 1 (the trivial H-comodule), whereas 1 is in the H-fixed subspace A1 = {a ∈ A δ(a) = a ⊗ 1}. (⇐) Suppose that for every simple H-comodule ρ, the unit 1 ∈ A belongs to A∗ ρAρ. This implies that for every ρ, 1 ⊗ Hρ is in the closure of A∗ ρAρ ⊗ Hρ, and hence A ⊗ Hρ is in the closure of AAρ ⊗ Hρ. The conclusion that the action is free then follows from the observation that for every simple H-comodule ρ, Aρ ⊗ Hρ is contained in δ(Aρ) ⊂ δ(A). (cid:4) When (H, ∆) is the Hopf C ∗-algebra C(G) of a compact group G with comultiplication dual to the multiplication of G, see also [18, Definition 5.2, Theorem 5.10]. We will henceforth focus almost exclusively on the case H = C(G), where equivariance in the sense of (4) is equivalent to the usual G-equivariance of morphisms. Further, in this setting it is known from [3] that there is no distinction between the classical join A ⊛ C(G) and the equivariant join A ⊛δ C(G). Indeed, applying the map A ⊗ C(G) δ⊗idC(G) A ⊗ C(G) ⊗ C(G) idA⊗mult A ⊗ C(G) (6) pointwise on C([0, 1], A ⊗ C(G)) identifies the subspace A ⊛ C(G) with the subspace A ⊛δ C(G), intertwining the diagonal G-action on the left hand side and the right-tensorand regular action on the right hand side. 3 When a compact group G acts freely on a compact Hausdorff space X, any orbit of X gives an equivariant embedding G ֒→ X. However, there is no analogous phenomenon in the C ∗-algebraic setting: a simple C ∗-algebra A may have a free C(G)-coaction even though A can have no quotient isomorphic to C(G). The embedding G ֒→ X is exactly why Conjecture 1.1 was not split into two types, and we will exploit this difference to produce counterexamples to Conjecture 1.2 Type 2. First, note that when H = C(G) is classical, the conjecture may be rephrased in a way that avoids the join entirely. Lemma 2.2 Let G be a compact group acting on a unital C ∗-algebra A. There is an equivariant unital ∗-homomorphism C(G) → A ⊛ C(G) ∼= A ⊛δ C(G) if and only if both conditions hold: • there is a G-equivariant unital ∗-homomorphism ϕ : C(G) → A, and • ϕ1 = ϕ can be connected to a one-dimensional representation ϕ0 : C(G) → C ⊂ A through a path ϕt, t ∈ [0, 1] in Hom(C(G), A). Further, the first condition guarantees that the associated coaction δ : A → A ⊗ C(G) is free. Proof If there is an equivariant map ψ : C(G) → A ⊛ C(G), then evaluation at any t ∈ [0, 1] produces equivariant maps ψt : C(G) → A ⊗ C(G) ∼= C(G, A). From the boundary conditions of the (classical) join, ψ1 maps to constant A-valued functions, and ψ0 maps to C(G). The equivariant map ϕ = eve ◦ψ1 is then connected within Hom(C(G), A) via eve ◦ψt to a one-dimensional representation, eve ◦ ψ0. If instead we assume that the conditions hold, we consider the two conclusions separately. 1: Freeness. The comultiplication ∆ on C(G) defines a free coaction of C(G) on itself, as the closure of finite sums in (3) is actually a closed ∗-subalgebra of C(G) ⊗ C(G) ∼= C(G × G) on which the Stone-Weierstrass theorem applies (see [4]). Moreover, we have a unital ∗-homomorphism φ : C(G) → A that is G-equivariant, so it satisfies the coaction equivariance identity (φ⊗id)◦∆ = δ ◦φ. Fix ε > 0, a ∈ A \ {0}, f ∈ C(G), and finitely many elements hi, ki ∈ C(G), 1 ≤ i ≤ m, so that ε (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (hi ⊗ 1)∆(ki) − 1 ⊗ f(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) mXi=1 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (aφ(hi) ⊗ 1)δ(φ(ki)) − a ⊗ f(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) =⇒ (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (φ(hi) ⊗ 1)δ(φ(ki)) − 1 ⊗ f(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) mXi=1 mXi=1 a < < ε a (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) by freeness of the comultiplication ∆. Applying φ ⊗ id and then left multiplying by a ⊗ 1 yields < ε, so the closure in (3) includes any a ⊗ f . The closed span of such elements is then all of A ⊗ C(G). 2: Existence of map into the join. Recasting the join in an equivalent form A ⊛ C(G) ∼= {f ∈ C([0, 1] × G, A) f {0}×G takes values in C ⊆ A, f {1}×G is constant on G}, (7) we can define ψ : C(G) → A ⊛ C(G) so that it sends f ∈ C(G) to the function ψ(f ) : [0, 1] × G → A, ψ(f )(t, g) = g−1 ⊲ ϕt(f (•g)), where f (•g) is the function x 7→ f (xg) and ⊲ denotes the group action. It is now immediate to check that 4 • ψ(f ){0}×G is in C(G) ⊂ A ⊗ C(G), as ϕ0 was assumed to be a one-dimensional representation; • ψ(f ){1}×G is constant on G and hence represents a single element of A ⊂ A ⊗ C(G), as ϕ = ϕ1 : C(G) → A was assumed to be equivariant; • ψ is G-equivariant with respect to the G-action on C([0, 1] × G, A) defined by (h ⊲ ζ)(t, g) = h ⊲ ζ(t, gh). In other words, ψ is the desired equivariant map C(G) → A ⊛ C(G). (cid:4) Equipped with Lemma 2.2, we can find counterexamples to the Type 2 conjecture. Theorem 2.3 Let H be an infinite-dimensional, separable Hilbert space and set A = B(H)⊗C(S1). If ∆ : C(S1) → C(S1) ⊗ C(S1) denotes the comultiplication, then define a coaction δ : A → A ⊗ C(S1) by δ = id ⊗ ∆. This coaction is free, and there is an equivariant, unital ∗-homomorphism ψ : C(S1) → A ⊛δ C(S1), so Conjecture 1.2 Type 2 fails. Proof First, there is an equivariant map ϕ = ϕ1 : C(S1) → A defined by φ(f ) = 1 ⊗ f . Because C(S1) is the universal C ∗-algebra generated by a single unitary character χ : z 7→ z, the map ϕ1 is determined by ϕ1(χ) = 1⊗χ. The unitary group U (H) with the norm topology is simply connected by Kuiper's theorem ([14]), so the unitary group U (A) ∼= U (C(S1, B(H))) ∼= C(S1, U (H)) is path connected. Let ut, t ∈ [0, 1] be a continuous path of unitaries in A connecting u1 = I ⊗ χ to u0 = I ⊗ 1, and define a continuous path of unital ∗-homomorphisms ϕt : C(S1) → A by ϕt(χ) = ut. Because ϕ0 is just a one-dimensional representation, we may apply Lemma 2.2, guaranteeing freeness of the coaction on A and existence of an equivariant map C(S1) → A ⊛δ C(S1). (cid:4) To see the result of the above proof more explicitly, note that the final equivariant map ψ : C(S1) → A ⊛δ C(S1) = (B(H) ⊗ C(S1)) ⊛δ C(S1) is determined by so its general form is given by ψ(χ)[t] = ut ⊗ χ, t ∈ [0, 1], ψ(f )[t] = f (ut ⊗ χ), t ∈ [0, 1], where f is applied using the continuous functional calculus. As before, χ ∈ C(S1) is the generator χ(z) = z, u1 = I ⊗ χ, u0 = I ⊗ 1, and ut is a continuous path of unitaries. The boundary conditions of the equivariant join are satisfied because ψ(f )[0] = f ((I ⊗ 1) ⊗ χ) = (I ⊗ 1) ⊗ f ∈ C ⊗ C(S1) and ψ(f )[1] = f ((I ⊗ χ) ⊗ χ) = f (δ(I ⊗ χ)) ∈ C ∗(δ(A)) = δ(A). We have used the fact that δ is a unital ∗-homomorphism. If δ∆ denotes the coaction on A ⊛δ C(S1) given by applying id ⊗ ∆ : A ⊗ C(S1) → A ⊗ C(S1) ⊗ C(S1) at each t ∈ [0, 1], then ψ is (∆, δ∆) equivariant. Indeed, the equations ∆(χ) = χ ⊗ χ and (id ⊗ ∆)(ψ(χ)[t]) = (id ⊗ ∆)(ut ⊗ χ) = ut ⊗ χ ⊗ χ = ψ(χ)[t] ⊗ χ show that the unital ∗-homomorphisms δ∆ ◦ ψ and (ψ ⊗ id) ◦ ∆ agree on χ, the generator of the domain C(S1). Theorem 2.3 leads to a wider class of counterexamples to Conjecture 1.2 Type 2. First, we make the following easy observation. 5 and the equivariant map C(Gi) C(G) ∼=OI ϕ =OI ϕi : C(G) → A. Lemma 2.4 If Conjecture 1.2 Type 2 fails for the compact groups Gi, i ∈ I, then it fails for the product G =QI Gi. Proof According to Lemma 2.2, the hypothesis ensures the existence of free actions of Gi on C ∗-algebras Ai together with equivariant morphisms ϕi : C(Gi) → Ai that are contractible to one-dimensional representations of C(Gi), respectively. Now we can apply Lemma 2.2 again in the opposite direction, using the C ∗-algebra A =NI Ai (universal tensor product) with the obvious tensor product coaction by The contractibility of ϕ follows from that of the individual ϕi. Failure of the conjecture for tori is then immediate. Corollary 2.5 Conjecture 1.2 Type 2 fails for tori TI . Proof This follows from applying Theorem 2.3 and Lemma 2.4 to S1. (cid:4) (cid:4) The counterexample presented in Theorem 2.3 uses a C ∗-algebra which is highly non-nuclear, but this can be avoided. Theorem 2.6 There is a counterexample to Conjecture 1.2 Type 2 for which H = C(S1) and A is a unital, separable, nuclear C ∗-algebra. nLi=1 Proof Let B be a unital, separable, nuclear C ∗-algebra with K0(B) ∼= {0} ∼= K1(B), such as the Cuntz Algebra O2 [6, Theorems 3.7 and 3.8]. Since the K-groups of B are trivially torsion free and the commutative C ∗-algebra C(S1) is certainly in the bootstrap class, the Kunneth formula [5, Theorem 23.1.3] shows that K1(B ⊗ C(S1)) ∼= (K0(B) ⊗ K1(C(S1))) ⊕ (K1(B) ⊗ K0(C(S1))) ∼= {0}. Let χ ∈ C(S1) denote the standard generating character and fix n such that there is a path of unitaries in Mn(B ⊗ C(S1)) connecting (1 ⊗ χ) ⊕ In−1 to In. Since the matrix unitary group Un(C) is path connected, it follows that I1 ⊕ (1 ⊗ χ) ⊕ In−2, I2 ⊕ (1 ⊗ χ) ⊕ In−3, . . . , In−1 ⊕ (1 ⊗ χ) are also connected to the identity, as is their product 1 ⊗ χ. That is, in the separable, unital, nuclear C ∗-algebra A := B ⊗ C(S1) ⊗ Mn(C), v1 := 1 ⊗ χ ⊗ In is connected via a path of unitaries vt to the identity element v0. Let S1 act on A via rotation in the C(S1) tensorand, so that v1 is a χ-eigenvector for this action. Then φt : χ ∈ C(S1) 7→ vt ∈ A defines a continuous path of morphisms. Since χ and v1 are χ-eigenvectors in C(S1) and A, re- spectively, φ1 is equivariant. Further, φ0 is a 1-dimensional representation since v0 is the identity element. Therefore, the conditions of Lemma 2.2 are satisfied, and the counterexample follows. (cid:4) Despite the above counterexamples, there are some circumstances under which Type 2 of Conjecture 1.2 holds. See for example [11, Corollary 2.7], in which the authors show that Type 2 holds for the compact quantum group C(SUq(2)) acting on its iterated joins, based on a general 6 result about finite-dimensional representations. In a different vein, the next theorem is dual to the following topological argument: an equivariant map X ∗ G → G is automatically surjective, and X ∗ G is connected, so such a map cannot exist if G is disconnected. To adapt this picture to the fully noncommutative setting (when neither A nor H is abelian), note that a compact group G is disconnected precisely when it admits a nontrivial finite quotient G → L. In turn, this corresponds to an embedding C(L) → C(G) of a finite-dimensional Hopf C ∗-algebra into that of G. For this reason, we regard such an embedding in the quantum case as an analogue of disconnectedness. Theorem 2.7 Suppose a compact quantum group H admits an equivariant embedding K ֒→ H of a nontrivial compact quantum group (K, ∆) whose underlying Hopf C ∗-algebra is finite-dimensional. Then Conjecture 1.2 Type 2 holds for any free coaction δ : A → A ⊗ H. Proof Suppose we have an H-equivariant map ψ : H → A ⊛δ H. The image of K must be contained in (C([0, 1], A ⊗ H))ρ (8) Mρ (A ⊛δ H)ρ ⊂Mρ as ρ ranges over the irreducible K-comodules. Since the H-coaction in (8) is the regular one on the H-tensorand, the right hand side of (8) is simply C([0, 1], A ⊗ K). Since K is a finite-dimensional Hopf C ∗-algebra, there exists a counit ε : K → C. Applying ε to the K tensorand of ψ(K) ⊂ C([0, 1], A ⊗ K) yields a path ψt : K → A of C ∗-algebra morphisms such that ψ1 is H-equivariant while ψ0 takes values in C ⊂ A. Now, K is a finite-dimensional C ∗-algebra, and hence it has a C-basis consisting of finite-order unitaries. Moreover, H-equivariance ensures that for some finite-order unitary u ∈ K, the element ψ1(u) ∈ A is not a scalar. But then the spectrum of ψ1(u) is a non-trivial finite subgroup of S1, and the continuity of the spectrum for the norm topology on the unitary subgroup of A implies that ψ1(u) cannot be connected by a path to ψ0(u) ∈ C ⊂ A. We have reached a contradiction, so there can be no equivariant map ψ. (cid:4) Restricting our attention once again to the commutative case H = C(G), we know from Theorem 2.7 that there are no counterexamples to Type 2 for which G has a nontrivial finite quotient group, so G must be connected to produce a counterexample. Further, there are coun- terexamples for all tori G = TI , which of course are path connected, from Corollary 2.5. Question 2.8 If G is a compact, abelian, (path) connected group, must a Type 2 counterexample exist for H = C(G)? A compact abelian group G is connected if and only if its discrete abelian Pontryagin dual Γ = bG is torsion-free ([12, Corollary 7.70]). Pontryagin duality is expressed in the identity C(G) ∼= C ∗(Γ), and for any discrete group Γ, abelian or not, there is a natural way to make C ∗(Γ) into a compact quantum group using the comultiplication ∆ :X aγγ ∈ C ∗(Γ) 7→X aγ γ ⊗ γ ∈ C ∗(Γ) ⊗ C ∗(Γ). We will see below that the analogue of Question 2.8 in the discrete torsion-free (non-abelian) setting can be resolved in the negative. That is, there are compact quantum groups of the form C ∗(Γ) with torsion-free non-abelian Γ for which Conjecture 1.2 Type 2 holds. Moreover, the groups 7 Γ in question will be amenable, so that there will be no ambiguity regarding which C ∗ completion C ∗(Γ) we are considering. Let n ≥ 2 be a positive integer, and equip the torus Tn ∼= Rn/Zn with an automorphism σ regarded simultaneously as an element of SL(n, Z) fixing the lattice Zn ∼= π1(Tn). Throughout the rest of Section 2 we assume that σ is hyperbolic, i.e. its eigenvalues as an element of SL(n, Z) have absolute value not equal to 1. For background on hyperbolic automorphisms on smooth manifolds we refer to [1, Chapter 6]; in reference to the eigenvalue condition on σ, see in particular [1, Definition 6.3 and Exercise 6.2]. Hyperbolicity also implies that σ is expansive in the sense that there is an ε > 0 such that Tn ∋ x 6= y ⇒ supn∈Z d(σn(x), σn(y)) > ε (9) for any distance function d on the torus (see e.g. [1, Definition 5.5]). construct the extension Zn ⋊σ Z. Now, σ induces an automorphism σ on the Pontryagin-dual lattice Zn ∼= cTn, allowing us to Lemma 2.9 If σ as above is hyperbolic, then Zn ⋊σ Z is amenable and torsion-free. Proof First, Zn ⋊σ Z is certainly amenable, as it is an extension of two abelian groups. Writing a generic nonzero element x of Zn ⋊σ Z as yσl for some the kth power of x is (0, 0) 6= (y, l) ∈ Zn × Z, (y + σly + · · · + σl(k−1)y)σlk. This element is nontrivial, as for nonzero l, σl ∈ SL(n, Z) has no root-of-unity eigenvalues. (cid:4) Since Zn ⋊σ Z as above is torsion-free, the compact quantum group C ∗(Zn ⋊σ Z) does not fit within the framework of Theorem 2.7. However, Conjecture 1.2 Type 2 still holds for C ∗(Zn ⋊σ Z). Theorem 2.10 Let Zn⋊σ Z be as in Lemma 2.9. Then Conjecture 1.2 Type 2 holds for the compact quantum group C ∗(Zn ⋊σ Z). Proof Let Γ := Zn ⋊σ Z, and suppose a counterexample does exist for C ∗(Γ), in the form of a free coaction δ : A → A ⊗ C ∗(Γ) and an equivariant morphism ψ : C ∗(Γ) → A ⊛δ C ∗(Γ). Regard ψ as a path ψt, t ∈ [0, 1], as in (2), and consider the C ∗-subalgebra C ∗(Zn) ∼= C(Tn) of C ∗(Γ). Restricting ψt produces a path of equivariant morphisms φt : C(Tn) → A ⊗ C(Tn), where C(Tn) acts on the right tensorand of the codomain. The boundary conditions Ran(φ0) ⊆ C ⊗ C(Tn) and Ran(φ1) ⊆ (A ⊗ C(Tn)) ∩ δ(A) ⊂ B ⊗ C(Tn) are also satisfied, where B is the closed direct sum of the isotypic subspaces of A corresponding to Zn ⊂ Γ. Applying the counit to the right tensorand produces a path of morphisms C(Tn) → A connecting a character C(Tn) → C to an equivariant morphism C(Tn) → B, which must be injec- tive. To show injectivity, we may first factor through the commutative range C ∗-algebra to obtain C(Tn) → C(X) ֒→ B. Then we note that because the original morphism is equivariant, there is a 8 corresponding coaction of C(Tn) on C(X), i.e. an action of Tn on X. Finally, any equivariant con- tinuous map X → Tn is surjective by direct examination of an orbit, so any equivariant morphism C(Tn) → C(X) is injective. The joint spectrum of the images of the n unitary generators of C ∗(Zn) ∼= C(Tn) changes along a path spt, t ∈ [0, 1] (continuous in the Hausdorff topology on closed subsets of Tn) from a singleton sp0 = {p} to sp1 = Tn. Moreover, because all of the homomorphisms are restrictions from C ∗(Γ), all of the spectra spt are σ-invariant. The eigenvalues of σ as an element of SL(n, Z) have absolute values not equal to 1, ensuring via the expansivity condition (9) the existence of some r > 0 such that the orbit of any non-trivial element of the open ball Br(p) (in the standard Euclidean metric on Tn ∼= Rn/Zn) under Z intersects the complement of Br(p). On the other hand, for small t the spectrum spt will be contained in the open ball Br(p), which is a contradiction. (cid:4) 3 Type 1: Reductions and Possible Approaches Corollary 2.4 of [17] implies that Type 1 of Conjecture 1.2 holds when H = C(G) for G a compact group with a nontrivial torsion element. Here we show a reduction of the remaining classical case H = C(G), with some comments on how it might be proved. The problem reduces easily to subgroups of G, so we may certainly assume our compact torsion-free G is abelian, and it is well-known that a copy of Zp = lim←− Z/pn, for some prime p, embeds into G. We include a proof in order to glean some information about the Pontryagin dual. Proposition 3.1 Suppose G is a nontrivial compact, torsion-free, abelian group, and fix any non- trivial character 1 6= τ ∈ bG. Then there is an embedding of Zp into G for which the restriction of τ to Zp is nontrivial. Proof First, we may replace G with the compact subgroup H generated by a single element g with τ (g) nontrivial. Every character on H is uniquely determined by its value at g, and hence the Pontryagin dual bH = Γ can be identified with a subgroup of the discrete circle group S1. Since H is torsion-free, Γ has no nontrivial finite quotients. The Pontryagin dual of Zp is the group Z/p∞ of roots of unity whose orders are powers of p, so it suffices to prove that there is a surjection Γ → Z/p∞ which does not annihilate τ H . Regarded as a discrete abelian group, S1 is the direct sum of one copy of Z/p∞ for each prime p, as well as continuum many copies of Q: (10) S1 ∼=Mp (Z/p∞) ⊕ Q⊕2ℵ0 . Since Γ embeds into S1, τ H has nontrivial image under a map from Γ to one of the summands in (10). There are now two cases to consider. (1) A morphism Γ → Z/p∞ does not annihilate τ H . Since Γ has no nontrivial finite quotients, and there are no proper infinite subgroups of Z/p∞, we obtain the desired surjection Γ → Z/p∞. (2) A morphism Γ → Q does not annihilate τ H . If necessary, we may rescale Q so that τ H is not mapped into Z. We then have a map Γ → Q/Z ∼=Mp (Z/p∞) which does not annihilate τ H . By selecting an appropriate summand, we can now continue as in case (1), completing the proof. (cid:4) 9 An embedding Zp ֒→ G provides a reduction of Conjecture 1.1 and the classical subcase of Conjecture 1.2 Type 1. Lemma 3.2 If Conjecture 1.1 holds for the compact groups Zp for all primes p, then it holds in general. Similarly, if Conjecture 1.2, Type 1 holds for all H = C(Zp), then it holds whenever H = C(G) for some nontrivial compact group G. Proof Let X, G, A, and H = C(G) be as in Conjecture 1.1 and Conjecture 1.2. If G has a nontrivial torsion element, then the conjectures already hold by [20, Corollary 3.1] and [17, Corollary 2.4]. Otherwise, a G-equivariant map X ∗ G → X restricts to a Zp-equivariant map X ∗ Zp → X for some subgroup Zp ≤ G provided by Proposition 3.1. Similarly, a G-equivariant morphism A → A ⊛ C(G) restricts to a G-equivariant morphism A → A ⊛ C(Zp), where we note that the classical join and equivariant join are equivariantly isomorphic because of (6). (cid:4) As in [17, Lemma 2.5], one way to approach the Type 1 conjecture is by iterating any proposed equivariant map A → A ⊛ C(Zp) with its joins A ⊛ C(Zp) → A ⊛ C(Zp) ⊛ C(Zp), etc., producing a chain A → A ⊛ C(Zp) → A ⊛ C(Zp) ⊛ C(Zp) → A ⊛ C(Zp) ⊛ C(Zp) ⊛ C(Zp) → . . . (11) of equivariant maps. Compositions then give A → A ⊛ C(Zp) ⊛ C(Zp) ⊛ · · · ⊛ C(Zp) ∼= A ⊛ C(Z∗n p ) ∀n with equivariant quotients Freeness of a Zp-action on A implies that the saturation condition of Proposition 2.1 (or, as A → C(Z∗n p ) ∀n. (12) eigenspace Aτ , 1 ∈ Aτ A∗ i=1 fig∗ i Zp is abelian, the condition described in [18, §5]) is met; that is, for any character τ ∈ cZp and that Pm τ . Therefore, there is a finite m and a list f1, . . . , fm, g1, . . . , gm ∈ Aτ such is invertible in A. The images of these functions under (12) then show that if an equivariant map A → A ⊛ C(Zp) is assumed, then for this fixed m, and for all n, there is a list of m τ -eigenfunctions in C(Z∗n p ) with no common zeroes. It would suffice to show that for some n much larger than m, this cannot happen. However, unlike in the torsion case, it is not obvious if such a contradiction actually occurs. It is therefore prudent to study the constraints of equivariant maps p → Cm \ {0}, where Zp acts on Cm through a particular character, as n increases. This may be Z∗n recast as a question about actions of finite groups. Lemma 3.3 Suppose G = lim←− Gα is a filtered limit of compact groups, with a matching limit X = lim←− Xα of compact Hausdorff spaces indexed by the same filtered poset. The ordering is such that α ≥ β implies that there exist maps Xα → Xβ and Gα → Gβ. Further, assume that each Gα acts continuously on Xα so that the diagram Gα × Xα Xα Gβ × Xβ Xβ (13) commutes for all α ≥ β. Let E be a finite-dimensional unitary representation of a fixed Gβ. Further, let G and Gα, α ≥ β, also act on E through the quotient maps πβ : G → Gβ and παβ : Gα → Gβ, respectively. Then for any continuous G-equivariant map f : X → E and ε > 0, there exist an α ≥ β and a continuous Gα-equivariant map fα : Xα → E such that kf − fα ◦ παk∞ < ε. 10 Proof Approximation by arbitrary functions Xα → E follows from the Stone-Weierstrass Theorem: the complex-valued, continuous functions on X that factor through some Xα (where the choice of α may depend on the function) form a complex unital ∗-algebra that separates points, because X = lim←− Xα. It follows that for any ε > 0, we can find some ψα : Xα → E for which kf − ψα ◦ παk∞ < ε, (14) and we may also assume α ≥ β. If µ denotes the Haar measure on G, then the averaging procedure ϕ 7→ ϕ ⊳ µ :=ZG gϕ(g−1•) dµ(g) ∈ C(X, E) produces G-equivariant maps, does not increase supremum norms, and fixes f (because f is already G-equivariant). In conclusion, applying ⊳µ to (14) produces kf − (ψα ◦ πα) ⊳ µk∞ < ε. Because α ≥ β and the G-action on E factors through πβ, (ψα ◦ πα) ⊳ µ must factor through πα as a composition fα ◦ πα. It follows that fα is Gα-equivariant and kf − fα ◦ παk∞ < ε. (cid:4) The assumptions of Lemma 3.3 are frequently satisfied when G is a pro-p group, such as Zp, as any finite-dimensional unitary representation E of G will factor through one of the p-group quotients Gα, such as Z/pk. The role of Xα may then be played by an iterated join of Gα, whose inverse limit is the iterated join of G. Moreover, following the techniques of Dold in [9], if we fix a primitive pk root of unity ω and note that the iterated join (Z/pk)∗2n is (2n − 2)-connected, it follows that there is a map S2n−1 → (Z/pk)∗2n which is equivariant for the rotation action ~z 7→ ω~z on the sphere and the diagonal action of Z/pk on its iterated join. Having replaced the relevant domain of an equivariant map with a sphere, we may also view the unit sphere S2m−1 ⊂ Cm \ {0} as the codomain through scaling, leading us to the following question. Question 3.4 For positive integers k, r, m, and n with k > r, let Z/pk act on the sphere S2n−1 freely through multiplication by e2πi/pk and non-freely on the sphere S2m−1 through multiplication by e2πi/pr . Call an equivariant map S2n−1 → S2m−1 for these actions a (pk, pr)-map. Fix m. Do there exist n and r such that for all k > r, (pk, pr)-maps S2n−1 → S2m−1 do not exist? If for all m ≥ 2 and primes p, the answer to Question 3.4 is yes, then the Type 1 conjecture immediately follows, so resolution of Question 3.4 is one possible approach to the conjecture. These types of questions have been studied, but to our knowledge no tight dimension bounds have been published that apply for large k. Precise bounds have been found for k = 2, r = 1, and scale information is known as k → ∞. We follow the notation of [15]: vp,k(n) = min{m ∈ Z+ : there exists a (pk, p)-map S2n−1 → S2m−1}. Theorem 4.8 of [15] states that for any odd prime p, vp,2(1) = 1 and & n − 2 p ' + 1 ≤ vp,2(n) ≤ & n − 2 vp,2(n) = & n − 2 p ' + 2 p ' + 2 for n 6≡ 2 mod p for n ≡ 2 mod p. (15) 11 Similar bounds for p = 2 may be found in [19]; the case p = 2 is unique, as the usual Z/2 antipodal action exists on even spheres as well as odd. Moreover, Theorem 5.5 of [15] is equivalent to the claim that (16) vp,k(n) lim n→∞ n = 1 pk−1, so for large spheres S2n−1, there is a large gap between n and the smallest m with (pk, p)-maps S2n−1 → S2m−1, roughly n/m ≈ pk−1. This does not answer Question 3.4 for r = 1, as this behavior might only manifest in large dimension, and the number dual to vp,k for fixed codomain, wp,k(m) = max{n ∈ Z+ : there exists a (pk, p)-map S2n−1 → S2m−1}, might remain bounded in k. It is also known from [2, Theorem 1.2] that vp,k(n) ≥& n − 1 pk−1' + 1, (17) (18) so boundedness of the set {wp,k(m)}k∈Z+ for individual m could follow, for instance, if the constant term 1 in (18) can be replaced with an unbounded, sublinear function of n alone. Even if Question 3.4 has a negative answer, it is possible that the Type 1 conjecture may still hold. In what follows, we consider a related approach that is less demanding than Question 3.4. Proposition 3.5 Conjecture 1.1 and the classical subcase H = C(G) of Conjecture 1.2 Type 1 are equivalent. Proof We follow the idea of [17]. Conjecture 1.2 Type 1 implies Conjecture 1.1, so assume that Conjecture 1.2 Type 1 fails for H = C(G). By Lemma 3.2 there is then a free action α of Zp on a unital C ∗-algebra A and an equivariant morphism A → A ⊛ C(Zp). This implies that the largest commutative quotient A ։ B is nontrivial, and the action α descends to an action β on B. From examination of spectral subspaces, we see that β is free. The composition A → A ⊛ C(Zp) ։ B ⊛ C(Zp) descends to an equivariant morphism B → B ⊛ C(Zp), as the codomain is commutative. This is dual to an equivariant map X ∗ Zp → X for the corresponding free action of Zp on X 6= ∅, so Conjecture 1.1 also fails. (cid:4) Question 3.6 Suppose X is a compact Hausdorff space with a free action of Zp. Equip Z∗n the diagonal action and take the direct limit Z∗∞ for an equivariant map Z∗∞ p with , with the diagonal action as well. Is it possible p → X to exist? p If the answer is always no, then the Type 1 conjecture follows, as the topological iteration dual to (11) takes an equivariant map X ∗ Zp → X and produces a chain · · · → X ∗ Zp ∗ Zp → X ∗ Zp → X, into Z∗n+1 p p = Zp ∗ Z∗n Because Z∗∞ is a non-compact Tychonoff space, its Stone- Cech compactification βZ∗∞ which gives equivariant maps Z∗n consistent with the inclusions of Z∗n p Z∗∞ p → X for all n. Similar to [17, Alternative Proof B], these maps are p , so they extend to an (equivariant) map p → X, which would give a contradiction based on the assumed negative answer to Question 3.6. exists. Now, βZ∗∞ is supplied with a not necessarily continuous diagonal action of Zp, as seen by following the universal property of the Stone- Cech compactification under each individual homeomorphism p → X extends to the Stone- Cech compactification, dual to a x 7→ x · g. Any equivariant map Z∗∞ morphism C(X) → Cb(Z∗∞ Question 3.7 Consider C(βZ∗∞ ). Perhaps this is ruled out by limitations on the spectral subspaces. ) with the (not necessarily continuous) diagonal action ) = Cb(Z∗∞ p p p p p of the compact group Zp. For some τ ∈cZp, does it hold that 1 6∈ Cb(Z∗∞ p 12 )∗ τ Cb(Z∗∞ p )τ ? 4 Correction of the Literature In a previous version of this manuscript, we proposed a solution to Type 1 of Conjecture 1.2 using Claim 2.6 of [8], which we repeat here. Claim 4.1 ([8] Claim 2.6, Erroneous) Let q ≥ 2 be a prime power, d ≥ 1 an odd integer, and N = (q − 1)(d + 1). For a group G of order q, let E be a unitary G-representation of (complex) dimension N/2 with no trivial subrepresentations. Then every G-equivariant map G∗(N +1) → E has a zero. (cid:7) We are grateful to Robert Edwards for pointing out to us that this claim is incorrect. Namely, the proof given in [8] only concerns the case G = Z/p, and for G = Z/pk the result fails with a kLi=1 counterexample that may be found by modifying [21, p. 153-154] (due to Floyd). For m ≥ 2, define a (pk, p)-map f on S2m−1 in polar form by f : (z1, . . . , zm) = (r1u1, . . . , rmum) ∈ S2m−1 7→ (r1upk−1 1 , . . . , rmupk−1 m ) ∈ S2m−1, so that f has degree pm(k−1). Since this degree is a multiple of pk, f may be modified along its free orbits to produce a (pk, p)-map h : S2m−1 → S2m−1 which is homotopically trivial. Let X = S2m−1 ∗ Z/pk and equip X with the free diagonal action of Z/pk, so that the homotopically trivial map h produces a (pk, p)-map from X to S2m−1. Further, X is (2m − 1)-connected, so there is an equivariant map (Z/pk)∗2m+1 → X, which may be constructed cell-by-cell as in [9]. Composition of these maps gives a (pk, p)-map (Z/pk)∗2m+1 → S2m−1. Finally, if m ≥ 2 and E = Cm is equipped with the representation of Z/pk through an order p character, there is an equivariant map (Z/pk)∗2m+1 → E \ {0}. Since E has no trivial subrepresen- tations, this produces many counterexamples to the claim, such as for p = 2, k = 2, q = 4, m = 3, d = 1, N = 6. Moreover, Claim 4.1 is actually inconsistent with results in [15] and [19], and a rea- sonably explicit construction of a (4, 2)-equivariant counterexample may be found by manipulating [13, Example 5.1], which produces a (4, 2)-map S3 → S2 and consequently a (4, 2)-map S3 → S3 which is homotopically trivial. The main issue with Claim 4.1 can be traced back to the proof in [8] for elementary abelian p-groups, making it clear how this proof fails to apply to other classes of p-groups (and specifically to G = Z/pk, the case we are interested in here). The line of reasoning followed in [8] translates Claim 4.1 into the language of the equivalent [8, Claim 4.9], which in our case demands the following. Claim 4.2 (Erroneous) Let G = Z/pk, χ a character of G with kernel Z/pk−1, and E = χ⊕ N for N as in Claim 4.1. Then the top Chern class of the bundle on G∗(N +1)/G associated to E is non-vanishing. (cid:7) 2 However, examining the proof following [8, Claim 4.9], we see that in fact the Chern class in question is (pk−1y) N 2 in the cohomology ring H ∗(Z/pk, Z) ∼= Z[y]/(pky), deg(y) = 2. For k ≥ 2 and N ≥ 2 this vanishes because pk divides (pk−1) strong negation of Claims 4.1 and 4.2, preserving the notation and conventions therein. 2 . In conclusion, we have the following N Proposition 4.3 Let q and N be as in Claim 4.1, where q is not prime. Then for an order-p character χ of Z/q, there exists a nowhere-vanishing (Z/q)-equivariant map (Z/q)∗(N +1) → χ⊕ N 2 . 13 Acknowledgments We are deeply grateful to Robert Edwards for his pivotal correction. B.P. would also like to thank his Ph.D. advisors John McCarthy and Xiang Tang, his postdoctoral mentors Baruch Solel and Orr Shalit, and Piotr M. Hajac for his hospitality at IM PAN. References [1] Luis Barreira. Ergodic theory, hyperbolic dynamics and dimension theory. Universitext. Springer, Heidelberg, 2012. [2] Thomas Bartsch. On the genus of representation spheres. Comment. Math. Helv., 65(1):85 -- 95, 1990. [3] Paul F. Baum, Ludwik D abrowski, and Piotr M. Hajac. Noncommutative Borsuk-Ulam-type ' conjectures. Banach Center Publ., 106:9 -- 18, 2015. [4] Paul F. Baum, Kenny De Commer, and Piotr M. Hajac. Free Actions of Compact Quantum Groups on Unital C ∗-algebras. Doc. Math., 22:825-849, 2017. [5] Bruce Blackadar. K-theory for operator algebras, volume 5 of Mathematical Sciences Research Institute Publications. Cambridge University Press, Cambridge, Second edition, 1998. [6] Joachim Cuntz. K-theory for certain C ∗-algebras. Ann. of Math. (2), 113(1):181 -- 197, 1981. [7] Ludwik D abrowski, Tom Hadfield, and Piotr M. Hajac. Equivariant join and fusion of non- ' commutative algebras. SIGMA Symmetry Integrability Geom. Methods Appl., 11:Paper 082, 7, 2015. [8] Mark de Longueville. Erratum to: "Notes on the topological Tverberg theorem". Discrete Math., 247(1-3):271 -- 297, 2002. [9] Albrecht Dold. Simple proofs of some Borsuk-Ulam results. In Proceedings of the Northwestern Homotopy Theory Conference (Evanston, Ill., 1982), volume 19 of Contemp. Math., pages 65 -- 69. Amer. Math. Soc., Providence, RI, 1983. [10] David Alexandre Ellwood. A new characterisation of principal actions. J. Funct. Anal., 173(1):49 -- 60, 2000. [11] Piotr M. Hajac and Tomasz Maszczyk. Pulling back noncommutative associated vector bundles and constructing quantum quaternionic projective spaces. ArXiv e-prints, December 2015. [12] Karl H. Hofmann and Sidney A. Morris. The Structure of Compact Groups, volume 25 of De Gruyter Studies in Mathematics. De Gruyter, Berlin, 2013. A primer for the student -- a handbook for the expert, Third edition, revised and augmented. [13] Jan Jaworowski. Involutions in lens spaces. Topology Appl., 94(1-3):155 -- 162, 1999. Special issue in memory of B. J. Ball. [14] Nicolaas H. Kuiper. The homotopy type of the unitary group of Hilbert space. Topology, 3:19 -- 30, 1965. 14 [15] D. M. Meyer. Z/p-equivariant maps between lens spaces and spheres. Math. Ann., 312(2):197 -- 214, 1998. [16] Oleg R. Musin and Alexey Yu. Volovikov. Tucker-Type Lemmas for G-spaces. ArXiv e-Prints, January 2017. [17] Benjamin Passer. Free Actions on C*-algebra Suspensions and Joins by Finite Cyclic Groups. To appear in IUMJ. [18] N. Christopher Phillips. Freeness of actions of finite groups on C ∗-algebras. In Operator structures and dynamical systems, volume 503 of Contemp. Math., pages 217 -- 257. Amer. Math. Soc., Providence, RI, 2009. [19] Stephan Stolz. The level of real projective spaces. Comment. Math. Helv., 64(4):661 -- 674, 1989. [20] A. Yu. Volovikov. Coincidence points of mappings of Z n p -spaces. Izv. Ross. Akad. Nauk Ser. Mat., 69(5):53 -- 106, 2005. [21] R. F. Williams. The construction of certain 0−dimensional transformation groups. Trans. Amer. Math. Soc., 129:140 -- 156, 1967. 15
1103.4111	2	1103	2011-05-30T13:04:54	Naturality of Symmetric Imprimitivity Theorems	[ "math.OA", "math.FA" ]	The first imprimitivity theorems identified the representations of groups or dynamical systems which are induced from representations of a subgroup. Symmetric imprimitivity theorems identify pairs of crossed products by different groups which are Morita equivalent, and hence have the same representation theory. Here we consider commuting actions of groups $H$ and $K$ on a $C^*$-algebra which are saturated and proper as defined by Rieffel in 1990. Our main result says that the resulting Morita equivalence of crossed products is natural in the sense that it is compatible with homomorphisms and induction processes.	math.OA	math	NATURALITY OF SYMMETRIC IMPRIMITIVITY THEOREMS ASTRID AN HUEF, S. KALISZEWSKI, IAIN RAEBURN, AND DANA P. WILLIAMS Abstract. The first imprimitivity theorems identified the representations of groups or dynamical systems which are induced from representations of a sub- group. Symmetric imprimitivity theorems identify pairs of crossed products by different groups which are Morita equivalent, and hence have the same repre- sentation theory. Here we consider commuting actions of groups H and K on a C ∗-algebra which are saturated and proper as defined by Rieffel in 1990. Our main result says that the resulting Morita equivalence of crossed products is natural in the sense that it is compatible with homomorphisms and induction processes. 1. Introduction G(B, α). Motivated by the idea that C0(T /G) and IndT Suppose that a locally compact group G acts freely and properly on the right of a locally compact space T , and rt is the induced action on C0(T ). Green proved in [4] that the crossed product C0(T ) ⋊rt G is Morita equivalent to C0(T /G). Raeburn and Williams considered diagonal actions rt ⊗ α on G on C0(T, B) = C0(T ) ⊗ B, and showed in [18] that C0(T, B) ⋊rt⊗α G is Morita equivalent to the induced algebra IndT G(B, α) are playing the role of a fixed-point algebra for the action of G, Rieffel studied a family of proper actions (A, α) for which there is a generalized fixed-point algebra Aα in M(A) [21]. Rieffel proved in particular that if α is an action of G on a C ∗-algebra A and if φ : C0(T ) → M(A) is an equivariant nondegenerate homomorphism, then the reduced crossed product A ⋊α,r G is Morita equivalent to Aα (see [22, Theorem 5.7] and [21, Corollary 1.7]); taking A = C0(T, B) gives the result in [18]. Rieffel's construction of Aα starts from a dense subalgebra A0 of A with properties like those of Cc(T ) in C0(T ); while in practice there always seems to be an obvious candidate for A0 which gives the "right answer", Aα does ostensibly depend on the choice of A0, and several authors have tried in vain to find a canonical choice [3, 22]. Alternatively, as in [10], we can try to say upfront what "right" means, and prove that Rieffel's construction has these properties. The idea is, loosely, that construc- tions should be functorial and isomorphisms, such as those implemented by Morita equivalences, should be natural. This program has already had some significant ap- plications, especially in nonabelian duality for crossed products of C ∗-algebras [10, 7]. Date: 25 March 2011. 2000 Mathematics Subject Classification. Primary: 46L55. Key words and phrases. Symmetric imprivitity theorem, proper actions on C ∗-algebras, natural- ity, fixed-point algebras, crossed products. This research was supported by the University of Otago and the Edward Shapiro fund at Dart- mouth College. 1 2 AN HUEF, KALISZEWSKI, RAEBURN, AND WILLIAMS In [6], we showed that the assigments (A, α, φ) 7→ A ⋊α,r G and (A, α, φ) 7→ Aα can be extended to functors RCP and Fix between certain categories whose morphisms are derived from right-Hilbert bimodules, and whose isomorphisms are given by Morita equivalences. The main result of [6] says that Rieffel's Morita equivalences give a natural isomorphism between RCP and Fix. All these Morita equivalences have symmetric versions. For Green's theorem, the symmetric version involves commuting free and proper actions of two groups H and K on the same space T , and says that C0(T /H)⋊K is Morita equivalent to C0(T /K)⋊H (this is marginally more general than the version proved in [20]). If in addition α and β are commuting actions of H and K on a C ∗-algebra B, then the symmetric imprimitivity theorem of [11, 17] gives a Morita equivalence between crossed products IndT rt⊗αH. Work of Quigg and Spielberg [16] implies that there is a similar Morita equivalence for the reduced crossed products. In [9, Corollary 3.8], we found a symmetric version of Rieffel's equivalence for a pair of commuting proper actions on a C ∗-algebra A, and then recovered the Quigg-Spielberg theorem by taking A = C0(T, B) (see [9, §4]). Symmetric imprimitivity theorems have found significant applications (see [12, 1], for example), and there are analogues for groupoids [13, 19], for graph algebras [15], and for Fell bundles [14]. rt⊗β K and IndT H(B, α)⋊ K(B, β)⋊ Here we consider commuting free and proper actions of H and K on T , and form a category whose objects (A, σ, τ, φ) consist of commuting actions σ : H → Aut A and τ : K → Aut A, and an equivariant nondegenerate homomorphism φ : C0(T ) → M(A). We show that there are functors FixH and FixK based on the assignments (A, σ, τ, φ) 7→ (Aσ, ¯τ ) and (A, σ, τ, φ) 7→ (Aτ , ¯σ), and prove that Rieffel's bimodules X(A, σ, τ, φ) from [9, Corollary 3.8] give a natural isomorphism between the functors RCP ◦ FixH and RCP ◦ FixK. This is interesting even in the situation of [17], where it gives the naturality of Quigg and Spielberg's symmetric imprimitivity theorem, and in the one-sided case, where it yields naturality of the Morita equivalence of [18]. 2. Preliminaries Throughout this paper, G, H, and K are locally compact groups, all of which act on the right of a locally compact space T . The actions of H and K are always assumed to commute. We denote by rt the action of any of them on C0(T ) by right translation: rtg(f )(t) = f (t · g) We use the same categories and notation as in [6]. In particular, C* is the category whose objects are C ∗-algebras and whose morphisms are isomorphism classes of right- Hilbert bimodules. In the category Cact(G), the objects (A, α) consist of an action α of G on a C ∗-algebra A, and the morphisms [X, u] : (A, α) → (B, β) are isomorphism classes of right-Hilbert A -- B bimodules X with an α -- β compatible action u of G. In the semi-comma category Cact(G, (C0(T ), rt)), the objects are triples (A, α, φ), where (A, α) is an object in Cact(G) and φ : C0(T ) → M(A) is an rt -- α equivariant nondegenerate homomorphism. The morphisms from (A, α, φ) to (B, β, ψ) are the same as the morphisms from (A, α) to (B, β) in Cact(G).1 1In [6], we were only interested in free and proper actions of G, but there is no reason to make this restriction when defining the semi-comma category Cact(G, (C0(T ), rt)). NATURALITY 3 Commuting actions σ, τ of H, K (on spaces, C ∗-algebras or Hilbert modules) are essentially the same as actions σ × τ of G = H × K, and hence we can apply the construction of the previous paragraph with G = H × K. However, to em- phasize the symmetry of our situation, we view the comma category as a category Cact(H, K, (C0(T ), rt)) in which the objects are quadruples (A, σ, τ, φ) such that (A, σ × τ, φ) is an object of the semi-comma category Cact(H × K, (C0(T ), rt)); the morphisms from (A, σ, τ, φ) to (B, µ, ν, ψ) are then triples [X, u, v] such that [X, u×v] is a morphism from (A, σ × τ, φ) to (B, µ × ν, ψ) in Cact(H × K, (C0(T ), rt)). Remark 2.1. We stress that, when we assume that H and K act freely and properly on T , we are not assuming that H ×K acts freely and properly, because then we would lose the main applications of the symmetric imprimitivity theorem. For example, let T = R, H = Z and K = Z, take an irrational number θ, and define r · h := r + hθ and r · k = r + k for r ∈ T , h ∈ H, k ∈ K: then the symmetric imprimitivity theorem implies that the irrational rotation algebra Aθ is Morita equivalent to Aθ−1. 3. Defining the Functors In this section we define functors on Cact(H, K, (C0(T ), rt)) analogous to RCP and Fix from [6], but which only deal with the H- or K-part of the action, and which take values in an equivariant category. Proposition 3.1. The assignments (A, σ, τ, φ) 7→ (A ⋊σ,r H, τ ⋊ idH ) and [X, u, v] 7→ [X ⋊u,r H, v ⋊ idH] define a functor RCPH from Cact(H, K, (C0(T ), rt)) to Cact(K), where τ ⋊ idH is the action of K given by (τ ⋊ idH)k(f )(h) = τk(f (h)) for f ∈ Cc(H, A) (and similarly for v ⋊ idH ). Proof. Let F : Cact(H × K) → Cact(H × K) be the subgroup-crossed-product functor from [2, Theorem 3.24], applied to the normal subgroup H ⊂ H × K. With αdec and wdec the "decomposition actions" of H × K as defined in [2, Section 3.3.1], F is given by (A, α) 7→ (A ⋊α,r H, αdec) and [X, w] 7→ [X ⋊w,r H, wdec]. With α = σ × τ , for k ∈ K and f ∈ Cc(H, A) we have αdec (e,k)(f )(h) = ∆H (e)σe(cid:0)τk(cid:0)f (e−1he)(cid:1)(cid:1) = τk(f (h)) = (τ ⋊ idH)k(f )(h), and similarly, if w = u×v, then wdec (e,k) = (v ⋊idH)k. So if we let Res : Cact(H ×K) → Cact(K) be the restriction functor from [2, Corollary 3.17] applied to the subgroup K ⊂ H × K, we see that the composition Res ◦ F is given by (A, σ × τ ) 7→ (A ⋊σ,r H, τ ⋊ idH ) and [X, u × v] 7→ [X ⋊u,r H, v ⋊ idH]. Defining RCPH to be the composition of Res ◦ F with the forgetful functor from Cact(H × K, (C0(T ), rt)) to Cact(H × K) which takes (A, α, φ) to (A, α) gives the result. (cid:3) 4 AN HUEF, KALISZEWSKI, RAEBURN, AND WILLIAMS For the next result, we observe that the proof of [7, Proposition 4.1] only requires that the normal subgroup N ⊂ G acts freely and properly; the basic factorization result in [6, Corollary 2.3], which is invoked in the proof, applies in the semi-comma category for arbitrary actions of G on T . So if H acts freely and properly on T , we can apply [7, Proposition 4.1] with G = H × K, N = H and G/N = K. This gives a fixed-point functor FixK H from Cact(H, K, (C0(T ), rt)) to Cact(K, (C0(T /H), rt)) with object and morphism maps (A, σ × τ, φ) 7→ (Aσ, (σ × τ )K, φH) and [X, u × v] 7→ [Fix(X, u), (u × v)K]. Now we introduce the notation ¯τ = (σ × τ )K, X u = Fix(X, u), and ¯v = (u × v)K. and define FixH to be the composition of FixK Cact(K, (C0(T /H), rt)) to Cact(K). Then we have: H with the forgetful functor from Proposition 3.2. Suppose that the action of H on T is free and proper. Then the assignments (3.1) (A, σ, τ, φ) 7→ (Aσ, ¯τ ) and [X, u, v] 7→ [X u, ¯v] define a functor FixH from Cact(H, K, (C0(T ), rt)) to Cact(K). 4. Naturality Suppose that the actions of H and K on T are free and proper. Then for each object (A, σ, τ, φ) of Cact(H, K, (C0(T ), rt)), the hypotheses of [9, Theorem 4.4] are satisfied; therefore A0 := span{φ(f )aφ(g) f, g ∈ Cc(T ), a ∈ A} can be completed to give an Aσ ⋊¯τ ,r K -- Aτ ⋊¯σ,r H imprimitivity bimodule, which we will denote here by X(A, σ, τ, φ). The isomorphism class [X(A, σ, τ, φ)] is an isomorphism in the category C* from Aσ ⋊¯τ ,r K = RCP ◦ FixH(A, σ, τ, φ) to Aτ ⋊¯σ,r H = RCP ◦ FixK(A, σ, τ, φ). Theorem 4.1. Suppose that H and K act freely and properly on T . Then (A, σ, τ, φ) 7→ [X(A, σ, τ, φ)] is a natural isomorphism between the functors RCP ◦ FixH and RCP ◦ FixK from Cact(H, K, (C0(T ), rt)) to C. Our strategy for proving Theorem 4.1 is as follows. First, we present a natural isomorphism between the functors FixH and RCPH defined in Section 3. This is done in Proposition 4.2, which is an equivariant version of [6, Theorem 3.5]. Composing with RCP gives a natural isomorphism between RCP ◦ FixH and RCP ◦ RCPH (Corollary 4.4). Since the iterated crossed-product functors RCP ◦ RCPH and RCP ◦ RCPK are easily seen to be naturally isomorphic, we obtain a natural isomorphism by composition: RCP ◦ FixH ∼ RCP ◦ RCPH ∼ RCP ◦ RCPK ∼ RCP ◦ FixK . The last step of the proof is to identify the bimodules underlying this composition with X(A, σ, τ, φ); this requires the reworking of [9, Theorem 4.4] in Proposition 4.6. For each object (A, σ, τ, φ) of Cact(H, K, (C0(T ), rt)), the triple (A, σ, φ) is an object of Cact(H, (C0(T ), rt)) with H acting freely and properly. Thus Rieffel's NATURALITY 5 theory provides an A ⋊σ,r H -- Aσ imprimitivity bimodule Z(A, σ, φ) which is a com- pletion of A0 := Cc(T )ACc(T ). Now τ restricts to an action of K on A0 which, by the proof of [9, Proposition 4.2], extends to an action (τ ⋊ idK, τ, ¯τ ) of K on Z(A, σ, φ). Thus [Z(A, σ, φ), τ ] is an isomorphism in the category Cact(K) between (A ⋊σ,r H, τ ⋊ idK) = RCPH(A, σ, τ, φ) and (Aσ, ¯τ ) = FixH(A, σ, τ, φ). Proposition 4.2. Suppose that the action of H on T is free and proper. Then (A, σ, τ, φ) 7→ [Z(A, σ, φ), τ ] is a natural isomorphism between the functors RCPH and FixH from the category Cact(H, K, (C0(T ), rt)) to Cact(K). Proof. We follow the "canonical decomposition" strategy of the proof of [6, The- orem 3.5]. So suppose [X, u, v] is a morphism from (A, σ, τ, φ) to (B, µ, ν, ψ) in Cact(H, K, (C0(T ), rt)). Then by [6, Corollary 2.3],2 there exists an isomor- phism [Y, u, v] : (K, ζ, η, χ) → (B, µ, ν, ψ) in Cact(H, K, (C0(T ), rt)) given by K -- B imprimitivity bimodule Y , and a morphism [κ] : (A, σ, τ, φ) → (K, ζ, η, χ) in Cact(H, K, (C0(T ), rt)) coming from a σ -- ζ and τ -- η equivariant nondegenerate homomorphism κ : A → M(K), such that [X, u, v] is the composition: (A, σ, τ, φ) κ / (K, ζ, η, χ) (Y,u,v) / (B, µ, ν, ψ). Consider the following diagram in Cact(K), which is an equivariant version of diagram (4.9) from the proof of [6, Theorem 3.5]: (A ⋊σ,r H, τ ⋊ idH) (Z(A,σ,φ),τ ) (Aσ, ¯τ ) (4.1) (X⋊u,rH,v⋊idH ) (K ⋊ζ,r H, η ⋊ idH ) κ⋊rH (PPPPPPPPPPPPPPPPPP vnnnnnnnnnnnnnnnnnn (Y ⋊u,rH,v⋊idH ) (Z(K,ζ,χ),η) (Kζ, ¯η) κ ~}}}}}}}}}}} AAAAAAAAAAA (Y u,¯v) (X u,¯v) (Z(B,µ,ψ),ν) / (Bµ, ¯ν). (B ⋊µ,r H, ν ⋊ idH) Here the arrow labeled κ ⋊r H denotes (the isomorphism class of) K ⋊ζ,r H viewed as an equivariant right-Hilbert (A ⋊σ,r H) -- (K ⋊ζ,r H)-bimodule with the left action given by κ ⋊r H, and similarly for κ. Thus the left and right triangles of (4.1) commute by functoriality of RCPH and FixH. The upper quadrilateral of (4.1) is an equivariant version of diagram (3.1) in [10, Theorem 3.2] (see also [10, Remark 3.3]). That result provides an isomorphism Φ : Z(A, σ, τ ) ⊗Aσ Kζ → Z(K, ζ, χ) such that Φ(a ⊗ EH(c)) = κ(a)EH (c) for a ∈ A0 and c ∈ K0 = span{χ(f )dχ(g) f, g ∈ Cc(T ), d ∈ K}, where EH is the averaging 2As we have observed earlier, we can apply [6, Corollary 2.3] to Cact(H × K, (C0(T ), rt)) without the assumption that H × K acts freely and properly on T . / / ( / / ~ / / v / 6 AN HUEF, KALISZEWSKI, RAEBURN, AND WILLIAMS process which maps K0 into Kζ (see [10, Section 2]). Hence to prove that this upper quadrilateral commutes, we need to check that (4.2) Φ ◦ (τ ⊗ ¯η)k = ηk ◦ Φ for k ∈ K. We break off our argument for a lemma: Lemma 4.3. For k ∈ K and c ∈ K0, ¯ηk(cid:0)EH (c)(cid:1) = EH(cid:0)η(c)(cid:1). Proof. By linearity, we may suppose c = f dg with f, g ∈ Cc(T ) and d ∈ K, where to reduce clutter we write f dg for χ(f )dχ(g). Then for h ∈ Cc(T ) and k ∈ K, we compute, using [10, Lemma 2.2], h¯ηk(cid:0)EH (f dg)(cid:1) = ¯ηk(cid:0)(ηk)−1(h)EH(f dg)(cid:1) = ¯ηk(cid:16)ZH (ηk)−1(h)ζs(f dg) ds(cid:17). Thus, since the integral is norm-convergent and since η and ζ commute, we have h¯ηk(cid:0)EH(f dg)(cid:1) = ZH hζs(cid:0)ηk(f dg)(cid:1) ds = hEH(cid:0)ηk(f dg)(cid:1). (cid:3) End of the proof of Proposition 4.2. For k ∈ K, a ∈ A0 and c ∈ K0, Lemma 4.3 gives Φ(cid:0)τk ⊗ ¯ηk(a ⊗ EH(c))(cid:1) = Φ(cid:0)τk(a) ⊗ ¯ηk(cid:0)EH (c)(cid:1)(cid:1) = Φ(cid:0)τk(a) ⊗ EH(cid:0)ηk(c)(cid:1)(cid:1) = κ(cid:0)τk(a)(cid:1)EH(cid:0)ηk(c)(cid:1) = ηk(cid:0)κ(a)(cid:1)¯ηk(cid:0)EH(c)(cid:1) = ηk(cid:0)κ(a)EH (c)(cid:1) = ηk(cid:0)Φ(a ⊗ EH(c)(cid:1). This establishes (4.2), and shows that the upper quadrilateral of (4.1) commutes. Now we turn to the bottom quadrilateral of (4.1). Here, all the morphisms arise from imprimitivity bimodules, and we showed in the proof of [6, Theorem 3.5] that such a diagram commutes in C* (that is, without the actions of K) by observing that L(Y )L(u) = L(Y u) and that L(Y ) ⋊L(u),r H is isomorphic to L(Y ⋊u,r H) by [8, Proposition 3.4]. Then we applied [2, Lemma 4.6] to get the requisite imprimitivity- bimodule isomorphisms. To see that the diagram commutes in Cact(K) just requires that these isomorphisms be equivariant, and this is proved in [2, Lemma 4.9]. This completes the proof of Proposition 4.2. (cid:3) Corollary 4.4. Suppose H and K act freely and properly. Then the assignment (A, σ, τ, φ) 7→ [Z(A, σ, φ) ⋊τ,r K] is a natural isomorphism between the functors RCP ◦ RCPH and RCP ◦ FixH from the category Cact(H, K, (C0(T ), rt)) to C. The corollary, which can be viewed as an asymmetric version of Theorem 4.1, is an immediate consequence of Proposition 4.2 and the following elementary lemma. Lemma 4.5. Suppose that the assigment A 7→ ζA is a natural isomorphism between functors F1 and F2 into a category C. If F3 is a functor defined on C, then the assigment A 7→ F3(ζA) is a natural isomorphism between F3 ◦ F1 and F3 ◦ F2. NATURALITY 7 Given an imprimitivity bimodule Y , we denote the dual bimodule by Y ∼ = {♭(y) y ∈ Y }. Proposition 4.6. Suppose H and K act freely and properly. For each object (A, σ, τ, φ) of Cact(H, K, (C0(T ), rt)), there is an (Aσ ⋊¯τ ,r K) -- (Aτ ⋊¯σ,r H) imprim- itivity bimodule isomorphism X(A, σ, τ, φ) ∼= (cid:0)Z(A, σ, φ) ⋊τ,r K(cid:1)∼ ⊗Σ (cid:0)Z(A, τ, φ) ⋊σ,r H(cid:1), where the balanced tensor product incorporates the natural isomorphism (4.3) Σ : (A ⋊σ,r H) ⋊τ ⋊id,r K ∼= (A ⋊τ,r K) ⋊σ⋊id,r H. Proof. Theorem 4.4 of [9] implies that X(A, σ, τ, φ) is isomorphic to the tensor product (4.4) (X ⋊τ,r K) ⊗Σ (cid:0)Z(A, τ, φ) ⋊σ,r H(cid:1), where the unfortunately-named X in (4.4) is a module built on A0 as defined in the beginning of Section 2 of [9], and is equivariantly isomorphic to Z(A, σ, φ)∼. The map on Cc(K, A0) given by f 7→ ♭(f ∗), where f ∗(k) := ∆K(k)−1τk(cid:0)f (k−1)∗(cid:1), extends to an imprimitivity bimodule isomorphism of X ⋊τ,r K onto (Z(A, σ, φ) ⋊τ,r K)∼, and the result follows. (cid:3) Proof of Theorem 4.1. It follows from Corollary 4.4 that the assignment (A, σ, τ, φ) 7→ (cid:2)Z(A, σ, φ) ⋊τ,r K(cid:3)−1 = (cid:2)(cid:0)Z(A, σ, φ) ⋊τ,r K(cid:1)∼(cid:3) is a natural isomorphism between RCP ◦ FixH and RCP ◦ RCPH; symmetrically, (A, σ, τ, φ) 7→ [Z(A, τ, φ) ⋊σ,r H] is a natural isomorphism between RCP ◦ RCPK and RCP ◦ FixK. It is a straightfor- ward computation to verify that the isomorphism Σ at (4.3) indeed gives a natural isomorphism between RCP ◦ RCPH and RCP ◦ RCPK; composing these three and using Proposition 4.6, we see that the assignment (A, σ, τ, φ) 7→ [Z(A, τ, φ) ⋊σ,r H] ◦ [Σ] ◦(cid:2)(cid:0)Z(A, σ, φ) ⋊τ,r K(cid:1)∼(cid:3) ⊗Σ (cid:0)Z(A, τ, φ) ⋊σ,r H(cid:1)(cid:3) = (cid:2)(cid:0)Z(A, σ, φ) ⋊τ,r K(cid:1)∼ = [X(A, σ, τ, φ)] is a natural isomorphism between RCP ◦ FixH and RCP ◦ FixK, as desired. (cid:3) 5. Applications We view objects in Cact(H × K) as triples (B, α, β) consisting of commuting actions α : H → Aut B and β : K → Aut B. Suppose that the actions of H and K on T are free and proper. Then the symmetric imprimitivity theorem of [17] says that IndT rt⊗α H (and provides a concrete bimodule which is a completion of Cc(T, B)). As explained in [5, Corollary 3], it follows from [16, Lemma 4.1] that the reduced crossed products are Morita equivalent via a quotient Y (B, α, β) of the bimodule of [17]. rt⊗β K is Morita equivalent to IndT H(B, α) ⋊ K(B, β) ⋊ 8 AN HUEF, KALISZEWSKI, RAEBURN, AND WILLIAMS Corollary 5.1. Suppose that H and K act freely and properly on T . Then the assignments (B, α, β) 7→ IndT H (B, α) ⋊ rt⊗β,r K and (B, α, β) 7→ IndT K(B, β) ⋊ rt⊗α,r H are the object maps for functors FH and FK on Cact(H × K), and the assignment is a natural isomorphism between FH and FK . (B, α, β) 7→ [Y (B, α, β)] Proof. Define φ : C0(T ) → M(C0(T, B)) by φ(f ) = f ⊗ 1. It is not hard to see that there is a functor CT : Cact(H × K) → Cact(H, K, (C0(T ), rt)) which sends the object (B, α, β) to (C0(T, B), rt ⊗ α, rt ⊗ β, φ). Then it follows from Theorem 4.1 (and general nonsense such as Lemma 4.5) that the assignment (B, α, β) 7→ [X(C0(T, B), rt ⊗ α, rt ⊗ β, φ)] is a natural isomorphism between RCP ◦ FixH ◦ CT and RCP ◦ FixK ◦ CT. Therefore we can complete the proof of the Corollary by showing that the relevant generalized fixed- point algebras coincide with the induced algebras, and that the modules Y (B, α, β) and X(C0(T, B), rt ⊗ α, rt ⊗ β, φ) are isomorphic as imprimitivity bimodules, hence define the same isomorphism in the category C. The dense subalgebra A0 = Cc(T )C0(T, B)Cc(T ) of C0(T, B) is Cc(T, B). Then for f, g ∈ A0, the Fix(C0(T, B), rt ⊗ α)-valued inner product hf , gi is multiplication by the function s 7→ ZH αt(f (st)∗g(st)) dt, which is the same as the IndT Fix(C0(T, B), rt ⊗ α) and IndT similar argument applies to IndT H(B, α)-valued inner product on Cc(T, B) in [17]. Hence H (B, α) are the same subalgebra of M(C0(T, B)). A K(B, β). To see that Y (B, α, β) is isomorphic to X(C0(T, B), rt ⊗ α, rt ⊗ β, φ), we just need to check that the two sets of actions and inner products on Cc(T, B) coincide, and this is a straightforward calculation. (cid:3) When K = {e}, IndT K(B, β) = C0(T, B) and [5, Corollary 4] implies that C0(T, B) ⋊rt⊗α H = C0(T, B) ⋊rt⊗α,r H. Both the bimodule X(C0(T, B), rt ⊗ α, id, φ) and the bimodule W (B, α) in [18] are completions of Cc(T, B), and again the formulas for the actions and inner products turn out to be the same. Hence we obtain the naturality of the Morita equivalence for diagonal actions from [18]. Corollary 5.2. Suppose that H acts freely and properly on T . Then there are functors on Cact(H) whose maps on objects are given by (B, α) 7→ Ind(C, α) and (B, α) 7→ C0(T, B) ⋊rt⊗α H, and the assignment (B, α) 7→ [W (B, α)] is a natural isomorphism between these two functors. NATURALITY References 9 [1] J. Chabert, S. Echterhoff and H. Oyono-Oyono, Shapiro's lemma for topological K-theory of groups, Comment. Math. Helv. 78 (2003), 203 -- 225. [2] S. Echterhoff, S. Kaliszewski, J. Quigg and I. Raeburn, A categorical approach to imprimitivity theorems for C ∗-dynamical systems, Mem. Amer. Math. Soc. 180 (2006), no. 850, viii+169 pages. [3] R. Exel, Morita-Rieffel equivalence and spectral theory for integrable automorphism groups of C ∗-algebras, J. Funct. Anal. 172 (2000), 404 -- 465. [4] P. Green, C ∗-algebras of transformation groups with smooth orbit space, Pacific J. Math. 72 (1977), 71 -- 97. [5] A. an Huef and I. Raeburn, Regularity of induced representations and a theorem of Quigg and Spielberg, Math. Proc. Camb. Phil. Soc. 133 (2002), 149 -- 159. [6] A. an Huef, S. Kaliszewski, I. Raeburn and D.P. Williams, Naturality of Rieffel's Morita equiv- alence for proper actions, to appear in Algebr. Represent. Theory. (arXiv:0810.2819) [7] A. an Huef, S. Kaliszewski, I. Raeburn and D.P. Williams, Fixed-point algebras for proper actions and crossed products by homogeneous spaces, to appear in Illinois J. Math. (arXiv:0907.0681) [8] A. an Huef, I. Raeburn and D.P. Williams, Proper actions on imprimitivity bimodules and decompositions of Morita equivalences, J. Funct. Anal. 200 (2003), 401 -- 428. [9] A. an Huef, I. Raeburn and D.P. Williams, A symmetric imprimitivity theorem for commuting proper actions, Canad. J. Math. 57 (2005), 983 -- 1011. [10] S. Kaliszewski, J. Quigg and I. Raeburn, Proper actions, fixed-point algebras and naturality in nonabelian duality, J. Funct. Anal. 254 (2008), 2949 -- 2968. [11] G.G. Kasparov, Equivariant KK-theory and the Novikov conjecture, Invent. Math. 91 (1988), 147 -- 201. [12] E. Kirchberg and S. Wassermann, Permanence properties of C ∗-exact groups, Doc. Math. 4 (1999), 513 -- 558. [13] P.S. Muhly, J. Renault and D.P. Williams, Equivalence and isomorphism for groupoid C ∗- algebras, J. Operator Theory 17 (1987), 3 -- 22. [14] P.S. Muhly and D.P. Williams, Equivalence and disintegration theorems for Fell bundles and their C ∗-algebras, Dissertationes Math. 56 (2008), 1 -- 57. [15] D. Pask and I. Raeburn, Symmetric imprimitivity theorems for graph C ∗-algebras, Internat. J. Math. 12 (2001), 609 -- 623. [16] J.C. Quigg and J. Spielberg, Regularity and hyporegularity in C ∗-dynamical systems, Houston J. Math. 18 (1992), 139 -- 152. [17] I. Raeburn, Induced C ∗-algebras and a symmetric imprimitivity theorem, Math. Ann. 280 (1988), 369 -- 387. [18] I. Raeburn and D.P. Williams, Pull-backs of C ∗-algebras and crossed products by certain diagonal actions, Trans. Amer. Math. Soc. 287 (1985), 755 -- 777. [19] J. Renault, Repr´esentation des produits crois´es d'alg`ebres de groupoıdes, J. Operator Theory 18 (1987), 67 -- 97. [20] M.A. Rieffel, Applications of strong Morita equivalence to transformation group C ∗-algebras, Operator Algebras and Applications, Proc. Symp. Pure Math., vol. 38, Part I, Amer. Math. Soc., Providence, 1982, pages 299 -- 310. [21] M.A. Rieffel, Proper actions of groups on C ∗-algebras, Mappings of Operator Algebras, Progress in Math., vol. 84, Birkhauser, Boston, 1990, pages 141 -- 182. [22] M.A. Rieffel, Integrable and proper actions on C ∗-algebras, and square-integrable representations of groups, Expositiones Math. 22 (2004), 1 -- 53. 10 AN HUEF, KALISZEWSKI, RAEBURN, AND WILLIAMS Department of Mathematics and Statistics, University of Otago, PO Box 56, Dunedin 9054, New Zealand E-mail address: [email protected] School of Mathematical and Statistical Sciences, Arizona State University, Tempe, AZ 85287-1804, USA E-mail address: [email protected] Department of Mathematics and Statistics, University of Otago, PO Box 56, Dunedin 9054, New Zealand E-mail address: [email protected] Department of Mathematics, Dartmouth College, Hanover, NH 03755, USA E-mail address: [email protected]
1409.4332	2	1409	2015-07-04T05:30:01	Exotic crossed products and the Baum-Connes conjecture	[ "math.OA", "math.GR", "math.KT", "math.RT" ]	We study general properties of exotic crossed-product functors and characterise those which extend to functors on equivariant C-algebra categories based on correspondences. We show that every such functor allows the construction of a descent in KK-theory and we use this to show that all crossed products by correspondence functors of K-amenable groups are KK-equivalent. We also show that for second countable groups the minimal exact Morita compatible crossed-product functor used in the new formulation of the Baum-Connes conjecture by Baum, Guentner and Willett extends to correspondences when restricted to separable G-C-algebras. It therefore allows a descent in KK-theory for separable systems.	math.OA	math	EXOTIC CROSSED PRODUCTS AND THE BAUM-CONNES CONJECTURE ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT Abstract. We study general properties of exotic crossed-product functors and characterise those which extend to functors on equivariant C ∗-algebra categories based on correspondences. We show that every such functor allows the construction of a descent in KK-theory and we use this to show that all crossed products by correspondence functors of K-amenable groups are KK-equivalent. We also show that for second countable groups the minimal exact Morita compatible crossed-product functor used in the new formulation of the Baum-Connes conjecture by Baum, Guentner and Willett ([4]) extends to correspondences when restricted to separable G-C ∗-algebras. It therefore allows a descent in KK-theory for separable systems. 1. Introduction The concept of a crossed product A ⋊α G by an action α : G → Aut(A) of a locally compact group G on a C ∗-Algebra A by -automorphisms plays a very important role in the fields of Operator Algebras and Noncommutative Geometry. Classically, there are two crossed-product constructions which have been used in the literature: the universal (or maximal) crossed product A ⋊α,u G and the reduced (or minimal/spatial) crossed product A ⋊α,r G. The universal crossed product satisfies a universal property for covariant representations (π, u) of the underlying system (A, G, α) in such a way that every such representation integrates to a unique representation π ⋊ u of A ⋊α,u G, while the reduced crossed product is the image of A ⋊α,u G under the integrated form of the regular covariant representation of (A, G, α). In case A = C we recover the construction of the maximal and reduced group algebras C ∗(G) = C ⋊u G and C ∗ r (G) = C ⋊r G of G, respectively. The constructions of maximal and reduced crossed product extend to functors be- tween categories with several sorts of morphisms (like equivariant ∗-homomorphisms or correspondences) and each one of these functors has its own special features. For instance the universal crossed-product functor always preserves short exact sequences of equivariant homomorphisms, while the reduced one only does if the group G is exact. On the other hand, the reduced crossed product always takes embeddings to embeddings, while the full one only does if the group G is amenable. More recently (e.g., see [4,6,8,25,26,31]), there has been a growing interest in the study of exotic (or intermediate) group C ∗-algebras C ∗ µ(G) and crossed products A ⋊α,µ G, where the word "intermediate" indicates that they "lie" between the universal and reduced group algebras or crossed products. To be more precise: 2010 Mathematics Subject Classification. 46L55, 46L08, 46L80. Key words and phrases. Exotic crossed products, Baum-Connes conjecture, correspondences, exotic group algebras. Supported by Deutsche Forschungsgemeinschaft (SFB 878, Groups, Geometry & Actions), by CNPq/CAPES -- Brazil, and by the US NSF (DMS 1401126). 1 2 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT both classical crossed products can be obtained as completions of the convolution algebra Cc(G, A) with respect to the universal norm k·ku or the reduced norm k·kr ≤ k · ku, respectively. The intermediate crossed products A ⋊α,µ G are completions of Cc(G, A) with respect to a C ∗-norm k · kµ such that kf kr ≤ kf kµ ≤ kf ku for all f ∈ Cc(G, A). This is equivalent to saying that the identity on Cc(G, A) induces canonical surjective -homomorphisms A ⋊α,u G ։ A ⋊α,µ G ։ A ⋊α,r G. The main goal of this paper is to study functors (A, α) 7→ A ⋊α,µ G from the category of G-C ∗-algebras into the category of C ∗-algebras which assign to each G-algebra (A, α) an intermediate crossed product A ⋊α,µ G. There are several reasons why researchers became interested in intermediate crossed products and group algebras. On one side intermediate group algebras provide interesting new examples of C ∗-algebras attached to certain representa- tion theoretic properties of the group. Important examples are the group algebras C ∗ Ep(G) attached to the unitary representations of G which have a dense set of matrix coefficients in Lp(G), for 1 ≤ p ≤ ∞, or group algebras attached to other growth conditions on their matrix coefficients. It is shown in [6, Proposition 2.11] that for every discrete group G and 1 ≤ p ≤ 2 we have C ∗ r (G). But if G = F2 (or any discrete group which contains F2) then Okayasu shows in [31] that all C ∗ Ep(G) are different for 2 ≤ p ≤ ∞. Very recently, a similar result has been shown by Wiersma for SL(2, R) ([38]). As one important outcome of the results of this paper we shall see that for every K-amenable group G, like G = F2 or G = SL(2, R), all these different group algebras are KK-equivalent. On the other hand, it has been already observed in [4, Example 6.4] that there are exotic group algebra completions C ∗ µ(G) of G = F2 which do not have the same K-theory as r (G) or C ∗(G). This means that the K-theory of such algebras depends not only C ∗ on the structure of the group G, but also on the structure of the completion C ∗ µ(G), or more precisely, on the structure of the crossed-product functor A 7→ A ⋊µ G. Our results will help us to understand which properties a crossed-product functor should have in order to behave well with K-theory and other constructions. Ep(G) = C ∗ This point also brings us to another important reason for the growing interest on intermediate crossed products, which is motivated by a very recent new formulation of the Baum-Connes conjecture due to Baum, Guentner, and Willett in [4]. Recall that the Baum-Connes conjecture predicted that the K-theory of a reduced crossed product A ⋊α,r G can be computed with the help of a canonical assembly map asr (A,G) : K top ∗ (G; A) → K∗(A ⋊α,r G), ∗ in which K top (G; A), the topological K-theory of G with coefficient A, can be com- puted (at least in principle) by more classical topological methods. The original Baum-Connes conjecture stated that this assembly map should always be an iso- morphism. But in [21] Higson, Lafforgue and Skandalis provided counter examples for the conjecture which are based on a construction of non-exact groups due to Gromov and others. Recall that a group is called exact if every short exact sequence of G-algebras descends to a short exact sequence of their reduced crossed products. EXOTIC BAUM-CONNES CONJECTURE 3 In [4] it is shown that for every group G, there exists a crossed-product functor (A, α) 7→ A ⋊α,E G which is minimal among all exact and (in a certain sense) Morita compatible crossed-product functors for G. Using the assembly map for the universal crossed product A ⋊α,u G together with the canonical quotient map, we may construct an assembly map (1.1) asµ (A,G) : K top ∗ (G; A) → K∗(A ⋊α,µ G) for every intermediate crossed-product functor ⋊µ. Using E-theory, Baum, Guent- ner and Willett show that all known counter-examples to the conjecture disappear if we replace reduced crossed products by E-crossed products in the formulation of the conjecture. If G is exact, the E-crossed product will always be the reduced one, hence the new formulation of the conjecture coincides with the classical one for such groups. Note that the authors of [4] use E-theory instead of KK-theory since the proof requires a direct construction of the assembly map, which in the world of KK-theory would involve a descent homomorphism G : KK G(A, B) → KK(A ⋊E G, B ⋊E G) J E and it was not clear whether such descent exists (due to exactness and Morita compatibility, it exists in E-theory). It is one consequence of our results (see §8) that a KK-descent does exist for ⋊E at least if we restrict to separable systems. Indeed, in §4 we give a systematic study of crossed-product functors ⋊µ which are functorial for equivariant correspondences. This means that for every pair of G-algebras A, B and for every G-equivariant Hilbert A− B-bimodule F in the sense of Kasparov, there is a canonical crossed-product construction F ⋊µ G as a Hilbert A ⋊µ G − B ⋊µ G bimodule. We call such functors correspondence functors. We show in §4 that there are a number of equivalent conditions which characterise correspondence functors. For instance, it turns out (see Theorem 4.9) that being a correspondence functor is equivalent to the property that functoriality extends to G-equivariant completely positive maps. But maybe the most convenient of the conditions which characterise correspondence functors is the projection prop- erty which requires that for every G-algebra A and every G-invariant projection p ∈ M(A), the descent pAp ⋊µ G → A ⋊µ G of the inclusion pAp ֒→ A is faith- ful (see Theorem 4.9). This property is easily checked in special situations and we observe that many natural examples of crossed-product functors, including all Kaliszewski-Landstad-Quigg functors (or KLQ-functors) attached to weak-closed ideals in the Fourier-Stieltjes algebra B(G) (see 2.3 for details of the construction) are correspondence functors and that correspondence functors are stable under cer- tain manipulations which construct new functors out of old ones. In particular, for every given family of functors {⋊µi : i ∈ I} there is a construction of an infimum ⋊inf µ of this family given in [4], and it is easy to check that if all ⋊µi satisfy the projection property (i.e. are correspondence functors), then so does ⋊inf µ. It fol- lows from this and the fact (shown in [4]) that exactness is preserved by taking the infimum of a family of functors that there exists a minimal exact correspondence functor ⋊ECorr for each given locally compact group G. We show in Proposition 8.10 that this functor coincides with the minimal exact Morita compatible functor ⋊E of [4] if G is second countable and if we restrict the functor to separable G-algebras, as one almost always does if one discusses the Baum-Connes conjecture. 4 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT In §6 we prove that every correspondence functor will allow the construction of a descent J µ G : KK G(A, B) → KK(A ⋊µ G, B ⋊µ G) in equivariant KK-theory, which now allows one to use the full force of equivariant KK-theory in the study of the new formulation of the Baum-Connes conjecture in [4]. Moreover, using this descent together with the deep results of [20] we can show that the assembly map (1.1) is an isomorphism whenever G is a-T-menable (or, slightly more general, if G has a γ-element equal to 1G ∈ KK G(C, C) in Kasparov's sense) and if ⋊µ is a correspondence functor on G. By a result of Tu in [36], all such groups are K-amenable, and we show in Theorem 6.6 that for every K-amenable group G and every correspondence crossed-product functor ⋊µ on G the quotient maps A ⋊u G ։ A ⋊µ G ։ A ⋊r G are KK-equivalences (the proof closely follows the line of arguments given by Julg and Valette in [24, Proposition 3.4], but the main ideas are due to Cuntz in [14]). In particular, since all Lp-group algebras C ∗ Ep(G) for p ∈ [1, ∞] are group algebras corresponding to KLQ-functors, which are correspondence functors, they all have In particular this isomorphic K-theory for a given fixed K-amenable group G. holds for interesting groups like G = F2 or G = SL(2, R), where C ∗ Ep(G) are all different for p ∈ [2, ∞]. The outline of the paper is given as follows: after this introduction we start in §2 with some preliminaries on intermediate crossed-product functors and the dis- cussion of various known examples of such functors. In §3 we isolate properties governing how crossed products behave with respect to ideals and multipliers that will be useful later. In §4 we give many useful characterisations of correspondence functors: the main result here is Theorem 4.9. In §5 we show that crossed prod- ucts by correspondence functors ⋊µ always admit dual coactions. In particular it follows that the group algebras C ∗ µ(G) := C ⋊µ G corresponding to any correspon- dence functor carries a dual coaction of G and hence corresponds to a translation invariant weak- closed ideal E ⊆ B(G) as described in §3. We also prove a version of Imai-Takai duality for the dual crossed-product. Then, in §6 we show that every correspondence functor allows a KK-descent and, as an application, we use this to show the isomorphism of the assembly map for correspondence functors on a-T- menable groups and the results on K-amenability. In §7 we give a short discussion on the Lp-group algebras and applications of our results to such algebras. In §8 we show that being a correspondence functor passes to the infimum of a family of correspondence functors and we study the relation of the minimal exact correspon- dence functor to the functor ⋊E constructed in [4]. Finally in §9 we finish with some remarks and open questions. Part of the work on this paper took place during visits of the second author to the Federal University of Santa Catarina, and of the third author to the Westfälische Wilhelms-Universität, Münster. We would like to thank these institutions for their hospitality. EXOTIC BAUM-CONNES CONJECTURE 5 2. Preliminaries on crossed-product functors Let (B, G, β) be a C ∗-dynamical system. By a covariant representation (π, u) into the multiplier algebra M(D) of some C ∗-algebra D, we understand a ∗-homo- morphism π : B → M(D) together with a strictly continuous ∗-homomorphism u : G → UM(D) such that π(βs(b)) = usπ(b)u∗ If E is a Hilbert module (or s. Hilbert space) then a covariant representation on E will be the same as a covariant representation into M(K(E)) ∼= L(E). We say that a covariant representation is nondegenerate if π is nondegenerate in the sense that π(B)D = D. If (B, G, β) is a C ∗-dynamical system, then Cc(G, B) becomes a ∗-algebra with respect to the usual convolution and involution: f ∗ g(t) :=ZG f (s)βs(g(s−1t)) ds and f ∗(t) := ∆(t)−1βt(f (t−1))∗. The full (or universal) crossed product B ⋊β,u G is the completion of Cc(G, B) with respect to the universal norm kf ku := sup(π,u) kπ ⋊ u(f )k in which (π, u) runs through all covariant representations of (B, G, β) and π ⋊ u(f ) =ZG π(f (s))us ds denotes the integrated form of a covariant representation (π, u). By definition of this norm, each integrated form π ⋊ u extends uniquely to a ∗-representation of B ⋊β,u G and this extension process gives a one-to-one correspondence between the nondegenerate covariant ∗-representations (π, u) of (B, G, β) and the nondegenerate ∗-representations of B ⋊β,u G. There is a canonical representation (ιB, ιG) of (B, G, β) into M(B ⋊β,u G) which is determined by (cid:0)ιB(b) · f(cid:1)(t) = bf (t), and (cid:0)ιG(s) · f(cid:1)(t) = βs(f (s−1t)), for all b ∈ B, s ∈ G, and f ∈ Cc(G, B). Then, for a given nondegenerate ∗-homomorphism Φ : B ⋊β,u G → M(D) we have Φ = π ⋊ u for π = Φ ◦ iB and u = Φ ◦ iG. In this sense we get the identity B ⋊β,u G = iB ⋊ iG(B ⋊β,u G). If B = C, then we recover the full group algebra C ∗(G) = C ⋊id,u G and since every nondegenerate covariant representation of (C, G, id) is of the form (1, u) for some unitary representation u of G, this gives the well-known correspondence be- tween unitary representations of G and nondegenerate representations of C ∗(G). In this case we denote the canonical representation of G into UM(C ∗(G)) by uG. For a locally compact group G, let λG : G → U(L2(G)) denote the left regu- lar representation given by (λG(s)ξ)(t) = ξ(s−1t) and let M : C0(G) → L(L2(G)) denote the representation of C0(G) by multiplication operators. Then, for each sys- tem (B, G, β), there is a canonical covariant homomorphism (ΛB, ΛG) into M(B ⊗ K(L2(G))) given as follows: Let β : B → Cb(G, B) ⊆ M(B ⊗ C0(G)) be the ∗-homomorphism which sends b ∈ B to (cid:0)s 7→ βs−1(b)(cid:1) ∈ Cb(G, B). Then (ΛB, ΛG) :=(cid:0)(idB ⊗ M ) ◦ β, 1B ⊗ λG). The reduced crossed product B ⋊β,r G is defined as the image of B ⋊β,u G under the integrated form Λ := ΛB ⋊ ΛG. Note that Λ is always faithful on the dense subalgebra Cc(G, B) of B ⋊β,u G, so that we may regard B ⋊β,r G also as the (2.1) 6 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT completion of Cc(G, B) given by the norm kf kr = kΛ(f )k for f ∈ Cc(G, B). B = C, we get C ⋊r G = λG(C ∗(G)) = C ∗ If r (G), the reduced group C ∗-algebra of G. 2.1. Exotic crossed products. We shall now consider general crossed-product functors (B, β) 7→ B ⋊β,µ G such that the µ-crossed product B ⋊β,µ G can be obtained as some C ∗-completion of the convolution algebra Cc(G, B). We al- ways want to require that the µ-crossed products lie between the full and reduced crossed products in the sense that the identity map on Cc(G, B) induces surjective ∗-homomorphisms B ⋊β,u G ։ B ⋊β,µ G ։ B ⋊β,r G for all G-algebras (B, β). Such crossed-product functors have been studied quite recently by several authors, and we shall later recall the most prominent construc- tions as discussed in [4, 6, 8, 9, 25]. Note that one basic requirement will be that this construction is functorial for G-equivariant ∗-homomorphisms, i.e., whenever we have a G-equivariant ∗-homomorphism φ : A → B between two G-algebras A and B, there will be a ∗-homomorphism φ ⋊µ G : A ⋊µ G → B ⋊µ G given on the dense subalgebra Cc(G, A) by f 7→ φ ◦ f . A, iµ For any crossed-product functor (A, α) 7→ A ⋊α,µ G there is a canonical covariant homomorphism (iµ G) of (A, G, α) into M(A ⋊α,µ G) given by the composition of the canonical covariant homomorphism (iA, iG) into the universal crossed product followed by the quotient map qA,µ : A ⋊α,u G → A ⋊α,µ G. If we compose it with the quotient map qµ G) into the reduced crossed product A ⋊α,r G. Since the latter are known to be injective on A and G, the same is true iµ A : A → M(A ⋊α,µ G) is nondegenerate, it uniquely extends to M(A) (compare with [4, Lemma 4.2]). r : A ⋊α,µ G → A ⋊α,r G, this will give the inclusions (ir A and iµ G. Moreover, since iµ A, ir Let us briefly discuss some particular examples of exotic crossed-product functors which have been introduced in the recent literature (e.g., see the appendix of [4]). For the discussion recall that if π : A → C and ρ : A → D are two ∗-homomorphisms of a given C ∗-algebra A into C ∗-algebras C and D, then π is said to be weakly contained in ρ, denoted π(cid:22)ρ, if ker ρ ⊆ ker π, or, equivalently, if kπ(a)k ≤ kρ(a)k for all a ∈ A. The homomorphisms π and ρ are called weakly equivalent, denoted π ≈ ρ if their kernels coincide, i.e., if π (cid:22) ρ and ρ (cid:22) π. Similarly, if {ρi : i ∈ I} is any collection of ∗-homomorphisms, we say that π is weakly contained in this collection, if kπ(a)k ≤ supi∈I kρi(a)k. One easily checks that in case of ∗-representations on Hilbert spaces, this coincides with the notion of weak containment as introduced by Fell in [17]. If G is a group and u, v are unitary representations of G, we write u (cid:22) v if and only if this holds for their integrated forms on C ∗(G). 2.2. Brown-Guentner crossed products. The Brown-Guentner crossed prod- ucts (or BG-crossed products for short) have been introduced by Brown and Guent- ner in [6] and can be described as follows: Let G be a locally compact group and fix any unitary representation v of G which weakly contains the regular representation λG. Then C ∗ v (G) := v(C ∗(G)) is a C ∗-algebra which lies between the maximal and reduced group algebras of G. Clearly, any such algebra is of this form for some v and is called an exotic group algebra of G. Suppose now that (B, G, β) is a system. Let Iβ,v := ∩{ker(π ⋊ u) : u (cid:22) v} ⊆ B ⋊β,u G. EXOTIC BAUM-CONNES CONJECTURE 7 Then B ⋊β,vBG G := (B ⋊β,u G)/Iβ,v is called the BG-crossed product corresponding to v. Note that if B = C, we recover the exotic group algebra C ∗ v (G) up to isomorphism. It is very easy to check that this always defines a crossed-product functor. Note that by the definition of the notion of weak containment, the BG-crossed product v (G) ∼= only depends on the ideal ker v ⊆ C ∗(G), i.e., on the exotic group algebra C ∗ C ∗(G)/ ker v. Notice that for the trivial coefficient B = C, we get C ⋊vBG G = C ∗ v (G). We refer to [4, 6] for more information on BG-crossed-product functors. Example 2.2 (cf.[4, Lemma A.6]). Let us consider the BG-crossed product corre- sponding to the regular representation v = λG. Although the corresponding group algebra C ∗ v (G) is the reduced group algebra, the BG-functor corresponding to λG does not coincide with the reduced crossed-product functor if G is not amenable. To see this consider the action α = AdλG of G on K(L2(G)). Then the BG- crossed product K(L2(G)) ⋊α,λBG G is the full crossed product K(L2(G)) ⋊α,u G ∼= C ∗(G) ⊗ K(L2(G)) since (1C ∗(G) ⊗ idK) ⋊ (uG ⊗ λ) is an isomorphism between K(L2(G)) ⋊α,u G and C ∗(G) ⊗ K(L2(G)) and uG ⊗ λ ≈ λ. It therefore differs from K(L2(G)) ⋊α,r G ∼= C ∗ r (G) ⊗ K(L2(G)) if G is not amenable. Moreover, since α is Morita equivalent to the trivial action id on C via (L2(G), λ) and since C ⋊λBG G = C ∗ r (G), we see that BG-crossed products do not preserve Morita equivalence! 2.3. Kaliszewski-Landstad-Quigg crossed products. The Kaliszewski-Land- stad-Quigg crossed products (or KLQ-crossed products for short) have been in- troduced by Kaliszewski, Landstad and Quigg in [25], and discussed since then in several papers (e.g., see [4,8,9,26]). The easiest way to introduce them is the follow- ing. Choose any unitary representation v of G and consider the universal covariant homomorphism (iB, iG) from (B, G, β) into M(B ⋊β,u G). Consider the covariant representation (iB ⊗ 1C ∗ v (G)). Let v (G), iG ⊗ v) of (B, G, β) into M(B ⋊β,u G ⊗ C ∗ Jβ,v := ker(cid:0)(iB ⊗ 1C ∗ v (G)) ⋊ (iG ⊗ v)(cid:1) ⊆ B ⋊β,u G. Then the KLQ-crossed product corresponding to v is defined as B ⋊β,vKLQ G := (B ⋊β,u G)/Jβ,v. A different characterisation of these KLQ-crossed products can be obtained by considering the dual coaction bβ := (iB ⊗ 1C ∗(G)) ⋊ (iG ⊗ uG) : B ⋊β,u G → M(B ⋊β,u G ⊗ C ∗(G)). It follows right from the definition that Jβ,v = ker(idB⋊G ⊗ v) ◦ bβ. Moreover, by the properties of minimal tensor products, this kernel only depends on the kernel ker v ⊆ C ∗(G), so that the KLQ-crossed product also only depends on the quotient v (G) ∼= C ∗(G)/ ker v. C ∗ Let us see what happens if we apply this to B = C. Using C ⋊u G ∼= C ∗(G), the construction gives C ⋊vKLQ G = C ∗(G)/ ker(uG ⊗ v) which equals C ∗ v (G) = C ∗(G)/ ker(v) if and only if v is weakly equivalent to uG ⊗ v. In this case we say that v absorbs the universal representation of G. Since bβ = (iB ⊗1C ∗(G))⋊(iG ⊗uG) is faithful on B ⋊α,u G, i.e., it is weakly equivalent to iB ⋊ iG, it follows that (iB ⊗ 1C ∗ v (G)) ⋊ (iG ⊗ uG ⊗ v). This follows from the properties of the minimal tensor product together with the fact that for any covariant representation (π, U ) of (B, G, β), we have (π ⊗1) ⋊(U ⊗v) = v (G)) ⋊ (iG ⊗ v) is weakly equivalent to (iB ⊗ 1C ∗(G) ⊗ 1C ∗ 8 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT ((π ⋊ U ) ⊗ v) ◦ bβ. But this implies that Jβ,v = Jβ,uG⊗v, hence B ⋊β,vKLQ G ∼= B ⋊β,(uG⊗v)KLQ G. Thus one can always restrict attention to representations v which absorb uG when considering KLQ-crossed products. It is proved in [8] that (B, G, β) 7→ B ⋊β,vKLQ G is a crossed-product functor. In [9] it is proved that functoriality even extends to (equivariant) correspondences. We will have much more to say about this property in §4 below. Another way to look at KLQ-functors is via the one-to-one correspondence be- tween ideals in C ∗(G) and G-invariant (for left and right translation) weak-closed subspaces of the Fourier-Stieltjes algebra B(G), which is the algebra of all bounded continuous functions on G which can be realised as matrix coefficients s 7→ hUsξ ηi of strongly continuous unitary Hilbert-space G-representations U . Then B(G) iden- tifies with the Banach-space dual C ∗(G)∗ if we map the function s 7→ hUsξ ηi to the linear functional on C ∗(G) given by x 7→ hU (x)ξ ηi, where we use the same letter for U and its integrated form. It is explained in [25] that the one-to-one corre- spondence between closed ideals in C ∗(G) and G-invariant weak-closed subspaces E in B(G) is given via E 7→ IE := ⊥E = {x ∈ C ∗(G) : f (x) = 0 ∀f ∈ E}. Hence, we obtain all quotients of C ∗(G) as C ∗ E(G) := C ∗(G)/IE for some E. The ideal IE will be contained in the kernel of the regular representation λG if and only if E contains the Fourier-algebra A(G) consisting of matrix coefficients of λG. Moreover, the quotient map vE : C ∗(G) → C ∗ E(G) absorbs the universal representation uG if and only if E is an ideal in B(G). As a consequence of this discussion we get: Proposition 2.3. Let G be a locally compact group. Then (1) there is a one-to-one correspondence between BG-crossed-product functors and G-invariant weak-closed subspaces E of B(G) which contain the Fourier algebra A(G) given by sending E to the BG-functor associated to the quo- tient map vE : C ∗(G) → C ∗ E(G). (2) Similarly, there is a one-to-one correspondence between KLQ-crossed-product functors and weak-closed ideals E in B(G) given by sending E to the KLQ- functor associated to the quotient map vE : C ∗(G) → C ∗ E(G). As for BG-crossed-product functors, we get C ⋊vKLQ G = C ∗ v (G), the exotic group C ∗-algebra for the trivial coefficient algebra B = C if v absorbs the universal representation uG, that is, if v = vE for some G-invariant ideal in B(G) as above. This is not true if E is not an ideal (compare also with [9, Proposition 2.2]). Example 2.4. For each p ∈ [2, ∞] we define Ep to be the weak-closure of B(G) ∩ Lp(G), which is an ideal in B(G). Let us write (B, β) 7→ B ⋊β,pKLQ G for the corresponding KLQ-functor. It then follows from [31] that for any discrete group G which contains the free group F2 with two generators, the group algebras C ∗ Ep(G) are all different. Since C ⋊pKLQ G = C ∗ Ep(G) it follows from this that for such groups G the KLQ-crossed-product functors associated to different p are also different. Note that for p = 2 we just get the reduced crossed-product functor and for p = ∞ we get the universal crossed-product functor as the associated KLQ-functors. The corresponding BG-crossed-product functors (B, β) 7→ B ⋊β,pBG G will also be different for 2 ≤ p ≤ ∞ because we also have C ⋊pBG G = C ∗ Ep(G). But Example 2.2 shows that these functors produce different results (and have different properties) for non-trivial coefficients. EXOTIC BAUM-CONNES CONJECTURE 9 2.4. New functors from old functors. It is shown in [4] that given any crossed- product functor (B, β) 7→ B ⋊β,µ G and any unital G-C ∗-algebra (D, δ), one can use D to construct a new functor ⋊µν by defining D B ⋊β,µν D G := jB ⋊µ G(B ⋊β,µ G) ⊆ (B ⊗ν D) ⋊β⊗δ,µ G, where ⊗ν either denotes the minimal or the maximal tensor product, and jB : B → B ⊗ν D, jB(b) = b ⊗ 1D denotes the canonical imbedding. A particularly interesting example is obtained in the case where G is discrete by taking D = ℓ∞(G) and ⋊µ = ⋊u, the universal crossed product. For general locally compact groups this could be replaced by the algebra U Cb(G) of uniformly continuous bounded functions on G: we will discuss this further in Section §9.1. For discrete G, this functor will coincide with the reduced crossed-product functor if and only if G is exact. Later, to non-unital C ∗-algebras in Corollary 4.20, we shall extend the definition of ⋊µν D. D 2.5. A general source for (counter) examples. The following general construc- tion is perhaps a little unnatural. It is, however, a very useful way to produce 'counterexamples': examples of crossed-product functors that do not have good properties. Let S be a collection of G-algebras, and for any given G-algebra (A, α), let Φ(A, S) denote the class of G-equivariant ∗-homomorphisms from A to an element of S. Now define A ⋊α,S G to be the completion of Cc(G, A) for the norm kf k := maxnkf kA⋊α,rG , sup (φ:A→B)∈Φ(A,S) kφ ◦ f kB⋊uGo; this is a supremum of C ∗-algebra norms, so a C ∗-algebra norm, and it is clearly between the reduced and maximal completions of Cc(G, A). Lemma 2.5. The collection of completions A ⋊S G defines a crossed-product func- tor. Proof. This follows immediately from the fact that if ψ : A → B is any G-equivariant ∗-homomorphism and φ : B → C is an element of Φ(B, S), then ψ ◦ φ is a member of Φ(A, S). (cid:3) Here is an example of the sort of 'bad property' the functors above can have. We will look at some similar cases in Example 3.5 and Example 8.12 below. Example 2.6. For a crossed-product functor ⋊µ, write C ∗ µ(G) for the exotic group algebra C ⋊id,µ G, and for a C ∗-algebra A let M2(A) denote the C ∗-algebra of 2 × 2 matrices over A. If ⋊µ is either a KLQ- or a BG-crossed product then it is not difficult to check that (2.7) M2(C ∗ µ(G)) ∼= M2(C) ⋊id,µ G. However, for any non-amenable G there is a crossed product where this isomorphism fails! Indeed, in the notation above take S = {(C, id)} and ⋊µ = ⋊S. Then M2(C ∗ µ(G)) = M2(C ∗(G)) but M2(C) ⋊id,µ G = M2(C ∗ r (G)) (the latter follows as M2(C) has no ∗-homomorphisms to C). As G is amenable if and only if C ∗ r (G) admits a non-zero finite dimensional representation, and as C ∗(G) always admits a non-zero one-dimensional representation, the isomorphism in line (2.7) is impossible for this crossed product and any non-amenable G. 10 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT 3. The ideal property and generalised homomorphisms In this section, we study two fundamental properties that a crossed product may or may not have: the ideal property, and functoriality under generalised homomor- phisms. In the next section we will use these ideas to investigate correspondence functors. Definition 3.1. A crossed-product functor (A, α) 7→ A ⋊α,µ G is said to be functo- rial for generalised homomorphisms if for any (possibly degenerate) G-equivariant ∗-homomorphism φ : A → M(B) there exists a ∗-homomorphism φ⋊µ G : A⋊µG → M(B ⋊µ G) which is given on the level of functions f ∈ Cc(G, A) by f 7→ φ ◦ f in the sense that φ ⋊µ G(f )g = (φ ◦ f ) ∗ g for all g ∈ Cc(G, B). An alternative (and equivalent) definition of functoriality for generalised homo- morphisms would be to ask that for every G-equivariant ∗-homomorphism φ : A → M(B) and every ∗-representation π ⋊ u of B ⋊β,u G which factors through B ⋊β,µ G, the representation (π ◦ φ) ⋊ u factors through A ⋊α,µ G. It follows from the definition that if a crossed-product functor is functorial for generalised homomorphisms, then it will automatically send nondegenerate G-equivariant ∗-homomorphisms φ : A → M(B) to nondegenerate ∗-homomorphisms φ ⋊µ G : A ⋊µ G → M(B ⋊µ G), since in this case φ ⋊µ G(Cc(G, A))Cc(G, B) will be inductive limit dense in Cc(G, B). Note also that whenever the composition of two generalised G-equivariant morphisms φ : A → M(B) and ψ : B → M(D) makes sense -- e.g., if the image of φ lies in B or if ψ is nondegenerate (in which case it uniquely extends to M(B)), we get (ψ ◦ φ) ⋊µ G = (ψ ⋊µ G) ◦ (φ ⋊µ G), since both morphisms agree on the dense subalgebra Cc(G, A) ⊆ A ⋊α,µ G. We shall see below that functoriality for generalised homomorphisms is equivalent to the following ideal property: Definition 3.2. A crossed-product functor (A, α) 7→ A ⋊α,µ G satisfies the ideal property, if for every G-invariant closed ideal in a G-algebra A, the inclusion map ι : I ֒→ A descends to an injective -homomorphism ι ⋊ G : I ⋊µ G ֒→ A ⋊µ G. Lemma 3.3. Let (A, α) 7→ A ⋊α,µ G be a crossed-product functor. Then the following are equivalent: (1) (A, α) 7→ A ⋊α,µ G is functorial for generalised homomorphisms; (2) (A, α) 7→ A ⋊α,µ G is functorial for nondegenerate generalised homomor- phisms; (3) (A, α) 7→ A ⋊α,µ G has the ideal property. Proof. (1) ⇒ (2) is trivial. (2)⇒ (3): Let A be a G-algebra and I ⊆ A a G-invariant closed ideal. Let ι : I ֒→ A denote the inclusion map. By functoriality, we get a ∗-homomorphism ι ⋊µ G : I ⋊µ G → A ⋊µ G. Let φ : A → M(I) be the canonical map, which is a nondegenerate G-equivariant ∗-homomorphism. By (2), it induces a (nondegenerate) ∗-homomorphism φ ⋊µ G : A ⋊µ G → M(I ⋊µ G). Notice that φ◦ι : I → M(I) is the canonical inclusion map, which is a nondegenerate generalised homomorphism, so it induces a ∗-homomorphism (φ◦ι) ⋊µ G : I ⋊µ G → M(I ⋊µ G). It is then easily checked on Cc(G, I) that (φ ◦ ι) ⋊µ G coincides with the the canoni- cal embedding I ⋊µ G ֒→ M(I ⋊µ G). In particular, (φ◦ι) ⋊µ G = (φ ⋊µ G)◦(ι ⋊µ G) EXOTIC BAUM-CONNES CONJECTURE 11 is injective, which implies the injectivity of ι ⋊µ G, as desired. (3) ⇒ (1): Assume now that the µ-crossed-product functor has the ideal property and let A and B be G-algebras and φ : A → M(B) a G-equivariant ∗-homomorphism. Let Mc(B) = {m ∈ M(B) : s 7→ βs(m) is norm continuous} be the subalgebra of M(B) of G-continuous elements. Then φ takes values in Mc(B) and by functorial- ity we see that we get a ∗-homomorphism φ ⋊µ G : A ⋊µ G → Mc(B) ⋊µ G. Now by the ideal property, we also have a faithful inclusion ι ⋊µ G : B ⋊µ G → Mc(B) ⋊µ G, hence we may regard B ⋊µ G as an ideal in Mc(B) ⋊µ G. But this implies that there is a canonical ∗-homomorphism Φ : Mc(B) ⋊µ G → M(B ⋊µ G). It is then straightforward to check that the composition Φ ◦ (φ ⋊µ G) : A ⋊µ G → M(B ⋊µ G) is given on functions in Cc(G, A) by sending f ∈ Cc(G, A) to φ ◦ f ∈ Cc(G, M(B)) as required. (cid:3) Remark 3.4. It is easy to see directly that all KLQ- and all BG-crossed products have the ideal property. Recall from [4] that a crossed-product functor A 7→ A ⋊µ G is exact if it preserves short exact sequences, that is, if every short exact sequence 0 → I → A → B → 0 of G-C ∗-algebras induces a short exact sequence 0 → I ⋊µ G → A ⋊µ G → B ⋊µ G → 0 of C ∗-algebras. In particular, all exact functors satisfy the ideal property, but the ideal property alone is far from being enough for exactness since the reduced crossed- product functor always satisfies the ideal property. By definition, the group G is exact if A 7→ A ⋊r G is exact. Example 3.5. Every non-amenable locally compact group admits a crossed-product functor which does not satisfy the ideal property. To see this, we use the construc- tion of Section 2.5. Let S be the collection of G-algebras consisting of only C0([0, 1)) equipped with the trivial action id. For each G-algebra (A, α) we define A ⋊α,S G as the completion of Cc(G, A) with respect to the norm kf kS := max{kf kr, sup φ kφ ◦ f ku}, where the supremum is taken over all G-equivariant ∗-homomorphisms φ : A → C0([0, 1)). As explained in Section 2.5, this is a crossed-product functor. To see that it does not have the ideal property, consider the short exact sequence of (trivial) G-algebras 0 → C0([0, 1)) → C([0, 1]) → C → 0. Observe that A ⋊µ G = A ⋊r G for every unital G-algebra A since there is no non- zero homomorphism A → C0([0, 1)). In particular, C([0, 1]) ⋊µ G = C([0, 1]) ⋊r G ∼= C([0, 1]) ⊗ C ∗ r (G). On the other hand, we obviously have C0([0, 1)) ⋊µ G = C0([0, 1)) ⋊u G ∼= C([0, 1)) ⊗ C ∗(G). Hence, if G is not amenable, the canonical map ι ⋊µ G : C0([0, 1)) ⋊µ G → C([0, 1]) ⋊µ G will not be injective. For crossed-product functors which enjoy the ideal property we have the impor- tant result: Lemma 3.6. Let (A, α) 7→ A ⋊α,µ G be a crossed-product functor with the ideal property and let X be a locally compact Hausdorff space. Let (jA, jX ) denote the canonical inclusions of A and C0(X) into M(A ⊗ C0(X)), respectively and let iA⊗C0(X) : A ⊗ C0(X) → M(cid:0)(A ⊗ C0(X)) ⋊α⊗idX ,µ G(cid:1) be the canonical map. Then ∼=−→ (A ⊗ C0(X)) ⋊α⊗idX ,µ G ΨX := (jA ⋊µ G) × (iA⊗C0(X) ◦ jX ) : A ⋊α,µ G ⊗ C0(X) ΨXy ΨX∞y∼= 12 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT is an isomorphism. Remark 3.7. We should remark that if X is compact, the result holds for any crossed-product functor ⋊µ. The proof of this fact is given for the case X = [0, 1] in [4, Lemma 4.3], but the same arguments will work for arbitrary compact Hausdorff spaces as well. We will use this fact in the proof below. Proof. Let X∞ = X ∪ {∞} denote the one-point compactification of X. Then the above remark implies that the lemma is true for X∞. Consider the diagram (A ⊗ C0(X)) ⋊µ G −−−−→ (A ⊗ C(X∞)) ⋊µ G A ⋊α,µ G ⊗ C0(X) −−−−→ A ⋊α,µ G ⊗ C(X∞) in which the horizontal maps are induced by the canonical inclusion C0(X) ֒→ C(X∞). One checks on elements of the form f ⊗g with f ∈ C0(X) and g ∈ Cc(G, A) that the diagram commutes and that ΨX has dense image in A ⋊α,µ G ⊗ C0(X). By the ideal property, the upper horizontal map is injective, which then implies that ΨX is injective as well. (cid:3) 4. Correspondence functors In this section we continue our study of properties that a crossed-product func- tor may or may not have. One particularly important property we would like to have is that the functor extends to a functor on the G-equivariant correspondence category, in which the objects are G-algebras and the morphisms are G-equivariant correspondences. The importance of this comes from the fact that functoriality for correspondences allows the construction of a descent in equivariant KK-theory J µ G : KK G(A, B) → KK(A ⋊µ G, B ⋊µ G) for our exotic functor ⋊µ; we discuss this in §6. The main result of this section is Theorem 4.9, which gives several equivalent characterisations of functors that extend to the correspondence category. We start by recalling some background about correspondences. Definition 4.1. If A and B are C ∗-algebras, a correspondence from A to B (or simply an A − B correspondence) is a pair (E, φ) consisting of a Hilbert B-module E together with a (possibly degenerate) ∗-homomorphism φ : A → L(E). If (A, α) and (B, β) are G-algebras, then a G-equivariant A − B correspondence (E, φ, γ) consists of an A − B correspondence (E, φ) together with a strongly contin- uous action γ : G → Aut(E) such that hγs(ξ) γs(η)iB = βs(hξ ηiB ), γs(ξ · b) = γs(ξ)βs(b) and γs(φ(a)ξ) = φ(αs(a))γs(ξ) for all a ∈ A, b ∈ B, ξ, η ∈ E and s ∈ G. Definition 4.2. We say that two G-equivariant A − B correspondences (E, φ, γ) and (E ′, φ′, γ′) are isomorphic if there exists an isomorphism U : E → E ′ which preserves all structures. We say that (E, φ, γ) and (E ′, φ′, γ′) are equivalent if there exists an isomorphism between (φ(A)E, φ, γ) and (φ′(A)E, φ′, γ′). In particular, every correspondence (E, φ, γ) is equivalent to the nondegenerate correspondence (φ(A)E, φ, γ). EXOTIC BAUM-CONNES CONJECTURE 13 We should note at this point that φ(A)E = {φ(a)ξ : a ∈ A, ξ ∈ E} will be a closed G-invariant Hilbert B-submodule of E -- just apply Cohen's factorisation theorem to the closed submodule span φ(A)E to see that this must coincide with φ(A)E. Note that allowing general (i.e., possibly degenerate) correspondences makes it more straightforward to view an arbitrary ∗-homomorphism φ : A → B as a corre- spondence: it will be represented by the correspondence (B, φ) (or (B, φ, β) in the equivariant setting) where B is regarded as a Hilbert B-module in the canonical way. Of course, by our definition of equivalence, φ : A → B will also be represented by the correspondence (φ(A)B, φ). Composition of an (A, α) − (B, β) correspondence (E, φ, γ) with a (B, β) − (D, δ) correspondence (F , ψ, τ ) is given by the internal tensor product construction (E ⊗ψ F , φ ⊗ 1, γ ⊗ τ ) (or just (E ⊗ψ F , φ ⊗ 1) in the non-equivariant case). The following lemma is folklore and we omit a proof. Lemma 4.3. The construction of internal tensor products is associative up to isomorphism of correspondences. Moreover, we have canonical isomorphisms and (φ(A)E) ⊗ψ F ∼= (φ ⊗ 1)(A)(E ⊗ψ F ). E ⊗ψ F ∼= E ⊗ψ (ψ(B)F ) The above lemma together with [16, Theorem 2.8] allows the following definition: Definition 4.4. There is a unique category Corr(G), which we call the G-equivariant correspondence category, in which the objects are G-C ∗-algebras and the morphisms between two objects (A, α) and (B, β) are equivalence classes of (A, α) − (B, β) correspondences (E, φ, γ) with composition of morphisms given by internal tensor products. We write Corr for the correspondence category of C ∗-algebras without group actions (i.e., where G = {e} is the trivial group). By our definition of equivalence of correspondences together with the above lemma, our category Corr(G) is isomorphic to the category A(G) defined in [16, The- orem 2.8] in which the objects are C ∗-algebras and the morphisms are isomorphism classes of nondegenerate G-equivariant correspondences. More precisely, the isomor- phism Corr(G) ∼−→ A(G) is given by the assignment [(E, φ, γ)] → [(φ(A)E, φ, γ)]. For every object (A, α) of Corr(G), the identity morphism on (A, α) is given by the equivalence class of the correspondence (A, idA, α). It has been shown in [16, Lemma 2.4 and Remark 2.9] that the invertible morphisms in Corr(G) ∼= A(G) are precisely the equivalence classes of equivariant Morita-equivalence bimodules (which we simply call equivalence bimodules below). We now want to discuss under what conditions one can extend a crossed-product functor (A, α) → A ⋊α,µ G, which is functorial for G-equivariant ∗-homomorphisms to a functor from the correspondence category Corr(G) to the correspondence cat- egory Corr = Corr({e}). For this it is necessary to say what the functor should do on morphisms. For this recall that if (E, φ, γ) is a G-equivariant correspondence from (A, α) to (B, β), then there is a canonical construction of a "correspondence" (Cc(G, E), φ ⋊c G) on the level of continuous functions with compact supports where the actions and inner products are given by hx yiCc(G,B)(t) :=ZG (x · ϕ)(t) :=ZG βs−1 (hx(s) y(st)iB ) ds x(s)βs(ϕ(s−1t)) ds 14 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT for all x, y ∈ Cc(G, E), ϕ ∈ Cc(G, B), and (4.5) (φ ⋊c G(f )x)(t) :=ZG φ(f (s))γs(x(s−1t)) ds, for all x ∈ Cc(G, E) and f ∈ Cc(G, A). It is well known (e.g., see [11, 27]) that this construction completes to a correspondence (E ⋊γ,u G, φ ⋊u G) between the maximal crossed products A ⋊α,u G and B ⋊β,u G. We simply regard hx yiCc(G,B) as an element of B ⋊β,u G and take completion with respect to the corresponding norm kxku :=pkhx xiCc(G,B)ku on Cc(G, E). On the right hand side we can do a similar procedure in complete generality by regarding hx yiCc(G,B) as an element in B ⋊β,µ G for any exotic crossed- product functor ⋊µ: the completion E ⋊γ,µ G of Cc(G, E) with respect to kxk := pkhx xiCc(G,B)kµ will always be a Hilbert B ⋊β,µ G-module. Moreover, we have a canonical isomorphism of Hilbert B ⋊β,µ G-modules: (4.6) (E ⊗γ,u G) ⊗B⋊β,uG (B ⋊β,µ G) ∼−→ E ⋊γ,µ G, sending x ⊗ b to x · b, where the universal crossed product acts on the µ-crossed product via the quotient map. The problem we need to address is the question whether the left action of Cc(G, A) on Cc(G, E) given by (4.5) will always extend to A ⋊α,µ G. We should also want that K(E) ⋊Adγ,µ G ∼= K(E ⋊γ,µ G) if the left action exists. Indeed, we shall see below (Remark 4.19) that this fails for all BG-functors different from the universal crossed product, which is the BG-functor corresponding to the full group algebra C ∗(G). On the other hand, it has been already shown in [9, Section 2] that all KLQ-functors extend to the correspondence category. We shall give a short alternative argument for this at the end of this section. We shall now introduce a few conditions that a functor may or may not have, which are related to these questions. Definition 4.7. Let (A, α) → A ⋊α,µ G be a crossed-product functor. Then we say (1) the functor is strongly Morita compatible if for every G-equivariant (A, α) − (B, β) equivalence bimodule (E, γ), the left action of Cc(G, A) on Cc(G, E) extends to an action of A ⋊α,µ G on E ⋊γ,µ G and makes E ⋊γ,µ G an A ⋊α,µ G − B ⋊β,µ G equivalence bimodule. (2) The functor has the projection property (resp. full projection property) if for every G-algebra A and every G-invariant (resp. full) projection p ∈ M(A), the inclusion ι : pAp ֒→ A descends to a faithful homomorphism ι ⋊µ G : pAp ⋊α,µ G → A ⋊α,µ G. (3) The functor has the hereditary-subalgebra property if for every hereditary G-invariant subalgebra B of A, the inclusion ι : B ֒→ A descends to a faithful map ι ⋊µ G : B ⋊α,µ G → A ⋊α,µ G. (4) The functor has the cp map property if for any completely positive and G-equivariant map φ : A → B of G-algebras, the map Cc(G, A) → Cc(G, B), f 7→ φ ◦ f extends to a completely positive map from A ⋊α,µ G to B ⋊β,µ G. Remark 4.8. The (full) projection property can be reformulated as follows: For every G-algebra A and for every G-invariant (full) projection p ∈ M(A), we have EXOTIC BAUM-CONNES CONJECTURE 15 pAp ⋊α,µ G = p(A ⋊α,µ G)p, where p denotes the canonical image of p in M(A ⋊α,µ G). This means that the closure of Cc(G, pAp) = pCc(G, A)p inside A ⋊α,µ G coincides with pAp ⋊α,µ G. Recall that the ideal property has been introduced in Definition 3.2. Since ideals are hereditary subalgebras it is weaker than the hereditary-subalgebra property. Recall also from Lemma 3.3 that the ideal property is equivalent to the property that the functor extends to generalised homomorphisms φ : A → M(B). Theorem 4.9. Let (A, α) → A ⋊α,µ G be a crossed-product functor. Then the following conditions are equivalent: (1) The functor extends to a correspondence functor from Corr(G) to Corr send- ing an equivalence class of a correspondence (E, φ, γ) to the equivalence class of the correspondence (E ⋊γ,µ G, φ ⋊µ G). (2) For every G-equivariant (right) Hilbert (B, β)-module (E, γ), the left action of K(E) on E descends to an isomorphism K(E) ⋊Adγ,µ G ∼= K(E ⋊γ,µ G). (3) The functor is strongly Morita compatible and has the ideal property. (4) The functor has the hereditary subalgebra property. (5) The functor has the projection property. (6) The functor has the cp map property. If the above equivalent conditions hold, we say that (A, α) 7→ A ⋊α,µ G is a corre- spondence crossed-product functor. We prepare the proof of Theorem 4.9 with some lemmas: Lemma 4.10. Suppose that the functor (A, α) → A ⋊α,µ G extends to a correspon- dence functor from Corr(G) to Corr. Then it has the ideal property. Proof. Let φ : A → M(B) be any G-equivariant generalised homomorphism from A to B. Consider the (A, α) − (B, β) correspondence (B, φ, β). It is clear that the µ-crossed product of B by G, if B is viewed as a Hilbert B-module, is equal to the C ∗-algebra-crossed product B ⋊µ G. Hence the left action of A ⋊µ G on the module exists if and only if our functor is functorial for generalised homomorphisms. By Lemma 3.3 this is equivalent to the ideal property. (cid:3) Lemma 4.11. Let (A, α) → A ⋊α,µ G be a crossed-product functor. Then the following are equivalent: (1) ⋊µ has the projection property. (2) ⋊µ has the full projection property and the ideal property. Proof. (2) ⇒ (1) Let p ∈ M(A) be a G-invariant projection. Then I := ApA is a G-invariant ideal and p can be viewed as a full G-invariant projection in M(I). Then pAp = pIp and the map ι ⋊µ G : pAp ⋊µ G → A ⋊µ G is the composition of the faithful inclusions pAp ⋊µ G ι⋊µG ֒→ I ⋊µ G ι⋊µG ֒→ A ⋊µ G. For the converse direction, assume that ⋊µ has the projection property. Let I be a G-invariant ideal in the G-algebra A. Consider the algebra L := (cid:18)I I A(cid:19) I with the obvious G-action. Then p = ( 1 0 0 1 ) are opposite G-invariant projections in M(L) and by the projection property we see that p and q map to 0 0 ) and q = ( 0 0 16 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT opposite projections p and q in M(L ⋊µ G) such that I ⋊µ G = p(L ⋊µ G)p and A ⋊µ G = q(L ⋊µ G)q. One easily checks that the right Hilbert A ⋊µ G-module p(L ⋊µ G)q is the completion I ⋊µ G of Cc(G, I) = Cc(G, pLq) with respect to the norm given by the inclusion Cc(G, I) ֒→ A ⋊µ G. On the other hand, regarding p(L ⋊µ G)q as a left Hilbert I ⋊µ G-module in the canonical way, it will be the completion of Cc(G, I) = Cc(G, pLq) with respect to the norm on I ⋊µ G. Hence I ⋊µ G = p(L ⋊µ G)q = I ⋊µ G and the result follows. (cid:3) The next lemma deals with various versions of Morita compatibility: Lemma 4.12. Let ⋊µ be a crossed-product functor for G. Then the following are equivalent: (1) For each full G-equivariant Hilbert (B, β)-module (E, γ) the left action of K(E) on E descends to an isomorphism K(E) ⋊Adγ,µ G ∼= K(E ⋊γ,µ G). (2) ⋊µ is strongly Morita compatible. (3) ⋊µ has the hereditary subalgebra property for full G-invariant hereditary subalgebras B ⊆ A ( i.e., span ABA = A). (4) ⋊µ has the full projection property. Proof. We prove (1) ⇔ (2) ⇒ (3) ⇒ (4) ⇒ (2). (1) ⇔ (2) follows from the fact that if (E, γ) is a G-equivariant (A, α) − (B, β) equivalence bimodule, then the left action of A on E induces a G-equivariant isomor- phism (A, α) ∼= (K(E), Adγ) and, conversely, that every full G-equivariant Hilbert B-module is a G-equivariant K(E) − B equivalence bimodule. (2) ⇒ (3): Recall that a closed subalgebra B ⊆ A is a full hereditary subalge- bra, if BAB ⊆ B and A = span ABA. Then BA will be a B − A-equivalence bimodule. Let B ⋊α,µ G := ι ⋊µ G(B ⋊β,µ G); this coincides with the closure of Cc(G, B) inside A ⋊α,µ G. Then a simple computation on the level of func- tions with compact supports shows that B ⋊α,µ G is a full hereditary subalgebra of A ⋊α,µ G. For functions f, g ∈ Cc(G, B) ∗ Cc(G, A) ⊆ Cc(G, BA) it is clear that the A ⋊α,µ G-valued inner products given by viewing f, g as elements of (B ⋊α,µ G)(A ⋊α,µ G) or as elements of the module (BA) ⋊µ G coincide. Thus we see that the inclusion Cc(G, B) ∗ Cc(G, A) ֒→ Cc(G, BA) extends to an isomor- phism (B ⋊α,µ G)(A ⋊α,µ G) ∼= (BA) ⋊µ G of Hilbert A ⋊α,µ G-modules. But this induces an isomorphism of the compact operators B ⋊α,µ G ∼= K((B ⋊β,µ G)(A ⋊α,µ G)) ∼= K((BA) ⋊µ G) ∼= B ⋊α,µ G, which extends the identity map on Cc(G, B), where the last isomorphism follows from the strong Morita invariance of our functor. (3) ⇒ (4): follows from the fact that if p ∈ M(A) is a full G-invariant projection, then pAp is a full G-invariant hereditary subalgebra of A. (4) ⇒ (2): Assume that (A, α) → A ⋊α,µ G satisfies the full projection property. Let (E, γ) be a (A, α) − (B, β) equivalence bimodule. Then we can form the linking algebra L =(cid:18) A E and q = (cid:18)0 0 0(cid:19) γ∗ β(cid:19). Then p =(cid:18)1 0 E ∗ B(cid:19) equipped with the action σ =(cid:18) α γ 1(cid:19) are opposite full projections in M(L) and by the projection 0 0 property we see that p and q map to opposite full projections p and q in M(L⋊σ,µG) EXOTIC BAUM-CONNES CONJECTURE 17 such that p(L ⋊σ,µ G)q is an equivalence bimodule between A ⋊α,µ G = p(L ⋊σ,µ G)p and B ⋊β,µ G = q(L ⋊σ,µ G)q. One easily checks that Cc(G, E) = Cc(G, pLq) equipped with the norm coming from the B ⋊β,µ G-valued inner product embeds isometrically and densely into p(L ⋊σ,µ G)q, hence we get E ⋊γ,µ G = q(L ⋊σ,µ G)q. Therefore E ⋊γ,µ G becomes an A ⋊α,µ G − B ⋊β,µ G equivalence bimodule. (cid:3) We are now ready for the Proof of Theorem 4.9. We prove (1) ⇒ (2) ⇒ (3) ⇒ (4) ⇒ (5) ⇒ (6) ⇒ (5) ⇒ (1). (1) ⇒ (2) Assume that (A, α) 7→ A ⋊α,µ G extends to a correspondence functor. Then by Lemma 4.10 we know that the functor has the ideal property. Let (E, γ) be a Hilbert (B, β)-module. Let I = spanhE EiB. Then the ideal property implies that viewing (E, γ) as a Hilbert (I, β)-module or as a Hilbert (B, β)-module does not change the induced norm k · kµ on Cc(G, E), and hence in both cases we get the same completion E ⋊γ,µ G. Thus we may assume without loss of generality that E is a full Hilbert B-module. But then the equivalence class of (E, γ) will be an isomorphism from (K(E), Adγ) to (B, β) in the equivariant correspondence category Corr(G), which by the properties of a functor must be sent to an isomorphism from K(E) ⋊Adγ,µ G to B ⋊β,µ G in Corr, which are equivalence classes of equivalence bimodules. Since the left action of K(E) ⋊Adγ,µ G on E ⋊γ,µ G is obviously non- degenerate, it follows that E ⋊γ,µ G is a K(E) ⋊Adγ,µ G − B ⋊β,µ G equivalence bimodule. This implies the desired isomorphism. (2) ⇒ (3) It follows clearly from (2) applied to full Hilbert modules that the functor (A, α) 7→ A ⋊α,µ G is strongly Morita compatible. To see that it has the ideal prop- erty, let I be a G-invariant closed ideal in A for some G-algebra (A, α). Regard (I, α) as a G-invariant Hilbert (A, α)-module in the canonical way. To avoid confusion, we write IA for the Hilbert A-module I. We then have (I, α) = (K(IA), Adγ). Now if the descent ι ⋊µ G : I ⋊µ G → A ⋊µ G is not injective, it induces a new norm k · kµ on Cc(G, I) which is (at least on some elements of Cc(G, I)) strictly smaller than the given norm k · kµ. It is then easily checked that IA ⋊µ G = I ⋊µ G. This also im- plies K(IA ⋊µ G) = I ⋊µ G which will be a proper quotient of K(IA) ⋊µ G = I ⋊µ G, which contradicts (2). (3) ⇒ (4) Let B ⊆ A be a G-invariant hereditary subalgebra of A. Then BA will be a B − ABA-equivalence bimodule, where ABA denotes the closed ideal of A generated by B. Since, by assumption, the µ-crossed-product functor satisfies the ideal property, we may assume without loss of generality that B is full in the sense that ABA = A. But then the result follows from part (2) ⇔ (4) of Lemma 4.12. (4) ⇒ (5) This follows from the fact that corners are hereditary subalgebras. (5) ⇒ (6) Let φ : A → B be a G-equivariant completely positive map between G-algebras; we may assume φ is non-zero. Multiplying φ by a suitable positive constant, we may assume that φ is contractive. It then follows from [7, Proposition 2.2.1] that the (unital) map defined on unitisations by eφ : eA → eB, eφ(a + z1A) = φ(a) + z1B is also completely positive and contractive; it is moreover clearly equivariant for the extended actions of G. 18 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT representation of G on H. Define a bilinear form on the algebraic tensor product Now, assume that eB is represented faithfully, non-degenerately, and covariantly on a Hilbert space H. Identify eB with its image in L(H), and write u for the given eA ⊙ H by ai ⊗ ξi, mXj=1 bj ⊗ ηjE :=Xi,j i bj)ηji. hξi,eφ(a∗ As eφ is completely positive, this form is positive semi-definite, so separation and completion gives a Hilbert space H′. If α is the action of G on eA, it follows from the fact that φ is equivariant that the formula D nXi=1 vg : ai ⊗ ξi 7→ αg(ai) ⊗ ugξi nXi=1 nXi=1 defines a unitary action of G on H′. Moreover, the action of eA on eA ⊙ H defined by π(a) ·(cid:16) nXi=1 ai ⊗ ξi(cid:17) := nXi=1 aai ⊗ ξi gives rise to a bounded representation of eA on H′ by the Cauchy-Schwarz inequality for φ. We also write π for the corresponding representation of eA on H′ and note that this is moreover covariant for v. The formula is easily seen to define an equivariant isometry with adjoint given by V : H → H′, ξ 7→ 1eA ⊗ ξ V ∗ : H′ → H, nXi=1 eφ(ai)ξi. Using these formulas, one sees that for all a ∈ eA we have nXi=1 ai ⊗ ξi 7→ (4.13) eφ(a) = V ∗π(a)V as elements of L(H). To complete the proof, let C be the C ∗-subalgebra of L(H′) generated by π(eA), V V ∗π(eA), π(eA)V V ∗ and V eBV ∗. Note that the action Adv of G on C induced by the unitary representation v is norm continuous, so C is a G-algebra with the induced action. Moreover, the ∗-homomorphism is equivariant and has image in C, so gives rise to a ∗-homomorphism on crossed products π : eA → L(H′) (4.14) π ⋊µ G : eA ⋊µ G → C ⋊µ G by functoriality of ⋊µ. Note that by construction of C, p := V V ∗ is in the multiplier algebra of C and pCp = V eBV ∗, so we have an equivariant ∗-isomorphism κ : pCp → B, x 7→ V ∗xV. (4.15) Now, if we denote by p the element of M(C ⋊µ G) induced by p, then by the projection property we have a ∗-isomorphism (4.16) p(C ⋊µ G)p ∼= pCp ⋊µ G EXOTIC BAUM-CONNES CONJECTURE 19 and so a ∗-isomorphism (4.17) ψ : p(C ⋊µ G)p → eB ⋊µ G defined as the map on crossed products induced by the composition of the isomor- phism in line (4.16) and the ∗-isomorphism on crossed products induced by the isomorphism in line (4.15). Now, consider the composition of maps A ⋊µ G → eA ⋊µ G π⋊µG → C ⋊µ G x7→px p −→ p(C ⋊µ G)p ψ → eB ⋊µ G where the first map is the ∗-homomorphism on crossed products induced from the equivariant inclusion A 7→ eA, the second is as in line (4.14), the third is compression by p, and the fourth is ψ as in line (4.17). Each map appearing in the sequence is completely positive, whence the composition is completely positive. Checking the image of Cc(G, A) using the formula in line (4.13) shows that the map agrees with the map f 7→ φ ◦ f from A ⋊µ G to D, where D is the image of B ⋊µ G under the canonical map of this C ∗-algebra into eB ⋊µ G. However, as ⋊µ has the projection property, it also has the ideal property, and so D is just a copy of B ⋊µ G. This completes the proof. (6) ⇒ (5) Let ⋊µ be a cp-functorial crossed product for G. Let p be a G-invariant projection in the multiplier algebra of a G-algebra A, and consider the G-equivariant maps pAp → A → pAp, where the first map is the canonical inclusion, and the second map is compression by p; note that the first map is a ∗-homomorphism, the second is completely positive, and the composition of the two is the identity on pAp. Functoriality then gives ∗-homomorphisms on crossed products pAp ⋊µ G → A ⋊µ G → pAp ⋊µ G whose composition is the identity; in particular, the first ∗-homomorphism is injec- tive, which is the projection property. (5) ⇒ (1) Assume that (A, α) → A ⋊α,µ G satisfies the projection property. We first show that this implies (2), i.e., for any Hilbert (B, β)-module (E, γ) we get K(E) ⋊Adγ,µ G ∼= K(E ⋊γ,µ G) in the canonical way. By Lemma 4.11 we know that ⋊µ satisfies the ideal property. Hence we may assume without loss of gener- ality that E is a full Hilbert B-module. The result then follows from (4) ⇔ (1) in Lemma 4.12. To see that (A, α) → A ⋊α,µ G extends to a correspondence functor we can now use the fact that by the ideal property we get functoriality for generalised homomor- phisms. Hence for any G-equivariant correspondence (E, φ, γ) from (A, α) to (B, β), the homomorphism φ : A → L(E) ∼= M(K(E)) descends to a ∗-homomorphism A ⋊µ G → M(K(E) ⋊Adγ,µ G) ∼= L(E ⋊γ,µ G), and therefore provides a correspondence (E ⋊γ,µ G, φ ⋊µ G) from A ⋊α,µ G to B ⋊β,µ G. We already saw above that it preserves isomorphisms (i.e., equivalence bimodules) and compatibility with compositions is proved on the level of functions with compact supports as in [16, Chapter 3]. (cid:3) ι⋊µGy (idpAp⋊G⊗v)◦bα (idA⋊G⊗v)◦bα A ⋊vKLQ G −−−−−−−−−→ M(A ⋊u G ⊗ C ∗ yι⋊uG⊗idC∗ v (G)) v (G) 20 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT We now give a few applications of Theorem 4.9. It is already shown in [9, Corollary 2.9] that all KLQ-functors extend to correspondences. We now give a very short argument for this: Corollary 4.18. Every KLQ-functor has the projection property and therefore ex- tends to a functor from Corr(G) to Corr. Proof. Let v : C ∗(G) → C ∗ v (G) be a quotient map which absorbs the universal representation uG. Then the corollary follows from commutativity of the diagram pAp ⋊vKLQ G −−−−−−−−−−→ M(pAp ⋊u G ⊗ C ∗ v (G)) together with the fact that the projection property holds for the universal crossed- product functor. (cid:3) Remark 4.19. On the other hand, the Brown-Guentner functors of Section 2.2 are almost never correspondence functors. In case of the BG-functor corresponding to the regular representation λG this is well illustrated in Example 2.2 above. In fact, it has been already observed in [4, Lemma A.6] that a BG-crossed product is Morita compatible in their sense (see Remark 5.5 and §8 for more discussion of this property) if and only if it coincides with the universal crossed product. Since every correspondence functor is also Morita compatible (see Corollary 5.4 and Remark 5.5 below), we also see that the universal crossed-product functors are the only BG-functors which are also correspondence functors. The next corollary shows that if we derive a functor from a correspondence functor via tensoring with a G-algebra (D, δ), then the new functor is also a corre- spondence functor. At the same time we extend the tensor-product construction of §2.4 to the case of non-unital G-algebras: Corollary 4.20. Suppose that (A, α) 7→ A ⋊α,µ G is a crossed-product functor which satisfies the ideal property and let (D, δ) be any G-algebra. Then there is a crossed-product functor ⋊µν , also satisfying the ideal property, defined as D A ⋊α,µν D G := jA ⋊µ G(A ⋊α,µ G) ⊆ M((A ⊗ν D) ⋊α⊗δ,µ G), where ⊗ν is either the minimal or the maximal tensor product of A with D and jA : A → M(A ⊗ν D) denotes the canonical inclusion. Moreover, if ⋊µ is a correspondence functor, then so is ⋊µν . D Proof. Let MD(A⊗ν D) denote the closed subalgebra of M(A⊗µ D) which consists of all elements m such that m(1⊗ν D) ⊆ A⊗ν D. Then for any (possibly degenerate) ∗-homomorphism φ : A → M(B) the ∗-homomorphism φ ⊗ idD : A ⊗ν D → M(B ⊗ν D) extends uniquely to MD(A ⊗ν D) (e.g., see [16, Proposition A.6]). Now let MD,c(A ⊗ν D) denote the set of G-continuous elements in MD(A ⊗ν D). Then it follows from the ideal property that (A ⊗ν D) ⋊µ G is an essential ideal in MD,c(A ⊗ν D) ⋊µ G, which implies that there is a canonical faithful inclusion of MD,c(A ⊗ν D) ⋊µ G into M((A ⊗ν D) ⋊µ G). Hence, we can alternatively define G as the image of jA ⋊µ G inside MD,c(A ⊗ν D) ⋊µ G. A similar argument A ⋊α,µν D EXOTIC BAUM-CONNES CONJECTURE 21 applies if we replace MD,c(A⊗ν D) by the C ∗-algebra Mc(A⊗ν D) of G-continuous elements in M(A ⊗ν D). Now let φ : A → M(B) be a G-equivariant ∗-homomorphism. Then the diagram A ⋊µ G jA⋊µGy MD,c(A ⊗ν D) ⋊µ G φ⋊µG −−−−−−−−−−−−−−−→ −−−−−−−−−→ (φ⊗ν idD )⋊µG M(B ⋊µ G) yjB ⋊µG Mc(B ⊗ν D) ⋊µ G ⊂ M((B ⊗ν D) ⋊µ G) commutes, since (jB ⊗ν idD) ◦ φ = (φ ⊗ν idD) ◦ jA. It follows that the bottom arrow restricts to a well-defined ∗-homomorphism from A ⋊µν G). It is a functor for generalised homomorphisms. By Lemma 3.3 this follows that ⋊µν also proves that the functor ⋊µν has the ideal property. G into M(B ⋊µν D D D D Assume now that ⋊µ is a correspondence functor. To see that this is then also , let p ∈ M(A) be any G-invariant projection and let ι : pAp → A be true for ⋊µν the inclusion map. Consider the commutative diagram D pAp ⋊µ G jpAp⋊µGy MD,c(pAp ⊗ν D) ⋊µ G ι⋊µG −−−−−−−−−−−−−−→ −−−−−−−−−→ (ι⊗ν idD )⋊µG A ⋊µ G yjA⋊µG Mc(A ⊗ν D) ⋊µ G ⊂ M((A ⊗ν D) ⋊µ G) Since ⋊µ satisfies the projection property, we have injectivity of (ι ⊗ idD) ⋊µ G : (pAp ⊗ν D) ⋊µ G ֒→ (A ⊗ν D) ⋊µ G which then extends to a unique injective ∗-homomorphism MD,c(pAp⊗ν D)⋊µG → M((A ⊗ν D) ⋊µ G). Thus the lower horizontal map in the diagram is injective. But G → then the commutativity of the diagram implies injectivity of ι⋊µν A ⋊µν (cid:3) G, as desired. G : pAp⋊µν D G G Remark 4.21. In [4, Lemma 5.4] it is shown that if ⋊µ is exact and D is unital, is also exact. We should note that for non-unital D this need not be then ⋊µmax true. To see an example, let G be any non-exact group, let D = C0(G) equipped with the translation action and let ⋊µ = ⋊u. Then D (A ⊗ C0(G)) ⋊u G ∼= (A ⊗ C0(G)) ⋊r G ∼= A ⊗ K(L2(G)) and the map jA ⋊u G : A ⋊u G → M(A ⊗ C0(G)) ⋊u G factors through the (faithful) map jA ⋊r G : A ⋊r G → M(A ⊗ C0(G)) ⋊u G). Therefore we get ⋊umax = ⋊r, which by the choice of G is not exact. C0(G) We conclude this section by showing that a correspondence crossed-product func- tor allows a nice description of K(E ⋊µ G, F ⋊µ G) in which (E, γ) and (F , τ ) are two G-equivariant Hilbert (B, β)-modules. This will be useful for our discussion of KK-theory in the next section. For this we first observe that K(E, F ) can be regarded as a K(F ) − K(E) correspondence with respect to the canonical left action of K(F ) on K(E, F ) given by composition of operators and with the K(E)-valued inner product given by The G-actions γ and ν induce an action Ad(γ, τ ) of G on K(E, F ) by hT SiK(E) = T ∗ ◦ S. Ad(γ, τ )s(T ) = τs−1 T γs. 22 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT Then (K(E, F ), Ad(γ, τ )) becomes a G-equivariant (K(E), Adγ)−(K(F ), Adτ ) corre- spondence. If ⋊µ is any crossed-product functor for G, we may consider the crossed product K(E, F ) ⋊Ad(γ,τ ),µ G as a completion of Cc(G, K(E, F )) as described at the beginning of this section. Lemma 4.22. Suppose that ⋊µ is a correspondence crossed-product functor. Then there is a canonical isomorphism K(E, F ) ⋊Ad(γ,τ ),µ G ∼= K(E ⋊γ,µ G, F ⋊τ,µ G) which sends a function f ∈ Cc(G, K(E, F )) ⊆ K(E, F ) ⋊µ G to the operator Tf ∈ K(E ⋊µ G, F ⋊µ G) given on the dense submodule Cc(G, E) ⊆ E ⋊γ,µ G by the convolution formula (Tf ξ)(s) =ZG f (t)γs(ξ(t−1s)) dt. Proof. Let E ⊕ F denote the direct sum of the Hilbert B-modules E and F . Let p, q ∈ L(E ⊕F) denote the orthogonal projections to E and F . This gives a canonical decomposition K(E ⊕ F) ∼=(cid:18) K(E) K(E, F ) K(F , E) K(F ) (cid:19) by identifying K(E) ∼= pK(E ⊕ F)p, K(F , E) = pK(E ⊕ F)q and so on. The projec- tions p and q are G-invariant and therefore map to opposite G-invariant projections p and q in M(K(E ⊕ F) ⋊Ad(γ⊕τ ),µ G) under the canonical map. Taking crossed products it follows from the properties of a correspondence functor shown in The- orem 4.9 that we get a decomposition K(cid:0)(E ⊕ F) ⋊γ⊕τ,µ G(cid:1) ∼= K(E ⊕ F) ⋊Ad(γ⊕τ ),µ G =(cid:18) K(E) ⋊Adγ,µ G q(cid:0)K(F ⊕ E) ⋊µ G(cid:1)p Notice that we have the identity p(cid:0)K(F ⊕ E) ⋊µ G(cid:1)q K(F ) ⋊Adτ,µ G (cid:19) (E ⊕ F) ⋊γ⊕τ,µ G ∼= E ⋊γ,µ G ⊕ F ⋊τ,µ G, which follows directly from the definition of the inner products on Cc(G, E ⊕ F). It implies the decomposition K((E ⊕ F ) ⋊γ⊕τ,µ G) ∼= K(E ⋊γ,µ G ⊕ F ⋊τ,µ G) ∼=(cid:18) ∼=(cid:18) K(E ⋊γ,µ G) K(E ⋊γ,µ G, F ⋊τ,µ G) K(F ⋊τ,µ G, E ⋊γ,µ G) K(F ⋊τ,µ G) K(E) ⋊Adγ,µ G K(E ⋊γ,µ G, F ⋊τ,µ G) K(F ⋊τ,µ G, E ⋊γ,µ G) K(F ) ⋊Adτ,µ G (cid:19) (cid:19) . Comparing both isomorphisms on functions in Cc(G, K(E ⊕ F)), we see that they agree and that the isomorphism of the upper right corner is given on the level of functions in Cc(G, K(E, F )) = pCc(G, K(E ⊕ F))q as in the statement. (cid:3) 5. Duality for correspondence functors We saw in Corollary 4.18 that all KLQ-functors are correspondence functors. Re- v (G) ∼= C ⋊vKLQ G corresponding to call from Section 2.3 that the group algebras C ∗ KLQ-functors are precisely those group algebras which correspond to weak equiva- lence classes of unitary representations v : G → U(Hv) which absorb the universal EXOTIC BAUM-CONNES CONJECTURE 23 representation uG : G → U(C ∗(G)) of G in the sense that v is weakly equivalent to v ⊗ uG. This just means that the comultiplication δG : C ∗(G) → M(C ∗(G) ⊗ C ∗(G)) on C ∗(G) which is given as the integrated form of uG⊗uG factors through a coaction δv : C ∗ v (G) → M(C ∗ v (G) ⊗ C ∗(G)) of C ∗(G) on C ∗ v (G). As discussed in Section 2.3, it is shown in [25] that these group algebras are in one-to-one correspondence to the translation invariant weak-* closed ideals E ⊆ B(G) of the Fourier-Stieltjes algebra B(G). Moreover, it is also shown in [25] that for any system (A, G, α) the dual coaction A ⊗ idG) ⋊ (iu G ⊗ uG) : A ⋊α,u G → M(A ⋊α,u G ⊗ C ∗(G)) bα = (iu of C ∗(G) on the full crossed product A ⋊α,u G factors through a dual coaction bαvKLQ : A ⋊α,vKLQ G → M(A ⋊α,vKLQ G ⊗ C ∗(G)). We refer to [16, Appendix A] for a survey on duality theory for actions and coactions of groups. We shall now show that the existence of a dual coaction holds true for every correspondence functor ⋊µ, so that versions of Imai-Takai and Katayama duality will hold for the corresponding double (or triple) crossed products. In particular, this implies that the group algebra C ∗ µ(G) = C ⋊µ G of a correspondence functor must correspond to a translation invariant weak-* closed ideal E in B(G). We start with a result about exterior equivalent actions, which is interesting in its own right. For this recall that two actions α, β : G → Aut(A) on a given C-algebra A are called exterior equivalent if there exists a strictly continuous map u : G → UM(A) such that for all s, t ∈ G we have (5.1) αs = Adus ◦ βs and ust = usβs(ut). It is a well-known fact that exterior equivalent actions give isomorphic full (and reduced) crossed products. Indeed, it is fairly straightforward to check that the map Φc : Cc(G, A) → Cc(G, A); Φc(f )(s) = f (s)us extends to an isomorphism Φu : A ⋊α,u G ∼−→ A ⋊β,u G. A short computation shows that it is a ∗-isomorphism on the level of Cc(G, A). Then one observes that there is a one-to-one correspondence between nondegenerate covariant homomor- phisms (A, G, β) and (A, G, α) such that for any covariant homomorphism (π, V ) for (A, G, β) the pair (π, (π ◦ u) · V ) is the corresponding covariant representation of (A, G, α). The result then follows from the fact that π ⋊ (π ◦ u · V ) = (π ⋊ V ) ◦ Φc for all f ∈ Cc(G, A). Alternatively, the isomorphism is given as the integrated form of the covariant homomorphism (iβ G) of (A, G, α) into M(A ⋊β,u G). We refer to [33, Lemma 3.3] for a detailed proof in the more general setting of twisted actions (with roles of α and β changed). We then have A ◦ u) · iβ A, (iβ Lemma 5.2. Suppose that ⋊µ is a strongly Morita compatible crossed-product func- tor for G. Suppose further that α, β : G → Aut(A) are exterior equivalent with re- spect to the one-cocycle u : G → UM(A). Then the isomorphism Φu : A ⋊α,u G ∼−→ A ⋊β,u G factors through an isomorphism Φµ : A ⋊α,µ G ∼−→ A ⋊β,µ G. 24 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT Proof. Using the above described correspondence of covariant representations of A ⋊α,u G and A ⋊β,u G it suffices to show that the integrated form of a covari- ant representation (π, V ) of (A, G, β) factors through A ⋊β,µ G if and only if the corresponding covariant representation (π, (π ◦ u) · V ) of (A, G, α) factors through A ⋊α,µ G. To see this consider A as the standard A−A equivalence bimodule. Then the action γ : G → Aut(A) given by γs(a) = usβs(a) makes A into a G-equivariant (A, α) − (A, β) equivalence bimodule. Since ⋊µ is strongly Morita compatible we obtain an A ⋊α,µ G − A ⋊β,µ G equivalence bimodule A ⋊γ,µ G, which is a quotient of the corresponding A ⋊α,u G − A ⋊β,u G equivalence bimodule A ⋊γ,u G. It fol- lows then from the Rieffel-correspondence that induction of representations from A⋊β,µG to A⋊β,µG via A⋊γ,µG is the same as induction from A⋊β,uG to A⋊β,uG via A ⋊γ,u G, if we regard representations of the µ-crossed products as representa- tions of the universal crossed product via pull-back with the quotient maps. This induction process can be described on the level of covariant representations by in- ducing a covariant representation (π, V ) of (A, G, β) via the G-equivariant module (A, γ) (see [11]): if (π, V ) acts on the Hilbert space H, then the induced covariant representation (indA π, indA V ) acts on the Hilbert space A ⊗A H via indA π(a)(b ⊗ ξ) = ab ⊗ ξ and indA Vs(b ⊗ ξ) = γs(b) ⊗ Vsξ for all a ∈ A, s ∈ G. But then a short calculation shows that the canonical isomorphism A ⊗A H ∼= H given by b ⊗ ξ 7→ π(b)ξ tranfers this representation to (π, (π ◦ u) · V ). Hence induction via A ⋊γ,µ G gives the desired correspondence of representations. (cid:3) Remark 5.3. (1) The conclusion of the above lemma can definitely fail for crossed- product functors ⋊µ which are not strongly Morita compatible. To see an example let G be any non-amenable group. Let λ = λG : G → U(L2(G)) = UM(K(L2(G))) denote the regular representation of G. Then λ induces an exterior equivalence between the actions Adλ and idK of G on K = K(L2(G)). Note that all irreducible covariant representations of (K, G, idK) are equivalent to one of the form (idK ⊗ 1H, 1L2(G) ⊗ V ) on L2(G) ⊗ H for some Hilbert space H, where V : G → U(H) is an irreducible unitary representation of G. This follows easily from the isomorphism K ⋊id,u G ∼= K ⊗ C ∗(G). Now let ⋊λBG be the Brown-Guentner functor corresponding to the regular rep- resentation (compare Section 2.2). Since the unitary part 1L2(G) ⊗ V ∼ V is weakly contained in λ if and only if V is weakly contained in λ it follows that K ⋊id,λBG G ∼= K ⊗ C ∗ r (G) ∼= K ⋊id,r G. On the other hand, the covariant rep- resentation of (K, G, Adλ) corresponding to (idK ⊗ 1H, 1L2(G) ⊗ V ) is given by (idK ⊗ 1H, λ ⊗ V ). Since by Fell's trick λ ⊗ V ∼ λ, we see that the integrated form of every irreducible covariant representation of (K, G, Adλ) factors through K ⋊Adλ,λBG G, which implies that this crossed product coincides with the universal crossed product K ⋊Adλ,u G ∼= K ⊗ C ∗(G). Since G is not amenable, this is not (canonically) isomorphic to K ⋊Adλ,λBG G ∼= K ⊗ C ∗ r (G). (2) Recall that two systems (A, G, α) and (B, G, β) are conjugate if there exists an isomorphism Ψ : A → B such that the action αΨ := Ψ ◦ α ◦ Ψ−1 on B coincides with β and they are called outer conjugate if αΨ is exterior equivalent to β. Since by functoriality, conjugate actions give isomorphic crossed products for any crossed- product functor ⋊µ, it follows from Lemma 5.2 that outer conjugate actions give isomorphic µ-crossed product if ⋊µ is strongly Morita compatible. EXOTIC BAUM-CONNES CONJECTURE 25 As a corollary of Lemma 5.2 we get Corollary 5.4. Suppose that α : G → Aut(A) is an action and that U : G → U(H) is a unitary representation of G. Assume further that ⋊µ is a strongly Morita compatible crossed-product functor for G and let (iµ G) be the canonical covariant homomorphism of (A, G, α) into M(A ⋊α,µ G). Then (iµ G ⊗ U ) is a covariant homomorphism of the system (A ⊗ K(H), G, α ⊗ AdU ) into M(A ⋊α,µ G ⊗ K(H) whose integrated form factors through an isomorphism A ⊗ idK(H), iµ A, iµ Φµ : (A ⊗ K(H)) ⋊α⊗AdU,µ G ∼−→ A ⋊α,µ G ⊗ K(H). Proof. We first observe that the homomorphism 1A ⊗ U : G → UM(A ⊗ K(H)) implements an exterior equivalence between (A⊗K(H), α⊗AdU ) and (A⊗K(H), α⊗ idK(H)), hence by Lemma 5.2 we get an isomorphism (A ⊗ K(H)) ⋊α⊗AdU,µ G ∼−→ (A ⊗ K(H)) ⋊α⊗idK,µ G A⊗K(H), iµ A⊗K(H) ◦ (1A ⊗ U ) · iµ given by the covariant homomorphism (iµ G). We also have an isomorphism (A ⊗ K(H)) ⋊α⊗idK,µ G ∼= A ⋊α,µ G ⊗ K(H) given by the integrated form of the covariant homomorphism (iµ G ⊗ 1H). To see this we consider the (A⊗K(H), α⊗idK)−(A, α) equivalence bimodule (A⊗H, α⊗1H). It is easy to check that the map Cc(G, A) ⊗ H → Cc(G, A ⊗ H); f ⊗ ξ 7→ [s 7→ f (s) ⊗ ξ] is isometric with respect to the A ⋊α,µ G-valued inner products and therefore extends to an isomorphism A ⋊α,µ G ⊗ H ∼= (A ⊗ H) ⋊α⊗1H,µ G. In this picture, the left action of (A ⊗ K(H)) ⋊α⊗idK,µ G on A ⋊α,µ G ⊗ H is given by the integrated form of (iµ G ⊗ 1H). By strong Morita compatibility this integrates to an isomorphism A ⊗ idK, iµ A ⊗ idK, iµ (A ⊗ K(H)) ⋊α⊗idK,µ G ∼= K(A ⋊α,µ G ⊗ H) = A ⋊α,µ G ⊗ K(H). The result now follows from composing this isomorphism with the integrated form of (iµ (cid:3) A⊗K(H) ◦ (1A ⊗ U )) · iµ A⊗K(H), (iµ G). Remark 5.5. Recall that in [4] Baum, Guentner and Willett define a crossed product functor ⋊µ to be Morita compatible if the isomorphism Φµ : (A ⊗ K(H)) ⋊α⊗AdU,µ G ∼−→ A ⋊α,µ G ⊗ K(H) holds in the special case where H = ℓ2(N) ⊗ L2(G) and U = 1ℓ2(N) ⊗ λG. Thus the above corollary shows that strong Morita compatibility in our sense implies Morita compatibility in the sense of [4]. We refer to Section 8 below for a more detailed comparison of the two notions of Morita compatibility. In particular we shall see in Proposition 8.10 below that both conditions are equivalent if we restrict our attention to systems (A, G, α) with A σ-unital. We are now ready for the main result of this section which shows that all crossed products coming from correspondence functors admit dual coactions and enjoy Imai- Takai duality. Recall for this that the dual coaction bα : A ⋊α,u G → M(A ⋊α,u G ⊗ C ∗(G)) is given as the integrated form of the covariant homomorphism (iu where uG denotes the universal representation of G. A ⊗ 1G, iu G ⊗ uG), 26 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT Theorem 5.6. Suppose that ⋊µ is a correspondence crossed-product functor for G. given by the integrated form of (iµ bαµ : A ⋊α,µ G → M(A ⋊α,µ G ⊗ C ∗(G)) Then the dual coaction bα of G on A ⋊α,u G factors to give a dual coaction crossed product A ⋊α,µ G ⋊bαµ bG is isomorphic to A ⊗ K(L2(G)) such that this isomorphism transfers the double dual action cbαµ : G → Aut(A ⋊α,µ G ⋊bαµ bG) to the action α ⊗ Adρ : G → Aut(A ⊗ K(L2(G))). In particular, it follows that G ⊗ uG). Moreover, the double dual A ⊗ 1C ∗(G), iµ G ∼= A ⋊α,µ G ⊗ K(L2(G)). A ⋊α,µ G ⋊bαµ bG ⋊bbαµ,µ A, iµ of (iµ G), hence the identity on A ⋊α,µ G. In the language of [8, Definition 4.5], Note that the map bαµ, if it exists, is automatically faithful, since the composition (cid:0)idA⋊α,µG ⊗ 1G) ◦bαG, with 1G the trivial representation of G, is the integrated form the above result says that bαµ satisfies µ-duality or is a µ-coaction. carry over to bαµ. For this let us represent C ∗(G) faithfully on Hilbert space via the Proof. For the first assertion it suffices to show that the integrated form of (iµ 1C ∗(G), iµ A ⊗ G ⊗ uG) factors through A ⋊α,µ G. Then all requirements for a coaction integrated form of a unitary representation U : G → U(H). Then it suffices to show that the covariant representation (iµ G ⊗ U ) of (A, G, α) into M(A ⋊α,µ G ⊗ K(H)) factors through A ⋊α,µ G. Since ⋊µ is Morita compatible, we know from Corollary 5.4 that the covariant representation (iµ G ⊗ U ) of (A ⊗ K(H), G, α ⊗ AdU ) into M(A ⋊α,µ G ⊗ K(H)) factors through an isomorphism A ⊗ idK, iµ A ⊗ 1H, iµ Φµ : (A ⊗ K(H)) ⋊α⊗AdU,µ G ∼−→ A ⋊α,µ G ⊗ K(H). Since ⋊µ is functorial for generalised homomorphisms, the G-equivariant ∗-homo- morphism idA ⊗ 1H : A → M(A ⊗ K(H)) induces a ∗-homomorphism A ⋊α,µ G → M((A ⊗ K(H)) ⋊α⊗AdU,µ G) given by sending a function f ∈ Cc(G, A) to the function [s 7→ (f (s) ⊗ 1H)] ∈ Cc(G, M(A ⊗ K(H))). The composition of both maps, which is well defined on A⋊α,µ G, is given by the integrated form of (iµ G ⊗U ). Imai-Takai duality now follows from the well-known Imai-Takai duality theorems for the full and reduced crossed products (see [22] for the reduced version or [16, Theorem A.67] for a discussion of the full and reduced cases): By construction, both quotient maps A⊗1H, iµ A ⋊α,u G qµ ։ A ⋊α,µ G qr ։ A ⋊α,r G are equivariant for the coactions bα, bαµ and bαr, respectively. They therefore induce quotient maps A ⋊α,u G ⋊bα bG qµ⋊bG ։ A ⋊α,µ G ⋊bαµ bG qr ⋊bG ։ A ⋊α,r G ⋊bαr bG. However, the composition is an isomorphism by [16, Proposition A.6.1], hence Imai- Takai duality for µ-crossed products follows from Imai-Takai duality for full and/or reduced crossed products. The last assertion of the theorem is then a direct conse- quence of Corollary 5.4. (cid:3) Corollary 5.7. Suppose that ⋊µ is a correspondence crossed-product functor for G. Then the group algebra C ∗ µ(G) := C ⋊µ G corresponds to a translation invariant weak- closed ideal E ⊆ B(G) as in Proposition 2.3. EXOTIC BAUM-CONNES CONJECTURE 27 Proof. Let v : G → UM(C ∗ µ(G)) be the homomorphism which integrates to the quotient map qµ : C ∗(G) → C ∗ µ(G). Then it follows from Theorem 5.6 that the integrated form of v ⊗uG factors through C ∗ µ(G) ⊗1G)◦(v ⊗uG) = v we see that v ∼ v ⊗ uG and result follows from the discussions in Section 2.3. (cid:3) µ(G). Since (idC ∗ 6. KK-descent and the Baum-Connes conjecture In this section we want to show that every crossed-product functor (A, α) 7→ A ⋊α,µ G which extends to a functor Corr(G) → Corr as discussed in the previ- ous section will always allow a descent in Kasparov's bivariant KK-theory. Let us briefly recall the cycles in Kasparov's equivariant KK-theory group KK G(A, B) in which (A, α) and (B, β) are G-algebras. As in [27] we allow Z/2Z-graded C ∗-al- gebras and Hilbert modules. In this case all G-actions have to commute with the given gradings, so the grading on the objects (A, α) and morphisms (E, γ) will al- ways induce canonical gradings on their crossed products A ⋊α,µ G and E ⋊γ,µ G, respectively. For convenience and to make sure that Kasparov products always exist we shall assume that A and B are separable and that G is second countable. If (A, α) and (B, β) are two G-algebras, we define EG(A, B) to be the set of all quadruples (E, γ, φ, T ) such that (E, φ, γ) is a correspondence from (A, α) to (B, β) such that E is countably generated, and T ∈ L(E) is an operator of degree one satisfying the following conditions: (1) s 7→ φ(αs(a))T is continuous for all a ∈ A, and (2) [T, φ(a)], (T − T ∗)φ(a), (T 2 − 1)φ(a), (Adγs(T ) − T )φ(a) ∈ K(E) for all a ∈ A, where we use the graded commutator [·, ·] in (2). Two Kasparov cycles (E1, γ1, φ1, T1) and (E2, γ2, φ2, T2) are isomorphic if there exists a bijection u : E1 → E2 which preserves all structures. If G is trivial, there is no action γ on E and the elements in E(A, B) are just given by triples (E, φ, T ) with the obvious properties. Write B[0, 1] for the algebra of continuous functions from the unit interval [0, 1] to B and ǫt : B[0, 1] → B for the evaluation at t ∈ [0, 1]. The action of G on B[0, 1] = B ⊗ C([0, 1]) is given by β ⊗ id. A homotopy between two elements (E0, γ0, φ0, T0) and (E1, γ1, φ1, T1) in EG(A, B) is an element (E, γ, φ, T ) ∈ EG(A, B[0, 1]) such that (Ei, γi, φi, Ti) ∼= (E ⊗εi B, γ ⊗1, φ ⊗1, T ⊗1), i = 0, 1, where, for each t ∈ [0, 1], E ⊗εt B denotes the balanced tensor product E ⊗B[0,1] B in which B[0, 1] acts on B via ǫt. Then Kasparov defines KK G(A, B) as the set of homotopy classes in EG(A, B). It is an abelian group with respect to addition given by taking direct sums of elements in EG(A, B). 0 (A, B) := KK G(A, B) and Recall also that for i = 0, 1 one may define KK G 1 (A, B) := KK G(C0(R) ⊗ A, B) ∼= KK G(A, C0(R) ⊗ B) where G acts trivially KK G on C0(R) (the last isomorphism follows from Kasparov's Bott-periodicity theorem for KK-theory). Note that if E is a graded module, then there are natural gradings on the crossed products determined by applying the given gradings pointwise to functions with compact supports. The balanced tensor product of two graded modules carries the diagonal grading. The following proposition extends the descent given by Kasparov in [27] for full and reduced crossed products to arbitrary correspondence crossed- product functors. 28 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT Proposition 6.1. Let (A, α) 7→ A ⋊α,µ G be a correspondence crossed-product functor. Then there is a well-defined descent homomorphism J µ G : KK G i (A, B) → KKi(A ⋊α,µ G, B ⋊β,µ G), i = 0, 1, given by sending a class [E, γ, φ, T ] ∈ KK G i (A, B) to the class [E ⋊γ,µ G, φ ⋊µ G, Tµ] ∈ KKi(A ⋊α,µ G, B ⋊β,µ G), where Tµ ∈ L(E ⋊γ,µ G) is the unique continuous extension of the operator T : Cc(G, E) → Cc(G, E) given by ( T ξ)(s) := T (ξ(s)) for all s ∈ G. The descent is compatible with Kasparov products, i.e., we have J µ G(x ⊗B y) = J µ G(x) ⊗B⋊β,µG J µ G(y) for all x ∈ KK G G(1B) = 1B⋊β,µG. i (A, B) and y ∈ KK G j (B, D) and J µ Proof. We basically follow the proof of [27, Theorem on p. 172]. First observe that since L(E ⋊γ,µ G) = M(K(E ⋊γ,µ G)) = M(K(E) ⋊Adγ,µ G), the operator Tµ is just the image of T under the extension of the canonical inclusion from K(E) into M(K(E) ⋊Adγ,µ G) to M(K(E)) ∼= L(E). It therefore exists. Next, the conditions for (E ⋊γ,µ G, φ ⋊µ G, Tµ) follow from the fact that for any a ∈ Cc(G, A) the elements [ T , φ(a)], ( T 2 − 1)φ(a), ( T ∗ − T )φ(a) (as operators on Cc(G, E)) lie in Cc(G, K(E)), and hence in K(E) ⋊Adγ,µ G ∼= K(E ⋊γ,µ G). This shows that (E ⋊β,µ G, φ ⋊µ G, Tµ) determines a class in KK(A ⋊α,µ G, B ⋊β,µ G). To see that this class does not depend on the choice of the representative (E, γ, φ, T ) ∈ EG(A, B) we need to observe that the procedure preserves homotopy. So assume now that (E, γ, φ, T ) ∈ EG(A, B[0, 1]). By Lemma 3.6 we know that B[0, 1]⋊β⊗id,µG ∼= (B⋊β,µG)[0, 1], hence (E ⋊γ,µG, φ⋊µG, Tµ) ∈ E(A⋊α,µG, B⋊β,µ G[0, 1]). It follows then from correspondence functoriality of our crossed-product functor that evaluation of (E ⋊γ,µ G, φ ⋊µ G, Tµ) at any t ∈ [0, 1] coincides with the descent of the evaluation of (E, γ, φ, T ) at t (this implies that the modules coincide, but a short look at the operator on functions with compact supports also shows that the operators coincide). This shows that the descent gives a well-defined homomor- phism of KK-groups. Using the isomorphism (C0(R)⊗ A) ⋊µ G ∼= C0(R)⊗ (A ⋊µ G) if G acts trivially on R, which follows from Lemma 3.6, it follows that it preserves the dimension of the KK-groups. Finally the fact that the descent is compatible with Kasparov products follows from the same arguments as used in the original proof of Kasparov as given on [27, page 173] for the full and reduced crossed products together with an application of Lemma 4.22 to K(E2, E1 ⊗BE2). (cid:3) Suppose now that (A, G, α) is a C ∗-dynamical system with A separable and G second countable. Suppose that A ⋊α,ν G is any quotient of the universal crossed product A ⋊α,u G with quotient map qν : A ⋊α,u G → A ⋊α,ν G. Using the Baum- Connes assembly map asu (A,G) : K top ∗ (G; A) → K∗(A ⋊α,u G) for the full crossed product, we obtain an assembly map asν (A,G) : K top ∗ (G; A) → K∗(A ⋊α,ν G) by putting (6.2) asν (A,G) = qν,∗ ◦ asu (A,G) . EXOTIC BAUM-CONNES CONJECTURE 29 Of course, in general we cannot assume that this map has any good properties, since we should assume that there is a huge variety of quotients of A ⋊α,u G with different K-theory groups. If the C ∗-algebra A ⋊α,ν G lies between the maximal and reduced crossed products, then the Baum-Connes conjecture predicts that asν is split injective, but we cannot say more than that. (A,G) On the other hand, if G is K-amenable, then the quotient map qr : A ⋊α,u G → A ⋊α,r G induces an isomorphism of K-theory groups, and there might be a chance that a similar result will hold for intermediate crossed products. For arbitrary intermediate crossed products this is not true, as follows from [4, Example 6.4], where an easy counter example is given that applies to any K-amenable1, non- amenable group. However, we shall see below that it holds for all correspondence crossed-product functors for any K-amenable group! We need to recall the construction of the assembly map: For this assume that G is a second countable locally compact group and that EG is a universal proper G-space, i.e., EG is a locally compact proper G-space such that for every locally compact G-space Y there is a (unique up to G-homotopy) continuous G-map ϕ : Y → EG. The topological K-theory of G with coefficient A is defined as K top ∗ (G; A) := lim X⊆EG KK G ∗ (C0(X), A), where X ⊆ EG runs through all G-compact (which means that X is a closed G-invariant subset with G\X compact) subsets of EG. Now, using properness, for any such G-compact subset X ⊆ EG we may choose a cut-off function c ∈ Cc(G)+ such that RG c2(s−1x) ds = 1 for all x ∈ X. Define pX (s, x) = ∆G(s)−1/2c(x)x(s−1x). It follows from the properties of c that the function pX ∈ Cc(G, C0(X)) ⊆ C0(X) ⋊ G is an orthogonal projection. As any two cut-off functions with the properties above are homotopic, the K-theory class [pX ] ∈ K∗(C0(X) ⋊G) does not depend on the choice of the cut-off function c. Note also that proper actions on spaces are always amenable, hence the universal and reduced crossed products coincide. This also implies that all intermediate crossed products coincide. Using this and the above defined descent in KK-theory, we may now construct a direct assembly map for any correspondence crossed-product functor (A, α) 7→ A ⋊α,µ G and separable G-algebra A given by the composition J-asµ X : KK G ∗ (C0(X), A) J µ −→ KK∗(C0(X) ⋊ G, A ⋊α,µ G) [pX ]⊗· −→ K∗(A ⋊α,µ G). Note that this is precisely the same construction as given by Baum, Connes and Higson in [3] for the universal or reduced crossed products, and as in those cases it is easy to check that the maps are compatible with the limit structure and therefore define a direct assembly map (6.3) J-asµ (A,G) : K top ∗ (G; A) → K∗(A ⋊α,µ G); the notation J-as is meant to recall that we used the descent functor in the definition. However, the next result shows that this assembly map is the same as that in line (6.2) above: it is a KK-theoretic version of [4, Proposition 4.5]. Lemma 6.4. Assume (A, G, α) is a C ∗-dynamical system with A separable and G second countable. Assume ⋊µ is a correspondence crossed-product functor for G. 1The discussion in [4, Example 6.4] is for a-T-menable, non-amenable groups, but the same argument applies if one assumes K-amenability in place of a-T-menability. 30 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT Then the diagram K∗(A ⋊u G) (A,G) (A,G) asu 5❦❦❦❦❦❦❦❦❦❦❦❦❦❦ J-asµ )❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙ asr (A,G) K top ∗ (G; A) K∗(A ⋊µ G) (qµ)∗ (qµ r )∗ K∗(A ⋊r G) commutes. In particular, the assembly maps asµ and (6.3) above agree. (A,G) and J-asµ (A,G) of lines (6.2) Proof. Let [qµ] ∈ KK(A ⋊u G, A ⋊µ G) be the KK-class induced by the quotient ∗-homomorphism qµ : A ⋊u G ։ A ⋊µ G. For commutativity of the top triangle, it suffices to show that for any G-compact subset X of EG the diagram KK G ∗ (C0(X), A) J u / KK∗(C0(X) ⋊ G, A ⋊α,u G) [pX ]⊗· J µ KK∗(C, A ⋊α,u G) ·⊗[qµ] KK∗(C0(X) ⋊ G, A ⋊α,µ G) [pX ]⊗· / KK∗(C, A ⋊α,µ G) commutes (we have used that C0(X) ⋊ G = C0(X) ⋊µ G to identify the group in the bottom left corner). In symbols, commutativity of this diagram means that for any x ∈ KK G ∗ (C0(X), A) we have ([pX ] ⊗ J u(x)) ⊗ [qµ] = [pX ] ⊗ J µ(x). Using associativity of the Kasparov product, it thus suffices to show that for any x ∈ KK G ∗ (C0(X), A), J u(x) ⊗ [qµ] = J µ(x). Indeed, say x is represented by the quadruple (E, γ, φ, T ). Then J u(x) ⊗ [qµ] and J µ(x) are represented by the triples (E ⋊γ,u G ⊗B⋊β G B ⋊β,µ G, φ ⋊µ G ⊗ 1, Tu ⊗ 1) and (E ⋊γ,µ G, φ ⋊µ G, Tµ) respectively. It is easy to see that the canonical isomorphism E ⋊γ,u G ⊗B⋊β,uG B ⋊β,µ G ∼−→ E ⋊γ,µ G of Equation (4.6) sends φ ⋊µ G ⊗ 1 to φ ⋊µ G and Tu ⊗ 1 to Tµ, and hence induces an isomorphism of these triples. This shows commutativity of the upper diagram. Commutativity of the lower (cid:3) one follows by similar arguments. Recall now that a group G is said to satisfy the strong Baum-Connes conjecture (or that G has a γ-element equal to one), if there exists an EG ⋊ G-algebra B (which means that there is a G-equivariant nondegenerate ∗-homomorphism C0(EG) into the center ZM(B) of the multiplier algebra M(B)), together with classes D ∈ KK G ∗ (C, B) such that η ⊗B D = 1G ∈ KK G(C, C). It was shown by Higson and Kasparov in [20] that all a-T-menable groups satisfy the strong Baum-Connes conjecture. We then get ∗ (B, C) and η ∈ KK G 5 / / ) / / EXOTIC BAUM-CONNES CONJECTURE 31 Theorem 6.5. Suppose that G is a locally compact group which satisfies the strong Baum-Connes conjecture and let (A, α) → A ⋊α,µ G be a correspondence crossed- product functor. Then the µ-assembly map asµ (A,G) : K top ∗ (G; A) → K∗(A ⋊α,µ G) is an isomorphism for all G-algebras (A, α). This can be proved directly in the same way as the special case dealing with the reduced crossed product. However, we shall obtain the result as a consequence of the known isomorphism of the assembly map for the reduced crossed products (e.g., see [36, Theorem 2.2] or [20, Theorem 1.1]) together with Lemma 6.4 above and Theorem 6.6 below, which in particular implies that for every K-amenable group G the quotient map A ⋊u G ։ A ⋊µ G induces an isomorphism in K-theory. Recall that a locally compact group G is called K-amenable in the sense of Cuntz and Julg-Valette ([14, 24]) if the unit element 1G ∈ KK G(C, C) can be represented by a cycle (H, γ, T ) in which the unitary representation γ : G → U(H) on the Hilbert space H is weakly contained in the regular representation. In other words, its integrated form factors through a ∗-representation γ : C ∗ r (G) → L(H). (The left action of C on H is assumed to be by scalar multiplication, so we omit it in our notation.) It is shown by Tu in [36, Theorem 2.2] that every locally compact group which satisfies the strong Baum-Connes conjecture is also K-amenable and it is shown by Julg and Valette in [24, Proposition 3.4] that for any K-amenable group the regular representation A ⋊α,u G ։ A ⋊α,r G induces a KK-equivalence for any G-C ∗-algebra (A, α). We shall now extend this result to correspondence functors: Theorem 6.6 (cf. [24, Proposition 3.4]). Suppose that G is K-amenable and let ⋊µ be any correspondence functor for G. Then both maps in the sequence A ⋊α,u G qµ ։ A ⋊α,µ G qµ r ։ A ⋊α,r G are KK-equivalences. Proof of Theorem 6.6. It is enough to show that the quotient map qµ : A ⋊α,µ r G → A ⋊α,r G is a KK-equivalence. Then the result for the first map follows from composing the KK-equivalence in case ⋊µ = ⋊u with the inverse of [qµ r ] in KK(A ⋊α,r G, A ⋊α,µ G). We closely follow the arguments given in the proof of [24, Proposition 3.4]. So assume that (H, γ, T ) is a cycle for 1G ∈ KK G(C, C) such that γ is weakly contained in λG and let (A, α) be any G-C ∗-algebra. Since 1G ⊗C 1A = 1A, it follows that 1A ∈ KK G(A, A) is represented by the cycle (H ⊗ A, 1H ⊗ idA, γ ⊗ α, T ⊗ 1). Using the isomorphism (H ⊗ A) ⋊µ G ∼= H ⊗ (A ⋊µ G) of Corollary 5.4 we see that Jµ(1A) = 1A⋊µG ∈ KK(A ⋊µ G, A ⋊µ G) is represented by the cycle The action Ψ of A ⋊α,µ G on H ⊗ (A ⋊µ G) is given by the integrated form of the covariant homomorphism (1H ⊗ iµ G) which is just the composition of the A, γ ⊗ iµ (cid:0)H ⊗ A ⋊µ G, Ψ, T ⊗ 1(cid:1) 32 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT homomorphism (1H ⊗ idA) ⋊µ G of A ⋊α,µ G into M(cid:0)(K(H) ⊗ A) ⋊Adγ⊗α,µ G(cid:1) with To see that this action factors through A ⋊α,r G we first observe that C ∗ the left action of (K(H)⊗A)⋊Adγ⊗α,µ G onto H⊗A⋊α,µ G, which, by Corollary 5.4, is given by the integrated form of the covariant homomorphism (idK(H) ⊗iµ A, γ ⊗iµ G). r (G) ⊗ A ⋊α,µ G acts on H ⊗ A ⋊α,µ G via γ ⊗ idA⋊µG, since γ is weakly contained in λG by assumption. Moreover, the covariant homomorphism G) : A ⋊α,u G → M(C ∗ r (G) ⊗ A ⋊α,µ G) A) ⋊ (λG ⊗ iµ r (G) ⊗ iµ (1C ∗ factors through A⋊α,rG, since this is already true if we replace A⋊α,µ G by A⋊α,uG (compare the discussion on KLQ-functors in §2.3). Now one checks on generators that Ψ = (γ ⊗ idA⋊µG) ◦(cid:0)(1C ∗ r (G) ⊗ iµ A) ⋊ (λG ⊗ iµ G)(cid:1), hence the result. Thus, replacing the left action of A ⋊α,µ G by the left action of A ⋊α,r G, we obtain an element xA ∈ KK(A ⋊α,r G, A ⋊α,µ G) such that 1A⋊µG = (qµ r ] = 1A⋊rG follows word for word as in [24, Proposition 3.4] and we omit further details. (cid:3) r ] ⊗A⋊rG xA. The converse equation xA ⊗A⋊uG [qµ r )∗(xA) = [qµ Remark 6.7. In case of a-T-menable groups a different line of argument could be used to obtain the above proposition. In fact, it is shown in [20, Theorems 8.5 and 8.6] that the elements D ∈ KK G(B, C) and η ∈ KK G(C, B) which implement the validity of the strong Baum-Connes conjecture can be chosen to be KK G- equivalences. Since for proper actions the full and reduced crossed products (and hence all exotic crossed products) coincide, we can use the descents for ⋊µ and ⋊u to obtain the following chain of KK-equivalences for a given G-C ∗-algebra (A, α): A ⋊µ G ∼KKG (A ⊗ B) ⋊µ G ∼= (A ⊗ B) ⋊u G ∼KKG A ⋊u G. One can deduce from this that the quotient maps A ⋊u G ։ A ⋊µ G ։ A ⋊r G are also KK-equivalences. For the special case of KLQ-functors, a similar line of argument as used above gives the following result: Theorem 6.8. Suppose that ⋊vKLQ is a KLQ-crossed product functor correspond- ing to the universally absorbing representation v : G → UM(C ∗ v (G)). Suppose that the unit 1G ∈ KK G(C, C) is represented by a cycle (H, γ, T ) such that γ is weakly contained in v. Then, for every system (A, G, α) the quotient map qv : A ⋊α,u G ։ A ⋊α,vKLQ G is a KK-equivalence. In particular, the quotient map v : C ∗(G) → C ∗ v (G) is a KK-equivalence. A, γ ⊗ iu Proof. As in the proof of Theorem 6.6 we can represent the descent of 1A ∈ KK G(A, A) for the maximal crossed products by the cycle (H ⊗ A ⋊α,u G, Ψ, T ⊗ 1) where A ⋊α,u G acts on H ⊗ A ⋊α,u G via the integrated form of the covariant ho- momorphism (1H ⊗ iu G). Since γ is weakly contained in v, this action can be written as the composition of the representation (1C ∗ G) of A ⋊α,u G into M(C ∗ v (G)⊗ A ⋊α,u G → B(H ⊗ A ⋊α,u G), where γ : C ∗ v (G) of the inte- grated form of γ -- which exists since γ is weakly contained in v. By the defi- G), and hence nition of KLQ-functors, the representation (1C ∗ also (1H ⊗ iu G) factors through A ⋊α,vKLQ G, which gives an element v (G) → B(H) denotes the unique factorisation to C ∗ v (G)⊗ A ⋊α,u G) with γ ⊗ idA⋊uG : C ∗ A) ⋊ (v ⊗ iu A) ⋊ (γ ⊗ iu A) ⋊ (v ⊗ iu v (G) ⊗ iu v (G) ⊗ iu EXOTIC BAUM-CONNES CONJECTURE 33 xA ∈ KK(A ⋊α,vKLQ G, A ⋊α,u G) which is a KK-inverse of the quotient map qv. (cid:3) Example 6.9. Suppose that N is a closed normal subgroup of G such that G/N is K-amenable. Let v = λG/N ⊗ uG be the interior tensor product of the regular representation of G/N with the universal representation of G. Then v is univer- sally absorbing and we may consider the corresponding KLQ-functor (A, G, α) 7→ A ⋊vKLQ G. Since G/N is K-amenable, there is a representative (H, γ, T ) of 1G/N ∈ KK G/N (C, C) such that γ is weakly contained in λG/N . Pulling this back to G via the quotient map q : G → G/N shows that (H, γ, T ) is also a rep- resentative of 1G ∈ KK G(C, C) if we regard γ as a representation of G. Since γ ≺ λG/N = λG/N ⊗ 1G ≺ λG/N ⊗ uG = v we see that γ is weakly contained in v and the theorem implies a KK-equivalence between A ⋊α,u G and A ⋊α,vKLQ G for any system (A, G, α). More generally, suppose that G is a locally compact group such that 1G ∈ KK G(C, C) is represented by some Kasparov cycle (H, γ, T ). Let v = γ ⊗uG. Then v is universally absorbing and the theorem applies to the KLQ-functor ⋊vKLQ. 7. Lp examples Some of the most interesting examples of exotic crossed products come from conditions on the decay of matrix coefficients. In this section, we use examples to explore some of the phenomena that can occur. There is no doubt much more to say here. For a locally compact group G and p ∈ [1, ∞], let Ep⊆ B(G) denote the weak- closure of the intersection Lp(G) ∩ B(G). This is clearly an ideal in B(G), and so gives rise to a KLQ-crossed product, and in particular a completion C ∗ Ep(G) of the group algebra. Recall from [6, Proposition 2.11 and Proposition 2.12]2 that C ∗ Ep (G) is always the reduced completion for p ≤ 2, and that C ∗ u(G) for some p < ∞ if and only if G is amenable. Of course, since E∞ = B(G), we always get C ∗ E∞ (G) = C ∗ Similarly, let E0 ⊆ B(G) denote the weak-closure of C0(G) ∩ B(G), which is also an ideal. For countable discrete groups, Brown and Guentner [6, Corollary 3.4] have shown that E0 = B(G) if and only if G is a-T-menable. In [23] Jolissaint extended this to general second countable3 locally compact groups. u (G). Ep(G) = C ∗ The following theorem records some properties of the algebras C ∗ Ep(G) for the free group on two generators. Theorem 7.1. Let G = F2 be the free group on two generators. Then one has the following results. (1) For p ∈ [2, ∞] the ideals Ep, or equivalently the C ∗-algebras C ∗ Ep(G), are all different (in the sense that the identity on Cc(G) does not extend to isomorphisms of these algebras for different parameters p). 2This reference only studies the discrete case, but the same proofs work for general locally compact groups. 3The techniques of Brown and Guentner, and of Jolissaint, in fact show that E0 = B(G) if and only if G has the Haagerup approximation property in the sense of [10, 1.1.1 (2)] without any second countability assumptions. The Haagerup approximation property is equivalent to all the other possible definitions of a-T-menability in the second countable case, but not in general: see [10, Section 1.1.1 and Theorem 2.1.1] for more details on this. 34 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT (2) The union of the ideals Ep for p ∈ [2, ∞) is weak-dense in B(G) = E0 = E∞. (3) The C ∗-algebras C ∗ Ep(G), p ∈ [2, ∞], all have the same K-theory (in the strong sense that the canonical quotient maps between them induce KK- equivalences). Proof. Properties (1) and (2) follow from Okayasu's work [31]. Property (3) is a consequence of Theorem 6.6 since F2 is a-T-menable and the corresponding KLQ- functors are correspondence functors. (cid:3) It seems interesting to ask exactly which other groups have properties (1), (2) and (3) from Theorem 7.1. We now discuss this. First, we look at property (1). If a group G has this property, then clearly G must be non-amenable, otherwise all the spaces Ep, p ∈ {0} ∪ [1, ∞], are the same. It is easy to see that property (1) for F2 implies that this property holds for all discrete groups that contain F2 as a subgroup. It is thus quite conceivable that it holds for all non-amenable discrete groups. Property (1) also holds for some non-discrete groups: it was observed recently by Wiersma (see [38, §7]), using earlier results of Elias Stein, that (1) holds for w∗ SL(2, R). Indeed, it is shown in [38, Theorem 7.3] that p 7→ Ep = B(G) ∩ Lp(G) is a bijection between [2, ∞] and the set of weak-closed ideals E ⊆ B(G) for G = SL(2, R). However, property (1) does not hold for all non-amenable locally compact groups as the following example shows. Example 7.2. Let G be a non-compact simple Lie group with finite center and rank at least two. It is well-known that such a group is non-amenable. Cowling [12, page 233]4 has shown that there exists p ∈ (2, ∞) (depending on G) such that all non-trivial irreducible unitary representations of G have all their matrix coefficients in Lp(G). Hence for all q ∈ [p, ∞) we have that C ∗ Ep(G). Indeed, this follows as the sets of irreducible representations of these C ∗-algebras are the same: they identify canonically with the set of all non-trivial irreducible unitary representations of G. Eq (G) = C ∗ We now look at property (2). As written, the property can only possibly hold for a-T-menable groups. Simple a-T-menable Lie groups of rank one (i.e. SU (n, 1) and SO(n, 1)) have the property by already cited results of Cowling [12, page 233]. For discrete groups, one has the following sufficient condition for property (2) to hold. Lemma 7.3. Say G is a finitely generated discrete group, and fix a word length l : G → N. Assume moreover there exists a negative type function ψ : G → R+ such that ψ(g) ≥ c · l(g) for some c > 0 and all g ∈ G. Then the union ∪Ep is weak-dense in B(G). Proof. Let φ be any element of B(G). For t > 0 let φt(g) = e−tψ(g), which is positive type by Schoenberg's theorem. For any fixed t > 0 and all p suitably large, the function φt will be in lp(G): indeed this follows from the estimate on ψ and as l(g) ≤ r} ≤ ebr for any r > 0 (the latter there exists b > 0 such that {g ∈ G : fact is an easy consequence of finite generation). Hence the functions φ · φt are all 4See [19, Corollary 3.3.13] for more detail in the special case when G = SL(n, R), n ≥ 3. EXOTIC BAUM-CONNES CONJECTURE 35 in some Ep (where p will in general vary as t does), and they converge pointwise, whence weakly, to φ. (cid:3) The condition in the lemma is closely related to the equivariant compression of G [18], and has been fairly well-studied. It is satisfied for example by finitely generated free groups, but also much more generally: for example, it can be easily deduced from [30] that any group acting properly cocompactly by isometries on a CAT(0) cube complex has this property, and from [5, Section 2.6] that any group acting properly cocompactly by isometries on a real hyperbolic space does. Note, however, that [2] implies that the lemma only applies to a subclass of finitely generated a-T- menable groups. Finally, we look at property (3). By Theorem 6.6 it holds for all second countable K-amenable groups (in particular, all a-T-menable groups). The following example shows that it fails in general. r (G)). Example 7.4. Say G = SP (n, 1), and n ≥ 3. Prudhon [35] has shown that there are infinitely many irreducible representations (π, Hπ) of G -- the so-called isolated series representations -- such that there is a direct summand of C ∗(G) isomorphic to K(Hπ), and moreover that the corresponding direct summand K0(K(Hπ)) ∼= Z of the group K0(C ∗(G)) is in the kernel of the natural map λ∗ : K0(C ∗(G)) → K0(C ∗ It follows from the previously cited work of Cowling [12, page 233] that any non-trivial irreducible representation of SP (n, 1) extends to C ∗ Ep(G) for some p < ∞. Combining all of this, it follows that there exists p ∈ (2, ∞) such that C ∗ Ep(G) has a direct summand isomorphic to K(H) for some separable Hilbert space H, and that the corresponding direct summand K0(K(H)) ∼= Z of K0(C ∗ Ep(G)) is in the kernel of the map on K-theory induced by the natural quotient C ∗ Ep(G) → C ∗ E2 (G). Hence property (3) fails for SP (n, 1) for n ≥ 3. One can deduce from work of Pierrot [34] that property (3) fails for SL(4, C) and SL(5, C) in a similar way. The work of Prudhon and Pierrot cited above uses detailed knowledge of the representation theory of the groups involved. It would be interesting to have more directly accessible examples, or techniques that showed similar results for discrete groups. For example, it is natural to guess that similar failures of property (3) occur for discrete hyperbolic groups with property (T), and for lattices in higher rank simple Lie groups; proving such results seems to require new ideas, however. 8. Morita compatibility and the minimal exact correspondence functor We now want to relate our results to the constructions of [4], in which a new version of the Baum-Connes conjecture is discussed. In fact, the authors consider a certain minimal exact and Morita compatible (in a sense explained below) crossed- product functor, which they call ⋊E , and the new version of the Baum-Connes conjecture formulated by Baum, Guentner and Willett in that paper asserts that the Baum-Connes assembly map should always be an isomorphism if we replace reduced crossed products by ⋊E crossed products. We should note that if G is exact, then ⋊E = ⋊r, so that in this case the new conjecture coincides with the old one. However, all known counter examples for the original conjecture are due to the existence of non-exact groups and cannot be shown to be counterexamples to (8.1) (A ⋊α,µinf G)b = \µ∈Σ (A ⋊α,µ G)b, 36 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT the reformulated conjecture. The most remarkable result of [4] shows that many such examples actually do satisfy the reformulated conjecture. Note that Baum, Guentner and Willett also constructed a direct assembly map for the crossed-product functor ⋊E where they used the descent in E-theory instead of KK-theory, since at the time of writing it was not clear to them whether the KK-theory descent exists (e.g., see the open question [4, Question 8.1 (vii)]). In this section we show that this is indeed the case, so that KK-methods do apply to the reformulation of the Baum-Connes conjecture. We start the discussion by showing that the projection property is inherited by the infimum of a collection of crossed-product functors which satisfy this prop- erty. Recall from [4, Lemma 3.7] that, starting with a collection of crossed-product functors {⋊µ : µ ∈ Σ} (where we do not assume that Σ will be a set), a new crossed-product functor (A, α) 7→ A ⋊α,µinf G can be constructed, which should be understood as the infimum of the functors in the collection {⋊µ : µ ∈ Σ}. The crossed product A ⋊α,µinf G is the unique quotient of A ⋊α,u G such that the set (A ⋊α,µinf G)b of equivalence classes of irreducible representations (denoted Sµinf (A) in [4]) is given by the formula where we view each dual space (A⋊α,µG)b as a subset of (A⋊α,uG)b via composition with the quotient maps qA,µ : A ⋊α,u G → A ⋊α,µ G. Alternatively, one may define A ⋊α,µinf G as the quotient (A ⋊α,u G)/Iµinf with Iµinf = Xµ∈Σ Iµ, the closed sum of all ideals Iµ := ker qA,µ, µ ∈ Σ. Note that this does not cause any set theoretic problems, since the collection of ideals {Iµ : µ ∈ Σ} is a set. It has been shown in [4, Lemma 3.7] that (A, α) 7→ A ⋊α,µinf G is indeed a crossed-product functor. Moreover, it is shown in [4, Theorem 3.8] that taking the infimum of a collection of exact crossed-product functors will again give an exact crossed-product functor. We now show Proposition 8.2. Let {⋊µ : µ ∈ Σ} be a collection of correspondence crossed- product functors. Then its infimum ⋊µinf will be a correspondence functor as well. We need a lemma: Lemma 8.3. Suppose that (A, α) is a G-algebra and B ⊆ A is a G-invariant hered- itary subalgebra. For each covariant representation (π, U ) of (A, G, α) on some Hilbert space (or Hilbert module) H let us write (πB, U ) for the covariant represen- tation of (B, G, α) which acts on the subspace (or submodule) HB := π(B)H via the restriction πB and the given unitary representation U . Then (8.4) (8.5) In particular, if p ∈ M(A) is a G-invariant projection, then (B ⋊α,u G)b = {πB ⋊ U : π ⋊ U ∈ (A ⋊α,u G)b , π(B) 6= {0}}. (pAp ⋊α,u G)b = {πpAp ⋊ U : π ⋊ U ∈ (A ⋊α,u G)b , π(p) 6= 0}. EXOTIC BAUM-CONNES CONJECTURE 37 Proof. We first consider the ideal I := ABA. It is then well known that (I ⋊α,u G)b = {πI ⋊ U : π ⋊ U ∈ (A ⋊α,u G)b, π(I) 6= {0}}. Note that we then have π(I)H = H if π(I) 6= {0}, since π ⋊ U is irreducible and π(I)H is π ⋊ U -invariant. Note that π(ABA) = π(A)π(B)π(A) = {0} if and only if π(B) = {0}, which follows by approximating b ∈ B by aibai, where (ai) runs through a bounded approximate unit of A. Using this and the above observation for the ideal I = ABA, we may now assume that B is full in the sense that ABA = A and to show that in this situation we have (B ⋊α,u G)b = {πB ⋊ U : π ⋊ U ∈ (A ⋊α,u G)b}. For this we observe that (BA, α) is a G-equivariant B − A equivalence bimodule with B- and A-valued inner products given by Bhc di = cd∗ and hc diA = c∗d for c, d ∈ BA. Thus, BA induces a bijection between the equivalence classes of irreducible covariant representations of (A, G, α) to those of (B, G, α) by sending a representation (π, U ) to the BA-induced representation (IndBA π, IndBA U ) on the Hilbert space (BA) ⊗π H on which IndBA π is given by the left action of B on the first factor BA and IndBA U = α ⊗ U (e.g., see [11, 15]). Now observe that there is a unique unitary operator V : BA ⊗π H → π(B)H given on elementary tensors by sending ba ⊗ ξ to π(ba)ξ, which clearly intertwines (IndBA π, IndBA U ) with (πB, U ). This proves (8.4) and (8.5) then follows from the fact that π(pAp) 6= 0 if and only if π(p) 6= 0. (cid:3) Corollary 8.6. A given crossed-product functor (A, α) → A ⋊α,µ G satisfies the projection property (hence is a correspondence functor) if and only if (8.7) (pAp ⋊α,µ G)b = {πpAp ⋊ U : π ⋊ U ∈ (A ⋊α,µ G)b, π(p) 6= 0}. holds for every G-invariant projection p ∈ M(A). Proof. Let p denote the image of p in M(A ⋊µ G) under the canonical map. Then p(A ⋊µ G)p is a corner of A ⋊α,µ G which implies that (p(A ⋊µ G)p)b = {(π ⋊ U ) p(A⋊µG) p : π ⋊ U ∈ (A ⋊α,µ G)b, π ⋊ U (p) 6= 0}. A computation on the level of functions in Cc(G, pAp) shows that (π ⋊ U ) p(A⋊µG) p is the integrated form of (πpAp, U ). Since π ⋊ U (p) = π(p) the homomorphism ι ⋊ G : pAp ⋊α,µ G ։ p(A ⋊µ G)p ⊆ A ⋊α,µ G will be faithful if and only if equation (8.7) holds. (cid:3) Proof of Proposition 8.2. By Theorem 4.9 it suffices to show that if all functors ⋊µ satisfy the projection property, the same holds for ⋊µinf . So we may assume that equation (8.7) holds for each µ. But then it also holds for µinf, since for every G-invariant projection p ∈ M(A) we have (pAp ⋊α,µinf G)b = \µ∈Σ (pAp ⋊α,µ G)b = \µ∈Σ {πpAp ⋊ U : π ⋊ U ∈ (A ⋊α,µ G)b, π(p) 6= 0} = {πpAp ⋊ U : π ⋊ U ∈ (A ⋊α,µinf G)b, π(p) 6= 0}. Hence the result follows from Corollary 8.6. (cid:3) 38 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT The following corollary gives a counterpart to the minimal exact Morita compat- ible crossed-product functor (A, α) 7→ A ⋊α,E G as constructed in [4]. Since it is a correspondence functor, it allows a descent in KK-theory and the results of §6 will apply. We shall see below that, at least for second countable groups and on the category of separable G-algebras, the functor ⋊E coincides with the minimal exact correspondence functor ⋊ECorr of Corollary 8.8. For every locally compact group G there exists a smallest exact correspondence crossed-product functor (A, α) 7→ A ⋊α,ECorr G. Proof. We simply define ⋊ECorr as the infimum of the collection of all exact corre- spondence functors ⋊µ. By Proposition 8.2 it will be a correspondence functor and by [4, Theorem 3.8] it will be exact. (cid:3) The minimal exact correspondence functor ⋊ECorr enjoys the following property: Corollary 8.9. Let (D, δ) be any unital G-C ∗-algebra. Then for each G-C ∗-algebra (A, α) the inclusion jA : A → A ⊗max D; a 7→ a ⊗ 1 descends to an injective ∗-homomorphism jA ⋊ECorr G : A ⋊ECorr G → (A ⊗max D) ⋊ECorr G. In other words, we have ⋊µ = ⋊µmax D if µ = ECorr. If D is exact, the same holds if we replace ⊗max by ⊗min. Proof. This follows from minimality of ⋊ECorr and the (easily verified) fact that is exact if ⋊µ is exact and either: ν = max; or both ν = min and D is ⋊µν exact. (cid:3) D In order to see that the functor ⋊ECorr coincides with the "minimal exact Morita compatible functor" ⋊E, we need to recall the notion of Morita compatibility as de- fined in [4] and then compare it with the properties of our correspondence functors. If G is a locally compact group we write KG := K(ℓ2(N))⊗ K(L2(G)) equipped with the G-action AdΛ, where Λ = 1 ⊗ λ : G → U(ℓ2(N) ⊗ L2(G)) and λ denotes the left regular representation of G. Then there is a canonical isomorphism of maximal crossed products Φ : (A ⊗ KG) ⋊α⊗AdΛ,u G ∼=−→ (A⋊α,uG) ⊗ KG given by the integrated form of the covariant homomorphism (iA ⊗ idKG, iG ⊗ Λ) of (A ⊗ KG, G, α ⊗ AdΛ) into M((A⋊α,uG) ⊗ KG). Then Baum, Guentner and Willett defined a crossed-product functor to be Morita compatible if the composition of Φ with the quotient map (A⋊α,uG) ⊗ KG → (A⋊α,µG) ⊗ KG factors through an isomorphism Φµ : (A ⊗ KG) ⋊α⊗AdΛ,µ G ∼=−→ (A⋊α,µG) ⊗ KG. Proposition 8.10. Suppose that G is a locally compact group and that (A, α) 7→ A ⋊α,µ G is a crossed-product functor on the category of σ-unital C ∗-algebras. Then the following three conditions are equivalent: (1) ⋊µ is Morita compatible as defined above. (2) ⋊µ has the full projection property. (3) ⋊µ is strongly Morita compatible. Note that the assumption that all G-algebras are σ-unital is only used in the proof (1) ⇒ (2). All other statements hold in full generality. EXOTIC BAUM-CONNES CONJECTURE 39 Proof. The proof of (2) ⇔ (3) is exactly the same as the proof of (2) ⇔ (4) in Lemma 4.12 and we omit it here. The proof of (3) ⇒ (1) follows from Corollary 5.4. (1) ⇒ (2): Let (A, α) be a σ-unital G-algebra and let p ∈ M(A) be a G-invariant full projection. We need to show that the inclusion ι : pAp ֒→ A descends to give a faithful inclusion ι ⋊ G : pAp ⋊µ G ֒→ A ⋊µ G. This will be the case if and only if (ι ⋊ G) ⊗ idKG : pAp ⋊µ G ⊗ KG → A ⋊µ G ⊗ KG Let α ⊗ AdΛ be the corresponding action of G on A ⊗ KG. is faithful, and we shall now deduce from Morita compatibility that this is the case. It follows from [29, Corollary 2.6] that there exists a G-invariant partial isometry v ∈ M(A ⊗ KG) such that v∗v = 1A ⊗ 1KG and vv∗ = p ⊗ 1KG. This implies that the mapping Adv∗ : pAp ⊗ KG → A ⊗ KG; (a ⊗ k) 7→ v∗(a ⊗ k)v; is a G-equivariant isomorphism. Hence it induces an isomorphism ∼=−→ (A ⊗ KG) ⋊µ G. Adv∗ ⋊ G : (pAp ⊗ KG) ⋊µ G Let v denote the image of v in M((A ⋊α,µ G) ⊗ KG) under the composition Φµ ◦ iA⊗KG, where iA⊗KG : M(A ⊗ KG) → M((A ⊗ KG) ⋊µ G)) denotes the canonical map. Then v is a partial isometry with v∗v = 1A⋊G⊗K and vv∗ = p ⊗ 1KG where p denotes the image of p in M(A ⋊µ G). Consider the diagram (pAp ⋊α,µ G) ⊗ KG ι⋊G⊗idK/ / p(A ⋊α,µ G)p ⊗ KG Adv∗ / (A ⋊α,µ G) ⊗ KG Φµ Φµ (pAp ⊗ KG) ⋊α⊗AdΛ,µ G Adv∗⋊G / (A ⊗ KG) ⋊α⊗AdΛ,µ G One easily checks on the generators that this diagram commutes. All maps but ι⋊G⊗idK are known to be isomorphisms, hence ι⋊G⊗idK must be an isomorphism as well. (cid:3) Lemma 8.11. Let G be a second countable group, and say ⋊µ is a crossed-product functor, defined on the category of separable G-C ∗-algebras. For each G-C ∗-algebra A, let (Ai)i∈I be the net of G-invariant separable sub C ∗-algebras of A ordered by inclusion. Define A ⋊sep,µ G by A ⋊sep,µ G := lim i∈I Ai ⋊µ G. Then A ⋊sep,µ G is a completion of Cc(G, A) with the following properties. (1) (A, α) 7→ A ⋊α,sep,µ G is a crossed-product functor for G. (2) If ⋊µ has the ideal property, the same holds for ⋊sep,µ. (3) If ⋊µ is a correspondence functor, the same holds for ⋊sep,µ. (4) If ⋊µ is exact, the same holds for ⋊sep,µ. (5) If ⋊µ is Morita compatible, then the same holds for ⋊sep,µ. (6) If ⋊ν is any crossed-product functor extending ⋊µ to all G-C ∗-algebras, then ⋊sep,µ ≥ ⋊ν. Proof. Note first that as G is second countable any f ∈ Cc(G, A) has image in some second countable G-invariant subalgebra of A, and thus A ⋊sep,µ G contains / O O / O O 40 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT Cc(G, A) as a dense subset as claimed. It is automatically smaller than the maximal completion, and it is larger than the reduced as the canonical maps Ai ⋊µ G → Ai ⋊r G → A ⋊r G give a compatible system, whence there is a map A ⋊sep,µ G → A ⋊r G (which is equal to the identity on Cc(G, A)) by the universal property of the direct limit. (1) Functoriality follows from functoriality of the direct limit construction. (2) If I is an ideal in A, then for each separable G-invariant C ∗-subalgebra Ai of A as in the definition of ⋊sep,µ, define Ii = I ∩ Ai. Then the directed system (Ii)i∈I is cofinal in the one used to define I ⋊sep,µ G, whence I ⋊sep,µ G = lim i (Ii ⋊µ G). The result now follows as each map Ii ⋊µ G → Ai ⋊µ G is injective by the ideal property for ⋊µ, and injectivity passes to direct limits. (3) We prove the projection property. Let p be a G-invariant projection in the multiplier algebra of A. The argument is similar to that of part (2). The only additional observation needed is that if we set Bi to be the C ∗-subalgebra of A generated by Aip and Ai, then the net (Bi)i∈I is cofinal in the net defining A⋊sep,µG. Moreover, p is in the multiplier algebra of each Bi, whence each of the injections pBip → Bi induces an injection on ⋊µ crossed products by the projection property for ⋊µ. The result follows now on passing to the direct limit of the inclusions (4) This is similar to parts (2) and (3). Given a short exact sequence pBip ⋊µ G → Bi ⋊µ G. 0 → I → A → B → 0 with quotient map π : A → B, let (Ai) be the net of separable G-invariant subalge- bras of A, and consider the net of short exact sequences 0 → Ai ∩ I → Ai → π(Ai) → 0. The nets (Ai ∩ I) and (π(Ai)) are cofinal in the nets defining the ⋊sep,µ crossed products for I and B respectively. On the other hand, exactness of µ gives that all the sequences 0 → (Ai ∩ I) ⋊µ G → Ai ⋊µ G → π(Ai) ⋊µ G → 0 are exact. The result follows as exactness passes to direct limits. (5) This again follows a similar pattern: the only point to observe is that the collection of separable G-invariant C ∗-algebras of A ⊗ KG of the form Ai ⊗ KG, where Ai is a separable G-invariant C ∗-subalgebra of A, is cofinal in the direct limit defining (A ⊗ KG) ⋊sep,µ G. (6) From functoriality of ⋊ν, there is a compatible system of ∗-homomorphisms Ai ⋊µ G = Ai ⋊ν G → A ⋊ν G. From the universal property of the direct limit there is a ∗-homomorphism A ⋊sep,µ G → A ⋊ν G which is clearly the identity on Cc(G, A), completing the proof. (cid:3) EXOTIC BAUM-CONNES CONJECTURE 41 Example 8.12. Every non-amenable second countable group has a crossed product ⋊µ for which the crossed-product functor ⋊sep,µ (in which we restrict ⋊µ to the category of separable G-algebras) does not coincide with ⋊µ. For this, we use the construction of §2.5. Let H be a Hilbert space with uncountable Hilbert space dimension. Let S denote the set of all separable subalgebras of K(H), all equipped with the trivial G-action. We define a crossed-product functor ⋊µ by defining A ⋊µ G as the completion of Cc(G, A) by the norm kf kµ := max(cid:8)kf kr, sup{kφ ◦ f kB⋊uG : B ∈ S, φ ∈ HomG(A, B)}(cid:9), where HomG(A, B) denotes the set of G-equivariant ∗-homomorphisms from A to B. As explained in §2.5, this is a functor. Now consider A = K(H) with the trivial G-action. Then K(H) ⋊µ G = K(H) ⋊r G ∼= K(H) ⊗ C ∗ r (G), since there are no nontrivial homomorphisms from K(H) into any of its separable subalgebras. On the other hand we have Bi ⋊µ G ∼= B ⋊u G for every separable subalgebra B ⊆ K(H), which implies K(H)⋊sep,µ G ∼= K(H)⋊u G ∼= K(H) ⊗ C ∗(G). Corollary 8.13. The following functors from the category of separable G-C ∗-alge- bras to the category of C ∗-algebras are the same. (1) The restriction to separable G-algebras of the minimum ⋊ECorr over all exact correspondence functors defined for all G-C ∗-algebras. (2) The restriction to separable G-algebras of the minimum ⋊E over all exact Morita compatible functors defined for all G-C ∗-algebras. (3) The minimum over all exact correspondence functors defined for separable G-C ∗-algebras. (4) The minimum over all exact Morita compatible functors defined for separa- ble G-C ∗-algebras. Proof. Label the crossed products (defined on separable G-C ∗-algebras) appearing in the points above as ⋊1, ⋊2, ⋊3 and ⋊4. By (2) ⇒ (1) of Proposition 8.10 (which holds in full generality) we clearly have ⋊1 ≥ ⋊2 ≥ ⋊4, and ⋊1 ≥ ⋊3 ≥ ⋊4. It thus suffices to show that ⋊4 ≥ ⋊1. Indeed, let ⋊sep,4 be the extension given by Lemma 8.11. Then ⋊sep,4 is an exact and Morita compatible functor on the category of σ-unital G-C ∗-algebras, whence it is an exact correspondence functor on this category by Proposition 8.10. Hence in particular ⋊4 was actually an exact correspondence functor on the category of separable G-algebras to begin with, and so ⋊sep,4 is an exact correspondence functor on the category of all G-algebras. It is thus one of the functors that ⋊1 is the minimum over, so ⋊1 ≤ ⋊sep,4 on the category of all G-algebras, and in particular ⋊1 ≤ ⋊4 on the category of separable G-algebras as required. (cid:3) Remark 8.14. The above corollary shows that (at least for second countable groups and separable G-algebras) the reformulation of the Baum-Connes conjecture in [4] is equivalent to the statement that asECorr (A,G) : K top ∗ (G; A) 7→ K∗(A ⋊α,ECorr G) 42 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT is always an isomorphism, where ⋊ECorr is the minimal exact correspondence func- tor. By the results of §6 we can use the full force of equivariant KK-theory to study this version of the conjecture. 9. Remarks and questions 9.1. Crossed products associated to U Cb(G). We only know one general con- struction of exact correspondence functors. Fix a unital G-algebra (D, δ). Let (A, α) be a G-algebra. The ∗-homomorphism induces an injective ∗-homomorphism A → A ⊗max D, a 7→ a ⊗ 1 Cc(G, A) → (A ⊗max D) ⋊α⊗δ,u G. Recall from Section 2.4 that the crossed product A ⋊umax closure of the image of Cc(G, A) inside (A ⊗ D) ⋊α⊗δ,u G. D G is by definition the The crossed product ⋊umax D is always exact by [4, Lemma 5.4], and always a correspondence functor by Corollary 4.20. Here we will show that the collection {⋊umax D : (D, δ) a G-algebra} of exact correspondence functors -- which are the only exact correspondence functors we know for general G -- has a concrete minimal element. Lemma 9.1. Let U Cb(G) denote the C ∗-algebra of bounded, left-uniformly contin- uous complex-valued functions on G, equipped with the G-action induced by trans- lation. Then for any unital G-algebra D, ⋊umax D ≥ ⋊umax U Cb(G) . Proof. Consider any state φ : D → C, and note that for any d ∈ D, the function dφ : G → C defined by dφ : g 7→ φ(δg(d)) is bounded and uniformly continuous, hence an element of U Cb(G). Consider the map Φ : D → U Cb(G), d 7→ dφ. This is clearly equivariant, unital and positive. Moreover, positive maps of C ∗- algebras with commutative codomains are automatically completely positive: in- deed, post-composing with multiplicative linear functionals reduces this to show- ing that a state is a completely positive map to C, and this follows from the GNS-construction which shows in particular that a state is a compression of a ∗-homomorphism by a one-dimensional projection. Hence Φ is completely positive. Now consider the diagram (A ⊗max D) ⋊u G (1⊗Φ)⋊uG / (A ⊗max U Cb(G)) ⋊u G , ιD A ⋊umax D G ιU Cb(G) A ⋊umax U Cb(G) G where the vertical maps are the injections given by definition of the crossed products on the bottom row, and the top line exists by Theorem 4.9 and the fact that 1 ⊗ Φ is completely positive. As Φ is unital, the composition (1 ⊗ Φ) ⋊u G ◦ ιD : A ⋊umax D G → ιUCb(G)(A ⋊umax U Cb(G) G) / O O O O EXOTIC BAUM-CONNES CONJECTURE 43 identifies with a ∗-homomorphism from A ⋊umax the identity map on Cc(G, A). This gives the desired conclusion. G to A ⋊umax D U Cb(G) G that extends (cid:3) The crossed product ⋊umax U Cb(G) the following from [1, Section 3]. has another interesting property. Indeed, recall Definition 9.2. A locally compact group G is amenable at infinity if it admits an amenable action on a compact topological space. Lemma 9.3. Say G is amenable at infinity. Then for any G-algebra (A, α), A ⋊α,umax U Cb(G) G = A ⋊α,r G. Proof. If G is amenable at infinity, then [1, Proposition 3.4] implies that the action of G on the spectrum X of U Cb(G) is amenable. Hence for any G-algebra A, the tensor product A ⊗max U Cb(G) is a G-C(X) algebra for an amenable G-space X. The result follows from [1, Theorem 5.4], which implies that (A ⊗max U Cb(G)) ⋊u G = (A ⊗max U Cb(G)) ⋊r G. (cid:3) A group that is amenable at infinity is always exact [1, Theorem 7.2]. For discrete groups, the converse is true by [32], but this is an open question in general; nonethe- less, many exact groups, for example all almost connected groups [1, Proposition 3.3], are known to be amenable at infinity. To summarise, we have shown that ⋊umax is an exact correspondence functor; that it is minimal among a large family of exact correspondence functors; and that for many (and possibly all) exact groups, it is equal to the reduced crossed product. The following question is thus very natural. U Cb(G) Question 9.4. Is the crossed product ⋊umax spondence crossed product ⋊ECorr? U Cb(G) equal to the minimal exact corre- There are other natural questions one could ask about ⋊umax : for example, is it a KLQ functor 'in disguise'? One could also ask this about any of the functors ⋊umax U Cb(G) . D 9.2. Questions about Lp functors. Recall from Section 7 that Ep denotes the weak-closure of B(G) ∩ Lp(G) in B(G) for p ∈ [2, ∞], and E0 is the weak-closure of B(G) ∩ C0(G). The following questions about these ideals and the corresponding group algebras and crossed products seem natural and interesting. (1) Say G is an exact group. Are all of the functors ⋊Ep exact? More generally, if G is an exact group, are all KLQ-crossed products exact? Note that Example 3.5 shows that any non-amenable exact group admits a non-exact crossed-product functor. (2) Are there non-exact groups and p ∈ {0} ∪ (2, ∞) such that ⋊Ep is exact? (3) For which locally compact groups does the canonical quotient q : C ∗ E0 (G) → C ∗ r (G) induce an isomorphism on K-theory (one could also ask this for other p)? This cannot be true in general by Example 7.4, but we do not know any discrete groups for which it fails. (4) For which groups G are all the exotic group C ∗-algebras C ∗ Ep(G) differ- ent? Example 7.2 shows that this cannot be true for general non-amenable groups, but it could in principal be true for all non-amenable discrete groups. 44 ALCIDES BUSS, SIEGFRIED ECHTERHOFF, AND RUFUS WILLETT (5) For which groups is E0 the weak-closure of ∪p<∞Ep? This holds for all simple Lie groups with finite center by results of Cowling [12], and we showed it holds for a fairly large class of discrete groups in Lemma 7.3. Conceivably, it could hold in full generality. More generally, one can ask: is Eq is the weak- closure of ∪p<qEp for all q ∈ [2, ∞)? Similarly, is it true that Ep = ∩q>pEq for all p ∈ [2, ∞)? Okayasu [31, Corollary 3.10] has shown that this is true for F2 and Cowling, Haagerup and Howe [13] have shown that E2 = ∩p>2Ep is always true. Not much else seems to be known here, however. References [1] Claire Anantharaman-Delaroche, Amenability and exactness for dynamical systems and their C ∗-algebras, Trans. Amer. Math. Soc. 354 (2002), no. 10, 4153 -- 4178. [2] Tim Austin, Amenable groups with very poor compression into Lebesgue spaces, Duke Math. J. 159 (2011), no. 2, 187 -- 222. [3] Paul Baum, Alain Connes, and Nigel Higson, Classifying space for proper actions and K-theory of group C ∗-algebras, C ∗-Algebras: 1943 -- 1993 (San Antonio, TX, 1993), Con- temp. Math., vol. 167, Amer. Math. Soc., Providence, RI, 1994, pp. 240 -- 291, DOI 10.1090/conm/167/1292018, (to appear in print). MR 1292018 [4] Paul Baum, Erik Guentner, and Rufus Willett, Expanders, exact crossed products, and the Baum-Connes conjecture (2013), To appear, Annals of K-theory. arXiv: 1311.2343. [5] Bachir Bekka, Pierre de la Harpe, and Alain Valette, Kazhdan's property (T), New Mathe- matical Monographs, vol. 11, Cambridge University Press, 2008. [6] Nathanial P. Brown and Erik Guentner, New C ∗-completions of discrete groups and related spaces, Bull. Lond. Math. Soc. 45 (2013), 1181-1193. [7] Nathanial P. Brown and Narutaka Ozawa, C ∗-algebras and finite-dimensional approxima- tions, Graduate Studies in Mathematics, vol. 88, American Mathematical Society, Providence, RI, 2008. [8] Alcides Buss and Siegfried Echterhoff, Universal and exotic generalized fixed-point algebras for weakly proper actions and duality, Indiana Univ. Math. J. 63 (2014), 1659 -- 1701, DOI 10.1512/iumj.2014.63.5405. [9] , Imprimitivity theorems for weakly proper actions of locally compact groups, Ergodic Theory Dynam. Systems, DOI 10.1017/etds.2014.36, (to appear in print). [10] Pierre-Alain Cherix, Michael Cowling, Paul Jolissaint, Pierre Julg, and Alain Valette, Groups with the Haagerup property (Gromov's a-T-menability), Progress in Mathematics, vol. 197, Birkhäuser, Basel, Boston, Berlin, 2001. [11] François Combes, Crossed products and Morita equivalence, Proc. London Math. Soc. (3) 49 (1984), no. 2, 289 -- 306. MR 748991 [12] Michael Cowling, The Kunze-Stein phenomenon, Ann. of Math. 107 (1984), 209 -- 324. [13] Michael Cowling, Uffe Haagerup, and Roger Howe, Almost L2 matrix coefficients, J. Reine Angew. Math. 387 (1988), 97 -- 100. [14] Joachim Cuntz, K-theoretic amenability for discrete groups, J. Reine Angew. Math. 344 (1983), 180 -- 195, DOI 10.1515/crll.1983.344.180. [15] Siegfried Echterhoff, Morita equivalent twisted actions and a new version of the Packer -- Raeburn stabilization trick, J. London Math. Soc. (2) 50 (1994), no. 1, 170 -- 186, DOI 10.1112/jlms/50.1.170. MR 1277761 [16] Siegfried Echterhoff, Steven P. Kaliszewski, John Quigg, and Iain Raeburn, A categorical approach to imprimitivity theorems for C ∗-dynamical systems, Mem. Amer. Math. Soc. 180 (2006), no. 850, viii+169. MR 2203930 [17] J. M. G. Fell, The dual spaces of C ∗-algebras, Trans. Amer. Math. Soc. 94 (1960), 365 -- 403. [18] Erik Guentner and Jerome Kaminker, Exactness and uniform embeddability of discrete groups, J. London Math. Soc. 70 (2004), 703 -- 718. EXOTIC BAUM-CONNES CONJECTURE 45 [19] Roger Howe and Eng Chye Tan, Non-abelian harmonic analysis, Universitext, Springer- Verlag, New York, NY, 1992. [20] Nigel Higson and Gennadi Kasparov, E-theory and KK-theory for groups which act prop- erly and isometrically on Hilbert space, Invent. Math. 144 (2001), no. 1, 23 -- 74, DOI 10.1007/s002220000118. [21] Nigel Higson, Vincent Lafforgue, and Georges Skandalis, Counterexamples to the Baum- Connes conjecture, Geom. Funct. Anal. 12 (2002), no. 2, 330 -- 354, DOI 10.1007/s00039-002- 8249-5. [22] Sh¯o and Takai Imai Hiroshi, On a duality for C ∗-crossed products by a locally compact group, J. Math. Soc. Japan 30 (1978), no. 3, 495 -- 504, DOI 10.2969/jmsj/03030495. [23] Paul Jolissaint, Notes on C0 representations and the Haagerup property, Bull. Belg. Math. Soc. Simon Stevin 21 (2014), 263 -- 274. [24] Pierre Julg and Alain Valette, K-theoretic amenability for SL2(Qp), and the action on the associated tree, J. Funct. Anal. 58 (1984), no. 2, 194 -- 215, DOI 10.1016/0022-1236(84)90039-9. [25] Steven P. Kaliszewski, Magnus B. Landstad, and John Quigg, Exotic group C ∗-algebras in noncommutative duality, New York J. Math. 19 (2013), 689 -- 711, available at http://nyjm.albany.edu/j/2013/19_689.html . MR 3141810 , Exotic coactions (2013), preprint. arXiv: 1305.5489. [26] [27] Gennadi G. Kasparov, Equivariant KK-theory and the Novikov conjecture, Invent. Math. 91 (1988), no. 1, 147 -- 201, DOI 10.1007/BF01404917. MR 918241 [28] Gennadi G. Kasparov and Georges Skandalis, Groups acting properly on "bolic" spaces and the Novikov conjecture, Ann. of Math. (2) 158 (2003), no. 1, 165 -- 206. MR 1998480 [29] James A. Mingo and William J. Phillips, Equivariant triviality theorems for Hilbert C ∗-modules, Proc. Amer. Math. Soc. 91 (1984), no. 2, 225 -- 230, available at http://www.jstor.org/stable/2044632. MR 740176 [30] Sarah Campbell and Graham Niblo, Hilbert space compression and exactness for discrete groups, J. Funct. Anal. 222 (2005), no. 2, 292 -- 305. [31] Rui Okayasu, Free group C-algebras associated with ℓp, Internat. J. Math. 25 (2014), no. 7, 1450065 (12 pages), DOI 10.1142/S0129167X14500657. MR 3238088 [32] Narutaka Ozawa, Amenable actions and exactness for discrete groups, C. R. Acad. Sci. Paris Sér I Math. 330 (2000), 691 -- 695. [33] Judith A. and Raeburn Packer Iain, Twisted crossed products of C ∗-algebras, Math. Proc. Cambridge Philos. Soc. 106 (1989), no. 2, 293 -- 311, DOI 10.1017/S0305004100078129. [34] F. Pierrot, Induction parabolique et K-théorie de C ∗-algèbres maximales, C. R. Acad. Sci. Paris Sér I Math. 332 (2001), no. 9, 805 -- 808. [35] N. Prudhon, K-theory for Sp(n, 1), J. Funct. Anal. 221 (2005), no. 7, 226 -- 249. [36] Jean-Louis Tu, The Baum-Connes conjecture and discrete group actions on trees, K-Theory 17 (1999), no. 4, 303 -- 318, DOI 10.1023/A:1007751625568. [37] Dana P. Williams, Crossed products of C ∗-algebras, Mathematical Surveys and Monographs, vol. 134, American Mathematical Society, Providence, RI, 2007. MR 2288954 [38] Matthew Wiersma, Lp-Fourier and Fourier-Stieltjes algebras for locally compact groups (2014), eprint. arXiv: 1409.2787v1. E-mail address: [email protected] Departamento de Matemática, Universidade Federal de Santa Catarina, 88.040-900 Florianópolis-SC, Brazil E-mail address: [email protected] Mathematisches Institut, Westfälische Wilhelms-Universität Münster, Einsteinstr. 62, 48149 Münster, Germany E-mail address: [email protected] Mathematics Department, University of Hawai'i at M¯anoa, Keller 401A, 2565 Mc- Carthy Mall, Honolulu, HI 96822, USA
1412.2085	3	1412	2017-05-13T16:12:22	$L_{p}$-improving convolution operators on finite quantum groups	[ "math.OA" ]	We characterize positive convolution operators on a finite quantum group $\mathbb{G}$ which are $L_{p}$-improving. More precisely, we prove that the convolution operator $T_{\varphi}:x\mapsto\varphi\star x$ given by a state $\varphi$ on $C(\mathbb{G})$ satisfies \[ \exists1<p<2,\quad\\|T_{\varphi}:L_{p}(\mathbb{G})\to L_{2}(\mathbb{G})\\|=1 \] if and only if the Fourier series $\hat{\varphi}$ satisfy $\\|\hat{\varphi}(\alpha)\\|<1$ for all nontrivial irreducible unitary representations $\alpha$, if and only if the state $(\varphi\circ S)\star\varphi$ is non-degenerate (where $S$ is the antipode). We also prove that these $L_{p}$-improving properties are stable under taking free products, which gives a method to construct $L_{p}$-improving multipliers on infinite compact quantum groups. Our methods for non-degenerate states yield a general formula for computing idempotent states associated to Hopf images, which generalizes earlier work of Banica, Franz and Skalski.	math.OA	math	Lp-IMPROVING CONVOLUTION OPERATORS ON FINITE QUANTUM GROUPS SIMENG WANG Abstract. We characterize positive convolution operators on a finite quantum group G which are Lp-improving. More precisely, we prove that the convolution operator Tϕ : x 7→ ϕ ⋆ x given by a state ϕ on C(G) satisfies ∃1 < p < 2, kTϕ : Lp(G) → L2(G)k = 1 if and only if the Fourier series ϕ satisfy k ϕ(α)k < 1 for all nontrivial irreducible unitary repre- sentations α, and if and only if the state (ϕ ◦ S) ⋆ ϕ is non-degenerate (where S is the antipode). We also prove that these Lp-improving properties are stable under taking free products, which gives a method to construct Lp-improving multipliers on infinite compact quantum groups. Our methods for non-degenerate states yield a general formula for computing idempotent states associated to Hopf images, which generalizes earlier work of Banica, Franz and Skalski. Introduction The convolution operators or multipliers constitute a central part of Fourier analysis. One among phenomena studied on the circle group T is the existence and behavior of positive Borel measures that convolve Lp(T) into Lq(T) with finite q > p for a given 1 < p < ∞, which are considered to be Lp-improving measures. An example due to Oberlin [Obe82] is the Cantor- Lebesgue measure supported by the usual middle-third Cantor set. Oberlin revealed that, after a careful analysis on the structure of this measure, this result can be reduced to proving that there exists p < 2 such that kµ ⋆ f k2 ≤ kf kp, f ∈ Lp(Z/3Z) where the Lp-norms are those taken with respect to the normalized counting measure on the cyclic group Z/3Z = {0, 1, 2} with three elements and µ is the probability measure with mass 1/2 at 0 and at 2. Motivated by these results, Ritter showed in [Rit84] that, if G is an arbitrary finite group and Tµ : f 7→ µ ⋆ f is the convolution operator associated to a probability measure µ on G, then (∃ p < 2, kTµ : Lp(G) → L2(G)k = 1) ⇔ G = hij−1 : i, j ∈ supp µi, which provides a more general method to construct Lp-improving measures on groups. In this paper we give an alternative approach related to these topics, in the context of quantum groups and noncommutative Lp-spaces. We show that, for a finite quantum group G and a state ϕ on C(G), denoting by ϕ the Fourier series of ϕ and writing ψ = (ϕ ◦ S) ⋆ ϕ, S being the antipode, the following assertions are equivalent (Theorem 4.4): (1) there exists 1 < p < 2 such that, ∀ x ∈ C(G), kϕ ⋆ xk2 ≤ kxkp ; (2) k ϕ(α)k < 1 for all α ∈ Irr(G) \ {1} ; (3) For any nonzero x ∈ C(G)+, there exists n ≥ 1 such that ψ⋆n(x) > 0. The last assertion should be interpreted as claiming that the "support" of ϕ "generates" the quantum group G, which will be explained in the last section. We will illustrate by example in Remark 4.7 that the finiteness condition in the above conclusion is rather crucial and cannot be removed. 2010 Mathematics Subject Classification. Primary: 20G42, 46L89. Secondary: 43A22, 46L30, 46L51. Key words and phrases. Lp-improving operators, compact quantum groups, positive convolution operators. 1 2 SIMENG WANG In particular, the result characterizes the Fourier-Schur multipliers on finite groups which have an Lp-improving property. Let Γ be a finite group and ϕ be a positive definite function on Γ. Let Mϕ be the associated Fourier-Schur multiplier operator determined by Mϕ(λ(γ)) = ϕ(γ)λ(γ) for all γ ∈ Γ. Then ∃1 < p < 2, kMϕxk2 ≤ kxkp, x ∈ C∗(Γ) if and only if ϕ(γ) < 1 for any γ ∈ Γ \ {e}. We should emphasize that our argument relies essentially on new and interesting properties on the unital trace preserving operators on noncommutative Lp-spaces, based on the recent work of Ricard and Xu [RX16]. In fact, the following fact proved in Theorem 1.6 plays a key role in our argument. For a finite dimensional C-algebra A equipped with a faithful tracial state τ , and T : A → A a unital trace preserving map, the Lp-improving property (0.1) ∃1 < p < 2, kT : Lp(A) → L2(A)k = 1 holds if and only if we have the following "spectral gap": kT xk2 kxk2 x∈A\{0},τ (x)=0 sup < 1. We provide two proofs of this result, where one is based on very elementary arguments with an additional assumption of 2-positivity and another, which is rather short, on [RX16]. In Theorem 1.9 we also show that the Lp-improving property (0.1) remains stable under the free products. This method permits us to give Lp-improving convolution operators for infinite quantum groups. In this paper we also include some simple properties of non-degenerate states on compact quantum groups with applications. We prove in Lemma 3.3 that the convolution Ces`aro limit of a non-degenerate state is the Haar state, which not only contributes to the proof of our main result, but also yields a generalization of [BFS12, Theorem 2.2] concerning the computation of idempotent states associated to Hopf images. We end this introduction with a brief description of the organization of the paper. Section 1 deals with the characterization of unital trace preserving Lp-improving operators on finite dimen- sional C-algebras and their free products. In Section 2 we present some preliminaries on compact quantum groups and the related Fourier analysis. Here we give a short and explicit calculation of Fourier series for compact quantum groups, parallel to the case of classical compact groups, which does not exist in other literature. In Section 3 we obtain some properties of non-degenerate states on a general compact quantum group. The last Section 4 is devoted to the positive convolution operators on finite quantum groups, and constructions of operators with similar properties on infinite compact quantum groups by free product. 1. Lp-improvement and spectral gaps 1.1. Basic notions. Let us firstly present some preliminaries and notations on noncommutative Lp-spaces and free products for later use. All the facts mentioned below are well-known. 1.1.1. Noncommutative Lp-spaces. Here we recall some basics of noncommutative Lp-spaces on finite von Neumann algebras. We refer to [Tak02] for the theory of von Neumann algebras and to [PX03] for more information on noncommutative Lp-spaces. Let M be a finite von Neumann algebra equipped with a normal faithful tracial state τ . Let 1 ≤ p < ∞. For each x ∈ M, we define kxkp = [τ (xp)]1/p . One can show that kkp is a norm on M. The completion of (M, kkp) is denoted by Lp(M, τ ) or simply by Lp(M). The elements of Lp(M) can be described by densely defined closed operators measurable with respect to (M, τ ), as in the commutative case. For convenience, we set L∞(M) = M equipped with the operator norm. Since τ (x) ≤ kxk1 for all x ∈ M, τ extends to a continuous functional on L1(M). Let 1 ≤ p, q, r ≤ ∞ be such that 1/p + 1/q = 1/r. If x ∈ Lp(M) and y ∈ Lq(M), then xy ∈ Lr(M) and the following Holder inequality holds: kxykr ≤ kxkpkykq. Lp-IMPROVING CONVOLUTION OPERATORS ON FINITE QUANTUM GROUPS 3 In particular, if r = 1, τ (xy) ≤ kxyk1 ≤ kxkpkykq for arbitrary x ∈ Lp(M) and y ∈ Lq(M). This defines a natural duality between Lp(M) and Lq(M): hx, yi = τ (xy). For any 1 ≤ p < ∞ we have Lp(M)∗ = Lq(M) isometrically. 1.1.2. Free products. We firstly recall some constructions of free product of C-algebras, for which we refer to [VDN92] and [NS06] for details. Consider a family of unital C-algebras (Ai, φi)i∈I with distinguished faithful states φi and associated GNS constructions (πi, Hi). Set Ai = ker φi and ai = ai − φi(ai)1 for each i and ai ∈ Ai. Construct a vector space (1.1) A = C1 ⊕Mn≥1(cid:16) Mi16=i26=···6=in Ai1 ⊗ Ai2 ⊗ · · · ⊗ Ain(cid:17). We equip A with an algebra structure such that 1 is the identity and the multiplication of a letter a ∈ Ai with an elementary tensor a1 ⊗ a2 ⊗ · · · ⊗ an in Ai1 ⊗ Ai2 ⊗ · · · ⊗ Ain is defined as a · (a1 ⊗ a2 ⊗ · · · ⊗ an) =  Moreover, we give an involution on A by a ⊗ a1 ⊗ a2 ⊗ · · · ⊗ an, (aa1) ⊗ a1 ⊗ a2 ⊗ · · · ⊗ an +φi1 (aa1)a2 ⊗ · · · ⊗ an, if i1 6= i, if i1 = i. (a1 ⊗ a2 ⊗ · · · ⊗ an)∗ = a∗ n ⊗ a∗ 2 ⊗ · · · ⊗ a∗ 1. In this sense A becomes a ∗-algebra, and each Ai can be viewed as a ∗-subalgebra in A by identifying Ai with C1 ⊕ Ai in the big direct sum. We call A the algebraic free product of (Ai)i∈I . It then can be shown that the algebra A admits a faithful ∗-representation (π, H, ξ) such that πAi = πi for each i ∈ I and φ(·) := hπ(·)ξ, ξi restricted on Ai coincides with φi. Moreover the state φ is faithful on A. Then the reduced C-algebraic free product of (Ai)i∈I is the C-algebra generated by π(A) in B(H), i.e., the norm closure of π(A) in B(H), denoted by ∗c0 i∈I Ai; and the state extends to ∗c0 i∈I Ai, called the free product state of (φi)i∈I and denoted by ∗i∈I φi. If moreover each Ai = Mi is a von Neumann algebra and each φi is normal, then the weak closure of π(A) in B(H), is defined to be the von Neumann algebraic free product of (Mi)i∈I , denoted by ¯∗i∈IMi, and the free product state φ = ∗i∈Iφi is also normal. Also, we remark that if each φi is a tracial state, then φ = ∗i∈I φi is also tracial. Let Ai and Bi be unital C-algebras with distinguished faithful states φi and ψi (i ∈ I) : Ai → Bi be a unital state preserving map for each i ∈ I. Set respectively, and let Ti (A, φ) = ∗i∈I (Ai, φi) and (B, ψ) = ∗i∈I (Bi, ψi). Then it is obvious that T (a1a2 · · · an) = Ti1(a1) · · · Tin(an) (ak ∈ Aik , ∀k, i1 6= i2 6= · · · 6= in) defines a unital state preserving map from the algebraic free products (A, φ) to (B, ψ). We denote by T = ∗i∈I Ti, and call it the free product map of the Ti. Similarly, we may define the c-free (conditionally free) product state in the sense of Bozejko, Leinert and Speicher [BLS96]. Let (Ai, φi) be as above and let ρi be further states respectively on Ai for each i. The conditional free product of (ρi)i is the functional ω := ∗(ψi)ρi on (A, φ) = ∗i∈I (Ai, φi) defined by the prescription ω(1) = 1 and ω(a1 · · · an) = ρi(1)(a1) · · · ρi(n)(an) for all n ≥ 1, i(1) 6= · · · 6= i(n) elements in I and aj ∈ ker φi(j) for j = 1, . . . , n. It is shown in [BLS96, Theorem 2.2] that the conditional free product of states is again a state. 1.2. Lp-improving operators. Let A be a finite dimensional C-algebra equipped with a faithful tracial state τ . The associated noncommutative Lp-spaces will be denoted by Lp(A). For a subset E ⊂ A, we denote by E+ the positive part of E. Recall that A can be identified with a direct sum of matrix algebras, that is, there exist some finite dimensional Hilbert spaces H1, . . . , Hm such that the following ∗-isomorphism holds A ≃ B(H1) ⊕ · · · ⊕ B(Hm). We will not distinguish the above two C-algebras in the sequel. For each i ∈ {1, . . . , m}, let ξi 1, . . . , ξi pq(v) = ni be an orthonormal basis for Hi, and define the operator ei pq ∈ B(Hi) by ei 4 SIMENG WANG qiHi ξi hv, ξi p for all v ∈ Hi and p, q ∈ {1, . . . , ni}. Take any x = x1 ⊕ · · · ⊕ xm ∈ A with xi ∈ B(Hi) for each i ∈ {1, . . . , m}, and let λi ni be the eigenvalues of xi ∈ B(Hi) (1 ≤ i ≤ m) ranged in non-increasing order and counted according to multiplicity. We can find a direct sum of unitaries u = u1 ⊕ u2 ⊕ · · · ⊕ um with ui ∈ B(Hi) for each i such that xi(uiξi k) for all kk) ∈ k) = λi kk. If we write βi [0, 1] for k ∈ {1, . . . , ni} and i ∈ {1, . . . , m}, then the Lp-norm of x for 1 ≤ p < ∞ is k(uiξi k = τ (ei 1, . . . , λi k=1 λi kei k ∈ {1, . . . , ni} and i ∈ {1, . . . , m}, that is, u∗xu = PiPni kk! = p = τ (u∗xpu) = τ m Xi=1 Xk=1 k)pei kxkp (1.2) (λi m Xi=1 Xk=1 (λi k)pβi k. n n We will prove in this section the result below. Theorem 1.1. Let A be a finite dimensional C-algebra equipped with a faithful tracial state τ , and T : A → A be a unital 2-positive trace preserving map on A. Then if and only if (1.3) ∃1 ≤ p < 2, kT xk2 ≤ kxkp, x ∈ A sup x∈A\{0},τ (x)=0 kT xk2 kxk2 < 1. Remark 1.2. Equivalently we can rewrite the above condition (1.3) as sup x∈A\{0},τ (x)=0 hT x, xi kxk2 2 < 1, which means exactly that the whole eigenspace of T for the eigenvalue 1 is just C1. In this sense we refer to the above inequality as a spectral gap phenomenon of T . Recall that the L2-norms assert some differential properties. The following lemma is elementary. Lemma 1.3. Let A be a C-algebra with a state ϕ and T : A → A be a positive map on A. Let O ⊂ Ah be an open set in the space Ah of all selfadjoint elements in A. The function f : O ∋ x 7→ ϕ((T x)2) is infinitely (Fr´echet) differentiable in O and for x ∈ O, f ′(x) = ϕ(T xT ·) + ϕ(T · T x), f ′′ ≡ 2ϕ(T · T ·), f (n) ≡ 0, n ≥ 3. In general a norm estimate can be reduced to the argument on positive cones. Lemma 1.4 ([RX16, Remark 9]). Let M be a von Neumann algebra and T : Lp(M) → Lq(M) be a bounded linear map for 1 ≤ p, q ≤ ∞. Assume that T is 2-positive in the sense that IdM2 ⊗ T maps the positive cone of Lp(M2 ⊗ M) to that of Lq(M2 ⊗ M). Then kT xkq ≤ kT (x)k1/2 q kT (x∗)k1/2 q , x ∈ Lp(M). Consequently, kT k = sup{kT xkq : x ∈ Lp(M)+, kxkp ≤ 1}. Now we give the proof of the theorem. Proof of the theorem. Assume firstly 1 ≤ p < 2 and kT xk2 ≤ kxkp for all x ∈ A. Note that kT xk2 = kT xk2. Observe that T ∗ is also a positive trace preserving map on A, and hence so is T . We choose an element x ∈ A such that τ (x) = 0 and T x = λx, with kT xk2 λ = sup kT xk2 kxk2 = x∈A\{0},τ (x)=0 x∈A\{0},τ (x)=0 sup . kxk2 Since T is a positive map on A and λ ∈ R, we may assume that x = x∗. For any self-adjoint element y ∈ A and 1 ≤ q < ∞, it is easy to compute that Note also that by assumption d2 dε2 k1 + εykq(cid:12)(cid:12)(cid:12)(cid:12)ε=0 = (q − 1)τ (y2) > 0. k1 + λεxk2 ≤ k1 + εxkp, ε > 0. Lp-IMPROVING CONVOLUTION OPERATORS ON FINITE QUANTUM GROUPS 5 Then taking the second derivative at ε = 0 we get λ2 ≤ (p − 1) < 1, as desired. Now we suppose (1.3) holds. Set A = {x ∈ A(cid:12)(cid:12) τ (x) = 0} and take σ = {x ∈ A+ (cid:12)(cid:12) τ (x) = 1} = (1 + A)+ which is exactly the set of positive elements in the unit sphere of L1(A). We first show that there exists 1 ≤ p < 2 and a neighborhood U of 1 such that (1.4) ∀x ∈ U ∩ σ, kT xk2 ≤ kxkp. To begin with, we consider F (x) = kT xk2 − kxk2, x ∈ A+. Using the previous lemma we see that F is infinitely differentiable at any x ∈ A+ \ {0} and F ′(x)(y) = kT xk−1 F ′′(x)(y1, y2) = −kT xk3 2 τ ((T x)(T y)) − kxk−1 2 τ (xy), y ∈ A 2τ ((T x)(T y1))τ ((T x)(T y2)) + kT xk−1 y1, y2 ∈ A. 2 τ (y1y2), 2 τ (xy1)τ (xy2) − kxk−1 + kxk−3 2 τ ((T y1)(T y2)) Since T is unital and preserves the trace, it follows that for y ∈ A, F ′(1)(y) = 0, F ′′(1)(y, y) = kT yk2 2 − kyk2 2. Then consider the second order Taylor expansion of F at 1. We can find a δ1 > 0 such that for all kyk2 ≤ δ1, y ∈ A, we have 1 + y ∈ A+ and F (1 + y) = F (1) + F ′(1)(y) + 1 2 F ′′(1)(y, y) + R1(y) = 1 2 (kT yk2 2 − kyk2 2) + R1(y), R1(y) = o(kyk2 2). Recall that by (1.3), kT yk2 2 − kyk2 c := sup{kT yk2 2 < 0 for y ∈ A. Thus by continuity, 2 : y ∈ A, kyk2 = 1} < 0. 2 − kyk2 Since the function y 7→ kT yk2 2 − kyk2 2 is 2-homogeneous, we get 2 ≤ ckyk2 2. 2 − kyk2 kT yk2 ∀y ∈ A, Take δ0 ∈ (0, δ1) such that ∀y ∈ A, kyk2 ≤ δ0, R1(y) kyk2 2 < c 4 . Then for y ∈ A, kyk2 ≤ δ0, (∗) F (1 + y) = 1 2 On the other hand, consider (kT yk2 2 − kyk2 2) + R1(y) ≤ c 4 kyk2 2. G(x) = kxk2 − kxkp, x = 1 + y, y = y∗ ∈ A, kyk2 < δ0. Let y = y∗ ∈ A with kyk2 < δ0, then by (1.2) we may take some K ∈ N and β1, . . . , βK ∈ [0, 1] such that the Lp-norm of x = 1 + y for 1 ≤ p < ∞ is exactly (∗∗) k1 + ykp = K Xi=1 p βi(1 + λi)p! 1 where (λi)i ⊂ R is the list of eigenvalues of y. So in order to estimate G, we consider the function g on RK defined as g(ξ) = K Xi=1 2 βi(1 + ξi)2! 1 A straightforward calculation gives − K Xi=1 p βi(1 + ξi)p! 1 , ξ = (ξ1, . . . , ξK) ∈ RK. ∂g ∂ξi (0) = 0, ∂2g ∂ξi∂ξj (0) = (p − 2)βiβj, ∂2g ∂ξ2 i (0) = (2 − p)(βi − β2 i ), 1 ≤ i 6= j ≤ K. If 2 − c 8 ≤ p ≤ 2 and 0 < δ < δ0 is such that R2(ξ) ≤ c i=1 βiξ2 i=1 βiξ2 i ≤ δ2, then for any ξ ∈ RK with PK Xi=1 g(ξ) ≤ (2 − p) 1 2 K i=1 βiξ2 i ≤ δ2, (βi − β2 i )ξ2 i + 1 2 (2 − p) β2 i ξ2 i + c 8 K Xi=1 8 PK i whenever PK Xi=1 Xi=1 c 4 βiξ2 i < K K βiξ2 i . 6 SIMENG WANG So by the Taylor formula g(ξ) = 1 2Xi (2 − p)(βi − β2 i )ξ2 i + 1 2Xj6=k (p − 2)βjβkξjξk + R2(ξ), R2(ξ) = o(kξk2). This, together with (∗∗), implies that, putting λ = (λ1, . . . , λK ), G(1 + y) = g(λ) ≤ Combined with (∗) we deduce c 4 K Xi=1 βiλ2 i = c 4 kyk2 2, kyk2 ≤ δ. kT xk2 − kxkp = F (1 + y) + G(1 + y) ≤ 0, x = 1 + y, y ∈ A, kyk2 ≤ δ, for all p ≥ 2 − c (1.4). 8 := p1. So U = {1 + y (cid:12)(cid:12) y = y∗ ∈ A, kyk2 < δ} is the desired neighborhood in Now we can derive the inequality for all x ∈ σ. For x ∈ σ \ U ⊂ (1 + A)+ \ {1}, we write x = 1 + y with y ∈ A\{0} and then by (1.3) and the trace preserving property we have kT xk2 2 = 1 + kT yk2 2. Note also that σ is compact, so we can find M < 1 such that kT xk2/kxk2 < M for all x ∈ σ\U . Given p < 2, let Cp be the optimal constant for the inequality kxk2 ≤ Cpkxkp for x ∈ A, then Cp → 1 when p → 2. Take p0 ≥ p1 such that Cp0 ≤ M −1. We get then 2 < 1 + kyk2 2 = kxk2 ∀x ∈ σ\U, ≤ M Cp ≤ 1, p0 ≤ p ≤ 2. kT xk2 kxkp As a result, for all p ∈ [p0, 2], it holds that kT xk2 ≤ kxkp, x ∈ σ. Since the norm is homogeneous and T is 2-positive, the above inequality holds for all x ∈ A as well. (cid:3) Apart from the above elementary proof, we would like to give an alternative simpler approach which yields a little bit stronger conclusion. The argument, however, depends heavily on the following recent and deep result on the convexity of Lp-spaces: Theorem 1.5 ([RX16, Theorem 1]). Let M be a von Neumann algebra equipped with a faithful semifinite normal trace φ. Let N be a von Neumann subalgebra such that the restriction of φ to N is semifinite. Denote by E the unique φ-preserving conditional expectation from M onto N . For 1 < p ≤ 2, we have kxk2 p ≥ kExk2 p + (p − 1)kx − Exk2 p, x ∈ Lp(M). For 2 < p < ∞, the inequality is reversed. Immediately we may deduce Theorem 1.1 as follows. Note that the result below is slightly stronger than the statement of Theorem 1.1. Theorem 1.6. Let A be a finite dimensional C-algebra equipped with a faithful tracial state τ , and T : A → A be a unital trace preserving map on A. Then (1.5) if and only if ∃1 < p < 2, ∀ x ∈ A, kT xk2 ≤ kxkp λ := sup x∈A\{0},τ (x)=0 kT xk2 kxk2 < 1. Lp-IMPROVING CONVOLUTION OPERATORS ON FINITE QUANTUM GROUPS 7 Moreover, if the above assertions are satisfied, then λ ≤ c−1 p pp − 1, where cp = sup x∈A\{0},τ (x)=0 kxk2 kxkp . Proof. The necessity has been already proved in the proof of Theorem 1.1. Now, assume λ < 1. Let x ∈ A and y = x − τ (x)1. Write a = τ (x). Since T is trace preserving, τ (T y) = τ (y) = 0. For p ≤ 2 we denote by cp the best constant with k · k2 ≤ cpk · kp. Then (p − 1)/c2 p → 1 when p → 2. Take p < 2 such that (p − 1)/c2 p > λ2, then we have kT xk2 2 = ka1 + T yk2 ≤ a2 + λ2c2 2 = a2 + kT yk2 pkyk2 2 ≤ a2 + λ2kyk2 2 p ≤ kxk2 p, p ≤ a2 + (p − 1)kyk2 whence (1.5). (cid:3) Remark 1.7. Let A be a finite dimensional C-algebra equipped with a faithful tracial state τ , and T : A → A be a unital trace preserving map on A. Consider the restriction of T on the subspace {x ∈ A : τ (x) = 0} of A and its adjoint, then we see that sup x∈A\{0},τ (x)=0 kT xk2 kxk2 = sup x∈A\{0},τ (x)=0 kT ∗xk2 kxk2 . Then the above theorem also implies that if there exists 1 < p < 2 such that then ∀ x ∈ A, kT xk2 ≤ kxkp, ∀ x ∈ A, kT ∗xk2 ≤ kxkp, and equivalently for 2 < q < ∞ with 1/p + 1/q = 1, ∀ x ∈ A, kT xkq ≤ kxk2. It is easy to see that the free product of unital trace preserving completely positive maps can be extended to the Lp-spaces on using the interpolation between L1 and L∞. But in general it is a delicate problem for the extension of algebraic free product of unital trace preserving maps onto the associated Lp-spaces. Here we provide a method to construct unital trace preserving Lp-improving operators on the free product of finite-dimensional C-algebras. To see this we need the following trivial claim. Claim 1.8. Let M be a finite von Neumann algebra equipped with a faithful tracial state τ . If the vectors e1, . . . , em ∈ M are orthonormal in L2(M, τ ) and denote c = max1≤k≤m kekk2 ∞, then for α1, . . . , αm ∈ C and 2 ≤ q ≤ ∞, m m k Xk=1 αkekkq ≤ (cm) 1 2 − 1 q k αkekk2. Xk=1 Proof. Note that k αkekk∞ ≤ c1/2 m Xk=1 m Xk=1 αk ≤ c1/2m1/2 m Xk=1 αk2!1/2 , which gives the claim for q = ∞. The inequality for 2 ≤ q ≤ ∞ then follows from the Holder inequality. (cid:3) Theorem 1.9. Let (Ai, τi), 1 ≤ i ≤ n be a finite family of finite dimensional C-algebras and set (A, τ ) = ¯∗1≤i≤n(Ai, τi) to be the von Neumann algebraic free product. For each 1 ≤ i ≤ n, Ti is a unital trace preserving map such that kTi : Lp(Ai) → L2(Ai)k = 1 for some 1 < p < 2. Then the (algebraic) free product map T = ∗1≤i≤nTi on ∗1≤i≤nAi extends to a map such that kT : Lp′(A) → L2(A)k = 1 8 SIMENG WANG for some 1 < p′ < 2. Proof. By the previous theorem and remark, (1.6) λ = max 1≤i≤n sup x∈ Ai kTixk2 kxk2 = max 1≤i≤n sup x∈ Ai kT ∗ i xk2 kxk2 < 1. Consider R = T ∗ and Ri = T ∗ i x ∈ ∗1≤i≤n(Ai, τi) in the algebraic free product and we will show that for all 1 ≤ i ≤ n, then R = R1 ∗ · · · ∗ Rn. By density, consider kRxkq ≤ kxk2 for some q > 2 independent of the choice of x. Now fix some r ≥ 1. For each i, choose a family (e(i) k=1 of eigenvectors of Ri which forms an orthonormal basis of Ai under τi, then k )ni k = e(i1) Er = {ei : 1 ≤ kj ≤ nj, 1 ≤ j ≤ r, i1 6= · · · 6= ir} forms an orthonormal basis of Ai1 ⊗· · ·⊗ Air which are also eigenvectors of R. Note that Er ≤ nrmr for m = maxj nj. ⊕i16=···6=ir Write additionally c = maxk,i ke(i) Ai1 ⊗ · · · ⊗ Air the above claim yields ∞. Then for any yr ∈ ⊕i16=···6=ir · · · e(ir) kr k k2 k1 (1.7) kyrkq ≤ (cnm)r( 1 2 − 1 q )kyrk2. Write x = τ (x)1 +Pr≥1 xr where xr ∈ ⊕i16=···6=ir according to (1.6) and the choice of Er. Together with Theorem 1.5 and (1.7), Ai1 ⊗ · · · ⊗ Air . Note that kRxrk2 ≤ λrkxrk2 Rxrk2 q ≤ τ (x)2 + (q − 1) Xr≥1 2 2 kRxrkq  kRxk2 q ≤ τ (x)2 + (q − 1)kXr≥1 ≤ τ (x)2 + (q − 1) Xr≥1 ≤ τ (x)2 + (q − 1) Xr≥1 1 2 − 1 (cnm)r( 1 2 − 1 (cnm)r( 1 2 − 1 q )kRxrk2  q )λrkxrk2  2 . Observe that (q − 1)(cnm) 2 < q < ∞ such that λ(cnm) q tends to 1 whenever q → 2 and that λ < 1, so we may choose 2 − 1 q ≤ (q − 1)−1. For such a q we then have 1 kRxk2 2 q ≤ τ (x)2 + (q − 1) (q − 1)−rkxrk2 Xr≥1  ≤ τ (x)2 + (q − 1)Xk≥1 (q − 1)−2kXr≥1 < τ (x)2 +Xr≥1 2 = kxk2 2. kxrk2 kxrk2 2 Take 1 < p′ < 2 such that 1/p′ + 1/q = 1. Then we get kT : Lp′(A) → L2(A)k = 1. (cid:3) 2. Preliminaries on quantum groups with Fourier analysis In this section we will do some preparations for discussing convolution operators in the quantum group framework. We will start with some preliminaries on compact quantum groups and then introduce the Fourier series in this setting. Lp-IMPROVING CONVOLUTION OPERATORS ON FINITE QUANTUM GROUPS 9 2.1. Compact quantum groups. In this short paragraph we recall some basic definitions and properties of compact quantum groups. All proofs of the facts mentioned below without references can be found in [Wor98] and [MVD98]. Definition 2.1. Consider a unital C-algebra A and a unital ∗-homomorphism ∆ : A → A ⊗ A called comultiplication on A such that (∆ ⊗ ι)∆ = (ι ⊗ ∆)∆ and {∆(a)(1 ⊗ b) : a, b ∈ A} and {∆(a)(b ⊗ 1) : a, b ∈ A} are linearly dense in A⊗A. Then (A, ∆) is called a compact quantum group. We denote G = (A, ∆) and A = C(G). We say that G is a finite quantum group if the space A = C(G) is finite dimensional. The following fact due to Woronowicz is fundamental in the quantum group theory. Proposition 2.2. Let G be a compact quantum group. There exists a unique state h on C(G) (called the Haar state of G) such that for all x ∈ C(G), (h ⊗ ι) ◦ ∆(x) = h(x)1 = (ι ⊗ h) ◦ ∆(x). Let G = (A, ∆) be a compact quantum group and consider an element u ∈ A ⊗ B(H) with i,j=1. Here, u is called an dim H = n. We identify A ⊗ B(H) = Mn(A), and write u = [uij]n n-dimensional representation of G if for all j, k = 1, ..., n, we have (2.1) ∆(ujk) = ujp ⊗ upk. n Xp=1 A representation u is said to be non-degenerate if u is invertible, unitary if u is unitary, and irreducible if the only matrices T ∈ Mn(C) with T u = uT are multiples of the identity matrix. Two representations u, v ∈ Mn(A) are called equivalent if there exists an invertible matrix T ∈ Mn(C) such that T u = vT . Denote by Irr(G) the set of unitary equivalence classes of irreducible unitary representations of G. For each α ∈ Irr(G), let uα ∈ C(G) ⊗ B(Hα) be a representative of the class α where Hα is the finite dimensional Hilbert space on which uα acts. With the notation above, the ∗-subalgebra A spanned by {uα i,j=1, α ∈ Irr(G)}, usually called the algebra of the polynomials on G, is dense in C(G) , and the Haar state h is faithful on this dense algebra. In the sequel we denote A = Pol(G). Consider the GNS representation (πh, Hh) of C(G), then Pol(G) can be viewed as a subalgebra of B(Hh). Define Cr(G) (resp., L∞(G)) to be the C-algebra (resp., the von Neumann algebra) generated by Pol(G) in B(Hh). Then h extends to a normal faithful state on L∞(G). ij : uα = [uα ij]nα It is known that there exists a linear antihomomorphism S on Pol(G) such that (2.2) determined by S(S(a)∗)∗ = a, a ∈ Pol(G), ij]nα S is called the antipode of G. For a, b ∈ Pol(G), we have ij) = (uα uα = [uα S(uα ji)∗, i,j=1, α ∈ Irr(G). (2.3) S((ι ⊗ h)(∆(b)(1 ⊗ a))) = (ι ⊗ h)((1 ⊗ b)∆(a)), S((h ⊗ ι)((b ⊗ 1)∆(a))) = (h ⊗ ι)(∆(b)(a ⊗ 1)). We will use the Sweedler notation for the comultiplication of an element a ∈ A, i.e. omit the summation and the index in the formula ∆(a) = Pi a(1),i ⊗ a(2),i and write simply ∆(a) = P a(1) ⊗ a(2). The Peter-Weyl theory for compact groups can be extended to the quantum case. In particular, it is known that for each α ∈ Irr(G) there exists a positive invertible operator Qα ∈ B(Hα) such that Tr(Qα) = Tr(Q−1 α ) := dα and (2.4) h(uα ij(uβ lm)∗) = δαβδil (Qα)mj dα , h((uα ij)∗uβ lm) = δαβδjm (Q−1 α )li dα where β ∈ Irr(G), 1 ≤ i, j ≤ dim Hα, 1 ≤ l, m ≤ dim Hβ. 10 SIMENG WANG The dual quantum group G of G is defined via its "algebra of functions" ℓ∞( G) = ⊕α∈Irr(G)B(Hα) where ⊕αB(Hα) refers to the direct sum of B(Hα), i.e. the bounded families (xα)α with each xα in B(Hα). We will not completely recall the quantum group structure on G as we do not need it in the following. We only remark that the (left) Haar weight h on G can be explicitly given by (see e.g. [VD96, Section 5]) h : ℓ∞( G) ∋ x 7→ Xα∈Irr(G) dα Tr(Qαpαx), where pα is the projection onto Hα and Tr denotes the usual trace on B(Hα) for each α. Our main result will only concentrate on the case where G is of Kac type, that is, its Haar state is tracial. Proposition 2.3 ([Wor98, Theorem 1.5]). Let G be a compact quantum group. The Haar state h on C(G) is tracial if and only if Qα = IdHα for all α ∈ Irr(G) in the formula (2.4) and if and only if the antipode S satisfies S2(x) = x for all x ∈ Pol(G). In particular, if the above conditions are satisfied and h is faithful on C(G), then S extends to a ∗-antihomomorphism on C(G) which is positive and bounded of norm one according to (2.2). Proposition 2.4 ([VD97]). If G is a finite quantum group, then the Haar state is tracial on C(G). For a compact quantum group G, we write L2(G) to be the Hilbert space associated to the GNS-construction with respect to the Haar state h. Similarly, denote by ℓ2( G) the Hilbert space given by the GNS-construction for ℓ∞( G) with respect to the Haar weight h. If G is of Kac type, for 1 ≤ p ≤ ∞, we denote additionally by Lp(G) the Lp-space associated to the pair (L∞(G), h), as defined in the previous subsection. Finally we turn to the dual free product of compact quantum groups. The following construction is given by [Wan95]. Proposition 2.5. Let G1 = (A, ∆A) and G2 = (B, ∆B) be two compact quantum groups with Haar states hA, hB respectively. There exists a unique comultiplication ∆ on A ∗c0 B such that the pair (A ∗c0 B, ∆) forms a compact quantum group, denoted by G = G1∗G2 and we have ∆A = (iA ⊗ iA) ◦ ∆A, ∆B = (iB ⊗ iB) ◦ ∆B, where iA and iB are the natural embedding of A and B into A ∗c0 B respectively. Moreover the Haar state on G is the free product state hA ∗ hB. 2.2. Fourier analysis. The Fourier transform for locally compact quantum groups has been defined in [Kah10], [Coo10] and [Cas13]. In the setting of compact quantum groups, we may give a more explicit description below. Let a compact quantum group G be fixed. For a linear functional ϕ on Pol(G), we define the Fourier transform ϕ = ( ϕ(α))α∈Irr(G) ∈ ⊕αB(Hα) by (2.5) ϕ(α) = (ϕ ⊗ ι)((uα)∗) ∈ B(Hα), α ∈ Irr(G). In particular, any x ∈ L∞(G) (or L2(G)) induces a continuous functional on L∞(G) defined by y 7→ h(yx), and the Fourier transform x = (x(α))α∈Irr(G) of x is given by x(α) = (h(·x) ⊗ ι)((uα)∗) ∈ B(Hα), α ∈ Irr(G). The above definition is slightly different from that of [Cas13] or [Kah10]. Indeed, we replace the unitary uα by (uα)∗ in the above formulas. This is just to be compatible with standard definitions in classical analysis on compact groups such as in [Fol95, Section 5.3], which will not cause essential difference. On the other hand, the notation ϕ has some slight conflict with the dual Haar weight h on G whereas one can distinguish them by the elements on which it acts, so we hope that this will not cause ambiguity for readers. Denote by F : x 7→ x the Fourier transform established above. It is easy to establish the Fourier inversion formula and the Plancherel theorem for L2(G). As we did not find them explicitly for compact quantum groups in the literature, we include the detailed calculation of the Fourier series in the following proposition. Lp-IMPROVING CONVOLUTION OPERATORS ON FINITE QUANTUM GROUPS 11 Proposition 2.6. (a) For all x ∈ L2(G), we have (2.6) x = Xα∈Irr(G) dα(ι ⊗ Tr)[(1 ⊗ x(α)Qα)uα], where the convergence of the series is in the L2-sense. For any α ∈ Irr(G), if we denote by Eα the orthogonal projection of L2(G) onto the subspace spanned by by the matrix coefficients (uα i,j=1, ij)nα and write Eαx =Pi,j xα ij uα ij with xα ij ∈ C, Xα = [xα ji]i,j, then x(α) = d−1 α XαQ−1 α . (b) F is a unitary from L2(G) onto ℓ2( G). Proof. (a) Denote by Eα the subspace spanned by the matrix units (uα i,j=1 for α ∈ Irr(G). Then Pol(G) is spanned by all the Eα, α ∈ Irr(G). It is easy to see from Holder's inequality that Pol(G) is k · k2-dense in L∞(G), and also recall that L2(G) is the k · k2-completion of L∞(G), so Pol(G) is k · k2-dense in L2(G) and L2(G) is a Hilbert direct sum of the orthogonal subspaces (Eα)α∈Irr(G). So each x ∈ L2(G) can be written as ij)nα (2.7) x = Xα∈Irr(G) Eαx = Xα∈Irr(G)Xi,j ij uα xα ij, (xα ij ∈ C) ij is the orthogonal projection of x onto Eα. where Eαx :=Pi,j xα Now for α ∈ Irr(G), write uα =Pl,m uα matrix units of B(Hα). Then ijuα lm⊗eα lm and Xα = [xα ji]i,j where eα lm denote the canonical (2.8) x(α) = (h(·x) ⊗ ι)((uα)∗) = (h(·Eαx) ⊗ ι)((uα)∗) + (h(·E ⊥ α x) ⊗ ι)((uα)∗) lm)∗ ⊗ eα ml) + 0 ij(cid:0)h(·uα ijh(cid:0)(uα ij ) ⊗ ι(cid:1) ((uα ij(cid:1) eα lm)∗uα ml xα xα = Xi,j,l,m = Xi,j,l,m α Xi,j,l = d−1 xα ij (Q−1 α )ileα jl = d−1 α XαQ−1 α . Hence dα(ι ⊗ Tr)[(1 ⊗ x(α)Qα)uα] = Xi,j,l,m = Xi,j,m (ι ⊗ Tr)[(1 ⊗ xα jieα ij)(uα lm ⊗ eα lm)] xα ji(ι ⊗ Tr)(uα jm ⊗ eα xα jiuα ji = Eαx. im) =Xi,j Combining the last equality with (2.7) proves the desired (2.6). (b) Let Eαx = Xα∈Irr(G)Xi,j x = Xα∈Irr(G) 2 = h((Eαx)∗(Eαx)) = Xi,j,l,m xα ij uα ij ∈ L2(G). ij xα xα lmh(cid:0)(uα ij )∗uα lm(cid:1) kEαxk2 = d−1 α Xi,j,l ij xα xα lj (Q−1 α )li For each α ∈ Irr(G), and also by (2.8) d2 αTr(Qα x(α)∗ x(α)) = Tr(X ∗ αXαQ−1 α ) =Xi,j,l ij xα xα lj (Q−1 α )li. 12 SIMENG WANG Hence by Parseval's identity, kxk2 kEαxk2 2 =Xα =Xα 2 =Xα d−1 α Xi,j,l ij xα xα lj (Q−1 α )li d−1 α d2 αTr(Qα x(α)∗ x(α)) = kxk2 2. Thus F maps isometrically L2(G) into ℓ2( G). From (2.6) and the isometric relation we see that the range of F contains the subset of all finitely supported families (aα) ∈ ⊕αB(Hα), which is dense in ℓ2( G). Therefore F is surjective and hence unitary. (cid:3) Example 2.7. (1) Let G be a compact group and define ∆(f )(s, t) = f (st), f ∈ C(G), s, t ∈ G. Then G = (C(G), ∆) is a compact quantum group. The elements in Irr(G) := Irr(G) coincide with the usual strongly continuous irreducible unitary representations of G. Any continuous functional ϕ on C(G) corresponds to a complex Radon measure µ on G by the Riesz representation theorem. By definition (2.5), the Fourier series of µ is given by µ(π) = (ϕ ⊗ ι)(π(·)∗) =ZG In particular for f ∈ L2(G), we have π(g)∗ dµ(g), π ∈ Irr(G). and we have the Fourier expansion and the Plancherel formula f (π) =ZG π(g)∗f (g)dg, π ∈ Irr(G) f = Xπ∈Irr(G) dπTr( f (π)π), kf k2 2 = Xπ∈Irr(G) dπk f (π)k2 HS where dπ is the dimension of the Hilbert space on which the representation π acts and kkHS denotes the usual Hilbert-Schmidt norm. We refer to [Fol95, Section 5.3] and [HR70, pp.77-87] for more information. (2) Let Γ be a discrete group with its neutral element e and C∗ r (Γ) be the associated reduced group C-algebra generated by λ(Γ) ⊂ B(ℓ2(Γ)), where λ denotes the left regular representation. The "dual" G = Γ of Γ is a compact quantum group such that C(G) is the group C-algebra C∗ r (Γ) equipped with the comultiplication ∆ : C∗ r (Γ) defined by r (Γ) → C∗ ∆(λ(γ)) = λ(γ) ⊗ λ(γ), r (Γ) ⊗ C∗ γ ∈ Γ. The Haar state of G is the unique trace τ on C∗ r (Γ) such that τ (1) = 1 and τ (λ(γ)) = 0 for γ ∈ Γ \ {e}. The elements of λ(Γ) give all irreducible unitary representations of G, which are all of dimension 1. It is easy to check from definition that for any f ∈ C∗ r (Γ), f (γ) = τ (f λ(γ)∗), γ ∈ Γ. And any f ∈ L2(G) has an expansion such that f (γ)λ(γ), f =Xγ∈Γ kf k2 2 =Xγ∈Γ f (γ)2. Let us end the section with a brief description of multipliers and convolutions in terms of Fourier series. Return back to a general compact quantum group G. For a = (aα)α ∈ ⊕αB(Hα), we define the left multiplier ma : Pol(G) → Pol(G) associated to a by (2.9) which means that (2.10) max = Xα∈Irr(G) dα(ι ⊗ Tr)[(1 ⊗ x(α)aαQα)uα], x ∈ Pol(G), (max)(α) = x(α)aα, α ∈ Irr(G). Lp-IMPROVING CONVOLUTION OPERATORS ON FINITE QUANTUM GROUPS 13 In the same way we may define the right multiplier m′ a : Pol(G) → Pol(G) such that m′ ax = Xα∈Irr(G) dα(ι ⊗ Tr)[(1 ⊗ aα x(α)Qα)uα], x ∈ Pol(G). Observe that the multiplier ma (or m′ only if a1 = 1. a resp.) is unital, i.e. ma(1) = 1 (m′ a(1) = 1 resp.) if and Remark 2.8. In case G is of Kac type, that is, Qα = IdHα for all α ∈ Irr(G) by Proposition 2.3, the multipliers ma and m′ a can be equivalently defined by (2.11) (ma ⊗ ι)(uα) = (1 ⊗ aα)(uα), (m′ a ⊗ ι)(uα) = (uα)(1 ⊗ aα), α ∈ Irr(G), which corresponds to the standard definition of left and right multipliers on locally compact quantum groups in [JNR09] and [Daw12]. If G is not of Kac type, the above formula (2.11) gives a similar equality corresponding to (2.10), that is, (max)(α) = x(α)QαaαQ−1 α , α ∈ Irr(G). We will use the standard definition of convolution products given by Woronowicz. Let x ∈ C(G) and ϕ, ϕ′ be linear functionals on C(G). We define ϕ ⋆ ϕ′ = (ϕ ⊗ ϕ′) ◦ ∆, x ⋆ ϕ = (ϕ ⊗ ι)∆(x), ϕ ⋆ x = (ι ⊗ ϕ)∆(x). Observing the embedding x 7→ h(· x) from C(G) into C(G)∗, the convolution products defined above are related as follows according to (2.3) (see also [VD07, Proposition 2.2]): on Pol(G) we have (2.12) We note that for α ∈ Irr(G) and uα = [uα ij]1≤i,j≤nα , h(· x) ⋆ ϕ = h(· [(ϕ ◦ S) ⋆ x]), ϕ ⋆ h(· x) = h(· [x ⋆ (ϕ ◦ S−1)]). Then by (2.1), a straightforward calculation shows that (cid:2)ϕ((uα ji)∗)(cid:3)i,j = (ϕ ⊗ ι)((uα)∗) = ϕ(α). (2.13) (ϕ ⋆ ϕ′)(α) = ϕ′(α) ϕ(α). Hence together with (2.12), (2.14) (x ⋆ ϕ)(α) = x(α)(ϕ ◦ S)(α), (ϕ ⋆ x)(α) = (ϕ ◦ S−1)(α)x(α). 3. Non-degenerate states and applications to Hopf images In this short section we give the key lemma on non-degenerate states, which will be of use for our main results. We need the following observation adapted from [Wor98, Lemma 2.1]. The result is mentioned in [So l05]. Lemma 3.1. Let G be a compact quantum group and A = C(G). Suppose that (ρi)i∈I is a family of states on A separating the points of A+, i.e., ∀x ∈ A+\{0}, ∃i ∈ I, ρi(x) > 0. If ρ is a state on A such that ∀i ∈ I, ρ ⋆ ρi = ρi ⋆ ρ = ρ, then ρ is the Haar state h of G. Proof. Set Then I is a closed left ideal of A ⊗ A. Define I = {q ∈ A ⊗ A : ∀i ∈ I, (ρi ⊗ ρ)(q∗q) = 0}. ΨL(x) = (ρ ⊗ ι)∆(x) − ρ(x)1, x ∈ A. Since ΨL is a difference of two unital completely positive maps, we see that ΨL is a completely bounded map with norm at most 2. We will prove that (ΨL ⊗ ι)∆(A) ⊂ I. 14 SIMENG WANG In fact, given x ∈ A, by the coassociativity of ∆ we have q := (ΨL ⊗ ι)∆(x) = (ρ ⊗ ι ⊗ ι)(ι ⊗ ∆)∆(x) − 1 ⊗ [(ρ ⊗ ι)∆(x)] = ∆((ρ ⊗ ι)∆(x)) − 1 ⊗ [(ρ ⊗ ι)∆(x)]. Thus q∗q = ∆ ([(ρ ⊗ ι)∆(x)]∗(ρ ⊗ ι)∆(x)) − ∆((ρ ⊗ ι)∆(x))∗[1 ⊗ (ρ ⊗ ι)∆(x)] − [1 ⊗ ((ρ ⊗ ι)∆(x))∗]∆((ρ ⊗ ι)∆(x)) + 1 ⊗ ([(ρ ⊗ ι)∆(x)]∗[(ρ ⊗ ι)∆(x)]) and hence for any i ∈ I we may write (ρi ⊗ ρ)(q∗q) = q1 − q2 − q3 + q4 where by the convolution invariance assumption and the coassociativity of ∆ we have q1 = (ρi ⊗ ρ)∆ ([(ρ ⊗ ι)∆(x)]∗(ρ ⊗ ι)∆(x)) = ρ ([(ρ ⊗ ι)∆(x)]∗(ρ ⊗ ι)∆(x)) , q2 = q∗ 3, q3 = (ρi ⊗ ρ)(cid:0)[1 ⊗ ((ρ ⊗ ι)∆(x))∗]∆((ρ ⊗ ι)∆(x))(cid:1) = ρ(cid:0)[((ρ ⊗ ι)∆(x))∗](ρi ⊗ ι)((ρ ⊗ ι ⊗ ι)(ι ⊗ ∆)∆(x))(cid:1) = ρ(cid:0)[((ρ ⊗ ι)∆(x))∗](ρ ⊗ ρi ⊗ ι)((∆ ⊗ ι)∆(x))(cid:1) = ρ ([(ρ ⊗ ι)∆(x)]∗(ρ ⊗ ι)∆(x)) , q4 = (ρi ⊗ ρ)(cid:0)1 ⊗ [((ρ ⊗ ι)∆(x))∗ (ρ ⊗ ι)∆(x)](cid:1) = ρ ([(ρ ⊗ ι)∆(x)]∗(ρ ⊗ ι)∆(x)) . Note that q1 = q2 = q3 = q4. So (ρi ⊗ ρ)(q∗q) = 0 and (ΨL ⊗ ι)∆(A) ⊂ I is proved. Now by the density of (1 ⊗ A)∆(A) in A ⊗ A and the complete boundedness of ΨL, it follows that ΨL(A) ⊗ 1 ⊂ (1 ⊗ A)(ΨL ⊗ ι)∆(A) is also contained in the closed left ideal I, which means that for any i ∈ I and x ∈ A, ρi(ΨL(x)∗ΨL(x)) = ρi ⊗ ρ(ΨL(x)∗ΨL(x) ⊗ 1) = 0. Recall that (ρi)i∈I separates the points of A+, so we have ΨL(x) = 0 and (ρ ⊗ ι)∆(x) = ρ(x)1 for all x ∈ A. A similar argument applies as well to the map ΨR(x) = (ι ⊗ ρ)∆(x) − ρ(x)1, x ∈ A. So ρ = h (cid:3) is the Haar state. Remark 3.2. We remark that in case G is a finite quantum group, we can provide a simpler proof of the above lemma. Indeed, since C(G) is finite-dimensional, its dual space C(G)∗ is also finite- dimensional, and we can take a maximal linear independent family {ρi1 , . . . , ρis } ⊂ (ρi)i∈I which form a basis of the subspace spanned by (ρi)i∈I in C(G)∗. Given a nonzero x ∈ A+, there is an k=1 akρik (ak ∈ C), then we see clearly that there exists k=1 ρik is at least one k ∈ {1, . . . , s} such that ρik (x) 6= 0 in order that ρi(x) > 0. Then ρ′ = 1 faithful on C(G) and ρ ⋆ ρ′ = ρ′ ⋆ ρ = ρ, thus ρ is the Haar state by [Wor98, Lemma 2.1]. i ∈ I such that ρi(x) > 0. Write ρi =Ps sPs We immediately obtain the following fact. Lemma 3.3. Let G be a compact quantum group and ϕ be a state on C(G). If ϕ is non-degenerate on C(G) in the sense that for all nonzero x ∈ C(G)+ there exists k ≥ 0 such that ϕ⋆k(x) > 0, then w∗- lim n→∞ 1 n ϕ⋆k = h. n Xk=1 If additionally h is faithful on C(G), then the converse also holds. Proof. If the above limit holds and h is faithful on C(G), then clearly ϕ is non-degenerate since if there existed a nonzero x ≥ 0 such that ϕ⋆n(x) = 0 for all n, then we would have k=1 ϕ⋆k(x) = 0, which contradicts the faithfulness of h. On the other hand, if ϕ is non- limn degenerate, the family of states { 1 k=1 ϕ⋆k : n ≥ 1} separates the points of C(G)+, so any accumulation point of this family becomes the unique Haar state by our previous lemma. (cid:3) nPn 1 nPn Lp-IMPROVING CONVOLUTION OPERATORS ON FINITE QUANTUM GROUPS 15 We would like to digress momentarily to see an application of above lemmas to some prob- lems concerning Hopf images. Let A be a unital C-algebra and π : C(G) → A be a unital ∗-homomorphism. The Hopf image of π, firstly introduced by Banica and Bichon in [BB10], is the largest algebra C(Gπ) for some compact quantum subgroup Gπ ⊂ G such that π factorizes through C(Gπ). In this paper we will use the following equivalent characterization recently given in [SS15]: for each k consider πk = (π ⊗ · · · ⊗ π ) ◦ ∆(k−1) : C(G) → A⊗k k {z } and let I = ∩∞ homomorphism πq : B → A such that k=1 ker πk. Then, there is a compact quantum group Gπ = (B, ∆π) with a ∗- (3.1) B = C(G)/I, ∆π ◦ q = (q ⊗ q) ◦ ∆, π = πq ◦ q where q : C(G) → C(G)/I denotes the quotient map. The algebra B = C(Gπ) is exactly the Hopf image of π. Now let hG, hGπ be the Haar states on G, Gπ respectively. A related question raised in [BFS12] is the computation of the associated idempotent state hGπ ◦ q on G. Simply based on Lemma 3.3, the following property generalizes the main result of [BFS12]. Theorem 3.4. Let G be a compact quantum group and A be a unital C-algebra with a unital ∗-homomorphism π : C(G) → A. Let Gπ be the compact quantum group constructed above. Then given any faithful state ϕ on A, hGπ ◦ q = w∗- lim n→∞ 1 n (ϕ ◦ π)⋆k. n Xk=1 Proof. Let ϕ be a faithful state on A and let I = ∩∞ k=1 ker πk with πk, k ≥ 1 constructed as above. Let us show that ϕ ◦ πq is non-degenerate on B = C(Gπ). Consider any x = q(y) with y ∈ C(G)+ satisfying ∀k ≥ 1, (ϕ ◦ πq)⋆k(x) = 0. Since, by (3.1), (ϕ ◦ πq)⋆k = [(ϕ ⊗ · · · ⊗ ϕ) ◦ (πq ⊗ ◦ ⊗ πq)]∆(k−1)(x) = [(ϕ ⊗ · · · ⊗ ϕ) ◦ ((πq ◦ q) ⊗ · · · ⊗ (πq ◦ q))]∆(k−1)(y) = (ϕ ⊗ · · · ⊗ ϕ)(πk(y)) and since ϕ is faithful, we get y ∈ ∩∞ is non-degenerate. Therefore we have hGπ = w∗- limn→∞ again, k=1 ker πk, which means that x = q(y) = 0. As a result ϕ ◦ πq k=1(ϕ ◦ πq)⋆k, and hence using (3.1) 1 nPn hGπ ◦ q = w∗- lim n→∞ 1 n as desired. n Xk=1 (ϕ ◦ πq)⋆k ◦ q = w∗- lim n→∞ (ϕ ◦ π)⋆k, 1 n n Xk=1 (cid:3) 4. Main results In this section we aim to give several characterizations of Lp-improving convolutions given by states on finite quantum groups, and also give the constructions for the free product of finite quantum groups. We will start with some discussions on multipliers on compact quantum groups. In this section we keep the notation of multipliers ma, m′ a and convolutions ϕ1 ⋆ϕ2 given in Section 2. Lemma 4.1. Let G be a compact quantum group of Kac type. Suppose a ∈ ℓ∞( G) such that a) extends to a unital left (resp., right) multiplier on L2(G) and b = aa∗. Then ma (resp. m′ b x = h(x)1 for all limn x ∈ L2(G) if and only if kaαk < 1 for all α ∈ Irr(G) \ {1}. b x = h(x)1 for all x ∈ L2(G) if and only if limn k=1 m′k k=1 mk 1 nPn 1 nPn 16 SIMENG WANG Proof. Without loss of generality we only discuss the left multiplier ma. Assume that kaαk < 1 for all α ∈ Irr(G) \ {1}. By the Plancherel theorem 2.6 and the formula (2.9) we note that mb extends to a bounded map of norm one on L2(G). We first consider the case x ∈ Pol(G), so that x is finitely supported. Let α ∈ Irr(G) \ {1} and kaαk < 1. Then n n n mk b x)(α)k2 = k x(α)(aαa∗ α)kk2 ≤ kaαk2kkx(α)k2 → 0 k( 1 n Xk=1 1 n Xk=1 whenever n → ∞. And for α = 1, 1 n Xk=1 Thus by the Plancherel theorem ( 1 n n Xk=1 mk b x)(1) = x(1) = h(x). k n 1 n Xk=1 nPn nPn mk b x − h(x)1k2 2 = Xα6=1 dαk( 1 n n Xk=1 mk b x)(α)k2 2 → 0 when n → ∞. Since 1 1 the convergence limn k=1 mk k=1 mk b is a contraction on L2(G) and Pol(G) is dense in L2(G), we get b x = h(x)1 for all x ∈ L2(G). Conversely, if ∃α0 ∈ Irr(G)\{1}, kaα0 k = 1, then viewing bα0 as a matrix in Mnα0 , we observe that 1 ∈ σ(bα0 ) and there exists a nonzero xα0 ∈ Mnα0 such that xα0 bα0 = xα0 . Take x ∈ L2(G) such that x(1) = 1, x(α0) = xα0 , x(α) = 0 for α ∈ Irr(G)\{1, α0}. Then mbx = x and hence 1 (cid:3) b x = x does not converge to h(x)1. k=1 mk Remark 4.2. In case the compact quantum group G is not of Kac type, the above argument still remains true for right multipliers. nPn The following first main result is now in reach. We will consider the case where G is a finite quantum group. Theorem 4.3. Let G be a finite quantum group. Suppose a ∈ ℓ∞( G) is such that ma (resp., m′ a) is a unital left (resp., right) multiplier on C(G). Then the following assertions are equivalent: (1) there exists 1 ≤ p < 2 such that, (2) there exists 1 ≤ p < 2 such that, ∀ x ∈ C(G), kmaxk2 ≤ kxkp ; ∀ x ∈ C(G), km′ axk2 ≤ kxkp ; (3) kaαk < 1 for all α ∈ Irr(G) \ {1} ; (4) limn (5) limn b x = h(x)1 for all x ∈ C(G) when b = aa∗; b x = h(x)1 for all x ∈ C(G) when b = aa∗. k=1 mk k=1 m′k 1 1 nPn nPn Proof. Without loss of generality we only discuss the left multiplier ma and prove the equivalence (1)⇔(3)⇔(4). It is easy to see from Plancherel's theorem that (3) is just (1.3) for T = ma. In fact note that for x ∈ C(G), h(x) = 0 if and only if x(1) = 0, so (3) implies (1.3) via Plancherel's theorem. On the other hand, suppose by contradiction that there exists α ∈ Irr(G) \ {1} such that kaαk = 1. By Proposition 2.6 we may take a nonzero x ∈ C(G) such that x(1) = 0, x(α) = 0 when kaαk < 1, and kx(α)aαk2 = kx(α)k2 when kaαk = 1. Then kmaxk2 = kxk2 with h(x) = 0. As a consequence the equivalence (1)⇔(3) follows from Theorem 1.6. The equivalence between (3) and (4) was proved in the previous lemma. Therefore the theorem (cid:3) is established. Now we turn to the corresponding convolution problems. Let ϕ ∈ C(G)∗ for a compact quantum group G. Recall the formula (2.13), and then we have (4.1) (cid:2)ϕ⋆n((uα ji)∗)(cid:3) = (ϕ⋆n)(α) = ϕ(α)n. n 1 n lim This is to say, n Xk=1 mk n nPn 1 Xk=1 ϕ∗(x)! (α) = lim mk n n ϕ(1) = 1, lim n 1 n n Xk=1 ϕ∗(x)! (1) = h(x) ϕ(1)∗ = h(x)1, ( ϕ(α)∗)k x(α) = 0, α ∈ Irr(G)\{1}. 1 n n Xk=1 ϕ(α)k = 0, α ∈ Irr(G)\{1}, Lp-IMPROVING CONVOLUTION OPERATORS ON FINITE QUANTUM GROUPS 17 Note that the convergence 1 k=1 mk lated in terms of Fourier coefficients as ϕ∗ (x) → h(x)1 for all x ∈ Pol(G), by (2.10) can be reformu- which, according to (4.1), is equivalent to 1 1 ≤ i, j ≤ nα, or in other words, nPn k=1 ϕ⋆k(uα ij) → h(uα ij ) for all α ∈ Irr(G) and ϕ⋆k(x) → h(x), n → ∞, x ∈ Pol(G). 1 n n Xk=1 Any state ϕ on C(G) induces two convolution operators on C(G) Tϕ : C(G) ∋ x 7→ x ⋆ ϕ = (ϕ ⊗ ι)∆(x), T ′ ϕ : C(G) ∋ x 7→ ϕ ⋆ x = (ι ⊗ ϕ)∆(x). If additionally G is of Kac type and the Haar state is faithful on C(G), then by Proposition 2.3 the antipode S extends to a positive linear operator on C(G) and S = S−1, and hence ϕ ◦ S = ϕ ◦ S−1 is also a state. In this case we have h ϕ(uα ji)ii,j = ϕ(α), (ϕ ◦ S)(α) =(cid:2) ϕ(uα ij )(cid:3)i,j = ϕ(α)∗ and by (2.14) we have (x ⋆ ϕ)(α) = x(α) ϕ(α)∗ and (ϕ ⋆ x)(α) = ϕ(α)∗ x(α) for α ∈ Irr(G), x ∈ C(G). So Tϕ = m ϕ∗ and T ′ ϕ∗ are unital completely positive left and right multipliers, respectively. ϕ = m′ Now with these remarks and Lemma 3.3 in hand, we may reformulate Theorem 4.3 in terms of convolution operators using the above arguments. Theorem 4.4. Let G be a finite quantum group and ϕ be a state on C(G). Denote ψ = (ϕ◦ S)⋆ ϕ. The following assertions are equivalent: (1) there exists 1 ≤ p < 2 such that, (2) there exists 1 ≤ p < 2 such that, ∀ x ∈ C(G), kϕ ⋆ xk2 ≤ kxkp ; ∀ x ∈ C(G), kx ⋆ ϕk2 ≤ kxkp ; (3) k ϕ(α)k < 1 for all α ∈ Irr(G) \ {1} ; (4) limn (5) For any nonzero x ∈ C(G)+, there exists n ≥ 1 such that ψ⋆n(x) > 0. k=1 ψ⋆k = h; 1 nPn Note that if C(G) is commutative, i.e. C(G) = C(G) where G is a finite group, then ϕ ∈ C(G)∗ corresponds to a Radon measure µ via the Riesz representation theorem. The above condition (5) in the theorem just asserts that G is the union of Dn := supp (ν⋆n), n ≥ 1, where ν denotes the Radon measure corresponding to ψ. It is easy to see that D1 = (cid:8)i−1j : i, j ∈ supp (µ)(cid:9) and 1 . So the above corollary covers Ritter's result [Rit84]. Dn = Dn Corollary 4.5. Let G be a finite group and µ be a probability measure on G. Then there is a 1 ≤ p < 2 such that if and only if G is equal to the subgroup generated by (cid:8)i−1j : i, j ∈ supp (µ)(cid:9). kx ⋆ µk2 ≤ kxkp, x ∈ Lp(G) 18 SIMENG WANG On the other hand, let Γ be a finite group with neutral element e and C∗(Γ) be the associated group C-algebra generated by λ(Γ) ⊂ B(ℓ2(Γ)), where λ denotes the left regular representation. Recall that if G = Γ, then C(G) is the group C*-algebra C∗(Γ) equipped with the comultiplication ∆ : C∗(Γ) → C∗(Γ) ⊗ C∗(Γ) defined by ∆(λ(γ)) = λ(γ) ⊗ λ(γ), γ ∈ Γ. Note that any state Φ on C(G) corresponds to a positive definite function ϕ on Γ with ϕ(e) = 1 via the relation Φ(λ(γ)) = ϕ(γ) for all γ ∈ Γ. Therefore we have Φ ⋆ λ(γ) = λ(γ) ⋆ Φ = (Φ ⊗ ι)∆(λ(γ)) = (Φ ⊗ ι)(λ(γ) ⊗ λ(γ)) = ϕ(γ)λ(γ), so the convolution operators associated to Φ are just the Fourier-Schur multiplier on Γ associated to ϕ. Our preceding argument in particular yields the following result extending [Rit84, Theorem 2(a)]. Corollary 4.6. Let Γ be a finite group and ϕ be a positive definite function on Γ with ϕ(e) = 1. Let Mϕ be the associated Fourier-Schur multiplier operator determined by Mϕ(λ(γ)) = ϕ(γ)λ(γ) for all γ ∈ Γ. Then there exists 1 ≤ p < 2 such that if and only if ϕ(γ) < 1 for any γ ∈ Γ \ {e}. kMϕxk2 ≤ kxkp, x ∈ C∗(Γ) Remark 4.7. We remark that the finite-dimensional condition cannot be removed in any of our results, including Theorem 1.1, Theorem 4.3, Corollary 4.4-4.6. Here we give a counterexample illustrating this. Let T be the unit circle in the complex plane, then T gives an infinite compact quantum group. Define an operator T : C(T) → C(T) by T (f ) = (1 − λ)τ (f ) + λf, f ∈ C(T), where 0 < λ < 1 and τ denote the usual integral against the normalized Lebesgue measure on T. Then T is obviously unital completely positive since so are τ and the identity map. It is a left multiplier satisfying T (f )(0) = f (0) and T (f )(n) = λ f (n) for n 6= 0. Also we may view T as a convolution operator associated to the state f 7→ (1 − λ)τ (f ) + λf (1) on C(T), which is faithful since τ is faithful. Note that T admits the spectral gap inequality (1.3) as well, and in fact, kT f k2 = λkf k2 < kf k2 for any f ∈ C(T) with τ (f ) = 0. However, there doesn't exist any p < 2 such that kT f k2 ≤ kf kp for all f ∈ C(T). Indeed if such a p existed, then for any f ∈ C(T), we would have kf k2 2 ≥ kf k2 p ≥ kT f k2 2 = τ (f )2 + λ2kf − τ (f )k2 2 which yields an impossible equivalence between the norms k · k2 and k · kp. ≥ λ2(τ (f )2 + kf − τ (f )k2 2) = λ2kf k2 2, In spite of the above general remark, Theorem 1.9 still gives constructions of Lp-improving positive convolution operators on infinite compact quantum groups. Let G1, . . . , Gn be finite quantum groups with Haar states h1, . . . , hn respectively and let each ϕi be a state on C(Gi), i ∈ {1, . . . , n}. Denote G = G1∗ · · · ∗ Gn with the Haar state h and consider the convolution operators Ti : x 7→ x ⋆ ϕi, x ∈ C(Gi). Note that the free product map T = ∗1≤i≤nTi on C(G) is just the convolution operator given by the c-free product state ϕ = ∗(h1,...,hn)ϕi, i.e., (4.2) T (x) = (ϕ ⊗ ι)∆(x), x ∈ C(G). In fact, we note that if h(a) = 0, then h(a(1)) = h(a(2)) = 0 by (2.1), where ∆(a) :=P a(1) ⊗ a(2) denotes the Sweedler notation. Now for a reduced word x = x1 · · · xm with xk ∈ C(Gik ) such that h(xk) = 0, i1 6= · · · 6= in, ik ∈ {1, . . . , n} for each k = 1, . . . , m,, we have T (x) = Ti1(x1) · · · Tim(xm) =X ϕi1 (x1 =X ϕ(x(1))x(2) = (ϕ ⊗ ι)∆(x) (1))x1 (2) · · ·X ϕim (xm (1))xm (2) Lp-IMPROVING CONVOLUTION OPERATORS ON FINITE QUANTUM GROUPS 19 where we have used the fact that the comultiplication ∆ is an homomorphism. Then the equality (4.2) follows from a standard density argument. Now taking in Theorem 1.9 each Ti to be a convolution operator on a finite quantum group, we get the following corollary: Corollary 4.8. Let G1, . . . , Gn be finite quantum groups and let each ϕi be a state on C(Gi), i ∈ {1, . . . , n}. Denote G = G1∗ · · · ∗ Gn and ϕ = ∗(h1,...,hn)ϕi. If each ϕi satisfies any one of the conditions (1)-(5) in Corollary 4.4, then the free product convolution operator given by T : x 7→ x ⋆ ϕ, x ∈ C(G) is a unital left multiplier on G satisfies for a certain 1 < p < 2. kT : Lp(G) → L2(G)k = 1 Example 4.9. Now we give a simple method to create nontrivial Lp-improving positive convo- lutions (i.e. the associated state is different from the Haar state) on finite and infinite compact quantum groups. Let G be a finite quantum group and h the Haar state on it. Given any state ϕ on C(G) and any 0 < λ < 1, we can define a state ρ on C(G) by ρ = λϕ + (1 − λ)h. This is a faithful state which is in particular non-degenerate, and hence by Theorem 4.4 the convolution operator Tρ : x 7→ x ⋆ ρ, x ∈ C(G) satisfies kTρ : Lp(G) → L2(G)k = 1 for a certain 1 < p < 2 according to Theorem 4.4. Moreover by Corollary 4.8, the convolution operator Tρ′ : x 7→ x ⋆ ρ′, x ∈ C(G∗G) given by the c-free product state ρ′ = ρ ∗(h,h) ρ satisfies for some 1 < p′ < 2. kTρ′ : Lp′ (G∗G) → L2(G∗G)k = 1 Acknowledgment. The author is indebted to his advisors Quanhua Xu and Adam Skalski for their helpful discussions and constant encouragement, and to Professor Gilles Pisier for his careful reading and pointing out a mistake in the preprint version. The author also thanks the referee for a careful reading of the manuscript and useful suggestions. This research was partially supported by the NCN (National Centre of Science), grant no. 2014/14/E/ST1/00525. References [BB10] T. Banica and J. Bichon. Hopf images and inner faithful representations. Glasg. Math. J., 52(3):677 -- 703, 2010. [BFS12] T. Banica, U. Franz, and A. Skalski. Idempotent states and the inner linearity property. Bull. Pol. Acad. Sci. Math., 60(2):123 -- 132, 2012. [BLS96] M. Bozejko, M. Leinert, and R. Speicher. Convolution and limit theorems for conditionally free random variables. Pacific J. Math., 175(2):357 -- 388, 1996. [Cas13] M. Caspers. The Lp-Fourier transform on locally compact quantum groups. J. Operator Theory, 69(1):161 -- 193, 2013. [Coo10] T. Cooney. A Hausdorff-Young inequality for locally compact quantum groups. Internat. J. Math., 21(12):1619 -- 1632, 2010. [Daw12] M. Daws. Completely positive multipliers of quantum groups. Internat. J. Math., 23(12):1250132, 23, [Fol95] [HR70] 2012. G. B. Folland. A course in abstract harmonic analysis. Studies in Advanced Mathematics. CRC Press, Boca Raton, FL, 1995. E. Hewitt and K. A. Ross. Abstract harmonic analysis. Vol. II: Structure and analysis for compact groups. Analysis on locally compact Abelian groups. Die Grundlehren der mathematischen Wissenschaften, Band 152. Springer-Verlag, New York-Berlin, 1970. [JNR09] M. Junge, M. Neufang, and Z.-J. Ruan. A representation theorem for locally compact quantum groups. Internat. J. Math., 20(3):377 -- 400, 2009. [Kah10] B.-J. Kahng. Fourier transform on locally compact quantum groups. J. Operator Theory, 64(1):69 -- 87, 2010. [MVD98] A. Maes and A. Van Daele. Notes on compact quantum groups. Nieuw Arch. Wisk. (4), 16(1-2):73 -- 112, [NS06] 1998. A. Nica and R. Speicher. Lectures on the combinatorics of free probability, volume 335 of London Math- ematical Society Lecture Note Series. Cambridge University Press, Cambridge, 2006. 20 SIMENG WANG [Obe82] D. M. Oberlin. A convolution property of the Cantor-Lebesgue measure. Colloq. Math., 47(1):113 -- 117, 1982. [PX03] G. Pisier and Q. Xu. Non-commutative Lp-spaces. In Handbook of the geometry of Banach spaces, Vol. [RX16] 2, pages 1459 -- 1517. North-Holland, Amsterdam, 2003. ´E. Ricard and Q. Xu. A noncommutative martingale convexity inequality. Ann. Probab., 44(2):867-882 (2016). [Rit84] D. L. Ritter. A convolution theorem for probability measures on finite groups. Illinois J. Math., 28(3):472 -- [SS15] 479, 1984. A. Skalski and P. M. So ltan. Quantum families of invertible maps and related problems. Canad. J. Math., 68(3):698-720, 2016. P. M. So ltan. Quantum Bohr compactification. Illinois J. Math., 49(4):1245 -- 1270, 2005. [So l05] [Tak02] M. Takesaki. Theory of operator algebras. I, volume 124 of Encyclopaedia of Mathematical Sci- ences. Springer-Verlag, Berlin, 2002. Reprint of the first (1979) edition, Operator Algebras and Non- commutative Geometry, 5. [VD96] A. Van Daele. Discrete quantum groups. J. Algebra, 180(2):431 -- 444, 1996. [VD97] A. Van Daele. The Haar measure on finite quantum groups. Proc. Amer. Math. Soc., 125(12):3489 -- 3500, 1997. [VD07] A. Van Daele. The Fourier transform in quantum group theory. In New techniques in Hopf algebras and graded ring theory, pages 187 -- 196. K. Vlaam. Acad. Belgie Wet. Kunsten (KVAB), Brussels, 2007. [VDN92] D.-V. Voiculescu, K. J. Dykema, and A. Nica. Free random variables, volume 1 of CRM Monograph Series. American Mathematical Society, Providence, RI, 1992. A noncommutative probability approach to free products with applications to random matrices, operator algebras and harmonic analysis on free groups. [Wan95] S. Wang. Free products of compact quantum groups. Comm. Math. Phys., 167(3):671 -- 692, 1995. [Wor98] S. L. Woronowicz. Compact quantum groups. In Sym´etries quantiques (Les Houches, 1995), pages 845 -- 884. North-Holland, Amsterdam, 1998. Laboratoire de Math´ematiques, Universit´e de Franche-Comt´e, 25030 Besanc¸on Cedex, France and Institute of Mathematics, Polish Academy of Sciences, ul. ´Sniadeckich 8, 00-956 Warszawa, Poland E-mail address: [email protected]
1506.06528	1	1506	2015-06-22T09:49:53	Stability of additive functional equation on discrete quantum semigroups	[ "math.OA", "math.FA" ]	We show that noncommutative analog of additive functional equation has Hyers-Ulam stability on amenable discrete quantum (semi)groups. This generalizes an old classical result.	math.OA	math	STABILITY OF ADDITIVE FUNCTIONAL EQUATION ON DISCRETE QUANTUM SEMIGROUPS MAYSAM MAYSAMI SADR Abstract. We show that noncommutative analog of additive functional equa- tion has Hyers-Ulam stability on amenable discrete quantum (semi)groups. This generalizes an old classical result. 1. Introduction Let G be a (semi)group. Consider the additive functional equation (AFE), F (xy) = F (x) + F (y), for functions F from G to the complex field C. AFE is said to have Hyers-Ulam stability (HUS) on G if the following property holds. Given r > 0, there is r′ > 0 such that if a function f on G satisfies f (xy) − f (x) − f (y) < r′ then there exists a function F on G satisfying F (xy) = F (x) + F (y) and F (x) − f (x) < r. The study of the above property goes back to a famous question of Ulam [9] for characterization of pairs (G, H), where H is a metric group, satisfying the above property with C replaced by H. In 1941, Hyers [5] showed that if G is the underlying additive group of a Banach space then AFE has HUS on G. Four decades later, Forti [4] extended the result of Hyers for amenable semigroups by a very simple method. Since the appearance of [5], the Ulam stability problem and its generalizations not only for AFE but also other types of functional equations has been considered and developed by many mathematicians. (See [6] for the history of developments.) Nowadays, this area of mathematics is generally named Hyers-Ulam stability. The main goal of this note that we wish it would be the first one of a series of papers is an invitation to stability theory of functional equations on noncom- mutative spaces. We start our program to study this subject by considering the same traditional problem of Ulam for quantum groups instead of ordinary groups. Indeed, we extend the above mentioned result of Forti [4] as follows. (For exact definitions of discrete quantum semigroups and amenability see Section 3.) Let G be a discrete quantum semigroup with comultiplication ∆. Denote by F(G) the function algebra on G and by Fb(G) the von Neumann subalgebra of bounded 2010 Mathematics Subject Classification. Primary 81R15, Secondary 81R60, 39B42, 34D99. Key words and phrases. Discrete quantum semigroup, Additive functional equation, Hyers- Ulam stability, Noncommutative geometry. 1 2 M. M. SADR functions on G. The 'sup-norm' of Fb(G) is denoted by k·k. The noncommutative analog of AFE becomes ∆(F ) = 1 ⊗ F + F ⊗ 1, for functions F in F(G). Similar the above mentioned stability property we make Definition 1. We say that noncommutative AFE has HUS on G if the following condition holds. For every r > 0 there is r′ > 0 such that if a function f ∈ F(G) satisfies the inequality k∆(f ) − 1 ⊗ f − f ⊗ 1k < r′, then there exists a function F ∈ F(G) for which ∆(F ) = 1 ⊗ F + F ⊗ 1 and kF − f k < r. The main result of this note is the following theorem that may be considered as an extension of [4, Theorem 7] to discrete quantum semigroups. Theorem 2. If G is a left or right amenable discrete quantum semigroup then noncommutative AFE has HUS on G. Our proof of Theorem 2 that will be given in Section 4 is the same proof of Forti [4, Theorem 7] but translated to the dual language of Hopf-algebras. As it would be clear for the reader by taking a quick look at Forti's proof, for this dualization we need to work with unbounded 'functions' of two variables which are bounded by fixing one of the variables. Moreover, we must have a machinery to apply bounded operators to spaces of such functions. In Section 2 following some ideas from [3] we introduce somewhat a new way for tensoring linear maps which is specially designated to overcome the mentioned difficulties. In Section 3 we consider definition of discrete quantum semigroups and amenability. Our definition is the same one of Van Daele [10] for discrete quantum groups but with weaker conditions which result semigroups. Also these can be considered as Hopf-von Neumann algebras of discrete type [2] expect that their coproducts need not to be injective. We end this section with three remarks. Remark 3. (i) Suppose that F ∈ F(G) as above satisfies in noncommutative AFE. Since F as a function takes values in finite dimensional matrix algebras we may consider E = exp(F ) as a member of F(G). Then it is straightforward to check that ∆(E) = E ⊗ E. Such elements in the language of Hopf-algebras are called group-like. (ii) One can consider Definition 2 for any locally compact quantum group G [7] where f and F are unbounded affiliated operators to the underlying von Neumann algebra of G. But the proof of Theorem 2 does not longer work in this more general case. (iii) Some works on stability of noncommutative analog of quadratic, Jensen and n-difference functional equations on quantum groups and Kac algebras are in process. ADDITIVE EQUATION ON QUANTUM SEMIGROUPS 3 2. A type of tensor products Throughout ι denotes the identity map and the C-algebra of n × n matrices is denoted by Mn. By an index system we mean a set I together with a positive integer valued function nI on I. If γ ∈ I then for simplicity we write Mγ In the following I, I ′, J, J ′ denote index systems. We denote by for MnI (γ). F(I) the -algebra of all functions f : I → ∪γ∈IMγ for which f (γ) ∈ Mγ, with pointwise operations. (In [3] F(I) is called multimatrix algebra.) The - subalgebra of all functions with finite support is denoted by Ff (I). So in the standard notation F(I) = Qγ∈I Mγ and Ff(I) = Lγ∈I Mγ. It is also simply verified that F(I) is identified with the multiplier algebra of Ff (I). We denote the unit of F(I) by 1 and hence 1(γ) = 1γ is the identity matrix in Mγ. Fb(I) is the - subalgebra of F(I) containing bounded functions i.e. those functions f for which kf k = supα∈I kf (α)k < ∞. This is a C-algebra with the sup-norm and is the dual space of absolutely sumable functions. So Fb(I) is a von Neumann algebra. Let Ii be an index system for i = 1, . . . , k. We consider the cartesian product set I1 × · · · × Ik as an index system with nI1×···×Ik(α1, · · · , αk) = n1(α1) · · · nk(αk). Let A = {i1, · · · , il} be a subset of {1, . . . , k}. Then we let Fb:i1···il(I1 × · · · × Ik) be the subspace of those functions f in F(I1 × · · · × Ik) such that for every fixed family {αi ∈ Ii}i∈{1,...,k}\A, the condition supαi∈Ii,i∈A kf (α1, · · · , αk)k < ∞ holds. Suppose that T is a linear map from F(I) (resp. Fb(I)) to F(I ′). We define a linear map T ⊗ι from F(I × J) (resp. Fb:1(I × J)) to F(I ′ × J) as follows. For β ∈ J let ιβ denote the identity linear map on Mβ. Let f be in F(I × J) (resp. Fb:1(I × J)). Since Mβ is finite dimensional the function α 7→ f (α, β) determines a unique member of F(I) ⊗ Mβ (resp. Fb(I) ⊗ Mβ). So (T ⊗ ιβ)(α 7→ f (α, β)) is in F(I ′) ⊗ Mβ. Considering this latter space as a space of functions from I ′ to ∪α′∈I ′Mα′ ⊗ Mβ we let [(T ⊗ι)(f )](α′, β) = [(T ⊗ ιβ)(α 7→ f (α, β))](α′). We may also write a more explicit formula for T ⊗ι as follows. Let {eij β }1≤i,j≤nJ (β) be the standard vector basis for Mβ. For f as above let the elements f ij β of F(I) (resp. Fb(I)) be such that f (α, β) =Pij f ij (1) β (α) ⊗ eij β . Then [(T ⊗ι)(f )](α′, β) =Xij [T (f ij β )](α′) ⊗ eij β . We remark that if T is a linear functional then the image of T ⊗ι canonically belongs to F(J). We may define similarly linear maps ι ⊗T and ι ⊗T ⊗ι. So, the latter is a map from F(J × I × J ′) (resp. Fb:2(J × I × J ′)) to F(J × I ′ × J ′). In below we list some properties of ⊗ which are used in next sections. (P0) If T is a linear map from F(I) (resp. Fb(I)) to F(I ′) then (T ⊗ι)(f ⊗ g) = T (f ) ⊗ g, where f is in F(I) (resp. Fb(I)) and g ∈ F(J), and f ⊗ g denotes the function (α, β) 7→ f (α) ⊗ g(β). Another trivial property of ⊗ is associativity: (P1) (ι ⊗T ) ⊗ι = ι ⊗T ⊗ι = ι ⊗(T ⊗ι) and (T ⊗ι) ⊗ι = T ⊗ι ⊗ι = T ⊗(ι ⊗ι). From (1) it follows easily that: 4 M. M. SADR (P2) If T is a linear map from F(I) or Fb(I) to Fb(I ′) then the image of T ⊗ι is contained in Fb:1(I ′ × J). The analogous statements are satisfied for ι ⊗T and ι ⊗T ⊗ι. For α ∈ I and β ∈ J let Pβ : F(J) → Mβ and Iα : Mα → F(I) denote canonical linear projection and imbedding respectively. For every linear map T : F(I) → F(J) we let T α β = PβT Iα. Now, suppose that T is a -homomorphism from F(I) to F(J). Since the kernel of PβT is a two-sided ideal with finite codimension in F(I), and since matrix algebras have no nontrivial two-sided It follows that ideals, there is a finite subset I0 of I with PβT F(I\I0) = 0. for every fixed β ∈ J there are only finitely many α in I with T α β 6= 0 and β (f (α)). Analogous statements are completely satisfied when [T (f )](β) = Pα T α the domain of T is the subalgebra Fb(I). (P3) If T is a -homomorphism from F(I) or Fb(I) to F(J) then [(T ⊗ι)(f )](β, β′) =Xα (T α β ⊗ ιβ ′)(f (α, β′)) (β′ ∈ J ′). The analogous statements are satisfied for ι ⊗T and ι ⊗T ⊗ι. (P4) Let T : F(I) → F(J) and T ′ : F(I ′) → F(J ′) be linear maps such that either T or T ′ is -homomorphism. Then (ι ⊗T ′)(T ⊗ι) = (T ⊗ι)(ι ⊗T ′) as linear maps from F(I × I ′) to F(J × J ′). α ⊗ f ij Proof. We suppose that T is -homomorphism. The other case is similar. Let f α ∈ F(I ′) (1 ≤ i, j ≤ nI(α)) be such that f (α, α′) = be in F(I × I ′) and let f ij α (α′). This α )](β′). On the α (α′). By (P3), [(T ⊗ι)(f )](β, α′) = PαPij T α Pij eij implies that [(ι ⊗T ′)(T ⊗ι)(f )](β, β′) = PαPij T α other hand [(ι ⊗T ′)(f )](α, β′) =Pij eij β ⊗ ιβ ′)(Xij [(T ⊗ι)(ι ⊗T ′)(f )](β, β′) =Xα α ) ⊗ [T ′(f ij α )](β′) and hence α ⊗ [T ′(f ij α ⊗ [T ′(f ij eij α )](β′)) α ) ⊗ f ij β (eij β (eij T α β (eij α ) ⊗ [T ′(f ij α )](β′). (T α =Xα Xij (cid:3) (P5) Suppose that T : F(I) → F(J) is a -homomorphism. If f belongs to Fb:2(I × J ′) then (T ⊗ι)(f ) ∈ Fb:2(J × J ′). Proof. Let f ∈ Fb:2(I × J ′). So for every α ∈ I we have supβ ′∈J ′ kf (α, β′)k < ∞. β ⊗ ιβ ′)(f (α, β′)). So (cid:3) Let β ∈ J be fixed. By (P3), [(T ⊗ι)(f )](β, β′) = Pα(T α supβ ′∈J ′ k[(T ⊗ι)(f )](β, β′)k ≤Pα∈I,T α β 6=0(supβ ′∈J ′ kf (α, β′)k) < ∞. 3. Discrete quantum semigroups Let I be an index system. A comultiplication for I is a collection of - β,γ : Mα → Mβ ⊗ Mγ for each ordered triple (α, β, γ) of homomorphisms ∆α elements of I, which satisfies the two conditions below. ADDITIVE EQUATION ON QUANTUM SEMIGROUPS 5 (i) ∆α β,γ(1α)∆α′ β,γ(1α′) = 0 for α 6= α′, and α,β ⊗ ι)∆λ (ii) the -homomorphismsPω(∆ω to Mα ⊗ Mβ ⊗ Mγ are equal. ω,γ andPω(ι ⊗ ∆ω β,γ)∆λ α,ω from Mλ Note that (i) implies that for fixed β and γ there are only finitely many α with ∆α β,γ(1α) 6= 0. Also (i) guarantees that the -homomorphisms in (ii) are well defined. Now, we may, and hence do, define a -homomorphism ∆ by [∆(f )](β, γ) =Xα ∆α β,γf (α), from F(I) to F(I × I). Then (ii) may be restated as (∆ ⊗ι)∆ = (ι ⊗∆)∆. Note that every ∆α β,γ may be recovered from ∆. So, from now on we do not distinguish between ∆ and the collection {∆α β,γ}. Definition 4. A discrete quantum semigroup is a pair G = (I, ∆) such that I is an index system and ∆ is a comultiplication for I. For a discrete quantum semigroup G = (I, ∆) we denote the algebras Fb(I) and F(I), respectively, by Fb(G) and F(G). Analogously, we let F(G × G) = F(I × I) and Fb(G × G) = Fb(I × I). The comultiplication ∆ of G transforms bounded functions to bounded functions i.e. ∆(Fb(G)) ⊆ Fb(G × G). It follows from the fact that the map f 7→ [∆(f )](β, γ) from Fb(G) to Mβ ⊗ Mγ is a -homomorphism between C-algebras and hence norm decreasing. Let G = (I, ∆) be a discrete quantum semigroup. Then G is called right amenable [1] if there is a state m on Fb(G), called right invariant mean, which satisfies (m ⊗ι)∆(f ) = m(f )1 for every f ∈ Fb(G). Left invariant means and left amenable discrete quantum semigroups are defined similarly. Example 5. Let G be a discrete semigroup. Then G gives rise to a discrete quan- tum semigroup G = (I, ∆) in which I = G and nI = 1. The *-homomorphisms ∆α β,γ : C → C ⊗ C = C are defined by ∆α β,γ =(ι α = βγ 0 otherwise In this case, G is right (resp. left) amenable iff G is right (resp. left) amenable as usual. Also, it is not so hard to see that every discrete quantum semigroup G = (I, ∆) for which nI = 1, is constructed from a discrete semigroup, as above. Discrete quantum groups [10] which are Pontryagin dual of compact quantum groups [8] (or Hopf-von Neumann algebras of discrete type [2]) are also discrete quantum semigroups in our sense. We will need the next lemmas in Section 4. Lemma 6. Let G = (I, ∆) be a discrete quantum semigroup and m be a right invariant mean for G. Then for every f ∈ Fb:1(G × G) the following holds. (m ⊗ι ⊗ι)(∆ ⊗ι)(f ) = 1 ⊗ [(m ⊗ι)(f )]. Proof. First of all, note that by (P2) the right hand side is well-defined. Let f be in Fb:1(G × G) and let f ij γ (ω) ⊗ eij γ . γ ∈ Fb(G) be such that f (ω, γ) = Pij f ij [(m ⊗ι ⊗ι)(∆ ⊗ι)(f )](β, γ) = [((m ⊗ι) ⊗ι)(∆ ⊗ι)(f )](β, γ) γ )](α, β) ⊗ eij γ and hence [(m ⊗ι)∆(f ij γ )](β) ⊗ eij γ 6 M. M. SADR Then we get [(∆ ⊗ι)(f )](α, β, γ) =Pij[∆(f ij =Xij =Xij =Xij m(f ij γ )1β ⊗ eij γ 1β ⊗ m(f ij γ )eij γ = 1β ⊗ [(m ⊗ι)(f )](γ) = (1 ⊗ [(m ⊗ι)(f )])(β, γ). Lemma 7. Let G = (I, ∆) be a discrete quantum semigroup, n be a linear func- tional on Fb(G), and f ∈ Fb:1(G × G). Then ∆(n ⊗ι)(f ) = (n ⊗ι ⊗ι)(ι ⊗∆)(f ). Proof. First of all, note that by (P5) the right hand side is well-defined. Let ¯n be an arbitrary linear extension of n to F(G). Then (cid:3) (n ⊗ι ⊗ι)(ι ⊗∆)(f ) = (¯n ⊗ι ⊗ι)(ι ⊗∆)(f ) = ∆(¯n ⊗ι)(f ) = ∆(n ⊗ι)(f ), where we have used (P4) to pass from the first equality to the second one. (cid:3) 4. The proof of Theorem 2 Suppose that G is right amenable. The proof of the other case is similar. Let m be a right invariant mean for G and let r > 0 be given. We show that the conditions of Definition 1 are satisfied for r′ = r. Let f ∈ F(G) be such that k∆(f ) − 1 ⊗ f − f ⊗ 1k < r, that is supβ,γ k[∆(f )](β, γ) − f (β) ⊗ 1γ − 1β ⊗ f (γ)k < r. It follows that sup k[∆(f )](β, γ) − f (β) ⊗ 1γk < r + k1β ⊗ f (γ)k = r + kf (γ)k. β So (∆(f ) − f ⊗ 1) ∈ Fb:1(G × G). We define a function F in F(G) by F = (m ⊗ι)(∆f − f ⊗ 1). By (P0) and (P1) we get (2) F ⊗ 1 = [(m ⊗ι)(∆(f ) − f ⊗ 1)] ⊗ ι(1) = (m ⊗ι ⊗ι)(∆(f ) ⊗ 1 − f ⊗ 1 ⊗ 1). It follows from Lemma 6 that 1 ⊗ F = (m ⊗ι ⊗ι)(∆ ⊗ι)(∆(f ) − f ⊗ 1) and hence by (P2) we get (3) 1 ⊗ F = (m ⊗ι ⊗ι)((∆ ⊗ι)∆(f ) − ∆(f ) ⊗ 1). ADDITIVE EQUATION ON QUANTUM SEMIGROUPS 7 It follows from (2) and (3) that F ⊗ 1 + 1 ⊗ F = (m ⊗ι ⊗ι)((∆ ⊗ι)∆(f ) − f ⊗ 1 ⊗ 1) = (m ⊗ι ⊗ι)((ι ⊗∆)∆(f ) − f ⊗ 1 ⊗ 1) = (m ⊗ι ⊗ι)(ι ⊗∆)(∆(f ) − f ⊗ 1), where we have used (P5) to pass from the second row to the third one. By Lemma 7 the third row is equal to ∆(m ⊗ι)(∆f − f ⊗ 1) = ∆(F ). So we shaw that ∆(F ) = 1 ⊗ F + F ⊗ 1. For the norm inequality we have kF − f k = k(m ⊗ι)(∆(f ) − f ⊗ 1) − f k = k(m ⊗ι)(∆(f ) − f ⊗ 1) − (m ⊗ι)(1 ⊗ f )k = k(m ⊗ι)(∆(f ) − f ⊗ 1 − 1 ⊗ f )k ≤ k(m ⊗ι)kk∆(f ) − f ⊗ 1 − 1 ⊗ f k < r. This completes the proof. References 1. E. B`edos, G.J. Murphy, L. Tuset, Amenability and co-amenability of algebraic quantum groups, Int. J. Math. Math. Sci. 31 (2002) 577-601. 2. M. Enock, J.-M. Schwartz, Kac algebras and duality of locally compact groups, Springer- Verlag, Berlin -- Heidelberg -- New York, 1992. 3. E.G. Effros, Z.-J. Ruan, Discrete Quantum Groups I, The Haar Measure, Int. J. Math. 5 (1994) 681 -- 723. 4. G.L. Forti, The stability of homomorphisms and amenability, with applications to functional equations, Abh. Math. Sem. Univ. Hamburg 57 (1987) 215-226. 5. D.H. Hyers, On the stability of the linear functional equation, Proc. Nat. Acad. Sci. U.S.A. 27 (1941) 222-224. 6. S.-M. Jung, Hyers-Ulam-Rassias stability of functional equations in nonlinear analysis, Springer, 2011. 7. J. Kustermans, S. Vaes, Locally compact quantum groups, Ann. Sci. ´Ecole Norm. Sup. (4) 33 (2000) 837-934. 8. A. Maes, A. Van Daele, Notes on compact quantum groups, Nieuw Arch. Wisk. 16 (1998) 73-112. 9. S.M. Ulam, A collection of the mathematical problems, Interscience Publ., New York, 1960. 10. A. Van Daele, Discrete quantum groups, J. Algebra 180 (1996) 431 -- 444. Department of Mathematics, Institute for Advanced Studies in Basic Sciences, P.O.Box 45195-1159, Zanjan 45137-66731, Iran E-mail address: [email protected]
1712.08824	2	1712	2018-01-19T17:30:18	$L^p$-operator algebras associated with oriented graphs	[ "math.OA" ]	For each $1\le p<\infty$ and each countable oriented graph $Q$ we introduce an $L^p$-operator algebra $\mathcal{O}^p(Q)$ which contains the Leavitt path $\mathbb{C}$-algebra $L_Q$ as a dense subalgebra and is universal for those $L^p$-representations of $L_Q$ which are spatial in the sense of N.C. Phillips. For $\mathcal{R}_n$ the graph with one vertex and $n$ loops ($2\le n\le \infty$), $\mathcal{O}^p(\mathcal{R}_n)=\mathcal{O}^p_n$, the $L^p$-Cuntz algebra introduced by Phillips. If $p\notin\{1,2\}$ and $\mathcal{S}(Q)$ is the inverse semigroup generated by $Q$, $\mathcal{O}^p(Q)=F_{\operatorname{tight}}^p(\mathcal{S}(Q))$ is the tight semigroup $L^p$-operator algebra introduced by Gardella and Lupini. We prove that $\mathcal{O}^p(Q)$ is simple as an $L^p$-operator algebra if and only if $L_Q$ is simple, and that in this case it is isometrically isomorphic to the closure $\overline{\rho(L_Q)}$ of the image of any nonzero spatial $L^p$-representation $\rho:L_Q\to\mathscr{L}(L^p(X))$. We also show that if $L_Q$ is purely infinite simple and $p\ne p'$, then there is no nonzero continuous homomorphism $\mathcal{O}^p(Q)\to\mathcal{O}^{p'}(Q)$. Our results generalize those obtained by Phillips for $L^p$-Cuntz algebras.	math.OA	math	Lp OPERATOR ALGEBRAS ASSOCIATED WITH ORIENTED GRAPHS GUILLERMO CORTI NAS AND MA. EUGENIA RODRIGUEZ Abstract. For each 1 ≤ p < ∞ and each countable oriented graph Q we introduce an Lp-operator algebra Op(Q) which contains the Leavitt path C-algebra LQ as a dense subal- gebra and is universal for those Lp-representations of LQ which are spatial in the sense of N.C. Phillips. For Rn the graph with one vertex and n loops (2 ≤ n ≤ ∞), Op(Rn) = Op n , the Lp-Cuntz algebra introduced by Phillips. If p < {1, 2} and S(Q) is the inverse semigroup generated by Q, Op(Q) = F tight(S(Q)) is the tight semigroup Lp-operator algebra intro- duced by Gardella and Lupini. We prove that Op(Q) is simple as an Lp-operator algebra if and only if LQ is simple, and that in this case it is isometrically isomorphic to the closure ρ(LQ) of the image of any nonzero spatial Lp-representation ρ : LQ → L(Lp(X)). We also show that if LQ is purely infinite simple and p , p′, then there is no nonzero continuous homomorphism Op(Q) → Op′ (Q). Our results generalize some similar results obtained by Phillips for Lp-Cuntz algebras. p 1. Introduction Let Q be a countable oriented graph, let Q0 and Q1 be the sets of vertices and edges, and let LQ be the Leavitt path C-algebra. For 1 ≤ p < ∞ we call a representation ρ : LQ → L(Lp(X)) spatial if X is a σ-finite measure space and ρ maps the elements of Q0⊔Q1⊔(Q1)∗ to partial isometries which are spatial in the sense of [12, Definition 6.4]. Each spatial representation ρ induces a seminorm on LQ via aρ = ρ(a); the supremum k k of these seminorms is a norm (Proposition 4.23) and we write Op(Q) for the completion of (LQ, k k). For p < {1, 2}, Op(Q) agrees with the tight semigroup algebra introduced by Gardella and Lupini in [8] (Proposition 7.12). We prove the following. Theorem 1.1. (Simplicity theorem) Let Q be a countable graph and let 1 ≤ p < ∞, p , 2. The following are equivalent. i) LQ is simple. ii) Every nonzero spatial Lp-representation of LQ is injective. iii) Every nondegenerate contractive nonzero Lp-representation of Op(Q) is injective. If furthermore we have either that Q0 is finite or that p > 1, then the above conditions are also equivalent to: iv) For every Lp-operator algebra B, every contractive, nonzero homomorphism Op(Q) → B is injective. Condition iv) says that Op(Q) is simple as an Lp-operator algebra. Since every Lp- operator algebra is isometrically embedded in L(Lp(X)) for some σ-finite measure space X, simplicity is equivalent to the condition that every contractive nonzero representation ρ : Op(Q) → L(Lp(X)), degenerate or not, be injective. We show (using a classical result of Ando [5] and a recent result of Gardella and Thiel [10]) that if either Q0 is finite or p > 1, Both authors were supported by grants UBACyT 20021030100481BA and PICT 2013-0454. Cortinas re- search was supported by Conicet and by grant MTM2015-65764-C3-1-P (Feder funds). 1 2 GUILLERMO CORTI NAS AND MA. EUGENIA RODRIGUEZ then the restriction of ρ to LQ factors through a nondegenerate spatial representation; this allows us to prove that iii) ⇐⇒ iv). To prove Theorem 1.1 we first show the following uniqueness theorem. Theorem 1.2. (Uniqueness theorem) Let Q be a countable graph such that LQ is simple. Let 1 ≤ p < ∞, X a σ-finite measure space and ρ : LQ → L(Lp(X)) a nonzero spatial representation. Then the canonical map Op(Q) → ρ(LQ) is an isometric isomorphism. Specializing Theorem 1.2 to the case when Q has only one vertex recovers N.C. Phillips' uniqueness result for Lp-analogues of Cuntz algebras [12, Theorem 8.7]. We also show (Theorem 11.2) that if Q and Q′ are countable graphs with LQ purely infinite simple and 1 ≤ p , p′ < ∞ then it is often the case that no nonzero continuous homomorphism Op(Q) → Op′ (Q′) exists. For example this is the case when LQ′ is simple. In particular, we have the following. Theorem 1.3. Let Q be a countable graph. If LQ is purely infinite simple then there is no nonzero continuous homomorphism Op(Q) → Op′ (Q). A similar result for Lp-Cuntz algebras was obtained by N.C. Phillips in [12, Theorem 9.2]. The rest of this paper is organized as follows. In Section 2 we recall some defini- tions and basic facts on Leavitt path algebras and prove some elementary technical lem- mas. In Section 3 we show (Lemma 3.3) that LQ is the universal algebra for tight alge- braic representations of the inverse semigroup S(Q) generated by Q. Spatial represen- tations of the Leavitt path algebra LQ of a countable graph Q are introduced in Section 4. We give examples of such representations and show in Proposition 4.23 that for every countable Q and 1 ≤ p < ∞ there is an injective, nondegenerate spatial representation LQ → L(ℓ p(N)). Spatial representations of matrix algebras MnLQ for 1 ≤ n ≤ ∞ are considered in Section 5 and it is shown that they are the same as spatial representations of the Leavitt path algebra over the graph MnQ (Remark 5.1) and that any such repre- sentation is equivalent to the matricial amplification Mnρ of a spatial representation ρ of LQ (Lemma 5.4). Section 6 is concerned with a characterization of spatiality of repre- sentations in terms of norm estimates. We prove a spatiality criterion which we shall presently explain. The subalgebra (LQ)0,1 = span{v ∈ Q0, ee∗, e ∈ Q1} ⊂ LQ is a direct sum of -- possibly infinite dimensional -- matrix algebras and is thus naturally equipped with a canonically equipped with an Lp-operator norm. The spatiality criterion, Theorem 6.4 -- which generalizes [12, Theorem 7.7] -- says that if p ∈ [1, ∞), p , 2, then a nondegener- ate representation ρ : LQ → L(Lp(X)) is spatial if and only if its restriction to (LQ)0,1 is contractive and kρ(x)k ≤ 1 for every x ∈ Q1`(Q1)∗ (cf. [12, Theorem 7.7]). Along the way we also prove a spatiality criterion for nondegenerate Lp-representations of matricial algebras (Proposition 6.3) which generalizes [12, Theorem 7.2]. Both spatiality criteria fail to be true if the nondegeneracy hypothesis is dropped (see Remark 6.5). In Section 7 we define Lp-operator algebras and introduce the Lp-operator algebra Op(Q). By defini- tion, any spatial representation of LQ → L(Lp(X)) factors uniquely through a contractive representation Op(Q) → L(Lp(X)) (1 ≤ p < ∞). Moreover we prove, using the spatial- ity criterion of Section 6, that for p , 2, any nondegenerate contractive representation Op(Q) → L(Lp(X)) induces a nondegenerate spatial representation LQ → L(Lp(X)) (The- orem 7.8). Using a result of E. Gardella and H. Thiel from [10], we show that if moreover p , 1, then the nondegeneracy hypothesis may be dropped (Theorem 7.9). We also show, using the material of Section 3, that if p < {1, 2} then Op(Q) is the same as the Lp-algebra Lp OPERATOR ALGEBRAS ASSOCIATED WITH ORIENTED GRAPHS 3 F p tight(S(Q)) introduced by E. Gardella and M. Lupini in [8], which is universal for tight Lp-spatial representations of S(Q). In the next section we show that adding heads and tails to a graph Q to obtain a new graph Q′ without sources, sinks or infinite emitters results in an isometric inclusion Op(Q) → Op(Q′) (Corollary 8.8). Section 9 is devoted to the proof of Theorem 1.2 (Theorem 9.1). The technical result of the previous section is used here to reduce the proof to the case of graphs without sources, sinks or infinite emmitters. After this reduction, the strategy of proof is similar to that of [12, Theorem 8.7], although it requires several nontrivial technical adjustments. Simplicity Theorem 1.1 is proved in Section 10. In fact we prove in Theorem 10.1 that the simplicity of LQ is equivalent not only to the conditions of Theorem 1.1, but also to other more restrictive conditions, e.g. that every nondegenerate spatial nonzero representation LQ → L(ℓ p(N)) be injective. The last section of this article is Section 11, where we prove Theorem 11.2, of which Theorem 1.3 is a particular case. Notation 1.4. In this paper N = Z≥1 and N0 = Z≥0. All algebras, vector spaces, and tensor products are over C. All identities pertaining measure spaces are to be interpreted up to sets of measure zero. For example we say that a family {Xn}n≥1 of measurable sets in a measurable space X = (X, B, µ) is disjoint if Xn ∩ Xm has measure zero for all n , m, and write `n Xn for their union. In case the latter agrees with X up to measure zero, we write X = `n Xn. This reflects the fact that under the above hypothesis (X, B, µ) is equivalent to set theoretic coproduct`n Xn equipped with the σ-algebra generated by `n Bn and the measure induced by the sequence of measures {µXn }. We write L0(X) for the vector space of classes of measurable functions X → C. Acknowledgements. This article has evolved from the PhD thesis of the second named author [15]. We are indebted to Chris Phillips for discussions on his paper [12]. Thanks also to our colleague Daniel Carando for several useful discussions and references on Lp- spaces. The first named author also wishes to thank Eusebio Gardella for an enlightening email exchange including several useful comments on a previous version of this paper. 2. Graphs and Leavitt path algebras An oriented graph or quiver Q = (Q0, Q1, r, s) consists of sets Q0 and Q1 of vertices and edges, and range and source functions r, s : Q1 → Q0 . We say that Q is finite or countable if Q0 and Q1 are both finite or countable. A vertex v ∈ Q0 is an infinite emitter if s−1(v) is infinite, and is a sink if s−1(v) = ∅. A vertex is singular if it is either a sink or an infinite emitter. We write sing(Q) = sink(Q) ∪ inf(Q) ⊂ Q0 for the set of singular vertices and reg(Q) = Q0 \ sing(Q). We call Q singular if sing(Q) , ∅ and nonsingular (or regular) otherwise. We call Q row-finite if it has no infinite emitters. A vertex v is a source if r−1(v) = ∅; we write sour(Q) ⊂ Q0 for the set of sources. Since all our graphs will be oriented, we shall use the term graph to mean oriented graph. A path α is a (finite or infinite) sequence of edges α = e1 . . . ei . . . such that r(ei) = s(ei+1) (i ≥ 1). For such α, we write s(α) = s(e1); if α is finite of length l, we put α = l and r(α) = r(el). Vertices are considered as paths of length 0. A finite path α is closed if s(α) = r(α). A closed path α = α1 . . . αn is a cycle if in addition s(ei) , s(e j) if i , j. Let P = P(Q) be the set of finite paths, and let Pn be the set of paths of length n. Thus, (2.1) P = an∈N0 Pn. We consider the following preorder in P: 4 GUILLERMO CORTI NAS AND MA. EUGENIA RODRIGUEZ (2.2) α ≤ β ⇐⇒ ∃ γ such that r(β) = s(γ) and α = βγ. Observe that (2.2) also makes sense when α is an infinite path. Definition 2.3. Let Q be a graph. The Leavitt path algebra LQ is the quotient of the free C-algebra on Q0 ∪ Q1 ∪ (Q1)∗, modulo the following relations: • vv′ = δv,v′v ∀ v, v′ ∈ Q0, • s(e)e = er(e) = e ∀ e ∈ Q1, • r(e)e∗ = e∗ s(e) = e∗ ∀ e ∈ Q1, • (CK1) e∗e′ = δe,e′ r(e) ∀ e, e′ ∈ Q1, • (CK2) v = X{e∈Q1:s(e)=v} ee∗, if v ∈ reg(Q). The Leavitt path algebra is a ∗-algebra with involution determined by v 7→ v, e 7→ e∗. It has a Z-grading where vertices have degree zero, edges have degree 1, and e∗ = −1 for e ∈ Q1 ([1, Corollary 2.1.5]). We write (2.4) (LQ)n = span{αβ∗ : α − β = n} for the n-th homogeneous component with respect to this grading. The elementary lemmas below shall be used later in the article. Lemma 2.5. Let Q be a nonsingular graph and a1, . . . , am ∈ LQ. Then there exist n ∈ N, a finite set F ⊂ P, and finitely supported functions λi : F × Pn → C, (α, β) 7→ λi (i = 1, . . . m, α ∈ F, β ∈ Pn), such that α,β ai = Xα∈F Xβ∈Pn α,βαβ∗, ∀i = 1, . . . , m. λi Proof. For each i = 1, . . . , m, we may write ai = Pni j}, using relation (CK2) of Definition 2.3. Put Fi := {αi n := max {βi j=1 λi jαi jβi j j : ∗ with paths βi j of length j = 1, . . . , ni}, i, j Gi := {βi j : j = 1, . . . , ni} and F := ai = Pα∈F Pβ∈Pn λi α,βαβ∗ with λi α,β m [i=1 Fi. Rewriting the sums for each i, we have = 0 if α < Fi or β < Gi. (cid:3) Lemma 2.6. Let Q be a graph, B a C-algebra, and ρ : LQ → B a homomorphism. Let u := {uv}v∈Q0 ⊂ B such that uv is invertible in ρ(v)Bρ(v) (v ∈ Q0). Then there is a unique homomorphism ρu : LQ → B such that ρu(e) = us(e)ρ(e), ρu(e∗) = ρ(e∗)u−1 s(e) and ρu(v) = ρ(v) (∀e ∈ Q1, v ∈ Q0). Proof. One checks that the elements ρ(x), x ∈ Q0 ∪ Q1 ∪ (Q1)∗ satisfy the relations of Definition 2.3. (cid:3) 3. Leavitt path algebras and semigroups Let Q be a graph and P = P(Q) the set of finite paths. Write (3.1) S = S(Q) = {0} ∪ {αβ∗ : α, β ∈ P} ⊂ LQ. S is the inverse semigroup associated with Q. The Cohn algebra of Q is the semigroup algebra CQ = C[S] of S; its elements are the finite linear combinations of the elements of Lp OPERATOR ALGEBRAS ASSOCIATED WITH ORIENTED GRAPHS 5 S with multiplication induced by that of S. Observe that LQ is the quotient of CQ modulo the relation CK2. Consider S ⊃ E = {0} ∪ {αα∗ : α ∈ P} the sub-semigroup of idempotent elements. The set E is partially ordered by p ≤ q ⇐⇒ pq = p and is a semilattice for this partial order. Observe that for the order of paths defined in (2.2), the bijection P → E \ {0}, α 7→ αα∗ is a poset isomorphism. Note also that p, q ∈ E are incomparable if and only if pq = 0. Let p ∈ E and Z ⊂ {q ∈ E : q ≤ p}. We call Z a cover of p if for every q ≤ p there exists z ∈ Z such that zq , 0. A representation of S in a vector space V is a semigroup homomorphism ρ : S → (End(V), ◦), where the latter is the set of linear endomorphisms considered as a semigroup under composition. The image of E under a representation ρ generates a boolean algebra Bρ with operations p ∧ q = pq, p ∨ q = p + q − pq. By [7, Proposition 11.8], the boolean representation ρ : E → Bρ is tight in the sense of [7, Definition 11.6] if and only if for every p ∈ E and every finite cover Z of p, we have (3.2) ρ(z) = ρ(p). _z∈Z Following [7, Definition 13.1], we call the representation ρ of S tight if its restriction to E is tight. Although the following lemma is well-known to experts, we have not been able to find it explicitly stated in the literature, so we include it here with proof. The particular case of Lemma 3.3 when Q has a single vertex is [10, Lemma 7.5]. See also [17, Corollary 5.3]. Lemma 3.3. Let ρ : S(Q) → End(V) be a representation. Then ρ is tight if and only if it extends to an algebra homomorphism LQ → End(V). Proof. If v ∈ reg(Q), then Z = {ee∗ : e ∈ Q1, s(e) = v} is a finite cover of v and the supremum in (3.2) equals Pe∈Z ρ(ee∗). It follows that if ρ is tight then it extends to an algebra homomorphism LQ → End(V). Assume conversely that ρ extends to LQ. We have to prove that (3.2) holds. Since the supremum in (3.2) depends only on the maximal elements of Z and any two of these are incomparable we may assume that no two distinct elements of Z are comparable. Hence _z∈Z ρ(z) = Xz∈Z ρ(z). If α ∈ P and r(α) = v, then W = α∗Zα is a cover of v and Pz∈Z = αPw∈W wα∗. Hence we may further assume that α = v. We must then prove that the following identity holds in LQ z = v Xz∈Z for each finite cover Z of v in which no two distinct elements are comparable. We do this by induction on n = m(Z) = max{α : αα∗ ∈ Z}. For n = 0 this is trivial. Assume n ≥ 1 and let A = {α ∈ Pn : αα∗ ∈ Z}. Each α ∈ A writes uniquely as αeα where α = n − 1 and eα ∈ Q1. For w ∈ B := {s(eα) : α ∈ A}, put Cw = {eα : s(eα) = w}; because Z is a cover, 6 GUILLERMO CORTI NAS AND MA. EUGENIA RODRIGUEZ Cw = s−1(w). Hence Xα∈A αα∗ =Xβ∈ AXα=β αα∗ βee∗β∗ =Xβ∈ A Xs(e)=r(β) =Xβ∈ A ββ∗. Let Z′ = (Z \ A) ∪ A; then m(Z′) = n − 1, any two distinct elements of Z′ are incomparable, (cid:3) and by the calculation above,Pz′∈Z′ z′ = Pz∈Z z. This concludes the proof. 4. Spatial representations of LQ Let E be a Banach space. We write L(E) for the Banach algebra of bounded linear maps E → E. A representation of LQ on E is an algebra homomorphism ρ : LQ → L(E). We say that ρ is nondegenerate if ρ(LQ)E ⊂ E is dense. In this paper we shall be mostly concerned with Lp-representations, that is, with representations on Banach spaces of the form Lp(X) (1 ≤ p < ∞) where X = (X, B, µ) is a σ-finite measure space. If A ∈ B, we write P(A) for the set of subsets of A and consider A as a measure space with σ-algebra BA := B ∩ P(A) and measure µBA ; thus A = (A, BA, µBA ). We write N(µ) = {A ∈ B : µ(A) = 0}, Bµ = B/N(µ). In what follows, we need to borrow several definitions from [12], pertaining to (partial) isometries between Lp-spaces. Let X = (X, B, µ) and (Y, C, ν) be σ-finite measure spaces. A measurable set transfor- mation from X to Y is homomorphism of σ-algebras S : Bµ → Cν. If S is bijective, then S ∗(µ) = µS −1 is a σ finite measure on C, absolutely continuous with respect to ν. By [12, Proposition 5.6], there is also a map S ∗ : L0(X) → L0(Y) such that S ∗(χE) = χS (E) (E ∈ Bµ). Let 1 ≤ p < ∞; to a bijective measurable set transformation S from X to Y and a measurable function h : Y → C such that h(x) = 1 for almost every x ∈ B one associates an isometric isomorphism u : Lp(X) → Lp(Y) as follows: (4.1) u(ξ)(y) = h(y)([ dS ∗(µ) dν(y) ])1/pS ∗(ξ)(y) (ξ ∈ Lp(X)). An isometric isomorphism u : Lp(X) → Lp(Y) is called spatial if there exist S and h such that u is of the form (4.1). If p , 2, then every isometric isomorphism in L(Lp(X), Lp(Y)) is spatial, by the Banach-Lamperti theorem ([12, Theorem 6.9 and Lemma 6.15]). A partial isometry s : Lp(X) → Lp(X) is spatial if there are A, B ∈ Bµ, called respectively the domain and the range support of s, such that for the projection πA : Lp(X) → Lp(A) and the inclusion ιB : Lp(B) → Lp(X) we have a factorization (4.2) s = ιBuπA where u : Lp(A) → Lp(B) is a spatial isometric isomorphism. If S and h are as in (4.1) we call s the spatial partial isometry associated with the spatial system (S , A, B, h); S and h are the spatial realization and the phase factor of the spatial system. The reverse of the spatial partial isometry (4.2) is the spatial partial isometry t = ιAu−1πB. If p = 2, then the reverse of a spatial partial isometry s is just its adjoint t = s∗. Lp OPERATOR ALGEBRAS ASSOCIATED WITH ORIENTED GRAPHS 7 Example 4.3. Let X = (X, B, µ) be a σ-finite measure space. Let E ∈ B and let χE be the characteristic function. Then the canonical projection πE : Lp(X) → Lp(E) ⊂ Lp(X), πE(ξ) = χE ξ is a spatial partial isometry with spatial system (IdBE , E, E, 1). Every idempotent spatial partial isometry is of this form, by [12, Lemma 6.18]. Remark 4.4. Spatial partial isometries in general and spatial idempotents in particular have norm 1. However the converse does not hold. For example, 1/2 1/2 1/2 1/2! ∈ M2 = B(ℓ p({1, 2}) is a norm one idempotent that is not spatial in our sense (which is that of [12]) for any p ≥ 1 ([12, Example 7.3]). However it is self-adjoint and therefore 2-spatial in the sense of [8, Definition 4.6]. A representation ρ : LQ → L(Lp(X)) is spatial if for each v ∈ Q0, ρ(v) is a spatial idempotent and for each e ∈ Q1, ρ(e) is a spatial partial isometry with reverse ρ(e∗). If ρ is spatial then ρ(x) is spatial for every x ∈ S(Q), whence by Lemma 3.3 a spatial representa- tion of LQ is the same as a tight spatial representation of S(Q), that is, a tight representation of S(Q) which takes values in the inverse semigroup S(Lp(X)) of spatial partial isometries. Remark 4.5. As we explained above, the reverse of a spatial isometry s ∈ L2(X) is just its adjoint. Hence any spatial representation LQ → L(L2(X)) is a ∗-representation. The converse does not hold. For example C is the Leavitt path algebra of the graph consisting of a single vertex and no edges, and the representation ρ : C → M2 = L(ℓ2(N)) that sends 1 to the self-adjoint idempotent of Remark 4.4 is a ∗-representation that is not spatial in our sense. Remark 4.6. If ρ is spatial and α, β ∈ P(Q) are paths with r(α) = r(β), then ρ(αβ∗) is a spatial partial isometry. In particular, ρ(αα∗) is an idempotent spatial partial isometry, and thus by Example 4.3, there is Xα ∈ B such that ρ(αα∗) is the canonical projection πXα : Lp(X) → Lp(Xα) ⊂ Lp(X). If S α is the measurable set transformation of ρ(α) then Xα = S α(Xr(α)), so the spatial system of ρ(α) is of the form (S α, Xr(α), Xα, gα) for some gα : Xα → C such that g(x) = 1 for almost all x ∈ Xα. If α ≥ β, say β = αγ, then Xβ ⊂ Xα because Xβ = S α(Xγ) ⊂ S α(Xr(α)) = Xα. On the other hand if α and β are not comparable then Xα and Xβ are disjoint. In particular, for each v ∈ Q0 the family {Xe : s(e) = v} ⊂ B ∩ P(Xv) is disjoint, and if v is regular its union is the whole Xv: (4.7) Xv = ae∈s−1(v) Xe (v ∈ reg(Q)). It follows from (4.7) that if Q is nonsingular then for each l ≥ 0 we have (4.8) Xv = aα∈vPl(Q) Xα (Q0 = reg(Q)). Conversely, if we are given disjoint families {Xv : v ∈ Q0} ⊂ B and {Xe : e ∈ Q1, s(e) = v} ⊂ B ∩ P(Xv) (v ∈ Q0) satisfying (4.7) and a family {se : e ∈ Q1} of spatial partial isometries in L(Lp(X)) with range and source projections πXe and πXv , then we have a unique algebra homomorphism ρ : LQ → L(Lp(X)) mapping ρ(v) = πXv , ρ(e) = se, and sending e∗ to the reverse of se. 8 GUILLERMO CORTI NAS AND MA. EUGENIA RODRIGUEZ Lemma 4.9. Let X be a σ-finite measure space. A spatial representation ρ : LQ → L(Lp(X)) is nondegenerate if and only if (4.10) X = av∈Q0 Xv. Proof. Immediate from the fact that ρ(LQ)Lp(X) = Xv∈Q0 ρ(v)Lp(X) = Mv∈Q0 Lp(Xv). (cid:3) It follows from (4.8) and Lemma 4.9 that if Q is nonsingular and ρ is nondegenerate, then for each l ≥ 0 we have (4.11) X = aα∈Pl (Q) Xα. Lemma 4.12. Let Q be a graph, 1 ≤ p < ∞, X = (X, B, µ) a σ-finite measure space, and ρ : LQ → L(Lp(X)) a spatial representation. Then there are X′ ∈ B and a nondegenerate spatial representation ρ′ : LQ → L(Lp(X′)) such that ρ factors as ρ′ followed by the inclusion L(Lp(X′)) ⊂ L(Lp(X)). Proof. Put X′ = `v∈Q0 Xv. Example 4.13. Let Q be a graph, and let (cid:3) (4.14) X = XQ = {α : infinite path in Q} ∪ {α ∈ P : r(α) ∈ sing(Q)}. For α ∈ P, let X ⊃ Zα = {x ∈ X : α ≥ x} = αX. The sets Zα are the basis of a topology which makes it a locally compact Hausdorff space; modulo our different conventions for ranges and sources, this is the space considered in [6, page 3]. The inverse semigroup S = S(Q) acts on X by partial homeomorphisms; an element u = αβ∗ ∈ S acts on X with domain Zβ and range Zα via (4.15) αβ∗(βx) = αx. Let B the the σ-algebra of all Borel subsets of X. The semigroup S of (3.1) acts on X via (4.15). If α, β ∈ P with r(α) = r(β), then (4.16) S αβ∗ : BZβ → BZα , A 7→ αβ∗(A) is a bijective homomorphism of σ-algebras. Let µ be a measure on B; µ is quasi-invariant under αβ∗ if µZβ and µZα ◦ βα∗ are equivalent measures (that is, if they are absolutely continuous with respect to each other); µ is quasi-invariant under S it is quasi-invariant under any element of S. One can show that X always has a σ-finite measure that is quasi- invariant under S. For example, in case X is countable we can take µ to be the counting measure. Assume that µ is a σ-finite measure on the Borel subsets of X, quasi-invariant under S, and let sαβ∗ be the spatial isometry of (4.1) with spatial realization S = S αβ∗ and constant phase factor h = 1. Then S → L(Lp(X, µ)), αβ∗ 7→ sαβ∗ is a tight nondegenerate spatial representation of S and thus induces a nondegenerate spa- tial representation ρµ : LQ → L(Lp(X, µ)). In general, ρµ is not injective. For example, if Q consists of one vertex and one loop, then LQ (cid:27) C[t, t−1] and ρµ is 1-dimensional. Lp OPERATOR ALGEBRAS ASSOCIATED WITH ORIENTED GRAPHS 9 Construction 4.17. Let X be a countable set, and let I(X) be the inverse semigroup of all partially defined injections X ⊃ dom f f −→ X. Let Q be a countable graph, S = S(Q) its associated inverse semigroup and S : S → I(X) a semigroup homomorphism. For each α ∈ P = P(Q), set Xα = dom(S α). We shall assume that S is tight, i.e. that the identities (4.7) and (4.10) are satisfied. Let G = G(S, X) be the groupoid of germs, as defined in [7, Section 4]. The elements of G are equivalence classes [αβ∗, x] where r(α) = r(β), x ∈ Xβ; the equivalence relation is determined by the prescription that [αβ∗, x] = [αγγ∗β∗, x] for any γ ∈ P with s(γ) = r(α). For αβ∗ ∈ S \ {0}, put Θα,β = {[αβ∗, x] : x ∈ Xβ} ⊂ G. Let A(G) ⊂ map(G, C) be the linear subspace generated by the characteristic functions χΘα,β , (αβ∗ ∈ S \ {0}). One checks that A(G) is an algebra under the convolution product (it is in fact the Steinberg algebra of G [16]) and that (4.18) ψ : LQ → A(G), φ(αβ∗) = χΘα,β is an algebra homomorphism. Let (4.19) L : A(G) → L(ℓ p(G)), L( f )(ξ)(h) = Xg∈G f (g)ξ(g−1h) This is well-defined because the domain and range functions are injective on each Θα,β. One checks that L is a monomorphism. Consider the composite (4.20) ρ = Lψ : LQ → L(ℓ p(G)). Let αβ∗ ∈ S(Q) and consider the following subsets of G: A = {[γδ∗, δx] : β ≥ γx}, B = {[αβ∗γδ∗, δx] : β ≥ γx}. The map A → B, [γδ∗, δx] 7→ [αβ∗, γx][γδ∗, δx] = [αβ∗γδ∗, δx] is bijective and thus induces a cardinality preserving bijection S α,β : P(A) → P(B). One checks that ρ(αβ∗) is the spatial isometry with spatial system (S α,β, A, B, 1). Hence ρ is a spatial, nondegenerate representation. Lemma 4.21. Assume that in Construction 4.17, one has Xv , ∅ for all v ∈ Q0. Then (4.18) is an isomorphism and (4.20) is an injective, nondegenerate spatial representation. Proof. Put A(G)n = span{ψ(αβ∗) : αβ∗ = n}; we have (4.22) A(G) = Xn A(G)n. Let c : G → Z, c([αβ∗, x]) = αβ∗; note that the elements of A(G)n are supported in c−1({n}). It follows from this that the sum in (4.22) is direct. Moreover, because c is a groupoid homomorphism, we have A(G)nA(G)m ⊂ A(G)n+m. Thus ψ is a homogeneous homomorphism of graded algebras. But for v ∈ Q0, ψ(v) is the characteristic function of {[v, x] : x ∈ Xv} which is nonempty by hypothesis, so ψ(v) , 0. By [1, Theorem 2.2.15] this implies that ψ is an isomorphism. (cid:3) Proposition 4.23. Let Q be a countable graph. Then LQ has an injective, nondegenerate spatial representation LQ → L(ℓ p(N)). 10 GUILLERMO CORTI NAS AND MA. EUGENIA RODRIGUEZ Proof. Let X be any countably infinite set. Because X is infinite and #Q0 ≤ #X, there exists a bijection φ : X → Q0 × X. For v ∈ Q0, set Xv = φ−1({v} × X); observe that (4.10) is satisfied by construction. Put Q1(v, −) = s−1({v}) ⊂ Q1 and let Rv = ( Q1(v, −) {v}` Q1(v, −) v ∈ reg(Q) v ∈ sing(Q). Because #Xv = #X is infinite and #Rv ≤ #Xv, there is a bijection ζv : Xv → Rv × X. Set Xe = ζ−1 s(e)({e} × Xs(e)) (e ∈ Q1). By construction, (4.7) is satisfied. For e ∈ Q1, let r−1 × 1 : {r(e)} × X → {e} × X be the obvious bijection. Define a semigroup homomorphism S : S(Q) → I(X) by setting S v = 1Xv , S e = ζ−1 s(e)(r−1 × 1)φ : Xr(e) → Xe, S e∗ = S −1 e (v ∈ Q0, e ∈ Q1). Let G be the groupoid of germs associated to this action of S on X, and consider the nondegenerate spatial representation ρ : LQ → L(ℓ p(G)) of (4.20). Then ρ is injective by Lemma 4.21; furthermore, #G = ℵ0 and any bijection G (cid:27) N induces a spatial isometric isomorphism ℓ p(G) (cid:27) ℓ p(N). (cid:3) 5. Matrix algebras and spatial representations Let 1 ≤ n ≤ ∞ and let A be an algebra. Write Mn for the algebra of n × n-matrices with finitely many nonzero entries, and MnA = Mn ⊗ A. If i, j ∈ N, we write Ei, j for the canonical matrix unit. Let Q be a countable graph, X a σ-finite measure space, and 1 ≤ p < ∞. Call a representation ρ : Mn(LQ) → L(Lp(X)) spatial if for every x ∈ Q0 ∪ Q1 and i, j, ρ(Ei, j ⊗ x) is a spatial partial isometry with reverse ρ(E j,i ⊗ x∗). Remark 5.1. Let n ≤ ∞ and let MnQ be the graph obtained by adding a head . . . / vi ev i / vi−1 ev i−1 / . . . ev 2 / v1 ev 1 / v for each v ∈ Q0 and i < n. By [3, Propositions 9.3 and 9.8], there is a ∗-isomorphism (5.2) LMn Q v 7→ E1,1 ⊗ v, (cid:27) −→ MnLQ, vi 7→ Ei+1,i+1 ⊗ v ev i 7→ Ei+1,i ⊗ e e 7→ E1,1 ⊗ e, It is clear that a representation MnLQ → L(Lp(X)) is spatial in the matricial sense above if and only if its composition with the map (5.2) is a spatial representation of LMn Q. Example 5.3. Let σ : LQ → L(Lp(X)) be a spatial representation. Let I = {1, . . . , n} if n is finite, and I = N if n = ∞. We have a canonical isometric isomorphism Lp(I × X) (cid:27) ℓ p(I, Lp(X)). Let σI : MnLQ → L(ℓ p(I, Lp(X))) σI (Ei, j ⊗ a)(ξ)(k) = δk,iσ(a)(ξ( j)). Then σI is spatial. Indeed if a ∈ S(Q) and σ(a) is a spatial isometry with domain support E and rank support F, then σI (Ei, j ⊗a) is a spatial isometry with domain support { j}×E and range support {i} × F. We remark that for I = {1, . . . , n}, σI is the representation induced by the amplification of σ in the sense of [8, Definition 4.10]. Lemma 5.4. Let Q be a countable graph, I a countable set, X a σ-finite measure space, 1 ≤ p < ∞, and ρ : MI LQ → L(Lp(X)) a nondegenerate spatial representation. Then there exist a σ-finite measure space Y, a spatial representation σ : LQ → L(Lp(Y)) and / / / / / Lp OPERATOR ALGEBRAS ASSOCIATED WITH ORIENTED GRAPHS 11 a spatial isometric isomorphism u : ℓ p(I, Lp(Y)) → Lp(X) such that ρ(a) = uσI (a)u−1 (a ∈ LQ). Proof. Let Xi,v be the domain support of the spatial idempotent ρ(Ei,i ⊗ v) (∈ I, v ∈ Q0). Set Xi = `v∈Q0 Xi,v; we have X = `i∈I Xi. Hence we have Lp-direct sum decompositions Lp(X) = Li∈I Lp(Xi) and Lp(Xi) = Lv∈Q0 Lp(Xi,v). Choose i0 ∈ I, and let Y = Xi0 . Then u = Li,v ρ(Ei,i0 ⊗ v) is a spatial isometric isomorphism ℓ p(I, Lp(Y)) = Li∈I Lp(Y) → Lp(X). Let σ : LQ → L(Lp(Y)), σ(a) = ρ(Ei0,i0 ⊗ a). One checks that u conjugates σI to ρ, concluding the proof. (cid:3) 6. A spatiality criterion We write Mn = MnC for the matrix algebra and M∞ = Sn Mn. We have a natural identification Mn = L(ℓ p({1, . . . , n}) for n < ∞ and a natural embedding M∞ → L(ℓ p(N)); by pulling back the operator norm, we get a norm p on Mn (1 ≤ n ≤ ∞) which makes the latter into a normed algebra M p n . If I is a set and (6.1) n = (ni)i∈I is a family with 1 ≤ ni ≤ ∞, we write (6.2) M p n = Mi∈I M p ni for the algebraic direct sum equipped with the supremum norm (ai) = supi∈I aip. We write Ei a,b (i ∈ I), 1 ≤ a, b ≤ ni for the canonical matrix unit. The following proposition generalizes [12, Theorem 7.2]. Proposition 6.3. Let p ∈ [1, ∞), p , 2, I a countable set, n as in (6.1), and M p n as in (6.2). Let X = (X, B, µ) be a σ-finite measure space with µ , 0. The following are equivalent for a nondegenerate representation ρ : M p i) ρ(Ei ii) ρ is contractive. a,b) is a spatial partial isometry for all i ∈ I, 1 ≤ a, b ≤ ni. n → L(Lp(X)). a ∈ B such that ρ(Ei a,a) = πXi a a. Put Xi = `a∈Ni Xi nondegenerate, we have X = `i∈I `a∈Ni Xi Proof. Assume that i) holds. Then each ρ(Ei a,a) is a spatial idempotent, whence by Ex- ample 4.3 there is Xi is the canonical projection. For each i ∈ I put Ni = N if ni = ∞ and Ni = {1, . . . , ni} if ni < ∞. Because ρ is a. By restriction, we : Mni → L(Lp(Xi)) satisfying i); hence we obtain a nondegenerate representation ρi may assume that I = {1} has only one element. If n < ∞, nondegeneracy implies that ρ(1) = 1, so ρ is contractive by [12, Theorem 7.2]. Assume n = ∞. Proceed as in loc. cit., using the partial isometries ρ(E1,a) : Lp(Xa) → Lp(X1) to construct an isometry u : Lp(X) → ℓ p(N, Lp(X1)) = ℓ p(N) ⊗p Lp(X1) (the Lp-tensor product) that conjugates ρ to the contractive representation T 7→ T ⊗ 1. It follows that ρ is contractive, concluding the proof that i)⇒ii). Assume now that ii) holds. Then {ρ(Ei a,a) : i ∈ I, a ∈ Ni} is a family of orthogonal idempotents. Let Bi a a is dense in Lp(X). For each z ∈ S1, i ∈ I and a ∈ Ni define an operator ui,a(z) : B → B as multiplication by z on Bi a and the identity on every other summand. Because ρ is contrac- tive, ui,a(z) has norm 1, so it extends to a norm 1 operator ui,a(z) ∈ L(Lp(X)). Since this also holds for ui,a(z−1), ui,a(z) is a bijective isometry. Hence it is spatial, by the Banach- Lamperti theorem. Now proceed as in [12, page 42] to deduce that ρ(Ei a,a) = (1−ui,a(−1))/2 is a spatial idempotent. Hence there exists Xi a. a,a)Lp(X); then the algebraic direct sum B = Li,a Bi a ∈ B such that Bi a = ρ(Ei = Lp(Xi a) and X = ` Xi 12 GUILLERMO CORTI NAS AND MA. EUGENIA RODRIGUEZ Since ρ(Ei shows that it is spatial. a,b) is an isometry Bi b → Bi a, another application of the Banach-Lamperti theorem (cid:3) Recall that the Leavitt path algebra is equipped with a Z-grading LQ = Ln(LQ)n where (LQ)n is as in (2.4). Write (LQ)0,n ⊂ (LQ)0 for the subalgebra linearly spanned by the elements of the form αβ∗ with r(α) = r(β) and α = β ≤ n. We have an increasing union (LQ)0 = ∞ (LQ)0,n. [n=0 Each (LQ)0,n is isomorphic to a direct sum of (possibly infinite dimensional) matrix alge- bras. Theorem 6.4. (cf. [8, Theorem 7.7] ) Let X = (X, B, µ) be a σ-finite measure space with µ , 0, p ∈ [1, ∞), p , 2, and Q a countable graph. The following are equivalent for a nondegenerate representation ρ : LQ → L(Lp(X)). i) ρ is spatial. ii) ρ(e), ρ(e∗) ≤ 1 (e ∈ Q1) and the restriction of ρ to ((LQ)0,1, p) is contractive. Proof. The implication i)⇒ ii) is clear using Proposition 6.3. Assume that ii) holds; then ρ(e) is a bijective isometry ρ(r(e))Lp(X) → ρ(ee∗)Lp(X) with inverse ρ(e∗). By Proposition 6.3, ρ(v) and ρ(ee∗) are spatial idempotents (v ∈ Q0), (e ∈ Q1). Hence it follows from the Banach-Lamperti theorem [12, Theorem 6.9] and from [12, Lemma 6.15] that ρ(e) and ρ(e∗) are spatial. This concludes the proof. (cid:3) Remark 6.5. The assumption that ρ be nondegenerate in necessary in both Proposition 6.3 and Theorem 6.4. For example the trivial graph on one vertex has Leavitt algebra C, which equals M p 2 that maps 1 to the idempotent of Remark 4.4 is contractive but not spatial. 1 for all 1 ≤ p < ∞, and the representation C → M p 7. The Lp-operator algebra Op(Q) Definition 7.1. Let p ∈ [1, ∞). An Lp-operator algebra is a Banach algebra B together with a norm on each Mn B that makes into a Banach algebra in such a way that there exists a nondegenerate representation ρ : B → L(Lp(X)) for some σ-finite measure space X, i=1 X)) is isometric for each 1 ≤ n < ∞. We call B standard if X can be chosen to be a standard Borel space. A homomorphism f : A → B between Lp-operator algebras is p-completely contractive (resp. isometric) if Mn f is contractive (resp. isometric) for every n. such that Mnρ : MnB → MnL(Lp(X)) = L(Lp(`n Remark 7.2. By [13, Proposition 1.25], any separable Lp-operator algebra admits an iso- metric representation in a separable, whence standard Lp-space. Thus a separable Lp- operator algebra is automatically standard. Remark 7.3. If p , 1 and B has a contractive approximate unit, then the condition that the isometric representation in Definition 7.1 be nondegenerate can be dropped, by [10, Theorem 3.19]. A spatial p-seminorm is a seminorm h : LQ → R≥0 such that there exist a σ-finite measure space X and spatial representation ρ : LQ → L(Lp(X)) such that h(a) = ρ(a) (a ∈ LQ). Observe that by Lemma 4.12, every spatial seminorm is induced by a nondegenerate Lp OPERATOR ALGEBRAS ASSOCIATED WITH ORIENTED GRAPHS 13 spatial representation. Put (7.4) (7.5) p-ssn(Q) ={h : LQ → R≥0 spatial p-seminorm}, a = sup{h(a) : h ∈ p-ssn(Q)}. is a norm. Write Op(Q) = LQ By Proposition 4.23, for the completion of LQ with respect to the norm (7.5); Op(Q) is a Banach algebra, and the canonical map LQ → Op(Q) is injective, again by Proposition 4.23. Since Q is countable, there is a countable family {ρn} of σ-finite nondegenerate spatial representations such that is the norm associated to the Lp-direct sum (7.6) ρ = Mn ρn : LQ → L(Lp(an Xn)) which is a nondegenerate spatial representation. Hence Op(Q) is isometrically isomorphic to the closure of ρ(LQ). Proposition 7.7. Let Q be a countable graph. Then Op(Q) has a canonical structure of Lp-operator algebra such that there is an isometric isomorphism MnOp(Q) (cid:27) Op(MnQ) (∞ > n ≥ 1). Proof. By Remark 5.1, the canonical map MnLQ (cid:27) LMn Q → Op(MnQ) is universal for Lp- spatial representations. By Lemma 5.4 and the discussion above, for each n there is a spatial representation ρn : LQ → L(Lp(Xn)) such that n := Mnρn( ) is the supremum of all p-spatial norms on LMn(Q). Let X = `n Xn and let ρ = Ln ρn : LQ → L(Lp(X)) be the n for all n ≥ 1, and we have isometric isomorphisms Lp-direct sum. Then Mnρ( ) = Op(MnQ) (cid:27) Mnρ(Mn(LQ)) = Mn(ρ(LQ)) = Mn(ρ(LQ)) (cid:27) MnOp(Q). (cid:3) Theorem 7.8. Let X be σ-finite measure space with nonzero measure, p ∈ [1, ∞), p , 2, Q a countable graph, ρ : Op(Q) → L(Lp(X)) a nondegenerate representation and ρ : LQ → L(Lp(X)) the restriction of ρ. Then the following conditions are equivalent: i) ρ is spatial. ii) ρ is contractive. Proof. If ρ is spatial then it induces a contractive homomorphism ρ′ : Op(Q) → L(Lp(X)) which agrees with ρ on LQ; since ρ does the same, we must have ρ = ρ′. This proves that i)⇒ii). Conversely if ii) holds, then ρ is spatial by Theorem 6.4. (cid:3) Theorem 7.9. Let X be σ-finite measure space with nonzero measure, p as in Theorem 7.8, Q a countable graph, ρ : Op(Q) → L(Lp(X)) a representation and ρ : LQ → L(Lp(X)) the restriction of ρ. Further assume either that p , 1 or that Q0 is finite. Then the following conditions are equivalent: i) There exist a σ-finite measure space Y, an isometry ι : Lp(Y) → Lp(X), a norm 1 operator π : Lp(X) → Lp(Y) such that πι = 1, and a spatial representation ρ′ : LQ → L(Lp(Y)), such that for f : L(Lp(Y)) → L(Lp(X)), f (T ) = ιT π, the following diagram commutes LQ ii) ρ is contractive. ❍❍ ❍❍ ρ′ ❍❍ ρ / L(Lp(X)) 8qqqqqqqqqq f ❍❍ $❍ L(Lp(Y)) $ / 8 14 GUILLERMO CORTI NAS AND MA. EUGENIA RODRIGUEZ Proof. If i) is satisfied, then ρ′ factors through a contractive representation ρ′ : Op(Q) → L(Lp(Y)). Thus f ρ′ = ρ is contractive. Assume conversely that ii) holds. Let E ⊂ L(Lp(X)) be the closure of ρ(LQ)L(Lp(X)). If Q0 is finite then Op(Q) is unital with unit 1 = Pv∈Q0 v which has norm 1; thus E is the image of the contractive idempotent ρ(1). For general Q, the family {Pv∈F v} indexed by the finite subsets of Q0 is a contractive approxi- mate unit of Op(Q); hence if p , 1, then again E is the image of a contractive idempotent, by [10, Corollary 3.13]. Hence under either hypothesis, by [5, Theorem 4] there are a con- tractive projection π′ : Lp(X) → E and an isometric isomorphism h : E → Lp(Y) for some standard Borel space Y. If p = 1 but Put π = hπ′, let ι be h−1 followed by the inclusion E ⊂ Lp(X), and set ρ′(a) = hρ(a)h−1. It is clear that the diagram commutes; moreover, ρ′ is spatial by Theorem 6.4. (cid:3) Remark 7.10. Let S(Q) be the semigroup of (3.1) and let 1 < p < ∞. Let F p tight(S(Q)) be the standard Lp-operator algebra of [8, Definition 6.7]; F p tight(S(Q)) is universal for tight Lp-representations of S(Q) which are spatial in the sense of [8, Definition 4.6] and take values in Lp-spaces of standard Borel spaces. As pointed out above, the spatiality notion of [8] agrees with ours for p , 2. Hence by Lemma 3.3 and the universal property of Op(Q), for p ∈ (1, ∞), p , 2, we have a canonical contractive homomorphism (7.11) Op(Q) → F p tight(S (Q)) (p ∈ (1, ∞), p , 2). Moreover, since the p-operator space structure on F p Mn(F p also contractive, by Proposition 7.7. In other words (7.11) is p-completely contractive. tight(S (MnQ)), the induced map Mn(Op(Q)) → Mn(F p tight(S (Q)) is defined in [8] so that tight(S (Q))) is tight(S (Q)) = F p Proposition 7.12. The map (7.11) is a p-completely isometric isomorphism. Proof. It suffices to show that F p tight(S(Q)) is universal for all σ-finite representations. Let X be a σ-finite measure space and let ρ : LQ → L(Lp(X)) be a spatial representation; we have to show that ρ factors through LQ → F p tight(S(Q)). An argument similar to that of the proof of Theorem 7.9 shows that ρ factors through a nondegenerate representation ρ′ : LQ → L(Lp(Y)) with Y standard Borel. Thus ρ factors through LQ → F p tight(S(Q)), as required. (cid:3) 8. Spatial seminorms, desingularization, and source removal Let Q be a countable, singular graph. Recall from [2, Section 5] that the desingulariza- tion of Q is a nonsingular graph Qd obtained from Q as follows. For each sink v, add an infinite tail (8.1) For each infinite emitter v, number the elements of s−1(v) = {e1, e2, . . . } and add a tail (8.1) and an arrow gi : vi → r(ei) (1 ≤ i). There is a canonical ∗-monomorphism [2, Proposition 5.5] v = v0 f3 → · · · f1 → v1 f2 → v2 (8.2) φd : LQ → LQd , φd(v) = v, φd(e) = ( e s(e) ∈ reg(Q) f1 · · · figi e = ei If Q is a graph such that sour(Q) , ∅, we may embed it in the source-free graph Qr obtained by adding an infinite head (8.3) w = w0 f1 ← w1 f2 ← w2 f3 ← · · · Lp OPERATOR ALGEBRAS ASSOCIATED WITH ORIENTED GRAPHS 15 at each w ∈ sour(Q). The obvious inclusion Q ⊂ Qr induces an algebra monomorphism (8.4) φr : LQ → LQr . Observe that for # ∈ {d, r}, composition with φ# sends spatial representations of LQ# to spatial representations of LQ. Hence we have an induced map (8.5) p-ssn(Q#) → p-ssn(Q), h 7→ h ◦ φ#. Proposition 8.6. Let Q be a countable graph, 1 ≤ p < ∞, and # ∈ {r, d}. Then the map (8.5) is surjective. Proof. It suffices to show that for every nonzero spatial representation ρ : LQ → L(Lp(X)) there exist a spatial representation ρ# : LQ# → L(Lp(Y)) and a spatial isometry s : Lp(X) → Lp(Y) with reverse t, with both Y and s depending on ρ and #, such that for the map σ : L(Lp(X)) → L(Lp(Y)), σ(A) = sAt, the following diagram commutes: (8.7) LQ φ# LQ# ρ / L(Lp(X)) σ / L(Lp(Y)) ρ# We begin by the case # = r. If α ∈ P(Q), we write Xα for the support of the spatial projection ρ(αα∗). Regard N as a measure space with counting measure; set Y := X ⊔ Fw∈sour(Q)(Xw × N). Let s and t be the inverse isometries induced by the inclusion X ⊂ Y. The canonical identification Xw → Xw × {n} induces an isometric spatial isomorphism τn : Lp(Xw) → Lp(Xw × {n}). Extend ρ along φr to a map ρr : LQr → L(Lp(Y)) by setting ρr(wn) := IdLp(Xw×{n}), ρr( fn) := τnτ−1 n . One checks that ρr is well-defined and makes (8.7) commute. Next we consider the case # = d. The measure space Y will be a coproduct n ) = τn−1τ−1 n−1, ρr( f ∗ Y = X ⊔ av∈sing(Q),n≥1 Yvn ; the isometries s, t will be those induced by the inclusion X ⊂ Y. For v ∈ sink(Q), we set n . If −→ Xv × {n} the obvious bijection, and put ρd( fn) = τn−1τ−1 = Xv × {n}, τn : Xv Yvn v ∈ inf(Q) and X′ v i=1 Xei, we set (cid:27) = Xv \`∞ Yvn = X′ v ⊔ai≥n Xei and let ρd( fn) be induced by the inclusion Yvn ⊂ Yvn−1 and ρd(gn) by the composite of ρ(en) : Lp(Xr(en)) → Lp(Xen ) followed by the inclusion Lp(Xen ) ⊂ Lp(Yvn ). One checks that this prescription defines a spatial representation ρd : LQd → L(Lp(Y)) that makes (8.7) commute. (cid:3) Corollary 8.8. The canonical homomorphisms (8.2) and (8.4) induce isometric homomor- phisms Op(Q) → Op(Qd) and Op(Q) → Op(Qr). The purpose of this section is to prove the following theorem. 9. A uniqueness theorem Theorem 9.1. Let Q be a countable graph and 1 ≤ p < ∞. If LQ is simple, then the set p-ssn(Q) of p-spatial seminorms on LQ has only one nonzero element. In particular, / / 16 GUILLERMO CORTI NAS AND MA. EUGENIA RODRIGUEZ if ρ : LQ → L(Lp(X)) is any nonzero spatial representation and ρ(LQ) ⊂ L(Lp(X)) the operator norm closure, then the natural map is an isometric isomorphism Op(Q) (cid:27) −→ ρ(LQ). The proof of Theorem 9.1 will be given at the end of the section, after a series of propositions, definitions, and lemmas, which adapt and extend those in [12, Section 8]. Definition 9.2. Let Q be a countable row-finite graph, p ∈ [1, ∞), X = (X, B, µ) a σ-finite measure space, and ρ : LQ → L(Lp(X)) a representation. (1) We say that ρ is free if there is a partition X = Fm∈Z Em, Em ∈ B, such that for all m ∈ Z, e ∈ Q1, we have (9.3) ρ(e)(Lp(Em)) ⊂ Lp(Em+1) and ρ(e∗)(Lp(Em)) ⊂ Lp(Em−1). (2) We say that ρ is approximately free if for every N ∈ N, there are n ≥ N and a partition X = n−1 Gm=0 Em, Em ∈ B, such that for m = 0, . . . , n − 1 and all e ∈ Q1 (9.3) holds if we set En = E0 and E−1 = En−1. Lemma 9.4. Let p ≥ 1, X = (X, B, µ) and Y = (Y, C, ν) σ-finite measure spaces, Q a row-finite graph, ρ : LQ → L(Lp(X)) a representation, and u ∈ L(Lp(Y)) an invertible operator. Then, there is unique representation ρu : LQ → L(Lp(X × Y)) such that, for all e ∈ Q1, we have ρu(e) = ρ(e) ⊗ u and ρu(e∗) = ρ(e∗) ⊗ u−1. Moreover, ρu has the following properties: (a) If α ∈ LQ is homogeneous of degree k with respect to the Z-grading of (2.4), then ρu(α) = ρ(α) ⊗ uk. (b) If u is isometric, p , 2 and ρ is spatial, then ρu is spatial. (c) If there is a partition Y = am∈Z Z, then ρu is free in the sense of Definition 9.2. Fm, Fm ∈ C, such that u(Lp(Fm)) = Lp(Fm+1) ∀m ∈ Proof. The proof is analogous to that of [12, Lemma 8.2] using Lemma 2.6 instead of [12, Lemmas 2.18, 2.19 and 2.20]. (cid:3) Proposition 9.5. Let p, X, Q, and ρ be as in Lemma 9.4. Let u ∈ L(ℓ p(Z)) be the shift operator, (u(x))(m) := x(m − 1) (x ∈ ℓ p(Z)). Let ρu be as in Lemma 9.4. Then, for all a ∈ LQ, we have kρu(a)k ≥ kρ(a)k. Proof. The proof is analogous to that of [12, Proposition 8.3], using Lemma 9.4 instead of [12, Lemma 8.2]. (cid:3) Lemma 9.6. Let Q be a nonsingular countable graph such that LQ is simple. Let X = (X, B, µ) be a σ-finite measure space. Let {Xv}v∈Q0 ⊂ B a family of sets of nonzero measure, {Xe}e∈Q1 ⊂ B a disjoint family such that X = av∈Q0 Xv and Xv = a{e:s(e)=v} Xe (∀v ∈ Q0), and S e : (Xr(e), BXr(e) , µXr(e) ) → (Xe, BXe , µXe ) (e ∈ Q1) a bijective measurable set transformation. If α = α1 · · · αm is a path, write S α = S α1 ◦ · · · ◦ S αm . Then, for each n ≥ 0 and each v ∈ Q0 there is a set Ev ∈ BXv such that µ(Ev) , 0, and such that the family {S α(Ev) : r(α) = v, α ≤ n} is disjoint. Lp OPERATOR ALGEBRAS ASSOCIATED WITH ORIENTED GRAPHS 17 Proof. We shall use the fact that, because LQ is simple, Q is cofinal, i.e. for every v ∈ Q0 and each cycle c there is a path starting at v and ending at some vertex in c (see [1, Theorem 2.9.7]). Let v ∈ Q0. If v ∈ Q0 is not in any cycle, we set Ev = Xv; observe that µ(Ev) , 0 by hypothesis. Because v is not in any cycle, any two distinct paths ending in v are incomparable, and so Ev satisfies the disjointness condition of the lemma. Next assume that v belongs to a cycle. Let α := αv be a cycle based at v and let β be a closed path with s(β) = v that agrees with α up to an exit, goes out following the exit, returns to c (which is possible by cofinality) and follows it till it gets back to v. Consider the infinite path γ := αβααββαααβββ . . . It is long, but straightforward to check that (9.7) Let n ∈ N and v ∈ E0. For i ≥ 1, let γi be the i-th edge of γ. Put ∄θ ∈ P(Q) such that θθ ≥ γ. B ∋ Ev := Xγ1...γ2n . Then µ(Ev) , 0 because µ(Xw) , 0 for all w ∈ Q0. Let η and τ be different paths such that r(η) = r(τ) = v, of lengths k and l respectively (k ≤ l ≤ n). We have to check that S η(Ev) and S τ(Ev) are disjoint. If k = 0 this is clear from Remark 4.6, because S τ(Ev) = Xτγ1...γ2n and the paths τγ1 . . . γ2n and γ1 . . . γ2n are incomparable, by (9.7). So assume that 0 < k ≤ l; if η and τ are incomparable, we are done. Otherwise, we must have η > τ; say τ = ηδ. Hence S η(Ev) ∩ S τ(Ev) = S η(Ev ∩ S δ(Ev)) has measure zero because Ev ∩ S δ(Ev) does. (cid:3) Let (X, B, µ) be a σ-finite measure space and τ1, . . . , τn ∈ L(Lp(X)) spatial partial isometries with reverses σ1, . . . , σn. Call τ1, . . . , τn orthogonal if τ jσi = σiτ j = 0 ∀i , j. Lemma 9.8. Let X be a σ-finite measure space, p ∈ [1, ∞), τ1, . . . , τn ∈ L(Lp(X)) orthog- onal spatial partial isometries, λ ∈ Cn, and τλ = λiτi. Then kτλk = kλk∞. n Xi=1 Proof. Straightforward. (cid:3) Proposition 9.9. Let Q be a nonsingular countable graph without sources. Let p ∈ [1, ∞), p , 2, and let X and Y be measure spaces and ρ : LQ → L(Lp(X)) and φ : LQ → L(Lp(Y)) spatial representations. Assume that LQ is simple and that ρ is approximately free. Then kρ(a)k ≤ kφ(a)k (a ∈ LQ). Proof. This proposition generalizes [12, Proposition 8.6]; we shall adapt the argument therein using Lemma 9.6 instead of [12, Lemma 8.5]. Let X′ = `v∈Q0 Xv; observe that the corestriction ρ′ of ρ to L(Lp(X′)) is approximately free. Hence by Lemma 4.12 we may assume that ρ and φ are both nondegenerate. For each α ∈ P = P(Q), let Rα and S α be the bijective measurable set transformations Xr(α) → Xα, Yr(α) → Yα associated to ρ(α) and φ(α), as in Remark 4.6. We have to show that if a ∈ LQ is such that kρ(a)k = 1, then kφ(a)k ≥ 1. By Lemma 2.5, there are N0 ≥ 0, a finite set F0 ⊂ P and a finitely supported function λ0 : F0 × PN0 → C such that a = Xα∈F0 Xβ∈PN0 α,βαβ∗ λ0 Because sour(Q) = ∅ by hypothesis, for each v ∈ s(F0) we may choose a path τv ∈ PN0 with r(τv) = v. Put x = Xv∈s(F0 ) τv, b = xa. 18 GUILLERMO CORTI NAS AND MA. EUGENIA RODRIGUEZ = {τv : v ∈ s(F0)} is of length N0, any two of them Because every path in the set τF0 are incomparable. Hence by Remark 4.6, the elements of ρ(τF0 ) are orthogonal spatial partial isometries. Therefore kρ(x)k = 1, by Lemma 9.8; similarly, kρ(x∗)k = 1. Hence kρ(b)k = kρ(a)k = 1 and by the same argument, kφ(b)k = kφ(a)k. Therefore it suffices to show that for every ǫ > 0, (9.10) kφ(b)k > 1 − ǫ. For β ∈ PN0 and α ∈ F0, let λτs(α)α,β = λ0 α,β. Put F = {τs(α)α : α ∈ F0}; the map F0 → F, α 7→ τs(α)α is clearly surjective. Moreover, because τv ∈ PN0 for all v ∈ s(F0), it is also injective. Using this in the third step, we obtain τv)(Xα∈F0 Xβ∈PN0 α,βαβ∗) λ0 b =( Xv∈s(F0) = Xv∈s(F0 ) Xα∈F0 =Xα∈F Xβ∈PN0 s(α)=v λτvα,βαβ∗ Xβ∈PN0 λα,βαβ∗ Let N1 = max{α : α ∈ F0}; then N0 ≤ α ≤ N0 + N1 for all α ∈ F. If N0 = N1 = 0, then b is a linear combination of vertices, b = Pv λvv, whence by Lemma 9.8 we have kφ(b)k = kλk∞ = kρ(b)k = 1. Hence (9.10) holds in this case. So we may assume N0 + N1 > 0, and take j > (N0 + N1)(2/ǫ)p. By our hypothesis on ρ, there are N ≥ j(N0 + N1) and a partition (9.11) X = N−1 an=0 Dn such that for the remainder ¯n of n modulo N, we have ρ(e)(Lp(D¯n)) ⊂ Lp(Dn+1) and ρ(e∗)(Lp(D¯n)) ⊂ Lp(Dn−1). By the argument of [12, pages 54 -- 55], upon cyclic permu- tation of the Dn if necessary, there exists ξ = N−1 Xm=0 ξm ∈ N−1 Mm=0 Lp(Dm) = Lp(X) with ξm = 0 for m ≤ N0 − 1 and for m ≥ N − N1, and such that kξk ≤ 1 and kρ(b)ξk > 1 − ǫ. For each γ ∈ P, put Dγ = Rγ(Xr(γ) ∩ D0) = Dγ ∩ Xγ. Because Q is nonsingular by hypothesis, and because we have assumed that ρ is nonde- generate, for each l ≥ 0 we have a decomposition (4.11). It follows from this that (9.12) Let Dm = aγ=m Dγ (0 ≤ m ≤ N − 1). W = P≤N−1 = a0≤l≤N−1 Pl. Lp OPERATOR ALGEBRAS ASSOCIATED WITH ORIENTED GRAPHS 19 It follows from (9.11) and (9.12) that X = aγ∈W Dγ. Hence we can write any η ∈ Lp(X) as a sum η = Xγ∈W ηγ (ηγ ∈ Lp(Dγ)). Next, by Lemma 9.6, for each v ∈ Q0 there is a measurable set Ev ⊂ Y of nonzero measure such that the family {S γ(Er(γ)) : γ ∈ W} is disjoint. Choose a norm-one element ζv ∈ Lp(Ev) for each v ∈ Q0. Let u : Lp(X) → Lp(X × Y), uη = Xγ∈W ρ(γ)ηγ ⊗ φ(γ)ζr(γ). One checks, as in the proof of [12, Proposition 8.6], that u is an isometry. Let ψ = 1 ⊗ φ : LQ → L(Lp(X × Y)), be as in Lemma 9.4. Observe that (9.13) kψ(b)k = kφ(b)k. A calculation similar to that of the proof of [12, Proposition 8.6] shows that for ξ as above, (9.14) uρ(b)ξ = ψ(b)uξ. It follows from (9.13) and (9.14) that (9.10) holds. This completes the proof. (cid:3) Proof of Theorem 9.1 Because LQ is simple by hypothesis, the C∗-algebra C∗(Q) is simple; thus every nonzero ∗-representation LQ → L(L2(X)) induces the same norm. But by Remark 4.5 every spatial representation is a ∗-representation, so the theorem is clear for p = 2. Assume p , 2. By Proposition 8.6 and Corollary 8.8, we may assume that Q is nonsingular and has no sources. By Lemma 9.4 and Propositions 9.5 and 9.9, every spatial seminorm is associated to a free spatial representation. Applying Proposition 9.9 again, we get that any two nonzero approximately free spatial representations lead to the same seminorm. (cid:3) 10. A simplicity theorem Theorem 10.1. Let p ∈ [1, ∞), p , 2. The following are equivalent for a countable graph Q. i) LQ is simple. ii) Every spatial nonzero Lp-representation of LQ is injective. ii') Every spatial nonzero representation LQ → L(ℓ p(N)) is injective. ii") Every nondegenerate spatial nonzero representation LQ → L(ℓ p(N)) is injective. iii) Every nondegenerate, contractive, nonzero Lp-representation of Op(Q) is injective. iii') Every nondegenerate, contractive, nonzero representation Op(Q) → L(ℓ p(N)) is injec- tive. If in addition we assume either that p , 1 or that Q0 is finite, then the above conditions are also equivalent to the following. iv) Every nonzero contractive homomorphism from Op(Q) to another Lp-operator algebra is injective. Proof. If either p , 1 or Q0 is finite, then iii) and iv) are equivalent, by Theorem 7.9. Let 2 , p ∈ [1, ∞). It follows from Theorems 7.8 and 9.1 that i)⇒iii). By Lemma 4.12 and Theorem 7.8, iii)⇒ii). Similarly, iii')⇒ii"). It is clear that ii)⇒ii')⇒ii") and that iii)⇒iii'). It remains to show that ii")⇒i). By [1, Theorem 2.9.1], LQ is simple if and only if Q0 is 20 GUILLERMO CORTI NAS AND MA. EUGENIA RODRIGUEZ the only nonempty hereditary and saturated subset of vertices, and every cycle in Q has an exit. We shall show that if any of these two conditions does not hold, then ii") does not hold either. So suppose there is a proper hereditary and saturated subset H ⊂ Q0. Let Q/H be the quotient graph of [1, Definition 2.4.11]. Then the natural map π : LQ → LQ/H is a nonzero surjection with nonzero kernel the ideal I(H) generated by H. Hence if ρ is an injective nondegenerate spatial representation LQ/H → L(ℓ p(N)) (which exists by Proposition 4.23) then ρπ is a nondegenerate nonzero spatial representation LQ → L(ℓ p(N)) which is not injective. So assume that Q0 is the only nonempty saturated and hereditary set of vertices, or equivalently, by [1, Lemma 2.9.6], that Q is cofinal in the sense of [1, Definitions 2.9.4] and that it has a cycle c without exits. Cofinality implies that c is the only cycle of Q modulo cycle rotation (by [1, Lemma 2.7.1 and Theorem 2.7.3]), and that sink(Q) = ∅ (by [1, Lemma 2.9.5]). Moreover, Q cannot have any infinite emmitters. For suppose v ∈ inf(Q); then v cannot be in any cycle, since any cycle containing v would have exits. In particular if e ∈ Q1 and s(e) = v then r(e) , v and by [1, Lemma 2.0.7] the hereditary and saturated closure of {r(e)} does not contain v, a contradiction. Hence Q = reg(Q), and therefore the space X of (4.14) consists of the infinite paths of Q. If s(c) = w, then any such path is of the form αc∞ for some finite path α ∈ P with r(α) = w. In particular X is countable and Xw = Xcn = {c∞} for all n ≥ 1. Hence for the counting measure µ, there is a spatial isometric isomorphism Lp(X, µ) (cid:27) ℓ p(N), and the nondegenerate representation ρµ of Example 4.13 maps c − c2 to zero, so it is not injective. This concludes the proof. (cid:3) Remark 10.2. By [9], an Lp-operator algebra may admit Banach algebra quotients which are not again Lp-operator algebras. Thus Phillips' theorem that the Lp-Cuntz algebra Op d is simple as a Banach algebra for 2 ≤ d < ∞ ([14, Theorem 5.14]) does not follow from Theorem 10.1 above. 11. Op(Q) vs. Op′ (Q) Let Rn be the countable graph with exactly one vertex and n loops, 1 ≤ n ≤ ∞. We write Ln = L(Rn), Op n = Op(Rn). In particular, L∞ = C{xi, x∗ i : 1 ≤ i}/hx∗ i x j − δi, ji. Lemma 11.1. Let Q be a countable graph and let 1 ≤ p < ∞. Assume that LQ is purely infinite simple. Then there is a homomorphism L∞ → LQ which induces an isometry Op ∞ → Op(Q). Proof. Let α be a cycle in Q and let v = s(α). Choose a closed path β with s(β) = v so that α and β are not comparable under the preorder of paths, as in the proof of Lemma 9.6. Then β∗α = α∗β = 0 and, of course, α∗α = β∗β = v. Hence there is a ∗-homomorphism φ : L∞ → LQ such that φ(xi) = βiα. Observe that if ρ : LQ → L(Lp(X)) is any spatial representation, then ρφ is again spatial. Hence φ induces a contractive homomorphism φ : Op ∞ → Op(Q). By Theorem 9.1, if ρ : LQ → L(Lp(X)) is a nonzero spatial representation, then φ agrees, up to isometric isomorphism, with the isometric inclusion ρφ(L∞) ⊂ ρ(LQ). (cid:3) Theorem 11.2. Let Q, Q′ be countable graphs and let 1 ≤ p , p′ < ∞ . Assume that LQ is purely infinite simple. If in addition, any of the following conditions holds, then there is no nonzero continuous homomorphism Op(Q) → Op′ i) LQ′ is simple. ii) p′ ≤ 2 and p < (p′, 2]. iii) p′ > 2 , p. (Q′). Lp OPERATOR ALGEBRAS ASSOCIATED WITH ORIENTED GRAPHS 21 Proof. Assume there is a nonzero continuous homomorphism f : Op(Q) → Op′ (Q′). Be- cause the inclusion LQ ⊂ Op(Q) is dense, f (LQ) , 0, which in view of the simplicity of LQ implies that f is injective on LQ. Let φ : L∞ → LQ be as in Lemma 11.1. Then f φ is injective, whence f φ : Op (Q′) is a nonzero continuous homomorphism. Hence there exists X ∈ {N, [0, 1]} and a spatial representation ρ′ : LQ′ → L(Lp′ (X)) such that ρ′ f φ : Op (X) con- tains a subspace isomorphic to ℓ p(N). If X = N, this cannot be, as noted in the proof of [12, Theorem 9.2], by [11, page 54]; if X = [0, 1] and either ii) or iii) holds, this cannot happen either, by [4, Theorem 6.4.19]. Thus parts ii) and iii) of the theorem are proved. Part i) also follows, using Proposition 4.23 and Theorem 9.1. (cid:3) (X)) is nonzero. By [12, Lemma 9.1] this implies that Lp′ ∞ → L(Lp′ ∞ → Op′ References [1] Gene Abrams, Pere Ara, and Mercedes Siles Molina, Leavitt path algebras, Lecture Notes in Math., vol. 2008, Springer, 2017. ↑4, 9, 17, 19, 20 [2] G. Abrams and G. Aranda Pino, The Leavitt path algebras of arbitrary graphs, Houston J. Math. 34 (2008), no. 2, 423 -- 442. MR2417402 ↑14 [3] Gene Abrams and Mark Tomforde, Isomorphism and Morita equivalence of graph algebras, Trans. Amer. Math. Soc. 363 (2011), no. 7, 3733 -- 3767. MR2775826 ↑10 [4] Fernando Albiac and Nigel J. Kalton, Topics in Banach space theory, Graduate Texts in Mathematics, vol. 233, Springer, New York, 2006. MR2192298 ↑21 [5] T. Ando, Contractive projections in Lp spaces, Pacific J. Math. 17 (1966), 391 -- 405. MR0192340 ↑1, 14 [6] Lisa Orloff Clark and Aidan Sims, Equivalent groupoids have Morita equivalent Steinberg algebras, J. Pure Appl. Algebra 219 (2015), no. 6, 2062 -- 2075. MR3299719 ↑8 [7] Ruy Exel, Inverse semigroups and combinatorial C∗-algebras, Bull. Braz. Math. Soc. (N.S.) 39 (2008), no. 2, 191 -- 313. MR2419901 ↑5, 9 [8] Eusebio Gardella and Martino Lupini, Representations of ´etale groupoids on Lp-spaces, Adv. Math. 318 (2017), 233 -- 278. MR3689741 ↑1, 3, 7, 10, 12, 14 [9] Eusebio Gardella and Hannes Thiel, Quotients of Banach algebras acting on Lp-spaces, Adv. Math. 296 (2016), 85 -- 92. MR3490763 ↑20 [10] , Extending representations arXiv:1703.00882v1. ↑1, 2, 5, 12, 14 of Banach algebras to their biduals, available at [11] Joram Lindenstrauss and Lior Tzafriri, Classical Banach spaces. I, Springer-Verlag, Berlin-New York, 1977. Sequence spaces; Ergebnisse der Mathematik und ihrer Grenzgebiete, Vol. 92. MR0500056 ↑21 [12] N. Christopher Phillips, Analogs of Cuntz algebras on Lp spaces, available at arXiv:1201.4196. ↑1, 2, 3, 6, 7, 11, 12, 16, 17, 18, 19, 21 [13] , Crossed products of Lp-operator algebras and the K-theory of Cuntz algebras on Lp-spaces, avail- able at arXiv:1309.6406. ↑12 , Simplicity of U HF and Cuntz algebras on Lp-spaces, available at arXiv:1309.0015. ↑20 [14] [15] Mar´ıa Eugenia Rodr´ıguez, ´Algebras de operadores a grafos asociadas http://cms.dm.uba.ar/academico/carreras/doctorado/tesis-doc_RODRIGUEZ,MARIA%20EUGENIA.pdf. ↑3 orientados, thesis, PhD en Buenos espacios Aires, Lp 2016, [16] Benjamin Steinberg, A groupoid approach to discrete inverse semigroup algebras, Adv. Math. 223 (2010), no. 2, 689 -- 727. MR2565546 ↑9 [17] , Simplicity, primitivity and semiprimitivity of ´etale groupoid algebras with applications to inverse semigroup algebras, J. Pure Appl. Algebra 220 (2016), no. 3, 1035 -- 1054. MR3414406 ↑5 Dept. Matem´atica-Inst. Santal´o, FCEyN, Universidad de Buenos Aires, Ciudad Universitaria, (1428) Buenos Aires, Argentina E-mail address: [email protected] Departamento de Ciencias Exactas, Ciclo B´asico Com ´un, Universidad de Buenos Aires, Ciudad Universi- taria, (1428) Buenos Aires, Argentina E-mail address: [email protected]
1105.2685	1	1105	2011-05-13T11:03:07	On the generalized quadratic mappings in quasi-Banach modules over a $C^*$--algebra	[ "math.OA", "math.FA" ]	Let $n>2$ be a positive integer. In this paper, we obtain the general solution of the following functional equation n \sum_{1 \le i<j \le n} Q(x_i-x_j)=\sum_{i=1}^{n}Q(\sum_{j =1}^n x_j -n x_i) which is derived from the centroid of the $n$ distinct vectors $x_1, ..., x_n$ in an inner product space. Furthermore, we prove that a mapping $f$ between quasi-Banach modules over a $C^*$-algebra satisfying approximately the equation can be approximated by a quadratic mapping $Q$ satisfying exactly the equation such that $\|f(x)-Q(x)\|$ is bounded.	math.OA	math	ON THE GENERALIZED QUADRATIC MAPPINGS IN QUASI-BANACH MODULES OVER A C ∗ -- ALGEBRA ∗ HARK-MAHN KIM AND DON O LEE † Abstract. Let n > 2 be a positive integer. In this paper, we obtain the general solution of the following functional equation n X1≤i<j≤n Q (xi − xj) = n Xi=1 xj − nxi  n Q Xj=1  which is derived from the centroid of the n distinct vectors x1, · · · , xn in an in- ner product space. Furthermore, we prove that a mapping f between quasi- Banach modules over a C ∗-algebra satisfying approximately the equation can be approximated by a quadratic mapping Q satisfying exactly the equation such that kf (x) − Q(x)k is bounded. 1. Introduction The stability problem of functional equations originated from a question of S.M. Ulam [25] concerning the stability of group homomorphisms: "When is it true that by slightly changing the hypotheses of a theorem one can still assert that the thesis of the theorem remains true or approximately true?" If the answer is affirmative, then we would say the equation of homomorphism H(x ∗ y) = H(x) ⋄ H(y) is stable. The concept of stability for a functional equation arises when we replace the functional equation by an inequality which acts as a perturbation of the equation. First, Ulam's question for approximately additive mappings was solved by D.H. Hyers [10]. In 1951, D.G. Bourgin [4] was the second author to treat the Ulam stability problem for additive mappings. Th.M. Rassias [18] succeeded in extending the result of Hyers' theorem by weakening the condition for the Cauchy difference to be unbounded. A number of mathematicians were attracted to this result of Th.M. Rassias and stimulated to investigate the stability problems of functional equations. Now, a square norm on an inner product space satisfies the important parallelo- gram equality kx + yk2 + kx − yk2 = 2(kxk2 + kyk2) for all vectors x, y. If △ABC is a triangle in a finite dimensional Euclidean space and I is the center of the side −→ CIk2) holds for BC, then the following identity k all vectors A, B and C. The following functional equation, which was motivated by −→ ACk2 = 2(k −→ ABk2 + k −→ AIk2 + k 1991 Mathematics Subject Classification. 39B82,46L05,30D05. Key words and phrases. Ulam stability problem, A-quadratic mapping, Unitary group, quasi- Banach modules, p-Banach modules. This Article was submitted in The Journal of Mathematical Analysis and Applications † Corresponding author:[email protected]. 1 2 H. KIM AND D. LEE these equations, (1.1) Q(x + y) + Q(x − y) = 2Q(x) + 2Q(y) is called a quadratic functional equation, and every solution of the equation (1.1) is said to be a quadratic mapping. A Hyers-Ulam stability problem for the quadratic functional equation (1.1) was first solved by F. Skof [23]. C. Borelli and G.L. Forti [3] generalized the stability result of the quadratic functional equation. The stability problems of several functional equations have been extensively investigated by a number of authors and there are many interesting results concerning this problem [1, 7, 9, 11, 19]. Furthermore, C. Park [17] have proved the Hyers-Ulam-Rassias stability problem for functional equations in Banach modules over a C ∗-algebra. Now, if △XY Z is a triangle in a finite dimensional Euclidean space and G := is the center of gravity of the triangle, then a simple direct calculation and X+Y +Z 3 the definition of the norm yields the following identity (1.2) −−→ XY k2 + k −→ Y Zk2 + k k −−→ XGk2 + k −−→ Y Gk2 + k Employing the above identity (1.2), we introduce the new functional equation, (1.3) 3Q(x − y) + 3Q(y − z) + 3Q(x − z) = Q(y + z − 2x) + Q(x + z − 2y) + Q(x + y − 2z) for a mapping Q : U → V and for all vectors x, y, z ∈ U, where U and V are linear spaces. More generally, let X1, X2, · · · , Xn (n ≥ 3) be distinct vectors in a finite dimensional Euclidean space E. Putting G := Pn , the centroid of the n distinct vectors, then we get the following identity by a simple direct calculation and the definition of the norm i=1 Xi n −−→ ZXk2 = 3(cid:16)k −→ ZGk2(cid:17) . −−−→ XiXjk2 = n k n X1≤i<j≤n which is equivalent to the equation (1.4) kXi − Xjk2 = n X1≤i<j≤n 2 n n k −−→ XiGk2 Xj=1 Xi=1 Xj − nXi(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi=1(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) xj − nxi! Q n Xj=1 Xi=1 n for any distinct vectors X1, X2, · · · , Xn. Employing the above equality (1.4), we introduce the new functional equation, (1.5) Q (xi − xj) = n X1≤i<j≤n for a mapping Q : U → V and for all vectors x1, · · · , xn ∈ U. We recall some basic facts concerning quasi-Banach spaces and some preliminary results. Definition 1.1. ([2, 20]) Let X be a linear space. A quasi-norm k·k is a real-valued function on X satisfying the following: (1) kxk ≥ 0 for all x ∈ X and kxk = 0 if and only if x = 0. ULAM STABILITY PROBLEM 3 (2) kλxk = λ · kxk for all λ ∈ R and all x ∈ X. (3) There is a constant K such that kx + yk ≤ K(kxk + kyk) for all x, y ∈ X. The smallest possible K is called the modulus of concavity of k · k. The pair (X, k · k) is called a quasi-normed space if k · k is a quasi-norm on X. A quasi- Banach space is a complete quasi-normed space. A quasi-norm k · k is called a p-norm (0 < p ≤ 1) if kx + ykp ≤ kxkp + kykp for all x, y ∈ X. In this case, a quasi-Banach space is called a p-Banach space. Clearly, p-norms are continuous, and in fact, if k · k is a p-norm on X, then the formula d(x, y) := kx − ykp defines an translation invariant metric for X and k · kp is a p-homogeneous F -norm. The Aoki -- Rolewicz theorem [2, 20] guarantees that each quasi-norm is equivalent to some p-norm for some 0 < p ≤ 1. Concerning the Ulam stability problem for functional equations, C. S´anchez [21] and J. Tabor [24] have investigated a version of the Hyers-Rassias-Gajda theorem (see [8, 18]) for approximate additive mappings in quasi-Banach spaces. In this paper, we are going to find the general solution of (1.5) for any fixed positive integer n ≥ 3 in the class of mappings between real vector spaces. Further- more, concerning the stability problem of Ulam for the functional equation (1.5) we are going to investigate the generalized Hyers-Ulam-Rassias stability problem for approximate mappings in quasi-Banach modules and p-Banach modules over a C ∗- algebra. Thus we generalize the stability results of the quadratic functional equation (1.5) in Banach spaces. 2. Solution of FE. (1.5) First of all, we find out the general solution of (1.3) in the class of mappings between real vector spaces. Lemma 2.1. Let U and V be real vector spaces. A mapping Q : U → V satisfies the functional equation (1.3) if and only if the mapping Q : U → V is quadratic. Proof. It is easy to see that the equation (1.1) implies the functional equation (1.3). Now let Q satisfy the equation (1.3). Putting y, z := 0 in (1.3) yields Q(2x) = 4Q(x) for all x ∈ U. By setting z := 0 in (1.3), we see (2.1) Q(x − 2y) + Q(2x − y) + Q(x + y) = 3Q(x − y) + 3Q(x) + 3Q(y) for all x, y ∈ U. In turn, substituting −y for y in (2.1) and then adding the resulting equation to (2.1), one obtains (2.2) Q(2x + y) + Q(2x − y) + Q(x + 2y) + Q(x − 2y) = 2Q(x − y) + 2Q(x + y) + 6Q(x) + 6Q(y) for any x, y ∈ U. Letting z := −y in (1.3), we obtain (2.3) Q(x + 3y) + Q(x − 3y) + 4Q(x) = 3Q(x − y) + 3Q(x + y) + 12Q(y) 4 H. KIM AND D. LEE for all x, y ∈ U. Replacing x by 2x in (2.3), we get (2.4) Q(2x + 3y) + Q(2x − 3y) + 16Q(x) = 3Q(2x − y) + 3Q(2x + y) + 12Q(y) for any x, y ∈ U. Now we substitute z := 2y in (1.3) to get (2.5) Q(x − 3y) + Q(2x − 3y) + Q(x) = 3Q(x − y) + 3Q(y) + 3Q(x − 2y) for any x, y ∈ U. Replacing y by −y in (2.5) and then adding (2.5) to the resulting expression, we obtain Q(x + 3y) + Q(x − 3y) + Q(2x + 3y) + Q(2x − 3y) + 2Q(x) = 3Q(x + y) + 3Q(x − y) + 3Q(x + 2y) + 3Q(x − 2y) + 6Q(y), which is rearranged in the following way by (2.4) (2.6) Q(x + 3y) + Q(x − 3y) + 3Q(2x + y) + 3Q(2x − y) + 6Q(y) = 3Q(x + y) + 3Q(x − y) + 3Q(x + 2y) + 3Q(x − 2y) + 14Q(x) for any x, y ∈ U. Now subtracting (2.3) from the equation (2.6) and then dividing it by 3, we have (2.7) Q(2x + y) + Q(2x − y) + 6Q(y) = Q(x + 2y) + Q(x − 2y) + 6Q(x) for any x, y ∈ U. Again we add (2.2) to (2.7) and then divide the resulting expression by 2 to obtain (2.8) Q(2x + y) + Q(2x − y) = Q(x + y) + Q(x − y) + 6Q(x), which is equivalent to the original quadratic functional equation Q(x + y) + Q(x − y) = 2Q(x) + 2Q(y) for any x, y ∈ U [5, Theorem 2.1]. (cid:3) Lemma 2.2. Assume that a mapping Q : U → V satisfies the functional equation (1.5). Then Q is even and (2.9) Q((n − 1)kx) = (n − 1)2kQ(x) for any vector x ∈ U. Proof. By setting xi := 0 for all i = 1, · · · , n in the equation (1.5), we see Q(0) = 0. Putting x1 = x and xi := 0 for all i = 2, · · · , n in (1.5), we get Q(−(n − 1)x) = (n − 1)2Q(x) for all x ∈ U. Substituting xi := x for all i = 1, · · · , n − 1 and xn := 0 in (1.5), one obtains n(n − 1)Q(x) = (n − 1)Q(−x) + Q((n − 1)x) = n(n − 1)Q(−x), which shows that Q is even, and hence Q((n − 1)x) = (n − 1)2Q(x) for all x ∈ U. Therefore we get the desired conclusion by induction on k. (cid:3) To find the general solution of (1.5), we need to prove the following lemma above all. ULAM STABILITY PROBLEM 5 Lemma 2.3. Let U and V be real vector spaces. For each integer a with a 6= 1, a mapping Q : U → V satisfies the functional equation (2.10) Q(ax + y) + Q(x + ay) + (a − 1)Q(x − y) = (a + 1)Q(x + y) + (a2 − 1)[Q(x) + Q(y)] for all x, y ∈ U if and only if a mapping Q : U → V is quadratic. Proof. Let Q satisfy the equation (2.10). It follows easily that Q is even, Q(ax) = a2Q(x) and Q(0) = 0. For a = 0, the equation (2.10) reduces to the equation (1.1). For any negative integer a < −1, by considering a as −a and applying the evenness of Q, we need to prove the lemma for the case a > 1 without loss of generality. Now we claim that if Q satisfies the equation (2.10), then Q also satisfies (1.1) by induction on positive integers a > 1. For a = 2, the equation (2.10) reduces to (2.11) Q(2x + y) + Q(x + 2y) + Q(x − y) = 3Q(x + y) + 3Q(x) + 3Q(y), which is exactly the equation (1.3), and hence it is equivalent to (1.1) by Lemma 2.1. Assume that the equation (2.10) implies the equation (1.1) for all a with a := 2, · · · , a. We are to show that if Q satisfies the equation (2.10) for a + 1, then Q is quadratic in the sequel. Letting y := x + y in (2.10), we obtain (2.12) Q((a + 1)x + y) + Q((a + 1)x + ay) + (a − 1)Q(y) = (a + 1)Q(2x + y) + (a2 − 1)[Q(x) + Q(x + y)] for all x, y ∈ U. Interchanging x with y in (2.12) and after that adding it to (2.12), we have (2.13) Q((a + 1)x + y) + Q(x + (a + 1)y) + Q((a + 1)x + ay) + Q(ax + (a + 1)y) = (2a2 + 3a + 1)Q(x + y) + (a2 + 2a + 3)[Q(x) + Q(y)] − (a + 1)Q(x − y) for all x, y ∈ U. Letting y := −x + y in (2.10), we obtain (2.14) Q((a − 1)x + y) + Q((a − 1)x − ay) + (a − 1)Q(2x − y) = (a + 1)Q(y) + (a2 − 1)[Q(x) + Q(x − y)] for all x, y ∈ U. Exchanging x and y in (2.14) and after that adding the resulting equation and (2.14), one has by induction (2.15) Q((a − 1)x + ay) + Q(ax + (a − 1)y) + Q(x − y) = (2a2 − 2a − 1)Q(x + y) + 3[Q(x) + Q(y)] for all x, y ∈ U. We observe from these inequalities that Q(λx) = λ2Q(x) for λ := a + 1, a − 1, 2, 2a − 1, and for all x ∈ U. Replacing y by ay in (2.15) and switching x with y in the resulting equation, and then adding two equations side by side, we obtain by inductive assumption that (2.16) Q((a − 1)x + a2y) + Q(a2x + (a − 1)y) + (a3 − 2a2 + 2a + 2)Q(x − y) = (a3 − 2a − 2)Q(x + y) + (a4 − a2 + 2a + 5)[Q(x) + Q(y)] 6 H. KIM AND D. LEE for all x, y ∈ U. Now we substitute y := ay − x in (2.10) to get (2.17) Q((a − 1)x + ay) + Q((a − 1)x − a2y) + (a − 1)Q(2x − ay) = a2(a + 1)Q(y) + (a2 − 1)[Q(x) + Q(x − ay)] for all x, y ∈ U. Switching x with y in (2.17), and then adding two equations side by side, we obtain by virtue of (2.15), (2.16) and (2.10) that (2.18) (a − 1)[Q(2x − ay) + Q(ax − 2y)] + (3a2 − 5a − 2)Q(x + y) = (a3 + a2 − 2a − 8)[Q(x) + Q(y)] + (a2 + a + 2)Q(x − y) holds for all x, y ∈ U. Note from (2.18) that Q(λx) = λ2Q(x) for λ := a − 2, a + 2, and for all x ∈ U Now substituting x for 2x in (2.10) yields (2.19) Q(2ax + y) + Q(2x + ay) + (a − 1)Q(2x − y) = (a + 1)Q(2x + y) + (a2 − 1)[4Q(x) + Q(y)] for all x, y ∈ U. Exchanging x and y in (2.19) and then adding two equations side by side, one obtains by (2.11), (2.18) (2.20) (a − 1)[Q(2ax + y) + Q(x + 2ay)] + (a2 − a + 4)Q(x − y) = (4a3 − 6a2 + 3a + 7)[Q(x) + Q(y)] + (3a2 − 3a − 4)Q(x + y) for all x, y ∈ U. We remark that Q(λx) = λ2Q(x) for λ := 2a − 1, 2a + 1, and for all x ∈ U Now, let's transform by variables like as x := 2ax + y and y := x + 2ay in (2.15). Then we can rewrite the equation (2.15) in the form (2.21) (2a − 1)2[Q((a + 1)x + ay) + Q(ax + (a + 1)y)] + (2a − 1)2Q(x − y) = (2a2 − 2a − 1)(2a + 1)2Q(x + y) + 3[Q(2ax + y) + Q(x + 2ay)] for all x, y ∈ U. Multiplying both sides of (2.21) by (a − 1) and applying (2.20) to the resulting expression, we get (2.22) (a − 1)(2a − 1)2[Q((a + 1)x + ay) + Q(ax + (a + 1)y)] +(a − 1)(2a − 1)2Q(x − y) = (8a5 − 8a4 − 10a3 + 13a2 − 4a − 11)Q(x + y) +3(4a3 − 6a2 + 3a + 7)[Q(x) + Q(y)] − (4a3 − 5a2 + 2a + 11)Q(x − y) for all x, y ∈ U. Multiplying (a − 1)(2a − 1)2 on both sides of (2.13) and applying (2.22) to the resulting expression, we get finally (a − 1)(2a − 1)2[Q((a + 1)x + y) + Q(x + (a + 1)y)] = (4a4 − 8a2 + 6a + 10)Q(x + y) − (4a4 − 8a3 + 2a2 + 2a − 12)Q(x − y) +(4a5 − 11a3 + 3a2 + 4a − 24)[Q(x) + Q(y)], ULAM STABILITY PROBLEM 7 which can be written in the form (2.23) (a − 1)(2a − 1)2hQ((a + 1)x + y) + Q(x + (a + 1)y) + aQ(x − y)i = (a − 1)(2a − 1)2h(a + 2)Q(x + y) + ((a + 1)2 − 1)[Q(x) + Q(y)]i +(3a2 − 3a + 12)[Q(x + y) + Q(x − y) − 2Q(x) − 2Q(y)] for all x, y ∈ U. Since Q satisfies the equation (2.10) for a + 1, the last equation reduces to Q(x + y) + Q(x − y) − 2Q(x) − 2Q(y) = 0. Consequently, we have proved that if Q satisfies the equation (2.10) for a + 1, then Q satisfies the equation (1.1). Therefore, by induction argument the equation (2.10) implies the equation (1.1) for each positive integer a > 1. Conversely, it is obvious that the equation (1.1) implies the functional equation (cid:3) (2.10). This completes the proof. Theorem 2.4. Let U and V be real vector spaces. A mapping Q : U → V satisfies the functional equation (1.5) for each positive integer n > 2 if and only if a mapping Q : U → V satisfies the functional equation (1.1). Thus there exists a symmetric biadditive mapping B : U × U → V such that Q(x) = B(x, x) for all x ∈ U. Proof. It is easy to see that the equation (1.1) implies the functional equation (1.5). Conversely, let Q satisfy the equation (1.5). Putting x1 := x, x2 := y and xi := 0 for all i = 3, · · · , n in (1.5), we get (2.24) Q(x − ay) + Q(ax − y) + (a − 1)Q(x + y) = (a + 1)Q(x − y) + (a2 − 1)[Q(x) + Q(y)] for all x, y ∈ U, where a := n − 1 is a positive integer with a ≥ 2. By the previous Lemma 2.3, the mapping Q : U → V satisfies the functional equation (1.1). (cid:3) The following result is interesting and useful characterization formulas for an inner product space among normed linear spaces. Corollary 2.5. Let U be a normed linear space. Then the following statements are equivalent: (a) U is an inner product space. (b) The norm in U satisfies the condition: kax + yk2 + kx + ayk2 + (a − 1)kx − yk2 = (a + 1)kx + yk2 + (a2 − 1)(kxk2 + kyk2) for all x, y ∈ U and for some fixed integer a with a 6= 1. (c) kxi − xjk2 = n X1≤i<j≤n for all xi(i = 1, · · · , n) ∈ U and a fixed n > 2. n Xi=1(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) n Xj=1 2 xj − nxi(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 8 H. KIM AND D. LEE Proof. The proof is obvious by Lemma 2.3 and Theorem 2.4. The inner product is defined as usual by (x, y) = 1/4(cid:0)kx + yk2 − kx − yk2 + ikx + iyk2 − ikx − iyk2(cid:1) , and (x, y) = 1/4(cid:0)kx + yk2 − kx − yk2(cid:1) for the complex and real spaces, respectively. (cid:3) 3. Stability of FE. (1.5) in quasi-Banach modules Now let A be a complex ∗-algebra with unit and let M be a left A-module. Let us call a mapping Q : M → A an A-quadratic mapping if both relations Q(ax) = aQ(x)a∗ and Q(x + y) + Q(x − y) = 2Q(x) + 2Q(y) are fulfilled [26]. A mapping Q : M → A is called a generalized A-quadratic mapping if Q(ax) = aQ(x)a∗ for all x ∈ M, and the following identity holds: aixi! + X1≤i<j≤n aiajQ(xi − xj) = n Xi=1 ai!" n Xi=1 aiQ(xi)# Q n Xi=1 such that Pn for all xi ∈ M, some fixed ai in R (i = 1, · · · , n) and at least two of them are nonzero i=1 ai 6= 0, and a fixed n ≥ 2 [15]. It was shown that the notion of A- quadratic mapping is equivalent to the notion of generalized A-quadratic mapping if all spaces are over the complex number field and a mapping B : M × M → A is defined in terms of the mapping Q as (3.1) B(x, y) = 1 4 [Q(x + y) − Q(x − y) + iQ(x + iy) − iQ(x − iy)] for all x, y in M [15]. It was indicated in [26] that if the relation (3.1) holds and Q is an A-quadratic form, then B is an A-sesquilinear form and Q(x) = B(x, x), and vice versa. Now it follows easily from Theorem 2.4 that a mapping Q is a generalized A-quadratic mapping if and only if Q(ax) = aQ(x)a∗, n X1≤i<j≤n Q(xi − xj) = n Xi=1 Q n Xj=1 xj − nxi! for all x and (x1, · · · , xn), where n ≥ 3. Now we are ready to investigate the generalized Hyers-Ulam-Rassias stability problem for approximate A-quadratic mappings acting on U(A) of the equation (1.5) in quasi-Banach modules over a C ∗-algebra. Let M1 and M2 be quasi-Banach A-bimodules and let K ≥ 1 be the modulus of concavity of k · k throughout this section unless we give any specific reference. Given a mapping f : M1 → M2, we ULAM STABILITY PROBLEM 9 define a difference Duf : M n 1 → M2 of the equation (1.5) as Duf (x1, · · · , xn) := n X1≤i<j≤n f (uxi − uxj) − n Xi=1 uf n Xj=1 xj − nxi! u∗, for all xi ∈ M1 and u ∈ U(A), which is called the approximate remainder of the functional equation (1.5) and acts as a perturbation of the equation. Theorem 3.1. Assume that there exists a mapping ϕ : M n which a mapping f : M1 → M2 satisfies the functional inequality 1 → [0, ∞) := R+ for (3.2) for all (x1, · · · , xn) ∈ M n (3.3) kDuf (x1, · · · , xn)k ≤ ϕ(x1, · · · , xn) 1 and for all u ∈ U(A), and the following series K iϕ((n − 1)ix1, · · · , (n − 1)ixn) < ∞ (n − 1)2i ∞ Xi=0 for all (x1, · · · , xn) ∈ M n 1 . If either f is measurable or f (tx) is continuous in t ∈ R for each fixed x ∈ M1, then there exists a unique generalized A-quadratic mapping Q : M1 → M2 which satisfies the equation (1.5) and the inequality (3.4) f (x) + (n − 1)f (0) for all x ∈ M1, where (cid:13)(cid:13)(cid:13)(cid:13) Φ(x) ϕi(x) and ϕ(x) K (n − 1)2 K iΦ((n − 1)ix) (n − 1)2i ∞ Xi=0 (n2 + 1) − (i + 1)n n ϕ(x)(cid:27) , , 0, · · · , 0), (i = 1, · · · , n), 2 ≤ − Q(x)(cid:13)(cid:13)(cid:13)(cid:13) 1≤i≤n(cid:26)ϕi(−x) + {z} 1≤i≤n−1 := ϕ(0, · · · , 0, x i−th := min := min {ϕi(x) + ϕi+1(x)} for all x ∈ M1. The mapping Q is defined by Q(x) = lim k→∞ f ((n − 1)kx) (n − 1)2k for all x ∈ M1. Proof. Put u := 1 ∈ U(A) in (3.2). Then for each i = 1, · · · , n − 1, interchanging xi for x and xj for 0 for all j 6= i in (3.2) and then comparing the sequent inequalities, we get nkf (x) − f (−x)k ≤ ϕi(x) + ϕi+1(x) for all x ∈ M1 and for all i = 1, · · · , n − 1. Thus one obtains the approximate even condition of f (3.5) kf (x) − f (−x)k ≤ 1 n ϕ(x), ϕ(x) := min 1≤i≤n−1 {ϕi(x) + ϕi+1(x)} 10 H. KIM AND D. LEE for all x ∈ M1. For each i = 1, · · · , n, replacing xi by −x and xj by 0 for all j 6= i we observe that (cid:13)(cid:13)(cid:13)(cid:13) (i − 1)nf (x) + [n2 − (i + 1)n + 1]f (−x) + n(cid:18)n − 1 ≤ ϕi(−x) 2 (cid:19)f (0) − f ((n − 1)x)(cid:13)(cid:13)(cid:13)(cid:13) for all x ∈ M1. Associating the last inequality with (3.5), we obtain (cid:13)(cid:13)(cid:13)(cid:13) (n − 1)2f (x) + n(cid:18)n − 1 2 (cid:19)f (0) − f ((n − 1)x)(cid:13)(cid:13)(cid:13)(cid:13) (n2 + 1) − (i + 1)n ≤ ϕi(−x) + ϕ(x) n for all x ∈ M1 and for all i = 1, · · · , n. Hence one has the following inequality for all x ∈ M1. Define a mapping g : M1 → M2 by g(x) := f (x) + (n−1)f (0) x ∈ M1. Then it follows from (3.6) that 2 for all ≤ Φ(x) (cid:13)(cid:13)(cid:13)(cid:13) (n − 1)2f (x) + n(cid:18)n − 1 2 (cid:19)f (0) − f ((n − 1)x)(cid:13)(cid:13)(cid:13)(cid:13) (n − 1)2 (cid:13)(cid:13)(cid:13)(cid:13) (n − 1)2 Φ(x) ≤ 1 g((n − 1)x) for all x ∈ M1, from which we obtain by applying a standard procedure of the induction argument on m that (3.6) (3.7) g(x) − (cid:13)(cid:13)(cid:13)(cid:13) (n − 1)2m (cid:13)(cid:13)(cid:13)(cid:13) g((n − 1)mx) (3.8) g(x) − (cid:13)(cid:13)(cid:13)(cid:13) ≤ + K m−2 (n − 1)2 (n − 1)2(cid:19)i Xi=0 (cid:18) K (n − 1)2(cid:19)m−1 (n − 1)2(cid:18) K 1 Φ((n − 1)ix) Φ((n − 1)m−1x) for all x ∈ M1 and all m ≥ 1, which is considered to be (3.7) for m = 1. In fact, we figure out by the inequality (3.7), g(x) − (cid:13)(cid:13)(cid:13)(cid:13) ≤ K(cid:13)(cid:13)(cid:13)(cid:13) ≤ K g(x) − g((n − 1)m+1x) g((n − 1)x) (n − 1)2(m+1) (cid:13)(cid:13)(cid:13)(cid:13) + K(cid:13)(cid:13)(cid:13)(cid:13) (n − 1)2 (cid:13)(cid:13)(cid:13)(cid:13) (n − 1)2(cid:13)(cid:13)(cid:13)(cid:13) K (n − 1)2 Φ(x) + g((n − 1)x) − g((n − 1)x) (n − 1)2 − g((n − 1)m+1x) g((n − 1)m+1x) (n − 1)2(m+1) (cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13) , (n − 1)2m ULAM STABILITY PROBLEM 11 which, in accordance with inductive assumption, yields (3.8) for m + 1. Thus one obtains that for all nonnegative integers m, l with m > l g((n − 1)m−l · (n − 1)lx) (n − 1)2(m−l) (cid:13)(cid:13)(cid:13)(cid:13) K iΦ((n − 1)l+ix) (n − 1)2i + 1 (n − 1)2l+2 K m−l−1Φ((n − 1)m−1x) (n − 1)2(m−l−1) K iΦ((n − 1)ix) (n − 1)2i + 1 K m−1Φ((n − 1)m−1x) K l(n − 1)2 (n − 1)2(m−1) , (3.9) g((n − 1)lx) (n − 1)2l − g((n − 1)mx) (n − 1)2m (cid:13)(cid:13)(cid:13)(cid:13) g((n − 1)lx) − (cid:13)(cid:13)(cid:13)(cid:13) = ≤ ≤ 1 (n − 1)2l(cid:13)(cid:13)(cid:13)(cid:13) (n − 1)2l+2 K K K l(n − 1)2 m−l−2 m−2 Xi=0 Xi=l which tends to zero by (3.3) as l → ∞. Hence the sequence n g((n−1)mx) is a Cauchy sequence for any x ∈ M1, and so it converges by the completeness of M2. Therefore we can define a mapping Q : M1 → M2 by (n−1)2m om∈N Q(x) = lim m→∞ g((n − 1)mx) (n − 1)2m = lim m→∞ f ((n − 1)mx) (n − 1)2m for all x ∈ M1. Taking the limit as m → ∞ in (3.8), we obtain the desired inequality (3.4). Exchanging (x1, · · · , xn) for ((n−1)mx1, · · · , (n−1)mxn) in (3.2) and dividing both sides by (n − 1)2m, we have (3.10) kD1Q(x1, · · · , xn)k 1 (n − 1)2m kDf ((n − 1)mx1, · · · , (n − 1)mxn)k (n − 1)2m ϕ((n − 1)mx1, · · · , (n − 1)mxn) K m = lim m→∞ ≤ lim m→∞ = 0. Therefore the mapping Q satisfies the equation (1.5) and hence Q is quadratic. To prove the uniqueness, let Q′ be another quadratic mapping satisfying (3.4). Then we get by Lemma 2.2 that Q′((n − 1)mx) = (n − 1)2mQ′(x) for all x ∈ M1 and all m ∈ N. Thus we have kQ(x) − Q′(x)k ≤ ≤ 1 (n − 1)2mnK(cid:13)(cid:13)(cid:13)Q((n − 1)mx) − f ((n − 1)mx) − (n − 1)f (0) (n − 1)f (0) 2 (cid:13)(cid:13)(cid:13) 2K 2 K m(n − 1)2 ∞ +K(cid:13)(cid:13)(cid:13)f ((n − 1)mx) + Xi=0 K m+iΦ((n − 1)k+ix) (n − 1)2(m+i) 2 − Q′((n − 1)mx)(cid:13)(cid:13)(cid:13)o for all x ∈ M1. Taking the limit as m → ∞, then we conclude that Q(x) = Q′(x) for all x ∈ M1. 12 H. KIM AND D. LEE Finally, we show that the quadratic mapping Q is A-quadratic. Under the as- sumption that either f is measurable or f (tx) is continuous in t ∈ R for each fixed x ∈ M1, the quadratic mapping Q satisfies Q(tx) = t2Q(x) for all x ∈ M1 and all t ∈ R by the same reasoning as the proof of [6]. That is, Q is R-quadratic. Putting x1 := −(n − 1)mx and xi := 0 for all i = 2, · · · , n in (3.2) and dividing the resulting inequality by (n − 1)2m, 1 (n − 1)2m(cid:13)(cid:13)(cid:13)n(n − 1)f (−(n − 1)mux) + n(cid:18)n − 1 2 (cid:19)f (0) −uf ((n − 1)m+1x)u∗ − (n − 1)uf (−(n − 1)mx)u∗(cid:13)(cid:13)(cid:13) (n − 1)2m ϕ(−(n − 1)mx, 0, · · · , 0). K m ≤ Taking the limit as m → ∞ and using the evenness of Q, we see that Q(ux) = uQ(x)u∗ for all x ∈ M1 and for each u ∈ U(A). The last relation is also true for u = 0. Now let a be a nonzero element in A and L a positive integer greater than 4a. Then we have a 3 . By [14, Theorem 1], there exist three elements u1, u2, u3 ∈ U(A) such that 3 a L = u1 + u2 + u3. Thus we calculate in conjunction with [13, Lemma 2.1] that 4 < 1 − 2 L < 1 3 Q(ax) = Q(cid:18)L 3(cid:19)2 = (cid:18)L = (cid:18)L 3(cid:19)2 3(cid:19)2 = (cid:18)L 3 a L 3(cid:19)2 x(cid:19) =(cid:18) L Q(u1x + u2x + u3x) B(u1x + u2x + u3x, u1x + u2x + u3x) (u1 + u2 + u3)B(x, x)(u∗ 1 + u∗ 2 + u∗ 3) 3 a L Q(x)3 a∗ L = aQ(x)a∗ for all a ∈ A(a 6= 0) and for all x ∈ M1. So the unique R-quadratic mapping Q is also generalized A-quadratic, as desired. This completes the proof. (cid:3) Theorem 3.2. Assume that the approximate remainder Duf of a mapping f : M1 → M2 satisfies the functional inequality (3.2) for all (x1, · · · , xn) ∈ M n 1 and all u ∈ U(A), and that the following series (3.11) ∞ Xi=1 K i(n − 1)2iϕ(cid:18) x1 (n − 1)i , · · · , xn (n − 1)i(cid:19) converges for all (x1, · · · , xn) ∈ M n 1 . If either f is measurable or f (tx) is continuous in t ∈ R for each fixed x ∈ M1, then there exists a unique generalized A-quadratic mapping Q : M1 → M2 which satisfies the equation (1.5) and the inequality (3.12) kf (x) − Q(x)k ≤ 1 (n − 1)2 ∞ Xi=1 K i(n − 1)2iΦ(cid:18) x (n − 1)i(cid:19) ULAM STABILITY PROBLEM 13 for all x ∈ M1, where Φ is defined as in Theorem 3.1. The mapping Q is defined by Q(x) = lim m→∞ (n − 1)2mf(cid:16) x (n − 1)m(cid:17) for all x ∈ M1. Proof. We use the same notations as those of Theorem 3.1. We observe that ϕ(0, · · · , 0) = 0 by the convergence (3.11), and thus we have f (0) = 0 by setting xi := 0 in (3.2) for all i = 1, · · · , n. Now we get by (3.7) which yields by induction x x (3.13) ≤ m−1 1 x ∈ M1, ≤ Φ(cid:18) x n − 1(cid:19) , n − 1(cid:19)(cid:13)(cid:13)(cid:13)(cid:13) (cid:13)(cid:13)(cid:13)(cid:13) f (x) − (n − 1)2f(cid:18) x (n − 1)m(cid:19)(cid:13)(cid:13)(cid:13)(cid:13) f (x) − (n − 1)2mf(cid:18) (n − 1)i(cid:19) + K i(n − 1)2iΦ(cid:18) (cid:13)(cid:13)(cid:13)(cid:13) It follows by (3.11) that the sequence n(n − 1)2mf(cid:16) x (n − 1)m(cid:19) , x ∈ M1. (n − 1)2mf(cid:18) Q(x) = lim m→∞ Xi=1 (n − 1)2 K m(n − 1)2m K(n − 1)2 Φ(cid:18) (n−1)m(cid:17)om∈N is a Cauchy sequence for any x ∈ M1. Since M2 is complete, we may define a mapping Q : M1 → M2 by for all x ∈ M1 and all integers m > 1. x (n − 1)m(cid:19) x n Xi=1 The rest of the proof goes through by the same way as that of Theorem 3.1. This (cid:3) completes the proof. From the main Theorem 3.1 and Theorem 3.2 we obtain the following corollary concerning the stability of the equation (1.5). Corollary 3.3. Let r, ε be positive real numbers with r − 2 < − logn−1 K or r − 2 > logn−1 K. Assume that a mapping f : M1 → M2 satisfies the inequality (3.14) kDuf (x1, · · · , xn)k ≤ ε kxikr for all xi ∈ M1 and for all u ∈ U(A). Then there exists a unique generalized A- quadratic mapping Q : M1 → M2 which satisfies the equation (1.5) and the inequality kf (x) − Q(x)k ≤( (n+2)Kεkxkr n[(n−1)2−K(n−1)r] , n[(n−1)r−K(n−1)2] , (n+2)Kεkxkr if if r − 2 < − logn−1 K r − 2 > logn−1 K ) for all x ∈ M1. Proof. Define ϕ(x1, · · · , xn) := ε(kx1kr + · · · + kxnkr) for all (x1, · · · , xn) ∈ M n 1 . Then we have the conditions ϕ(x) := 2εkxkp and Φ(x) := εkxkp + 2 n εkxkp for all x ∈ M1. Applying Theorem 3.1 and Theorem 3.2, we obtain the desired results 14 H. KIM AND D. LEE according to the cases of r. Exchanging xi for 0 in (3.14) for all i = 1, · · · , n yields f (0) = 0. (cid:3) Problem 3.4. It is an open problem to investigate the stability problem of Ulam for the case of K and r with − logn−1 K ≤ r − 2 ≤ logn−1 K in Corollary 3.3. Corollary 3.5. Assume that there exists a nonnegative number θ for which a map- ping f : M1 → M2 satisfies the inequality kDuf (x1, · · · , xn)k ≤ θ for all xi ∈ M1 and for all u ∈ U(A). Then there exists a unique generalized A- quadratic mapping Q : M1 → M2 which satisfies the equation (1.5) and the inequality f (x) + (n − 1)f (0) 2 for all x ∈ M1. (cid:13)(cid:13)(cid:13)(cid:13) ≤ (n + 2)Kθ n[(n − 1)2 − K] , if K < (n − 1)2 − Q(x)(cid:13)(cid:13)(cid:13)(cid:13) Problem 3.6. If K is so large a constant that (n−1)2 ≤ K, then we can't guarantee that the functional equation (1.5) is stable on concerning the Ulam stability problem. So, it is interesting to investigate the stability problem of Ulam for the case of n, K with (n − 1)2 ≤ K in Corollary 3.5. 4. Stability of FE. (1.5) in p-Banach modules We now prove the Hyers -- Ulam -- Rassias stability of the functional equation (1.5) in p-Banach A-bimodules. Theorem 4.1. Let M1 and M2 be p-Banach A-bimodules. Assume that the approx- imate remainder Duf of a mapping f : M1 → M2 satisfies the functional inequality (3.2) for all (x1, · · · , xn) ∈ M n 1 and all u ∈ U(A), and that the following series ϕ((n − 1)ix1, · · · , (n − 1)ixn)p (n − 1)2ip < ∞ ∞ Xi=0 for all (x1, · · · , xn) ∈ M n 1 . If either f is measurable or f (tx) is continuous in t ∈ R for each fixed x ∈ M1, then there exists a unique generalized A-quadratic mapping Q : M1 → M2 which satisfies the equation (1.5) and the inequality for all x ∈ M1, where Q and Φ are defined as in Theorem 3.1. Proof. It follows by the inequality (3.7) and the definition of p-norm that ≤ 1 (n − 1)2" ∞ Xi=0 Φ((n − 1)ix)p (n − 1)2ip #1/p ≤ 1 (n − 1)2p 1 (n − 1)2pi Φ((n − 1)ix)p, (n − 1)f (0) f (x) + (cid:13)(cid:13)(cid:13)(cid:13) 2 − Q(x)(cid:13)(cid:13)(cid:13)(cid:13) (n − 1)2(i+1) (cid:13)(cid:13)(cid:13)(cid:13) g((n − 1)i+1x) p g((n − 1)ix) (n − 1)2i − (cid:13)(cid:13)(cid:13)(cid:13) ULAM STABILITY PROBLEM 15 and so g((n − 1)lx) (n − 1)2l − (cid:13)(cid:13)(cid:13)(cid:13) p g((n − 1)mx) (n − 1)2m (cid:13)(cid:13)(cid:13)(cid:13) ≤ ≤ m−1 Xi=l (cid:13)(cid:13)(cid:13)(cid:13) 1 g((n − 1)ix) (n − 1)2i − Xi=l m−1 1 p g((n − 1)i+1x) (n − 1)2(i+1) (cid:13)(cid:13)(cid:13)(cid:13) (n − 1)2p (n − 1)2pi Φ((n − 1)ix)p Φ((n−1)ix)p for all x ∈ M1 and all integers l, m with m > l ≥ 0. Note that the series converges for all x ∈ M1. Thus we obtain the desired results us- (cid:3) ing the similar argument to Theorem 3.1. (n−1)2ip i=0 P∞ Theorem 4.2. Let M1 and M2 be p-Banach A-bimodules. Assume that the approx- imate remainder Duf of a mapping f : M1 → M2 satisfies the functional inequality (3.2) for all (x1, · · · , xn) ∈ M n 1 and all u ∈ U(A), and that the following series ∞ Xi=1 (n − 1)2ipϕ(cid:18) x1 (n − 1)i , · · · , xn (n − 1)i(cid:19)p < ∞ for all (x1, · · · , xn) ∈ M n 1 . If either f is measurable or f (tx) is continuous in t ∈ R for each fixed x ∈ M1, then there exists a unique generalized A-quadratic mapping Q : M1 → M2 which satisfies the equation (1.5) and the inequality kf (x) − Q(x)k ≤ 1 (n − 1)2" ∞ Xi=1 (n − 1)2ipΦ(cid:18) x (n − 1)i(cid:19)p#1/p for all x ∈ M1, where Q and Φ are defined as in Theorem 3.2. Corollary 4.3. Let M1 and M2 be p-Banach A-bimodules. Let r, ε be positive real numbers with r 6= 2. Assume that a mapping f : M1 → M2 satisfies the inequality kDuf (x1, · · · , xn)k ≤ ε kxikr n Xi=1 for all xi ∈ M1 and for all u ∈ U(A). Then there exists a unique generalized A- quadratic mapping Q : M1 → M2 which satisfies the equation (1.5) and the inequality kf (x) − Q(x)k ≤( (n+2)εkxkr n[(n−1)2p−(n−1)rp]1/p , n[(n−1)rp−(n−1)2p]1/p , (n+2)εkxkr if if r < 2 r > 2 ) for all x ∈ M1. Remark 4.4. The result for the case K = 1 in Theorem 3.1 (Theorem 3.2, Corollary 3.3, respectively) is the same as the result for the case p = 1 in Theorem 4.1 (Theorem 4.2, Corollary 4.3, respectively). Let M1 and M2 be Banach left A-modules and let a := aa∗, a∗a, or aa∗+a∗a for each a ∈ A. A mapping Q : M1 → M2 is called Asa-quadratic if Q(x + y) + Q(x − y) = 2Q(x) + 2Q(y) and Q(ax) = aQ(x) for all a ∈ A and all x, y ∈ M1 [16]. Since two Banach spaces E1 and E2 are considered as Banach modules over A := C, the Asa-quadratic mapping Q : E1 → E2 implies Q(ax) = a2Q(x) for all a ∈ C. 2 16 H. KIM AND D. LEE Theorem 4.5. Let M1 and M2 be quasi-Banach A-bimodules. Assume that there exists a mapping ϕ : M n 1 → R+ for which a mapping f : M1 → M2 satisfies the functional inequality f (uxi − uxj) − n X1≤i<j≤n (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) n Xi=1 uf n Xj=1 xj − nxi!(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ≤ ϕ(x1, · · · , xn), ∀xi ∈ M1, ∀u ∈ A(u = 1), and the series (3.3) converges for all xi ∈ M1, i = 1, · · · , n. If either f is measurable or f (tx) is continuous in t ∈ R for each fixed x ∈ M1, then there exists a unique f ((n−1)mx) Asa-quadratic mapping Q : M1 → M2, defined by Q(x) = lim (n−1)2m , which m→∞ satisfies the equation (1.5) and the inequality (3.4) for all x ∈ M1. Proof. By the same reasoning as the proof of Theorem 3.1, it follows from u = 1 ∈ A(u = 1) that there exists a unique R-quadratic mapping Q : M1 → M2, defined by f ((n−1)mx) (n−1)2m , which satisfies the equation (1.5) and the inequality (3.4). Q(x) = lim m→∞ By the similar manner to the proof of Theorem 3.1 we obtain that Q(ux) = uQ(x) for all x ∈ M1 and each u ∈ A(u = 1). The last relation is also true for u = 0. Since Q is R-quadratic, for each element a(a 6= 0) ∈ A Q(ax) = Q(cid:18)a = aQ(x) a a x(cid:19) = a2Q(cid:18) a a x(cid:19) = a2 a a2 Q(x) for all x ∈ M1 and for all a ∈ A. So the unique R-quadratic mapping Q is also Asa-quadratic, as desired. This completes the proof. (cid:3) Theorem 4.6. Let M1 and M2 be quasi-Banach A-bimodules. Assume that there exists a mapping ϕ : M n 1 → R+ for which a mapping f : M1 → M2 satisfies the functional inequality f (uxi − uxj) − n Xi=1 uf n Xj=1 n X1≤i<j≤n (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) ≤ ϕ(x1, · · · , xn), ∀xi ∈ M1, ∀u ∈ A(u = 1), xj − nxi!(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (n − 1)2mf(cid:16) x and the series (3.11) converges for all xi ∈ M1, i = 1, · · · , n. If either f is measur- able or f (tx) is continuous in t ∈ R for each fixed x ∈ M1, then there exists a unique Asa-quadratic mapping Q : M1 → M2, defined by Q(x) = lim m→∞ which satisfies the equation (1.5) and the inequality (3.12) for all x ∈ M1. (n−1)m(cid:17), References [1] J. Bae, K. Jun and Y. Lee, On the Hyers-Ulam-Rassias stability of an n-dimensional Pexider- ized quadratic equation, Math. Ineq. Appl. 7(1)(2004), 63-77. [2] Y. Benyamini and J. Lindenstrauss, Geometric Nonlinear Functional Analysis, Vol. 1, Colloq. Publ. 48, Amer. Math. Soc. Providence, 2000. [3] C. Borelli and G.L. Forti, On a general Hyers-Ulam stability result, Internat. J. Math. Math. Sci. 18(1995), 229-236. ULAM STABILITY PROBLEM 17 [4] D.G. Bourgin, Classes of transformations and bordering transformations, Bull. Amer. Math. Soc. 57(1951), 223-237. [5] I. Chang and H. Kim, On the Hyers-Ulam stability of quadratic functional equations, J. Ineq. Pure Appl. Math. 3(2002), 1-12. [6] S. Czerwik, On the stability of the quadratic mapping in normed spaces, Abh. Math. Sem. Univ. Hamburg, 62(1992), 59-64. [7] G.L. Forti, Comments on the core of the direct method for proving Hyers-Ulam stability of functional equations, J. Math. Anal. Appl. 295(2004), 127-133. [8] Z. Gajda, On stability of additive mappings, Internat. J. Math. Math. Sci. 14 (1991), 431 -- 434. [9] P. Gavruta, A generalization of the Hyers-Ulam-Rassias stability of approximately additive mappings, J. Math. Anal. Appl. 184(1994), 431-436. [10] D.H. Hyers, On the stability of the linear functional equation, Proc. Natl. Acad. Sci. 27(1941), 222-224. [11] D.H. Hyers, G. Isac and Th.M. Rassias, Stability of Functional Equations in Several Variables, Birkhauser, Basel, 1998. [12] K. Jun and H. Kim, Ulam stability problem for generalized A-quadratic mappings, J. Math. Anal. Appl. 305(2005), 466-476. [13] K. Jun and H. Kim, On the generalized A-quadratic mappings associated with the variance of a discrete-type distribution, Nonlinear Anal. 62(2005), 975-987. [14] R.V. Kadison and G. Pedersen, Means and convex combinations of unitary operators, Math. Scand. 57(1985), 249-266. [15] C.S. Lin, Sesquilinear and quadratic forms on modules over ∗-algebra, Publ. Inst. Math. 51(1992), 81-86. [16] C. Park, On the stability of the quadratic mapping in Banach modules, J. Math. Anal. Appl. 276(2002), 135-144. [17] C. Park, Multilinear mappings in Banach modules over a C ∗-algebra, Indian J. pure appl. Math. 35(2004), 183-192. [18] Th.M. Rassias, On the stability of the linear mapping in Banach spaces, Proc. Amer. Math. Soc. 72(1978), 297-300. [19] Th.M. Rassias, On the stability of functional equations in Banach spaces, J. Math. Anal. Appl. 251(2000), 264-284. [20] S. Rolewicz, Metric Linear Spaces, Reidel and Dordrecht, and PWN-Polish Sci. Publ. 1984. [21] S.C. S´anchez, The singular case in the stability of additive functions, J. Math. Anal. Appl. 268(2002), 498-516. [22] P. Semrl, On quadratic functionals, Bull. Austral. Math. Soc. 37(1987), 27-28. [23] F. Skof, Local properties and approximations of operators(Italian), Rend. Sem. Mat. Fis. Mi- lano 53(1983), 113-129. [24] J. Tabor, Stability of the Cauchy functional equation in quasi-Banach spaces, Ann. Polon. Math. 83 (2004), 243 -- 255. [25] S.M. Ulam, A collection of the mathematical problems, Interscience Publ. New York, 1960. [26] J. Vukman, Some functional equations in Banach algebras and an application, Proc. Amer. Math. Soc. 100(1987), 133-136. Department of Mathematics, Chungnam National University, 220 Yuseong-Gu, Daejeon, 305-764, Republic of Korea E-mail address: [email protected] Information Center for Mathematical Sciences, Korea Advanced Institute of Science and Technology, 373-1 Guseong-dong, Yuseong-Gu, Daejeon, 305-701, Re- public of Korea E-mail address: [email protected]
1007.2331	2	1007	2012-02-19T22:02:09	The representation theory of C*-algebras associated to groupoids	[ "math.OA" ]	Let E be a second-countable, locally compact, Hausdorff groupoid equipped with an action of T such that G:=E/T is a principal groupoid with Haar system \lambda. The twisted groupoid C-algebra C(E;G,\lambda) is a quotient of the C-algebra of E obtained by completing the space of T-equivariant functions on E. We show that C(E;G,\lambda) is postliminal if and only if the orbit space of G is T_0 and that C(E;G, \lambda) is liminal if and only if the orbit space is T_1. We also show that C(E;G, \lambda) has bounded trace if and only if G is integrable and that C(E;G, \lambda) is a Fell algebra if and only if G is Cartan. Let \G be a second-countable, locally compact, Hausdorff groupoid with Haar system \lambda and continuously varying, abelian isotropy groups. Let A be the isotropy groupoid and R := \G/A. Using the results about twisted groupoid C-algebras, we show that the C-algebra C(\G, \lambda) has bounded trace if and only if R is integrable and that C*(\G, \lambda) is a Fell algebra if and only if R is Cartan. We illustrate our theorems with examples of groupoids associated to directed graphs.	math.OA	math	THE REPRESENTATION THEORY OF C ∗-ALGEBRAS ASSOCIATED TO GROUPOIDS LISA ORLOFF CLARK AND ASTRID AN HUEF Abstract. Let E be a second-countable, locally compact, Hausdorff groupoid equipped with an action of T such that G := E/T is a principal groupoid with Haar system λ. The twisted groupoid C ∗-algebra C ∗(E; G, λ) is a quotient of the C ∗-algebra of E obtained by completing the space of T-equivariant functions on E. We show that C ∗(E; G, λ) is postliminal if and only if the orbit space of G is T0 and that C ∗(E; G, λ) is liminal if and only if the orbit space is T1. We also show that C ∗(E; G, λ) has bounded trace if and only if G is integrable and that C ∗(E; G, λ) is a Fell algebra if and only if G is Cartan. Let G be a second-countable, locally compact, Hausdorff groupoid with Haar system λ and continuously varying, abelian isotropy groups. Let A be the isotropy groupoid and R := G/A. Using the results about twisted groupoid C ∗-algebras, we show that the C ∗-algebra C ∗(G, λ) has bounded trace if and only if R is integrable and that C ∗(G, λ) is a Fell algebra if and only if R is Cartan. We illustrate our theorems with examples of groupoids associated to directed graphs. 1. Introduction Let H be a locally compact, Hausdorff group acting continuously on a locally compact, Hausdorff space X. When the orbit space X/H is reasonable, for example if X/H is T0, then every irreducible representation of the transformation-group C ∗-algebra C0(X) ⋊ H is induced from an irreducible representation of an isotropy subgroup Sx = {h ∈ H : h· x = x}. In particular, if the action of H on X is free then the spectrum of C0(X) ⋊H is homeomorphic to the orbit space by [12], or if H is abelian then the spectrum of C0(X)⋊H is homeomorphic to a quotient of (X/H) × H by [28]. Many of the postliminal (Type I) properties of the transformation-group C ∗-algebra can be deduced from the dynamics of the transformation group (H, X). For example, C0(X) ⋊ H is postliminal if and only if the orbit space is T0 and all the isotropy subgroups are postliminal [11]. There are many more results of this nature in the literature: [12, 29, 8] investigate when C0(X) ⋊ H has continuous trace, [15, 16, 1] when C0(X) ⋊ H is a Fell algebra or has bounded trace, and [28] when C0(X) ⋊ H is liminal. Usually the results are first proved for free actions and then generalized to non-free actions; but even when the isotropy groups are abelian the level of technical difficulty is much greater, and to get general results assumptions on the isotropy subgroups (for example, amenability or that they vary continuously) often seem unavoidable. The theorems above have served as a template for establishing similar Date: 12 July 2010. 2000 Mathematics Subject Classification. 46L05, 46L25, 46L55. Key words and phrases. Locally compact groupoid, twisted groupoid, C ∗-algebra, bounded trace, Fell algebra, graph groupoid. This research was supported by the Australian Research Council and the Association for Women in Mathematics. 1 2 CLARK AND AN HUEF theorems for the C ∗-algebras of directed graphs [9, 13] and the C ∗-algebras of groupoids [19, 20, 21, 4, 5, 6]. Let G be a locally compact, Hausdorff groupoid with abelian isotropy subgroups, and let A be the isotropy groupoid. The main theorem of [21] says that C ∗(G) has continuous trace if and only if the isotropy groups vary continuously and G/A is a proper groupoid. The proof strategy, quickly, is to show that all the irreducible representations of C ∗(G) are induced, and to use the dual isotropy groupoid A to construct a T-groupoid whose associated twisted groupoid C ∗-algebra is isomorphic to C ∗(G). Then the characterization of when twisted groupoid C ∗-algebras have continuous trace from [20] completes the proof. In this paper we generalize first the results from [20] to characterize when a twisted groupoid C ∗-algebra has bounded trace or is a Fell algebra (Theorems 4.3 and 5.2), and second, the results from [21] to characterize when a groupoid with continuously varying, abelian isotropy groups has bounded trace or is a Fell algebra (Theorems 6.4 and 6.5). To do this we had to deal with non-Hausdorff spectra, which led to a sharpening of [20, Proposition 3.3] and [21, Proposition 4.5] (see Theorem 3.4 and Proposition 6.3). The- orem 3.4 says that when the orbit space of the groupoid is T0, then the spectrum of the twisted groupoid C ∗-algebra is homeomorphic to the orbit space and Proposition 6.3 establishes the isomorphism of C ∗(G) with the twisted groupoid C ∗-algebra of [21] men- tioned above under weaker hypothesis. Finally we illustrate our theorems with examples of groupoids associated to directed graphs. 2. Preliminaries Let G be a locally compact, Hausdorff groupoid. We denote the unit space of G by G(0), the range and source maps r, s : G → G(0) are r(γ) = γγ−1 and s(γ) = γ−1γ, respectively, and the set of composable pairs by G(2). Recall that G is principal if the map γ 7→ (r(γ), s(γ)) is injective. Let N ⊆ G(0). The saturation of N is r(s−1(N)) = s(r−1(N)), and if N = r(s−1(N)) then we say that N is saturated. We define the restriction of G to N to be GN := {γ ∈ G : s(γ) ∈ N and r(γ) ∈ N}. The latter is not to be confused with GN := {γ ∈ G : s(γ) ∈ N}. If u ∈ G(0), we call the saturation of {u} the orbit of u ∈ G(0) and denote it by [u]; we also write Gu instead of G{u}. 2.1. T-groupoids. A T-groupoid E is a topological groupoid E with a continuous free action of the circle group T on E such that (1) if (γ1, γ2) ∈ E(2) and s, t ∈ T then (sγ1, tγ2) ∈ E(2) and (sγ1)(tγ2) = (ts)(γ1γ2); (2) G := E/T is a principal groupoid. In what follows, we will always assume that E is second-countable, locally compact and Hausdorff. Note that the composibility condition (1) implies that s(γ) = s(t · γ) and r(γ) = r(t · γ) for all γ ∈ E and t ∈ T; in particular, E(0) = G(0). That G is principal implies that there is an exact sequence (2.1) E(0) −→ E(0) × T i−→ E j−→ G −→ E(0) THE REPRESENTATION THEORY OF C ∗-ALGEBRAS ASSOCIATED TO GROUPOIDS 3 where i is the homeomorphism i(u, t) = t · u onto a closed subgroupoid and j is the quotient map. Conversely, starting with a sequence (2.1), there is a free action of T on E defined by t · γ = i(r(γ), t)γ, and the quotient E/T can be identified with G. Remark 2.1. Since T is compact, G = E/T is Hausdorff, and since T is a compact Lie group, E is a locally trivial bundle over G by [23, Proposition 4.65 and Hooptedoodle 4.68]. That the sequence (2.1) is exact is equivalent to: every γ in the isotropy groupoid A := {γ ∈ E : s(γ) = r(γ)} can be written as t · u for some t ∈ T and u ∈ E(0). Thus our T-groupoid is what is called a "proper T-groupoid" in [17, Definition 2.2]. But since we do not assume that G = E/T is ´etale, E is not a "twist" in the sense of [17, Definition 2.4]; T-groupoids are more general. In particular, our assumption that E is T-groupoid such that G = E/T is a principal groupoid puts us in the situation of [20]. Construction of the twisted groupoid C ∗-algebra. We briefly outline the construc- tion of the twisted groupoid C ∗-algebra from [20]. Let E be a T-groupoid over a principal groupoid G equipped with a left Haar system {λu : u ∈ G(0)}. Then there is a left Haar system {σu : u ∈ E(0) = G(0)} on E characterized by (2.2) ZE f (α) dσu(α) =ZGZT f (t · α) dt dλu(j(α)) (f ∈ Cc(E)). A left Haar system {λu : u ∈ G(0)} gives a right Haar system {λu : u ∈ G(0)} via λu(E) := λu(E−1), and we will move freely between the left and right systems when convenient. The usual groupoid C ∗-algebra C ∗(E, σ) of E is the C ∗-algebra which is universal for continuous nondegenerate ∗-representations L : Cc(E) → B(HL), where Cc(E) has the inductive limit topology, B(HL) the weak operator topology, and Cc(E) is a ∗-algebra via f ∗ g(γ) =ZE for f, g ∈ Cc(E). f (γα)g(α−1) dσs(γ)(α) and f ∗(γ) = f (γ−1) The twisted groupoid C ∗-algebra C ∗(E; G, λ) is a quotient of C ∗(E, σ) obtained as fol- lows. Let Cc(E; G) be the collection of f ∈ Cc(E) such that f (t · γ) = tf (γ). Note that for f, g ∈ Cc(E; G) and γ ∈ E, the function α 7→ f (γα)g(α−1) depends only on the class j(α) of α. So we can equip Cc(E; G) with a *-algebra structure via f ∗ g(γ) =ZG f (γα)g(α−1) dλs(γ)(j(α)) (cid:18)=ZE f (γα)g(α−1) dσs(γ)(α)(cid:19) and f ∗(γ) = f (γ−1) for f, g ∈ Cc(E; G); using (2.2) it is straightforward to check that the formulae for f ∗ g in Cc(E) and Cc(E; G) coincide. Let Rep(E; G) be the collection of non-degenerate ∗-representations L : Cc(E; G) → B(HL) which are continuous when Cc(E; G) has the inductive limit topology and B(HL) has the weak operator topology. It follows from [26, Proposition 3.5 and Th`eor´eme 4.1] that for f ∈ Cc(E; G) kf k = sup{kL(f )k : L ∈ Rep(E; G)} is finite and defines a pre-C ∗-norm on Cc(E; G). The completion of Cc(E; G) in this norm is the twisted groupoid C ∗-algebra C ∗(E; G, λ). That C ∗(E; G, λ) is a quotient of C ∗(E, σ) 4 CLARK AND AN HUEF follows because Rep(E; G) is a subset of the representations considered when constructing C ∗(E, σ). By Lemma 3.3 of [26], the surjective homomorphism Υ : Cc(E) → Cc(E; G) defined by (2.3) Υ(f )(γ) =ZT f (t · γ)¯t dt is continuous in the inductive limit topology, and hence extends to a homomorphism Υ : C ∗(E, σ) → C ∗(E; G, λ) called the quotient map. The reasons for calling C ∗(E; G, λ) the "twisted groupoid C ∗-algebra" are outlined in [20, §2]. 2.2. Postliminal properties of C ∗-algebras. Let A be a C ∗-algebra and A its spec- trum. If π is an irreducible representation of A then we write Hπ for the Hilbert space on which π(A) acts. If π(A) ⊇ K(Hπ) for every irreducible representation π of A, then A is postliminal ; if π(A) = K(Hπ) for every irreducible representation π of A, then A is limi- nal. In the literature postliminal and liminal C ∗-algebras are also called GCR and CCR C ∗-algebras, respectively. A positive element b ∈ A is called a bounded-trace element if the map [π] 7→ tr(π(b)) is bounded on A. Then A has bounded trace if the ideal consisting of the linear span of bounded-trace elements is dense in A. An irreducible representation π of A satisfies Fell's condition if there is a positive a ∈ A and a neighbourhood U of [π] in A such that σ(a) is a rank-one projection whenever [σ] ∈ U. If every irreducible representation of A satisfies Fell's condition then A is a Fell algebra. A Fell algebra A has Hausdorff spectrum if and only if A has continuous trace. Each of the properties above are listed in order of reverse containment. 3. The spectrum of a twisted groupoid C ∗-algebra We start by investigating ideals in C ∗(E; G, λ) associated to open saturated subsets of the unit space of G. Lemma 3.1. Suppose that E is a second-countable, locally compact, Hausdorff, T-groupoid such that G := E/T is a principal groupoid with Haar system λ. Let σ be the Haar system on E defined at (2.2) and U an open saturated subset of G(0) with F := G(0) \ U . Then the short exact sequence (3.1) 0 → C ∗(EU , σ) i→ C ∗(E, σ) p → C ∗(EF , σ) → 0 of [21, Lemma 2.10] induces a short exact sequence (3.2) 0 → C ∗(EU ; GU , λ) k→ C ∗(E; G, λ) r→ C ∗(EF ; GF , λ) → 0 such that k is isometric and Υ ◦ i = k ◦ Υ and Υ ◦ p = r ◦ Υ. On continuous functions the maps k and p are extension by 0 and restriction, respectively. Proof. Note that we write just Υ for both the homomorphisms Υ : C ∗(E, σ) → C ∗(E; G, λ) and Υ : C ∗(EU , σ) → C ∗(EU ; GU , λ). Since Υ ◦ i = Υ ◦ i ◦ Υ on Cc(EU ) we have ker Υ ⊆ ker(Υ ◦ i), and hence there exists a unique homomorphism k : C ∗(EU ; GU , λ) → C ∗(E; G, λ) such that Υ ◦ i = k ◦ Υ. Similarly, Υ ◦ p = Υ ◦ p ◦ Υ on Cc(E), so ker Υ ⊆ ker(Υ ◦ p), and hence there exists a unique homomorphism r : C ∗(E; G, λ) → C ∗(EF ; GF , λ) such that r ◦ Υ = Υ ◦ p. Note that r is surjective because p and Υ are. THE REPRESENTATION THEORY OF C ∗-ALGEBRAS ASSOCIATED TO GROUPOIDS 5 To see that k is isometric, fix a representation π of C ∗(EU ; GU , λ). It suffices to see that π determines a representation π of Cc(E; G) such that kπ(f )k = kπ(k(f ))k for f ∈ Cc(EU ; GU ); this will give kk(f )k ≥ kf k and hence kk(f )k = kf k. By [21, Lemma 2.10], i is an isometric isomorphism of C ∗(EU ) onto an ideal I of C ∗(E). Let π : C ∗(E) → B(Hπ) be the canonical extension of π ◦ Υ ◦ i−1 : I → B(Hπ). Note that, for g ∈ Cc(E) and h ∈ Cc(EU ) ⊆ Cc(E) we have π(Υ(g))π ◦ Υ ◦ i−1(h) = π ◦ Υ ◦ i−1(Υ(g)h) = π ◦ k−1 ◦ Υ(Υ(g)h) = π ◦ k−1 ◦ Υ(gh) = π ◦ Υ ◦ i−1(gh) = π(g)π ◦ Υ ◦ i−1(h). Thus π(Υ(g)) = π(g) and hence π factors through C ∗(E; G, λ) and gives a representation π : C ∗(E; G, λ) → B(Hπ) such that π = π ◦ Υ. Finally, if f ∈ Cc(EU ; GU ) then for all h ∈ Cc(EU ) we have π(k(f ))π ◦ Υ ◦ i−1(h) = π ◦ Υ ◦ i−1(k(f )h) = π ◦ Υ(f h) = π(f )π ◦ Υ ◦ i−1(h), and hence π(k(f )) = π(f ). Thus k is isometric. Since k is isometric, the image of k is the completion of Cc(EU ; GU ) viewed as functions on E. Since U and F are disjoint r(Cc(EU ; GU )) = 0 and hence range k ⊆ ker r. Conversely, if h ∈ Cc(E) ∩ ker r then h has support in U and hence is in the range of i. Thus range k = ker r. (cid:3) Fix u ∈ G(0) and let H0 u be the collection of bounded Borel functions f on E with compact support in Eu = s−1({u}) satisfying f (t · γ) = tf (γ) for all t ∈ T and γ ∈ E. For each ξ, η ∈ H0 u define (ξ η)u =ZG ξ(γ)η(γ) dλu(j(γ)) (cid:18)=ZE ξ(γ)η(γ) dσu(γ)(cid:19) to get an inner product on H0 u with respect to this inner product; note that Hu is a closed subspace of L2(Eu, σu). Moreover, the functions obtained by restricting elements of Cc(E; G) to Eu form a dense subset of Hu (see [20, Page 133]). u. Denote by Hu the Hilbert space completion of H0 Let f, ξ ∈ Cc(E; G). By [20, §3], the formula Lu(f )ξ(γ) = f ∗ ξ(γ) =ZG f (γα)ξ(α−1) dλu(j(α)) (3.3) defines an appropriately continuous representation Lu(E; G) = Lu of Cc(E; G) on a dense subspace of Hu, whence Lu extends to a representation Lu of C ∗(E; G, λ) on Hu. By [20, Lemma 3.2], Lu is irreducible, and if [u] = [v] then Lu and Lv are unitarily equivalent. In Proposition 3.3 of [20], Muhly and Williams prove that if C ∗(E; G, λ) has Hausdorff spectrum, then L : u 7→ [Lu] induces a homeomorphism Ψ from the orbit space G(0)/G onto the spectrum of C ∗(E; G, λ); it seems from the application of [20, Proposition 3.3] in the proof of [21, Proposition 4.5] that its authors knew that the proof goes through using only that C ∗(E; G, λ) has T1 spectrum (see [21, middle of p. 3638] and the applications of [21, Proposition 4.5] in the proof of [21, Theorem 1.1]). The original proof of [20, Proposition 3.3] refers the reader to the proof of [19, Propo- sition 25] to see that Ψ induces a continuous injection; since the notations of [20] and [19] 6 CLARK AND AN HUEF don't match up, we had to carefully go through the details to verify that the Hausdorff condition wasn't needed, and we record these details here. The proof that Ψ is open onto its range given in [20, Proposition 3.3] used that C ∗(E; G, λ) has T1 spectrum; our argument below does not require this hypothesis. We strengthen Proposition 3.2 further in Theorem 3.4 below. Proposition 3.2 (Muhly-Williams). Let E be a second-countable, locally compact, Haus- dorff, T-groupoid such that G := E/T is a principal groupoid with Haar system λ. For u ∈ G(0) let Lu be the irreducible representation defined at (3.3). (1) Then the map u 7→ [Lu] induces a continuous injection Ψ : G(0)/G → C ∗(E; G, λ)∧ which is open onto its range. (2) If G(0)/G is T1 then Ψ is a homeomorphism of G(0)/G onto C ∗(E; G, λ)∧. Proof. (1) We start by showing that Ψ is continuous. Fix ξ, η ∈ Cc(E; G) and a ∈ C ∗(E; G, λ). We claim that the map x 7→ (Lx(a)ξ η)x is continuous. To see why this is so, first consider f ∈ Cc(E; G) and note that (Lx(f )ξ η)x =ZG (f ∗ ξ)(γ)η(γ) dλx(j(γ)), where the convolution f ∗ ξ is taking place in Cc(E; G). Since (f ∗ ξ)(γ)η(γ) has compact support, the continuity of x 7→ (Lx(f )ξ η)x follows from the continuity of the Haar system. It now follows from an ǫ/3 argument that the map x 7→ (Lx(a)ξ η)x is also continuous. Now suppose that un → u in G(0); we will show by way of contradiction that [Lun] → [Lu]. Suppose that [Lun] does not converge to [Lu]. Then there exists a neighbourhood O of [Lu] such that [Lun] /∈ O frequently. By passing to a subsequence and relabeling we may assume [Lun] /∈ O for all n. Let J be the ideal of C ∗(E; G, λ) such that O = {ρ ∈ C ∗(E; G, λ)∧ ρ(J) 6= 0}. So there exists a ∈ J such that and Lu(a) 6= 0 and Lun(a) = 0 for all n. Now choose functions ξ, η ∈ Cc(G) such that (Lu(a)ξ η)u 6= 0; but now 0 = (Lun(a)ξ η)un → (Lu(a)ξ η)u 6= 0, contradicting the continuity of x 7→ (Lx(a)ξ η)x. So [Lun] → [Lu], and it follows that Ψ is continuous. Next we show that Ψ is injective. Let u, v ∈ G(0) and suppose that Lu and Lv are unitar- ily equivalent. We will show that [u] = [v]. By [20, Lemma 3.1] there is a homomorphism R : C0(G(0)) → M(C ∗(E; G, λ)) defined by (R(φ)f )(γ) = φ(r(γ))f (γ) for f ∈ Cc(E; G). In the proof of [20, Lemma 3.2] Muhly and Williams show that Lu is unitarily equivalent to a representation T u : C ∗(E; G, λ) → B(L2([u], µ[u]), and that Nu := T u ◦ R : C0(G(0)) → B(L2([u], µ[u]) has formula Nu(φ)η(x) = φ(x)η(x) for η ∈ L2([u], µ[u]). Since Lu and Lv are unitarily equivalent so are T u and T v, and thus so are Nu and Nv. But now if [u] 6= [v] then [u] ∩ [v] = ∅ and [28, Lemma 4.15] implies that N u and N v are not unitarily equivalent, a contradiction. So [u] = [v], and hence Ψ is injective. THE REPRESENTATION THEORY OF C ∗-ALGEBRAS ASSOCIATED TO GROUPOIDS 7 To see that Ψ is open onto its range we show that Ψ−1 : range Ψ → G(0)/G is continuous. We argue by contradiction: suppose that [Lun] → [Lu] in C ∗(E; G, λ) but [un] 6→ [u] in G(0)/G. Let U0 be an open neighbourhood of [u] in G(0)/G; let q : G(0) → G(0)/G be the quotient map and set U = q−1(U0). By passing to a subsequence and relabeling we may assume that un /∈ U for all n. By Lemma 3.1 C ∗(EU ; GU ) is isomorphic to an ideal I of C ∗(E; G). Set F := G(0) \ U. Fix f ∈ Cc(E; G) such that f (γ) = 0 for γ ∈ EF (such f are dense in I) and fix ξ ∈ Hun. Then kLun(f )ξk2 Lun(f )ξ(γ)2 dλun(j(γ)) un =ZG =ZG(cid:16)ZG f (γα−1)ξ(α) dλun(j(α))(cid:17)2 dλun(j(γ)). Now consider the inner integrand: there r(γα−1) = r(γ) and s(γ) = un, so γα−1 ∈ E[un]. But U is saturated with un /∈ U, so γα−1 ∈ EF and f (γα−1) = 0. Thus kLun(f )ξk2 un = 0, and since ξ was fixed we have Lun(f ) = 0. It now follows that I ⊆ ker Lun for all n. But now [Lun] /∈ I for all n, contradicting that I is an open neighbourhood of [Lu] and [Lun] → [Lu]. Thus Ψ is open. (2) In view of (1) it suffices to show that Ψ is surjective. See the proof of [20, Proposi- (cid:3) tion 3.3] (the proof given there only uses that G(0)/G is T1). Proposition 3.3. Suppose that E is a second-countable, locally compact, Hausdorff, T- groupoid such that G := E/T is a principal groupoid with Haar system λ. Then (1) C ∗(E; G, λ) is liminal if and only if the orbit space G(0)/G is T1; and (2) C ∗(E; G, λ) is postliminal if and only if the orbit space G(0)/G is T0. Proof. Let σ be the Haar system for E from (2.2). (1) Suppose C ∗(E; G, λ) is liminal. Since C ∗(E; G, λ) is separable, its spectrum is T1 by [7, 9.5.3]. By Proposition 3.2(1), Ψ : G(0)/G → C ∗(E; G, λ)∧, [u] 7→ [Lu] is a continuous injection. So for each u ∈ G(0), {[u]} = Ψ−1({[Lu]}) is closed in G(0)/G. Thus G(0)/G is T1. Conversely, suppose G(0)/G is T1. Since all the isotropy groups of E are amenable, C ∗(E, σ) is liminal by [5, Theorem 6.1]. Now C ∗(E; G, λ) is liminal because quotients of liminal algebras are liminal. (2) Proceed as in (1) using [7, 9.5.2] and [5, Theorem 6.1]. (cid:3) We now improve Proposition 3.2 by using a composition series to reduce to the T1 case: Theorem 3.4. Suppose that E is a second-countable, locally compact, Hausdorff, T- groupoid such that G := E/T is a principal groupoid with Haar system λ. If G(0)/G is T0 then Ψ is a homeomorphism of G(0)/G onto C ∗(E; G, λ)∧. Proof. We adapt the argument of [4, Proposition 5.1]. By Proposition 3.2(1) it suffices to show that Ψ is onto. Since G is σ-compact the equivalence relation R = {(r(γ), s(γ)) : γ ∈ G} is an Fσ set, so G(0)/G is almost Hausdorff by [24, Theorem 2.1]. A Zorn's lemma argument (see the discussion on page 25 of [10]) gives an ordinal γ and a collection {Vα : α ≤ γ} of open subsets of G(0)/G such that V0 = ∅, Vγ = G(0)/G, β < α implies Vβ ⊆ Vα, if α is a limit ordinal then Vα = ∪β<αVβ, and if α is not a limit ordinal or 0 then Vα \ Vα−1 is an open Hausdorff subset of (G(0)/G) \ Vα−1. Let q : G(0) → G(0)/G 8 CLARK AND AN HUEF be the quotient map and set Uα := q−1(Vα). By Lemma 3.1, for each α ≤ γ, there is an isomorphism kα of C ∗(EUα; GUα, λ) onto an ideal Iα of C ∗(E; G, λ). Now fix an irreducible representation π of C ∗(E; G, λ). Let α be the smallest element of the set {λ ≤ γ : π(Iλ) 6= 0}. Note that α is not a limit ordinal (if α were a limit ordinal then Uα = ∪β<αUβ and Iα is the ideal generated by the Iβ with β < α. But then π(Iα) 6= 0 implies π(Iβ) 6= 0 for some β < α, contradicting the minimality of α.) So π(Iα) 6= 0 and π(Iα−1) = 0. It follows that π is the canonical extension of the representation πIα and that πIα factors through a representation of Iα/Iα−1. By Lemma 3.1, Iα/Iα−1 is isomorphic to C ∗(EUα\Uα−1; GUα\Uα−1, λ). The orbit space Vα \Vα−1 of GUα\Uα−1 is Hausdorff, hence T1. Now apply Proposition 3.2(2) to get that πIα is unitarily equivalent Lu(EUα; GUα) ◦ k−1 α for some u ∈ Uα \ Uα−1. Note that Lu(EUα; GUα) ◦ k−1 α = Lu(E; G)Iα. Since πIα is unitarily equivalent to Lu(EUα; GUα) ◦ k−1 α it follows that their canonical extensions are unitarily equivalent. Thus π is unitarily equivalent to Lu(E; G). So Ψ is onto. (cid:3) 4. The twisted groupoid C ∗-algebras with bounded trace We recall [6, Definition 3.1]: a locally compact, Hausdorff groupoid G is integrable if for every compact subset N of G(0), (4.1) {λx(s−1(N))} < ∞. sup x∈N Equivalently, by [6, Lemma 3.5], G is integrable if and only if for each z ∈ G(0), there exists an open neighbourhood U of z in G(0) such that (4.2) {λx(s−1(U))} < ∞. sup x∈U So if a groupoid fails to be integrable, then there exists a z ∈ G(0) so that (4.3) {λx(s−1(U))} = ∞, sup x∈U for all open neighbourhoods U of z, and we say G fails to be integrable at z. In Proposition 4.2 we will prove that if G is integrable, then C ∗(E; G, λ) has bounded trace. To do so, we need to know that all irreducible representations of C ∗(E; G, λ) are equivalent to Lu for some u ∈ G(0). We proved in [6, Lemma 3.9] that if G is integrable, then all the orbits are closed; unfortunately, there is a gap in the proof (the proof assumes implicitly that orbits are locally closed at Equation 3.2 in [6]). Lemma 4.1 establishes that if G is a principal integrable groupoid, then the orbits are indeed locally closed. The proof of [6, Lemma 3.9] then goes through as written. The proof of Lemma 4.1 is based on the proof of [1, Lemma 2.1] which establishes similar results in the transformation-group setting. Lemma 4.1. Let G be a second-countable, locally compact, Hausdorff, principal groupoid and let z ∈ G(0). (1) If the orbit [z] is not locally closed then for every open neighbourhood V of z in G(0), λz(r−1(V )) = ∞. (2) If G is integrable then the orbits are locally closed. Proof. (1) Let W be any open neighbourhood of z in G(0). We claim that for every compact neighbourhood L of z in G there exists γL ∈ G \ L such that s(γL) = z and THE REPRESENTATION THEORY OF C ∗-ALGEBRAS ASSOCIATED TO GROUPOIDS 9 r(γL) ∈ W . Suppose there exists an L for which no such γ exists. Since [z] is not locally closed, ([z] \ [z]) ∩ W 6= ∅. Let y ∈ ([z] \ [z]) ∩ W . Then there exists {γi} ⊆ G such that s(γi) = z, r(γi) → y. Then r(γi) ∈ W eventually. So by assumption, γi ∈ L eventually. By passing to a subsequence we may assume that γi → γ ∈ L. But now s(γ) = z and r(γ) = y, and hence y ∈ [z], a contradiction. This proves the claim. Let V be an open neighbourhood of z in G(0) and let M ∈ P. There exists an open neighbourhood U of z in G0 and a compact symmetric neighbourhood K of z in G such that r(KU) ⊆ V . Let c > 0 such that λu(K) ≥ c for u ∈ U by [6, Lemma 3.10(1)]. Choose k ∈ P such that kc > M and γ(1), . . . , γ(k) as follows. Set γ(1) = z. By the claim there exists γ(2) ∈ G \ (K 2γ(1)) such that s(γ(2)) = z and r(γ(2)) ∈ U. Next, note that K 2γ(1) ∪ K 2γ(2) is compact, so by the claim there exists γ(3) ∈ G \ (K 2γ(1) ∪ K 2γ(2)) with s(γ(3)) = z and r(γ(3)) ∈ U. Continue. Now r(γ(i)) ∈ U for 1 ≤ i ≤ k and γ(j)(γ(i))−1 ∈ G \ K 2 when i 6= j. Thus r(Kγ(i)) ⊆ r(KU) ⊆ V and Kγ(i) ∩ Kγ(j) = ∅ when i 6= j. We have λz(r−1(V )) ≥ λz(cid:16) k [i=1 Kγ(i)(cid:17) = k Xi=1 λz(Kγ(i)) = k Xi=1 λr(γ(i))(K) ≥ kc > M. Since M was arbitrary, λz(r−1(V )) = ∞. (2) Suppose there exists z ∈ G(0) such that [z] is not locally closed. Let V be any open, relatively compact neighbourhood ofz in G. By (1), λz(r−1(V )) = ∞ thus sup{λx(s−1(V )) : x ∈ V } ≥ sup{λx(r−1(V )) : x ∈ V } = ∞ so G fails to integrable at z. (cid:3) Proposition 4.2. Suppose that E is a second-countable, locally compact, Hausdorff, T- groupoid such that G := E/T is a principal groupoid with Haar system λ. If G is integrable then the twisted groupoid C ∗-algebra C ∗(E; G, λ) has bounded trace. Proof. Since G is an integrable principal groupoid, the orbits of G are locally closed by Lemma 4.1. By Theorem 3.4, [u] 7→ [Lu] is a homeomorphism of G(0)/G onto the spectrum of C ∗(E; G, λ). Fix u ∈ G(0) and f ∈ Cc(E; G)+. Then, by [20, Proposition 4.1], Lu(f ) is trace class with Thus tr(Lu(f )) =ZG f (r(γ)) dλu(j(γ)). tr(Lu(f )) ≤ kf k∞λu{j(γ) : r(γ) ∈ supp f } = kf k∞λu{j(γ) : γ ∈ r−1(supp f )} = kf k∞λu{j(γ) : j(γ) ∈ j(r−1(supp f ))} = kf k∞λu{j(γ) : j(γ) ∈ r−1(j(supp f ))} ≤ kf k∞ sup {λu(r−1(j(supp f )))} < ∞ u∈j(supp f ) because supp f is compact and G is integrable. Thus Cc(E; G)+ is contained in the ideal spanned by the bounded-trace elements, and hence C ∗(E; G, λ) has bounded trace. (cid:3) In Theorem 4.3 below we show that if C ∗(E; G, λ) has bounded trace then G is inte- grable. The proof is modeled on [20, Theorem 4.3], where Muhly and Williams prove that 10 CLARK AND AN HUEF if C ∗(E; G, λ) has continuous trace then G is a proper groupoid. Their proof strategy is the following. Suppose that G is not proper. Then G fails to be proper at some z ∈ G(0). This gives a sequence {xn} ⊆ G which is eventually disjoint from every compact subset of G and r(xn), s(xn) → z. (In the terminology of [6, Definition 3.6], un := s(xn) converges 2-times in G(0)/G to z.) To show that C ∗(E; G, λ) does not have continuous trace, they construct a function F and show that F ∗F has certain tracial properties. In particular, they partition the Hilbert space Hu of Lu into a direct sum Hu,1 ⊕ Hu,2 and write Pu,i for the projection of Hu onto Hu,i. First they show that u 7→ tr(Lu(F ∗F )Pu,1) is continuous at z. Second, they have a really clever and technical argument to show that there is a constant a > 0 such that tr(Lun(F ∗F )Pun,2) ≥ kLun(F ∗F )Pun,2k ≥ a eventually. Next they push F ∗F into the Pedersen ideal of [22, Theorem 5.6.1] using a function q ∈ Cc(0, ∞) such that q(t) = t for t ∈ [a, kF ∗F k]. This gives an element d = q(F ∗F ) in the Pedersen ideal such that tr(Lun(d)Pun,2) ≥ a > 0 eventually. It follows that [Lu] 7→ tr(Lu(d)) is not continuous at [Lz]. Thus d is not a continuous-trace element. But the Pedersen ideal is the minimal dense ideal of a C ∗-algebra, so the ideal spanned by the continuous-trace elements cannot be dense in C ∗(E; G, λ). Thus C ∗(E; G, λ) does not have continuous trace. Theorem 4.3. Suppose that E is a second-countable, locally compact, Hausdorff, T- groupoid such that G := E/T is a principal groupoid with Haar system λ. The following are equivalent: (1) the twisted groupoid C ∗-algebra C ∗(E; G, λ) has bounded trace; (2) G is integrable; and (3) C ∗(G) has bounded trace. Since G is principal, (2) and (3) are equivalent by [6, Theorem 4.4]. By Proposition 4.2, if G is integrable then C ∗(E; G, λ) has bounded trace, so it remains to show that (1) implies (2). We prove the contrapositive. So suppose that G is not integrable, say G fails to be integrable at z ∈ G(0). Then by [6, Proposition 3.11] there exists a sequence {un} in G(0) so that {un} converges to z, un 6= z for all n, and un converges k-times in G(0)/G to z, for every k ∈ P. That is, there exist k sequences {γ(1) n } ⊆ G such that n }, . . . , {γ(k) n }, {γ(2) n ) → z and s(γ(i) (1) r(γ(i) (2) if 1 ≤ i < j ≤ k then γ(j) n (γ(i) n ) = un for 1 ≤ i ≤ k; n )−1 → ∞ as n → ∞, in the sense that {γ(j) n (γ(i) n )−1} admits no convergent subsequence. We will prove that C ∗(E; G, λ) does not have bounded trace. Since C ∗-algebras with bounded trace are liminal, we may assume that the orbits are closed in G(0) by Proposition 3.3. Let M > 0 be given. In order to show that C ∗(E; G, λ) does not have bounded trace, we will show that there exists an element d of the Pedersen ideal of C ∗(E; G, λ) such that tr(Lun(d)) > M eventually (see (3.3) for the definition of the irreducible representations Lun). Since M is arbitrary and the Pedersen ideal is the minimal dense ideal [22, Theorem 5.6.1], this shows that the ideal of bounded-trace elements cannot be dense. We will use the same function F as Muhly and Williams and adapt their proof as follows. (1) Show that there exists a constant a > 0 such that kLun(F ∗F )Pun,1k ≥ a. THE REPRESENTATION THEORY OF C ∗-ALGEBRAS ASSOCIATED TO GROUPOIDS 11 (2) Fix l ∈ N such that la > M. Since {un} converges k-times to z in G(0)/G for any n } ⊆ G satisfying the items n }, ..., {γ(l) n }, {γ(2) k, there exist l sequences {un = γ(1) (1) and (2) listed above. (3) Using that γ(j) (4) Write Pun,i for the projection of Hun onto Hun,i. Show that, for every 1 ≤ i ≤ l, n )−1 → ∞, partition Hun into l summands Hun,i. n (γ(i) kLun(F ∗F )Pun,ik ≥ a eventually. (5) Push F ∗F into the Pedersen ideal. Note that the order of events is subtle. We have to find the constant a before we can choose an appropriate l and then get l sequences in G which witness the l-times convergence of the sequence {un}. We retain, as much as possible, the notation of [20]. We start by explaining the function F (see (4.5) below). Fix a function g ∈ C + c (G(0)) so that 0 ≤ g ≤ 1 and g is identically one on a neighbourhood U of z. By [19, Lemma 2.7] there exist symmetric, open, conditionally compact neighbourhoods W0 and W1 in G such that G(0) ⊆ W0 ⊆ W 0 ⊆ W1 and W1z \ W0z ⊆ r−1(G(0) \ supp(g)). Choose symmetric, relatively compact, open neighbourhoods V0 and V1 of z in G such that V0 ⊆ V1. Apply Lemma A.1 to obtain a compact neighbourhood A of z in G(0) such that Thus (4.4) For γ ∈ EA, set 7 (W1 V1 \ W0V0) ∩ GA ⊆ r−1(G(0) \ supp g). 7 (W1 V1 W1 7 \ W0V0W0) ∩ GA ⊆ r−1(G(0) \ supp g). g(1)(γ) =(g(r(γ)), 0, 7 if j(γ) ∈ W1 V1 W1 if j(γ) /∈ W0V0W0, 7 where j : E → E/T is the quotient map. This gives a well-defined function g(1) on EA which, by Tietze's extension theorem extends to a well-defined function, also called g(1), in Cc(E). 0 V0W0 and b vanishes off W1 Next choose a self-adjoint b ∈ Cc(G) such that 0 ≤ b ≤ 1, b is identically one on W0V0W 2 . Also choose a compact neighbourhood C of z in G(0) = E(0) such that i(C × T) ⊇ G(0) ∩ supp g(1). Define l : i(C × T) → T by l(i(u, t)) = 1 and extend l to a function l ∈ Cc(E). By replacing l by (l + l∗)/2 we may assume l is self-adjoint. Set h = Υ(l) to obtain a self-adjoint h ∈ Cc(E; G) such that h(i(u, 1)) = 1 for all u ∈ C. Finally, define V1 W1 4 4 (4.5) F (γ) = g(r(γ))g(s(γ))b(j(γ))h(γ). Note that the h ensures F ∈ Cc(E; G), and that F is self-adjoint because h and b are. For each n, define Hun,1 = Hun ∩ L2(Eun ∩ j−1(W0V0W0), σun) and let Pun,1 be the projection onto Hun,1. The calculation [20, pages 140 -- 141] shows that Lun(F )Hun,1 ⊆ Hun,1. By [19, Lemma 2.9] there is a neighbourhood V2 of z in G and a conditionally compact, symmetric neighbourhood Y of G(0) in G such that V2 ⊆ V0, and (4.6) γ ∈ V2 =⇒ r(Y γ) ⊆ U. 12 CLARK AND AN HUEF In particular, since r(Y γ) = r(Y r(γ)), we get r(Y γ) ⊆ U whenever r(γ) ∈ V2. Since un → z we may assume that un ∈ V2 ∩ C for all n. The following lemma closely resembles [20, Lemma 4.5] and our proof is similar; we replace the unbounded sequence {xn} appearing in [20, Lemma 4.5] with the sequence {un} (corresponding to {r(xn)}) which causes us to consider the projection onto Hun,1 rather than the projection onto H⊥ un,1). Lemma 4.4. Let F be the function defined at (4.5). There exist an a > 0 and a neigh- bourhood V3 ⊆ V2 ∩ C of z in G such that whenever un ∈ V3. kLun(F ∗F )Pun,1k ≥ a Proof. Let Y be as above. Let O1, O2, c, Y0 be as in [20]. Thus O1 and O2 are open neighbourhoods of i(C × 1) in E so that for every η ∈ O1, Re(h(η)) > 1 2 and for every η ∈ O2, Re(h(η)) > 1 4, and c is a regular cross section of j (see [20, Proof of Lemma 3.2] for definition of regular). The set Y0 is a conditionally compact, symmetric neighbourhood Y0 of G(0) in G such that CY0 ⊆ j(O1) and Y ⊆ Y0. Let T0 = {ti}∞ i=1 be a countable dense subset of T. If x ∈ Y un, then j(c(un)c(x)−1) = unx−1 ∈ CY ⊆ CY0 ⊆ j(O1). So there exists a t ∈ T0 so that t · c(un)c(x)−1 ∈ O1. Define ζn : Y un → T0 by ζn(y) = tj where j = min{k : tk · c(un)c(y)−1 ∈ O1}. Thus apart from its domain ζ is the function defined in [20, Lemma 4.5]; that our ζn is Borel is proved as is done for the function in [20, Lemma 4.6]. By another argument very similar to that of [19, Lemma 2.9], there exists a conditionally compact, symmetric neighbourhood Y of E(0) in E such that Y j−1(CY ) ⊆ O2 and j( Y ) = Y . Since we are assuming that un ∈ V2 ∩ C for all n, if x ∈ Y un the claim gives (4.7) for all n. Y ζn(x)unc(x)−1 ⊆ Y C Y ⊆ Y j−1(CY ) ⊆ O2 Muhly and Williams use a function t : E → T defined as follows: for each γ ∈ E, consider the element γc(j(γ))−1. This is in the image of i and equals i(u, s) for some s ∈ T; then t(γ) := s. Let χn be the characteristic function of Y un and define ξn : E → C by ξn(γ) = t(γ)ζn(j(γ))χn(j(γ)). Since all of the functions involved in defining ξn are Borel, so is ξn. Also, it is clear that ξn is bounded and has compact support contained in supp(ξn) ⊆ j−1(Y un). Notice that t(s · γ) = st(γ) so that ξn ∈ Hun; since also Y un ⊆ W0V0W0 we have ξn ∈ Hun,1. (This is where we have departed from the Muhly-Williams proof - their unbounded sequence {xn} used in place of our {un} ensures their ξn has support in the orthogonal complement of Hun,1.) THE REPRESENTATION THEORY OF C ∗-ALGEBRAS ASSOCIATED TO GROUPOIDS 13 Now fix γ ∈ Y un and compute: Lun(F )(ξn)(γ) = F ∗ ξn(γ) =ZG F (γα−1)ξn(α) dλun(j(α)) g(r(γα−1))g(s(γα−1))b(j(γα−1))h(γα−1)ξn(α) dλun(j(α)) =ZG = g(r(γ))ZG (4.8) g(r(α))b(j(γα−1))h(γα−1)ξn(α) dλun(j(α)). Note that the integrand is zero unless j(α) ∈ Y un. Let j(α) ∈ Y un; then j(γα−1) ∈ Y unu−1 n Y ⊆ Y V0Y ⊆ W0V0W0, and hence b(j(γα−1) = 1. Also j(supp ξn) ⊆ Y un, so (4.8) = g(r(γ))ZY un = g(r(γ))ZY un g(r(α))h(γα−1)ξn(α) dλun(j(α)) g(r(x))h(γc(x)−1)ξn(c(x)) dλun(x) by letting x = j(α) and noting that r(α) = r(x) and c(x) = c(j(α)) = α. Since r( Y un) = r(Y un) ⊆ U by our choice of Y at (4.6) and since g is identically one on U, this is (4.9) =ZY un h(γc(x)−1)ξn(c(x)) dλun(x). By the definition of ξn and using that t ◦ c = 1 we get ξn(c(x)) = ζn(x) for x ∈ Y un, and since h is T-equivariant we get (4.9) =ZY un h(γζn(x)c(x)−1) dλun(x). But for x ∈ Y un, γζn(x)c(x)−1 ∈ O2 by (4.7), so Re(h(γζn(x)c(x)−1)) > 1 4. So Re(cid:0)Lun(F )(ξn)(γ)(cid:1) ≥ 1 4 λun(Y un) = 1 4 λun(Y ) Lun(F )(ξn)(γ)2 ≥ 1 16 λun(Y )2. and hence Now kLun(F )ξnk2 =ZG Lun(F )(ξn)(γ)2 dλun(j(γ)) ≥Z{j(γ):γ∈ Y un} ≥ZY un 1 16 Lun(F )(ξn)(γ)2dλun(j(γ)) λun(Y )2 dλun(j(γ)) = 1 16 λun(Y )3. By [6, Lemma 3.10(2)], applied to the conditionally compact neighbourhood Y , there exists a neighbourhood V3 of z and k ∈ N such that if v ∈ V3, λv(Y ) ≥ k > 0. By shrinking, we may take V3 ⊆ V2 ∩ C. 14 CLARK AND AN HUEF Let un ∈ V3. Then λun(Y ) ≥ k. Now let a be a real number so that 0 < a ≤ 1 16 k2. Since kξnk2 = λun(Y ) we get kLun(F ∗F )Pun,1k = kLun(F )Pun,1k2 = sup kηk=1 1 16 ≥ kLun(F ) ξn kξnk k2 = kLun(F )ηk2 λun(Y )2 ≥ a. (cid:3) Now that we have our a, choose l ∈ P so that la > M. Since {un} converges k times n } satisfying to z in G(0)/G for every k, there exist l sequences {un = γ(1) the two items on page 10. n }, ..., {γ(l) n }, {γ(2) For each n and 1 ≤ i ≤ l, we define the subspace Hun,i = Hun ∩ L2(Eun ∩ j−1(W0V0W0γ(i) n ), σun). Let Pun,i be the projection onto Hun,i for 1 ≤ i ≤ l. Lemma 4.5. The Hun,i (1 ≤ i ≤ l) are invariant under Lun(F ), and the Hun,i are eventually pairwise disjoint. Proof. Fix 1 ≤ i ≤ l. To see that Hun,i is invariant under Lun(F ) it suffices to show that supp(Lun(F )ψ) ⊆ j−1(W0V0W0γ(i) n ) for ψ ∈ Hun,i with compact support. Fix γ ∈ supp(Lun(F )ψ). Thus 0 6= Lun(F )ψ(γ) = g(r(γ))ZG g(r(α))b(j(γα−1))h(γα−1)ψ(α) dλun(j(α)). For the integral to be non-zero, there must exist α ∈ s−1({un}) in the support of the integrand. Then s(γ) = s(α) = un and, in particular, j(γ) ∈ Gun ⊆ GA. Also j(α) ∈ W0V0W0γ(i) n . Suppose, by way of contradiction, that j(γ) /∈ W1 V1 W1 supp b ⊆ W1 j(γ) /∈ W1 7 V1 W1 . But now j(γ) = j(γα−1)j(α) ⊆ W1 γ(i) n . So j(γ) ∈ W1 V1 W1 γ(i) n . 7 7 7 4 4 V1 W1 7 7 γ(i) n . We have j(γα−1) ∈ 7 γ(i) V1 W1 n , contradicting that 4 Now suppose, again by way of contradiction, that j(γ) /∈ W0V0W0γ(i) n . Then 7 7 V1 W1 γ(i) n \ W0V0W0γ(i) n (cid:1) ∩ GA(cid:1) ⊆ r−1(G(0) \ supp g) r(γ) = r(j(γ)) ∈ r(cid:0)(cid:0)W1 by (4.4). But now g(r(γ)) = 0, contradicting that Lun(F )ψ(γ) 6= 0. Thus j(γ) ∈ W0V0W0γ(i) n . Hence Hun,i is invariant under Lun(F ) for 1 ≤ i ≤ l. Next, suppose that 1 ≤ i < j ≤ l and that Hun,j and Hun,i are not eventually disjoint. n ) are not eventually disjoint. So there exists n ) and j−1(W0V0W0γ(i) nk }, {γ(j) nk } of {γ(i) n }, {γ(j) n }, and a sequence {αk} ⊆ G such that Then j−1(W0V0W0γ(j) subsequences {γ(i) Thus αk(γ(i) So nk )−1 ∈ W0V0W0r(γ(i) αk ∈ W0V0W0γ(i) nk ∩ W0V0W0γ(j) nk . nk ) ⊆ W0V0W0V3 and αk(γ(j) nk )−1 ∈ W0V0W0V3 eventually. γ(j) nk (γ(i) nk )−1 = γ(j) 3 W0V0W 2 nk (γ(i) eventually. But V −1 gent subsequence. But this contradicts the l-times convergence of {un}. nk )−1 ∈ V −1 0 V0W0V3 is relatively compact, so {γ(j) nk (αk)−1αk(γ(i) nk s(αk)(γ(i) nk )−1 = γ(j) 3 W0V0W 2 nk )−1} has a conver- (cid:3) 0 V0W0V3 THE REPRESENTATION THEORY OF C ∗-ALGEBRAS ASSOCIATED TO GROUPOIDS 15 Lemma 4.6. Let F be the function defined at (4.5) and a > 0 be as in Lemma 4.4. Suppose that un ∈ V3 and r(γ(i) n ) ∈ V3 ∩ C for 1 ≤ i ≤ l. Then for 1 ≤ i ≤ l. kLun(F ∗F )Pun,ik ≥ a Notice that in Lemma 4.4 above, we proved Lemma 4.6 in the special case where i = 1; we needed to do the base case i = 1 to find the constant a. The proof of Lemma 4.6 is similar to that of Lemma 4.4. Proof. Let Y, O1, O2, c, Y0, Y , T0 be as in Lemma 4.4. Fix i and x ∈ Y γ(i) n . Then j(c(γ(i) n )c(x)−1) = γ(i) n x−1 ∈ r(γ(i) n )Y ⊆ CY ⊆ CY0 ⊆ j(O1). There exists a t ∈ T0 so that t · c(γ(i) in Lemma 4.4 we now define ζ i c(γ(i) n )c(y)−1 ∈ O1}. Since r(γ(i) n : Y γ(i) n ) ∈ C, we have n )c(x)−1 ∈ O1. Just as we defined ζn : Y un → T0 n(y) = tj where j = min{k : tk · n → T by ζ i Let χi Y ζ i n(x)γ(i) n c(x−1) ⊆ Y r(γ(i) n be the characteristic function of Y γ(i) n(γ) = t(γ)ζ i ξi n ) Y ⊆ Y j−1(CY ) ⊆ O2. n and define ξi n : E → C by n(j(γ))χi n(j(γ)). Since all of the functions involved in defining ξi bounded, T-invariant and has compact support in j−1(Y γ(i) supp ξi n ) ⊆ Eun ∩ j−1(W0V0W0γ(i) n ⊆ Eun ∩ j−1(Y γ(i) n are Borel, so is ξi n. It is clear that ξi n is n ) = un we have n ). Since s(γ(i) n ). Thus ξi n ) then j(γα−1) ∈ Y r(γ(i) n ∈ Hun,i. n )Y ⊆ W0V0W0 and hence n , so the same calculation as done on page 13 n is Y γ(i) Fix γ ∈ j−1(Y γ(i) n ). If α ∈ j−1(Y γ(i) b(j(γα−1)) = 1. The support of ξi gives Lun(F )(ξi n)(γ) = g(r(γ))ZY γ (i) n g(r(y))h(γc(y)−1)ξi n(c(y)) dλun(y) which, since r(γ(i) n ) ∈ V2 implies r(Y γ(i) n ) ⊆ U and since g is identically one on U, is =ZY γ =ZY γ (i) n (i) n h(γc(y)−1)ξi n(c(y)) dλun(y) h(γζ i n(y)c(y)−1) dλun(y). But γζ i n(y)c(y)−1 ∈ O2, so Re(h(γζ i n(y)c(y)−1)) > 1 4 and Re(cid:0)Lun(F )(ξi n)(γ)(cid:1) ≥ 1 4 λun(Y γ(i) n ) = 1 4 λr(γ (i) n )(Y ). Since r(γ(i) Lemma 4.4 gives kLun(F ∗F )Pun,ik ≥ a. n ) ∈ V3 ∩ C by assumption, the same calculation as at the end of the proof of (cid:3) 16 CLARK AND AN HUEF To push F ∗F into the Pedersen ideal, let q ∈ Cc(0, ∞) be any function satisfying 0, 2t − t, 2a 3 , if t < a 3 , if a if 2a 3 ≤ t < 2a 3 , 3 ≤ t ≤ kF ∗F k. q(t) =  Set d := q(F ∗F ). We will show that tr(Lun(d)) ≥ la > M eventually. Fix un ∈ V3 such that r(γ(i) n ) ∈ V3 ∩ C for 1 ≤ i ≤ l. By Lemma 4.6, each Lun(F ∗F )Pun,i is positive with an eigenvalue at least as large as a, and by choice of q each q(Lun(F ∗F )Pun,i) is positive with norm at least as large as a. We claim that q(Lun(F ∗F )Pun,i) = q(Lun(F ∗F ))Pun,i. To see the claim, let p be a polynomial that vanishes at 0. Since Lun(F ∗F ) leaves Hun,i invariant, Lun(F ∗F ) and Pun,i commute. If we plug the operator Lun(F ∗F )Pun,i into p and simplify, we see that p(Lun(F ∗F )Pun,i) = p(Lun(F ∗F ))Pun,i. Because q can be uniformly approximated by polynomials p, each vanishing at 0, the claim follows. Now kLun(d)k = k(Lun(q(F ∗F ))k ≥ = kLun(q(F ∗F ))Pun,ik = kq(Lun(F ∗F ))Pun,ik l Xi=1 kq(Lun(F ∗F )Pun,i)k ≥ la > M l l Xi=1 Xi=1 by the choice of l. Thus tr(Lun(d)) > M, and since M was arbitrary d is not a bounded- trace element. But d is an element of the Pedersen ideal of C ∗(E; G, λ), so C ∗(E; G, λ) does not have bounded trace. This completes the proof of Theorem 4.3. 5. The twisted groupoid C ∗-algebras that are Fell Algebras Recall from [19] that a groupoid G is proper if the map π : G → G(0) × G(0), defined by π(γ) = (r(γ), s(γ)) for γ ∈ G, is a proper map. A subset U of G(0) is wandering if GU = π−1(U × U) is relatively compact. Thus G is proper if and only if every compact subset of G(0) is wandering. A groupoid G where each unit has a wandering neighbourhood is called Cartan [4, Definition 7.3]. The following lemma illustrates the relationship between a Cartan groupoid and 2- times convergence in the orbit space of the groupoid; it is similar to one direction of [2, Lemma 2.3]. Lemma 5.1 will be used in Example 7.1 below. Lemma 5.1. Let G be a topological groupoid. If there exists a sequence {un} ⊆ G(0) which converges 2-times in G(0)/G to z ∈ G(0), then G is not Cartan. Proof. We argue by contradiction. Suppose that {un} ⊆ G(0) converges 2-times in G(0)/G to z and that G is Cartan. Let U be a wandering neighbourhood of z in G(0), so that GU is relatively compact. There exist sequences {γ(1) n ) → z, s(γ(i) n ∈ GU eventually, and hence γ(2) n )−1 ∈ GU GU , eventually. But GU GU is compact, contradicting that γ(2) n (γ(1) n )−1 → ∞ as n → ∞. Thus γ(2) n ) = un for i = 1, 2 and γ(2) n )−1 → ∞ as n → ∞. n }, {γ(2) n } ⊆ G such that r(γ(i) n , γ(1) n (γ(1) n (γ(1) THE REPRESENTATION THEORY OF C ∗-ALGEBRAS ASSOCIATED TO GROUPOIDS 17 (cid:3) Theorem 5.2. Let E be a second-countable, locally compact, Hausdorff, T-groupoid such that G := E/T is a principal groupoid with Haar system λ. The following are equivalent: (1) the twisted groupoid C ∗-algebra C ∗(E; G, λ) is a Fell algebra; (2) G is Cartan; (3) C ∗(G) is a Fell algebra. Proof. Since G is principal, (2) and (3) are equivalent by [4, Theorem 7.9]; we will now prove the equivalence of (1) and (2). Suppose that G is Cartan. Fix an irreducible representation ρ of C ∗(E; G, λ); we will show that ρ satisfies Fell's condition. Since G is Cartan, G(0)/G is T1 by [4, Lemma 7.4]. So by Proposition 3.2, ρ is unitarily equivalent to Lu(E; G) for some u ∈ G(0). It suffices to show Lu(E; G) satisfies Fell's condition. Let U0 be a wandering neighbour- hood of u in G(0) and U = r(s−1(U0)) its saturation. Since G has a Haar system, r is open and hence U is open. By [4, Lemma 7.8], GU is a proper groupoid, so by [19, Theorem 4.2], C ∗(EU ; GU , λ) has continuous trace. By Lemma 3.1, the inclusion k : Cc(EU ; GU ) → Cc(E; G) induces an isometric isomorphism k of C ∗(EU ; GU , λ) onto an ideal I of C ∗(E; G, λ). Thus I has continuous trace and Lu(E; G)I = Lu(EU ; Gu) ◦ k−1. Since I has continuous trace, Lu(E; G)I satisfies Fell's condition in I, and hence Lu(E; G) satisfies Fell's condition in C ∗(E; G, λ)∧. Thus ρ satisfies Fell's condition in C ∗(E; G, λ)∧ as well and C ∗(E; G, λ) is a Fell algebra. Conversely, suppose that C ∗(E; G, λ) is a Fell algebra. Fix u ∈ G(0); we will show that u has a wandering neighbourhood in G(0). Since C ∗(E; G, λ) is liminal, G(0)/G is T1 by Proposition 3.3 and L : G(0)/G → C ∗(E; G, λ)∧, [u] 7→ [Lu] is a homeomorphism by Proposition 3.2. By [3, Corollary 3.4], [Lu] has an open Hausdorff neighbourhood O in C ∗(E; G, λ)∧. Let q : G(0) → G(0)/G the quotient map and set U = q−1(L−1(O)). Then U is an open saturated subset of G(0) and C ∗(EU ; GU , λ) is isomorphic to an ideal I of C ∗(E; G, λ) with spectrum O. Thus C ∗(EU ; GU , λ) has continuous trace (because I has) and hence GU is a proper groupoid by [19, Theorem 4.3]. So any relatively compact neighbourhood contained in U is a wandering neighbourhood of u in G(0). Thus G is Cartan. (cid:3) 6. Groupoids with abelian isotropy groups. Throughout this section G is a second-countable, locally compact, Hausdorff groupoid with Haar system λ. The change in notation from G to G is to emphasize that we are no longer assuming that the groupoid G is principal. Let Au = {γ ∈ G : r(γ) = s(γ) = u} be the isotropy group at u ∈ G(0) and let A = {γ ∈ G : r(γ) = s(γ)} be the isotropy groupoid; we also assume throughout this section that the isotropy groups are abelian and vary continuously, that is, that the map u 7→ Au from G(0) to the space of closed subsets of G(0), is continuous in the Fell topology. The isotropy groupoid acts on the left and right of G and the quotient R := G/A is a principal groupoid. The main results of this section, Theorems 6.4 and 6.5, say that C ∗(G, λ) has bounded trace if and only if R is an integrable groupoid, and that C ∗(G, λ) is a Fell algebra if and only if R is a Cartan groupoid. Once again, our proofs are modeled after the analogous result [21, Theorem 1.1] for groupoid C ∗-algebras with continuous trace. 18 CLARK AND AN HUEF Since the isotropy groups vary continuously, A has a Haar system β [27, Lemmas 1.1 and 1.2]. Write A for the spectrum of C ∗(A, β). Then R acts on the right of A (see (6.1) and (6.2) below). In [21] Muhly, Renault, and Williams show that if A/R is Hausdorff, then C ∗(G, λ) is isomorphic to a particular twisted groupoid C ∗-algebra [21, Proposi- tion 4.5]. They then apply the characterization of when twisted groupoid C ∗-algebras have continuous trace from [20] to prove [21, Theorem 1.1]. Our strategy is similar. We prove in Lemma 6.1 that G(0)/G is T1 if and only if A/R is T1. This allows us to show that the isomorphism of [21, Proposition 4.5] holds even if A/R is only T1. Then we use the isomorphism and our characterizations in Theorems 4.3 and 5.2 of when twisted groupoid C ∗-algebras have bounded trace or are Fell algebras to get results for C ∗(G, λ). We need some background before we can proceed to Lemma 6.1. Since C ∗(A, β) is a separable commutative C ∗-algebra, the discussion on [21, p. 3630] shows that (6.1) A = {(χ, u) : u ∈ G(0), χ ∈ Au} where (χ, u)(f ) = RAu χ(a)f (a) dβu(a) for f ∈ Cc(A). Proposition 3.3 of [21] describes criteria for convergence in A: (χn, un) → (χ, u) in A if and only if (1) un → u in A(0)(= G(0)), and (2) if an ∈ Aun, a ∈ Au and an → a in A, then χn(an) → χ(a). If χ ∈ Au and γ ∈ G with r(γ) = u, then χ · γ is the character of As(γ) defined by χ · γ(a) = χ(γ−1aγ). Note that χ · γ depends only on γ. There is a groupoid action of R (and G) on the right of A via (6.2) (χ, u) · γ = (χ · γ, s(γ)) for γ ∈ G with r(γ) = u. Lemma 6.1. Suppose that G is a second-countable, locally compact, Hausdorff groupoid with Haar system. Also assume that the isotropy groups are abelian and vary continuously. Then G(0)/G is T1 if and only if A/R is T1. Proof. First suppose that G(0)/G is T1. Fix (ρ, v) ∈ A. It suffices to show that [(ρ, v)] is closed. Let (χn, un) ∈ [(ρ, v)] and suppose that (χn, un) → (χ, u) in A. Thus there exists γ ∈ G with s(γ) = u and r(γ) = v, and, for each n, there exists γn ∈ G with s(γn) = un, r(γn) = v such that (χn, un) = (ρ · γn, un) = (ρ, v) · γn. Note u ∈ [v] since un ∈ [v] and [v] is closed by assumption, and that γ, γn ∈ G[v]. Since G has a Haar system, r and s are open maps [25, Proposition 2.4] and this puts us in the setting of [21]. Since G[v] is a transitive groupoid, the map π : G[v] → [v] × [v], π(α) = (r(α), s(α)) is open by [21, Theorems 2.2A and 2.2B]. Since π(γn) = (v, un) → (v, u) = π(γ) and π is open, there exists a subsequence {γnk} and a sequence {ηk} ⊆ G such that π(γnk) = π(ηk) and ηk → γ in G[v] (see, for example, [30, Proposition 1.15]). Thus ηk → γ in G as well. Note that γnk = ηk. such that ak → a in A. Then by the continuity of Fix a sequence {ak} with ak ∈ Aunk multiplication, χk(ak) = (ρ · γnk)(ak) = ρ(η−1 k akηk) → ρ(γ−1aγ) = (ρ · γ)(a). THE REPRESENTATION THEORY OF C ∗-ALGEBRAS ASSOCIATED TO GROUPOIDS 19 Thus {(χnk, unk)} converges to both (χ, u) and (ρ · γ, u) in A. Since A is Hausdorff we have (χ, u) = (ρ · γ, u) = (ρ, v) · γ ∈ [(ρ, v)]. So [(ρ, v)] is closed. Hence A/R is T1. For the converse, first consider φ : G(0) → A/R defined by φ(u) = [(1u, u)], where 1u is the trivial character a 7→ 1 for a ∈ Au. If un → u in G(0) then, using the convergence criteria for sequences in A of [21, Proposition 3.3], it is clear that (1un, un) → (1u, u) in A, so that φ is continuous. Now suppose that φ(u) = φ(v). Then there exists γ ∈ G with r(γ) = u such that (1u, u) · γ = (1v, v). Thus v = s(γ) and hence [u] = [v]. So φ induces a continuous injection φ : G(0)/G → A/R. It follows that G(0)/G is T1 if A/R is. (cid:3) In [21], Muhly, Renault, and Williams define a groupoid A ⋊ R as follows. As a set A ⋊ R = {(χ, r(γ), γ) ∈ A × R}, but an element (χ, r(γ), γ) is abbreviated to just (χ, γ); the topology on A ⋊ R is the product topology. The unit space is A with range and source maps r((χ, γ)) = (χ, r(γ)) and s((χ, γ)) = (χ · γ, s(γ)). The multiplication and inverse in A ⋊ R is given by (χ, γ)(χ · γ, α) = (χ, γ α) and (χ, γ)−1 = (χ · γ, γ−1). It is straightforward to see that A ⋊ R is principal. Note that R is proper if and only if A ⋊ R is proper; similarly one is Cartan or integrable if and only if the other is: Lemma 6.2. Suppose that G is a second-countable, locally compact, Hausdorff groupoid with abelian isotropy. Also assume that the isotropy groupoid A has a Haar system. (1) If G has a Haar system, then R and A ⋊ R have Haar systems α and δ × α, respectively; and with respect to these Haar systems, R is integrable if and only if A ⋊ R is integrable. (2) R is Cartan if and only if A ⋊ R is Cartan. Proof. (1) Since G and A have Haar systems, R has a Haar system α by [21, Lemma 4.2]. It is straightforward to check that if (χ, u) ∈ A and δ(χ,u) is point-mass measure, then δ × α(χ,u) := δ(χ,u) × αu gives a Haar system on A ⋊ R. Suppose R is integrable. Fix a compact subset K in ( A⋊R)(0) = A. Let p2 : A⋊R → R be the projection onto the second coordinate; note that p2(K) is a compact subset of R(0) = G(0). We have (δ × α)(χ,u)(s−1(K)) = δ(χ,u) × αu(cid:0){(η, γ) ∈ A ⋊ R : (η · γ, s(γ)) ∈ K}(cid:1) = αu(cid:0){ γ ∈ R : r(γ) = u, (χ · γ, s(γ)) ∈ K}(cid:1) ≤ αu(cid:0){ γ ∈ R : s(γ) ∈ p2(K)}(cid:1) = αu(cid:0)s−1(p2(K))(cid:1). (χ,u)∈K(cid:8)(δ × α)(χ,u)(s−1(K))(cid:9) ≤ sup u∈p2(K)(cid:8)αu(s−1(p2(K))})(cid:9) < ∞. Since R is integrable, this gives sup So A ⋊ R is integrable. 20 CLARK AND AN HUEF Conversely, suppose that A ⋊ R is integrable. Fix a compact set L in R(0) = G(0). For each u ∈ L, let 1u be the trivial character of Au, so that a 7→ 1 for all a ∈ Au. Set L = {(1u, u) : u ∈ L} ⊆ A. We claim that L is a compact subset of ( A ⋊ R)(0) = A. To see this, let {(1vn, vn)} be a sequence in L. Then {vn} is a sequence in L and hence has a convergent subsequence vnk → v ∈ L. Using the convergence criteria for sequences in A , vnk) → (1v, v) in A. Thus L is compact in A. of [21, Proposition 3.3] it is clear that (1vnk Now note that s((1u, γ)) = (1u, s(γ)), so that s((1u, γ)) ∈ L if and only if s( γ) ∈ L. Thus sup u∈L(cid:8)αu(s−1(L))(cid:9) ≤ sup (1u,u)∈ L {(δ × α)(1u,u)(s−1( L))} < ∞ because A⋊R is integrable and L is a compact subset of its unit space. So R is integrable. (2) First suppose that R is Cartan. Fix (χ, u) ∈ A = ( A ⋊ R)(0). Let K be a relatively compact, wandering neighbourhood of u in R(0). Let p1 : A ⋊ R → A, p2 : A ⋊ R → R be the projections onto the first and second coordinate, respectively. Let N be a relatively compact neighbourhood of (χ, u) in A such that p2(N) = K. Let {(ηn, γn)} be a sequence in π−1(N × N) = {(χ, γ) : (χ, r(γ)) ∈ N, (χ · γ, s(γ)) ∈ N}. Then { γn} ⊆ π−1(p2(N) × p2(N)) = π−1(K × K), hence has a convergent subsequence { γnk}. Note {ηnk} ⊆ p1(N), a relatively compact set. So there exists a convergent subsequence } is a convergent subsequence of {(ηn, γn)}. Thus π−1(N × N) is {ηnki relatively compact. Hence A ⋊ R is Cartan. }. So {(ηnki γnki Conversely, suppose that A ⋊ R is Cartan. Fix u ∈ R(0). There exists a wandering neighbourhood N of (1u, u) in A. Let K = p2(N); then K is a neighbourhood of u. Let { γn} ⊆ π−1(K × K). For each n there exists ηn such that (ηn, γn) ∈ π−1(N × N). But π−1(N ×N) is relatively compact, so {(ηn, γn)} has a convergent subsequence {(ηnk, γnk)}. Thus { γnk} is a convergent subsequence of { γn}. Thus π−1(K × K) is relatively compact. Hence R is Cartan. (cid:3) We will now briefly describe the T-groupoid D of [21, §4]. There D := {(χ, z, γ) : χ ∈ Ar(γ), z ∈ T, γ ∈ G}/ ∼, (6.3) where (χ, χ(a)z, γ) ∼ (χ, z, a · γ); the unit space is A with r([(χ, z, γ)]) = (χ, r(γ)) and s([(χ, z, γ)]) = (χ · γ, s(γ)) and multiplication and inverse [(χ, z, γ)][(χ · γ, z′, γ′)] = [(χ, zz′, γγ′)] and [(χ, z, γ)]−1 = [(χ · γ, ¯z, γ−1)]. That D is indeed a T-groupoid over A ⋊ R is established on [21, p. 3636]. Proposition 4.5 of [21] says that if A/R is Hausdorff, then C ∗(G) and C ∗(D; A⋊R, δ×α) are isomorphic. We now establish that the given proof works almost as is written even if A/R is only T1. Proposition 4.5 of [21] uses the Hausdorff assumption in three places. The first use is in Lemma 4.8 to establish that the G-orbits in G(0) are closed; so assuming A/R is T1 suffices by Lemma 6.1. The second use is to establish again that the G-orbits are closed in G(0) so that [21, Lemma 2.11] applies. The third use is to establish that every irreducible representation of C ∗(D; A ⋊ R, δ × α) is of the form [L(χ,u)]; here we THE REPRESENTATION THEORY OF C ∗-ALGEBRAS ASSOCIATED TO GROUPOIDS 21 note that ( A ⋊ R)(0)/( A ⋊ R) is homeomorphic to A/R, so we can use Proposition 3.2 for this if A/R is T1. Thus we have: Proposition 6.3. Suppose G is a second-countable, locally compact, Hausdorff groupoid with Haar system λ. Also suppose that the isotropy groups of G are abelian and vary continuously. If G(0)/G is T1, then C ∗(G, λ) and C ∗(D; A ⋊ R, δ × α) are isomorphic. Theorem 6.4. Suppose that G is a second-countable, locally compact, Hausdorff groupoid with Haar system λ. Also suppose that the isotropy groups of G are abelian and vary continuously. Let A be the isotropy groupoid. The following are equivalent: (1) C ∗(G, λ) has bounded trace; (2) R := G/A is integrable; (3) C ∗(R) has bounded trace. Proof. Since R is principal, the equivalence of (2) and (3) is [6, Theorem 4.4]; we will now prove the equivalence of (1) and (2). Note that the isotropy groups vary continuously if and only if the isotropy groupoid A has a Haar system by [27, Lemmas 1.1 and 1.2]. First suppose that C ∗(G, λ) has bounded trace. Then C ∗(G, λ) is liminal and hence the orbits of G are closed by [5, Theorem 6.1]. Now C ∗(G, λ) and C ∗(D; A ⋊ R, δ × α) are isomorphic by Proposition 6.3. Thus C ∗(D; A ⋊ R, δ × α) has bounded trace as well. Thus A ⋊ R is integrable by Theorem 4.3, and hence R is integrable by Lemma 6.2(1). Conversely, suppose R is integrable. Then the orbits in R are closed by [6, Lemma 3.9] and Lemma 4.1. Since R and G have the same orbit space, orbits are closed in G. By Proposition 6.3, C ∗(G, λ) and C ∗(D; A ⋊ R, α) are isomorphic. Since R is integrable, A ⋊ R is integrable by Lemma 6.2(1). Thus C ∗(D; A ⋊ R, α), and hence C ∗(G, λ), has bounded trace by Theorem 4.3. (cid:3) Theorem 6.5. Suppose that G is a second-countable, locally compact, Hausdorff groupoid with Haar system λ. Also suppose that the isotropy groups of G are abelian and vary continuously. Let A be the isotropy groupoid. The following are equivalent: (1) C ∗(G, λ) is a Fell algebra; (2) R := G/A is Cartan; (3) C ∗(R) is a Fell algebra. Proof. Since R is principal, the equivalence of (2) and (3) is [4, Theorem 7.9]. The proof of the equivalence of (1) and (2) is similar to the proof of Theorem 6.4, using Lemma 6.2(2), Theorem 5.2 and [4, Lemma 4.7] in place of Lemma 6.2(1),Theorem 4.3 and [6, Lemma 3.9], respectively. (cid:3) 7. Examples Our examples use groupoids constructed from directed graphs. We start with some background. Let E = (E0, E1, r, s) be a directed graph. Thus E0 and E1 are countable sets of vertices and edges, respectively, and r, s : E1 → E0 are the range and source map, respectively. For e ∈ E1, call s(e) the source of e and r(e) the range of e. A directed graph E is row-finite if r−1(v) is finite for every v ∈ E0. A finite path is a finite sequence α = α1α2 · · · αk of edges αi ∈ E1 with s(αj) = r(αj+1) for 1 ≤ j ≤ k − 1; write s(α) = s(αk) and r(α) = r(α1), and call α := k the length of α. An infinite 22 CLARK AND AN HUEF path x = x1x2 · · · is defined similarly, although s(x) remains undefined. Let E∗ and E∞ If α = α1 · · · αk denote the set of all finite paths and infinite paths in E respectively. and β = β1 · · · βj are finite paths with s(α) = r(β), then αβ is the path α1 · · · αkβ1 · · · βj. When x ∈ E∞ with s(α) = r(x) define αx similarly. A cycle is a finite path α of non-zero length such that s(α) = r(α). By [18, Corollary 2.2], the cylinder sets Z(α) := {x ∈ E∞ : x1 = α1, . . . , xα = αα}, parameterized by α ∈ E∗, form a basis of compact, open sets for a locally compact, σ-compact, totally disconnected, Hausdorff topology on E∞. In [18], Kumjian, Pask, Raeburn and Renault built a groupoid GE, called the path groupoid, from a row-finite directed graph E as follows. Two paths x, y ∈ E∞ are shift equivalent with lag k ∈ Z (written x ∼k y) if there exists N ∈ N such that xi = yi+k for all i ≥ N. Then the groupoid is GE := {(x, k, y) ∈ E∞ × Z × E∞ : x ∼k y}. with composable pairs G(2) and composition and inverse given by E := {(cid:0)(x, k, y), (y, l, z)(cid:1) : (x, k, y), (y, l, z) ∈ GE}, (x, k, y) · (y, l, z) := (x, k + l, z) and (x, k, y)−1 := (y, −k, x). For each α, β ∈ E∗ with s(α) = s(β), let Z(α, β) be the set {(x, k, y) : x ∈ Z(α), y ∈ Z(β), k = β − α, xi = yi+k for i > α}. By [18, Proposition 2.6], the collection of sets {Z(α, β) : α, β ∈ E∗, s(α) = s(β)} is a basis of compact, open sets for a second-countable, locally compact, Hausdorff topol- ogy on GE such that GE is an r-discrete groupoid with a Haar system of counting measures. After identifying each (x, 0, x) ∈ G(0) E with x ∈ E∞, [18, Proposition 2.6] says that the topology on G(0) E is identical to the topology on E∞. We caution that in our notation (which is now standard) the sources and ranges are swapped from the notation used in [18]. Example 7.1. Let E be the graph v1• v0• v2• f0,0 f0,1 f1,0 f1,1 f2,0 f2,1 w0,0 • e0,0 w1,0 • e1,1 e1,0 w2,0 • e2,2 e2,0 • w1,1 • w2,1 2 • w2,2 e2,1 . . . . . . . . . Let x ∈ E∞. If x = ααα... for some cycle α with r(α) = wn,k for some 0 ≤ k ≤ n, then the isotropy subgroup of x in GE is Ax = (n + 1)Z; otherwise Ax = {0}. It is straightforward to check that the isotropy subgroups vary continuously in the Fell topology. We claim o o o o o o E E I I U U I I U U I I U U U U 2 e e THE REPRESENTATION THEORY OF C ∗-ALGEBRAS ASSOCIATED TO GROUPOIDS 23 that the groupoid C ∗-algebra C ∗(GE) has bounded trace but is not a Fell algebra. To see this, by Theorems 6.4 and 6.5, we need to show that R := GE/A is integrable but not Cartan. We start by considering the following graph F from [14, §8]: v′ 0• v′ 1• v′ 2• f ′ 0,0 f ′ 0,1 f ′ 1,0 f ′ 1,1 f ′ 2,0 f ′ 2,1 0,0 w′ • e′ 0,0 • • ... 1,0 w′ • e′ 1,0 1,1 w′ • e′ 1,1 • ... 2,0 w′ • e′ 2,0 2,1 w′ • e′ 2,1 2,2 w′ • e′ 2,2 ... . . . . . . . . . . . . There are no cycles in F , so GF is a principal groupoid by [14, Proposition 8.1]. By [13], the groupoid GF is integrable. So C ∗(GF ) has bounded trace by [6, Theorem 4.4]. For n ≥ 0, let xn be the unique infinite path with range v′ n,0 as an edge, let yn be the unique infinite path with range v′ n,1 as an edge, and let z be the infinite path going through each v′ It is shown in [14, Example 8.2] that the i. sequence {xn} converges 2-times in G(0) F /GF to z; the sequences in GF witnessing this 2-times convergence are γ(1) n = (yn, 0, xn). It follows that GF is not a Cartan groupoid by Lemma 5.1. Since GF is principal, C ∗(GF ) is not a Fell algebra by [4, Theorem 7.9]. n = (xn, 0, xn) and γ(2) 0 which has f ′ 0 which has f ′ Now consider the open subset i) ∪j≤i Z(w′ i,j)(cid:1) U = [i≥0(cid:0)Z(v′ of F ∞. Let G be the groupoid obtained by restricting GF to U. Then G is a principal, integrable groupoid which is not Cartan (because the two sequences witnessing the 2- times convergence of {xn} in GF are also in G). Thus C ∗(G) has bounded trace but is not a Fell algebra. We claim that G is isomorphic to R = GE/A. To see this, first note that "unwrapping" cycles in E∗ sets up a bijection φ between E∗ and the set of finite paths in U; similarly "unwrapping" cycles in E∞ sets up a bijection ψ between E∞ and the set of infinite paths in U. If α is a finite path in E∗ then Z(φ(α)) = ψ(Z(α)). Since the cylinder sets form a basis for the topology on E∞, ψ : E∞ → U is a homeomorphism. Second, fix (x, k, y) ∈ GE so that x ∼k y. Then either (1) x and y are of the form x = αγγ..., y = βγγ.... where γ is a cycle with r(γ) = wn,0 for some n ∈ N, and α, β ∈ E∗ don't contain γ, or (2) both x and y do not contain cycles. In (1), ψ(x) and ψ(y) are o o o o o o I I U U I I U U I I U U O O O O O O O O O O O O O O O O O O 24 CLARK AND AN HUEF shift-equivalent with lag α − β, and in (2) ψ(x) and ψ(y) are shift-equivalent with lag k. Thus, since G is principal, if (x, k, y) ∈ GE then there exists a unique lk such that (ψ(x), lk, ψ(y)) ∈ G. Finally, it is now straightforward to verify that ρ : GE → G defined by ρ((x, k, y)) = (ψ(x), lk, ψ(y)) is a groupoid homomorphism which is continuous, open and surjective, and that ρ induces a homeomorphism ρ : R → G. Thus R is an integrable groupoid which is not Cartan, and hence C ∗(GE) has bounded trace but is not a Fell algebra by Theorems 6.4 and 6.5. Example 7.2. Let GE be the groupoid from Example 7.1. Let DE be the associated T- groupoid over A ⋊ R defined by Muhly-Williams-Renault (see (6.3)). Note that the orbit space G(0) E /GE is T1, so C ∗(GE) and C ∗(DE; A⋊R, δ×α) are isomorphic by Proposition 6.3. Thus C ∗(DE; A ⋊ R, δ × α) has bounded trace but is not a Fell algebra, and hence by Theorems 4.3 and 5.2, A ⋊ R is an integrable groupoid but is not Cartan. Appendix A. Corrections to the proof of Theorem 2.3 of [19] contributed by Dana P. Williams. Robert Hazlewood pointed out that there is a problem with the proof of Theorem 3.2 in [19]. On the bottom of page 237, we assert that we can find neighbourhoods V0 and V1 of z such that V0 ⊆ V1 with the property that1 (A.1) 7 W1 V1 W1 7 \ W0V0W0 ⊆ r−1(G(0) \ N). Unfortunately, if V1 is larger than V0, then we see no reason such neighbourhoods should exist. In fact, we now suspect that it is not possible to find such neighbourhoods -- let alone via a "straightforward compactness argument". However, (A.1) does hold provided we restrict to elements with source sufficiently close to z.2 Namely, we can prove the following. Lemma A.1. Given neighbourhoods V0 and V1 of z in G with V0 open and V1 relatively compact, there is a compact neighbourhood A of z in G(0) such that (A.2) where GA := s−1(A). 7 (W1 V1 \ W0V0) ∩ GA ⊆ r−1(G(0) \ N), Proof. If no such A exists, we can let { An } be a neighbourhood basis of z with each An compact and An+1 ⊆ An. Then, by assumption, for each n we can find γn ∈ GAn belonging to the closed set r−1(N) ∩ (W1 V1 is compact, we can pass to a subsequence, relabel, and assume that γn → γ. Notice that we must have γ ∈ r−1(N) ∩ (W1 V1 \ W0V0). Since s(γn) ∈ An and s(γn) → s(γ), we must have V1 \ W0V0). Since W1 7 7 7 1We are retaining the notations of [19] except we have dropped the fraktur font for groupoids and written G in place of G for clarity. This is more of an issue in [20] where our readers have been frustrated trying to distinguish between G, S and E -- rather than between G, S and E. 2A similar restriction was required in [29] -- the function f 1 x defined on the bottom of [29, p. 61] is only well-defined on U0 (even though I failed to mention this). This is reflected in the statement of [29, Lemma 4.4]. THE REPRESENTATION THEORY OF C ∗-ALGEBRAS ASSOCIATED TO GROUPOIDS 25 s(γ) = z. Since γ /∈ W0V0, we have γ ∈ Gz \ W0z. Since Fz ⊆ W0z, our construction of Fz forces r(γ) /∈ N. But this is a contradiction. This completes the proof of the Lemma. (cid:3) Now, if γ ∈ GA, then (A.3) g(1)(γ) :=(g(cid:0)r(γ)(cid:1) 0 7 if γ ∈ W1 if γ /∈ W0V0W0 V1 W1 7 and is a well defined function on GA. Consequently (A.3) defines an element of Cc(GA). We can use the Tietze-Extension Theorem to extend g(1) to an element of Cc(G) provided we keep in mind that (A.3) holds only for γ ∈ GA. Next, we must modify [19, Lemma 2.8] to hold only near z; specifically, we have the following. Lemma A.2. With the choices above, provided γ, α ∈ GA. g(cid:0)r(γ)(cid:1)g(cid:0)r(α)(cid:1)b(γα−1)g(1)(α) = g(1)(γ)g(cid:0)r(α)(cid:1)g(1)(α) Then with the given restriction on γ and α, the proof of Lemma A.2 goes through as written in [19]. Now it is straightforward to check that the rest of the proof of [19, Theorem 2.3] goes through with the observation that (1) we only need consider the rep- resentations Lu with u close to z, and that (2) Lu acts on L2(Gu, λu). This allows us to apply Lemma A.2 at the appropriate time. Remark A.3. The existence of the neighbourhoods V0 and V1 such that (A.1) holds is used in the proof of [20, Theorem 2.3] (see page 237 of [20]) and in the proof of [6, Proposition 4.1]; both results are saved by Lemma A.1 since we can restrict to elements with source sufficiently close to z. References [1] R.J. Archbold and A. an Huef, Strength of convergence in the orbit space of a transformation group, J. Funct. Anal. 235 (2006), 90 -- 121. [2] R.J. Archbold and A. an Huef, Strength of convergence and multiplicities in the spectrum of a C ∗- dynamical system, Proc. London Math. Soc. 96 (2008), 545 -- 581. [3] R.J. Archbold and D.W.B. Somerset, Transition probabilities and trace functions for C ∗-algebras, Math. Scand. 73 (1993), 81 -- 111. [4] L.O. Clark, Classifying the type of principal groupoid C ∗-algebras, J. Operator Theory 57 (2007), 251 -- 266. [5] L.O. Clark, CCR and GCR Groupoid C ∗-algebras, Indiana Univ. Math. J. 56 (2007), 2087 -- 2110. [6] L.O. Clark and A. an Huef, Principal groupoid C ∗-algebras with bounded trace, Proc. Amer. Math. Soc. 136 (2008), 623 -- 634. [7] J. Dixmier, C ∗-algebras, North-Holland, New York, 1977. [8] S. Echterhoff, On transformation group C ∗-algebras with continuous trace, Trans. Amer. Math. Soc. 343 (1994), 117 -- 133. [9] M. Ephrem, Characterizing liminal and type I graph C ∗-algebras, J. Operator Theory 52 (2004), 303 -- 323. [10] J. Glimm, Locally compact transformation groups, Trans. Amer. Math. Soc. 101 (1961), 124 -- 128. [11] E. C. Gootman, The type of some C ∗- and W ∗-algebras associated with transformation groups, Pacific J. Math. 48 (1973), 93 -- 106. [12] P. Green, C ∗-algebras of transformation groups with smooth orbit space, Pacific J. Math. 72 (1977), 71 -- 97. 26 CLARK AND AN HUEF [13] R. Hazlewood, Continuous trace, Fell, bounded trace, liminal and postliminal graph algebras, in preparation. [14] R. Hazlewood and A. an Huef, The strength of convergence in the orbit space of a groupoid, preprint, arXiv:1006.3115. [15] A. an Huef, The transformation groups whose C ∗-algebras are Fell algebras, Bull. London Math. Soc. 33 (2001), 73 -- 76. [16] A. an Huef, Integrable actions and the transformation groups whose C ∗-algebras have bounded trace, Indiana Univ. Math. J. 51 (2002), 1197 -- 1233. [17] A. Kumjian, On C ∗-diagonals, Canad. J. Math. 38 (1986), 969 -- 1008. [18] A. Kumjian, D. Pask, I. Raeburn and J. Renault, Graphs, groupoids and Cuntz-Krieger algebras, J. Funct. Anal. 144 (1997), 505 -- 541. [19] P.S. Muhly and D.P. Williams, Continuous trace groupoid C ∗-algebras, Math. Scand. 66 (1990), 231 -- 241. [20] P.S. Muhly and D.P. Williams, Continuous trace groupoid C ∗-algebras II, Math. Scand. 70 (1992), 127 -- 145. [21] P.S. Muhly, J. Renault and D.P. Williams, Continuous trace groupoid C ∗-algebras, III, Trans. Amer. Math. Soc. 348 (1996), 3621 -- 3641. [22] G.K. Pedersen, C ∗-algebras and their automorphism groups, Academic Press, London, 1979. [23] I. Raeburn and D.P. Williams, Morita equivalence and continuous-trace C ∗-Algebras, Math. Surveys and Monographs, vol. 60, Amer. Math. Soc., Providence, 1998. [24] A. Ramsay, The Mackey-Glimm dichotomy for foliations and other Polish groupoids, J. Funct. Anal. 94 (1990), 358 -- 374. [25] J. Renault, A groupoid approach to C ∗-algebras, Lecture Notes in Mathematics, No. 793, Springer- Verlag, New York, 1980. [26] J. Renault, Repr´esentations des produits crois´es d'alg`ebres de groupoides, J. Operator Theory 18 (1987), 67 -- 97. [27] J. Renault, The ideal structure of groupoid crossed product C ∗-algebras, J. Operator Theory 25 (1991), 3 -- 36. [28] D.P. Williams, The topology on the primitive ideal space of transformation group C ∗-algebras and CCR transformation group C ∗-algebras, Trans. Amer. Math. Soc. 266 (1981), 335 -- 359. [29] D.P. Williams, Transformation group C ∗-algebras with continuous trace, J. Funct. Anal. 41 (1981), 40 -- 76. [30] D.P. Williams, Crossed products of C ∗-algebras, Math. Surveys and Monographs, vol. 134, Amer. Math. Soc., Providence, 2007. Dept of Mathematical Sciences, Susquehanna University, Selinsgrove, PA 17870, USA E-mail address: [email protected] Department of Mathematics and Statistics, University of Otago, PO Box 56, Dunedin 9054, New Zealand E-mail address: [email protected]
1809.04932	1	1809	2018-09-13T13:18:28	Higher rank graphs, k-subshifts and k-automata	[ "math.OA", "math.DS" ]	Given a $k$-graph $\Lambda $ we construct a Markov space $M_\Lambda $, and a collection of $k$ pairwise commuting cellular automata on $M_\Lambda $, providing for a factorization of Markov's shift. Iterating these maps we obtain an action of ${\mathbb N}^k$ on $M_\Lambda $ which is then used to form a semidirect product groupoid $M_\Lambda \rtimes {\mathbb N}^k$. This groupoid turns out to be identical to the path groupoid constructed by Kumjian and Pask, and hence its C-algebra is isomorphic to the higher rank graph C-algebra of $\Lambda $.	math.OA	math	Higher rank graphs, k-subshifts and k-automata R. Exel∗ and B. Steinberg∗∗ Given a k-graph Λ we construct a Markov space MΛ, and a collection of k pairwise commuting cellular automata on MΛ, providing for a factorization of Markov's shift. k on MΛ which is then used to form Iterating these maps we obtain an action of k. This groupoid turns out to be identical a semidirect product groupoid MΛ ⋊ to the path groupoid constructed by Kumjian and Pask, and hence its C-algebra is isomorphic to the higher rank graph C-algebra of Λ. 1. Introduction. Given a row-finite k-graph Λ (see [4]), Kumjian and Pask have constructed a path space k on Λ∞ such that the C-algebra of the path groupoid Λ∞ k is Λ∞ and an action of canonically isomorphic to the corresponding higher rank graph C-algebra C ∗(Λ). ⋊ Recall from [4] that a path in Λ consists of a map x : Ωk → Λ, where Ωk =(cid:8)(n, m) ∈ k × k, n ≤ m(cid:9), satisfying suitable conditions. In the paragraph after [4: Remarks 2.2], the authors observe that each path x in Λ∞ is uniquely determined by a very small subset of its values, such as, for example, the values of the form y(n) = x(cid:0) (n, n, . . . , n) , (n+1, n+1, . . . , n+1)(cid:1), for every n ∈ . Noting that each y(n) above is an element of Λ of degree d(cid:0)y(n)(cid:1) = (1, 1, . . . , 1), we consider the subset Σ of Λ formed by all elements possessing the above degree. Viewing Σ as an alphabet, in the spirit of Symbolic Dynamics, one may easily see that each y given above is in fact an element of a certain Markov subspace MΛ ⊆ Σ, the correspondence x → y in fact being a homeomorphism from Λ∞ to MΛ. We thus have two homeomorphic spaces, each carrying an action of a different monoid, k in case of Λ∞, while acts on MΛ by means of iterating Markov's shift. namely of These actions are compatible in the sense that the above correspondence is covariant for the action of the submonoid (cid:8)(n, n, . . . , n) : n ∈ (cid:9) ⊆ k, ∗ Universidade Federal de Santa Catarina and University of Nebraska -- Lincoln. ∗∗ The City University of New York. 1 on Λ∞, but one may also use the above homeomorphism to extend the action of on MΛ k. Alternatively, considering the action of the to an action of the much larger monoid canonical basis vectors ei of k on Λ∞, we may define continuous maps Si : MΛ → MΛ, for i = 1, . . . , k, which commute among themselves, giving a factorization of Markov's shift S on MΛ, in the sense that S1S2 · · · Sk = S. The well known Curtis-Hedlund-Lyndon Theorem in fact states that, when the alpha- bet is finite, every continuous map commuting with the shift on Σ is a cellular automaton, given by means of a sliding block code. Regardless of the size of our alphabet, we indeed show that the Si above are given by sliding block codes closely linked to the unique fac- torization property of Λ. Motivated by this example, we introduce the notion of a weak k-automaton over a given alphabet Σ, as being a k+1-tuple (Y ; S1, S2, . . . , Sk), where Y is a classical subshift (i.e., a closed subset of Σ, invariant under the shift S), and the Si are pairwise commuting continuous maps from Y to Y , providing for a factorization of the shift. ⋊ Since the path groupoid Λ∞ k may be built from nothing more than the information k on the path space, one sees that this groupoid is identical in contained in the action of all respects to the groupoid constructed from the associated weak k-automaton, and hence that the higher rank graph C-algebra may be constructed solely based on the the latter. As an auxiliar gadget we also define a notion of a k-subshift, as being a closed subset k, and which is isomorphic to its image of Σ under the restriction to the diagonal. See (2.1) for the precise definition. The relevance of this notion resides in the fact that it has a more geometrical appeal, while being essentially the same thing as a weak k-automaton, as proved in (7.4). , invariant under the natural action of k The adjective "weak" above is nothing but a disclaimer highlighting the fact that the Si involved are not actually supposed to be cellular automata, although they share with the latter the important property of commuting with the shift. In (7.9) we then improve on (7.4) by precisely characterizing the k-subshifts giving rise to weak k-automata involving actual cellular automata. The first named author was partially supported by CNPq. The second named author thanks the Fulbright Commission for supporting his recent visit to Florianopolis during which part of the research for this paper was conducted. 2. P -subshifts. Let Σ be a set, henceforth called the alphabet, viewed as a topological space with the discrete topology. 2 Given a monoid P , we will consider the set ΣP equipped with the product topology. Although we will not assume that Σ is finite here, we observe that when Σ is finite, then ΣP is a compact space by Tychonov's Theorem. For each p in P , we will moreover denote by θp : ΣP → ΣP , the map given by θp(ξ) t = ξ(pt), ∀ ξ ∈ ΣP , ∀ t ∈ P. It is then easy to prove that θ is a right action of P on ΣP , that is, θpθq = θqp, ∀ p, q ∈ P, henceforth referred to as the Bernoulli action, or the full P -shift, on the alphabet Σ. ◮ From now on the alphabet Σ and the semigroup P will be considered fixed. 2.1. Definition. A P -subshift is any closed subset X ⊆ ΣP which is invariant under the Bernoulli action in the sense that θp(X) ⊆ X, for every p in P . We will next describe an important source of P -subshifts given in terms of forbidden patterns. 2.2. Definition. By a pattern we shall mean a pair (π, Dπ), where Dπ is a finite subset of P , and π : Dπ → Σ is any function. We shall frequently refer to π as a pattern without mentioning Dπ explicitly. Given a pattern π and an element ξ in ΣP , we will say that π occurs in ξ, provided there exists some p0 in P such that π(t) = ξ(p0t), for all t in Dπ. 2.3. Proposition. If the pattern π occurs in θp(ξ), for some ξ in ΣP , then π occurs in ξ. Proof. By hypothesis there exists p0 in P such that π(t) = θp(ξ) p0t = ξ(pp0t), ∀ t ∈ Dπ, whence the conclusion. (cid:3) Given a pattern π, it is easy to see that the set of all ξ in ΣP such that π occurs in ξ is open in ΣP . 2.4. Proposition. Given a collection Π of patterns, let XΠ be the set of all elements ξ in ΣP such that no pattern in Π occurs in ξ. Then XΠ is a P -subshift. Proof. XΠ is closed by the observation made just before the statement, and it is invariant under the Bernoulli action by the contrapositive of (2.3). (cid:3) The following is a well known result in the theory of classical subshifts. It is usually stated for finite alphabets, but it works just as well for infinite ones. 2.5. Proposition. If X is any P -subshift then there exists a collection Π of patterns such that X = XΠ. 3 Proof. Let Π be the collection of all patterns which do not occur in any ξ in X. It is then obvious that X ⊆ XΠ, and we next claim that X is dense in XΠ. To see this, choose any η in XΠ, and let V be any open subset of ΣP containing η. Since ΣP has the product topology, there exists a finite set D ⊆ P such that η ∈ W := {ζ ∈ ΣP : ζ(t) = η(t), for all t ∈ D} ⊆ V. Setting π = ηD, we have that π is a pattern, obviously occuring in η, whence π is certainly not in Π. By definition of Π, it follows that π occurs in some ξ in X, so there exists p0 in P such that π(t) = ξ(p0t) = θp0 (ξ) , t for every t in D. This says that θp0(ξ) ∈ W ∩ X, proving the desired density, namely that XΠ ⊆ X. Since X is closed by hypothesis, the proof is concluded. (cid:3) If Q is a submonoid of P , always assumed to share the neutral element, we may consider the restriction mapping ρQ from ΣP to ΣQ, namely ρQ(ξ) = ξQ, ∀ ξ ∈ ΣP . (2.6) Clearly ρQ is a continuous map. 2.7. Proposition. Let X ⊆ ΣP be a P -subshift, and let Q be a submonoid of P . Then (i) ρQ(X) is invariant under the Bernoulli action of Q, (ii) if Σ is finite, then ρQ(X) is a Q-subshift. Proof. Letting θ′ denote the full Q-shift, observe that, for all q in Q, and all ξ in X, one has that θ′ q(ξQ) = θq(ξ)Q, from where it easily follows that ρQ(X) is invariant under the Bernoulli action of Q. Assum- ing that Σ is finite, we have that ΣP is compact, whence so is X. Observing that ρQ is con- tinuous, we see that ρQ(X) is compact, hence closed in ΣQ. This concludes the proof. (cid:3) We will soon discuss an important class of examples in which ρQ(X) is closed, even It will then follow from (2.7.i) that ρQ(X) is a though the alphabet might be infinite. Q-subshift. Incidentally, we do not have any example in which ρQ(X) fails to be closed. 3. Cellular Automata. As before we let Σ be any set, which we view as a discrete topological space. From now on we shall be concerned with metric aspects, most notably with the notion of uniform continuity, so we shall equip Σ with the metric defined by d(a, b) =( 0, if a = b, 1, otherwise, for all a and b in Σ. The topology on Σ induced by this metric is clearly the discrete topology. For this reason d is sometimes called the discrete metric. However, while many 4 other metrics on Σ also induce the discrete topology, we observe that d induces the discrete topology in a uniform way, meaning that there exists r > 0, namely r = 1/2 such that for every a in Σ, the ball centered at a with radius r coincides with the singleton {a}. The fact that r does not depend on a is what makes d a uniformly discrete metric. In this section we shall be concerned with the monoid , formed by all natural num- bers, including zero, and hence we will be working with -subshifts, also known simply as subshifts. The most popular metric considered on the product space Σ (as usual also denoted by d, by abuse of language) is as follows: given x and y in Σ, one puts d(x, y) = 2−k, where k is the largest integer such that xi = yi, for all i ≤ k. If x = y, then obviously no such k exists, in which case we set d(x, y) = 0. It is well known that d defines a metric on Σ, which is compatible with the product topology. The role of uniform continuity is evidenced by our next result. 3.1. Lemma. (i) Each projection pk : Σ → Σ is uniformly continuous. (ii) For every nonempty X ⊆ Σ, and for every uniformly continuous map ϕ : X → Σ, one has that ϕ depends only on finitely many coordinates, meaning that there exists some k ∈ , and a map ψ : Σk+1 → Σ such that ϕ(x) = ψ(x0, x1, . . . , xk), ∀ x ∈ X. Proof. Regarding the projection pk, and given ε > 0, choose δ = 2−k. Then, for every x and y in Σ, with d(x, y) ≤ δ, we necessarily have that xi = yi, for all i ≤ k, hence proving (i). With respect to (ii), let δ > 0 be such that d(cid:0)pk(x), pk(y)(cid:1) = d(xk, yk) = 0 < ε, d(x, y) < δ ⇒ d(cid:0)ϕ(x), ϕ(y)(cid:1) < 1/2, for every x, y ∈ Σ, and choose an integer k such that 2−k < δ. Given x and y in Σ such that (x0, x1, . . . , xk) = (y0, y1, . . . , yk), we then have that d(x, y) ≤ 2−k < δ, so d(cid:0)ϕ(x), ϕ(y)(cid:1) < 1/2, which clearly implies that ϕ(x) = ϕ(y), since Σ has the 0-1 metric. This proves that ϕ(x) depends only on (x0, x1, . . . , xk). (cid:3) Observe that when Σ is finite and X is closed in Σ, then X is compact by Tychonov's Theorem, so every continuous function on X is necessarily uniformly continuous. Conse- quently the conclusion of (3.1.ii) holds for every continuous function ϕ. As usual, we denote by S the shift on Σ, defined by S(x0, x1, x2, . . .) = (x1, x2, x3, . . .), ∀ x = (x0, x1, x2, . . .) ∈ Σ, so that a closed subspace X ⊆ Σ is an -subshift if and only if X is invariant under S in the sense that S(X) ⊆ X. 5 3.2. Definition. Let X ⊆ Σ be a subshift. (i) The language of X, denoted L(X), is the set of all finite words occurring as a con- tiguous block of characters in some x in X. (ii) Given k ∈ , the subset formed by all words in L(X) of length k + 1 will be denoted by Lk(X). (iii) By a (sliding block) code for X we mean any function ψ : Lk(X) → Σ. The integer k is called the anticipation of ψ. (iv) Given a code ψ, we define Tψ : (x0, x1, . . .) ∈ X 7→ (y0, y1, . . .) ∈ Σ, where yn = ψ(xn, xn+1, . . . , xn+k), and k is the anticipation of ψ. One says that Tψ is the cellular automaton associated to the code ψ. Another reason for our interest in uniform continuity is in order: 3.3. Proposition. Let X ⊆ Σ be a subshift and let ψ : Lk(X) → Σ be a code for X. The the cellular automaton Tψ is a uniformly continuous map. Proof. Given ε > 0, choose p such that 2−p < ε. For x and x′ in X, let y = Tψ(x), and y′ = Tψ(x′). If d(x, x′) ≤ 2−(p+k), we have that xi = x′ i, for all i ≤ p + k, from where one deduces that yi = y′ i, for all i ≤ p, whence d(y, y′) ≤ 2−p < ε. This proves that Tψ is uniformly continuous. (cid:3) If ψ is defined on Σ2 by ψ(a, b) = b, then Tψ is clearly the shift itself. From the above we then deduce the elementary fact that the shift is uniformly continuous. Given a subshift X and a code ψ for X, it easy to prove that Tψ(cid:0)S(x)(cid:1) = S(cid:0)Tψ(x)(cid:1), ∀ x ∈ X. (3.4) The next result, known as the Curtis-Hedlund-Lyndon Theorem, says that the only uniformly continuous maps commuting with the shift are the cellular automata. The corresponding result for the bilateral shift is discussed in [5: 13.9]. 3.5. Proposition. Let X be a subshift and let T : X → X be a continuous mapping which commutes with S. Then T is uniformly continuous if and only if T is a cellular automaton. 6 Proof. The "if" part having already been dealt with in (3.3), we move on to the "only if" part. Consider the function ϕ : X → Σ given by ϕ = p0 ◦ T , where p0 is the projection on the leftmost coordinate. By (3.1.i) we have that ϕ is uniformly continuous, and hence p0(cid:0)T (x)(cid:1) = ψ(x0, x1, . . . , xk), ∀ x ∈ X, for some k and some ψ : Σk+1 → Σ, by (3.1.ii). For every x in X it then follows that pn(cid:0)T (x)(cid:1) = p0(cid:0)Sn(T (x))(cid:1) = p0(cid:0)T (Sn(x))(cid:1) = = p0(cid:0)T (xn, xn+1, . . .)(cid:1) = ψ(xn, xn+1, . . . , xn+k). Noticing that (xn, xn+1, . . . , xn+k) necessarily lies in Lk(X), and upon restricting ψ (cid:3) to Lk(X), we then have that T is the cellular automaton associated to ψ. 4. Higher rank graphs. Given any integer k ≥ 1, let Λ be a k-graph (see [4]). That is, Λ is a (small) category k, satisfying the unique factorization property, equipped with a degree functor d: Λ → namely, if d(λ) = m + n, then there are unique α, β ∈ Λ with λ = αβ and d(α) = m and d(β) = n. Recall that Ωk is the k-graph consisting of all pairs (m, n) ∈ k such that m ≤ n, equipped with the degree map defined by d(m, n) = n − m and with allowed products (n, r)(m, n) = (m, r). By definition, a path in Λ is a functor from Ωk to Λ, compatible with the degree maps. The set of all paths in Λ is denoted by Λ∞. k × Recall from [4: Definitions 2.1] that, for each p in k, one defines σp : Λ∞ → Λ∞, by σp(x)(m, n) = x(p + m, p + n), ∀ x ∈ Λ∞, ∀ (m, n) ∈ Ωk. In what follows we would like to relate the path space Λ∞ to an k-subshift. In order to do so we begin by introducing the notation 1 := (1, 1, . . . , 1), which we use in defining our alphabet Σ, by Σ =(cid:8)λ ∈ Λ : d(λ) = 1(cid:9). Given any path x ∈ Λ∞, we may consider the element ξx of Σ k , defined by ξx(n) = x(n, n+1), ∀ n ∈ k, 7 ........................... ........................... ........................... ........................... ........................... •......................................................................................... •......................................................................................... •......................................................................................... •......................................................................................... ξx(0,3) ξx(1,3) ξx(2,3) ξx(3,3) •......................................................................................... •......................................................................................... •......................................................................................... •......................................................................................... ξx(0,2) ξx(1,2) ξx(2,2) ξx(3,2) •......................................................................................... •......................................................................................... •......................................................................................... •......................................................................................... ξx(0,1) ξx(1,1) ξx(2,1) ξx(3,1) •......................................................................................... •......................................................................................... •......................................................................................... •......................................................................................... ξx(0,0) ξx(1,0) ξx(2,0) ξx(3,0) •......................................................................................... •......................................................................................... •......................................................................................... •......................................................................................... .........................................................................................•........................... .........................................................................................•........................... .........................................................................................•........................... .........................................................................................•........................... •........................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... A representation of the path ξx For every p in k, and every x in Λ∞, one easily checks that from where one deduces that the subset of Σ k , given by θp(ξx) = ξσp(x), (4.1) is invariant under the Bernoulli action of k. XΛ =(cid:8)ξx : x ∈ Λ∞(cid:9) 4.2. Proposition. The set XΛ introduced above is an dence k-subshift and the correspon- Ξ : x ∈ Λ∞ 7→ ξx ∈ XΛ is a homeomorphism. In addition, Ξ is covariant relative to the action σ on Λ∞ and the Bernoulli action on XΛ. Proof. Having already observed that XΛ is invariant under the Bernoulli action, we will next prove that XΛ is closed. We then pick any ξ in the closure of XΛ, so there exists a sequence {xi}i in Λ∞ such that {ξxi}i converges to ξ. For each j in , write j 1 for the element (j, j, . . . , j) of k, and put λj = ξ(j 1). We then claim that s(λj) = r(λj+1), for every j, where s and r refer to the source and range maps relative to the category Λ. To see this, let i0 be large enough, so that ξxi(j 1) = ξ(j 1), and ξxi (j 1+1) = ξ(j 1+1), for every i ≥ i0. Choosing any i ≥ i0, we then have that s(λj) = s(cid:0)ξ(j 1)(cid:1) = s(cid:0)ξxi (j 1)(cid:1) = s(cid:0)xi(j 1, j 1+1)(cid:1) = · · · 8 Recall that xi is a path, hence a functor from Ωk to Λ. Since the morphisms (j 1, j 1+1) and (j 1+1, j 1+21) may be composed in Ωk, we have that xi(j 1, j 1+1) and xi(j 1+1, j 1+21) may be composed in Λ, so the source of the former must coincide with the range of the latter, whence the above equals · · · = r(cid:0)xi(j 1+1, j 1 + 21)(cid:1) = r(cid:0)ξxi (j 1+1)(cid:1) = r(cid:0)ξ(j 1+1)(cid:1) = r(λj), thus proving our claim. By the paragraph after [4: Remarks 2.2], we conclude that there exists a path x in Λ∞ such that x(j 1, j 1+1) = λj, for all j, and we next claim that ξx = ξ. Given any n in k, choose some j in such that n ≤ j 1, and observe that λ0λ1 . . . λj = x(01, 1) x(1, 21) · · · x(j 1, j 1+1) = x(01, j 1+1) = x(01, n) x(n, n+1) x(n+1, j 1+1) = x(01, n) ξx(n) x(n+1, j 1+1). We next choose i0 large enough, so that ξxi coincides with ξ on {01, 1, . . . , j 1, n}, for every i ≥ i0. Therefore λ0λ1 . . . λj = ξ(01)ξ(1) . . . ξ(j 1) = ξxi(01) ξxi(1) . . . ξxi(j 1) = xi(01, 1) xi(1, 21) . . . xi(j 1, j 1+1) = xi(01, j 1+1) = xi(01, n) xi(n, n+1) xi(n+1, j 1+1). Contrasting our last two calculations, and invoking the unique factorization property, we deduce that ξx(n) = xi(n, n+1) = ξxi(n) = ξ(n). This concludes the proof of the claim according to which ξx = ξ, whence ξ lies in XΛ, and so we see that XΛ is closed. The last sentence of the statement has already been verified in (4.1), so we are finally left with the task of proving our correspondence Ξ to be a homeomorphism. In order to do this we first observe that Ξ is injective by the paragraph after [4: Remarks 2.2]. We will next prove that Ξ is an open mapping. For this, recall from [4: Definitions 2.4] that the topology of Λ∞ is generated by the cylinders, Z(λ) = {x ∈ Λ∞ : x(cid:0)0, d(λ)(cid:1) = λ}, as λ range in Λ. Given any y in Λ∞, and given any open set U ⊆ Λ∞ containing y, one may then find some λ such that y ∈ Z(λ) ⊆ U. 9 Choosing j such that j 1 ≥ d(λ), and setting µ = y(0, j 1), we then have that µ = y(0, j 1) = y(cid:0)0, d(λ)(cid:1) y(cid:0)d(λ), j 1(cid:1) = λ y(cid:0)d(λ), j 1(cid:1), from where we see that y ∈ Z(µ) ⊆ Z(λ), and we conclude that the cylinders of the form Z(µ), with d(µ) a multiple of 1, also form a basis for the topology of Λ∞. In order to prove that Ξ is an open map, it is therefore enough to verify that Ξ(cid:0)Z(µ)(cid:1) is open for every such µ. Given µ in Λ as above, i.e. with d(µ) = j 1, use the factorization property to write with d(λi) = 1, and observe that µ = λ0λ1 . . . λj−1, x ∈ Z(µ) ⇔ x(0, j 1) = λ0λ1 . . . λj−1 ⇔ x(i1, i1+1) = λi, ∀ i = 0, · · · , j−1 ⇔ ξx(i1) = λi, ∀ i = 0, · · · , j−1. Setting k V =(cid:8)ξ ∈ Σ k : ξ(i1) = λi, ∀i = 0, · · · , j−1(cid:9), which is clearly open in Σ , we then deduce that x ∈ Z(µ) if and only if ξx ∈ V . It then follows that Ξ(cid:0)Z(µ)(cid:1) = XΛ ∩ V , proving that Ξ is an open mapping, as claimed. Finally, leaving for the reader the easy task of verifying that Ξ is continuous, the proof (cid:3) is concluded. Having twice resorted to the paragraph after [4: Remarks 2.2], we nave not yet ex- hausted its consequences from our point of view. A further, and major consequence is the content of our next result. Identifying the monoid as a submonoid of k via the correspondence i ↔ i1, we will shortly refer to the restriction map ρ, introduced in (2.6). 4.3. Proposition. Let A = {Aλµ}λ,µ∈Σ be the matrix given by Aλµ =( 1, if s(λ) = r(µ), 0, otherwise, and let XA ⊆ Σ be the Markov space for A. Then ρ(XΛ) = XA, and ρ is a homeo- morphism from XΛ onto XA. Proof. We first observe that, for every x in Λ∞, one has ρ(ξx) =(cid:0)x(i1, i1+1)(cid:1)i∈ 10 , which is evidently in XA, so we see that ρ(XΛ) ⊆ XA. Given any λ = (λi)i∈ ∈ XA, the paragraph after [4: Remarks 2.2] produces a path x in Λ∞ such that x(i1, i1+1) = λi, for all i in , so that ρ(ξx) = λ, which in turn proves that ρ(XΛ) = XA. The uniqueness of the path x obtained above implies that ρ is one-to-one, so it remains to prove that ρ is a homeomorphism. Continuity not being an issue, we focus on proving continuity of the inverse map. In order to do so let us review the above construction of x from any given λ in XA: given any (m, n) ∈ Ωk, one chooses j such that j 1 ≥ n, and uses the unique factorization property to write λ0λ1 . . . λj−1 = µx(m, n)ν, with d(µ) = m, d(cid:0)x(m, n)(cid:1) = n − m, and d(ν) = j 1 − n. The resulting map x : Ωk → Λ is then a path satisfying ρ(ξx) = λ, whence ξx = ρ−1 (λ), and we must then prove that , all we need to ξx varies continuously with λ. Since ξx lives in the product space Σ k. Recalling that do is show that ξx(n) is continuous as a function of λ, for every n in ξx(n) = x(n, n+1), notice that the above recipe to construct x(n, n+1) depends only on (λ0, λ1, . . . , λj−1), where j is any integer such that j 1 ≥ n+1. The function producing ξx(n) from λ therefore factors as the composition k XA → Σj → Σ, where the leftmost arrow is the projection on the first j coordinates while the rightmost one corresponds to the above recipe producing x(n, n+1) from (λ0, λ1, . . . , λj). Since the projection is continuous and Σj carries the discrete topology, continuity follows. (cid:3) 5. Higher rank subshifts. Motivated by the example of the the following: k-subshift discussed in the previous section, we make k-subshift X on a given alphabet Σ such that 5.1. Definition. Let k be a positive integer. By a subshift of rank k, or a k-subshift, we shall mean a (i) the restriction map ρ is injective on X, (ii) ρ(X) is closed in Σ, and (iii) ρ is a homeomorphism from X to ρ(X). ........................... ........................... ........................... ........................... •......................................................................................... •......................................................................................... •......................................................................................... ξ(0,2) ξ(1,2) ξ(2,2) •......................................................................................... •......................................................................................... •......................................................................................... ξ(0,1) ξ(1,1) ξ(2,1) •......................................................................................... •......................................................................................... •......................................................................................... ξ(0,0) ξ(1,0) ξ(2,0) •......................................................................................... •......................................................................................... •......................................................................................... .........................................................................................•........................... .........................................................................................•........................... .........................................................................................•........................... •........................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ......................................................................................... ρ ....................... ................................................................................................................ ( ξ(0,0), ξ(1,1), ξ(2,2), ... ) 11 Since Markov spaces are automatically closed, we conclude from (4.3) that the k- subshift XΛ built from a k-graph Λ is an example of a k-subshift. Observe also that, in case the alphabet Σ is finite, then any k-subshift is compact, In other words, hence (5.1.ii) is automatically true, while (5.1.iii) follows from (5.1.i). when the alphabet is finite, (5.1.ii-iii) could be omitted from the above definition without any consequences. ◮ Let us now fix a subshift X of rank k on the alphabet Σ. Setting Y = ρ(X), we have by (2.7.i) that Y is invariant under the Bernoulli action of , that is, invariant under the usual shift map S : Σ → Σ, and since Y is also closed by assumption, we have that Y is a -subshift, that is, a classical subshift. A priori, it does not make sense to ask whether or not ρ is covariant, since the and Σ are not the same. But if we consider monoids acting on the Bernoulli spaces Σ only the smaller monoid, namely , then covariance clearly holds, and in particular k ρθ1 = Sρ, (5.2) as the reader may easily verify. Furthermore, since ρ is a homeomorphism from X to Y , the Bernoulli action of k on Y such that the diagram on X may be transfered to Y , so we get an action τ of k θn X .................................................................................................. ............ X ρ .................................................................................................. ............ .................................................................................................. ............ ρ τn Y .................................................................................................. ............ Y Diagram (5.3) commutes for every n in k. 5.4. Theorem. Let X be a k-subshift on the alphabet Σ, and put Y = ρ(X). Then: (i) Y is a classical subshift. (ii) For any integer i with 1 ≤ i ≤ k, let ei be ith canonical basis vector of k, and put Si = τei. Then the Si are pairwise commuting, continuous maps from Y to Y , and S1S2 · · · Sk = S, where S is the restriction of the shift map to Y (here denoted simply by S, by abuse of language). 12 Proof. The first point is an obvious consequence of the definitions and of (2.7.i). It was included here only for future reference. Regarding (ii) we have S1S2 · · · Sk = τ (e1) · · · τ (ek) = τ (e1 + · · · + ek) = τ (1) = ρθ1ρ−1 (5.2) = S. (cid:3) Given that the Si commute among themselves, it follows that Si also commutes with S, so this is turns out to be strongly related to (3.4), and hence also to cellular automata by (3.5). 6. Cellular automaton factorization of Markov subshifs associated to k-graphs. We have already mentioned that the space XΛ built from a k-graph Λ is a subshift of rank k. Moreover, by (4.3), we have that ρ(XΛ) is a Markov space. We may then use (5.4) to produce a factorization of Markov's shift and we will now show that each Si occuring in (5.4) is in fact a cellular automaton with anticipation 1. This could be obtained by (3.5), but we may in fact produce the block codes directly. We first observe that the language of a Markov subshift, such as ρ(XΛ), is governed by its matrix, and in particular L1(cid:0)ρ(XΛ)(cid:1) = {(λ0, λ1) ∈ Σ2 : s(λ0) = r(λ1)}. We therefore define, for each i = 1, . . . , k, as follows: given (λ, µ) in L1(cid:0)ρ(XΛ)(cid:1), we have that λµ ∈ Λ, and clearly d(λµ) = 1 + 1. Writing ϕi : L1(cid:0)ρ(XΛ)(cid:1) → Σ 1 + 1 = ei + 1 + (1 − ei), the unique factorization property allows us to write λµ = αβγ, where α, β and γ lie in Λ, d(α) = ei, d(β) = 1, and d(γ) = 1 − ei. We then set ϕi(λ, µ) = β. We will now show that each Si coincides with the cellular automaton relative to the sliding block code ϕi. In order to do this, observe that each Si is officially defined as Si = τei = ρθeiρ−1 . Given any y in ρ(XΛ), we may write y = ρ(ξ), for some ξ in XΛ, and we may further write ξ = ξx, for some x ∈ Λ∞. In other words, y = ρ(ξx). We then have for every j in , that Si(y) j = ρθeiρ−1 (y) = θeiρ−1 (y) j j1 = ρ−1 (y) = j1+ei = ξx(j 1 + ei) = x(j 1 + ei, j 1 + ei+1). (6.1) 13 On the other hand, notice that y(j) y(j+1) = ξx(j 1) ξx(j 1+1) = x(j 1, j 1+1) x(j 1+1, j 1+21) = x(j 1, j 1+21) = By the unique factorization property we then have that = x(j 1, j 1+ei) x(cid:0)j 1+ei, j 1+ei+1) x(j 1+ei+1, j 1+21). ϕi(cid:0)y(j), y(j+1)(cid:1) = x(j 1+ei, j 1+ei+1) (6.1) = Si(y) , j thus proving that indeed Si is the cellular automaton associated to the block code ϕi, as claimed. 7. Constructing k-subshifts from factorizations of the shift. Motivated by (5.4) we introduce the following concept: 7.1. Definition. A weak k-automaton over a given alphabet Σ is a k+1-tuple (Y ; S1, S2, . . . , Sk), where Y is a closed subset of Σ, invariant under the shift S (i.e., Y is a classical subshift), and the Si are pairwise commuting continuous maps from Y to Y such that S1S2 · · · Sk = S. As seen in (5.4), every k-subshift leads to a weak k-automaton, and it is our plan to show that k-subshifts are essentially the same thing as weak k-automata. As a first step let us show how to construct a k-subshift given a weak k-automata (Y ; S1, S2, . . . , Sk), which we consider fixed for the time being. k, put For each n = (n1, n2, . . . , nk) ∈ αn = Sn1 1 Sn2 2 · · · Snk k , so that α is an action of k on Y . We next let Ψ : Y → Σ k be defined by Ψ(y) n = αn(y) , 0 ∀ y ∈ Y, ∀ n ∈ k. 7.2. Proposition. Setting X = Ψ(Y ), one has that X is a subshift of rank k. In addition we have that ρ(X) = Y and, regarding the canonical action τ introduced in Diagram (5.3), one has that τei = Si, for every i = 1, . . . , k. Proof. Given y in Y , and given p and t in k, observe that θp(cid:0)Ψ(y)(cid:1) t = Ψ(y) = αt+p(y) 0 t+p = αt(cid:0)αp(y)(cid:1) 0 = Ψ(cid:0)αp(y)(cid:1) t , (7.2.1) so we deduce that θp(cid:0)Ψ(y)(cid:1) = Ψ(cid:0)αp(y)(cid:1), from where we see that the range of Ψ, also known as X, is invariant under the Bernoulli action of k. 14 In order to prove that ρ(X) = Y , it suffices to verify that ρ(Ψ(y)) = y, ∀ y ∈ Y, (7.2.2) but this follows from the following computation, where j ∈ : ρ(Ψ(y)) = Ψ(y) j j1 = αj1(y) 0 = Sj 1Sj 2 · · · Sj k(y) = Sj(y) 0 = y(j). 0 In order to prove that X is closed, suppose that {yi}i is a sequence in Y such that {Ψ(yi)}i converges to some x in Σ k . Then y := ρ(x) = lim i ρ(Ψ(yi)) (7.2.2) = lim yi, i so y ∈ Y , and we claim that x = Ψ(y). In fact, for every n in k, we have that x(n) = lim i Ψ(yi) = lim i n αn(yi) 0 = αn(y) = Ψ(y) , n 0 proving the claim. So far we have thus proven that X is a k-subshift. In order to show that it is a subshift of rank k, we must verify (5.1.i -- iii). By (7.2.2) we see that Ψ is one-to-one, and since it is onto X, by definition, it follows that Ψ is bijective. Employing (7.2.2) once more we deduce that ρX = Ψ−1, from where (5.1.i) follows. Noticing that both ρ and Ψ are clearly continuous, we obtain (5.1.iii), while (5.1.ii) follows from the facts that ρ(X) = Y , and that Y is closed by assumption. The last part of the statement may now be proved as follows: for y ∈ Y , and j ∈ , one has τei(y) j = ρθeiρ−1 (y) = θeiρ−1 (y) j j1 = θei Ψ(y) j1 (7.2.1) = Ψ(cid:0)αei(y)(cid:1) j1 = (cid:3) so τei = Si. = Ψ(cid:0)Si(y)(cid:1) j1 = αj1(Si(y)) 0 = Sj(cid:0)Si(y)(cid:1) 0 = Si(y) , j 7.3. Proposition. Suppose that X and X ′ are k-subshifts over the same alphabet Σ such that ρ(X) = ρ(X ′), and such that the canonical actions τ and τ ′ coincide. Then X = X ′. Proof. Given x ∈ X, and n ∈ k, we have seen in (5.3) that τnρ(x) = ρθn(x). Therefore x(n) = θn(x) = ρθn(x) = τnρ(x) . 0 0 01 This says that x may be reconstructed from ρ(x) together with the canonical action τ , from where the result follows. (cid:3) Summarizing our last two results we have: 15 7.4. Corollary. Given k ≥ 1 and an alphabet Σ, the corresponence establishes a one-to-one correspondence from the collection of all k-subshifts X ⊆ Σ onto the collection of all weak k-automata on the alphabet Σ. X 7→(cid:0)ρ(X); τe1, τe2, . . . , τek(cid:1) k The adjective "weak" employed in Definition (7.1) is meant to highlight the fact that the Si mentioned there are not actually cellular automata, although they share with the latter the important property of commuting with the shift, a property we saw in (3.5) to characterize true cellular automata in the uniformly continuous case. Nevertheless it is interesting to determine necessary and sufficient conditions on a given k-subshift for the maps Si in the weak k-automata associated to it by (7.4) to be as actual cellular automata. In order to do this we must first consider a metric on Σ follows. k agree on a given subset k , whenever x(n) = y(n), for all n in A. If p is the largest integer such that x and 7.5. Definition. We shall say that two elements x and y of Σ A ⊆ y agree on the subset k we put Bp :=(cid:8)n = (n1, n2, . . . , nk) ∈ k : ni ≤ p, for all i(cid:9), d(x, y) = 2−p, with the understanding that if x = y, then p = ∞, in which case d(x, y) = 0. As in the case of section (3), one proves that d is a metric on Σ , compatible with the product topology, and we may then speak of uniformly continuous functions defined, or taking values, in Σ . k k The above choice of the Bp is not so crucial except for the fact that the Bp form an k. Any other increasing sequence of finite subsets of , choice of finite subsets with these properties may also be used to define a metric on Σ . The common which in turn induce the same uniform structure [3: Chapter 6] on Σ underlying uniform structure is in fact what really matters here, and it says that two points x and y are close if and only if they agree on a large finite subset of k, whose union coincides with k. k k We shall however not make any explicit use of uniform structures in this work, beyond the elementary observation that d(x, y) ≤ 2−p ⇐⇒ x and y agree on Bp, (7.6) for every x and y in Σ k . 7.7. Proposition. The restriction map k ρ : Σ → Σ, defined in (2.6) is uniformly continuous. 16 Proof. Given x and y in Σ Therefore ρ(x) and ρ(y) obviously agree on {0, 1, . . . , p}, so , let p be the largest integer such that x and y agree on Bp. k This proves that ρ is in fact contractive, hence uniformly continuous. d(cid:0)ρ(x), ρ(y)(cid:1) ≤ 2−p = d(x, y). (cid:3) If X ⊆ Σ k is a k-subshift, then ρ is a homeomorphism on X by definition, so ρ−1 is continuous on ρ(X), although perhaps not uniformly. 7.8. Definition. (a) By a uniform k-subshift we shall mean a k-subshift X ⊆ Σ k such that the inverse of ρ is uniformly continuous on ρ(X). (b) By a k-automaton we shall mean a weak k-automaton (Y ; S1, S2, . . . , Sk) such that each Si is actually a cellular automaton. In what follows we will show that the two concepts just defined are related to each other by the same process involved in (7.4). be a k-subshift. Then X is a uniform k-subshift if and 7.9. Proposition. Let X ⊆ Σ only if its associated weak k-automaton (ρ(X); τe1, τe2, . . . , τek) is an actual k-automaton. k Proof. Recall from Diagram (5.3) that τn = ρθnρ−1 , k. Assuming that X is a uniform k-subshift, we have that ρ−1 for every n in is uniformly continuous. Leaving for the reader the easy task of proving that θn is also uniformly continuous, we deduce that τn is likewise uniformly continuous. Since we have already seen that τn commutes with the shift, we deduce from (3.5) that τn is a cellular automaton, and in particular so are the τei. This completes the proof of the "only if" part of the statement. Writing Si for τei, assume now that each Si is a cellular automaton, hence a uniformly continuous map. Setting Y = ρ(X), recall from the proof of (7.2) that the inverse of ρ on Y is the map Ψ given by Ψ(y) n = αn(y) , 0 ∀ y ∈ Y, ∀ n ∈ k, where αn = Sn1 1 Sn2 2 · · · Snk k , for each n = (n1, n2, . . . , nk) ∈ k. Since each Si, is uniformly continuous, the same is true for αn, so it follows from (3.1) that Ψ(y) depends only on finitely many coordinates of y. n Given ε > 0, choose p ∈ such that 2−p < ε. Based on the conclusion of the above paragraph, let q be a positive integer such that for every n ∈ Bp, one has that Ψ(y) depends only on n (y(0), y(1), . . . y(q)). 17 Setting δ = 2−q, assume that y, y′ ∈ Y are such that d(y, y′) ≤ δ. Then y and y′ agree on {0, 1, . . . , q}, whence Ψ(y) n = Ψ(y′) , n ∀ n ∈ Bp, and we then conclude that This shows that Ψ, and hence also ρ−1 that X is a uniform k-subshift. d(cid:0)Ψ(y), Ψ(y′)(cid:1) ≤ 2−p < ε. is uniformly continuous, which in turn says (cid:3) Given a k-graph Λ, recall from the paragraph immediately after (5.1), that the as- k-subshift XΛ built from Λ according to (4.2) is an example of a k-subshift. sociated Moreover, as seen in Section (6), the weak k-automaton (cid:0)ρ(XΛ); S1, . . . , Sk(cid:1) associated to XΛ via (7.4) is such that the Si are cellular automata with anticipation 1, and hence we in fact have a k-automaton, according to Definition (7.8.b). It then follows from (7.9) that XΛ is a uniform k-subshift. It would therefore be interesting to determine conditions on a given k-automaton (Y ; S1, . . . , Sk), which would imply that it comes from a k-graph in the sense that the k on Y is conjugate to one coming from the k-automaton arising associated action of from a k-graph. By (4.3), requiring Y to be a finite type subshift will likely be among these conditions. 8. Renault-Deaconu groupoids. A submonoid P of a discrete group G satisfying P −1P ⊆ P P −1 is called an Ore monoid. Given a right action α of P on a locally compact, Hausdorff, topological space X by means of local homeomorphisms, as in [2: Section 2] (see also [6]), one may build the corresponding Renault-Deaconu groupoid X ⋊α P =(cid:8)(x, g, y) ∈ X × G × X : ∃ n, m ∈ P, αn(x) = αm(y), g = nm−1(cid:9), also called the transformation, or semidirect product groupoid. By [2: 3.2], one has that X ⋊α P is an ´etale groupoid with the topology generated by the sets of the form U (n, m, A, B) =(cid:8)(x, nm−1, y) : x ∈ A, y ∈ B, αn(x) = αm(y)(cid:9), where A, B ⊆ X are open subsets, and n, m ∈ P . If X is a given subshift of rank k, we of course have a natural action of the Ore k on X, but this might not put us in the situation of the paragraph above since monoid k by local k-subshifts are not necessarily locally compact, and neither is the action of homeomorphisms. In fact, even when k = 1 and the alphabet is finite (in which case at least X is compact), namely in the case of a classical subshift, the shift map itself may 18 fail to be a local homeomorphisms; even worse, it may fail to be an open mapping. By [1 : 2.5], only subshifts of finite type are open. The k-subshift XΛ arising from a row-finite k-graph Λ fortunately does not suffer from such tribulations: the path space Λ∞ is locally compact by [4: Lemma 2.6], and each σp is a local homeomorphisms by [4: Remarks 2.5]. In view of the commutative diagram Ξ .............................................................................................. ............ Λ∞ ρ XΛ .................................................................................................. ............ MΛ σp .................................................................................................. ............ θp .................................................................................................. ............ .................................................................................................. ............ τp Λ∞ .............................................................................................. ............ Ξ XΛ .................................................................................................. ............ ρ MΛ where we write MΛ for ρ(XΛ), in which all vertical arrows are homeomorphisms, we then have that XΛ and MΛ are locally-compact spaces, and both θp and τp are local homeomorphisms. We may therefore form the Renault-Deaconu groupoids relative to the actions σ, θ and τ , obtaining the following three evidently isomorphic groupoids: Λ∞ ⋊σ , XΛ ⋊θ k, and MΛ ⋊τ k. The first one above has already explicitly appeared in [4: Definition 2.7], where it was called the path groupoid of Λ, while playing a prominent role given that its groupoid C-algebra is isomorphic [4: Corollary 3.5] to the higher rank graph C-algebra C ∗(Λ). Evidently we now see that C ∗(Λ) may also be modeled by the weak k-automaton MΛ. So let us formally state this as one of our main conclusions. 8.1. Theorem. Given a row-finite k-graph Λ, let Σ be the alphabet consisting of all morphisms λ with d(λ) = (1, 1, . . . , 1), and let A = {Aλµ}λ,µ∈Σ be the 0-1 matrix such that Aλµ = 1, if and only if s(λ) = r(µ). Then there are k pairwise commuting cellular automata S1, S2, . . . , Sk, whose product coincide with Markov's shift on the Markov space k on XA. Each Si is moreover a local homeomorphism and, denoting by τ the action of k XA obtained by iterating the Si, one has that the semidirect product groupoid XA ⋊τ k) and C ∗(Λ) is a model for the higher rank graph C-algebra in the sense that C ∗(XA ⋊τ are isomorphic. References [1] M. Dokuchaev and R. Exel, "Partial actions and subshifts", J. Funct. Analysis, 272 (2017), 5038 -- 5106. [2] R. Exel and J. Renault, "Semigroups of local homeomorphisms and interaction groups", Ergodic Theory Dynam. Systems, 27 (2007), 1737 -- 1771. [3] J. Kelley, "General topology", Springer-Verlag, 1975. [4] A. Kumjian and D. Pask, "Higher-rank graph C*-algebras", New York J. Math., 6 (2000), 1 -- 20 (electronic). [5] D. Lind and B. Marcus, "An introduction to symbolic dynamics and coding", Cambridge University Press, 1995. [6] J. Renault and S. Sundar, "Groupoids associated to Ore semigroup actions", J. Operator Theory, 73 (2015), no. 2, 491 -- 514. 19
1710.06257	2	1710	2018-03-03T15:56:01	Derivations and Spectral Triples on Quantum Domains II: Quantum Annulus	[ "math.OA" ]	Continuing our study of spectral triples on quantum domains, we look at unbounded invariant and covariant derivations in the quantum annulus. In particular, we investigate whether such derivations can be implemented by operators with compact parametrices, a necessary condition in the definition of a spectral triple.	math.OA	math	DERIVATIONS AND SPECTRAL TRIPLES ON QUANTUM DOMAINS II: QUANTUM ANNULUS SLAWOMIR KLIMEK, MATT MCBRIDE, AND SUMEDHA RATHNAYAKE Abstract. Continuing our study of spectral triples on quantum domains, we look at un- bounded invariant and covariant derivations in the quantum annulus. In particular, we investigate whether such derivations can be implemented by operators with compact para- metrices, a necessary condition in the definition of a spectral triple. 1. Introduction Derivations in operator algebras are natural objects to investigate from the point of view of noncommutative geometry. In [12] we studied unbounded rotationally invariant and co- variant derivations in the quantum disk. We classified such derivations and looked at their implementations in various Hilbert spaces obtained from the GNS construction with respect to an invariant state. The question when such implementations are operators with com- pact parametrices was answered and, surprisingly, we found out that no implementation of a covariant derivation in any GNS Hilbert space for a faithful normal invariant state has compact parametrices for a large class of reasonable boundary conditions. This paper (and its prequel) arose from an attempt to construct geometric even spectral It is motivated by the classical example of the two- triples on quantum plane domains. dimensional Dirac operator of the form: D =(cid:20) 0 D 0 (cid:21) , D∗ where D is the d-bar operator on a disk or an annulus in the complex plane, acting in L2 space of the domain. Thus, it is very natural in the noncommutative case to look for D as an implementation in some GNS Hilbert space of a covariant derivation in the underly- ing algebra, since there is no canonical analog of the d-bar operator, which is a covariant derivation, and since the choice of such D would guarantee that the commutator condition of the spectral triple is satisfied. Unfortunately, the results in this paper show that spectral triples cannot be defined in such a way. This is in contrast with classical analysis where for the Dirac operator above, subject to APS-like boundary conditions, the parametrices are compact. In noncommutative case however, no boundary conditions help. The key reason of the no-go theorem is the fundamental difference between classical differential operators and finite-difference operators with unbounded coefficients, the later appearing in quantum case. Nonetheless, since the assumptions of those constructions, while very natural, are not the most general possible, it remains an interesting challenge to construct meaningful geometric spectral triples for quantum domains in some other way. Date: March 6, 2018. 1 2 KLIMEK, MCBRIDE, AND RATHNAYAKE Here we look at unbounded invariant and covariant derivations in the quantum annulus, which is the C∗−algebra generated by a special weighted bilateral shift. This algebra was studied in detail in [10] and [11]. Much like the work presented in [12] we first classify such derivations and then look at their implementations in various Hilbert spaces obtained from the GNS construction with respect to an invariant state. However, even though the quantum annulus is singly generated, the classification is done differently than in [12] because the relations are more complicated and not so useful. We then answer the question when such implementations are operators with compact parametrices and thus can be used to define spectral triples. Like the quantum disk situation, no implementation of a covariant derivation in any GNS Hilbert space for a faithful normal invariant state has compact parametrices for a large class of boundary conditions. Unlike the quantum disk however, the technical reasons are quite different for why this fails. In [12], we were able to study Fourier components of an implementation of a covari- ant derivation and show that their spectral properties contradict the implementation having compact parametrices. The argument there followed similar calculations of the spectrum of the Ces`aro operator [4]. For the annulus the spectral argument for components is not con- clusive. The coefficients, rather than polynomially bounded as in the disk case, now have to be rapidly decaying and rapidly increasing at different infinities. In one case the components of an implementation fail to have decaying norms while in the other case the implement- ing operator has an infinite multiplicity eigenvalue. Thus, even though the conclusions are similar in the two papers, the techniques are quite different. Recall that if A is a Banach algebra and if A is a dense subalgebra of A then a linear map d : A → A is called a derivation if the Leibniz rule holds: d(ab) = ad(b) + d(a)b for all a, b ∈ A. The following example is of the d-bar operator on the annulus, and it is the motivating example for the rest of the paper. Let A = C(Ar) be the C∗-algebra of continuous functions on the annulus: 0 < r < 1. If A is the algebra of polynomials in z and ¯z then Ar := {z ∈ C : r ≤ z ≤ 1}, (da)(z) := ∂a(z) ∂ ¯z is an unbounded, closable derivation in A. Let ρθ : A → A be the one-parameter family of automorphisms of A given by the rotation z → eiθz on the annulus. Notice that ρθ : A → A. Moreover, d is a covariant derivation in A in the sense that it satisfies: The map τ : A → C given by d(ρθ(a)) = e−iθρθ(d(a)), a ∈ A. τ (a) = 1 π(1 − r2)ZAr a(z)d2z, is a ρθ-invariant, faithful state on A. The GNS Hilbert space Hτ , obtained using the state τ , is naturally identified with L2(Ar, d2z), the completion of A with respect to the usual inner DERIVATIONS AND SPECTRAL TRIPLES ON QUANTUM DOMAINS II: QUANTUM ANNULUS 3 product: kak2 τ = τ (a∗a) = 1 π(1 − r2)ZAr a(z)2d2z. The representation πτ : A → B(Hτ ) is given by multiplication: The unitary operator: πτ (a)f (z) = a(z)f (z). Uτ,θf (z) := f (eiθz) in Hτ implements ρθ in the sense that: Then the covariant operator: Uτ,θπτ (a)U −1 τ,θ = πτ (ρθ(a)). (Dτ f )(z) = ∂f (z) ∂ ¯z on domain Dτ = A ⊂ A ⊂ Hτ is an implementation of d in Hτ , i.e. [Dτ , πτ (a)] = πτ (d(a)), for all a ∈ A. The operator Dτ has an infinite dimensional kernel, so the operator (1 + D∗ τ Dτ )−1/2 is not compact. This is because the annulus is a manifold with boundary and we need to impose elliptic-type boundary conditions to make Dτ elliptic, so that it has compact parametrices. . On the other hand, let Dmin is infinite dimensional. There exist however closed operators Dτ with compact parametrices, such that: the closure of Dτ , hence there are no boundary conditions on Dmax be the closure of Dτ defined on C ∞ Denote by Dmax 0 (Ar). The cokernel of Dmin τ τ τ τ Dmin τ ⊂ Dτ ⊂ Dmax τ ; as in the disk case, examples are given by Atiyah-Patodi-Singer (APS) type boundary con- ditions, see [2]. A more general discussion of spectral triples for manifolds with boundary using operators with APS boundary conditions is contained in [1]. Additionally, recent paper [5] studies generalities of spectral triples on noncommutative manifolds with boundary. The paper is organized as follows. In section 2 we review the quantum annulus. Section 3 contains a classification of invariant and covariant derivations in the quantum annulus. In section 4 we classify invariant states on the quantum annulus and describe the correspond- ing GNS Hilbert spaces and representations. Section 5 describes implementations of those derivations in the GNS Hilbert spaces of section 4. Finally in section 6 we analyze when those implementations have compact parametrices. Let {Ek} be the canonical basis for ℓ2(Z) and let U be the bilateral shift, i.e. 2. Quantum Annulus We use the diagonal label operator: UEk = Ek+1. KEk = kEk, 4 KLIMEK, MCBRIDE, AND RATHNAYAKE so that, for a bounded function a : Z → C, we can write: a(K)Ek = a(k)Ek. We have the following crucial commutation relation for a diagonal operator a(K): a(K) U = Ua(K + 1). (2.1) Consider the following special weighted bilateral shift: UrEk =(cid:26) rEk+1 k < 0 Ek+1 k ≥ 0, 0 < r < 1. Let A = C ∗(Ur), the unital C∗−algebra generated by Ur. Reference [10] argues, with explicit details in [11], that this algebra can be thought of as a quantum annulus, see also [9]. Its structure is described by the following short exact sequence: 0 −→ K −→ A −→ C(S1) ⊕ C(S1) −→ 0, where K is the ideal of compact operators in ℓ2(Z). In fact K is the commutator ideal of the algebra A. The generators in this algebra are a bit difficult to work with, and we prefer to think of A as a crossed product. To see how this works we need the following observation. Lemma 2.1. U = 1 + r + r2 (1 + r)r Ur − 1 (1 + r)r UrU ∗ r Ur Additionally, the orthogonal projection P0 onto E0 can be written in terms of Ur and U ∗ r as: P0 = 1 1 − r2 [U ∗ r , Ur] . Proof. This is a straightforward calculation, applying both sides of those identities to the basis elements. (cid:3) As a direct consequence we get the following formulas: U −1 = U ∗ = 1 + r + r2 (1 + r)r U ∗ r − 1 (1 + r)r U ∗ r UrU ∗ r Pk = U kP0U −k 1 P≥0 = 1 − r2 (cid:0)U ∗ r Ur − r2I(cid:1) P<0 = I − P≥0, (2.2) where P≥0 is the orthogonal projection onto span{Ek}k≥0, P<0 is the orthogonal projection onto span{Ek}k<0 and Pk is the orthogonal projection onto Ek. We call a function a : Z → C eventually constant, if there exists a natural number k0 such that a(k) are constants for k ≥ k0. The smallest such k0 is called the domain constant. The set of all such functions will be denoted by c± 00(Z). Let P ol(Ur, U ∗ r ) be the set of all polynomials in Ur and U ∗ r and define A =(a =Xn U nan(K) : a(k) ∈ c± 00(Z), finite sum) . DERIVATIONS AND SPECTRAL TRIPLES ON QUANTUM DOMAINS II: QUANTUM ANNULUS 5 We have the following observation. Proposition 2.2. A = P ol(Ur, U ∗ Proof. It is clear that A is a ∗-subalgebra of A. We need to check A ⊂ P ol(Ur, U ∗ r ) and vice versa. Notice that Ur = Ua(K) where a(k) = r for k < 0 and a(k) = 1 for k ≥ 0. This function is obviously eventually constant so we have that Ur ∈ A. Thus P ol(Ur, U ∗ r ), the algebra of polynomials in generators Ur and U ∗ r . Next, by Lemma 2.1 we have U ∈ P ol(Ur, U ∗ r ). Moreover we have P0, U −1, Pk, P≥0, P<0 ∈ P ol(Ur, U ∗ 00(Z) with domain constant L and set a− = a(k) for k ≤ −L and a+ = a(k) for k ≥ L. We can decompose a(K) as follows: r ), from equations (2.2). Suppose that a(k) ∈ c± r ) ⊂ A. L−1 Notice that we have: and a(K) = a−P≤−L + Xk=−L+1 a(k)Pk + a+P≥L. (2.3) P≥L = P≥0 − P0 − P1 − . . . − PL−1 P≤−L = P<0 − P−1 − P−2 − . . . − P−L+1. It follows that P≥L and P≤−L are both in P ol(Ur, U ∗ a(K) ∈ P ol(Ur, U ∗ r ). Thus it follows that A ⊂ P ol(Ur, U ∗ r ) and consequently, by (2.3), we have (cid:3) r ), completing the proof. Let c(Z) be the space of convergent sequences, and consider the abelian algebra: Adiag = {a(K) : {a(k)} ∈ c(Z)} . It follows from the above proposition that this is precisely the subalgebra of all diagonal operators in A and we have: A = C ∗(Ur) = C ∗(Adiag, U). Additionally, because of formula (2.1), we can view the quantum annulus as the group crossed product of Adiag with Z acting on Adiag via shifts (translation by n ∈ Z), that is: Also, we see that (A, Adiag) is a Cartan pair [13]. A = Adiag ⋊shif t Z. 3. Derivations in the Quantum Annulus 3.1. Classification of derivations. For each θ ∈ [0, 2π), let ρθ : A → A be an au- tomorphism defined by ρθ(a(K)) = a(K) for a diagonal operator a(K), ρθ(U) = eiθU and ρθ(U ∗) = e−iθU ∗. It is well defined on all of A because it preserves the relations U ∗U = UU ∗ = I as well as (2.1). Alternatively, the action of ρθ can be written down using the label operator K as: It follows that ρθ : A → A. Any derivation d : A → A that satisfies the relation: ρθ(a) = eiθKae−iθK. ρθ(d(a)) = d(ρθ(a)) will be referred to as a ρθ-invariant derivation. Similarly, any derivation d : A → A that satisfies the relation d(ρθ(a)) = e−iθρθ(d(a)) 6 KLIMEK, MCBRIDE, AND RATHNAYAKE for all a ∈ A will be referred to as a ρθ-covariant derivation. Notice that we have the identifications: and similarly Adiag :=(cid:8)a(K) : {a(k)} ∈ c± 00(Z)(cid:9) = {a ∈ A : ρθ(a) = a}, Adiag = {a(K) : {a(k)} ∈ c(Z)} = {a ∈ A : ρθ(a) = a}. Additionally, we need the following sets and the identifications: and Acov := {a ∈ A : ρθ(a) = eiθa} = UAdiag, Acov := {a ∈ A : ρθ(a) = eiθa} = UAdiag. Below we use the following terminology: we say that a function β : Z → C has convergent increments, if the sequence of differences {β(k) − β(k − 1)} is convergent i.e. {β(k) − β(k − 1)} ∈ c(Z). The set of all such functions will be denoted by cinc(Z). Similarly the set of eventually linear functions is the set of β : Z → C such that: {β(k) − β(k − 1)} ∈ c± 00(Z). The following two propositions classify all invariant and covariant derivations d : A → A. Proposition 3.1. If d is a ρθ-invariant derivation d : A → A, then there exists a function β ∈ cinc(Z), which is unique modulo an additive constant, such that for a ∈ A. If d : A → A, then the corresponding function β(k) is eventually linear. d(a) = [β(K), a] Proof. Since d is ρθ-invariant we have that d : Adiag → Adiag, because If P ∈ Adiag is a projection then, applying d to P 2 = P gives: ρθ(d(a(K)) = d(ρθ(a(K))) = d(a(K)). d(P ) = d(P )P + P d(P ), or, equivalently, using commutativity of Adiag: (2P − 1)d(P ) = 0. This implies that d(P ) = 0, and in fact that d(a(K)) = 0 for every a ∈ Adiag because Adiag is linearly generated by projections. Next, consider the formula: ρθ(d(U)) = d(ρθ(U)) = eiθd(U). It says that d(U) ∈ Acov, and thus we must have d(U) = Ua(K) for some a(K) ∈ Adiag by the above identifications. Next we define β(k) so we can write: a(K) = β(K + 1) − β(K). DERIVATIONS AND SPECTRAL TRIPLES ON QUANTUM DOMAINS II: QUANTUM ANNULUS 7 In fact, the above formula determines β(K) uniquely up to a constant. Moreover if a(K) ∈ Adiag, then a(k) is eventually constant and so β(k) must be eventually linear. Then the commutation relation (2.1) implies that: d(U) = U(β(K + 1) − β(K)) = β(K)U − Uβ(K) = [β(K), U]. Using inductive reasoning we obtain: d (U n) = [β(K), U n] , then, using linearity and the decomposition of any a ∈ A, we obtain the result. (cid:3) Proposition 3.2. If d is a ρθ-covariant derivation d : A → A, then there exists a unique function β ∈ cinc(Z), such that d(a) = [Uβ(K), a] for all a ∈ A. If d : A → A then the corresponding function β(k) is eventually linear. Proof. Consider the following equation: ρθ(d(U ∗)) = eiθd(ρθ(U ∗)) = eiθe−iθd(U ∗) = d(U ∗), which says that d(U ∗) ∈ Adiag. Consequently, this implies that d(U ∗) = α(K) for some α(K) ∈ Adiag, which then gives: d(U) = −U 2α(K + 1). (3.1) Next we define a linear map d : A → A via the formula: d(a) := U −1d(a). Notice that we have ρθ( d(a)) = ρθ(U −1)ρθ(d(a)) = e−iθU −1eiθd(ρθ(a)) = d(ρθ(a)), which shows that d is ρθ-invariant. In particular this implies that d : Adiag → Adiag. Addi- tionally, d satisfies a twisted Leibniz rule of the form: d(a(K)b(K)) = d(a(K))b(K) + a(K + 1) d(b(K)). We claim that there is a diagonal operator β(K) such that: d(a(K)) = β(K)(a(K) − a(K + 1)). To construct β(k) we apply d to both sides of the equality P 2 ≥k = P≥k and obtain: (3.2) or, equivalently, d(P≥k) = d(P≥k)P≥k + P≥k−1 d(P≥k), P≤k−1 d(P≥k) = P≥k−1 d(P≥k). This implies that d(P≥k)(l) is zero unless l = k − 1. Define the sequence β(k) by: β(k − 1) = − d(P≥k)(k − 1). As a consequence we have: d(P≥k)(K) = −β(K)Pk−1(K) = β(K)(P≥k(K) − P≥k−1(K)) = β(K)(P≥k(K) − P≥k(K + 1)). Using the decomposition of a(K) as a sum of orthogonal projections P≥k we see that (3.2) holds for any a(K) ∈ Adiag. Also, using the relation d(a) = U −1d(a), we get d(a(K)) = U d(a(K)) = Uβ(K)(a(K) − a(K + 1)) = [Uβ(K), a(K)]. 8 KLIMEK, MCBRIDE, AND RATHNAYAKE Now we need to relate α and β. Applying d to the commutation relation (2.1) and using (3.1) and (3.2) yields: [Uβ(K), a(K)]U − a(K)U 2α(K + 1) = −U 2α(K + 1)a(K + 1) + U[Uβ(K), a(K + 1)], implying that for all a ∈ Adiag we have: (β(K + 1) − β(K) + α(K + 1))(a(K + 2) − a(K + 1)) = 0. We conclude that α(K) = β(K − 1) − β(K), thus d(U) = −U 2α(K + 1) = [Uβ(K), U]. Since α ∈ c(Z) and α(k) = β(k − 1) − β(k) we must have β ∈ cinc(Z). The result follows from the standard inductive reasoning in powers of U and linearity. (cid:3) 3.2. Structure of derivations. In [12] we studied when invariant derivations in the quan- tum disk are approximately bounded/approximately inner. This was motivated by the ques- tions arising in [3], [6], and [7]. We continue this discussion for the quantum annulus. Recall that d is called approximately inner if there are an ∈ A such that d(a) = lim n→∞ [an, a] for a ∈ A. If d(a) = limn→∞ dn(a) for bounded derivations dn in A then d is called ap- proximately bounded. Note also that any bounded derivation d on A can be written as a commutator d(a) = [a, x] with x in a weak closure of A, see [8], [14]. Recall also, from the proof of Proposition 3.1, that if d : A → A is an invariant derivation then d(a(K)) = 0 for every a(K) ∈ Adiag and so d is completely determined by its value on U. Lemma 3.3. Let d be a ρθ-invariant derivation in A with domain A. If d is approximately bounded then there exists a sequence {µn(k)} ∈ ℓ∞(Z) such that d(a) = lim n→∞ [µn(K), a] for all a ∈ A, i.e. d can be approximated by bounded ρθ-invariant derivations. Proof. Given an element a ∈ A we define its ρθ-average aav ∈ A by: aav := ρθ(a) dθ. 1 2πZ 2π 0 It follows that aav is ρθ-invariant since the Lebesgue measure dθ is translation invariant. Additionally, all ρθ-invariant operators in ℓ2(Z) are diagonal with respect to the basis {Ek} so that aav ∈ Adiag. Since by assumption d is approximately bounded, there exists a sequence of bounded operators bn such that: for all a ∈ A. It suffices to show that: d(a) = lim n→∞ [bn, a] lim n→∞ [(bn)av, a] = d(a), (3.3) DERIVATIONS AND SPECTRAL TRIPLES ON QUANTUM DOMAINS II: QUANTUM ANNULUS 9 since (bn)av is ρθ-invariant for every θ and hence by Proposition 3.1 it is given by the commu- tator with a diagonal operator µn(K) such that {µn(k)} ∈ ℓ∞(Z) because of the assumption of boundedness. By the remark before the statement of the lemma it is enough to verify (3.3) on the generator U. This computation is essentially identical to the similar one in the corresponding proof for the quantum disk in [12]. This completes the proof. (cid:3) The following results answer the question when is a ρθ-invariant derivation in d : A → A approximately inner/bounded. They are stated here without a proof because the only verifications needed are for the action of the derivations on U, and those are exactly the same as in the proofs in [12]. Proposition 3.4. Let d(a) = [β(K), a] be a ρθ-invariant derivation in A with domain A. If d is approximately bounded then {β(k) −β(k −1)} ∈ c0(Z), the space of sequences convergent to zero. We also have the following converse result. Proposition 3.5. If d(a) = [β(K), a] is a ρθ-invariant derivation in A with domain A such that {β(k) − β(k − 1)} ∈ c0(Z), then d is approximately inner. 4. Invariant States 4.1. Classification of states. Next we describe all the invariant states on A. If τ : A → C is a state, then τ is called a ρθ-invariant state on A if it satisfies: τ (ρθ(a)) = τ (a). The results in this section closely parallel similar analysis in [12]. Since A = Adiag ⋊shif t Z, there is a natural expectation E : A → Adiag, i.e. E is positive, unital and idempotent linear map defined, for a ∈ A, by the formula: E(a) = E Xn U nan(K)! = a0(K) and a0(K) ∈ Adiag. Since Adiag is the fixed point algebra for ρθ, we immediately obtain the following lemma: Lemma 4.1. Suppose τ : A → C is a ρθ-invariant state on A, then there exists a state t : Adiag → C such that τ (a) = t(E(a)) where E is the natural expectation. Conversely given a state t : Adiag → C, the formula τ (a) = t(E(a)) defines a ρθ-invariant state τ on A. To parametrize all invariant states we need to first identify the pure states. Lemma 4.2. The pure states on Adiag, denoted by tk for k ∈ N and t±∞, are given by: tk(a(K)) = a(k) = hEk, aEki, t±∞(a(K)) = lim k→±∞ a(k) = lim k→±∞ tk(a(K)). Proof. Adiag is a commutative C ∗−algebra that is isomorphic to the algebra of continuous functions on the two-point compactification of Z, that is Adiag ∼= C({−∞} ∪ Z ∪ {∞}). 10 KLIMEK, MCBRIDE, AND RATHNAYAKE So by general theory, see [8] for details, the pure states are the Dirac measures (or point mass measures). (cid:3) We have the following classification theorem of the ρθ-invariant states on A. Proposition 4.3. The ρθ-invariant states on A are in the closed convex hull of the states τk and τ±∞ where τk(a) = tk(E(a)) and τ±∞(a) = t±∞(E(a)). Explicitly, if τ is a ρθ-invariant state, there exists weights w(k) ≥ 0 satisfying: and non-negative numbers λ0 and λ±∞, with: w(k) = 1, Xk∈Z λ0 + λ∞ + λ−∞ = 1, such that: and also: In the formula above we have: w(k)τk. 0 τ (Pk), τ = λ∞τ∞ + λ−∞τ−∞ + λ0Xk∈Z λ0 =Xk∈Z λ−∞ = τ (P≤0) −Xk≤0 τ (Pk), w(k) = λ−1 τ (Pk), λ∞ = τ (P≥0) −Xk≥0 τ (Pk), where again Pk is the orthogonal projection onto the one dimensional subspace spanned by Ek and P≥0 is the orthogonal projection onto span{Ek}k≥0. Proof. This is a straightforward consequence of Lemma 4.2, for similar details see also [12]. (cid:3) Below we use the following notation with w(k) ≥ 0 such that Pk∈Z w(k) = 1: w(k)τk(a). τw(a) := tr(w(K)a) =Xk∈Z 4.2. GNS representations. Given a state τ on A let Hτ be the GNS Hilbert space and let πτ : A → B(Hτ ) be the corresponding representation. We describe the Hilbert spaces and representations coming from the following ρθ-invariant states: τw with all w(k) 6= 0, τ0, and τ±∞. The states τw with all w(k) 6= 0 are general ρθ-invariant faithful normal states on A. Proposition 4.4. The GNS Hilbert spaces with respect to the ρθ-invariant states τw with all w(k) 6= 0, τ0, and τ±∞ can be naturally identified with the following Hilbert spaces, respectively: (1) Hτw is the Hilbert space whose elements are power series such that U nfn(K) f =Xn∈Z kf k2 τw = τw(f ∗f ) =Xn∈ZXk∈Z w(k)fn(k)2 < ∞. (4.1) DERIVATIONS AND SPECTRAL TRIPLES ON QUANTUM DOMAINS II: QUANTUM ANNULUS 11 (2) Hτ0 (3) Hτ±∞ ∼= ℓ2(Z), πτ0 is the defining representation of A. ∼= L2(S1), πτ±∞(U) is the multiplication by eix, and πτ±∞(a(K)) is the multi- plication by the constant limk→±∞ a(k). Proof. The first Hilbert space is just the completion of A with respect to the inner product given by (4.1) and was introduced in [10]. It is the natural analog of the classical space of square-integrable functions L2(Ar) for the quantum annulus. The Hilbert space Hτ0 comes from the state τ0(a) = hE0, aE0i. To describe it we first need to find the algebra: A simple calculation yields: Aτ0 = {a ∈ A : τ0(a∗a) = 0}. thus, if τ0(a∗a) = 0, we get that an(0) = 0 for all n ∈ Z. Then, we have: τ0(a∗a) =Xn ∼=(a =Xn A/Aτ0 an(0)2, U nan(0)P0) , and let kak2 identify A/Aτ0 with a dense subspace of ℓ2(Z). τ0 := τ0(a∗a). Using the basis {En := U nP0} in A/Aτ0 for n ∈ Z, we can naturally It is easy to describe the representation πτ0 : A → B(Hτ0) of A in the bounded operators on Hτ0. We have: and πτ0(U)En = En+1, πτ0(a(K))En = a(K)U nP0 = U na(K + n)P0 = U na(n)P0 = a(n)En. Notice also that we have: A/Aτ0 ∋ [I] 7→ P0 := E0. In other words, πτ0 is the defining representation of the algebra A. Next we look at the GNS spaces associated with τ±∞(a) = limk→±∞hEk, aEki. If a(K) ∈ A, we set a±∞ = lim k→±∞ a(k). Again we want to find the subalgebra Aτ±∞ of a ∈ A such that τ±∞(a∗a) = 0. A direct computation shows that: so τ±∞(a∗a) = 0 if and only if an,±∞ = 0 for all n. Now A/Aτ±∞ can be identified with a dense subspace of L2(S1) by τ∞(a∗a) =Xn an,±∞2, A/Aτ±∞ ∋ [a] ="a =Xn We also have the formula: U nan,±∞# 7→Xn 2π Z 2π 1 0 τ±∞([a]) = fa(x) dx. an,±∞einx =: fa(x) ∈ L2(S1). 12 KLIMEK, MCBRIDE, AND RATHNAYAKE The representation πτ±∞ : A → B(Hτ±∞) is clearly seen to be given by: and πτ±∞(U)f (x) = eixf (x), πτ±∞(a(K))f (x) = a±∞f (x). This completes the proof. (cid:3) 5. Implementations of Derivations Let Hτ be the Hilbert space formed from the GNS construction on A using a ρθ-invariant state τ and let πτ : A → B(Hτ ) be the representation of A in the bounded operators on Hτ via left multiplication, that is πτ (a)f = [af ]. We have that [A] ⊂ Hτ is dense in Hτ and [I] ∈ Hτ is cyclic. Let Dτ be the subspace of Hτ given by: Dτ = πτ (A) · [I]. Then Dτ is dense in Hτ . Define Uτ,θ : Dτ → Dτ via Uτ,θ[a] = [ρθ(a)]. Notice that for every θ the operator Uτ,θ extends to a unitary operator in Hτ . Moreover, by direct calculation we get: Uτ,θπτ (a)U −1 τ,θ = πτ (ρθ(a)). It follows from the definitions that Uτ,θ(Dτ ) ⊂ Dτ and πτ (A)(Dτ ) ⊂ Dτ . An operator Dτ : Dτ → Hτ is called an implementation of a derivation d : A → A if [Dτ , πτ (a)] = πτ (d(a)). In this section we construct implementations of the derivations in A in the GNS Hilbert spaces: Hτw , Hτ0, and Hτ±∞. The results are completely analogous to those in [12], however we present here a more general approach to implementation questions. It should be noted that an implementation of a derivation may not exist. In fact, we have the following useful criterion. Proposition 5.1. Let A be a C ∗-algebra and let π : A → B(H) be its representation in some Hilbert space H. Let A ⊂ A be a dense subalgebra of A. Let d : A → A be a derivation and suppose that x0 is a cyclic vector for π(A). There is an operator D : π(A)x0 → H implementing d, i.e. for every a ∈ A and x ∈ π(A)x0, if and only if there is z ∈ H such that: [D, π(a)](x) = π(d(a))(x) if π(a)x0 = 0 then π(d(a))x0 = −π(a)z. Moreover, any implementation D of d is of the form: for z := D(x0) ∈ H satisfying the condition 5.1 above. D(π(a)x0) = π(d(a))x0 + π(a)D(x0), (5.1) (5.2) DERIVATIONS AND SPECTRAL TRIPLES ON QUANTUM DOMAINS II: QUANTUM ANNULUS 13 Proof. We first prove the sufficient condition. If there exists a D such that [D, π(a)] = π(da), then in particular we have [D, π(a)]x0 = π(da)x0. Suppose that π(a)x0 = 0, then π(da)x0 = D(π(a)x0) − π(a)D(x0) = −π(a)D(x0). with z = D(x0). This proves the sufficiency. On the other hand, suppose there exists such a z with the desired properties in the statement of the theorem. Define D by the formula above: D(π(a)x0) = π(da)x0 + π(a)z. We need to check that D is well-defined. If π(a)x0 = π(a)x0 then, by the properties of z, we have: Consequently, this gives: π(d(a − a))x0 = −π(a − a)z. D(π(a)x0) = π(d(a))x0 + π(a)z = π(d(a − a))x0 + π(a − a)z + π(d(a))x0 + π(a)z = −π(a − a)z + π(a − a)z + D(π(a)x0) = D(π(a)x0), and so the definition of D makes sense. Given a and b in A, consider the following calculation: [D, π(b)] π(a)x0 = D(π(b)π(a)x0) − π(b)D(π(a)x0) = d(π(ba))x0 + π(b)π(a)z − π(b)d(π(a))x0 − π(ba)z = d(π(b))π(a)x0 + π(b)d(π(a))x0 − π(b)d(π(a))x0 = d(π(b))π(a)x0. It shows that D is in fact an implementation of d, verifying the necessity and completing the proof. (cid:3) We want to apply the above proposition to our unital algebra A and GNS representations πτ coming from invariant states τ . The cyclic vector x0 can be chosen to be the class of the identity: x0 = [I]. Notice that, in condition (5.1) we have: {a ∈ A : πτ (a)x0 = 0} = {a ∈ A : τ (a∗a) = 0} = Aτ , and algebras Aτ were described in proofs in the previous section. 5.1. Invariant derivations. We first consider implementations of ρθ-invariant derivations. Let dβ be a ρθ-invariant derivation dβ : A → A, dβ(a) = [β(K), a], as described in Proposition 3.1. Since β(k) is defined up to a constant, we normalize it so that β(0) = 0. Definition: Dτ : Dτ → Hτ is called an invariant implementation of a ρθ-invariant deriva- tion dβ if [Dτ , πτ (a)] = πτ (dβ(a)) and Uτ,θDτ U −1 τ,θ = Dτ . Below we discuss invariant implementations of invariant derivations in A in various GNS Hilbert spaces. 14 KLIMEK, MCBRIDE, AND RATHNAYAKE Proposition 5.2. There exists a function α(k), satisfying: such that any invariant implementation Dβ,τw : Dτw → Hτw of dβ is given by: β(k) − α(k)2w(k) < ∞, Xk∈Z Proof. We start by computing Uτw,θ. From the definitions we have: Dβ,τwa = β(K)a − aα(K). Uτw,θ(a) =Xn U neinθan(K). Since τw is faithful we have that Aτw = 0, which implies tha z = Dβ,τw(I) is arbitrary, since both sides of condition (5.1) are equal to zero. However, it follows from the assumptions of invariant implementations that Dβ,τw(I) must be invariant with respect to Uτw,θ. This implies that Dβ,τw(I) = η(K) for some diagonal operator η(K) ∈ Hτw . Thus the formula (5.2) gives: Dβ,τwa = Dβ,τwπτw(a) · I = [β(K), a] + aη(K) = β(K)a − aα(K), where α(k) = β(k) − η(k). Notice also that η(K) ∈ Hτw implies that: kη(K)k2 τw =Xk∈Z β(k) − α(k)2w(k) < ∞. Thus the result follows. (cid:3) Proposition 5.3. There is a number c such that any invariant implementation Dβ,τ0 : Dτ0 → ℓ2(Z) of dβ is of the form: Proof. First we find Uτ0,θ. Since ρθ(U nP0) = einθU nP0, we have: Dβ,τ0 = c · I + β(K). Uτ0,θEn = einθEn. Because Dβ,τ0E0 must be invariant with respect to Uτ0,θ, we have Dβ,τ0E0 = cE0 for some constant c. We need to check that z = cE0 satisfies condition (5.1) to guarantee the existence of Dβ,τ0. In ℓ2(Z), [I] = E0, therefore πτ0(a)E0 = 0 if and only if an(0) = 0 for every n as in the proof of Proposition 4.4. We have: πτ0(d(a))E0 = β(K)aE0 − aβ(K)E0 = −β(0)aE0 = 0. Consequently, z has to satisfy az = 0 whenever aE0 = 0, which is clearly true for z = cE0. By using the formula (5.2) we get: Dβ,τ0En = [β(K), U n]E0 + cEn = (β(K) − β(K − n))En + cEn = (β(n) + c)En. Notice that the above Dβ,τ0 are the only implementations even if we don't assume their invariance. Indeed, consider any z satisfying condition (5.1). Define the following operator: aN = U −N δN (K) DERIVATIONS AND SPECTRAL TRIPLES ON QUANTUM DOMAINS II: QUANTUM ANNULUS 15 for N 6= 0, where δN (k) = 1 for k = N and zero otherwise. Clearly we have aN E0 = 0. Consequently, we obtain: 0 = hE0, aN zi = hEN , zi for all N 6= 0, thus z = cE0 for some constant c. This completes the proof. (cid:3) Proposition 5.4. There exists a number c such that all invariant implementations Dβ,τ±∞ : Dτ±∞ → L2(S1) of dβ are of the form: where Dβ,τ±∞ = β±∞ 1 i d dx + c, β±∞ := lim k→±∞ (β(k) − β(k − 1)) . Proof. Like in the other proofs we need to understand what is the value of Dβ,τ±∞ on [I], the constant function 1. A simple calculation shows that: (Uτ±∞,θf )(x) = f (x + θ). It is clear by the invariance properties that there exists a constant c such that: Notice that πτ±∞(a)1 = 0 if and only if an,±∞ = 0. Using the formula: Dβ,τ±∞(1) = c · 1. dβ(a) =Xn U n(β(K + n) − β(K))an(K), we see that if πτ±∞(a)1 = 0 then also πτ±∞(dβ(a))1 = 0. Consequently, z = c · 1 satisfies condition (5.1). Notice that Dτ±∞ is the space of trigonometric polynomials on S1. By linearity we only need to look at Dβ,τ±∞ on einx. We have: Dβ,τ±∞(einx) = πτ±∞(d(U n)) + πτ±∞(U n)Dβ,τ±∞(1) = πτ±∞(U n) · lim k→±∞ (β(k + n) − β(k)) + cπτ±∞(U n)(1) = einx (nβ±∞ + c) = β±∞ 1 i d dx (einx) + ceinx. This completes the proof. (cid:3) 5.2. Covariant derivations. Now let dβ be a ρθ-covariant derivation dβ : A → A, of the form dβ(a) = [Uβ(K), a], as proved in Proposition 3.2. Let τ be a ρθ-invariant state. Definition: Dτ : Dτ → Hτ is called an implementation of a ρθ-covariant derivation dβ if [ Dτ , πτ (a)] = πτ ( dβ(a)) and Uτ,θ Dτ U −1 τ,θ = eiθ Dτ for every a and θ. We state without proofs the analogs of the above implementation results for covariant derivations; the verifications are simple modifications of the arguments for invariant deriva- tions. 16 KLIMEK, MCBRIDE, AND RATHNAYAKE Proposition 5.5. There exists a function α(k), Pk∈Z β(k) − α(k)2w(k) < ∞, such that any implementation Dβ,τw : Dτw → Hτw of dβ is uniquely represented by: Dβ,τwf = Uβ(K)f − f Uα(K). Proposition 5.6. Any implementation Dβ,τ0 : Dτ0 → ℓ2(Z) of dβ is of the form: Dβ,τ0a = Uβ(K)a, i.e. on basis elements Dβ,τ0En = β(n)En+1. Proposition 5.7. There exists a number c such that any implementation Dβ,τ±∞ : Dτ±∞ → L2(S1) of dβ is of the form: Dβ,τ∞ = eix(cid:18)β±∞ 1 i d dx + c(cid:19) . 6. Compactness of Parametrices 6.1. Spectral triples. We say that a closed operator D has compact parametrices if the operators (1+D∗D)−1/2 and (1+DD∗)−1/2 are compact. Other equivalent formulations were summarized in the appendix of [12]. In what follows we analyze when do the implementations of the derivations in A have compact parametrices. The cases below are very similar to the examples in [12]. Consequently, all those cases give even spectral triples for the algebra A. We use the same notation for the closure of the operators constructed in the previous section. Proposition 6.1. The operators Dβ,τ0, Dβ,τ0 have compact parametrices if and only if β(k) → ∞ as k → ∞. Proof. The operators Dβ,τ0 are diagonal with eigenvalues β(k) − β(0) + c, which must go to infinity for the operators to have compact parametrices. The operators Dβ,τ0 differ from the operators Dβ,τ0 by a shift, so they behave in the same way. (cid:3) Proposition 6.2. The operators Dβ,τ∞, Dβ,τ∞ have compact parametrices if and only if β±∞ 6= 0. Proof. Similarly to the proof of the proposition above, the operators Dβ,τ∞ are diagonal with eigenvalues ±β±∞n + c, which go to infinity if and only if β±∞ 6= 0. (cid:3) Proposition 6.3. The operators Dβ,τw have compact parametrices if and only if as n, k → ∞. β(k + n) − α(k) → ∞ Proof. The operators Dβ,τw can be diagonalized using the Fourier series: Computing Dβ,τwf = β(K)f − f α(K) we get: U nfn(K). f =Xn∈Z Dβ,τwf =Xn∈Z U n(β(K + n) − α(K))fn(K). DERIVATIONS AND SPECTRAL TRIPLES ON QUANTUM DOMAINS II: QUANTUM ANNULUS 17 It follows that the numbers β(k + n) − α(k) are the eigenvalues of the diagonal operator, and must diverge for the operator to have compact parametrices. (cid:3) 6.2. Covariant derivations and normal states. Here we study the parametrices of the operators which implement ρθ-covariant derivations in GNS Hilbert spaces Hτw corresponding to faithful normal states. In this section, like in [12], we use a different notation for Dβ,τw; we use instead: Dβ,α,wf = Uβ(K)f − f Uα(K), a notation that indicates the coefficients of the operator. Denote by Dmax Dβ,α,w defined on Dmax Define the ∗−algebra: τ = πτ (A) · [I]. β,α,w the closure of A0 =na =X U nan(K) : an(k) ∈ c± 00(Z)o , 00(Z) are the sequences with compact support, i.e. eventually zero, and let Dmin β,α,w be τ = πτ (A0) · [I]. Finally, we use the symbol Dβ,α,w for where c± the closure of Dβ,α,w defined on Dmin any closed operator in Hτw such that: Dmin β,α,w ⊂ Dβ,α,w ⊂ Dmax β,α,w. The main objective of this section is to prove the following no-go result, analogous to the non-existence theorem in the case for the quantum disk, see [12]. Theorem 6.4. There is no closed operator Dβ,α,w in Hτw , Dmin β±∞ 6= 0, such that Dβ,α,w has compact parametrices. β,α,w ⊂ Dβ,α,w ⊂ Dmax β,α,w, with Proof. The proof proceeds through a number of steps, similar to those in [12]. Initially, by a sequence of equivalences, we show that the operator Dβ,α,w has compact parametrices if and only if a simplified version of it has compact parametrices. Since in particular an operator with compact parametrices has to be Fredholm, the finiteness of kernel and cokernel, as well as closedness of the range, imply certain growth estimates on the parameters. As mentioned before, these growth estimates on the parameters are very different from what occurred in the disk situation. First we show that the coefficients β(k) can be replaced by its absolute values. We need the following information. Lemma 6.5. Let {β(k)} be sequence of complex numbers. If β(k + 1) − β(k) → β∞ as k → ∞ and β(k) − β(k + 1) → β−∞ as k → −∞ with β±∞ 6= 0, then there exists positive constants c1 and c2, and a nonnegative constant c3 such that: c2(k + 1) − c3 ≤ β(k) ≤ c1(k + 1). Moreover β(k + 1) − β(k) is bounded. Proof. We prove first that: β(k) = β±∞(k + 1)(1 + o(1)) as k → ±∞. From this the first inequality follows immediately. We decompose β(k) as follows: β(k) =(cid:26) β∞ · (k + 1) + β0(k) β−∞ · k + β1(k) for k ≥ 0 for k < 0, 18 so that as k → ∞ and KLIMEK, MCBRIDE, AND RATHNAYAKE β0(k) − β0(k − 1) =: ψ0(k) → 0 β1(k) − β1(k + 1) =: ψ1(k) → 0 as k → −∞. We want to show that: β0(k) k + 1 → 0 and β1(k) k + 1 → 0 as k → ±∞ respectively. We study k ≥ 0 first, k < 0 is similar. Notice that we have: β0(k) k + 1 = β0(k) − β0(−1) k + 1 + β0(−1) k + 1 = 1 k + 1 ψ0(j) + β0(−1) k + 1 . k Xj=0 It is clear that the second term above goes to 0 as k → ∞. For the first term, given ε > 0 we choose jε so that ψ0(j) < ε for j ≥ jε. Then we split the sum as follows: 1 k + 1 k Xj=0 ψ0(j) = 1 k + 1 and then choose kε so that: ψ0(j) + jε−1 Xj=0 1 k + 1 k Xj=jε ψ0(j) ≤ 1 k + 1 jε−1 Xj=0 ψ0(j) + ε, for k ≥ kε. Therefore it follows that: (supl ψ0(l)) jε k + 1 < ε β0(k) k + 1 = o(1). The second part of the lemma follows from the estimate: β(k + 1) − β(k) ≤ β(k + 1) − β(k) < ∞. Lemma 6.6. Under the assumption of Theorem 6.4 the operator Dβ,α,w has compact para- metrices if and only if the operator Dβ,α,w, satisfying: (cid:3) Dmin β,α,w ⊂ Dβ,α,w ⊂ Dmax β,α,w, has compact parametrices. Proof. Define the unitary operator V (K) by: V (k) = exp i Arg(β(j))! , k ≥ 1, and V (k) = exp −i Arg(β(j))! , k ≤ 0, k−1 Xj=0 0 Xj=k and consider the map φ : f 7→ V (K)f for f ∈ Hτw . This map preserves the domains Dmin and Dmax . A direct computation gives that: τ τ This shows that Dβ,α,w and Dβ,α,w are unitarily equivalent, thus completing the proof. (cid:3) Dβ,α,w = φDβ,α,wφ−1. DERIVATIONS AND SPECTRAL TRIPLES ON QUANTUM DOMAINS II: QUANTUM ANNULUS 19 Lemma 6.7. The operator Dβ,α,w has compact parametrices if and only if for any bounded sequences γ1(k) and γ2(k) the operator Dβ+γ1,α+γ2,w satisfying: Dmin β+γ1,α+γ2,w ⊂ Dβ+γ1,α+γ2,w ⊂ Dmax β+γ1,α+γ2,w, has compact parametrices. Proof. Notice that the difference Dβ+γ1,α+γ2,w − Dβ,α,w is bounded, hence the two operators both either have or do not have compact parametrices simultaneously, see the appendix of [12]. (cid:3) It follows from those lemmas that, without loss of generality, we may assume that β(k) > 0, where β(k) satisfies inequalities: c2(k + 1) ≤ β(k) ≤ c1(k + 1), β(k + 1) − β(k) < ∞, (6.1) where c1 and c2 are positive constants. Next we look at properties of the coefficient α(k). If f is in the domain of the operator Dβ,α,w and: we can write Dβ,α,wf = Uβ(K)f − f Uα(K) in Fourier components as: U n+1(Dnfn)(K), (6.2) U nfn(K), f =Xn Dβ,α,wf =Xn where (Dnf )(k) = β(k + n)f (k) − α(k)f (k + 1). Lemma 6.8. If the operator Dβ,α,w has compact parametrices, then α(k) has at most finitely many zeros. Proof. Suppose that α(k) has infinitely many zeros. Let k0 be one of the zeroes of α(k). Then the sequence δk0+1(k) is an eigenvector of Dn with eigenvalue β(k0 + 1 + n), where δj(k) = 1 for k = j and zero otherwise. Indeed, we have: (Dnδk0+1)(k) = β(k + n)δk0+1(k) − α(k)δk0+1(k + 1) = = β(k0 + 1 + n)δk0+1(k) − α(k0)δk0+1(k + 1) = β(k0 + 1 + n)δk0+1(k). Using this observation we obtain that: U −1Dβ,α,wU −(k0+1)δk0+1(K) = β(0)U −(k0+1)δk0+1(K). Notice that U −(k0+1)δk0+1(K) has finite support and hence belongs to the domain of Dmin β,α,w, and so it is in the domain of any Dβ,α,w. Thus, if α(k) has infinitely many zeros, the operator: U −1Dβ,α,w − β(0)I has an infinite dimensional kernel, which is impossible since it has compact parametrices and hence it is Fredholm by the results in the appendix of [12]. Therefore α(k) can only have finitely many zeros. This completes the proof. (cid:3) 20 KLIMEK, MCBRIDE, AND RATHNAYAKE As a consequence of the above lemma, and also Lemma 6.7, we assume from now on that α(k) 6= 0 for every k. While that conclusion is the same as in [12], the argument above is different. It is more convenient to work in the Hilbert space with no weights. The corresponding equivalence is described by the following lemma. Lemma 6.9. Let Hτw be the weighted Hilbert space of Proposition 4.4, part 1, and let H be that Hilbert space for weight w(k) = 1. The operator Dβ,α,w in Hτw has compact parametrices if and only if the operator Dβ, α,1 in H, also satisfying: has compact parametrices, where Proof. In Hτw write the norm as: Dmin β, α,1 ⊂ Dβ, α,1 ⊂ Dmax β, α,1 α(k) = α(k) pw(k) pw(k + 1) . kf k2 w = tr(w(K)f ∗f ) = tr(cid:0)f w(K)1/2(cid:1)∗(cid:0)w(K)1/2f(cid:1) , and set ϕ(f ) = f w(K)1/2 : Hτw → H. Then ϕ is an isomorphism of Hilbert spaces. Moreover, we have: ϕ(Dβ,α,wϕ−1f ) = Uβ(K)f − f Uα(K) pw(K) pw(K + 1) = Dβ, α,1f, and ϕDβ,α,wϕ−1 is an unbounded operator in H, and so Dβ,α,w and Dβ, α,1 are unitarily equivalent, thus completing the proof. Notice also that α(k) 6= 0, because α(k) 6= 0. (cid:3) From now on we work with operators Dmin β,α,1 ⊂ Dβ,α,1 ⊂ Dmax space H. For convenience we define a sequence {µ(k)} such that β,α,1 in the unweighted Hilbert α(k) = β(k) µ(k + 1) µ(k) . The sequence µ, normalized by µ(0) = 1, is completely determined by the above equation in terms of α and β and will be used as a coefficient instead of α. Using µ(k) we rewrite the two main operators as follows: (Dnf )(k) = β(k + n)(cid:18)f (k) − (Dn)∗f (k) = β(k + n)(cid:18)f (k) − β(k) µ(k + 1) β(k + n) µ(k) f (k + 1)(cid:19) , f (k − 1)(cid:19) . β(k − 1) β(k + n) µ(k) µ(k − 1) The next goal is to compute the kernel and the cokernel of Dβ,α,1. This is done by using Fourier decomposition (6.2) and the operators Dn and (Dn)∗ above. DERIVATIONS AND SPECTRAL TRIPLES ON QUANTUM DOMAINS II: QUANTUM ANNULUS 21 Lemma 6.10. The formal kernels of Dn and (Dn)∗ are one dimensional and are spanned by, correspondingly: β(j + k) for n > 0 1 β(j + k) 1 β(j + k) for n = 0 for n < 0, for n > 0 for n = 0 1 µ(k) 1 µ(k) 1 µ(k) n−1 Yj=0 −1 Yj=n Yj=0 n     µ(k) µ(k) µ(k) and hn(k) = hn(k) = −1 Yj=n+1 β(j + k) for n < 0. Proof. This is a result of straightforward calculations similar to those in [10] and [12]. (cid:3) The computations above were formal; to actually compute the kernel and the cokernel of Dβ,α,1 we need to look at only those solutions which are in the domain/codomain of Dβ,α,1. We have the following inclusions: and ker Dmin β,α,1 ⊂ ker Dβ,α,1 ⊂ ker Dmax β,α,1, coker Dmax β,α,1 ⊂ coker Dβ,α,1 ⊂ coker Dmin β,α,1. The following lemma exhibits the key departure from the analogous classical analysis of the d-bar operator. Lemma 6.11. If the operator Dβ,α,1 has compact parametrices, then both ker Dmax coker Dmin β,α,1 are trivial. Moreover, we have: β,α,1 and Xk∈Z hn(k)2 =Xk∈Z hn(k)2 = ∞ (6.3) for every n, where hn and hn are the formal kernels from Lemma 6.10. Proof. Let hn and hn be the formal solutions to the equations Dnhn = 0 and (Dn)∗hn = 0 respectively, as described in Lemma 6.10. First we study the ℓ2(Z) kernel of Dn. There are two possibilities: (1) there exists n0 ∈ Z such that khn0k < ∞, or (2) khnk = ∞ for all n. Consider the first case. It is clear from the growth conditions (6.1) and the formulas for hn that if there exists n0 ∈ Z such that khn0k < ∞ then khnk < ∞ for all n ≤ n0. Those inequalities imply that Dmax β,α,1 has an infinite dimensional kernel. We argue below that in this case the kernel of Dmin β,α,1 is also infinite dimensional, in contrast to classical theory. 22 KLIMEK, MCBRIDE, AND RATHNAYAKE For any n < n0 consider the sequence: hN n (k) =(cid:26) hn(k) 0 for k ≤ N else. DnhN n (k) =  Notice that, because it is eventually zero, the sequence hN also hN n → hn in ℓ2(Z) as N → ∞. Moreover, a key direct calculation shows that: for k = N β(n + N)hn(N) = hn+1(N) n (k) is in the domain of Dmin n and − β(−N − 1) µ(−N) µ(−N − 1) 0 hn(−N) = −hn+1(−N − 1) for k = −N − 1 else. From this we see that DnhN that the formal kernel of Dn is contained in the domain of Dmin Dβ,α,1 has an infinite dimensional kernel, contradicting the fact that Dβ,α,1 is Fredholm. n → 0 as N → ∞ since khnk < ∞ for all n ≤ n0. This shows . In turn, this implies that n A similar argument produces an infinite dimensional cokernel for Dβ,α,1 assuming that there exists n0 ∈ Z such that khn0k < ∞. Consequently, this does not happen in our case, and khnk = khnk = ∞ for all n. But that means that the ℓ2(Z) kernels of Dn and (Dn)∗ are all trivial for any n. This implies that both ker Dmax β,α,1 are trivial. Thus the proof is complete. (cid:3) β,α,1 and coker Dmin It follows from the above lemma, and from the "sandwich" proposition in appendix of [12], that all three operators Dmin β,α,1 ⊂ Dβ,α,1 ⊂ Dmax β,α,1 have compact parametrices. Using the above result we show that existence of compact parametrices for Dβ,α,1 implies specific growth estimates on coefficients µ(k) in the following lemma. Notice that, while the methods used here are similar to those in [12], the outcomes are quite different; instead of being polynomially increasing or decreasing the coefficients µ(k) now have to be rapidly increasing or decreasing. Lemma 6.12. If the operator Dβ,α,1 such that Dmin β,α,1 has compact para- metrices then the Fourier components (Dn)min and ((Dn)∗)max are invertible operators with bounded inverses. Moreover, for every N > 0 there exists constants CN > 0 such that exactly one of the following two holds: β,α,1 ⊂ Dβ,α,1 ⊂ Dmax (1) µ(k) ≥ CN (1 + k)N for k > 0 and µ(k) ≤ CN (1 + k)−N for k < 0, or (2) µ(k) ≤ CN (1 + k)−N for k > 0 and µ(k) ≥ CN (1 + k)N for k < 0. β,α,1 and ((Dβ,α,1)∗)max Proof. The Fredholm property of Dβ,α,1 implies that the ranges of Dmax are closed. By the proof of Lemma 6.11 the ℓ2(Z) kernels of Dn and (Dn)∗ are trivial for all n. It follows that we have: Ran Dmax n = Ran (D∗ n)max = ℓ2(Z). In particular this says that we have inclusions: ηn := (Dmax n )−1χ0(k) ∈ ℓ2(Z) and ηn := ((D∗ n)max)−1χ0(k) ∈ ℓ2(Z), where, as before, χ0(k) = 1 for k = 0 and zero for all other k. Since the right-hand sides of the equations: (Dnηn)(k) = χ0(k) and (Dn)∗ ηn(k) = χ0(k) (6.4) DERIVATIONS AND SPECTRAL TRIPLES ON QUANTUM DOMAINS II: QUANTUM ANNULUS 23 are zero when k 6= 0, the solutions are multiples of the kernel solutions from Lemma 6.10: ηn(k) =( c+ 1 (n)hn(k) c− 1 (n)hn(k) for k > 0 for k < 0, and ηn(k) =( c+ 2 (n)hn(k) 2 (n)hn(k) c− for k > 0 for k < 0. When k = 0 in equations (6.4) the right-hand side is one and a simple calculation shows that the constants c± 2 (n) satisfy the following: 1 (n), c± 1 (n) − c− c+ 1 (n) = µ(0) c+ 1 (n) − c− 1 (n) = µ(0) c+ 2 (n) − c− 2 (n) = c+ 2 (n) − c− 2 (n) = 1 µ(0) 1 µ(0) n −1 n−1 Yj=0 Yj=n+1 Yj=0 Yj=n −1 1 β(j) for n ≥ 0, β(j) for n < 0, β(j) β(j) for n ≥ 0, for n < 0. In particular for any fixed n the constants c± for c± 2 (n). 1 (n) cannot be both equal to zero, and the same Next, from Lemma 6.11 we know that the two-sided sums from formula (6.3) are infinite. This can only be reconciled with the expressions for ηn and ηn, which are in ℓ2(Z), if some one-sided sums of squares of ηn and ηn are finite and some coefficients c± 2 (n) are zero. In fact we will show below that there are only two options that can occur. 1 (n), c± Notice that for any n exactly one of the two sums: hn2 + :=Xk≥0 hn(k)2, hn2 hn(k)2 − :=Xk<0 must be finite, or otherwise ηn will not be in ℓ2(Z) as at least one of c± 1 (n) is not zero. Assume then that for some n0 we have hn0+ < ∞. Because hn0 = ∞ we must have hn0− = ∞. Now it follows from the growth conditions (6.1) and the formulas for hn that khnk+ < ∞ for all n ≤ n0, and similarly khnk− = ∞ for all n ≥ n0. But whenever khnk− = ∞ we must have khnk+ < ∞ by the square summability of ηn. Thus if hn0+ < ∞ for some n0, we have khnk+ < ∞ for all n. Clearly, the same arguments can be made for hn. Notice however that: hn+hn+ ≥Xk≥0 hn(k)hn(k) =Xk≥0 1 β(n + k) = ∞ by the growth conditions on β(k), and similarly for sums with k < 0. Thus, if khnk+ < ∞ then khnk+ = ∞ and the other way around. This leads to the following two options: Option 1: For all n we have: Xk>0 hn(k)2 < ∞,Xk<0 hn(k)2 = ∞,Xk>0 hn(k)2 = ∞,Xk<0 hn(k)2 < ∞. 24 KLIMEK, MCBRIDE, AND RATHNAYAKE Option 2: For all n we have: for every n: Xk>0 hn(k)2 = ∞,Xk<0 Xk>0(cid:12)(cid:12)(cid:12)(cid:12) hn(k)2 < ∞,Xk>0 (cid:12)(cid:12)(cid:12)(cid:12) < ∞,Xk<0(cid:12)(cid:12)(cid:12)(cid:12) 2 hn(k)2 < ∞,Xk<0 (k2 + 1)n+1(cid:12)(cid:12)(cid:12)(cid:12) µ(k) 2 (k2 + 1)n µ(k) < ∞. hn(k)2 = ∞. Using the estimates (6.1) for option 1 we see that the following two sums are convergent This clearly implies the growth estimates in the statement of the lemma. Option 2 is similar, thus the result follows, completing the proof. (cid:3) In light of this lemma, we see that the coefficients µ(k) are functions of rapid decay or rapid growth at different infinities. Now that we have control over the coefficients of Dβ,α,1 we can finish the main result. Interestingly, the two options from the previous lemma require different arguments. Lemma 6.13. Suppose µ(k) is a function of rapid growth for k > 0 and a function of rapid decay for k < 0. Then the operator Dβ,α,1 cannot have compact parametrices. Proof. By Lemma 6.11, the operator Dn has no kernel or cokernel for all n. Therefore we can solve the equation: (Dnf )(k) = β(k + n)(cid:18)f (k) − β(k) µ(k + 1) β(k + n) µ(k) f (k + 1)(cid:19) = g(k) for any g ∈ ℓ2(Z). Using the fact that µ(k) is a function of rapid decay for k < 0, and solving recursively, we get the following formulas: (D−1 n g)(k) = − − k−1 k−1 Xj=−∞ Xj=−∞   β(k) · · · β(k + n − 1) β(j) · · · β(j + n) · µ(j) µ(k) g(j) = − ∆+ n (k, j)g(j) for n > 0 β(j − 1) · · · β(j + n + 1) β(k − 1) · · · β(k + n) · µ(j) µ(k) g(j) = − ∆− n (k, j)g(j) for n ≤ 0, k−1 Xj=−∞ k−1 Xj=−∞ where we introduced the notation ∆± contradiction we consider the sub-diagonal matrix coefficients: n for the integral kernels of operators D−1 n . To obtain a ∆+ n (k, k − 1) = β(k) · · · β(k + n − 1) β(k − 1)β(k) · · · β(k + n − 1) · µ(k − 1) µ(k) = µ(k − 1) β(k − 1)µ(k) which are independent of n. On the other hand we have ∆+ n (k, k − 1) = hχk, D−1 n χk−1i ≤ kD−1 n kkχkkkχk−1k = kD−1 n k, where, as before, χk(j) = 1 when j = k and 0 otherwise. However, if Dβ,α,1 has compact parametrices then kD−1 n k goes to 0 as n → ±∞ as U −1Dβ,α,1 is a direct sum of operators Dn. This is in clear contradiction with the above calculation. (cid:3) Now we consider option 2 of Lemma 6.12. DERIVATIONS AND SPECTRAL TRIPLES ON QUANTUM DOMAINS II: QUANTUM ANNULUS 25 Lemma 6.14. Suppose µ(k) is a function of rapid decay for k > 0 and a function of rapid growth for k < 0. Then the operator Dβ,α,1 cannot have compact parametrices. Proof. To obtain a contradiction we study the spectrum of the operator Dmax compact parametrices, the Fredholm property of Dmax is closed, which means that the continuous spectrum σc((D+ β,α,1 − I implies that Ran ((D+ n )max) is empty. β,α,1. Since it has n )max − λI) Next we study the eigenvalue equation (Dnf )(k) = λf (k), that is: β(k + n)f (k) − β(k) µ(k + 1) µ(k) f (k + 1) = λf (k). (6.5) Pick a value β(l) of the coefficient β such that β(j) 6= β(l) for all j < l. This is possible since β(j) → ∞ as j → −∞. Consider equation (6.5) for eigenvalue λl = β(l). Then, substituting k = l − n into (6.5), we get that f (l − n + 1) = 0. Consequently, equation (6.5) can be solved recursively, yielding a one-parameter formal solution generated by: 1 µ(k) l−n−1 Yj=k 0 1 + β(j) − β(j + n) + λl β(j + n) − λl ! k ≤ l − n k > l − n. fl(k) =  Notice that by our assumption on β(l) the denominators in the above equation are never zero. The question then is when does fl ∈ ℓ2(Z)? First we estimate from above each factor in the formula for fl as follows: This implies that for k ≤ l − n 1 + (cid:12)(cid:12)(cid:12)(cid:12) β(j + n) − λl β(j) − β(j + n) + λl (cid:12)(cid:12)(cid:12)(cid:12) fl(k) ≤ exp"l−n−1 Xj=k (cid:12)(cid:12)(cid:12)(cid:12) 2 β(j) − β(j + n) + λl β(j + n) − λl ≤ exp (cid:12)(cid:12)(cid:12)(cid:12) 2! . (cid:12)(cid:12)(cid:12)(cid:12) β(j) − β(j + n) + λl β(j + n) − λl 1 µ(k) . 2# (cid:12)(cid:12)(cid:12)(cid:12) Notice that we have β(j) − β(j + n) + λl ≤ const uniformly in j by (6.1), so, using additionally (6.1) in the denominator, we obtain: Therefore we have β(j) − β(j + n) + λl β(j + n) − λl l−n−1 ≤ 2 (cid:12)(cid:12)(cid:12)(cid:12) l−n−1 Xj=k l−n−1 fl(k)2 = fl(k)2 ≤ const j2 + 1 < const. const µ(k)2 < ∞, l−n−1 Xj=k (cid:12)(cid:12)(cid:12)(cid:12) Xk∈Z Xk=−∞ Xk=−∞ and so fl ∈ ℓ2(Z). Thus λl is an eigenvalue for Dmax β,α,1). Crucially, since λl is independent of n, we see that λl has an infinite degeneracy. As ex- plained in the appendix of [12], if Dmax β,α,1 has compact parametrices then its spectrum is either empty, or is the whole plane C, or consist of eigenvalues going to infinity, which is in direct contradiction with the infinite degeneracy. Therefore Dβ,α,1 does not have compact parametrices. (cid:3) for all n ∈ Z. So λl ∈ σ(U −1Dmax n 26 KLIMEK, MCBRIDE, AND RATHNAYAKE To summarize, we reduced the existence of an operator Dβ,α,w in Hτw, Dmin β,α,w ⊂ Dβ,α,w ⊂ Dmax β,α,w and having compact parametrices to the existence of an operator Dβ,α,1 in ℓ2(Z) with coefficients satisfying (6.1) and conditions of Lemma 6.12 and having compact parametrices. However, as seen in the last two lemmas, that leads to contradictions and the proof of the theorem is complete. (cid:3) References [1] Battisti, U., Seiler, J., Boundary value problems with Atiyah-Patodi-Singer type conditions and spectral triples, J. Noncomm. Geom., 11, 887 - 917, 2017 [2] Boo-Bavnbek, B. and Wojciechowski, K.P., Elliptic Boundary Problems for Dirac Operators, Birkhauser, Boston, 1993. [3] Bratteli, O., Elliott, G. A., and Jorgensen, P. E. T., Decomposition of unbounded derivations into invariant and approximately inner parts, Jour. Reine Ang. Math., 346, 166 - 193, 1984. [4] Brown, A., Halmos, P.R., Shields A.L., Ces`aro operators, Acta Sci. Math. (Szeged), 26, 125 - 137, 1965. [5] Forsyth, I., Goffeng, M., Mesland, B., Rennie, A., Boundaries, spectral triples and K-homology, arXiv:1607.07143 [math.KT]. [6] Hadfield, T., The noncommutative geometry of the discrete Heisenberg group, Houston J. Math., 29, 453 - 481, 2002. [7] Jorgensen, P., Approximately inner derivations, decompositions and vector fields of simple C∗-algebras, in Mappings of operator algebras: Proceedings of the Japan-U.S. Joint Seminar (Philadelphia, 1988), pp. 15 - 113, (H. Araki and R.V. Kadison, eds.), Progr. Math., vol. 84, Birkhauser, Boston, 1990. [8] Kadison, R. V.; Ringrose, J. R., Fundamentals of the Theory of Operator Algebras, Academic Press 1986. [9] Klimek, S. and Lesniewski, A., Quantum Riemann surfaces, III. The Exceptional Cases, Lett. Mat. Phys., 32, 45 - 61, 1994. [10] Klimek, S. and McBride, M., D-bar Operators on Quantum Domains, Math. Phys. Anal. Geom., 13, 357 - 390, 2010. [11] Klimek, S., McBride, M., Rathnayake, S., Sakai, K., The Quantum Pair of Pants, SIGMA, 11, 012, 1 - 22, 2015. [12] Klimek, S. and McBride, M., Rathnayake, S., Sakai, K., Wang, H., Derivations and Spectral Triples on Quantum Domains I: Quantum Disk, SIGMA, 13, 075, 1 - 26, 2017. [13] Renault J., Cartan subalgebras in C*-algebras, Irish Math. Soc. Bull., 61, 29 - 63, 2008. [14] Sakai, S. Operator Algebras in Dynamical Systems, Cambridge University Press, 1991. Department of Mathematical Sciences, Indiana University-Purdue University Indianapo- lis, 402 N. Blackford St., Indianapolis, IN 46202, U.S.A. E-mail address: [email protected] Department of Mathematics and Statistics, Mississippi State University, 175 President's Cir., Mississippi State, MS 39762, U.S.A. E-mail address: [email protected] Department of Mathematics, University of Michigan, 530 Church St., Ann Arbor, MI 48109, U.S.A. E-mail address: [email protected]
1905.01889	1	1905	2019-05-06T09:05:03	Gabor Duality Theory for Morita Equivalent $C^*$-algebras	[ "math.OA", "math.FA" ]	The duality principle for Gabor frames is one of the pillars of Gabor analysis. We establish a far-reaching generalization to Morita equivalent $C^$-algebras where the equivalence bimodule is a finitely generated projective Hilbert $C^$-module. These Hilbert $C^$-modules are equipped with some extra structure and are called Gabor bimodules. We formulate a duality principle for standard module frames for Gabor bimodules which reduces to the well-known Gabor duality principle for twisted group $C^$-algebras of a lattice in phase space. We lift all these results to the matrix algebra level and in the description of the module frames associated to a matrix Gabor bimodule we introduce $(n,d)$-matrix frames, which generalize superframes and multi-window frames. Density theorems for $(n,d)$-matrix frames are established, which extend the ones for multi-window and super Gabor frames. Our approach is based on the localization of a Hilbert $C^*$-module with respect to a trace.	math.OA	math	GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS ARE AUSTAD, MADS S. JAKOBSEN, AND FRANZ LUEF Abstract. The duality principle for Gabor frames is one of the pillars of Gabor analysis. We establish a far-reaching generalization to Morita equivalent C ∗-algebras where the equivalence bimodule is a finitely generated projective Hilbert C ∗-module. These Hilbert C ∗-modules are equipped with some extra structure and are called Gabor bimodules. We formulate a duality principle for standard module frames for Gabor bimodules which reduces to the well-known Gabor duality principle for twisted group C ∗-algebras of a lattice in phase space. We lift all these results to the matrix algebra level and in the description of the module frames associated to a matrix Gabor bimodule we introduce (n, d)-matrix frames, which generalize superframes and multi-window frames. Density theorems for (n, d)-matrix frames are established, which extend the ones for multi-window and super Gabor frames. Our approach is based on the localization of a Hilbert C ∗- module with respect to a trace. Contents Introduction 1. 2. Preliminaries on C ∗-algebras and Hilbert C ∗-modules 3. Gabor bimodules 3.1. The single generator case 3.2. Extending to several generators 3.3. From a Gabor bimodule to its localization 4. Applications to Gabor analysis Acknowledgement References 1 3 7 7 13 17 20 34 35 1. Introduction Hilbert C ∗-modules are well-studied objects in the theory of operator algebras and Rieffel made the crucial observation that they provide the correct framework for the extension of Morita equivalence of rings to C ∗-algebras. In his seminal work [27] he noted that the equivalence bimodules between two C ∗-algebras are bimodules where the left and right Hilbert C ∗-module structures are compatible and the respective C ∗-valued inner products satisfy an associativity condition. We are interested in the features of these equivalence bimodules from the perspective of frame theory. In [12] the notion of standard module frame was introduced for countably generated Hilbert C ∗-modules. Rieffel has already in [28] observed that finitely generated equivalence bimodules may be described in terms (Are Austad, Franz Luef) Norwegian University of Science and Technology, Department of Mathematical Sciences, Trondheim, Norway. (Mads S. Jakobsen) Weibel Scientific A/S, Solvang 30, 3450 Allerød, Denmark E-mail addresses: [email protected], [email protected], [email protected]. 1 2 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS of finite standard module frames and used it in his study of Heisenberg modules, which is a class of projective Hilbert C ∗-modules over twisted group C ∗-algebras. In [18] the properties of standard module frames for Heisenberg modules have been studied from the perspective of duality theory, which was motivated by the observation in [21] that these module frames are closely related to Gabor frames for an associated Hilbert space. Gabor frames have some additional features not shared by wavelets and shearlets that is due to the seminal contributions [9, 19, 29], where they developed the duality theory of Gabor frames. Theorem (Duality Theorem). The Gabor system {e2πiβltg(t − αk)}k,l∈Z generated by a function g ∈ L2(R) is a frame for L2(R) if and only if {e2πilt/αg(t − k/β)}k,l∈Z is a Riesz sequence for the closed span of {e2πilt/αg(t − k/β)}k,l∈Z in L2(R) Here t ∈ R. Due to its far-reaching implications there have been attempts to extend the duality principle to other classes of frames [1, 15, 3, 4], see [6, 8, 31, 32] for the theory of R-duals and [10]. Motivated by the link between the duality theory of Gabor frames and the Morita equivalence of noncommutative tori [21, 18] we extend the duality theory of Gabor frames to module frames for equivalence bimodules between Morita equivalent C ∗-algebras. The setup for our duality theory has its roots in [18] and is as follows: Let A and B be C ∗- algebras where B is assumed to have a unit and be equipped with a faithful finite trace trB. We define a left Gabor bimodule to be a quadruple (1.1) (A, B, E, trB) where E is a Morita equivalence A-B-bimodule. We show that module frames for Gabor bimodules admit a duality theorem and by localization with respect to a trace we are able to connect these module frame statements to results on frames in Hilbert spaces. Note that in [11] a different notion of localization of frames was introduced which constructs frames with additional regularity, which we also establish in our general setting. The main application of our duality results is a concise treatment of Gabor frames for closed cocompact subgroups of locally compact abelian phase spaces. Our general approach to duality principles has led us to the introduction of (n, d)-matrix Gabor frames that is a joint generalization of multi-window superframes and Riesz bases and we prove that their Gabor dual systems are (d, n)-matrix Gabor frames. Let G be a second countable LCA group and let Λ be a closed subgroup of G × bG. For g ∈ L2(G × Zn × Zd), let (1.2) G(g; Λ) := {π(λ)gi,j λ ∈ Λ}i∈Zn,j∈Zd. We say G(g; Λ) is an (n, d)-matrix Gabor frame for L2(G) if the collection of time-frequency shifts G(g; Λ) is a frame for L2(G× Zn × Zd). Equivalently, there exists h ∈ L2(G× Zn × Zd) such that for all f ∈ L2(G × Zn × Zd) we have (1.3) fr,s = Xk∈Zd Xl∈Zn ZΛ hfr,k, π(λ)gl,kiL2(G)π(λ)hl,sdλ, for all r ∈ Zn and s ∈ Zd. We develop the theory of these (n, d)-matrix Gabor frames and prove a duality theorem for this novel type of frames. Let us summarize the content of this paper. In Section 2 we collect some facts about C ∗- algebras, Hilbert C ∗-modules, the localization of Hilbert C ∗-modules and finitely generated projective Hilbert C ∗-modules. In Section 3 we introduce Gabor bimodules, and study GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 3 the case when these have a single generator in terms of module frames. We establish the analog of the duality theorem for Gabor frames for Gabor bimodules with one generator. In Section 3.2 we extend all of these results to the case of finitely many generators which leads one naturally to matrix-valued extensions of the statements and definitions in the preceding section. We also prove a density theorem for module frames. In the final section, Section 4, we discuss applications to Gabor frames for closed subgroups of the time-frequency plane of locally compact abelian groups. 2. Preliminaries on C ∗-algebras and Hilbert C ∗-modules We assume basic knowledge about Banach ∗-algebras, C ∗-algebras, and of Banach mod- ules and Hilbert C ∗-modules. In this section we collect definitions and basic facts of concepts crucial for this paper, such as positivity in C ∗-algebras, Morita equivalence of C ∗-algebras, and localization of Hilbert C ∗-modules. For these topics we refer to [20], [25], and [24]. For a C ∗-algebra A and a ∈ A, we denote by σA(a) the spectrum of a in A. We will need the following important result. Proposition 2.1 ([24], Theorem 2.1.11). Let A be a unital C ∗-algebra and let B be a C ∗-subalgebra of A containing the unit of A. If b ∈ B, then σB(b) = σA(b). Equivalently, if b is invertible in A, then b−1 ∈ B. Definition 2.2. Let A be a unital C ∗-algebra and let B ⊂ A be a Banach ∗-subalgebra of A with the same unit. We say that B is spectral invariant in A if whenever b ∈ B is invertible with b−1 ∈ A, we have b−1 ∈ B. Now recall that a selfadjoint element a in a C ∗-algebra with σA(a) ⊂ [0, ∞) is called positive. We state a useful characterization of positivity. Proposition 2.3. Let A be a C ∗-algebra. For a = a∗ ∈ A we have σA(a) ⊂ [0, ∞) if and only if a = b∗b for some b ∈ A. Denote by A+ the set of positive elements in the C ∗-algebra A. The positive elements form a cone. In particular, if a ∈ A+ then ka ∈ A+ for all k ∈ [0, ∞), and if a1, a2 ∈ A+ then a1 + a2 ∈ A+. We also obtain a partial order on A+ by a ≤ b if and only if b − a ∈ A+. Note that not all elements of A+ are comparable, but all elements are comparable to 1A in the case A is unital. Central to our results in Section 3 will be the localization of a Hilbert C ∗-module. For this we need positive linear functionals. Definition 2.4. A positive linear functional on a C ∗-algebra A is a linear functional φ such that φ(A+) ⊂ [0, ∞). If kφk = 1 we say φ is a state. Remark 2.5. If φ : A → C is a positive linear functional and A is unital, it is known that φ is a state if and only if φ(1A) = 1. We will denote the set of adjointable operators on the Hilbert A-module E by EndA(E), and the set of compact module operators by K(E). The following two results will be of great importance in our approach to duality theorems. Proposition 2.6 ([20], Proposition 1.1). Let A be a C ∗-algebra. If E is an inner product A-module and f, g ∈ E, then Ahg, f iAhf, gi ≤ kAhf, f ikAhg, gi, 4 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS where Ah·, ·i is the A-valued inner product. Also, whenever c ≥ 0 in A, we have a∗ca ≤ kcka∗a for all a ∈ A. Proposition 2.7 ([25], Corollary 2.22). Let A be a C ∗-algebra. If E is a Hilbert A-module and T ∈ EndA(E), then for any f ∈ E AhT f, T f i ≤ kT k2 Ahf, f i as elements of the C ∗-algebra A, where Ah·, ·i is the A-valued inner product. Suppose φ is a positive linear functional on a C ∗-algebra B, and that E is a right Hilbert B-module. We define an inner product h·, ·iφ : E × E → C (f, g) 7→ φ(hg, f iB), where h·, ·iB is the B-valued inner product. We may have to factor out the subspace Nφ := {f ∈ E hf, f iB = 0}, and complete E/Nφ with respect to h·, ·iφ. This yields a Hilbert space which we will denote by HE. This is known as the localization of E in φ. There is a natural map ρφ : E → HE which induces a map ρφ : EndB(E) → B(HE). We will focus entirely on the case in which φ is a faithful positive linear functional, that is, when b ∈ B+ and φ(b) = 0 implies b = 0. In that case Nφ = {0} and we have the following useful result from [20, p. 57-58]. Proposition 2.8. Let A be a C ∗-algebra equipped with a faithful positive linear functional φ : A → C, and let E be a left Hilbert A-module. Then the map ρφ : EndA(E) → B(HE) is an injective ∗-homomorphism. The Hilbert C ∗-modules of interest will be of a very particular form in that they will be A-B-equivalence bimodules for C ∗-algebras A and B. We will denote the A-valued inner product by •h ·, ·i, and the B-valued inner product by h·, · i•. Definition 2.9. Let A and B be C ∗-algebras. A Morita equivalence bimodule between A and B, or an A-B-equivalence bimodule, is a Hilbert C ∗-module E satisfying the following conditions. (1) •h E, Ei = A and hE, E i• = B, where •h E, Ei = spanC{•h f, gi f, g ∈ E} and likewise for hE, E i•. (2) For all f, g ∈ E, a ∈ A and b ∈ B, haf, g i• = hf, a∗g i• and •h f b, gi = •h f, gb∗i. (3) For all f, g, h ∈ E, •h f, gih = f hg, h i• . Now let A ⊂ A and B ⊂ B be dense Banach ∗-subalgebras such that the enveloping C ∗-algebra of A is A, and the enveloping C ∗-algebra of B is B. Suppose further there is a dense A-B-inner product submodule E ⊂ E such that the conditions above hold with A, B, E instead of A, B, E. In that case we say E is an A-B-pre-equivalence bimodule. We will make repeated use of the following fact in the sequel without mention. Proposition 2.10 ([25], Proposition 3.11). Let A and B be C ∗-algebras and let E be an A-B-equivalence bimodule. Then k •h f, f ik = khf, f i• k for all f ∈ E. GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 5 It is a well-known result that if E is an A-B-equivalence bimodule, then B ∼= KA(E) through the identification Θf,g 7→ hf, g i•. Here Θf,g is the compact module operator Θf,g : h 7→ •h h, f ig. We make particular note of the case when E is a finitely generated Hilbert A-module. Proposition 2.11. Let E be an A-B-equivalence bimodule. Then E is a finitely generated projective A-module if and only if B is unital. Proof. Suppose first A is finitely generated and projective as a Hilbert A-module. As E is finitely generated, any A-endomorphism on E is determined by its action on a finite set of generators. Hence EndA(E) = KA(E), and the former is unital. Since B ∼= KA(E), have that B is unital. Conversely, we assume that B is unital. As B ∼= KA(E), and the latter is an ideal in EndA(E), it follows that KA(E) = EndA(E). As B is unital and hE, E i• is dense in B, i=1hfi, gi i• = 1B. The maps we can find elements f1, . . . , fn, g1, . . . , gn ∈ E such that Pn s : E → An and h 7→ (•h h, fii)n i=1, r : An → E (ai)n i=1 7→ nXi=1 aigi, are A-module maps satisfying r ◦ s(z) = nXi=1 •h h, fiigi = nXi=1 hhfi, gi i• = h nXi=1 hfi, gi i• = h · 1B = h for all h ∈ E. It follows that E is a finitely generated projective A-module. (cid:3) pendent, but they still provide a reconstruction formula: z = Pn Note that the systems {f1, ..., fn} and {g1, ..., gn} are not necessarily A-linearly inde- i=1 •h h, fiigi. Motivated by spanning sets in finite-dimensional vector spaces, also called frames, we call the system {f1, ..., fn} a module frame for E and {g1, ..., gn} is referred to as a dual module frame. The properties of module frames for equivalence bimodules are the main objective of this work. The following two results concern properties of module frames consisting of a single element, though we do not formally introduce module frames until Definition 3.8. For our setup it will turn out that it is enough to consider module frames consisting of only one element, see Section 3. The results will come into play when we relate module frames and Gabor frames in Section 4. Lemma 2.12. Let A be any C ∗-algebra, and let E be a (left) Hilbert A-module. Suppose T ∈ EndA(E) is such that there exist C, D > 0 such that (2.1) C •h f, f i ≤ •h T f, f i ≤ D •h f, f i, for all f ∈ E. Then T is invertible, and 1 D •h f, f i ≤ •h T −1f, f i ≤ 1 C •h f, f i, for all f ∈ E. 6 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS Proof. By (2.1), we see that T is positive and invertible with C IdE ≤ T ≤ D IdE. Pos- itivity is preserved when multiplying by positive commuting operators, so it follows that CT −1 ≤ T −1T ≤ DT −1, from which we get 1 (cid:3) Lemma 2.13. Let A be any C ∗-algebra, and let E be a (left) Hilbert A-module. Let T ∈ EndA(E) be such that there exist C, D > 0 such that D IdE ≤ T −1 ≤ 1 C IdE. (2.2) C •h f, f i ≤ •h T f, f i ≤ D •h f, f i. for all f ∈ E. Then the smallest possible value of D is kT k, and the largest possible value for C is kT −1k−1. It follows that the Proof. Since T is positive we have kT k = supkf k=1{k •h T f, f ik}. smallest value for D is kT k. By Lemma 2.12 we see by the same argument that the minimal value for 1 (cid:3) C is kT −1k. Hence the largest value for C is kT −1k−1. It is an interesting question when the property of being finitely generated projective passes to dense subalgebras and corresponding dense submodules. Proposition 2.14. Let E be an A-B-equivalence bimodule as in Definition 2.9, with B unital. Suppose there are dense Banach ∗-subalgebras A ⊂ A and B ⊂ B, where B is spectral invariant in B and has the same unit as B. Suppose further that E ⊂ E is an A-B-pre-equivalence bimodule. If E is a finitely generated projective A-module, then E is a finitely generated projective A-module. Proof. The latter part of the proof of Proposition 2.11 can be adapted to this situation. The full proof can be found in [28, Proposition 3.7]. (cid:3) Since we aim to mimic the situation of Gabor analysis, which we will treat in Section 4, the positive linear functional which we localize our Morita equivalence bimodule with respect to will have a particular form. In particular it will be a faithful trace. For unital Morita equivalent C ∗-algebras A and B Rieffel showed in [26] that there is a bijection between non-normalized finite traces on A and non-normalized finite traces on B under which to a trace trB on B there is an associated trace trA on A satisfying (2.3) trA(•h f, gi) = trB(hg, f i•) for all f, g ∈ E. Here E is the Morita equivalence bimodule. We will in the sequel almost always consider A or B unital, and so instead we will suppose the existence of a finite faithful trace on one C ∗-algebra (the unital one) and induce a possibly unbounded trace on the other C ∗-algebra. The following was proved in [2, Proposition 2.7] and ensures that this procedure works. Proposition 2.15. Let E be an A-B-equivalence bimodule, and suppose trB is a faithful finite trace on B. Then the following hold: i) There is a unique lower semi-continuous trace on A, denoted trA, for which (2.4) (2.5) trA(•hf, gi) = trB(hg, f i•) for all f, g ∈ E. Moreover, trA is faithful, and densely defined since it is finite on span{•hf, gi : f, g ∈ E}. Setting hf, gitrA = trA(•hf, gi), hf, gitrB = trB(hg, f i•), for f, g ∈ E defines inner products on E, with hf, gitrA = hf, gitrB for all f, g ∈ E. Consequently, the Hilbert space obtained by completing E in the norm kf k′ = trA(•hf, f i)1/2 is just the localization of E with respect to trB. GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 7 ii) If E and F are equivalence A-B-bimodules, then every adjointable A-linear operator E → F has a unique extension to a bounded linear operator HE → HF . Further- more, the map EndA(E, F ) → End(HE, HF ) given by sending T to its unique extension is a norm-decreasing linear map of Banach spaces. Finally, if E = F , the map EndA(E) → B(HE) is an isometric ∗-homomorphism of C ∗-algebras. If both C ∗-algebras are unital then the induced trace is also a finite trace as in [26], see [2, p. 8]. Convention 2.16. We have the following as a standing assumption for the rest of the manuscript unless otherwise specified: Suppose we have a faithful trace trB on a unital C ∗-algebra B, and an A-B-equivalence bimodule E. If A is a unital C ∗-algebra, we pick the normalization of trB such that trA becomes a state. That is, we normalize the trace on the C ∗-algebra on the left in the Morita equivalence. 3. Gabor bimodules 3.1. The single generator case. Throughout this section we discuss properties of equiv- alence bimodules of the following type. Definition 3.1. Let A and B be C ∗-algebras where B is assumed to have a unit and is equipped with a faithful finite trace trB. We define a left Gabor bimodule to be a quadruple (3.1) (A, B, E, trB) where E is an A-B-equivalence bimodule. We define a right Gabor bimodule analogously, that is, it is a quadruple (3.2) (A, B, E, trA) where A is unital, trA is a faithful finite trace on A, and E is an A-B-equivalence bimodule. Remark 3.2. By Proposition 2.15 we may always induce a possibly unbounded trace trA on A given a left Gabor bimodule (A, B, E, trB). Indeed, this will be of great importance in the sequel, and we will use trA without mentioning that it is induced by trB. Likewise with trB if we treat the case of right Gabor bimodules. We are also interested in Gabor bimodules possessing some regularity, see Section 4. Definition 3.3. A left Gabor bimodule with regularity is a septuple (3.3) such that (A, B, E, trB, A, B, E), (1) (A, B, E, trB) is a left Gabor bimodule. (2) A ⊂ A and B ⊂ B are dense Banach ∗-subalgebras. (3) E ⊂ E is an A-B-pre-equivalence bimodule. (4) B is spectral invariant in B with the same unit. We define a right Gabor bimodule with regularity analogously. The rest of this section will be devoted to exploring properties of Gabor bimodules, mostly left Gabor bimodules. In Section 4 we show that Gabor bimodules over twisted group C ∗-algebras for LCA groups are Rieffel's Heisenberg modules and provide a different approach to Gabor analysis. We start with some basic definitions from [12]. We restrict to a single generator in this section, and extend the results to finitely many generators in 8 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS Section 3.2. Indeed, we will see in that section that even the case of finitely many gener- ators can be reduced to the case of a single generator of an associated Morita equivalence bimodule. Definition 3.4. For g ∈ E we define the analysis operator by (3.4) and the synthesis operator : (3.5) Φg : E → A f 7→ •h f, gi, Ψg : A → E a 7→ a · g. An elementary computation shows that Φ∗ g = Ψg. Remark 3.5. As E is an A-B-bimodule, we could just as well have defined the analysis operator and the synthesis operator with respect to the B-valued inner product. Indeed we will need this later, but it will then be indicated by writing ΦB g . Unless otherwise indicated the analysis operator and synthesis operator will be with respect to the left inner product module structure. Definition 3.6. For g, h ∈ E we define the frame-like operator Θg,h to be (3.6) Θg,h : E → E f 7→ •h f, gih. In other words, Θg,h = ΨhΦg = Φ∗ hΦg. The frame operator of g is the operator (3.7) Θg := Θg,g = Φ∗ gΦg : E → E f 7→ •h f, gig. Remark 3.7. The module frame operator Θg is a positive operator since Θg = Φ∗ gΦg. Definition 3.8. We say g ∈ E generates a (single) module frame for E if Θg is an invertible operator E → E. Equivalently, there exist constants C, D > 0 such that (3.8) C •h f, f i ≤ •h f, gi •h g, f i ≤ D •h f, f i, holds for all f ∈ E. Remark 3.9. When {g} is a module frame for E, Θg is a positive invertible operator on E. What follows will largely be a study of g and h in E such that Θg,h is invertible, and how this relates to inner product inequalities for a localization HE of E. Our interest in this question is due to the fact that in certain cases module frames can be localized to obtain Hilbert space frames, see Section 4. We begin with a result which generalizes the Wexler-Raz biorthogonality condition for Gabor frames, which we also look at in Section 4. Proposition 3.10 (Wexler-Raz for Gabor modules). Let g, h in E. Then f = Θg,hf = Θh,gf for all f ∈ E if and only if B is unital and hg, h i• = hh, g i• = 1B. Proof. Suppose f = Θh,gf = Θg,hf for all f ∈ E. Then E is a finitely generated projective A-module as it is generated by g, and we can use g and h to make the maps r and s from the proof of Proposition 2.11. Hence B ∼= KA(E) = EndA(E), and as EndA(E) is unital, we deduce B is unital. By Morita equivalence f = Θg,hf = •h f, gih = f hg, h i• = f hg, h i• GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 9 for all f ∈ E. Since B acts faithfully on E we deduce hg, h i• = 1B. Then also hh, g i• = hg, h i• ∗ = 1∗ B = 1B. Conversely, suppose B is unital and hg, h i• = hh, g i• = 1B. Then f = f 1B = f hg, h i• = •h f, gih = Θg,hf, and f = f 1B = f hh, g i• = •h f, hig = Θh,gf, which finishes the proof. (cid:3) The following result showcases a duality between certain A-submodules of E and B- submodules of E, which is very particular to our setting of Morita equivalence bimodules. Proposition 3.11. For any g, h ∈ E the following two statements are equivalent. (i) •h f, hig = f for all f ∈ Ag. (ii) f = ghh, f i• for all f ∈ gB. Proof. Suppose first •h f, hig = f for all f ∈ Ag. By Morita equivalence of A and B, f = f hh, g i• for all f ∈ Ag, hence hh, g i• fixes all elements in Ag. In particular, since A has an approximate unit, g ∈ Ag, so ghh, g i• = g. Now let f ′ ∈ gB. We then write f ′ = gb for some b ∈ B, and so we deduce ghh, f ′ i• = ghh, gb i• = ghh, g i• b = gb = f ′, since ghh, g i• = g by the above. We extend the reconstruction formula to all of gB by continuity. The proof of the converse is completely analogous. (cid:3) In the special case where Proposition 3.11 (i) holds for all f ∈ E we get another recon- struction formula. Note the (subtle) difference in the placement of g and h in the statement compared to statement (ii) in the preceding proposition. Proposition 3.12. Let g, h ∈ E be so that •h f, hig = f for all f ∈ E. Then f = hhg, f i• for all f ∈ hB. Proof. Suppose that •h f, hig = f for all f ∈ E. Then E is finitely generated and projective as an A-module as before, since it is singly generated by g, and we may use g and h to make the maps r and s from the proof of Proposition 2.11. Hence B ∼= KA(E) = EndA(E) and B is unital. We may rewrite the equality to f = f hh, g i• for all f ∈ E, which implies hh, g i• = 1B as B acts faithfully on E. But then hg, h i• = hh, g i• ∗ = 1∗ B = 1B as well. Then if we let f ∈ hB we may write f = hb for some b ∈ B, and we get hhg, f i• = hhg, hb i• = hhg, h i• b = h1Bb = hb = f. We extend the reconstruction formula to hB by continuity. (cid:3) Note that in the setting of Proposition 3.11 we may of course interchange g and h. However the subspaces Ah and Ag do not need to coincide. We may however guarantee Ag = Ah when h has a special form. 10 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS Lemma 3.13. Let g ∈ E be such that ΘgAg is invertible as a map Ag → Ag. For h = Θg−1 Ag g we have Ag = Ah. Proof. Let f ∈ Ag. As Θg, and hence also ΘgAg, is an A-module operator, so is Θg−1 Thus we get Ag . f = Θg−1 Ag Θgf = Θg−1 Ag (•h f, gig) = •h f, giΘg−1 Ag g = •h f, gih ∈ Ah. Hence we have Ag ⊂ Ah. Also g ∈ Ag as A has a left approximate unit, and as Θg is invertible as a map Ag → Ag it follows that h = Θg−1 (cid:3) Ag g ∈ Ag. Hence Ag = Ah. In the remaining part of the article we focus mostly on the case where Θg,h is invertible as a map E → E. Definition 3.14. If g ∈ E is such that Θg is invertible, then h = Θ−1 canonical dual atom of g. g g is called the Remark 3.15. Note that if g is such that Θg : E → E is invertible, then Ag = E. To see this, let f ∈ E. Then f = ΘgΘ−1 g f = •h Θ−1 g f, gig ∈ Ag. Given g ∈ E such that Θg : E → E is invertible and with h = Θ−1 g g the canonical dual atom, one may ask what the canonical dual atom of h is. The following lemma tells us that it is exactly what one would expect from Hilbert space frame theory. Lemma 3.16. Let g ∈ E be such that Θg : E → E is invertible, and let h = Θ−1 Θhg = h. Moreover, Θh : E → E is invertible with Θ−1 atom of h is g. g g. Then h = Θg, and the canonical dual Proof. With h = Θ−1 g g the identity Θhg = h is established as follows: Θhg = •h g, hih = •h g, Θ−1 g giΘ−1 g g g ΘgΘ−1 g g, gig) = Θ−1 g g = Θ−1 = Θ−1 g (•h Θ−1 g g = h. We proceed to show that Θh is invertible. By Lemma 3.13 Ag = Ah, and we know Ag = E. Now let f ∈ Ag, and write f = ag for some a ∈ A. We then have ΘgΘhf = Θg(•h f, hih) = Θg(•h ag, hih) g giΘ−1 g g) = aΘgΘ−1 g (•h Θ−1 g g, gig) = aΘg(•h g, Θ−1 = a •h Θ−1 = ag = f. g g, gig = aΘgΘ−1 g g Likewise we have ΘhΘgf = ΘhΘg(ag) = aΘhΘg(g) = aΘh(•h g, gig) = a •h •h g, gig, hih = aΘ−1 = aΘ−1 g (•h Θ−1 g (•h g, gig) = aΘ−1 g (•h g, gig), gig) = aΘ−1 g Θgg g (•h Θ−1 g Θgg, gig) = ag = f. GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 11 Ag is dense in E, and by extending the reconstruction formulas to all of E by continuity it follows that Θ−1 h = Θg. Then the canonical dual atom of h is Θ−1 h h = Θgh = •h h, gig = ΘgΘ−1 g g = g, which proves the result. (cid:3) The following proposition tells us that g and Θ−1 g g then indeed have the desired prop- erties as described in Proposition 3.11. Proposition 3.17. Let g ∈ E be such that Θg is invertible and let h = Θ−1 g g. Then (1) •h f, gih = f for all f ∈ E = Ah = Ag. (2) f ′ = hhg, f ′ i• for all f ′ ∈ hB (3) •h f, hig = f for all f ∈ E = Ah = Ag. Proof. Note first that E = Ag = Ah by Lemma 3.13. Now, let f ∈ Ah and write f = ah for some a ∈ A. Then we have •h f, gih = •h ah, gih = a •h h, gih = a •h Θ−1 g g, gih = a •h g, Θ−1 g gih = a •h g, hih = aΘhg = ah = f, where we have used Θhg = h, which holds by Lemma 3.16. We extend the reconstruction formula by continuity so it is valid for all f ∈ E. By Proposition 3.11 this also implies f ′ = hhg, f ′ i• for all f ′ ∈ hB. Hence items (1) and (2) are true. To show (3), let f ∈ E. Then •h f, hig = •h f, Θ−1 g gig = •h Θ−1 g f, gig = ΘgΘ−1 g f = f, so item (3) is also true. (cid:3) We may also prove the following additional reconstruction formula when h is the canon- ical dual atom. Note the (subtle) difference in where h and g are in the reconstruction formula compared to Proposition 3.11. Proposition 3.18. Let g ∈ E be such that Θg is invertible, and let h = Θ−1 g g. Then f = hhg, f i• for all f ∈ gB and f ′ = ghh, f ′ i• for all f ′ ∈ hB. As a consequence, gB = hB. Proof. Suppose f in gB. For g and h as stated, we have f = •h f, hig for all f ∈ E. By Proposition 3.11 we then have f = ghh, f i• for all f ∈ gB. Then f = ghh, f i• = •h g, hif = •h g, Θ−1 g gif = •h Θ−1 g g, gif = hhg, f i• . The second statement follows from noting that our assumptions imply Ag = Ah = E by Lemma 3.13 and the fact that the canonical dual of h is g by Lemma 3.16. Then we may simply interchange g and h in the argument for the first assertion. Lastly we prove gB = hB. We know g ∈ gB as B has an approximate unit, so Likewise, h ∈ hB and so This finishes the proof. g = Θgh = hhg, g i• ∈ hB. h = Θhg = ghh, h i• ∈ gB. (cid:3) 12 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS There is a correspondence between projections in Morita equivalent C ∗-algebras, see for example [28]. We formulate the following variant. Let E be an A-B-equivalence bimodule, and let B be unital. Then there is a way of constructing idempotents in A. This is the content of the following proposition. Proposition 3.19. Let E be an A-B-equivalence bimodule between a C ∗-algebra A and a unital C ∗-algebra B. If g, h ∈ E are such that hg, h i• = 1B, then •h g, hi is an idempotent in A. In particular, the canonical dual atom h = Θ−1 g g yields a projection •h g, hi in A. Proof. From hg, h i• = 1B = 1∗ B = hh, g i•, we get •h g, hi •h g, hi = •h •h g, hig, hi = •h ghh, g i•, hi = •h g · 1B, hi = •h g, hi, so •h g, hi is an idempotent in A. If h = Θ−1 g g, we also have •h g, hi = •h g, Θ−1 g gi = •h Θ−1 g g, gi = •h h, gi = •h g, hi∗, so •h g, hi is a projection in A. (cid:3) One of the cornerstones of Gabor analysis is the duality principle, see for example [9, 19, 29]. One of the main intentions of this investigation is a reformulation of this duality principle in our module framework. To this end we introduce the following operator. For an element g ∈ E we define the B-coefficient operator by (3.9) ΦB g : E → B f 7→ hg, f i• . Note that this operator is B-adjointable with adjoint g )∗b 7→ g · b. (3.10) (ΦB We are now in the position to state and prove the module version of the duality principle. Proposition 3.20 (Module Duality Principle). Let g ∈ E. The following are equivalent. (1) Θg : E → E is invertible. (2) ΦB g )∗ : B → B is an isomorphism. g (ΦB Proof. We show that both statements are equivalent to hg, g i• being invertible in B. Sup- pose Θg is invertible. Then E is finitely generated and projective as an A-module, as we can make the maps r and s from the proof of Proposition 2.11 using g and Θ−1 g g. Thus B is unital. As statement (1) is equivalent to hg, g i• being invertible in B. On the other hand, Θgf = f hg, g i•, (3.11) ΦB g (ΦB g )∗b = ΦB g (g · b) = hg, g · b i• = hg, g i• b. g (ΦB Since ΦB is unital and the statement is equivalent to hg, g i• being invertible in B. g )∗ ∈ EndB(B) and B is an ideal in EndB(B), statement (2) implies that B (cid:3) In Gabor analysis one is often concerned with the regularity of the atoms generating a Gabor frame, see Section 4. In case g is so that Θg is invertible on all of E with g ∈ E, and B ⊂ B is spectral invariant Banach ∗-subalgebra with the same unit as B, the canonical dual atom has the following important property. Proposition 3.21. Let E be an A-B-equivalence bimodule, with an A-B-pre-equivalence bimodule E ⊂ E. Suppose B ⊂ B is spectral invariant with the same unit. If g ∈ E is such that Θg : E → E is invertible, then the canonical dual Θ−1 g g is in E as well. GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 13 Proof. For f ∈ E we have We deduce that hg, g i• is invertible in B and Θgf = •h f, gig = f hg, g i• . Θ−1 g g = ghg, g i• −1 . But as g ∈ E we have hg, g i• ∈ B. By spectral invariance of B in B it follows that hg, g i• −1 ∈ B. Then, since EB ⊂ E, it follows that Θ−1 g g = ghg, g i• −1 ∈ E, which is the desired assertion. (cid:3) 3.2. Extending to several generators. We extend the above theory to several genera- tors. Indeed we will lift the A-B-equivalence bimodule E to an Mn(A)-Md(B)-equivalence bimodule, for d, n ∈ N, and consider a type of module frame in this matrix setting. We will see in Section 4 that this generalizes n-multiwindow d-super Gabor frames of [18]. Note that we will index an n × d-matrix by (i, j), i ∈ Zn, j ∈ Zd, that is, we start indexing at 0. The reason for this is that in Section 4 we will need to incorporate the groups Zk, k ∈ N. Here Zk denotes the group Z/(kZ). We will consider Mn,d(E) as an Mn(A)-Md(B)-bimodule. Define an Mn(A)-valued inner product on Mn,d(E) by •[ −, −] : Mn,d(E) × Mn,d(E) → Mn(A) (3.12) (f, g) 7→ Xk∈Zd •h f0,k, g0,ki •h f1,k, g0,ki ... •h fn−1,k, g0,ki •h f0,k, g1,ki •h f1,k, g1,ki ... •h fn−1,k, g1,ki . . . . . . . . . . . . •h f0,k, gn−1,ki •h f1,k, gn−1,ki ... •h fn−1,k, gn−1,ki   . The action of Mn(A) on Mn,d(E) is defined in the natural way, that is (af )i,j = Xk∈Zn ai,kfk,j, for a ∈ Mn(A) and f ∈ Mn,d(E). Likewise we define an Md(B)-valued inner product on Mn,d(E) in the following way [−, − ]• : Mn,d(E) × Mn,d(E) → Md(B) (3.14) (f, g) 7→ Xk∈Zn hfk,0, gk,0 i• hfk,1, gk,0 i• ... hfk,0, gk,1 i• hfk,1, gk,1 i• ... hfk,d−1, gk,0 i• hfk,d−1, gk,1 i• . . . . . . . . . . . . hfk,0, gk,d−1 i• hfk,1, gk,d−1 i• ... hfk,d−1, gk,d−1 i• The right action of Md(B) on Mn,d(E) is defined by   . (f b)i,j = Xk∈Zd fi,kbk,j for f ∈ Mn,d(E) and b ∈ Md(B). With this setup, Mn,d(E) becomes an Mn(A)-Md(B)-equivalence bimodule. Indeed it is not hard to verify the three conditions of Definition 2.9. Verifying conditions ii) and iii) is a matter of verifying the statements in each matrix element using that E is an A-B- equivalence bimodule. Verifying condition i) is a matter of getting density in each matrix entry by choosing elements of Mn,d(E) in the correct way. Namely, if we want to get the     (3.13) (3.15) 14 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS elements in place (i, j) in Mn(A), then we may for example pair elements of Mn,d(E) with nonzero entry only in place (i, k) with elements of Mn,d(E) with nonzero entry only in place (j, k), for some k ∈ Zd. The analogous procedure holds for Md(B). Density then follows by •h E, Ei = A and hE, E i• = B. In particular, we have for f, g, h ∈ Mn,d(E) that (3.16) and also (3.17) •[ f, g]h = f [g, h ]•, Mn(A) = KMd(B)(Mn,d(E)), Md(B) = K Mn(A)(Mn,d(E)). Also, since the new inner products are defined using the inner products •h −, −i and h−, − i•, we see that in case we have Banach ∗-subalgebras A ⊂ A and B ⊂ B, as well as an A-B-subbimodule E ⊂ E as above, we get •[ Mn,d(E), Mn,d(E)] ⊂ Mn(A), [Mn,d(E), Mn,d(E) ]• ⊂ Md(B), as well as Mn(A)Mn,d(E) ⊂ Mn,d(E), Mn,d(E)Md(B) ⊂ Md(E). We wish to reduce the matrix algebra case to the Gabor bimodule case of Section 3.1, so we need to guarantee that spectral invariance of Banach ∗-subalgebras lifts to matrices. For convenience we include the following result. Lemma 3.22 ([30]). If B is a spectral invariant Banach subalgebra of a Banach algebra B, then Mm(B) is a spectral invariant Banach subalgebra of Mm(B), for all m ∈ N. For g ∈ Mn,d(E) we define as in Section 3.1 the analysis operator (3.18) Φg : Mn,d(E) → Mn(A) f 7→ •[ f, g] which has as adjoint the operator (3.19) Φg : Mn(A) → Mn,d(E) a 7→ ag. Using these we also define the frame-like operator Φ∗ hΦg =: Θg,h : Mn,d(E) → Mn,d(E) by (3.20) Θg,hf = •[ f, g]h, for f ∈ Mn,d(E), and the frame operator Φ∗ gΦg =: Θg : Mn,d(E) → Mn,d(E) by (3.21) Θgf = •[ f, g]g for f ∈ Mn,d(E). As noted in Section 3.1, Θg is a positive operator. For simplicity, and since it is the case we will most often consider, suppose in the following that B is unital with a faithful finite trace. There is then an induced (possibly unbounded) trace on A as in Section 2. We may lift these traces to the matrix algebras. Indeed, there are traces on Mn(A) and Md(B) satisfying (3.22) trMn(A)(•[ f, g]) = trMd(B)([g, f ]•) GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 15 for all f, g ∈ Mn,d(E). They are given by (3.23) trMn(A)(•[ f, g]) = trMd(B)([f, g ]•) = 1 n Xi∈Zn n Xi∈Zd 1 trA(•[ f, g]i,i), trB([f, g ]•i,i). The trace on Md(B) extends to a finite trace on the whole algebra, but the same might not be true for the densely defined trace on Mn(A). It is however true if A, and hence also Mn(A), is unital. Remark 3.23. The normalization on the traces in (3.23) is so that if A is unital, then trMn(A)(1Mn(A)) = 1, that is, trMn(A) is a faithful tracial state. In general trMd(B) will not be a state, even if Md(B) is unital. The following lemma may be verified by elementary computations. Lemma 3.24. Let B be a unital C ∗-algebra. induced mapping trMm(B) on the matrix algebra Mm(B), m ∈ N, given by If trB is a faithful trace on B, then the (3.24) trMm(B)(b) = Xi∈Zm trB(bi,i), for b ∈ Mm(B) is also a faithful trace. We summarize the preceding discussion in the following proposition which allows us to study the Mn(A)-Md(B)-equivalence bimodule Mn,d(E) by studying the A-B-bimodule E. Proposition 3.25. Let (A, B, E, trB) be a left Gabor bimodule. Then for all n, d ∈ N, the quadruple (Mn(A), Md(B), Mn,d(E), trMd(B)) with the above defined actions, inner products, and traces is also a left Gabor bimodule. Furthermore, if (A, B, E, trA, trB, A, B, E) is a left Gabor bimodule with regularity, then for all n, d ∈ N, the septuple (Mn(A), Md(B), Mn,d(E), trMd(B), Mn(A), Md(B), Mn,d(E)), with the above defined actions, inner products, and traces is also a left Gabor bimodule with regularity. The analogous statements hold for right Gabor bimodules. Since our focus is on the description of frames in equivalence bimodules for Morita equiv- alent C ∗-algebras, we want to do this now on the matrix algebra level and thus introduce an appropriate notion of module frames for the matrix-valued equivalence bimodules. Definition 3.26. Let g = (gi,j)i∈Zn,j∈Zd ∈ Mn,d(E). We say g generates a module (n, d)- matrix frame for E with respect to A if there exists h = (hi,j)i∈Zn,j∈Zd ∈ Mn,d(E) for which (3.25) fr,s = Xk∈Zd Xl∈Zn •h fr,k, gl,kihl,s, holds for all f = (fi,j)i∈Zn,j∈Zd ∈ Mn,d(E), r ∈ Zn, and s ∈ Zd. 16 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS By definition of the above Hilbert Mn(A)-module structure on Mn,d(E), we see that g ∈ Mn,d(E) generates a module (n, d)-matrix frame for E with respect to A if and only if there is h ∈ Mn,d(E) such that (3.26) f = •[ f, g]h for all f ∈ Mn,d(E). In other words, g generates a module (n, d)-matrix frame for E with respect to A if and only if g generates a single module frame for Mn,d(E) with respect to Mn(A). When (3.26) is satisfied Mn,d(E) is finitely generated projective as an Mn(A)- module, so as before it follows by Proposition 2.11 that Md(B) is unital. Then B is also unital. By the identity f = •[ f, g]h = f [g, h ]•, we deduce that (3.26) is satisfied if and only if Md(B) is unital and [g, h ]• = 1Md(B). Remark 3.27. By the above discussion it follows that finding module (n, d)-matrix frames for the A-B-equivalence bimodule E is the same as finding g, h ∈ Mn,d(E) such that [g, h ]• = 1Md(B). That is, it is the same as finding single module frames for Mn,d(E) as an Mn(A)-module. By Lemma 3.22, the corresponding statement is true of finding module frames with regularity. Hence all results of Section 3.1 can be carried over to the setup in this section. Even though all results of Section 3.1 lift to the induced matrix algebra setup, we want to discuss explicitly two results relating the lifted traces. We show in Section 4 that these two results extend the density theorems of Gabor analysis to Gabor bimodules. Since we in Theorem 3.28 talk about left Gabor bimodules and in Theorem 3.29 talk about right Gabor bimodules, we will for the sake of avoiding confusion not have any convention on the normalization of traces in the two results. Theorem 3.28. Let (A, B, E, trB) be a left Gabor bimodule. If, in addition, A is unital and g ∈ Mn,d(E) is such that Θg : E → E is invertible, then (3.27) d trB(1B) ≤ n trA(1A). Proof. The assumption that Θg is invertible implies [g, g ]• is invertible. Then (3.28) u = Θ−1 g g = g[g, g ]• −1 is the canonical dual frame for Mn,d(E). We have [g, u ]• = [u, g ]• = 1Md(B), and by Proposition 3.19, •[ g, u] is a projection in Mn(A). Using Lemma 3.24 and •[ g, u] ≤ 1Mn(A) in Mn(A), as well as (3.23), we get d trB(1B) = n · 1 n dXi=1 trB(1B) = n trMd(B)(1Md(B)) = n trMd(B)([u, g ]•) = n trMn(A)(•[ g, u]) ≤ n trMn(A)(1Mn(A)) = n · 1 n nXi=1 trA(1A) = n trA(1A). (cid:3) Theorem 3.29. Let (A, B, E, trA) be a right Gabor bimodule. If, in addition, B is unital and g ∈ Mn,d(E) is such that ΦgΦ∗ g : Mn(A) → Mn(A) is an isomorphism, then (3.29) d trB(1B) ≥ n trA(1A). GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 17 Proof. The assumptions imply •[ g, g]−1 ∈ Mn(A), so it follows as in Section 3.1 that 1Mn(A) = •[ g, g]−1 •[ g, g] = •[ •[ g, g]−1g, g], and [•[ g, g]−1g, g ]• is a projection in Md(B) by Proposition 3.19. Since B is unital, then, using Lemma 3.24 together with [•[ g, g]−1g, g ]• ≤ 1Md(B) in Md(B), as well as (3.23), we get n trA(1A) = n · 1 n nXi=1 trA(1A) = n trMn(A)(1Mn(A)) = n trMn(A)(•[ •[ g, g]−1g, g]) = n trMd(B)([g, •[ g, g]−1g ]•) ≤ n trMd(B)(1Md(B)) = n · 1 n dXi=1 trB(1B) = d trB(1B). (cid:3) 3.3. From a Gabor bimodule to its localization. In [21] the existence of multi-window Gabor frames for L2(Rd) with windows in Feichtinger's algebra was proved through con- siderations on a related Hilbert C ∗-module. Furthermore, in [22] projections in noncom- mutative tori were constructed from Gabor frames with sufficiently regular windows. Thus being able to pass from an equivalence bimodule E to a localization HE and back is quite important, and we dedicate this section to results on this procedure. We will interpret this in terms of standard Gabor analysis in Section 4, and we will explain how L2(G), for G a second countable LCA group, relates to HE for specific modules E which arise in the study of twisted group C ∗-algebras. We denote by (−, −)E the inner product on the localization of E in trA. Concretely, we have (f, g)E = trA(•h f, gi). Proposition 3.30. Let (A, B, E, trB) be a left Gabor bimodule, and let g ∈ E. Then there exists an h ∈ E such that we have •h f, gih = f for all f ∈ E if and only if there exist constants C, D > 0 such that (3.30) C(f, f )E ≤ (f hg, g i•, f )E ≤ D(f, f )E for all f ∈ HE. In other words, g is a module frame for E if and only if the inequalities in (3.30) are satisfied for some C, D > 0. Proof. Suppose first that there is an h ∈ E such that •h f, gih = f for all f ∈ E. By Morita equivalence this implies f = •h f, gih = f hg, h i• 18 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS for all f ∈ E. As before, this implies 1B = hg, h i• = hh, g i•. Since trB is a positive linear functional we obtain (f, f )E = trA(•h f, f i) = trA(•h f hg, h i•hh, g i•, f i) = trA(•h f hg, hhh, g i• i•, f ) = trA(•h f hg, •h h, hig i•, f i) = trB(hf, f hg, •h h, hig i• i•) ≤ trB(hf, f hg, g i• k •h h, hik i•) = k •h h, hik trB(hf, f hg, g i• i•) = k •h h, hik trA(•h f hg, g i•, f i) = k •h h, hik(f hg, g i•, f )E, for all f ∈ E, where we have used Proposition 2.7 to deduce We then get the lower frame bound with C = k •h h, hik−1, that is hg, •h h, hig i• ≤ k •h h, hikhg, g i• . 1 k •h h, hik (f, f )E ≤ (f hg, g i•, f )E for all f ∈ E. By Proposition 2.8 all intermediate steps involve operators that extend to bounded operators on HE, so we may extend by continuity. We get the upper frame bound by use of Proposition 2.7 in the following manner (f hg, g i•, f )E = trA(•h f hg, g i•, f i) = trA(•h f hg, g i• ≤ khg, g i• 1/2, f hg, g i• 1/2 k2 trA(•h f, f i) 1/2i) = khg, g i• k trA(•h f, f i) = k •h g, gik(f, f )E , for all f ∈ E. Once again all intermediate steps involve operators that extend to bounded operators on HE by Proposition 2.8, so we may extend the result to all of HE. Thus we have shown that 1 k •h h, hik (f, f )E ≤ (f hg, g i•, f )E ≤ k •h g, gik(f, f )E for all f ∈ HE. Conversely, suppose there are C, D > 0 such that C(f, f )E ≤ (f hg, g i•, f )E ≤ D(f, f )E for all f ∈ HE. We wish to show that this implies there exist h ∈ E such that •h f, gih = f for all f ∈ E. The assumption implies that f 7→ f hg, g i• is a positive, invertible operator on HE. By Proposition 2.1 it follows that hg, g i• is invertible in B. Thus f 7→ f hg, g i• is a positive, invertible operator on E as well. Hence the operator is invertible with inverse Θg : E → E f 7→ •h f, gig = f hg, g i• Θ−1 g f = f hg, g i• −1 . GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 19 Define h := Θ−1 g g, and let f ∈ E be arbitrary. Then we have •h f, gih = •h f, giΘ−1 g Θgf = f, g g = Θ−1 g (•h f, gig) = Θ−1 from which the result follows. (cid:3) We are interested in module frames and module Riesz sequences, and their relationship to frames and Riesz sequences in Gabor analysis for LCA groups. To get results on Riesz sequences in Section 4 we need a module version of Riesz sequences which, when localized, yields the Riesz sequences we know from Gabor analysis. For this we let A be unital with a faithful trace trA, and we need to localize A as a Hilbert A-module in the trace trA. We let (a1, a2)A := trA(a1a∗ 2). The completion of A in this inner product will be denoted HA, and the action of A on HA is the continuous extension of the multiplication action (from the right) of A on itself. Proposition 3.31. Let (A, B, E, trA) be a right Gabor bimodule, and let g ∈ E. Then ΦgΦ∗ g : A → A is an isomorphism if and only if there exist C, D > 0 such that for all a ∈ A it holds that (3.31) C(a, a)A ≤ (ag, ag)E ≤ D(a, a)A. Proof. First suppose ΦgΦ∗ g : A → A is an isomorphism. Then, since by Proposition 2.6 •h ag, agi = a •h g, gia∗ ≤ k •h g, gikaa∗, we may deduce (ag, ag)A = trA(•h ag, agi) ≤ k •h g, gik trA(aa∗) = k •h g, gik(a, a)A. Hence in (3.31) we may set D = k •h g, gik. Since ΦgΦ∗ ΦgΦ∗ ga = a •h g, gi, it follows that there is •h g, gi−1 ∈ A. Then g : A → A is an isomorphism and (a, a)A = trA(aa∗)A •h g, gi1/2a∗) •h g, gi−1 = trA(a •h g, gi1/2 ≤ k •h g, gi−1k trA(a •h g, gia∗) = k •h g, gi−1k trA(•h ag, agi) = k •h g, gi−1k(ag, ag)E , which implies that we may set C = k •h g, gi−1k−1 in (3.31). All intermediate steps extend to HA by Proposition 2.8. Suppose now that (3.31) is satisfied. The lower inequality in (3.31) tells us that for all a ∈ A, (a(•h g, gi − C), a)A = trA(a(•h g, gi − C)a∗) = trA(a •h g, gia∗) − C trA(aa∗) = trA(•h ag, agi) − C trA(aa∗) = (ag, ag)E − C(a, a)A ≥ 0. Note that we need the upper inequality of (3.31) to extend all intermediate steps to HA via Proposition 2.8. It follows that •h g, gi is a positive invertible operator on HA ⊃ A. By Proposition 2.1 it follows that •h g, gi is invertible in A. Then, since it follows that ΦgΦ∗ g : A → A is an isomorphism. (cid:3) ΦgΦ∗ ga = a •h g, gi, 20 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS Both Proposition 3.30 and Proposition 3.31 were proved for Gabor bimodules, so by Remark 3.27 the results lift to the corresponding matrix setting of Section 3.2. Remark 3.32. Note that in the proofs of the two preceding results the upper bounds in (3.30) and (3.31) were both satisfied with D = k •h g, gik. We will see in Section 4 that in the Gabor analysis setting, this means that all atoms coming from the Hilbert C ∗-module are Bessel vectors for the localized frame system. Remark 3.33. The two preceding results actually have shorter proofs using Proposition 2.1 in a more direct way, but these proofs would not give us values for C and D in (3.30) and (3.31), only the existence. The values of the constants are of interest on their own, see Section 4. For use in Section 4, we introduce the following notion. Definition 3.34. Let (A, B, E, trA) be a right Gabor bimodule, and let g ∈ E. If ΦgΦ∗ g : A → A is an isomorphism, we say g generates a module Riesz sequence for E with respect to A. If h ∈ Mn,d(E) generates a module Riesz sequence for Mn,d(E) with respect to Mn(A), we will also say that h generates a module (n, d)-matrix Riesz sequence for E with respect to A. 4. Applications to Gabor analysis In this section we show how the above results reproduce some of the core results of Gabor analysis for LCA groups. We will see how some of the cornerstones of Gabor analysis on LCA groups are trivial consequences of the above framework. Of particular interest is the reproduction of some of the main results of [18] on n-multiwindow d-super Gabor frames with windows in the Feichtinger algebra. Indeed we will show the corresponding results for localized module (n, d)-matrix frames, which generalize n-multiwindow d-super Gabor frames. To present the results we will need to explain how time frequency analysis on LCA groups relates to Morita equivalence of twisted group C ∗-algebras. In the interest of brevity, we refer the reader to [18] for a more in-depth treatment of time frequency analysis and its relation to twisted group C ∗-algebras, and to [16] for a survey on the Feichtinger algebra. turn them into LCA groups as well, and we may equip them with their respective Haar the Plancherel theorem holds. By Λ we denote a closed subgroup of the time-frequency Throughout this section, we fix a second countable LCA group G and let bG be its dual group. We fix a Haar measure µG on G and normalize the Haar measure µ bG on bG such that plane G × bG. The induced topologies and group multiplications on Λ and (G × bG)/Λ measures. Having fixed the Haar measures on G, bG, and Λ, we will assume (G × bG)/Λ is for all f ∈ L1(G × bG) we have equipped with the unique Haar measure such that Weil's formula holds, that is, such that ZG× bG f (ξ)dµG× bG =Z(G× bG)/ΛZΛ f (ξ + λ)dµΛ(λ)dµ(G× bG)/Λ( ξ), ξ = ξ + Λ. In this setting we can define the size of Λ by s(Λ) :=Z(G× bG)/Λ 1dµ(G× bG)/Λ. Note that s(Λ) is finite if and only if Λ is cocompact in G × bG. GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 21 For any x ∈ G and ω ∈ bG we define the translation operator (or time shift) Tx by Txf (t) = f (t − x), t ∈ G, and the modulation operator (or frequency shift) Eω by Eωf (t) = ω(t)f (t), t ∈ G. These operators are unitary on L2(G), and satisfy the commutation relation (4.1) EωTx = ω(x)TxEω. For any ξ = (x, ω) ∈ G × bG we may then define the time-frequency shift operator π(ξ) = π(x, ω) = EωTx. (4.2) We define the 2-cocycle (4.3) c : (G × bG) × (G × bG) → T for ξ1 = (x1, ω1), ξ2 = (x2, ω2) ∈ G × bG. Note then that (4.4) π(ξ1)π(ξ2) = c(ξ1, ξ2)π(ξ1 + ξ2), (ξ1, ξ2) 7→ ω2(x1) and (4.5) π(ξ1)π(ξ2) = c(ξ1, ξ2)c(ξ2, ξ1)π(ξ2)π(ξ1) = cs(ξ1, ξ2)π(ξ2)π(ξ1), where we have introduced the symplectic cocycle cs by (4.6) We also remark that (4.7) cs : (G × bG) × (G × bG) → T (ξ1, ξ2) 7→ ω2(x1)ω1(x2). π(ξ)∗ = c(ξ, ξ)π(−ξ) for all ξ ∈ G × bG. Using the symplectic cocycle cs we define for a closed subgroup Λ ⊂ G × bG the adjoint subgroup Λ◦ by (4.8) Λ◦ := {ξ ∈ G × bG cs(ξ, λ) = 1 ∀λ ∈ Λ}. and only if Λ◦ is discrete. With these identifications we put on Λ◦ the Haar measure such Then (Λ◦)◦ = Λ and cΛ◦ ∼= (G × bG)/Λ, see for example [17]. Note that Λ is cocompact if that the Plancherel theorem holds with respect to Λ◦ and (G × bG)/Λ. We define the short time Fourier transform with respect to g ∈ L2(G) as the operator (4.9) Vg : L2(G) → L2(G × bG) Vgf (ξ) = hf, π(ξ)gi, for ξ ∈ G × bG. The Feichtinger algebra S0(G) is then defined by S0(G) := {f ∈ L2(G) Vf f ∈ L1(G × bG)}. A norm on S0(G) is given by (4.10) (4.11) kf kS0(G) := kVgf kL1(G× bG), for some g ∈ S0(G) \ {0}. It is a nontrivial fact that all elements of S0(G)\{0} determine equivalent norms on S0(G). In case G is discrete it is known that S0(G) = ℓ1(G) with equivalent norms. Furthermore, S0(G) consists of continuous functions and is dense in both L1(G) and L2(G). 22 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS For two functions F1, F2 over Λ ⊂ G × bG we define the twisted convolution by F1(λ′)F2(λ − λ′)c(λ′, λ − λ′) dλ′, (4.12) F1♮F2(λ) :=ZΛ and the twisted involution F ∗ 1 (λ) := c(λ, λ)F1(−λ). In [18] it was shown that S0(Λ) is a Banach ∗-D-algebra for some D > 0 when equipped with twisted convolution, and indeed it is possible to choose an equivalent norm on S0(G) such that it becomes a Banach ∗- algebra. We denote the resulting Banach ∗-algebra by S0(Λ, c). Using this we may then define two Banach ∗-algebras (4.13) A := {a ∈ B(L2(G)) a =ZΛ B := {b ∈ B(L2(G)) b =ZΛ◦ a(λ)π(λ) dλ, a ∈ S0(Λ)}, b(λ◦)π(λ◦)∗ dλ◦, b ∈ S0(Λ◦)}. Note that A ∼= S0(Λ, c) and B ∼= S0(Λ◦, c) via the natural maps. We will use these identifications without mention in the sequel. The following was proved in [18]. Lemma 4.1. Suppose Λ ⊂ G × bG is a closed subgroup. Then ξ → π(ξ) is a faithful unitary c-projective representation of Λ. As a result, the integrated representation is a non-degenerate ∗-representation of S0(Λ, c). We may then obtain the minimal universal enveloping algebra C ∗ r (Λ, c) of S0(Λ, c) through the integrated representation of S0(Λ, c) on L2(G), that is, the representation (4.14) a · f =ZΛ a(λ)π(λ)f dλ, for a ∈ S0(Λ, c) and f ∈ L2(G). As Λ is abelian, hence amenable, the minimal and maximal enveloping algebras coincide, so we write C ∗(Λ, c) for the universal enveloping algebra of S0(Λ, c). We do the same for S0(Λ◦, c), and denote its universal enveloping C ∗-algebra by C ∗(Λ◦, c). Theorem 4.2 ([28]). The twisted group C ∗-algebras C ∗(Λ, c) and C ∗(Λ◦, c) are Morita equivalent. Indeed, S0(G) becomes a pre-equivalence bimodule between A and B as in Definition 2.9 when equipped with the inner products (4.15) and (4.16) and the actions (4.17) and (4.18) •h f, gi =ZΛ hf, g i• =ZΛ◦ a · f =ZΛ f · b =ZΛ◦ hf, π(λ)giπ(λ)dλ hg, π(λ◦)∗f iπ(λ◦)∗dλ◦, a(λ)π(λ)f dλ b(λ◦)π(λ◦)∗f dλ◦, with a ∈ A, b ∈ B, and f, g ∈ S0(G). That these are well-defined was noted in Section 3 of [18]. In the remainder of the section we denote by A the C ∗-completion of A, B the GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 23 C ∗-completion of B, E = S0(G), and by E the Hilbert C ∗-module completion of E. Hilbert C ∗-modules E as in this setting are called Heisenberg modules. Remark 4.3. The fact that we get the same twisted group C ∗-algebras by using S0(Λ, c) as we get when using the more traditional approach with L1(Λ, c) was noted in [2]. Since S0-functions are continuous, there are also well-defined canonical faithful traces on A and B given by (4.19) and (4.20) trA : A → C a 7→ a(0), trB : B → C b 7→ b(0). In general, these traces do not extend to A and B, but we will nonetheless denote them by trA and trB. These are indeed related as in (2.3), which can be seen from (4.21) trA(•h f, gi) = •h f, gi(0) = hf, giL2(G) = hg, f i•(0) = trB(hg, f i•). In our discussion the following two results are crucial. The first follows immediately by [23], and the second is a consequence of [14]. Lemma 4.4. C ∗(Λ, c) is unital if and only if Λ is discrete. Proposition 4.5. For a discrete subgroup Λ in G × bG the involutive Banach algebra S0(Λ, c) is spectral invariant in C ∗(Λ, c). Remark 4.6. Although the traces trA and trB do not in general extend to the algebras A and B, we can guarantee they extend in one case. Namely, trA extends to all of A if A is unital, which is equivalent to Λ being discrete. The same is of course true for B and trB, with the discreteness condition on Λ◦. This is due to the fact that the trace given by evaluation in the identity extends to twisted group C ∗-algebras when the underlying group is discrete [5, p. 951]. The case of Λ or Λ◦ being discrete is the case we will almost exclusively restrict to after Proposition 4.8. The following is now an immediate consequence. Proposition 4.7. Let Λ be cocompact, which implies Λ◦ is discrete. Then under the above conditions on A, B, E, trB the quadruple (A, B, E, trB) is a left Gabor bimodule. In addition, the septuple (A, B, E, trB, A, B, E) is a left Gabor bimodule with regularity. If Λ◦ is cocompact and thus Λ is discrete, then we obtain a right Gabor bimodule with regularity analogously. We may then reprove Theorem 3.9 of [18] in this framework. Proposition 4.8. Let Λ ⊂ G × bG be a closed subgroup. Then E is a finitely generated projective A-module if and only if Λ ⊂ G × bG is a cocompact subgroup. Also, E is a finitely generated projective A-module if and only if Λ ⊂ G × bG is cocompact. Proof. E is finitely generated and projective over A if and only if KA(E) = EndA(E). As E is an A-B-equivalence bimodule, this is equivalent to B being unital by Proposition 2.11. B is unital if and only if Λ◦ is discrete by Lemma 4.4, so equivalently (4.22) cΛ◦ ∼= (G × bG)/Λ 24 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS is compact, that is, Λ is cocompact in G × bG. Now, if Λ is cocompact, then B is unital, so we are in the situation of Proposition 2.14 by Proposition 4.5. Hence by the first part of this proposition it follows that E is a finitely generated projective A-module. Conversely, suppose E is a finitely generated projective A-module. Then E ∼= Anp isometrically for some n ∈ N and some p ∈ Mn(A). Passing to the completions we obtain E ∼= Anp, so E is a finitely generated projective A-module. By the first part of this proposition it follows that Λ is cocompact. (cid:3) Remark 4.9. Proposition 4.8 shows that we can only have finite module frames for E as an A-module if Λ is cocompact in G × bG. Since we wish to study the relationship between finite module frames and Gabor frames this is the case we care most about in the sequel. To get results on Gabor frames for L2(G) with windows in E from the above setup, we will need to localize certain subsets of the C ∗-algebras A and B, as well as the Morita equivalence bimodule E, just as explained in Section 2. For simplicity, let Λ be cocompact in G × bG from now on, unless otherwise specified. Then Λ◦ is discrete and trB is defined on all of B. The localization of B in trB is induced by the inner product (−, −)B given by (b1, b2)B := trB(b∗ 1b2). Since B is dense in B and trB is continuous, it follows that their localizations in trB are the same. For b1, b2 ∈ B we then have (b1, b2)B = trB(b∗ 1b2) b1(λ◦)π(λ◦)∗ Xξ∈Λ◦ b2(ξ)π(ξ)) b1(λ◦)b2(ξ)c(λ◦, λ◦)π(−λ◦)π(ξ)) b1(λ◦)b2(ξ)c(λ◦, λ◦)c(−λ◦, ξ)π(−λ◦ + ξ)) b1(λ◦ + ξ)b2(ξ)c(λ◦ + ξ, λ◦ + ξ)c(−λ◦ − ξ, ξ)π(−λ◦)) = trB( Xλ◦∈Λ◦ = trB( Xλ◦∈Λ◦ Xξ∈Λ◦ = trB( Xλ◦∈Λ◦ Xξ∈Λ◦ = trB( Xλ◦∈Λ◦ Xξ∈Λ◦ = Xξ∈Λ◦ = Xξ∈Λ◦ b1(ξ)b2(ξ) b1(ξ)b2(ξ)c(ξ, ξ)c(−ξ, ξ) = hb1, b2iℓ2(Λ◦). As B = S0(Λ◦, c) = ℓ1(Λ◦, c) is dense in ℓ2(Λ◦), we may identify the localization HB of B with ℓ2(Λ◦). By [2, Proposition 3.2] we also obtain that the localization of E in trB is L2(G). Note that this is the same as the localization of E in trA by construction, and that there is an action of A on L2(G) by extending the action of A on E. It is slightly more tricky to localize subsets of A. Indeed, it is not in general possible as the trace might not be defined everywhere. However, even if A is not unital we may localize the algebraic ideal •h E, Ei ⊂ A in the trace trA. Indeed, by [2, Theorem 3.5], elements of E are such that whenever g ∈ E and f ∈ L2(G), then {hf, π(λ)gi}λ∈Λ ∈ L2(Λ). This is the property of being a Bessel vector, which we will discuss in more detail below. Hence GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 25 for any f, g ∈ E, we may identify •h f, gi ∈ A with (hf, π(λ)gi)λ∈Λ in L2(Λ) by doing the analogous procedure with trA as for trB above. We may do the same for the matrix algebras and matrix modules considered in Section 3.2. Note that •[ Mn,d(E), Mn,d(E)] = Mn,d(•h E, Ei). Adapting the setting of twisted group C ∗-algebras and Heisenberg modules above to the matrix algebra setting of Section 3.2 we see that we obtain the following identifications (4.23) HMd(B) = ℓ2(Λ◦ × Zd × Zd) HMn(•h E,Ei) = L2(Λ × Zn × Zn) HMn,d(E) = L2(G × Zn × Zd). Remark 4.10. Should Λ◦ be cocompact and therefore Λ discrete, we do the obvious changes. Also if both Λ and Λ◦ are discrete, that is, they are both lattices, then we may localize all of Mn(A) and all of Md(B). We can finally treat the analogs of Gabor frames in our framework. In what follows we will consider a novel type of Gabor frames. To ease notation we will for f ∈ L2(G × Zn × Zd) write fi,j instead of f (·, i, j), and the same for elements of L2(Λ × Zn × Zn) and L2(Λ◦ × Zd × Zd). Definition 4.11. Let Λ be a closed subgroup of G × bG. For g ∈ L2(G × Zn × Zd) we define the coefficient operator Cg by (4.24) Cg : L2(G × Zn × Zd) → L2(Λ × Zn × Zn) Cg(f ) = { Xm∈Zd hfk,m, π(λ)gl,mi}λ∈Λ,k,l∈Zn and the synthesis operator Dg by Dg : L2(Λ × Zn × Zn) → L2(G × Zn × Zd) (4.25) Dga = { Xm∈Zn ZΛ ak,m(λ)π(λ)gm,l dλ}k∈Zn,l∈Zd. Furthermore, we define the frame-like operator Sg,h = DhCg, and for brevity we write Sg for DgCg. We say Sg is the frame operator associated to g. We say g generates an (n, d)-matrix Gabor frame for L2(G) with respect to Λ if Sg : L2(G × Zn × Zd) → L2(G × Zn × Zd) is an isomorphism. Equivalently, the collection of time-frequency shifts (4.26) G(g; Λ) := {π(λ)gi,j λ ∈ Λ}i∈Zn,j∈Zd is a frame for L2(G×Zn ×Zd). We then say that G(g; Λ) is an (n, d)-matrix Gabor frame for L2(G). Equivalently, there exists h ∈ L2(G× Zn × Zd) such that for all f ∈ L2(G× Zn × Zd) we have (4.27) hfr,k, π(λ)gl,kiπ(λ)hl,sdλ, fr,s = Xk∈Zd Xl∈Zn ZΛ for all r ∈ Zn and s ∈ Zd. When g and h satisfy (4.27) we say G(g; Λ) and G(h; Λ) are a dual pair of (n, d)-matrix Gabor frames. If Λ is implicit, we may also say h is a dual (n, d)-matrix Gabor atom for g, or just a dual atom of g. Remark 4.12. The equivalence of the definitions of (n, d)-matrix Gabor frames given in Definition 4.11 follows by [7, Lemma 6.3.2] and Proposition 4.16 below. 26 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS Remark 4.13. When G(g; Λ) is an (n, d)-matrix Gabor frame for L2(G), there is always a dual (n, d)-matrix Gabor atom for g, namely h = S−1 g g. This is known as the canonical dual of g. Remark 4.14. One can verify that Cg = D∗ Hilbert spaces, just as for the module frame operator in Section 3. g. Thus Sg is always a positive operator between For general g ∈ L2(G × Zn × Zd) the operator Cg will not be bounded. Elements g such that Cg is bounded are of interest on their own. Definition 4.15. If g ∈ L2(G×Zn×Zd) is so that Cg : L2(G×Zn×Zd) → L2(Λ×Zn×Zn) is a bounded operator we say g is an (n, d)-matrix Gabor Bessel vector for L2(G) with respect to Λ, or that G(g; Λ) is an (n, d)-matrix Gabor Bessel system for L2(G). Equivalently, there is D > 0 such that for all f ∈ L2(G × Zn × Zd) we have (4.28) hf, f i ≤ DhCgf, Cgf i, which may also be written as (4.29) Xi∈Zn Xj∈Zd ZG fi,j(ξ)2dξ ≤ D Xk,l∈Zn ZΛ Xm∈Zd hfk,m, π(λ)gl,mi2dλ. The smallest D > 0 such that the condition of (4.28) holds is called the optimal Bessel bound of G(g; Λ), or just the optimal Bessel bound of g if the set Λ is clear from the context. The Gabor frames of Definition 4.11 seemingly generalize the n-multiwindow d-super Gabor frames of [18]. Indeed, we obtain n-multiwindow d-super Gabor frames if we only require reconstruction of f ∈ L2(G × Zd) and we identify L2(G × Zd) ⊂ L2(G × Zn × Zd) by embedding along a single element of Zn. Hence (4.27) generalizes both multiwindow Gabor frames and super Gabor frames as well, setting d = 1 or n = 1, respectively. However, we will in Proposition 4.34 show that any n-multiwindow d-super Gabor frame for L2(G) with respect to Λ is an (n, d)-matrix Gabor frame for L2(G) with respect to Λ. However, we continue to call them by separate names, since, as mentioned above, they are used for reconstruction in different Hilbert spaces. The following proposition was noted in the (n, 1)-matrix case in [2, Theorem 3.11], and its proof in the (n, d)-matrix Gabor case goes through the same except with more bookkeeping. Proposition 4.16. Let Λ ⊂ G × bG be closed and cocompact. For every g ∈ Mn,d(E), Cg : L2(G × Zn × Zd) → L2(Λ × Zn × Zn) is a bounded operator. In other words, every g ∈ Mn,d(E) is a Bessel vector. For ease of notation, the localization map in in Mn(A) will be denoted by ρMn(A), even though we might not be able to localize all of Mn(A). With the above definitions, the following calculation is justified for f, g ∈ Mn,d(E) ⊂ L2(G × Zn × Zd) by Proposition 4.16 ρMn(A)Φg(f ) = ρMn(A)(•[ f, g]) ZΛ = ρMn(A)({ Xm∈Zd = { Xm∈Zd hfk,m, π(λ)gl,miπ(λ)}k,l∈Zn ) hfk,m, π(λ)gl,mi}λ∈Λ,k,l∈Zn = CgρMn,d(E)(f ). Hence we obtain the following result. GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 27 Lemma 4.17. Let Λ ⊂ G × bG be closed and cocompact. For every g ∈ Mn,d(E), the module coefficient operator Φg localizes to give the coefficient operator Cg. Equivalently, the diagram Mn,d(E) ρMn,d(E) L2(G × Zn × Zd) Φg Cg Mn(A) ρMn(A) L2(Λ × Zn × Zn) commutes for all g ∈ Mn,d(E). Likewise one may obtain C ∗ g : Mn(•h E, Ei) → L2(G × Zn × Zd) for all g ∈ Mn,d(E). Note that the domain might be larger, but we cannot guarantee this unless A is unital, that is, when Λ is discrete. g ρMn(A) = ρMn,d(E)Φ∗ Lemma 4.18. Let Λ ⊂ G × bG be closed and cocompact. For every g ∈ Mn,d(E), the g localizes to the Gabor synthesis operator C ∗ g . Equivalently, module synthesis operator Φ∗ the diagram Mn(•h E, Ei) ρMn(A) L2(Λ × Zn × Zn) Φ∗ g C ∗ g Mn,d(E) ρMn,d(E) L2(G × Zn × Zd) commutes for every g ∈ Mn,d(E). Combining Lemma 4.17 and Lemma 4.18 we then obtain Proposition 4.19. Let Λ ⊂ G × bG be closed and cocompact. For all g, h ∈ Mn,d(E), Sg,hρMn,d(E) = ρMn,d(E)Θg,h, meaning the module frame-like operator Θg,h localizes to the frame-like operator Sg,h. Equivalently, the diagram Mn,d(E) ρMn,d(E) L2(G × Zn × Zd) Θg,h Sg,h Mn,d(E) ρMn,d(E) L2(G × Zn × Zd) commutes for all g ∈ Mn,d(E). As ρMn,d(E) : Mn,d(E) → ρMn,d(E)(Mn,d(E)) is a linear bijection intertwining both the A-actions and the B-actions, we see by Proposition 4.19 that for g ∈ Mn,d(E), Θg is invertible if and only if SgρMn,d (E)(Mn,d(E)) is invertible. But we also have the following result. Lemma 4.20. Let Λ ⊂ G × bG be closed and cocompact, and let g ∈ Mn,d(E). Then SgρMn,d(E)(Mn,d(E)) : ρMn,d(E)(Mn,d(E)) → ρMn,d(E)(Mn,d(E)) is invertible if and only if Sg : L2(G × Zn × Zd) → L2(G × Zn × Zd) 28 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS is invertible. Proof. Suppose first SgρMn,d(E)(Mn,d(E)) : ρMn,d(E)(Mn,d(E)) → ρMn,d(E)(Mn,d(E)) is in- vertible. Since any g ∈ Mn,d(E) is a Bessel vector by Proposition 4.16, we may extend the operator by continuity to obtain that Sg : L2(G × Zn × Zd) → L2(G × Zn × Zd) is invertible as well. Conversely, suppose Sg : L2(G × Zn × Zd) → L2(G × Zn × Zd) is invertible. Since Sg is the continuous extension of Θg, it then follows by Proposition 2.8 and Proposition 2.1 that Θg is invertible, which implies SgρMn,d (E)(Mn,d(E)) is invertible. (cid:3) Remark 4.21. From now on we will identify Mn,d(E) and its image in the localization, and we will do this without mention. Combining Proposition 4.19 and Lemma 4.20 we obtain the following important result. Proposition 4.22. Let Λ ⊂ G × bG be closed and cocompact. For g ∈ Mn,d(E) we have that Θg is invertible if and only if Sg is invertible. In other words, g generates a module (n, d)-matrix frame for E with respect to A if and only G(g; Λ) is an (n, d)-matrix Gabor frame for L2(G). We also have the following important corollary. Corollary 4.23. Let Λ ⊂ G × bG be closed and cocompact, and let g, h ∈ Mn,d(E). Then g and h generate dual (n,d)-matrix Gabor frames for L2(G) with respect to Λ if and only if [g, h ]• extends to the identity operator on L2(G × Zn × Zd). Proof. Suppose first g, h ∈ Mn,d(E) generate dual (n, d)-matrix Gabor frames for L2(G) with respect to Λ. Then we know that for all f ∈ Mn,d(E) we have f = •[ f, g]h = f [g, h ]•, from which we as before deduce that [g, h ]• = 1Md(B). This extends by continuity to the identity operator on all of L2(G × Zn × Zd). Conversely, if [g, h ]• extends to the identity operator on L2(G × Zn × Zd), then [g, h ]• acts as the identity on Mn,d(E). For any f ∈ Mn,d(E) we then have f = f [g, h ]• = •[ f, g]h, hence (4.27) holds for all f ∈ Mn,d(E). But this extends to L2(G × Zn × Zd) by continuity, which implies that g and h generate dual (n, d)-matrix Gabor frames. (cid:3) We wish to establish a duality principle for (n, d)-matrix Gabor frames. For this we also need to treat (n, d)-matrix Gabor Riesz sequences and relate them to Definition 3.34. Definition 4.24. Let g ∈ L2(G × Zn × Zd). We say g generates an (n, d)-matrix Gabor Riesz sequence for L2(G) with respect to Λ, or that G(g; Λ) is an (n, d)-matrix Gabor Riesz sequence for L2(G), if (4.30) is an isomorphism. Equivalently, there exists h ∈ L2(G × Zn × Zd) such that for all a ∈ L2(Λ × Zn × Zn) we have g : L2(Λ × Zn × Zn) → L2(Λ × Zn × Zn) CgC ∗ (4.31) ar,s(µ) = Xi∈Zd Xj∈Zn hZΛ ar,j(λ)π(λ)gj,idλ, π(µ)hs,ii for all r, s ∈ Zn and all µ ∈ Λ. (n, d)-matrix Gabor Riesz sequence of g. If (4.31) is satisfied we will say h generates a dual GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 29 Remark 4.25. Note that the equivalence of the definitions of (n, d)-matrix Gabor Riesz sequences in Definition 4.24 follows by [7, Theorem 3.6.6] and Proposition 4.16. Remark 4.26. (4.31) can be seen to be equivalent to CgC ∗ h = ChC ∗ g = IdL2(Λ×Zn×Zn). Before treating localization of module matrix Riesz sequences and how they relate to matrix Gabor Riesz sequences, we do a necessary but justified simplification. Recall that existence of finite module matrix Riesz sequences for Mn,d(E) with respect to Mn(A) requires A to be unital by Proposition 3.20. In the following we therefore let Λ be discrete, but not necessarily cocompact. Hence in the following, A is unital with a faithful trace, but B might not have that property. By [17, p. 251] we know that G(g; Λ) is a Bessel system with Bessel bound D if and only if G(g; Λ◦) is a Bessel system with Bessel bound D. Applying Proposition 4.16 we immediately get the following from Lemma 4.17 and Lemma 4.18. Proposition 4.27. Let Λ ⊂ G × bG be discrete. For all g, h ∈ Mn,d(E) we have (ChC ∗ g). Equivalently, the diagram ρMn(A) = ρMn(A) ◦ (ΦhΦ∗ g ) ◦ Mn(A) ρMn(A) L2(Λ × Zn × Zn) ΦhΦ∗ g ChC ∗ g Mn(A) ρMn(A) L2(Λ × Zn × Zn) commutes. As ρMn(A) : Mn(A) → ρMn(A)(Mn(A)) is a linear bijection respecting the actions of g is an isomorphism if and g )ρMn(A)(Mn(A)) is an isomorphism. In analogy with Lemma 4.20 we have the A, we see by Proposition 4.27 that for g ∈ Mn,d(E), ΦgΦ∗ only if (CgC ∗ following result. Lemma 4.28. Let Λ ⊂ G × bG be discrete. For g ∈ Mn,d(E) we have that g )ρMn(A)(Mn(A)) : ρMn(A)(Mn(A)) → ρMn(A)(Mn(A)) (CgC ∗ (4.32) is invertible if and only if (4.33) is invertible. CgC ∗ g : L2(Λ × Zn × Zn) → L2(Λ × Zn × Zn) Proof. Suppose first that (CgC ∗ g )ρMn(A)(Mn(A)) : ρMn(A)(Mn(A)) → ρMn(A)(Mn(A)) is in- vertible. Since any g ∈ Mn,d(E) is a Bessel vector by Proposition 4.16, we may extend g : L2(Λ × Zn × Zn) → L2(Λ × Zn × Zn) is the operator by continuity to obtain that CgC ∗ invertible as well. Conversely, suppose CgC ∗ is the continuous extension of ΦgΦ∗ that ΦgΦ∗ ρMn(A)(Mn(A)) is invertible. g is invertible as well, which implies (CgC ∗ g : L2(Λ×Zn ×Zn) → L2(Λ×Zn ×Zn) is invertible. Since CgC ∗ g g, it then follows by Proposition 2.8 and Proposition 2.1 g )ρMn(A)(Mn(A)) : ρMn(A)(Mn(A)) → (cid:3) Remark 4.29. From now on we will identify Mn(•h E, Ei) (and potentially a larger domain) and its localization. The same goes for Md(B). Now the following is an immediate consequence. 30 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS Proposition 4.30. Let Λ ⊂ G × bG be discrete. For g ∈ Mn,d(E) we have that ΦgΦ∗ g is invertible if and only if CgC ∗ g is invertible. In other words, g generates a module (n, d)- matrix Riesz sequence for E with respect to A if and only if G(g; Λ) is an (n, d)-matrix Gabor Riesz sequence for L2(G). By the proof of Lemma 4.28 we then have the following statement. Corollary 4.31. Let Λ ⊂ G × bG be discrete. Suppose g, h ∈ Mn,d(E). Then g and h generate dual (n, d)-matrix Gabor Riesz sequences for L2(G) with respect to Λ if and only if •[ g, h] extends to the identity operator on L2(Λ × Zn × Zn). Proof. Suppose first that g and h generate dual (n, d)-matrix Gabor Riesz sequences for L2(G) with respect to Λ. Then for all a ∈ Mn(A) we have (ar,s) = {Xi∈Zd Xj∈Zn hZΛ ar,j(λ)π(λ)gj,idλ, π(µ)hs,ii}µ∈Λ,r,s∈Zn, which is equivalent to a = a •[ g, h] for all a ∈ Mn(A). But the first expression extends by continuity to L2(Λ × Zn × Zn), so •[ g, h] extends to the identity on L2(Λ × Zn × Zn). Conversely, suppose •[ g, h] extends to the identity on L2(Λ × Zn × Zn). Once again, for all a ∈ Mn(A) we then have (ar,s) = {Xi∈Zd Xj∈Zn hZΛ ar,j(λ)π(λ)gj,idλ, π(µ)hs,ii}µ∈Λ,r,s∈Zn, which again extends to L2(G × Zn × Zn). Hence g and h are dual (n, d)-matrix Gabor Riesz sequences for L2(G) with respect to Λ. (cid:3) Note how the above results guarantee that when Λ ⊂ G×bG is closed and cocompact and g ∈ Mn,d(E) is such that G(g; Λ) is an (n, d)-matrix Gabor frame for L2(G), the canonical dual frame S−1 g g ∈ Mn,d(E). Indeed, g g = Θ−1 S−1 g g = [g, g ]• −1 ∈ Mn,d(E). Likewise, for Riesz sequences there is the notion of canonical biorthogonal atom, see for example [7, p. 160]. Restricting to Λ discrete, it is given by (SB is the frame operator with respect to the right hand side, that is, with respect to Λ◦. We see that for all f ∈ Mn,d(E) g )−1g, where SB g SB g f = (ΦB g )∗ΦB g f = (ΦB g )∗([g, f ]•) = g[g, f ]• = •[ g, g]f. Thus it follows that (SB g )−1g = (ΘB g )−1g = •[ g, g]−1g ∈ Mn,d(E). Hence for both matrix Gabor frames and matrix Gabor Riesz sequences with generating atom in Mn,d(E), the canonically associated dual atoms are also in Mn,d(E). We have the following result which shows that in the cases we are interested in, if the generating atom is regular, the canonical dual atom has the same regularity. Proposition 4.32. Let g ∈ Mn,d(E). i) If G(g; Λ) is an (n, d)-matrix Gabor frame for L2(G) and Λ is closed and cocompact in G × bG, then the canonical dual atom is in Mn,d(E). ii) If G(g; Λ) is an (n, d)-matrix Gabor Riesz sequence for L2(G) and Λ is discrete, then the canonical biorthogonal atom is also in Mn,d(E). GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 31 Proof. For the proof of i), note that the assumption that Λ is cocompact implies that Λ◦ is discrete, so by Lemma 4.4 we get that Md(B) is unital. Also Md(B) is a C ∗-subalgebra of B(HMn,d(E)) by Proposition 2.8. That G(g; Λ) is an (n, d)-matrix Gabor frame for L2(G) then means that (3.30) is satisfied for our current setting. We deduce, as in the proof of Proposition 3.30, that [g, g ]• is invertible in Md(B). Since [g, g ]• ∈ Md(B) and Md(B) is spectral invariant in Md(B) by Proposition 4.5 and Lemma 3.22, it follows that the canonical dual atom is g[g, g ]• −1 ∈ Mn,d(E). For the proof of ii), note that the assumption that Λ is discrete implies Mn(A) is unital. Also, Mn(A) is a C ∗-subalgebra of B(HMn(A)) by Proposition 2.8. That G(g; Λ) determines an (n, d)-matrix Gabor Riesz sequence for L2(G) then means that (3.31) is satisfied for our current setting. The middle term of (3.31) can be written (a •[ g, g], a)A, so •[ g, g] extends to a positive, invertible operator on HMn(A). We deduce as in the proof of Proposition 3.31 that •[ g, g] is invertible in Mn(A). Since g ∈ Mn,d(E) we have •[ g, g] ∈ Mn(A), and by Proposition 4.5 and Lemma 3.22 Mn(A) is spectral invariant in Mn(A). It follows that the canonical dual atom h := •[ g, g]−1g is in Mn,d(E). (cid:3) Remark 4.33. In the special case n = d = 1 Proposition 4.32 gives a proof of the fact that the canonical dual atom of a Gabor frame vector in Feichtinger's algebra S0(G) is also in Feichtinger's algebra whenever Λ is cocompact. When applying the module setup of Section 3 to Gabor analysis, we take as a pre- equivalence bimodule E = S0(G × Zn × Zd), which is a proper subspace of L2(G × Zn × Zd) unless G is a finite group. Even the Hilbert C ∗-module completion E is properly contained in L2(G × Zn × Zd) for most choices of Λ, see [2, Example 3.8]. As such, we cannot hope to treat general atoms in L2(G × Zn × Zd) by applying just this method. But indeed the module reformulation is made exactly to guarantee some regularity of the atoms generating frames. From Definition 4.11 we see that (n, d)-matrix Gabor frames generalize n-multiwindow d-super Gabor frames considered in [18]. However, we now make clear how they fit into the module framework. As mentioned earlier, we obtain n-multiwindow d-super Gabor frames if we only require reconstruction of f ∈ L2(G × Zd) and we identify L2(G × Zd) ⊂ L2(G × Zn × Zd) by embedding it along a single element in Zn. The module reformulation of this is that g, h ∈ Mn,d(E) are dual n-multiwindow d-super Gabor frames if for all f ∈ Mn,d(E) supported only one row we have (4.34) f = •[ f, g]h = f [g, h ]• . Likewise, it is clear that the (n, d)-matrix Gabor Riesz sequences of Definition 4.24 gen- eralize the n-multiwindow d-super Gabor Riesz sequences also considered in [18]. Indeed, we obtain n-multiwindow d-super Gabor Riesz sequences if we only require reconstruction of a ∈ L2(Λ × Zn) and we identify L2(Λ × Zn) ⊂ L2(Λ × Zn × Zn) by embedding it along a single element in the middle copy of Zn. The module reformulation of this is that g, h ∈ Mn,d(E) are dual n-multiwindow d-super Gabor Riesz sequences if for all a ∈ Mn(A) supported only one row we have (4.35) a = •[ ag, h] = a •[ g, h]. We proceed to prove that all n-multiwindow d-super Gabor frames for L2(G) with respect to Λ are (n, d)-matrix Gabor frames for L2(G) with respect to Λ, as well as the analogous statement for Riesz sequences. The converse statement is true as well. 32 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS Proposition 4.34. Let g be in Mn,d(E). i) If G(g; Λ) is an n-multiwindow d-super Gabor frame for L2(G) with a dual window h ∈ Mn,d(E), then G(g; Λ) is an (n, d)-matrix Gabor frame for L2(G) with dual window h. ii) If G(g; Λ) is an n-multiwindow d-super Gabor Riesz sequence for L2(G) with a dual Gabor Riesz sequence G(h; Λ) with h ∈ Mn,d(E), then G(g; Λ) is an (n, d)-matrix Gabor Riesz sequence for L2(G) with dual Gabor Riesz sequence G(h; Λ). Proof. If G(g; Λ) is an n-multiwindow d-super Gabor frame for L2(G) with respect to Λ with a dual window h ∈ Mn,d(E), we can, as noted above, reconstruct any f ∈ Mn,d(E) supported on a single row. In other words, f = f [g, h ]• for all f ∈ Mn,d(E) supported on a single row. Given arbitrary f ′ ∈ Mn,d(E) we may then just write f ′ as a sum of n matrices f ′ i, i = 0, . . . , n − 1, with only one nonzero row, namely the kth row of f ′ i is given by (fi,0, . . . , fi,d−1)δik for k ∈ Zn, and we can reconstruct each of these rows. Hence we can reconstruct arbitrary elements of Mn,d(E). This passes to the localization L2(G × Zn × Zd), and thus finishes the proof of (i). The proof of (ii) is completely analogous, writing elements a ∈ Mn(A) as a sum of matrices with only one nonzero row and then using that we can reconstruct such matrices. This will also pass to the localization L2(Λ × Zn × Zn). (cid:3) Given a closed and cocompact subgroup Λ, we may ask if there are restrictions on n, d ∈ N for there to possibly exist (n, d)-matrix Gabor frames for L2(G) with respect to Λ. Conversely, if we fix n and d, we may ask if there are restrictions on the size of the subgroup Λ for there to possibly exist (n, d)-matrix Gabor frames for L2(G) with respect to Λ. When Λ is a lattice, we have the following proposition. Proposition 4.35. Let Λ ⊂ G × bG be a lattice. If there is g ∈ Mn,d(E) such that G(g; Λ) is an (n, d)-matrix Gabor frame for L2(G), then n d s(Λ) ≤ . Proof. Since Λ is discrete and cocompact, both A and B are unital. We also know by Proposition 4.32 that the canonical dual of g is in Mn,d(E). Hence we are in the setting of Theorem 3.28. Since module (n, d)-matrix frames localize to (n, d)-matrix Gabor frames for the localization, and we have trA(1A) = 1, and trB(1B) = s(Λ) (since the identity on B is s(Λ)δ0, where δ0 is the indicator function in the group identity, see for example [28]), the result is immediate by Theorem 3.28. (cid:3) Likewise, given a lattice Λ, we may ask if there is a relationship between the size of Λ and the integers n and d such that there can possibly exist (n, d)-matrix Gabor Riesz sequences for L2(G) with respect to Λ. This is the content of the following proposition. Proposition 4.36. Let Λ ⊂ G × bG be a lattice. If g ∈ Mn,d(E) is such that G(g; Λ) is an (n, d)-matrix Gabor Riesz sequence for L2(G), then s(Λ) ≥ n d . GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 33 Proof. As before we know by the conditions on Λ that both A and B are unital, and by Proposition 4.32 the canonical dual of g is in Mn,d(E). Thus we are in the setting of Theorem 3.29. Since module (n, d)-matrix Riesz sequences localize to (n, d)-matrix Gabor Riesz sequences for the localization, and trA(1A) = 1 and trB(1B) = s(Λ) (once again since the identity on B is s(Λ)δ0), the result is immediate by Theorem 3.29. (cid:3) Remark 4.37. The two preceding propositions contain statements known as density theo- rems in Gabor analysis. This is due to the fact that they give conditions on the density of a lattice for there to possibly exist Gabor frames. Now let Λ ⊂ G × bG be cocompact again. In this framework two of the cornerstones of Gabor analysis, namely the Wexler-Raz biorthogonality relations and the duality principle for Gabor frames, are then quite easy to prove for (n, d)-matrix Gabor frames for L2(G) with respect to Λ with atoms in Mn,d(E). Proposition 4.38 (Wexler-Raz Biorthogonality Relations). Let Λ ⊂ G × bG be a closed and cocompact subgroup, and let g, h ∈ Mn,d(E). Then the following are equivalent: i) G(g; Λ) and G(h; Λ) are dual (n, d)-matrix Gabor frames for L2(G). ii) hg, π(λ◦)hiℓ2(Λ◦×Zn×Zd) = s(Λ) · δ0,λ◦. Proof. As Λ is cocompact we know Λ◦ is discrete, so Md(B) is unital. Knowing this, we can see that both the above statements are equivalent to the statement [g, h ]• = [h, g ]• = 1Md(B). (cid:3) Theorem 4.39 (Duality principle). Let Λ ⊂ G × bG be a closed and cocompact subgroup, and let g ∈ Mn,d(E). Then the following are equivalent. i) G(g; Λ) is an (n, d)-matrix Gabor frame for L2(G). ii) G(g; Λ◦) is a (d, n)-matrix Gabor Riesz sequence for L2(G). Proof. Statement i) can be seen to be equivalent to [g, g ]• being invertible by Proposition 4.22. But statement ii) is also equivalent to [g, g ]• being invertible by Proposition 4.30 since we consider G(g; Λ◦), that is, we work over Md(B). This finishes the proof. (cid:3) For completeness we also include the following result related to the duality principle. This is a strengthening of the corresponding result in [18]. such that [g, h ]• extends to the identity operator on L2(G × Zn × Zd). Then •[ g, h] is an Proposition 4.40. Let Λ ⊂ G × bG be closed and cocompact, and let g, h ∈ Mn,d(E) be idempotent operator from L2(G × Zn × Zd) onto span{Li∈ZnLj∈Zd π(λ◦)gi,j}. Proof. Since [g, h ]• extends to the identity operator, we have [g, h ]• = [h, g ]• = 1Md(B). That •[ g, h] is an idempotent then follows by Proposition 3.19. By Proposition 3.12 we get that •[ g, h] is an idempotent from Mn,d(E) onto gMd(B). But this passes to the localization, and the localization of gMd(B) is span{Mi∈Zn Mj∈Zd π(λ◦)gi,j} ⊂ L2(G × Zn × Zd). (cid:3) Lastly in this section, we prove that whenever Λ is cocompact, there is a close relationship between the module frame bounds and the Gabor frame bounds in the localization. 34 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS Proposition 4.41. Let Λ ⊂ G×bG be a closed and cocompact subgroup. Then g ∈ Mn,d(E) generates a module (n, d)-matrix frame for E as an A-module with lower frame bound C and upper frame bound D if and only if G(g; Λ) is an (n, d)-matrix Gabor frame for L2(G) with lower frame bound C and upper frame bound D. Proof. By Lemma 2.13 it suffices to prove that the optimal frame bounds are equal for both the module frame and the Gabor frame. We know that the localization of a module (n, d)- matrix frame for E as an A-module becomes an (n, d)-matrix Gabor frame for L2(G) with respect to Λ. Since Λ is cocompact, we also know that if g ∈ Mn,d(E) is such that G(g; Λ) is an (n, d)-matrix Gabor frame for L2(G), then the canonical dual S−1 g g ∈ Mn,d(E) also. By Proposition 4.22 we have ρ(Θg) = Sg. From standard Hilbert space frame theory we know that the optimal upper frame bound for Sg is kSgk, and the optimal lower frame g k−1, see for example Section 5.1 of [13]. We know by Proposition 2.8 bound for Sg is kS−1 that kΘgk = kρ(Θg)k = kSgk and kΘ−1 g k. The result then follows by Lemma 2.13. (cid:3) g k = kρ(Θ−1 g )k = kS−1 Remark 4.42. A straightforward calculation will show that kΘgk = k •[ g, g]k. Indeed, as Θgx = x[g, g ]• and [g, g ]• is positive, we see that kΘgk = sup kf k=1 {k •[ f [g, g ]•, f ]k}, and it follows immediately that kΘgk ≤ k[g, g ]• k = k •[ g, g]k. Inserting f = k •[ g, g]k−1/2g we obtain the equality. Note that this is the same upper bound we obtained in Proposition 3.30. Corollary 4.43. Let g ∈ Mn,d(E) and let Λ ⊂ G × bG be a closed and cocompact subgroup. Then g has the same Bessel bound both as an (n, d)-matrix Gabor atom for L2(G) with respect to Λ and as a (d, n)-matrix Gabor atom for L2(G) with respect to Λ◦. Proof. It suffices to prove that the optimal Bessel bounds agree. Let DΛ be the opti- mal Bessel bound for G(g; Λ), and let DΛ◦ be the optimal Bessel bound for G(g; Λ◦). By Proposition 4.41 and Remark 4.42 it follows that DΛ = kSgk = k •[ g, g]k. But the anal- ogous argument works with Λ◦ instead of Λ, since the important part for the setup with localization as done in this paper is that Λ or Λ◦ is cocompact. Indeed, this is really a consequence of Proposition 2.8. Hence we may just as well apply the frame operator SB g , which is given by the continuous extension of multiplication by •[ g, g] on the left. That is, for all f ∈ Mn,d(E) we have SB g f = •[ g, g]f. An analogous argument to the one in Remark 4.42 will show that DΛ◦ = k •[ g, g]k, and hence DΛ = DΛ◦ . (cid:3) Acknowledgement The authors wish to thank Ulrik Enstad for his valuable input and for finding an error in an earlier draft of the manuscript. We would also like to thank the Erwin Schrödinger Institute for their hospitality and support since part of this research was conducted while the authors attended the program "Bivariant K-theory in Geometry and Physics". GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS 35 References [1] Luis Daniel Abreu. Sampling and interpolation in Bargmann-Fock spaces of polyanalytic functions. Appl. Comput. Harmon. Anal., 29(3):287 -- 302, 2010. [2] Are Austad and Ulrik Enstad. The Heisenberg module as a function space. arXiv preprint arXiv:1904.10826, 2019. [3] Davide Barbieri, Eugenio Hernández, and Javier Parcet. Riesz and frame systems generated by unitary actions of discrete groups. Applied and Computational Harmonic Analysis, 39(3):369 -- 399, 2015. [4] Davide Barbieri, Eugenio Hernández, and Victoria Paternostro. Invariant spaces under unitary repre- sentations of discrete groups. arXiv preprint arXiv:1811.02993, 2018. [5] Erik Bédos and Tron Omland. On reduced twisted group C ∗-algebras that are simple and/or have a unique trace. J. Noncommut. Geom., 12(3):947 -- 996, 2018. [6] Peter G. Casazza, Gitta Kutyniok, and Mark C. Lammers. Duality principles, localization of frames, and Gabor theory. In Wavelets XI (San Diego, CA, 2005), volume 5914, pages 389 -- 398, 2005. [7] Ole Christensen. An introduction to frames and Riesz bases. Applied and Numerical Harmonic Anal- ysis. Birkhäuser/Springer, [Cham], second edition, 2016. [8] Ole Christensen, Xiang Xiao, and Yu Zhu. Characterizing R-duality in Banach spaces. Acta Math. Sin. (Engl. Ser.), 29(1):75 -- 84, 2013. [9] Ingrid Daubechies, Henry J. Landau, and Zeph Landau. Gabor time-frequency lattices and the Wexler- Raz identity. J. Fourier Anal. Appl., 1(4):437 -- 478, 1995. [10] Zhitao Fan, Andreas Heinecke, and Zuowei Shen. Duality for frames. Journal of Fourier Analysis and Applications, 22(1):71 -- 136, 2016. [11] Massimo Fornasier and Karlheinz Gröchenig. Intrinsic localization of frames. Constr. Approx., 22(3):395 -- 415, 2005. [12] Michael Frank and David R. Larson. Frames in Hilbert C ∗-modules and C ∗-algebras. J. Operator Theory, 48(2):273 -- 314, 2002. [13] Karlheinz Gröchenig. Foundations of Time-Frequency Analysis. Appl. Numer. Harmon. Anal. Birkhäuser, 2001. [14] Karlheinz Gröchenig and Michael Leinert. Wiener's lemma for twisted convolution and Gabor frames. J. Amer. Math. Soc., 17:1 -- 18, 2004. [15] Karlheinz Gröchenig and Yurii Lyubarskii. Gabor (super)frames with Hermite functions. Math. Ann., 345(2):267 -- 286, 2009. [16] Mads S. Jakobsen. On a (no longer) New Segal Algebra: A Review of the Feichtinger Algebra. J. Fourier Anal. Appl., 24(6):1579 -- 1660, 2018. [17] Mads S. Jakobsen and Jakob Lemvig. Density and duality theorems for regular Gabor frames. J. Funct. Anal., 270(1):229 -- 263, 2016. [18] Mads S. Jakobsen and Franz Luef. Duality of Gabor frames and Heisenberg modules. arXiv preprint arXiv:1806.05616v3, 2018. [19] A. J. E. M. Janssen. Duality and biorthogonality for Weyl-Heisenberg frames. J. Fourier Anal. Appl., 1(4):403 -- 436, 1995. [20] E. C. Lance. Hilbert C-modules: a toolkit for operator algebraists, volume 210. Cambridge University Press, 1995. [21] Franz Luef. Projective modules over non-commutative tori are multi-window Gabor frames for mod- ulation spaces. J. Funct. Anal., 257(6):1921 -- 1946, 2009. [22] Franz Luef. Projections in noncommutative tori and Gabor frames. Proceedings of the American Mathematical Society, 139(2):571 -- 582, 2011. [23] Paul Milnes. Identities of group algebras. Proceedings of the American Mathematical Society, 29(2):421 -- 422, 1971. [24] Gerald J. Murphy. C-algebras and operator theory. Academic press, 2014. [25] Iain Raeburn and Dana P. Williams. Morita Equivalence and Continuous-trace C ∗-algebras. American Mathematical Society (AMS), 1998. [26] Marc Rieffel. C ∗-algebras associated with irrational rotations. Pacific Journal of Mathematics, 93(2):415 -- 429, 1981. [27] Marc A. Rieffel. Morita equivalence for C ∗-algebras and W ∗-algebras. J. Pure Appl. Algebra, 5:51 -- 96, 1974. 36 GABOR DUALITY THEORY FOR MORITA EQUIVALENT C ∗-ALGEBRAS [28] Marc A. Rieffel. Projective modules over higher-dimensional noncommutative tori. Canad. J. Math., 40(2):257 -- 338, 1988. [29] Amos Ron and Zuowei Shen. Affine systems in L2(Rd): The analysis of the analysis operator. J. Funct. Anal., 148(2):408 -- 447, 1997. [30] Larry B. Schweitzer. A short proof that Mn(A) is local if A is local and Fréchet. Int. J. Math., 3:581 -- 589, 1992. [31] Diana Stoeva and Ole Christensen. On R-Duals and the Duality Principle in Gabor Analysis. J. Fourier Anal. Appl., 21(2,):383 -- 400, 2015. [32] Diana Stoeva and Ole Christensen. On Various R-duals and the Duality Principle. Integral Equations and Operator Theory, 84(4,):577 -- 590, 2016.
1107.3415	5	1107	2012-02-18T10:42:05	Square functions for Ritt operators on noncommutative $L^p$-spaces	[ "math.OA", "math.FA" ]	For any Ritt operator $T$ acting on a noncommutative $L^p$-space, we define the notion of \textit{completely} bounded functional calculus $H^\infty(B_\gamma)$ where $B_\gamma$ is a Stolz domain. Moreover, we introduce the `column square functions' $\norm{x}_{T,c,\alpha}=\Bnorm{\Big(\sum_{k=1}^{+\infty}k^{2\alpha-1}\|T^{k-1}(I-T)^{\alpha}(x)\|^2\Big)^{1/2}}_{L^p(M)}$ and the `row square functions' $\norm{x}_{T,r,\alpha}=\Bnorm{\Big(\sum_{k=1}^{+\infty}k^{2\alpha-1} \|\Big(T^{k-1}(I-T)^{\alpha}(x)\Big)^*\|^2\Big)^{1/2}}_{L^p(M)}$ for any $\alpha>0$ and any $x\in L^p(M)$. Then, we provide an example of Ritt operator which admits a completely bounded $H^\infty(B_\gamma)$ functional calculus for some $\gamma \in \big]0,\frac{\pi}{2}\big[$ such that the square functions $\norm{\cdot}_{T,c,\alpha}$ and $\norm{\cdot}_{T,r,\alpha}$ are not equivalent. Moreover, assuming $1<p<2$ and $\alpha>0$, we prove that if $\Ran (I-T)$ is dense and $T$ admits a completely bounded $H^\infty(B_\gamma)$ functional calculus for some $\gamma \in \big]0,\frac{\pi}{2}\big[$ then there exists a positive constant $C$ such that for any $x \in L^p(M)$, there exists $x_1, x_2 \in L^p(M)$ satisfying $x=x_1+x_2$ and $\norm{x_1}_{T,c,\alpha}+\norm{x_2}_{T,r,\alpha}\leq C \norm{x}_{L^p(M)}$. Finally, we observe that this result applies to a suitable class of selfadjoint Markov maps on noncommutative $L^p$-spaces.	math.OA	math	Square functions for Ritt operators on noncommutative Lp-spaces Cédric Arhancet Abstract For any Ritt operator T acting on a noncommutative Lp-space, we define the notion of com- pletely bounded functional calculus H ∞(Bγ ) where Bγ is a Stolz domain. Moreover, we intro- 1 (cid:16)P+∞ k=1 k2α−1 (cid:12)(cid:12)(cid:12) k=1 k2α−1 (cid:12)(cid:12)T k−1(I − T )α(x)(cid:12)(cid:12) 2 (cid:17) (cid:16)T k−1(I − T )α(x)(cid:17)∗(cid:12)(cid:12)(cid:12) duce the 'column square functions' kxkp,T ,c,α = (cid:13)(cid:13)(cid:13) the 'row square functions' kxkp,T ,r,α = (cid:13)(cid:13)(cid:13) (cid:16)P+∞ α > 0 and any x ∈ Lp(M ). Then, we provide an example of Ritt operator which admits a com- pletely bounded H ∞(Bγ ) functional calculus for some γ ∈ (cid:3)0, π 2 (cid:2) such that the square functions k · kp,T ,c,α and k · kp,T ,r,α are not equivalent. Moreover, assuming 1 < p < 2 and α > 0, we prove that if Ran(I − T ) is dense and T admits a completely bounded H ∞(Bγ ) functional calculus for 2 (cid:2) then there exists a positive constant C such that for any x ∈ Lp(M ), there exists some γ ∈ (cid:3)0, π x1, x2 ∈ Lp(M ) satisfying x = x1 + x2 and kx1kp,T ,c,α + kx2kp,T ,r,α 6 CkxkLp(M ). Finally, we observe that this result applies to a suitable class of selfadjoint Markov maps on noncommutative Lp-spaces. 2 (cid:17) 2 (cid:13)(cid:13)(cid:13)Lp(M ) 2 (cid:13)(cid:13)(cid:13)Lp(M ) for any and 1 1 Introduction Let M be a semifinite von Neumann algebra equipped with a normal semifinite faithful trace. For any 1 6 p < ∞, we let Lp(M ) denote the associated (noncommutative) Lp-space. Let T be a bounded operator on Lp(M ). Consider the following 'square function' (1.1) : uk + vk = k uk2(cid:19) if 1 < p 6 2 and (1.2) kxkp,T,1 = inf((cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) +∞ Xk=1 kxkp,T,1 = max((cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) +∞ Xk=1 v∗k2(cid:19) 1 +(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) +∞ Xk=1 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp k(cid:12)(cid:12)T k(x) − T k−1(x)(cid:12)(cid:12) 1 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp 2(cid:19) 1 ,(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) +∞ Xk=1 1 2(cid:0)T k(x) − T k−1(x)(cid:1) for any k) 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp) k(cid:12)(cid:12)(cid:12)(cid:16)T k(x) − T k−1(x)(cid:17)∗(cid:12)(cid:12)(cid:12) 2(cid:19) 1 if 2 6 p < ∞, defined for any x ∈ Lp(M ). Such quantities were introduced in [LM2] and studied in this paper and in [ALM]. Similar square functions for continuous semigroups played a key role in the recent development of H∞-calculus and its applications. See in particular the paper [JMX], the survey [LM1] and the references therein. This work is partially supported by ANR 06-BLAN-0015. 2010 Mathematics subject classification: Primary 46L52; Secondary, 46L51, 47A60. Key words and phrases: noncommutative Lp-space, Ritt operator, square function, Stolz domain, functional calculus. 1 For any γ ∈(cid:3)0, π 2(cid:2), let Bγ be the interior of the convex hull of 1 and the disc D(0, sin γ). Suppose 1 < p < ∞. Let T be a Ritt operator with Ran(I − T ) dense in Lp(M ) which admits a bounded H∞(Bγ) functional calculus for some γ ∈ (cid:3)0, π 2(cid:2) and a positive constant K such that (cid:13)(cid:13)ϕ(T )(cid:13)(cid:13)Lp(M)→Lp(M) 6 KkϕkH∞(Bγ ) for any complex polynomial ϕ. A result 2(cid:2), i.e. there exists an angle γ ∈ (cid:3)0, π of [LM2] essentially says that (1.3) kxkLp(M) ≈ kxkp,T,1, x ∈ Lp(M ) (see also [ALM, Remark 6.4]). Now, consider the following 'column and row square functions' (1.4) 2(cid:19) 1 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp (cid:18) +∞ Xk=1 k(cid:12)(cid:12)T k(x) − T k−1(x)(cid:12)(cid:12) and kxkp,T,r,1 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) kxkp,T,c,1 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) +∞ Xk=1 k(cid:12)(cid:12)(cid:12)(cid:16)T k(x) − T k−1(x)(cid:17)∗(cid:12)(cid:12)(cid:12) 2(cid:19) defined for any x ∈ Lp(M ). Assume 1 < p < 2. In this context, if x ∈ Lp(M ), it is natural to search sufficient conditions to find a decomposition x = x1 +x2 such that kx1kp,T,c,1 and kx2kp,T,r,1 are finite. The first main result of this paper is the next theorem. It strengthens the above equivalence (1.3) in the case where T actually admits a completely bounded H∞(Bγ) functional calculus, i.e. there exists a positive constant K such that (cid:13)(cid:13)ϕ(T )(cid:13)(cid:13)cb,Lp(M)→Lp(M) 6 KkϕkH∞(Bγ ) for any complex polynomial ϕ. Theorem 1.1 Suppose 1 < p < 2. Let T be a Ritt operator on Lp(M ) with Ran(I − T ) dense in Lp(M ). Assume that T admits a completely bounded H∞(Bγ) functional calculus for some γ ∈(cid:3)0, π 2(cid:2). Then we have 1 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp kxkLp(M) ≈ infnkx1kp,T,c,1 + kx2kp,T,r,1 : x = x1 + x2o, x ∈ Lp(M ). In this context, it is natural to compare the both quantities of (1.4). The second principal result of this paper is the following theorem. It says that in general, 'column and row square functions' are not equivalent. 2(cid:2) such that Theorem 1.2 Suppose 1 < p 6= 2 < ∞. Then there exists a Ritt operator T on the Schatten space Sp, with Ran(I − T ) dense in Sp, which admits a completely bounded H∞(Bγ) functional calculus for some γ ∈(cid:3)0, π (1.5) sup(kxkp,T,c,1 : x ∈ Sp) = ∞ if 2 < p < ∞ and sup(kxkp,T,r,1 kxkp,T,c,1 kxkp,T,r,1 Moreover, the same result holds with k · kp,T,c,1 and k · kp,T,r,1 switched. : x ∈ Sp) = ∞ if 1 < p < 2. The paper is organized as follows. Section 2 gives a brief presentation of noncommutative Lp-spaces and Ritt operators and we introduce the notions of Col-Ritt and Row-Ritt operators and completely bounded H∞(Bγ) functional calculus which are relevant to our paper. The next section 3 mostly contains preliminary results concerning Col-Ritt and Row-Ritt operators. Section 4 is devoted to prove Theorems 1.2. In section 5, wepresent a proof of Theorem 1.1. We end this section by giving some natural examples to which this result can be applied. In the above presentation and later on in the paper we will use . to indicate an inequality up to a constant which does not depend to the particular element to which it applies. Then A(x) ≈ B(x) will mean that we both have A(x) . B(x) and B(x) . A(x). 2 2 Background and preliminaries We start with a few preliminaries on noncommutative Lp-spaces. Let M be a von Neumann algebra equipped with a normal semifinite faithful trace τ. Let M+ be the set of all positive elements of M and let S+ be the set of all x in M+ such that τ (x) < ∞. Then let S be the linear span of S+. For any 1 6 p < ∞, define 1 1 kxkLp(M) = (cid:0)τ (xp)(cid:1) p , x ∈ S, p 1 p 2(cid:1) 1 q = 1 are regarded as matrices A = [aij ]i,j>1 in the usual way. We refer to [Pis3] for more about these spaces and complements. If we equip the space B(ℓ2) with the operator norm and the canonical trace r (noncommutative Hölder's inequality). Using trace duality, we then have Lp(M )∗ = Lp∗ 2 is the modulus of x. Then (cid:0)S,k · kLp(M)(cid:1) is a normed space. The corresponding where x = (x∗x) completion is the noncommutative Lp-space associated with (M, τ ) and is denoted by Lp(M ). By convention, we set L∞(M ) = M , equipped with the operator norm. The elements of Lp(M ) can also be described as measurable operators with respect to (M, τ ). Further multiplication of measurable op- erators leads to contractive bilinear maps Lp(M )× Lq(M ) → Lr(M ) for any 1 6 p, q, r 6 ∞ such that p + 1 (M ) isometrically for any 1 6 p < ∞. Moreover, complex interpolation yields Lp(M ) =(cid:2)L∞(M ), L1(M )(cid:3) 1 for any 1 6 p 6 ∞. We refer the reader to [PX] for details and complements. Let 1 6 p < ∞. Tr , the space Lp(cid:0)B(ℓ2)(cid:1) identifies to the Schatten-von Neumann class Sp. This is the space of those compact operators x from ℓ2 into ℓ2 such that kxkSp =(cid:0) Tr (x∗x) p < ∞. Elements of B(ℓ2) or Sp If the von Neumann algebra B(ℓ2)⊗M is equipped with the semifinite normal faithful trace Tr ⊗τ, the space Lp(cid:0)B(ℓ2)⊗M(cid:1) canonically identifies to a space Sp(cid:0)Lp(M )(cid:1) of matrices with entries in Lp(M ). Moreover, under this identification, the algebraic tensor product Sp ⊗ Lp(M ) is dense in Sp(cid:0)Lp(M )(cid:1). If 1 6 p < ∞, we say that a linear map on Lp(M ) is completely bounded if ISp ⊗ T extends to a bounded operator on Sp(cid:0)Lp(M )(cid:1). In this case, the completely bounded norm kTkcb,Lp(M)−→Lp(M) of T is defined by kTkcb,Lp(M)−→Lp(M) = (cid:13)(cid:13)ISp ⊗ T(cid:13)(cid:13)Sp(Lp(M))−→Sp(Lp(M)). We use the convention to define kTkcb,Lp(M)−→Lp(M) by +∞ if T is not completely bounded. information on these spaces. For any Pn xk ⊗ ak(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M,ℓ2 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) n Xk=1 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M) ek1 ⊗ xk(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Sp(Lp(M)) Xk=1 Xk=1 c) is the completion of Lp(M ) ⊗ ℓ2 for this norm. r) similarly. For any finite family (xk)16k6n in Lp(M ), we have sequences (xk)k>1 in Lp(M ) such that P+∞k=1 xk ⊗ ek is convergent for the above norm. We define Lp(M, ℓ2 n Xk=1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) xk ⊗ ek(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M,ℓ2 We shall use various ℓ2-valued noncommutative Lp spaces. We refer to [JMX, Chapter 2] for more =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) n Xi,j=1 xk2(cid:19) 1 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M) =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) k=1 xk ⊗ ak ∈ Lp(M ) ⊗ ℓ2, we set We have for any family (xk)k>1 in Lp(M ) 1 haj , aiix∗i xj(cid:19) It identifies to the space of n . n (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (2.1) The space Lp(M, ℓ2 c) c) . n Xk=1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) xk ⊗ ek(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M,ℓ2 r) =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) n Xk=1 x∗k2(cid:19) 1 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M) =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) n Xk=1 e1k ⊗ xk(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Sp(Lp(M)) . 3 For any 1 6 p < ∞ and for any x1, . . . , xn ∈ Lp(M ), we have n c) (2.2) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xk=1 n Xk=1 n Xk=1 : (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) xk ⊗ ek(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M,ℓ2 yk ⊗ ek(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp∗ (M,ℓ2 6 1). c) for Lp(cid:0)B(ℓ2), ℓ2 c(cid:1). If r). For any u ∈ Lp(M, ℓ2 hxk, ykiLp(M),Lp∗ (M)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = sup((cid:12)(cid:12)(cid:12)(cid:12)(cid:12) rad) = maxnkukLp(M,ℓ2 rad) = infnku1kLp(M,ℓ2 rad) = Lp(M, ℓ2 r). For simplicity, we write Sp(ℓ2 rad) = Lp(M, ℓ2 A similar formula holds for the space Lp(M, ℓ2 2 6 p < ∞ we define the Banach space Lp(M, ℓ2 we have If 1 6 p 6 2 we define the Banach space Lp(M, ℓ2 we have c)∩ Lp(M, ℓ2 r)o. c),kukLp(M,ℓ2 r). For any u ∈ Lp(M, ℓ2 r)o. c) + ku2kLp(M,ℓ2 r). Recall that, if 1 < p < ∞, we have an isometric identification where the infimum runs over all possible decompositions u = u1 + u2 with u1 ∈ Lp(M, ℓ2 u2 ∈ Lp(M, ℓ2 (2.3) kukLp(M,ℓ2 c)+Lp(M, ℓ2 kukLp(ℓ2 Lp(M, ℓ2 rad)∗ = Lp∗ (M, ℓ2 rad). c) and rad), rad), r) Let X be a Banach space and let (εk)k>1 be a sequence of independent Rademacher variables on some probability space Ω. Let Rad(X) ⊂ L2(Ω; X) be the closure of Span(cid:8)εk ⊗ x : k > 1, x ∈ X(cid:9) in the Bochner space L2(Ω; X). Thus for any finite family x1, . . . , xn in X, we have n Xk=1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) εk ⊗ xk(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Rad(X) = ZΩ(cid:13)(cid:13)(cid:13)(cid:13) n Xk=1 εk(ω) xk(cid:13)(cid:13)(cid:13)(cid:13) 2 X 1 2 . dω! If 1 6 p < ∞, the noncommutative Khintchine's inequalities (see [LPP] and [PX]) implies (2.4) rad). We say that a set F ⊂ B(X) is R-bounded if there is a constant C > 0 such that for any finite families T1, . . . , Tn in F, and x1, . . . , xn in X, we have Rad(cid:0)Lp(M )(cid:1) ≈ Lp(M, ℓ2 6 C(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) εk ⊗ Tk(xk)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Rad(X) Xk=1 n n Xk=1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) . εk ⊗ xk(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Rad(X) In this case, we let R(F ) denote the smallest possible C, which is called the R-bound of F. R- boundedness was introduced in [BeG] and then developed in the fundamental paper [ClP]. We refer to the latter paper and to [KW, Section 2] for a detailed presentation. On noncommutative Lp-spaces, it will be convenient to consider two naturals variants of this notion, introduced in [JMX, Chapter 4]. Let 1 < p < ∞. A subset F of B(cid:0)Lp(M )(cid:1) is Col-bounded (resp. Row-bounded) if there exists a constant C > 0 such that for any finite families T1, . . . , Tn in F, and x1, . . . , xn in Lp(M ), we have (2.5) (2.6) xk2(cid:19) 1 2(cid:19) 6 C(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) n (cid:18) n Xk=1 Xk=1(cid:12)(cid:12)Tk(xk)(cid:12)(cid:12) 6 C(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) resp. (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M) (cid:18) n (cid:18) n 2(cid:19) Xk=1 Xk=1(cid:12)(cid:12)Tk(xk)∗(cid:12)(cid:12) 1 1 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M) 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M)!. x∗k2(cid:19) 1 4 The least constant C satisfying (2.5) will be denoted by Col(F ). Obviously any Rad-bounded (resp. Col-bounded, resp. Row-bounded) set is bounded. It follows from (2.4) that if a subset F of B(cid:0)Lp(M )(cid:1) is both Col-bounded and Row-bounded, then it is Rad-bounded. Note that contrary to the case of R-boundedness, a singleton {T} is not automatically Col-bounded or Row-bounded. Indeed, {T} is Col-bounded (resp. Row-bounded) if and only if T ⊗ Iℓ2 extends to a bounded operator on Lp(M, ℓ2 r)). And it turns out that if 1 < p 6= 2 < ∞, according to [JMX, Example 4.1], there exists a bounded operator T on Sp such that T ⊗ Iℓ2 does not extend to a bounded operator on Sp(ℓ2 c). Moreover, T ⊗ Iℓ2 extends to a bounded operator on Sp(ℓ2 r). Then, we also deduce that there are sets F which are Rad-bounded and Col-bounded without being Row-bounded. Similarly, one may find sets which are Rad-bounded and Row-bounded without being Col-bounded, or which are Rad-bounded without being either Row-bounded or Col-bounded. c) (resp. Lp(M, ℓ2 We turn to Ritt operators, the key class of this paper, and recall some of their main features. Details and complements can be found in [ALM], [Bl1], [Bl2], [LM2], [Lyu], [NaZ], [Nev] and [Vit]. Let X be a Banach space. We say that an operator T ∈ B(X) is a Ritt operator if the two sets (2.7) and (cid:8)T n : n > 0(cid:9) (cid:8)n(T n − T n−1) : n > 1(cid:9) are bounded. This is equivalent to the spectral inclusion (2.8) and the boundedness of the set (2.9) σ(T ) ⊂ D (cid:8)(λ − 1)R(λ, T ) : λ > 1(cid:9) where R(λ, T ) = (λI − T )−1 denotes the resolvent operator and D denotes the open unit disc centered at 0. Likewise we say that T is an R-Ritt operator if the two sets in (2.7) are R-bounded. This is equivalent to the inclusion (2.8) and the R-boundedness of the set (2.9). Let T be a Ritt operator. The boundedness of (2.9) implies the existence of a constant K > 0 such that λ − 1(cid:13)(cid:13)R(λ, T )(cid:13)(cid:13)X→X 6 K whenever Re(λ) > 1. This means that I − T is a sectorial operator. Thus for any α > 0, one can consider the fractional power (I − T )α. We refer to [Haa, Chapter 3], [KW] and [MCS] for various definitions of these (bounded) operators and their basic properties. We will use the following two naturals variants of the notion of R-Ritt operator. Definition 2.1 Suppose 1 < p < ∞. Let T be a bounded operator on Lp(M ). We say that T is a Col-Ritt (resp. Row-Ritt) operator if the two sets (2.7) are Col-bounded (resp. Row-bounded). Remark 2.2 Assume that 1 < p < ∞. Let T be a bounded operator on Lp(M ). Using (2.2), it is easy to see that T is Col-Ritt if and only if T ∗ is Row-Ritt on Lp∗ (M ). We let P denote the algebra of all complex polynomials. Let T be a bounded operator on a Banach 2(cid:2). Accordingly with [LM2], we say that T has a bounded H∞(Bγ) functional space X. Let γ ∈ (cid:3)0, π calculus if and only if there exists a constant K > 1 such that (cid:13)(cid:13)ϕ(T )(cid:13)(cid:13)X→X 6 KkϕkH∞(Bγ ) for any ϕ ∈ P. Naturally, we let: Definition 2.3 Suppose 1 < p < ∞. Let T be a bounded operator on Lp(M ). Let γ ∈ (cid:3)0, π if ISp ⊗ T admits a bounded H∞(Bγ) functional calculus on Sp(cid:0)Lp(M )(cid:1). say that T admits a completely bounded H∞(Bγ) functional calculus if T is completely bounded and 2(cid:2). We 5 Let T be a bounded operator on Lp(M ) and γ ∈ (cid:3)0, π H∞(Bγ) functional calculus if and only if there exists a constant K > 1 such that 2(cid:2). Note that T admits a completely bounded for any ϕ ∈ P. (cid:13)(cid:13)ϕ(T )(cid:13)(cid:13)cb,Lp(M)→Lp(M) 6 KkϕkH∞(Bγ ) 3 Results related to Col-Ritt or Row-Ritt operators In the subsequent sections, we need some preliminary results on Col-Ritt or Row-Ritt operators that we present here. Some of them are analogues of existing results in the context of R-Ritt operators, for which we will omit proofs. We start with a variant of [ALM, Proposition 2.8] suitable with our context. The proof is similar, using [JMX, Lemma 4.2] instead of [ALM, Lemma 2.1]. Proposition 3.1 Suppose 1 < p < ∞. Let T be a Col-Ritt operator on Lp(M ). For any α > 0, the set is Col-bounded. Moreover, a similar result holds for Row-Ritt operators. nnα(T )n−1(I − T )α : n > 1, ∈]0, 1]o Moreover, we need the following result [LM2]. In the next statement, we establish a variant of the above result. Theorem 3.2 Suppose 1 < p < ∞. Let T be a bounded operator on Lp(M ) with a bounded H∞(Bγ) functional calculus for some γ ∈(cid:3)0, π Theorem 3.3 Suppose 1 < p < ∞. Let T be a bounded operator on Lp(M ). Assume that T admits a completely bounded H∞(Bγ) functional calculus for some γ ∈ (cid:3)0, π 2(cid:2). Then the operator T is both Col-Ritt and Row-Ritt. 2(cid:2). Then T is R-Ritt. : We will only show the 'column' result, the proof for the 'row' one being the same. We wish Proof to show that the sets F =(cid:8)T m : m > 0(cid:9) and G =(cid:8)m(T m − T m−1) : m > 1(cid:9) are Col-bounded. We consider the operator I⊗T on the noncommutative Lp-space Sp(cid:0)Lp(M )(cid:1). Then, applying Theorem 3.2, we obtain that the sets and T =(cid:8)ISp ⊗ T m : m > 0(cid:9) are Rad-bounded. Now consider x1, . . . , xn in Lp(M ) and T1, . . . , Tn in F. For any finite sequence (εk)16k6n valued in {−1, 1}, we have xk2(cid:19) K =(cid:8)mISp ⊗ (T m − T m−1) : m > 1(cid:9) 1 1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) n Xk=1 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M) n =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M) (cid:18) n (εkxk)∗(εkxk)(cid:19) Xk=1 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) εkek1 ⊗ xk(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Sp(Lp(M)) Xk=1 . 6 Then passing to the average over all possible choices of εk = ±1, we obtain that By a similar computation, we have (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) n Xk=1 It follows that 1 1 n (cid:18) n Xk=1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Tk(xk)2(cid:19) (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M) xk2(cid:19) =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M) Xk=1 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M) Tk(xk)2(cid:19) (cid:18) n Xk=1 1 . n εk ⊗ ek1 ⊗ xk(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Rad(Sp(Lp(M))) Xk=1 εk ⊗ (ISp ⊗ Tk)(ek1 ⊗ xk)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Rad(Sp(Lp(M))) 6 Rad(T )(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M) (cid:18) n Xk=1 xk2(cid:19) . 1 . This concludes the proof of Col-boundedness of F with Col(F ) 6 Rad(T ). The proof for the set G is identical. Remark 3.4 Suppose 1 < p 6= 2 < ∞. The complete boundedness assumption in Theorem 3.3 cannot be replaced by a boundedness assumption. Proof : We have already recalled that, there exists a bounded operator T on Sp such that {T} is not Col-bounded. Let us fix γ ∈ (cid:3)0, π 2(cid:2). We may clearly assume that σ(T ) is included in the open set Bγ. Using the Dunford calculus, it is easy to prove that T is a Ritt operator which admits a bounded H∞(Bγ) functional calculus. The set {T} is not Col-bounded. Hence T cannot be Col-Ritt. few comments. Let T a Ritt operator on Lp(M ). For any α > 0, let us consider Now, we give a precise definition of 'square functions' which clarifies (1.1), (1.2) and (1.4) and a xk = kα− 1 2 T k−1(I − T )α(x) for any k > 1. If the sequence belongs to the space Lp(M, ℓ2 c), then kxkp,T,c,α is defined as the norm of (xk)k>1 in that space. Otherwise, we set kxkp,T,c,α = ∞. In particular, kxkp,T,c,α can be infinite. We define the quantities kxkp,T,r,α by the same way. The quantities kxkp,T,α are defined similarly in [ALM], using the space Lp(M, ℓ2 rad) instead of Lp(M, ℓ2 c). Finally, note that, if 2 6 p < ∞, we have and if 1 6 p 6 2, we have kxkp,T,α = infnkukLp(M,ℓ2 kxkp,T,α = max(cid:8)kxkp,T,c,α,kxkp,T,r,α(cid:9). c) + kvkLp(M,ℓ2 r) : uk + vk = kα− 1 2 T k−1(I − T )αx for any integer ko. In [LM2], the following connection between the boundedness of square functions and functional calculus is established. Theorem 3.5 Suppose 1 < p < ∞. Let T be a bounded operator on Lp(M ). The following assertions are equivalent. 1. The operator T is R-Ritt and T and its adjoint T ∗ both satisfy uniform estimates kxkp,T,1 . kxkLp(M) and kykp∗,T ∗,1 . kykLp∗ (M) for any x ∈ Lp(M ) and y ∈ Lp∗ (M ). 7 2. The operator T admits a bounded H∞(Bγ) functional calculus for some γ ∈(cid:3)0, π 2(cid:2). Recall a special case of the principal result of [ALM]. Theorem 3.6 Let T be an R-Ritt operator on Lp(M ) with 1 < p < ∞. For any α, β > 0 we have an equivalence We shall now present a variant suitable to our context. kxkp,T,α ≈ kxkp,T,β, x ∈ Lp(M ). For any integer n > 1, we identify the algebra Mn of all n × n matrices with the space of linear n. For any infinite matrix [cij]i,j>1, we set maps ℓ2 n → ℓ2 This is the so-called 'regular norm'. We refer to [Pis1] and [Pis5] for more information on regular norms. (cid:13)(cid:13)[cij](cid:13)(cid:13)reg = sup n>1(cid:13)(cid:13)(cid:13)(cid:2)cij(cid:3)16i,j6n(cid:13)(cid:13)(cid:13)B(ℓ2 n) The next proposition will be useful. This result is similar to [ALM, Proposition 2.6]. Proposition 3.7 Suppose 1 < p < ∞. Let [cij ]i,j>1 be an infinite matrix with (cid:13)(cid:13)[cij](cid:13)(cid:13)reg < ∞. Suppose that (cid:8)Tij : i, j > 1(cid:9) is a Col-bounded set of operators on Lp(M ). Then the linear map (cid:2)cijTij(cid:3) : Lp(cid:0)M, ℓ2 xj ⊗ ej c(cid:1) −→ 7−→ +∞ Xj=1 +∞ Lp(cid:0)M, ℓ2 c(cid:1) Xi=1 +∞ cijTij(xj )! ⊗ ei Xj=1 is well-defined and bounded. Moreover, we have a similar result for Row-bounded sets. Proof [ALM, Lemma 2.2], we can write cij = aijbij for any 1 6 i, j 6 n with : We shall only prove the 'Col' result. We can assume that (cid:13)(cid:13)[cij](cid:13)(cid:13)reg 6 1. Let n > 1. By n n sup 16i6n Xj=1 aij2 6 1 and sup 16j6n Xi=1 bij2 6 1. Let x1, . . . , xn ∈ Lp(M ) and y1, . . . , yn ∈ Lp∗ exists a positive constant C such that n n n (M ). Since the set (cid:8)Tij i, j > 1(cid:9) is Col-bounded, there Xi,j=1(cid:10)aij bijTij(xj ), yi(cid:11)Lp(M),Lp∗ (M)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) cijTij(xj ), yi+Lp(M),Lp∗ (M) Xi=1* n Xj=1 Xi,j=1(cid:10)Tij(bij xj), aij yi(cid:11)Lp(M),Lp∗ (M)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 6(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp∗ (M) (cid:18) n (cid:18) n Tij (bijxj)2(cid:19) 2(cid:19) Xi,j=1 Xi,j=1(cid:12)(cid:12)(aij yi)∗(cid:12)(cid:12) 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 6 C(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp∗ (M) (cid:18) n (cid:18) n aij y∗i 2(cid:19) bijxj2(cid:19) Xi,j=1 Xi,j=1 . 1 1 1 1 8 Now, we have Similarly, we have Consequently n Xi,j=1 bijxj2 = n Xj=1 xj2 n Xi=1 bij2! 6 n Xj=1 xj2. n Xi,j=1 aijy∗i 2 6 n Xi=1 y∗i 2. n Xi=1* n Xj=1 cijTij(xj ), yi+Lp(M),Lp∗ (M) Taking the supremum over all y1, . . . , yn ∈ Lp∗ (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 1 1 n xj2(cid:19) 6 C(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) n (cid:18) n y∗i 2(cid:19) Xj=1 Xi=1 (M ) such that (cid:13)(cid:13)(Pn 2(cid:13)(cid:13)Lp∗ (M) 6 1, we obtain i=1 y∗i 2) xj ⊗ ej(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M,ℓ2 6 C(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) cijTij(xj )(cid:19) ⊗ ei(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp(M,ℓ2 Xj=1 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp∗ (M) c) c) . 1 n Xi=1(cid:18) n Xj=1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) by (2.2). We conclude with [JMX, Corollary 2.12]. Now, we state a result which allows to estimate square functions k·kp,T,c,α and k·kp,T,r,α by means of approximation processes, whose proof is similar to [ALM, Lemma 3.2]. Lemma 3.8 Suppose 1 < p < ∞. Assume that T is a Col-Ritt operator on Lp(M ). Let α > 0. 1. Let V be an operator on Lp(M ) such that T V = V T with {V } Col-bounded. Then, for any x ∈ Lp(M ), we have kV (x)kp,T,c,α 6 Col(cid:0){V }(cid:1)kxkp,T,c,α. 2. Let ν > α + 1 be an integer and let x ∈ Ran(cid:0)(I − T )ν(cid:1). Then kxkp,T,c,α −−−−→→1− kxkp,T,c,α. Moreover, the same result holds with k · kp,T,c,α replaced by k · kp,T,r,α for Row-Ritt operators. Now we state an equivalence result in our context similar to Theorem 3.6. Theorem 3.9 Let T be a bounded operator on Lp(M ) with 1 < p < ∞. Let α, β > 0. 1. If T is Col-Ritt, we have an equivalence kxkp,T,c,α ≈ kxkp,T,c,β, x ∈ Lp(M ). 2. If T is Row-Ritt, we have an equivalence kxkp,T,r,α ≈ kxkp,T,r,β, x ∈ Lp(M ). : The proof is similar to the one of [ALM, Theorem 3.3], using Proposition 3.1, Proposition Proof 3.7, Lemma 3.8 and [JMX, Corollary 2.12]. 9 4 Comparison between squares functions and the usual norm We aim at showing Theorem 1.2. We will provide an example on the Schatten space Sp. This example also prove that in general, row and column square functions are not equivalent (Theorem 4.3). Let a a bounded operator on ℓ2. Assume 1 < p < ∞. We let La : Sp → Sp the left multiplication by a on Sp defined by La(x) = ax and we denote Ra : Sp → Sp the right multiplication. It is clear that L∗a and R∗a are the right multiplication and the left multiplication by a on Sp∗. Note that, by [JMX, Proposition 8.4 (4)], if I − a has dense range then Ran(I − La) is dense in Sp. The next statement gives a link between properties of a and its associated multiplication operators. Proposition 4.1 Suppose 1 < p < ∞. Assume that a is a bounded operator on ℓ2. 1. If a is a Ritt operator then the left multiplication La is a Ritt operator on Sp. 2. Let γ ∈(cid:3)0, π that case, La actually has a completely bounded H∞(Bγ) functional calculus. 2(cid:2). Then La has a bounded H∞(Bγ) functional calculus if and only if a has one. In Moreover, we have a similar result for right multiplication. Proof clearly follows. The statement (2) is a straightforward consequence of : We have σ(La) ⊂ σ(a). Moreover, if λ ∈ ρ(a) we have R(λ,La) = LR(λ,a). The first assertion ISp ⊗ La = LIℓ2⊗a The proof of the 'right' result is identical. and f (La) = Lf (a), f ∈ P. We denote by (ek)k>1 the canonical basis of ℓ2. Now, for any integer k > 1, we fix ak = 1 − 1 2k . We consider the selfadjoint bounded diagonal operator a on ℓ2 defined by (4.1) a +∞ Xk=1 xkek! = +∞ Xk=1 akxkek. It follows from the Spectral Theorem for normal operators, that the operator a admits a bounded H∞(Bγ) functional calculus for any γ ∈ (cid:3)0, π 2(cid:2). Thus La and Ra admit a completely bounded H∞(Bγ) functional calculus for any γ ∈(cid:3)0, π 2(cid:2) (hence La and Ra are Ritt operators). Lemma 4.2 Assume that 2 6 p < ∞. Let a be the bounded operator on ℓ2 defined by (4.1). If La : Sp → Sp and Ra : Sp → Sp are the left and right multiplication operators associated to a, we have (4.2) and kxkp,La,c,1 ≈ kxkSp kxkp,Ra,r,1 ≈ kxkSp, x ∈ Sp. : We will only show the result for the operator La, the proof for Ra being the same. For any Proof x ∈ Sp and any ∈]0, 1[, we have k(cid:0)(La)k−1(I − La)(x)(cid:1)∗(cid:0)(La)k−1(I − La)(x)(cid:1) = k(cid:0)(a)k−1(I − a)x(cid:1)∗(cid:0)(a)k−1(I − a)x(cid:1) = kx∗(I − a)(a)2(k−1)(I − a)x = kx∗(I − La)2(La)2(k−1)(x). Now, for any z ∈ D, we have (4.3) +∞ Xk=1 kzk−1 = (1 − z)−2. 10 Since the operator La is a contraction, we deduce that, for every ∈]0, 1[, the operator I − (La)2 is invertible and that we have (4.4) +∞ Xk=1 k(La)2(k−1) =(cid:0)I − (La)2(cid:1)−2 , the series being absolutely convergent. Then we deduce that the series +∞ Xk=1 k(cid:0)(La)k−1(I − La)(x)(cid:1)∗(cid:0)(La)k−1(I − La)(x)(cid:1) is convergent in the Banach space S 2 and that p +∞ Xk=1 k(cid:0)(La)k−1(I − La)(x)(cid:1)∗(cid:0)(La)k−1(I − La)(x)(cid:1) = x∗(I − La)2(cid:0)I − (La)2(cid:1)−2 = x∗(I + a)−2x. x We deduce that Then, for any x ∈ Sp, we obtain the estimate 6 kxkSp . By a similar computation, for any x ∈ Sp, we have 1 kxkLa,c,1 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:16)x∗(I + a)−2x(cid:17) 2(cid:13)(cid:13)(cid:13)(cid:13)Sp =(cid:13)(cid:13)(I + a)−1x(cid:13)(cid:13)Sp . kxkp,La,c,1 6(cid:13)(cid:13)(I + a)−1(cid:13)(cid:13)B(ℓ2)kxkSp 1 2kxkSp 6 kxkp,La,c,1. Applying Lemma 3.8 (2), we deduce an equivalence 1 2kxkSp 6 kxkp,La,c,1 6 kxkSp , x ∈ Ran(cid:0)(I − La)2(cid:1). For any integer n > 1, we let dn the bounded diagonal operator on ℓ2 defined by the matrix It is not difficult to see that, for any integer n > 1, the range of Ldn is a diag(1, . . . , 1, 0, . . .). subspace of Ran(cid:0)(I − La)2(cid:1). Hence we actually have 1 Then, on the one hand, we obtain 2(cid:13)(cid:13)Ldn (x)(cid:13)(cid:13)Sp 6(cid:13)(cid:13)Ldn(x)(cid:13)(cid:13)p,La,c,1 6(cid:13)(cid:13)Ldn (x)(cid:13)(cid:13)Sp , x ∈ Sp, n > 1. x ∈ Sp, n > 1. By [JMX, Corollary 2.12] and (2.1), this latter inequality is equivalent to (cid:13)(cid:13)Ldn (x)(cid:13)(cid:13)p,La,c,1 6 kxkSp, (I − La)(cid:0)Ldn (x)(cid:1)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Sp(Sp) 11 1 2 Lk−1 a ek1 ⊗ k l Xk=1 (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) . kxkSp , x ∈ Sp, n > 1, l > 1. Passing to the limit in the above inequality, we infer that (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) l Xk=1 ek1 ⊗ k 1 2Lk−1 a (I − La)(x)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Sp(Sp) Using again [JMX, Corollary 2.12], we obtain that . kxkSp, x ∈ Sp, l > 1. kxkp,La,c,1 6 kxkSp, x ∈ Sp. Note, in particular that, for any x ∈ Sp, we have kxkp,La,c,1 < ∞. On the other hand, note that, for any integer n > 1, the operators La and Ldn commute. Hence, for any x ∈ Sp and any integer n > 1, we have 1 +∞ ek1 ⊗ k kLdn(x)kSp .(cid:13)(cid:13)Ldn (x)(cid:13)(cid:13)p,La,c,1 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xk=1 2 Lk−1 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (ISp ⊗ Ldn )(cid:18) +∞ Xk=1 a (I − La)(cid:0)Ldn (x)(cid:1)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Sp(Sp) 1 ek1 ⊗ k 2Lk−1 a (I − La)(x)(cid:19)(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Sp(Sp) . Letting n to the infinity, we deduce that The proof is complete. kxkSp . kxkp,La,c,1, x ∈ Sp. Theorem 4.3 Let α > 0. Let a be the bounded operator on ℓ2 defined by (4.1). Let La : Sp → Sp and Ra : Sp → Sp be the left and right multiplication operators associated to a. Assume that 2 < p < ∞. Then (4.5) sup(kxkp,La,c,α kxkp,La,r,α : x ∈ Sp) = ∞ and sup(kxkp,Ra,r,α kxkp,Ra,c,α : x ∈ Sp) = ∞. Assume that 1 < p < 2. Then (4.6) sup(kxkp,La,r,α kxkp,La,c,α : x ∈ Sp) = ∞ and sup(kxkp,Ra,c,α kxkp,Ra,r,α : x ∈ Sp) = ∞. : By Theorem 3.9, it suffices to prove the result for one specific real α. Throughout the Proof proof, we will use α = 1. We first assume that 2 < p < ∞. Given an integer n > 1, we consider e = e1 + ··· + en ∈ ℓ2 n and x = 1√n e ⊗ e ∈ Sp. Clearly, we have xx∗ = eij. n Xi,j=1 12 Now, we have a k(cid:0)Lk−1 (I − La)(x)(cid:1)(cid:0)Lk−1 a n (I − La)(x)(cid:1)∗ = k(cid:0)ak−1(I − a)x(cid:1)(cid:0)ak−1(I − a)x(cid:1)∗ kak−1(I − a)eij(I − a)ak−1 = kak−1(I − a)xx∗(I − a)ak−1 Xi,j=1 = Xi,j=1 (1 − ai)(1 − aj)k(aiaj)k−1eij. = n Using the equality (4.3), we obtain that the series +∞ Xk=1 is convergent in S 2 and that p a k(cid:0)Lk−1 (I − La)(x)(cid:1)(cid:0)Lk−1 a (I − La)(x)(cid:1)∗ +∞ Xk=1 a k(cid:0)Lk−1 Now, note that We deduce that n a (I − La)(x)(cid:1)(cid:0)Lk−1 (I − La)(x)(cid:1)∗ = (1 − ai)(1 − aj)(1 − aiaj)−2 = Xi,j=1 2i+j (2i + 2j − 1)2 . (1 − ai)(1 − aj)(1 − aiaj)−2eij. We let A =h We have 2i+j (2i+2j−1)2i16i,j6n be the n × n matrix in the last right member of the above equations. 1 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Sp . p 2 S n 1 2 2i+j 2i+j kxkp,La,r,1 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) n (2i + 2j − 1)2 eij(cid:19) Xi,j=1 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (2i + 2j − 1)2 eij(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) Xi,j=1 (2i + 2j − 1)2!2 Xi,j=1 Xi,j=1 (2i + 2j − 1)4 . kAk2 2i+j 4i+j S2 n = = n n Moreover, note that 4i+j (2i + 2j)4 4i+j (2i + 2j − 1)4 6 16 = 16 1 2i−j + 2j−i + 2!2 6 16 4i−j . 13 Thus we have If 4 6 p < ∞, we obtain kAk2 S2 n 6 32(cid:18)Xk∈Z 1 4k(cid:19)n ≈ n. kxkp,La,r,1 = kAk 1 2 S p 2 n 6 kAk 1 2 S2 n . n 1 4 . Since x = 1√n e⊗ e is rank one, its norm in Sp does not depend on p, and it is equal to 1√nkek2 Then, by Lemma 4.2, we have kxkp,La,c,1 ≈ √n. We obtain the first equality of (4.5) in that case. ℓ2 n If 2 < p 6 4, we can write 1 = 1−θ 1 + θ 2 with 0 < θ 6 1. Then p 2 = √n. By construction, we have A > 0, hence we have kxk2 p,La,r,1 = kAkS p 2 n 6 kAk1−θ n kAkθ S1 S2 n . kAkS1 n n = 4i = Tr n Xi,j=1 Xi=1 (2i+1 − 1)2 6 p,La,r,1 . n1−θn kxk2 2i+j (2i + 2j − 1)2 eij! 4i (2i)2 = n. n Xi=1 θ 2 = n1− θ 2 . Thus Recall that kxkp,La,c,1 ≈ √n. We obtain that kxkp,La,c,1 kxkp,La,r,1 & 1 2 n 2 − θ n 1 4 = n θ 4 . Since n was arbitrary and θ > 0, we obtain the first part of (4.5) in this case. Likewise, the above proof has a 'right analog' which proves the second equality of (4.5). We now turn to the proof of (4.6). We assume that 1 < p < 2. The second part of (4.5) says (4.7) sup(kykp∗,L∗ kykp∗,L∗ a,r,1 a,c,1 : y ∈ Sp∗) = ∞. To prove the first equality of (4.6), assume on the contrary that there is a constant K > 0 such that for any x ∈ Sp (4.8) kxkp,La,r,1 6 Kkxkp,La,c,1. We begin by showing a duality relation between k · kp∗,L∗ a,c,1 and k · kp,La,r,1. Let y ∈ Sp∗ and x ∈ Sp. For any integer n > 1, recall that dn is the bounded diagonal operator on ℓ2 defined by the 14 matrix diag(1, . . . , 1, 0, . . .). By (4.4), for any 0 < < 1 and any integer n > 1, we have Ldn(x)+Sp∗ ,Sp (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) +∞ y, 1 +∞ +∞ Xk=1 2 (L∗a)k−1(I − L∗a)(I + L∗a)2y, k (cid:12)(cid:12)(cid:12)(cid:10)y,Ldn (x)(cid:11)Sp∗ ,Sp(cid:12)(cid:12)(cid:12) k(La)2(k−1)(cid:0)I − (La)2(cid:1)2 =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Ldn (x)ESp∗ ,Sp(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Xk=1Dy, k(La)2(k−1)(cid:0)I − (La)2(cid:1)2 =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Xk=1Dk 2 (L∗a)k−1(I − L∗a)(I + L∗a)2y(cid:17)k>1(cid:13)(cid:13)(cid:13)(cid:13)Sp(ℓ2 6(cid:13)(cid:13)(cid:13)(cid:13)(cid:16)k (cid:12)(cid:12)(cid:12)(cid:10)y,Ldn (x)(cid:11)Sp∗ ,Sp(cid:12)(cid:12)(cid:12) 1 Now, it is easy to see that {L∗a} is Col-bounded. We infer that 1 c)(cid:13)(cid:13)Ldn (x)(cid:13)(cid:13)p,La,r,1 Assume for a while that y ∈ Ran(cid:0)(I − L∗a)2(cid:1). By Lemma 3.8 (2), letting to 1, we obtain 2 (L∗a)k−1(cid:0)I − L∗a)y(cid:17)k>1(cid:13)(cid:13)(cid:13)(cid:13)Sp(ℓ2 a,c,1(cid:13)(cid:13)Ldn (x)(cid:13)(cid:13)p,La,r,1. .(cid:13)(cid:13)(cid:13)(cid:13)(cid:16)k = kykp∗,L∗ 1 2 (La)k−1(I − La)Ldn (x)ESp∗ ,Sp(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) c)(cid:13)(cid:13)Ldn (x)(cid:13)(cid:13)p,La,r,1. Letting n to the infinity, we obtain (cid:12)(cid:12)(cid:12)(cid:10)y,Ldn(x)(cid:11)Sp∗ ,Sp(cid:12)(cid:12)(cid:12) . kykp∗,L∗ a,c,1(cid:13)(cid:13)Ldn (x)(cid:13)(cid:13)p,La,r,1. According to (4.8) and the first part of (4.2), we deduce that (cid:12)(cid:12)hy, xiSp∗ ,Sp(cid:12)(cid:12) . kykp∗,L∗ (cid:12)(cid:12)hy, xiSp∗ ,Sp(cid:12)(cid:12) . kykp∗,L∗ . kykp∗,L∗ a,c,1kxkp,La,r,1. a,c,1kxkp,La,c,1 a,c,1kxkSp. By duality, we finally obtain that (4.9) kykSp∗ . kykp∗,L∗ a,c,1. For an arbitrary y ∈ Sp∗, we also obtain (4.9) by applying it to L∗dn The second equivalence of (4.2) says that kykp∗,L∗ and completes the proof of the first part of (4.6). The proof of the second part is similar. (y) and then passing to the limit. a,r,1 ≈ kykSp∗ for any y ∈ Sp∗. This contradicts (4.7) For a operator admitting a completely bounded H∞(Bγ) functional calculus, it also seems interest- ing, in view of the equivalence (1.3), to compare the column and row square functions with the usual norm k·kLp(M). If T is a operator with Ran(I − T ) dense in Lp(M ) which admits a bounded H∞(Bγ) functional calculus for some γ ∈(cid:3)0, π 2(cid:2), the equivalence (1.3) and Theorems 3.5 and 3.6 implies that and kxkLp(M) . kxkp,T,c,1 kxkLp(M) . kxkp,T,r,1 if 1 < p 6 2 and kxkp,T,c,1 . kxkLp(M) and kxkp,T,r,1 . kxkLp(M) if 2 6 p < ∞, for any x ∈ Lp(M ). The following result says that except for p = 2, these estimates cannot be reversed: 15 Corollary 4.4 Suppose that 2 < p < ∞ (resp. 1 < p < 2). Let α > 0. There exists a Ritt operator T on the Schatten space Sp, with Ran(I − T ) dense in Sp, which admits a completely bounded H∞(Bγ) functional calculus with γ ∈(cid:3)0, π sup(cid:26) kxkSp kxkp,T,c,α 2(cid:2) such that : x ∈ Sp(cid:27) = ∞ resp. sup(cid:26)kxkp,T,c,α kxkSp Moreover, the same result holds with k · kp,T,c,α replaced by k · kp,T,r,α. Proof 4.2 and Theorem 4.3. : One more time, we only need to prove this result for α = 1. Then, this follows from Lemma : x ∈ Sp(cid:27) = ∞!. 5 An alternative square function for 1 < p < 2 Let T be a Ritt operator on Lp(M ), with 1 < p < 2. For any α > 0, we may consider an alternative square function by letting kxkp,T,0,α = infnkx1kp,T,c,α + kx2kp,T,r,α : x = x1 + x2o for any x ∈ Lp(M ). kxkp,T,0,β are equivalent for any α, β > 0. and kx2kp,T,r,α < ∞. Letting uk = kα− 1 Note that if T is both Col-Ritt and Row-Ritt, by Theorem 3.9, the square functions kxkp,T,0,α and Suppose that kxkp,T,0,α is finite and that we have a decomposition x = x1+x2 with kx1kp,T,c,α < ∞ 2 T k−1(I − T )αx2, we have 2 T k−1(I − T )αx1 and vk = kα− 1 k > 1. 2 T k−1(I − T )αx = uk + vk, kα− 1 Moreover, the sequences u and v belong to Lp(cid:0)M, ℓ2 r(cid:1) respectively. We deduce that kxkp,T,α 6 kxkp,T,0,α, We do not know if the two square functions k · kp,T,α and k · kp,T,0,α are equivalent in general. In the next statement, we give a sufficient condition for an such equivalence to hold true. c(cid:1) and Lp(cid:0)M, ℓ2 x ∈ Lp(M ). Theorem 5.1 Suppose 1 < p < 2. Let T be a bounded operator on Lp(M ) with Ran(I − T ) dense in Lp(M ). Assume that T is both Col-Ritt and Row-Ritt. Let α, η > 0. Suppose that T satisfies a 'dual square function estimate' (5.1) kykp∗,T ∗,η . kykLp∗(M), y ∈ Lp∗ (M ). Then we have an equivalence kxkp,T,α ≈ kxkp,T,0,α, x ∈ Lp(M ). Indeed, there is a positive constant C such that whenever x ∈ Lp(M ) satisfies kxkp,T,α < ∞, then there exists x1, x2 ∈ Lp(M ) such that x = x1 + x2 and kx1kp,T,c,α + kx2kp,T,r,α 6 Ckxkp,T,α. 16 : Since T is both Col-Ritt and Row-Ritt, it is also an R-Ritt operator. Then, by Theorem Proof 3.6 and Theorem 3.9, we only need to prove this result for α = 1 and η = 1. Observe that, for any y ∈ Lp∗ (M ), we have (cid:13)(cid:13)(cid:13)(cid:13)(cid:16)k We let 1 2 (T ∗)k−1(I + T ∗)2(I − T ∗)y(cid:17)k>1(cid:13)(cid:13)(cid:13)(cid:13)Lp∗ (M,ℓ2 .(cid:13)(cid:13)(I + T ∗)2(cid:13)(cid:13)Lp∗ (M)→Lp∗ (M)(cid:13)(cid:13)(cid:13)(cid:13)(cid:16)k . kykLp∗ (M) by (5.1). 1 rad) 2 (T ∗)k−1(I − T ∗)y(cid:17)k>1(cid:13)(cid:13)(cid:13)(cid:13)Lp∗ (M,ℓ2 Lp∗(cid:0)M, ℓ2 rad(cid:1) rad) Z : Lp∗ (M ) −→ y 7−→ (cid:16)k 1 2 (T ∗)k−1(I + T ∗)2(I − T ∗)y(cid:17)k>1 denote the resulting bounded map. Let x ∈ Lp(M ) such that kxkp,T,1 < ∞. There exists two elements u ∈ Lp(cid:0)M, ℓ2 r(cid:1) such that for any positive integer k c(cid:1) and v ∈ Lp(cid:0)M, ℓ2 1 (5.2) uk + vk = k 2 T k−1(I − T )x and such that c) + kvkLp(M,ℓ2 Recall that we have contractive inclusions Lp(cid:0)M, ℓ2 Thus, by (2.3), we can define x1 and x2 of Lp(M ) by r) 6 2kxkp,T,1. c(cid:1) ⊂ Lp(cid:0)M, ℓ2 kukLp(M,ℓ2 rad(cid:1) and Lp(cid:0)M, ℓ2 r(cid:1) ⊂ Lp(cid:0)M, ℓ2 rad(cid:1). x1 = Z∗u and x2 = Z∗v. We will show that x = x1 + x2. Since T is a Col-Ritt-operator, by Proposition 3.1 (or by [ALM, Proposition 2.8]), we infer that there exists a positive constant C such that k +∞ Xk=1(cid:13)(cid:13)(cid:13) 2 1 2 T k−1(I − T )2(cid:13)(cid:13)(cid:13) Lp(M)→Lp(M) 2 Lp(M)→Lp(M) +∞ = k(cid:13)(cid:13)(cid:13) T k−1(I − T )2(cid:13)(cid:13)(cid:13) Xk=1 Xk=1 1 k3 < ∞. +∞ 6 C2 deduce that the series For any 1 < p < 2, by [JMX, Proposition 2.5], we have the contractive inclusion Lp(cid:0)M, ℓ2 ℓ2(cid:0)Lp(M )(cid:1). We deduce that P+∞k=1 kukk2 2 T k−1(cid:0)I − T 2(cid:1)2 c(cid:1) ⊂ Lp(M) < ∞. According to the Cauchy-Schwarz inequality, we 2 T k−1(I − T )2uk uk = (I + T )2 Xk=1 Xk=1 +∞ +∞ k k 1 1 17 converges absolutely in Lp(M ). Now, for any y ∈ Lp∗ (M ), we have D(I − T )x1, yELp(M),Lp∗ (M) 1 rad) rad),Lp∗ (M,ℓ2 =D(I − T )Z∗u, yELp(M),Lp∗ (M) =Du, Z(I − T ∗)yELp(M,ℓ2 =(cid:28)u,(cid:16)k Xk=1Duk, k = +∞ Xk=1 2 (T ∗)k−1(I + T ∗)2(I − T ∗)2y(cid:17)k>1(cid:29)Lp(M,ℓ2 yELp(M),Lp∗ (M) 2 (T ∗)k−1(cid:0)I − (T ∗)2(cid:1)2 2 T k−1(cid:0)I − T 2(cid:1)2 uk, y+Lp(M),Lp∗ (M) +∞ = k . 1 1 rad),Lp∗ (M,ℓ2 rad) Thus, we deduce that (5.3) Similarly we have Now, we infer that (I − T )x1 = (I − T )x2 = +∞ Xk=1 +∞ Xk=1 k 1 2 T k−1(cid:0)I − T 2(cid:1)2 uk. k 1 2 T k−1(cid:0)I − T 2(cid:1)2 vk. +∞ k 1 2 T k−1(cid:0)I − T 2(cid:1)2 vk 1 2 T k−1(I − T )x by (5.2) (I − T )(x1 + x2) = = = = By (4.3), for any z ∈ D, we have +∞ +∞ Xk=1 Xk=1 Xk=1 Xk=1 +∞ +∞ 1 1 (uk + vk) k k uk + Xk=1 2 T k−1(cid:0)I − T 2(cid:1)2 2 T k−1(cid:0)I − T 2(cid:1)2 2 T k−1(cid:0)I − T 2(cid:1)2 kT 2k−2(I + T )2(I − T )3x. k k 1 +∞ Xk=1 kz2k−2(1 − z2)2 = 1. Since the operator T is power bounded, we note that for every ∈]0, 1[ we have (5.4) I = +∞ Xk=1 k(T )2k−2(cid:0)I − (T )2(cid:1)2 , 18 the series being absolutely convergent. Hence, for any ∈]0, 1[, we have k(T )2k−2(cid:0)I − (T )2(cid:1)2 k(T )2k−2(I + T )2(I − T )3x. (I − T )x = (I − T ) Xk=1 +∞ +∞ = x Xk=1 +∞ Xk=1 It is not difficult to see that the latter series is normally convergent on [0,1]. Hence, letting to 1, we deduce that Then we obtain (I − T )x = kT 2k−2(I + T )2(I − T )3x. (I − T )x = (I − T )(x1 + x2). Since the space Ran(I − T ) is dense in Lp(M ), by the Mean Ergodic Theorem (see [Kre, Section 2.1]), the operator I − T is injective. Consequently, we have x = x1 + x2. Now, it remains to estimate kx1kp,T,1,c and kx2kp,T,1,r. According to (5.3), we have for any integer m > 1. It is convenient to write this as m m 1 2 T m−1(I − T )x1 = 1 2 m k +∞ Xk=1 1 uk 2 T k+m−2(cid:0)I − T 2(cid:1)2 2 T m−1(I − T )x1 = (I + T )2ym with 1 (5.5) Now, observe that ym = +∞ Xk=1 1 2 m k 1 2 T k+m−2(I − T )2uk. 1 2 m k 1 2 T k+m−2(I − T )2 = (k + m − 1)2 · (k + m − 1)2T k+m−2(I − T )2. 1 k 2 m 1 2 According to [ALM, Proposition 2.3] and [ALM, Lemma 2.4], the matrix represents an element of B(ℓ2). Moreover, by Proposition 3.1, the set 1 1 2 (cid:20) k 2 m (k + m − 1)2(cid:21)k,m>1 n(k + m − 1)2T k+m−2(I − T )2 : k, m > 1o is Col-bounded. By Proposition 3.7, we deduce that (ym)m>1 ∈ Lp(cid:0)M, ℓ2 Since {T} is Col-bounded, we have (cid:13)(cid:13)(ym)m>1(cid:13)(cid:13)Lp(M,ℓ2 c) . kukLp(M,ℓ2 c). c(cid:1) and that kx1kp,T,c,1 =(cid:13)(cid:13)(cid:13)(cid:13)(cid:16)m 1 2 T m−1(I − T )x1(cid:17)m>1(cid:13)(cid:13)(cid:13)(cid:13)Lp(M,ℓ2 c) by (5.5) =(cid:13)(cid:13)(cid:13)(cid:13)(cid:16)(I + T )2ym(cid:17)m>1(cid:13)(cid:13)(cid:13)(cid:13)Lp(M,ℓ2 .(cid:13)(cid:13)(ym)m>1(cid:13)(cid:13)Lp(M,ℓ2 c). c) 19 Then, by Theorem 5.1 above, the norms k · kp,T,α and k · kp,T,0,α are equivalent. Furthermore, by Theorem 3.6 and (1.3), k·kp,T,α is equivalent to the usual norm k·kLp(M), which proves the result. Assume now that τ is finite and normalized, that is, τ (1) = 1. Following [HaM] and [Ric] (see also [AD]), we say that a linear map T on M is a Markov map if T is unital, completely positive and trace preserving. As is well known, such a map is necessarily normal and for any 1 6 p < ∞, it extends to a contraction Tp on Lp(M ). We say that T is selfadjoint if, for any x, x′ ∈ M , we have This is equivalent to T2 being selfadjoint in the Hilbertian sense. We also consider the operator τ(cid:0)T (x)x′(cid:1) = τ(cid:0)xT (x′)(cid:1). Ap = I − Tp. Finally, we deduce that there exists a positive constant C such that Moreover, we have a similar result for x2. Finally, we have kx1kp,T,c,1 6 CkukLp(M,ℓ2 c). kx1kp,T,c,1 + kx2kp,T,r,1 6 CkukLp(M,ℓ2 6 Ckxkp,T,1. c) + CkvkLp(M,ℓ2 r) Corollary 5.2 Suppose 1 < p < 2. Let T be a bounded operator on Lp(M ) with Ran(I − T ) dense in Lp(M ) and let α > 0. Assume that T admits a completely bounded H∞(Bγ) functional calculus for some γ ∈(cid:3)0, π 2(cid:2). Then we have an equivalence infnkx1kp,T,c,α + kx2kp,T,r,α : x = x1 + x2o ≈ kxkLp(M), : By Theorem 3.3, the operator T is both Col-Ritt and Row-Ritt (hence R-Ritt). Moreover, x ∈ Lp(M ). Proof by Theorem 3.5, it satisfies a 'dual square estimate' kykp∗,T ∗,1 . kykLp∗ (M), y ∈ Lp∗ (M ). The following result is proved in the proof of [LM2, Proposition 8.7] with bounded instead of completely bounded. But a careful reading of the proof shows that we have this stronger result. We refer to [Haa], [JMX], [LM1] and [LM2] for information on H∞(Σθ) functional calculus. Proposition 5.3 Suppose 1 < p < ∞. Let T be a selfadjoint Markov map on M . Then the operator Ap is sectorial and admits a completely bounded H∞(Σθ) functional calculus for some θ ∈(cid:3)0, π 2(cid:2). Assume 1 < p < ∞. At this point, it is crucial to recall that Lp-realizations Tp of Markov maps T on M such that −1 /∈ σ(T2) are Ritt operators, as noticed by C. Le Merdy in [LM2]. Let T be a selfadjoint Markov map on M . According to [LM2] and Proposition 5.3, we obtain that Tp admits a completely bounded H∞(Bγ) functional calculus for some γ ∈ (cid:3)0, π 2(cid:2). Hence, by Corollary 5.2, we deduce the following result which strengthens a result of [LM2]. Corollary 5.4 Suppose 1 < p < 2. Let T be a selfadjoint Markov map on M such that −1 /∈ σ(T2) with Ran(I − Tp) dense in Lp(M ). Then, for any α > 0 there exists a positive constant C such that for any x ∈ Lp(M ), there exists x1, x2 ∈ Lp(M ) satisfying x = x1 + x2 and (cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) +∞ Xk=1 k2α−1(cid:12)(cid:12)(cid:12)(cid:16)T k−1(I − T )α(x2)(cid:17)∗(cid:12)(cid:12)(cid:12) Acknowledgement. We wish to thank my thesis adviser Christian Le Merdy for his support and k2α−1(cid:12)(cid:12)T k−1(I − T )α(x1)(cid:12)(cid:12) +(cid:13)(cid:13)(cid:13)(cid:13)(cid:13) (cid:18) +∞ Xk=1 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp 1 2(cid:13)(cid:13)(cid:13)(cid:13)(cid:13)Lp 2(cid:19) 1 2(cid:19) 6 CkxkLp(M). advice and Éric Ricard for simplifications in some approximation arguments. 20 References [AD] C. Anantharaman-Delaroche. On ergodic theorems for free group actions on noncommutative spaces. Probab. Theory Related Fields 135 (2006), no. 4, 520 -- 546. [ALM] C. Arhancet and C. Le Merdy. Dilation of Ritt operators on Lp-spaces. Preprint, arXiv:1106.1513. [BeG] E. Berkson and T. A. Gillespie. Spectral decompositions and harmonic analysis on UMD spaces. Studia Math. 112 (1994), no. 1, 13 -- 49. [Bl1] S. Blunck. Maximal regularity of discrete and continuous time evolution equations. Studia Math. 146 (2001), no. 2, 157 -- 176. [Bl2] S. Blunck. Analyticity and discrete maximal regularity on Lp-spaces. J. Funct. Anal. 183 (2001), no. 1, 211 -- 230. [ClP] P. Clement, B. de Pagter, F. A. Sukochev and H. Witvliet. Schauder decomposition and multiplier theorems. Studia Math. 138 (2000), no. 2, 135 -- 163. [HaM] U. Haagerup and M. Musat. Factorization and dilation problems for completely positive maps on von Neumann algebras. Preprint, arXiv:1009.0778. [Haa] M. Haase. The functional calculus for sectorial operators. Operator Theory: Advances and Applications, 169. Birkhäuser Verlag, 2006. [JMX] M. Junge, C. Le Merdy, and Q. Xu. H ∞ functional calculus and square functions on noncommutative Lp-spaces. Astérisque No. 305, 2006. [Kre] U. Krengel. Ergodic theorems. With a supplement by Antoine Brunel. de Gruyter Studies in Mathe- matics, Berlin, 1985. [KW] P. Kunstmann and L. Weis. Lp-regularity for parabolic equations, Fourier multiplier theorems and H ∞-functional calculus. Functional analytic methods for evolution equations. Lecture Notes in Math. 1855, Springer, Berlin, 2004. [LM1] C. Le Merdy. Square functions, bounded analytic semigroups, and applications. Perspectives in operator theory, 191 -- 220, Banach Center Publ., 75, Polish Acad. Sci., Warsaw, 2007. [LM2] C. Le Merdy. H ∞ functional calculus and square function estimates for Ritt operators. Preprint. [LMX] C. Le Merdy, Q. Xu. Maximal theorems and square functions for analytic operators on Lp-spaces. Preprint, arXiv:1011.1360. [LPP] F. Lust-Piquard and G. Pisier. Noncommutative Khintchine and Paley inequalities. Ark. Mat. 29 (1991), no. 2, 241 -- 260. [Lyu] Y. Lyubich. Spectral localization, power boundedness and invariant subspaces under Ritt's type condi- tion. Studia Math. 134 (1999), no. 2, 153 -- 167. [MCS] C. Martinez Carracedo and M. Sanz Alix. The theory of fractional powers of operators. North-Holland Mathematics Studies, 187. North-Holland Publishing, Amsterdam, 2001. [NaZ] B. Nagy and J. Zemanek. A resolvent condition implying power boundedness. Studia Math. 134 (1999), no. 2, 143 -- 151. [Nev] O. Nevanlinna. Convergence of iterations for linear equations. Lectures in Mathematics ETH Zürich, Birkhäuser Verlag, Basel, 1993. [Pis1] G. Pisier. Complex interpolation and regular operators between Banach lattices. Arch. Math. 62 (1994), no. 3, 261 -- 269. [Pis2] G. Pisier. Regular operators between non-commutative Lp-spaces. Bull. Sci. Math. 119 (1995), no. 2, 95 -- 118. [Pis3] G. Pisier. Non-commutative vector valued Lp-spaces and completely p-summing maps. Astérisque, 247, 1998. [Pis4] G. Pisier. Introduction to operator space theory. Cambridge University Press, Cambridge, 2003. 21 [Pis5] G. Pisier. Complex interpolation between Hilbert, Banach and operator spaces. Mem. Amer. Math. Soc. 208 (2010), no. 978. [PX] G. Pisier and Q. Xu. Non-commutative Lp-spaces. Handbook of the Geometry of Banach Spaces volume II: 1459 -- 1517, 2003. [Ric] E. Ricard. A Markov dilation for self-adjoint Schur multipliers. Proc. Amer. Math. Soc. 136 (2008), no. 12, 4365 -- 4372. [Vit] P. Vitse. A band limited and Besov class functional calculus for Tadmor-Ritt operators. Arch. Math. 85 (2005), no. 4, 374 -- 385. Laboratoire de Mathématiques, Université de Franche-Comté, 25030 Besançon Cedex, France [email protected] 22
1502.04290	1	1502	2015-02-15T08:03:39	$K$-theory for the crossed products of infinite tensor product of $C^*$-algebras by the shift	[ "math.OA" ]	C. Schochet shows K\"unneth theorem for the $C^$-algebras in the smallest class of nuclear $C^$-algebras which contains the separable Type I algebras and is closed under some operations. We calculate the $K$-theory for the crossed product of the infinite tensor product of a unital $C^$-algebra in this class by the shift. In particular, we calculate the $K$-theory of the lamplighter group $C^$-algebra.	math.OA	math	K-theory for the crossed products of infinite tensor product of C ∗-algebras by the shift Issei Ohhashi April 2, 2019 Abstract C. Schochet shows Kunneth theorem for the C ∗-algebras in the smallest class of nuclear C ∗-algebras which contains the separable Type I algebras and is closed under some oper- ations. We calculate the K-theory for the crossed product of the infinite tensor product of a unital C ∗-algebra in this class by the shift. In particular, we calculate the K-theory of the lamplighter group C ∗-algebra. 1 Introduction Let A be a unital C ∗-algebra and A⊗n denote the (minimal) tensor product of n copies of A for each n ≥ 1. An infinite tensor product A⊗ Z of A is the inductive limit of the sequence {A⊗2n−1}n with the embeddings a 7→ 1A ⊗ a ⊗ 1A. Note that A⊗ Z is represented as the closed linear span of ai : ai ∈ A, aj = 1A for all but finitely many j(cid:27) . A shift s on A⊗ Z is the automorphism of A⊗ Z satisfying (cid:26)Ni∈Z s(cid:18)Ni∈Z ai(cid:19) = Ni∈Z ai−1 ai ∈ A⊗ Z. for all Ni∈Z We consider the crossed product A⊗ Z ⋊s Z of A⊗ Z by Z under the action by s. Example 1. (i) Let C(X) denote the C ∗-algebra of all continuous functions on a compact Hausdorff space X and θ be an automorphism on X Z defined by (xi)i∈Z 7→ (xi+1)i∈Z. Then C(X Z) ⋊θ Z ∼= C(X)⊗ Z ⋊s Z. (ii) Let Γ ≀ Z denote the wreath product of an amenable discrete group Γ with Z. Then, since the group C ∗-algebra C ∗(Γ) of Γ is nuclear, C ∗(Γ ≀ Z) ∼= C ∗(Γ)⊗ Z ⋊s Z. 1 For a C ∗-algebra B, K∗(B) is the (Z2-)graded K-theory K0(B) ⊕ K1(B), where Z2 := Z /2 Z. Let N denote the smallest class of separable nuclear C ∗-algebras which contains the separable Type I algebras and is closed under the operations of taking ideals, quotients, extensions, inductive limits, stable isomorphism, and crossed products by Z or R. The class N plays an essential role in Kunneth theorem by C. Schochet: Theorem 1 (the Spatial Case of Kunneth Theorem [S, Theorem 2.14]). Let A and B be C ∗- algebras such that A is contained in N . If K∗(B) is torsion free, then there exists a natural isomorphism α : K∗(A) ⊗ K∗(B) −→ K∗(A ⊗ B), that is, there exist a natural isomorphisms (K0(A) ⊗ K0(B)) ⊕ (K1(A) ⊗ K1(B)) −→ K0(A ⊗ B), (K0(A) ⊗ K1(B)) ⊕ (K1(A) ⊗ K0(B)) −→ K1(A ⊗ B). We introduce the concept of infinite tensor product of graded abelian group. Let G = G0 ⊕ G1 be an graded abelian group and G⊗n denote the tensor product of n copies of G (as graded Z-module). For an elements e of G0, an infinite tensor product (G, e)⊗ Z of G with respect to e is the inductive limit of the sequence {G⊗2n−1}n with the embeddings x 7→ e ⊗ x ⊗ e. We often denote (G, e)⊗ Z by G⊗ Z. Note that (G, e)⊗ Z is represented as the linear span of with the grading (G⊗ Z)0 ⊕ (G⊗ Z)1, where (G⊗ Z)p is the linear span of A shift λ on (G, e)⊗ Z is the automorphism of (G, e)⊗ Z satisfying xi : xi ∈ G, xj = e for all but finitely many j(cid:27) (cid:26)Ni∈Z (cid:26)Ni∈Z xi ∈ (G, e)⊗ Z : xi ∈ Gpi, pi ∈ Z2, Pi∈Z xi(cid:19) = Ni∈Z λ(cid:18)Ni∈Z for all Ni∈Z xi−1 pi = p(cid:27) . xi ∈ (G, e)⊗ Z. Remark 1. Let A be a unital C ∗-algebra in N , and assume that K∗(A) is a graded free abelian group. (i) Kunneth theorem gives that there exists a natural isomorphism αn from K∗(A)⊗n to K∗(A⊗n) for each n ≥ 1. Form the definition of α [S, p.446], there exists a natural isomorphism α∞ from (K∗(A), [1A]0)⊗ Z to K∗(A⊗ Z) such that the following diagram is commutative: K∗(A) K∗(A)⊗3 K∗(A)⊗5 · · · / K∗(A)⊗ Z (1) α1 α3 α5 α∞ K∗(A) / K∗(A⊗3) / K∗(A⊗5) / · · · / K∗(A⊗ Z). 2 / / / / / / / / / / / (ii) The automorphism λ := α−1 ∞ ◦ s∗ ◦ α∞ is the shift on (K∗(A), [1A]0)⊗ Z. (iii) Applying A⊗ Z ⋊s Z to Pimsner -- Voiculescu theorem [B, Theorem 10.2.1], we obtain the following exact triangle: K∗(A⊗ Z) id −s∗ / K∗(A⊗ Z) ∂ f▲▲▲▲▲▲ xrrrrrr ι∗ K∗(A⊗ Z ⋊s Z), where id −s∗ and ι∗ have degree 0 and ∂ has degree 1. Therefore, by (i) and (ii), we obtain the following exact triangle: K∗(A)⊗ Z id −λ / K∗(A)⊗ Z α−1 ∞ ◦∂ f▲▲▲▲▲▲ xrrrrrr K∗(A⊗ Z ⋊s Z). ι∗◦α∞ (2) Let G be a free abelian group which has a basis E containing e. Let G be the subgroup of G such that G = Z e ⊕ G. (3) Let G ⊗ G⊗ N denote the linear span of (cid:26)Ni∈Z xi ∈ G⊗ Z : x0 ∈ G, xi = e for all i < 0(cid:27) . Set H(G, e) := Z e⊗ Z + G ⊗ G⊗ N where e⊗ Z :=Ni∈Z e. Then, we prove in Lemma 3 that G⊗ Z = (id −λ)(G⊗ Z) ⊕ H(G, e), where λ is the shift on G⊗ Z, In particular, H(G, e) is naturally isomorphic to the cokernel of id −λ. Remark 2. (i) If G is an ordered free abelian group with the positive cone G+ and e ∈ G+, + , all sums of then (G, e)⊗ Z is also an ordered abelian group with the positive cone G⊗ Z the elements of Set H(G, e)+ := H(G, e) ∩ G⊗ Z + . (cid:26)Ni∈Z (ii) Let E⊗ Z denote the basis(cid:8)Ni∈Z e(i) xi ∈ (G, e)⊗ Z : xi ∈ G+(cid:27) . (G, e)⊗ Z. Then, there exists a natural isomorphism : e(i) ∈ E, e(j) = e for all but finitely many j(cid:9) of Φ(G, e, E) : (G, e)⊗ Z −→ Z⊕(E⊗ Z) . But, Φ(G, e, E) may not be order-preserving, where the positive cone of Z⊕I is Z⊕I {(km)m∈I ∈ Z⊕I : km ≥ 0} for a set I. + := 3 / x f / x f Example 2. (i) (Z, 1)⊗ Z = H(Z, 1) = Z 1⊗ Z. (ii) For a natural number d ≥ 2, let Ed be the standard basis {e(k)}d k=1 of Zd. Then, Φ(Zd, e(1), Ed) is order-preserving. On the other hand, let Ed be the basis {e(0)} ∩ k=2 where e(0) := (1, 1, · · · , 1) ∈ Zd. Then, Φ(Zd, e(0), Ed) is not order-preserving. {e(k)}d The following is our main theorem: Main Theorem. Let A be a C ∗-algebra in N and assume that K∗(A) is a free abelian group which has a basis containing [1A]0. Then there exists the following split exact sequence: 0 / H(K∗(A), [1A]0) ϕ / K∗(A⊗ Z ⋊s Z) ψ / Z [1A]⊗ Z 0 / 0, (4) where ϕ has degree 0 and ψ has degree 1. In particular, ϕ is the restriction of ι∗ ◦ α∞ and ψ is the homomorphism defined by x 7→ (α−1 ∞ ◦ ∂)(x). Corollary. Let A be a C ∗-algebra satisfying the assumption of the above theorem. If K1(A) = 0, then H(K0(A), [1A]0) ϕ / K0(A⊗ Z ⋊s Z), K1(A⊗ Z ⋊s Z) ψ / Z [1A]⊗ Z 0 are isomorphisms. Moreover, if (K0(A), K0(A)+) is ordered abelian group, then ϕ is order- preserving. Example 3. We calculate the K-theory for the group C ∗-algebra of the lamplighter group: Z2 ≀ Z := Z⊕ Z 2 ⋊ Z . Note that (i) C ∗(Z2 ≀ Z) ∼= (C2)⊗ Z ⋊s Z (Example 1), (ii) C2 ∈ N , (iii) (K0(C2), K0(C2)+, [1C2]0) ∼= (Z2, Z2 Applying the Main Theorem in the case of A = C2, we obtain the following isomorphisms: +, (1, 1)) and K1(C2) = 0. (K0(C ∗(Z2 ≀ Z)), K0(C ∗(Z2 ≀ Z))+) ∼= (H(Z2, (1, 1)), H(Z2, (1, 1))+), K1(C ∗(Z2 ≀ Z)) ∼= Z . 2 Proof In this section, let G be a graded free abelian group which has a basis E containing e. Set j=1 e ∈ G⊗n and G⊗0 := Z. Notice that, by (3), e⊗n :=Nn G⊗n = (e ⊗ G⊗n−1) ⊕ ( G ⊗ G⊗n−1) (5) for each n ≥ 1. 4 / / / / / / Lemma 1. For n ≥ 1, y ∈ G⊗n, k ∈ Z, and x ∈ G ⊗ G⊗n, if the equality y ⊗ e − e ⊗ y = ke⊗n+1 + x (6) holds, then y ∈ Z e⊗n, k = 0, and x = 0. Proof. Note that, by (5), y = e ⊗ y1 + x1 for some y1 ∈ G⊗n−1 and x1 ∈ G ⊗ G⊗n−1. Then, from (6), we get the following: e ⊗ (y1 ⊗ e − y − ke⊗n) + (x1 ⊗ e − x) = 0. Hence, by (5), y1 ⊗ e − y − ke⊗n = 0 and x1 ⊗ e − x = 0, so y = y1 ⊗ e − ke⊗n, x = x1 ⊗ e. Therefore, from (6), we get (y1 ⊗ e − e ⊗ y1) ⊗ e = (ke⊗n + x1) ⊗ e, so y1 ⊗ e − e ⊗ y1 = ke⊗n + x1. Note that, replacing (n − 1, y1, x1) of this equation with (n, y, x), we get (6). Repeating this argument, we can find yj ∈ G⊗n−j for each integer j (1 ≤ j ≤ n) such that yj−1 = yj ⊗ e − ke⊗n−j+1, where y0 := y. Inductively, we obtain y = (yn −nk)e⊗n ∈ Z e⊗n. Hence, from (6), e⊗(ke⊗n)+ x = 0. Thus, by (5), k = 0 and x = 0. For each n ≥ 1, identify G⊗2n−1 and the linear span of (cid:26)Ni∈Z xi ∈ G⊗ Z : xi = e for all i > 2n − 1(cid:27) . Lemma 2. Let λ be the shift on G⊗ Z. For z ∈ G⊗ Z and w ∈ H(G, e), if the equality (id −λ)(z) = w (7) holds, then z ∈ Z e⊗ Z and w = 0. Proof. Take n ≥ 1, satisfying (id −λ)(z) (= w) ∈ G⊗2n+1. Note that (id −λ)(G⊗ Z) ∩ G⊗2n+1 = {e ⊗ z′ − z′ ⊗ e : z′ ∈ G⊗2n}, H(G, e) ∩ G⊗2n+1 = Z e⊗2n+1 + e⊗n ⊗ G ⊗ G⊗n. Hence, (id −λ)(z) = e ⊗ z′ − z′ ⊗ e and w = ke⊗2n+1 + e⊗n ⊗ x for some z′ ∈ G⊗2n, k ∈ Z, and x ∈ G ⊗ G⊗n. Therefore, by (7), Inductively, this equality follows that z′ = e⊗n ⊗ y for some y ∈ G⊗n. Hence z′ ⊗ e − e ⊗ z′ = ke⊗n+1 + e⊗n ⊗ x. y ⊗ e − e ⊗ y = ke⊗n+1 + x. By Lemma 1, we obtain y ∈ Z e⊗n, k = 0, and x = 0. Thus, z ∈ Z e⊗ Z and w = 0. 5 Lemma 3. Let λ be the shift on G⊗ Z. Then, (i) Ker(id −λ) = Z e⊗ Z, (ii) G⊗ Z = (id −λ)(G⊗ Z) ⊕ H(G, e). Proof. (i) At first, it is clear that Ker(id −λ) ⊃ Z e⊗ Z. Conversely, take z ∈ G⊗ Z such that (id −λ)(z) = 0. Applying Lemma 2 to the case of w = 0, we obtain that z ∈ Z e⊗ Z. Thus Ker(id −λ) = Z e⊗ Z. (ii) By Lemma 1, (id −λ)(G⊗ Z)∩H(G, e) = 0, so it is sufficient to show E⊗ Z ⊂ (id −λ)(G⊗ Z)⊕ H(G, e). Take f = Ni∈Z ei ∈ E⊗ Z. Since e⊗ Z ∈ H(G, e), we may assume f 6= e⊗ Z. Then, there exists an integer i0 ∈ Z such that ei0 6= e and ei = e for all i < i0. If i0 = 0, then f ∈ H(G, e), so we may assume i0 6= 0. Note that λ−i0(f ) ∈ G ⊗ G⊗ N and f =   i0 λ−j(f )! + λ−i0(f ) (id −λ) − Pj=1 (id −λ) −i0 λ−j+1(f )! + λ−i0(f ) Pj=1 for i0 > 0 for i0 < 0, Thus, f ∈ (id −λ)(G⊗ Z) ⊕ H(G, e). Proof of the Main Theorem. Applying Lemma 3 in the case of G = K∗(A) and e = [1A]0, we obtain Ker(id −λ) = Z [1A]⊗ Z 0 , K∗(A)⊗ Z = (id −λ)(K∗(A)⊗ Z) ⊕ H(K∗(A), [1A]0). Therefore, from the exactness of (2), the sequence (4) is exact. Moreover, since Z [1A]⊗ Z (4) is split. 0 ∼= Z, Acknowlegement: I would like to thank Tsuyoshi Kato for introducing to this subject and for encouraging discusses. References [B] B. Blackader, K-theory for Operator Algebras, Springer-Verlag, New York, Berlin, Hei- delberg, London, Paris, Tokyo, 1896. [S] C. Schochet, Topological methods for C ∗-algebras II: geometric homology, Pacific J. Math 114 (1984), 447 -- 468. 6
1006.5397	3	1006	2012-08-30T13:15:49	A simple, monotracial, stably projectionless C*-algebra	[ "math.OA" ]	We construct a simple, nuclear, stably projectionless C-algebra W which has trivial K-theory and a unique tracial state, and we investigate the extent to which W might fit into the hierarchy of strongly self-absorbing C-algebras as an analogue of the Cuntz algebra O_2. In this context, we show that every nondegenerate endomorphism of W is approximately inner and we construct a trace-preserving embedding of W into the central sequences algebra M(W)_\infty \cap W'. We conjecture that W\otimes W is isomorphic to W and we note some implications of this, for example that W would be absorbed tensorially by a certain class of nuclear, stably projectionless C*-algebras. Finally, we explain why W may play some role in the classification of such algebras.	math.OA	math	A SIMPLE, MONOTRACIAL, STABLY PROJECTIONLESS C∗-ALGEBRA BHISHAN JACELON Abstract. We construct a simple, nuclear, stably projectionless C ∗-algebra W which has trivial K- theory and a unique tracial state, and we investigate the extent to which W might fit into the hierarchy of strongly self-absorbing C ∗-algebras as an analogue of the Cuntz algebra O2. In this context, we show that every nondegenerate endomorphism of W is approximately inner and we construct a trace- preserving embedding of W into the central sequences algebra M (W )∞ ∩ W ′. We conjecture that W ⊗W ∼= W and we note some implications of this, for example that W would be absorbed tensorially by a certain class of nuclear, stably projectionless C ∗-algebras. Finally, we explain why W may play some role in the classification of such algebras. 1. Introduction In the study of operator algebras, classification has always been a central theme. The classification of nuclear C∗-algebras was initiated by Elliott, who used ordered K-theory to classify approximately finite dimensional (AF) C∗-algebras in [Ell76b] (building on earlier work of Glimm [Gli60] and Bratteli [Bra72]) and approximately circle (AT) algebras of real rank zero in [Ell93]. Around 1990, Elliott conjectured that the class of all separable, nuclear C∗-algebras could be classified by an invariant Ell(·) based on K-theory. Augmenting the invariant to include tracial data (and the natural pairing between traces and K-theory), this has come to be known as the Elliott Conjecture, and the project to establish its veracity, the Elliott Programme. The survey article [ET08] gives an excellent overview of the history and recent developments of the Programme, and a more detailed exposition can be found in Rørdam's monograph [Rør02]. Today, it is known that the Elliott Conjecture does not hold in full generality. The ideas of [Vil98] were used by Rørdam [Rør03] and Toms [Tom08] to construct simple, separable, nuclear non-isomorphic C∗-algebras with the same Elliott invariant. These counterexamples to the Conjecture can be dealt with in two ways: (1) enlarge the invariant to include the Cuntz semigroup (which is sensitive enough to be able to distinguish between the counterexamples of Rørdam and Toms, and, for well-behaved C∗-algebras, can be recovered functorially from K-theory and traces -- see [BPT08]), or (2) impose further regularity conditions on the C∗-algebras to be classified. This article is in the spirit of option (2), and the relevant regularity property is Z-stability. The Jiang-Su algebra Z is a simple, separable, nuclear, infinite dimensional, projectionless C∗- algebra which has the same Elliott invariant as C (see [JS99] and also [RW09] for some alternative descriptions of Z). A C∗-algebra A is 'Z-stable' if A ⊗ Z ∼= A; the counterexamples of Rørdam and Toms each involve a pair of non-isomorphic C∗-algebras which have the same Elliott invariant, but one of which is not Z-stable. Thus the class of Z-stable C∗-algebras is the largest class in which the Elliott Conjecture can be expected to hold. The term 'strongly self-absorbing' was coined by Toms and Winter in [TW07] to describe a handful of algebras which, like Z, have played pivotal roles in the classification programme. Specifically, a separable unital C∗-algebra D 6= C is strongly self-absorbing if there is an isomorphism ϕ : D → D ⊗ D such that ϕ is approximately unitarily equivalent to the first factor embedding idD ⊗ 1D (written ϕ ∼a.u. idD ⊗ 1D), which means that there is a sequence of unitaries (un)∞ n=1 in D ⊗ D such that ϕ(a) = limn→∞ un(a ⊗ 1)u∗ n for every a ∈ D. As for Z, a C∗-algebra A is 'D-stable' if A ⊗ D ∼= A. Date: October 27, 2018. 2000 Mathematics Subject Classification. 46L35, 46L05. Key words and phrases. Strongly self-absorbing C ∗-algebra, stably projectionless C ∗-algebra, classification. Research supported by the ERC through AdG 267079. 1 2 BHISHAN JACELON If A is strongly self-absorbing then A is simple and nuclear, and is either purely infinite or stably finite with a unique tracial state. Moreover, the list of known strongly self-absorbing algebras is short (and closed under ⊗): the Cuntz algebras O2 and O∞, UHF algebras of infinite type Mp∞ (such as the CAR algebra M2∞ ), O∞ ⊗ Mp∞, and the Jiang-Su algebra Z. One of the reasons that these algebras are important is that they provide localized versions of the Elliott Conjecture in the sense of [Win07], i.e. theorems of the form Ell(A) ∼= Ell(B) =⇒ A ⊗ D ∼= B ⊗ D. For example, the classification of Kirchberg algebras ([Kir], [Phi00]) can be interpreted as the classification up to O∞-stability of all simple, separable, unital, nuclear C∗-algebras that satisfy the UCT. At the other end of the spectrum, Winter proves in [Win07] and [Win10] that the Elliott Conjecture holds for a large class of stably finite Z-stable C∗-algebras (which have finite decomposition rank, as defined in [KW04]). The known strongly self-absorbing algebras form a hierarchy, with Z at the bottom (every strongly self-absorbing algebra is Z-stable -- see [Win09]) and O2 at the top (O2 absorbs every strongly self- absorbing algebra -- see [KP00]). Every purely infinite algebra in this hierarchy has a stably finite analogue (most notably, Z corresponds to O∞), except for O2. That is, there is no stably finite strongly self-absorbing C∗-algebra A with K∗(A) = 0. In fact, it is not hard to see that if A is separable and stably finite with K0(A) = 0 then A ⊗ K cannot have any full projections; in particular, A must be nonunital (in fact, stably projectionless if A is also simple). This explains the gap since the definition of 'strongly self-absorbing' conspicuously involves a unit. What is therefore needed to fill this gap is a well-behaved notion of 'strongly self-absorbing' that makes sense for nonunital C∗-algebras and agrees with the existing definition for unital algebras. There are a few equivalent characterizations of strongly self-absorbing C∗-algebras, some more amenable than others to a sensible interpretation in the nonunital case. The above definition is obvi- ously troublesome and highlights a general problem: If A is a nonunital C∗-algebra without projections then there are no obvious ∗-homomorphisms from A to A ⊗ A; in particular, there is no obvious way of making sense of an infinite tensor product A⊗∞. On the other hand, if A 6= C is separable and unital, then A is strongly self-absorbing if and only if (with some redundancy) (i) A ∼= A ⊗ A; (ii) every unital endomorphism of A is approximately inner; (iii) A admits an asymptotically central sequence of unital endomorphisms (see section 6). These conditions also make sense if A is nonunital (provided that we replace 'unital' by 'nondegen- erate' where appropriate) and we could think of taking some subset of these as a general definition of strongly self-absorbing. It is not clear how well-behaved such a definition would be, but the goal of this paper is to find a stably projectionless C∗-algebra W with trivial K-theory and a unique tracial state, and at least see how far these properties hold for W . The hope is that W will play a role in the classification of stably projectionless C∗-algebras, similar to the roles played by O∞ and Z in the classification of purely infinite and stably finite algebras respectively. Indeed, W itself arises from a class of stably projectionless algebras which have been completely classified (by Razak [Raz02, Theorem 1.1] and Tsang [Tsa05, Theorem 5.1]). These algebras are inductive limits of building blocks of the form A(n, n′) := {f ∈ C([0, 1], Mn′) : f (0) = diag( c, . . . , c, 0n) and f (1) = diag( c, . . . , c), c ∈ Mn}, where n and n′ are natural numbers with nn′ and a := n′ projectionless and has trivial K-theory, so the Elliott invariant is purely tracial. n − 1 > 0. Each building block is stably a z } { a+1 z } { Theorem 1.1 (Razak). Let A and B be simple inductive limits of (countably many) building blocks. If (T +A, ΣA) is isomorphic to (T +B, ΣB) then A is isomorphic to B. Here, T +A is the cone of densely defined lower semicontinuous traces on A and ΣA is the compact convex subset of T +A consisting of those (bounded) traces of norm ≤ 1. We will call an inductive limit of countably many building blocks a Razak algebra. (We may always assume that the connecting maps are injective -- see section 2 below.) A SIMPLE, MONOTRACIAL, STABLY PROJECTIONLESS C ∗-ALGEBRA 3 Theorem 1.1 has been superseded by the main result of [Rob10]. There, a functor based on the Cuntz semigroup is used to classify (not necessarily simple) inductive limits of one-dimensional NCCW- complexes with trivial K1-groups. In the case of simple C∗-algebras with trivial K-theory, this reduces to the following (see [Rob10, Corollary 6.2.4]). Theorem 1.2 (Robert). Let A and B be inductive limits of one-dimensional NCCW-complexes which have trivial K1-groups, such that A and B are simple, nonunital and have trivial K-theory. If (T +A, ΣA) is isomorphic to (T +B, ΣB) then A is isomorphic to B. For the range of the invariant, we still refer to [Tsa05] (although an alternative proof will be provided in section 5 of this article). Theorem 1.3 (Tsang). Let C be a topological convex cone that has a metrizable Choquet simplex as a base and let ω be a faithful lower semicontinuous linear map C → [0, ∞]. Then there exists a simple Razak algebra A such that (T +A, ΣA) = (C, ω−1([0, 1])). Suppose that C = R+ and let ω(x) = x. By Theorems 1.3 and 1.1, there exists a unique simple Razak algebra W which has a tracial state τ such that T +W = R+τ . That is, W has a unique tracial state and every trace on W is bounded. Note that W ⊗ K is also a simple Razak algebra whose trace is unique up to scalar multiples, except that in this case the trace is unbounded. (This corresponds to taking ω(x) = 0 if x = 0 and ω(x) = ∞ otherwise.) Again, Theorem 1.1 says that W ⊗ K is the unique simple Razak algebra with this property. It is easy to find a simple, nuclear, nonunital C∗-algebra with a (unique) tracial state (for example, any proper hereditary subalgebra A of Z has a tracial state, which is unique since A is stably isomorphic to Z), but to the author's knowledge, W may be the first example of such an algebra which is stably projectionless. In particular, W lives outside of the realm of C∗-algebras for which projections separate traces. Note that Theorem 1.1 also shows that W absorbs the universal UHF algebra Q, which implies that W absorbs Z (see also section 6), and the Kirchberg-Phillips classification theorem ([Kir], [Phi00]) shows that W ⊗ A ∼= O2 ⊗ K for any simple, separable, nuclear, stably infinite C∗-algebra A that satisfies the UCT (see Theorem 8.4.1 and also Theorems 4.1.10 and 4.1.3 of [Rør02]). In particular, W ⊗ O2 ∼= W ⊗ O∞ ∼= O2 ⊗ K. Finally, it should be noted that [Rob10] shows that W also arises from certain crossed products of O2 by R. For λ ∈ R, consider the action α of R on O2 = C∗(s1, s2) given by αt(s1) = eits1, αt(s2) = eλits2. These actions, and the associated crossed products O2 ⋊λ R, have been studied by many authors including Evans, Kishimoto and Kumjian (see [Eva80], [KK96] and [KK97]). Kishimoto and Kumjian proved that O2 ⋊λ R is simple if and only if λ is irrational; in this case O2 ⋊λ R is stable, and is purely infinite if λ < 0 and projectionless with a unique (unbounded) trace if λ > 0. Moreover, Dean proves in [Dea01] that for generic positive irrational λ, O2 ⋊λ R can be written as a countable inductive limit of certain one-dimensional subhomogeneous C∗-algebras called NCCW complexes (see sections 2 and 4 below), which are shown in [Rob10] to have trivial K1-groups. Since, by [Con81, Theorem IV.2], these crossed products have trivial K-theory, we can apply the Kirchberg-Phillips classification theorem in the former case, and the classification theorem of [Rob10] in the latter, to deduce that O2 ⋊λ R ∼=(cid:26) O2 ⊗ K for every λ ∈ R−\Q W ⊗ K for generic λ ∈ R+\Q. The paper is organized as follows. We first recall some basic facts about building blocks and establish notation in section 2. In section 3 we will explicitly exhibit W in a manner analogous to the construction of the Jiang-Su algebra Z. Next, in section 4 we characterize W as the unique terminal object in a certain category, from which it will follow that every nondegenerate endomorphism of W In section 5, we make some remarks on the conjecture that W ⊗ W ∼= W is approximately inner. (an incorrect proof of which was offered in an earlier version of this article). We also prove here that, if this conjecture were to be true (and the Elliott Conjecture certainly predicts that it is), then 4 BHISHAN JACELON among the C∗-algebras classified in [Rob10], those that are simple and have trivial K-theory would all absorb W tensorially. Section 6 contains a complete proof of [TW05, Proposition 4.1], which says that every simple Razak algebra is approximately divisible, and we show how the proof can be adapted to construct a trace-preserving embedding of W into the central sequences algebra M (W )∞ ∩ W ′ (where M (W ) is the multiplier algebra of W ). Acknowledgements. This work was part of my doctoral research at the University of Glasgow. I am indebted to: my supervisor Simon Wassermann, and also Stuart White, Rob Archbold and Ulrich Kraehmer, for carefully reading earlier versions of the manuscript; Wilhelm Winter for suggesting the topic and for many illuminating discussions; and Andrew Toms and George Elliott for their very helpful correspondence. I am also enormously grateful to Luis Santiago and Leonel Robert for helping to uncover the error in the attempted proof of Conjecture 5.1. Throughout this section, let A be the building block 2. The building blocks A(n, n′) := {f ∈ C([0, 1], Mn′) : f (0) = diag( c, · · · , c, 0n) and f (1) = diag( c, · · · , c), c ∈ Mn}, where n and n′ are natural numbers with nn′ and a := n′ dimensional NCCW-complex, i.e. a pullback of the form n − 1 > 0. Each building block is a one- a z } { a+1 z } { A Mn C[0, 1] ⊗ Mn′ / Mn′ ⊕ Mn′ ∂ (see [ELP98] and also section 4 below), is stably projectionless (because of the multiplicity differences at the endpoints), and has trivial K-theory (which can be seen from the Mayer-Vietoris sequence of [Sch84]). We will call an inductive limit of countably many building blocks a Razak algebra. Razak algebras are separable, nuclear, satisfy the UCT and are completely classified by tracial data when simple (Theorems 1.1 and 1.3). We now establish some basic notation that will be used throughout the paper. Irreducible representations. Every ideal I of A is of the form I = {f ∈ A : f Γ(I) = 0} for some unique closed subset Γ(I) ⊆ T = [0, 1]/{0, 1}. In particular, the primitive ideals of A are precisely those of the form Ix = {f ∈ A : f (x) = 0} for x ∈ [0, 1], and we can describe all the irreducible representations of A. They are (up to unitary equivalence) the evaluation maps evs : A → Mn′ for s ∈ (0, 1) together with 'evaluation at the irreducible fibre at infinity', i.e. ev∞ : A → Mn is such that ev0 =La i=1 ev∞. Connecting maps. Given the above characterization of ideals of A, it is easy to see that any proper quotient of A has nontrivial projections. Therefore, if B is another projectionless C∗-algebra then every nonzero ∗-homomorphism ϕ : A → B is injective. (This means that we may always assume that Razak algebras have injective connecting maps.) If B is also a building block then for each x ∈ [0, 1] we will denote by ϕx : A → Bx the morphism ϕx(f ) = ϕ(f )(x). Traces. We write T +A for the cone of densely defined lower semicontinuous traces on A. Every trace on the building block A is bounded and has the form τ = tr ⊗ µ, where tr is the normalized trace on Mn′ and µ is some positive Borel measure on (0, 1]. We will write T + 1 A for the simplex of tracial states on A (corresponding to Borel probability measures on (0, 1]), ΣA for those traces of norm at most one, and we let Aff0T +A denote the ordered vector space of continuous real-valued linear functionals on the cone T +A. Following [Raz02], we equip Aff0T +A with two different norms. First, we define k · kA by kf kA := sup{f (τ ) : τ ∈ ΣA}. Second, we use the order unit norm given by the following: fix η ∈ Aff0T +A with inf{η(τ ) : kτ k = 1} > 0, so that η is an order unit of Aff0T +A, and denote by Ση the closed / / / A SIMPLE, MONOTRACIAL, STABLY PROJECTIONLESS C ∗-ALGEBRA 5 convex set {τ ∈ T +A : η(τ ) = 1}; we then get a corresponding order unit norm k · kη given by kf kη := sup{f (τ ) : τ ∈ Ση}. It is not hard to show that the norms k · kA and k · kη are equivalent for the building block A, so in particular Ση is a compact base of T +A. In fact, we can say exactly what the spaces (Aff0T +A, k · kA) and (Aff0T +A, k · kη) look like. (i) Define C[0, 1]a := {f ∈ C[0, 1] : f (0) = a a+1 f (1)}. Then there is an isometric isomorphism of ordered Banach spaces ι : (Aff0T +A, k · kA) → C[0, 1]a, ι(f )(t) = f (tr ⊗ δt) (where δt denotes the point mass at t), which moreover preserves infima: inf{f (τ ) : τ ∈ T + 1 A} = inf ι(f ) for every f ∈ Aff0T +A. (ii) There is an isomorphism of order unit spaces ι′ η : (Aff0T +A, η) → (CR(T), 1) given by ι′ η(f ) = ι(f ) ι(η) , which again preserves infima: inf{f (τ ) : τ ∈ Ση} = inf ι′ η(f ) for f ∈ Aff0T +A. Finally, there is an embedding ψA of Aff0T +A into the set Asa of self-adjoint elements of A given by ψA(f )(t) = a + 1 a + t   a f (tr ⊗ δt)1n ⊕ · · · ⊕ f (tr ⊗ δt)1n ⊕tf (tr ⊗ δt)1n z  , } { which is right-inverse to the usual map ρA : Asa → Aff0T +A. That is, ψA satisfies τ (ψA(f )) = f (τ ) for every τ ∈ T +A. The embedding is natural in the sense that if B is another building block and ϕ : A → B is a ∗-homomorphism then τ (ϕ ◦ ψA(f )) = ϕ∗τ (ψA(f )) = f (ϕ∗τ ) = ϕ∗(f )(τ ) = τ (ψB ◦ ϕ∗(f )) for every τ ∈ T +B and f ∈ Aff0T +A. This justifies suppressing the notation ψA and ρA, and in the sequel we will do so without comment (particularly in section 4). 3. The construction of W In this section we construct explicit connecting maps for W by adapting the procedure of [JS99, Proposition 2.5]. Proposition 3.1. There exists an inductive sequence (Ai, ϕi) of building blocks Ai = A(ni, (ai + 1)ni) such that each connecting map ϕij : Ai → Aj is a ∗-homomorphism of the form f ◦ ξ1 f ◦ ξ2 ϕij (f ) = u     u∗ . . . f ◦ ξm (3.1) for some unitary u ∈ C([0, 1], Mn′ j ) and continuous maps ξk : [0, 1] → [0, 1] that satisfy ξk(x) − ξk(y) ≤ (1/2)j−i for every x, y ∈ [0, 1] and 1 ≤ k ≤ m; (3.2) m[k=1 ξk([0, 1]) = [0, 1]. (3.3) Proof. Let A1 be some building block A(n1, (a + 1)n1). We will find a building block A2 = A(n2, (b + 1)n2) and an injective ∗-homomorphism ϕ : A1 → A2 of the form (3.1) where each ξk is one of the maps ξ(x) = x/2, ξ(x) ≡ 1/2 or ξ(x) = (x + 1)/2. (3.4) Repeating this process then gives an inductive sequence where each ϕi : Ai → Ai+1 has the right form. Note also that as defined, ϕi(f ) makes sense for any f ∈ C([0, 1], Mn′ ), so ϕi extends to a unital ∗-homomorphism C([0, 1], Mn′ ) and in particular ϕi(u) is unitary whenever u is. ) → C([0, 1], Mn′ i i i+1 6 BHISHAN JACELON Therefore, each connecting map ϕij : Ai → Aj will be of the form (3.1) with each ξk a composition of j − i functions from the list (3.4), so will be one of the maps ξ(x) = l 2j−i or ξ(x) = x + l 2j−i , for some integer l with 0 < l < 2j−i in the former case and 0 ≤ l < 2j−i in the latter. Hence (3.2) is satisfied. Let b = 2a + 1, n2 = bn1 and m = 2b. Let f ∈ A1, so that f (0) = diag( c, · · · , c, 0n1) and c, · · · , c) (in M(a+1)n1) for some c ∈ Mn1. Write df := diag(f (1/2), c, · · · , c) ∈ Mn2. f (1) = diag( Then, in M(b+1)n2 , the matrix a+1 z } { df ⊗ 1b =   df . . . df 0n2   consists (up to permutation) of ab copies of c, b copies of f (1/2), and a zero matrix of size n2. On the other hand, the matrix diag( f (0), . . . , f (0), f (1/2), . . . , f (1/2)) ∈ Mm(a+1)n1 = M(b+1)n2 z b } { z b } { also consists of ab copies of c, b copies of f (1/2), and a zero matrix of size bn1 = n2. Therefore, there is a permutation unitary u0 ∈ M(b+1)n2 such that df . . .     = u0 df 0n2 f (0) . . . f (0) f (1/2) . . . f (1/2) f (1), . . . , f (1)) and df ⊗1b+1 Similarly, again in M(b+1)n2, both of the matrices diag( consist up to permutation of a(b + 1) copies of c and b + 1 copies of f (1/2). Therefore, there is a permutation unitary u1 ∈ M(b+1)n2 such that f (1/2), . . . , f (1/2), z b+1 } { z b−1 } df df     = u1 . . . df f (1/2) . . . f (1/2) f (1) . . . f (1) Now we just connect the endpoints: take u to be any continuous path of unitaries in M(b+1)n2 from u0 to u1, and define functions ξ1, . . . , ξm : [0, 1] → [0, 1] by ξk(x) =  x/2 1/2 (x + 1)/2 if 1 ≤ k ≤ b, if if k = b + 1, b + 1 < k ≤ m. Then the map ϕ as defined in (3.1) is by construction a ∗-homomorphism from A1 to A2. Finally, note that ϕij is injective if and only if condition (3.3) holds (since ϕij (f ) = 0 if and only if k=1 ξk[0, 1]). By construction, (3.3) holds for the ξk used to define ϕ above, and it therefore follows that for every i and j, ϕij is injective and satisfies (3.3). (cid:3) f ∈ Ai is supported on the open set [0, 1]\Sm a a z } { z } {   {   u∗ 0. u∗ 1.     A SIMPLE, MONOTRACIAL, STABLY PROJECTIONLESS C ∗-ALGEBRA 7 Remark 3.2. For the above inductive sequence we have ai → ∞ (and also ni/ai = ni−1 → ∞) as i → ∞; we will make use of this in section 4. Lemma 3.3. If (Ai, ϕi) is any inductive sequence as in Proposition 3.1 then the connecting maps ϕij are nondegenerate. In particular, ϕ∗ 1 (Ai) for every j ≥ i. Moreover, if A = lim−→(Ai, ϕi) then every τ ∈ T +A is bounded. 1 (Aj )) ⊆ T + ij (T + Proof. Here, 'nondegenerate' means that an, and hence every, approximate unit of Ai is mapped to an z a } { i 1 ⊕ · · · ⊕ 1 ⊕t) ⊗ 1n1. approximate unit of Aj. Let h ∈ A1 be the canonical self-adjoint element h(t) = ( We claim that for every i ≥ 1, hi := ϕ1i(h) is strictly positive in Ai, which is equivalent to saying that (h1/n )n∈N is an approximate unit for Ai. To make the analysis easier, we may assume by induction that i = 2. To prove the Lemma, let f ∈ Ai and let ǫ > 0, and for convenience write p = 1n′ i−ni and . Certainly we have khik = 1, and we may assume that kf k = 1 as well. Choose δ > 0 such q = 1n′ that if 0 ≤ t < δ then kf (t) − f (0)k < ǫ/5 and kut − u0k < ǫ/5 (where u is as in (3.1) of Proposition 3.1). It is easy to see that as n → ∞, h1/n converges locally uniformly to q on (0, 1]. Hence we can find some N such that kh1/n (t) − qk < ǫ for every δ ≤ t ≤ 1 and n ≥ N . For δ ≤ t ≤ 1 we therefore have (0) converges to p and h1/n (0) − pk < ǫ/5 and kh1/n i i i i i kh1/n i (t)f (t) − f (t)k = k(h1/n i (t) − q)f (t)k < ǫ ∀n ≥ N. t h1/n Now let 0 ≤ t < δ and write gn(t) = u0u∗ are increasing, we have kpgn(t)p − pk ≤ kh1/n i i 0. Then gn(t) commutes with p, and since the ξk (t)utu∗ (0) − pk < ǫ/5 for n ≥ N . Thus kh1/n i (t)f (t) − f (t)k ≤ kh1/n i (t)f (t) − h1/n i (t)f (0)k + kh1/n i (t)f (0) − gn(t)f (0)k + kgn(t)f (0) − pgn(t)pf (0)k + k(pgn(t)p − p)f (0)k + kf (0) − f (t)k < (ǫ + 2ǫ + 0 + ǫ + ǫ)/5 = ǫ for 0 ≤ t < δ and n ≥ N . Therefore, kh1/n approximate unit for Ai. It follows that the connecting maps ϕij are nondegenerate. f − f k < ǫ for every n ≥ N and hence (h1/n i i )n∈N is an Now suppose that j ≥ i and let ρ be a state on Aj . Then ρ ◦ ϕij is a positive linear functional on Ai and moreover kρ ◦ ϕij k = lim n→∞ ρ ◦ ϕij (h1/n i ) = lim n→∞ ρ(ϕij (hi)1/n) = lim n→∞ ρ(h1/n j ) = kρk = 1, so ϕ∗ ijρ = ρ ◦ ϕij ∈ S(Ai). Finally, let τ ∈ T +A, which we identify with lim←−(T +Ai, ϕ∗ bounded trace ϕ∗ i∞τ on Ai, and we have i ). Then for every i, τ restricts to a kϕ∗ i∞τ k = lim n→∞ (ϕ∗ i∞τ )(h1/n i ) = lim n→∞ τ (ϕi∞(hi)1/n) = lim n→∞ τ (ϕ1∞(h)1/n) = kϕ∗ 1∞τ k. Hence τ is bounded. (cid:3) To show that A is simple, we use the following well-known lemma (see [EGJS97, §4] for a unital version). Lemma 3.4. Let A = lim−→(Ai, ϕij ) be an inductive limit of building blocks. Then A is simple if and only if for every i ∈ N and every nonzero element a of Ai, the image ϕij (a) generates Aj as a closed two-sided ideal for all but finitely many j ≥ i. Proposition 3.5. Let A = lim−→(Ai, ϕi) for any inductive sequence (Ai, ϕi) as in Proposition 3.1. Then A is simple and has a unique tracial state. 8 BHISHAN JACELON Proof. If f is a nonzero element of Ai then there is an interval U ⊂ [0, 1] on which f is nonzero. By (3.2) and (3.3) of Proposition 3.1, if j ≥ i is large enough then there is some 1 ≤ k ≤ m such that ξk([0, 1]) ⊂ U . Then f ◦ ξk, and hence ϕij (f ), is nonzero on all of [0, 1], and so ϕij (f ) generates Aj as a closed two-sided ideal. Lemma 3.4 therefore implies that A is simple. Next, note that A has a nonzero trace: the cone T +A has a compact base Σ which can be written as the inverse limit of bases Σi of T +Ai; since these are nonempty compact Hausdorff spaces it follows that Σ is nonempty (and does not contain zero since it is a base). By Lemma 3.3, this trace is bounded and we now show that it is unique. Let f ∈ Ai and ǫ > 0 be given and choose δ > 0 such that kf (y) − f (z)k ≤ ǫ/2 whenever y − z ≤ δ. Then provided 2j−i > 1/δ, it follows from (3.2) that kf (ξk(x)) − f (ξk(y))k ≤ ǫ/2 for every x, y ∈ [0, 1] and 1 ≤ k ≤ m. It follows that for all sufficiently large j, every τ = tr ⊗ µ ∈ T + 1 Aj and for fixed y ∈ (0, 1], we have τ (ϕij (f )) − tr ⊗ δy(ϕij (f )) =(cid:12)(cid:12)(cid:12)(cid:12) Z tr(ϕij (f )(x) − ϕij (f )(y))dµ(x)(cid:12)(cid:12)(cid:12)(cid:12) ≤ ǫ/2 τj,1(ϕij (f )) − τj,2(ϕij (f )) ≤ ǫ for every τj,1, τj,2 ∈ T + 1 Aj. and so (3.5) (cid:3) Hence A has at most one tracial state. As in the introduction, we will denote the unique such inductive limit by W and its unique tracial state by τ . 4. A categorical description of W In this section we characterize (W, τ ) as a terminal object (Theorem 4.5), and use this description to prove that every trace-preserving endomorphism of W is approximately inner (Corollary 4.6), and that every simple Razak algebra embeds into W ⊗ K (Corollary 4.7). We accomplish this via the following adaptation of [Rør04, Theorem 2.1]. Lemma 4.1. Let B be a building block and let τ be the unique tracial state on W . (i) For every faithful trace τ0 on B with kτ0k ≤ 1 there exists a ∗-homomorphism ψ : B → W such that τ ◦ ψ = τ0. (ii) Two ∗-homomorphisms ψ1, ψ2 : B → W are approximately unitarily equivalent if and only if τ ◦ ψ1 = τ ◦ ψ2. To prove this, we need to use Razak's local existence and local uniqueness theorems, which appear as [Raz02, Theorem 3.1] and [Raz02, Theorem 4.1] respectively. We restate them here for convenience, referring to section 2 for notation. Proposition 4.2 (Local existence). Let B be a building block, and fix some finite subset F ⊂ Aff0T +B and some ǫ > 0. Then there is a natural number N and some η ∈ Aff0T +B with kηkB ≤ 1 and inf ι(η) ≥ 1/2 such that the following property holds. For any building block A = A(n, (a + 1)n) and contractive positive linear map ξ : (Aff0T +B, k · kB) → (Aff0T +A, k · kA), if n ≥ N a/ inf ι(ξ(η)) then there is a ∗-homomorphism ψ : B → A with kξ(f ) − ψ∗(f )kξ(η) < ǫ for every f ∈ F . Proposition 4.3 (Local uniqueness). Let B be a building block and let h be the canonical self-adjoint element of B (as in Lemma 3.3). Fix a finite subset F ⊂ B and a tolerance ǫ > 0. Then there exists a natural number M and two families of positive functions {ζj}M j=1 in the unit ball of (Aff0T +B, k · kB) such that for any building block A and any two ∗-homomorphisms ϕ, ψ : B → A that satisfy j=1, {σj}M (i) ϕ∗(ζj)(τ ) > m, ψ∗(ζj )(τ ) > m, and ϕ∗(σj )(τ ) − ψ∗(σj )(τ ) < m for some m > 0 and every τ ∈ T + 1 A and 1 ≤ j ≤ M ; (ii) ϕt(h) and ψt(h) have at least three distinct eigenvalues for every t ∈ [0, 1]; there exists a unitary u ∈ eA such that kϕ(f ) − uψ(f )u∗k < ǫ for every f ∈ F . Typically, the eigenvalue condition (ii) of Proposition 4.3 is handled by the following standard consequence of Lemma 3.4. A SIMPLE, MONOTRACIAL, STABLY PROJECTIONLESS C ∗-ALGEBRA 9 z { a } Lemma 4.4 (δ-density). Let A = lim−→(Ai, ϕij) be a simple inductive limit of building blocks Ai = A(ni, (ai + 1)ni), and let h ∈ A1 be the canonical self-adjoint element h(t) = ( 1 ⊕ · · · ⊕ 1 ⊕t) ⊗ 1n1 . Then for every δ > 0, there exists an integer N such that for every j ≥ N and every x ∈ [0, 1], the eigenvalues of ϕx 1j (h) are δ-dense in [0, 1]. Note that Lemma 4.4 implies that ni → ∞ as i → ∞ for any simple inductive limit of building blocks Ai = A(ni, (ai + 1)ni). Of course, we already know this for W by construction; either way, this will allow us to deal with the hypothesis of Proposition 4.2 (see also Remark 3.2). We also need to use the fact that, since the building blocks are (nonunital) one-dimensional NCCW complexes, they are semiprojective, and hence can be finitely presented with stable relations (see [ELP98]). We can therefore use [Lor93, Lemma 3.7] to deduce that whenever B is a building block, i=1 Ci is a ∗-homomorphism and we have fixed some finite subset F ⊂ B and some tolerance ǫ > 0, for all sufficiently large k ∈ N there exists a ∗-homomorphism ψ : B → Ck such that kθ(f ) − ψ(f )k < ǫ for every f ∈ F . We will use this in the proof of Lemma 4.1 (ii). θ : B → C = S∞ Proof of Lemma 4.1. As usual, write W =S∞ self-adjoint elements of the unit ball B1 of B such thatS∞ (i) Let τ0 be a faithful trace in ΣB and fix an increasing sequence F1 ⊂ F2 ⊂ · · · of finite sets of k=1 Fk is dense in the self-adjoint part of B1. i=1 Ai with Ai = A(ni, (ai + 1)ni). k=1, together with ∗-homomorphisms ψk : B → Aik and unitaries uk ∈ gAik We will find a sequence (ik)∞ (with u1 = 1) such that kψk(f ) − uk+1ψk+1(f )u∗ k+1k < 2−k and τ (ψk(f )) − τ0(f ) < 1/k for every f ∈ Fk. We will then have an approximately commutative diagram B id B id B id · · · ψ1 Adu2 ◦ψ2 Adu2 u3 ◦ψ3 Ai1 / Ai2 / Ai3 / · · · / B ψ / W such that τ (Adu1···uk ◦ ψk(f )) − τ0(f ) < 1/k for every f ∈ Fk, and we get a ∗-homomorphism ψ : B → W that satisfies ψ(f ) = limk→∞ u1 · · · ukψk(f )u∗ 1 for every f ∈ B. This will imply that τ ◦ ψ = τ0. k · · · u∗ The idea is to use local existence (Proposition 4.2) to find the ψk and local uniqueness (Proposition 4.3) to find the uk. Working inductively, fix k ≥ 1, and let {ζj}M j=1 ⊂ Aff0T +B be the test functions in Proposition 4.3 corresponding to the finite set Fk and the tolerance ǫk = 2−k. Define Gk to be the finite set Fk ∪ {ζj}M j=1 ∪ Gk−1 ⊂ Bsa (with G0 := ∅). Set ck := 2/3 min{τ0(ζj) : 1 ≤ j ≤ M } and δk := min{1/k, ck/2, ck−1/2}. Since τ0 is faithful, we have ck > 0 and δk > 0. j=1 ∪ {σj}M j=1, {σj}M Let ηk ∈ Aff0T +B and Nk ∈ N be as in Proposition 4.2, corresponding to the finite set Gk and the tolerance δk/2. For each i ∈ N, fix νi ∈ Aff0T +Ai ∼= C[0, 1]ai with, say, kνikAi = 1 and inf ι(νi) = ai/(ai + 1). By construction of W (Proposition 3.1, see also Remark 3.2) we have ai, ni/ai → ∞ as i → ∞, so we may choose ik > ik−1 (where i0 := 1) such that nik /aik > 4Nk/kτ0k and aik /(aik + 1) > 1 − δk/2. Define ξk : Aff0T +B → Aff0T +Aik by ξk(f )(τ ′) := νik (τ ′)f (τ0). This ξk is positive and linear, and we have f (τ0) ≤ kf kB (since kτ0k ≤ 1), so ξk is a contraction from (Aff0T +B, k · kB) to (Aff0T +Aik , k · kAik = sup{ξk(f )(τ ′) : kτ ′k = 1} = kνik kAik kξk(f )kAik ). Since inf ι(ξk(ηk)) = inf τ ′∈T + 1 Aik ξk(ηk)(τ ′) = ηk(τ0) inf ι(νik ) ≥ ηk(τ0) 2 ≥ kτ0k 4 ≥ Nkaik nik , / / / / / / / ✤ ✤ ✤ ✤ ✤ / / / / 10 BHISHAN JACELON Proposition 4.2 implies that there exists a ∗-homomorphism ψk : B → Aik such that kξk(f ) − (ψk)∗(f )kξk(ηk) < δk/2 for every f ∈ Gk. Moreover, we have for every τ ′ ∈ T + 1 Aik . Hence νik (τ ′) − 1 ≤ 1 − aik /(aik + 1) < δk/2 δk/2 > sup ξk(ηk)(τ ′)=1 νik (τ ′)f (τ0) − (ψk)∗(f )(τ ′) = sup νik (τ ′)=1/ηk(τ0) νik (τ ′)f (τ0) − (ψk)∗(f )(τ ′) ≥ sup kτ ′k=1 νik (τ ′)f (τ0) − (ψk)∗(f )(τ ′) ≥ sup f (τ0) − (ψk)∗(f )(τ ′) − δk/2 kτ ′k=1 for every f ∈ Gk. (For the penultimate inequality we have used the fact that ηk(τ0) ≤ 1 and kνik kAik 1.) Thus = τ ′(ψk(f )) − τ0(f ) < δk ∀f ∈ Gk ∀τ ′ ∈ T + 1 Aik . (4.1) In particular, noting Lemma 3.3 and its proof, τ (ψk(f )) − τ0(f ) < 1/k for every f ∈ Fk, and for every τ ′ ∈ T + 1 Aik+1 ⊆ T + (ψk)∗(ζj)(τ ′) > ck, 1 Aik and 1 ≤ j ≤ M we have (ψk+1)∗(ζj )(τ ′) > ck, and (ψk+1)∗(σj)(τ ′) − (ψk)∗(σj )(τ ′) < ck. By Lemma 4.4, we may also assume (by replacing each ψk by ϕik,l ◦ ψk as necessary) that for every k, ψt k(h) has at least three distinct eigenvalues for every t ∈ [0, 1]. Hence, by Proposition 4.3, there exists a unitary uk+1 ∈ ^Aik+1 such that kψk(f ) − uk+1ψk+1(f )u∗ k+1k < 2−k for every f ∈ Fk, as required. (ii) The 'only if' part is obvious. Suppose conversely that ψ1, ψ2 : B → W are ∗-homomorphisms with τ ◦ ψ1 = τ ◦ ψ2. If either map is zero, then the statement is trivial, so we may assume that both ψ1 and ψ2 are injective (see section 2). Fix a finite subset F ⊂ Bsa and a tolerance ǫ > 0, and let j=1 ⊂ Aff0T +B be the test functions in Proposition 4.3 corresponding to F and ǫ/3. Set {ζj}M j=1, {σj}M j=1 ∪ {σj}M F ′ := F ∪ {ζj}M j=1 and δ := min{ǫ/3, τ (ψ1(ζj ))/6 : 1 ≤ j ≤ M }; since ψ1 and ψ2 are injective, we have δ > 0. By semiprojectivity of B, for all sufficiently large k there exist ∗-homomorphisms ψ(k) 1 , ψ(k) : B → Ak such that 2 Note in particular that kψ(k) m (f ) − ψm(f )k < δ ∀f ∈ F ′, m = 1, 2. 1 (f ) − τ ◦ ψ(k) We may also assume that k is large enough so that τ ◦ ψ(k) 2 (f ) < 2δ ∀f ∈ F ′. sup{τ (x) − τ ′(x) : τ ′ ∈ T + 1 Ak} < δ ∀x ∈ ψ(k) 1 (F ′) ∪ ψ(k) 2 (F ′) (4.2) (4.3) (4.4) (as in (3.5) of Proposition 3.5) and so that condition (ii) of Proposition 4.3 holds. By (4.3) and (4.4) we have (ψ(k) 2 )∗(σj )(τ ′) < 4δ and by (4.4), (4.2) and our choice of δ we have (ψ(k) 1 Ak. Hence, by Proposition 4.3, there exists m )∗(ζj )(τ ′) > 4δ for m = 1, 2, 1 ≤ j ≤ M and τ ′ ∈ T + 1 )∗(σj )(τ ′) − (ψ(k) 1 (f ) − uψ(k) 2 (f )u∗k < ǫ/3 for every f ∈ F , which implies that a unitary u ∈ fAk such that kψ(k) kψ1(f ) − uψ2(f )u∗k < ǫ ∀f ∈ F. This proves that ψ1 and ψ2 are approximately unitarily equivalent. (cid:3) An object T in a category C is terminal if for every object X in C there exists a unique morphism from X to T ; such objects are unique up to isomorphism. For example, the Cuntz algebra O2 is the unique terminal object in the category of strongly self-absorbing C∗-algebras, where the morphisms are approximate unitary equivalence classes of unital ∗-homomorphisms (see [KP00] and [TW07]). We now use Lemma 4.1 and an intertwining argument to characterize (W, τ ) as a terminal object. A SIMPLE, MONOTRACIAL, STABLY PROJECTIONLESS C ∗-ALGEBRA 11 Theorem 4.5. (W, τ ) is the unique terminal object in the category whose objects are pairs (A, τA) with A a simple Razak algebra and τA ∈ ΣA, and where a morphism from (A, τA) to (B, τB) is (the approximate unitary equivalence class of ) a ∗-homomorphism ψ : A → B with ψ∗τB = τA. Proof. Let B = lim−→(Bi, βi) be a simple Razak algebra and let τ0 ∈ ΣB. We need to show that there is a ∗-homomorphism ψ : B → W with ψ∗τ = τ0, and prove that, up to approximate unitary equivalence, ψ is the unique map B → W with this property. This is obvious if τ0 = 0 (since τ is faithful), so we i=1, where for i ∈ N, may assume that τ0 is nonzero, hence faithful (since B is simple). Write τ0 = (τi)∞ τi is a faithful trace on Bi with kτik ≤ 1 and τi+1 ◦ βi = τi. By Lemma 4.1(i), for each i there exists a ∗-homomorphism ψi : Bi → W with τ ◦ ψi = τi. But τ ◦ ψi+1 ◦ βi = τi+1 ◦ βi = τi, so by Lemma 4.1(ii), we have ψi+1 ◦ βi ∼a.u. ψi. It then follows from a (one-sided) approximate intertwining (as in the proof of Lemma 4.1(i)) that there is a ∗-homomorphism ψ : B → W which, by construction, satisfies ψ∗τ = τ0. By restricting ψ to the building blocks Bi, Lemma 4.1(ii) says that, up to approximate unitary equivalence, ψ is the unique ∗-homomorphism B → W with this property. (cid:3) The next two results are immediate corollaries of Theorem 4.5. Corollary 4.6. Every trace-preserving endomorphism (hence every nondegenerate endomorphism) of W is approximately inner. That is, for any such endomorphism θ, there is a sequence of unitaries un in fW such that θ(a) = limn→∞ unau∗ n for every a ∈ W . Corollary 4.7. Let B be a simple Razak algebra. Then B admits a tracial state if and only if B is isomorphic to a subalgebra of W . If B has no nonzero bounded traces, then B is stable and is isomorphic to a subalgebra of W ⊗ K. Proof. The first assertion follows from Theorem 4.5. For the second assertion, note that if B is a simple Razak algebra then so is B ⊗ K, and (T +(B ⊗ K), ΣB⊗K) ∼= (T +B, 0). Therefore, Theorem 1.1 implies that B ∼= B ⊗ K whenever B has no nonzero bounded traces, and Theorems 1.3 and 1.1 imply that every simple Razak algebra is stably isomorphic to a simple Razak algebra which has no unbounded traces. The second statement therefore follows from the first. (cid:3) 5. W -stability Let us first make a remark on the proof that Z is strongly self-absorbing. Jiang and Su adopt the following strategy. (i) Prove that the two maps id ⊗ 1, 1 ⊗ id : Z → Z ⊗ Z are approximately unitarily equivalent. (ii) Show that there exists a unital ∗-homomorphism ψ : Z ⊗ Z → Z. (iii) Prove that (id ⊗ 1) ◦ ψ ∼a.u. idZ⊗Z and note that ψ ◦ (id ⊗ 1) ∼a.u. idZ since every unital endo- morphism of Z is approximately inner. A standard intertwining argument [Rør94, Proposition A] then shows that there exists an isomorphism ϕ : Z → Z ⊗ Z. Again, since every unital endomorphism of Z is approximately inner, it follows easily that ϕ ∼a.u. id ⊗ 1. Step (ii) in this procedure goes roughly as follows: Write the Jiang-Su algebra as Z = lim−→(An, ϕn), where each An = I[pn, dn, qn] is a prime dimension drop algebra (i.e. pn and qn are coprime, dn = pnqn and An = {f ∈ C([0, 1], Mdn) : f (0) ∈ Mpn ⊗ 1qn , f (1) ∈ 1pn ⊗ Mqn}). Define Bn to be the diagonal of such that f (0) ∈ (Mpn ⊗ 1qn )⊗2 An ⊗ An, i.e. Bn consists of all continuous functions f : [0, 1] → Md2 and f (1) ∈ (1pn ⊗ Mqn )⊗2. Then Bn ∼= I[p2 n] is a prime dimension drop algebra and we have a ∗-homomorphism ρn : An ⊗ An → Bn given by restriction: ρn(f )(x) = f (x, x) for f ∈ An ⊗ An and x ∈ [0, 1]. Jiang and Su construct connecting maps ψn : Bn → Bn+1 such that the diagram n, d2 n, q2 n A1 ⊗ A1 ϕ1⊗ϕ1 A2 ⊗ A2 ϕ2⊗ϕ2 A3 ⊗ A3 · · · / Z ⊗ Z ρ1 B1 ρ2 ρ3 ψ1 / B2 ψ2 / B3 / · · · ρ / B / / / / / / / ✤ ✤ ✤ ✤ ✤ / / / / 12 BHISHAN JACELON commutes approximately (where B := lim−→(Bn, ψn)), so there is an induced morphism ρ : Z ⊗ Z → B. They also show that B is simple, and since it is an inductive limit of (prime) dimension drop algebras, their Theorem 6.2 then shows that there exists a morphism from B to Z; the composition of this morphism with ρ is the required morphism ψ : Z ⊗ Z → Z. It turns out that, even though all of the restriction maps ρj are surjective, the induced morphism In particular, B may not be isomorphic to Z ⊗ Z, and steps (i) and (iii) seem to ρ need not be. be unavoidable. This is unfortunate for us because it is difficult to make sense of these steps for the nonunital algebra W . Perhaps it would be possible to construct a two-sided approximate intertwining between W ⊗ W = lim−→(Ai ⊗ Ai, ϕi ⊗ ϕi) and some limit of suitable one-dimensional NCCW complexes, but it is currently unclear how this might be done. We therefore leave the following as a conjecture, and remark that classification certainly predicts it to be true. Conjecture 5.1. W ⊗ W ∼= W . Remark 5.2. If Conjecture 5.1 were to hold, then it would follow immediately from Corollary 4.6 that W would have approximately inner flip, i.e. there would be a sequence (un)∞ n=1 of unitaries in the unitization of W ⊗ W such that limn→∞ un(a ⊗ b)u∗ n = b ⊗ a for a, b ∈ W . Next, we prove that, were Conjecture 5.1 to hold, every simple Razak algebra would absorb W tensorially. We need the following extension of a theorem of Blackadar [Bla80] and Goodearl [Goo78], which is likely to be already known by experts. The reader is referred to [BPT08] or [APT09] for details of the Cuntz semigroup Cu(·). Proposition 5.3. Let ∆ be a metrizable Choquet simplex, let C be the cone with ∆ as its base, and let ω be a lower semicontinuous affine map ∆ → (0, ∞]. Then there exists a simple AF algebra A and an isomorphism T +A ∼= C under which ω corresponds to the norm map on T +A. Proof. By [Bla80, Theorem 3.10], there exists a simple, unital AF algebra B with T + 1 B = ∆. (Tensoring B with a UHF algebra if necessary, we may assume that B is infinite dimensional.) We will produce a hereditary sublagebra A of B ⊗ K with the required properties. Since every trace on B extends uniquely to one on B ⊗ K, we identify T +(B ⊗ K) with C. As in [BT07], denote by SAff(∆) the space of strictly positive lower semicontinuous affine maps on ∆. The key fact is that the natural map Cu(B)\V (B) → SAff(T + 1 B) is surjective. What this means is that, given ω ∈ SAff(∆), there exists a positive element b ∈ B ⊗ K (which we may assume to be of norm 1) such that ω(τ ) = limn→∞ τ (b1/n) for every τ ∈ ∆. This follows from [BPT08, Theorem 5.3] and [BT07, Theorem 2.6], but is not difficult to prove using the well-documented structure of the K-theory of simple, unital AF algebras. Let A = b(B ⊗ K)b. Then A is a (full) hereditary subalgebra of B ⊗ K, so is a simple AF algebra which is stably isomorphic to B (for example by [Bro77, Theorem 2.8] and [Ell76a, Theorem 3.1]). Moreover, every trace on A extends uniquely to a trace on B ⊗ K (see [BK04, Remark 2.27 (viii)]), so T +A can also be identified with C. Since (b1/n)∞ n=1 is an approximate unit for A, we then have kτ k = lim n→∞ τ (b1/n) = ω(τ ) for every τ ∈ ∆ (and hence, by linearity, for every τ ∈ T +A). (cid:3) Corollary 5.4. Let A be a simple inductive limit of one-dimensional NCCW complexes Ai such that K1(Ai) = 0 for every i and K0(A) = 0. Then there exists a simple AF algebra B such that A ∼= B ⊗ W . In particular, if Conjecture 5.1 were to hold, it would follow that A ⊗ W ∼= A. Proof. Taking C = T +A and ω = k · k, Theorem 5.3 gives a simple AF algebra B and an isomorphism between T +B and T +A that maps the tracial states of B onto the tracial states of A. The C∗-algebra W is by construction a simple inductive limit of one-dimensional NCCW complexes each of which has trivial K-theory, so B ⊗ W is also of this form. Moreover, T +W is generated by a tracial state, so there is an isomorphism between T +(B ⊗ W ) and T +B that maps the tracial states of B ⊗ W onto the tracial states of B. (This is easy to see for F ⊗ W whenever F is finite dimensional, and the assertion for B follows from the continuity of the functor T +.) Theorem 1.2 then implies that A ∼= B ⊗ W , and the second statement would follow directly from Conjecture 5.1. (cid:3) A SIMPLE, MONOTRACIAL, STABLY PROJECTIONLESS C ∗-ALGEBRA 13 Remark 5.5. This shows that every inductive limit as in the statement of Corollary 5.4 is isomorphic to an inductive limit of finite direct sums of Razak building blocks. Moreover, the full range of the invariant is exhausted, so we obtain a version of Theorem 1.3. In the opposite direction, we note that a certain dichotomy holds for well-behaved, simple, W -stable C∗-algebras. Proposition 5.6. Suppose that A is a simple, separable, nuclear C∗-algebra that satisfies the UCT, such that A ⊗ W ∼= A. Then either A ∼= O2 ⊗ K or A is stably projectionless. Proof. Since W lies in the UCT class, we can apply the Kunneth Theorem to deduce that K∗(A) = K∗(A ⊗ W ) = 0. Suppose that A ⊗ K contains a nonzero projection p, and let B := p(A ⊗ K)p. Then B is a unital C∗-algebra which, by Brown's Theorem [Bro77, Theorem 2.8], is stably isomorphic to A; in particular, K∗(B) = 0. Hence [p] = [0] in K0(B), so p ⊕ q ∼ 0 ⊕ q for some projection q ∈ Mn(B). Hence p ⊕ q is infinite, so B is 'stably infinite' and so is A. Then, since A ∼= A ⊗ W is tensorial non- prime, A must in fact be purely infinite [Rør02, Theorem 4.1.10] and stable [Rør02, Theorem 4.1.3]. The Kirchberg-Phillips classification theorem [Rør02, Theorem 8.4.1] then shows that A ∼= O2 ⊗ K. (cid:3) 6. Asymptotically central sequences If A and B are C∗-algebras and C is a sub-C∗-algebra of B then a sequence of ∗-homomorphisms ϕn : A → B is said to be asymptotically central for C if k[ϕn(a), c]k → 0 as n → ∞ for every a ∈ A and c ∈ C. Such a sequence induces a ∗-homomorphism ϕ : A → B∞ ∩ C′. Here, B∞ is the limit algebra l∞(B)/c0(B), and there is a canonical inclusion of B into B∞ as constant sequences; B∞ ∩ C′ denotes the relative commutant of C in B∞. Every strongly self-absorbing C∗-algebra D admits an asymptotically central sequence of unital endo- morphisms [TW07, Proposition 1.10]. Conversely, exhibiting asymptotically central ∗-homomorphisms can be used to prove D-stability -- see [Rør94], [Rør02, Theorem 7.2.2] and [TW07, Theorem 2.3]. It is not clear whether there exists such a relationship between asymptotically central sequences and tensor products in the nonunital case, essentially because we lack the notion of an infinite tensor product. Nevertheless, asymptotically central sequences are still interesting and useful in their own right, and we show below that we can adapt the approximate divisibility of W to at least find an embedding of W into the limit algebra M (W )∞ ∩ W ′. Definition 6.1. A C∗-algebra A is said to be approximately divisible if for any N ∈ N there is a sequence of unital ∗-homomorphisms µn from MN ⊕ MN +1 into the multiplier algebra M (A) which is asymptotically central for A. Approximate divisibility for unital C∗-algebras was studied in [BKR92], and the nonunital case appears for example in [Rør02, Definition 3.1.10]. Toms and Winter prove in [TW05, §2] that separable approximately divisible C∗-algebras are Z-stable. Moreover, their Proposition 4.1 says that every simple Razak algebra is approximately divisible (hence Z-stable). We show below that this is in fact an immediate consequence of classification (Theorem 1.1). Let Q denote the universal UHF algebra (characterized by K0(Q) = Q); Q is isomorphic to its infinite tensor product Q⊗∞ and (so) is strongly self-absorbing. Proposition 6.2. Every simple Razak algebra is Q-stable (so absorbs every UHF algebra). Proof. Let B = lim−→(Bi, βi) be a simple Razak algebra and let U = lim−→(Mki , αi) be a UHF algebra. Note that if A is a building block then so is A ⊗ Mk for every k, so B ⊗ U ∼= lim−→(Bi ⊗ Mki, βi ⊗ αi) is also a simple Razak algebra. Moreover, (T +(B ⊗ U ), ΣB⊗U ) ∼= (T +B, ΣB) since U has a unique tracial state. Therefore, Theorem 1.1 implies that B ⊗ U ∼= B. (cid:3) Corollary 6.3. Let A be a simple Razak algebra. Then A is approximately divisible and there is an isomorphism ϕ : A → A ⊗ Z such that ϕ ∼a.u. idA ⊗ 1Z. 14 BHISHAN JACELON Proof. For each k ∈ N, let ιk be a unital embedding of Mk into Q and define a sequence of unital ∗-homomorphisms µk,m : Mk → M (A ⊗ Q⊗∞) by µk,m := 1 eA ⊗ 1Q ⊗ · · · ⊗ 1Q ⊗ ιk{z}m ⊗1Q ⊗ · · · (6.1) (where eA is the minimal unitization of A). Then the sequence (µk,m)∞ m=1 is asymptotically central for A ⊗ Q⊗∞. Thus A ∼= A ⊗ Q⊗∞ is approximately divisible. The second claim follows directly from Proposition 6.2 since UHF algebras are Z-stable, or alternatively from [TW05, Theorem 2.3], which says that separable approximately divisible C∗-algebras are Z-stable. That the isomorphism ϕ satisfies ϕ ∼a.u. idA ⊗ 1Z is automatic (see [TW07, Theorem 2.2]). (cid:3) Next, we combine the proofs of Corollary 6.3 and [TW05, Proposition 2.2] to construct an embedding of W into the central sequences algebra M (W )∞ ∩ W ′. Theorem 6.4. Let B be a separable Q-stable C∗-algebra. Then there exists a ∗-homomorphism σ = (σi)∞ satisfies the nondegeneracy condition i=1 : W → (eB ⊗ Q⊗∞)∞ ∩ (B ⊗ Q⊗∞)′ ⊂ M (B ⊗ Q⊗∞)∞ ∩ (B ⊗ Q⊗∞)′ ∼= M (B)∞ ∩ B′ which kσ(a)bk = kakkbk for every a ∈ W and b ∈ B, (6.2) and which is trace-preserving in the sense that lim i→∞ ρ(σi(a)) = τ (a) for every a ∈ W and every ρ ∈ T + 1 (B) (6.3) (where τ is the unique tracial state on W ). Remark 6.5. If B is a Q-stable C∗-algebra then every tracial state ρ on B ⊗ Q⊗∞ (i) is of the form τ ⊗ τQ for some tracial state τ on B, where τQ is the unique tracial state on Q, and This is what is meant in (6.3). It may also be more natural to replace B∞ with an ultrapower Bω for some free ultrafilter ω, and the proof works equally well in this setting. (ii) extends uniquely to a tracial state on eB ⊗ Q⊗∞ ⊂ M (B ⊗ Q⊗∞) ∼= M (B). Proof. Write W = S∞ define ∗-homomorphisms µk,m : Mk → eB ⊗ Q⊗∞ ⊂ M (B ⊗ Q⊗∞) by i=1 Ai with the building blocks Ai = A(ni, (ai + 1)ni) and inclusion maps ϕij given by Proposition 3.1. As in (6.1), for each k ∈ N let ιk be a unital embedding of Mk into Q, and ⊗1Q ⊗ · · · . µk,m := 1 eB ⊗ 1Q ⊗ · · · ⊗ 1Q ⊗ ιk{z}m Note that the sequence (µk,m)∞ m=1 is asymptotically central for B⊗Q⊗∞ and that kµk,m(a)bk → kakkbk as m → ∞ for every a ∈ Mk and b ∈ B ⊗ Q⊗∞ (since this is true for finite tensors). Moreover, if ρ is a tracial state on B ⊗ Q⊗∞ and trk denotes the unique tracial state on Mk, then in the sense of Remark 6.5 we have ρ(µk,m(x)) = trk(x) for every x ∈ Mk and m ∈ N. For every i ∈ N, let πi : Ai → Mni be the irreducible representation ev∞ (actually, any point i=1 be dense in B ⊗ Q⊗∞ i=1 Fi = W . For each i, we can use the properties of the ∗-homomorphisms µni,m to choose mi ≥ mi−1 such that for a ∈ Fi and 1 ≤ j ≤ i we have evaluation will do), and define σi,m := µni,m ◦ πi : Ai → eB ⊗ Q⊗∞. Let (bi)∞ and fix finite subsets Fi ⊂ Ai such that Fi ⊂ Fi+1 and S∞ and k[σi,mi(a), bj]k ≤ 1/i (cid:12)(cid:12)(cid:12)(cid:12)kσi,mi(a)bjk − kπi(a)kkbjk(cid:12)(cid:12)(cid:12)(cid:12) ≤ 1/i. Note also that, for every a ∈ Ai and as j → ∞ we have by (3.2) and (3.3) of Proposition 3.1 and kπj ◦ ϕij (a)k → kak trnj (πj ◦ ϕij (a)) → τ (a) (6.4) (6.5) (6.6) (6.7) A SIMPLE, MONOTRACIAL, STABLY PROJECTIONLESS C ∗-ALGEBRA 15 by (3.5) of Proposition 3.5. Now we just patch together the σi,mi to get the desired map σ (as in the proof of [TW05, Proposition 2.2]). Since σi,mi is finite rank, by Arveson's extension theorem we the map σ := (σi,mi)∞ can extend it to a linear, contractive (in fact c.c.p.) map σi,mi : W → eB ⊗ Q⊗∞. Define σ to be i=1 : W → (eB ⊗ Q⊗∞)∞. Then σ is linear, contractive and (since the σi,mi are ∗-homomorphisms) is multiplicative on S∞ By (6.4), σ(W ) commutes with B ⊗ Q⊗∞. For a ∈ Fi and fixed bj we have i=1 Ai, hence on all of W . That is, σ is a ∗-homomorphism. kσ(a)bjk = lim sup k kσk,mk (ϕik(a))bjk = lim sup k kπk(ϕik(a))kkbjk = kakkbjk by (6.5) and (6.6). Finally, for a ∈ Ai and ρ ∈ T + 1 (B ⊗ Q⊗∞) we have lim k→∞ ρ(σk,mk (a)) = lim k→∞ ρ(µnk,mk ◦ πk(ϕik(a))) = lim k→∞ trnk (πk(ϕik(a))) = τ (a) by (6.7). Therefore, σ is a ∗-homomorphism W → (eB ⊗ Q⊗∞)∞ ∩ (B ⊗ Q⊗∞)′ that satisfies (6.2) and (6.3). (cid:3) It is easy to check that, by taking an approximate unit (ei)∞ appropriate subsequence of (ei⊗1Q⊗∞)∞ map from W into B∞ ∩ B′ which preserves orthogonality (i.e. is 'order zero' -- see [WZ09]). i=1 of B and cutting σ down by an i=1, we can get a trace-preserving completely positive contractive Corollary 6.6. There exists a ∗-homomorphism σ = (σi)∞ kσ(a)bk = kakkbk for every a, b ∈ W and limi→∞ τ (σi(a)) = τ (a) for every a ∈ W . i=1 : W → M (W )∞ ∩ W ′ which satisfies References [APT09] Pere Ara, Francesc Perera and Andrew S. Toms. K-Theory for operator algebras. Classification of C ∗-algebras. arXiv preprint math.OA/0902.3381, 2009. [BKR92] Bruce Blackadar, Alexander Kumjian, and Mikael Rørdam. Approximately central matrix units and the struc- ture of noncommutative tori. K-Theory, 6(3):267 -- 284, 1992. [Bla80] Bruce Blackadar. Traces on simple AF C ∗-algebras. J. Funct. Anal., 38(2):156 -- 168, 1980. [BK04] Etienne Blanchard and Eberhard Kirchberg. Non-simple purely infinite C ∗-algebras: the Hausdorff case. J. Funct. Anal., 207(2):461 -- 513, 2004. [BPT08] Nathanial P. Brown, Francesc Perera and Andrew S. Toms. The Cuntz semigroup, the Elliott conjecture, and dimension functions on C ∗-algebras. J. Reine Angew. Math., 621:191 -- 211, 2008. [Bra72] Ola Bratteli. Inductive limits of finite dimensional C ∗-algebras. Trans. Amer. Math. Soc., 171:195 -- 234, 1972. [Bro77] Lawrence G. Brown. Stable isomorphism of hereditary subalgebras of C ∗-algebras. Pacific J. Math., 71(2):335 -- 348, 1977. [BT07] Nathanial P. Brown and Andrew S. Toms. Three applications of the Cuntz semigroup. Int. Math. Res. Not. IMRN, no. 19, Art. ID rnm068, 14pp, 2007. [Con81] Alain Connes. An analogue of the Thom isomorphism for crossed products of a C ∗-algebra by an action of R. Adv. in Math., 39(1):31 -- 55, 1981. [Dea01] Andrew Dean. A continuous field of projectionless C ∗-algebras. Canad. J. Math., 53(1):51 -- 72, 2001. [EGJS97] George A. Elliott, Guihua Gong, Xinhui Jiang, and Hongbing Su. A classification of simple limits of dimension drop C ∗-algebras. In Operator algebras and their applications (Waterloo, ON, 1994/1995), volume 13 of Fields Inst. Commun., pages 125 -- 143. Amer. Math. Soc., Providence, RI, 1997. [Ell76a] George A. Elliott. Automorphisms determined by multipliers on ideals of a C ∗-algebra. J. Funct. Anal., 23(1):1 -- 10, 1976. [Ell76b] George A. Elliott. On the classification of inductive limits of sequences of semisimple finite-dimensional algebras. J. Algebra., 38(1):29 -- 44, 1976. [Ell93] George A. Elliott. On the classification of C ∗∗-algebras of real rank zero. J. Reine Angew. Math., 443:179 -- 219, 1993. [ELP98] Søren Eilers, Terry A. Loring and Gert K. Pedersen. Stability of anticommutation relations: an application of noncommutative CW complexes. J. Reine Angew. Math., 499:101 -- 143, 1998. [ET08] George A. Elliott and Andrew S. Toms. Regularity properties in the classification program for separable amenable C ∗-algebras. Bull. Amer. Math. Soc. (N.S.), 45(2):229 -- 245, 2008. [Eva80] David E. Evans. On On. Publ. Res. Inst. Math. Sci., 16(3):915 -- 927, 1980. [Gli60] James G. Glimm. On a certain class of operator algebras. Trans. Amer. Math. Soc., 95:318 -- 340, 1960. [Goo78] K. R. Goodearl. Algebraic representations of Choquet simplexes. J. Pure Appl. Algebra, 11(1 -- 3):111 -- 130, 1977/78. [JS99] Xinhui Jiang and Hongbing Su. On a simple unital projectionless C ∗-algebra. Amer. J. Math., 121(2):359 -- 413, 1999. 16 BHISHAN JACELON [Kir] Eberhard Kirchberg. The classification of purely infinite C ∗-algebras using Kasparov's theory. in preparation for Fields Inst. Monograph. [KK96] Akitaka Kishimoto and Alexander Kumjian. Simple stably projectionless C ∗-algebras arising as crossed products. Canad. J. Math., 48(5):980 -- 996, 1996. [KK97] Akitaka Kishimoto and Alexander Kumjian. Crossed products of Cuntz algebras by quasi-free automorphisms. In Operator algebras and their applications (Waterloo, ON, 1994/1995), volume 13 of Fields Inst. Commun., pages 173 -- 192. Amer. Math. Soc., Providence, RI, 1997. [KP00] Eberhard Kirchberg and Christopher N. Phillips. Embedding of exact C ∗-algebras in the Cuntz algebra O2. J. Reine Angew. Math., 525:17 -- 53, 2000. [KW04] Eberhard Kirchberg and Wilhelm Winter. Covering dimension and quasidiagonality. Internat. J. Math., 15(1):63 -- 85, 2004. [Lor93] Terry A. Loring. C ∗-algebras generated by stable relations. J. Funct. Anal., 112(1):159 -- 203, 1993. [Phi00] Christopher N. Phillips. A classification theorem for nuclear purely infinite simple C ∗-algebras. Doc. Math., 5:49 -- 114 (electronic), 2000. [Raz02] Shaloub Razak. On the classification of simple stably projectionless C ∗-algebras. Canad. J. Math., 54(1):138 -- 224, 2002. [Rob10] Leonel Robert. Classification of inductive limits of 1-dimensional NCCW complexes. arXiv preprint math.OA/1007.1964, 2010. [Rør94] Mikael Rørdam. A short proof of Elliott's theorem: O2 ⊗ O2 ∼= O2. C. R. Math. Rep. Acad. Sci. Canada, 16(1):31 -- 36, 1994. [Rør02] Mikael Rørdam. Classification of nuclear C ∗-algebras, volume 126 of Encyclopaedia of Mathematical Sciences, Springer-Verlag, Berlin, 2002. Operator Algebras and Non-commutative Geometry, 7. [Rør03] Mikael Rørdam. A simple C ∗-algebra with a finite and an infinite projection. Acta Math., 191(1):109 -- 142, 2003. [Rør04] Mikael Rørdam. The stable and the real rank of Z-absorbing C ∗-algebras. Internat. J. Math., 15(10):1065 -- 1084, 2004. [RW09] Mikael Rørdam and Wilhelm Winter. The Jiang-Su algebra revisited. J. Reine Angew. Math., 642:129 -- 155, 2010. [Sch84] Claude Schochet. Topological methods for C ∗-algebras. III. Axiomatic homology. Pacific J. Math., 114(2):399 -- 445, 1984. [Tom08] Andrew S. Toms. An infinite family of non-isomorphic C ∗-algebras with identical K-theory. Trans. Amer. Math. Soc., 360(10):5343 -- 5354, 2008. [Tsa05] Kin-Wai Tsang. On the positive tracial cones of simple stably projectionless C ∗-algebras. J. Funct. Anal., 227(1):188 -- 199, 2005. [TW05] Andrew S. Toms and Wilhelm Winter. Z-stable ASH algebras. Canad. J. Math., 60(3):703 -- 720, 2005. [TW07] Andrew S. Toms and Wilhelm Winter. Strongly self-absorbing C ∗-algebras. Trans. Amer. Math. Soc., 359(8):3999 -- 4029 (electronic), 2007. [Vil98] Jesper Villadsen. Simple C ∗-algebras with perforation. J. Funct. Anal., 154(1):110 -- 116, 1998. [Win07] Wilhelm Winter. Localizing the Elliott conjecture at strongly self-absorbing C ∗-algebras. arXiv preprint math.OA/0708.0283, 2007. [Win09] Wilhelm Winter. Strongly self-absorbing C ∗-algebras are Z-stable. J. Noncommut. Geom., 5(2):253 -- 264, 2011. [Win10] Wilhelm Winter. Decomposition rank and Z-stability. Invent. Math., 179(2):229 -- 301, 2010. [WZ09] Wilhelm Winter and Joachim Zacharias. Completely positive maps of order zero. Munster J. Math., 2:311 -- 324, 2009. Mathematisches Institut, Einsteinstr. 62, 48149 Munster, Germany E-mail address: [email protected]
1905.03504	1	1905	2019-05-09T09:37:38	Some remarks in $C^*$- and $K$-theory	[ "math.OA", "math.KT" ]	This note consists of three unrelated remarks. First, we demonstrate how roughly speaking $$-homomorphisms between matrix stable $C^$-algebras are exactly the uniformly continuous $*$-preserving group homomorphisms between their genral linear groups. Second, using the Cuntz picture in $KK$-theory we bring morphisms in $KK$-theory represented by generators and relations to a particular simple form. Third, we show that for an inverse semigroup its associated groupoid is Hausdorff if and only if the inverse semigroup is $E$-continuous.	math.OA	math	SOME REMARKS IN C∗- AND K-THEORY BERNHARD BURGSTALLER Abstract. This note consists of three unrelated remarks. First, we demon- strate how roughly speaking ∗-homomorphisms between matrix stable C ∗- algebras are exactly the uniformly continuous ∗-preserving group homomor- phisms between their genral linear groups. Second, using the Cuntz picture in KK-theory we bring morphisms in KK-theory represented by generators and relations to a particular simple form. Third, we show that for an inverse semi- group its associated groupoid is Hausdorff if and only if the inverse semigroup is E-continuous. . A O h t a m [ 1 v 4 0 5 3 0 . 5 0 9 1 : v i X r a 1. Introduction In this note we present three unrelated results in C∗-theory and K-theory. The first result is demonstrated in Section 2 and shows that for all unital C∗-algebras A and B, every uniformly continuous, ∗-preserving group homomorphism ϕ : GL(A ⊗ M2) → GL(B) can be extended to a ∗-homomorphism A⊗M2 → B, provided a very light additional technical condition for the restriction of ϕ to the complex numbers is satisfied, see Corollary 2.3 and Section 2. Actually, we have demonstrated a similar result already in [5], but the improvement, thanks to some trick by L. Moln´ar [12], is that the additional technical condition is here subjectively somewhat easier, even if not strictly logically comparable with the one in [5]. In the next Section 3, we make a turn to KK-theory [9]. J. Cuntz [6] and N. Higson [8] found out that Kasparov's KK-theory is the universal stable, homotopy invariant, split-exact functor from the C∗-category to an additive category. This makes it possible to describe KK-theory as a localization of the category of C∗- algebras, or expressed in less technical terms, by adding certain synthetical inverses to the category of C∗-algebras and moding out certain relations to form KK-theory. We slightly simplify the representation of KK-elements in this picture at first, but make the most dramatical simplification by using the Cuntz-picture [6, 7] of KK- theory elements. This picture of KK-theory may also be analogously and readily defined equivariantly for other equivariant structures than groups, say semigroups, categories and so on, and even the category of C∗-algebras may be changed to other (topological) algebras. In the last Section 4 we observe that a discrete inverse semigroup induces a Hausdorff groupoid if and only if the inverse semigroup is E-continuous. We also note that both equivalent technical conditions appear necessary to define a non- degenerate, C0(X)-compatible C0(X)-valued L2(G)-module, see Definition 4.4 and Example 4.14 for more on this. Such a module is a useful tool for the computation of the K-theory of inverse semigroup crossed products. However, the lack of such 1991 Mathematics Subject Classification. 46L05, 19K35, 20M18. Key words and phrases. C ∗-algebra, general linear group, homomorphism, KK-theory, gener- ators, universal property, inverse semigroup, Hausdorff, L2. 1 2 B. BURGSTALLER a module in the non-Hausdorff case hinders the computation of beformentioned K-theory groups of crossed products by non-applicability of parallel methods suc- cessful in the group case. The difficulty of computation has been already observed by Tu [15] for the more general setting of non-Hausdorff groupoids in the context of Baum -- Connes theory. All chapters in this note can be read completely independently. 2. Group and algebra homomoprhisms In this section we show how certain group homomorphisms between the group of invertible elements of C∗-algebras can be extended to ∗-homomorphisms. A map ϕ : A → B between C∗-algebras A and B is called a ∗-semigroup homomorphism if it is multiplicative (i.e. ϕ(ab) = ϕ(a)ϕ(b)) and ∗-preserving (i.e. ϕ(a∗) = ϕ(a)∗). As usual, Mn denotes the C∗-algebra of all complex-valued n × n-matrices, and GL(A) the general linear group of A. Proposition 2.1. Let ϕ : GL(A ⊗ M2) → B be an arbitrary function where A and B are C∗-algebras and A is unital. Then the following are equivalent: (a) ϕ extends to a ∗-homomorphism A ⊗ M2 → B. (b) ϕ is a uniformly continuous, ∗-semigroup homomorphism with (1) kϕ(1/2)k < 1, ϕ(i1) = i1. Remark 2.2. Alternatively, instead of requiring kϕ(1/2)k < 1 in Proposition 2.1.(b), we may equivalently require that kϕ(z)k < 1 for any single fixed z ∈ GL(A ⊗ Mn) with kzk < 1. Proof. (a) to (b) is clear. To show (b) to (a), we are going to apply [, Propo- sition 2.6]. At first we continuously extend ϕ to an equally denoted function ϕ : GL(A ⊗ M2) → B (norm closure) by using Cauchy sequences and the uni- form continuity of ϕ. Then ϕ is a ∗-semigroup homomorphism. Notice that ϕ(0) = limn ϕ(zn) = 0 by Remark 2.2. By applying Proposition 2.6 of [5] we are done when showing the ortho-additivity relation ϕ(e11 + e22) = ϕ(e11) + ϕ(e22), where eii are the standard matrix corners. To this end, we use the following trick by L. Moln´ar [12] by means of the exponential function, which we are going to recall for convenience of the reader. Consider the C∗-subalgebra B′ of B generated by the image of ϕ. It is unital with unit ϕ(1). Represent B′ faithfully on a Hilbert space H such that 1B(H) is the unit of B′. In the following, identify now B′ as a subalgebra of B(H). Let P be a projection in M2(A). Clearly eλP is invertible for every λ ∈ R and so in the domain of ϕ. Consider the map λ 7→ ϕ(eλP ) = ϕ(1 − P + eλP ) from R into GL(B(H)). This is a one-parameter group. Thus there exists an operator T ∈ B(H) such that ϕ(1 − P + eλP ) = eλT . Since ϕ is ∗-preserving, eλT is self-adjoint for all λ ∈ R. This implies that T is also self-adjoint. By the uniform continuity of ϕ, for every ε > 0 there exists a δ > 0 such that keλT − eµT k = sup t∈σ(T ) eλt − eµt < ε SOME REMARKS IN C ∗- AND K-THEORY 3 if eλ − eµ < δ. The last identity is by standard functional calculus. Therefore, the function x 7→ xt is uniformly continuous on the positive half-line for all t ∈ σ(T ). Hence σ(T ) ⊆ {0, 1} and so T is a projection. Consequently, ϕ(1 − P + eλP ) = 1 − T + eλT. For λ → −∞ we get ϕ(1 − P ) = 1 − T . Setting P = 1 and using ϕ(0) = 0 this implies T = 1, and consequently ϕ(eλ1) = eλ1. In particular, ϕ is R+-homogeneous. Hence the above equality divided by eλ and letting λ → ∞ yields ϕ(P ) = T . Thus, putting λ = 1, Now set P = e11. ϕ(1) = ϕ(1 − P ) + ϕ(P ). (cid:3) We remark that in Proposition 2.1.(b) ϕ is obviously actually a group homomor- phism into the image of ϕ. So let us also state the following variant to emphasize this fact: Corollary 2.3. Let ϕ : GL(A ⊗ M2) → GL(B) be an arbitrary function where A and B are unital C∗-algebras. Then the following are equivalent: (a) ϕ extends to a unital ∗-homomorphism A ⊗ M2 → B. (b) ϕ is a uniformly continuous, ∗-preserving group homomorphism satisfying (1). Examples 2.4. • The determinante det : GL(Mn(C))) → GL(C), though a continuous ∗-preserving group homomorphism, cannot be extended to a ∗-homomorphism because det(λ1) = λn, which is not uniformly continuous. • The trivial group homomorphism ϕ : GL(Mn(A)) → GL(B), ϕ(x) = 1, though a uniformly continuous ∗-preserving group homomorphism, cannot be extended to a ∗-homomorphism because kϕ(1/2)k = 1. 3. KK -theory and generators In this section we deal with the Kasparov category KK. This is the category with object class being the C∗-algebras, and morphism class from C∗-algebra A to C∗-algebra B being the Kasparov group KK(A, B). Composition of morphisms is defined to be the Kasparov product KK(A, B) × KK(B, C) → KK(A, C) : (f, g) 7→ f g := f ⊗B g. Analogously, we have the Kasparov category KK G in the group equivariant setting with respect to a given second-countable locally compact group G. By the work of J. Cuntz [6] and N. Higson [8] it became clear that Kasparov's KK-theory allows a very elegant characterization when restricted to the class of ungraded separable C∗-algebras. Cuntz noted that if F is a stable, homotopy invariant, split-exact functor F from the C∗-category C∗ to the abelian groups Ab, then each KK-theory element of KK(A, B) induces a map F (A) → F (B). Higson brought these findings to its final form by showing that the Kasparov category KK is universal in this respect in the sense that every such functor F factorizes over the Kasparov category KK. This fact is called the universal property of KK-theory. K. Thomsen has generalized this result to the group equivariant setting, that is, to the category KK G. 4 B. BURGSTALLER Quite straightforward, in [3] we described KK G-theory by generators and rela- tions based on Cuntz and Higsons's findings. We denoted it by GK-theory ('gen- erators K-theory', the group G is not indicated) for better clearity. One advantage of this basic construction is that it may be straightforwardly generalized to other modes of equivariance, that is, to other objects than groups G, for example semi- groups G, categories G and so on. Also, one may change the category C∗ to another category of (topological) algebras under adaption of the stability property, say. Another advantage is that it is more elementary than Kasparov's original defi- nition. Its definition is also clearer motivated by its relative naturality, whereas the definition of the original KK-theory appears highly unmotivated at first (without further background like the Atyiah -- Singer index theory). Also Cuntz's picture of KK-theory by quasi isomorphisms in [7] appears still rather technical and difficult. A disadvantage of GK-theory is that the Kasparov product is not computed. It remains a formal, uncomputed product f · g. On the other hand, this makes GK- theory also easy, again. Also, the general construction of the Kasparov product in KK-theory uses the indirect, unexplicit axiom of choice. In concrete computations the product has to be guessed, which is rather difficult. We are going to briefly recall GK-theory. For more details see [3]. Definition 3.1 (C∗-category C∗). Let G be a second-countable locally compact group, or a discrete countable inverse semigroup. Denote by C∗ the category with objects being the C∗-algebras equipped with an action by G, and morphisms being the G-equivariant ∗-homomorphisms. If nothing else is said, we could also allow that G is another equivariance-inducing object like a general topological group, or a groupoid, or a category, or a semigroup and so on. Definition 3.2 (Synthetical morphisms). We introduce two types of synthetical morphisms. (a) For each corner embedding c ∈ C∗(A, A ⊗ K), that is a map defined by c(a) = a ⊗ e for a one-dimensional projection e ∈ K (where the G-action on A ⊗ K need not be diagonal but may be any) introduce one synthetical morphism (inverse map, localization) c−1 : A ⊗ K → A. (b) For each short split exact sequence (2) S : 0 / A i / D f s / B 0 in C∗ introduce one synthetical morphism P −1 localization). S : D → A ⊕ B (inverse map, Definition 3.3 (Preadditive Category W ). Let W be the preadditive category with object class Obj(C∗). The morphism class W (A, B) from object A to object B let be the collection of all formal expressions ± a11a12 · · · a1n1 ± · · · · · · ± ak,1ak,2 · · · ak,nk , (3) where each letter aij is either a morphism in C∗ or one of the synthetical morphisms c−1 or P −1 of Definition 3.2. Each ± stands here either for a single +-sign or a single −-sign. S We think of a word ai1 · · · ai,ni as a composition of morphisms (=arrows) aij going from the left to the right with start point A and end point B, that is, as a / / / / / o o SOME REMARKS IN C ∗- AND K-THEORY 5 picture A = Ai1 ai1 / Ai2 ai2 / Ai3 ai3 / · · · ai,ni/ / Ai,ni = B for objects Aij . We require here that the range object Ai,j+1 of the morphism aij coincides with the source object of the morphism ai,j+1 for all ij. Composition and addition of morphisms in W is given formally (i.e. freely). That is, we add and multiply morphisms of the from (3) like in a ring by using the distributive law. Definition 3.4 (GK-theory). The category GK is defined to be additive cate- gory which comes out when dividing the preadditive category W by the following relations: (a) The canonical assignment C∗ → GK is a functor, i.e. we require f g = g ◦ f in GK(A, C) for all elements f ∈ C∗(A, B) and g ∈ C∗(B, C). (b) The category GK is additive, i.e. we require pAiA + pBiB = 1A⊕B in GK(A ⊕ B, A ⊕ B) for all natural diagrams A (canonical injections and projections) in C∗. iA / pA / A ⊕ B pB iB B (c) The category GK is homotopy invariant, that is, every pair of homotopic G-equivariant ∗-homomorphisms f0, f1 : A → B (homotopic within C∗) satisfies the identity f0 = f1 in GK. (d) The category GK is stable, that is, every corner embedding c is invertible in GK with inverse c−1 as introduced in Definition 3.2.(a). (e) The category GK is split exact, that is, for every split exact sequence (2) in C∗ the morphism PS := pAi + pBs in the following diagramm A ⊕ B (4) A pA =④④④④④④④④④④④④④④④④④ }④④④④④④④④④④④④④④④④④ PS iA −1 i tS pB !❈❈❈❈❈❈❈❈❈❈❈❈❈❈❈❈❈ a❈❈❈❈❈❈❈❈❈❈❈❈❈❈❈❈❈ iB f PS D B s is invertible in GK with inverse P −1 as introduced in Definition 3.2.(b). (Here, pA, pB, iA, iB are the canonical projections and injections, and the dotted arrow tS may be ignorred here.) S The category GK is just another model for Kasparov's KK G-theory: Proposition 3.5 ([3]). Let G be a locally compact second-countable group, or a discrete countable inverse semigroup. Let C∗ be restricted to the subcategory of separable C∗-algebras. Then, the categories KK G and GK are isomorphic. Proof. Almost evident as KK G-theory and GK-theory are characterized by the same universal property. See [3, Theorem 5.1] for more details. (cid:3) In this section we are going to show that expression (3) of a morphism in GK may be considerably simplified. A first simplification will be reduction of sum, where the notion word is defined in Definition 3.3: / / / / / o o o o } ! / / = / / o o O O o o a 6 B. BURGSTALLER Lemma 3.6. In GK we may rewrite any plus-signed sum x1 + . . . + xn of words xi as a single word x. In particular, any morphism in GK is presentable as a difference x − y of some words x, y ∈ GK. Proof. By induction, it clearly suffices to show that any sum x + y of two words x, y ∈ GK is presentable as a single word. Assume that we have given a split exact sequence S, see (2), for which we consider ϑ := PS = pAi + pBs ∈ GK(X, Y ) of Definition 3.4. Define (ϑ ⊕ idX ) : X ⊕ X → Y ⊕ X : ϑ ⊕ idX := pAi ⊕ idX + pBs ⊕ 0X = (pA ⊕ idX )(i ⊕ idX ) + (pB ⊕ 0X )(s ⊕ 0X ). Notice that ϑ ⊕ idX is just PT for the split exact sequence T : 0 / A ⊕ X i⊕idX/ / D ⊕ X f ⊕0 s⊕0 / B 0 . Consider the canonical projections and embeddings X i1 / p1 / X ⊕ X p2 i2 / X, Y p′ 1 Y ⊕ X p′ 2 X . Set ϑ−1 := P −1 S . Then observe that p1i1 = (ϑ ⊕ idX )p′ 1ϑ−1i1, p2i2 = (ϑ ⊕ idX )p′ 2i2, so that with p1i1 + p2i2 = idX⊕X we get (5) (6) (ϑ ⊕ idX )−1 = p′ 1ϑ−1i1 + p′ 2i2, (ϑ ⊕ idX ) = (idX⊕X )(ϑ ⊕ idX ) = p1ϑi1 + p2i′ 2. If we have given a corner embedding ϑ := c ∈ C∗(X := A, Y := A ⊗ K) then we set (ϑ ⊕ idX ) : X ⊕ X → Y ⊕ X obvious and get again relations (5) and (6). Notice that in this case (ϑ ⊕ idX )−1 is just the word (idA⊗K ⊕ e)d−1 for the corner embeddings d ∈ C∗(A ⊕ X, A ⊗ K ⊕ X ⊗ K) and e ∈ C∗(X, X ⊗ K). By some abuse of notation, in the sequel we shall omit notating the primes in p′ 1 and p′ 2 and simply write p1 and p2 instead. In other words, we shall not indicate the involved spaces X and Y in our notation, even when we are going to have different spaces. As already above, the index 1 will mean projection or embedding on the first (left hand sided) coordinate, and 2 on the second (right hand sided) coordinate. Let us be given two words xε1 m in GK(X, Y ), where xi ∈ GK(Xi, Xi+1) and yj ∈ GK(Yj , Yj+1) are either morphisms in C∗ or morphisms PS, and let εi, ǫj ∈ {1, −1} present exponents in case letters are invertible by synthetical inverses as defined in Definition 3.2. The expression x1 S is not allowed, because PS can be expressed by morphisms in C∗. n and yǫ1 1 . . . yǫm 1 . . . xεn i = P 1 Let j : X → X ⊕ X be defined by j(x) = (x, x). Let d : Y ⊕ Y → M2(Y ) be the diagonal embedding d(x, y) = (cid:18)x 0 embedding k(x) =(cid:18)x 0 y(cid:19) and k : B → M2(Y ) the corner 0(cid:19). Using the identities (5) and (6) and their analogs, and 0 0 / / / / o o o o / o o / / o o SOME REMARKS IN C ∗- AND K-THEORY 7 the orthogonality relations i2p1 = 0 and i1p2 = 0, the following computation shows our claim. Simply consider the word j(x1 ⊕ idX )ε1 · · · (xn ⊕ idX )εn (idX ⊕ y1)ǫ1 · · · (idX ⊕ ym)ǫmdk−1 = j(p1xε1 i1 + p2i2) · · · (p1xεn n i1 + p2i2) ·(p1i1 + p2yǫ1 = j(p1xε1 · · · xεn = xε1 · · · xεn 1 i2) · · · (p1i1 + p2yǫm n i1 + p2yǫ1 1 · · · yǫm m , 1 · · · yǫm n + yǫ1 m i2)dk−1 m i2)dk−1 where for the last identity we have used that the ∗-homomorphism i2d is homotopic to the ∗-homomorphism i1d by rotation, and i1dk−1 = idY . (cid:3) Instead of the split exactness axiom in the definition of GK we may use alter- natively the following axiom without difference. Lemma 3.7. Instead of introducing the synthetical arrows P −1 in Defintion 3.2.(b) and using axiom 3.4.(e) we may alternatively introduce the dotted arrow tS for each split exact sequence (2) and the axiomatic relations S itS = 1A, tSi + f s = 1D (as a replacement of Definition 3.4.(e)) without changing the definition of GK. It would not make any difference in the definition of GK if we added both P −1 and tS simultaneously, because they automatically define each other as follows in GK: S Lemma 3.8. P −1 S and tS of diagram (4) define each other as follows: tS = P −1 S pA, P −1 S = tSiA + f iB Proof of Lemmas 3.7 and 3.8. Let GK be the category with the usual split exact- ness axiom involving PS, and GK ′ the category with the alternative split exactness axiom involving tS. Let Φ : GK → GK ′ and Ψ : GK ′ → GK be the functors which are identical on C∗ and on the synthetical inverses of corner embeddings, and according to the 'transformation' rules defined to be Φ(P −1 S ) = tSiA + f iB, Ψ(tS) = P −1 S pA for each split exact sequence S. We remark that stS = 0 because stS = stSitS = s(1 − f s)tS = 0. To see that Φ is well-defined we compute Φ(PS )Φ(P −1 S ) = (pAi + pBs)(tS iA + f iB) = 1A⊕B, Φ(P −1 S )Φ(PS ) = 1D. To show that Ψ is well-defined we calculate Ψ(i)Ψ(tS) = iP −1 S pA = iApAiP −1 S pA = iA(PS − pBs)P −1 S pA = iApA = 1A, Ψ(tS)Ψ(i) + Ψ(f )Ψ(s) = P −1 S (PS − pBs + PS f s) = 1D. That Ψ and Φ are inverses to each other follows then from the observation S pAi + f s = P −1 Ψ ◦ Φ(P −1 S ) = P −1 S (pAiA + PS f iB) = P −1 S , Φ ◦ Ψ(tS) = tS. (cid:3) 8 B. BURGSTALLER We remark that we have also shown in the last proof that stS = 0. (That shows even more more clearly that D ∼= A ⊕ B in GK.) Again, the element tS is uniquely defined by its defining relations. Also, Lemma 3.6 would hold if we had introduced tS instead of P −1 S . All these follows immediately as a corollary from the formula tS = P −1 S pA of Lemma 3.8. We can always move the inverse c−1 of a corner embedding c ∈ C∗ to the right in a word: Lemma 3.9. If f is a morphism in C∗, c a corner embedding and the composition c−1f admissible, then there exists a corner embedding c′ and a morphism f ′ in C∗ such that c−1f = f ′c′−1. (Analogously, c−1tS = tS ′ c′−1. Similarly, c−1P −1 S = P −1 S ′ ϕ−1c′−1, where ϕ is the canonical isomorphism (A ⊕ B) ⊗ K → A ⊗ K ⊕ B ⊗ K.) Proof. This follows from the commutation relation c(f ⊗ idK) = f c′ for the corner embeddings c : A → A⊗K and c′ : B → B⊗K and a morphism f : A → B. The case P −1 is analog: since K is an exact C∗-algebra we can tensor the diagrams (2) and (4) with K, then check PSc = c′ϕPS ′ , where c : D → D⊗K, c′ : A⊕B → (A⊕B)⊗K and S′ = S ⊗ K (also with additivity, Definition 3.4.(b)). The case tS follows from that and tS = P −1 (cid:3) S pA of Lemma 3.8. S A drastical simplification of morphisms in GK goes by the Cuntz picture: Proposition 3.10. Let G be a locally compact second-countable group or a count- able inverse semigroup and the category C∗ be restricted to separable C∗-algebras. Every morphism z in GK may be written in the form z = (ad−1 − b) · tS etT · c−1 for some homomorphisms a, b ∈ C∗, some split exact sequences S and T , and some corner embeddings c, d, e ∈ C∗. If the morphism z is in GK(A, B) and B is unital we can omit tT (i.e. tT = 1). If G is the trivial group then d−1 and e can be omitted (i.e. d−1 = e = 1). Both simplifications can be combined simultaneously. Proof. By the universal property of KK G and GK there is an isomorphism of categories G : KK G → GK, see Proposition 3.5. The idea is now to keep track of the formulas appearing in the proof of this fact and see how a morphism z ∈ KK G is presented as G(z) in GK. The original proof of the universal property of KK is by Cuntz [6] and Higson [8], and by Thomsen [14] in the group equivariant setting for KK G. We shall refer here to our exposition in the inverse semigroup equivariant setting [4]. All we shall do here may be read verbatim topological group equivariantly. Let us be given fixed objects A, B ∈ C∗. Assume at first that B is stable, i.e. B ∼= B ⊗ K in C∗ (K equipped with the trivial G-action). In [4, Theorem 8.5], there is stated an isomorphism Φ : FG(A, B) → KK G(A, B). Here, FG(A, B) is just the Cuntz-picture of G-equivariant KK-theory by quasi homomorphisms and G-cocycles, see [4, Def. 7.1 and Def. 7.8]. To recall it, an element x = [ϕ+, ϕ−, u+, u−] ∈ FG(A, B) is given by two G-equivariant ∗- homomorphisms ϕ± : A → M(B) and two α-cocycles u± : G → M(B), see [4, Def. 5.1]. SOME REMARKS IN C ∗- AND K-THEORY 9 One has two split-exact sequences (for + and −) S : 0 / B j / Ax p s±o / A 0 for Ax := {A ⊕ M(B) ϕ+(a) = m mod B} by [4, Def. 9.1 and 9.4]. Define the split-exact, homotopy invariant, stable functor F from C∗ to the abelian groups by F (B) = GK(A, B) and F (f : B → C) : GK(A, B) → GK(A, C) : z 7→ zf. For an α-cocycle u ∈ M(A), recall [4, Def. 5.4, 6.1 and 6.2] for the definition of an abelian group isomorphism u# = F (Tu,A)−1 ◦ F (Su,A) : F (A, α) → F (A, uαu∗) and corner embeddings Su,A, Tu,A : A → M2(A, δu). As in [4, Def. 9.4], define an abelian group homomorphism −1 # ◦ F (j)−1 ◦(cid:0)u# ◦ F (s+) − F (s−)(cid:1) (7) Ψx : F (A) → F (B) : Ψx = u− (here u is the cocycle for Ax of [4, Def. 9.1]!). Now assume that B is not necessarily stable. In [4, Def. 10.2] there appears a similar variant Ψ′ z : F (A) → F (B) : Ψ′ z = F (cB)−1 ◦ F (jB)−1 ◦ ΨjB ∗cB ∗(Φ−1(z)) of Ψx, where z ∈ KK G(A, B). Here cB : B → B ⊗ K is the corner embedding, see [4, Def. 10.1], and jB appears in some split exact sequence T : 0 / B ⊗ K jB / / B+ ⊗ K pB / / C∗(E) ⊗ K / 0 in [4, Def. 10.2]. The stars in jB ∗ and cB ∗ are defined in [4, Def. 8.6]. By [4, Def. 11.1] there is a natural transformation ξ : KK(A, −) → F (−) : ξB(z) = Ψ′ z(1GK(A,A)). We are now applying [4, Thm. 1.3] (= [4, Thm. 12.4]) to the canonical quotient functor G : C∗ → GK, which is split-exact, homotopy invariant and stable. The claim and proof of [4, Thm. 12.4] show that there is a functor G : KK G → GK defined by G(z) = ξB(z) for all z ∈ KK G(A, B) such that G factorizes over G (i.e. G = G ◦ G2 for the canonical quotient functor G2 : C∗ → KK G). This functor is an isomorphism, since GK itself has the universal properties of KK G, confer [3, 5.1]. In details we get G(z) = ξB(z) = Ψ′ z(1GK(A,A)) = F (cB)−1 ◦ F (jB)−1 ◦ ΨjB ∗cB ∗(Φ−1(z))(1GK(A,A)). Now observe that for the corner embedding cB, the inverse map F (cB)−1 is just realized by right multiplication with the synthetical inverse c−1 B in GK. Similarly, according to the split-exactness of GK the (one-sided) inverse map F (jB)−1 is just right multiplication with the synthetical (one-sided) inverse tT . We choose now the x from above as x := jB ∗cB ∗(Φ−1(z)) ∈ FG(A, B+ ⊗ K) and put formula (7) into the formula of G(z). Here, z is the given morphism in KK G that we want to present in GK via G. Then we have G(z) = F (cB)−1 ◦ F (jB)−1 ◦ u− # ◦ F (j)−1 ◦(cid:0)u# ◦ F (s+) − F (s−)(cid:1) (1GK(A,A)) −1 / / / / / o / / 10 B. BURGSTALLER = 1GK(A,A) · (s+Su,AT −1 u,A − s−) · tSTu−,AS−1 u−,A · tT c−1 B = (ad−1 − b) · tSetU · f −1c−1 in GK(A, B) by Lemma 3.9 for suitable homomorphisms a, b ∈ C∗, corner embed- dings c, d, e and split-exact sequence U. If B is unital we can omit jB in the definition of Ψ′ satisfy u = 1 and thus all u# = 1. z. If G is trivial all cocycles (cid:3) It is however rather difficult to bring a product of such standardized elements as in Proposition 3.10 again to such a standard form, see Cuntz [6]. It is not really easier than forming the Kasparov product of Kasparov cycles. Remark 3.11. A further slight simplification of the split exactness axiom could be done by observing that the split exact sequence (2) is isomorphic in C∗ to an idempotent ∗-homomorphisms P : D → D (translation is P = f s). Then split exactness just says that every idempotent P ∈ C∗ has an orthogonal split tS : D → ker(P ) in GK (orthogonal projection: tSi = 1D − P ). 4. E-continuity and Hausdorff property In this section we shall see that the groupoid associated to an inverse semigroup is Hausdorff if and only if the inverse semigroup is E-continuous. This condition is technically easier and more intrinsic to the inverse semigroup. We shall see that E-continuity is a necessary and sufficient condition to define a non-degenerate C0(X)-compatible C0(X)-valued L2(G)-module. Let G be a discrete inverse semigroup. Definition 4.1 (E and X). Let E denote the subset of idempotent elements of G. The free universal abelian C∗-algebra C∗(E) generated by the commuting self- adjoint projections of E has a totally disconnected Gelfand spectrum X. That is we have C∗(E) ∼= C0(X). Under this isomorphism we identify E as a subset of C0(X) (under the formula e(x) = x(e)). To this end, we also use the suggestive notation 1e ∈ C0(X) for the corresponding element of e ∈ E in C0(X). We write "x ∈ e" for x ∈ X and e ∈ E iff x is an element of the support of 1e ∈ C0(X) (also denoted by carrier(1e)). For e, f ∈ E we use the usual order e ≤ f in a C∗-algebra. This order can be extended to G by saying that g ≤ h for g, h ∈ G iff g = hg∗g (or equivalently iff g = gg∗h). Definition 4.2 (G-action). In this note we understand under a G-action on a C∗- algebra A a semigroup homomorphism α : G → End(A) such that αe(a)b = aαe(b) (compatibility) for all e in E. In this case, A is called a G-algebra. A G-action on a Hilbert A-module E is a semigroup homomorphism U : G → LinMaps(E) (linear maps) such that Ue is an adjoint-able operator for all e ∈ E, and hUg(ξ), Ug(η)i = g(hξ, ηi), Ug(ξa) = Ug(ξ)αg(a), Ue(ξ)a = ξαe(a) (the last identity being called compatibility or C0(X)-compatibility of U ) for all ξ, η ∈ E, a ∈ A, g ∈ G and e ∈ E. Then E is called a (compatible) G-Hilbert A- module. Often we write the G-action in the form g(ξ) := Ug(ξ) and g(a) := αg(a). Definition 4.3 (G-action on X). The C∗-algebra C0(X) is equipped with the G- action g(1e) := 1geg∗ for e ∈ E, g ∈ G. This G-action may be extended to the bigger C∗-algebra ℓ∞(X) by setting (g(f ))(x) := 1{x·g6=0}f (x · g) for g ∈ G, f ∈ ℓ∞(X) SOME REMARKS IN C ∗- AND K-THEORY 11 and characters x ∈ X, where the (possibly zero) character x · g : C∗(E) → C is defined by (x · g)(e) = x(geg∗) for all e ∈ E. We are going to recall the E-continuity property of an inverse semigroup. For more details see [2]. In the next few paragraphs (until Lemma 4.7) we shall identify elements e ∈ E with their corresponding characteristic functions 1e in C0(X). Write Alg∗(E) for the dense ∗-subalgebra of C0(X) generated by the characteristic : X → C for the pointwise functions 1e for all e ∈ E. Moreover, write Wi fi W{e ∈ E e ≤ g} ∈ CX (in precise notation: W{1e ∈ C0(X) e ∈ E, e ≤ g} ∈ CX ) Definition 4.4. An inverse semigroup G is called E-continuous if the function is a continuous function in C0(X) for all g ∈ G. supremum of a family of functions fi : X → C. A simple compactness argument shows the following, see [2]: Lemma 4.5. An inverse semigroup G is E-continuous if and only if for every g ∈ G there exists a finite subset F ⊆ E such that W{e ∈ E e ≤ g} =W{e ∈ F e ≤ g}. Definition 4.6 (Compatible C0(X)-valued L2(G)-module). Let G be an E- continuous inverse semigroup. Write c for the linear span of all functions ϕg : G → C (in the linear space CG) defined by ϕg(t) := 1{t≤g} (characteristic function) for all g, t ∈ G. Endow c with the G-action g(ϕh) := ϕgh for all g, h ∈ G. Turn c to an Alg∗(E)-module by setting ξe := e(ξ) for all ξ ∈ c and e ∈ E. Define an Alg∗(E)-valued inner product on c by (8) hϕg, ϕhi := _{e ∈ E eg = eh, e ≤ gg∗hh∗}. The norm completion of c is a G-Hilbert C0(X)-module denoted by bℓ2(G). Lemma 4.7 ([2]). The vectors (ϕg)g∈G ⊆ bℓ2(G) are linearly independent. We recall the well-known topological groupoid associated to an inverse semigroup by Paterson [13]: Definition 4.8 (Groupoid associated to an inverse semigroup). Let G be a discrete inverse semigroup and X the Gelfand spectrum of C∗(E). Consider the topological subspace G ∗ X = {(g, x) ∈ G × X g ∈ G, x ∈ g∗g} of the topological space G × X (product topology with G having the discrete topology). Two points (g, x), (h, y) in G ∗ X are called equivalent, also denoted (g, x) ≡ (h, y), iff x = y and ge = he for some e ∈ E with x ∈ e. Let π : G∗X → G∗X/ ≡ denote the set-theoretical quotient map. The quotient is a groupoid under the multiplication: π(g, x)π(h, y) = π(gh, y) if and only if for all e ∈ E such that y ∈ e one has x ∈ (he)(he)∗. Otherwise the composition is declared to be undefined. We now regard the quotient G ∗ X/ ≡ as a topological groupoid under the quotient topology and call it the groupoid asscociated to the inverse semigroup G. (Recall that a subset Y ⊆ G ∗ X/ ≡ is declared to be open if and only if π−1(Y ) is open.) Usually the groupoid associated to G is a non-Hausdorff topological space. We are going to prove that the Hausdorff condition is equivalent to E-continuity of G. 12 B. BURGSTALLER Lemma 4.9. The sets of the form π(g × U ), where g ∈ G and U ⊆ X is an open subset of X with U ⊆ carrier(g∗g), are open and generate the topology of G ∗ X/ ≡. (Here g × U := {g} × U .) Proof. We claim that the inverse π−1(π(g×U )) is open. Indeed if (h, x) ∈ π−1(π(g× U )) then it is equivalent to some (g, x) ∈ g × U . Hence there exists some e ∈ E with x ∈ e and he = ge. Let V = carrier(e) ∩ U ∩ carrier(h∗h). Then h × V is an open subset of π−1(π(g × U )) containing (h, x). If π−1(O) is open and contains the point (g, x) together with its open neighbor- hood g × U then π−1(π(g × U )) ⊆ π−1(O). Thus π(g × U ) ⊆ O. Hence such sets generate the topology. (cid:3) We call π(g × U ) the open set in G ∗ X/ ≡ generated by g × U . Lemma 4.10. If G is E-continuous then its associated groupoid is Hausdorff. Proof. Let (g, x), (h, x) ∈ G ∗ X be two points such that (g, x) 6≡ (h, x). Then for all e ∈ E with x ∈ e and e ≤ g∗gh∗h one has ge 6= he, and so e 6≤ h∗g. Since G is E-continuous the function F :=Wf ∈E,f ≤h∗g f is continuous. Note that x /∈ F . Let t ⊆ X be the (open!) complement of the carrier of F . Consider Ug := {g} × t ∩ carrier(g∗g) and Uh := {h} × t ∩ carrier(h∗h). Clearly x ∈ t and so (g, x) ∈ Ug and (h, x) ∈ Uh. Consider the open subsets Wg and Wh that Ug and Uh generate in G ∗ X/ ≡. Assume Wg and Wh would intersect. Then there are (g, y) ∈ Ug, (h, z) ∈ Uh such that (g, y) ≡ (h, z). That is, there is a e ∈ E such that y = z ∈ e, e ≤ g∗gh∗h and ge = he. Hence y ∈ e ≤ F . By definition of Ug one has also certainly y ∈ t. A contradiction. This shows that Wg and Wh are disjoint neighborhoods which separate (g, x) and (h, x). (cid:3) Lemma 4.11. If its associated groupoid is Hausdorff then G is E-continuous. Proof. Let g ∈ G. Assume the projection F :=Wf ∈E, f ≤g f would be discontinuous, say in the point x ∈ X. Then for any neighborhoods U of x there is at least one f ≤ g (f ∈ E) such that U has nonempty intersection with the carrier of f . On the other hand x is not in the carrier of any f ∈ E with f ≤ g, because there F is continuous. Consider the points (g, x) and (g∗g, x) in G ∗ X. They must be distinct in the quotient G ∗ X/ ≡ because assuming to the contrary the existence of some e ∈ E with x ∈ e and g∗ge = ge would imply g∗ge ≤ g; a contradiction to what we said above. Let U ⊆ carrier(g∗g) ⊆ X be an open neighborhood of x. Consider the open neighborhoods Wg and Wg∗g in G ∗ X/ ≡ generated by {g} × U and {g∗g} × U . As remarked above we may choose y ∈ U, f ∈ E such that y ∈ f and f ≤ g. Then (g, y) and (g∗g, y) are equivalent because y ∈ f and g∗gf = gf . Hence Wg, Wg∗g intersect. Hence (g, x) and (g∗g, x) cannot be separated. Con- (cid:3) tradiction. Corollary 4.12. An inverse semigroup is E-continuous if and only if its associated groupoid is Hausdorff. We have seen in Definition 4.6 that for E-continuous inverse semigroups there exist non-degenerate compatible L2(G)-modules with coefficients in C0(X). The SOME REMARKS IN C ∗- AND K-THEORY 13 next example indicates that we cannot construct such L2(G)-modules for E- discontinuous inverse semigroups. Before that, for the discussion of another L2(G)-module, we recall the following discretized coefficient algebra ε(E) of C0(X). Definition 4.13 (Discretized coefficient algebra ε(E) of C0(X)). Recall that there exists a map ǫ : E → X assigning to each e ∈ E the character ǫe on C∗(E) determined by the formula ǫe(f ) = 1{f ≥e} for every f ∈ E. The image ǫ(E) is dense in X, see [13] or [10, 3.2]. We have a G-invariant sub-C∗-algebra ε(E) := c0(cid:0)ǫ(E)(cid:1) ⊆ ℓ∞(X) (complex-valued functions on the image of ǫ vanishing at infinity). Given e ∈ E, we write εe for the characteristic one-point supported function 1{ǫe} ∈ ε(E) ⊆ ℓ∞(X). One checks that G acts through g(εe) = εgeg∗ if e ≤ g∗g, and g(εe) = 0 otherwise. Example 4.14 (Elementary abelian E-discontinuous example). Let us discuss one of the most simplest examples of an (even abelian) inverse semigroup G which is not E-continuous. Let G = {1, S, e1, e2, e3, . . .} consist of an identity element 1, a strictly increasing sequence of projections e1 < e2 < e3 < . . . < 1, and a symmetry S 6= 1 (i.e. S2 = 1, S∗ = S) such that Sen = enS = en for all n ≥ 1. (A concrete representation of G on a direct sum Hilbert space H ⊕ H may be given as 1 = idH ⊕ idH , S = s ⊕ idH with s a symmetry and en ≤ 0 ⊕ idH .) The associated C∗-algebra C∗(G) is an AF-algebra. Indeed it is the union of its finite-dimensional sub-C∗-algebras An generated by {1, S, e1, . . . , en}. One has An ∼= Cn+2 for all n ≥ 0. The two generating projections of A0 ∼= C2 are (1 ± S)/2. The projection (1 − S)/2 is orthogonal to all projections en, and en < (1 + S)/2. Hence K0(C∗(G)) =LN But K0(ε(E) ⋊ G) = LN projections εen ⋊ en. Z ⊔ {1} (here 1 denotes an adjoint unit). algebras Bn generated by ε1 ⋊ 1, ε1 ⋊ S, εe1 If we compare this with ε(E) ⋊ G then we have that it is the union of the sub-C∗- ⋊ en. Again Bn ∼= Cn+2. Z as the projection ε1 ⋊ (1 + S)/2 is orthogonal to all ⋊ e1, . . . , εen We are now coming to the most important point, namely that it appears not possible to construct a non-degenerate C0(X)-compatible C0(X)-valued L2(G)- module. Somehow we should have some sort of generators δ1, δS, δe1 , . . . , δen , . . . of the module. The G-action should be g(δh) = δgh to be regarded as an L2(G)-module. By compatibility of the module product we naturally have δge = δg · e(1) = e(δg) · 1 = δeg for e ∈ E ⊆ C0(X). Naturally we should choose hδen , δen i = en for the inner product. By compatibility of the inner product we have hδS, δSien = hδSen, δSeni = en for all n ≥ 1. Consequently C0(X) ∋ hδS, δSi = 1 (because the carriers of the elements e ∈ E generate the topology of X) and simi- larly hδS, δ1i = hδ1, δ1i = 1. But then kδ1 − δSk = 0 and the module degenerates. Let us discuss another module. We may construct the non-degenerate ε(E)- valued L2(G)-module of [1, Def. 5.5]. The generators are the characteristic func- tions δg : G → C with δg(h) = 1{g=h} for g, h ∈ G. The G-action is given by h(δg) = 1{h∗h≥gg∗}δhg. The inner product is determined by hδg, δhi = 1{g=h}, hδp, δpi = εp for the projections p in G, and hδS, δSi = ε1. The module product computes as δpεq = 1{p=q} for projections p, q, and δSε1 = δS and δSεen = 0. Example 4.15 (Dense E-discontinuity example). In Example 4.14 we had some kind of E-discontinuity only at S (or we may say at 1). We may construct such an 14 B. BURGSTALLER E-discontinuity at every e in E by the same method. Start with a given inverse semigroup G = E consisting only of projections. Adjoin to G for every e in E a symmetry Se such that Sef = f Se = f for all f < e. Other relations we do not add. The resulting inverse semigroup G is E-discontinuous in those Se in the sense f is discontinuous where e has no precursor f < e. If no element of E has a precursor in E then the E-discontinuity points are dense in X (at the points ǫe we may say, which form a dense subset of X). that Wf ∈E,f ≤Se Example 4.16 (Finitely presented E-discontinuous inverse semigroup). A finitely presented E-discontinuous inverse semigroup may be defined as follows. Consider the finitely presented inverse semigroup G = ht, l, e tl = lt, t∗l = lt∗, te = e, t∗e = ei. That is t and t∗ commute with l and l∗, and e absorbs t and t∗. Between l and e we have no relations, they are free in G, so that we get infinitely many distinct projections p0 := e, p1 := lee∗l∗, p2 := llee∗l∗l∗, . . . , pn := lnee∗l∗n, . . . in G. Now tpn = pn by the defining relations of G. The projections pn cannot be compared among each other, i.e. pn ≤ pm implies n = m. Hence the criterion for E-continuity of Lemma 4.5 fails for t, as the supremum of {e ∈ E e ≤ t} will not be attained at a finite set of projections of E. To see this, let us first note that we have no single projection q ∈ E such that t ≥ q ≥ p0, p1, p2, . . .. Indeed, every such projection q would require to include the letter e to obtain t ≥ q, and consequently any letter t or t∗ in q would be absorbed by e. So q would allow a presentation with letters l and e and their adjoints only, and such a q ≥ p1, p2, p3, . . . as required does not exist. One can similarly argue that we also cannot choose q1, . . . , qn ∈ E such that t ≥ q1 ∨ . . . ∨ qn ≥ p0, p1, p2, . . .. So by Lemma 4.5 we get that G is not E-continuous. Remark 4.17 (Baum -- Connes map for inverse semigroups). In [2] we have tried to define a Baum -- Connes map for inverse semigroup crossed products parallel to the method of Meyer and Nest in [11] for group crossed products, which automat- ically would include some theoretical method to compute the left hand side of the Baum -- Connes map. On that way, C0(X)-compatible Hilbert modules and their KK-theory appeared the better choice than the corresponding, C0(X)-structure ignorring incompatible tools. Thus C0(X) is the natural coefficient algebra. But since L2(G)-spaces are in the center and the core of any Baum -- Connes theory, and Example 4.14 shows that a compatible C0(X)-valued L2(G)-module requires E-continuity of G, it appears not possible to overcome the E-discontinuity barrier when defining a Baum -- Connes map, at least not with the known (group) L2(G)- space methods. That is, as soon as the associated groupoid of G is non-Hausdorff the method fails. More generally, Tu [15] has tried to develop a Baum -- Connes theory for non-Hausdorff groupoids, and came to the same conclusion that for non- Hausdorff groupoids the known methods fail, even one may be able to formally write down the Baum -- Connes map also for non-Hausdorff groupoids. Remark 4.18 (Baum -- Connes theory for discreticized crossed products). Whereas we have no approach to handle the K-theory of a crossed product A ⋊ G for an inverse semigroup G, we have a Baum -- Connes map and additionally at least the- oretically an approach to treat the K-theory of (ε(E) ⊗C0(X) A) ⋊ G by [1]. Even SOME REMARKS IN C ∗- AND K-THEORY 15 though the K-theories of the latter two crossed products are obviously different in general, they might have some aspects in common in certain good interesting cases as the latter two crossed products are also similar. For example, if G = E consists only of projections then both C0(X) ⋊ E and ε(E) ⋊ E are the direct limit of canonically ∗-isomorphic finite dimensional sub-C∗-algebras. Only the direct limit embedding maps are different in both cases. Hence their K-theories are still similar. For example, in both cases the K0-groups are infinitely generated and the K1 groups are zero if E is infinite. That is, being infinitely generated is a common quality of the K-theory of both crossed products. Example 4.19. That being said, let us remark that the discretized crossed product and the usual crossed product may however also be rather distinct. Write for example the Cuntz algebra On as the inverse semigroup crossed product On ∼= A⋊G (Sieben's crossed product, which is the universal crossed product subject to the relations e(a) ⋊ g ≡ a ⋊ eg for all a ∈ A, e ∈ E, g ∈ G), where G is defined to be the inverse semigroup G ⊆ On generated by the standard generators S1, . . . , Sn of the Cuntz algebra, and A ⊆ On denotes the smallest G-invariant C∗-subalgebra of the Cuntz algebra generated by the identity 1 ∈ On under the (incompatible) G-action g(a) = gag∗ for a ∈ On, g ∈ G. Note that A is the commutative G-algebra (in the sense of Definition 4.2) generated by the elements of the form gg∗ for g ∈ G. The isomorphism is ϕ : On → A ⋊ G : ϕ(Si) = 1 ⋊ Si. Then we have that 0 = (ε(E) ⊗C0(X) A) ⋊ G 6= A ⋊ G = On, because in the left hand sided crossed product we have 1 + . . . + SnS∗ (ε1 ⊗ 1) ⋊ 1 =(cid:0)ε1 ⊗ (S1S∗ n)(cid:1) ⋊ 1 = 0 i (ε1) = 0 (action of SiS∗ as SiS∗ i on ε1) for all i, and by similar reasoning (εe ⊗ 1) ⋊ 1 = 0 for all e ∈ E. We see thus that the discretized crossed prodcut is not an approximation of the crossed product A ⋊ G at all as it collapses to zero. (As already the discretized coefficient algebra ε(E) ⊗C0(X) A is zero). Still the K-theory of both crossed products is finitely generated. But this need not be in general true, as we may replace A by an infinite sum of copies of A, and so K0((LN A) ⋊ G) =LN K0(A ⋊ G) is infinitely generated whereas the K-theory of the discretized crossed product is an infinite sum of zeros, so zero and thus finitely generated. References [1] B. Burgstaller. A note on a certain Baum -- Connes map for inverse semigroups. preprint. arXiv:1609.01913. [2] B. Burgstaller. Attempts to define a Baum -- Connes map for inverse semigroups via localization of categories. preprint. arXiv:1506.08412. [3] B. Burgstaller. The generators and relations picture of KK-theory. preprint. arXiv:1602.03034v2. [4] B. Burgstaller. The universal property of inverse semigroup equivariant KK-theory. preprint. arXiv:1405.1613v2. [5] B. Burgstaller. Semigroup homomorphisms on matrix algebras. Adv. Operat. Th., 2(3):287 -- 292, 2017. [6] J. Cuntz. K-theory and C ∗-algebras. Algebraic K-theory, number theory, geometry and anal- ysis, Proc. int. Conf., Bielefeld/Ger. 1982., Lect. Notes Math. 1046, 55-79 (1984)., 1984. [7] J. Cuntz. A new look at KK-theory. K-Theory, 1(1):31 -- 51, 1987. [8] N. Higson. A characterization of KK-theory. Pac. J. Math., 126(2):253 -- 276, 1987. 16 B. BURGSTALLER [9] G.G. Kasparov. Equivariant KK-theory and the Novikov conjecture. Invent. Math., 91(1):147 -- 201, 1988. [10] M. Khoshkam and G. Skandalis. Regular representation of groupoid C ∗-algebras and appli- cations to inverse semigroups. J. Reine Angew. Math., 546:47 -- 72, 2002. [11] R. Meyer and R. Nest. The Baum-Connes conjecture via localisation of categories. Topology, 45(2):209 -- 259, 2006. [12] L. Moln´ar. *-semigroup endomorphisms of B(H). In Recent advances in operator theory and related topics. The B´ela Szokefalvi-Nagy memorial volume. Proceedings of the memorial conference, Szeged, Hungary, August 2 -- 6, 1999, pages 465 -- 472. Basel: Birkhauser, 2001. [13] A.L.T. Paterson. Groupoids, inverse semigroups, and their operator algebras., volume 170. Boston, MA: Birkhauser, 1999. [14] K. Thomsen. The universal property of equivariant KK-theory. J. Reine Angew. Math., 504:55 -- 71, 1998. [15] J.L. Tu. Non-Hausdorff groupoids, proper actions and K-theory. Doc. Math., 9:565 -- 597, 2004. E-mail address: [email protected]
1603.02105	5	1603	2017-03-17T14:21:14	A new bicommutant theorem	[ "math.OA", "math.LO" ]	We prove an analogue of Voiculescu's theorem: Relative bicommutant of a separable unital subalgebra $A$ of an ultraproduct of simple unital C*-algebras is equal to $A$.	math.OA	math	A NEW BICOMMUTANT THEOREM I. FARAH Abstract. We prove an analogue of Voiculescu's theorem: Relative bicommutant of a separable unital subalgebra A of an ultraproduct of simple unital C∗-algebras is equal to A. Ultrapowers1 AU of separable algebras are, being subject to well-developed model-theoretic methods, reasonably well-understood (see e.g. [12, Theo- rem 1.2] and §2). Since the early 1970s and the influential work of McDuff and Connes central sequence algebras A′ ∩ AU play an even more important role than ultrapowers in the classification of II1 factors and (more recently) C∗-algebras. While they do not have a well-studied abstract analogue, in [12, Theorem 1] it was shown that the central sequence algebra of a strongly self-absorbing algebra ([27]) is isomorphic to its ultrapower if the Continuum Hypothesis holds. Relative commutants B′ ∩ DU of separable subalgebras of ultrapowers of strongly self-absorbing C∗-algebras play an increasingly important role in classification program for separable C∗-algebras ([22, §3], [5]; see also [26], [29]). In the present note we make a step towards better understanding of these algebras. C∗-algebra is primitive if it has representation that is both faithful and ir- reducible. We prove an analogue of the well-known consequence of Voiculescu's theorem ([28, Corollary 1.9]) and von Neumann's bicommutant theorem ([4, §I.9.1.2]). A is a separable C∗-subalgebra. In addition assume A is a unital subalgebra if computed in the ultraproduct of faithful Theorem 1. Assume QU Bj is an ultraproduct of primitive C∗-algebras and QU Bj is unital. Then (with A A = (cid:18)A′ ∩YU irreducible representations of Bjs) Bj(cid:19)′ WOT = A WOT Bj. ∩YU A slightly weaker version of the following corollary to Theorem 1 (stated here with Aaron Tikuisis's kind permission) was originally proved by using very different methods (Z(A) denotes the center of A). Date: October 17, 2018. 1991 Mathematics Subject Classification. 46L05, 03C20, 03C98. 1Throughout U denotes a nonprincipal ultrafilter on N. 1 2 I. FARAH simple unital C∗-algebras and A is a separable unital subalgebra. Then (cid:3) Corollary 2 (Farah -- Tikuisis, 2015). Assume QU Bj is an ultraproduct of Z(A′ ∩QU Bj) = Z(A). At least two open problems are concerned with bicommutants of separable subalgebras of massive operator algebras. As is well-known, central sequence algebras M ′ ∩ M U of II1 factors in tracial ultrapowers behave differently from the central sequence algebras of C∗-algebras. For a II1 factor M with separable predual the central sequence algebra M ′ ∩ M U can be abelian or even trivial. Popa conjectured that if P is a separable subalgebra of an ultraproduct of II1 factors then (P ′ ∩ QU Ni)′ = P implies P is amenable ([25, Conjecture 2.3.1]). In the domain of C∗-algebras, Pedersen asked ([24, Remark 10.11]) whether the following variant of Theorem 1 is true: If the corona M (B)/B of a σ-unital C∗-algebra B is simple and A is a separable unital subalgebra, is (A′ ∩ M (B)/B)′ = A? (For the connection between ultraproducts and coronas see the last paragraph of §3.) The proof of Theorem 1 uses logic of metric structures ([3], [14]) and an analysis of the interplay between C∗-algebra B and its second dual B∗∗. Acknowledgments. Theorem 1 was inspired by conversations with Aaron Tikuisis and Stuart White and I use this opportunity to thank them. Corol- lary 2 was proved during a very inspirational visit to the University of Ab- erdeen in July 2015. I am indebted to Aaron Tikuisis for warm hospitality and stimulating discussions and to the London Mathematical Society for funding my visit to Aberdeen. The original proof of a weaker form of The- orem 1 was presented in a three-hour seminar at the Fields Institute in February 2016. I would like to thank the audience, and to George Elliott and Alessandro Vignati in particular, for numerous sharp observations. Af- ter the completion of the present paper Stuart White and Dan Voiculescu pointed out that its results are related to Hadwin's asymptotic double com- mutant theorem ([18], see also [16] and [17]), and Martino Lupini pointed out that Theorem 1 also holds in the nonunital case. I am indebted to the anonymous referee for several useful remarks. Last, but not least, I would like to thank Leonel Robert for pointing out an error in an early draft. 1. Model theory of representations We expand the language of C∗-algebras ([14, §2.3.1]) to representations of C∗-algebra. Reader's familiarity with, or at least easy access to, [14, §2] is assumed. A structure in the expanded language Lrep is a C∗-algebra together with its representation on a Hilbert space. As in [14], the domains of quantification on C∗-algebra are Dn for n ∈ N and are interpreted as the n-balls. The domains of quantification on Hilbert space are DH n for n ∈ N and are also interpreted as the n-balls. On all domains the metric is d(x, y) = kx − yk (we shall denote both the operator norm on C∗-algebra and the ℓ2-norm on Hilbert space by k · k). As in [14, §2.3.1], for every λ ∈ C we have a unary function symbol λ to be interpreted as multiplication A NEW BICOMMUTANT THEOREM 3 n to DH by λ. We also have binary function + whose interpretation sends DH m × DH m+n. As the scalar product (··) is definable from the norm via the polarization identity, we shall freely use it in our formulas, with the understanding that (ξη) is an abbreviation for 1 j=0 ijkξ + ijηk. 4 P3 Language Lrep also contains a binary function symbol π whose interpre- mn for all m and n. It is interpreted as an action m to DH tation sends Dn × DH of A on H. Every variable is associated with a sort. In particular variables x, y, z range over the C∗-algebra and variables ξ, η, ζ range over the Hilbert space, all of them decorated with subscripts when needed. We shall write ¯x for a tuple ¯x = (x1, . . . , xn) (with n either clear from the context or irrelevant). Terms come in two varieties. On the C∗-algebra side, term is a noncommutative ∗-polynomial in C∗-variables. On the Hilbert space side terms are linear combinations of Hilbert space variables and ex- pressions of the form π(α(¯x))ξ where α(¯x) is a term in the language of C∗- algebras. Formulas are defined recursively. Atomic formulas are expressions of the form ktk where t is a term. The set of all formulas is the smallest set F containing all atomic formulas with the following properties. (i) for every n, all continuous f : [0, ∞)n → [0, ∞) and all ϕ1, . . . , ϕn in F the expression f (ϕ1, . . . , ϕn) belongs to F, and (ii) if ϕ ∈ F and x, ξ are variable symbols then each of supkξk≤m ϕ, inf kξk≤m ϕ, supkxk≤m ϕ and infkxk≤m ϕ belongs to F (see [14, §2.4] or [11, Definition 2.1.1]). Suppose π : A → B(H) is a representation of a C∗-algebra A on Hilbert space H. To (A, H, π) we associate the natural metric structure M(A, H, π) in the above language. Suppose ϕ(¯x, ¯ξ) is a formula whose free variables are included among ¯x and ¯ξ. If π : A → B(H) is a representation of a C∗-algebra on Hilbert space, ¯a are elements of A and ¯ξ are elements of H,2 then the interpretation ϕ(¯a, ¯ξ)M(A,H,π) is defined by recursion on the complexity of ϕ in the obvious way (see [3, §3]). Proposition 1.1. Triples (A, H, π) such that π is a representation of A on H form an axiomatizable class. Proof. As in [14, Definition 3.1], we need to define a Lrep-theory Trep such that that the category of triples (A, H, π) where π : A → B(H) is a repre- sentation of a C∗-algebra A is equivalent to the category of metric structures that are models of Trep, via the map (A, H, π) 7→ M(A, H, π). 2Symbols ξ, η, ζ,. . . denote both Hilbert space variables and vectors in Hilbert space due to the font shortage; this shall not lead to a confusion. 4 I. FARAH We use the axiomatization of C∗-algebras from [14, §3.1]. In addition to the standard Hilbert space axioms, we need two axioms assuring that the interpretation of DH n equals the n-ball of the underlying Hilbert space for all n: sup ξ∈Dn and (writing s .− t := max(s − t, 0)), kξk ≤ n () supξ∈Dn(1 .− kξk) inf η∈D1 kξ − ηk. The standard axioms, (π(x)ξη) = (ξπ(x∗)η) π(xy)ξ = π(x)π(y)ξ, are expressible as first-order sentences.3 The axioms described here comprise theory Trep. π(x + y)ξ = π(x)ξ + π(y)ξ One needs to check that the category of models of Trep is equivalent to the category of triples (A, H, π). Every triple (A, H, π) uniquely defines a model M(A, H, π). Conversely, assume M is a model of Trep. The algebra AM obtained from the first component of M is a C∗-algebra by [14, Proposi- tion 3.2]. Also, the linear space HM obtained from the second component of M is a Hilbert space and the third component gives a representation πM of A on H. To see that this provides an equivalence of categories, we need to check that M(AM, HM, πM) ∼= M for every model M of Trep. We need to show that the domains on M are determined by AM and HM. The former was proved in the second paragraph of [14, Proposition 3.2], and the latter follows by (). (cid:3) Proposition 1.1 gives us full access to the model-theoretic toolbox, such as the Lo´s' theorem (see §2) and the Lowenheim -- Skolem Theorem ([14, Theorem 4.6]). From now on, we shall identify triple (A, H, π) with the associated metric structure M(A, H, π) and stop using the latter notation. We shall also write supkξk≤n and infkξk≤n instead of supξ∈Dn and inf ξ∈Dn, respectively. Lemma 1.2. The following properties of a representation π of A are ax- iomatizable: (1) π is faithful (2) π is irreducible. Proof. We shall explicitly write the axioms for each of the properties of π. Fix a representation π. It is faithful if and only if it is isometric, which can be expressed as sup kxk≤1 inf kξk≤1 kxk − kπ(x)ξk = 0. 3Our conventions are as described in [14, p. 485]. In particular α(x, ξ) = β(x, ξ) is an abbreviation for supξ∈Dn supξ kα(x, ξ) − β(x, ξ)k = 0, for all n. A NEW BICOMMUTANT THEOREM 5 Representation π is irreducible if and only if for all vectors ξ and η in H such that kηk ≤ 1 and kξk = 1 the expression kη − π(a)ξk can be made arbitrarily small when a ranges over the unit ball of A. In symbols sup kξk≤1 sup kηk≤1 inf kxk≤1 kξk .− 1kη − π(x)ξk = 0. The interpretation of this sentence in (A, H, π) is 0 if and only if π is irre- ducible. (cid:3) Triple (D, θ, K) is an elementary submodel of (B, π, H), and (B, π, H) is an elementary extension of (D, θ, K, if D ⊆ B, K ⊆ H θ(d) = π(d) ↾ H for all d ∈ D, and ϕ(¯a)(D,θ,K) = ϕ(¯a)(B,π,H) for all formulas ϕ and all ¯a in (D, θ, K) of the appropriate sort. Axiomati- zable properties, such as being irreducible or faithful, transfer between ele- mentary submodels and elementary extensions. Therefore the Downwards Lowenheim -- Skolem Theorem ([14, Theorem 4.6]) and Lemma 1.2 together imply e.g. that if ϕ is a pure state of a nonseparable C∗-algebra B then B is an inductive limit of separable subalgebras D such that the restriction of ϕ to D is pure. This fact was proved in [1] and its slightly more precise version will be used in the proof of Lemma 3.2. Some other properties of representations (such as not being faithful) are axiomatizable but we shall concentrate on proving Theorem 1. 2. Saturation and representations It was known to logicians since the 1960s that the two defining properties of ultraproducts associated with nonprincipal ultrafilters on N in axioma- tizable categories are Lo´s' Theorem ([14, Proposition 4.3]) and countable saturation ([14, Proposition 4.11]). By the former, the diagonal embedding of metric structure M into its ultrapower is elementary. More generally, if ϕ(¯x) is a formula and ¯a(j) ∈ Mj are of the appropriate sort then ϕ(¯a)QU Mj = lim j→U ϕ(¯a(j))Mj . In order to define countable saturation, we recall the notion of a type from the logic of metric structures ([14, §4.3]). A closed condition (or simply a condition; we shall not need any other conditions) is any expression of the form ϕ ≤ r for formula ϕ and r ≥ 0 and type is a set of conditions ([14, §4.3]). As every expression of the form ϕ = r is equivalent to the condition max(ϕ, r) ≤ r and every expression of the form ϕ ≥ r is equivalent to the condition min(0, r − ϕ) ≤ 0, we shall freely refer to such expressions as conditions. For m and n in N such that m + n ≥ 1, an (m, n)-type is a type t such that all free variables occurring in conditions of t are among {x1, . . . , xm} ∪ {ξ1, . . . , ξn}. Given a structure (A, H, π) and a subset X of A ∪ H, we expand the language Lrep by adding constants for the elements of X (as in [14, §2.4.1]). 6 I. FARAH The new language is denoted (Lrep)X . C∗-terms in (Lrep)X are ∗-polynomials in C∗-variables and constants from X ∩ A. Hilbert space terms are linear combinations of Hilbert space variables, constants in X ∩H, and expressions of the form π(α)ξ where α is a C∗-term in the expanded language. The interpretation of a (Lrep)X -formula is defined recursively in the natural way (see e.g. [11], paragraph after Definition 2.1.1). Type over X is a type in (Lrep)X. Such type is realized in some ele- mentary extension of (A, H, π) if the latter contains a tuple satisfying all conditions from the type. A type is consistent if it is realized in some ul- trapower of (A, H, π), where the ultrafilter is taken over an arbitrary, not necessarily countable set. This is equivalent to the type being realized in some elementary extension of (A, H, π). By Lo´s' Theorem, type t is consistent if and only every finite subset of t is ε-realized in (A, H, π) for every ε > 0 ([14, Proposition 4.8]). A structure (A, H, π) is said to be countably saturated if every consistent type over a countable (or equivalently, norm-separable) set is realized in (A, H, π). Ultraproducts associated with nonprincipal ultrafilters on N are always countably saturated ([14, Proposition 4.11]). A standard transfinite back-and-forth argument shows that a structure of density character ℵ1 is countably saturated if and only if it is an ultraproduct. (Density character is the smallest cardinality of a dense subset.) In the case when A = B(H) we have (B(H), H)U = (B(H)U , H U ), in particular B(H)U is identified with a subalgebra of B(H U ). These two algebras are equal (still assuming U is a nonprincipal ultrafilter on N) if and only if H is finite-dimensional. As a matter of fact, no projection p ∈ B(H U ) with a separable, infinite-dimensional, range belongs to B(H)U (this is proved by a standard argument, see e.g. the last two paragraphs of the proof of [13, Proposition 4.6]). In the following, π will always be faithful and clear from the context and we shall identify A with π(A) and suppress writing π. We shall therefore write (A, H) in place of (A, H, id). The following two lemmas are standard (they were used by Arveson in the proof of Corollary 2 on p 344 of [2]) but we sketch the proofs for the reader's convenience. Lemma 2.1. Suppose A is a C∗-algebra and ϕ is a functional on A. Then there are a representation π : A → B(K) and vectors ξ and η in K such that ϕ(a) = (π(a)ξη) for all a ∈ A. Proof. Let ¯ϕ be the unique extension of ϕ to a normal functional of the von Neumann algebra A∗∗. By Sakai's polar decomposition for normal linear functionals (see e.g. [23, Proposition 3.6.7]) there exists a normal state ψ of A∗∗ and a partial isometry v such that ϕ(a) = ψ(av) for all a ∈ A∗∗. Let π : A∗∗ → B(K) be the GNS representation corresponding to ψ. If η is the corresponding cyclic vector and ξ = vη, then the restriction of π to A is as required. (cid:3) A NEW BICOMMUTANT THEOREM 7 Lemma 2.2. Suppose A is a proper unital subalgebra of C = C∗(A, b). Then there exists a representation π : C → B(K) and a projection q in π(A)′ ∩ B(K) such that [q, b] 6= 0. Proof. By the Hahn -- Banach separation theorem there exists a functional ϕ on C of norm 1 such that ϕ annihilates A and ϕ(b) = dist(A, b). Let π : C → B(K), η and ξ be as guaranteed by Lemma 2.1. Let L be the norm-closure of π(A)ξ. Since A is unital L 6= {0}. As 0 = ϕ(a) = (π(a)ξη) for all a ∈ A, η is orthogonal to L and therefore the projection p to L is nontrivial. Clearly p ∈ π(A)′ ∩ B(K). Since (π(b)ξη) = ϕ(b) 6= 0, π(b) does not commute with p and we therefore have q ∈ π(A)′ ∩ B(K) such that k[π(b), q]k > 0. (cid:3) The proof of Theorem 1 would be much simpler if Lemma 2.2 provided an irreducible representation. This is impossible in general as the following example shows. Let A be the unitization of the algebra of compact opera- tors K(H) on an infinite-dimensional Hilbert space and let b be a projection in B(H) Murray-von Neumann equivalent to 1 − b. Then C = C∗(A, b) has (up to equivalence) three irreducible representations. Two of those repre- sentations annihilate A and send b to a scalar, and the third representation is faithful and the image of b is in the weak operator closure of the image of A. It is well-known that for a Banach space X the second dual X ∗∗ can be embedded into an ultrapower of X ([19, Proposition 6.7]). In general, the second dual A∗∗ of a C∗-algebra A cannot be embedded into an ultrapower of A for at least two reasons. First, A∗∗ is a von Neumann algebra ([4, §III.5.2]) and it therefore has real rank zero, while A may have no nontrivial projections at all. Since being projectionless is axiomatizable ([11, Theo- rem 2.5.1]), if A is projectionless then Lo´s's Theorem implies that AU is projectionless as well and A∗∗ cannot be embeded into it. The anonymous referee pointed out an another, much subtler, obstruction. In the context of Banach spaces, the embeddability of X ∗∗ into X U is equivalent to a finitary statement, the so-called local reflexivity of Banach spaces the C∗-algebraic version of which does not hold for all C∗-algebras [8, §5]. In particular, for a large class of C∗-algebras the diagonal embedding of A into AU cannot be extended even to a unital completely positive map from A∗∗ into AU . The anonymous referee also pointed out that a result of J. M. G. Fell is closely related to results of the present section. It is a standard fact that a repre- sentation of a discrete group is weakly contained in another representation of the same group if and only if it can be embedded into an ultrapower of the direct sum of infinitely many copies of the latter representation. In [15, Theorem 1.2] it was essentially proved that this equivalence carries over to arbitrary C∗-algebras. All this said, Lemma 2.3 below is a poor man's C∗-algebraic variant of the fact that Banach space X ∗∗ embeds into X U . As in [23, 3.3.6], we say that 8 I. FARAH two representations π1 and π2 of A are said to be equivalent if the identity map on A extends to an isomorphism between π1(A)′′ and π2(A)′′. Lemma 2.3. Assume (QU Bj,QU Hj) is an ultraproduct of faithful irre- ducible representations of unital C∗-algebras and C is a unital separable subalgebra of BU . (1) If C ∩ K(QU Hj) = {0} then the induced representation of C on QU Hj is equivalent to the universal representation of C. (2) In general, if p = _{q : q is a projection in C ∩ K(YU Hj)} then p ∈ C ′∩B(QU Hj) and c 7→ (1−p)c is equivalent to the universal representation of C/(C ∩ K(QU Hj)) on (1 − p)QU Hj. Proof. For a state ψ on C the (0, 1)-type tψ(ξ) of a vector ξ implementing ψ consists of all conditions of the form (aξξ) = ψ(a) for a ∈ C and kξk = 1. (1) Fix a state ψ on C. By Glimm's Lemma ([6, Lemma II.5.1]) type tψ is consistent with the theory of (QU Bj,QU Hj). By the separability of C and countable saturation, there exists a unit vector η ∈ QU Hj such that ψ(c) = (cηη) for all c ∈ C. Let L be the norm-closure of Cη in QU Hj. Then L is a reducing subspace for C and the induced representation of C on L is spatially isomorphic to the GNS representation of C corresponding to ψ. Since ψ was arbitrary, by [23, Theorem 3.8.2] this completes the proof. (2) For every a ∈ C we have pa ∈ C ∩K(QU Hj) and therefore pa(1−p) = 0 Similarly (1−p)ap = 0, and hence p ∈ C ′∩B(QU Hj). Let pn, for n ∈ N, be a maximal family of orthogonal projections in C ∩K(QU Hj). It is countable by the separability of C and p = Wn pn. Let ψ be a state of C that annihilates C ∩ K(QU Hj). Let t+ ψ (ξ) be the type obtained from tψ(ξ) by adding to it all conditions of the form pnξ = 0 for n ∈ N. By Glimm's Lemma (as stated in [6, Lemma II.5.1]) the type t+ ψ (ξ) is consistent, and by the countable saturation we can find ξ1 ∈ QU Hj that realizes this type. Then pξ1 = 0 and therefore ξ1 ∈ (1 − p)QU Hj. Therefore every GNS representation of C/(C ∩ K(QU Hj)) is spatially equivalent to a subrepresentation of c 7→ (1 − p)c, and by [23, Theorem 3.8.2] this concludes the proof. (cid:3) 3. Second dual and Day's trick The natural embedding of a C∗-algebra B into its second dual B∗∗ is rarely elementary. For example, having real rank zero is axiomatizable ([11, Theorem 2.5.1]) and B∗∗, being a von Neumann algebra, has real rank zero while B may have no nontrivial projections at all. However, we shall see that there is a restricted degree of elementarity between B and B∗∗, and it will suffice for our purposes. We shall consider the language (Lrep)B obtained by adding new constants for parameters in B (see §2). Term α(x) in the extended language is linear A NEW BICOMMUTANT THEOREM 9 if it is of the form α(x) = xa + bx for some parameters a and b. A restricted B-linear formula is a formula of the form (1) maxj≤m kαj(x) − bjk + maxj≤n(rj .− kβj (x)k) where (2) all bj, for 1 ≤ j ≤ m are parameters in B, (3) all rj, for 1 ≤ j ≤ n are positive real numbers, (4) all αj for 1 ≤ j ≤ m are linear terms with parameters in B, and (5) all βj for 1 ≤ j ≤ n are linear terms with parameters in B. Proof of the following is based on an application of the Hahn -- Banach sep- aration theorem first used by Day ([7]; see also [9, Section 2] for some uses of this method in the theory of C∗-algebras). Lemma 3.1. Suppose B is a unital C∗-algebra and γ(x) = max j≤m kαj(x) − bjk + max j≤n (rj .− kβj(x)k). is a restricted B-linear formula. Then the following are equivalent. (6) inf x∈B γ(x) = 0 (7) inf x∈B∗∗ γ(x) = 0. Proof. (6) implies (7) because B is isomorphic to a unital subalgebra of B∗∗ and therefore inf x∈B∗∗ γ(x) ≤ inf x∈B γ(x). Assume (7) holds. Let aj and cj, for j ≤ n, be such that αj(x) = ajx+xcj. For each j we shall identify αj with its interpretation, linear map from B to B. The second adjoint α∗∗ j (x) = ajx + xcj, hence α∗∗ j (x) is the interpretation of the term αj(x) in B∗∗. The set j : B∗∗ → B∗∗ also satisfies α∗∗ Z := hαj(x) : x ∈ B≤1i is, being an image of a convex set under a linear map, a convex subset of Bm and by the Hahn -- Banach theorem Z1 := Bm ∩ hαj(x) : x ∈ B∗∗ ≤1i is included in the norm-closure of Z. By (7) we have (b1, . . . , bm) ∈ Z1. Fix ε > 0 and let X1 := {x ∈ B≤1 : maxj≤m kαj(x) − bjk ≤ ε}. By the above this is a convex subset of the unit ball of B and (by using the Hahn -- Banach separation theorem again) the weak∗-closure of X1 in B∗∗ is equal to {x ∈ B∗∗ ≤1 : maxj≤m kαj(x) − bjk ≤ ε}. Let c ∈ B∗∗ ≤1 be such that γ(c) < ε. Then c belongs to the weak∗-closure of X1. For each j ≤ n we have kβj (c)k > rj − ε. Fix a norming functional ϕj ∈ B∗ such that kϕjk = 1 and ϕj(βj(c)) > rj − ε. Then U := {x ∈ B∗∗ : ϕj(βj(x)) > rj − ε for all j} 10 I. FARAH is a weak∗-open neighbourhood of c and, as c belongs to the weak∗-closure of X1, U ∩ X1 is a nonempty subset of B≤1. Any b ∈ U ∩ X1 satisfies γ(b) < ε. As ε > 0 was arbitrary, this shows that (6) holds. (cid:3) Lemma 3.2. Suppose (Bj, Hj) is an irreducible representation of Bj on Hj In the following A ⊆ QU Bj is identified with a subalgebra of B(QU Hj). for j ∈ N and A is a separable subalgebra of QU Bj. (1) For every b ∈ QU Bj we have b ∈ (A′ ∩ B(QU Hj))′ if and only if b ∈ (A′ ∩QU Bj)′. Equivalently, (A′ ∩ B(QU Hj))′ ∩QU Bj = (A′ ∩QU Bj)′ ∩QU Bj. WOT (2) A ∩QU Bj = (A′ ∩QU Bj)′. Proof. (1) Since QU Bj ⊆ B(QU Hj) we clearly have (A′ ∩ B(QU Hj))′ ⊆ (A′ ∩QU Bj)′. In order to prove the converse inclusion, fix b ∈ QU Bj and suppose that there exists q ∈ A′ ∩ B(QU Hj) such that k[q, b]k = r > 0. We need to find d ∈ A′ ∩QU Bj satisfying [d, b] 6= 0. Consider the (1, 0)-type t(x) consisting of all conditions of the form k[x, b]k ≥ r and [x, a] = 0 for a ∈ A. This type is satisfied in B(QU Hj) by q. Since all formulas in t(x) are quantifier-free, their interpretation is unchanged when passing to a larger algebra. Fix a finite subset of t(x) and let F ⊆ A be the set of parameters occurring in this subset. Then γF (x) := inf x max a∈F k[x, a]k + (r .− k[x, b]k) projection to H0. By the Downward Lowenheim -- Skolem Theorem ([14, Theorem 4.6]) there satisfies k[q1, b]k = r. To find p as required, take a separable elementary is a restricted QU Bj-linear formula. Since A is separable, we can find projection p in C∗(A, q)′ ∩B(QU Hj) with separable range such that q1 := pq submodel (C, H0) of (B(QU Hj),QU Hj) such that A ⊆ C and let p be the exists a separable elementary submodel (D, K) of (QU Bj,QU Hj) such = B(QU Hj) and Lemma 1.2 and Lo´s' Theorem imply that QU Bj = B(pK QU Hj). (with pK denoting the projection to K) that pKDpK We can therefore identify pK with a minimal central projection in D∗∗. Via this identification we have q1 ∈ D∗∗. Since γF (q1) = 0, Lemma 3.1 im- plies inf x∈D,kxk≤1 γF (x) = 0 and inf x∈QU Bj ,kxk≤1 γF (x) = 0 (since γF is quantifier-free). that C∗(A, b) ⊆ D and the range of p is included in K. Part (2) of WOT WOT Since F was an arbitrary finite subset of A type t(x) is consistent with the theory of QU Bj. Since A is separable, by the countable saturation there exists d ∈ A′ ∩QU Bj satisfying k[d, b]k ≥ r. A NEW BICOMMUTANT THEOREM 11 (2) By the von Neumann bicommutant theorem A WOT and therefore (1) implies A WOT ∩QU Bj = (A′ ∩QU Bj)′. 4. Proof of Theorem 1 = (A′∩B(QU Hj))′ (cid:3) Suppose (Bj, Hj) is a faithful irreducible representation of Bj on Hj WOT By (2) of Lemma 3.2 we have A such that r := dist(b, A) > 0. By (1) of Lemma 3.2 it suffices to find for j ∈ N and A is a separable subalgebra of QU Bj. By Lemma 1.2 (QU Bj,QU Hj) is an irreducible faithful representation of QU Bj. ∩ QU Bj = (A′ ∩ QU Bj)′. Since A ⊆ (A′ ∩QU Bj)′, it remains to prove (A′ ∩QU Bj)′ ⊆ A. Fix b ∈ QU Bj d ∈ A′ ∩ B(QU Hj) such that [d, b] 6= 0. Let Lemma 4.1. With QU Bj, A, b, C and r as above, there exists a repre- sentation π : C/(C ∩ K(QU Hj)) → B(K) and q ∈ π(A)′ ∩ B(K) such that C := C∗(A, b). [q, π(b)] 6= 0. Since the proof of Lemma 4.1 is on the long side, let us show how it completes the proof of Theorem 1. Lemma 2.3 implies that if p = W{q : q is a projection in C ∩ K(QU Hj)} then p ∈ C ′ ∩ B(QU Hj) and c 7→ (1 − p)c is equivalent to the universal representation of C/(C ∩ K(QU Hj)) on (1 − p)QU Hj. Therefore q as in the conclusion of Lemma 4.1 can be found in A′ ∩ B(QU Hj), implying b /∈ (A′ ∩ B(QU Hj))′. By Lemma 3.2 this implies b /∈ (A′ ∩ QU Bj)′, reducing the proof of Theorem 1 to the following. Proof of Lemma 4.1. An easy special case may be worth noting. If C ∩ π : C → B(K) and q ∈ π(A)′ ∩ B(K) such that [q, π(b)] 6= 0. is countable (and possibly finite or even empty) since A is separable. Let In the general case, let qn, for n ∈ J, be an enumeration of a maximal K(QU Hj) = {0} then Lemma 2.2 implies the existence of a representation orthogonal set of minimal projections in A ∩ K(QU Hj). The index-set J pn := Wj≤n qj. Suppose for a moment that there exists n such that pnbpn /∈ A. Since the range of pn is finite-dimensional, by von Neumann's Bicommutant The- orem ([4, §I.9.1.2]). and the Kadison Transitivity Theorem ([4, Theorem II.6.1.13]) there exists d ∈ A′ ∩ B(pnQU Hj) such that [d, b] 6= 0. Lemma 3.2 now implies pnbpn /∈ (A′ ∩QU Bj)′ and b /∈ (A′ ∩QU Bj)′. We may therefore assume pnbpn ∈ A for all n. Let p = Wn pn. Lemma 2.3 (2) implies p ∈ A′ ∩ B(QU Hj), and we may therefore assume [b, p] = 0. Since C = C∗(A, b) this implies p ∈ C ′ ∩ B(QU Hj). Since pnbpn ∈ A for all n we have A ∩ K(QU Hj)pCp ∩ K(QU Hj). If c ∈ C then for every n pn, for n ∈ N, is an approximate unit for A ∩ K(QU Hj), the latter is we have pnc(1 − p) = 0 and similarly (1 − p)cpn = 0. Since the sequence 12 I. FARAH an ideal of C. Let θ : C → C/(A ∩ K) be the quotient map. We claim that dist(θ(b), θ(A)) = dist(b, A) > 0. Fix a ∈ A. We need to show that kθ(a − b)k ≥ r. Consider the (0, 1)-type t(ξ) consisting of all conditions of the form kξk = 1, k(a − b)ξk ≥ r, pnξ = 0, for n ∈ J. To see that this type is consistent fix a finite F ⊆ J. Let m ≥ max(F ) and a′ := (1 − pm)a(1 − pm) + pmbpm. As both summands belong to A, a′ ∈ A and therefore ka′ − bk ≥ r. Fix then ξ′ = (1 − pm)ξ has the same property since (a′ − b)pm = 0. Since ε > 0 was arbitrary, t(ξ) is consistent. By the countable saturation there exists a ε > 0. If ξ ∈ QU Hj is a vector of norm ≤ 1 such that k(a′ − b)ξk > r − ε unit vector ξ ∈ QU Hj which realizes t(ξ). Since pnξ = 0 for all n we have Suppose for a moment that (1 − p)C(1 − p) ∩ K(QU Hj) = {0}. By (2) of pξ = 0 and therefore kθ(a − b)k ≥ k(1 − p)(a − b)(1 − p)k ≥ r. Since a ∈ A was arbitrary, we conclude that dist(θ(b), θ(A)) = r. Lemma 2.3 the representation is equivalent to the universal representation of C. Hence by Lemma 2.2 we C ∋ c 7→ (1 − p)c ∈ B((1 − p)QU Hj) the above this concludes the proof in this case. can find d ∈ (1 − p)(A′ ∩ B(QU Hj)) that does not commute with b, and by We may therefore assume that (1 − p)C(1 − p) ∩ K(QU Hj) 6= {0}. By the spectral theorem for self-adjoint compact operators and continuous func- tional calculus, there exists a nonzero projection q ∈ (1 − p)C(1 − p) of finite rank. Fix c ∈ C such that (1 − p)c(1 − p) = q. WOT By Lemma 3.2 it suffices to find q ∈ A′ ∩ (1 − p)B(QU Hj)(1 − p) such that [q, c] 6= 0. Suppose otherwise, that c ∈ (A′ ∩QU Bj)′. (2) of Lemma 3.2 implies that c ∈ A . By the Kaplansky Density Theorem ([4, Theorem I.9.1.3]) there is a net of positive contractions in A converging to c in the weak operator topology. By the continuous functional calculus and the Kadison Transitivity Theorem ([4, Theorem II.6.1.13]) we may choose this net among the members of Z := {a ∈ A+ : kak = 1, qaq = q}. Consider the (0, 1)-type t1(ξ) consisting of all conditions of the form kξk = 1, aξ = ξ, qξ = 0, pnξ = 0 for n ∈ N and a ∈ Z. We claim that t1(ξ) is consistent. Fix ε > 0 and a1, a2, . . . , an in Z. Let a := a1a2 . . . an−1anan−1 . . . a2a1. Then a ∈ Z and q ≤ a. By the choice of p the operator (1 − p)(a − s)+ is not compact for any s < 1. Therefore there exists a unit vector ξ0 in A NEW BICOMMUTANT THEOREM 13 (1 − p − q)QU Hj such that kξ0 − aξ0k is arbitrarily small. By the countable saturation there exists a unit vector ξ1 ∈ (1 − (p + q))QU Hj such that aξ1 = ξ1. As each aj is a positive contraction, we have ajξ1 = ξ1 for 1 ≤ j ≤ n. Since a1, . . . , an was an arbitrary subset of Z, this shows that t1(ξ) is consistent. aξ = ξ for all a ∈ Z. As cξ = 0, this contradicts c being in the weak operator topology closure of Z. Since Z is separable, by the countable saturation there exists ξ ∈ QU Hj realizing t1(ξ). Then ξ is a unit vector in (1 − (p + q))QU Hj such that Therefore there exists q ∈ A′ ∩(1−p)B(QU Hj)(1−p) such that [q, c] 6= 0. Since c ∈ C = C∗(A, b) we have [q, b] 6= 0, and this concludes the proof. (cid:3) 5. Concluding remarks In the following infinitary form of the Kadison Transitivity Theorem pK denotes projection to a closed subspace K of QU Hj. Proposition 5.1. Assume (QU Bj,QU Hj) is an ultraproduct of faithful separable closed subspace of QU Hj and T ∈ B(K). and irreducible representations of unital C∗-algebras. Also assume K is a (2) If T is self-adjoint (or positive, or unitary in B(K)) then b can be (1) There exists b ∈ QU Bj such that kbk = kT k and pKbpK = T . chosen to be self-adjoint (or positive, or unitary in B(QU Hj)). Proof. (1) is a consequence of the Kadison transitivity theorem and count- able saturation of the structure (QU Bj,QU Hj). Let pn, for n ∈ N, be an increasing sequence of finite-dimensional projections converging to pK in the strong operator topology and let an, for n ∈ N, be a dense subset of A. We need to check that the type t(x) consisting of all conditions of the form kpn(x − T )pnk = 0, kxk = kT k ducible by Lemma 1.2, every finite subset of t(x) is consistent by the Kadison for n ∈ N is consistent. Since the representation of QU Bj on QU Hj is irre- Transitivity Theorem. We can therefore find b ∈ QU Bj that satisfies t(x) and therefore pKbpK = T and kbk = kT k. (2) If T is self-adjoint, add the condition x = x∗ to t(x). By [23, The- orem 2.7.5] the corresponding type is consistent, and the assertion again follows by countable saturation. The case when T is a unitary also uses [23, Theorem 2.7.5]. (cid:3) An important consequence of full Voiculescu's theorem is that any two unital representations πj : A → B(H) of a separable unital C∗-algebra A on H such that ker(π1) = ker(π2) and π1(A) ∩ K(H) = π2(A) ∩ K(H) = {0} are approximately unitarily equivalent ([28, Corollary 1.4]). The analogous statement is in general false for the ultraproducts. Let Bn = Mn(C) for n ∈ N and let A = C2. Group K0(QU Mn(C)) is isomorphic to ZN with the natural ordering and the identity function id as the order-unit. Every 14 I. FARAH unital representation of A corresponds to an element of this group that lies between 0 and id, and there are 2ℵ0 inequivalent representations. Also, Z and 2ℵ0 of these K0(QU Mn(C)) is isomorphic to the ultraproduct QU extensions remain inequivalent even after passing to the ultraproduct. We return to G. K. Pedersen's question ([24, Remark 10.11]), whether a bicommutant theorem (A′ ∩ M (B)/B)′ = A is true for a separable unital subalgebra A of a corona M (B)/B of a σ-unital C∗-algebra B? A simple and unital C∗-algebra C is purely infinite if for every nonzero a ∈ C there are x and y such that xay = 1. Question 5.2. Suppose C is a unital, simple, purely infinite, and separable and A is a unital subalgebra of C. Is (A′ ∩ C ∗∗)′ ∩ C = A? Let us prove that a positive answer to Question 5.2 would imply a positive answer to Pedersen's question. If A is a separable and unital subalgebra of M (B)/B and b ∈ (M (B)/B) \ A, then there exists a separable elementary submodel C of M (B)/B containing b. By [21], M (B)/B is simple if and only if it is purely infinite, Since being simple and purely infinite is axiomatizable ([11, Theorem 2.5.1]), C is simple and purely infinite. If (A′ ∩ C ∗∗)′ ∩ C = A then Proposition 5.3 below implies that there exists d ∈ A′ ∩ M (B)/B such that [d, b] 6= 0. Proposition 5.3. Suppose B is a C∗-algebra, A is a separable subalgebra of B, b ∈ B and r ≥ 0. If B is an ultraproduct or a corona of a σ-unital, non-unital C∗-algebra then sup k[d, b]k = sup k[d, b]k. d∈(A′∩B)+,kdk≤1 d∈(A′∩B∗∗)+,kdk≤1 Proof. The only property of B used in the proof of Proposition 5.3 (given at the end of this section) is that of being countably degree-1 saturated ([10, Theorem 1]). Since B ⊆ B∗∗, it suffices to prove '≥' in the above inequality. Suppose b ∈ B and d ∈ (A′ ∩ B∗∗)+ are such that kdk = 1 and r .− k[b, d]k. Consider the type t(x) consisting of conditions kxk = 1, x ≥ 0, kxb − bxk ≥ r, and k[x, a]k = 0 for a in a countable dense subset of A, This is a countable degree-1 type. If φj = 0, for j < n, is a finite subset of t(x) then γ(x) := maxj<n φj(x) is a restricted B-linear formula and Lemma 2.3 implies that it is approximately satisfied in B. By the countable degree-1 saturation of B ([10, Theorem 1]) we can find a realization d′ of t(x) in B. Clearly d′ ∈ (A′ ∩ B)+, kd′k = 1, and k[d′, b]k ≥ r, completing the proof. (cid:3) Some information on a special case of Pedersen's conjecture can also be found in [20]. References [1] C. Akemann and N. Weaver, Consistency of a counterexample to Naimark's problem, Proc. Natl. Acad. Sci. USA 101 (2004), no. 20, 7522 -- 7525. [2] W. Arveson, Notes on extensions of C∗-algebras, Duke Math. J. 44 (1977), 329 -- 355. A NEW BICOMMUTANT THEOREM 15 [3] I. Ben Yaacov, A. Berenstein, C.W. Henson, and A. Usvyatsov, Model theory for metric structures, Model Theory with Applications to Algebra and Analysis, Vol. II (Z. Chatzidakis et al., eds.), London Math. Soc. Lecture Notes Series, no. 350, Cambridge University Press, 2008, pp. 315 -- 427. [4] B. Blackadar, Operator algebras, Encyclopaedia of Mathematical Sciences, vol. 122, Springer-Verlag, Berlin, 2006, Theory of C∗-algebras and von Neumann algebras, Operator Algebras and Non-commutative Geometry, III. [5] J. Bosa, N.P. Brown, Y. Sato, A.P. Tikuisis, S. White, and W. Winter, Covering dimension of C∗-algebras and 2-coloured classification, Memoirs of the Amer. Math. Soc. (to appear). [6] K.R. Davidson, C∗-algebras by example, Fields Institute Monographs, vol. 6, Ameri- can Mathematical Society, Providence, RI, 1996. [7] M. M. Day, Amenable semigroups, Illinois J. Math. 1 (1957), no. 4, 509 -- 544. [8] E. G. Effros and U. Haagerup, Lifting problems and local reflexivity for C∗-algebras, Duke Math. J. 52 (1985), 103 -- 128. [9] G. A. Elliott, Some C∗-algebras with outer derivations. III, Ann. Math. (2) 106 (1977), no. 1, 121 -- 143. [10] I. Farah and B. Hart, Countable saturation of corona algebras, C.R. Math. Rep. Acad. Sci. Canada 35 (2013), 35 -- 56. [11] I. Farah, B. Hart, M. Lupini, L. Robert, A. Tikuisis, A. Vignati, and W. Winter, Model theory of nuclear C∗-algebras, arXiv preprint arXiv:1602.08072 (2016). [12] I. Farah, B. Hart, M. Rørdam, and A. Tikuisis, Relative commutants of strongly self-absorbing C∗-algebras, Selecta Math. (to appear). [13] I. Farah, B. Hart, and D. Sherman, Model theory of operator algebras I: Stability, Bull. London Math. Soc. 45 (2013), 825 -- 838. [14] , Model theory of operator algebras II: Model theory, Israel J. Math. 201 (2014), 477 -- 505. [15] J. M. G. Fell, The dual spaces of C∗-algebras, Trans. Amer. Math. Soc. 94 (1960), no. 3, 365 -- 403. [16] D. Hadwin, Approximate double commutants in von neumann algebras, arXiv preprint arXiv:1108.5021 (2011). [17] D. Hadwin and J. Shen, Approximate double commutants and distance formulas, arXiv preprint arXiv:1210.5285 (2012). [18] D. W. Hadwin, An asymptotic double commutant theorem for C∗-algebras, Trans. Amer. Math. Soc. 244 (1978), 273 -- 297. [19] S. Heinrich, Ultraproducts in Banach space theory., J. Reine Angew. Math. 313 (1980), 72 -- 104. [20] D. Kucerovsky and G. A. Elliott, A relative double commutant theorem for hereditary sub-C∗-algebras, (2007), preprint, Centre de Recerca Matem`atica. [21] H. Lin, Simple corona C∗-algebras, Proc. Amer. Math. Soc. 132 (2004), no. 11, 3215 -- 3224. [22] H. Matui and Y. Sato, Decomposition rank of UHF-absorbing C∗-algebras, Duke Math. J. 163 (2014), no. 14, 2687 -- 2708. [23] G. K. Pedersen, C∗-algebras and their automorphism groups, London Mathemati- cal Society Monographs, vol. 14, Academic Press Inc. [Harcourt Brace Jovanovich Publishers], London, 1979. [24] , The corona construction, Operator Theory: Proceedings of the 1988 GPOTS- Wabash Conference (Indianapolis, IN, 1988), Pitman Res. Notes Math. Ser., vol. 225, Longman Sci. Tech., Harlow, 1990, pp. 49 -- 92. [25] S. Popa, Independence properties in subalgebras of ultraproduct II1 factors, J. Funct. Anal. 266 (2014), no. 9, 5818 -- 5846. [26] A. Tikuisis, S. White, and W. Winter, Quasidiagonality of nuclear C∗-algebras, An- nals of Math. (to appear). 16 I. FARAH [27] A.S. Toms and W. Winter, Strongly self-absorbing C ∗-algebras, Trans. Amer. Math. Soc. 359 (2007), no. 8, 3999 -- 4029. [28] D. V. Voiculescu, A non-commutative Weyl-von Neumann theorem, Rev. Roumaine Math. Pures Appl 21 (1976), no. 1, 97 -- 113. [29] W. Winter, QDQ vs. UCT, Abel Symposia 12: Operator Algebras and Applications: The Abel Symposium 2015, Springer, 2016, pp. 321 -- 342. Department of Mathematics and Statistics, York University, 4700 Keele Street, North York, Ontario, Canada, M3J 1P3 E-mail address: [email protected] URL: http://www.math.yorku.ca/∼ifarah

Subsets and Splits

No saved queries yet

Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.